Multiphysics Modelling and Simulation For Systems Design and Monitoring

Applied Condition Monitoring
Mohamed Haddar · Mohamed Slim Abbes

Jean-Yves Choley · Taoufik Boukharouba
Tamer Elnady · Andrei Kanaev
Mounir Ben Amar · Fakher Chaari Editors
Multiphysics Modelling
and Simulation for
Systems Design and
Monitoring
Proceedings of the Multiphysics Modelling
and Simulation for Systems Design Conference,
MMSSD 2014, 17–19 December, Sousse, Tunisia
Volume 2
Series editors
Mohamed Haddar, National School of Engineers of Sfax, Tunisia
Walter Bartelmus, Wroclaw University of Technology, Poland
Fakher Chaari, National School of Engineers of Sfax, Tunisia
e-mail: fakher.chaari@gmail.com
Radoslaw Zimroz, Wroclaw University of Technology, Poland
About this Series
The book series Applied Condition Monitoring publishes the latest research and
developments in the field of condition monitoring, with a special focus on indus-
trial applications. It covers both theoretical and experimental approaches, as well
as a range of monitoring conditioning techniques and new trends and challenges in
the field. Topics of interest include, but are not limited to: vibration measurement
and analysis; infrared thermography; oil analysis and tribology; acoustic emissions
and ultrasonics; and motor current analysis. Books published in the series deal with
root cause analysis, failure and degradation scenarios, proactive and predictive tech-
niques, and many other aspects related to condition monitoring. Applications con-
cern different industrial sectors: automotive engineering, power engineering, civil
engineering, geoengineering, bioengineering, etc. The series publishes monographs,
edited books, and selected conference proceedings, as well as textbooks for ad-
vanced students.
More information about this series at http://www.springer.com/series/13418

Mohamed Haddar · Mohamed Slim Abbes
Jean-Yves Choley · Taoufik Boukharouba
Tamer Elnady · Andrei Kanaev
Mounir Ben Amar · Fakher Chaari
Editors
Multiphysics Modelling
and Simulation for Systems
Design and Monitoring
Proceedings of the Multiphysics Modelling
and Simulation for Systems Design
Conference, MMSSD 2014, 17–19 December,
Sousse, Tunisia
ABC
Editors
Mohamed Haddar Tamer Elnady
National School of Engineers Faculty of Engineering
of Sfax ASU Sound and Vibration Lab
Sfax Ain Shams University
Tunisia Cairo
Egypt
Mohamed Slim Abbes
National School of Engineers Andrei Kanaev
of Sfax Laboratory of Materials Sci. & Process
Sfax University of Paris 13
Tunisia Villetaneuse
France
Jean-Yves Choley
Superior Engineering Institute Mounir Ben Amar
of Paris University of Paris 13
Saint-Ouen Laboratory of Materials Sci & Process
France Villetaneuse
France
Taoufik Boukharouba
Laboratory for Advanced Mechanics Fakher Chaari
University of Sciences and Technology National School of Engineers
Houari Boumediene (USTHB) of Sfax
Alger Sfax
Algeria Tunisia
ISSN 2363-698X ISSN 2363-6998 (electronic)

ISBN 978-3-319-14531-0 ISBN 978-3-319-14532-7 (eBook)
DOI 10.1007/978-3-319-14532-7
Library of Congress Control Number: 2014958074
Springer Cham Heidelberg New York Dordrecht London

c Springer International Publishing Switzerland 2015
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, express or implied, with respect to the material contained herein or for any
errors or omissions that may have been made.
Printed on acid-free paper
Springer International Publishing AG Switzerland is part of Springer Science+Business Media

(www.springer.com)
Preface
Multiphysics problems are usually encountered when the behaviour of a system is

affected by several physical phenomena. The interaction between these phenomena
leads to a completely different behaviour than the one resulting if each physical
phenomenon would be treated separately from the others. In recent years, studying
multiphysics systems has become a hot research field and has attracted researchers
from different areas.
This book is the proceedings of MMSSD 2014 (Multiphysics Modelling and Sim-
ulation for Systems Design and Monitoring) which was held on December 17–19,
2014, in Sousse, Tunisia. The first edition of the conference attracted about 140 par-
ticipants who had the opportunity to discuss a number of important topics in the field
of multiphysics modelling and simulation, and establish important collaborations. A
few chapters in this book are devoted to modeling and simulation of multiphysics
systems. They describes several kind of interactions, such as thermomechanical and
electromechanical interactions, fluid-structure interactions, acoustic-structure inter-
actions, and thermofluidic interactions. Specific contributions deal with mechatronic
design and their applications, for example, in the field of energy harvesting or in
robotics. Other contributions describe advances in systems engineering, mainly in
the area of complex systems design. They reports on the implementation of ad-
vanced numerical and analytical simulation methods, and include case studies. An-
other topic is concerning with material characterization of multiphysics systems.
The corresponding chapters discuss several structural applications, giving a special
emphasis to the analysis of material behavior.
A few other chapters cover monitoring methods of multiphysics systems. Failure
characterization and faults identification represent the mostly-discussed subjects in
this section.
VI
But, what is the future of multiphysics systems? After three days of lively dis-
cussion all the participants concurred on the necessity of increasing the level of
collaboration between researchers and professionals with different expertise. This
synergy is fundamental to deal with the increasing complexity of future systems.
December 2014 Mohamed Haddar (Sfax-Tunisia)

Mohamed Slim Abbes (Sfax-Tunisia)
Jean-Yves Choley (Paris - France)
Taoufik Boukharouba (Algiers, Algeria)
Tamer Elnady (Cairo, Egypt)
Andrei Kanaev (Paris - France)
Mounir Ben Amar (Paris - France)
Fakher Chaari (Sfax-Tunisia)
Contents
Influence of the Plies Orientation on the Rigidity of the Laminated

Composites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Abdelghani Belkadi, Ali Ahmed Benyahia, Nourdine Ouali
Simplified Calculation Methods on Smoke and Temperature
Stratification in Ventilated Compartments . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Souhila Agred, Abdallah Benarous, Djamel Karmed, Larbi Loukarfi
Analysis of the Notched Specimens on the Fracture Behavior by the
Volumetric Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Mustafa Moussaoui, Salim Meziani
Temperature Effect on Shear Flow and Time Dependant Modelling of
Cutting Oil Emulsion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Larbi Hammadi, Nasr-Eddine Boudjenane, Mansour Belhadri
A Comparison between Two Reliability Based Optimization Strategies
of Tuned Mass Damper Parameters under Stochastic Loading . . . . . . . . . 37
Elyes Mrabet, Mohamed Guedri, Samir Ghanmi, Mohamed Soula,
Mohamed Ichchou
Modeling and Simulation for Vertical Rail Vehicle Dynamic Vibration
with Comfort Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
Mortadha Graa, Mohamed Nejlaoui, Ajmi Houidi, Zouhaier Affi,
Lotfi Romdhane
Localization of Impact on a Beam by Time Reversal Method . . . . . . . . . . 59
Omar Chaterbache, Abdelhamid Miloudi
VIII Contents
Early Detection of Gear Failure by Vibration Analysis . . . . . . . . . . . . . . . 69

Mustapha Merzoug, Khalid Ait-Sghir, Abdelhamid Miloudi,
Jean Paul Dron, Fabrice Bolaers
Effect of Water Hammer on Relatively Long Inclined Pipelines . . . . . . . . 81
Manel Dalleli, Ezzeddine Hadj-Taïeb, Christian Schmitt, Zitouni Azari
Experimental Study of the 60◦ PBT6 Pitching Blade Effect with a PIV
Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
Bilel Ben Amira, Zied Driss, Mohamed Salah Abid
Incidence Angle Effect on the Aerodynamic Structure of an Incurved
Savonius Wind Rotor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
Zied Driss, Olfa Mlayeh, Dorra Driss, Makram Maaloul,
Mohamed Salah Abid
Experimental Characterization of a NACA2415 Airfoil Wind
Turbine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
Zied Driss, Tarek Chelbi, Ahmed Kaffel,
Mohamed Salah Abid
DFMM Approach: Modelling, Formalization and Choice of Material
and Manufacturing Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
Mohamed Saidi, Achraf Skander, Slim Kaddeche
Study of the Aerodynamic Structure around an Obstacle with Inclined
Roof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
Slah Driss, Zied Driss, Imen Kallel Kammoun
Contribution to Inverse Kinematic Modeling of a Planar Continuum
Robot Using a Particle Swarm Optimization . . . . . . . . . . . . . . . . . . . . . . . . 141
Ammar Amouri, Chawki Mahfoudi, Abdelouahab Zaatri
Experimental and Numerical Study of Beams under Low-Velocity
Impact . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
Mahmoud Neder, Abdelhamid Miloudi
Modeling of Cutting Forces and Roughness by the Response Surface
Method (RSM) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
Fayçal Bentaleb, Idriss Amara
Analysis of Inelastic Behavior of Amorphous Polymers . . . . . . . . . . . . . . . 169
Nourdine Ouali, Ali Ahmed Benyahia
Contents IX
First Principal Calculations of Optical Properties of InGaN2 Using in

Solar Cells Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
Hamza Bennacer, S. Berrah, A. Boukortt, Mohamed Issam Ziane
Agent-Based Approach for the Optimal Design of Mechatronic
Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
Amir Guizani, Moncef Hammadi, Jean-Yves Choley, Thierry Soriano,
Mohamed Slim Abbes, Mohamed Haddar
A Digital Pattern Approach to the Design of an Automotive Power
Window by means of Object-Oriented Modelling . . . . . . . . . . . . . . . . . . . . 199
Stanislao Patalano, Ferdinando Vitolo, Antonio Lanzotti
Numerical Study of the External Excitation Amplitude Effect on
Liquid Sloshing Phenomenon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
Bouabidi Abdallah, Zied Driss, Abid Mohamed Salah
The Study of the Hardening Precipitates and the Kinetic Precipitation.
Its Influence on the Mechanical Behavior of 2024 and 7075 Aluminum
Alloys Used in Aeronautics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
Ahmed Ben Mohamed, Amna Znaidi, Moez Baganna, Rachid Nasri
Multidisciplinary Design Optimization in Small Satellite . . . . . . . . . . . . . . 229

Abdelmadjid Boudjemai, Abdelmoumen Bacetti,
Mohammed Amine Zafrane, Rachida Hocine
Mechanical Vibration Cancellation Using Impact Absorber . . . . . . . . . . . 239
Kaouther Chehaibi, Charfeddine Mrad, Rachid Nasri
RFLP Approach in the Designing of Power-Trains for Road Electric
Vehicles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
Clemente Capasso, Moncef Hammadi, Stanislao Patalano,
Ruixian Renaud, Ottorino Veneri
Mechanical Vibration Elimination Using Friction Absorber . . . . . . . . . . . 259
Aymen Nasr, Charfeddine Mrad, Rachid Nasri
AC Flashover: An Analysis with Influence of the Pollution, Potential
and Electric Field Distribution on High Voltage Insulator . . . . . . . . . . . . 269
Hani Benguesmia, Nassima M’Ziou, Ahmed Boubakeur
Evaluation by Wide-Angle X-ray Scattering of the Particle Evolution
in Polypropylene Processed by Equal Channel Angular Extrusion . . . . . . 281
Ramdane Boulahia, Taoufik Boukharouba, Jean-Michel Gloaguen
A Multi-site Supply Chain Planning Using Multi-stage Stochastic
Programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
Houssem Felfel, Omar Ayadi, Faouzi Masmoudi
X Contents
A System Engineering Conception of Multi-objective Optimization

for Multi-physics System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
Mian Chen, Omar Hammami
Falling Film in a Heated Micro-channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
Sabrine Mejri, Olivier Fudym, Jalila Sghaier, Ahmed Bellagi
Interfacial Stresses in FRP-Plated RC Beams: Effect of Adherend
Shear Deformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317
Abderezak Rabahi, Belkacem Adim, Selma Chargui,
Tahar Hassaine Daouadji
A Sensitivity Analysis of Multi-objective Cooperative Planning
Optimization Using NSGA-II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
Wafa Ben Yahia, Omar Ayadi, Faouzi Masmoudi
Transient Flow Analysis through Centrifugal Pumps . . . . . . . . . . . . . . . . . 339
Issa Chalghoum, Sami Elaoud, Mohsen Akrout, Ezzeddine Hadj-Taïeb
Vibro-Acoustic Analysis of Laminated Double-Wall: Finite Element
Formulation and Reduced-Order Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 349
Walid Larbi, Jean-François Deü, Roger Ohayon
Time Step Size Effect on the Hydrodynamic Structure around a
Water Darrieus Rotor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
Ibrahim Mabrouki, Ahmed Samet, Zied Driss, Mohamed Salah Abid
Simulation of the Dynamic Behavior of Multi-stage Geared Systems
with Tooth Shape Deviations and External Excitations . . . . . . . . . . . . . . . 369
Hassen Fakhfakh, Jérome Bruyère, Philippe Velex, Samuel Becquerelle
Fuzzy Modeling and Control of Centrifugal Compressor Used in Gas

Pipelines Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 379
Ahmed Hafaifa, Guemana Mouloud, Belhadef Rachid
Burst Test and J-Integral Crack Growth Criterion in High Density
Poly-Ethylene Pipe Subjected to Internal Pressure . . . . . . . . . . . . . . . . . . . 391
Mohamed Amine Guidara, Mohamed Ali Bouaziz, Christian Schmitt,
Julien Capelle, Ezzeddine Hadj-Taïeb, Zitouni Azari, Said Hariri
Solving the Three-Dimensional Time-Harmonic Maxwell Equations
by Discontinuous Galerkin Methods Coupled to an Integral
Representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 401
Nabil Gmati, Stéphane Lanteri, Anis Mohamed
Contents XI
Iterative Methods for Steady State Looped Network Analysis . . . . . . . . . . 409

Zahreddine Hafsi, Sami Elaoud, Mohsen Akrout, Ezzeddine Hadj-Taïeb
Investigation and Modeling of Surface Roughness of Hard Turned
Aisi 52100 Steel: Tool Vibration Consideration . . . . . . . . . . . . . . . . . . . . . . 419
Ikhlas Meddour, Mohamed Athmane Yallese, Hamdi Aouici
Finite Element Simulation of Fatigue Damage Accumulation for
Repaired Component by Cold Expansion Method . . . . . . . . . . . . . . . . . . . 433
Abdelkrim Aid, Mostefa Bendouba, Mohamed Benguediab,
Abdewahab Amrouche
Geometrically Non-linear Free Vibration of Fully Clamped
Symmetrically Laminated Composite Skew Plates . . . . . . . . . . . . . . . . . . . 443
Hanane Moulay Abdelali, Bilal Harras, Rhali Benamar
Scratch Adhesion Characteristics of PVD Cr/CrN Deposited on High
Speed Steel and Silicon Substrates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453
Kaouthar Khlifi, Hafedh Dhifelaoui, Ahmed Ben Cheikh Larbi
Load Sharing Behavior in Planetary Gear Set . . . . . . . . . . . . . . . . . . . . . . . 459
Ahmed Hammami, Miguel Iglesias Santamaria,
Alfonso Fernandez Del Rincon, Fakher Chaari,
Fernando Viadero Rueda, Mohamed Haddar
Low Velocity Impact Behavior of Glass Fibre-Reinforced Polyamide . . . . 469
Jamel Mars, Mondher Wali, Remi Delille, Fakhreddine Dammak
Burst Pressure Estimation of Corroded Pipeline Using Damage
Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 481
Djebbara Benzerga
Numerical Simulation of Incremental Sheet Metal Forming Process . . . . 489
Lotfi Ben Said, Mondher Wali, Fakhreddine Dammak
A Higher Order Shear Strain Enhanced Solid-Shell Element for
Laminated Composites Structures Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 497
Abdessalem Hajlaoui, Abdessalem Jarraya, Mondher Wali,
Fakhreddine Dammak
New Approch of High Cycle Fatigue Behaviour of Defective Material
under Multiaxial Loading in 1045 Steel . . . . . . . . . . . . . . . . . . . . . . . . . . . . 507
Hassine Wannes, Anouar Nasr, Chokri Bouraoui
Determination of Stress Concentration for Orthotropic and Isotropic
Materials Using Digital Image Correlation (DIC) . . . . . . . . . . . . . . . . . . . . 517
Mhalla Mohamed Makki, Bouraoui Chokri
XII Contents
The Extended Finite Element Method for Cracked Incompressible

Hyperelastic Structures Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 531
Mehrez Zaafouri, Mondher Wali, Said Abid, Mohammad Jamal,
Fakhreddine Dammak
Displacement Influence on Frequencies and Modal Deformations of a
Sandwich Beam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 541
Idris Chenini, Youssef Abdelli, Rachid Nasri
Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 553

Influence of the Plies Orientation on the
Rigidity of the Laminated Composites
Abdelghani Belkadi, Ali Ahmed Benyahia, and Nourdine Ouali
Laboratoire de Mécanique Avancée (LMA) – USTHB – Bab Ezzouar / Alger

{belkhadi_pg,abenyahiaa,nourouali}@yahoo.fr
Abstract. The overall properties of a laminated composite structure depend on de-

sign thereof, i.e., they depend on the thickness, on the orientation and the material of
each ply in the stacking sequence, that is why optimization tools are used to assist
decision making and designing structures with optimized properties. In this study,
the design variables are only the plies orientation because the plies number, the ma-
terial and the thickness are already selected. This problem will be treated as an opti-
mization problem under constraints, whose objective function is the modulus of
transverse stiffness of laminated composite to maximize, and the optimization va-
riables are the plies orientation. The Monte Carlo method is applied as a constrained
stochastic optimization, based on stratification parameters for the determination of
optimal fiber orientations laminate structures to find the maximum longitudinal
modulus of rigidity (Ex). Validation results of this approach have been compared
with the results of Girard [4] who has used a genetic algorithm to optimize laminates
by using equations of the classical theory of laminates and, by referring to the results
of Luersen [3].
Keywords: Laminated composites, optimization, stiffness, plies orientation.
1 Introduction
The mechanical properties of a laminated structure are related to the design pa-
rameters, including the number of layers, thickness, orientation, and the material
of each fold and the stacking sequence of these folds. This is very advantageous
compared to conventional materials because it gives the possibility of designing
structures according to the properties that meet the needs and requirements. Typi-
cal problems of optimal design concern particularly a laminate, the rigidity,
strength, critical load, the natural frequencies and the minimization of thermal ex-
pansion coefficients. However, mechanical properties of a laminate cannot gener-
ally be predicted trivially, even less optimized. Pedersen (1987) showed that in
term of fibers orientation, the composite optimization problems tend to have mul-
tiple local optima which make them difficult to approach by using local optimition
methods based on gradients. Luersen et al. (2004) used an optimization method
© Springer International Publishing Switzerland 2015 1

M. Haddar et al. (eds.), Multiphysics Modelling and Simulation for Systems Design
and Monitoring, Applied Condition Monitoring 2, DOI: 10.1007/978-3-319-14532-7_1
2 A. Belkadi, A.A. Benyahia, and N. Ouali
called Globalized and Bounded Nelder-Mead. As for Girard (2006), he has used a
genetic algorithm for maximizing the longitudinal rigidity module with con-
straints on the shear modulus and Poisson's ratio. The design variables are the ply
orientations which vary in a continuous or discrete manner.
The purpose of this study is to determine the optimal fibers orientations, for each
ply of a laminated structure, which give the modulus of longitudinal stiffness with
maximum constraints on the shear modulus and Poisson's ratio. The study will use the
Monte Carlo method as a method of stochastic optimization under constraints
2 Modelling
For a laminate consisting of n layers numbered from the bottom to the top, aver-
age surface is chosen as the reference plane. The constitutive equation of a lami-
nated plate expressing the resultants and moments based on the membrane
deformations and bends is given by:
(1)
Where and k correspond to vectors of deformations and bends of the average

plan. This constitutive equation can also be written in the following contracted form:
…. …. ….. …. (2)
Where A is the membrane stiffness matrix, D flexural stiffness matrix and B

matrix membrane-bending-torsion coupling and their terms are given by the fol-
lowing expressions:
• ∑
• 1/2 ∑ (3)
• 1/3 ∑
The relationships of the off-axis rigidity constants and the main axes for
orthotropic, symmetrical and balanced material can be written in the following forms:
Influence of the Plies Orientation on the Rigidity of the Laminated Composites 3
1 0 0
1 0 0
0 0 1 0 (4)
0 0 0 1
0 0 0
0 0 0
1 0 0
1 0 0
0 0 1 0
(5)
0 0 0 1
0 0 0
0 0 0
With the parameters of lamination:

/
1
, , , cos 2 , cos 4 , sin 2 , sin 4
/
/
, , , /
cos 2 , cos 4 , sin 2 , sin 4 6
Furthermore, for a design of a balanced and decoupled laminate of which the

material is chosen beforehand, stratification parameters define fully this design by
searching the optimal orientation of plies in the interval -90° to 90°. Thus, the op-
timal solutions can be determined from the stratification parameters, as follows:
1 1 and 2ξ 1 ξ 1
3 Method of Resolution
The optimization method used to solve this problem is that of Monte-Carlo, but
with reduction of interval research. The method involves introducing a reduction
factor of the interval research (the best solution found neighbourhood) in which
are carried out random tests. At the beginning of the process, the size of the
neighbourhood is made wide enough but when the number of iterations increases,
there is a narrowing of this neighborhood. This allows well explore the field of re-
search in the initial phase of the calculations. The reduction comes to allow re-
finements of the solution. On the other hand, one introduces a number of successive
chess (E) which will be counted since the last improvement. E will remain small as
long as the process of optimization improves the solution, but when E begins to take
large values, this means that the process brings no improvement, then, it will be more
profitable to reduce the research interval to allow refinements by introducing a pa-
rameter _ which is a threshold of decision-making. The effectiveness of the re-
duction technique of research interval depends on essentially the wise choice of the
reduction factor of research interval and of the failures threshold parameter.
Consider the example treated by Girard (2006). It is a symmetrical and balanced
laminate with 16 glass/epoxy unidirectional plies (UD). The laminate must be de-
signed to have a modulus of elasticity longitudinal (maximum) while ensuring that
its module of shear ( ) higher or equal to 12 GPa, and that its poisson’s ratio ( )
is smaller than 0,5. The elastic properties of the glass-epoxy UD plies are: E1 = 45
GPa, E2 = 10 GPa, G12 = 4.5 GPa and = 0.31. Since the laminate is symmetrical
and balanced, so there are 4 possible orientations of the UD plies limited between 0°
and 90°. Thus, the optimization problem can be posed as follows:
Maximize the modulus of rigidity by changing the orientation of the 16
plies under the constraints:
- symmetrical and balanced Stacking.
- With : 12 and ν 0.5
4 Results and Discussions
The idea of solving this problem is to work in the search space , as shown
in figure 1 and then back to the directions of the plies.
If one takes ξ and ξ as design variables, the field of feasibility is a parable
and each point in this area (point lamination) corresponds to a tensor A. Indeed,
the shear modulus and the poisson’s ratio of the optimal solution were re-
spectively 12 GPa and 0.5, either the lower and upper limits of these constraints.
In this case, a solution violating one of the constraints will probably have a longi-
tudinal Young modulus greater than the optimal solution. The mode of evo-
lution of the Young module based on the change in the orientation of the
plies is illustrated by figure 2. That is why it must penalise enough this solution,
this violates one of the constraints, in order to make it less attractive compared to
solutions that do not violate any constraints. Otherwise, the algorithm will focus
on a no-workable solution and, therefore, will generate a solution violated one or
more constraints.
In the case of this study, a programming our algorithm was made under
MATLAB, where we have considered 0.85 as value of reduction factor, with a
maximum number of chess factor _ 250. The program was rolled out 100
times to ensure the optimum values.
Optimal solutions obtained by the proposed algorithm are given in table 1.
Fig. 1 Influence de l’orientation des plis sur le module de rigidité longitudinal
This table includes, in addition to results concerning the laminate studied by

Girard (2006), results of optimization for two other laminates of the same type
(symmetric and balanced) but with 4 plies and 8 plies.
Table 1 Optimization results by algorithm proposed
plies num-
Strafication sequence optimized Ex max.
ber
[ ± 48.0047 ] 12.8259
04 [ ± 48.0048 ] 12.8258
[ ± 48.0056 ] 12.8256
[ ± 53.1224 ± 39.3220 ] 14.5268
08 [ ± 53.1239 ± 39.3366 ] 14.5216
[ ± 39.3249 ± 53.1152 ] 14.5262
[ ±53.9459 ±35.1962 ±45.9894 ±49.6245] 14.5225
16 [±35.4489 ±45.3720 ±54.2212 ±49.6980] 14.5276
[±58.3783 ±40.9708 ±42.2751 ± 43.5982] 14.5239
A comparison between the results, obtained for a 16 plies, by our algorithm and
those obtained by Girard are grouped in table 2
Table 2 Comparison of the results obtained with those reported by Girard [4]
To check the performance of our algorithm, we will apply it to another laminate

which has been studied by Vanucci (2007). It's a unidirectional carbon/epoxy lami-
nate of 8 folds 128Gpa, E = 13 GPa, = 6.4 GPa, ν = 0.3) where
we want to have a Young's modulus maximal by changing the orientation of the
plies while complying with the following conditions: 25 and. 1
. For optimization, Vanucci has used the method of Miki (1982), which is a grapho-
analytical method based on the use of lamination parameters. The results of our op-
timization are compared to those found by Vanucci (2007) in table 3.
Table 3 Optimization of the sequence of an 8 ply carbon/epoxy laminate
5 Conclusion
In this work a stratified structure optimization technique based on the parameters
of stratification (lamination parameters) is presented, this technique is based on
the decomposition of the parameters that influence the behavior of an elastic lami-
nate invariant parts and dependent parts of the stacking (the parameters of stratifi-
cation). Thus, everything that depends on the material is separated of what is es-
sentially geometric (sequence of stacking and orientations). For this purpose,
the base material is preselected, then the parameters that influence the change in
elastic behavior are precisely the lamination parameters, which are the only para-
meters which describe the design of the structure, because the stacking sequence,
the thickness and the number of layers are already selected. To this end, the Monte
Carlo method is applied as a constrained stochastic optimization, based on stratifi-
cation parameters for the determination of optimal fiber orientations laminate
structures to find the maximum longitudinal modulus of rigidity (Ex ). Validation

results of this approach have been compared with the results of Girard [4] who has
used a genetic algorithm to optimize laminates by using equations of the classical
theory of laminates and, by referring to the results of Luersen [3]. The obtained re-
sults are also compared with the results of Vanucci [8], who has used the method
of Miki (1982) which is a grapheme method - analytical based on the use of lami-
nation parameters. This method also shows that for a desired modulus of elasticity,
there may be several sequences stacking, which allows handling of the manufac-
turing processes for the composite structures
References
Pedersen, P.: On sensitivity analysis and optimal design of specially orthotropic laminates.
Engineering Optimization 11, 305–316 (1987)
Vanucci, P.: Méthodes Avancées en Optimisation des Structures, Conception optimale des
composites. Université de Versaille (2007)
Luersen, M.A., Le Riche, R., Guyon, F.: A constrained, globalized, and bounded Nelder-
Mead method for engineering optimization. Thèse de doctorat, Ecole Doctorale SPMI,
pp. 52–59 (Décembre 2004)
Girard, F.: Optimisation de stratifiés en utilisant un algorithme génétique. Thèse de
maîtrise, Université LAVAL, pp. 9–20, 49–69 (2006)
Voyiadjis, G.Z., Kattan, P.I.: Mechanics of Composite Materials with MATLAB. Louisiana
State University, pp. 9–190 (2005)
Irisari, F.-X.: Stratégies de calcul pour l’optimisation multiobjectif des structures compo-
sites. Thèse de doctorat, l’Université Toulouse III (2009)
Allaire, G.: Conception optimale de structures. Springer (2007)
Vannucci, P.: Matériaux composites structuraux: Méthode d’analyse et de conception. Un-
iversité de Versailles et Saint-Quentin-en-Yvelines (2007)
Simplified Calculation Methods on Smoke
and Temperature Stratification in Ventilated
Compartments
Souhila Agred1, Abdallah Benarous2,*, Djamel Karmed3, and Larbi Loukarfi4
1
LCEMSM Laboratory, Hassiba Benbouali University, Po Box 151, UHB Chlef, Algeria
agrsousou@yahoo.fr
2
Mechanical Department, Faculty of Technology, UHB Chlef, Algeria
a.benarous@univ-chlef.dz
3
ISAE-ENSMA, Po Box 40109, Futuroscope Chasseneuil, France
Djamel.karmed@isae.ensma.fr
4
LCEMSM Laboratory, UHB Chlef, Algeria
l.loukarfi@yahoo.fr
Abstract. Using different calculation methods with respect to smoke and heat
evacuation systems (SHEVS), a comparative study is presented with regards to
enclosure fire in large single storey compartments. A focus is made on European
and American manual methods for which a set of graphical Matlab routines are
developed. The results obtained with several modeling approaches are discussed
in this paper.
Keywords: smoke and heat evacuation, calculation methods, natural/mechanical

ventilation, fire engineering software.
1 Introduction
A fire can be defined as undesirable burning of materials with release of heat and
toxic gases, causing hazards to people and structures. In this context, it has been
shown that human causalities are much more due to smoke exposure than burns by
flames (Stec et al., 2010). Consequently, some efforts are needed to explore fire
safety engineering issues, more particularly those related with heat and smoke dy-
namics. In this direction, there is a specific need for fire simulation softwares, as
well as predictive tools for the design of smoke and heat evacuating systems
(SHEVS). These devices are supposed to allow for thermal comfort under safe (no
fire) conditions while providing optimal circumstances for survival and egress of
occupants. Design procedures for SHEVS use manual-based methods, which obey
to regionally standardizations and safety regulations.
*
Corresponding author.

10 S. Agred et al.
Although the calculation methods deal with analytical /empirical formulations,

they have the ability to meet quickly specifications for fire safety projects and help
providing start-up data for zone and field modelling (Paranthoen et al., 2010).
In the present work, one proposes to explore some formulations involving
SHEVS design for enclosure fires. Two standard calculation methods (European,
American) are applied in the case of single storey compartments subjected to natu-
ral and mechanical ventilation (CR 12010-5, 2005), (Klote et al., 2012). For com-
parison purposes, additional calculation using a zone modelling software (CFAST)
is performed (Peacock et al., 2011). Towards a technological reliability regarding
fire safety engineering, calculation methods are automated in a form of computer
routines for which a set of graphical user interfaces (GUI) are developed.
2 Calculation Methodology
2.1 Input Variables and Calculation Assumptions

For the calculation methods, temperature and smoke behaviour is supposed to rely
on a steady state assumption. Since the convective and radiative transfer to walls
are not considered, the formulation is not suitable for studies on fire resistance of
structures. In fact, thermal losses from the flame to the plume are only quantified
by a relative fraction α R of the heat release rate Q f . Prior to any SHEVS design, a
conventional fire should be defined in terms of heat release ( HRR = Q f ) and pool
dimensions ( A f , D f , Pf ). The compartment is characterized by the floor area,
ceiling height ( H ceiling ) as well as opening area ( Ai ) for gates /windows. Depend-
ing on the user choice, stratification is characterised by the smoke (hot layer)
thickness ( δ s ) or the temperature rise ΔT of the cold layer. These informations
are indicated by means of the main graphical user interface (Fig.1).
Fig. 1 The main graphical user interface (GUI)

Simplified Calculation Methods on Smoke and Temperature Stratification 11
2.2 The European Method

A threshold value of 20% is specified for radiative losses. The convective thermal
power is given by:
Q c = (1 − α R )Q f = 0.8Q f (1)
If this value induces an absolute thermal rise ΔT , the mass flow rate of the
smoke can be expressed by:
Q c
M s = (2)
cΔ T
where c is the specific heat capacity for the air-smoke mixture.
Owing to buoyancy forces, a stratification of the hot layer is created and the
smoke lies at a vertical distance Ys from the floor:
(
Ys = M s / C e Pf )
2/3
(3)
This value is retained if the regularity condition Ys ≤ A f is fulfilled, so that

the smoke layer thickness is given by:
δ s = H ceiling − Ys (4)
For large compartments, the empirical constant C e is set to 0.19.

If the numerical value of Ys exceeds the ceiling height, one considers that the
smoke flow rate is too high and the user may impose an “appropriate” value for
the thermal rise (Fig.2).
Fig. 2 Graphical user interface for smoke calculation

12 S. Agred et al.
As the static pressure is supposed to be constant throughout the hot and cold
layers, the smoke volumetric flow rate can be written as:
Vs = ( M s .Ts ) /( ρ ∞ T∞ ) (5)
where Ts = ΔT + T∞ stands for the static temperature within the smoke layer.
t
In the case of natural ventilation, a set of N NV vents whose total area ANV are
expected in the compartment, so that:
t M s Ts
ANV C v = (6)
M s Ts .T∞
2
2
2 ρ ∞ gδ s T∞ ΔT −
( Ai C i )2
where C v , C i denote loss coefficients of ventilation holes and air openings (gates,
windows) respectively. The cross sectional area of each vent is easily obtained
from the following relation :
n t
ANV = ANV / N NV (7)
In the case of a mechanical ventilation, a set of N MV ventilators /jet fans are to

be expected for the compartment. The corresponding mass flow rate for each de-
vice should not exceed a critical value M
crit
to avoid the plug-holing phenome-
MV
non (Klote et al., 2012). Depending on whether the component is considered as
‘far’ or ‘adjacent’ to the wall, two expressions for the threshold flow rate are used:
⎧1.3 δ s2
M MV = ⎨
crit
g .δ s T∞ ΔT if X VW < Dv
⎩ Ts
(8)
⎧ 2.05 δ s2 ρ ∞
M MV =⎨
crit
g .T∞ ΔT D v if X VW ≥ Dv
⎩ Ts
Dv , X VW are respectively the hydraulic diameter and the distance to the wall of
each device. The number of elements (Fig.3) that should be set to extract the
smoke is given by:
N MV ≥ M s / M MV
crit
(9)
Fig. 3 SHEVS calculation GUI (mechanical ventilation case)
2.3 The American Method

The American method specifies two rates of radiative losses depending on the
presence (20%) or absence (50%) of sprinklers, so that the convective thermal
power is expressed as:
Q c = (1 − α R )Q f ; α R = 0.2 or 0.5 (10)
The same procedure is used for the calculation of the smoke layer thickness,
according to a slight different value for the constant C e = 0.188 .
The total area required for a natural ventilation case is obtained by solving the
nonlinear equation:
2
⎛ At C ⎞
Ts2 +
⎜ NV v ⎟ T T
⎜ ⎟ ∞ s
t M s ⎝ Ai C i ⎠
ANV C v = (11)
ρ∞ 2 g δ s T s ΔT
For a mechanical ventilation case, a critical volume flow rate is imposed to

avoid plug-holing. For each device (ventilator, jet-fan) the volume flow rate is
given by:
2
VMV =
crit 5
gδ s T∞ ΔT (12)
Ts
The expected number of components (Fig.4) is obtained from:
N MV ≥ Vs / VMV
crit
(13)
14 S. Agred et al.
Fig. 4 Graphical user interface for SHEVS calculation
The first test case concerns a polyvalent hall which dimensions are 95 m length,
66m width and 11m height. The hall features ten gates whose total opening area is
109 m2.
The fire source is taken to be a square 9m×9m pool (category 2) with a total
heat release rate of HRR = 2.02 Megawatts. A natural ventilation configuration is
considered here. Basing on 20% radiative loss, the temperature rise is estimated at
ΔT = 150 °C for both methods, with a corresponding smoke flow rate of
M = 108 kg/s. In this case, the average smoke temperature is almost 170 °C
s
which shows an overestimation of 42% from that of the hot layer, as predicted by
the CFAST solver (Fig.5). The important discrepancy is certainly, due to the fact
that the “manual” formulations do not account for thermal losses to the ceiling
wall.
Furthermore, the CFAST solver makes use of a well-known entrainment model
that accurately predicts the air-smoke mixing (Mc Caffrey, 1983). A model that
justifies the moderate temperature, in the vicinity of the ceiling.
180
160
European + American method
CFAST calculation
140
Hot layer temperature (C°)
120
100
80
60
40
20
0 50 100 150 200 250 300 350 400 450 500 550
Time (s)
Fig. 5 Hot layer temperature (natural ventilation)
Consequently, the steady state temperature (120°C) of the hot layer seems to be
most revealing of the thermal behaviour. Regarding the smoke stratification, the
European as well as the American methods recover approximately the same value
Ys ≈ 6.3 m against Ys = 8.0 m for the CFAST solver (Fig.6).
Fig. 6 Hot layer height for various calculation methods (natural ventilation)
16 S. Agred et al.
Figure 6 shows a relative deviation of 27% on the Ys values obtained by the

CFAST software. The discrepancy was expected since a one-dimensional (1D)
development is assumed for the smoke dynamics in the zone modelling solver.
This assumption is strongly questionable for such geometrical (large compart-
ment) and physical (natural convection) configurations.
t
The two “manual” methods succeed in predicting the value ANV = 53 m2 for
the total ventilation area. This quantity is mandatory to build a geometrical model
for any field-based (CFD) calculation.
The second test case is a simplified supermarket model of length 70m, width
35m and 4m height. Gates and windows occupy an area of 84 m2. The design fire
of 4.5 Megawatts represents a square pool of 3m×3m, yielding a thermal power of
500 Kw/m2. Owing to the important heat release flux, a mechanical ventilation de-
sign will be considered.
On a basis of a prescribed temperature rise 150 °C, a mass flow
rate M s = 24 kg/s is obtained for both methods. However, equation (3) yields a
non realistic value Ys = 4.8 m of the smoke layer height, since the ceiling height
is only 4m. This result indicates that a stable thermal steady state is unlikely. Ac-
cording to the regulatory condition (see eqn. 3), the value Ys = 3 m will be retained
in the following calculations.
The European method gives a global critical mass flow rate M MV = 3.65 kg/s.
crit
If one assumes that the ventilation components are sufficiently remote from the
walls, a minimum of N MV = 7 ventilators having 1m diameter, are required within
the compartment.
The American method considers the existence of a threshold for the volume
flow rate. The calculated value VMV = 3 m3/s corresponds to a minimum
crit
of N MV = 11 ventilators. This latter value seems more realistic since ventilator

characteristics are based on a constant volume flow rate rather than a mass flow
rate, which is a fluid-dependent quantity.
Both methods significantly overestimate the smoke layer temperature by 55%
comparing with the CFAST solver calculation (Fig.7).
Figure 7 depicts the time evolution of the average hot layer temperature. It is
noticed that an established steady state cannot be reached throughout the curve. A
slight asymptotic behaviour is observed after 10 minutes of burning. The limiting
hot layer temperature (110°C) seems to be fairly moderate thanks to ventilation
devices that contribute to improve the mixing of air and smoke. In this mechanical
ventilation case, the CFAST solver exhibits a slight overestimation of only 09%
on the smoke layer height, comparing with manual methods (Fig.8).
180
160
European + American method
CFAST calculation
140
Hot layer temperature (C°)
120
100
80
60
40
20
0 50 100 150 200 250 300 350 400 450 500 550 600 650 700 750
Time (s)
Fig. 7 Hot layer temperature (mechanical ventilation)
Fig. 8 Hot layer height for various calculation methods
Although the compartment has a large surface where flow of air and smoke
may develop in a three dimensional pattern, the mechanical ventilation contributes
to favour a spreading (blowing) direction. Consequently, the ceiling flow (at least
at the blowing level) can be considered as quasi one-dimensional. The present
configuration certainly enhances the applicability of the zone modelling calcula-
tion while allowing for a reduction of fire simulations cost.
18 S. Agred et al.
4 Conclusion
Different calculation methods have been applied for two test cases of large single
storey compartments. Both European and American method recover almost the
same values for SHEVS characteristics. Indeed, these “manual” procedures con-
tain quasi-identical formulations while basing upon a similar design philosophy. It
has been shown that the American method seems more conservative with respect
to critical volume flow rate, which is a more appropriate yardstick of a ventilator
performance. Mixing of air and smoke in the plume region were accurately quanti-
fied by means of an entrainment sub-model that exists in the zone modelling
solver. Smoke and temperature stratification were predicted fairly good using the
CFAST solver. This was not the case of the manual methods for which a common
drawback is the steady state assumption. Several fire issues can be investigated by
means of zone/field modelling formulations where much pertinent informations
can be provided. Nevertheless, field/zone calculations are much time and resource
consuming and require more skills from the user to deal with realistic applications.
References
CR 12010-5: Guidelines on functional recommendations and calculation methods for
smoke and heat evacuating systems. CEN Technical Report No.191, Brussels, Belgium
(2005)
Klote, J.H., Milke, J.A., Turnbull, P.G., Kashef, A., Ferreira, M.J.: The handbook of smoke
control engineering, 1st edn. American Society of Heating, Refrigerating and Air Condi-
tioning Engineers (ASHRAE), Washington (2012)
Mc Caffrey, B.J.: Momentum implications for buoyant diffusion flames. Combustion and
Flame (1983), doi: 0010-2180(83) 90129-3
Paranthoen, P., Gonzalez, M.: Mixed convection in a ventilated enclosure. International
Journal of Heat and Fluid Flow (2010), doi:10.1016/2010.01.006
Peacock, R.D., Forney, G.P., Reneke, P.A.: Consolidated model for fire growth and smoke
transport. NIST Technical Report No.1026 R1, USA (2011)
Stec, A.A., Hull, T.R.: Fire toxicity, Part 2, 1st edn. Woodhead Publishing, Cambridge
(2010)
Analysis of the Notched Specimens on the
Fracture Behavior by the Volumetric Method
Mustafa Moussaoui and Salimpop Meziani
Laboratory of Mechanic, University Constantine 1,

Campus Chaab Erssas, Constantine 25000 – ALGERIA
{moussaoui_must,meziani_salim}@yahoo.fr
salim.meziani@labomecanique-umc.org
Abstract. A single edge notched specimen is commonly used in materials of steel

construction for the determination the notch stress intensity factor. CT- specimen
is considered as a finite element model containing an elliptical notch under a uni-
form uni-axial tensile loading. The volumetric method is applied in perfect elastic-
plastic behavior. Changing made to notch parameters influences the evaluation of
the effective stress accordingly the results of the notch stress intensity factor.
Keywords: Weight function, Notch stress intensity factor, Notch effect, Effective
stress.
1 Introduction
Voluntarily created or induced inadvertently, the geometric discontinuities always

exist in mechanical components and engineering structures. They can take differ-
ent forms such as the notches, the voids, the micro-cracks, etc… They induce
strong local stresses and strains in their roots. Accordingly, the cracks initiate at
their roots and they become responsible for failures of the notched components.
It is well established that the concentration of localized stresses associated to a
geometric discontinuity, reduces the static and cyclic strength of a structure (Shin.
C 1994). Accordingly, the evaluation of the stress intensity factor associated to the
various types of geometrical discontinuities or notch becomes paramount for the
engineer during the design in order to guard against the risk of sudden fracture.
The stress distribution near the tip of such a cutting is characterized by a stress
singularity. The dimensional quantities of elliptical notch are major axis and minor
axis, affect the stress intensity factor values. Local approaches are introduced to
describe the magnitude of stress field, and evaluate the factors that influence the
behaviour of structures under various loading conditions. According to this study,
a new concept, based on the volumetric method focus on the notches effect.
Analysis of stress distribution at a notch’s root shows a pseudo-singularity stress
distribution governed by notch stress intensity factor (NSIF), KΙρ (Pluvinage et al

20 M. Moussaoui and S. Meziani
1999). The result of this works and others indicate that this approach gives a good
description in relation with the notch effect.
Under a plane stress mode, two methods are used especially volumetric ap-
proach and ones of Irwin for analysis of the stresses field using FEM methods.
This present study focus on the effect of the variation of dimensional quantities of
the notch on the fracture behaviour of notched specimens by the calculation of
notch stress intensity factor and analysis the stresses field that reigns at the notch
root.
2 Finite Element Modeling of Elliptical Notch
A semi-elliptical notch is applied to a plain CT specimen (Fig. 1) having the di-

mensions 20x20 [mm], subjected to a uniform uniaxial tension loading σ at its two
ends of value 125 [MPa]. Four notch configurations are used, with different values
of the minor axis (b=0.5, 4 and 5[mm]) and the major axis ‘a’ takes all the follow-
ing values 0.5,1,2,3,5,6,8,10 [mm]. To get a better accuracy in the results, a
much more refined mesh was created around the edge of the notch.
Fig. 1 Finite element model
Considering that the material has perfect elastic plastic behavior and having the
following mechanical properties in Table 1.
Table 1 Mechanical characteristics of the material
Young's modulus yield strength Poisson’s coeffi-

E [MPa] σE [MPa] cient ν
230E03 670 0.293
Analysis of the Notched Specimens on the Fracture Behavior 21
3 Behavior Analysis of Elliptical Notch Root by Analytical

Methods
Irwin’s approach has set a goal: the determination of the stress intensity factor in
mode I, taking into account the presence of a plastic zone that is assumed to be
small compared to the crack length. The resulting value of this factor, KI , will
establish a critical “intrinsic” value corresponding to the elastic–plastic fracture
(Irwin 1964).
Irwin therefore assumes an extension of the fracture of ‘re << a’ and considers
that the stress, obtained from this zone does not exceed the elastic limit, (the law
of perfect elastic–plastic behavior, without extension). It is therefore required to
estimate the value re of the stress applied at a distance (Naman. R 2012)
From the stress field, which is determined in linear elasticity, for θ = 0 and in
mode I, we can write:
Κ Ι = σ π (a + re ) (1)
The volumetric method (VM) is a local fracture criterion (H. Meliani et al

2011) which assumes that the fracture process requires a certain volume of cylin-
drical shape with its diameter and the effective distance. This kind of methodology
is classified as Theory of Critical Distance (TCD). The main objective of the VM
is to calculate the effective distance and its corresponding effective stress, ob-
tained from the stress distribution at the notch root.
The effective distance for the proposed elastic-plastic case, can be found via the
minimum point of relative stress gradient (Adib and Jeong 2007).The relative
stress gradient, in the volumetric method is defined by:
1 ∂σ yy ( x )
χ( x ) = (2)
σ yy (x ) ∂x
Analysis by finite element method of volumetric method enables to determine

two fundamental elements which are effective distance and effective stress. Fig. 2
shows the bi-logarithmic diagram of elastic-plastic stresses distribution from notch
root along the ligament, relating to a thin semi-elliptical notch with its minor axis
‘b’ equal to 0.5 [mm] and semi-major axis ‘a’ equal to 6 [mm].
It shows three distinct zones, the first zone (zone I) is near to notch tip, where
stress reaches the peak value (Shi. S 1995), then it gradually decreases in the elas-
tic-plastic material regime (zone II).
Fig. 2 Bi-logarithmic diagram of elastic-plastic stresses distribution, and relative stress

gradient
The third area which can be assimilated as stresses pseudo-singularity (Minor et

al 2003) is the site of fracture initiation (H. Meliani et al 2011). The notch stress
intensity factor is expressed as function of Xeff and σeff:
Κ Ιρ = σ eff 2π . X eff (3)
Where Κ Ιρ , σ eff and Xeff are respectively the notch stress intensity factor, the
effective stress, and the effective distance. The effective distance represents the
upper limit of this zone. The latter starts at the inflection point of the curve, which
is given by the minimum of relative stress gradient.
4 Elastic Stresses Distribution for Thinned and Opened

Notch
Fig. 6 shows an elastic stress distribution analysis of an open and thin notch hav-
ing semi-elliptical shape with dimension b equal to 0.5 and 4 [mm] respectively, a
notch having a depth equal to 6 [mm]. It shows a comparison of analytically ob-
tained results with those obtained by the finite element method. Various formalisms
representing the stress distribution can be found in literature (Pluvinage1998).
The results are similar near the end of notch where the stress concentration is
high. However at remote distances from notch bottom, i.e. where the stress gradi-
ent is lower, the results obtained by finite element method decrease rapidly
towards lower values.
Fig. 3 Elastic stresses distribution for thinned and opened notch
5 Stress Intensity Factor Emanating from Notch
The phenomenon of rupture must indeed be considered in its physical dimension.

It requires a certain elaboration volume of process failure (Pluvinage 2003). In this
volume, a zone is governed by the notch stress intensity factor, which depends at
the same time on the effective distance, the effective stress, the relative stress
gradient and on the weight function.
Fig. 4 Evolution of the notch stress intensity factor for thinned and opened semi elliptical
notch
The results obtained are compared with those calculated according to the two
Irwin models (Nui et al 1994, Chaboche1988). Fig. 4, show the notch stress inten-
sity factor evolution versus the ratio b/a, calculated using the two Irwin models,
and the three weight functions: unit, delta and gradient functions and considering
semi-elliptical notch.
The obtained results of the stress intensity factor for the thin semi-elliptical
notch (b = 0.5 [mm]) show , The Irwin corrected model gives converging results
with the results of unit weight function and the results obtained from delta weight
function represented the upper limit and gradient weight function is lower limit
(Fig.4a). At increased value of ration (b/a), the results obtained from Irwin without
corrected model tend to delta function.
Near to the notch root, the results obtained by the three weight functions move
away from those obtained by the Irwin corrected model and are close to those of
Irwin without correction model (Fig.4b).
The increase of the major axis "a" and the reduction of the minor axis "b" lead
to a flattened semi-elliptical notch, corresponding to a deep notch, create a region
where the stress concentration is high, and generate an increased stress gradient as
a result of the increase of the stress intensity factor values.
The thin and deep semi-elliptical notches are much more dangerous than the
short semi-elliptical and semi-circular notches
6 Notch Effect on the Elastic-Plastic Stress
The effective stress is the average value of the stress distribution over the effective
distance, weighted by the relative stress gradient inside the elaboration volume of
fracture, in fracture elaboration volume is expressed according of a weighting
function, in order to take into account the stress gradient due to geometry and
loading mode (Minor et al 2003, Adib and Jeong.2007) by the following relation:
1 x eff
σ eff =
x eff ∫
0
σ ij ϕ ( x , χ ) dx (4)
Three weight functions are used in this study, the unit weight function is
ϕ( ) ( )
x , χ = 1 , delta weight function is δ x − X eff and gradient is (1− x.χ ) .
Fig. 5 show the evolution of effective stress calculated using the three weight
functions (unit, delta and gradient) with the maximum elastic-plastic stress, ac-
cording to the ratio (b/a), relating to the thinned and opened semi-elliptical
notches. In the case of an opened notch with a lower ratio (b/a), the elastic-plastic
stresses value (Fig. 5b) is greater than the effective stresses value calculated by the
volumetric approach. And for thinned notch where the gradient is strong, the val-
ues of effective stresses calculated using the three weight functions; unit, delta and
gradient are identical (Fig. 5a).
The more the minor axis 'b' of semi-elliptical notch increases the more the
results obtained by the three weight functions converge between them (Fig. 5).
For both thin and open notch configurations, all the results obtained using
weight delta function are situated in a superior position and are close to the values
of the maximum elastic-plastic stress. The increase of the minor axis 'b' reduces
the error gap between the results obtained using delta weight function and those
obtained using the maximum elastic-plastic stress.
Fig. 5 Evolution of effective stress & maximum elastic-plastic stress for thinned and
opened semi elliptical notch
7 Conclusion
The analysis presented that is supported, by a volumetric approach has a SFI

(stress fields intensity) base and its advantages are consideration of elastic–plastic
stress distribution at notch roots. The changes made to notch parameters with dif-
ferent weight functions affect the stress field near the notch tip. The volumetric
method considers, its fundamentals elements, the effective distance, effective
stress, and stress gradient their change affects the computation of notch stress
intensity factor.
In the region of low stress gradient, the values of SIF obtained using the Irwin
corrected model are intermediate between two limits, an upper limit corresponding
to SIF calculated using the weight delta function, and a lower limit which corre-
sponds to the SIF calculated according to the gradient function. The variation of
the dimensional quantities in thin elliptical notches affects the amplitude of the
stress gradient. Near the notch root there is convergence between SIF obtained
using the volumetric method and those obtained using the Irwin without correction
model.
In case of thin elliptical notches, the volumetric method and Irwin models give
similar results and describe together all variations introduced in the notch parame-
ters. While for open elliptical notches, the effective stress and Maximum Elastic-
Plastic Stress gives some closest results. The tendency to deep thin elliptical notch
is much more dangerous compared to short elliptical and circular notches.
References
Shi, S.Q., Puls, M.P.: A simple method of estimating the maximum normal stress and plas-
tic zone size at a shallow notch. International Journal of Pressure Vessels and Pip-
ing 64(1), 67–71 (1995)
Pluvinage, G.: Fatigue and fracture emanating from notch; the use of the notch stress inten-
sity factor. Nuclear Engineering and Design 185, 173–184 (1998)
Pluvinage, G.: Fracture and Fatigue Emanating from Stress Concentrators, Université de
Metz, France. Kluwer Academic Publishers (2003)
Nui, L.S., Chehimi, C., Pluvinage, G.: Stress filed near a large blunted tip V-notch and
application of the concept of the critical notch stress intensity factor (NSIF) to the frac-
ture toughness of very brittle materials, Engineering Fracture Mechanics. Perga-
mos 49(3), 325–335 (1994)
Lemaitre, J., Chaboche, J.-L.: Mécanique des matériaux solides, © Bordas, Paris (1988)
Hadj Meliani, M., Matvienko, Y.G., Pluvinage, G.: Corrosion defect assessment on pipes
using limit analysis and notch fracture mechanics. Engineering Failure Analysis 18,
271–283 (2011)
El Minor, H., Kifaini, A., Louah, M., Azari, Z., Pluvinage, G.: Fracture toughness of high
strength steel-using the notch stress intensity factor and volumetric approach. Structural
Safety 25, 35–45 (2003)
Adib-Rammezani, H., Jeong, J.: Advanced volumetric method for fatigue life prediction
using stress gradient effects at notch roots. Computational Materials Science 39,
649–663 (2007)
Pluvinage, G., Azari, Z., Kadi, N., Dlouhy, I., Kozak, V.: Effect of ferritic micro-structure
on local damage zone distance associated with fracture near notch. Theoretical and Ap-
plied Fracture Mechanics 31, 149–156 (1999)
Irwin, G.R.: Structural aspects of brittle fracture. Applied Material Research 3, 65–81
(1964)
Shin, C.S., Man, K.C., Wang, C.M.: A practical method to estimate the stress concentration
of notches. Fatigue 16 (June 1994)
Recho, N.: Fracture Mechanics and Crack Growth. ISTE Ltd and John Wiley & Sons, Inc.
(first published, 2012)
Temperature Effect on Shear Flow and Time
Dependant Modelling of Cutting Oil Emulsion
Larbi Hammadi, Nesr-Eddine Boudjenane, and Mansour Belhadri
Laboratory of Rheology, Transport and Treatment of the Complex Fluids, University of Science
and Technology, Mohamed Boudiaf, B.P. 1505, El M’Naour, Oran 31000, Algeria
Hammadi7280@yahoo.fr
Abstract. The effect of temperature and shear rate on rheological behaviour of

cutting oil emulsion was investigated in this work. The model of Herschel-Bulkley
was used to fit the shear stress dependence of the shear rate for temperatures be-
tween 20°C and 100°C. The temperature increase induced not only a decrease in
the yield stress and the consistency index of cutting oil emulsion but also an in-
crease of the flow index of cutting oil emulsion. The temperature dependence of
limit viscosity at high shear rate of the cutting oil emulsion was fitted by an Ar-
rhenius equation. For constant shear rate applied on the cutting oil emulsion at
20°C a thixotropic behaviour was observed and analysed using a modified model
of Herschel-Bulkley in which a structural parameter λ was included in order to ac-
count for time dependent effect.
Keywords: Yield stress, Structural parameter, Thixotropy, Cutting oil emulsion.
1 Introduction
During mechanical operations of cutting and machining metals, a considerable

quantity of heat is generated, mainly due to the high plastic deformation in the
primary shear zone, and to the friction of the chip on the tool rake face (Bensadok
et al. 2008), in this case this cutting oil emulsions is used to reduce the detrimental
effects of heat and friction on both tool and work-piece. The cutting oil emulsions
produce three positive effects in the process heat elimination on the chip tool in-
terface and chip removal (Pottirayil et al. 2011). The emulsion term is generally
utilized to describe the blending mixture of two or more immiscible liquids to-
gether. The dispersion of droplets of one liquid phase such as oil thoroughly into
an immiscible continuous phase of another liquid such as water will lead to oil-in-
water emulsion (O/W emulsion) (Liu et al. 2011). The Oil–water emulsion is a
complex mixture of multi-component and multi-phase system. The emulsion cha-
racteristics have been investigated in several papers from different points of view.
Literature survey on the rheological behavior of oil emulsions showed that many in-
vestigations have been studied on cutting oil emulsions in which the continuous
phase is Non-Newtonian fluids and thixotropic behavior (Stachowiak and Batchelor

28 L. Hammadi, N-E. Boudjenane, and M. Belhadri
2006 ). The thixotropy is generally defined as a progressive decrease in viscosity

over time under constant applied shear stress, followed by gradual recovery when
this stress is removed (Hammadi et al. 2013). Mujumdar et al (2002) describe the
thixotropy as reversible breakdown of the particulate structure under shear, the
structure formed at rest is often (but not necessarily) associated with a flow yield.
In this case, the variable viscosity is explicitly associated with a variable micro-
structure, characterized by a structure parameter the kinetics of which governs the
time-dependent behavior (Dullaert and Mewis 2005). In this work the influences
of temperature and shear rate on the cutting oil emulsions rheological properties
were evaluated, and a modified model of Herschel-Bulkley was employed to
describe the thixotropic behavior of cutting oil emulsions.
2 Materials and Methods
2.1 Sample Preparation and Experimental Set Up

Cutting oil emulsion is a mixture of water and cutting oil. The latter is made of
mineral oil to which is added around surfactant called TASFLOUT 22-B. Homo-
genization was obtained by continuous mechanical agitation for 2 hours at
ambient temperature. The rheological measurements were performed by using a
torque controlled rheometer (RS600 from Thermo-Fischer), equipped with a cone-
plate geometry (diameter: 60 mm; angle: 2 degrees; gap: 105 µm). In order to pre-
vent changes in composition during measurements due to water evaporation, a
solvent trap was placed around the measuring device. In order to avoid any memo-
ry effect, the sample was pre-sheared at a frequency of 200s-1 for 60 seconds in the
measuring device. The sample was then kept at rest for 600 s prior to measure-
ments in order to allow the material to recover its initial structure at least partially.
The imposed shear rate range depends on the volume concentration of cutting oil
emulsions. Consequently, for the following experiments, a continuous ramp of
shear rate ranging from 0.001 to 200 s-1has been applied on each sample during
600 s. The experiments were performed at different temperature between 20°C
and 100°C.A new fresh sample was used for each temperature in order to avoid
any irreversible evolution of the cutting oil emulsion. In order to investigate the
reproducibility of results, two replicates were made for most of the experiments.
After a rest time of 600s, the samples were sheared during 120s at different constant
shear rates (50 s-1, 100 s-1 ,150 s-1 ,200 s-1 and 250 s-1) at a constant temperature of
20°C.
Temperature Effect on Shear Flow and Time Dependant Modelling 29
3 Results and Discussion
3.1 Effect of the Temperature on Rheological behavior of

Cutting Oil Emulsion
The variation of the shear stress τ as a function of the shear rate γ at different
temperatures (20°C to 100°C) for the 60% of cutting oil emulsion is shown in
Figure 1. It shows clearly Non-Newtonian behavior after a yield stress. Therefore
experimental data were fitted to the classical model of Herschel-Bulkley:
(1)
Where τ0 is the yield stress in Pa, K the consistency index in Pa sn and n is the
flow index.
The fitted parameters of the model of Herschel-Bulkley as a function of tem-
perature are presented in Figs 2 and 3. It appears that yield stress and consistency
index decreases with increasing temperature but the flow index increases.
12
11 20°C
40°C
10 60°C
80°C
9 100°C
8 Model
Shear stress (Pa)
7
0,5
6 20°C
0,4
40°C
5 60°C
Viscosity (Pa.s)
0,3
80°C
4 0,2
100°C
3 0,1
2 0,0
0 50 100 150 200
1 Shear rate (s-1)
0
0 50 100 150 200 250 3 00 350
-1
She ar rate (s )
Fig. 1 Shear stress as a function of shear rate at different temperatures (concentration

60% of cutting oil emulsion). The solid lines correspond to the curve fitting to Eq.1.
3,5
3,0
2,5
Yield stress (Pa)
2,0
1,5
1,0
20 30 40 50 60 70 80 90 100 110
Temperature (°C)
Fig. 2 Variation of yield stress as a function of temperature (60% of cutting oil emul-
sion)
0,85 0,40
0,80
0,75
0,35
0,70 n
0,65
Consistency index (Pa.s )

n
0,60 0,30
Flow index (-)

0,55 K
0,50
0,25
0,45
0,40
0,35 0,20
0,30
0,25
0,15
0,20
0,15
0,10 0,10
20 30 40 50 60 70 80 90 100 110
Temperature (°C)
Fig. 3 Variation of consistency index and flow index as a function of temperature
3.2 Effect of Temperature on Limit Viscosity of Cutting Oil

Emulsion
It is also clearly revealed on the insets of Figs. 1 that the apparent viscosity steeply
decreases and evolves toward a constant value at high shear rate, called as the lim-
it viscosity . Fig 4 shows that the limit viscosity of the cutting oil emulsion
decreased from 0.069 to 0.036Pa.s for a volume concentration of 60% as the tem-
perature was increased from 293 to 373 °K. A likely explanation is a weakening of
the network strength between particles of the dispersions due to thermal motion.
This temperature dependence on the limiting viscosity can be described by an
Arrhenius type equation involving the absolute temperature (T), universal gas
constant (R), energy of activation for viscosity (Ea) and a constant A:
(2)
With A=0.0041 and Ea=0.72 J/mol
0,070
0,065 Experimental
Arrhenius equation
0,060
Limit viscosity (Pa.s)
0,055
0,050
0,045
0,040
0,035
2,8 3,0 3,2 3,4 3,6 3,8 4,0
(1/T)*10 3 (1/°K)
Fig. 4 Limit viscosity as a function of 1/T at volume concentration 60% of cutting oil emul-
sion. The solid lines correspond to the curve fitting to Eq.2.
3.3 Apparent Viscosity Evolution under Constant Shear Rate

and Determination the Structural Parameter
Fig.5 shows the apparent viscosity evolution as a function of time for different
shear rates varying from 50 to 250 s-1. For all values of shear rates, the viscosity
decreases significantly with time then tends to stay constant. In order to analyze
the structural evaluation of the emulsion, we applied the phenomenological model

of Tiu and Boger 1974 based on an approach initially suggested by Moore (1959)
and refined by Cheng and Evans (1965). This choice was motivated by the sim-
plicity (minimum number of parameters) and proven efficacy of this model.
Cheng and Evans (1965) argued that the stress-strain equation depends on the flu-
id model and must be completed with a kinetic equation of structural parameter
changing with time. This parameter describes the current state of agglomeration.
As pointed out before, the flow curves of cutting oil emulsion were found to be
well fitted by the model of Herschel-Bulkley, with satisfactory agreement (see
Fig. 1 ). Following the model proposed by Tiu and Boger (1974), the Herschel–
Bulkley model was then modified to take into account the structural changes by
introducing a structural decay parameter . Thus, the equation of state is assumed
to be:
(3)
The structural parameter ranges from an initial value of unity at zero shear time
(agglomerated structure at rest).
0 ,09 5
0 ,09 0
0 ,08 5
. -1
0 ,08 0 γ = 50 S
. -1
γ = 100 S
0 ,07 5 . -1
γ = 150 S
Viscosity (Pa.s)
0 ,07 0 .
γ = 200 S
-1
. -1
0 ,06 5 γ = 250 S
Equation (10)
0 ,06 0
0 ,05 5
0 ,05 0
0 ,04 5
0 ,04 0
0 ,03 5
0 ,03 0
0 20 40 60 80 100 120 140
Time of shears (s)
Fig. 5 Viscosity of cutting oil emulsion as a function of time shears at 20°C for constant
shear. The solid line corresponds to curve fitting to eq.10.
For their rate equation Tiu and Boger (1974) employed a second-order kinetic
equation:
for (4)
Where k2 is the rate constant, function of shear rate, to be determined experi-

mentally and can be considered as a measure of the rate of structure breakdown,
i.e., the degree of thixotropy (Butler and McNulty 1995).Then entire time-
dependent behavior of our cutting oil emulsion might be completely described by
Eq. (3) and (4) with the parameters , , and k2 evaluated from experi-
mental data. The rheological constants , and are determined with ini-
tial shearing conditions ( 1 and t=0), and Eq.(4) reduces to the classical
Herschel-Bulkley equation. In other words, they are determined from the initial
shear stress in the material, observe the beginning of a test, for each shear rate
considered. Fig.6 shows the zero shear stress at a function of shear rate for volume
concentration of cutting oil emulsion equal 60%. The curve were fitted to the
model of Herschel-Bulkley and the apparent yield stress value obtained by extra-
polation to 0. The resulting equation was:
.
0.547 0.6 , 0.994 (5)
10
60% Vol
9
Herschel-Bulkley model
Zero time shear stress (Pa)

8
4
50 100 1 50 200 25 0
-1
Shear ra te (s )
Fig. 6 Zero time shear stress as a function of shear rate for 60% of cutting oil emulsion and
comparison with the Hershel Bulkley model (temperature 20°C)
The structural parameters and are expressed in terms of the apparent vis-
cosity in order to find k2. Equating the rheological model of Eq. (3) to the
definition of the apparent viscosity which, in this case, is a function of both shear
and time, it is found:
(6)
Equation (6) is valid for all values of including at (the equilibrium of

the apparent viscosity). Differentiating Eq. (6), with respect to time at a constant
shear rate and combining the resulting equation with Eq. (4), yields:
, (7)
Where
(8)
, ,
Integrating equation (7) at constant shear rate from at 0
yields:
(9)
The equilibrium viscosity was found by fitting the time dependence of the
viscosity (Fig. 5) by an exponential decay curve (Hammadi et al. 2014) of the
form:
+ exp (10)
Where β is the rate constant in (s-1)

Fig.7 shows the variation of initial viscosity and equilibrium viscosity

as a function of shear rate
0,09
initial viscosity η0 and equilibrium viscosity η e (Pa.s)

0,08
η0
ηe
0,07
0,06
0,05
0,04
0,03
50 1 00 150 200 25 0
-1
shear rate (s )
Fig. 7 Initial viscosity and equilibrium viscosity determined by equation (10) as a function
of shear rate (60% of cutting oil emulsion)
Using Eq. (9), a plot of versus t’, at constant shear rate, should yield a
straight line with slope equal to . The same procedure is repeated at different
shear rate. Fig.8 shows the data for 60% cutting oil emulsion.
10000
-1
250 s
-1
200 s
8000 -1
150 s
-1
100 s
1/(η-ηe) [1/(Pa.s)]
-1
6000 50 s
Fitting
4000
2000
0 5 10 15 20 25 30
Time(s)
Fig. 8 Linear relationship between and time for different shear rates (60% cutting oil
emulsion) at temperature 20 °C. The solid line corresponds to a fitting of data by linear
regression.
Fig.9 shows the evolution of the parameters a1 and k2 as a function of shear rate
for 60 % of cutting oil emulsion. It was found that both a1 and k2 can be described
by a power function of shear rate [7]:
(11)
(12)
With
1.567 0.334; 0.961 0.0390, 0.996
0.965 0.190; 0.460 0.039, 0.982
The value of the structural parameter was determined at each shear rate by
using Eq. (13).
(13)
Where are the initial viscosity and the equilibrium viscosity deter-
mined by equation (10).
350
12
K2
300
Fitting
10
250
8
a1 [1/(Pa.s2)]
K2 (s )
-1
200
6
150
4
100
a1
2
50
0
50 100 150 2 00 250
-1
shear rate (s )
Fig. 9 Parameters a1 and k2 of Tiu and Boger model as a function of shear rate for 60 % cut-
ting oil emulsion
Fig.10 shows the evolution of equilibrium structural parameter as a function of

shear rate.
The Herschel-Bulkley model was found to describe well the relationship be-
tween equilibrium structural parameter and shear rate:
.
λ 0.934 0.0142 , 0.999 (14)
As the rate equation Eq.(4), cannot be integrated analytically for varying shear
rate conditions, further progress is still necessary to predict shear stress at given
constant shear rate. In order to derive an equation that would predict the structural
parameter, for a constant shear rate after a given time, Eq. (6) was rearranged to
yield
(15)
Fig.11 shows the evolution of structural parameter as a function of time for

each values of shear rate. It can also be seen that structural parameter changes are
most significant during the initial shear period, after which they remain constant.
0,9 95
Equilibrium structural λ e (-)
0,9 90
0,9 85
60% Vol
Herschel-Bulkley model
0,9 80
0,9 75
50 1 00 150 200 25 0
-1
Shear rate (s )
Fig. 10 Equilibrium structural parameter versus shear rate at 20°C together with Herschel-
Bulkley model (60% of cutting oil emulsion)
1,0 00
. -1
γ =250 s
0,9 95 . -1
γ =200 s
Structural parameter λe (-)

. -1
γ =150 s
0,9 90
. -1
γ =100 s
0,9 85
0,9 80 . -1
γ =50 s
0,9 75
0 20 40 60 80 100 12 0 140
Time (s)
Fig. 11 Structural parameter as a function of time at 20° (60% of cutting oil emulsion)
4 Conclusion
In this study the effect of temperature on the rheological behavior of cutting oil
emulsion and the effect of shear rate on structure buildup and breakdown of cut-
ting oil emulsion have been performed. The model of Herschel-Bulkley, whose
parameters are temperature dependent, was found to be well correlated with the
behavior of cutting oil emulsion. With increased temperature, the yield stress and
the consistency index decreased, while the flow index increased. The limit viscosi-
ty of the cutting oil emulsion decreased with increasing temperature, and their
relationship could be modeled by the Arrhenius equation.
The time dependant viscosity decreased rapidly with shearing time and reached
an equilibrium stage. In order to take into account the structural evolution of the
cutting oil emulsion, the model of Herschel-Bulkley was modified by the introduc-
tion of a structural parameter which allows a good description of the thixotropic
behavior of the cutting oil emulsion.
References
Bensadok, K., Benammar, S., Lapicque, F., Nezzal, G.: Electrocoagulation of cutting oil
emulsions using aluminium plate electrodes. Journal of Hazardous Materials 152,
423–430 (2008)
Butler, F., McNulty, P.: Time dependent rheological characterization of buttermilk at 5°C.
Journal of Food Engineering 25, 559–580 (1995)
Cheng, D.H., Evans, F.: Phenomenological characterization of the rheological be-havoiur
of inelastic reversible thixotropic and antithixotropic fluids. Br. J. Appl. Phys. 16,
1599–1617 (1965)
Dullaert, K., Mewis, J.: Thixotropy: Build-up and breakdown curves during flow. Journal
of Rheology 49, 1213–1230 (2005)
Hammadi, L., Ponton, A., Belhadri, M.: Temperature effect on shear flow and thixo-tropic
behavior of residual sludge from wastewater treatment plant. Mechanics of Time-
Dependent Materials 17, 401–412 (2013)
Hammadi, L., Boudjnane, N., Belhadri, M.: Effect of polyethylene oxide (PEO) and shear
rate on rheological properties of bentonite clay. Applied Clay Science 99, 306–311
(2014)
Liu, J., Han, R., Zhang, L., Guo, H.: Study on lubricating characteristic and tool wear with
water vapor as coolant and lubricant in green cutting. Wear 262, 442–452 (2007)
Moore, F.: The rheology of ceramic and slip bodies. Trans. Br. Ceram. Soc. 58, 470–494
(1959)
Mujumdar, A., Beris, A.N., Metzner, A.B.: Transient phenomena in thixotropic systems.
Journal of Non-Newtonian Fluid Mech. 102, 157–178 (2002)
Pottirayil, A., Kailas, S.V., Biswas, S.K.: Lubricity of an oil in water emulsion in metal cut-
ting: The effect of hydrophilic/lypophilic balance of emulsifiers. Colloids and Surfaces
A: Physicochem. Eng. Aspects 384, 323–330 (2011)
Stachowiak, G.W., Batchelor, A.W.: Physical Properties of Lubricants. Engineering Tribol-
ogy, 3rd edn., pp. 11–50 (2006)
Tiu, C., Boger, D.V.: Complete rheological characterisation of time dependent food prod-
ucts. J. Texture Studies 5, 329–338 (1974)
A Comparison between Two Reliability Based
Optimization Strategies of Tuned Mass Damper
Parameters under Stochastic Loading
Elyes Mrabet1, Mohamed Guedri1, Samir Ghanmi1, Mohamed Soula2,

and Mohamed Ichchou3
1
Unité de Recherche en Dynamique des Structures, Modélisation et Ingénierie des Systèmes
Multi-physiques, Preparatory Engineering Institute of Nabeul (IPEIN), 8000 M’rezgua,
Nabeul, Tunisia
2
Laboratory of applied mechanics and engineering, ENIT. Mechanical engineering department,
École Nationale Supérieure des Ingénieurs de Tunis, University of Tunis, 5, Av.
Taha Hussein Montfleury 1008 Bab Menara TUNIS Tunisia
3
LTDS UMR5513 Ecole Centrale de Lyon, 36 Avenue Guy de Collongue,
69134 Ecully, France
Abstract. In this work two reliability based optimization (RBO) strategies of

Tuned Mass Damper (TMD) parameters are presented and compared. The TMD
device is attached on a primary structure modeled by a generic one degree of free-
dom system. The first strategy is based on the displacement response of a primary
structure whereas the second one is based on a new approach related to the dissi-
pated power (DPO). Analytic expressions of the failure probabilities are given
based on the Rice's formula. The obtained results proved that both strategies are
earthquake dependant for earthquake frequency close to the main system frequen-
cy. A correlation analysis is made and it has been found that the new RBO strate-
gy is more robust in the sense that it can provide the result of the displacement
RBO with a small relative error.
Keywords: Reliability based optimization of TMD, Energy approach, Non-

Gaussian processes, Displacement approach, Rice's formula.
1 Introduction
The use of the TMD devices in engineering to mitigate vibrations in structures sub-
ject to stochastic loading has taken wide interests in last years. One of the main ques-
tions asked to the designer is how to choose the optimum parameters of the TMD so
that a certain criterion is met. In the literature (Yu et al. 2013) and (Marano et al.
2010), several strategies can be used such as the RBO (Chakraborty and Roy 2011)
where the objective function is a failure probability related to a certain failure mode.
A failure mode can be the exceedence, for the first time, of the displacement response
(Chakraborty and Roy 2011) or the Von Mises stress (Gupta and Manohar 2005)
through a certain given thresholds during a time interval.

38 E. Mrabet et al.
The evaluation of the failure probabilities in the context of RBO constitutes a

time depending problem of reliability analysis and one of the most common tech-
niques used when handling the RBO is the transformation of the problem into time
invariant format (Gupta and Manohar 2005). Consequently and by making the
Poisson assumptions (Crandall 1970), the classical Rice's formula (Li and Chen
2009) can be used and then the failure probability is easily expressed.
In the present work a new RBO strategy is introduced using the DPO stochastic
process which is strictly related to the damage in vibrating structures (Greco and
arano 2013). The strategy consists to get optimum TMD parameters so that the
failure probability characterized by the out-crossing of the DPO, for the first time,
across a certain threshold during a time interval. The expression of the failure
probability is presented in this work for a generic system of one degree of freedom
subject to stochastic loading. Once the RBO strategy is formulated, the effective-
ness (vs. earthquake frequency) of the TMD device, for both strategies is dis-
cussed and a correlation analysis is made. The obtained results show up a strong
correlation between these strategies and equivalence can be made.
2 Structural Model and Dissipated Power Expression
Figure 1 shows an idealized mechanical model of a TMD attached to a primary

structure. The TMD device is described by mass mT , spring stiffness kT and
damping cT . The single degree of freedom primary structure is characterized by
mass ms , spring stiffness k s and damping c s .
mT
TMD
kT cT yT
ms
Primary
ks cs ys
structure
yb
Fig. 1 The primary structure with TMD device
The system is subjected to a base acceleration yb due to seismic motion. Let
y(t ) = ( yT , ys )T be the relative displacement vector of the TMD device and the
A Comparison between Two Reliability Based Optimization Strategies 39
primary structure, respectively, the governing equation (Greco and Marano 2013)
of the dynamic system can be expressed as follows:
My(t ) + Cy (t ) + Ky (t ) = − Mryb (1)
⎛ mT 0 ⎞
where the overdot represents the time derivative, M = ⎜⎜ ⎟,
⎝ 0 ms ⎟⎠
⎛ cT − cT ⎞ ⎛ k − kT ⎞
C = ⎜⎜ ⎟ , K = ⎜⎜ T ⎟ and r = [1 1]T .
⎝ − cT cT + c s ⎟⎠ ⎝ − kT kT + k s ⎟⎠
Introducing the state space vector Z 0 = ( yT , y s , yT , y s )T , Eq.(1) can be ex-

pressed as follows: Z 0 = A 0 Z 0 + r0 yb where, r0 = [0 0 1 1] and
T
⎛ 0 0 1 0 ⎞
⎜ ⎟
⎜ 0 0 0 1 ⎟
A0 = ⎜ ⎟
− ωT2 ωT2 − 2ξT ωT 2ξT ωT
⎜ ⎟
⎜ μω 2 − ( μωT2 + ωs2 ) 2μξT ωT − (2μξT ωT + 2ξ sω s ) ⎟⎠
⎝ T
In the above expression, μ = mT ms is the mass ratio, ωT = kT mT and

ξT = cT 2 kT mT are the natural frequency and the damping ratio, respectively,
of the TMD device, whereas ω s = k s ms and ξ s = cs 2 k s ms are the natural
frequency and the damping ratio, respectively, of the primary structure.
The system is excited by a base acceleration yb due to seismic motion which is
modeled by the well known Kanai-Tajimi stationary stochastic process (Marano et
al. 2010). Then, the base acceleration satisfies the following equations:
y f + 2ξ f ω f y f + ω 2f y f = − w(t ) and yb = y f + w(t ) = − (2ξ f ω f y f + ω 2f y f ) (2)
where w(t ) is a stationary zero mean white noise process applied at the bed rock.
It represents the excitation of the dynamic system and it has intensity S 0 . ξ f , ω f
and y f are the damping ratio, the natural frequency and the relative response
(with respect to the ground) of the elastic filter (Chakraborty and Roy 2011). Con-
sidering the Kanai-Tajimi model, the global space state vector of the mechanical
( )
system is given by Z = yT , y s , y f , yT , y s , y f T and the equation of motion can be
40 E. Mrabet et al.
written in the space state form (Yu et al. 2013) as Z = AZ + f where

f = [0 0 0 0 0 − w(t )] and
T
⎛ 0 0 0 1 0 0 ⎞
⎜ ⎟
⎜ 0 0 0 0 1 0 ⎟
⎜ 0 0 0 0 0 1 ⎟
A=⎜ ⎟
⎜ − ωT 2 ωT 2 ωf 2 − 2ξT ωT 2ξ T ωT 2ξ f ω f ⎟
⎜ ⎟
⎜ μωT − ( μωT + ω s ) ωf 2 2μξ T ωT − 2(ξ sω s + μξT ωT ) 2ξ f ω f ⎟
2 2 2
⎜ 0 0 −ωf
2
0 0 − 2ξ f ω f ⎟⎠
⎝
The stochastic response of the mechanical system is completely known by the

space state covariance matrix R which is the solution of the Lyapunov equation
given, in stationary condition, as follows:
AR + RA T + B = 0 (3)
where B has all zero elements, except the last one [B]( 6, 6 ) = 2πS0 . The root mean
square displacement (rmsd) of the primary system can be obtained as
σ y = R (2,2) . The root mean square velocity (rmsv) is expressed as
s
σ y = R (5,5) . In stationary condition, the DPO in the primary structure and its
s
time derivative can be written as follows (Greco and Marano 2013):
e(t ) = c s y s2 (t ) = 2ξ s ω s ms y s2 (t ) and e(t ) = de(t ) dt = 4ξ s ω s ms y s (t ) ys (t ) (4)
Let . denotes the expectation, the acceleration covariance matrix is related to

the covariance matrix R and it is given by:
⎛ σ y2T yT ys ⎞

R yy = yyT = ⎜ ⎟ = (D1 D ) R (D1 D )T ,where
⎜ ys yT σ ys ⎟⎠
2 2 2
⎝
⎛ − ωT 2 ωT 2 ω f 2 ⎞⎟
D1 = ⎜ and
⎜ μωT − ( μωT 2 + ω s 2 ) ω f 2 ⎟
⎝ ⎠
⎛ − 2ξT ωT 2ξT ωT 2ξ f ω f ⎞
D 2 = ⎜⎜ ⎟.
⎟
⎝ 2μξT ωT − 2(ξ sω s + μξT ωT ) 2ξ f ω f ⎠
Hence, the root mean square σ y of the relative acceleration ys can be imme-
s
diately deduced and written as σ y = R yy (2,2) .

s
3 The Proposed Reliability Based Optimization Strategies
In the present work, two optimization strategies are considered in order to get op-
timum design parameters of the TMD. The first strategy is the classical approach
for the RBO problems (Chakraborty and Roy 2011) where the failure probability
is related to the primary structure displacement ys , representing the performance
quantity. This RBO strategy consists to find the optimum design parameters
d = (ωT , ξT ) of the TMD device that minimizes the failure probability Pf , D cha-
racterized by the exceedence, for the first time, of the displacement ys of a certain
threshold value β during some interval time [0, T]. The second strategy is a new
RBO problem. It's related to the DPO in the primary structure e(t ) . This RBO
problem consists to find d that minimizes a failure probability Pf , E characterized
by the up-crossing, for the first time, of e(t ) across a certain threshold value
β E during some interval time [0, T]. Since the system described in Fig.1 is subject
to stochastic loading, the evaluation of both quantities Pf , D and Pf , E constitute a
time dependant problem of reliability analysis that can be transformed into time
invariant problem (Gupta and Manohar 2005) by introducing the random variables
y s , max = max y s (t ) for the displacement approach and emax = max e(t ) for the
0≤ t ≤T 0≤ t ≤T
DPO approach, respectively. Considering the Poisson assumptions (Li and Chen
2009) and adopting the classical Rice's formula, the failure probabilities can be
approximated by:
(
Pf , D ≈ 1 − exp − υ β ( β )T ) (
and Pf , E ≈ 1 − exp − υ β ( β E )TE
) (5)
where υ β ( β ) and υ β ( β E ) are the mean out-crossing rates of the displacement

E
and the DPO processes, respectively. These mean out-crossing rates can be
obtained using the Rice's formula as follows:
∞ ∞
υ β ( β ) = ∫ y s . pY Y (β , y s )dy s
s s
and υ β ( β E ) = ∫ e. p EE (β E , e)de
E
(6)
0 0
where pY Y and pEE are the joint probabilities functions (PDF) of the considered
s s
processes and their time derivatives. The main difficulty when handling reliability
analysis using the Rice's formula resides in the computation of mean out-crossing
rates which are function of the joint PDF that cannot be always easy to obtain ana-
lytically (Gupta and Manohar 2005).
42 E. Mrabet et al.
3.1 Formulation of the RBO Problem with Displacement

Approach
For the RBO problem using the displacement approach the mean out-crossing rate
is easily obtained because both processes Ys (t ) and Ys (t ) are independents statio-
nary zeros means Gaussians processes therefore their corresponding joint PDF
pY Y can be easily calculated and the mean out-crossing is immediately given
s s
s s
(
by υ β = (σ y πσ y ) exp − β 2 2σ y2 s
) and the failure probability is expressed as
{ (
Pf , D = 1 − exp − T (σ y s πσ ys ) exp − β 2 2σ y2s )} . Hence, the RBO problem with
displacement approach can be formulated as follows:
Find d = (ωT , ξT ) To minimize Pf , D (7)
3.2 Formulation of the RBO Problem with Dissipated Power

Approach
Unlike the first RBO strategy (displacement approach), the analytic computation
of the mean out-crossing rate υ β ( β E ) needs some mathematical manipulations
E
because both processes E and E , describing the DPO in the primary structure
and its time derivative, are non-Gaussian processes.
After manipulations, the mean out-crossing rate can be written as
E s s
{ s
}
υ β ( β E ) = (σ y πσ y ) exp − β E 4ξ sω s msσ y2 and the failure probability is im-
mediately given by:
{ s s
(
Pf , E = 1 − exp − T (σ y πσ y ) exp − β E 4ξ sω s msσ y2
s
)} (8)
Eventually, the RBO problem with DPO approach can be formulated as

follows:
Find d = (ωT , ξT ) To minimize Pf , E (9)
4 Numerical Study
The purpose of the numerical study is to analysis the correlation between the pro-
posed RBO problems. The correlation analysis allows establishing relationships
and possible equivalences between the proposed optimization approaches. To this
end, a dimensionless quantity ψ = ω f ω s is introduced to make comparison be-
tween the obtained results of the proposed strategies. The structural parameters are
assumed to be ms = 1 , ω s = 8π , ξ f = 0.4 , ξ s = 0.03 , S 0 = 0.03 , μ = 2% ,
β = 0.045 , β E = 2.6 and T = 10s . For the optimization, the genetic algorithm
routine available in Matlab is used.
Figure 2 gives the optimum design parameters of TMD when both approaches
are applied. From Fig. 2(a) one can observe that, for both strategies, the optimum
natural frequencies are earthquake dependant for ψ < 2 . Indeed, the optimized
natural frequencies are increasing functions until roughly ψ ≈ 1.5 and then
decreasing function until values neighboring ψ ≈ 2 . Besides, it seems that the

natural frequencies become approximately constant for values ψ > 2 which
means an earthquake independence.
26
25.8
25.6
ωT(rd/s)
25.4
25.2
25
( a)
24.8
0.5 1 1.5 2 2.5 3 3.5
ψ
Displacement approach Diss. power approach
0.074
(b)
0.072
ξT
0.07
0.068
0.5 1 1.5 2 2.5 3 3.5
ψ
Fig. 2 Optimum natural frequencies (a) and damping ratios (b), obtained using the
displacement and the DPO approaches
Figure 2 (b) shows that the optimized damping ratios obtained using the
proposed strategies are also earthquake dependent for ψ < 2 and they become in-
dependent for ψ > 2 . In effect, the optimized damping ratios are roughly linear
decreasing function until approximately ψ ≈ 1 where a minimum is reached, then
increasing function until ψ ≈ 2 . In addition, the observed curve slope of the ob-
tained results shown in Fig. 2(b) for ψ < 1 also prove that the TMD performances
are strongly dependant on earthquake. The optimized failure probabilities Pf*, E ob-
tained using the DPO approach and the failure probabilities Pf , E evaluated using
the optimum values when the displacement approach is applied as well as their
44 E. Mrabet et al.
corresponding relative error are shown in Fig.3. The obtained results shown in
Fig. 3(a) confirm the remarks made previously in Fig.2. Indeed, the obtained val-
ues of Pf*, E present large variability vs. ψ and the TMD performance is strongly
earthquake dependant especially for the values neighboring ψ ≈ 1.2 where a max-
imum is reached. It is also observed that the optimized failure probabilities are less
sensitive, to earthquake, for ψ > 2 . Besides, one can see that the curve corres-
ponding to the optimized values are below the curve corresponding to the failure
probability evaluated with the optimum parameters when the displacement ap-
proach is applied.
Figure 3(b) shows the relative error between the optimized failure probabilities
obtained when the DPO is applied and the failure probabilities evaluated using the
optimum TMD parameters obtained when the displacement approach is applied.
The examination of this figure demonstrates that a minimum is reached for
ψ ≈ 1.25 with an error of 3.5%. Besides it has been also found that the maximum
error can attain 9.5% for ψ ≈ 0.6 and it is roughly constant for ψ > 2 .
-3
x 10
1.2
1
Failure probability
RBO (Diss. power approach)

0.8 Pf,E (Displacement optimum)
0.6
0.4
( a)
0.2
0
0.5 1 1.5 2 2.5 3 3.5
ψ
10
Relative error (%)
8
(b)
6
2
0.5 1 1.5 2 2.5 3 3.5
ψ
Fig. 3 (a) Pf*, E obtained using the DPO approach compared with those evaluated using the
optimum values of the displacement approach; (b) the relative error.
Figure 4(a) shows the optimized failure probabilities Pf*, D when the displace-
ment approach is applied compared with the failure probabilities Pf , D evaluated
using the optimum TMD parameters if the DPO approach is applied. The same
observations made previously for Fig. 3(a) can be done for Fig. 4(a). Indeed the
TMD performance is strongly dependant on earthquake for values around ψ ≈ 1.2

then it's less sensitive for ψ > 2 . The curve corresponding to the optimized values
are below the curve corresponding to the failure probability evaluated with the op-
timum parameters when the DPO approach is applied. The relative error between
Pf*, D and Pf , D is given in Fig.4 (b). The inspection of this figure shows that a
minimum of error equal to 3% is reached for ψ ≈ 1.2 and the maximum error is
less than 6% over the entire range of ψ . The comparison of the obtained results in
Fig.3 (b) and Fig.4 (b) demonstrate clearly that the RBO using the DPO approach
is more robust than the RBO using the displacement approach. The robustness
should be understood in the sense that the RBO using the DPO can provide the re-
sults obtained with the RBO using the displacement approach with a maximum
error, over the considered range of ψ , less than 6% which is less that the obtained
error of 9.5% in the reciprocal case.
-3
x 10
8
6 RBO (displacement approach)

Failure probability
Pf,D (Diss.power optimum)
2
( a)
0
0.5 1 1.5 2 2.5 3 3.5
ψ
6
Relative error (%)
5
(b)
4
3
0.5 1 1.5 2 2.5 3 3.5
ψ
Fig. 4 (a) Pf*, D obtained using the displacement approach compared with those evaluated
using the optimum values obtained with the DPO approach; (b) the relative error.
5 Concluding Remarks
In this paper two RBO problem of TMD parameters under stochastic loading are
presented and compared. The selection of one of these strategies totally depends
on the designer aim. The first strategy is based on a displacement approach so that
the optimal TMD parameters are obtained by minimizing the failure probability
46 E. Mrabet et al.
explicitly related to the covariance responses of the primary structure. The second
RBO strategy is a new approach in RBO problems. The approach uses the DPO in
the primary structure. The introduction of this strategy is motivated by the strictly
relationship between the DPO in the primary structure and the damage level of the
vibrating structure.
The expression of the failure probability, for the DPO approach, is presented
and a comparison between the presented RBO strategies is made. The obtained re-
sults show that both strategies are strongly earthquake dependent on values of
earthquake frequencies close to main system frequency and they are less sensitive
outside of this frequency.
The correlation analysis made between the presented strategies showed that
the RBO with DPO approach is more robust in the sense that it can provide the
optimized TMD parameters if the RBO displacement is applied with a maximum
relative error less than 6%. Besides, both RBO strategies are equivalent, from ro-
bustness point of view, for frequencies close to the main system one because the
relative error is roughly the same.
References
Yu, H., Gillot, F., Ichchou, M.N.: Reliability based robust design optimization for tuned
mass damper in passive vibration control of deterministic/uncertain structures. J. Sound
Vib. 332, 2222–2238 (2013)
Marano, G.C., Greco, R., Sgobba, S.: A comparison between different robust optimum de-
sign approaches: Application to tuned mass dampers. Probab. Eng. Mech. 25, 108–118
(2010)
Chakraborty, S., Roy, B.K.: Reliability based optimum design of tuned mass damper in
seismic vibration control of structures with bounded uncertain parameters. Probab. Eng.
Mech. 26, 215–221 (2011)
Gupta, S., Manohar, C.S.: Probability Distribution of Extremes of Von Mises Stress in ran-
domly vibrating structures. J. Vib. Acoust. 127, 547–555 (2005)
Crandall, S.H.: First crossing probabilities of the linear oscillator. J. Sound Vib. 12(3),
285–299 (1970)
Li, J., Chen, J.: Stochastic dynamics of structures. John Wiley, Asia (2009)
Greco, R., Marano, G.C.: Optimum design of Tuned Mass Dampers by displacement and
energy perspectives. Soil Dyn. Earthq. Eng. 49, 243–253 (2013)
Modeling and Simulation for Vertical Rail
Vehicle Dynamic Vibration with Comfort
Evaluation
Mortadha Graa1, Mohamed Nejlaoui1, Ajmi Houidi2, Zouhaier Affi1,

and Lotfi Romdhane2,3
1
Laboratoire de Génie Mécanique(LGM),
Ecole Nationale d’Ingénieurs de Monastir, Université de Monastir, Tunisie
2
Laboratoire de Mécanique de Sousse (LMS),
Ecole Nationale d’Ingénieurs de Sousse, Université de Sousse, Tunisie
3
On leave at the College of Engineering, The American University of Sharjah, UAE
Abstract. Investigation of vibration is an important topic for the purposes of ride

comfort in railway engineering. The vibration of rail vehicles becomes very com-
plex because it is affected by the condition of vehicles, including suspensions and
wheel profile, condition of track sections, including rail profile, rail irregularities,
cant and curvature. The present study deals with the effects of railway track im-
perfections on dynamic behavior, and investigates the effect of vehicle speed and
the rail irregularity on ride comfort through numerical simulation. The numerical
simulation of the vertical dynamic behavior of a typical railroad vehicle will be
performed using Largrangian dynamics. The model consists of 17 degrees of free-
dom with 4 wheelsets, 2 bogies and a car body. For the assessment of the ride
comfort, the Sperling ride index (ISO2631) is calculated using filtered RMS acce-
lerations. The ride characteristics of the vehicle provide an assessment of the
dynamic behavior of the vehicle through the analysis of the accelerations at the
vehicle body, whereas the ride comfort assesses the influence of the vehicle dy-
namic behavior on the human body. A parametric study was carried out to suggest
design modifications in order to improve the level Sperling index.
Keywords: Sperling index, ride quality, rail vehicle, dynamic behavior, ride com-
fort, vertical dynamic vibration.
1 Introduction
Rail transport is one of the major modes of transportation, so it must offer a high
comfort level for passengers and crew. However, the comfort that passengers ex-
perience is usually perceived differently from one individual to another. In several
research works, noise and vibration have been identified as the most important
factors for high comfort. The main sources of vibration in a train are track defects

48 M. Graa et al.
in welding or rolling defects, rail joints, etc. The nature of vibration itself is ran-
dom and covers a wide frequency range [1]. The improvement of the passenger
comfort while travelling has been the subject of intense interest for many train
manufacturers, researchers and companies all over the world. Although new tech-
niques in manufacturing and design ensure better ride quality in railway carriages,
it is sometimes impossible to completely eliminate track defects or various ground
irregularities. The dynamic behavior of a train also depends on the load and and its
mechanical components, such as springs, dampers, etc., which interact with the
wheels, the train body and bogies. The dynamic performance of a rail road vehicle
as related to safety and comfort is evaluated in terms of specific performance in-
dices. The quantitative measurement of the ride quality is one of such performance
indices. Ride quality is interpreted as the capability of the rail road suspension to
maintain the motion within the range of human comfort. Sperling’s ride index is a
measure of the ride quality and ride comfort used by ISO 2631[2]. Due to the
complex dynamics that exists between the rail and wheel, rail vehicle dynamics
are often difficult to model accurately. This velocity-dependent dynamics justifies
the importance of the track input to railcar modeling. In the physical system, the
input comes from the actual track. In a model, a user-defined input is used to pre-
dict the actual track characteristics. The user-defined input can be created analyti-
cally or can be based on actual measurements. Measured track data are obtained
by running a specialized railcar down the track. Analytic track data are created us-
ing mathematical shapes, such as cusps, bends, and harmonic functions, to
represent the track geometry [3].
There have been several studies, which dealt with the dynamic analysis of rail
vehicles in order to enhance the ride comfort while travelling. Nejlaoui et al [4]
optimized the structural design of passive suspensions in order to ensure simulta-
neously passenger safety and comfort. Abood et al [5] investigated the Railway
carriage simulation model to study the influence of vertical secondary suspension
stiffness on ride comfort of railway car body. Kumar and Sujata [2] presented the
numerical simulation of the vertical dynamic behavior of a railway vehicle and
calculated Sperling ride index for comfort evaluation. Nielsen and Igeland [6] in-
vestigated the vertical dynamic behavior for a railway bogie moving on a rail
which is discretely supported by sleepers resting on an elastic foundation. Effects
of imperfections on the running surfaces of wheel and rail were studied by assign-
ing irregularity functions to these surfaces.
2 Modeling of Rail Road Vehicle

To analyze the dynamic behavior of railway vehicles, usually the vehicle (and if
necessary the environment) is represented as a multi body system. A multi body
system consists of rigid bodies, interconnected via massless force elements and
joints. Due to the relative motion of the system’s bodies, the force elements gener-
ate applied forces and torques. Typical examples of such force elements are
springs, dampers, and actuators combined in primary and secondary suspensions
of railway vehicles.
Modeling and Simulation for Vertical Rail Vehicle Dynamic Vibration 49
2.1 Assumptions
The assumptions made in formulating the model are as follows:
• Bogie and car body component masses are rigid.

• The springs and dampers of the suspension system elements have linear
characteristics.
• Friction does not exist between the axle and the bearing.
• The vehicle is moving with constant velocity on a rigid and constant
gauge.
• All wheel profiles are identical from left to right on a given axle and from
axle to axle and all wheel remain in contact with the rails.
• Straight track.
• An irregularity in the vertical direction with the same shape for left and
right rails.
2.2 Rail Road Vehicle Model

Figure.1 illustrates the train vehicle model adopted in this study. It consists of a
vehicle body, two bogies frames and four wheelsets. Each bogie consists of the
bogie frame, and two wheel sets. The car body is modeled as a rigid body having a
mass Mc; and having moment of inertia Jbx and Jcy about the transverse and longi-
tudinal axes, respectively. Similarly, each bogie frame is considered as a rigid
body with a mass Mb (Mb1 and Mb2) with moment of inertia Jbx and Jby about the
transverse and longitudinal axes, respectively. Each axle along with the wheel set
has a mass Mw (for four axles Mw1; Mw2; Mw3 and Mw4). The spring and the shock
absorber in the primary suspension for each axle are characterized by a spring
stiffness Kp and a damping coefficient Cp, respectively. Likewise, the secondary
suspension is characterized by a spring stiffness Ks and a damping coefficient Cs,
respectively. As the vehicle car body is assumed to be rigid, its motion may be de-
scribed by the vertical displacement (bounce or Zc) and rotations about the trans-
verse horizontal axis (pitch or Φc) and about the longitudinal horizontal axis (roll
or θc). Similarly, the movements of the three bogie units are described by three
degrees of freedom Zb; θb and Φb, each about their centers. Each axle set is de-
scribed by two degrees of freedom Zw; and Φw. about their centers. Totally, 17 de-
grees of freedom have been considered in this study for the vehicle model shown
in Figure.1. The detailed parameters regarding the moment of inertia and mass of
different component are given in Table 1.
Table 1 Detailed parameter of rigid bodies
Moment of Inertia( Kg.m² )

Name of Rigid Bodies Mass ( Kg ) IXX IZZ
Car Body 6,7 ×105 105 106
Bogie-I and II 105 105 105
Wheel-set-I,II,III and IV 4000 4000 4000
50 M. Graa et al.
Some parameters regarding the rigid bodies are already given in Table 1; how-
ever, the other parameters, which are essential for the simulation of the vehicle,
are presented in Table 2.
2Lb
2b2
2Ld Y
h3
Scondary
Ks Cs h2 Suspension
h1 Kp Cp Primary
R Suspension
Z
2b1 K Hertizian
y(lateral) 2a
(pitching)
Z
Speed V
x (logitudinal)
(rolling)
(yawing)
z(vertical)
Fig. 1 physical model of railway vehicle
A typical rail road vehicle system is composed of various components such as car
body, springs, dampers, Bogies, Wheel-set, and so forth. When such dynamic sys-
tems are put together from these components, one must interconnect rotating and
translating inertial elements with axial and rotational springs and dampers, and al-
so appropriately account for the kinematics of the system structure.
3 Track Inputs to Rail Road Vehicle

The dynamic wheel loads generated by a moving train are mainly due to various
wheel/track imperfections. These imperfections are considered as the primary
source of dynamic track input to the railroad vehicles. Normally, the imperfections
that exist in the rail-track structure are associated with the vertical track profile,
cross level, rail joint, wheel flatness, wheel/rail surface corrugations and some-
times uneven support of the sleepers.
In actual practice different types of periodic, a-periodic or random track irregu-
larities may exist on the track, but in the present study bump type of irregularity is
considered as shown in Figure.2 [3]. The shape of the irregularity is assumed to be
similar on the left and the right rails.
Table 2 Vehicle parameters
Parameter Nomenclature Values

Primary spring stiffness Kp 106 N/m
Secondary spring stiffness Ks 1,7×10 6N/m
Primary damping coefficient Cp 6×104 Ns/m
Secondary damping coefficient Cs 105 Ns/m
Vertical hertz spring stiffness Khz 35×109 N/m
Longitudinal distance between bogies I and II and car body mass center Lb 6m
Longitudinal distance between wheel-set and corresponding bogie origin Ld 1,4 m
Lateral distance between a longitudinal primary suspension and corresponding wheel-set dp 1m
Lateral distance between longitudinal secondary suspension and corresponding bogie origin ds 1m
Lateral distance between contact point of wheel–rail and corresponding wheel-set origin a 0,7163 m
Lateral distance between vertical primary suspension and corresponding wheel-set origin b1 1m
Lateral distance between vertical secondary suspension and car body mass center b2 1m
Nominal wheel radius R1 0,61 m
Vertical distance between wheel-set and bogie mass centers h1 0,3 m
Vertical distance between bogie mass center and lateral secondary suspension h2 0,2 m
Vertical distance between lateral secondary suspension and car body mass center h3 1,3 m
X
H
Fig. 2 Model of track irregularity
The bump excitations of the left wheels (Figure. 3) of leading bogies are as
follows:
⎧H ⎡ ⎛ V ⎞⎤
⎢1 − cos ⎜⎝ 2π L ( t − tdi ) ⎟⎠⎥
L
⎪ for tdi ≤ t ≤ + tdi
Z ri = ⎨ 2 ⎣ ⎦ V (i = 1..4) (1)
⎪0
⎩ otherwise
⎡ 2 Ld 2 Lb 2 Lb + 2 Ld ⎤
Where [t d 1 , t d 2 , td 3 , td 4 ] = 0, (2)
⎢⎣ V , V , V ⎥⎦
In the present study, H is taken as 0.03m and L is taken as 1m.
4 Simulation Study
The models were built in the MATLAB/ Simulink® environment. The fixed step
solver ODE-45 (Dormand-Prince) was utilized, with the sampling time
Ts=0.0001. Ts is smaller than the fastest half-car Active Vehicle Suspension Sys-
tems (AVSS) model dynamics, enabling observation of all model dynamics [7, 8].
52 M. Graa et al.
Dynamic analysis was carried out for the vehicle at different speeds: 15m/s,
30m/s, 45m/s and 60m/s.
-3
0.03 2.5
x 10
Zr1 15 m/s²
Bum p e xcita tion (m )
Zr2 30 m/s²
0.025 2 45 m/s²
Zr3 60 m/s²
Car body displacement (m)

0.02 Zr4 1.5
Zr5
0.015 Zr6 1
0.5
0.01
0
0.005
-0.5
0
0 0.2 0.4 0.6 0.8 1
-1
Time(s) 0 0.5 1 1.5 2 2.5
Time(s)
Fig. 3 The bump excitations of the left Fig. 4 The vertical Car body displacement
wheels vehicle speed of 15m/s for differents vehicle speeds
4.1 Dynamics Analysis

The following output parameters are evaluated:
• Vertical displacement and acceleration at the floor of the car-body center

of mass.
Vertical acceleration at the front and the rear bogie center pivot.
The displacement and acceleration responses of the carbody at speeds of 15m/s,
30m/s, 45m/s, 60m/s are shown, respectively, in figure.4 and figure.5. Plots show
that initially, the value of acceleration is nearly equal to 0 m/s2, which is mainly
the acceleration without gravity. Finally it goes to zero, when the vibration of the
car body ceases and stabilizes. The acceleration is generally within an acceptable
range and does not show any instability.
The acceleration response of front and the rear bogie with time is presented at
different velocities of vehicle in figure.6.a) and figure.6.b), respectively. It is clear
from the plots that initially the wheels of the front bogies come in contact with the
track irregularity and the vibration starts in the front bogie and later these vibra-
tions are shifted to the rear bogies. The amplitude of the vehicle vibration also
increased with the vehicle speed.
4.2 Sperling Ride Index

Sperling’s ride index is defined as [9, 10, 11]:
1
⎛ n f 10 ⎞10
WZ = ⎜⎜ ∑ WZi ⎟⎟ (3)
⎝ i =1 ⎠
Where nf is the total number of discrete frequencies of the acceleration response

of the railway vehicle identified by the FFT and WZi is the comfort index corres-
ponding to the itch discrete frequency, given by:
1
WZ = ⎡⎣ ai2 B( fi ) 2 ⎤⎦ 6.67 (4)
Where ai denotes the amplitude of the peak acceleration response (in cm/s²)
measured on the floor of the itch frequency identified by the FFT and B(fi) a
weighting factor, given by:
1.911 f 2 + ( 0.25 f 2 )
2
B( f ) = k (5)
(1 − 0.277 f 2 )2 + (1.563 f − 0.0368 f 3 )2
Where k = 0.737 for horizontal vibration and 0.588 for vertical vibration.
4.3 Comfort Evaluation

Acceleration frequency response plots were generated for car body at vehicle
speeds of 15m/s to 60 m/s and are shown in figure.5.b) to calculate the Sperling
ride comfort index. The FFT plot is generated for a frequency range between 0 to
25 Hz, as the human beings are most sensitive in the frequency range of 4 to 12.5
Hz. Ride comfort analysis has been performed for speeds ranging from 15m/s to
60m/s.
The analysis has been performed on the system model to calculate the vertical
acceleration of the system. FFT output is taken to get the peak acceleration fre-
quency component. Comfort index has been calculated through Eqs[3-5], which
are presented in Table.3.
1000
1 15 m/s² 15 m/s²
30 m/s² 30 m/s²
45 m/s² 45 m/s²
Car body acceleration (cm/s²)
60 m/s² 60 m/s²
Car body Acceleration (m/s²)
0.5 500
0 0
-0.5
-500 0 1 2 3
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 10 10 10 10
Time(s) Frequency (Hz)
(a) (b)
Fig. 5 The vertical Car body acceleration in time domain (a) and frequency domain (b) for
vehicle speed of 15m/s, 30m/s, 45m/s and 60m/s
54 M. Graa et al.
Table 3 Sperling’s ride comfort index an evaluation for different vehicle velocities
Vehicle Speed Sperling Index Ride comfort evaluation

(m/s) (Wz)
15 2.53 More pronounced but not unpleasant
30 2.03 Clearly noticeable
45 1.72 Just noticeable
60 1.62 Just noticeable
200 200
15 m/s²
15 m/s² 150 30 m/s²
150
30 m/s² 45 m/s²
Front bogie acceleration (m/s²)
Rear bogie acceleration (m/s²)

45 m/s² 60 m/s²
100 100
60 m/s²
50 50
0 0
-50 -50
-100 -100
-150 -150
-200 -200
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0 0.5 1 1.5
Time(s) Time(s)
(a) (b)
Fig. 6 The vertical Front (a) and Rear (b) bogie acceleration for vehicle speed of 15m/s,
30m/s, 45m/s and 60m/s
The maximum and minimum ISO Sperling Index values are respectively 2.53
and 1.62 for the rail vehicle speed respectively 15 m/s and 60 m/s. These values
respectively indicate “The more pronounced but not unpleasant” and “just notice-
able” zones. Figure 7 show that the comfort index decreases as the vehicle speed
increases while maintaining an acceptable level of comfort. This means that the
passengers are not much affected by the vibration as they are exposed to low level
of vibrations.
2.6
More pronounced but not unpleasant

2.4
Ride index Wz
2.2
2
Clearly noticeable
1.8
1.6
10 20 30 40 50 60
Speed(m/s)
Fig. 7 The Sperling index comfort variation for different speeds

4.4 Parametric Study

A parametric study was undertaken to find the influence of different suspension
parameters on the Sperling index (Wz). The parameters considered for the analysis
were primary stiffness, secondary stiffness, primary damping and secondary
damping. The results of the parametric study have been plotted in terms of per-
formance index such as the Wz index versus vehicle parameter. Figures 8 and 9
show the variation of Wz index as a function of stiffness and damping respectively
(primary and secondary). The increase in the primary suspension stiffness reduces
the Wz index at the carbody marginally up to a speed of 60 m/s. The influence of
secondary stiffness has been found to be just the opposite of the primary stiffness.
A reduction of the secondary stiffness value from the present value reduces the
Wz index at the carbody at all the speeds. The variation of the primary damping is
seen to have little influence on the Wz index at the carbody. At speeds above 45
m/s, increase in the primary damping is shown to produce marginal reduction in
the Wz index at the carbody. The secondary damping has great influence at speeds
higher than 30 m/s.
2.8 2.8
C arb o d y Sp erlin g In d ex W z
2.6 2.6
Carbody Sperling Index Wz
2.4 15 m/s 2.4

30 m/s
2.2 45 m/s 2.2
60 m/s
2 2
15 m/s
1.8 1.8 30 m/s
45 m/s
1.6 1.6 60 m/s
1.4 1.4
0 500 1000 1500 2000 2500 1.65 1.7 1.75 1.8 1.85 1.9
Primary stiffness Kp (N/mm) Secondary stiffness Ks (N/mm) x 10
4
a) b)
Fig. 8 The influence of Primary a) and secondary b) stiffness
An increase in the secondary damping reduces the Wz index at speeds greater

than 30 m/s, whereas up to 30 m/s speed, the secondary damping has little influ-
ence. It has also been observed that all the primary damping have very limited
influence at low speeds.
56 M. Graa et al.
2.6 3
Carbody Sperling index Wz 2.4
Carbody Sperling index Wz

2.5
2.2
15 m/s
2 30 m/s
2
45 m/s
1.8 15 m/s 60 m/s
30 m/s
1.6 1.5
45 m/s
60 m/s
1.4
1
1.2
1 0.5
500 600 700 800 900 1000 6 8 10 12 14 16 18 20 22
Primary damping Cp (Ns/mm) Secondary damping (Ns/mm)
a) b)
Fig. 9 The influence of Primary a) and secondary b) damping
5 Conclusion
Vertical dynamic analysis has been carried out for a Railway Vehicle. A 17 degree
of freedom model is used for the analysis. Velocity input at all the wheelset is giv-
en by considering similar bump irregularities at both right and left rail. A vertical
acceleration response at the car body has been calculated in the frequency domain.
Sperling Ride index has been calculated for the above vehicle. The Sperling Ride
Index values at different speeds are presented. The calculated values of the Sperl-
ing index are found well in the satisfactory limits defined by the ISO 2631 stan-
dard which means that the passengers are not much affected by the vibration as
they are exposed to low level of vibrations. A parametric study has been carried
out with emphasis on better ride index. The parametric study has brought out poss-
ible design changes required in different parameters to deliver better Sperling
comfort index. It should be noticed that the parametric study was carried out to
suggest design modifications to improve Wz index, but others dynamic and con-
trol behaviors like stability, control of secondary suspension were to be considered
when implementing the design modifications.
References
[1] Skarlatos, D., et al.: Railway fault diagnosis using a fuzzy logic method. Applied
Acoustic 65(10), 951–966 (2004)
[2] Kumar, H., Sujata: Vertical dynamic analysis of a typical indian rail road vehicle. In:
Proceedings of Computational Mechanics and Simulation, IIT, India, pp. 8–10 (2006)
[3] Dukkipati, V., Amyot, J.: Computer Aided Simulation in Railway Dynamics. Marcel
Dekker, New York (1988)
[4] Nejlaoui, M., et al.: Multiobjective robust design optimization of rail vehicle moving
in short radius curved tracks based on the safety and comfort criteria. Simulation
Modelling Practice and Theory 30, 21–34 (2013)
[5] Abood, K.H.A., Khan, R.A.: The Railway carriage simulation model to study the in-
fluence of vertical secondary suspension stiffness on ride comfort of railway carboy.
Journal of Mechanical Engineering Science 225, 1349–1359 (2011)
[6] Nielsen, J., Igeland, A.: Vertical dynamic interaction between train and track-
influence of wheel and track imperfections. Journal of Sound and Vibration 187,
25–26 (1995)
[7] Chen, H., et al.: Application of Constrained H∞ Control to Active Suspension Sys-
tems on Half-Car Models. Journal of Dynamic Systems, Measurement and Control,
Transactions of the ASME 127(3), 345–354 (2005)
[8] Pedro, J., Dahunsi, O.: Neural Network-based Feedback Linearization Control of a
Servo-Hydraulic Vehicle Suspension System. International Journal of Applied Ma-
thematics and Computer Science 21(1), 137–147 (2011)
[9] Forstberg, J.: Ride comfort and motion sickness in tilting trains (2000) - Human res-
ponses to motion environments in train and simulator experiments. Master’s Thesis,
KTH/ FKT/D-00/28-SE, Division of Railway Technology, KTH (2000)
[10] Carg, V.K.: Dynamics of Railway Vehicle Systems. Academic Press, NY (1984)
[11] B153/RP21 Application of ISO Standard to Railway Vehicles: Comfort Index Nmv
Comparison with the ISO/SNCF Comfort Note and with the Wz (European Rail Re-
search Institute) (1993)
Localization of Impact on a Beam by Time
Reversal Method
Omar Chaterbache1 and Abdelhamid Miloudi2
1
Laboratory for Advanced Mechanics (LMA), USTHB, BP32 El Alia,
Bab Ezzouar, Algiers, Algeria
ochaterbache@usthb.dz
2
Laboratory for Advanced Mechanics (LMA), USTHB, BP32 El Alia,
Bab Ezzouar, Algiers, Algeria
amiloudi@usthb.dz
Abstract. This paper deals with a numerical simulation of time reversal (TR)
process applied to a cantilever beam. To conduct this process we assume a finite
difference scheme based on Euler-Bernoulli theory of transversal vibration of the
beam. The TR process is then used to localize the position of an impact on
the beam. To complete the TR process we need one or more measurements of the
field on different positions. We had study the effect of the measurements number
and their positions on the localization with the TR process. The objective of this
study is to reduce the number of measurements and hopefully use it in an experi-
ment with few measuring instruments.
Keywords: Transverse vibrations, finite difference, impact source, time reversal,

localization.
1 Introduction
The time reversal idea is based on the time invariance of theory formulation of a
physical phenomenon (Sachs, 1987). Robert G. Sachs (Sachs, 1987) has explained
in his book ”The Physics of Time Reversal” that the time invariance concept is
purely theoretical and only applicable to a physical governing ”equations”. Also
the time invariance can’t have a physical sense only if it’s tested experimentally
(Sachs, 1987). Mathias Fink was interested on ultrasonic field time invariance
(Fink et al., 1989) and he used it to ultrasound field focusing on an object through
an inhomogeneous (Fink, 1992). First he supposes the existence of substance act-
ing like an acoustic source when it receives an ultrasonic field. So, the reflected
pressure from that is measured at different positions and then remitted in the in-
verted sense (i.e. time reversed) at the same measurement positions simultaneous-
ly. This last step should focus all the pressure on the body (Fink, 1992). Our work
was inspired from the study by G. Ribay (Ribay, 2006) who compares the time

60 O. Chaterbache and A. Miloudi
reversal to a simple correlation. When a database exists on the behavior of an im-

pact plate at several positions. A step similar to the D.O.R.T method proposed by
C. Prada (Prada et al., 1996). By determining the correlation coefficients between
the response to an impact in any position and the signals of the database G. Ribay
establishes a map of the correlation coefficients as a function of position and notes
that this ratio is close to 1 near the position of the original source. Recently D.
Francoeur and A. Berry tried to carry out the time-reversal mirror experimentally
in the case of an infinite beam (Francoeur and Berry, 2008). Using a network of
exciters piezoceramic (PZT) and piezopolymer sensors (PVDF) to simulate the
time reversal process. Then the DORT method is used to locate heterogeneities
(point masses). An acoustic source at the audible frequency was used. R. Ernst and
J. Dual uses the time reversal method to localize an acoustic source applied to a
simply supported beam (Ernst and Dual, 2014). Using a single measurement, he
located discontinuities in the section of the beam.
2 Numerical Model
Let’s consider a finite beam of length L, thickness h, a cross section A, density

ρ , a moment of inertia I and Young modulus E. The response of the cantilever
beam subjected to a transverse impact force applied at position x = x0 is the dis-
placement w(x, t). It’s a solution to the Euler-Bernoulli equation (Graff 1975) with
the corresponding boundary conditions:
⎧ ∂ 2 w( x, t )
⎪⎪M x = − E I ∂x 2
⎨ (1)
⎪ ρ A ∂ w( x, t ) = ∂ Mx( x, t ) + δ ( x − x ) δ (t )
2 2
⎪⎩ ∂t 2 ∂x 2
0
The system (1) contains only second order derivatives. It will be discretized
with fourth order finite difference in space and second order in time. The discrete-
zation detail is reported to the appendix.
The impact force is of the form shown in figure 1. As needed in our work we
construct the function (2) of the impact force to be as close as possible to a real
impact (caused by a hammer impactor for example (Fig. 2)) :
2 ⎛ −8t ⎞
⎛ 4 t ⎞ ⎜⎝ τ ⎠
⎟
δ (t ) = e 2 ⎜ ⎟ e (2)
⎝τ ⎠
τ is the impact duration.
Localization of Impact on a Beam by Time Reversal Method 61
Fig. 1 Impact form τ = 4 ms Fig. 2 Hammer impact form
The figure 3 shows the response to an impact at the middle of the beam in
terms of transverse displacement. The impact duration was τ = 0.2 ms
Fig. 3 Transversal displacement at L / 2

(steal beam, L = 1 m, h = 6 mm, A = 180 mm²)
In the following we noted that working in terms of accelerations (Fig. 4) gave

better results. Acceleration is more sensitive to the vibrations than displacement.
Fig. 4 Transversal acceleration at L / 2

(steal beam, L = 1 m, h = 6 mm, A = 180 mm²)
3 Time Reversal Principle
The time reversal process consists of three steps :

(1) perform the impulse responses hij(t) between each impact positions xi and
measurement positions xj to the same impact force and time interval T ;
(2) time reverse hij(t) to obtain hij(T-t) (noted hij(-t) in the following) ;
(3) and the last step is to send back hij(-t) from the same positions (initially
measuring positions).
The response after the third step at a position xl is given by:
wl ( xl , t ) = ∑ hik (−t ) ∗ hkl (t ) (3)

k
* represent the convolution product. Equation (3) shows that the time reversal
gives a maximum at the initial impact position.
wmax ( xl ) = max( wl ( xl , t ) ) (4)
4 Impact Localization
The trace of the responses’ maxima after retransmission versus the position allows
us to determine the initial impact position. First we define a set of measurement
positions as the “measuring network” with m positions on the beam and an “exci-
tation network” with n positions. For both defined networks the positions are
equally spaced. A localization of an impact at position x0 = 0.2 m is performed

with same impact and measuring positions for both networks (Fig. 5). We note a
maximum at the impact position and two maxima around at a distance, d, from the
λ
maximum representing the half-wavelength corresponding to the impact dura-
2
tion τ . The distance d is given by:
λ EI
d= =4 πτ (3)
2 ρA
Fig. 5 Impact localization at x0 = 0.2 m
The measuring network (Fig. 5) was formed by a measuring position every 0.01
m (i.e. m = 100) which can be considered as an ideal case for localization. But for
further experiments consideration it was in our interest to reduce the number of
measuring points. (See sections 5 and 6)
5 Influence of the Measuring Positions Number
To study the effect of measuring position number on localization we propose a

comparison of three cases (Fig. 6).
1st case: the positions are same for both networks which means m = n. In this
case a square transfer matrix is built containing all the impulse responses hij(t);
2nd case: a transfer matrix is built with only two points of measurement and n
impact positions;
3rd case: Only one measuring point is chosen here. So we get a (1 x n) transfer
matrix.
Fig. 6 Impact localization at x0 = 0.2 m with three different measuring networks
Figure 6 shows the emergence of other maxima; parasites. Given the nature of
the problem, the impact spreads in both directions giving rise to two types of
waves; progressive and evanescent. When the measurement is made on either
sides of the impact position (second case with two measurement points, for exam-
ple), the parasites are less apparent. In the case of a single measurement point the
information on the impact position is not lost but the parasites are more important.
6 Influence of the Measuring Position
Here we are interested on the third case with only one measuring point. We study
the influence of the measuring position from the initial impact position; left or
right side, and the effect of the single measuring position itself.
Fig. 7 Impact localization at x0 = 0.2 m with a single point measurement

Figure 7 shows two cases for a single measuring point; left (x = 0.1 m) and
right (x = 0.8 m) of the impact position (x0 = 0.2 m). In both cases we distinguish
the position of impact. The maxima parasites probably are related to the
interaction between the wave caused by the impact and boundary conditions. The
reflections from the boundaries act as virtual sources.
Fig. 8 Impact localization with measuring at x = 0:5m
Using a single measuring point doesn’t affect the time reversal localization but
we need to select carefully its position. In the case of transverse waves in a beam
and because of the reciprocity principle (hij = hji) the impact localization fails if the
measurement is made at x = 0.5 m (the middle of the beam). Figure 8 shows that
for this particular point localization is completely disrupted; two maxima of the
same level have emerged.
7 Conclusion
This study allowed us to analyze the time reversal process for the impact localiza-
tion on a cantilever beam. We demonstrate the effect of the number of measure-
ment points as well as the choice of their positions and allow us to capitalize on
the efficiency of time reversal. The time reversal method shows its robustness for
impact localization even with partial information about the behavior of the beam
(one or two measuring points). The ideal case is to measure the response of the
beam along its entire length which needs advanced experimental technique of
measuring. So by reducing the number of measurement points to one we can con-
sider an experiment with a single accelerometer.
References
Ernst, R., Dual, J.: Acoustic emission localization in beams based on time reversed disper-
sion. Ultrasonics 54(6), 1522–1533 (2014)
Fink, M.: Time reversal of ultrasonic fields. i. basic principles. IEEE Transactions on Ultra-
sonics, Ferroelectrics and Frequency Control (1992)
Fink, M., Prada, C., Wu, F., Cassereau, D.: Self focusing in inhomogeneous media with
time reversal acoustic mirrors. In: Proceedings of the IEEE Ultrasonics Symposium
(1989)
Francoeur, D., Berry, A.: Time reversal of flexural waves in a beam at audible frequency.
The Journal of the Acoustical Society of America 124(2), 1006–1017 (2008)
Graff, K.: Wave Motion in Elastic Solids. Dover Publications (1975)
Prada, C., Manneville, S., Spoliansky, D., Fink, M.: Decomposition of the time reversal op-
erator: Detection and selective focusing on two scatterers. The Journal of the Acoustical
Society of America 99(4), 2067–2076 (1996)
Ribay, G.: Localisation de source en milieu réverbérant par Retournement Temporel. PhD
thesis (2006)
Sachs, R.: The Physics of Time Reversal. University of Chicago Press (1987)
Appendix
Numeric Scheme
Let’s differentiate the system of equations (1) using a centered scheme with a
constant time step Δt and space step Δx . We write then in second order and using
the notation wi = w(k Δt , i Δx ) :
k
∂ 2 w( x, t ) wik +1 − 2wik + wik −1

∂t 2
=
Δt 2
+ O Δt 2 ( ) (6)
And in fourth order:

∂ 2 w( x, t ) − wik− 2 + 16wik−1 − 30wik + 16wik+1 − wik+ 2
= ( )
+ O Δx 4 (7)
∂x 2 12 Δx 2
The system (1) leads to the following iterative process :

⎧ k − wik−2 + 16wik−1 − 30wik + 16wik+1 − wik+2
⎪Mxi = − E I
⎪ 12 Δx 2
⎪ wk +1 − 2wik + wik −1
⎪ρ A i =
⎨ Δt 2 (8)
⎪ − Mx k + 16Mx k − 30Mx k + 16Mx k − Mx k
⎪ i −2 i −1 i i +1 i+2
⎪ 12 Δx 2
⎪ k
⎩+ f i
This iterative process is to assess the displacement at time k + 1 knowing the

displacement and bending moment at time k (and k-1 for displacement) and the
force applied at a point i.
Stability Criteria
The stability condition for this process is determined using the Fourier method by
k
replacing wi by ξ k e I kx i Δx . The condition ξ <1 should be checked by the
process (8) to make it do not diverge over time. This imposes a condition on the
choice of the time step according to the space step:
3 ρA 2
Δt < Δx (9)
8 EI
Boundary Conditions
The boundary conditions we imposed are Fixed - Free and it results:
Fixed: zero deflection and zero rotation.
⎧w(0, t ) = 0
⎪
⎨ ∂w (10)
⎪⎩ ∂x (0, t ) = 0
By differentiating the first space derivative with a centered second order

scheme:
⎧w0k = 0
⎪ k
⎨ w+1 − w−k1 (11)
⎪ =0
⎩ Δx
The second equation from (11), w−k1 = w+k1 , states that the fixed boundary condi-
tion is a line of symmetry between w+k1 , which is a point from the beam, and a
fictive point w−k1 . So, we can extend this property to the fictive point w−k 2 by
writing:
w−k2 = w+k2 (12)
Free boundary; zero bending moment and zero shear force.

⎧ ∂2w
⎪⎪Mx(L, t ) = 2 (L, t ) = 0
∂x
⎨ (13)
⎪Qx(L, t ) = ∂Mx (L, t ) = ∂ w (L, t ) = 0
3
⎪⎩ ∂x ∂x 3
Or;
⎧MxNk = 0
⎪ k
⎨ MxN +1 − MxNk −1 (14)
⎪ =0
⎩ Δx
For this condition we differentiate the second derivative with a second order
scheme instead of fourth order. So we get only one fictive point from the second
equation of (14)
wNk +1 = 2wNk − wNk −1 (15)
The process (8) is used for all points i = 0..N to evaluate the momentum Mx
and the deflection w. For the boundaries we use the second order instead of the
fourth to evaluate the deflection at instant k+1. The bending momentum is always
calculated at the instant k.
Early Detection of Gear Failure by Vibration
Analysis
Mustapha Merzoug1, Khalid Ait-Sghir2, Abdelhamid Miloudi1, Jean Paul Dron2,

and Fabrice Bolaers2
1
Advanced Mechanics Laboratory, USTHB, Algiers, Algeria
moustapha_merzoug@yahoo.fr, amiloudi@usthb.dz
2
Advanced Mechanics Laboratory, Champagne Ardène University, Reims, France
{khalid.ait-sghir,fabrice.bolaers,jp.dron}@univ-reims.fr
Abstract. Gear mechanisms are an important element in a variety of mechanical

systems, such as industrial machinery and automotive. Health monitoring of rotat-
ing machines is important to avoid failure of the system in advance. Principally,
this paper consists of two parts: in the first part, a gear dynamic model including
localized tooth defect has been developed. The model consists of a spur gear pair,
two inertias. The model incorporates the effects of time-varying mesh stiffness
and damping, excitation due to gear errors. The results of a dynamic modeling of
the gears transmission are calculated by using the Newmark integration scheme.
The second part consists of signal processing of simulated and experimental sig-
nals using the wavelet transform. It is shown that the kurtosis of the vibration
signal is a sensitive indicator of the existence of damage in the gear pair.
Keywords: Gears, wavelet analysis, spalling, Kurtosis.
1 Introduction
Gears are commonly used in power transmission achines for many applications,
such as industrial machinery and automotive. Health monitoring of rotating ma-
chines is important to avoid failure of the system in advance. A detailed examina-
tion of models published in 1988 by Ozguven (Ozguven 1988) allowed classified
these models into several groups.
• Simple dynamic factor models. These models include most of the early studies
in which dynamic factors that can be used in gear root stress formula are de-
termined,
• Models with tooth compliance. These models include the tooth stiffness as the
only potential energy storing element in the system,
• Global models of gear transmissions.
These models are currently heavily used, and can include shaft torsional vibra-
tion, shaft bending stiffness, gear tooth bending stiffness, bearings stiffness. In the
defect presence, the mesh stiffness variations give important information about the

70 M. Merzoug et al.
transmission. It also reflects the damage severity and can be obtained by calcula-
ting the variation of the tooth mesh stiffness so that preventive action can be envi-
saged. Weber [2], Cornell [3] computed the gear mesh stiffness analytically, and
Kasuba [4] calculated it with a digitization approach. Yang [5] used the potential
energy principle to calculate the total mesh stiffness of meshing gear pairs as a
function of the rotation angle of the gear. Chaari [6] developed an analytical mo-
del to calculate the gear mesh stiffness based on the Weber's tooth bending equa-
tion. In addition, some researchers attempted to extend the gear mesh stiffness
model to involve the presence of the some gear faults, such as tooth breakage and
spalling, gear tooth crack. A comprehensive review of the methods of signal pro-
cessing published in 2006 can be found in [7]. Time-domain analysis is based on
the time waveform itself. Traditional time-domain analysis calculates characteris-
tic features from time waveform signals as descriptive statistics such as mean,
peak, peak-to-peak interval, standard deviation, crest factor, high-order statistics:
root mean square, skewness, kurtosis, etc. These features are usually called time-
domain features, which have proved to be effective in fault diagnosis and are now
well established [8] and [9]. The advantage of frequency domain analysis over
time domain analysis is its ability to easily identify and isolate certain frequency
components of interest. The most widely used conventional analysis is the spec-
trum analysis by means of fast Fourier transform (FFT). One limitation of fre-
quency-domain analysis is its inability to handle non-stationary waveform signals,
which are very common when machinery faults occur. Thus, time-frequency ana-
lysis, which investigates waveform signals in both time and frequency domain,
has been developed for non-stationary waveform signals. Traditional time-
frequency analysis uses time-frequency distributions, which represent the energy
or power of waveform signals in two-dimensional functions of both time and fre-
quency to better reveal fault patterns for more accurate diagnosis. Short Time Fou-
rier Transform (STFT) or spectrogram (the power of STFT) [10] and Wigner-Ville
distribution) [11,12] are the most popular time–frequency distributions. The time
frequency representation of non stationary signals was represented using the new
type of wavelet transform, in which discrete wavelet transform was used to get de-
tails and approximations information about the signal to identity the fault [13,14].
The discretes wavelets transform was also used for energy evaluation of known
bandwidth with time scale analysis and acoustic emission signals in detection of
faults in gearbox [15,16]. The gear fault identification based on Hilbert-Hung
transform was proposed to overcome the problem of reconstructing a feature ma-
trix of singular value decomposition [7]. Djebala [17] proposed an approach that
benefits the local nature of wavelet analysis and its ability to detect transients.
They first calculated the Hilbert transform of the signal then calculate an envelope
of the signal will be processed by wavelet analysis. It then extracts the detail cove-
ring the rotation frequency of the defective wheel. Parey [18] using Empirical
Mode Decomposition (EMD) used as a processing tool for non-stationary and
nonlinear signals. In their work, they expose its operation from a real signal. They
exploit this method to detect an early stage the presence of a fault using global vi-
bratory indicators that are based on statistical analysis of a time signal such that
Early Detection of Gear Failure by Vibration Analysis 71
the kurtosis and the crest factor. Indicators are sometimes limited and may not
give a reliable verdict on the state of the gear. Random noise and other machine
components pollute the signal and make it difficult or impossible to detect in some
cases. One solution is to measure signals in narrow bands at high frequencies with
a band pass filter [19]. A second approach is to use modern tools noising vibratory
signals developed from the wavelet transform [20,21], spectral subtraction [22] or
adaptive filtering [23]. A comparison of the three methods was applied to the si-
gnals of defective bearings [24]. Indeed, the use of these methods allows to denoi-
sed and filtered signals whose kurtosis is more significant. This will help detect
any defects that the detection was not possible in the raw signals or broadband.
2 Numerical Simulation
The analytical model used in the present paper is available in literature and the
corresponding graphical representation is shown in fig. 1.
Fig. 1 Model for gear dynamic behaviour
The equation of motion is rearranged to the form shown in the Eq. 1. The dy-
namic response is simulated by numerically solving the second order term (accele-
rations) for each step of incremented time. The effect of the mesh stiffness varia-
tion is implemented in the model by updating its value for each time increment.
∆
1 cos (1)
Where variable is the analysed degree of freedom, named dynamic transmis-

sion error, defined as:
(2)
Where and are the base radius, and the torques, and the
rotations, respectively, of input and output gears. Further, is an equivalent
mass that will be described in Eq. (4); the stiffness term is mathematically treated
by means of a Fourier development, thanks to a constant factor .
is the contact force between teeth and the last term is the second derivative
of static transmission error, treated as an external input acting on the system. The
force is calculated as the rapport between the torque and the corresponding base
radius :
(3)
The second parameter to be assessed is the equivalent mass , it is a function

of the two gears moments of inertia, and , and of the corresponding base ra-
dius, and :
·
(4)
2.1 Effect of Mesh Stiffness Variation on the Gear Pair System

Lin and Parker [25] assume that stiffness varies along the shaft rotation, which
follows the equation :
2 sin cos (5)
and
2
2 (6)
2 (7)
2
where,
: teeth stiffness (mean), 2 peak-to-peak teeth stiffness, : contact ratio
and : phase angle
Gear tooth crack is a defect that has severe impact on healthy machine functio-
ning. Detecting this defect as soon as possible has an extreme importance. In tooth
root crack modeling, most of the studies consider some modifications in meshing
stiffness profile [26]. For instance, faulty tooth stiffness decreases and it is repea-
ted once per contact cycle of this tooth. Therefore, the authors proposed to model
a localized crack on a tooth taking into account its angular position on the base
circle. On this position, meshing stiffness decrease is considered and this pheno-
menon is repeated at each time the cracked tooth enters in contact with the teeth of
the opposed gear. In this case, the condition that governs the stiffness expression
in the presence of a localized fault on a tooth is as follows:
• If the contact is between two healthy teeth, the meshing stiffness is given by
.
• If the contact is between the cracked tooth and the tooth of the opposed gear,
the stiffness is given by 1 .
where is a periodic function representing the amount of stiffness degradation
expressed as:
sin (8)
1
· cos Ω
where Ω 2 , 1,2, and represents the rotation frequency of the gear af-
fected by the crack.
is the amplitude of the function , it represents the ratio stiffness degrada-
tion or the crack evolution.
represents the cyclic ratio.
Figure 2 shows the kurtosis variation according to the default severity taken for
different values of in the expression of the mesh stiffness.
Fig. 2 Kurtosis evolution vs. the degradation ratio
The results obtained show that it is difficult to detect the defect at birth by the
kurtosis (for a previously fixed level). An analysis of this indicator shows that the
fault only appears when the degradation ratio reaches a value of 3,18 %. It is the-
refore necessary to decompose the signal into several approximations using the
wavelet transform.
2.2 Kurtosis Evolution for Different Approximation

In several studies the kurtosis was more sensitive than other scalar indicators,
which the privileged place as an indicator when it comes to shocks such defects,
including bearings and gears [19]. Several simulations have shown that the most
appropriate indicator in this case is the kurtosis since it is the indicator who can
detect early as the default for the first approximations.
Several simulations have shown that the most appropriate indicator in this case
is the kurtosis since he can detect an early stage failure for the first approxima-
tions. Indeed Fig. 3; highlights this conclusion. After two approximations, we can
detect the fault when the degradation ratio reaches a value of 1,36 % if we set a
threshold of 3 for kurtosis. Effectively Fig. 3, demonstrates this conclusion.
Fig. 3 Kurtosis evolution for different approximations
Fig. 4 Kurtosis comparison between original signal and the first two approximations
A comparison between the values of Kurtosis before and after the signal de-
composition (Fig. 4) shows the contribution of wavelet transforms in improving
the sensitivity of these indicators with respect to the conventional case.
3 Experimental Validation
The vibratory signals on which we made the calculation come from a gearbox (test
bench) installed at the CETIM (France). The dimensioning of the gear wheels of
the tempered hardened, and the operating conditions (speed, torque) were set so as
to obtain a peeling over the entire length of a tooth. [27,28]. Every day after the
acquisition of vibration signals, the bench is stopped, to appraise the state of the
teeth of wheels. Testing of reducing spread over 12 days with registration of a vi-
bratory signal daily. The gear consists of a pinion 20 and teeth of a tooth wheel
21. The speed of the drive motor is 1000 or approximately 16.67 , the
rotational speed of the wheel is 15.87 and the meshing frequency is about
330 and the samplinng frequency is 20 . Each record contains $600000$

points and has 3 secondss of duration. During the experiment, the gear changees
from working correctly inn a deteriorated condition.
In 2006 Parey and [18 8], used the method of Empirical Mode Decompositioon
and showed that it is posssible to detect the defect the 10th day (Fig. 5) in place oof
the 11th day as the expertt report CETIM mentionned.
Fig. 5 Kurtosis evolution off the of Fig. 6 Evolution of cepstral peaks A1 andd
experimental signal [18] A2, their sum and the indicator vs. the dayys
acquisition. [29]
El-Badaoui [29] found d that the peak of the cepstral peak of the wheel whicch
develops the defect (A1) increases
i at the expense of the other. The sum of the firrst
peak close to 0,5, tells theem that the processed signals are not very noisy. The evoo-
lution of the indicator clearly shows that from the 8th day, the gear develops a dee-
fect. This result is consisstent with the observation made on the expertise of thhe
gearbox, which indicates that the 8th day of occurrence of chipping on the toop
tooth 15/16 (Fig. 6).
Fig. 7 Peaks evolution during the days (Prony Technic) [30].
In their paper, Haloui and

a al. [30] proposed a technique based on Prony modeel.
They showed that vibratio on analysis through technical Prony can detect this defeect
on the 6th day deteriorateed Wheel (fig. 7), while the synchronous Cepstre deteccts
the fault on the 8th day as specified above. To demonstrate the performance of this
technique in the diagnosis of gear systems compared with those of the method of
synchronous cepstrum and the expert report was completed.
3.1 Results
A preliminary analysis of the temporal signals allowed us to see the occurrence of
a fault very marked the 12th day, it is not possible to early detect the occurrence of
a fault with the times signal (Fig. 8).
Fig. 8 Kurtosis comparison of original signals and approximations
The decomposition of the experimental signal approximations and evolution of

kurtosis show the fault on the 4th day after the approximation 3 (Fig. 5). To de-
monstrate the effectiveness of our approach using signal decomposition of the ap-
proximations, we tested experimental noised signals. The decomposition of the
experimental noised signal into several approximations and the kurtosis evolution
show the appearance of the defect on the 4th day at the approximation 4 (Fig. 6).
3.2 Suggested Indicator

The indicator we used is determined from the spectral representation. It is an indi-
cator inspired from the work of Breneur [31] which used in the case of the
combined deficiencies of bearings and gears. The author calculated the sum of
amplitudes corresponding to the first two harmonic of the rotation frequency of
the shaft carring the affected gear in the case of wheel default gear only, then he
calculated the difference from the case without defect. In our work, before calcula-
ting the sum of the amplitudes we have decomposed signals by wavelets transform
and we have calculated the spectrum of the envelopes obtained by the Hilbert
transform.
Fig. 9 Indicators before and after decomposition by wavelets transform
4 Conclusion
The approach just described can trace the origin of some defects after decomposi-
tion of the original signal details and approximations because a preliminary analy-
sis of temporal signals allowed us to visualize the occurrence of a defect very
marked in the 12th day. Therefore it’s not possible to early detect how the defect at
the time signals. Also, we hope that all the results on the monitoring of gears, seals
those obtained with respect to other machines, the engineer will make a better
diagnosis and that in order to improve the transmission quality.
References
1. Ozguven, H.N., Houser, D.R.: Mathematical models used in gear dynamics a review.
Journal of Sound and Vibrations 121, 383–411 (1988)
2. Weber, C.: The deformation of loaded gears and the effect on their load carrying ca-
pacity. Report Number 3, Spons. Res. (Germany). British Dept. of Sci. and Ind. Res.
(1949)
3. Cornell, R.W.: Compliance and stress sensitivity of spur gear teeth. Journal of Me-
chanical Design, Transaction of the ASME 103, 447–459 (1981)
4. Kasuba, R., Evans, J.W.: An extended model for determining dynamic loads in spur
gearing. Journal of Mechanical Design, Transaction of the ASME 103, 398–409
(1981)
5. Yang, D.C.H., Lin, J.Y.: Hertzian damping, tooth friction and bending elasticity in
gear impact dynamics. Journal of Mechanical Design, Transaction of the
ASME 109(2), 189–196 (1987)
6. Chaari, F., Fakhfakh, T., Haddar, M.: Analytical modelling of spur gear tooth crack
and influence on gearmesh stiffness. Europ. Jou. of Mech. A/Solids 28, 461–468
(2009)
7. Jardine, K.S., Lin, D., Banjevic, D.: A review on machinery diagnostics and prognos-
tics implementing condition-based maintenance. Journal of Mechanical Systems and
Signal Processing 28, 1483–1510 (2006)
8. McFadden, P.D., Smith, J.D.: A signal processing technique for detecting local defects
in a gear from the signal average of the vibration. Proceedings of the Institute of Me-
chanical Engineers 28, 287–292 (1985)
9. Randall, R.B.: A new method of modelling gear faults. Journal of Mechanical De-
sign 104, 259–267 (1982)
10. Wang, W.J., McFadden, P.D.: Early detection of gear failure by vibration analysis I.
Calculation of the time–frequency distribution. J. Mec. Design 7, 193–203 (1993)
11. Meng, Q., Qu, L.: Early detection of gear failure by vibration analysis I. Calculation of
the time–frequency distribution. J. of Mec. Sys. and Sig. Proc. 5, 155–166 (1991)
12. Pan, M.C., Van Brussel, H., Sas, P., Verbeure, B.: Fault diagnosis of joint backlash.
Journal of Vibration and Acoustics. Transactions of the ASME 120, 13–24 (1998)
13. Peng, Z.K., Chu, F.L.: Application of the wavelet transform In machine condition
monitoring and fault diagnostics: a review with bibliography. Journal of Mechanical
Systems and Signal Processing 18, 199–221 (2004)
14. Douglas, H., Pillay, P., Ziarani, A.K.: A new algorithm for transient motor current sig-
nature analysis using wavelets. IEEE Trans. on Ind. App. 40, 1361–1368 (2004)
15. Kia, S.H., Henao, H., Capolino, G.: A Diagnosis of broken-bar fault in induction ma-
chines using discrete wavelet transform without slip estimation. IEEE Trans. on Indus-
trial Applications 45, 1395–1403 (2009)
16. Gu, D., Kim, J., An, Y., Choi, B.: Detection of faults in gearboxes using acoustic emis-
sion signal. Journal of Mechanical Science and Technology 45, 1279–1286 (2009)
17. Djebala, A., Ouleaa, A., Kebbabsa, T., Ramdane, Y.: Détection des défauts non sta-
tionnaires de dentures d’engrenages par application de l’analyse en ondelettes et la
transformée de Hilbert. In: Congrés Algérien de Mécanique, Mascara, Algeria, No-
vember 27-29 (2013)
18. Parey, A., El Badaoui, M., Guillet, F., Tandon, N.: Dynamic modelling of spur gear
pair and application of empirical mode decomposition-based statistical analysis for
early detection of localized tooth defect. Journal of Sound and Vibrations 294,
547–561 (2006)
19. Pachaud, C., Salvetat, R., Fray, C.: Crest factor and kurtosis contributions to identify
defects inducing periodical impulsive forces. Journal of Mechanical Systems and Sig-
nal Processing 11, 903–916 (1997)
20. Donoho, D.L.: De-noising by soft-thresholding. IEEE Transactions on Information
Theory 41, 613–627 (1995)
21. Ling, J., Qu, L.: Feature extraction based on morlet wavelet and its application for me-
chanical fault diagnosis. Journal of Sound and Vibrations 234, 613–627 (2000)
22. Dron, J.P., Bolaers, F., Rasolofondraibe, L.: Improvement of the sensibility of scalars
indicators thanks to de-noising method by spectacle subtraction. Application to the de-
tection ball bearing defects. Journal of Sound and Vibrations 270, 61–73 (2004)
23. Lin, J., Zuo, M.J.: Gearbox fault diagnosis using adaptive wavelet filter. Journal of
Mechanical Systems and Signal Processing 17, 1259–1269 (2003)
24. Dron, J.P., Rasolofondraibe, L., Chiementin, X., Bolaers, F.: A comparative experi-
mental study on the use of three denoising methods for bearing defect detection. Jour-
nal Meccanica 45, 265–277 (2010)
25. Lin, J., Parker, R.G.: Mesh stiffness variation instabilities in two-stage gear system.
Journal of Vibration and Acoustics 124, 68–76 (2002)
26. Abbes, M.S., Trigui, M., Chaari, F., Fakhfakh, T., Haddar, M.: Dynamic behaviour
modelling of a flexible gear system by the elastic foundation theory in presence of de-
fects. Mechanics A/Solids 29, 887–896 (2010)
27. Drouiche, K., Menad, S., Grenier, Y.: Détection de défauts d’engrenages par analyse
vibratoire. Journal du Traitement du Signal 8, 333–343 (1991)
28. Laspi: Documents du Laboratoire d’Analyse des Signaux et des Processus Industriels,
Université Jean Monnet, St Etienne, France, http://www.laspi.fr
29. El-Badaoui, M.: Contribution au diagnostic vibratoire des réducteur complexes à en-
grenages par l’analyse cepstrale. Thèse de doc., Université Jean Monnet St Etienne
(1999)
30. Haloui, N., Chikouche, D., Benidir, M.: Diagnosis of gear systems by spectral analysis
of vibration signals. Int. J. of Comp. Sci. and Network Security 7, 285–293 (2007)
31. Breneur, C.: Eléments de maintenance préventive de machines tournantes dans le cas
de défauts combinés d’engrenages et de roulements. Thèse de doc. INSA de Lyon
(2002)
Effect of Water Hammer on Relatively Long
Inclined Pipelines
Manel Dalleli1, Ezzeddine Hadj-Taïeb1, Christian Schmitt2, and Zitouni Azari2
1
Laboratoire de Mécanique des Fluides Appliqués, Génie des Procédés et Environnement,
Ecole Nationale d’Ingénieurs de Sfax, BP 1173, 3038, Sfax, Tunisie
manel.dalleli@gmail.com, ezed.hadj@enis.rnu.tn
2
Laboratoire de Biomécanique, Polymères et Structures (LaBPS),
Ecole Nationale d’Ingénieurs de Metz, 57070 Metz, France
{Schmitt,azari}@enim.fr
Abstract. Consideration is given in this paper to the numerical solution of tran-

sient flows in relatively long pipeline provoked by the water hammer phenome-
non. Equations which describe these one-dimensional flows in a cylindrical pipe,
of linear elastic behavior according to the Hooke’s law, are solved in the time do-
main by the method of characteristics using linear integration. The obtained results
due to a rapid closure valve at the downstream end of a long pipeline show that the
gravity lift may have an important effect on the maximum pressures, which may
become very important near the valve and provoke the failure of the pipe especial-
ly in presence of defect. The safety factor, computed at equidistant sections of the
pipe, determines the distance between the supplying reservoir and the defect from
which the failure may happen.
Keywords: Water hammer, Long pipeline, Method of characteristics, pressure,

safety factor.
1 Introduction
The analysis of transient flows in relatively long pipelines is rather complex be-
cause the pipeline friction head losses are large compared to the instantaneous
head rise caused by a sudden variation of the flow velocity that is by the water
hammer phenomenon.
In this paper, transient flows are accurately analysed to examine if, under unfa-
vourable circumstances damage due to water hammer may occur in relatively long
pipelines.
A computing procedure based on the method of characteristics is developed.
Transient flow is provoked by an instantaneous valve closure at the downstream
end of a relatively long pipeline supplied by a reservoir at a constant level. In this
case, the gravity lift may have an important effect on the maximum pressures,

82 M. Dalleli et al.
computed in order to select the pipe-wall thickness necessary to withstand the

pipeline failure. Indeed, the strength of the pressure wave may become very im-
portant especially somewhere in the pipe and provoke, there, the failure of the
pipe. To analyse the security of the pipe, the safety factor, which should be greater
than two, is computed to define the distance between the reservoir and the defect
from which the failure may happen
2 Mathematical Model
The equations which describe transient one-dimensional flow in a cylindrical pipe,
of linear elastic behaviour according to the Hooke law, can be adapted from the
analytical model developed by (Wylie et al. 1993). Applying the law of conserva-
tion of mass and momentum yields to the following equations of continuity and
motion:
2
∂H Q ⎛ ∂H ⎞ C ∂Q
+ ⎜ − sin γ ⎟ + =0 (1)
∂t S ⎝ ∂x ⎠ gS ∂x
∂Q Q ∂Q ⎛ ∂H ⎞
+ + gS ⎜ +J⎟=0 (2)
∂t S ∂x ⎝ ∂x ⎠
where Q is the fluid discharge, H is the head, C is the pressure waves celerity,
( )
J = (λ D)Q Q 2gS 2 : is the linear head loss by unit of length of the pipe, λ is
the coefficient of friction, S is the cross section area of the pipe, γ is the pipe
slope, D is the circle pipe diameter inside, t is the time and x is the distance along
the pipe.
3 Resolution by the Method of Characteristics
The method of characteristics (Fox, 1977) is applied to transform the system of

partial differential equations (1) and (2) into a system of ordinary differential
equations that can be integrated numerically without difficulty. One gets:
gS g ⎛Q ⎞
dQ ± dH ∓ Q sin γ + gSJdt = 0 and dx = ⎜ ± C ⎟dt (3)
C C ⎝S ⎠
the + is for the waves coming from the upstream while the – is for the waves com-
ing from the downstream.
Equations (3) determine the evolution of the head pressure and the discharge
according to the time and the space. They are much appropriated to be solved nu-
merically on a microcomputer. The obtained solution constitutes a solution to the
original system of the equations (1) and (2).
Effect of Water Hammer in Relatively Long Inclined Pipelines 83
The transient flow is generated by a discontinuity introduced in the flow by the

instantaneous valve closure. This discontinuity, known as the water hammer phe-
nomenon, propagates along the pipe and its displacement is presented in the plan
(x, t) by the characteristic lines. In the case of the quasi-rigid pipes, as metallic or
concrete conducts, the pressure waves propagation speed C is constant.
Δx
P
t+Δt
C+ C-
Δt
t
i-1 R i T i+1
Fig. 1 Characteristic lines: Regular grid
To determine H ( x , t ) and Q ( x ,t ) , consider a pipe of length L subdivided in

N equal reaches. If we start with known steady-state conditions at t = 0, then we
know Q and H at the N+1 sections of the pipe. If we specify the time interval Δt ,
such as Δ t (C + Q S ) ≤ Δ x , the characteristic lines from the sections R and T
intersect at P, which is a section i of the pipe (figure 1). In these conditions, the in-
tegration of the equations (3) permits to write (Wylie and Streeter,1978):
HPi = CP − BQPi (4)
HPi = CM + BQPi (5)
where B = C (gS ) , S = πD 2 4 , r = λ Δ t (2 DS ) , G = gΔt C ,

QR = Qi − ξ (Qi − Qi−1 ) , QT = Qi −ξ (Qi − Qi +1) , HR = Hi − ξ (Hi − Hi−1 ) ,
HT = Hi −ξ(Hi − Hi+1) , ξ = Δ tC Δ x , C P = H R + BQR (1 + G sin γ − r QR )
(
and CM = HT − BQT 1 − G sin γ − r QT . )
The calculation of the discharge at any section i is obtained if we solve equa-
tions (4) and (5) for i = 2,3, , N :
HPi = (CP + CM ) 2 and QPi = (CP −CM ) 2B (6)

The pressure at each node i is then calculated by the formula:
pPi = ρg(HPi − zi ) (7)
where ρ is the fluid density and zi is the elevation at the node i.

After the first time step, the extreme points of the system begin to influence the
calculation of the interior points. Thus, in order to get the solution until an ele-
vated time level, it is necessary to introduce the appropriate boundary conditions.
At either end of a single pipe only one of the characteristic equations (4) or (5) is
available in the two unknowns Q P and H P . An auxiliary equation is therefore
necessary in every case to specify Q P or H P .
4 Numerical Results
An example is presented illustrating the accurate solution of the water hammer phe-
nomenon in a relatively long pipe. A computer program was developed using the
FORTRAN language. The transient flow is produced in a long cast iron pipe which
data are taken from the textbook of (Wylie and Streeter, 1978) (see figure 2):
L = 201 km, D = 0.762 m, e = 10.2 mm, λ = 0.018, H0 = 426.72m ,
sinγ = 0.002 , V0 =1.2954m s , Q 0 = 0 . 59075 m 3 s and C = 1005 m s .
Resevoir Final hydraulic grade line
Initial hydraulic grade line
H0
L
Pipeline
Valve
Fig. 2 Single pipeline
Initial conditions
The initial conditions at t=0 s are given by the initial steady-state flow. They are
defined, for i = 1,2, , N + 1 , by: Q i = Q 0 = 0 .59075 m 3 s and
H i = H 0 − λ (i − 1)Δ xQ 02 (2 gDS 2 )
Boundary conditions
The transient flow depends on the conditions imposed at the ends of the pipe. In
this study, the pipe is supplied, at the upstream end, by a reservoir at fixed level,
so HP1 = Hresevoir(t ) = 426,72m , ∀t ≥ 0 . At the downstream end, the pipe is
equipped with a valve closed suddenly, so QPN+1 = Qvalve(t) = 0 , ∀t > 0 .
Results
By using the method of characteristics (MOC) for the time step (increment)
( )
Δt = L / N * C (N = 4000), one can determine the changes in pressure at dif-
ferent sections of the pipe and then the maximum stresses can be readily calcu-
lated via thin-walled hollow cylinder assumptions. In figure 3, the pressures are
plotted at some equidistant sections of the pipe. The maximum pressures and
stresses at these sections are presented in table 1.
50
p (Bar)
x =0
40 x = L/8
x = L/4
30 x = 3L/8
x = L/2
x = 5L/8
20
x = 3L/4
x = 7L/8
10 x =L
0
0 2 4 6 8 10 t*C/L 12
Fig. 3 Pressure evolution at some equidistant sections of the pipe
Table 1 Maximum pressures and stress along the pipe
x 0 L/8 L/4 3L/8 L/2 5L/8 3L/4 7L/8 L

Pmax 2.381 8.874 14.184 19.749 25.069 30.32 35.562 40.680 45.676
(bar)
σmax 8.9 33.1 52.9 73.8 93.6 113.2 132.8 151.9 170.6
(MPa)
5 Security Analysis
The risk of failure of the long pipeline in presence of a notch-type defect, under
the effect of water hammer, can now be enjoyed by the value of the safety factor.
This failure may occur in case of the presence of a critical defect, of size a/e = 0.5,
subjected to critical stress corresponding to the breaking strength of the pipe mate-
rial or tenacity (Pluvinage , 2003) .
The break condition can be written:
f (K r , L r ) = 0 (8)
Where K r is the no dimensional stress intensity factor, being defined as the

applied stress intensity factor K app divided by the fracture toughness KIc:
K app
Kr = (9)
K Ic
L r is the dimensionless applied stress defined as the quotient between the
maximum applied stress σ max and the flow stress Rc :
σ max
Lr = (10)
Rc
Any loading condition in the structure containing a defect, leading to a non-
critical situation, is described by the calculated point in plan ( L r , K r ).
The coordinates of this point are ( Lr * , K r * ). The failure begins or occurs
when the coordinates of this point increases until it reaches the critical values
( Lrc , Krc).
All critical operating points leading to failure are located on a curve given out
by the code SINTAP level 1 (SINTAP, 1999) and defined by the relationship:
− 0 .5
⎛
⎜
L2
f (l rc ) = ⎜ 1 + rc
2
⎞
⎟
⎟
[0.3 + 0.7 exp (− 0.6 L )]
6
rc
0 ≤ Lrc ≤ Lr max (11)
⎝ ⎠
2.5
⎡150 ⎤
where Lr max = 1 + ⎢ ⎥ .
⎣σY ⎦
As illustrated in Figure 4, the Failure Assessment Diagram sets the safety factor
defined by the report OC on OA for an operating point A. If this point is inside the
boundary lines of the diagram which is limited by the interpolation curve, the
structure is safe. If not, the assessment point is situated outside of the interpolation
curve and failure may occur (Rouabeh et al. 2012).
Fig. 4 Typical failure assessment diagram (FAD)
In this study, the pipes are considered as ductile cast iron having the mechani-
cal properties shown in Table 2.
Table 2 Mecanical properties of cast iron material
Yield stress Ultimo stress Elongation % Density fracture toughness

σY (MPa) σu (MPa) (Kg/m3)
KIc (MPa√m)
300 420 10 7050 14.90
Breaking pipes is initiated from a semi elliptical defect, for instance a corrosion
crater. The facture toughness is determined by a local failure criterion, and not by
a conventional criterion of fracture mechanics which assumes that the defect is a
crack (Pluvinage, 2001). This criterion assumes that the fracture process requires a
certain level of development of the failure process, Figure 5. This volume is as-
sumed cylindrical of diameter equal to the effective distance X eff . Inside this vo-
lume weighted mean stress prevails, equal to the effective opening stress.
The elastic–plastic stress and gradient constraints are shown in this diagram.
Three specific areas can be distinguished (Bouaziz et al. 2014):
Zone I: The elastic–plastic stress notch root increases to the maximum stress
σ Max. Note that the maximum stress is not at the bottom notch but at a dis-
tance X Max.
Zone II: The elastic–plastic stress decreases until the effective distance
X eff which corresponds to the minimum of stress gradient.
Zone III: The evolution of the elastic–plastic strain has a linear behavior in the
log–log graph.
This stress distribution, before failure, is described by the equation:
Kρ
σ yy = (12)
(2πr )α
where σ yy is the opening stress, K ρ the notch stress intensity factor, r the dis-
tance and α the exponent of the power stress distribution.
The effective distance is defined as the boundary between zones II and III
(figure 5).
The toughness is then defined as equal to the critical notch stress intensity
c
factor Kρ :
( )
Kρc = 2πX eff α σ eff (13)
For brittle materials such as cast iron and a failure to low notch radius, tough-
C
ness Kρ is equated to toughness KIc.
Log(σ yy (r))
σ eff Stress distribution along defect root
I II III
Fracture Process Zone (FPZ)
X eff Log(r)
Geometrical defect
Fig. 5 Typical stress distribution around Fig. 6 Mesh density around the semi-
notch tip in bi-logarithmic diagram elliptical defect
Calculations of stress distributions at the head of a corrosion crater, assumed to

be semi elliptical of depth equal to half the thickness, have been performed by the
finite element method using ABAQUS software (figure 6) (Zienkiewicz C,1988) .
Table 3 k r , L r and safety factor for semi-elliptical defects via SINTAP
Position Pressure Lr kr Safety

(MPa) Factor
0 2.381 0.0247 0.1291 7.486
L/8 8.874 0.0921 0.4834 1.999
L/4 14.184 0.1472 0.7702 1.255
3L/8 19.749 0.2049 1.0601 0.911
L/2 25.069 0.2601 1.3574 0.712
5L/8 30.32 0.3146 1.6226 0.595
3L/4 35.562 0.369 1.9124 0.505
7L/8 40.680 0.422 2.2815 0.424
L 45.676 0.474 2.5139 0.385
The effective stress and the effective distance are determined by the volumetric
method. Then using equations (9) and (10), the coordinates of evaluation points
( k r and L r ) are determined, see Table 3
Figure 7 presents the Failure Assessment Diagram using the SINTAP code for
a long pipeline. All the assessment points are given on the diagram for different
positions of the defect along the pipe for defects place on different positions along
the pipe.
Fig. 7 Failure assesment diagram
Examining Figure 7, it appears that the points placed after the point X = L / 4
are in a critical situation (located in the failure zone) while the other points lie in
the Zone of security. In fact, the greatest value of safety factor obtained is equal to
7.49 which is the point of coordinate X = 0 (the lowest pressure).
6 Conclusion
In this study, pressure wave propagation in a long pipeline system was investi-
gated numerically by the method of the characteristics. These waves were caused
by the sudden closing of a valve at the downstream end of an inclined long pipe. A
mathematical model, represented by a system of two partial differential equations,
taking into account friction and gravity forces, was developed. The mathematical
equations were solved in the time domain by the method of characteristics using
linear integration.
The results indicate that the gravity action has an important effect in the predic-
tion of the maximum pressures, which are very important near the valve and may
provoke the failure of the pipe especially in presence of defect. Maximum pres-
sures are calculated to see if the chosen pipe-wall thickness is reasonable to resist
these pressures and thus to avoid the pipeline failure. In this study the safety factor
has been computed at equidistant sections of the pipe. It determines the distance
between the reservoir and the defect from which the failure may happen. A simu-
lation of a grey cast iron pipe with a surface defect was made to analyze the ef-
fects of the change in pressure. An example of application in respect of a long pipe
was proposed. The criticality of a semi-elliptical defect in the cast iron pipes and
the depth half the thickness was studied using the finite element method pro-
grammed on the ABAQUS software. The results drawn in the diagram DIR, based
on the code SINTAP level 1, show that all points within the three quarters of the
downstream conduit are located outside the field of integrity. It can be concluded
that the studied system does not work in the field of security and we must
therefore install protection elements to control the phenomenon of water hammer.

To validate these results, Intensity Factor Constraints calculated numerically have
been successfully compared with that obtained by the volumetric approach.
Results, obtained by the two methods agree perfectly.
Acknowledgements. This work is partially supported by PHC-UTIQUE program managed

by the CMCU.
I would like to thank my committee members, for serving as my committee members. I
also want to thank you for letting my defense be an enjoyable moment, and for your bril-
liant comments and suggestions, thanks to you.
References
Journal article only by DOI

Bouaziz, M., Guidara, M., Schmitt, C., Hadj-Taïeb, E., Azari, Z.: Water hammer effects on
a gray cast iron water network after adding pumps. Engineering Failure Analysis (2014),
doi:10-1016/j.engfailanal.2014.04.023
Rouabeh, K., Schmitt, C., Elaoud, S., Hadj-Taïeb, E., Pluvinage, G.: Failure of grey cast
iron water pipe due to resonance phenomenon. Engineering Failure Analysis (2012),
doi:10.1016/j.engfailanal.2012.06.009
Book
Fox, J.A.: Hydraulic Analysis of Unsteady Flow in Pipe Networks. The Mac Millan Press
LTD, London (1977)
Pluvinage, G.: Fracture and Fatigue emanating from stress concentrators. Kluwer Academic
Publishers, Dordrecht (2003)
Wylie, E.B., Streeter, V.L.: Fluid Transients. McGraw-Hill Company, New York (1978)
Wylie, E.B., Streeter, V.L., Suo, L.: Fluid Transients in Systems. Prentice Hall, Englewood
Cliffs (1993)
Zienkiewicz, C., Taylor, R.L.: The finite element method, 4th edn. McGraw-Hill Book
Company, UK (1988)
Dissertation
SINTAP: Structural integrity assessment procedure. Final report E-U project. Brite Euram
Programme secrets; BE95-1462, Bruxelle (1999)
Experimental Study of the 60° PBT6 Pitching
Blade Effect with a PIV Application
Bilel Ben Amira, Zied Driss, and Mohamed Salah Abid
Laboratory of Electromechanical Systems (LASEM), National School of Engineers of Sfax

(ENIS), University of Sfax, B.P. 1173, Road Soukra, 3038, Sfax, Tunisia
benamira.bilel@yahoo.fr, {Zied.Driss,MohamedSalah.Abid}@enis.rnu.tn
Abstract. The aim of this paper is to study experimentally by particle image velo-
cimetry system (PIV) the effect of the pumping direction introduced with a 60°
pitched blade turbine. In fact, an Up- and a down-pitched blades turbine were
used. The particle image velocimetry technique is equipped with a double pulsed
Nd:YAG laser, a CCD camera resolution and a mini-synchronizer. Therefore, sev-
eral results were carried out to investigate the hydrodynamic structure and the
energetic parameters in a stirred vessel.
Keywords: pumping direction, 60° pitched blade turbine, PIV, hydrodynamic

structure.
1 Introduction
Since the efficiency of an agitated vessel is dependent on some geometric parame-

ters, type and dimensions of impeller, internals and physical properties of the me-
dium processed, it looks necessary to analyze the effect of one or more parameter
on the vessel performance. Several numerical process and experimental apparatus
have been used to investigate the hydrodynamic structure of the flow generated by
different impellers geometries. For example Mohan et al. (1992) used numerous
empirical heat transfers correlations to have a fundamental understanding of the
flow fields in stirred vessels. La Fontaine and Shepherd (1996) investigated the
time-averaged velocity vectors and the standard deviation of the fluctuations flow
in a stirred vessel using a scanning particle image velocimetry (PIV) system. Guil-
lard et al. (1998) used Laser Induced Fluorescence (LIF) and Planar Laser Induced
Fluorescence (PLIF) techniques to determine mixing and entrainment effects in
many flow configurations. Guillard et al. (2000) used Laser Induced Fluorescence
(PLIF) to investigate the turbulent mixing process in a model stirred tank equipped
with two Rushton turbines of standard geometry. Paglianti and Pintus (2001) de-
veloped an electrical impudence measurement to measure hod-up and mixing time
of a stirred tank equipped with a six blade Rushton turbine. Aubin et al. (2004)

92 B. Ben Amira, Z. Driss, and M.S. Abid
used laser doppler velocimetry (LDV) data to validate results of the single phase
turbulent flow in a stirred tank generated by a down- and an up-pumping pitched
blade turbine. Gabriele et al. (2009) used angle resolved particle image velocime-
try to compare the turbulent flow characteristics generated with a 45° up- and
down-pumping pitched blade. Driss et al. (2010) developed specific computational
fluid dynamics (CFD) code to analyze numerically the turbulent flow field in a
stirred tank equipped with the pitched blade turbines. Driss et al. (2011) used spe-
cific computational fluid dynamics code to compare the laminar hydrodynamic
flow generated by mono and twin screw profile with simple and modified profiles.
Ben Amira et al. (2011) used particle image velocimetry to analyze the Reynolds
number effect in a cylindrical stirred tank generated with a six flat blade turbine.
Feng et al. (2012) used an explicit algebraic stress model (EASM) to simulate ani-
sotropic turbulent flows in baffled stirred tanks equipped with a standard Rushton
turbine.
This work is an attempt to enhance the studies of the pitched blades turbine to
improve the mixing in a stirred vessel instead the 45° turbine. PBTU 60° and the
PBTD 60° are proposed according to these previous studies and as a continuation
of a previous works.
2 Experimental Setup
The schematic of the experimental setup is shown in figure 1. Obviously, the ex-
periments were carried out in a cylindrical stirred vessel. The diameter of the
Fig. 1 Experimental setup

Experimental Study of the 60° PBT6 Pitching Blade Effect with a PIV Application 93
stirred tank is equal to T = 0.3 m. Four baffles with width of 0.03 m are equipped
equally-spaced at the wall. The liquid height for all experiments is set at H=T.
The off-bottom clearance (from the middle of the impeller to the vessel bottom) is
T/3 in experiments. Two different types of stirrers are used, namely, 60° pitched
blade turbine down flow (PBTD) and 60° pitched blade turbine up flow (PBTU).
The experimental data were taken from a particle image velocimetry (PIV) system
which equipped with a 532 nm green light Nd:YAG laser, a 1600 x 1200 pixel² of
CCD camera resolution and a mini-synchronizer. In this study, we are interested in
the analysis of the hydrodynamic structure of turbulent flow. Up- and down-
pumping pitched six-blade turbines with a blade inclination equal to 60° were used
(figure 2). In fact, we present results associated to the vertical measurement plan
which located midway between two blades. The diameter of the impeller is equal
to 102 mm and the tangential velocity at the tip of the blade is equal to 0.8 m.s-1.
(a) PBTU 60° (b) PBTD 60°
Fig. 2 Stirred tanks configurations
3 Experimental Results
3.1 Velocity Field

Figure 3 shows the velocity field for both the up- and down- pumping pitched six-
blade turbines. Two different zones are presented in the vessel. In fact, one of
these zones is characterized by the vortex attached to the blade and the passage of
the vortex from the previous blade, but the other zone is characterized with the ab-
sence of vortices. In the up-pumping configuration, the volume upper the impeller
is more extended than in the down-pumping configuration. However, in the down-
pumping configuration, the volume lower the impeller is more extended than in
the up-pumping configuration. Two main recirculation loops are presented in the
up-pumping configuration. Thus, one is localized in the third upper zone of
the vessel and attached to the free surface. Otherwise, the second one is attached
the wall of the vessel, which it is larger than the first recirculation loop. However,
one main recirculation loop is presented in the down-pumping configuration
which localized in the bottom of the vessel and near the vessel wall. In addition,
there are no vortices presented in the volume upper the impeller which characte-
rizes by the lowest velocity. Close to the impeller, the stream discharged from the
blade as the fluid accelerates from a downward to an upward direction in the upper
reaches of the vessel. In fact, the sloping stream turns vertically close the impeller,
then, included velocity direction appears between the main recirculation loops.
Otherwise, the stream discharged from the blade causing a local velocity in the
down-pumping configuration. Whereas, the velocity increases as the fluid accele-
rates from an upward to a downward direction in the inferior reaches of the vessel.
The sloping stream confined by the bottom wall and the side wall.
V=49.9 mm/s V=51.2 mm/s
Fig. 3 Velocity field
3.2 Mean Instantaneous Velocity

Figure 4 shows the variation of mean instantaneous velocity with time for both the
up- and down- pumping pitched six-blade turbines. Subsequently, for both the two
applications, the x-axis presents the time axis in which the unit of time reflects to
one shaft rotation. Thereby, it is clearly evident that the flow oscillates in a sinu-
soidal form. In addition, it has been seen that for the up pumping configuration,
the velocity oscillation is the same along the experience. However, for the down
pumping configuration, the velocity oscillation decreases and increase along the
experience.
Velocity (mm.s-
Velocity (mm.s-
Fig. 4 Variation of mean instantaneous velocity with time
3.3 Vortices
Figure 5 shows the distribution of the vortices for both up- and down- pumping
pitched six-blade turbines. The maximum values are observed in two locations
corresponding to the location of highest value and the lowest values for both up-
and down- pumping configuration. Consequently, the highest and the lowest local
vortex values were found at the impeller level and become medium moving both
upward and downward from this level. For the up-pumping configuration, the
trailing vortices cross the plane to the back side at the impeller tip and to the front
side above the blade. However, for the down-pumping configuration, the trailing
vortices cross the plane to the back side above the blade and to the front side at the
impeller tip. Indeed, it has been observed that the high vortex value of the down-
pumping configuration is greater than the high vortex value of the up-pumping
configuration. However, the values of the lowest vortex of the down-pumping
configuration and the up-pumping configuration are almost the same.
30.9 s-1 55 s-1
1.8 s-1 12.7 s-1
-27.3 s-1 -29.7 s-1
Fig. 5 Vortices
3.4 Turbulent Kinetic Energy

Figure 6 shows the distribution of the turbulent kinetic energy for both the up- and
down- pumping pitched six-blade turbines. The turbulent kinetic energy in the up-
pumping configuration has two maximum regions. The faraway region is localized
in the same direction of the velocity acceleration between the two main recircula-
tion loops. Moreover, it has been noted that the impeller defuse the energy to the
vessel at the same direction of the discharge flow. Otherwise, it is clear that the
down-pumping turbine transfers the kinetic energy to the region near the bottom
of the tank. Therefore, in this case it has been observed that the turbulent kinetic
energy region which generated with one main circulation loop is different from
that in the double main circulation loops which transferred to the upper-right
region of the turbine.
0.16 0.35
0.14
0.3
0.12
0.25
0.1
0.2
0.08
0.15
0.06
0.1
0.04
0.02 0.05
Fig. 6 Distribution of the turbulent kinetic energy
3.5 Dissipation Rate of the Turbulent Kinetic Energy

Figure 7 shows the dissipation rate of the turbulent kinetic energy for both the up-
and down- pumping pitched six-blade turbines. The dissipation turbulent kinetic
energy rate in the up-pumping configuration is localized close to the impeller,
which has the same direction of the velocity acceleration between the two main
recirculation loops. However, the down-pumping turbine transfers the energy to
the region near the bottom of the tank.
0.045 0.11
0.04 0.1
0.09
0.035
0.08
0.03
0.07
0.025
0.06
0.02 0.05
0.015 0.04
0.03
0.01
0.02
0.005
0.01
0
Fig. 7 Distribution of the dissipation rate of the turbulent kinetic energy
3.6 Profiles
In this section we are interested to the radial revolution velocity profile, the radial
velocity component profile, the radial rms velocity profile, the turbulent kinetic
energy profile, the dissipation rate of the turbulent kinetic energy profile and the
turbulent viscosity profile in several positions in the bulk tank for both the up- and
down- pumping pitched six-blade turbines. Figure 8 shows the radial revolution of
the profiles at the axial position equal to z=1/3 H from the free surface. Therefore,
for all configurations, it has been seen that the hydrodynamic structure of the ves-
sel has the same distribution along the vessel. The up- and the down-pumping con-
figurations have the similar effect at this region. However, it has been seen that the
up-pumping turbine is greeter that the down-pumping turbine at the upper region
of the vessel. Then, the highest value of the up-pumping configuration were pre-
sented at the half of the vessel near the vessel wall which described by the main
recirculation loop. A quick decrease of the velocity near the vessel wall has been
noted. However, the first half region besides the shaft is described by the absence
of the recirculation loop for both configurations. Consequently, at this region, the
up-and the down-pumping configuration were looked confounded.Figure 9 shows
the radial revolution of the profiles at the axial position equal to z=1/2 H. It has
been seen that the up-pumping turbine is greeter that the down-pumping turbine at
the medium of the vessel. Moreover, the down-pumping configuration has sinu-
soidal reparation along the vessel. Then, the highest value of the up-pumping con-
figuration were presented at the medium region of the vessel which described by
the discharge flow region. A progressive decrease of the velocity has been noted.
(a) Velocity (b) rms (c) vorticity
(d) Turbulent kinetic energy (e) Dissipation rate of the (f) Turbulent viscosity
turbulent kinetic energy
Fig. 8 High third position (z=2/3 H)
(d) Turbulent kinetic ener- (e) Dissipation rate of the (f) Turbulent viscosity
gy turbulent kinetic energy
Fig. 9 Half position (z=1/2 H)
Figure 10. shows the radial revolution of the profiles at the axial position equal
to z=1/3 H from the vessel bottom. Therefore, for all configurations, it has been
seen that the hydrodynamic structure of the vessel has the same distribution along
the vessel unless both velocity and vortex change direction near the vessel wall
which described by the second recirculation loop.
(d) Turbulent kinetic energy (e) Dissipation rate of the (f) Turbulent viscosity
turbulent kinetic energy
Fig. 10 Lowest third position (z=1/3 H)
4 Conclusion
In this work, we have studied the effect of the pumping direction on the hydrody-
namic structure of a stirred vessel. In fact, it has been noted that the vessel which
equipped by the PBTU configuration is more stirred than the PBTD configuration.
In addition, the velocity change direction only for the up-pumping configuration
unless at the upper region of the vessel. Moreover, the trailing vortices and the
highest values were found in the same direction of the flow discharge.
In the future, we propose to develop a numerical simulation to study a new
geometrical parameters.
References
Aubin, J., Fletcher, D.F., Xuereb, C.: Modeling turbulent flow in stirred tanks with CFD:
the influence of the modeling approach, turbulence model and numerical scheme. Expe-
rimental Thermal and Fluid Science 28, 431–445 (2004)
Ben Amira, B.: Experimental characterization of a stirred vessel equipped with eight flat
blades turbine using a PIV system: Effect of the Reynolds number. Récents Progrès en
Génie des Procédés, Numéro 101–2011. Ed. SFGP, Paris, ISSN 1775-335X - ISBN
2-910239-75-6
Driss, Z., Bouzgarrou, G., Kchaou, H., Abid, M.S.: Simulation numérique de l’écoulement
laminaire induit dans des cuves agitées générées par les mobiles de proximité de type
mono et double vis à profils simple et modifié. Mécanique & Industries 12, 109–121
(2011)
Driss, Z., Bouzgarrou, G., Chtourou, W., Kchaou, H., Abid, M.S.: Computational studies of
the pitched blade turbines design effect on the stirred tank flow characteristics. European
Journal of Mechanics B/Fluids 29, 236–245 (2010)
Feng, X., Cheng, J., Li, X., Yang, C., Mao, Z.S.: Numerical simulation of turbulent flow in
a baffled stirred tank with an explicit algebraic stress model. Chemical Engineering
Science 69, 30–44 (2012)
Gabriele, A., Nienow, A.W., Simmons, M.J.H.: Use of angle resolved PIV to estimate local
specific energy dissipation rates for up-and down-pumping pitched blade agitators in a
stirred tank. Chemical Engineering Science 64, 126–143 (2009)
Guillard, F., Fritzon, R., Revstedt, J., TraKgasrdh, C., AldeHn, M., Fuchs, L.: Mixing in a
confined turbulent impinging jet using planar laser induced fluorescence. Exp. Flu-
ids 25, 143–150 (1998)
Guillard, F., TraKgasrdh, C., Fuchs, L.: A study of turbulent mixing in a turbine-agitated
tank using a fluorescence Technique. Experiments in Fluids 28, 225–235 (2000)
La Fontaine, R.F., Shepherd, I.C.: Particle Image Velocimetry Applied to a Stirred Vessel.
Experimental Thermal and Fluid Science 12, 256–264 (1996)
Mohan, P., Emery, A.N., AI-Hassan, T.: Review Heat Transfer to Newtonian Fluids in Me-
chanically Agitated Vessels. Experimental Thermal and Fluid Science 5, 861–883
(1992)
Paglianti, A., Pintus, S.: An impedance probe for the measurements of liquid hold-up and
mixing time in two/three-phase stirred tank reactors. Experiments in Fluids 31, 417–427
(2001)
Incidence Angle Effect on the Aerodynamic
Structure of an Incurved Savonius Wind Rotor
Zied Driss*, Olfa Mlayeh, Dorra Driss, Makram Maaloul,

and Mohamed Salah Abid
Laboratory of Electro-Mechanic Systems (LASEM), National School of Engineers of Sfax

(ENIS), Univrsity of Sfax, B.P. 1173, Road Soukra km 3.5, 3038 Sfax, TUNISIA
Zied.Driss@enis.rnu.tn
Abstract. In this paper, a computational fluid dynamics investigation has been

developed to study the incidence angle effect on the aerodynamic structure of an
incurved Savonius wind rotor. The software "SolidWorks Flow Simulation" has
been used to present the local characteristics. The numerical model considered is
based on the resolution of the Navier-Stokes equations in conjunction with the
standard k-ε turbulence model. These equations were solved by a finite volume
discretization method. An experimental validation has been done using a wind
tunnel to confirm the computer method validity.
Keywords: Savonius wind rotor, incidence angle, fluid dynamic, aerodynamic.
1 Introduction
Savonius wind rotor has many advantages over others in that its constructions are
simpler and cheaper. It is independent of the wind direction and has a good start-
ing torque at lower wind speeds (Altan et al. 2008, Saha et al. 2008, Mohamed et
al. 2010, Damak et al. 2013, Driss et al. 2014). Therefore, available in the litera-
ture are a lot of studies that have been conducted to increase the performance of a
Savonius wind rotor. For example, Grinspan et al. (2001) developed a new blade
shape with a twist for the Savonius rotor. Saha and Rajkumar (2006) performed
work on twist bladed metallic Savonius rotor and compared the performance
with conventional semi-circular blades having no twist. Saha et al. (2008)
conducted wind tunnel tests to assess the aerodynamic performance of single,
two and three-stage Savonius rotor systems. Both semicircular and twisted
blades have been used. In this paper, we are interested on the study of the inci-
dence angle effect on the aerodynamic structure characteristics of an incurved
Savonius wind rotor. Results have been visualized and commented.
*

102 Z. Driss et al.
2 Cases Study
Using the software "SolidWorks FloWorks”, a computer simulation of the aerody-

namic structure was achieved due to the variation of the incidence angle θ. Four
different values of the incidence angle θ equal to 0°, 45°, 90° and 135° were con-
sidered (figure 1).
(a) θ=0° (b) θ=45°
(c) θ=90° (d) θ=135°

Fig. 1 Incidence angles
3 Numerical Results
3.1 Velocity Field

Figure 2 shows the distribution of the velocity field in the longitudinal plane de-
fined by x=0 mm. According to these results, the velocity field in the tunnel intake
is uniform. It increases gradually with the decrease of the tunnel section. Upstream
of the rotor, a deceleration of the velocity has been observed in the region located
in the concave surface of the rotor. In the other hand, a rapid deceleration down-
stream of the rotor has been observed. This fact is due to the rotor considered as
an obstacle in front of the air flow that captures the kinetic energy of the wind.
The distribution of the velocity field presents the formation of recirculation zones
on the advancing and returning buckets that depends on the incidence angle.
Incidence Angle Effect on the Aerodynamic Structure 103
For θ=45°, a large recirculation zone is formed in the downstream of the rotor.
However, for θ=90° two recirculation zones have been observed in the down-
stream of the advancing and returning buckets. The direction of the infiltration has
an effect on the formation of the recirculation zones. Indeed, the wake characteris-
tic of the maximum values has been observed above and under the rotor both for
incidence angle θ=45° and θ=90°. Two acceleration zones has been observed in
the downstream of the rotor where the maximum velocity values are recorded. The
largest wake is formed for θ=45° and is located in the exterior endplates of the re-
turning bucket, while the wakes formed in the exterior sides of the two buckets for
θ=90° and θ=135° and that of the advancing bucket for θ=45° are less important.
For θ=0°, an acceleration zone where the maximum velocity values are recorded is
located in the convex surface of the two buckets.
(a) θ=0° (b) θ=45°
(c) θ=90° (d) θ=135°

Fig. 2 Distribution of the velocity field
3.2 Total Pressure

Figure 3 shows the distribution of the total pressure in the longitudinal plane de-
fined by x=0 mm. According to these results, the total pressure is globally uniform
in the wind tunnel upstream. In the downstream, it presents a depression zone
104 Z. Driss et al.
which is weak for θ=0° and relatively weak for θ=135°, while it becomes impor-
tant for θ=45° and θ=90°. In the other hand, the total pressure is found to be great-
er whenever the convex surface of the advancing bucket of the rotor is facing the
air flow. For an incidence angle θ=0°, the total pressure is uniform and globally
low with a zone of minimum pressure value located in the concave surface of the
rotor buckets. While for θ=45° and θ=135°, the total pressure increases and is
greater for an incidence angle θ=45°. This fact can be explained by the bucket po-
sition. In fact, in this case the returning bucket is facing the flow while for θ=135°
it’s the advancing bucket. For θ=90° the configuration of the rotor is allowing it to
play the role of a large obstacle and the total pressure is high in the collector and
in the upstream of the rotor. A rapid decrease in the total pressure in the down-
stream of the rotor causes a depression zone and the total pressure keeps increas-
ing in the way out of the wind tunnel. The formation of an important compression
zone has been noted for θ=90°, θ=45° and is less important for θ=135°. However,
for θ=0° the wake formed is restricted to the zone located in the concave surface
of the rotor buckets.
(a) θ=0° (b) θ=45°
(c) θ=90° (d) θ=135°

Fig. 3 Distribution of the total pressure
3.3 Dynamic Pressure

Figure 4 shows the distribution of the dynamic pressure in the longitudinal plane
defined respectively by x=0 mm. According to these results, it has been noted that
the dynamic pressure value is found to be low. During the cross collector, a pro-
gressive increase of the dynamic pressure has been observed. This fact is due to
the reduction of the transverse surface of the collector. In the upstream of the ro-
tor, it has been noted that the dynamic pressure decreases along its crossing
through the distributor. The four configurations present a minimum values area lo-
cated in the interior zone of the rotor. For θ=45° and θ=90°, two important com-
pression zones are located in the endplates where the flow is throttled. For θ=0°,
the distribution of the dynamic pressure is uniform and get to its maximum in the
convex surface of the rotor buckets. For θ=45°, a compression zone where the dy-
namic pressure is at its maximum is formed in the endplates of both the advancing
and the returning buckets. For θ=90° and θ=135°, two similar compression zones
are formed, the first is located in convex surface of the returning bucket and the
second is located in the endplate of the advancing bucket. The distribution of the
pressure is found to be almost symmetrical for both θ=0° and θ=90° as a result of
the symmetrical aspect of the configuration.
(a) θ=0° (b) θ=45°
(c) θ=90° (d) θ=135°
Fig. 4 Distribution of the dynamic pressure

106 Z. Driss et al.
However the dissymmetrical configuration for θ =45° and θ=135° generates a

dissymmetrical distribution of the dynamic pressure. After the exit of the test vein,
it has been noted that the dynamic pressure decreases along its crossing through
the distributor until reaching atmospheric pressure.

Figure 5 shows the distribution of the turbulent kinetic energy in the longitudinal
plane defined respectively by x=0 mm. According to these results, It has been
noted that the turbulent kinetic energy is very weak out of the test vein for θ=0°.
For this angle, the maximum energy’s value is recorded in the exterior endplates
of the rotor and the convex surface of the advancing bucket. In the diffuser, the
maximum turbulent kinetic energy values are recorded for θ=90° and θ=45°.
However, for θ=0° and θ=135° the maximum turbulent kinetic energy values are
located in the endplates and around the shaft. Two symmetrical wakes are formed
in the downstream of the rotor where they get mixed for θ =45°. The configuration
for which the greater zone where the maximum values of turbulent kinetic energy
are located is found for θ=90°.
(a) θ=0° (b) θ=45°
(c) θ=90° (d) θ=135°

In this case the diffuser is globally considered as a zone that holds the maxi-
mum value of the turbulent kinetic energy with the exception of the lateral surface
of the diffuser’s exit in which the energy’s value drops. For θ=45° as well as
θ=135°, the turbulent kinetic energy’s maximum value is located in the exterior
endplates of the rotor’s buckets and around the shaft. Although the turbulent kinet-
ic energy is greater in the rotor’s area for θ=135°, the opposite is noted in the zone
of the diffuser.
3.5 Turbulent Dissipation Rate

Figure 6 shows the distribution of the turbulent dissipation rate in the longitudinal
plane defined by x=0 mm. According to these results, it has been noted that the
turbulent dissipation rate is very weak out of the test vein for all values of the in-
cidence angle. In addition, it is clear that these wakes created in the end plates of
the rotor’s buckets increase with the incidence angle θ. For θ=90°, the wake is ex-
tending downstream of the rotor and the maximum values of the turbulent dissipa-
tion rate are recorded in the diffuser. However, the maximum values of the turbulent
dissipation rate are located in the exterior endplates for θ=0° and in the exterior and
the interior endplates and around the shaft for θ=45° and θ=135°.
(a) θ=0° (b) θ=45°
(c) θ=90° (d) θ=135°

Fig. 6 Distribution of the turbulent dissipation rate
108 Z. Driss et al.
Indeed, two symmetrical wakes are formed in the downstream of the rotor for
θ=45° and θ=90°. The greater one is found to be for θ=90°. In this case, the test
section in the rotor downstream is globally considered as a zone that holds the
maximum value of the turbulent dissipation rate. Then, it drops through the diffus-
er’s exit. For θ=45° as well as θ=135°, the maximum value of the turbulent dissi-
pation rate is located in the exterior endplates of the rotor buckets and around the
shaft.
3.6 Turbulent Viscosity

Figure 7 shows the distribution of the turbulent viscosity in the plane defined by
x=0 mm. According to these results, it has been noted that the turbulence viscosity
is very weak in the collector inlet and in the upstream of the rotor. After the exit of
the test vein, it’s noted that the turbulent viscosity increases in the exit of the dif-
fuser both for θ=0° and θ=135°. However for θ=45° and θ=90°, the wake that
holds the maximum value of turbulent viscosity is extending among the diffuser.
The distribution of the turbulent viscosity depends strongly on the rotor configura-
tion that is generated by the incidence angle.
(a) θ=0° (b) θ=45°
(c) θ=90° (d) θ=135°

Fig. 7 Distribution of the turbulent viscosity
For θ=0°, the viscosity increases slightly in the concave surface of the two
buckets, keeps decreasing out of the test section and increases again forming a
wake characteristic of the maximum values of the turbulent viscosity in the lat-
eral walls of the diffuser’s exit. A relatively similar behavior has been noted for
θ=135°. In fact, the viscosity increases in the concave surface of the buckets and
again in the diffuser exit. Moreover, another wake is found to be formed in the
convex surface of the advancing rotor bucket and in the diffuser exit. This wake
is greater than in θ=0° and it covers not only the walls but the whole exit area.
For θ=45° and θ=90°, the wake holds the maximum viscosity values and extends
on the diffuser. For θ=45°, the wake is created in the exterior endplate of the ad-
vancing bucket while for θ=90° the wake covers the whole convex surface of the
advancing bucket.
4 Comparison with Experimental Results
Figure 8 compares the numerical values of velocity obtained by numerical simula-

tion with the experimental values gathered from the experiments conducted in the
LASEM laboratory (Driss et al., 2014). The velocity profile was chosen for points
situated in the test section in the direction defined by z=-150 mm and y=-150 mm.
This profile gives symmetrical values relative to the middle plane (x=0 mm). The
velocity decreases due to the presence of the rotor. Also, it has been observed that
the velocity is globally constant in the region away from both the test vein walls,
in which the velocity is set to zero. Due to the rotor shape for which the curvature
is not very important, the velocity values are slightly affected by the rotor. The
comparison between the experimental and numerical velocity values lead us to the
conclusion that despite some unconformities, the values are comparable. The nu-
merical model seems to be able to predict the aerodynamic characteristics of the
air flow around the incurved Savonius wind rotor.
V (m/s)
15
10
Experimental
Numerical
0 x(mm)
-200 -100 0 100 200
Fig. 8 Velocity profile

110 Z. Driss et al.
5 Conclusion
Numerical investigation has been developed to study the incidence angle effect
on the aerodynamic structure of an incurved Savonius wind rotor. Numerical re-
sults, such as velocity fields, pressure and turbulent characteristics are presented in
the wind tunnel, considered as a control volume. According to the obtained re-
sults, the the incidence angle of the incurved Savonis wind rotor has a direct
effect on the local characteristics.
In the future, we intend to change the bucket geometrical parameters and pro-
pose new configurations in order to increase the performance of the Savonius
wind rotor.
References
Altan, B.D., Atılgan, M., Özdamar, A.: An experimental study on improvement of a Savo-
nius rotor performance with curtaining. Experimental Thermal and Fluid Science 32,
1673–1678 (2008)
Damak, A., Driss, Z., Abid, M.S.: Experimental investigation of helical Savonius rotor with
a twist of 180°. Renewable Energy 52, 136–142 (2013)
Driss, Z., Mlayeh, O., Driss, D., Maaloul, M., Abid, M.S.: Numerical simulation and expe-
rimental validation of the turbulent flow around a small incurved Savonius wind rotor.
Energy 74, 506–517 (2014)
Mohamed, M.H., Janiga, G., Pap, E., Thévenin, D.: Optimization of Savonius turbines us-
ing an obstacle shielding there turning blade. Renewable Energy 35, 2618–2626 (2010)
Saha, U.K., Thotla, S., Maity, D.: Optimum design configuration of Savonius rotor through
wind tunnel experiments. Journal of Wind Engineering and Industrial Aerodynamics 96,
1359–1375 (2008)
Grinspan, A.S., Kumar, P.S., Saha, U.K., Mahanta, P., Ratnarao, D.V., Veda Bhanu, G.:
Design development & testing of Savonius wind turbine rotor with twisted blades. In:
Proceedings of National Conference on Fluid Mechanics & Fluid Power, India, vol. 28,
pp. 28–31 (2001)
Saha, U.K., Jaya Rajkumar, M.: On the performance analysis of Savonius rotor with twisted
blades. Renewable Energy 31, 1776–1788 (2006)
Saha, U.K., Thotla, S., Maity, D.: Optimum design configuration of Savonius rotor through
wind tunnel experiments. Journal of Wind Engineering and Industrial Aerodynamics 96,
1359–1375 (2008)
Experimental Characterization of a NACA2415
Airfoil Wind Turbine
Zied Driss1,*, Tarek Chelbi1, Ahmed Kaffel2, and Mohamed Salah Abid1
1
Laboratory of Electro-Mechanic Systems (LASEM), National School of Engineers of Sfax
(ENIS), Univrsity of Sfax, B.P. 1173, Road Soukra km 3.5, 3038 Sfax, TUNISIA
2
University of Maryland College Park, MD 20742, USA
Zied.Driss@enis.rnu.tn
Abstract. In this paper, an experimental characterization of a horizontal axis

NACA2415 airfoil type wind turbine has been developed. Particularly, a detailed
description of the used wind tunnel and the various manipulations performed are
presented to study the aerodynamic characteristics of the wind turbine. The ex-
perimental setup is developed to estimate the velocity profiles and the torque me-
ter variation for different Reynolds numbers and wedging angles of the blade. The
main objective is to develop results to validate the numerical results established
with Computational Fluid Dynamics (CFD) codes.
Keywords: Wind turbine, NACA2415 airfoil, experimental.
1 Introduction
To prepare a truly sustainable development, the share of renewable resources for

electricity generation is now crucial. In this context, many scientists have experi-
mentally and numerically examined the effects of parameters design such as
blades number, wedging angle and airfoil profile. For example, Hu et al. (2006)
showed that Coriolis and centrifugal forces play important roles in 3D stall-delay.
At the root area of the blade, where the high angles of attack occur, the effect of
the Coriolis and centrifugal forces is dominant. Thus, it shows apparent stall-delay
phenomenon at the inner part of the blade. However, by increasing the Reynolds
number, the separation position has a stronger effect than by increasing the Cori-
olis and centrifugal forces. Sicot et al. (2008) investigated the aerodynamic prop-
erties of a wind turbine airfoil. Particularly, they studied the influence of the
inflow turbulence level (from 4.5% to 12%) and of the rotation on the stall me-
chanisms in the blade. A local approach was used to study the influence of these
parameters on the separation point position on the suction surface of the airfoil,
through simultaneous surface pressure measurements around the airfoil.
*

112 Z. Driss et al.
Hirahara et al. (2005) developed a unique and very small wind turbine with a di-
ameter of 500 mm and a small aspect ratio for wide use in urban space. The basic
performance of μF500 was tested for various free stream and load resistance. The
airflow around the turbine was investigated using a particle image velocimetry
(PIV). Schreck and Robinson (2007) showed that wind turbine blade aerodynamic
phenomena can be broadly categorized according to the operating state of the ma-
chine, and two particular aerodynamic phenomena assume crucial importance. At
zero and low rotor yaw angles, increasing rotation determines blade aerodynamic
response. At moderate to high yaw angles, dynamic stall dominates blade aerody-
namic. Wright and Wood (2004) showed that the acceleration and deceleration of
the rotor at speeds below its controlled maximum speed for a range of wind speeds
were calculated and compared with data. The main goal of the Mirzaei et al.
(2009) investigation was to understand the flow field structure of the separation
bubble formed on NLF-0414 airfoil with glaze-ice accretions using CFD and hot-
wire anemometry and comparing these results with previous researches performed
on NACA 0012 airfoil. Driss et al. (20014) investigated the complex flow field
developing around a NACA2415 airfoil wind turbine and studied the wedging an-
gle effect of the airfoil. This will help improving the aerodynamic efficiency in the
design of packaged installations of the NACA2415 airfoil type wind turbine. In
other paper, Driss et al. (20014) studied the turbulent flow around a small in-
curved Savonius wind rotor. Experimental results are conducted on an open wind
tunnel equipped by a small incurved Savonius wind rotor to validate the numerical
method.
In this paper, we are interested on the study of the experimental characteriza-
tion of the horizontal axis NACA2415 airfoil type wind turbine. Particularly, a de-
tailed description of the used wind tunnel and the various manipulations per-
formed will be presented to study the aerodynamic characteristics of the wind
turbine.
2 Materials and Method
2.1 Wind Turbine

In the experimental study, we are interested in a NACA2415 airfoil wind turbine.
This turbine consists of three blades with a length L and a width C. The rotor is
equipped with a mechanism for varying the wedging angle. This angle is meas-
ured between the rotation plane of the wind turbine and the chord. Figure 1
presents the geometric parameters of horizontal axis wind turbine, defined by:
• Rotor radius R=157 mm,
• Length of the blade L=110 mm,
• Chord length C=45 mm,
• Variable wedging angle β.
Experimental Characterization of a NACA2415 Airfoil Wind Turbine 113
Fig. 1 Geometric parameters
2.2 Wind Tunnel

The wind tunnel is an experimental device used to study and determine the aero-
dynamic behavior of the NACA2415 airfoil wind turbine installed in a test sec-
tion. Our wind tunnel is open type. It allows providing a flow of air stable and
uniform in the test section through a vacuum downstream of the flow.
7 Wind turbine
6 Support
5 Ventilation chamber
4 Diffuser
3 Test vein
2 Collector
1 Plenum
Ref. Designation
Fig. 2 Wind tunnel

114 Z. Driss et al.
2.3 Anemometer
To initiate the various processes necessary for the characterization of the rotor
with the wind tunnel, several measuring instruments were used. Among which in-
clude the tachometer, the anemometer, the static torque meter and the dynamome-
ter. The experimental study conducted as part of this work involved the realisation
of a horizontal axis wind turbine, a wind tunnel. Experimental trials, allowing the
determination of the static torque Ms, the dynamic torque Md and the velocities
profiles in different directions pre-selected in the test section. Such a system is
equipped with a specific instrument for the development of various experimental
tests required at the laboratory scale. The anemometer measures wind speed in dif-
ferent positions in the pre-selected test section or outside. This is feasible, since
the range of variation of speed is between 0.2 and 20 m.s-1 with a resolution which
reaches 0.1 m.s-1. The different characteristics of the anemometer used are
summarized in table 1.
Fig. 3 Measuring of the average speed using an anemometer
Table 1 Characteristics of the anemometer
Description Anemometer type AM 4204

Maker Lutron
Probe type telescopic
Measurement parameters Air velocity , temperature, gas flow
Air velocity 0.1 m.s-1
Resolution
Temperature 0.1 ° C
Air velocity 5%
Precision
Temperature ±0,8 °C
Velocity 0.1 m.s-1
Measuring range
Temperature from -20 °C to +70 °C
2.4 Torque Meter

Figure 4 presents the static torque meter used for measuring the static torque ex-
erted by the rotor. Under these conditions, the wind turbine is held fixed and can-
not turn. Therefore, our experience consists on the measure of the aerodynamic
force exerted by air on the wings at different angular positions. The characteristics
of the used torque meter are summarized in table 2.
Fig. 4 Measurement method of the static torque meter
Table 2 Characteristics of the torque meter
Description Static torquemeter TQ 8800

Maker Lutron
Measurement parameters Static torque
Maximum measurement range 0-147.1 N.cm
Resolution 0,1 N.cm
Precision ± (1.5 % + 5d)
Overload capacity 220.1 N.cm max
Dimension 180 x 72 x 32 mm
3.1 Static Torque Variation

In this section, we are interested on the study of the Reynolds number effect on the
static torque variation. Particularly, static torque is measured using the torque
116 Z. Driss et al.
meter types TQ 8800 mounted directly on the rotor axis. Particularly, we chose to
study different flow regimes defined by Reynolds numbers equal to Re=165093,
Re=194350, Re=217338, Re=242415, Re=257044 and Re=265403. Figures 5 and
6 present the variation of the static torque Ms according to the wedging angle β.
These results relate well to the study of the influence of pitch angle on the static
torque for different Reynolds numbers. According to these results, it has been
noted that these curves have the same allure. With the increase of the wedging an-
gle β, the value of the static torque increases and reaches a maximum and then it
decreases. Figure 6 shows the variation of static torque as a function of the wedg-
ing angle β for different Reynolds numbers. From these results, it has been noted
that the maximum static torque Ms is obtained for a Reynolds number equal to
Re=265403. In this case, the value of the static torque is equal to Ms=5 N.cm at
the corresponding wedging angle equal to β=50 °. All these curves reach their
maximum for the same wedging angle. Moreover, it has been noted that the mini-
mum value of the static torque Ms is obtained for a Reynolds number equal to
Re=165093. In this case, the static torque value is equal to Ms=0.5 N.cm with a
wedging angle β=5°.
Fig. 5 Results superposition

(a) Re=165093 (b) Re=194350
(c) Re=217338 (d) Re=242415
(e) Re=257044 (f) Re=265403
Fig. 6 Variation of the static torque Ms as a function of β

118 Z. Driss et al.
3.2 Angular Velocity

Figure 7 shows the variation of the angular velocity as a function of the wedging
angle β. According to these results it has been noted that the angular velocity of
the wind turbine, measured by the tachometer, gets increases by increasing the
wedging angle. It gets its maximum with β=25°. After that when the wedging
angle increases, the angular velocity decreases.
Fig. 7 Angular velocity variation as a function of β
3.3 Average Velocity

Figure 8 shows the velocity profiles in the transverse plane defined by the axial
position equal to z=100 mm. This plane is placed in the test vein of the wind tun-
nel in the downstream of the wind turbine. Each of these profiles are situated on
the directions defined by y=0 mm, y=50 mm, y=100 mm and y=150 mm. By set-
ting the two directions y and z, the work reduces to a measure of the average
velocity at different altitudes x. This is measured after positioning of the anemo-
meter in the correct position and reading the value displayed on the screen. Ac-
cording to these results, It has been noted that far from the wind turbine, the value
of speed is almost constant over the entire length of the test section. It was only at
the walls that the velocity is zero. As he approached the wind turbine, the average
velocity profile begins to undergo some oscillating indicating that the speed is no
longer uniform. The average velocity is very weak on the meeting of the wind tur-
bine blades. Moreover, there is a slight increase in average velocity in the wind
turbine downstream. Under these conditions, the speed can reach a value equal to
V= 15.8 m.s-1.
(a) y=0 mm (b) y=50 mm
(c) y=100 mm (d) y=150 mm
Fig. 8 Velocity profile in the transverse plane z =100 mm
4 Conclusion
In this paper, we have developed an experimental method to characterize a hori-

zontal axis NACA2415 airfoil type wind turbine. For thus, we have presented a
detailed description of the used wind tunnel and the considered manipulations.
Particularly, we are interested on the measuring of the average velocity of the flow
in different positions of the test vein. The static torque measurements are also per-
formed for different Reynolds numbers and wedging angles of the blade. In the
future, we propose to study others geometrical parameters.
120 Z. Driss et al.
References
Driss, Z., Chelbi, T., Kaffel, A., Abid, M.S.: Static Study of the Wedging Angle Effect of A
NACA2415 Airfoil Type Wind Turbine. World Research Journal of Engineering and
Technology 3, 1–8 (2014)
Driss, Z., Mlayeh, O., Driss, D., Maaloul, M., Abid, M.S.: Numerical simulation and expe-
rimental validation of the turbulent flow around a small incurved Savonius wind rotor.
Energy 74, 506–517 (2014)
Hu, D., Hu, O., Du, Z.: A study on stall-delay for horizontal axis wind turbine. Renewable
Energy 31, 821–836 (2006)
Hirahara, H., Hossain, M.Z., Kawahashia, M., Nonomura, Y.: Testing basic performance of
a very small wind turbine designed for multi-purposes. Renewable Energy 30,
1279–1297 (2005)
Mirzaei, M., Ardekani, M.A., Doosttalab, M.: Numerical and experimental study of flow
field characteristics of an iced airfoil. Aerospace Science and Technology 13, 267–276
(2009)
Schreck, S.J., Robinson, M.C.: Horizontal Axis Wind Turbine Blade Aerodynamics in Ex-
periments and Modeling. IEEE Transactions on Energy Conversion 22, 61–70 (2007)
Sicot, C., Devinant, P., Loyer, S., Hureau, J.: Rotational and turbulence effects on a wind
turbine blade: investigation of the stall mechanisms. Journal of Wind Engineering and
Industrial Aerodynamics 96, 1320–1331 (2008)
Wrigh, A.K., Wood, D.H.: The starting and low wind speed behaviour of a small horizontal
axis wind turbine. Journal of Wind Engineering and Industrial Aerodynamics 92,
1265–1279 (2004)
DFMM Approach: Modelling, Formalization
and Choice of Material and Manufacturing
Processes
Mohamed Saidi, Achraf Skander, and Slim Kaddeche
Laboratory of Materials, Measurements and Applications (MMA),

National institute of Applied Sciences and Technology (INSAT),
University of Carthage, Centre Urbain Nord- BP N° 676 - 1080 Tunis Cedex-Tunisia
Tunisia Polytechnic School, University of Carthage, Tunisia
saidi.mohamed.tunis@gmail.com,
{achraf_skander,slimkaddeche}@yahoo.fr
Abstract. In a very competitive industrial world, the integrated design and par-
ticularly the design for manufacturing make progress the profitability of the prod-
uct. The DFM (design for manufacturing) improves the transition between the
design and the manufacturing phases and reduces the time and the costs of product
manufacturing. Within this framework of research, the DFMM approach (design
for material and manufacturing) takes into account the constraints of manufactur-
ing, as of the phase of design, and makes it possible to choose consequently the
most adequate material and processes for the manufacturing of the mechanical
parts. This approach is composed primarily of the modelling of the product, the
manufacturing knowledge management, the choice of material and the choice of
manufacturing processes.
Keywords: Integrated design, DFMM, Modelling, Choice of the material, Choice

of the processes.
1 Introduction
Separation between design and manufacturing phases generates the realization of
not manufacturable concept. Then, the product is taken back to the design phase
for making the necessary modifications to be manufacturable. With these iterative
loops, the companies lose much time and money. As part of the integrated design
and simultaneous engineering, design for manufacturing solves this problem by
integrating the manufacturing parameters at the design stage. According to (Ker-
brat and Al 2011), the DFM treats the objectives of design and the constraints of
manufacturing simultaneously such is the case of the modular hybrid concept.
Other DFM applications help the designer in the choice of the manufacturing
processes such as the data-processing demonstrator “DFM_Peaux_Squelettes”,
(Skander and Al 2008). In this context, the DFMM is a more complete approach

122 M. Saidi, A. Skander, and S. Kaddeche
which makes it possible to choose the most adequate material and processes for
the manufacturing of the mechanical parts.
2 The DFMM Approach

The design for material and manufacturing helps the designer in his decision mak-
ing. Several software presents a reliable support for the choice of material and
processes like the “DFMA product-costing software”, (Assembly magazine 2013)
and the “CES Selector 2014”, (Granta Design 2014).
The DFMM approach takes into account the constraints of manufacturing and
allows to choose consequently material and processes for the manufacturing of the
mechanical parts, in this way it makes possible the good transition between the de-
sign of the product and its manufacture. With the research of (Ashby and Al 2004)
and (Gupta and Al 2003), the general structure of the approach is composed pri-
marily of the modelling of the part, the manufacturing knowledge management,
the choice of material and the choice of the manufacturing processes (cf Fig.1).
Fig. 1 Principal stages of the DFMM approach
2.1 Modelling of the Part

The modelling of the product, in occurrence the part, makes it possible to take and
transform all the properties of the part into exploitable data. The modelling of the part
is composed of the classification of the attributes of the material which is useful for the
choice of material and modelling by entities skin and skeleton which is used for the
choice of the processes.
DFMM Approach 123
• Classification of the attributes of material

The classification of the attributes of the material (Kalpesh and Bhatt 2009) de-
termines the properties of material by classifying the useful attributes for the
choice of the material into beneficial (whose the increase of the value would be
beneficial for the part) and not beneficial (whose the increase of the value would
be not beneficial for the part). Each material is defined by fourteen attributes
which will be classified (cf Table 1, in our example of use).
• Modelling by entities skin and skeleton

Modelling by entities skin and skeleton (Skander 2006) allows to the designer
to model all the part and to translate its geometrical and technological properties
into entities skin and skeleton. Each entity skeleton will be defined, moreover, by
a “thickness section skeleton max” and a “thickness section skeleton min”. The re-
sultants entities skin and skeleton of the modelling will be classified by the user in
two tables of modelling (cf Fig.3 and Fig.4, in our example of use).
2.2 Manufacturing Knowledge Management

The manufacturing knowledge management ensures the capitalization and the
formalization of the useful data for the choice of material and the processes. For
instance, (Ashby and Johnson 2009) classifies materials of manufacture while re-
ferring to their physical and mechanical properties.
For DFMM approach, the carried out database is composed of three nonexhaus-
tive tables: materials, manufacturing processes and compatibility material-
processes (Ashby and al 2001). By determining essentials information to the
DFMM (Ferrer and al. 2010), we formalized materials and the manufacturing
processes by choosing and adapting their attributes according to the attributes of
the modelling phase.
2.3 Choice of Material of Manufacturing

The choice of material of manufacturing allows to identify the most adequate ma-
terial for the manufacture of the part. During this stage, the classification of the at-
tributes of material and the database of materials will be used.
The MADM (multiple attribute decision making) is a concept which classifies
in a systematic way materials according to their competences by treating several
attributes for each material (Jahan and Al 2010). For the choice of material, we
preferred the PSI method (Preference selection index) where a preferably index is
calculated for each material (Kalpesh and Bhatt 2009). This method, which be-
longs to MADM concept, is composed of seven stages of calculation. At the end,
the materials will be ranked according to their preferably index. The material with
the highest index is selected like better alternative. Therefore, the designer can
choose the most adequate material while basing himself on this classification.
2.4 Choice of the Manufacturing Processes

The choice of the manufacturing processes depends on the choice of material and
allows to identify and classify the most adequate processes for the manufacture of
the part (Ashby and Al 2004), (Esawi and Ashby 2003). This stage uses modelling
by entities skin and skeleton and the database of the manufacturing processes.
The method of choice is composed of the following operations:
• Identification of the adequate processes for the manufacturing of all the

entities skeleton and all the entities skin of the part,
• Identification of the primary processes,
• Ranking of all the selected processes according to their relative means
cost.
The choice of the processes is based on the comparison of the attributes of the enti-
ties skin and skeleton of the part with the attributes of the compatibles processes of
the selected material, this method is composed of the following stages (cf Fig.2):
Fig. 2 Choice of the manufacturing processes
While following these stages (cf Fig.2), the designer can identify all the proc-
esses able to manufacture the totality of the part. Then, for each entity skin and
each entity skeleton, there is a list of ranked appropriate processes. Similarly, the
primary processes are identified and classified.
3 Example of Use
To implement this work, we used the example of the piston of a Schrader grip.
While following all the stages of the DFMM approach, we will determine the most
adequate material and processes for the manufacture of this part.
DFMM Approach 125
3.1 Modelling of the Part

The modelling of the part is the first stage of DFMM approach, it is composed of
the classification of the attributes of material and modelling by entities skin and
skeleton.
• Classification of the attributes of material

This stage classifies the necessary attributes to the choice of material in benefi-
cial and not beneficial attribute. In this example, the part must support a repetitive
compressive forces without a high deformation while remaining in the elastic
range. It must be light, not expensive and non-polluting. The table of classification
of the attributes of material includes the whole of these requirements (cf Table 1).
Table 1 Classification of the material attributes of the part
Attributes Beneficial Not beneficial

Young's modulus X
Shear modulus
Poisson's ratio X
Yield stress X
Vickers hardness
Thermal conductivity
Electrical resistivity
CO2 footprint X
Density X
Price X
Tensile strength
Compressive strength X
Mechanical loss coefficient
Fatigue strength X
• Modelling by entities skin and skeleton

By identifying all the entities skin and skeletons of use of the part, the piston is
modelled by seven entities skeleton and twelve entities skin. The tables of model-
ling by entities skin and skeleton brings together all the geometrical and techno-
logical properties of modelled entities. Then we take the example of the two enti-
ties skeleton 1 and 7 (cf Fig.3) and the two entities skin 1 and 2 (cf Fig.4).
Fig. 3 Modelling of the two entities skeleton 1 and 7
Fig. 4 Modelling of the two entities skin 1 and 2
3.2 Choice of the Material of Manufacturing

The choice of the material of manufacturing is carried out with the PSI method.
After the calculation of the preferably index for each material, the classification of
the highest indices ensures the identification of the most powerful materials for the
manufacture of the part (cf. Table 2). In our case, the first three materials are most
powerful but they have a short list of compatible processes which does not allows
to manufacture the totality of the part. Then the selected material is the nickel-
based super alloys which has a high preferably index and which has a wide range
of compatible processes.
DFMM Approach 127
Table 2 Choice of the material of manufacturing

classified materials Preference index Ranking
Alumina 0,68886 1
Tungsten carbide 0,681352 2
Silicon nitride 0,611825 3
Nickel-based super alloys 0,606125 4
Boron carbide 0,52545 5
3.3 Choice of the Manufacturing Processes

It is the last stage of DFMM approach, it makes it possible to identify:
• Manufacturing processes for each entity skin of the part,

• Manufacturing processes for each entity skeleton of the part,
• Primary processes for the manufacture of the part.
Table 3 Choice of the manufacturing processes for the two entities skin (1, 2) and the two
entities skeleton (1, 7)
The entities skin and skeleton of the part Primary processes

Manufacturing Average Manufacturing Average
Entity adequate processes relative adequate processes relative
cost cost
Sand casting 0.43 Sand casting 0.43
Lost wax Casting 0.49 Lost wax Casting 0.49
Powder injection
Skeleton 1 0.55 Powder injection molding 0.55
molding
Forging 0.55 Forging 0.55
Laser sintering 0.55 Laser sintering 0.55
Water-jet cutting 0.37
Drilling 0.49
Skeleton 7 Grinding 0.49
Laser cutting 0.52
Turning 0.56
Milling 0.49
Skin 1 Turning 0.56
Punching, perforating 0.49
Milling 0.49
Skin 2 Electric discharge
0.55
wire cutting
Turning 0.56
The processes compatible with selected material (Super alloys based on nickel)
will be used like database for the choice of the processes. The comparison be-
tween the attributes of these processes and the attributes of the entities skin and
skeleton allows to identify the processes able to carry out these entities. For exam-
ple, the capacities (attributes) of the processes of milling and turning make it
possible to manufacture the entity skin 1. Thereafter, the primary processes are
identified by taking the matter addition processes present in all the matter addition
entities (sens of material =1).
While following these stages, we identified the processes able to manufacture
the part such is the case of the two entities skin (1, 2) and the two entities skeleton
(1, 7). Thereafter, these processes are classified according to their average relative
cost (cf Table 3).
4 Discussion and Conclusion

DFMM approach is composed of the modelling of the part, the manufacturing
knowledge management, the choice of material and the choice of the manufactur-
ing processes. While following these stages for the example of the piston, the most
adequate material and processes to manufacture the part were given with effec-
tiveness and precision. On another side, for some stages, the implementation is a
little long such is the case of calculations for the choice of material. In this way,
DFMM approach allows to identify with effectiveness and precision the appropri-
ate material and processes but remains a little long to apply.
This article presents the DFMM approach which uses the concept of design for
manufacturing. The approach integrates the constraints of manufacturing as of the
phase of design and allows to the designer to choose, consequently, the most ade-
quate material and processes for the manufacture of its part. Therefore, DFMM
approach ensures the transition between the design phase and manufacturing
phase. Also, it reduces time and costs of the manufacturing of the product.
In prospects, this approach can generate other similar approaches like the design
for the environment or the design for assembly. We add the possibility of carrying
out a computer application which incarnates and reproduces DFMM approach.
Thereby the task of the designer will be faster and his work more effective.
References
Kerbrat, O., Mognol, P., Hascoët, J.Y.: A new DFM approach to combines machining
and additive manufacturing. Computers in Industry 62(2011), 684–692 (2011),
ISSN: 0166-3615
Skander, A., Roucoules, L., Meyer, J.S.K.: Design and manufacturing interface modelling
for manufacturing processes selection and knowledge Synthesis in design. The Interna-
tional Journal of Advanced Manufacturing Technology 37(5-6), 443–454 (2008)
Assembly magazine, DFMA Software Generates Cost-Efficient Product Design (2013),
http://www.assemblymag.com/articles/91090-dfma-software-
generates-cost-efficient-product-design (accessed April 1, 2013)
DFMM Approach 129
Granta Design. CES Selector 2014 (2014),

http://www.grantadesign.com/fr/ces.htm
Ashby, M.F., Brechet, Y.J.M., Cebon, D., Salvo, L.: Selection strategies for materials and
processes. Materials and Design 25, 51–67 (2004)
Gupta, S.K., Chen, Y.S., Feng, S., Sriram, R.: A system for generating process and ma-
terial selection advice during embodiment design of mechanical components. Journal of
Manufacturing Systems 22(1), 28–45 (2003) ISSN: 0278-6125
Kalpesh, M., Bhatt, M.G.: A selection of material using a novel type decision-making me-
thod: Preference selection index method. Materials and Design 31(2010), 1785–1789
(2009) ISSN: 0261-3069
Skander, A.: Méthode et modèle DFM pour le choix des procédés et l’intégration des con-
traintes de fabrication vers l’émergence de la solution produit. Thesis of doctorate of the
University of Technology of Troyes (2006)
Ashby, M.F., Johnson, K.: Materials and design: the art and science of material selec-tion
in product design. Published by Elsevier Ltd. (2009) ISBN: 978-1-85617-497-8
Ashby, M.F., Bréchet, Y., Salvo, L.: Sélection des matériaux et des procédés de mise en
œuvre. Presses polytechniques et universitaires romandes, CH-1015 Lausanne (2001)
ISBN 2-88074-473-3
Ferrer, I., Rios, J., Ciurana, J., Garcia-Romeu, M.L.: Methodology for capturing and forma-
lizing DFM Knowledge. Robotics and Computer-Integrated Manufacturing 26, 420–429
(2010)
Jahan, A., Ismail Md, Y., Mustapha, F., Mohd Sapuan, S.: Material selection based on or-
dinal data. Materials and Design 31, 3180–3187 (2010) ISSN: 0261-3069
Esawi, A.M.K., Ashby, M.F.: Cost estimates to guide preselection of processes. Materials
and Design 24, 605–616 (2003)
Study of the Aerodynamic Structure around an
Obstacle with Inclined Roof
Slah Driss, Zied Driss, and Imen Kallel Kammoun
Laboratory of Electro-Mechanic Systems (LASEM),

National School of Engineers of Sfax (ENIS), University of Sfax,
B.P. 1173, km 3.5 Soukra, 3038 Sfax, TUNISIA
Archs1@hotmail.com
Abstract. In this work, we are interested on the study of the aerodynamic struc-
ture around an obstacle with inclined roof for a Reynolds numbers equal to
Re=2666. The software "SolidWorks Flow Simulation" has been used to present
the local characteristics. The numerical model considered is based on the resolu-
tion of the Navier-Stokes equations in conjunction with the standard k-ε turbu-
lence model. These equations were solved by a finite volume discretization
method. The numerical model is validated with experimental results conducted on
an open wind tunnel.
Keywords: CFD, Modeling, Airflow, Obstacle, Inclined roof, Wind tunnel.
1 Introduction
The charactristics of the flows around building-shaped obstacles are different from
those in laminar boundary immersed in turbulent boundary layers. It is important
for the appropriate estimation of wind loading to investigate these characterestics
of flows observed under oncoming turbulence (Nozawa and Tamura, 2002; Ka-
wai, 1999). Computational simulations have an advantage to clarify the process
and mechanism of these localized phenomena. In recent years, several methods to
generate turbulent inflow data were proposed and some of them were used as in-
flow data (Kondo and Murakami,1997; Kataoka, and Mizuno,1999; Noda and
Nakayama, 1989). In this context, Nasrollahei et al. (2013) used a builder software
to study the thermal and cryogenic performance. The results of the research show
that the temperature of Shevaduns in summer is less than the average maximum
temperature and the average minimum temperature of outdoor, and the thermal
condition of the Shevadun space during the measurement is lower than the thermal
comfort limit which is specified in the standard. Ould Said et al. (2014) dedicated
to the numerical simulation of thermal convection in a two dimensional vertical
conical cylinder partially annular space. The governing equations of mass,

132 S. Driss, Z. Driss, and I.K. Kammoun
momentum and energy are solved using the CFD FLUENT code. Tominaga and
Stathopoulos (2013) reviewed current modeling techniques in computational fluid
dynamics (CFD) simulation of near-field pollutant dispersion in urban environ-
ments and discussed the findings to give insight into future applications. De Paepe
et al. (2013) simulated five different wind incidence angles using a turntable, in
order to quantify their effect on indoor air velocities. The responses in local air ve-
locities could largely be attributed to the relative position of the end walls of the
scale models orientated towards the wind. This crucial position allows the meas-
ured air velocity trends to be explained. The estimated airflow rates gradually de-
creased for larger wind incidence angles. Lim et al. (2009) presented a numerical
simulation of flow around a surface mounted cube placed in a turbulent boundary
layer which, although representing a typical wind environment, to match a series
of wind tunnel observations. The simulations were carried out at a Reynolds num-
ber, based on the velocity at the cube height, of 20,000. The results presented in-
clude detailed comparison between measurements and large eddy simulation
(LES) computations of both the inflow boundary layer and the flow field around
the cube including mean and fluctuating surface pressures. Melo et al. (2012) de-
veloped two Gaussian atmospheric dispersion models, AERMOD and CALPUFF.
Both incorporating the PRIME algorithm for plume rise and building downwash,
are intercompared and validated using wind tunnel data on odour dispersion
around a complex pig farm facility comprising of two attached buildings. The re-
sults show that concentrations predicted by AERMOD are in general higher than
those predicted by CALPUFF, especially regarding the maximum mean concen-
trations observed in the near field. Comparison of the model results with wind
tunnel data showed that both models adequately predict mean concentrations fur-
ther downwind from the facility. However, closer to the buildings, the models may
over-predict or under-predict concentrations by a factor of two, and in certain cas-
es even larger, depending on the conditions. Meslem et al. (2014) observed
changes in the prediction of local and global mean-flow quantities as a function of
the considered turbulence model and by the lack of consensus in the literature on
their performance to predict jet flows with significant three-dimensionality. The
study reveals that none of the turbulence models is able to predict well all jet cha-
racteristics in the same time.
According to these anteriors studies, it’s clear that the study of the aerodynamic
around the obstacle is very interesting. For thus, we have studied the aerodynamic
structure around an obstacle with inclined roof.
2 Geometrical System
The computational domain is shown in figure 1. It is defined by the interior vo-
lume of the wind tunnel blocked by two planes. The first one is in the tranquilliza-
tion chamber entry and the second one is in the exit of the diffuser. The test vein is
equipped by the considered inclined roof obstacle.
Study of the Aerodynamic Structure around an Obstacle with Inclined Roof 133
Parameters Value (mm)

H 153
h 118
l 82
Fig. 1 Geometrical arrangement
3 Numerical Results
3.1 Magnitude Velocity

Figure 2 presents the distribution of the magnitude velocity in the longitudinal
plane defined by X=0 mm. According to these results, it has been noted that the
velocity is weak in the inlet of the collector. It is indeed governed by the boundary
condition values of the inlet velocity. In this region, the velocity field is found to
be uniform and increases progressively downstream of the collector. At the test
vein, an important increase has been noted due to the reduction of the tunnel sec-
tion that causes the throttling of the flow. While the upper side of the obstacle is
characterized by the high velocity, a brutal drop is located behind the obstacle and
this is due to the deceleration of the velocity field while passing through the ob-
stacle. In the test vein, the velocity keeps increasing till the out of the test section.
Then, a decrease has been noted through the diffuser where the minimum velocity
values are recorded in the lateral walls of the diffuser.
Fig. 2 . Distribution of the average Velocity

3.2 Velocity Streamlines

Figure 3 presents the distribution of the velocity streamlines in the longitudinal
plane defined by X=0 mm. According to these results, it has been noted that the
velocity is weak in the inlet of the collector. It is indeed governed by the boundary
condition value of the inlet velocity. In this region the velocity is found to be uni-
form and increases progressively downstream of the collector. At the test vein, an
important increase has been noted due to the reduction of the tunnel section that
causes the throttling of the flow. While the upper side of the obstacle is characte-
rized by the high velocity, a brutal drop is located behind the obstacle. This is due
to the deceleration of the velocity field while passing by the obstacle. In the test
vein the velocity keeps increasing till the out of the test section. Then, a decrease
on the maximum values has been noted through the diffuser where the minimum
velocity values are recorded in the lateral walls of the diffuser. The flow circula-
tion appears in the dead zones where the velocity presents a weak value. In our case,
the flow circulations are located behind the obstacle and in the diffuser outlet.
Fig. 3 Distribution of the velocity streamlines
3.3 Static Pressure

Figure 4 presents the distribution of the static pressure in the longitudinal plane
defined by X=0 mm. According to these results, it has been noted that the static
pressure is on its maximum in the inlet of the collector. Besides, it has been ob-
served a depression above the obstacle. The static pressure continues decreasing
the way out of the test vein. A brutal drop of the pressure has been noted just
behind the obstacle.
Fig. 4 Distribution of the static pressure

Figure 5 presents the distribution of the dynamic pressure in the longitudinal plane
defined by X=0 mm According to these results, the dynamic pressure is found to
be weak in the collector inlet and increases gradually through the collector as long
as the tunnel section gets smaller. When it gets to the test section, the dynamic
pressure keeps increasing in the upstream of the obstacle. A compression zone is
recorded in the region located behind the obstacle and is developed through the
diffuser. The distribution of the dynamic pressure in the shows also a zone located
in the downside of the wind tunnel and a compression zone located in its upper
side. For example, at longitudinal plane X=0 mm, the maximum value of the dy-
namic pressure is equal to pd=0.43 Pa for the Reynolds number equal to Re= 2666.
Fig. 5 Distribution of the dynamic pressure

Figure 6 presents the distribution of the turbulent kinetic energy in the longitudinal
plane defined by X=0 mm. From these results, it has been noted that the turbulent
kinetic energy is found to be very weak in the first half of the wind tunnel in the
obstacle upstream. A wake characteristic of the maximum value of the turbulent
kinetic energy appears upstream of the obstacle. This wake starts in the obstacle
corner until the outlet of the diffuser. In these conditions, the maximum value of
the turbulent kinetic energy is equal to k=0.03 m2.s-2 for a Reynolds number equal
to Re=2666.
3.6 Turbulent Dissipation Rate

Figure 7 presents the distribution of the turbulent dissipation rate on the longitu-
dinal plans defined by X=0 mm. From these results, it has been noted that the tur-
bulent dissipation rate is found to be very weak in the first half of the wind tunnel
in the obstacle upstream. A wake characteristic of the maximum value of the tur-
bulent dissipation rate appears upstream of the obstacle. This wake starts in the
obstacle corner until the outlet of the diffuser. In these conditions, the maximum
value of the turbulent dissipation rate is equal to ε=0.04 W/Kg for a Reynolds
number equal to Re=2666.
Fig. 7 Distribution of the turbulent dissipation rate
3.7 Turbulent Viscosity

Figure 8 presents the distribution of the turbulent viscosity in the longitudinal
plane defined by X=0 mm. According to thes results; it’s clear that the viscosity is
at its minimum in the collector region but starts increasing after crossing the ob-
stacle. The wake characteristic of the maximum values is located in the outlet of
the diffuser. In these conditions the maximum value of the turbulent viscosity is
equal to µ t=0.005 Pa.s for a Reynolds number equal to Re=2666.
Fig. 8 Distribution of the turbulent viscosity
3.8 Vorticity
Figure 9 presents the distribution of the vorticity in the longitudinal plane defined by
X=0 mm. According to these results, the vorticity is at its minimum in the collector
region and starts increasing after crossing the obstacle. The greatest vorticity values
are reached after hitting the obstacle blade. In these conditions, the maximum value
of the vorticity is equal to 6.2 s-1 for a Reynolds number equal to Re=2666.
Fig. 9 Distribution of the vorticity

In this section, we are interested on the comparison of the numerical results with
the experimental results conducted in the LASEM laboratory using a wind tunnel
(Figure 10). The velocity profiles are chosen for points situated in the test section.
The considered planes are defined by Z=0 mm, Z=150 mm and Z=-150 mm. The
results for each plane are shown respectively in figure 11. For each transverse
plane, values are taken along the directions defined by X=0 mm. Near the obsta-
cle, it’s clear that the velocity value is very weak. Outside, the velocity has a
maximum value. For example, for the direction defined by Z=0 mm and X=0 mm,
the maximum value is equal to V=0.6 m.s-1 for the Reynolds number equal to
Re=2666. The comparison between the numerical and experimental velocity
values leads us to the conclusion that despite some unconformities, the values are
comparable. The numerical model seems to be able to predict the aerodynamic
characteristics of the air flow around the inclined roof obstacle.
Fig. 10 Wind tunnel equipped by an inclined roof obstacle
(a) Z=0 mm and X=0 mm
(b) Z=+150 mm and X=0 mm
Fig. 11 Velocity profiles

(c) Z=-150 mm and X=0 mm

Fig. 11 (continued)
5 Conclusion
In this paper, we have performed a numerical simulation to study the aerodynamic

characteristics around an obstacle with inclined roof. Numerical results, such as
velocity fields, pressure and turbulent characteristics are presented in the wind
tunnel, considered as a control volume. According to the obtained results, the cir-
culation zones of the velocity streamlines and the wakes characteristics of the
maximum values of the turbulent characteristics are located after hitting the obsta-
cle. Use of this knowledge will assist the design of packaged installations of the
wind rotors in the buildings. In the future, we propose to study the geometrical
effect of the obstacle.
References
De Melo, A.M.V., Santos, J.M., Mavroidis, I., Reis Junior, N.C.: Modelling of odour dis-
persion around a pig farm building complex using AERMOD and CALPUFF. Compari-
son with wind tunnel results, Building and Environment 56, 8–20 (2012)
De Paepe, M., Pieters, J.G., Cornelis, W.M., Gabriels, D., Merci, B., Demeyer, P.: Airflow
measurements in and around scale-model cattle barns in a wind tunnel: Effect of wind
incidence angle. Biosystems Engineering 115, 211–219 (2013)
Driss, Z., Bouzgarrou, G., Chtourou, W., Kchaou, H., Abid, M.S.: Computational studies of
the pitched blade turbines design effect on the stirred tank flow characteristics. European
Journal of Mechanics B/Fluids 29, 236–245 (2010)
Lim, H.C., Thomas, T.G., Castro, I.P.: Flow around a cube in a turbulent boundary layer:
LES and experiment. Journal of Wind Engineering and Industrial Aerodynamics 97,
96–109 (2009)
Meslem, A., Bode, F., Croitoru, C., Nastase, I.: Comparison of turbulence models in simu-
lating jet flow from a cross-shaped orifice. European Journal of Mechanics B/Fluids 44,
100–120 (2014)
Nozawa, K., Tamura, T.: Large eddy simulation of the flow around a low-rise building im-
mersed in a a rough-wall turbulent boundary layer. J. Wind Eng. Ind. Aerodyn. 90,
1151–1162 (2002)
Kawai, H.: Local peak pressure and conical vortex on building. In: Proceedings of the 10th
International Conference on Wind Engineering, pp. 1807–1812 (1999)
Kataoka, H., Mizuno, M.: Numerical flow computation around 3D square cylinder using in-
flow turbulence. J. Archit. Plan Environ. Eng., AIJ 523, 71–77 (1999)
Kondo, K., Murakami, S., Mochida, A.: Generation of velocity fluctuation for inflow boun-
dary condition of LES. J. Wind Eng. Ind. 67&68, 51–64 (1997)
Nasrollahei, N., Mahdavinejad, M., Hadiyanpour, M.: Studying the Thermal and Cryogenic
Performance of Shevadun in Native (Local) Buildings of Dezful Based on Modeling and
Environmental Measuring. American Journal of Energy Research 1(3), 45–53 (2013)
Noda, H., Nakayama, A.: Reproducibility of flow past two-dimensional rectangular cylind-
ers in a homogeneous turbulent flow by LES. J. Wind Eng. 89, 321–324 (2001)
Ould Said, B., Retiel, N., Bouguerra, E.H.: Numerical Simulation of Natural Convection in
a Vertical Conical Cylinder Partially Annular Space. American Journal of Energy Re-
search 2(2), 24–29 (2014)
Tominaga, Y., Stathopoulos, T.: CFD simulation of near-field pollutant dispersion in the
urban environment: A review of current modeling techniques. Atmospheric Environ-
ment 79, 716–730 (2013)
Contribution to Inverse Kinematic Modeling
of a Planar Continuum Robot Using a Particle
Swarm Optimization
Ammar Amouri1, Chawki Mahfoudi2, and Abdelouahab Zaatri1
1
Department of Mechanical Engineering, University1, Constantine, Algeria
{ammar_amouri,azaatri}@yahoo.fr
2
Department of Mechanical Engineering,
University Labri Ben M’ Hidi, Oum el Bouaghi, Algeria
mahfoudi.chawki@gmail.com
Abstract. According to the literature, research on modeling continuum robots is

focused on ways to develop the kinematic models, because of the lack of analyti-
cal models for these robots and the complexity of the problem which reside in the
coupling of operational variables and infinite of possible solutions for a desired
configuration. This paper presents a numerical approach for solving the inverse
kinematic model of a planar continuum robot (PCR), assuming that each section
of the manipulator is curved as a circular arc, with an inextensible central axis of
the structure. At first, this paper presents an inverse kinematic model solution for
one bending section, whereas the extreme points, of each section, used in calculat-
ing the inverse kinematic model for multi-sections is calculated numerically using
a particle swarm optimization (PSO) technique. Finally, Simulation examples of
this method are carried to validate the proposed approach.
Keywords: Planar continuum robot, Modeling, Inverse kinematic model, Optimi-

zation, Particle swarm optimization.
1 Introduction
An important research theme for study the continuum robots is the calculation of
kinematic models, where the traditional approaches to modeling, in which frames
are associated with each joint, are inappropriate for this case, because of the ab-
sence of discrete and rigid links in their architecture and the complexity of the
problem which reside in the coupling of operational variables. However, the hy-
per-redundant robotic systems, discrete links or continuum form cannot be control
by considering a finite number of degrees of freedom, thus admitting a reduced set
of physical solutions (Robinson and Davies 1999). Among contributions on the
inverse kinematic model of these structures, we can cite the model for one bending
section with a special configuration proposed by (Hannan and Walker 2003), to

142 A. Amouri, C. Mahfoudi, and A. Zaatri
validate the model; the authors used a specific software symbolic computation and
have simulated the kinematic behavior of the robot at two sections of Air-Octor
and Oct-Arm. The same authors studied the workspace and configuration space
describing the constraints due to limitations with respect to the lengths of robots.
(Escande et all. 2011), synthesized and validated an inverse kinematic model for
one bending section. The experimental results of this model show an accurate fol-
low-up identification of all existing uncertainties on the bench test and measure-
ment systems for both models and also in (Lakhal et all. 2014), the authors used
an optimization algorithm, based on a Sequential Quadratic Program (SQP). The
case of a multi-section is studied in (Neppalli et all. 2009), where the end-point of
each section is assumed to be known. The overall points can be chosen by solving
a system of inequalities. The authors did not validate experimentally the model but
they simulated the overall behavior using appropriate software. (Iqbal et all.
2009), presenting a complete inverse kinematic model for one bending section
based on the development realized by (Escande et all. 2011).This inverse kinemat-
ic model is formulated as an optimization problem with a cost function and con-
straints using the principle of interval analysis. The validation of this model has
done on a micro-robot and showed that the model was not sufficiently robust
against uncertainties of environment and inaccuracies of materials. The solution of
inverse kinematic model is posed also in terms of optimizing an objective function
under inequality constraints by (Bailly and Amirat 2005). The validation of this
model was made on a micro-robot dedicated to surgery. Finally, a state of the art
report in modeling these classes of robots is presented by (Robert et all. 2010).
The lack of analytical models for the kinematic of continuum robots, guide us to
use a powerful numerical method such as Particle Swarm Optimization (PSO).
The first idea based on swarm intelligence algorithm was introduced in 1995
(Kennedy and Eberhart 1995), who studied social relationships in a group to de-
velop new methods of calculation. This method has showed its capacity to solve
different optimization problems compared to other heuristics (Elbeltagi et all.
2005). The main contribution of this paper is the calculation of operational coor-
dinates of each section, by using an optimization method, under the constraints of
conserving the length of each section. The rest of the paper is organized as fol-
lows: Section 2 describes the bending section modeling. Section 3 shows the op-
timization technique for inverse kinematic modeling of a PCR. Finally, simulation
examples are performed to validate the proposed approach.
2 Inverse Kinematic Model for One Bending Section

One bending section of a PCR is modeled as an inextensible arc of circle, with one
end point o k −1 fixed to the origin of the reference frame R k −1 , the other end
point Pk located at anywhere in the workspace, where the first tangent of this
bending section is collinear with the y k −1 axis (physical constraints) and indivi-
dually controlled . This section of a PCR is parameterized by its arc length s k , its
curvature κ k , and its orientation θ k as shown in Fig.1.
Contribution to Inverse Kinematic Modeling of a Planar Continuum Robot 143
From the Fig.1 and using trigonometrically relations we have:
y k −1
Pk ( x k , y k )
yk
sk
xk
lth ,k
κk
δk θk
x k −1
o k −1
Fig. 1 Circle arc parameters
⎧δ k = a tan 2 ( y k , x k )
⎨ (1)
⎩ θ k = π − 2δ k
⎛θ ⎞
2sin ⎜ k ⎟
κk = ⎝ 2 ⎠ (2)
x k + y k2
2
θk
sk = (3)
κk
3 PSO and Inverse Kinematic Model of Multi-sections
3.1 PSO
Particle Swarm Optimization (PSO) is a biologically inspired computational
search and optimization method based on the social behaviors of some biological
organisms, especially the group's ability of some animal species to locate a desira-
ble position in the given area. It was proposed first by (Kennedy and Eberhart
1995) and its fast convergence compared to other Evolutionary Algorithms (Elbel-
tagi et all. 2005). Thus, using the simple rules of displacement (in the space of the
solutions), the particles can progressively converge to a global optimum. During
each iteration t, the velocity changes by applying equation (4) to each particle:
( ) (
v p i (t + 1) = ω v p i (t ) + c1 ρ1 Pi ,best (t ) − Pi (t ) + c 2 ρ 2 Pg ,best (t ) − Pi (t ) ) (4)
The position update is applied by equation (5) based on the new velocity and
the current position.
Pi (t + 1) = Pi (t ) + v Pi (t + 1) (5)
Where Pi (t ) , v pi (t ) represent the current position and the velocity of the

particle i at iteration t , respectively; ω is called the inertia weight, c1 and c 2
are weighting factors, also called the cognitive and social parameters, respectively;
and ρ1 , ρ2 are random variables uniformly distributed within interval [0, 1]. The
basic algorithm is as follows, from which we can obtain the good solution of our
problem:
1) Initialize each particle of the swarm, with random values for position in
the search space.
2) For each particle, evaluate of the cost function.
3) Compare the value obtained by the cost function from particle i , with the
value of Pi ,best . If the current value is better than Pi ,best , than set Pi ,best
equal to the current value.
4) If the value Pi ,best is better in the swarm, then Pg ,best = Pi ,best .
5) Change the velocity and position of the particle using equations (4) and
(5), respectively.
6) If the maximum number of iterations or the ending condition is not
achieved, return to step 2.
3.2 Problem Formulation

In 2D research space, the particle i can be represented by a two-dimensional posi-
{ }
T
tion vector Pi (t ) = Px ,i Py ,i and the velocity vec-
{ }
T
tor v Pi (t ) = v Px ,i v Py ,i . The best position visited by the particle is denoted
{ }
T
Pi ,best (t ) = Px ,i ,best Py ,i ,best and the best position found by the swarm
{ }
T
is Pg ,best (t ) = Px , g ,best Py , g ,best .
Given the desired position Pn x n∗ , y n∗ ( ) and following the above assumptions,

the operational Cartesian coordinates Pk ( x k∗ , y k∗ ) of each section, with respect
to the reference frame R 0 , located on the virtual central axis, are calculated nu-
merically by the PSO method, where the cost function is given by:
n
(
F lth ,k , l r ,k =) ∑ f k ( lth ,k , l r ,k ) (6)
k =1
With:
( ) 1
( )
2
f lth ,k , l r ,k = lth ,k − l r ,k (7)
2
Such that:
⎧ θk
⎪lth ,k = ( x k − x k −1 ) + ( y k − y k −1 ) sin (θ
2 2
⎨ k ) (8)
⎪ θ = a tan 2 ( ( y − y ) , ( x − x ) )
⎩ k k k −1 k k −1
Where lth ,k , l r ,k : the real length and the theoretical length calculated using eq-
uation (8) of a bending section k respectively and where n is the number of
bending section.
The intermediate operational coordinates x k∗ , y k∗ ( ) of each section, are calcu-
lated by solving the following optimization problem:
⎧ min
⎪ x ∗ , y ∗ ; k =1,…,n (
F x k∗ , y k∗ )
⎪ k k
⎨ under the constraints g (9)
⎪ j
⎪ g j ≤ 0; j = 0,1,… , m
⎩
Where g j the constraints are could be used for example: obstacle avoidance,
taking objects, energy minimization, etc. And m is the number of added con-
straints.
Initially, the population of particles is located with random positions in the
problem space. During each iteration t, each particle changes their velocity ac-
cording to equation (4) and move to a new position using (5). Once the displace-
ment of the particles carried out, the new positions are evaluated using equation
(6) and the vector Pi ,best and Pg ,best are determined. The algorithm stops if
maximum number of iterations is achieved or any other stopping criteria are satis-
fied. The solution of this algorithm represents the intermediate Cartesian coordi-
nate researched with respect to the reference frame R 0 .
3.3 Inverse Kinematic Model of Multi-sections

Knowing the intermediate Cartesian coordinates of each bending section, the in-
verse kinematic model presented in the previous section can be iteratively applied
to several sections connecting in series to model a PCR of n sections. Where each
found point (calculated by optimization method PSO) will be transformed to a lo-

cal reference frame of each section, by the following equation:
T
⎡ x k , y k , 0 ⎤ = T 0k −1Pk0 (10)
⎣ ⎦
Such that:
T k0 = T10T 21 "T kk −1 (11)
Where T kk −1 ( k = 1, 2… , n ) , is a homogeneous transformation matrix, contains

both rotational and translational terms, defines the origin of the reference
frame R k in the reference frame R k −1 and n is the total number of the bending
sections.
4 Simulation Results
The effectiveness of proposed approach is verified by some simulations through a
personal computer with i3 3.30 GHz processor. The first example was performed
for a planar robot consists of two bending sections for the following settings of
⎛ 1 ⎞
the trajectory in operational space: ⎜ x 2∗ = t 2 , y 2∗ = 200 − 4 t ⎟ The trajectory
⎝ 2 ⎠
has been accomplished in 10 sec of time duration, corresponding to 51 positions
captures. Fig. 2 represents the configurations (solutions) for a desired point of the
given trajectory, while respecting the imposed assumptions. The execution time of
the optimization process for each point found is shown in Fig. 3. Fig. 4 shows the
convergence of cost function. In this simulation, the tolerance error is fixed to
10−6 while the time execution is equal to 23.91 msec.
For the second case, knowing the general movement of a superior mobile plat-
form generated by the trajectory in operational space ( x 2∗ = 0, y 2∗ = 200 − 10t )
where t is time variable varying from 0 to 10 sec, with a step equal to 2 sec.
Fig.5.a represents the possible configurations for a desired point of the given tra-
jectory. The problem of infinite solutions is solved by adding some other con-
straints. To illustrate this, two cases be imposed. The first case uses the constraint
of g1 = θ1 〈 0 and the second that uses g 2 = θ1 〉 0 , as shown in Fig.5.b and 5.c,
respectively.
(
In the third example, the trajectory of equation: x 2∗ = 0, y 2∗ = 200 − 10 t ) is
used as input in the proposed model and lengths variation is determined. This tra-
jectory has been accomplished in 10 sec of time duration, corresponding to 121
position captures. Fig. 6 shows the length profiles obtained by unconstrained
( )
model g j = 0 . Fig. 7 show the length profiles, where the constraint is given
by g1 = θ1 〈 0 .
200
150
Y ( mm ) 100
50
0
-40 -20 0 20 40 60
X ( mm)
Fig. 2 Virtual axis of the planar robot
30
Time ( msec )
20
10
0
0 2 4 6 8 10
Found Position
Fig. 3 Execution time of the optimization algorithm
200
Cost Function
150
100
50
0
0 100 200 300 400 500
No. of Iterations
Fig. 4 Convergence of the cost function
200 200 200
150 150 150
Y ( mm )
Y ( mm )
Y ( mm )
100 100 100
50 50 50
0 0 0
-100 -50 0 50 -100 -50 0 50 -50 0 50 100
X ( mm ) X ( mm ) X ( mm )
(a) Without constraint (b) With: g 1 = θ1 〈 0 (c) With: g 2 = θ1 〉 0
Fig. 5 Virtual axis of the planar robot
First bending section

110
L11
len g th ( m m )
L21
100
90
0 1 2 3 4 5 6 7 8 9
Time (sec )
Second bending section
120
len g th ( m m )
L12
100
L22
80
0 1 2 3 4 5 6 7 8 9
Time ( sec )
Fig. 6 Lengths variation without constraint

First bending section

len g th ( m m ) 110
L11
100
L21
90
0 1 2 3 4 5 6 7 8 9
Time ( sec )
Second bending section
120
len g th ( m m )
L12
100
L22
80
0 1 2 3 4 5 6 7 8 9
Time ( sec )
Fig. 7 Lengths variation with constraint
5 Conclusion
This article presents a new approach for the synthesis of the inverse kinematic
model for a planar continuum robot. The mathematical formulations defining the
kinematic of a bending section are given. This approach allows us to calculate the
Cartesian coordinates of the origin of intermediate platforms using the PSO opti-
mization method, where the search space that is chosen arbitrarily. Future work is
to improve the execution time, one taking the space of research in the workspace
of each section, in order to find a solution close to real time and the generalization
of the approach in the three-dimensional.
References
Robinson, G., Davies, J.B.C.: Continuum Robots: A State of the Art. In: IEEE International
Conference on Robotics and Automation, Detroit, Michigan, pp. 2849–2854 (1999)
Hannan, M., Walker, I.D.: Kinematics and the Implementation of an Elephant’s Trunk Ma-
nipulator and Other Continuum Style Robots. Journal of Robotic Systems 20(2), 45–63
(2003)
Escande, C., Pathak, P.M., Merzouki, R., Coelen, V.: Modeling of Multi-section Bionic
Manipulator: application to Robotino XT. In: 2011 IEEE International Conference on
Robotics and Biomimetics (ROBIO), Phuket, Thailand, pp. 92–97 (2011)
Lakhal, O., Chibani, A., Escande, C., Merzouki, R.: Contribution to Inverse Kinematic
Modeling of a class of Continuum Bionic Handling Arm. In: 2014 IEEE/ASME, Intern-
tional Conference on Advanced Intelligent Mechatronics (AIM), Besançon, France
(2014)
Neppalli, S., Csencsits, M.A., Jones, B.A., Walker, I.D.: Closed-Form Inverse Kinematics
for Continuum Manipulators. Advanced Robotics 23, 2077–2091 (2009)
Iqbal, S., Mohammed, S., Amirat, Y.: A guaranteed approach for kinematic analysis of con-
tinuum robot based catheter. In: IEEE International Conference on Robotics and Bio-
mimetics (ROBIO), Guilin, China, December 19-23, pp. 1573–1578 (2009)
Bailly, Y., Amirat, Y.: Modeling and control of a hybrid continuum active catheter for aor-
tic aneurysm treatment. In: Proc. of the International Conference on Robotic and Auto-
mation, Barcelona, Spain, pp. 936–941 (2005)
Webster III, R.J., Jones, B.A.: Design and Kinematic Modeling of Constant Cur-vature
Continuum Robots: A Review. The International Journal of Robotics Research 29(13),
1661–1683 (2010)
Kennedy, J., Eberhart, R.C.: Particle Swarm Optimization. In: Proceeding of the IEEE In-
ternational Conference on Neural Networks IV, pp. 1942–1948 (1995)
Elbeltagi, E., Hegazy, T., Grierson, D.: Comparison among five evolutionary-based optimi-
zation algorithms. Advanced Engineering Informatics 19(1), 43–53 (2005)
Experimental and Numerical Study of Beams
under Low-Velocity Impact
Mahmoud Neder and Abdelhamid Miloudi
Advanced Mechanics Laboratory, USTHB, Algiers, Algeria

nedmah_2000@yahoo.fr, amiloudi@usthb.dz
Abstract. This paper is an experimental and numerical study of the low velocity
impact on beams. It is shown that the numerical model adopted, for the impact
force history, produces numerical results in good agreement with experimental da-
ta. The analysis establishes that the contact point position is insignificant on the
main impact and rebounds are closer in time when we approach the clamped edge.
The increasing of the incident energy (mass or celerity of the striker) has the effect
of amplifying the impact force without changing its duration in the case of the ve-
locity and with increasing the amplitude of the impact force in the case of the
mass. However elevated hardness of the contact aggravates the impact effects by
increasing the magnitude and decreasing the contact force duration.
Keywords: Impacted beam, striker, shock.
1 Introduction
The study of the structures behaviour under impact remains an area of interest to
many researchers in recent years because of the complexity of modeling these
structures and the phenomena occurring during impact. Determining the impact
force is very difficult. Knowledge of the evolution of the impact force is necessary
to study the behaviour of structures in shock. Several researches has been done us-
ing numerical approaches or experimental studies but still unable to properly
define the exact evolution of the impact force.
Fig. 1 Transversal impact at the center of a rectangular plate

152 M. Neder and A. Miloudi
Several modelling of the impact have been adopted, we can note those consist
in representing the impact globally by an energy balance. The main advantage of
this type this presentation type, is the fact that the input parameters are parameters
characteristic of the impact: incident speed, the mass of the striker , geome-
tric and mechanical properties of the plate and impactor. In its work, Greszczuk
[1] modeled the impact by summing the contact energy than bending Shi-
vakumar [2] improved the earlier model by matching the energy of fexion, the
energy of shear and introducing the energy of membrane in the case of
thin plates. Thus the energy balance becomes:
1 2 (1)
2
Caprino and al. [3] presented the impact by a discrete mass-spring model ha-
ving a single degree of freedom, where the spring stiffness corresponds to the
static stiffness of the plate. A good correlation is observed up to 2.5 / speed,
between the calculated by the model and the measured.
In this model type, the effects of contact and plate inertia are not considered,
hence the usage limit for higher speeds. Sjoblom [4] then suggested a model with
two degrees of freedom that allows the integration of non-linearity in the contact.
With this model, it is possible to uncouple the rigidity of the contact plate. Never-
theless, the effects of shear localized around the point of impact of the large, for
thick plates or for very high speeds, are not taken into account. Therefore Lal [5]
developed a model with three , and representing rigidities punching,
bending and shear springs respectively. This model is valid in the case of small
displacements. Shivakumar and al. [2] then took into account the membranes ef-
fects using mass models-springs of one or two degrees of freedom with four
springs stiffness , , and . The indices , , and respectively
represent the membrane, shear, and bending of the plate contact with the striker.
Fig. 2 Mass-springs models: (a) one degree of freedom, Caprino and al. [3], (b) two degrees
of freedom model, Shivakumar and al. [2]
In this model, the two masses and are connected by a spring represen-
ting the law wireless loading-penetration of the contact. The strength properties-
transverse displacement are represented by the combination of the flexural
springs, and shear membrane. A previous study [6] on the plate free vibrations
with a concentrated mass attached to the plate center showed that the effective
mass, contributing to the plate inertia effect is a quarter of the total plate mass. In
the present document, we present an experimental study of the low-speed impact
Experimental and Numerical Study of Beams under Low-Velocity Impact 153
of a projectile ball end on homogeneous beams. Results will be compared with

those obtained by a numerical method based on modal decomposition and mode-
ling of the impact by the Hertz contact low. Subsequently we present the influence
of different impact parameters on the impact force history.
2 Formulations of Contact-Impact of Beams
To treat the impact problem, normally the contact force method and energy ba-
lance method are used. Among various contact force methods, Hertzian method is
the simplest in terms of formulation by using Hertzian’s contact law or its modifi-
cation. An homogeneous beam subjected to bending vibration is governed by the
equation of motion follows:
, ,
, (2)
where and are respectively the area and the moment of inertia of the cross
section, and are the density and Young's modulus of the beam.
The impact is caused by a spherical striker of mass end radius animated
by an incident velocity , as shown in fig. 1.
Fig. 3 The contact discretization
On contact with the beam, the striker will be modeled by a mass attached to
by non-linear beam spring cf. fig. 3, describing the law of Hertzian contact,
given by:
/ (3)
where is the indentation of the sphere in the beam and the contact stiffness
is given by:
4√
(4)
3
where is the striker radius and the index is related to the striker.
The dynamic balance of the spherical striker and the beam is given respectively
by the equation of dynamic equilibrium of the projectile, driven by an incident
velocity upon contact and initial conditions are written:
0
(5-a)
0 0 0
Those of the beam, initially at rest, are written:
, ,
. (5-b)
,0 0 ,0 0
is the displacement of the projectile, is the contact force of Hertz eq. (3)
with .
is the deflection of the beam at the impact point.
is the Dirac pulse which introduces a concentrated force equation of
the beam.
These two coupled equations (5-a) and (5-b) are non-linear differential equa-
tions which can be solved by a numeric integration method (Newmark integration
scheme) until 0.
3 Experimental Study
The experimental device consists of two parts: structural and acquisition parts.
1. The structural part consists in turn of two compartments:
• The compartment for positioning of the structure, subject of study, which
allows for several combinations of boundary conditions,
• The fund implementation force that is a pendulum which, by the law of po-
tential energy to kinetic energy conversion, to estimate the impact speed.
2. The acquisition part allows to increase the signal strength and the response of
the structure. It consists of:
• Acceleration sensor "accelerometer" piezoelectric brand Wilcoxon-
Research; model 736,
• A hammer pulse "modal hammer" brand Endvico model: 2302-10,
• A data acquisition card 4-input, National Instrument Model: NI9234,
• Acquisition and data processing Labview software.
Fig. 4 Experimental device

(1): modal hammer, (2): acccelerometer, (3): beam, (4): acquisition card,
(5):PC+data acquisition and analysis software.
The structure studied is an homogeneous beam dimension. The

hammer will strike the structure at a distance from the right end, the impact
velocity is calculated from the horizontal distance " " that initially separates the
hammer and the beam by the formula:
2 (6)
4 Results
This study concerns an homogeneous steel beam and aluminum strucked by a

spherical striker (pendulum hammer). The characteristics of the plate are given
below in Table 1:
Table 1 Mechanical characteristics of plates
Parameters Steel Aluminium

Young’s Modulus E (N/m2) 2.1 x 1011 0.69 x 1011
Masse volumique ρ (Kg/m3) 7800 2700
Coefficient de Poisson 0.3 0.33
Largeur b (m) 0.04 0.0265
Epaisseur h (m) 0.003 0.0065
4.1 Impact Force Evolution

The impact test was performed by a ball end projectile weight 0.375 (it
is the rod-hammer whose center of gravity coincides with the impact point). The
spherical aluminum end of radius 0.0095 , strike the clamped-clamped
steel beam of length 0.66 , in its center with an incident velocity
0.327 / .
Fig. 5 Comparison between calculated and measured impact force

Fig. 5, shows a comparison between the measured force and that determined by
a numerical approach. It can be seen that the force calculated numerically reached
it's peak faster than the measurement one. It may be caused by the boudaries con-
ditions in the experimental study whitch it is pratically impossible to realise per-
fect clamped edges. Also we note that the main impact is generally followed by a
less important amplitude rebounds.
4.2 Influence of the Impact Force Position

Fig. 6 and 7 showed that the first impact is pratically identical in amplitude and
duration. However, we can note that, far from the clamped edges, the rebounds are
more spaced in time.
Fig. 6 Influence of the impact force position: steel clamped-clamped beam with 0.66m
length
Fig. 7 Influence of the impact force position: aluminum cantilever beam with 0.5m length
4.3 Influence of the Incident Energy

To study the influence of energy on impact force, we took the clamped-clamped
steel beam with the same impact parameters. The strike had been in the beam center.
4.3.1 Variation of the Striker Mass
We have retained three different masses for the same incidental velocity impact.
Fig 8. Show the influence of the masses on impact force evolution.
Fig. 8 Influence of the striker mass; , /
4.3.2 Variation of the Incident Velocity
We have retained four different masses for the same incidental velocity impact.
Fig 8. Show the influence of the masses on impact force evolution.
Fig. 9 Influence of the incident velocity; ,

Fig. 8 and 9 showed that when the incident energy is more important, it has the
effect of increasing the intensity of the impact force. The duration of the impact is
the same if the velocity changes but increase with increasing the mass of the pro-
jectile.
4.4 Influence of the Contact Stiffness

On an aluminum cantilever beam, the impact of different projectiles showed that
when the striker hardness is increased, it has the effect of increasing the force mag-
nitude and decrease the duration of the impact making the shock most damaging.
Fig. 10 Influence of the striker material
5 Conclusion
An experimental and a numerical study were conducted to estimate the history of

a spherical end projectile impact on a homogeneous beam. A comparison of the
results with those calculated by a numerical method, based on modal decomposi-
tion and modeling of the impact by the Hertz contact law was made. This study
showed that the main impact is usually followed by a series of smaller intensity
rebounds. The study of the influence of various parameters of the impact allowed
us to make the following observations: the influence of the contact point is insig-
nificant on the main impact and rebounds are closer in time approaching edges.
Increasing the mass or velocity of the projectile (incident energy) has the effect of
amplifying the impact force without changing its duration in the case of speed and
with increased duration for the case of mass. The high hardness of the contact ag-
gravates the effects of the impact by increasing the magnitude and decreasing the
duration of the contact force.
References
1. Greszczuk, L.B.: Damage in Composite Materials due to Low Velocity Impact. In:
Zukas, Z.A., et al. (eds.) Impact Dynamics, pp. 55–94. Wiley editions, New York
(1982)
2. Shivakumar, K.N., Elber, W., Illg, W.: Prediction of Impact Force and Duration due to
Low Velocity Impact on Circular Composite Laminates. Journal of Applied Mechan-
ics 52, 674–680 (1985)
3. Caprino, G., Crivelli-Visconti, I., Di Ilio, A.: Elastic Behaviour of Composite Struc-
tures Under Low Velocity Impact. Composites 15(3), 213–234 (1984)
4. Sjoblom, P.: Simple Design Approach Against Low Velocity Impact Damage. In: Pro-
ceedings of the 32nd SAMPE Symposium, Anaheim, CA, pp. 30–52 (1987)
5. Lal, K.M.: Residual Strength Assessment of Low Velocity Impact Damage of Graphite
Epoxy Laminates. Journal Reinf. Plast. Comp. 2, 226–238 (1984)
6. Leissa, A.W.: Vibrations of Plates, NASA SP-160 (1969)
7. Blevins, R.D.: Formulas for Natural Frequency and Mode Shape. R.E. Kriegen Pub-
lishing (1984)
8. Miloudi, A.: M., Comportement Dynamique de Plaques Impactées. In: 8ème Congrès
de Mécanique, El Jadida, Maroc., Avril 17-20, pp. 17–20 (2007)
9. Wali, M., Abdennadher, M., Fakhfakh, T., Haddar M.: Comportement Dynami-que des
Plaques Sandwichs Impactées: Estimation du Coefficient de Restitution. In: 3ème
Congrès de Conception et Modélisation des Systèmes Mécaniques CMSM 2009,
Hammamet, Tunisie, Mars 16-18 (2009)
10. Neder, M., Miloudi, A.: Dynamique des plaques sandwichs sous impact à faible vi-
tesse: Approche numérique. In: 5ème Congrès de Conception et Modélisation des
Systèmes Mécaniques, CMSM 2013, Tunisie (2013)
Modeling of Cutting Forces and Roughness
by the Response Surface Method (RS)
Fayçal Bentaleb1 and Idriss Amara2

1
Department of Mechanical Engineering, Faculty of applied Science,
University of Ouargla, Algeria.
bentaleb_fmec@yahoo.fr
2
Laboratory of Transportation Engineering and Environment, Faculty of Technology,
University 1 Constantine, Algeria.
a.idriss.2011@gmail.com
Abstract. Longitudinal turning results in a cutting force whose decomposition in

three preferred directions can be the basis for the definition of cutting forces for all
machining operations.
The work presented in this paper aims to modeled phenomena that occur when
shooting steel Fk 20 Mn Cr 5 with a metal coated carbide tool P40.
The main objectives focused on the delineation of the area of the shooting and
to study the relationship between the cutting parameters (cutting speed, feed and
depth of cut) and the response variables cutting forces and roughness of the ma-
chined parts through the response surface methodology (RSM). The resulting
models are types: quadratic, linear, exponential, Gilbert and genetic algorithm.
Keywords: Cutting Forces, Roughness, Cut Settings, Response surface method

RSM, Genetic Algorithm.
1 Introduction
Metal cutting, machining processes whose origin is very old, has expanded greatly
with the mechanization of increasingly sophisticated manufacturing facilities. De-
spite the emergence of new technologies in recent years, it remains today one of
the methods most widely used (Bouacha 2011).
Machining by cutting means all techniques to obtain a surface by removing ma-
terial with a sharp tool. This ancient technique is often called cutting expensive in
so far as the shaping of the part causes a noble processing waste material (chip).
However, it remains an important manufacturing technique and answered. In shap-
ing metal by cutting, lathing, in our study, alone accounts for 33% of the area of
machining by chip removal (CETIM 2000).
Metal cutting is strongly coupled thermo mechanical process in which plastic
deformation, heat and friction phenomena plays a critical role in terms of wear
(Gautier 2004).

162 F. Bentaleb and I. Amara
These occurring in a very small region around the tip of the tool, the phenome-
na related to tribology of cutting are difficult to study and model [Gokkaya and
Nalbant ).
The objective of this study is to examine the behavior do matter-torque tool when
shooting Dry Steel FK 20 Mn Cr 5 with a metal coated carbide tool P40 Brand
Sandvik (reference TPMN 16 03 08).
The experiment was done as follows:
• The tests are performed on a lathe roughing and model Maximat Emco V13
thread.
• Measurement of cutting forces during machining consists of an acquisition
channel consists of a dynamometer Kistler 9257 B and a charge amplifier Kistler
5019 B three channels.
To study the criteria roughness Ra, Rt and Rz, we used a roughness (2D) Mitu-
toyo, with printer profile roughness. The experimenter, whatever their field of
study, is still facing the problem difficult for the optimal organization of his es-
says. How to get the right information as soon as possible and for the least cost?
(Remadna 2001).
The use of experimental designs, especially Taguchi orthogonal tables (Taguchi
and konishi 1987) allows the determination of the influence of the simultaneous
effects and interaction of operating parameters on the cutting forces and rough-
ness.
In this work, the standard orthogonal table L27 (313) Taguchi is suitable as an
experimental plane. Parameter levels were chosen as recommended by the cutting
tool manufacturer intervals.
The parameters to be studied and the allocation of the respective levels are
shown in Table 1.
Table 1 Values cutting parameters and their levels
Cutting speed Feed Depth of cut

Level
Vc(m / min) f (mm/rd) ap (mm)
1 85 0,08 0,1
2 110 0,12 0,15
3 155 0,16 0,2
Modeling of Cutting Forces and Roughness by the Response Surface Method (RS) 163
3 Resultats
3.1 Cutting Forces

The cutting forces results obtained during the tests are shown in figure 1.
Fig. 1 Variation of cutting forces based testing
From this figure, we see that the lowest value of cutting forces obtained with
different combinations of the parameters of cutting regime that is provided with
Test No19 (Vc = 155 m / min, f = 0, 08 mm / rd and ap = 0.1 mm).
In this section we will determine the mathematical models of the cutting forces.
These models highlight the relationship between the elements of the regime cut-
ting (cutting speed, feed and depth of cut) and responses studied (cutting force).
To achieve these objectives, several methods and models are available.
3.1.1 Quadratic Model (Q. M)
Initial analysis of responses from the response surface method RSM includes all
parameters and their interactions. Models are reduced by eliminating the terms
that have no significant effect on responses.
By this process of elimination back end quadratic models of the equation of
responses in terms of actual factors are presented as follows:
2
Fa=343 – 2 * Vc – 2130 * f - 249 * ap + 9716 *f + 300 * ap2 +
Vc * ap + 1096 * f * ap (1)
Fc = 657 – 3 * Vc – 5136 * f – 806 * ap + 21243 * f2 + 753 * ap2 +
2 * Vc * ap + 3668 * f * ap (2)
Fp = 842 – 4 * Vc – 5033 * f – 743 * ap + 18663 * f2 + 1357 *
ap2 + 3 * Vc * ap + 4445 * f * ap (3)
3.1.2 Linear Model (M. L)
Fa = 13,98 - 0,17 * Vc + 484,11 * f + 223,63 * ap (4)

Fc = - 59,16 - 0,26 * Vc + 1019,33 * f + 390,49 * ap (5)
Fp = 100,38 - 0,67 * Vc + 733,32 * f + 563,95 * ap (6)
3.1.3 Exponential Model
Fa = e15,102 * Vc 0,654 * ap - 3,216 * f ln (f 2,403) * apln (ap 0,301) * Vc ln (ap 0,79) (7)
Fa = e41,02 * Vc-10,68 * f 9,19 * ap- 3,59 * f ln (f 1,99)

* apln (ap 0,31) * Vc ln (ap 0,76)
* Vcln (Vc0,76) (8)
Fa = e39,592 * Vc-9,441 * f 3,939 * ap- 2,592 * f ln (f 0,794)

*
apln (ap 0,336) * Vc ln (Vc 0,409) * fln (ap0,138) (9)
3.1.4 Gilbert Model
Fa = e 8,767 * Vc -0,273 * f 0,273 * ap 0,609 (10)
Fc = e 8,164 * Vc -0,081 * f 0,293 * ap 0,402 (11)
Fp = e 7,243 * Vc -0,058 * f 0,273 * ap 0,639 (12)
3.1.5 Genetic Algorithm
Fa = e 7,764 * Vc 0,317 * f 6,79 * ap -3,287 * f ln(f 1,031) * ap ln(ap 0,323) *

Vc ln(ap 0,565) (13)
Fc = e 11,836 * Vc 0,43 * f 10,171 * ap -3,23 * f ln(f 1,754) * ap ln(ap 0,5) *

Vc ln(ap 0,662) (14)
Fp = e 11,255 * Vc -0,116 * f 4,659 * ap 0,295 * f ln(f 0,595) * ap ln(ap 0,542) *

Vc ln(ap 0,359) (15)
The comparison between the experimental values of the cutting forces and
those estimated by the models obtained by the response surface methodology
RSM is presented graphically in Figure 2.
(a) (b)
(c)
Fig. 2 Curves of experimental efforts and estimated cut: (a) F, (b) Fc and (c) Fp.
3.2 Roughness
Results of the criteria of roughness obtained during the tests are shown in figure 3.
Fig. 3 Change in roughness (Ra, Rt and Rz) based on the parameters of cutting regime
In this section we will determine the mathematical models of the criteria chosen
roughness. These models highlight the relationship between the elements of the
regime cutting (cutting speed, feed per revolution and depth of cut) and responses
studied (roughness of the machined surfaces). For a number of methods and mod-
els are proposed.
3.2.1 Quadratic Model (Q. M)
By process of elimination back end quadratic models of response equation in

terms of actual factors are presented as follows:
Ra = 1,54 - 0,02 * Vc + 11,91 * f – 31,18 * f2 – 0,02 *Vc* f (16)

Rt = 6,72 - 0,1 * Vc + 83,52 * f – 204,83 * f2 – 0,1 *Vc* f (17)
Rz = 2,95 - 0,03 * Vc + 37,57 * f – 74,94 * f2 – 0,03 *Vc* f (18)
3.2.2 Linear Model (L. M)
The analysis of responses obtained from RSM includes only linear terms. The
final equation linear response models in terms of actual factors were as follows:
Ra = -2,1479 - 0,0018 * Vc + 34,3606 * f + 0,0204 *ap (19)
Rt = 1,5255 - 0,0011 * Vc + 18,31106 * f + 0,2913 *ap (20)
Rz = 3,0169 - 0,0108 * Vc + 7,2217 * f + 0,5834 *ap (21)
3.2.3 Exponential Model
In order to explain the variation in cutting forces based on the factors identified in
the study (Vc, f and ap) in an exponential form, a linear regression (multiple) can
be used.
opRa = e17,743 * Vc -6,485 * f -2,163 * ap - 0,529 * Vc ln(Vc 0,806) *f ln (f -1,013) *

Vcln (f 0,177) * Vc ln (ap 0,063) (22)
Rt = e20,325 * Vc -6,953 * f -2,784 * ap - 0,262 * Vc ln(Vc 0,849) *f ln (f -1,007) *

Vcln (f 0,257) * Vc ln (ap -0,115) (23)
Rz = e10,181 * Vc -2,664 * f -2,352 * Vc ln(Vc 0,478) *f ln (f -0,518) * Vcln (f 0,604) (24)
3.2.4 Gilbert Model
Ra = e 3,2533 * Vc -0,5453 * f 0,9105 * ap 0,0233 (25)
Rt = e 5,2424 * Vc -0,5923 * f 0,8945 * ap 0,0489 (26)
Rt = e 6,1698 * Vc -0,8777 * f 0,7597 * ap 0,0755 (27)
3.2.5 Genetic Algorithm
Ra = e 1,4361 * Vc -1,2666 * f -3,0991 * ap -1,4891 * Vc ln(Vc 0,1584) * f ln(f -0,7301)

* Vc ln(f 0,2421) * Vc ln(ap 0,3436) (28)
Rt = e -0,7641 * Vc -0,4096 * f -5,7115 * ap -0,1255 * Vc ln(Vc 0,0199) * f ln(f -1,0136)

* Vc ln(f 0,3009) * Vc ln(ap 0,0252) (29)
Rt = e 5,4966 * Vc -1,6171 * f -2,9041 * ap -0,3551 * Vc ln(Vc 0,1815) * f ln(f -0,3689)

* Vc ln(f 0,4798) * Vc ln(ap 0,0839) (30)
The comparison between the experimental values of the different roughness pa-
rameters Ra, Rt and Rz and those estimated by the models obtained by the
response surface methodology is presented graphically in Figure 4, although qua-
dratic and exponential models give very good results. It was found that clearly
those following the methodology of genetic algorithms are better.
(a) (b)
(c)
Fig. 4 Curves of different criteria of experimental and estimated roughness: (a) Ra, (b) and
Rt (c) Rz
4 Conclusion
The objective is reached because optimizing a response or finding a compromise

between several answers is to define within the experimental range of adjustment
factors to best meet the requirements set out in terms of response. The choose
from three levels for each variable is required by this design to explore the region
of the response surface near the optimum. Optimization of responses by RSM
helps to identify the combination of parameters input variables that optimize a
single answer or set of answers.
Although the models or the terms of the interactions are present give very good
results. It was found that clearly models from the genetic algorithm methodology
give excellent results.
References
Journal article only by DOI
Taguchi, G., Konishi, S.: Taguchi metods, orthogonal arrays and lineair graphs, tools for
quality engineering. American supplier institute (1987)
Gokkaya, H., Nalbant, M.: The effects of cutting tool geometry and processing parameters
on the surface of AISI 1030 steel of roughness. Materials and Design
Book
Documentation CETIM, hard turning and high-speed machining (HSM), a technical center
for mechanical industries (AAJ), Report, p. 86 (2000)
Dissertation
Bouacha, K.: Behavior of tool-material pair when machining hard materials. PhD thesis,
defended in January 2011 in the University of May 8, 1945, Guelma-Algeria
Gautier, L.: Study of damage carbide tools WC-CO by the characterization of the mechan-
isms chip tool interface application to dry machining of aerospace aluminum alloy
AA2024 T351. PhD thesis submitted in December 2004 to the National Higher School
of Arts and Crafts Centre in Bordeaux-France (2004)
Remadna, M.: The system behavior in machining hard turning. Application to the case of a
hardened steel machined with CBN. Doctoral thesis submitted in 2001 at INSA Lyon
France (2001)
Analysis of Inelastic Behavior of Amorphous
Polymers
Nourdine Ouali and Ali Ahmed Benyahia
Laboratoire de Mécanique Avancée. Université des Sciences et de la Technologie Houari

Boumediene. B.P. 32 El-Alia, 16111 Bab-Ezzouar, Alger, Algérie
nourouali@yahoo.fr
Abstract. A physically constitutive model based on the concepts of quasipoint de-

fects and hierarchically constrained molecular dynamics is applied to describe the
sub-Tg inelastic behavior of amorphous polymers over a wide range of tempera-
ture and strain rates. Temperature and strain rate greatly influence the mechanical
response of amorphous polymers. The results satisfactorily reproduce experimen-
tal observations and provide a general physical explanation for the occurrence of
the change in deformation regime observed at a temperature lower than Tg. The
kinetic equilibrium occurring between the creation of defects due to plastic defor-
mation and their annihilation through structural relaxation is shown to play a ma-
jor role in the existence of the two regimes of deformation.
Keywords: Amorphous polymer, PMMA, DMA, Plastic deformations, Thermo-

mechanical behavior
1 Introduction
The sub-Tg plastic deformations of amorphous polymers have been successfully
described using a theory [Perez J 1984, Perez J et al 1988, Cavaillé JY et al 1989]
based on:
• The concept of quasipoint defects qpds in concentration Cd corresponding to
positive or negative microfluctuations of specific volume ;
• The assumption of hierarchically constrained molecular motions. Once
activated, these qpd may be transformed into shear microdomains smd after an av-
erage time period τ 1 . As long as the stress is applied, the smd grow larger and ul-
timately merge irreversibly into one another after a average time period τ2,
which can expressed as [Cavaillé JY et al 1989]:
dn(t ) n(t ) − n(∞ ) n(0) − n(t ) (1)
=− +
dt τ1 τ2
where n(t) is the number of inactivated defects at time t. τ 2 is the mean time
of the order of the time characterising the elementary molecular motions responsi-
ble to the secondary, or β , relaxations, τ β and the time τ 2 was found to be of

170 N. Ouali and A. Ahmed Benyahia
the order of the mean time needed for a translational elementary molecular motion
τ mol , which is related to τ β according to:
1
⎛ τβ ⎞κ
τ mol = ⎜ 1−κ ⎟ (2)
⎜t ⎟
⎝ 0 ⎠
where t0 is a scaling parameter. κ is a measure of the degree of hierarchical
constraints on the correlated molecular motions (0< κ <1): κ =0 corresponds to a
fully constrained situation (closely packed molecular arrangements) and κ =1
corresponds to a constrained free situation, that is τ mol = τ β . After τ 2 the ma-
terial begins to deform plastically. The fact that the correlation parameter κ de-
pends on the polymers microstructure. A simple linear relationship was originally
proposed [Cavaillé et al 1991] to relate κ and the structural parameter Cd:
κ = a 0 .C d (3)
where a0 is a proportionality factor.
⎡U ⎤
τ β = τ 0 β . exp .⎢ β ⎥ is the characteristic time for the elementary molecular mo-
⎣ KT ⎦
tions, which has been related to the β relaxation ; where U β is the molecucular
activation energy and τ 0 β is a time of vibration ( τ 0 β =5.2 10-16 s). The time was
found to lie between τ β and τ 2 . On the basis of these arguments, the time de-
pendent shear compliance was expressed as:
⎧ ⎡ ⎛ t ⎞
κ
⎤ ⎫⎪ ⎛ t ⎞
κ'
1 ⎪ (4)
J (t ) = + Aan ⎨1 − exp ⎢− ⎜⎜ ⎟⎟ ⎥ ⎬ + Avp ⎜⎜ ⎟⎟
Gu ⎪⎩ ⎢⎣ ⎝ τ mol ⎠ ⎥⎦ ⎪⎭ ⎝ τ mol ⎠
where Gu is the unrelaxed modulus and Aan and Avp are proportionality factors
that depend mainly on the concentration Cd of defects and κ ' (0< κ ' <1) is an
exponent that takes into consideration the fact that segmental mobility is affected
by strain above a certain threshold of strain. Calculating the Fourier transform of
equation (4) and adding the relaxed modulus, Gr , for the primary relaxation which
corresponds to the value of the modulus at the rubbery plateau, yields the dynamic
modulus:
Gu − G r
G * (iω ) = Gu + (5)
1 + (iωτ mol ) + C.(iωτ mol )
−κ −κ '
where C is a constant of the order of unity which is related to Aan .

The Dynamic Mechanical Analysis (DMA) were used to characterise the time-
temperature dependence of the elastic storage moduli and the loss factors of
amorphous polymers. The DMA testing provided a quantification of the rate de-
pendent shift of material transitions, and led to an analytical description of
the elastic modulus at several testing rates and temperatures. Equation (5) was
Analysis of Inelastic Behavior of Amorphous Polymers 171
successfully used to describe the primary relaxation of several polymer materials

including amorphous homopolymers, blends, one random copolymer and one
cross-linked thermosetting resin [Cavaillé JY 1991]. This theory, using the viscoe-
lastic material transitions quantified in DMA testing, as well an understanding of
the molecular mechanisms of resistance to deformation, is extended and applied to
the case of large stress mechanical tests [Ouali N 1992, Mangion MBM 1993,
Haward RN et al 1997, Perez J 1998, Cady C et al 2003, Mulliken AD et al 2004].
Theoretically calculated curves compared with the experimental data are presented
and their implications discussed.
2 Theoretical Considerations
In previous developments, the elementary molecular motions were assumed to be
independent of stress, for the stress levels applied in mechanical spectrometry
were low. If strain is no longer linear with stress ( σ 〉 106 Pa ), a part of the basic
hypothesis must be completed: these elementary molecular motions become ther-
momechanically activated and the amount of purely thermal energy needed has
been shown to decrease, this leads to the expression:
⎧ ⎡ ⎤3 ⎫
2
⎪U ⎢1 − ⎛⎜ σ ⎞⎟⎥ ⎪
⎜ ⎟
⎪⎪ ⎢ ⎝ σ 0 ⎠⎦⎥ ⎪⎪
β
(6)
⎛ ΔH a ⎞
τ β = τ 0β . exp.⎨ ⎣ ⎬ = τ 0β . exp.⎜ ⎟
⎪ KT ⎪ ⎝ KT ⎠
⎪ ⎪
⎪⎩ ⎪⎭
where ΔH a is the thermomechanical activation energy of the β relaxation, and

is the limit of the yield stress when T becomes 0 K. The effects of the large de-
formations on the microstructure of the material, that is on the concentration Cd of
defects and on the correlation parameter κ . Very large plastic deformations are
expected to cause:
• An increase in Cd from an initial value, Cd eq(Tg), due to the propagation of
sessile dislocation lines that border the smd. This increase is proportional to the
additional amount of viscoplastic deformation, ε vp , with a proportionality factor
that decreases as the temperature increases ;
• A decrease in Cd towards an equilibrium value, Cd eq (T), due to structural re-
laxation [Perez J 1998].
The rate of decrease in Cd is proportional to (Cd – Cd eq) with a proportionality
factor that depends on both thermal, and mechanical, diffusion mechanisms. These
considerations are included in the following relationship [Mangion MBM 1993]:
2
dCd ⎡ dε vp ⎤ ⎡ 3 C − C (T ) ⎤ (7)
= ⎢ B0 (1 − Θ)⎥ − ⎢ Cd d d eq (1 + Θ)⎥
dt ⎣ dt ⎦ ⎢⎣ 2 τ 2 (σ ) ⎥⎦
where B0 and Θ = τ 2 (σ ) / τ 2 (0) is a constant.
The (1 − Θ ) expresses the fact that, as the temperature is increased, that is as Θ

approaches units, the diffusion of defects becomes an increasingly important me-
chanism for plastic deformation, a mechanism which is not associated with defect
creation. The (1 + Θ ) factor indicates that, while towards lower temperatures,
that is as Θ approaches zero, molecular diffusion is mainly mechanically and
thermally driven.
Finally, the effect of large strains on κ which was taken into account through the
definition of κ' is included here by modifying equation (3) as follows :
κ = a 0 .C d (1 − a1 .ε vp ) (8)
where a1 is a constant.
During a mechanical test with a constant speed of the compressing cross-head Vt,
the experimental considerations lead to the following relationship:
dσ ⎡ dε dε ⎤ (9)
Vt = R + L0 ⎢ an + vp ⎥
dt ⎣ dt dt ⎦
from which dσ can be isolated as follows:
dσ 1 ⎡ ⎛ dε dε ⎞ ⎤
= ⎢Vt − L0 ⎜⎜ an + vp ⎟⎟⎥ (10)
dt R⎣ ⎝ dt dt ⎠⎦
where L0 is the sample length and R is a factor including the rigidity of both test
machine and the sample. Thus, incrementing the time and calculating the incre-
mental values of anelastic deformation, dε an , and viscoplastic deformation, dε vp ,
from equation (4) modified, it becomes possible to numerically obtain σ (t ) .
A computer program was accordingly written to calculate the time evolutions of
σ and ε .
3 Mechanical Test and Material
A uniaxial compression test was performed for the study of the large-stress de-
formation plastic response of PMMA. In order to apply a uniaxial stress, cylin-
drical samples were selected. Samples 12.00 mm thick were thus machined with a
turn from a PMMA rod of 8.02 mm diameter. A good parallelism between the two
circular faces was necessary to avoid bending during compression. The two faces
were carefully polished in order to minimise interfacial friction between the sam-
ple and the press. The simples thus prepared were placed inside a thermoregulated
cell (200 – 400 K).
The True stress, σ , and the strain, ε , were directly obtained from the meas-
ured values of compressive force, F, and the displacement, ΔL , according to the
following relationships:
F
σ= [1 − ε ] (11)
S0
ΔL
ε= (12)
L0
where S0 and L0 are respectively the cross-section and the length of the sample be-
fore testing.
Figure 1 shows the stress-strain curves during compression tests at a constant de-
formation rate ε =14.10-5 s-1, measured at several temperatures between 293 and
363 K. These curves resemble those usually found in the literature for glassy po-
lymers [Ward IM & Sweeney J 2004, Richeton J 2006, Richeton J 2007] and can
be divided in four parts:
• An initially linear part corresponding to the elastic response ;
• A part where the curve reaches a maximum in stress, indicating a softening of
the simple with an accompanying drop in stress ;
• A part where the stress is minimum, corresponding to the stationary plastic
flow mode ;
• A final part where the stress increases, corresponding to homogeneous defor-
mation with increasing work hardening prior to breaking.
Figure 1 further presents a comparison between data and the theoretical model.
We remark that all the features of the experimental plastic responses are remarka-
bly well predicted by the molecular model which is therefore able to describe a
full set of large-stress plastic response data.
Figure 2 shows the evolutions of σ , Cd corresponding to T=293 K and
ε =14.10-5 s-1 together with the experimental activation volume, Vexp, determined
according to:
⎡ ⎛ dε vp ⎞⎤
∂ ⎢ln⎜⎜ ⎟⎥
⎟
⎣⎢ ⎝ dt ⎠ ⎦⎥
Vexp = KT (13)
∂σ
Fig. 1 Effect of the temperature on the plastic behavior of the PMMA in compression:
Results of simulations (solid line), compared with the experimental curves.
Fig. 2 Variation compared during a test of compression calculated at T=293 K and

ε =14.10-5 s-1 (curve a) of the activation volume (curve b) and concentration of quasipoint
defects (curve c)
Cd is seen to increase right after the appearance of the peak in stress, implying
that, no plastic deformation occurs prior to it. Both Cd and Vexp tend to stabilise
following parallel trends around a time when σ is minimum in the plastic flow
region. The Cd and Vexp values corresponding to such stationary conditions are re-
ferred to as Cd stat and Vexp stat, respectively, and are plotted as a function of temper-
ature in figure 3 together with the stationary experimental activation enthalpy
values, ΔH exp stat , which was calculated from:
⎡ ⎛ dε vp ⎞⎤
∂ ⎢ln⎜⎜ ⎟⎥
⎟
⎣⎢ ⎝ dt ⎠⎦⎥ (14)
ΔH exp stat = KT 2
∂T
Both the time and temperature evolutions of Vexp stat and ΔH exp stat closely re-
semble those measured by LEFEBVRE JM 1982.
Fig. 3 Variation of the stationary values: (a) Cd stat , (b) Vexp stat , (c) ΔH exp stat with the
temperature. Comparison with the experimental values (curves d and e) obtained by
LEFEBVRE [13]
A careful analysis of figures 2 and 3 provides a physical interpretation for the

widely observed existence of two modes of deformation:
• A low temperature mode where Vexp stat is fairly constant and ΔH exp stat
linearly increases;.
• A high temperature mode where both Vexp stat and ΔH exp stat increase from
their low temperature trends.
The Cd stat evolutions convincingly show that, while at low temperatures the dy-
namic equilibrium between creation and annihilation of defects stabilises around a
value mainly independent of temperature, that equilibrium becomes more and
more biased towards the annihilation of defects as the temperature increases so
that Cd stat rapidly decreases towards its equilibrium value Cd eq (T) when T ap-
proaches the glass transition temperature, Tg (Tg=380K). The model is thus capa-
ble to explain both regimes using only one set of equations, as is clearly seen in
figure 4 where the stationary value of compression plastic flow stress, , norma-
lised to the shear modulus, G (G=2500 MPa), is plotted as a function of tempera-
ture normalised to Tg.
Figure 4 shows that log (σ stat ) initially decrease linearly with the increasing
G
temperature and decreases more rapidly as T approaches Tg. The good agreement
between the calculated curve and the experimental data points obtained from
figure 1 confirms the validity of the model. The temperature which separates the
two deformation regimes (figure 4) indicates the same separation temperature as
that shown in figure 3 and formally separates the low from the high temperature
behavior.
Fig. 4 Variation of the plastic flow stress with the temperature: comparison of the
calculated results (solid line) with the experimental results of figure 1
5 Conclusion
The model Based on the concepts of quasipoint defects, qpds, nucleation and ex-
pansion of shear microdomains smd, and hierarchically constrained molecular dy-
namics, provides a satisfactory description both of the dynamic mechanical
response, and the sub-Tg mechanical behavior of amorphous polymers when sub-
jected to large stresses under varying conditions of temperature and deformation
rates. Plastic deformation of amorphous polymers could thus be described in terms
of dynamic recovery phenomena.
References
Cady, C., Blumenthal, W., Gray III, G., Idar, D.: Journal de Physique IV 110, 27–32 (2003)
Cavaillé, J., Perez, J., Johari, J.P.: Phys. Rev. B 39(4), 2411 (1989)
Cavaillé, J., Perez, J., Johari, J.P.: J. Non-Crystalline Solids 133, 935 (1991)
Haward, R.N., Young, R.J.: The physics of glassy polymers, 2nd edn. Chapman & Hall,
London (1997)
Lefebvre, J.M.: Ph. D. Thesis, UST Lille, France (1982)
Mangion, M.B.M., Cavaillé, J.Y., Perez, J.: Phil. Mag. A 66, 773 (1992)
Mulliken, A.D.: M.S. Thesis, Dept. of Mech. Eng. Massachusetts Institute of Technology,
Cambridge, MA (2004)
Ouali, N.: Ph. D. Thesis, INSA-Lyon, France (1992)
Perez, J.: Acta. Mat. Sci. 32(12), 2163 (1984)
Perez, J., Cavaillé, J.Y., Etienne, S., Jourdan, C.: Rev. Phys. Appl. 23, 125 (1988)
Perez, J.: Physics and mechanics of amorphous polymers, 324 p. Editions CRC Press, Be-
kema AA, Rotterdam (1998)
Richeton, J., Sahzi, S., Vecchio, K.S., Jiang, F.C., Adharapurapu, R.R.: International Jour-
nal of Solids and Structures 43, 2318 (2006)
Richeton, J., Sahzi, S., Vecchio, K.S., Jiang, F.C., Makradi, A.: International Journal of
Solids and Structures 44, 7938
Ward, I.M., Sweeney, J.: An introduction to the mechanical properties of solid polymers,
2nd edn. Wiley, Chichester (2004)
First Principal Calculations of Optical
Properties of InGaN2 Using in Solar Cells
Applications
Hamza Bennacer1, S. Berrah2, A. Boukortt 3, Mohamed Issam Ziane4

1
LMER Laboratory, University of Bejaia 06000, Algeria Elaboration and Characterization
Physical Mechanics and Metallurgical of Material, Laboratory,
Electrical Engineering Department, University of Mostaganem 27000, Algeria
Hamza_max28@yahoo.fr
2
LMER Laboratory, University of Bejaia 06000, Algeria
3
Elaboration and Characterization Physical Mechanics and Metallurgical of Material,
Laboratory, Electrical Engineering Department, University of Mostaganem 27000,
Algeria
4
Applied materials Laboratory, University of Sidi bel abbesse, 22000, Algeria
Abstract. The aim behind this work is to look into the optical properties of In-
GaN2 with chalcopyrite structure in ternaries compounds. To obtain precise
results, we have used the first-principles calculations, using the full potential-
linearized augmented plane wave method (FP-LAPW) within the density func-
tional theory (DFT). in order to get best and accurate values of the band gap, as
exchange–correlation potential, , we have used the modified Becke – Johnson
(mBJ) of Tran and Blaha, which are based on the optimization of total energy and
corresponding potential. The InGaN2 chalcopyrite demonstrates semiconducting
behavior, with the direct-band gap of 1.68 eV. We have studied the dielectric
function, refractive index, reflectivity, absorption coefficient, and the optical con-
ductivity function. The results we have obtained indicate that InGaN2 is an attrac-
tive and promising material for optoelectronic and photovoltaic applications.
Keywords: First principal, Ab- initio, FP-LAPW, Wien2k, mBJ-LDA, chalcopy-

rite, linear optical proprieties.
1 Introduction
Recently, the group-III nitride-based semiconductors have proved their potential

application in optoelectronic devices from the ultraviolet to near-infrared (A. La-
ref et al, 2013). These categories of wide band gap (> 2.4 eV) semiconductors
have been discerned to be excellent and promising candidates for laser and short-
wavelength light emitting diodes and the absorber layers in solar cells, (A. Laref et
al, 2013).

180 H. Bennacer et al.
Currently wide-band-gap materials are required in multiple-junction solar cells to

achieve ultrahigh efficiency (A. De Vos, 1992). This is due to the modulation of
the energy band gaps. InxGa1-xN ternary alloy is one of the most important mate-
rials in III-nitride family with direct band gap over than 2.4 eV, and has a strong
absorption coefficients. This makes InxGa1-xN ternary alloys are emerged as new
solar cell materials with record high open-circuit voltage of 2.0 V (O. Jani et al,
2007) (Balakrishnam R et al, 2010).
In Our study we try to extract the optical properties of the chalcopyrite ternary
InGaN2 which is normally crystallize in the chalcopyrite structure showed in fig.1
and have physical properties similar to those of the In0.5Ga0.5N in zinc-blende
structure (Bin Xu et al, 2009). Although two crystal structures are quite similar,
the anisotropy of chalcopyrite crystals gives rise to many interesting properties not
possible in zinc-blende crystals (Yang Mao et al, 2012).
Fig. 1 The structure of InGaN2.
2 Computational Details
The present calculations are performed using the full-potential linearized aug-
mented plane wave (FP-LAPW) method (G. K. H. Madsen et al, 2001) (K.
Schwarz et al, 2002) integrated in Wien2k package (P. Blaha et al, 2001).
We take into consideration the InGaN2 ternary compounds in body-centered te-
tragonal chalcopyrite structure (space group I ¯ 42d) with 8 atoms in the unit cell,
shown in fig 1, the lattice constant a corresponding to the lattice constant of a cu-
bic zincblende structure, the c/a ratio, and the internal displacement parameter u
revealing the distortion of the anion sublattice due to different surroundings. In the
case of ideal structure, c/a = 2 and u = ¼ (J.P. Perdew et al, 2008).
In order to achieve energy eigenvalues convergence, the wave functions in the
interstitial regions were expanded in plane wave with a cutoff RMT x Kmax = 8,
First Principal Calculations of Optical Properties of InGaN2 181
where RMT is the minimum radius of the muffin-tin spheres and Kmax gives the
magnitude of the largest k vector in the plane wave expansion. Inside the spheres,
the valence wave functions were expanded up to lmax =10. In order to keep the
same degree of convergence for all the lattice constants, we kept the values of the
sphere radii and Kmax constant over all the range of lattice spacing considered. The
RMT chosen values in our calculations for Ga, In and N are respectively 1.9, 2.0
and 1.6 a.u.
The exchange-correlation potential for electronic and linear optical properties
was treated by the modified Becke-Johnson (mBJ-LDA) (F. Tran and P. Blaha,
2009) (D.Koller, F. Tran, and P. Blaha, 2011) (A. D. Becke and E. R. Johnson,
2006) which is seen as a promising tool for accurate determination of the funda-
mental band gaps of wide-band-gap materials (D.Koller et al, 2011) (A. D. Becke
and E. R. Johnson, 2006) (P. E. Blochl et al, 1994)( V. L. Shaposhnikov et al,
2012).
The self-consistent calculations are considered to be converged when the total
energy of the system is stable (P. E. Blochl et al, 1994) (F. Annane et al, 2010)
within 10-4 Ryd. The integrals over the Brillouin zone (BZ) are performed up to
102 k-points, (grid of 10_10_10 meshes, and equivalent to 1000 k-points in the
entire BZ).
In the optical properties calculations we used the equivalent to 20000 k-points
in the entire BZ by the OPTIC code (C. Ambrosch-Draxl and J.O. Sofo, 2006),
implemented in the all-electron Wien2K method (P. Blaha et al, 2001). We have
therefore made a correction ∆ 0.057 eV to the gaps.
3.1 Optical Properties

In this section, we have studied the optical linear properties of the chalcopyrite
ternary InGaN2 semiconductor, by using the full potential linearized augmented
plane wave (FPLAPW) method (H.J.F. Jansen and A.J. Freeman, 1984) based on
density functional theory (DFT) (P. Hohenberg and W. Kohn, 1964) (L.J. Sham
and W. Kohn, 1965) (J.P. Perdew and Y. Wang, 1992) within mBJ-LDA approx-
imation. The calculated band structure and total density of state are shown in fig.2.
The calculation shows that InGaN2 is semiconductor with direct energy gap of
1.68 eV. In the other hand we present the dielectric function, refractive index, ref-
lectivity, the absorption coefficient, and the optical conductivity. As scissor ener-
gy (energy shift), we used the values of 0.057 eV. The most representative results
are shown below.
DOS
16
14 InGaN2
12 mBJ-LDA
10
8
6
4
Energy (eV)
2
0
-2
-4
-6
-8
-10
-12
-14
-16
Z Γ XP N Γ 20 40 60 80 100
Fig. 2 Electronic band structure of density of state of InGaN2 calculated within mBJ-LDA.
3.1.1 Dielectric Function
The optical properties are deduced from the dielectric function of the semiconduc-
tor, given by:
ε (ω) = ε1 (ω)+iε 2 (ω) (1)
Respectively 1(ω) and 2(ω), are, its real and imaginary parts. The dielectric
function can be calculated based on the band-structure results. The imaginary part
of the dielectric function 2(ω) is directly obtained from the electronic structure,
using the joint density of states and the optical matrix overlap. The real part of the
dielectric function 1(ω) is then calculated by the Kramers-Kronig relation (S.
Berrah et al, 2009) (M. Fox, 2001).
For calculation of frequency dependent dielectric function we need energy
eigenvalues and electron wave functions. The energy eigenvalues and electron wave
function are the natural output for the calculation of electronic band structure.
The imaginary dielectric function is an indispensable quantity; it indicates the
varied transitions inter-band in the semiconductor (S. Berrah et al, 2009).
The real and imaginary parts of InGaN2 between 0 and 18 eV are shown in Fig-
ures 3 and 4 respectively, along two main crystallographic directions E//c and
E c. Our calculated ɛ2 (ω) curve of InGaN2 shows a sharp increase at about 1.74
eV at the first onset E0 of the direct optical transitions. The first main peak is si-
tuated at about 2.02 eV, the second one at 5.92 eV, whereas the third, the fourth
and the fifth respectively located at 6.87, 9.82 and 12.21 eV. Our calculated static
value of ɛ1(0) is 5.060.
8
InGaN2
E⊥c
6
E//c
ε1(ω)
2
-2
mBJ-LDA
-4
0 2 4 6 8 10 12 14 16 18
Energy (eV)
Fig. 3 Reel part of dielectric function of InGaN2
Table 1 The calculations peaks (E c) in the imaginary dielectric function ε2(ω), static
optical parameters ε1(ω) and n (ω)
E0 E1 E2 E3 E4 E5 ε1(0) n(0)
InGaN2 1.74 eV 2.02 eV 5.92 eV 6.87 eV 9.82 eV 12.21 eV 5.060 2.253
10
mBJ-LDA E4 InGaN2
E⊥c
8 E//c
E5
E3
6
ε2(ω)
E2
4
E1
E0
2
0
0 2 4 6 8 10 12 14 16 18
Energy (eV)
Fig. 4 Imaginary part of dielectric function of InGaN2
3.1.2 Refractive Index
The refractive index n is a very important physical parameter related to the micro-
scopic atomic interactions. Its values are often required to interpret various types
of spectroscopic data. The refractive index of a semiconductor, n (ω) is computed
through the real dielectric function (S. Berrah et al, 2009):
n = ε1 (2)
We have presented on fig 5 the refractive index of InGaN2. The calculated static
components of refractive index are presented in Table 1. The static values of the
refractive index for low frequency described by the relation n(0)=√ε1(0), n(ω)
having value 2.253 and n( ) having value 2.248 satisfy the relation n(0)=√ε1(0).
The two polarizations non zero components of refractive index show maximum
value at 5.77 eV at n(ω) and 9.74 eV at n(ω) .
3.0
InGaN2 E⊥c
E//c
refraction index 2.5
2.0
1.5
1.0
0.5
mBJ-LDA
0.0
0 2 4 6 8 10 12 14 16 18 20 22 24 26
Energy (eV)
Fig. 5 The refraction index of InGaN2
3.1.3 Reflectivity
In fact, reflectivity of light given by equation (3) (Sadao Adachi, 2009), is one of
the most important parameters in linear optical calculations, it is sensitive to a
complicated combination of 1(ω) and 2(ω) (Mohamed Issam Ziane et al, 2015).
We plot the reflectivity R(ω) calculated within mBJ-LDA functional as a function
of photon energy in Fig. 6. The zero frequency reflectivity is 0.148. To be used in
photovoltaics, a material should satisfy some requirements: to be a direct-gap sem-
iconductor, and to have a sizable band gap value (more than 1.2 eV), small
reflectance, and large absorption coefficient to absorb the essential part of the visi-
ble light spectrum (V.L.Shaposhnikov et al, 2012).
R (ω ) = ((n (ω ) − 1)2 + k (ω ) 2 ((n (ω ) + 1)2 + k (ω ) 2 (3)

0,6
InGaN2 E⊥c
mBJ-LDA E//c
0,5
0,4
Reflectivity
0,3
0,2
0,1
0,0
0 2 4 6 8 10 12 14 16 18 20
Energy (eV)
Fig. 6 The calculated reflectivity of InGaN2 using mBJ-LDA

However, the compound with a band gap higher than 1.5 eV may absorb less of
visible spectrum (V.L.Shaposhnikov et al, 2012). Obviously from Fig. 6, the max-
imum static value of reflectivity (reflectivity at 0 eV) located at about 7.42 eV.
3.1.4 Absorption Coefficient
The optical absorption coefficient α(ω) is one of the most important evaluation
criterions for the photoelectric materials (A. Boukortt et al, 2010) (A. Bouhema-
dou et al, 2013) (Omkar Jani et al, 2005) (B. Amin et al, 2012). Our calculated ab-
sorption coefficient spectra α(ω) show in fig 7, that InGaN2 have a good optical
absorption in a wide energy range (2-17 eV). The absorption is determined by the
expression:
4π
α (ω ) = k (ω ) (4)
λ
The maximum value of the absorption coefficient shifts from 12.36 eV (E c) to
12.98 eV (E//c). There is a high absorption between 1.76 eV and 13 eV, corres-
ponding to wave lengths between 704.455 to 95.372 nm.
275
InGaN2 E⊥c
250
E//c
225
absorption in [10^4 / cm)]
200
175
150
125
100
75
50
25
mBJ-LDA
0
0 2 4 6 8 10 12 14 16 18 20 22
Energy (eV)
Fig. 7 The absorption coefficient of InGaN2 calculated within mBJ-LDA.
3.1.5 Optical Conductivity
The optical conductivity given by equation (5) is deduced from the dielectric func-
tion (S. Berrah et al, 2009):
iω
σ (ω ) = − ε (ω ) (5)
4π
The curve of the optical conductivity versus energy is shown in fig 8. We
represent the optical conductivity spectra using the imaginary part of the dielectric
function in range of energies between 0 and 18 eV. The curve presents several
peaks corresponding to the bulk plasmon excitations which are caused by elec-
trons crossing from the valence to the conduction band. The main peak is situated
in 9.80 eV and the total width of the optical conductivity is 13.33 eV.
14000
InGaN2 E⊥c
12000 mBJ-LDA E//c
σ (ω) in [1 / (Ohm cm)]

10000
8000
6000
4000
2000
0
0 2 4 6 8 10 12 14 16 18
Energy (eV)
Fig. 8 The calculated optical conductivity of InGaN2
4 Conclusion
In our study, we simulated the optical properties of InGaN2 ternary chalcopyrite.

Our theoretical contribution has been shown previously that the InGaN2 systems
possess excellent and promising potential optical properties workable in a lot of
optoelectronic applications. So, our goal was to modulate these proprieties that are
convenient to design theoretically InGaN2-based solar cells.
References
Laref, A., Altujar, A., Luo, S.J.: Eur. Phys. J. B 86, 475 (2013)
De Vos, A.: Endoreversible Thermodynamics of Solar Energy Conversion, p. 90. Oxford
Univ. Press, Oxford (1992)
Jani, O., Ferguson, I., Honsberg, C., Kurtz, S.: Design and characterization of GaN/InGaN
solar cells. Appl. Phys. Lett. 91(13), 132117 (2007)
Jampana, B.R., Melton, A.G., Jamil, M., Faleev, N.N., Opila, R.L., Ferguson, I.T., Hons-
berg, C.B.: IEEE Electron Device Letters 31(1) (January 2010)
Xu, B., Han, H., Sun, J., Yi, L.: Physica B 404, 1326–1331 (2009)
Mao, Y., Jun-Jie, S., Wei-Ning, M., Lu, W., Min, Z.: Chin. Phys. Lett. 29(10), 107301
(2012)
Madsen, G.K.H., Blaha, P., Schwarz, K., Sjostedt, E., Nordström, L.: Phys. Rev. B 64, 64
(2001)
Schwarz, K., Blaha, P., Madsen, G.K.H.: Compute. Phys. Commun. 147, 147 (2002)
Blaha, P., Schwarz, K., Madsen, G.K.H., Kvasnicka, D., Luitz, J.: WIEN2k, An Augmented
Plane Wave Plus Local Orbital’s Program for Calculating Crystal Properties (Vienna
University of Technology, Vienna, Austria (2001) ISBN 3-9501031-1-2
Perdew, J.P., Ruzsinszky, A., Csonka, G.I., Vydrov, O.A., Scuseria, G.E., Constantin, L.A.,
Zhou, X., Burke, K.: Phys. Rev. Lett. 100, 136406 (2008)
Tran, F., Blaha, P.: Phys. Rev. Lett. 102, 226401 (2009)
Koller, D., Tran, F., Blaha, P.: Phys. Rev. B 83, 195134 (2011)
Becke, A.D., Johnson, E.R.: J. Chem. Phys. 124, 221101 (2006)
Blochl, P.E., Jepsen, O., Anderson, O.K.: Phys. Rev. B 49, 16223 (1994)
Shaposhnikov, V.L., Krivosheeva, A.V., Borisenko, V.E.: Phys. Rev. B 85, 205201 (2012)
Annane, F., Meradji, H., Ghemid, S., El Haj Hassan, F.: Computational Materials
Science 50, 274–278 (2010)
Ambrosch-Draxl, C., Sofo, J.O.: Computer Physics Communications 175, 1–14 (2006)
Jansen, H.J.F., Freeman, A.J.: Phys. Rev. B 30, 561 (1984)
Hohenberg, P., Kohn, W.: Phys. Rev. 136, B64 (1964)
Sham, L.J., Kohn, W.: Phys. Rev. 145, 561 (1965)
Perdew, J.P., Wang, Y.: Phys. Rev. B 45, 13244 (1992)
Berrah, S., Boukortt, A., Abid, H.: Physica E 41, 701–704 (2009)
Fox, M.: Optical Properties of Solids, p. 6. Oxford University Press, New York (2001)
Boukortt, A., et al.: Physica B 405, 763–769 (2010)
Bouhemadou, A., et al.: Solid State Sciences 20, 127–134 (2013)
Jani, O., Honsberg, C., Asghar, A., Nicol, D., Ferguson, I., Doolittle, A., Kurtz, S.: The
31st IEEE Photovoltaic Specialists Conference, January 3-7 (2005)
Amin, B., et al.: Opto-electronic response of spinels MgAl2O4 and MgGa2O4 through
modified Becke-Johnson exchange potential. Physica B 407, 2588–2592 (2012)
Adachi, S.: Properties of Semiconductor Alloys: Group-IV, III–V and II–VI Semiconduc-
tors. John Wiley & Sons, Ltd. (2009) ISBN: 978-0-470-74369-0
Ziane, M.I., Bensaad, Z.: Tarik Ouahrani and Hamza Bennacer. Materials Science in Semi-
conductor Processing 30, 181–196 (2015)
Agent-Based Approach for the Optimal Design
of Mechatronic Systems
Amir Guizani1, Moncef Hammadi1, Jean-Yves Choley1, Thierry Soriano1,

Mohamed Slim Abbes2, and Mohamed Haddar2
1
Engineering Laboratory of Mechanical Systems and Materials LISMMA, SUPMECA-
Paris, 3 rue Fernand Hainaut, 93407 Saint-Ouen, France
amir.guizani@live.fr,
{moncef.hammadi,jean-yves.choley,thierry.soriano}@supmeca.fr
2
Mechanics Modeling and Production Research Laboratory LA2MP,
National School of Engineers of Sfax (ENIS), B.P, 1173, 3038 Sfax, Tunisia
{slim.abbes,mohamed.haddar}@enis.rnu.tn
Abstract. Mechatronic design optimization is a hard task especially if performed

by several different designers geographically distributed and using different mod-
eling tools. Therefore, the design tasks should be divided into partitions easier to
manage to reduce this complexity. However, the optimization of the overall design
requires incorporating the relevant partitions in order to find the optimum mecha-
tronic design. Efficient strategies of partitioning and coordination should be speci-
fied at the conceptual level to have a successful optimization process. In this
study, a new efficient collaborative optimization approach based on multi-agent
paradigm is proposed for mechatronic design optimization. The proposed method
is applied to the preliminary design case of an electric vehicle to demonstrate its
validity and effectiveness.
Keywords: Mechatronic design, Modeling, Simulation, Optimization, Multi-agent

approach.
1 Introduction
A typical mechatronic design involves multidisciplinary teams to work together on

the design of a mechatronic system. Challenges associated to collaborative design
of multidisciplinary teams require dividing the work into smaller design problems
easier to manage and coordination between different sub-problems of design in
order to avoid misleading conclusions that could make mechatronic system expen-
sive or useless.

190 A. Guizani et al.
In this paper we present an agent-based approach to facilitate a mechatronic

system design for the purpose of its optimization. The presented approach has
been applied to the case of an electric vehicle design optimization to illustrate its
efficiency in dividing the complex design of mechatronic systems.
2 Review of Existing Methods
The optimization of mechatronic systems requires performance evaluation taking

into account the coupling effect between the parties composing the overall mecha-
tronic system. (Plume and Mitchell 2007) stressed that good design can only be
achieved by focusing on the parties, because the parameters of a party may affect
the level of other parties. There is therefore a risk that one party is evaluated
incorrectly.
Several approaches have been proposed in previous works to address this prob-
lem. Multidisciplinary design optimization (MDO) seems to be the most promis-
ing solution for the optimization of mechatronic systems, because it allows
designers to incorporate all relevant disciplines simultaneously. MDO has been
used in a number of fields, including automobile design, electronics, aerospace de-
sign, etc. For automobile design, (Guizani et al. 2014) proposed new approach
combining Modelica and ModelCenter to optimize mechatronic systems, model-
ing language Modelica has been used to model and simulate the electric vehicle
and ModelCenter has been used for the multidisciplinary optimization of the elec-
tric motor and the transmission gear ratio. In power electronics, (Hammadi et al.
2011) presented an application of MDO techniques for the layout optimization of
power modules. MDO allows designers to find the optimum of the collaborative
design which is better compared to the design found by optimizing each discipline
sequentially, since it can exploit the interactions between the disciplines. Howev-
er, including all disciplines simultaneously increases significantly the complexity
of the problem and computation time.
To reduce the complexity of the problem some authors proposed to integrate
surrogate modeling technique with multidisciplinary optimization. For instance,
(Hammadi et al. 2012) proposed to combine surrogate modeling technique with
Modelica language to optimize electric vehicles. To reduce computation time, de-
composition-based design optimization requires the definition of a system parti-
tion and the construction of an efficient strategy for coordinating the solution of
the resulting sub-problems (Allison et al. 2009).
Agent-Based Approach for the Optimal Design of Mechatronic Systems 191
3 Description of the Multi-agent Approach for Mechatronic

Design Optimization
To optimize a mechatronic system, design is divided into two levels (partition and
coordination), these levels are related to the division of the problem into multiple
partitions according to certain attributes. For each partition, we allocate an agent
called the design agent DA (partition level), whose goal is to make a local optimi-
zation with respect to a specific partition. To find the global optimal design of me-
chatronic systems, we consider another type of agent called the coordinating agent
CA (coordination level). The objective of CA is to find the best way to partition
the global optimization problem and how to coordinate between DA(s) during the
design optimization process. DA ensures communication with CA. In case no solu-
tion is found in the process of optimizing a partition, the role of DA is to propose a
modification of the search space of design variables or constraints of local optimi-
zation and find the optimal solutions with respect to these changes.
Each agent is associated to a team of engineering and a development platform.
The engineering team may be composed of engineers and experts in the various
fields of engineering system. The platform consists of various development tools
for modeling, simulation and optimization.
Each DA is related to two types of model: analysis model and knowledge mod-
el. The analysis model is used to evaluate the design assigned to a partition (CAD
model, finite element model). The knowledge model is based on rules of modeling
and control and context information of the agent. The modeling rules help design-
ers for example in the definition of the mathematical models that can be used in
the modeling of the mechatronic system. Control rules are used to check the con-
sistency of the analysis model. The context of the agent is based on information in
relation to the design requirements, constraints and optimization objectives.
The coordination model is assigned to CA. This model contains information for
the cooperation and communication between DA(s) such as shared design va-
riables, coupling variables.
To illustrate the coordination process, we consider a case of a mechatronic de-
sign optimization with three design agents DA1, DA2 and DA3 as represented in
figure 1, where every DA is responsible for solving a sub-problem. These DA(s)
are related with linking variables yij, i; j є{1;2;3}, which represent the values sent
by DAj to DAi. yij can be a vector of design variables in the case of a one-objective
function associated to DAj. In the case DAi has more than one objective function,
yij represents an interval of values (Pareto fronts). CA, which is not represented in
the figure 1, defines the order in which every DA performs its local optimization.
Fig. 1 Example of coordination between three design agents

We suppose that this process begins with DA1, then DA2 and terminates with
DA3. In the first step, the coordinating agent defines the initial values of y13 and y12
for DA1 to perform the first local optimization and generate y31 and y21. Then DA2
uses y21 and an initial value of y23 for its local optimization to generate y12 and y32.
Then, DA3 uses y31 and y32 to generate y13 and y23. In the next iteration, every DA
uses the previously generated values for the new local optimizations. For mono-
objective problems, CA calculates the difference between yij and yji (||yij-yji||2) at
very iteration and the optimization process stops when the error is less than a pre-
defined minimum error. For multi-objective partitions, the Pareto-front is nar-
rowed by determining the intersection of fronts between the previous iteration and
the next one until convergence. For both cases of mono-objective or multi-
objective partitions, if no solution is found the stop criteria could be a number of
iterations not to be exceeded.
4 Application to the Case of a Preliminary Design of an

Electric Vehicle
4.1 Introduction
We consider the case of a preliminary design of an electric vehicle. The objective
of this study is to optimize the electric motor and a one-ratio gear box to fulfill the
performance requirements related to the maximum velocity, gradeability and acce-
leration test.
4.2 Mathematical Modeling

The total resistive force applied on the vehicle is defined with:
Ftotal = Fa + Fr + Fg + Fi
1 dV (1)
= ρair .Cdrag . Afront .V + Cro .M .g + M .g .sin α + M .
2 dt
Where Fa is the resistive aerodynamic drag force, Fr is the rolling resistive
force, Fg is the resistive force due to the grade angle of the road, Fi is the inertial
force, ρair is the air density, Cdrag is the aerodynamic drag coefficient, Afront is the
vehicle frontal area, V is the vehicle velocity, Cro is rolling coefficient, M is the
vehicle mass, g is the gravitational acceleration constant and α is the grade angle.
The electric machine is divided into an electric part and mechanic part. The ma-
thematical model of the electric machine is developed in (Guizani et al. 2014).
To make modelization of the system more realistic, we found it useful to model
the mechanical gearbox more accurately. We constructed a simplified model of
gear, valid only for an external cylindrical gearbox. We searched to find the num-
ber of teeth Z1 of the gear and its diameter d1, which allow us to calculate the
width b and the number of teeth of the toothed wheel d2.
The diameter d1 and the number of teeth on the pinion Z1 are given by:
2τ m .KV .K B .Kαβ 1/3
d1 ≥ ( )
Rg
Ω0 .ψ . 2 2
.Z p .ZV
Rg + 1 (2)
⎛ σ ⎞ Rg + 1 1
Z1 < Z1max = ⎜ 0 ⎟ . . 2 2
⎝ Ω 0 ⎠ Rg Z p .ZV
Where τm is the resultant torque, Ω0 is the criterion of superficial resistance of a
gear tooth, σ0 is the criterion of resistance to the root, ψ is the length coefficient of
a pinion, Rg is the gearbox ratio, KV is the dynamic speed factor, KB is the service
factor, Kαβ is the load distribution factor, ZV is the oil film factor and ZP is the
roughness factor.
The diameter d2 and the width b of the gear are determined by:
d 2 = Rg .d1
(3)
b = ψ .d1
In our case, cementation alloy steel 20 MC 5 is chosen. The characteristics of the
material used to make the gear are shown in (Le Borzec 1992).
4.3 Modeling the Electric Vehicle with Modelica

After developing the mathematical models of the electric vehicle, the system has
been modeled using Modelica language (Fritzson 2010). Figure 2 shows the de-
veloped analysis model.
The electric vehicle model is composed of the following components: an input
model for velocity demand, a controller, a power sensor, a current sensor, a torque
sensor, an electric motor, a transmission and a force model. The controller, is
made of the association of two PID controllers (PID Current and PID Voltage), to
respectively control current and voltage.
To validate the analysis model that will be used later in the optimization process,
we will compare the velocity at the output of the electric vehicle with a standard
driving cycle provided as input. This allows us defining default values for the gear
ratio ( Rg ) and EMF ( K m ), as well as setting the controller parameters.
Fig. 2 Electric vehicle model developed with Modelica

4.4 Simulation of the NEDC Profile

For the sake of simulation and for reproducing a road path with different driving
conditions, we will be using the New European Driving Cycle NEDC. To simulate
the NEDC profile the parameters given in (Guizani et al. 2014) are used.
Results shown in figure 3 indicate a good agreement between the input driving
cycle and the output vehicle velocity. By modifying the input to constant speed,
we can simulate different other test cases such as the maximum velocity and the
velocity in a grade road. For instance, the pick power given by simulation for a
test of maximum velocity equal to 140 km/h is 191.69 kW, a maximum torque is
equal to 803 N.m and a maximum volume occupied by the gearbox is equal to
0.00314 m3. These results need to be optimized to reduce the torque, the electric
power consumption and the volume of the gearbox.
Fig. 3 Comparison between input NEDC profile and output vehicle velocity
Using a classic optimization approach, by defining one analysis model for the
three test cases (Acceleration test, Maximum velocity, Gradeability), is a complex
task and computationally expensive. Therefore we will use the multi-agent ap-
proach for this purpose.
4.5 Use of Multi-agent Approach

We consider the optimization of an electric vehicle with three objective functions
and three design constraints:
• Objective1: minimizing the maximum electric power in order to choose
the optimal battery.
• Objective2: minimizing the maximum torque used on the transmission
line in order to choose the optimal electric motor.
• Objective3: minimizing the maximum volume of the gearbox.
• Constraint1: Acceleration test: Vehicle velocity on a 0% grade road af-
ter 10 seconds of start-up (V10@0%) shall be equal to 100 ± 2 km/h.
• Constraint2: Maximum velocity: Vehicle maximum velocity on a ho-

rizontal road (Vmax@0%) shall be equal to 140 ± 2 km/h.
• Constraint3: Gradeability: Vehicle velocity in a road with a grade
equal to 30% after 10 seconds of start-up (V10@30%) shall be equal to
20 ± 2 km/h.
For the design variables, we will limit our study to the gear ratio Rg and the
electromotive force Km of the electric motor.
The first step of the coordinating agent is to find the possible configurations for
partitioning the optimization problem and the distribution of constraints and de-
sign variables between its partitions.
One important metric to be used is the degree of coupling between objective
functions, constraints and design variables. In our example, the two objective
functions are strongly coupled, so they should belong to the same partition. How-
ever, the degree of coupling between constraints is not as strong as the case of ob-
jective functions. Several combinations can therefore be studied; few of them
however are feasible. The three possible cases are: one partition, two partitions
and three partitions.
One partition configuration corresponds to the all-at-one (AAO) case. In the
configuration with two partitions, the degree of coupling between vehicle veloci-
ties V10@0% and Vmax@0% is loose and they are not coupled with V10@30%.So that,
Constraint1 and Constraint2 can be associated to the same partition because they
can be evaluated by the same analysis model (0% grade). In this case Constraint3
will be associated to the second partition. In the configuration with three parti-
tions, each constraint is associated to one partition.
The second task of the coordinating agent is to find the best configuration to be
used for solving the optimization problem. A metric that can be used here is the
computing time of the overall solution for each configuration. Another criterion that
could also be considered is the number of partitions that should be minimized. In our
case the choice is trivial since the two selection criteria (number of partitions and
computing time) are in favor of the configuration with two partitions. For the solu-
tion with two partitions, the global optimization problem is divided into two design
agents. We allocate the same three objective functions to each agent. Design va-
riables gear ratio Rg and electromotive force Km are shared by the two agents.
The third step is to develop the analysis models to be used in the optimization
process. We affected two Modelica models to each agent by modifying the inputs
of the basic model showed in figure 2 to respond to the constraints of each agent.
The next step is to define an optimizer for every agent. For this, we will
allocate the same algorithm of optimization Non-dominated Sorting Genetic
Algorithm II (NSGA II) with ModelCenter11 software to each agent.
The last step of the coordinating agent is to coordinate the optimization process
with two design agents DA1 and DA2. In our case, the linking variable y shown in
figure1 represents the two design variables Km and Rg. After the local optimization
of DA1, y21 is generated which are a list of optimal solutions (Pareto fronts). Then
DA2 performs a local optimization to generate y12 using y21. The coordinating
1
http://www.phoenix-int.com/software/phx-modelcenter.php
agent compares the two optimal Pareto fronts generated by DA1 and DA2 and
keeps the common solution for both design agents.
The process can be repeated until convergence. If the problem has no solution
an empty Pareto front will be returned by the coordinating agent.
4.6 Results and Discussion

To develop our approach, it is convenient to choose a platform, adopting open-
sourced JADE22 as the software development environment fully implemented in
JAVA language. This framework is provided of charge by TILabs and runs entire-
ly on the Java runtime environment.
Figure 4 shows the implementation of the first iteration in the JADE develop-
ment environment.
After the first iteration of optimization, the Pareto fronts are generated in the
partition levels. The best design solutions returned by DA(s) are summarized as
following: (0.265 ≤ Km1 ≤ 0.438; 5.52 ≤ Rg1 ≤ 11.35) for DA1 and (0.248 ≤ Km2 ≤
2.791; 6.83 ≤ Rg2≤ 13.83) for DA2 .
These results are sent to the CA which therefore synthesizes the solutions by
reducing the range of intervals of the design variables. A simple design rule con-
sists in determining the minimum of interval maximums and the maximum of in-
terval minimums. In our case, this gives (0.265 ≤ Km≤ 0.438; 6.83 ≤ Rg ≤ 11.35).
These values are then sent to DA to find the best designs that satisfy all the objec-
tives by narrowing furthermore the design variables range.
The first iteration limits the search space of design variables, but it is very diffi-
cult to determine the vector that satisfies the three requirements of DA(s), hence
the need for a second iteration of optimization. By running a second optimization
processes for every DA with the new design intervals, the new synthesized results
are (0.295 ≤ Km ≤ 0.312; 10.25 ≤ Rg ≤ 10.68).
Fig. 4 Implementation of the first iteration in the JADE development environment
2
http://jade.tilab.com/
Fig. 5 Results of simulation after optimization
The designer could choose one optimal configuration in the Pareto-front. For
example, results of simulation for the optimal configuration corresponding to
Km=0.31 and Rg=10.5, for which we have V10@0%=100.1 km/h, Vmax@0%=140.08
km/h and V10@30%=20.02 km/h (figure 5). The maximum pick of the required elec-
tric power is 190.6 kW, the maximum torque is 202 N.m and the maximum volume
of the gearbox is 0.00205 m3.
5 Conclusion
In this paper we have presented a new approach for optimizing mechatronic sys-
tems based on multi-agent technology. We described our approach and we
presented an example on how to deal with the optimization of an electric vehicle.
Instead of integrating all models relative to all test cases, we have decomposed the
design process to two design agents and we have affected to each agent its objec-
tives and constraints. The coordinating agent has also been considered to
coordinate between design agents to find the overall optimal design. Our main
contribution in this study is to propose an agent approach that allows designers to
determine optimal mechatronic design by decomposing the complex problem and
by coordinating the local optimizations performed by design agents.
Acknowledgements. The authors would like to thank the researchers of the Linz Center of
Mechatronics (LCM), who have contributed in this work with their helpful comments and
suggestions.
References
Allison, J., Kokkolaras, M., Papalambros, P.Y.: Optimal partitioning and coordination deci-
sions in decomposition-based design optimization. Journal of Mechanical De-
sign 131(8), 081008 (2009)
Fritzson, P.: Principles of object-oriented modeling and simulation with Modelica 2.1. John
Wiley & Sons (2010)
Hammadi, M., Choley, J.Y., Penas, O., Louati, J., Rivière, A., Haddar, M.: A and Haddar
M Layout optimization of power modules using a sequentially coupled approach. Inter-
national Journal of Simulation Modelling (IJSIMM) 10(3) (2011)
Hammadi, M., Choley, J.Y., Penas, O., Rivière, A.: Mechatronic system optimization based
on surrogate models-application to an electric vehicle. In: SIMULTECH, pp. 11–16
(2012)
Guizani, A., Hammadi, M., Choley, J.Y., Soriano, T., Abbes, M.S., Haddar, M.: Multidis-
ciplinary approach for optimizing mechatronic systems: Application to the optimal de-
sign of an electric vehicle. In: 2014 IEEE/ASME International Conference on Advanced
Intelligent Mechatronics (AIM), pp. 56–61 (2014)
Guizani, A., Hammadi, M., Choley, J.Y., Soriano, T., Abbes, M.S., Haddar, M.: Multidis-
ciplinary optimization of mechatronic systems: Application to an electric vehicle. In:
Mechatronic Systems: Theory and Applications, pp. 1–14. Springer (2014)
Le Borzec, R.: Réducteurs de vitesse à engrenages. Ed. Techniques Ingénieur (1992)
Plume, J., Mitchell, J.: Collaborative design using a shared ifc building modellearning from
experience. Automation in Construction 16(1), 28–36 (2007)
A Digital Pattern Approach to the Design
of an Automotive Power Window
by means of Object-Oriented Modelling
Stanislao Patalano, Ferdinando Vitolo, and Antonio Lanzotti
University of Naples Federico II - Department of Industrial Engineering, P.le V. Tecchio

80, 80125 Naples (Italy)
{antonio.lanzotti,stanislao.patalano,
ferdinando.vitolo}@unina.it
Abstract. The paper deals with a digital pattern (DP) approach to the design of an
automotive power window, using object-oriented modelling. Therefore, the paper
faces the designing of a mechatronic system by using an integrated approach to
product development. Then, Dymola/Modelica environment is used as a tool of a
decision support system that makes possible the DP approach. The paper briefly
sum up the results of simulations related to a power window system characterized
by a double bowden sliding mechanism. Finally, the paper highlights the parame-
ters that could be easily integrated in a graphical user interface, aimed to reduce
both the development time of new power window system and to increase the accu-
racy of design activities.
Keywords: mechatronic systems, digital pattern for product development, object-

oriented modeling, power window systems.
1 Introduction
The object-oriented modelling (OOM) aimed to the design of multi-domain systems
is, today, widely used. The OOM, in fact, is used to model and simulate possible de-
sign solutions related to different domains (e.g. a mechanical or a controller
domain), not only in a bounded domain. The OOM of multi-domain systems is per-
formed by continuously switching between the domains. Then, simulations allow
easily exploring and comparing several design changes (van Amerongen and
Breedveld, 2003).
The OOM is significant, for example, in case of mechatronic systems defined
as “engineering systems with mechanical, electrical, hydraulic or pneumatic com-
ponents, including interactions of physical effects from various energy domains”
(Borutzky, 2009). Such systems, in fact, due to their multi-disciplinary nature in-
volve several competencies and need for a team of designers with different back-
grounds. Therefore, the modelling of mechatronic systems must support standard-
ized languages and tools for model management and model abstraction (Sinha

200 S. Patalano, F. Vitolo, and A. Lanzotti
et al, 2001). In the last year, several research activities have been focused on such
integration of languages and tools, as well as on the possibility to design a mecha-
tronic system as a whole (Chhabra and Reza Emami, 2011). In particular, the
powerful bond graph representations have been used to deem mechatronic systems
as energy systems and to define three holistic design criteria (namely energy, en-
tropy and agility) to be used in the designing of the mechatronic system when the
power energy, the thermodynamic aspects or the operative aspects of the system,
respectively, are the most significant.
On the other side, industrial companies do prefer to use OOM and multi-
domain tools based on Modelica language for their design activities, without ac-
complishing a deep insight into modelling and simulation (Zupancic and
Sodja, 2011). In this context, also the availability of powerful libraries can help
companies to develop complex models using Dymola-Modelica environment, but
methodological tools are necessary to reduce times and improve target of design-
ing activities. Some large companies, in fact, in automotive and railway fields,
have deepen the tools and methods to accomplish “digital patterns for product de-
velopment” viz. geometrical data and models, to be reconfigured and re-used in
different but similar design activities (Lanzotti, 2013) (Patalano, 2013). These pat-
terns are generally supported by a series of tools, as for example, “quality and
standard checkers”, “predictive engineering wizards” and “cost advisors” that are
part of a decision support system. Then, Dymola-Modelica environment and re-
lated tools could be used to accomplish the decision support system able to guide
the designer during his activities.
A typical mechatronic system to be frequently developed in automotive field is
the power window system. Some research activities were carried out on general
methodological approaches aimed to design the power window controller, as for
example in (Mustafiz et al, 2012). On the opposite, further activities were carried
out for the development of pinch detection algorithm related to the control system
(Ra et al, 2008). No contributes are known as regards the OOM of the whole sys-
tem, according the above mentioned digital pattern approach for product devel-
opment.
The paper is arranged as follows. Section 2 presents the approach and simula-
tion tools used to accomplish the simulations of a mechatronic system. Section 3
summarizes the specifications of the power window system taken into account.
Finally, Sections 4 and 5 draw results of simulations and conclusions.
2 The Digital Pattern Approach and the OOM as a Tool of

the Support Decision System
The Digital Pattern (DP) approach aims to develop models, to be reconfigured and
re-used in similar designing tasks. Therefore, the DP models are geometrical and
numerical models, preconfigured and parametrical, able to be adapted according
the constraints of a new designing task. Then, the designer, by means of DP
models, reduces the development time of new complex products and increases the
A Digital Pattern Approach to the Design of an Automotive Power Window 201
accuracy of his results as the tools of a decision support system drive him. Specif-
ic models, international standards and company rules are used to develop such
tools. Figure 1 depicts the DP approach. In particular, the exchange of contents be-
tween parametrical CAD models and the tools of the decision support system is
cyclic as it deals with the continuous development of each product aimed to fulfill
the system requirements. A graphical user interface (GUI) could be developed to
support designers’ choices, to collect company knowledge and to accomplish veri-
fications.
Geometrical Data Product/

Process Design
Parametrical CAD models
Numerical Data develop Output
ment
⊆X
Mathematical Models
Tools for Decision
Company Rules Support System
Quality and standard checkers
International Predictive Engineering Wizard
Standards
⊆X Cost advisor
Dymola/Modelica simulations
.....
Digital Pattern (DP) approach
Fig. 1 Contents and operational flow of the DP approach for product/process development
In particular, for complex and multi-domain systems as mechatronic systems,

OOM could be used as a tool of the decision support system. The output of simu-
lations, carried out in a Dymola-Modelica environment, in fact, are the mean to
accomplish design choices in terms of dimensioning of components and relative
positioning (as, for example, for power and masses involved). Furthermore, the
output of simulations represents the key to determine, in automatic way, further
characteristics of components (as, for example, overall dimensions, related dis-
tances and displacements). Therefore, a quick verification of a new designing so-
lution, i.e. the components of the mechatronic system, could be easily perform by
the designer as well as the effects of several choices on the assembly in which the
mechatronic system will be assembled.
The procedure to accomplish the OOM of a mechatronic system could be a top-
down approach to complex system. In particular, the present work uses the ap-
proach presented in (Chhabra and Reza Emami, 2011). The mechatronic system is
composed by three main subsystems: the generalized executive subsystem A, the
sensor subsystem B and the control subsystem C. The subsystems exchange en-
ergy (between A and B as well as between C and A), and general flow/effort (be-
tween C and B). The subsystem C exhibits an inner exchange of information while
the whole mechatronic system releases effective work towards external systems.
Using this holistic vision is then possible to decompose each subsystem to accom-
plish the modeling of groups and components, respectively, taken into account es-
sential simplification hypotheses.
3 Case Study: The Automotive Power Window System
The power window system is a mechatronic system i.e. an electrically powered

window aimed to lift and lower the car window by depressing a human interface.
Today, the system is manufactured and assembled in a large variety of automotive
doors. The power windows system is generally composed by human interfaces
(button or remote control), a control system, a DC motor, a gear reduction, a slid-
ing mechanism and a window which are fixed on door framework. This system is
mainly characterized by the sliding mechanism. The architectures used for the
sliding mechanisms are the following: cable spiral, crossed levers, bowden type,
double bowden type.
In the present work we focus on double bowden type (Figure 2).
Fig. 2 Scheme of double bowden mechanism for power window system
The window is fixed on two supports respectively constrained along two rails.
The user, moving upwards or downwards the button, generates a signal, which is
processed by controller that drives the DC motor. The motor wraps two bowden
cables, which move two supports and, consequently, the automotive window. The
basic operations of the motor are accomplished by reversing the polarity of its
power and ground input.
The power window system could be modelled by applying a top down ap-
proach to OOM. A series of levels of detail could be taken into account (Figure 3).
First, the system was divided into three subsystems related to Control, Sensor and
Physic subsystem, respectively. Then, each subsystem was further detailed using
functional groups. Finally, the level related to the components was defined.
Level 0
system
Power Window
System
subsystems
Level 1
Level 2
groups
components
Level 3
Fig. 3 Functional levels related to power window system using a top-down approach
The power window system was modeled using Modelica language within Dy-
mola environment. Modelica Standard Libraries and ad hoc developed objects
were used. The model of the power window system is showed in Figure 4.
Fig. 4 Power window model in Dymola environment

Several assumptions were done to model the power window system:

- no energy losses, therefore the model is ideal;
- no double curvature of window; only the magnitudes in the window
movement direction were taken into account;
- window mass center was computed by means of Varignon’s Theorem;
- human interaction was introduced through a timetable.
The DC motor is driven by means of an H bridge transistor, which is triggered
by a switch that processes the human actions.
Fig. 5 Object-oriented model of the DC motor switch
The model of power window system is characterized by independent and de-

pendent parameters (Table 1). The independent parameters control the behavior of
the model. Their change entails the variation of the lifting and/or lower law.
Table 1 Independent and dependent parameters in OOM of power window system
Independent Parameters Dependent Parameters
Gear Speed ratio Pulley diameter Torque output
DC motor Power Current Torque

Sliding Force on front Force on rear
mechanism support support
Window Mass Displacement Speed Acceleration
A Digital Pattern Approach to
t the Design of an Automotive Power Window 2005
4 Results
The results of simulationss are briefly summed up in this section. In particular, thhe
trends of some dependentt parameters are represented when the power of DC moo-
tor and speed ratio changee.
The trends of dependeent parameters were obtained through a simulation of thhe
button tap and release, vizz. when the window goes all the way up and down. Winn-
dow mass has been set on o 3 kg, max supply voltage on 12V, speed ratio on 1.8,
winding pulley diameter on 130mm and window stroke on 400mm. The requireed
time of transition of the window
w is about 3s.
Figure 6 shows, respecctively, the window displacement and speed as well as thhe
forces acting on front andd rear support. It can be seen how the lifting (lower) tim
me
decreases with increasing g powerful, thus also for the speed and force amplitudees.
Then, designers could diirectly use values of simulated DC motor power, to ac-
complish the dimensionin ng of the system. Overall dimensions of DC motor changge
when power increases. Th herefore, also fixture shape and pins change, consequennt-
ly. The characteristics of assembling surfaces and pins are then automatically proo-
vided to designers, by meeans of a datasheet, accomplished in Excel environmennt
(Figure 7).
Fig. 6 a) Window displaacement; b) Window speed; c) Force on front supporrt;

d) Force on rear support. Curves related to different values of DC Motor powerfuul:
hed line); 28W (dotted line).
72W (solid line); 48W (dash
Fig. 7 Operative flow existiing between DC motor independent parameters and dependeent
parameters plus assembly ch
haracteristics
Figure 8 depicts the trrends of the parameters for three different values of thhe
speed ratios. It highlights the need for a more powerful DC motor when speed ra-
tio decreases. Powers of 96W,
9 50W and 38W are necessary when the speed ratiio
is equal to 1, 2 and 3, resspectively. This behavior is due to the fact that the exteer-
nal loads do not vary; therrefore, it is necessary to balance the external loads to fuul-
fill the requirement of lifting and lower of the window within 3s.
Fig. 8 a) Window displacem ment; b) Window speed; c) Force on front support; d) Force oon
rear support. Curves related to speed ratio values equal to 1 (solid line), 2 (dashed line) orr 3
(dotted line), respectively.
5 Conclusions an
nd Further Works
The DP approach to product development could be applied in the designing of auu-

tomotive power window w systems by using object-oriented modelling. Dymoo-
m
la/Modelica environment,, in fact, could act as a tool of a decision support system
in order to guide designerr during the preliminary development tasks aimed to thhe
designing of a power wind dow systems. The simulations of independent parameters
of the power window system, characterized by a double bowden sliding mechan-

ism, have been accomplished. All the relationships existing between independent
and dependent parameters were set as a basis for the development of a graphical
user interface aimed to support the designer during designing activities.
Further works has to be carried out, by means of experimental tests on physical
systems, in order to validate the mathematical models of the power window sys-
tem here developed.
Acknowledgements. The present work was developed with the economic support of MIUR
(Italian Ministry of University and Research) performing the activities of the project
PON01_01268 “Digital pattern product development: a pattern driven approach for indus-
trial product design”.
References
Borutzky, W.: Bond graph modelling and simulation of multidisciplinary systems – An in-
troduction. Simulation Modelling Practice and Theory 17, 3–21 (2009),
doi:10.1016/j.simpat.2007.08.008
Chhabra, R., Reza Emami, M.: Holistic system modeling in mechatronics. Mechatron-
ics 21, 166–175 (2011), doi:10.1016/j.mechatronics.2010.10.003
Lanzotti, A., Patalano, S., Vitolo, F.: A Graph-based Software Tool for the CAD Modeling
of Mechanical Assemblies. In: Proc. of 8th International Conference on Computer
Graphics, Theory and Applications, GRAPP 2013 IVAPP 2013, Barcelona, Spain, Feb-
ruary 21-24, pp. 60–69. SciTePress (2013), doi:10.5220/0004299000600069
Mustafiz, S., Joachim Denil, J., Lucio, L., Vangheluwe, H.: The FTG+PM framework for
multi-paradigm modelling: An automotive case study. In: Proceedings of 6th Interna-
tional Workshop on Multi-Paradigm Modeling – MPM 2012, Innsbruck, Austria (2012)
Patalano, S., Vitolo, F., Lanzotti, A.: A graph-based approach to CAD modeling: a digital
pattern application to the sizing and modeling of manual transverse gearboxes. In: Proc.
of International Conference on Graphic Engineering, Madrid, Spain, June 19-21 (2013)
Ra, W.S., Lee, H.J., Park, J.B.: Practical pinch detection algorithm for smart automotive
power window control systems. IEEE Transactions on Industrial Electronics 55, 1376–
1384 (2008)
Shina, R., Peredis, C.J.J., Liang, V.C., Khosla, P.K.: Modeling and simulation methods for
design of engineering systems. Journal of Computing and Information Science in Engi-
neering 1, 84–91 (2001), doi:10.1115/1.1344877
Van Amerongen, J., Breedveld, P.: Modelling of physical systems for the design and con-
trol of mechatronic systems. Annual Review Control 27, 87–117 (2003),
doi:10.1016/S1367-5788(03)00010-5
Zupančič, B., Sodja, A.: Advanced multi-domain modelling: advantages and dangers. In:
UKSim 13th International Conference on Modelling and Simulation (2011),
doi:10.1109/UKSIM.2011.56
Numerical Study of the External Excitation
Amplitude Effect on Liquid Sloshing
Phenomenon
Bouabidi Abdallah, Zied Driss, and Abid Mohamed Salah
National School of Engineers of Sfax (ENIS), Laboratory of Electro-Mechanic Systems

(LASEM), B.P. 1173, km 3.5 Soukra, 3038 Sfax, TUNISIA
bouabidi_abdallah@yahoo.fr,
{zied.Driss,mohamedsalah.Abid}@enis.rnu.tn
Abstract. In this work, we are interested on the study the liquid sloshing pheno-
mena in a tank subjected to a sinusoidal excitation. Particularly, we have investi-
gated the amplitude movement effect. The hydrodynamic parameters describing
the flow like the velocity and the static pressure are presented. The numerical re-
sults are investigated for four different values of sinusoidal excitation amplitude.
The results show that the behavior of the sloshing phenomenon strongly depends
on the amplitude value of the external excitation. A good agreement has been
shown by comparing the numerical and the experimental results.
Keywords: sloshing, amplitude, external excitation, VOF.
1 Introduction
The liquid sloshing in rigid containers has been the interest of many studies be-
cause of its frequent application in many engineering sections such as aerospace
vehicles, transportation engineering and maritime applications. The sloshing phe-
nomenon in the tank subjected to external movement was investigated analytical-
ly, numerically and experimentally. Many researchers are interested to studying
the effect of the baffle on the sloshing. For example, Liu et al. (2009) studied the
effect of the vertical and horizontal baffle in the liquid sloshing in a tank. They
showed that sloshing is more reduced using the vertical baffle than using the hori-
zontal baffle. Bouabidi et al. (2013) studied the effect of the vertical baffle height
on liquid sloshing in an accelerating rectangular tank. They showed that the liquid
sloshing decrease with the increase of the vertical baffle height. Panigrahy et al.
(2009) studied experimentally the effect of the three different baffles, horizontal,
vertical and ring baffle. They showed that the ring baffle reduced the sloshing
more than the conventional baffle. The type of the external excitation also affects
the behavior of the liquid motion. Therefore, the sloshing is studied for many
types of excitations. Yan et al. (2009) investigated experimentally the effect a lat-
eral and longitudinal excitations on the sloshing of the liquid in a partially filled

210 B. Abdallah, Z. Driss, and A. Mohamed Salah
tank. Koh et al. (2013) studied the phenomena of the liquid sloshing in a container
subjected to a sinusoidal movement. Akyıldız et al. (2013) analyzed the sloshing
phenomena in a tank subjected to rotational flow for different values of rolling
frequencies. Ji et al. (2012) investigated the motion of the liquid in rectangular
tank subjected to a lateral oscillation. Jin et al. (2014) carried out a series of expe-
riments to study the phenomenon of sloshing subjected to a sinusoidal excitation.
They studied the effect of the frequency and the amplitude of the external excita-
tion on the liquid sloshing. They showed that the characteristic of the excitation
affects considerably the sloshing in the tank.
In this paper, we are interested to study numerically the effect of the amplitude
excitation on the sloshing phenomena in a rectangular tank subjected to a sinu-
soidal movement. The VoF method is used to predict the free surface elevation
and the Navier- Stockes equations are solved in conjuction with k-ε turbulence
models. The free surface elevation and the magnitude velocity are predicted and
analyzed for the different cases.
2 Geometry System
As shown in figure 1, the considered geometry is a rectangular tank. The geometry

is defined by the length L=0.6 m and the height equal to H=0.6 m. The tank
subjected to a sinusoidal excitation is partially filled with liquid with the height
h=0.1 m.
3 Numerical Method
The commercial CFD code "Fluent" has been used in the numerical simulation of
the liquid sloshing phenomenon in the tank. The VoF method (Volume of Fluid) is
Numerical Study of the External Excitation Amplitude Effect 211
used to define the deformation of the free surface over time. For the motion of the
tank, a User Defined Function (UDF) is developed on language “C++” and inter-
preted in “Fluent”. The Navier-Stokes equations and the standard k-ε turbulence
model equations were solved using a finite volume discretization method.
The problem of liquid sloshing is governed by the continuity and the momen-
tum equations as follows:
∂ρ ∂
+ ( ρ ui ) = 0 (1)
∂t ∂xi
∂ ∂ ∂p ∂ ∂u ∂u j
( ρ ui ) + ( ρ ui u j ) = − + μ( i + ) + Fi (2)
∂t ∂x j ∂xi ∂x j ∂x j ∂xi
Where xi are Cartesian coordinates, ui are the corresponding velocity components,

p is the pressure, ρ is the density, µ is the viscosity and Fi is the external body
force due to the gravity and the external excitation. The sinusoidal movement is
defined in the x direction as follow:
X = Asin( wt ) (3)
In this work, the effect of the amplitude A has been investigated. Four different
values of A are considered (table 1).
Table 1 Amplitude values
Amplitude Value (m)

A1 0.1
A2 0.2
A3 0.3
A4 0.4
The angular frequency of the sinusoidal movement in this study is equal to w=

3.14 rad.s-1. According to Akyildiz et al. (2006), the naturel frequency is given by
the equation (4):
nΠ g nΠ h
wn = tanh( ) (4)
L L
Where n is the mode number, g represents the gravity, L represents the length of
the tank and h is the water depth. The first natural frequency is obtained for n=1.
The ratio between the first natural frequency and the chosen frequency in this pa-
per is:
w
= 0.63 (5)
w1
The free surface is defined as the zero level set of a level-set function Ф: in air,
is set to be positive and in water, to be a negative function as follows:
Ф ( xi , t ) > 0 , in air
Ф ( xi , t ) = 0 , on surface (6)
Ф ( xi , t ) < 0 , in water
Differentiating Ф with respect to time, a transport equation is derived to de-

scribe the free surface motion in the form:
∂Ф ∂Ф
+ ui =0 (7)
∂t ∂xi
Where ui is the local fluid velocity. Moving the interface is equivalent to updating
Ф by solving equation (7).
For the boundary conditions, a no-slip boundary condition is imposed by set-
ting the components of velocity on the wall to zero as given by:
u = v = w = 0 (8)
4 Numerical Results
4.1 Free Surface Evolution

Figures 2, 3, 4 and 5 shows the evolution of the free surface over time respectively
at t=T/4, t=T/2, t=3T/4 and t=T for the different value of amplitude considered in
this study. According to these results, it appears that the evolution of the free sur-
face depends on the amplitude excitation. For the different instances, the results
(a) A=A1 (b) A=A2
(c) A=A3 (d) A=A4

Fig. 2 Free surface evolution at t=T/4
show that the violence of the sloshing increases with the increase of the amplitude
value. The most free surface deformation has been observed for A=A4 and the
minimum for A=A1. For the t=T/4, the liquid moves from right to left. For t=T/2
and t=T, when the external excitation changes the sense, the liquid appears stable
and the free surface evolution has been observed minimal. However for t=3T/4, it
moves from the left to the right.
(a) A=A1 (b) A=A2
(c) A=A3 (d) A=A4

Fig. 3 Free surface evolution at t=T/2
(a) A=A1 (b) A=A2
(c) A=A3 (d) A=A4

Fig. 4 free surface evolution at t=3T/4
(a) A=A1 (b) A=A2
(c) A=A3 (d) A=A4

Fig. 5 free surface evolution at t=T
4.2 Magnitude Velocity

The magnitude velocity is presented in figures 6, 7, 8 and 9 at t=T/4, t=T/2, t=3T/4
and t=T respectively and for different values of the amplitude. According to these
results, it has been noted that the motion of the liquid in the tank generates the ap-
pearance of a recirculation zones. Globally, the recirculation zones appear in the
free surface. Over time, the motion of the liquid in the tank becomes more violent
(a) A=A1 (b) A=A2
(c) A=A3 (d) A=A4

Fig. 6 Velocity field at t=T/4
and the recirculation zones become important for the different amplitudes. The re-
sults show that the value of the velocity depends on the amplitude value. It in-
creases with the increase of the amplitude. Except at t=T/4, the velocity for A1 is
more important than A2 and A3. The location of the maximum value of the veloci-
ty varies over time and also according to the external excitation amplitude.
(a) A=A1 (b) A=A2
(c) A=A3 (d) A=A4

Fig. 7 Velocity field at t=T/2
(a) A=A1 (b) A=A2
(c) A=A3 (d) A=A4

Fig. 8 Velocity field at t=3T/4
(a) A=A1 (b) A=A2
(c) A=A3 (d) A=A4

Fig. 9 Velocity field at t=T
The evolution of the static pressure over time is presented in figure 10 for the dif-
ferent values of amplitude considered in this study A1, A2, A3 and A4. According
to these results, it has been observed that the static pressure varies periodically
with time. These results are expected since the tank is subject to a periodic sinu-
soidal movement. Indeed, it is clear that the static pressure increases with the
increase of the external excitation amplitude. The maximum value has been ob-
served for A=A4. However, the minimum pressure is obtained for A=A1.
p (Pa)
t (s)
Fig. 10 Static pressure profile

The period of the static pressure variation depends also on the external excitation
amplitude. For the case of A=A1, the numerical results are compared with the ex-
perimental results of Panigrahy at al. (2009). According to these results, a good
agreement was shown which confirms the validity of our numerical model.
6 Conclusion
In this work, we have studied the external excitation amplitude effect on liquid
sloshing phenomenon in a rectangular tank. The tank is subjected to an external
sinusoidal excitation. Our numerical results are compared with experimental re-
sults and a good agreement is shown. The free surface evolution, the distribution
of the magnitude velocity and the static pressure profile are presented and ana-
lyzed. It appeared that the motion of the liquid in the tank subjected to the exter-
nal excitation considerably depends to the value of the amplitude. When the
amplitude value of the external motion increase, the sloshing become more impor-
tant. The static pressure also depends on the amplitude value. It increases with the
increase of the amplitude value. For the velocity, it appears that the recirculation
zones become more important over time and generally located on the free surface
of the liquid. The location of maximum value of the velocity changes over time
and with the amplitude value. The highest value of the velocity has been observed
for the maximum value of amplitude. However, the minimum value has been
noted for the minimum amplitude value.
In the future, we propose to study the liquid sloshing phenomenon in a battery cell
with a mixing element.
References
Akyildiz, H., Ünal, N.E.: Sloshing in a three-dimensional rectangular tank: numerical simu-
lation and experimental validation. Ocean Engineering 33, 2135–2149 (2006)
Bouabidi, A., Driss, Z., Abid, M.S.: Vertical Baffles Height Effect on Liquid Sloshing in an
Accelerating Rectangular Tank. International Journal of Mechanics and Applications 3,
105–111 (2013)
Ji, Y.M., Shi, Y.S., Park, J.S., Hyun, J.M.: Experiments on non-resonant sloshing in a rec-
tangular tank with large amplitude lateral oscillation. Ocean Engineering 50, 10–22
(2012)
Liu, D., Lin, P.: Three dimensional liquid sloshing in a tank with baffles. Ocean Engineer-
ing 36, 202–212 (2009)
Jin, H., Liu, Y., Li, J.H.: Experimental study on sloshing in a tank with an inner horizontal
perforated plate. Ocean Engineering 82, 75–84 (2014)
Koh, C.G., Luo, M., Gao, M., Bai, W.: Modelling of liquid sloshing with constrained float-
ing baffle. Computers and Structures 122, 270–279 (2013)
Panigrahy, P.K., Saha, P.K., Maity, U.K.: Experimental studies on sloshing behavior due to
horizontal movement of liquids in baffled tanks. Ocean Engineering 36, 213–222 (2009)
Yan, G., Rakheja, S., Siddiqui, K.: Experimental study of liquid slosh dynamics in a partial-
ly-filled tank. J. Fluids Eng. 131(7) (2009)
The Study of the Hardening Precipitates and the
Kinetic Precipitation. Its Influence on the
Mechanical Behavior of 2024 and 7075
Aluminum Alloys Used in Aeronautics
Ahmed Ben Mohamed, Amna Znaidi, Moez Baganna, and Rachid Nasri
LR-MAI-ENIT BP37 Le Belvedere 1002 Tunis TUNISIE

Ahmed82enit@yahoo.fr, Amna.znaidi@laposte.net,
Rachid.nasri@enit.rnu.tn
Abstract. The Research about the aluminum alloys has been in progress. Alumi-
num alloys, in contrast to the pure aluminum have very high mechanical characte-
ristics of substantial resistance. These mechanical performances combined with
lightness of the structure there of has a variety of uses especially in the transport
sector. The aluminum alloys of the 2000 series (Cu is a main part of addition) and
7000 (Zn is the main part of addition) are essentially aeronautics alloys, hence its
strong presence Air in the design of bus where Al 2024 and Al 7075 are the major
part (Figure 1).
Both alloys (2024-7075) are hardening structural alloys and not by strain work-
ing. Their mechanical properties are of particular microstructure obtained from
thermo-mechanical treatments. Among other things, this is a complicated micro-
structure which is responsible for these mechanical performances. It is interesting
to examine the description of this complex phenomenon of the hardening precipi-
tation, including areas of Guinier-Preston (GP Zone) which are very dependent on
the alloy’s composition and thermo-mechanical treatments they undergo. The
hardening phases (A.Deschamps2001) of 2024 and 7075 aluminum alloys are re-
spectively (Al2Cu - Al2CuMg) and (MgZn2). These phases are obtained by preci-
pitation of the sequence which depends on the proportions of elements alloying
elements. An experimental tensile tests companion of these heat treated aluminum
alloys (quenching) will allow us to understand the kinetic precipitation and its in-
fluence on the mechanical properties (Rp0.2, Rm and A%), dice quenching cool
until maturation (T4).
Keywords: 2024 and 7075 Al, thermo-mechanical treatments, experimental

mechanical testing, kinetic precipitation, maturation T4.
1 Introduction
To study the influence of thermal treatment on the mechanical properties of alu-

minum alloys, a companion tensile test is performed on specimens cut of Al 2024

220 A.B. Mohamed et al.
7075 following materials and different load tests, which will later on visualize the
anisotropy of these materials and to accurately describe the evolution of hardening
precipitates that are responsible for the evolution of the mechanical properties of
the aluminum alloys. These experimental tests are carried out for different direc-
tions of test pieces in the following sequences ( fresh quenching, after 1 hour of
quenching, after 2 hours of quenching, one day after quenching and after 7 days of
quenching). The 2024 and 7075 aluminum alloys are hardening alloys (age har-
dening) which makes interesting description of the evolution of hardening precipi-
tates and their effects on the mechanical properties of alloys to achieve maturation
and stabilization.
Fig. 1 Presence of aluminum alloys in the airplane

The Study of the Hardening Precipitates and the Kinetic Precipitation 221
2 Study and Synthesis
2.1 Material Description

Table 1 Chemical composition and mechanical properties of Al 2024 and 7075 1
Aluminum Chemical compoisition mechanical properties
Al 2024 Cu Mn Mg Rp0.2 Rm A%
T3 3.8- 4.9 0.3-0.9 1.2-1.8 300MPa 440MPa 9
Al 7075 Cu Mg Zn Rp0.2 Rm A%
T7 1.2- 2.0 2.1-2.9 5.1-6.1 390MPa 475MPa 7
2.2 Precipitates and Kinetic Precipitation

The oversaturated solid solution of the aluminum alloy obtained by solution
treatment, followed by quenching, never gives almost direct effect (fresh quench)
to the precipitation of the stable equilibri the um of intermetallic phase. To reach
this stage of maturation or natural aging, the alloy gradually hardens and the solid
solution decomposes to give rise to the formation of small clusters of solute atoms
(GP zone) in high density: it’s the precipitation. The GP zone introducing elastic
deformations in the crystal lattice of the matrix (CFCs) promotes the precipitation
the presence of crystal defects. The hardening at room temperature usually results
in a sequential decomposition of the solid solution and a progressive depletion of
the solute thereof.
It is worth mentioning that the sequence of precipitation of the hardening 2024
depend the proportions of copper grades of aluminum Al alloys. For the aluminum
alloy Al 2024 (4% Cu) precipitation sequence is as follows (J. BARRALIS 2002):
Initial solid solution → GP zones → Ө'' → Ө' → Ө-Al2Cu (Figure 2).
• The GP zones: disc plans are similar parallel to the {100} plans of the alumi-
num matrix.
• Phase Ө'': coherent phase precipitates as parallel plates to {100} plans of the
aluminum matrix.
• Phase Ө': semi-coherent Phase semi-coherent.
• Phase Al2Cu: incoherent Phase.
Fig. 2 Precipitation sequence
3 Results of the Mechanical Tests and Interpretations
Our source of experimental information is a range of tensile tests carried out on

samples of Al 2024 and Al 7075 after different directions of loading (0 ° direc-
tion: rolling, direction 45 ° and direction 90 °) and from the fresh quenching to
the maturation (T4), and this after heat treatment quenching. Tractions curves of
aluminum alloys in the commercialized state (Al 2024 and Al 7075 T351 T73)
have been completed and will be represented as well.
The frequency of production of mechanical tensile tests for each direction and
for each material is as follows:
fresh quenching (T fraiche).
After 1 hour of quenching (Ap 1h).
After 2 hours of quenching (Ap 2h).
After one day of quenching (Ap 1j).
After 7 days of quenching (Ap 7j).
We will reach 7 days only for the rolling direction (0 ° direction). For the other
directions, we will follow the tests to two hours after quenching.
3.1 Experimental Results

- Curves traction Al 2024 for 0 ° direction :
Fig. 3a
- Curves of Al traction in 2024 for 45 ° direction:
Fig. 3b
- Curves of Al traction in 2024 for 90 ° direction :
Fig. 3c
- Curves of Al traction in 7075 for 0° direction :
Fig. 4a
Fig. 4b
Fig. 4c
- Evolution of Rp0,2 for Al 2024 :
Fig. 5a
- Evolution of Rm for Al 2024 :
Fig. 5b
- Evolution of A % for Al 2024 :
Fig. 5c
- Evolution of Rp0,2 for Al 7075 :
Fig. 6a
- Evolution of Rm for Al 7075 :
Fig. 6b
- Evolution of A % for Al 7075 :
Fig. 6c
3.2 Interpretation of Experimental Results

The change of direction of loading specimens for experimental tensile tests per-
formed (Fig. 3 and 4) have a remarkable influence on the mechanical characteris-
tics (Rp0.2, Rm and A%). Microstructural evolution and precipitation hardening
in the time even showed a relevant effect of the studied mechanical properties (fig.
5 and 6).
Indeed, this regression is due to the dispersion of the main elements of addition
(Cu and Zn in 2024 to 7075) in the weight of the aluminum alloy material ,
atoms to form consistent with the alloy matrix of substitution (fig. 2).
During curing and from the fresh quench up to 7 days of aging (T4) atoms of
the alloying elements are clustered to form incoherent precipitates with the matrix
(Fig. 2). These precipitates cause the microstructural evolution by generating cur-
ing increased studied as figures (fig. 5 and 6) shows mechanical properties. We
can say that the response of aluminum alloys is dependent on the loading direction
of the specimens in the first place and time of curing and maturation parameter in
the second place.
The curves of figures (Fig. 5 and 6) show that during the evolution of the me-
chanical properties of aluminum alloys, the increase is almost absent between 1
hour and 2 hours of fresh quenching. This bearing is reduced due to the number of
dislocations in the crystal structure of aluminum. This slowed the movement of al-
loying elements to the agglomeration, which weakened the evolution of mechani-
cal properties.
After 2 hours of fresh quenching, increasing the mechanical characteristics be-
comes more pronounced to achieve maximum values at 7 days of maturation in
which the materials have reached their equilibrium microstructure. For against, Al
7075 behaves differently. Ductility (A%) of this material greatly increases the
fresh quench. The ductility of the Al 7075 reaches its maximum value between 1
hour and fresh quenching by varying the loading direction of the specimens. The
change in mechanical properties and especially the ductility (A%) allowed us to
know the answers as well as the plastic deformation behavior of its aluminum al-
loys.
4 Conclusion
In this paper, we studied the behavior of 2024 and 7075 aluminum alloys sub-
jected to tensile tests in different directions. The change in directions of loading of
the specimens and the precipitation kinetics showed a significant influence on the
mechanical characteristics of these aluminum alloys.
From these interpretations, we can summarize the behavior of aluminum Al
2024 and Al 7075 plastic deformation which is a basic process for making alloys
in the aerospace industry:
Al 2024 is a material forming and deep drawing by excellence and especially to
its equilibrium state T4 (after 7 days of fresh quenching). While the 7075 Al is
better for plastic forming and deep drawing between fresh quenching and 1 hour
to fresh quench.
This work will allow us later to better optimize the behavior of aluminum al-
loys to plastic deformation and forming by deep drawing by adopting new identi-
fication behavior laws and calculating the coefficients of anisotropy strategies and
hardenings. Further visualizations and by transmission electron microscopy mi-
crographs explorations (MET) in these types of alloys will be discussed.
References
Barralis, J., Maeder, G.: Précis métallurgie: élaboration, structure, propriété, normalisation.
Édition Afnor/Nathan, Paris (2002)
Wilm, A.: Recherche sur la métallurgie physique des alliages d’aluminium-magnésium.
Métallurgie 8, 225 (1911)
Deschamps, A., Militzer, M., Poole, W.J.: Precipitation kinetics and hardening in a Fe-Cu
alloy. ISIJ International (Japan) 41(2), 196–205 (2001)
Deschamps, A., Bigot, A., Auger, P., Bréchet, Y., Livet, F.: A comparative study of preci-
pitate composition in Al-Zn-Mg using Tomographic Atom Probe and Small-Angle X-
Rays Scattering. Phil. Mag. A 81(10), 2391–2414 (2001)
Dumont, D., Deschamps, A., Bréchet, Y.: On the relationship between microstructure,
strength and toughness in AA7050 aluminum alloy. Materials Science and Engineering:
A00 1-/11 (2003)
Znaidi, A.: Stratégie d’identification de lois de comportements anisotropes: Applications
aux alliages d’aluminium. Toulouse (2010)
Znaidi, A.: Formulation et identifications des lois de comportements anisotropes d’un al-
liage d’aluminium 2024T3. Marseil (2009)
Baganna, M., Ben Mohamed, A., Znaidi, A., Guellouz, S.: Nasri, Etude des alliages
d’aluminium utilisés dans le domaine aéronautique: comportement mécanique de
l’alliage d’aluminium, T351. Cotume, Tunisie (2012)
Multidisciplinary Design Optimization in Small
Satellite
Abdelmadjid Boudjemai1, Abdelmoumen Bacetti2, Mohammed Amine Zafrane2,

and Rachida Hocine3
1
Centre of Satellite Development (CDS), Space Technology Research Division BP.: 4065
Ibn Rochd USTO Oran, Algeria
a_boudjemai@yahoo.fr
2
Department of Electronics Engineering, Faculty of Electrical Engineering, BP 1505 EL
M’Naour, University of Sciences and Technology of Oran Algeria
bmoumen27@yahoo.fr,
amine.eln@hotmail.com
3
Department of Automation Engineering, Faculty of Electrical Engineering, BP 1505 EL
M’Naour, University of Sciences and Technology of Oran Algeria
rak_hocine@yahoo.fr
Abstract. Many industrial tools for simulation developed during these last years
still have an impact limited on the complex systems design such as satellites, air-
crafts, engines or rockets. Whereas the modeling tools of each discipline or sub-
systems are refined gradually, the total optimization of these systems encounters
difficulties of orders at the same time methodological, organizational, data-
processing and numerical. Actually and in the best of the cases, the complex sys-
tems are optimized by subsystems. Savings of time of design and performances
are accessible if one manages to optimize in a collaborative and robust way the
unit of the subsystems. The aerospace structures generally require light designs.
The goal of these later is to be optimized the force by weight or effectiveness of
the design. The satellite structural design developed considerably during the last
four decades. Traditionally, the effectiveness was achieved by a combination of
various designs and structural materials. Our objective is to build methods and
tools for the robust design and multidisciplinary optimization in the small satel-
lites mechanics design, taking into account the aspects related to attitude and orbit
control subsystem, thermal control, static, dynamic and thermo-elastic analysis
and including the design of mechanical housing for RF systems. Our proposed
method is demonstrated through multidisciplinary design optimization application.
In this study, the multidisciplinary design optimization in small satellite mechani-
cal design is formulated as optimization of sub problem involving the appropriate
selection of a configuration. The different results obtained in this paper show that
the multidisciplinary design optimization methods present a large interest in the
space design with an aim of having a reliable space system with lower cost.
Keywords: Small Satellite, Multidisciplinary Design Optimization, Genetic Algo-

rithm, Honeycomb, QFH Antenna, Subsystem.

230 A. Boudjemai et al.
1 Introduction
The space vehicle design process is complex and difficult due to the physical and
operating constraints, uncertain environments, and extreme operating conditions].
The Multidisciplinary design optimization (MDO) approaches have significant ef-
fects on aerospace vehicle design methodology]. The complexity and the multidis-
ciplinary design make it difficult to have a complete survey and a deep under-
standing of the whole design process. A recent step forward towards a
multidisciplinary approach in the early phases of space system design has been
achieved by concurrent engineering (Blaine 2004).
One of the most important considerations in designing the space vehicle is
weight. By reducing the weight of a space vehicle, it is possible to increase the
payload. In many industrial applications, reducing the weight of a structure with-
out compromising its strength and stiffness is considered as one of the most im-
portant design criteria (Kalavalapally et al. 2006). Today, the search of the best
performance, best quality, and the best cost for space vehicles became a complex
process. The required optimum in a total way, on the level of the system, implies
choices of compromise between the various elements which make it up in order to
answer increasingly many and sometimes contradictory requirements. This con-
cern the Multidisciplinary Optimization (MDO) hence of many disciplines (me-
chanisms, thermal, control of flexible structure, propulsion, cost, structure, opto-
mechanical design, functional architecture, and architecture system) must be taken
into account. These disciplines interact in a very complex way, often according to
some criteria's, and the objective of the MDO consists in optimizing them simul-
taneously. The optimum must be found globally, at the system level, by trade off
choices among the various systems’ components, in order to achieve homogeneity
and perfect integration. Multidisciplinary design optimization (MDO) has
emerged as an engineering discipline that focuses on the development of new de-
sign and optimization strategies for complex systems. MDO researchers strive to
reduce the time and cost associated with the coupling interaction among several
disciplines. As stated in (AIAA Technical Committee for MDO 1991) decomposition
approaches provide many advantages for the solution of complex MDO problems,
as they enable a partitioning of a large coupled problem into smaller, more mana-
geable sub-problems. The resulting computational benefits, besides the obvious
one associated with the solution of smaller problems, include creating a potential
distributed processing environment.
The primary benefit, however, pertains to the savings in personal hours, be-
cause groups are no longer required to wait around for other groups in the process
to complete their design tasks. Considering the number of disciplines involved and
the level of knowledge acquired in each of them, the "engineer’s trade" in the
sense of man’s capability to master all of the disciplines and their interactions is
no longer enough. It is now necessary to develop design and optimization tech-
niques based on mathematical tools, which allow effective integration of the me-
thods used and developed in the various disciplines. These techniques can then be
Multidisciplinary Design Optimization in Small Satellite 231
used in searching for the best tradeoffs leading to a complete optimized system.
This study develops a system sensitivity analysis method to optimize multidiscip-
linary design of space vehicle. Suitable design variables, technological, and func-
tional constraints are considered. Appropriate combinations of disciplines such as
space vehicle layout, thermal, structure; opto-mechanical design, thermo elastic
analysis and electronics are used.
2 Multidisciplinary Design Optimization for Space Vehicles
The application of multidisciplinary design optimization to space vehicle design

has the potential to significantly improve decision making capabilities early in the
design process. Successful integration of optimization and system-level space ve-
hicle tools allow the concept architecture, technology choices, and performance
requirements to be adjusted to meet an overall estimated cost goal; thereby enabl-
ing space vehicle design to be completed in a reliable, economical and timely
manner (AIAA Technical Committee for MDO 1991). Successful engineering design
processes of complex systems require that numerous design variables and con-
straints be taken into account across multiple disciplines. The systematic modifi-
cation of design parameters relying on judgement in a manual design process is
often ineffective, and the benefits of computer-aided design optimisation can re-
duce design time, improve design through improved methodology, solve complex
interactions and ultimately reduce the cost of design.
Multidisciplinary design problems require an optimiser capable of efficiently
handling local minima in non-linear and discontinuous design spaces of high di-
mensionality. Traditionally optimisers rely on a good starting point to obtain a
solution or even to converge, thus an additional requirement is that an acceptable
solution be found without a good initial starting point to the optimisation. Fur-
thermore, the optimiser must be robust, as the computational expense of objective
function calculation makes convergence on non-optimal solutions unacceptable.
Typically the multidisciplinary design optimization in space vehicle mechanical
design approach is shown on the figure 1. The mechanical engineering discipline
addresses all aspects of the mechanical design of space products, where mechani-
cal in this context includes structural, thermal and material selection aspects,
propulsion for spacecraft and launch vehicles, pyrotechnic and environmental con-
trol/life support functions, and mechanical parts, interfaces and interconnections.
Considering the system design process the main challenge is managing the sys-
tem dependencies as shown in figure 2 to get a focus on the crucial matters and
subsystems.
Structural
Structural Attitude, orbit determination Software
mechanics and control
mechanics Reliability
Reliability
Optimization
Optimization Operations and
Geometry
Geometry andand Data handling ground systems
satellite
satellite layout
layout Instruments
Fluid
Fluid Electrical power Telemetry
mechanics
mechanics Wet mass Tracking and
Dry mass command
multidisciplinary
multidisciplinary design
design
optimization
optimization Thermal control
Materials
Materials
Mission analysis Launch mass
Mechanisms
Mechanisms andand
Structure and
control
control systems
systems mechanisms
Propulsion Adapter
Opto-mechanics
Opto-mechanics Heat
Heat transfer
transfer
Propellant mass
Fig. 1 Mechanical Multidisciplinary De-

Fig. 2 System dependencies
sign Optimization Approach
3 Multidisciplinary Design Optimization Approach
In this study, the small satellite design problem is formulated as optimisation of

sub problem involving the appropriate selection of a configuration as well as the
analysis of mission and performance objectives (ground resolution, elevation an-
gle, etc.) The mathematical model is developed specifically for space vehicle
such as small satellite remote sensing missions. The problem is represented
graphically in figure 3. The objective function of maximizing performance per
unit cost within specific launch vehicle constraints is chosen for formal optimiza-
tion. In general form, the small satellite multidisciplinary Optimization design
problem can be written as:
Multidisciplinary Design Optimization (MDO) is a methodology for the design
of complex engineering systems and subsystems that coherently exploits the syn-
ergism of mutually interacting phenomena (AIAA Technical Committee for MDO
1991).
Δ Design = (∑ Δ
i Discipline i
)+ Δ MDO
(1)
The proposed algorithm is given on figure 3. This makes our process a powerful
MDO technique because of:
- The quality of the solution obtained compared to the selected criteria of
optimization,
- The robustness of the solution with respect to uncertainties of the models
and the variations of parameters of design,
- The CPU time in the process of optimization.
The general form of a traditional MDO problem can be summarized in the
following way:
min imize the function : f(z, x, y(x, y, z))

given : z, x
such us : g(z, x, y(x, y, z)) ≤ 0
h(z, x, y(x, y, z)) = 0 (2)
Where:
– f : minimization criterion,
– z : shared variables between different disciplines,
– x : local variables of each discipline,
– y : coupling variables between different disciplines,
– g : problem equality constraints,
− h : problem inequality constraints.
Several algorithms already exist are used to solve MDO problems like gradient
based algorithms, game theory, genetic and evolutionary algorithms, multi-agent
systems and many other numerical methods which existed before even the appear-
ance of the MDO.
We can gather the various methods in two categories: single level and multi-
level. For the Single level methods, these methods use only one optimizer and
consider the disciplines as analysis tools.
Among these methods (Clément 2009): MDF (Multi-Discipline Feasible)], IDF
(Individual Discipline Feasible), AAO (All At Once). The multilevel methods
use a global optimizer on the system level and a local optimizer for each disci-
pline. We finds in this category: CSSO (Concurrent SubSpace Optimization),
BLISS (Bi-Level Integrated Systems Synthesis), Collaborative Optimization.
systemlevel
system levelDesign
Design
System level analysis Coordination
System level analysis Coordination
Design Vector
Experts
Experts
⎡x1 ⎤ TTC & command
⎢ ⎥
⎢x 2 ⎥
Thermal
Knowledge Acquisition
⎢ . ⎥
Knowledge Acquisition
Component
Propulsion Component
⎢ . ⎥
Input
Input
⎢ ⎥
AODCS
Interactive Component
Interactive Component
Explanation
⎢ . ⎥
Explanation
Power Component
Component
Users
⎢ . ⎥
Users
Knowledge
⎢ ⎥
Structures & Knowledge
mechanisms base
base
Inference
⎣x n ⎦ Inference
Component
Component
Opto-Mechanical (Instruments)
Output
Output
Objectif
Objectif vector
vector
Fig. 3 Multidisciplinary System Design Optimization of Small Satellite
4 MDO of a QFH antenna for Alsat-1 first Algerian satellite
Alsat-1 is an Algerian earth observation satellite (90kg) which evolves in a sun-

synchronous retrograde circular orbit (SSO). Designed to be part of a constellation
for daily disaster monitoring, Alsat-1is equipped with two banks of cameras
giving a total of 600km field of view at 32 meters ground sampling distance in

three spectral bands: Red, Green and Near Infra-Red. In November 2002, the
28th, at 06:07am GMT, Alsat-1was successfully launched by COSMOS-3M
from the cosmodrome of Plesetsk in Russia into a 700 km SSO orbit and is now
fully operational with five-year satellite mission lifetime (Boudjemai 2007).
Alsat-1 is three axes stabilized earth observation satellite in imaging mode and
evolves in a BBQ mode out of imaging time. The attitude determination and con-
trol subsystem gives a good attitude pitch/roll/yaw stability during imaging (=5
m°/s) and the Orbit filter provides a maximum track error of the scene position of
5km (GPS On during one orbit a day).
The QFH Antenna was used in commercial, military and amateurs satellites.
The basic form of the quadrafilar antenna was developed by Dr C. C. Kilgus of the
Applied Physics Laboratory, Johns Hopkins University, who has published several
papers that establish the theoretical basis for its operation (Kilgus, 1968). The ad-
vantage of the QFH antenna is due to (1) its small geometrical occupation, (2) it
practically radiates in a hemisphere, (3) it results a light weight antenna compared
to antennas sharing the same radiation characteristics (e.g. circularly polarized
crossed dipole over a ground plane system), (4) A very significant characteristic is
that the use of the two types of antennas one in emission and other in reception
makes possible to obtain a relatively constant field during the visibility of the sat-
ellite to the ground station.
Figure 4 shows QFH antenna for Alsat-1 first Algerian microsatellite built and
constructed by SSTL team (Aglietti 2002). The Alsat-1 QFH antenna is essentially
composed of: Conical base (PEEK), antenna mast, collar with viscoelastic mate-
rial, plastic cap.
Plastic cup
Mast
Helixes
Central helix spacer
Brass tube
Collar (with RTV potting

For base of mast)
PEEK base cone
Fig. 4 QFH antenna flayed on Alsat-1
Our proposed method is demonstrated through multidisciplinary design optimiza-

tion application.
The figure 5 shows the MDO framework of Alsat-1 QFH antenna. We consider
in this problem the radiation aspect, mechanics, launcher specification, the aero-
thermodynamics effect, mass and cost.
The decomposition of the previous problem with respect to the CO approach is

illustrated in figure 6.
Fig. 5 QFH antenna design problem Fig. 6 QFH antenna CO optimizer architecture
The global optimizer developed in order to minimize the total mass of the an-
tenna (mass of helices and mass of the support supposed as a hollow cylinder –
monocoque structure). The formulas used for the mechanical optimizer are de-
scribed in (Maxwell 2001) by using the Matlab built-in function “fmincon” used
for such constrained nonlinear problems. The radiation pattern optimizer is based
on a FORTRAN genetic algorithm combined with a nonofficial Numerical Elec-
tromagnetic Code which is largely used for antenna calculation (Burke, 1981).
Our optimization tool use two databases, one for launchers (Falcon1, Delta
II/IV, Soyuz, Dnepr, Start1 and LM2C) and a second for materials (1- Al 7075, 2-
Al Alloy 2024 T, 3- Austenitic Stainless Steel 304, 4- Ti Alloy Ti-6Al-4V and 5-
Graphite Epoxy).
Fig. 8 Geometry of the quadrifilar

helix antenna with cardbord sup-
Fig. 7 QFH Optimizer GUI
port
The figure 7 presents the GUI designed to enter the design variables (antenna
operating frequency, orbit height and launcher) and shows values used in this test.
With the values shown in figure 7, we obtained the configuration shown in figure
8 and a radiation pattern illustrated in figure 9.
Figures 10 to 12 present parameters evolution in 9 iterations. We can note that
all constraints are respected except the VSWR value which is equal to 1.2. In spite
of that, the resultant antenna presents a small footprint (3.6cm of diameter and
18.75cm of height) and has a light weight (294.7gr). We also remarque that the
thickness of the support is equal to its diameter, it means that this support is not a
monocoque structure but it’s a cylinder. With this configuration (figure 8), we got
natural frequencies equal to 84.66 Hz/lateral and 665.43Hz/axial which are greater
than launcher natural frequencies. The material of the support is graphite epoxy.
We should remarque those three parameters still constant during all the 9 itera-
tion. Those parameters are the antenna helices diameter, the antenna pitch distance
and the antenna number of turns. According to (Larson and Wertz 2005), the two
parameters are dependant and the number of turns must be greater than 1.25 to
have a conical shaped pattern. This configuration is sufficient to resist to mechani-
cal choc due to launcher’s rocket motors load and can be used practically
independently to available space both in the satellite and the launcher’s fairing be-
cause the small dimensions, it results a reduced cost compared to other types of
antennas.
Fig. 9 Alsat-1 QFH antenna radiation pat-Fig. 10 Evolution of system parameters.

tern.
Fig. 11 Evolution of antenna parameters
Fig. 12 Evolution of support parameters
6 Conclusion
Recognizing the need for multidisciplinary optimisation method, the research pre-
sented in this paper introduced a new approach for multidisciplinary optimization
in space vehicle and in the particular in small satellite mechanical design for re-
mote sensing missions. The multidisciplinary design optimisation methods present
a large interest in space vehicle design and in particular the related aspects with
the small satellite mechanical design. This is an important characteristic of the
conceptual design for high-cost systems. Collaborative Optimization approach
was used to solve a QFH antenna design problem. To have a reliable configuration
of the antenna, space environment constraints were taken into account. The an-
tenna design respects both radiation pattern and mechanical constraints. The opti-
miser developed during this study presents an environmental aided tool for the
multidisciplinary design optimization of small satellite by integrating all the ex-
perience gained from the previous space projects in order to obtain a powerful
product and minimize the design risk.
References
Aglietti, G.S.: QFH qualification vibration test report, SSTL internal report (2002)
AIAA Technical Committee for MDO. Current State of the Art of Multidisciplinary Design
optimization, AIAA White Paper, approved by AIAA Technical Activities Committee,
Washington, D.C. (September 1991)
Boudjemai, A., Bouanane, M.H., Merad, L., Si Mohammed, M.: A Small Satellite Structur-
al Optimisation Using Genetic Algorithm Approach. In: 3rd International Conference on
Recent Advances in Space Technologies (2007), doi:10.1109/RAST.2007.4284021, 1-
4244-1057-6
Blaine, B.T.: Multifunction structures in satellite design: thermoelectric elements embedded
into structural honeycomb panels. Dissertation, Santa Clara University (2004)
Burke, G. (1981), Numerical Electromagnetic Code developed at Lawrence Livermore Lab
& Livermore CA (January 1981), http://www.si-list.net
Clément, J.: Optimisation multidisciplinaire: étude théorique et application à la conception
des avions en phase d’avant projet. Dissertation, Toulouse University (2009)
Kalavalapally, R., Penmetsa, R., Grandhi, R.: Multidisciplinary optimization of a
lightweight torpedo structure subjected to an underwater explosion. Finite Elements in
Analysis and Design (2006), doi:10.1016/j.finel.2006.07.005.
Kilgus, C.C.: Multielement Fractional Turn Helices. IEEE Transactions on Antennas and
Propagation (1968), doi:10.1109/TAP.1968.1139231
Kilgus, C.C.: Resonant Quadrafilar Helix. IEEE Transactions on Antennas and Propagation
(1969), doi:10.1109/TAP.1969.1139459
Larson, W.J., Wertz, J.R.: Space Mission Analysis and Design, the space technology li-
brary. Published jointly by Microcosm Press and Kluwer Academic Publishers (2005)
Mechanical Vibration Cancellation
Using Impact Absorber
Kaouther Chehaibi, Charfeddine Mrad, and Rachid Nasri
Laboratory of Applied Mechanics and Engineering (LMAI),

National School of Engineers of Tunis (ENIT), University of Tunis El Manar (UTM),
BP 37, Le Belvedere, 1002, Tunis, Tunisia
ChehaibiKaouther@yahoo.fr,
{Charfeddine.Mrad,Rachid.Nasri}@enit.rnu.tn
Abstract. The free and forced vibrations of a mechanical system equipped with a
ball absorber were studied. The modeling of the impact damping and the problem
formulation and resolution were conducted. The effect of the absorber parameters
on the vibration attenuation of the primary system was then examined. The absor-
ber parameters considered are: the mass, the stiffness, the damping, the clearance,
and the impact velocity. The excitation parameters were also evoked. Physical and
mathematical modeling, then numerical simulation led to determine the free and
forced responses of the considered system equipped with the impact absorber, for
different situations. This helped to identify the absorber favorable parameters,
leading to reduce considerably the primary system vibration, and to advance de-
sign recommendations.
Keywords: mechanical systems, free vibration, forced vibration, impact absorber,

vibration attenuation.
1 Introduction
The ball absorber is considered as a small mass linked to a primary system. The
vibration attenuation is due to the impacts between the two masses leading to vi-
bratory energy dissipation (Marhadi and Kinra). The impacts are modeled by a
spring and a viscous damper (Cheng and Xu).
Ball absorbers are used to damp excessive vibrations, as for turbine blades and
machine tools: analytical, numerical, and experimental studies were conducted
(Cheng and Wang, Ema and Marui). Also, ball absorbers were used to decrease
industrial machine noise (Xu et al.). A well designed ball absorber is more effi-
cient, less complicated to realize, and more robust to parameters changes then a
classic dynamic absorber (Sankar and Bapat). Other experimental studies on ball
absorbers, also called impact absorbers, were also elaborated (Masri, Desen).
In this work, we are interested in studying the free and forced responses of a
mechanical system equipped with an impact absorber. Firstly, we present the vi-

240 K. Chehaibi, C. Mrad, and R. Nasri
bration equations and their resolution. Secondly, we exhibit the free and forced
responses numerically obtained using Matlab software. Thirdly, we examine the
influence of the impact absorber parameters variation on the vibration reduction to
identify the favorable design. Finally, we advance some recommendations.
The results are displacement-based responses, in contrast with known works
(energy-based), which are more useful for mechanical applications.
2 Free Vibration
2.1 Formulation
We model the mechanical system equipped with the ball absorber in free vibration
as a 2ddl vibratory system, as shows Fig.1:
Fig. 1 The free system equipped with the absorber
The impacts are modeled by two springs, with a stiffness k1, and two viscous
dampers, with a damping coefficient c1. x1 and x2 are respectively the displace-
ments of the primary mass (M) and the impact mass (m), and d represents the
clearance (the width of the free zone).
Three cases are possible:
• Case 1: x2(t) – x1(t) < 0
The impact mass hits the primary mass left side, the vibration equations are as fol-
lows:
⎡M 0 ⎤ ⎧ x1 ⎫ ⎡c + c1 − c1 ⎤ ⎧ x1 ⎫ ⎡k + k1 − k1 ⎤ ⎧ x1 ⎫
⎨ ⎬+ ⎨ ⎬+ ⎨ ⎬=0
⎢0
⎣ m⎥⎦ ⎩x2 ⎭ ⎢⎣ − c1 c1 ⎥⎦ ⎩x2 ⎭ ⎢⎣ − k1 k1 ⎥⎦ ⎩x2 ⎭
The velocities of masses M and m just before and immediately after the collision,
verify these equations:
(1 − μα ) − μ (1 + α ) − + (1 + α ) − (μ − α ) −
x + 1 = x 1 + x 2 ; x 2 = x 1 + x 2
(1 + μ ) (1 + μ ) (1 + μ ) (1 + μ )
Where the mass ratio and the restitution coefficient are respectively:
m ( x + 2 − x +1 )
μ= ;α = − −
M ( x 2 − x −1 )
Mechanical Vibration Cancellation Using Impact Absorber 241
The indices – and + correspond respectively to just before and immediately after
the collision.
Therefore, the vibration equations become:
x1 (t ) + 2ξω x1 (t ) + ω ² x1 (t ) = μ [2ξ 1ω1 y (t ) + ω1 ² y (t ) ]
y(t ) + 2ξ 1ω1 y (t ) + ω1 ² y (t ) = − x1 (t )
Where:
c c
ω = k M ; ω1 = k1 m ; ξ = ; ξ 1 = 1 ; y (t ) = x 2 (t ) − x1 (t )
2Mω 2 mω 1
The system initial conditions, at t=0, are:
x1 (0) = x10 ; x2 (0) = x20 ; y(0) = y0 = x20 − x10
x1 (0) = x10 ; x2 (0) = x20 ; y (0) = y 0 = x20 − x10
To simplify the study, we consider the following hypotheses:
a. The two natural frequencies can be approximated by their average; b. The im-
pact mass is small relatively to the primary mass; c. The primary mass damping
factor (ξ) is small relatively to the impact mass damping factor (ξ1).
Three cases show up:
ξ1 < μ ; The approximate response is:
⎛1 ⎞ ( − ξ1ω a t )
1
x1 (t ) = A cos ⎜ μ − ξ12 ω a t − θ ⎟e 2 sin(ω a t )
⎝2 ⎠
1/ 2
⎡ ⎛ x ⎞
2
⎛ μ x ⎞
2
⎤
A = ⎢ ⎜⎜ 10 ⎟⎟ + ⎜⎜ μ y 0 + ⎛⎜ ξ 1 + ⎞⎟ 10 ⎟⎟ (μ − ξ 1 ² )− 1 ⎥
⎢⎣ ⎝ ω a ⎠ ⎝ ⎝ 2 ⎠ ωa ⎠ ⎥⎦
⎡⎛ μω a y 0 μ⎞ ⎤
θ = tan −1 ⎢⎜⎜ + ξ1 + ⎟(μ − ξ1 ² )−1 / 2 ⎥
⎢⎣⎝ x10 2 ⎟⎠ ⎥⎦
ξ1 = μ ; The approximate response is:
⎡ x μ x10 + μωa y0 ⎤ (−
1
μω a t )
x1 (t ) = ⎢ 10 + t ⎥e 2
sin(ωa t )
⎢⎣ ωa 2 ⎥⎦
ξ1 ; μ ; The approximate response is:
⎧⎡ ξ + ξ ²−μ ⎫
(
⎪ ⎢ − μy0 + − ξ1 + ξ1 ² − μ 10 ⎥ e
ωa ⎦
)
x ⎤ − 1 21 ωat
+⎪
⎪⎣ ⎪ sin(ω a t )
x1 (t ) = ⎨ ⎬
⎪ 2 ξ1 ² − μ
−ξ + ξ ² − μ
⎪⎡
( )
x10 ⎤ 1 2 1 ωat
⎪ ⎢ μy0 + − ξ1 + ξ1 ² − μ ω ⎥ e ⎪
⎩⎣ a ⎦ ⎭
• Case 2: 0 ≤ x2(t) – x1(t) ≤ d

The impact mass is in free motion with constant velocity and without any col-
lision, the vibration equations are as follows:
Mx1 + cx1 + kx1 = 0
mx2 = 0
The primary mass response is:

⎡ x + ξω x10 ⎤
x1 ( t ) = e − ξω t ⎢ x10 cos( ω d t ) + 10 sin( ω d t ) ⎥
⎣ ωd ⎦
The impact mass response is:
x 2 (t ) = x 20 t + x 20
Where, ωd = 1 − ξ 2 .ω
And x10 , x10 , x20 , x20 are the primary mass and the impact mass initial displacements
and velocities, respectively.
• Case 3: x2(t) --- x1(t) > d
The impact mass hits the primary mass right side, the responses are similar to
those obtained in Case 1, when the impact mass hits the primary mass left side.
2.2 Optimization
To examine the effect of the absorber parameters on the primary system vibration
amplitudes, we conduct a numerical study. The basic characteristics of the primary
system equipped with the absorber are:
M=1Kg; m=0.4Kg; K=1N/mm; k1=0.4N/mm; d=6mm; α=0.2
The initial conditions are:
x10=0; x20 =0; x 10 =15.10-3m/s; x 20 =0
• Undamped System
We consider a negligible damping either for the primary system or the absor-
ber: C=c1=0.01Ns/m.
Figure 2, shows the effect of the impact absorber on the primary system re-
sponse. The primary mass vibration amplitudes are neatly reduced.
-4
x 10
5
1
X1(m)
-1
-2
-3
avec absorbeur
-4 sans absorbeur
-5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
t(s)
Fig. 2 Primary free system response without (Black) and with (Red) absorber
The numerical results show that the favorable stiffness ratio is: k1/K=0.8 (Fig.
3). Other results, not presented, show that the favorable mass ratio is: m/M=0.8
(maximal), whereas the clearance (d) and the restitution coefficient (α) have no
clear effect.
Fig. 3 Primary free mass displacement for k1/K= a)0.2; b)0.4; c)0.8
• Damped System
We consider a damped primary system: ξ=2.5%.
The system mass maximal displacement is the lowest when the damping ratio
is: c1/C=3 (Fig. 4); and when the clearance is: d=12mm (Fig. 5).
Other results, not presented, show that the effect of d depends on ξ, whereas the
effect of α is notable and depends also on ξ.
Fig. 4 Primary free mass displacement for c1/C= a)0.3; b)1; c)3
Fig. 5 Primary free mass displacement for d= a)3mm; b)12mm

3 Forced Vibration
3.1 Formulation
The mechanical system equipped with the ball absorber in forced vibration is
modeled as a 2ddl vibratory system, as shows Fig.6:
Fig. 6 The forced system equipped with the absorber
f1 is the applied excitation, of modulus F0 and pulsation ω.

Three situations are to consider:
• Situation 1: x2(t) – x1(t) < 0
The impact mass hits the primary mass left side, the vibration equations are:
⎡M 0 ⎤ ⎧ x1 ⎫ ⎡c + c1 − c1 ⎤ ⎧ x1 ⎫ ⎡k + k1 − k1 ⎤ ⎧ x1 ⎫ ⎧ f1 ⎫
⎨ ⎬+ ⎨ ⎬+ ⎨ ⎬=⎨ ⎬
⎢0
⎣ m ⎥⎦ ⎩ x2 ⎭ ⎢⎣ − c1 c1 ⎥⎦ ⎩ x 2 ⎭ ⎢⎣ − k1 k1 ⎥⎦ ⎩ x2 ⎭ ⎩0 ⎭
In another form:
[M ]{x} + [C ]{x} + [K ]{x} = { f1 }avecf 1 (t ) = F0 e iωt
The response amplitude is:
{X } = (− ω ²[M ] + iω [C ] + [K ])−1 {F0 }
• Situation 2: 0 ≤ x2(t) – x1(t) ≤ d
The impact mass is in free motion with constant velocity and without any col-
lision, the vibration equations are as follows:
Mx1 + cx1 + kx1 = f1
Using Laplace transform, the response amplitude is:
F0 ( s )
X 1 (s) =
Ms ² + Cs + K
Returning to the time domain, with replacing the Laplace variable s by jω, and
while using the damping factor ξ, the response is:
x1(t) = X1 sin(ωt +ϕ)
Where:
⎡ ⎛ω ⎞ ⎤
⎢ 2ξ ⎜⎜ ⎟⎟ ⎥
F0 / K ⎢ ⎝ ω0 ⎠ ⎥
X1 = and ϕ = − arctg ⎢ 2 ⎥
⎢1 − ⎛⎜ ω ⎞⎟ ⎥
2
⎡ ⎛ ω ⎞2 ⎤ ⎡ ⎛ ω ⎞⎤
2
⎢1 − ⎜⎜ ⎟⎟ ⎥ + ⎢2 jξ ⎜⎜ ⎟⎟⎥ ⎢ ⎜ω ⎟ ⎥
ω ⎝ ω 0 ⎠⎦⎥ ⎣ ⎝ 0⎠ ⎦
⎣⎢ ⎝ 0 ⎠ ⎥⎦ ⎣⎢
• Situation 3: x2(t) – x1(t) > d
The impact mass hits the primary mass right side, the responses are similar to
those obtained in Case 1, when the impact mass hits the primary mass left side.
3.2 Optimization
We examine the effect of the absorber parameters on the primary system vibration
amplitudes through a numerical study. The basic characteristics of the primary
system equipped with the absorber are:
M=1Kg; m=0.4Kg; K=1N/mm; k1=0.4N/mm; d=6mm; F0=1N; f=ω/2π=10Hz
The initial conditions, at t=0, are null.
• Undamped System
Negligible damping is considered either for the primary system or the absorber:
C=c1=0.01Ns/m.
Figure 7, shows the effect of the impact absorber on the primary system re-
sponse. The primary mass vibration amplitudes are clearly reduced.
0.02
0.015
0.01
0.005
X1(m)
-0.005
-0.01
-0.015
avec absorbeur
sans absorbeur
-0.02
0 5 10 15
t(s)
Fig. 7 Primary forced system response without (Black) and with (Red) absorber
The numerical results show that the favorable stiffness ratio is: k1/K=0.8 (Fig.
8). Other results, not presented, show that the favorable mass ratio is: m/M=0.8
(maximal), the clearance (d) has no clear effect, and the favorable excitation fre-
quency and modulus are: f=50Hz (maximal) and F0=0.2N (minimal).
Fig. 8 Primary forced mass displacement for k1/K= a)0.2; b)0.4; c)0.8
• Damped System
We consider a damped primary system: ξ=5%.
The system mass maximal displacement is the lowest when the damping ratio
is: c1/C=1 (Fig. 9); the clearance is: d=12mm (Fig. 10); and when the excitation
frequency is: f=50Hz (Fig. 11).
Other results, not presented, show that the favorable excitation modulus is:
F0=0.2N (minimal).
Fig. 9 Primary forced mass displacement for c1/C= a)0.3; b)1; c)3
Fig. 10 Primary forced mass displacement for d= a)3mm; b)12mm
Fig. 11 Primary forced mass displacement for f= a)2Hz; b)50Hz
4 Conclusion
The ball absorber is found efficient in reducing free and forced vibrations of me-
chanical systems especially when the absorber parameters are optimized.
When the system is free, large mass and stiffness ratios lead to small primary
mass displacements. Besides, a large relative damping improves the primary sys-
tem response. The incidence of the clearance and the impact quality is not clear
and needs to be examined closely.
When the system is forced, also large mass and stiffness ratios lead to small
primary mass displacements. A medium relative damping is favorable but masks
the impact quality, and needs further examination. Besides, the clearance effect is
clearer when the primary system is damped, a large clearance is favorable. In ad-
dition, excitation with large frequency and small modulus is favorable.
Acknowledgements. This work was conducted within the Systems Mechanics Team di-
rected by Charfeddine MRAD. The authors gratefully acknowledge the helpful comments
and suggestions of the reviewers, which have improved the presentation.
References
Marhadi, K.S., Kinra, V.K.: Particle impact damping: effect of mass ratio, material, and
shape. Texas Agricultural and Mechanical University, USA (2004)
Cheng, J., Xu, H.: Inner mass impact damper for attenuating structure vibration. Interna-
tional Journal of Solids and Structures 43, 5355–5369 (2006)
Cheng, C.C., Wang, J.Y.: Free vibration analysis of a resilient impact damper. International
Journal of Mechanical Sciences 45, 589–604 (2003)
Xu, Z., Wang, M.Y., Chen, T.: A particle damper for vibration and noise reduction. Journal
of Sound and Vibration 270, 1033–1040 (2004)
Ema, S., Marui, E.: Damping characteristics of an impact damper and its application. Inter-
national Journal of Machine Tools and Manufacturing 36, 293–306 (1996)
Sankar, S., Bapat, C.N.: Single unit impact damper in free and forced vibration. Journal of
Sound and Vibration 99, 85–94 (1985)
Masri, S.F.: General motion of impact dampers. University of Southern California, USA
(1972)
Masri, S.F.: Analytical and experimental studies of impact dampers. Dissertation, Califor-
nia Institute of Technology, USA (1965)
Desen, I.C.: Experimental study on an impact vibration absorber. Dissertation, Texas Tech-
nical University, USA (2000)
RFLP Approach in the Designing of Power-
Trains for Road Electric Vehicles
Clemente Capasso1, Moncef Hammadi2, Stanislao Patalano3, Ruixian Renaud2,

and Ottorino Veneri1
1
Istituto Motori - CNR National Research Council of Italy, Via Marconi 4,
80125 Naples (Italy)
c.capasso@im.cnr.it, o.veneri@im.cnr.it
2
SUPMECA, Higher Institute of Mechanical Engineering, Paris 3 rue Fernand Hainaut
93407, Saint-Ouen (France)
moncef.hammadi@supmeca.fr, ruixian.liu@gmail.com
3
University of Naples Federico II - Department of Industrial Engineering, P.le V. Tecchio
80, 80125 Naples (Italy)
patalano@unina.it
Abstract. This paper is focused on the designing process of propulsion systems

for road electric vehicles, by means of the RFLP approach for System Engineer-
ing. The process starts from the analysis of the main requirements for the vehicle
considered, in relation to its specific mission. The vehicle behavior is then simu-
lated on standard driving cycles, evaluating the performance figures of different
power-train configurations, under different operative conditions. The presented
designing procedure reaches the 3D CAD model of the identified propulsion sys-
tem, coupled with a specific laboratory test bench, based on an eddy current brake
and flywheel for the simulation of the vehicle inertia. The obtained simulation re-
sults show the good performance of the power-train in terms of vehicle speed fol-
lowing its reference on driving cycle and vehicle autonomy.
Keywords: Systems Engineering, RFLP approach, Electric Vehicles.
1 Introduction
The System Engineering (SE) approach is today widely used in current designing
activities as well as in research activities. During last years, the approach has been
extended also to product development lifecycle to gain the contents of Axiomatic
Design and cover early factors that affect the entire product lifecycle (Gumus et
al., 2008). Furthermore, SE was joined to Model Based (MB) approach as a
unique and integrated methodology to be used especially when designers have to
accomplish modelling and simulation integration, during the whole design process
(Graignic et al., 2013). Typical complex systems to be developed using the MBSE
approach are the mechatronic systems because they act on multi-domains and ex-
hibit interactions of physical effects due to different energy domains. For these

250 C. Capasso et al.
systems it is possible to highlight several and different contributes in literature.

From one side, the proposal of methodologies to support the whole design process,
as the one in (Hehenberger et al., 2010) that foresees the use of hierarchical models
or the one in (Mhenni et al., 2014) that uses the general purpose and multi-view
SysML language for systems modeling to propose a multidisciplinary approach
aimed to the designing of mechatronic system. On the other side, different topics
deal with the use of RFLP approach to MBSE. Several efforts have to be accom-
plished, for example, to deep the way to use the integrated design environment and
the way to obtain continuous tests and validations for the successive development
steps. The present works deals with the use of RFLP approach in an integrated and
commercial environment for the design of a mechatronics system. It is arranged as
follows. Section 1 resumes the RFLP approach. Section 2 illustrates the case study
related to the designing of power-trains for urban electric busses. Section 3 states
simulation results and, finally, Section 4 draws the conclusions.
2 RFLP Approach in Systems Engineering
The RFLP approach represents the baseline for Systems Engineering in model-based
designing processes, enabling a strong interaction and collaboration between differ-
ent engineering disciplines. The RFLP can be considered as a top-down approach,
which starts analyzing the whole complex system. Then the system is divided into
different sub-systems and the approach proceeds with the analysis of the single sub-
systems (Lanzotti et al., 2013). The acronym RFLP stands for Requirements engi-
neering, Functional design, Logical design and Physical design, i.e. the 3D CAD
modelling (Kleiner and Kramer, 2013). The first step consists in the definition of the
system requirements and is characterized by the analysis of customer needs and
technical requirements. During this phase the main features and services offered by
the analyzed system are defined in a specific documentation. The second step of the
RFLP approach is represented by the functional analysis. This phase is focused on
the analysis of the specific functions that each component needs to perform in order
to satisfy the requirements described in the above requirement phase. In the third
step of the RFLP approach, represented by the logical analysis, the logical connec-
tions between the subsystems are analyzed in order to carry out a logical model of
the system. This logical step represents the first quantitative analysis of the complex
system. In particular the behavior of the system components is modeled, starting
from the knowledge of the expected inputs and outputs of each subsystem as already
analyzed in the functional diagram. The last step of the RFLP approach consists in
the physical model, in fact the design process of a system is not considered complete
without its geometric and dimensional model. During this phase a tridimensional
models of sub-systems are created and their parametrical relationships, which allow
the physical design, are defined.
RFLP Approach in the Designing of Power-Trains for Road Electric Vehicles 251
3 Case Study: Designing of Power-Trains for Urban Electric

Busses
In this work, the electric power-train of an electric urban bus is considered as case
study. In order to simplify the designing procedure, the propulsion system under
study is considered mechanically coupled to a laboratory test bench for the simula-
tion of the base operative conditions of the vehicle. The main characteristic of the
vehicle are reported in Table 1.
Table 1 Main characteristics of the electric bus
Vehicle Weight (kg) 6185

2
Vehicle Main Section (m ) 4.664
Transmission Ratio 0.166
Aerodynamic Drag Coefficient 0.7
Effective Rolling Radius 0.4 m
Number of Passengers 18
The definition of Requirements starts with the need for the propulsion systems
to power a specific electric vehicle. This requirement can be divided in the follow-
ing sub-requirements: the test-bench has to represent the mechanical load of the
vehicle on the road; the energy storage system has to power the electric drive; the
power-train on the test-bench has to follow the driving cycle in terms of speed and
resistant torque references. These requirements can be further divided in more de-
tailed sub-requirements. In particular the simulation of the real mechanical load
has to represent the rolling, advancing, aerodynamic and inertia resistant forces,
which depend on the road slope and resistant coefficient, air density, weight, fron-
tal area, speed and acceleration of the vehicle. Moreover, the energy storage sys-
tem has to guarantee both the required driving range and dynamic performance for
the specific mission of the vehicle. In addition, the test-bench simulating requires
the proper control of resistant torque, motor torque and speed.
The functional map, realized using CATIA V6, as result of this functional anal-
ysis described in Section 2, is reported in Figure 1 for the considered urban bus.
In order to be concise, only the highest level of the functional diagram is
represented in the functional map of Figure 1. According to this diagram the main
functionalities the propulsion system has to perform are: storing and supplying
electric energy; supplying mechanical energy; simulating the driver; simulating
the resistant forces; monitoring and controlling the operative parameters. The
functionalities represented in the above diagram are connected each others through
lines and arrows, which represent different type of incoming and outgoing flows.
In particular the energy flows are represented by red lines, and flows of informa-
tion are represented by green lines.
Simulate the driver

Resistant Torque
Simulate resistant
forces
Actual resistant torque Reference
torque
Reference speed
Store electric energy
Recharge
energy Voltage and
Electric current measure
energy
Torque and Monitor, control and
speed supervise working
Actual speed and torque
Supply mechanical Reference speed and torque

energy
Fig. 1 Functional map of the propulsion system working on the test-bench
The logical analysis aims to characterize the behavior of the whole propulsion
system of the urban bus and of its components on the test-bench. The logical mod-
el is firstly implemented by means of Dymola software, which is based on the
Modelica simulation environment and is specifically devoted to modeling and si-
mulating integrated and complex systems. Then, the integration into the logical
environment of CATIA v6 is obtained by means of DBM Workbench of CATIA
V6 (Smolek et al, 2013).
In order to model and simulate the behavior of the whole power-train, the first
component taken into consideration is the energy storage system. In this regard,
the energy storage modeling procedure is based on simplifying hypothesis, using
the specific ‘Electric Energy Storage’ Modelica library (Einhorn et al. 2011).
The input parameters of this block are set on the base of the main operative
conditions related to the Li(NiCoMn)O2 battery pack considered to power the pro-
pulsion system under study. The main characteristics and recommended operative
conditions of a single Li(NiCoMn)O2 cell are reported in Table 2 (Capasso and
Veneri, 2014).
Table 2 Main characteristics of Li(NiCoMn)O2 cell
Cell Type Li(NiCoMn)O2

Nominal Voltage (V) 3.65
Recommended Voltage Limit For Discharge (V) 3.0
Maximum Charge Voltage (V) 4.15
Specific Energy (Wh/Kg) [Energy Density (Wh/l)] 174 (370)

Dimensions (mm) 217 X 129 X 7.2
In a similar way the electric drive is implemented using components available

in the Modelica library.
As stated before, the power-train is considered connected to an eddy current

brake in order to simulate in laboratory the real mechanical load representative of
the resistant forces. The main characteristics of the eddy current brake, considered
in this case study, are reported in Table 3 (Veneri et al., 2012).
Table 3 Main Characteristics of the Eddy Current Brake
Rotor diameter (mm) 506

Max. Torque (Nm) 1400
Max. speed (rpm) 8000
Power (kW) [Power (CV)] 257.4 [350]
Inertia (kg m²) 0.618
Weight (kg) 723
In this paper, the electric brake model is implemented representing only the
mechanical behavior, without taking into account thermal and electrical aspects.
Moreover the maximum brake mechanical torque is considered constant with the
angular speed. According to these hypotheses the electric brake model is built us-
ing the Dymola mechanical libraries (Hammadi et al., 2012) taking into account
that the vehicle inertia is simulated considering a flywheel of equivalent inertia in-
cluded inside the "brake" block. The related inertia value is evaluated at the elec-
tric motor shaft using a Modelica script.
Battery
Electric Drive
Ground
Driving Cycle
Parameters Electric Drive Control Brake Control
Fig. 2 Logical of the propulsion system working on the test-bench
Before starting the physical step, a verification phase of the whole design pro-
cedure is required to optimize performance and size of the power-train. The flow-
chart of this verification procedure is reported in Figure 3.
As described in the flow-chart of Figure 3, after the design of the logical model,
the input parameters of the specific case study are required to be properly set.
Then it is possible to run the simulations and the results are evaluated. The first
evaluation is related to the relative difference between actual speed reached by the
vehicle and reference speed imposed by the required driving cycle. If this differ-
ence is higher than 15% the propulsion system is not considered suitable to satisfy
the dynamic performance requirements and a better solution for the electric drive
Fig. 3 Flow-chart of the verification for the design procedure
needs to be evaluated. The next step consists in evaluating the vehicle autonomy,
in terms of hours, on a specific driving cycle, which is required to be greater than
4 hours when the vehicle runs on a plane road and greater than 2 hours on a road
slope of 10%. This last evaluation is directly related to the design of the vehicle
battery pack in terms of capacity. The simulation is stopped when the battery pack
SoC reaches the value of 20%, in order to avoid damages to the storage system. At
the end of this iterative verification procedure, the power-train is considered well
designed and its dimensions can be used in the parametrical 3D CAD model to
complete the physical step of the RFLP process.
4 Simulation Results and Discussion

The simulations carried out in this work aim to characterize the behavior of the
electric bus in different operative conditions representative of its specific mission.
The first set of simulations is carried out using the parameters of first configu-
ration reported in Table 4, in order to start the verification procedure of Fig. 3.
Table 4 Parameters of First Configuration of propulsion system for the Trapezoidal Driving
Cycle 0% Slope Simulation
Battery Cells in Series Cells in Parallel Cell Capacity Cell Weight Battery Pack Weight
Pack 140 1 50 Ah 1.25 kg 175 kg
Electric Power Maximum Torque Rated Current Rotor Inertia Weight
Drive 54.6 kW 470 Nm 101 A 0.123 kgm2 130 kg
The first analysis, reported in Figure 4, is related to the actual vehicle speed
profile compared to the reference speed profile for a trapezoidal driving cycle on a
plane road, derived by a standard ECE driving cycle.
40
Reference Speed
Actual Speed
30
Speed (km/h)
20
10
0
0 20 40 60 80
Time (s)
Fig. 4 Comparison among the actual and reference vehicle speed on a trapezoidal driving
cycle for the first configuration
In this case it is clear that the urban bus, in this configuration is not able to sa-
tisfy the dynamic performance required by the driving cycle, because of a percen-
tage error of about 20%. This result depends on the motor torque reported to the
wheels, which results too low if compared with the vehicle inertia combined with
the resistant forces.
For the above reasons, following the designing verification procedure, a second
test is performed choosing an electric drive with a nominal power of 63.4 kW and
a battery cell capacity of 150 Ah. In this case, the electric bus follows, with a per-
centage error of 12%, the dynamic requirements of the driving cycle of Figure 4,
although the parameters chosen for this second solution affect the vehicle weight.
For this last configuration, Figure 5A shows the vehicle mechanical power in
comparison with the battery pack electric power during the driving cycle. For this
set of parameters, as reported in Fig. 5B, the vehicle autonomy on the driving
cycle, with a road slope of 0%, is evaluated equal to 5.2 hours. This means that the
dynamic performance and autonomy requirements are satisfied.
100 100
A B
Battery Electric Power

80
50 Motor Shaft Mechanical Power
Battery SoC (%)
Power (kW)
60
40
-50
20
-100 0
0 20 40 60 80 0 1 2 3 4 5 6
Time (s) Time (h)
Fig. 5 Vehicle mechanical and electric power (A) reported to the motor shaft and battery
SoC (B) vs time during a trapezoidal driving cycle
The second set of simulations is related to the vehicle running on a positive

slope of 10% on the same driving cycle mentioned above. Simulation results evi-
denced that the road slope cannot be performed with the last configuration identi-
fied for the urban bus propulsion system. For this reason, after different iterations
of the design verification procedure, another configuration with a 162kW/1400Nm
electric motor and a 300Ah battery pack is considered. In this case the reference
speed is followed by the vehicle actual speed with a percentage error value of
about 7%. In Figure 6A the vehicle mechanical power reported to the motor shaft
and the battery electric power during the performance of the trapezoidal driving
cycle with a slope of 10% are depicted. In this case, a maximum value of electric
power of about 190 kW is reached during the acceleration phase. As reported in
Figure 6B, a travel range of 2.1 hours is evaluated.
200 100
A B
Battery Electric Power
Motor Shaft Mechanical Power
150 80
Battery SoC (%)
Power (kW)
100 60
50 40
0 20
-50 0
0 20 40 60 80 0 0.5 1 1.5 2 2.5
Time (s) Time (h)
Fig. 6 Vehicle mechanical and electric power (A) reported to the motor shaft and battery
SoC (B) vs time during a trapezoidal driving cycle with a road slope of 10%
At the end of the design verification process, the geometries of 3D CAD mod-
els related to the electric motor, joint, flywheel and brake are automatically rege-
nerated. These models are used to design the supports for the motor and flywheel.
Fig. 7 Test Bench 3D CAD Model

Figure 7 shows the 3D CAD models of the whole test bench of the power-train
powered by the battery pack and connected to the eddy current brake, running in
CATIA V6 environment.
The designing approach presented in this paper, enriched by means of parame-
tric relations building the final test bench CAD model implemented in CATIA V6,
represents a design procedure which is expected to achieve a drastic reduction of
designing times and costs for the whole development process of propulsion sys-
tems for road electric vehicles.
5 Conclusions
The present paper briefly summarizes the preliminary method and main steps ac-
complished in the designing process of a propulsion system for road electric ve-
hicles, by means of the RFLP approach. The results obtained through simulations,
compared to those related to reference driving cycles, evidence the possibility to
accomplish adequate simulation models. Also the verification and improvement
procedures, in the sense of RFLP approach, as well as the possibility to directly
link the changes of physical models to the ones of logical model were well
accomplished. At the present, further developments need for the management of
system requirements, as regards the availability of tools related to the educational
environment CATIA V6 used. Further activities will deal with the validation of
actual simulation models by means of a comparison with experimental data. In
this way it will be possible to extend the same RFLP approach and related models
to the designing of different propulsion systems for road electric vehicles.
Acknowledgements. The present work was developed with the contribution of the French
Ministry for Higher Education performing the activities of the PLACIS (Plateforme Colla-
borative d’Ingénierie Système, Collaborative Platform for Systems Engineering) project.
The authors gratefully acknowledge Mr. Antonio Rossi, technician of Istituto Motori, for
his cooperation in the drawings of this paper, and engineers Antonio Amodio and Daniela
De Vivo for their assistance in carrying out preliminary aspects of the simulation phase.
References
Capasso, C., Veneri, O.: Experimental analysis on the performance of lithium based batte-
ries for road full electric and hybrid vehicles. J. Applied Energy (2014),
doi:10.1016/j.apenergy.2014.04.013
Einhorn, M., Conte, F.V., Kral, C., Nikla, S.C.: A Modelica Library for Simulation of Elec-
tric Energy Storages. In: Proc. of the 8th International Modelica Conference, Technical
Univeristy, Dresden, Germany, March 20-22 (2011), doi:10.3384/ecp11063436
Graignic, P., Vosgien, T., Jankovic, M., Tuloup, V., Berquet, J., Troussier, N.: Complex
System Simulation: Proposition of a MBSE Framework for Design-Analysis Integration.
Procedia Computer Science 16, 59–68 (2013)
Gumus, B., Ertas, A., Tate, D., Cicek, I.: Trans disciplinary Product Development Life-
cycle. Journal of Engineering Design 19(03), 185–200 (2008),
doi:10.1080/09544820701232436
Hammadi, M., Choley, J., Penas, O., Riviere, A.: Multidisciplinary approach for modelling
and optimization of Road Electric Vehicles in conceptual design level. In: Proc. of Elec-
trical Systems for Aircraft, Railway and Ship Propulsion (2012),
doi:10.1109/ESARS.2012.6387488
Hehenberger, P., Poltschak, F., Zeman, K., Amrhein, W.: Hierarchical design models in the
mechatronic product development process of synchronous machines. Mechatronics 20,
864–875 (2010)
Kleiner, S., Kramer, C.: Model Based Design with Systems Engineering Based on RFLP
Using V6. In: Proc. of the 23rd CIRP Design Conference (2013), doi:10.1007/978-3-
642-30817
Lanzotti A, Patalano S, Vitolo F (2013) A Graph-based Software Tool for the CAD Model-
ing of Mechanical Assemblies. Proc. of 8th International Conference on Computer
Graphics, Theory and Applications, GRAPP 2013 IVAPP 2013, pp. 60-69, Barcelona,
Spain, 21th-24th February,SciTePress Digital Library, doi 10.5220/0004299000600069
Mhenni, F., Choley, J.-Y., Penas, O., Plateaux, R., Hammadi, M.: A SysML-based metho-
dology for mechatronic systems architectural design. Advanced Engineering Informat-
ics 28, 218–231 (2014)
Smolek, P., Heinzl, B., Ecker, B., Breitenecker, B.: Exploring the Possibilities of Co-
Simulation with CATIA V6 Dynamic Behavior Modeling. In: Proc. of Electrotechnical
and Computer Science Conference ERK (2013)
Veneri, O., Migliardini, F., Capasso, C., Corbo, P.: Experimental performance assessment
of Pb, Li[NiCoMn]O2 and LiFePO4 batteries for road vehicles. In: Proc. of Power Elec-
tronics, Electrical Drives, Automation and Motion, SPEEDAM (2012),
doi:10.1109/SPEEDAM.2012.6264626
Mechanical Vibration Elimination Using
Friction Absorber
Aymen Nasr, Charfeddine Mrad, and Rachid Nasri
Laboratory of Applied Mechanics and Engineering (LMAI), National School of Engineers

of Tunis (ENIT), University of Tunis El Manar (UTM), BP 37, Le Belvedere, 1002, Tunis,
Tunisia
AymenNasr01@gmail.com, Charfeddine.Mrad@enit.rnu.tn,
Rachid.Nasri@enit.rnu.tn
Abstract. The work consists on studying the vibration reduction of mechanical

systems using friction absorber, with and without viscous damping. The aim is to
determine the influence of the friction absorber parameters on the mechanical sys-
tem response. A study using a nonlinear friction force is conducted while varying
the absorber parameters, which are: the mass, the stiffness, the friction coefficient,
and the damping coefficient. The linearization of the friction force is then consi-
dered, and the results are compared. The effect of the primary system damping
and the excitation parameters are also examined. The influence evaluation of the
friction absorber parameters on the mechanical system vibration leads to conclude
on the favorable absorber giving significant vibration decrease, and to advance de-
sign recommendations.
Keywords: mechanical vibration, friction absorber, vibration reduction, parame-

tric optimization.
1 Introduction
Various studies were done to reduce mechanical systems vibration using friction
absorbers, however the absorber parameters effect on the mechanical system vi-
bration amplitude is barely considered.
The friction presents a damping behavior (Balachandran and Magrab, Lopeza et
al.), the vibration reduction by friction needs understanding the excited system (Hi-
nrichs et al., Stein et al.), this behavior is nonlinear relatively to velocity (Pratt,
Liang), whereas the Coulomb model is basic to dry friction (Hundal): the friction
force is proportional to the velocity, it is confined within two limits to form an im-
pair function. To study 2ddl vibratory systems containing friction, the velocity re-
sponse is periodic due to physical considerations, verified experimentally (Hundal).
The work consists on conducting a parametric study on the vibratory response
of a mechanical system equipped with a friction absorber, with and without damp-
ing. The effect of the absorber parameters: mass, stiffness, friction coefficient, and

260 A. Nasr, C. Mrad, and R. Nasri
damping coefficient; is examined. The influence of the primary system damping

and excitation parameters is considered. The linearization consists on replacing all
forms of damping by a viscous damper, which is also investigated.
The results are displacement-based responses, unlike known works (energy-
based), which are more useful for mechanical applications.
Fig. 1 Coulomb friction model
2 Undamped Absorber
2.1 Formulation
We consider a vibratory mechanical system with a base excitation. The system is
modeled as a mass (m1) linked to the base through a spring (K1). To the primary
mass, a secondary mass is added (m2), it is linked through a spring (K2). The sec-
ondary mass slides with dry friction (Fig. 2).
Fig. 2 The vibratory model, without absorber damping
The vibration equations are as follows:
m 0 x K K K x f cos ωt F sign x x
0 m x K K x F sign x x
f1 et ω are respectively the excitation modulus and pulsation.
Mechanical Vibration Elimination Using Friction Absorber 261
2.2 Optimization
A parametric study was conducted to determine the absorber parameters effect on
the primary system response.
Figure 3, shows the increase of mass m2 conducts to a decrease of the primary
mass vibration amplitude. Figure 4, shows the primary system vibration amplitude
increases while increasing the stiffness K2. Figure 5, shows the rise of friction
conducts to a drop of the primary mass vibration amplitude.
Fig. 3 X1 function of m2
Fig. 4 X1 function of K2
2.3 Equivalence
• Existance
The linearization consists on replacing all forms of energy dissipation by a
viscous damper, dissipating the same quantity of energy.
Fig. 5 X1 function of
The equivalence existence is investigated using the force ratios R1 and R2, link-
ing friction and equivalent forces. Figures 6, 7, 8, show the development of R1 and
R2 with m2 and K2, for m1=1Kg, K1=1N/mm, f1=1N, ω=20πrad/sec, and =0.1.
R1 and R2 are always under 1, which indicates equivalence when the secondary
mass slides without stopping.
Fig. 6 R1 and R2 for K2=0.2N/mm
• Comparaison
Figure 9, shows the primary mass vibration amplitude using the linear and non-
linear methods; when m2=0.8Kg, K2=0.4N/mm, and m2 slides without stopping.
The two responses are close, which confirms that the linear method can replace
the nonlinear method when the secondary mass slides without stopping.
Fig. 9 Primary system linerar and nonlinear undamped responses

3 Damped Absorber
3.1 Formulation
The mechanical system vibrates due to a base excitation. To the primary mass, a
secondary mass is added (m2), it is linked through a spring (K2) and a viscous
damper (C). The secondary mass slides with dry friction (Fig. 11).
Fig. 10 The vibratory model, with absorber damping
The vibration equations are as follows:
m 0 x C C x K K K x
0 m x C C x K K x
f cos ωt F sign x x
F sign x x
f1 et ω are respectively the excitation modulus and pulsation.
3.2 Optimization
A parametric study was also conducted to determine the absorber parameters ef-
fect on the primary system response.
Figure 12, shows the rise of mass m2 conducts to a drop of the primary mass
vibration amplitude. Figure 13, shows the primary system vibration amplitude de-
creases while decreasing the stiffness K2. Figure 14, shows the augmentation of
friction conducts to a diminishment of the primary mass vibration amplitude.
Fig. 11 X1 function of m2
Fig. 12 X1 function of K2
Fig. 13 X1 function of
3.3 Linearization
• Existence
The equivalence existence is examined using the force ratios R1 and R2, linking
friction and equivalent forces. Figures 15, 16, show the change of R1 and R2 with
m2 and K2, for m1=1Kg, K1=1N/mm, f1=1N, ω=20πrad/sec, =0.1, and with an
absorber relative damping of 15%.
R1 and R2 are continually under 1, which indicates equivalence when the sec-
ondary mass slides without stopping.
• Comparaison
Figure 17, shows the primary mass vibration amplitude utilizing the linear and
nonlinear methods; when m2=0.8Kg, K2=0.4N/mm, and m2 slides without stop-
ping.
The two responses are close, which prove that the linear method can replace the
nonlinear method when the secondary mass slides without stopping.
Fig. 16 Primary system linerar and nonlinear damped responses
4 Conclusion
The friction absorber is found efficient in reducing mechanical systems vibration

especially when the absorber parameters are optimized.
An absorber with large mass and small stiffness is favorable. The increase of
friction is favorable but its effect is not regular. The absorber damping is suppor-
tive, it gives a positive effect. Other results, not presented, show the primary sys-
tem damping is supportive, and an excitation with large frequency and small
modulus is favorable.
The equivalence approach is interesting, it simplifies enormously studying dry
friction, but it is only valid when the secondary mass can be considered as it slides
without stopping relatively to the primary mass.
Acknowledgements. This work was conducted within the Systems Mechanics Team di-
rected by Charfeddine MRAD. The authors gratefully acknowledge the helpful comments
and suggestions of the reviewers, which have improved the presentation.
References
Balachandran, B., Magrab, E.B.: Vibrations. Toronto, Canada (2009)
Lopeza, I., Busturiab, J.M., Nijmeijera, H.: Energy dissipation of a friction damper. Journal
Hinrichs, N., Oestreich, M., Popp, K.: On the modeling of friction oscillators. Journal of
Sound and Vibration 216, 435–459 (1998)
Stein, G.J., Zahoransky, R., Mucka, P.: On dry friction modelling and simulation in kine-
matically excited oscillatory systems. Journal of Sound and Vibration 311, 74–96 (2008)
Pratt, T.K., Williams, R.: Non-linear analysis of stick/slip motion. Journal of Sound and
Vibration 74, 531–542 (1981)
Liang, J.: Damping estimation via energy-dissipation method. Journal of Sound and Vibra-
tion 307, 349–364 (2007)
Hundal, M.S.: Response of a base excited system with Coulomb and viscous friction. Jour-
nal of Sound and Vibration 64, 371–378 (1979)
AC Flashover: An Analysis with Influence
of the Pollution, Potential and Electric Field
Distribution on High Voltage Insulator
Hani Benguesmia1, Nassima M’ziou1, and Ahmed Boubakeur2
1
L’I3CUB university of Biskra, Algeria
hanibenguesmia@yahoo.fr
2
Laboratoire de Recherche en Electrotechnique (L.R.E).
Ecole Nationale Polytechnique, Algiers, Algeria
Abstract. This paper presents a work devoted to the impact study of the pollution
severity on the energy level dissipated on high voltage insulator. Pollution of high
voltage insulators is a factor of prime importance in the quality and reliability of
power transmission. It causes the circumvention of high voltage insulators. Howev-
er, when it is wet, it causes the dissolution of salts and the formation of an electro-
lyte layer on the surface of insulators, thus reducing dielectric strength. Pollution
generally poses few problems. However, when it is humid, it causes the dissolution
of salts and the formation of an electrolytic layer on the insulator surfaces; as a
result, it reduces their dielectric rigidity.
The aim of this paper is the experimental study of the flashover process and the
distributions of the potential and the electric field pollution carried out under 50
Hz applied voltage on a circular model simulating the 1512 L outdoor insulator
largely used by the Algerian company of electricity and gas (SONELGAZ). This
circular model is under different surface conductivity and different level of uni-
form discontinues pollution.
The computer simulations are carried out by using the COMSOL Multiphysics
3.5 software. This program uses the finite element method to solve the partial dif-
ferential equations that describe the field. Original experimental results made in
the laboratory are presented and compared with a real insulator, and simulations
presented in this paper are originals.
Keywords: insulator, flashover, superficial conductivity, high voltage, electric

potential, electric field.
1 Introduction
High voltage insulators form an essential part of high voltage electric power
transmission systems. Any failure in the performance of high voltage insulators
will cause considerable loss of capital. (B.Moula et al).

270 H. Benguesmia, N. M’ziou, and A. Boubakeur
Outdoor insulation systems are exposed to the most unfavourable atmospheric

conditions, as well as pollution, from coastal areas and industrial emissions. Under
wet conditions, the contamination layer causes the flow of leakage current, which in
turn can lead to the development of surface discharges. (B. Moula, A. Mekhaldi et al).
The insulators entre for a very modest percentage 7%, in the price of a medium
voltage aerial line. However, they are an essential element in which the security of
exploitation, the quality, and the continuity of service depend.
The pollution of insulators is a phenomenon that acts negatively on the high
voltage insulators “behavior”.(S.Jaafar et al)
The flashover voltage mechanism was investigated by many researchers. In
particular, the final stages of the complex mechanism that occur when an interme-
diate pollution band forms on theinsulating surface. [(B. Moula, A. Mekhaldi et al).
The non-homogeneity is due to the presence of different polluting agents in the
same region, and the non-uniformity of the distribution on the insulator surfaces is
because of the insulator profile, wind direction, and the position of the insulator
chains in relation to the ground (vertical, horizontal, inclined). The insulator
position in the chain, the degree of the site’s pollution where the upper or lower
insulator surface is found and the unknown effect of humidity on the pollution
(L.E Zaffanella et al), (N. Dhahbi-Meghriche et al).
The electric field distribution on the surface and within insulators string is a
function of numerous parameters including applied voltage, insulator design,
tower configuration, corona ring, hardware design, phase spacing. To approach the
ideal conditions obtained for a linear potential distribution along an insulators
string, corona rings are required. Unfortunately, there exist no specific standards
for the design and placement of corona rings. Each manufacturer makes indepen-
dent recommendations for the use of their corona rings. (B.M’hamdi et al).
In the present work, experimental results, electric field and potential distribu-
tion under normal operating conditions are simulated by a two dimensional finite
element model using the COMSOL Multiphysics 3.5 software.
2 Experimental Techniques
The laboratory’s model used in this study is a glass disc (400 “mm” diameter and
5 “mm” thick) having the feature that resists to the heat due to the electric dis-
charge. This model is equipped with two cylindrical electrodes made of alumi-
num. The first (26 “mm” diameter) placed at the high voltage and the second (172
“mm” diameter) placed at the ground (fig.1(a)). The disposition of the clean and
polluted zones is illustrated on (fig.1(b)). (p: indicates polluted zone, c: indicates
clean zone). The longitudinal dimensions of the clean “zone (c)” and “polluted
ones (p)” (fig.2) are indicated in the table1.
The tension of test is measured using a capacitive tension divider connected to
the secondary of a transformer of test 140kV (fig. 3).
AC Flashover: An Analysis with Influence of the Pollution, Potential and Electric Field 271
(a) (b)
Fig. 1 Experimental laboratory model (a), Determination of clean and polluted areas (b)
Fig. 2 Distribution of clean and polluted areas

HV Transformer
Measurement
capability
Test object
(model insulator)
Fig. 3 Test circuit
Table 1 Longitudinal dimension areas clean and pollution
Polluted area Cleaned area

(mm) (mm)
P1 P2 P3 P4 C1 C2 C3 C4
N1 51 19 17 18 9 38 40 8
N2 53 23 21 22 5 34 36 6
N3 83 0 25 26 0 30 32 4
N4 85 0 29 30 0 26 28 2
N5 87 0 33 34 0 22 24 0
N6 89 0 37 34 0 18 22 0
N7 91 0 41 34 0 14 20 0
N8 93 0 45 34 0 10 18 0
The degree of pollution used in the laboratory’s model surface is a saline solu-
tion (water and NaC1). The different conductivities studied are: 1.823 “mS/cm”,
3.33 “ms/cm”, 8.02 “ms/cm”, and 12.61 “ms/cm”, 16.32 “mS/cm”, 30.05
“ms/cm”, 50.04 “ms/cm”, and 93.70 “ms/cm”.
The software (COMSOL Multiphysics 3.5) determines the potential by solving

the electrostatic equation in addition to the conduction equation:
(1)
0 2
0 3
The profile of the insulator is introduced in the software. The mesh density is
higher in the critical regions of the insulators where higher accuracy is required. The
number of meshes for insulator string is 8644 elements. Fig(4) (B.M’hamdi et al).
Fig. 4 Discretization in finite elements and determination of mush elements of the insulator
1512L
2.1 The Process Flashover on Glass Disc Surface

For a layer of discontinuous uniform pollution, the process of flashover on surfac-
es, as observed in the present study, can be divided into three steps. First, during
the set-up of the applied voltage, initial partial discharges occurred randomly on
the surface, generates a leakage current, (fig.5(a)); second, the strong density of
the current in the active vicinity of the electrode causes an evaporation of the sa-
line solution, by Joule effect, and a dry zone appears. The increase in the tension
applied causes the lengthening of the arcs in the direction of the opposite electrode
(fig.5(b)). L’arc in the surface melted continuously and the arc either eventually
turned into a flashover, (fig.5(c)).
(a)
(b) (c)
Fig. 5 initiation of arcs (a), evollution of arcs (b) and flashover total (c)
2.2 Simulation of the Real Model

We calculated the electric field and the electric potential; we used the software
comsol mutiphysics. This last is a continuation of the programs making it possible
to give the distribution of the electric field in two dimensions. For that, we intro-
duced, in this software, our model with all its specifications (forms and nature of
the electrodes and the various mediums, tension applied, boundary conditions,…).
The configuration of our model gives a relatively difficult representation for
that uses of them the FEMM (Finite Element Magnetism Method) then one makes
the call of the model, fig.(6(a)).
We have to make the call, in this software, (comsol) our real model with all its
specifications thus, we fixed the tension applied to the electrode activates with
30kV, fig.(6(b)).
In this section we focused on the variation of the flashover according to the con-
ductivity and the pollution width (polluting layer). This circumvention is characte-
rized by the non-generation of partial arcs, is due to the pollution used and to the
fact that the total width of the clean equivalent tape (representing the amount of
different clean partial tape) exceeds the critical value from no stable discharge is
propagated (R. J Wilkins et al), (H. Streubel),( A. Mekhaldi et al).
Cap (Ground electrode)
Length of the line of

creepage (leak) (cm)
Pin (high voltage electrode)

(b) (a)
Fig. 6 Presentation of the real model on the software (clean model), (a) FEMM, (b) Comsol
3.1 Influence of the Polluting Layer

Figures (7) presenting the variation of the flashover according to width pollution
for different conductivities for a model insulator and for a real one.
We expected a flashover increase following the diminution of the circular of
the polluting layer. Either for the model insulator or the real one increase in insig-
nificant and does not exceed in the extreme case 13% of the initial flashover.
We note that the higher the conductivity is, the weaker the flashover will be.
This is true whatever the pollution width. When the conductivity increases, the
flashover decreases rapidly for an inferior conductivity to 8.02 “ms/cm” and slow-
er beyond this conductivity, in making towards a clean limit to each solution.
In fact, it is obvious that the flashover taken where the pollution is weaker
(conductivity=1,823 “ms/cm”) is more important. We could deduce that the isolat-
ing system (insulator) is more rigid when the conductivity is weak.
4 Result of Simulation
4.1 Distribution of the Electric Field on the Real Model

In the case of the clean plate, the electric field is intense meadows of the active
electrode. It decreases as one move towards the ground electrode along the
creepage distance. In addition, the lines of the field are divergent (outgoing of the
electrode of high voltage).
It comes out from these characteristics the following remarks: (fig.9)
• The electric field is intense at the end of the electrode of high voltage.
• The field will not vanish in the glass but gives a very low value.
• The electric field is practically null inside the two electrodes, because the two
electrodes are conductive.
• The electric field is important in the zone meadows of the active electrode.
• The electric field is null in the clean zones.
The lines of the electric field being perpendicular to equipotential surfaces
leave the electrode under tension and arrive at the level of the electrode ground.
The vectors represented on fig.9 (a) are tangential with the lines of electric field,
we notice that the electric field is more important with dimensions interior of the
electrodes.
64.00 80.00
1.823 1.823
3.33
8.02 3.33
8.02
62.00 12.61
16.32 12.61
30.5 16.32
70.00 30.5
50.4
60.00 93.7 50.4
mS/cm 93.7
mS/cm
58.00
60.00
56.00
Flashover voltage (kV)
54.00 50.00
Flashover voltage (kV)
52.00
40.00
50.00
48.00
30.00
46.00
44.00 20.00
0.00
1.00 3.00 5.00 7.00
Width2.00
1.00
of layer
3.00
pollution (mm)6.00
5.00 7.00
Width 2.00 4.00
of layer pollution 6.00
(mm) 8.00 0.00 4.00 8.00
(a) (b)
Fig. 7 Flashover voltage-width of the layer pollution for different surface conductivity (a)
model insulator), (b) real insulator
4.2 Distribution of the Electric Potential on the Real Model

We were interested in the determination of the distribution of the potential on the
insulators 1512 L. the tension was maintained constant and equal to 30 Kv. This
allows simulating the behavior of insulators of line 220 Kv.
(a)
(b)
Electric field (kV/cm)
(c) leakage distance (cm)

(d)
Fig. 8 Electric field, (a) vector of the electric field,(b) distribution (echograms) of the
electric field, (c)Electric field along the leakage distance,(d) equipotential lines
Electric potential (V)
(b) leakage distance (cm)
(a)
Fig. 9 Distribution (echograms) of the potential (a), electric potential along the
cleakage distance (b)
The figure10 respectively illustrate the variation of the potential along the cree-
page distance and the distribution of the potential of the insulators. The potential is
very important on the level of the electrode of high voltage then decreases as one
moves away from the electrode credit.
• In a dry state, the potential takes the maximum value 30 Kv on the active
level of the electrode then decrease in a linear way as one moves away from this
electrode until the attack of the electrode of mass where the potential is cancelled.
5 Conclusion
In this paper, we have presented the results of different tests on the laboratory’s
model high voltage insulator as well as the real one and simulation of electric field
and potential distribution under normal operating conditions in 220kV AC trans-
mission lines has been carried out using a two dimensional finite element model.
According to the results:
• The flashover reduces with the conductivity of the polluted environment. Oth-
erwise the insulator is more rigid when the conductivity of the surface
increases as well as its width.
• For the dry state, the potential takes the maximum value 30 Kv on the active
level of the electrode then decrease in an almost linear way as one moves
away from this electrode until the attack of the electrode of mass where the
potential is cancelled.
• We also saw that the potential decreases as one moves away from the active
electrode.
• The maximum value of the electric field is obtained on the level of the clean
layers of lower width.
• Field calculations help in improving the design and identification of vulnera-
ble areas.
• Usually in electric field distribution, the electric field magnitude is high at the
energized end and decreases exponentially along the creepage distance.
References
Moula, B., Mekhaldi, A., Teguar, M.: Wavelet Packet Analysis of the Applied Voltage
Waveform as a Monitoring Technique of the Outdoor Insulation Surface. In: IEEE In-
ternational Conference on Solid Dielectrics, Bologna, Italy, pp. 174–177 (2013)
Moula, B., Mekhaldi, A., Teguar, M., Haddad, A.: Characterization of Discharges on non-
Uniformly Polluted Glass Surfaces Using a Wavelet Transform Approach. IEEE Trans-
actions on Dielectrics and Electrical Insulation 20(4), 1457–1466 (2013)
Jaafar, S., Ahmad, A.S., Ghosh, P.S., Aljunid, S.A.K.: A new approach in modeling AC
flashover voltage for polluted insulator. In: 2002 IEEE Conference on Electrical Insula-
tion and Dielectric Phenomena, Universiti Tenaga Nasional, Kajang, Selangor, Malay-
sia, pp. 558–561 (2002)
Zaffanella, L.E., Scheinder, H.M., Dunlap, J.H.: Perfermances des isolateurs polluées pour
lignes ccht, CIGRE, rapport 33-05 (1986)
Dhahbi-Meghriche, N., Beroual, A.: Flashover dynamic model of polluted insulators under
ac voltage. IEEE Transactions on Dielectrics and Electrical Insulation 7(2), 283–289
(2000)
M’hamdi, B., Teguar, M., Mekhaldi, A.: Potential and Electric Field Distributions on HV
Insulators String Used in The 400 kV Novel Transmission Line in Algeria. In: 2013
IEEE International Conference on Solid Dielectrics, Bologna, Italy, June 30-July 4,
pp. 190–193 (2013)
Huang, D., Ruan, J., Cai, W., Li, T., Wei, Y., Liu, J.: Flashover Prevention on High-altitude
HVAC Transmission Line Insulator Strings. IEEE Transactions on Dielectrics and Elec-
trical Insulation 16(1), 88–98 (2009)
Jowkar, M., Gholami, A.: Corona ring designation for 400 kv ac transmission line with
com-posite insulators by FEM. In: 26th International Power System Conference, PSC
2011, Tehran, Iran, October 31-November 02 (November 2011)
Wilkins, R.J., Al-Baghdadi, A.A.J.: Arc Propagation along an electrolyte surface. Proc.
IEEE 118(12), 1886–1894 (1971)
Streubel, H.: The influence of non-uniform contamination on the flashover voltage of insu-
lators. In: 14th International Symposium on High Voltage Engineering, Athens, rapport
46-10 (September 1983)
Mekhaldi, A., Namane, D., Bouazabia, S., Beroual, A.: Flashover of discontinuous pollu-
tion laye ron hv insulators. IEEE Trans. on Dielec. and Elect. Insul. 6(6), 900–906
(1999)
Evaluation by Wide-Angle X-ray Scattering
of the Particle Evolution in Polypropylene
Processed by Equal Channel Angular Extrusion
Ramdane Boulahia1, Taoufik Boukharouba1, and Jean-Michel Gloaguen2
1
LMA, FGMGP/USTHB, BP.32 El-Alia, Bab-Ezzouar, 16111 Alger, Algérie
{ramdane.boulahia,t-boukha}@gmail.com
2
UMET, Université de Lille 1, Rue Longevain, Lille, France
jean-michel.gloaguen@univ-lille1.fr
Abstract. The equal channel angular extrusion (ECAE) is a process used to im-
prove the mechanical properties of materials. Key parameters to achieve this
objective are mainly the geometrical dimensions of the extrusion die and the ex-
trusion parameters such as extrusion rate, the back pressure, the applied load,
etc…. The characterization by the use of X-ray showed that the sample extruded at
45 mm/min, and which has light and dark bands due to the stick-slip phenomenon
polymer, has sheared in the direction of approximately 45° to the extrusion direc-
tion. The using of the Fit2d software can locate the families of crystal planes (hkl)
of the PP. Observations by X-ray in the direction of flow have shown the exis-
tence of three planes (040), (130) and (111). On the other hand, by the observation
in the transverse direction revealed the existence of another plane (131).
Keywords: Equal channel angular extrusion, Polypropylene, Wide-angle X-ray

scattering.
1 Introduction
For the last 20 years, a lot of research efforts in materials science-related fields of
metallurgy, mechanical engineering, and chemical engineering have been focused
on wider development, implementation, commercialization and improvement of
new material forming methods known as severe plastic deformation (SPD)
schemes (Alexander V. Perig et al, 2014). The classical SPD processing method is
Segal’s equal channel angular extrusion (ECAE) material forming technique,
which have found application to processing of metals, alloys, composites, powder
and polymer semi manufactured articles, etc. The ECAE method, developed in the
beginning of the 1980s (Segal et al 1981), is based on one or several extrusion
passes of a lubricated metal, polymer, powder, or composite material through a die
with two intersecting channels of equal cross-section (Fig 1). The works on ECAE
of polymers were initiated relatively recently (Sue H-J, Li CK-Y 1998). In this

282 R. Boulahia, T. Boukharouba, and J.-M. Gloaguen
case, as opposite to metals, ECAE is applied in order to form an oriented structure

that will promote obtaining articles with higher plastic and strength characteristics.
Plastic deformation mechanism of semi-crystalline polymer materials is different
in deformation types. In uniaxial tensile deformation, interlamellar was separated
at initial stage and then crystalline phase was tilted along the tensile direction by
fine slip. Coarse slip leads the separation of crystalline lamellae into segments at
large deformation (Jianhui Qiu et al,2012).
Fig. 1 Schematic illustration of a 90° ECAE die (flow direction: FD, load direction: LD,
transverse direction: TD)
For example, Zuo et al. (2007) evaluated the microstructure change of

polypropylene during uniaxial stretching by X-ray diffraction and observed the fibril
structure performed by atomic forced microscopy. For the simple shear deformation,
the tie molecule extended at initial stage. The chain axis tilted along the shear
direction due to the coarse slip at large deformation. Further deformation, fibril
rotated along the shear direction. Bartczak et al. (1994) reported chain axis tilted
close to shear direction on high-density polyethylene under simple shear. Research
on the application of the ECAE process to polymer materials had mainly focused on
the changes of crystalline deformation (Sue and Li, 1998), mechanical properties (Li
et al., 2000), the effect of processing conditions on the extrusion quality and material
shape (Kitagawa and Kozaki, 2006). Phillips et al. (2006) had reported that the
change of morphology was measured by wide and small angle X-ray scattering and
reflected optical microscopy.
In the present work, we studied the plastic deformation mechanism of
polypropylene by the ECAE process. Our analysis included crystalline
morphology, microstructure of polypropylene after being extruded. Crystalline
microstructure was evaluated using wide-angle X-ray scattering.
Evaluation by Wide-Angle X-ray Scattering of the Particle Evolution 283
2 Analytical Approach of the ECAE Process

The plastic shear strain assigned to a sample as it passes through the ECAE die
depends upon the two angles φ and ψ, and can be given by the following
theoretical expression (Iwahashi et al, 1996):
1 ⎡ ⎛ϕ ψ ⎞ ⎛ ϕ ψ ⎞⎤
εp = ⎢2 cot⎜ 2 + 2 ⎟ + ψ cos ec⎜ 2 + 2 ⎟⎥
(1)
3⎣ ⎝ ⎠ ⎝ ⎠⎦
3 Equipment Used for ECAE with Controlled Back-Pressure

The ECAE die was made of stainless steel. It is schematically given in (Fig. 1). An
internal angle φ of 90° between the two channels, an outer corner angle ψ of 10°
and an inner radius r of 2 mm were adopted. The channels had cross-sectional
dimensions of 10x10mm. The ECAE tooling was adapted on an Instron_ model
5800 screw-testing machine (Fig. 2).
4 Experimental Procedures and Material
4.1 ECAE Experiments

ECAE experiments were conducted at room temperature (about 23°C) and under
constant ram speeds in the range of 0.45–45 mm/min. The material used in this
investigation is a polypropylene (PP) of weight-average molar weight of
180 kg/mol purchased from Goodfellow. The material was supplied in the form of
10 mm thick compression-moulded plates. Specimens were cut from the as
received PP plates along the same direction. The samples were 75 mm in length
with square crosssections 10x10 mm.
(Fig. 3) presents the deformation behaviour of PP specimens at the end of the
ECAE operation performed at a ram speed of 45mm/min. The strain pattern
appears in a very particular manner. Without backpressure (Fig. 3.), the sample
does not fill up the outer corner of the die and exhibits a wavy shape on the top
surface in the exit channel: the summit of the waves is in contact with the surface
of the die whereas the bottom of the waves is not. This suggests a heterogeneous
strain field in the bulk, which is confirmed by the occurrence of dark grey and
light grey alternated stripes inclined at about 45° from FD along the sample
length. The dark and light stripes in reflected light are respectively translucent and
opaque in transmitted light. This means that the former ones have undergone high
shear that imparted severe deformation or destruction of the spherulitic structure
whereas the later ones went undeformed through the transition zone of the ECAE,
keeping intact the initial light-diffusing superstructure. This kind of ≪stick-slip≫
phenomenon, never reported in previous ECAE studies on polymers, may be due
to the combined effects of the specific viscoplastic behaviour of the material and
the friction between the sample and the die, as a result of an unstable balance
between the yield stress and the friction force. This interpretation borrows from
previous studies regarding solid-state extrusion (Krjutchkov et al. 1986).
The effect of backpressure on the plastic flow instabilities is shown in (Fig. 4).
The application of back-pressure significantly contributes to reduce the gap
between the sample and the outer corner of the die. The wave formation is
concomitantly reduced and the alternating stripes tend to disappear with increasing
back-pressure.
Fig. 2 Equipment used for ECAE
Fig. 3 Macrographs of PP samples after ECAE at a ram speed of 45 mm/min without back-
pressure
Fig. 4 Macrographs of PP samples after ECAE at a back-pressure of 900 N and ram speeds
of 0.45 mm/min
4.2 Structural Characterizations

Wide-angle X-ray scattering (WAXS) experiments were performed using the Ni-
filtered Cu Kα radiation from a 2 kW Panalytical sealed tube operated on an INEL
2000 generator at 40 kV and 20 mA. The 2D-WAXS patterns recorded on a digital
CCD camera from Photonic Science Ltd were corrected for background scattering.

After the investigation of the extrusion behaviour of PP samples under various
processing conditions, WAXS experiments are carried out for characterizing the
structural evolution of the materials as a function of processing parameters. Then,
the mechanical properties of extruded specimens are presented and discussed.
5.1 2D-WAXS Patterns of an as-received PP

Similarly as for the extruded material, we prepared lamellae of an as-received PP
to be analyzed by X-ray. We observed the structure in different places of the
sample: at the highest point A, B in the middle of the sample C and at the bottom
(Fig. 5.). The X-ray (WAXS) of these lamellas gave images of (Fig. 6.). The
WAXS patterns recorded with the X-ray beam along the FD show that the
material is isotropic, since the lines observed at points A, B and C are identical.
(a)
(b)
Fig. 5 (a) Lamellae of PP for WAXS measurements, (b) Samples after ECAE
Fig. 6 2D-WAXS patterns, viewing along TD axe of an as-received PP (a) top, (b) middle,
(c) bottom
5.2 2D-WAXS Patterns of PP After ECAE at a Ram Speed of 45

mm/min and Without Back-Pressure
The WAXS patterns recorded with the X-ray beam along the FD are quite
isotropic, whatever the experimental conditions of the ECAE process. This is
illustrated in (Fig. 7.) in the case of a slice taken from a dark stripe of the sample
extruded at 45 mm/min without back-pressure, for three different positions from
top to bottom of the slice, as sketched in (Fig. 5.b). However, much different is the
situation for the edge-view with the X-ray beam normal to the FD-LD plane. In
the case of the higher ram velocity 45 mm/min and no back-pressure, (Fig. 8.)
displays three 2D-patterns recorded from consecutive dark-light-dark stripes
according to the sketch of (Fig. 5b.). The middle light stripe looks roughly
isotropic whereas the dark ones exhibit a strong diagonal reinforcement of the 3
first reflections. This is relevant to a preferred orientation of the (110), (040) and
(130) planes containing the chain axis parallel to the shear direction.
Fig. 7 2D-WAXS patterns, viewing along FD axe, of PP after ECAE at a ram speed of 45
mm/min and without back-pressure: (a) top, (b) middle, (c) bottom.
Fig. 8 2D-WAXS patterns, viewing along TD axe, of PP after ECAE at a ram speed of 45
mm/min and without back-pressure: (a) dark stripe, (b) light stripe, (c) dark stripe (the shear
direction is parallel to the beam stop arm).
6 Processing of 2D-WAXS Patterns by the Using of Fit2d

Software
The 2D-WAXS patterns were recorded on a Photonic Science VHR CCD camera
and stored as digital images using software “Image-Pro”. The Fit2d software is an
investigative tool that can tell us about the degree of crystallinity, microstructure
and orientation of the long chain structure of polymers.
6.1 X-ray Patterns of as-Received PP

Fig. 9 shows the X-ray patterns of as-received PP obtained using Fit2d software.
The calculation by the Bragg allowed us to calculate the inter-planar distance d
and identify the different crystal planes of the material. Miller indices are
deducted from the table given by the International Data Centre of Diffractions RX.
The results show that the PP material is isotropic because it diffracts the x-rays in
the same manner as the top, bottom and middle of the sample.
Fig. 9 X-ray diffraction patterns of as-recevied PP
6.2 X-ray Patterns of PP after ECAE at a Ram Speed of 45

mm/min and Without Back-Pressure
Fig. 10 shows the X-ray patterns of the three observations in the transverse
direction of a strip cut from a sample extruded at a speed of 45mm/min.
Observations in the transverse direction did appear another peak 2θ = 24.6 °.
In addition to the previously observed peaks in the X-ray patterns of the
samples cut in the longitudinal direction, the observation in a transverse direction
allows the plane (131) to diffract with an intensity, which revealed the presence of
this latter. Whatever the light or dark stripes showed no difference in structure.
(a) (b)
Fig. 10 X-ray diffraction patterns of PP after ECAE at a ram speed of 45 mm/min
Table 1 Tab. Directions 1 and corresponding plans for each diffraction peaks
θ° d (cps) Plane (hkl)

1st peak 8.1 5.4348 (040)
2rd peak 9.7 4.57 (130)
3rd peak 10.7 4.14 (111)
7 Conclusion
The characterization by means of X-ray showed that the polymer extruded with a
speed of the 45mm/min, which has light and dark bands due to the phenomenon
stick-slip has sheared in the direction of approximately 45° to the direction of
extrusion. The treatment with Fit2d software allowed us to locate the families of
crystal planes (hkl) of the PP. The observations in the flow direction showed the
existence of three planes (040), (130) and (111). Against this, the observation in
the transverse direction revealed the existence of another plane (131)."
References
Perig, A.V., Golodenko, N.N.: CFD 2D simulation of viscous flow during ECAE through a
rectangular die with parallel slants. Int. J. Adv. Manuf. Technol. (2014),
doi:10.1007/s00170-014-5827-2
Sue, H.-J., Li, C.K.-Y.: Control of orientation of lamellar structure in linear low-density po-
lyethylene via a novel equal channel angular extrusion process. J. Mater. Sci. Lett. 17,
853–856 (1998)
Segal, V.M., Reznikov, V.I., Drobyshevskiy, A.E., Kopylov, V.I.: Russ. Metall. 1, 99–105
(1981)
Qiu, J., Murataa, T., Wu, X., Kitagawa, M., Kudo, M.: Plastic deformation mechanism of
crystalline polymer materials in the equal channel angular extrusion process. Journal of
Materials Processing Technology 212(2012), 1528–1536 (2012),
doi:10.1016/j.jmatprotec.2012.02.015
Zuo, F., Keum, J.K., Chen, X., Hsiao, B.S., Chen, H., Lai, S.Y., Wevers, R., Li, J.: The role
of interlamellar chain entanglement in deformation-induced structure changes during un-
iaxial stretching of isotactic polypropylene. Polymer 48, 6867–6880 (2007)
Phillips, A., Zhu, P.-W., Edward, G.: Simple shear deformation of polypropylene via the
equal channel angular extrusion process. Macromolecules 39, 5796–5803 (2006)
Bartczak, Z., Argon, A.S., Cohen, R.E.: Texture evolution in large strain simple shear de-
formation of high-density polyethylene. Polymer 35, 3427–3441 (1994)
Li, C.K.Y., Xia, Z.-Y., Sue, H.J.: Simple shear plastic deformation behavior of polycarbo-
nate plate II. Mechanical property characterization. Polymer 41, 6285–6293 (2000)
Kitagawa, M., Kozaki, T.: ECAP process in polypropylene. J. Soc. Mater. Sci. 55, 37–41
(2006)
Iwahashi, Y., Wang, J., Horita, Z., Nemoto, M., Langdon, T.: Principle of ECA pressing for
the processing of ultra-fine grained materials. Scripta Mater. 35, 143–146 (1996)
Krjutchkov, A.N., Dorfman, I.Y., Prut, E.V., Enikolopyan, N.: Solid state extrusion of se-
micrystalline polymers. Polym. Compos. 7, 413–420 (1986)
A Multi-site Supply Chain Planning Using
Multi-stage Stochastic Programming
Houssem Felfel, Omar Ayadi, and Faouzi Masmoudi
University of Sfax, National School of Engineers of Sfax, Laboratory of Mechanics,

Modeling and Production,
Road of Soukra km 3.5, BP1173 - 3038 Sfax, Tunisia
houssem.felfel@gmail.com,
omar.ayadi@yahoo.fr,
faouzi.masmoudi@enis.rnu.tn
Abstract. This study focuses on a multi-product, multi-period, multi-site supply

chain planning problem under demand uncertainty. A multi-stage stochastic linear
programming model is proposed to maximize the expected profit. The decisions to
be made comprise the production amount, the inventory and backorder sizes as
well as the quantity of products to be transported between upstream and down-
stream plants and customers in each period. A numerical example is presented in
order to illustrate the effectiveness of the proposed model. Results indicate that
the solution of the multi-stage stochastic model outperforms the deterministic and
the two-stage stochastic models.
Keywords: multi-site, supply chain planning, multi-stage stochastic program-

ming.
1 Introduction
To cope with the continuous highly competition, manufacturing environments

have changed from traditional single-plant to multi-site structure. In the last dec-
ades, multi-site planning problem has received a remarkable interest in the litera-
ture (Lin and Chen 2006, Shah and Ierapetritou 2012, Felfel et al. 2014). Most of
these works have focused on deterministic models. However, real production
planning problems are characterized by many sources of uncertainty such as
deterministic assumptions will lead to non-optimal results.
To cope with uncertainty, researchers have used different methodologies such
as stochastic programming (Kall and Wallace, 1994); fuzzy set theory (Wang and
Fang, 2000); robust optimization (Leung et al. 2007) and stochastic dynamic pro-
gramming (Lin and al. 2014). Methods based on two-stage stochastic program-
ming is one of the most widely spread technique in the literature to incorporate
uncertainty. In this approach, the decision variables of the optimization problem

290 H. Felfel, O. Ayadi, and F. Masmoudi
are divided into two sets. The decision variables of the first stage called “here and
now” decisions have to be made before the realization of uncertainty. Subsequent-
ly, the second stage decision variables are chosen after the presence of uncertain
parameters in order to correct the infeasibilities caused by uncertainty realization
(“wait and see” decisions). Therefore, the value of the objective function is the sum
of first stage decision variables and the second stage expected recourse variables.
Many works in the literature have been interested in two-stage stochastic pro-
gramming model for supply chain planning problem. Gupta and Maranas (2003)
studied a multi-product, multi-period multisite supply chain tactical planning
problem under demand uncertainty based on a two-stage stochastic programming
approach. Leung et al. (2006) proposed a two-stage stochastic programming mod-
el in order to optimize a multi-site aggregate medium-term production planning
problem under an uncertain environment. Mirzapour Al-e-hashem et al. (2013)
proposed a multi-site multi-period multi-product aggregate production medium-
term planning problem in a green supply chain under demand uncertainty using a
two-stage stochastic programming approach.
However, adopting a two-stage stochastic approach to address the uncertain
multi-period production planning problem cannot effectively reflect the dynamic
variations of the system condition. The multi-stage stochastic programming for-
mulation is an extension of the two-stage stochastic formulation (Birge and Lou-
veaux, 1997). The multi-stage stochastic model allows the decision maker to
revise and to update the production plan through time period evolution and when
more information about the customer demand scenarios becomes available. Many
papers have addressed such situation in the literature. Ahmed et al. (2003) pro-
posed a multi-period capacity expansion problem under demand and investment
costs uncertainties. A multi-stage stochastic programming model was developed in
order to incorporate the uncertain parameters. Karabuk (2008) developed a multi-
stage stochastic programming formulation to deal with a yarn production planning
problem under demand uncertainty in textile industry. Kazemi Zanjani et al.
(2010) proposed a multi-product, multi-period production sawmill planning prob-
lem under quality of raw materials and demand uncertainty using a multi-stage
stochastic programming model. To the best of our knowledge, the multi-stage sto-
chastic programming approach has been rarely applied in multi-site supply chain
planning problem.
The objective of this paper is to develop a new mathematical model in order to
maximize the net profit of a multi-product, multi-period, multi-site supply chain
planning problem under demand uncertainty using a multi-stage stochastic pro-
gramming approach. A numerical example is illustrated to demonstrate the effec-
tiveness of the proposed model compared to the traditional deterministic model.
The remainder of the paper is organized as follows. In the next section, we
present the problem statement. Section 3 details the stochastic mathematical for-
mulation of the multi-site manufacturing network planning problem. Section 4
provides an illustrative example and corresponding analyses followed by the
conclusion and future work in section 5.
A Multi-site Supply Chain Planning Using Multi-stage Stochastic Programming 291
2 Problem Statement
In this paper, we consider a multi-site supply chain wherein the finished product is
processed by means of different production stages. Each production stage may in-
clude more than one plant forming a multi-site supply network manufacturing
environment as illustrated in Fig. 1. Within such multi-plant supply network, pro-
duction and supply decisions should be coordinated in order to achieve high per-
formance. Since these plants are situated in different areas, the transportation of
semi-finished and finished products between the entities of the supply chain is
considered. It should be noted, that a delivery lead time is taken into account in
the transportation of semi-finished product.
The fluctuation of customers’ demand is one of the most important sources of
uncertainty that should be considered in supply chain planning problems. Indeed,
the exact customer demand cannot be determined before the end of the planning
horizon. The under-estimation of overall demand leads either to loss sales and un-
satisfied customer. However, the over-estimation of the products demand results
in high production and inventory costs.
The objective of the supply network planning problem is to maximize the ex-
pected profits, computed by subtracting the expected total costs from the total rev-
enues. Decisions to be decided include the production amount, the inventory and
backorder sizes and the flows of materials between the different plants and cus-
tomers taking into account the demand uncertainty.
Fig. 1 Multi-site manufacturing network environment
3 Multi-stage Stochastic Programming Formulation
3.1 Multi-stage Stochastic Programming Approach

The multi-stage stochastic programming approach is an extension of the two-stage
stochastic approach to deal with multi-period optimization problem under uncer-
tainty. Indeed, the stages correspond to time periods in the planning horizon and
every stage may include more than one period. The evolution of the uncertainties
is represented using a scenario tree. The scenario tree is formed of many nodes
and arcs at each stage as illustrated in Fig. 2. The different nodes in each stage are
used to represent the stochastic parameters and the arcs denote the scenarios for
the following stage. Every arc of the scenario tree is associated with a probability,
which is the probability of scenario that correspond to this arc.
Fig. 2 Scenario tree
A multi-stage stochastic programming formulation is proposed based on this

scenario tree for the uncertain demand. The production amounts ( Pikt ) and the
quantity of products to be transported between the different plants (TRi −>i ', kt )
represent the control decision while the inventory ( S ikt ) and backorder size ( BDkt )
as well as the amounts of finished products shipped to the customer (TRi −>CUS ,kt )
are the consequences or state variables. In this problem, we assume that there is
enough information available of customer demand at the beginning of each stage
that allows the decision maker to adjust the decision variables ( Pikt ) and
(TRi −>i ',kt ) for the overall demand scenarios. On the other hand, the consequences
decisions variables ( Sikt ) , ( BDkt ) and (TRi −>CUS ,kt ) should be fixed for the different
demand scenarios at each stage of the scenario tree.
4 Multi-stage Stochastic Mathematical Formulation
Indices
Li Set of direct successor plant of plant i.
ST j Set of stages (j= 1,2, ..., N).
i, i’ Production plant index (i,i’ = 1, 2 , ...,I) where i belongs to stage n
and i’ belongs to stage n+1.
k Product index (k = 1,2, ..., K).

t Period index (t = 1,2, ..., T).
S Scenario index (s = 1,2,. . .,S).
tn Set of time periods related to node n.
n, m Node of scenario tree (n = 1,2,. . .,Nn) where
⎛ Pre(n), t − 1∉ tn
m=⎜
⎝ n, t − 1 ∈ tn
Pre() Predecessor of the node n in the scenario tree.
Decision variables
Pikt (n) Production amounts of product k at plant i in period t at node n.
Amounts of end of period inventory of product k for scenario s at
S ikts ( n ) plant i in period t at node n.
s Amounts of end of period inventory of semi-finished product k
JSikt (n) for scenario s at plant i in period t at node n.
s Backorder amounts of finished product k for scenario s in period t
BDkt (n) at node n.
TRi −>i ',kt (n) Amounts of product k transported from plant i to i’ in period t at
node n.
s Amounts of product k transported from the last plant i to customer
TRi −>CUS ,kt (n) for scenario s in period t at node n.
Amounts of product k received by plant i for scenario s in period t
Qi ,k ( n)
at node n.
Parameters
s
D kt Demand of finished product k for scenario s in period t.
The occurrence probability of scenario s for the random demand
S
π s
where ∑π s
=1
s =1
πn Probability of the node n in the scenario tree
The multi-stage stochastic model

Nn S T K I
Max E [ NPV ] = ∑π n ∑π s ∑∑∑ prk TRis−>CUS ,kt (n) − cbk BDks,t (n)
n =1 s =1 t =1 k =1 i =1
−csik (S (n) + JS (n)) − cti −>CUS ,kTRis−>CUS ,kt (n) − cti −>i ',k TRi −>i ',kt (n)
s
ikt
s
ikt (1)
Nn T K I
−∑π n ∑∑∑ cpik Pikt (n)
n =1 t =1 k =1 i =1
Siks ,t (n) = Siks ,t −1 (m) + Pikt (n) − ∑ TRi −>i ',kt (n), ∀i ∈ STj < N , ∀k , t, s, n (2)
i '∈Li
I I
∑ Siks ,t (n) = ∑ Siks ,t −1 (m) + Pikt (n) − TRis−>CUS ,kt (n),

i =1 i =1 (3)
∀i ∈ ST j= N , k , t , s, n
JSiks ,t ( n) = JSiks ,t −1 ( m) + Qikt ( n) − Pikt ( n), ∀i, k , t , s, n (4)
BDkts ( n) = BDks,t −1 ( m) + Dkts ( n) − TRis−>CUS ,kt ( n), ∀k , t , s, n (5)
Qi ' k ,t + DL (n) = ∑ TRi −>i ',kt (n), ∀i, k , t , s, n (6)

i '∈Li
K
∑b P
k =1
k ikt (n) ≤ cappit , ∀i, t , n (7)
∑S
k =1
s
ikt (n) + JSikts (n) ≤ capsit , ∀i, t , s, n (8)
∑ TR
k =1
i −> i ', kt (n) ≤ captrit , ∀i, t , s, n (9)
Pikt ( n), S ( n), JSikts ( n), TRi −>i ',kt ( n), TRis−>CUS ,kt ( n), Qi ,k ( n),
s
ikt
(10)
BDkts ( n) ≥ 0, ∀i , k , t , s , n
The objective function (1) aims to maximize the expected profit for different
scenarios of product demand by subtracting the total expected cost from the ex-
pected revenue. The equation (2) provides the balance for the inventory of prod-
ucts in every production stage except for the last stage. Constraint (3) represents
the balance for the inventory for the last production stage, considering customer
demands. Constraint (4) is the inventory balance equation for the semi-finished
products. Constraint (5) provides the balance equation for shortage in end product
demand. Equation (6) represents the balance equations for the transportation be-
tween the different plants. The set of constraints (7), (8) and (9) makes sure that
the production capacity, the storage capacity and the transportation capacity
are respected. Constraint (10) is the non-negativity restriction on the decision
variables.
5 Illustrative Example
In this section, an illustrative example involving a multi-site supply chain network

is presented in order to show the effectiveness of the proposed model. The consi-
dered supply chain network consists of 5 production stages with 8 serial and paral-
lel plants and 2 products. The planning horizon includes 10 periods and the length
of a period is one week. Four scenarios are considered for the random demand of
two finished products P1 and P2 under period T6, T8 and T10 as reported in
Table 1. The total number of demand scenarios is equal to 64 scenarios ( 43 = 64 ).
The proposed model is then solved and the quality of the obtained solutions is
compared as detailed in section 4.1. The experiments are conducted using LINGO
14.0 package program and MS-Excel 2010 with an INTEL(R) Core (TM) and 2
GB RAM.
Table 1 Random finished product demand with corresponding probability
T6 T8 T10
Scenario
P1 P2 PROB P1 P2 PROB P1 P2 PROB
S1 3350 3230 0.25 3300 3310 0.25 3430 3210 0.2
S2 2650 2520 0.35 2730 2770 0.4 2720 2530 0.3
S3 1860 1550 0.2 1930 2040 0.15 1950 1860 0.25
S4 1240 1020 0.2 1180 1060 0.2 1010 930 0.25
5.1 Results of Stochastic and Deterministic Programming

Models
In this section, we compare the solution of the four-stage stochastic programming
model (4SM) to the solution of the three-stage (3SM) and two-stage stochastic
programming model (2SM) as well as the solution of deterministic model
(DTM).The solution of deterministic model is calculated based on the mean value
of the uncertain demand. For the four-stage programming model, the planning ho-
rizon is clustered into 3 levels and the multi-stage decision process is approx-
imated by a four-stage one. The first stage is the time period zero, the second stage
consists of periods 1-6, the third stage includes periods 7-8 and the fourth period
includes periods 9-10. For the three-stage programming model, the first stage con-
sists of the time period zero, the second stage consists of periods 1-6, the third
stage includes periods 7-10. However, for the two-stage programming approach
all the time periods are considered as a one stage. In the two-stage stochastic mod-
el, the production amounts in each plant and the product amounts to be transported
between upstream and downstream plants are taken “here and now” before the
realization of the uncertainty. Other decision variables such as inventory and
backorder size, flow of finished products to be shipped to the customer are made
in a “wait and see” mode until the realization of random demand.
According to Fig. 3 and Table 2, the expected net profit of the supply chain
production plan of (4SM), (3SM) and (2SM) models are significantly higher than
those of the deterministic model. Moreover, the expected inventory and backorder
costs of the deterministic model are considerably greater than those of the (4SM),
(3SM) and (2SM) models. It is worth remarking that when the number of stages
increases in the demand scenario tree, the expected net profit increases too. In-
deed, the expected profit of (4SM) model is higher than those of (3SM) and (2SM)
models. Furthermore, Fig. 3 shows that the expected profit of (4SM) and (3SM)
models are close to each other. So, there is no need to consider more stages in the
scenario tree.
Fig. 3 Expected values of different supply chain planning models
In order to evaluate the influence of uncertainty on the planning decisions two

stochastic measures was used. The first metric is the expected value of perfect in-
formation (EVPI). This parameter helps to determine the expected profit loss
under uncertainty (Birge and Louveaus 1997). It can be calculated as:
EVPI= WS –2SP (11)
Where WS represents the objective value of the “wait and see” model and 2SP
is the objective value of two-stage stochastic programming model.
The second measure used in order to evaluate the performance of multi-stage
programming model is the relative value of the multi-stage stochastic solution
(RVMSS) (Nickel et al. 2012). This measure shows how much saving can be rea-
lized using the stochastic programming model in comparison with the determinis-
tic model. It is formulated as follows:
SP − EEV (12)
RVMSS = 100
EEV
Where EEV represents the optimal solution of deterministic model.
The obtained results are reported in Table 2 and Table 3. As shown in Table 2,
the EVPI/WS ratio shows a big influence of demand uncertainty on the obtained
solution (19.26%). So, it is worth to have better forecast about the demand scena-
rios. We can see also from Table 3 that the relative value of the multi-stage sto-
chastic solution RVMSS of the 4SM, 3SM and 2SM models are 20.33%, 19.19%
and 16.53%, respectively. It is clear that the multi-stage programming solution of
4SM and 3SM outperforms the two-stage programming solution for the supply
chain planning problem.
Table 2 Expected net profit of different models
WS EEV 2SP 3SP 4SP

141720 98193.06 114419.9 117035.02 118153.26
Table 3 Stochastic programming metrics
EPVI EPVI/WS RVMSS (2SP) RVMSS (3SP) RVMSS (4SP)

27300.09 19.26% 16.53% 19.19% 20.33%
6 Conclusion
In this paper, we have developed a multi-site supply chain production planning
model under demand uncertainty. A multi-stage stochastic programming approach
is applied to incorporate the effects of uncertainty in the production planning prob-
lem. The computational results show that the quality of the four-stage and three
stage stochastic model solutions is higher than those of the two-stage stochastic
and deterministic models. Managing risk by incorporating a risk metric into the
stochastic programming model is one of the directions for future research.
Acknowledgements. We would like to acknowledge the financial support provided by the

Mobility of researchers and research for the creation of value (MOBIDOC) as well as
LINDO Systems, Inc for giving us a free educational research license of the extended ver-
sion of LINGO 14.0 software package.
References
Ahmed, S., King, A., Parija, G.: A multi-stage stochastic integer programming approach for
capacity expansion under uncertainty. Journal of Global Optimization 26, 3–24 (2003)
Birge, J.R., Louveaux, F.: Introduction to stochastic programming. Springer, New York
(1997)
Felfel, H., Ayadi, O., Masmoudi, F.: Multi-objective Optimization of a Multi-site Manufac-
turing Network. In: Mechatronic Systems: Theory and Applications. Springer (2014)
Gupta, A., Maranas, C.D.: Managing demand uncertainty in supply chain planning. Com-
puters and Chemical Engineering 27(8-9), 1219–1227 (2003)
Kall, P., Wallace, S.W.: Stochastic Programming. John Wiley and Sons, New York (1994)
Kazemi Zanjani, M., Nourelfath, M., Ait-Kadi, D.: A multi-stage stochastic programming
approach for production planning with uncertainty in the quality of raw materials and
demand. International Journal of Production Research 48(16), 4701–4723 (2010)
Leung, S.C.H., Tsang, O.S., Ng, W.L., Wu, Y.: A robust optimization model for multi-site
production planning problem in an uncertain environment. European Journal of Opera-
tional Research 181, 224–238 (2007)
Leung, S.C.H., Wu, Y., Lai, K.K.: A stochastic programming approach for multi-site ag-
gregate production planning. Journal of the Operational Research Society 57, 123–132
(2006)
Lin, J.T., Chen, Y.Y.: A multi-site supply network planning problem considering variable
time buckets– A TFT-LCD industry case. The International Journal of Advanced Manu-
facturing Technology 33(9-10), 1031–1044 (2006)
Lin, J.T., Chen, T.L., Chu, H.C.: A stochastic dynamic programming approach for multi-
site capacity planning in TFT-LCD manufacturing under demand uncertainty. Interna-
tional Journal of Production Economics 148, 21–36 (2014)
Mirzapour Al-e-hashem, S.M.J., Baboli, A., Sazvar, Z.: A stochastic aggregate production
planning model in a green supply chain: considering flexible lead times, nonlinear pur-
chase and shortage cost functions. European Journal of Operational Research (ISI) 230,
26–41 (2013)
Nagar, L., Jain, K.: Supply chain planning using multi-stage stochastic programming.
Supply Chain Management: An International Journal 13(3), 251–256 (2008)
Nickel, S., Saldanha-da-Gamac, F., Zieglera, H.P.: A multi-stage stochastic supply network
design problem with financial decisions and risk management. Omega 40(5), 511–524
(2012)
Shah, N.K., Ierapetritou, M.G.: Integrated production planning and scheduling optimization
of multisite, multiproduct process industry. Computers and Chemical Engineering 37,
214–226 (2012)
Wang, R.C., Fang, H.H.: Aggregate production planning in a fuzzy environment. Interna-
tional Journal of Industrial Engineering-Theory, Applications, and Practice 7(1), 5–14
(2000)
A System Engineering Conception of Multi-
objective Optimization for Multi-physics System
Mian Chen1 and Omar Hammami2
1
IRT SystemX, France
mian.chen@iry-systemx.fr
2
ENSTA ParisTech, France
omar.hammami@ensta-paristech.fr
Abstract. At present, a system involves more and more domains and becomes
more and more complex. In this paper, a system engineering conception of multi-
objective optimization for a multi-physics system is formed. An example of multi-
physics system design illustrates the presented approach.
Keywords: System Engineering, Multi-Physics Simulation, Multi-Objective
Optimization.
1 Introduction
A system recently deals with more and more domains like mechanical, electronic
and thermal etc. In addition, the interaction and information exchanges among
them increase rapidly. Therefore, the complexity of system is increasing gradual-
ly. As the complexity of a system augments, the work of system optimization be-
comes not only heavier but also more difficult. Besides, the cost and time for the
whole system engineering design augment [9].
System engineering flow plays a crucial role because it should have a global
view of the whole system and optimize the whole system in consideration of mul-
tiple objectives in different domains. Also, it ought to insure that the system satis-
fies some requirements [8].
For the moment, in order to develop a multi-physics system, the experts in dif-
ferent domains have to hold a lot of meetings to reach agreement. However, a
method of automatic system engineering working flow with multi-objective opti-
mization simultaneously is expected.
2 Multi-objective Optimization System Engineering Flow for

a Multi-physics System
In Fig.1, a multi-objective system engineering flow for multi-physics system is
represented. At the top level, by SysML, a system engineering model which
includes all relative system information is defined. At the bottom level, there are a
lot of different physical system models and each model could be simulated sepa-
rately. At central level, a model of multi-objective optimization is built and it
connects the top level and the bottom level.

300 M. Chen and O. Hammami
Fig. 1 Multi-objective optimization system engineering flow
In order to realize the expected system engineering flow and take into account
the feasibility, several software are chosen as Fig.2. Rhapsody is chosen as a tool
to deal with system engineering model which presents the system’s requirements
and describes all the structural, parametric information by language SysML [6].
At the bottom level, several different software could be applied. Catia is chosen
to simulate the geometrical model, Dymola is used to simulate the mechatronic
model and Comsol is chosen to do the mechanical and thermal simulation.
Certainly, other software of physical simulation would be added in this multi-
objective system engineering working flow to realize different simulations.
Besides, the software ModelCenter is taken use to do the multi-objective opti-
mization, which would harmonize multiple simulations in different software and
on the other side, it could collaborate with Rhapsody [7].
Fig. 2 Multi-objective optimization system engineering flow with different software
3 Multi-objective Optimization System Engineering Flow for

a Study Case – Power Closure System
In this paper, a study case of Power Closure System [10] is chosen to illustrate the
proposed multi-objective optimization system engineering working flow. The goal
of this system is that by electronic button control, two little motors could drive
A System Engineering Conception of Multi-objective Optimization 301
respectively two jacks to open or close the whole liftgate automatically which
could replace traditionally manual operation.
3.1 Rhapsody Model

In Fig.3, among all of the system engineering model, only the system structure de-
finition is shown. Overall, the power closure system has four parts: control part,
mechanical part, motor and liftgate. And the liftgate includes a piece of glass and a
frame made of metal [11] [12].
Fig. 3 Rhapsody model
3.2 Catia Model

For the geometrical model, according to those
existing liftgate in market, a very simple 3D
CAD model of liftgate is designed and its form
is shown in Fig.4. This model has two different
parts, one part is the frame which is made of
metal and the other part is the window which is
made of transparent glass. Fig. 4 Catia model
3.3 Comsol Model

To create the mechanical model in Comsol, the geometrical model of Catia could be
imported directly in Comsol. In Comsol, a moment when the liftgate has already
opened to the maximal angle is simulated. Supporting that at the moment, the force is
still perpendicular to the principal liftgate and the value is 200N. Then, the stress and
displacement distribution of the whole liftgate could be seen in Fig.5 and Fig.6.
Fig. 5 Comsol model - Stress distribution Fig. 6 Comsol model - Displacement distri-
bution
3.4 Dymola Model

The motor PLE-29 is chosen as model which is furnished by Johnson Electric and
applied in Power Lift Gate System. According to the parameters sheet and several
important indices, a model of Motor is obtained and shown in Fig.7.
Fig. 7 Dymola model - Motor

In order to do the liftgate model, multi-body library is used. A reduction gear

box and a ball screw is chosen as one part of the jack. The whole liftgate with two
arms is shown in Fig.8.
Fig. 8 Dymola model - Liftgate
Fig. 9 Dymola model - 3D Animation open- Fig. 10 Dymola model - 3D Animation

ing closing
After modeling of motor and liftgate, those two parts could be combined. In the
Dymola model, the motor control part is simplified, a constant tension is supposed
to supply those two motors which drive the whole liftgate. Finally, a 3D animation
simulation has been done. The whole process of opening or closing is between 7
seconds and 8 seconds (Fig.9 and Fig.10).
4 Multi-objective Optimization
Before running the whole system multi-objective optimization, one important con-
ception should be emphasized: an optimization problem with more than one objec-
tive, there is no single optimum solution. It exists a number of solutions which are
all optimal. Without any further information, no solution from the set of optimal
solutions can be said to be better than any other.
4.1 NSGA-II
Compared with other multi-objective algorithms like Multi-Objective Simulated
Annealing (MOSA) [5], Multi-Objective Optimization using Evolutionary Algo-
rithms (MOEA) [4] and Multi-Objective Genetic Algorithm II (MOGA-II) [3], an
algorithm called Non-dominated Sorting Genetic Algorithm II (NSGA-II) [1] [2]
is selected to achieve the multi-objective optimization in this paper.
The algorithm NSGA-II involves with three main conceptions: non-dominated
sort, crowding distance and genetic operators. Its main loop of optimization
process is presented in Fig.11.
Fig. 11 NSGA-II algorithm
4.2 ModelCenter Model

After imposing the Rhapsody model into ModelCenter, the model lists all the in-
put variables and output variables.
In the tool ModelCenter, for running the optimization process, at first, accord-
ing to the different goals, the objective function should be defined. Secondly, the
constraints are indicated as Fig.12. Thirdly, the decision variables should be de-
fined. Finally, the NSGA-II has been selected as algorithm to do the optimization.
After defining all the parameters of multi-objective optimization as above, the
optimization process could be launched, and the results would be presented in
Fig.13. The red points represent infeasible solutions (which don’t meet con-
straints), the green ones represent feasible Pareto solutions and the blue ones
represent the optimal Pareto (which represents optimal solutions and no one could
be better than the others).
Fig. 12 NSGA-II objectives and constrains
Fig. 13 Results
5 Conclusion
This work proposes a multi-objective optimization system engineering working

flow for multi-physics system which could be regarded as an automatic new me-
thodology to reduce design cost and cycle time. Although there are still some de-
tails which have to be completed and improved, the main conception has been
realized.
Even though the main conception is realized in this paper, from the industrial
viewpoint, this study case is too simple. Therefore, in the future work, some more
connections problems between those software and models synchronization should
be solved.
Acknowledgment. For the members in the laboratory U2IS and SUPMECA, here we ex-
press all our esteem and sincere acknowledgement to them for their encouragement, friend-
ship moral support and their scientific assistants for all the time.
References
[1] Deb, K.: Multi-Objective Optimization Using Evolutionary Algorithms, Chichester,
U.K. (2001)
[2] Deb, K., Pratap, A., Agarwal, S., Meyarivan, T.: A Fast and Elitist Multiobjective
Genetic Algorithm: NSGA-II. IEEE Trans. Evol., Comput. 6(2), 182–197 (2002)
[3] Poles, S.: MOGA-II An improved Multi-Objective Genetic Algorithm. Esteco, Italy
(2003)
[4] Seshadri, A.: Multi-Objective Optimization using Evolutionary Algorithms (MOEA)
[5] Rigoni, E.: MOSA Multi-Objective Simulated Annealing, Esteco, Italy (2003)
[6] Systems Engineering Tutorial for Rational Rhapsody, IBM (2009)
[7] MBSE PAK for Rhapsody USER GUIDE, Phoenix Integration, Blacksburg, USA
(2014)
[8] Hammami, O.: SYNSYS-ME: Seamless System Engineering to mechanical flow
through multiobjective optimization and requirements analysis. In: 8th Annual IEEE
Systems Conference (SysCon), Ottawa, Canada (2014)
[9] Hammadi, M., Choley, J., Penas, O., Riviere, A., Louati, J., Haddar, M.: A new mul-
ti-criteria indicator for mechatronic system performance evaluation in preliminary de-
sign level. In: 9th France-Japan Mechatronics, Paris, France (2012)
[10] Raka, S.: Methodes et outils ensemblistes pour le pre-dimensionnement de systemes
mecatroniques. Ensea, Cergy-Pontoise, France (2011)
[11] Peak, R.S., Burkhart, R., Friedenthal, S., Wilson, M., Bajaj, M., Kim, I.: Simulation-
Based Design Using SysML Part 1: A Parametrics Primer. In: INCOSE International
Symposium, San Diego, U.S.A. (2007)
[12] Peak, R., Burkhart, R.M., Friedenthal, S.A., Wilson, M.W., Bajaj, M., Kim, I.: Simu-
lation-Based Design Using SysML Part 2: Celebrating Diversity by Example. In:
INCOSE Intl. Symposium, San Diego, U.S.A. (2007)
[13] Hammadi, M.: Contribution a l’integration de la modelisation et la simulation multi-
physique pour la conception des systemes mecatroniques. In: LISMMA and
SUPMECA, Paris, France (2012)
Falling Film in a Heated Micro-channel
Sabrine Mejri1, Olivier Fudym2, Jalila Sghaier1, and Ahmed Bellagi1
1
Research unit of heat and thermodynamics of industrial processes
National Engineering School of Monastir
Rue Ibn Al Jazzar 5000 Monastir, Tunisia
sabrynener@yahoo.fr
2
Research Center of Albi in Process Engineering, Divided Solids, Energy and Environment
Ecole des Mines d'Albi Campus Jarlard - Route de Teillet
81013 Albi CT Cedex 09, France
fudym@mines-albi.fr
Abstract. The main objective of this work is to study experimentally and numeri-
cally a falling film in a micro-channel. The experimental section involves in creat-
ing a temperature gradient within the liquid, while monitoring the temperature
using an infrared camera. A numerical model is established and solved by a semi-
analytical method called the thermal quadrupole method. Finally, we conclude
with a comparison between the experiments and the numerical study.
Keywords: falling film, micro-channel, thermal quadrupole.
1 Introduction
Miniaturization is currently a promising field of both theoretical and experimental
research (Aiello et al. 2009; Foerster et al. 2013). It is to reduce the dimensions of
some industrial equipment in order to reduce the technological constraints in favor of
environmental chemistry and processes safety (Brendner et al. 2013). It is targeting,
in general, refrigeration equipments, air conditioning and heat pumps which have
their effects on the environment, the layer ozone and global warming (Luke et al.
2010). Thus, comes the idea of increasing processes. This means: developing and
creating methods and devices offering improved production quality, reducing the size
/ capacity, lowering effluent quantities and / or catalysts, and reducing costs and
energy losses with a greater respect for environmental constraints (Mhiri.N 2006).
Micro-reactors operate on the principle of a continuous process and are highly
different from those several traditional characteristics such that the increase of the
exchange surface through the ratio S / V (Perry et al. 2010), the improvement of heat
and mass transfer and a high temperature gradient and pressure (Baldassari et al.
2013; Conti et al. 2012). They help maintain better control of reaction conditions,
improved security conditions, the gain in energy and reducing congestion (Hung et al.
2013). The falling films are generally used as an intermediary for the transfer of heat

308 S. Mejri et al.
and mass transfer (Zhang 2008; Bu et al 2011) in industrial equipment such as

condensers and the evaporation towers (Marenzana et al. 2004; Buonomo et al.
2012). An extension of thermal quadrupole formalism is proposed for the modeling
of heat conduction in heterogeneous environments, where the change in thermal
properties is one-dimensional (Fudym 2002). The main problem was to find an
intrinsic relationship between generalized temperature and heat flux at the boundary
of a heterogeneous medium.
2 Experimental Set Up and Procedure
Our experiment consists in: creating a temperature gradient, in the back of the plate,
and then managing the liquid flow (water and ethanol). An infrared camera is used in
order to visualize the effect of the temperature gradient on the flow.
2.1 Preparation
First of all, the plate was cleaned well and then sprayed with a black spray to create a
black thin layer in order to maximize radiation, since an infrared camera is used for
the visualization of the heat transfer on the surface of the falling film. Then, an
electrical resistance is implanted in the back of the plate to generate heat.
Fig. 1 Experimental Set up

Falling Film in a Heated Micro-channel 309
In order to create a temperature gradient, a heat exchanger is placed in contact with

the back face of the plate to maintain a low temperature.
The infrared camera is a FLIR A320 (COMPACT range 19,600 pixels), which
records the various infrared radiation emitted by the body and which vary with their
temperature. This thermal imaging camera does not capture what is behind a wall or
an obstacle. Its spectral range is from 12 to 8 microns, and since water is almost
opaque in this field then it is certain that the radiations captured by this camera are
those of the liquid surface.
To achieve the flow in the micro-channel, we used a syringe and a very thin tube.
In this experiment we used two liquids: water and ethanol. The flow of ethanol was
smoother because its surface tension is lower than that of water.
2.2 Results of the Experiments

After processing the images, the results obtained are as follows: In figures (Fig. 2 and
Fig. 3), we visualized the ethanol falling film flow in a micro-channel, wherein it is
noted that initially the temperature of the ethanol at room temperature (20 ° C),
increases, once the flow reaches the heating zone to 32 ° C and then begins to
decrease through the cooling zone to 23 ° C.
Fig. 2 IR image: ethanol falling film on a channel of the micro-structured plate

310 S. Mejri et al.
Ethanol surface Temperature in a

micro-channel
Ethanol surface temperature (°C)
40
30
20
10 Ethanol surface
0 Temperature in a
micro-channel
10.0625
10.9375
11.8125
12.6875
13.5625
14.4375
2.1875
3.0625
3.9375
4.8125
5.6875
6.5625
7.4375
8.3125
9.1875
Lenght (cm)
Fig. 3 Curve of the surface temperature of the ethanol falling film in a single channel
Fig. 4 IR image: water falling film on a channel of the micro-structured plate
For the water (Fig. 4 and Fig. 5), the temperature reached 33°C, and then went
down to 18°C. In both experiments, we calculated the temperature gradient and found
it 2.8°C per centimeter which is close to 3°C/cm, the value we wanted to reach.
Water surface Temperature in a

Water surface temperature (°C)
micro-channel
40
30
20
10 Water surface
Temperature in a
0 micro-channel
0.924
1.848
2.772
3.696
5.544
6.468
7.392
8.316
0
10.164
11.088
12.012
12.936
4.62
9.24
Lenght (cm)
Fig. 5 Curve of the surface temperature of the falling film of ethanol for one channel
3 Mathematical Model
3.1 Governing Equations

The plate we are using is a stainless micro-plate in which the micro-structured
channels have the following dimensions:
Width: 1000 microns
Depth: 150 microns
Length: 15 cm
The fluid flow is in a free fall flow governed by gravity and in permanent re-
gime. The flow assumptions are as follows:
• 2D permanent regime.
• Ideal fluid (no shear stresses).
• Constant physical properties.
• No effect of surface tension.
• Very thin liquid film.
Applying the Navier-Stocks equations on our system and taking the boundary
conditions, we get these equations:
y
u = um (1 − (1 − )²) (1)
δ
∂T ∂ ²T ∂ ²T
u = α( + ) (2)
∂x ∂x ² ∂y ²
312 S. Mejri et al.
∂ ²T ∂ ²T
+ =0 (3)
∂x² ∂y ²
Equation (2) is the result of the energy equation in the falling film and Equation
(3) is the result of the energy equation in the micro-channel.
3.2 Solving with the Thermal Quadrupoles Method

We will adopt the method of thermal quadrupoles to solve this problem, which is
mainly finding the relationship between flow and temperature at input and output by
applying the Fourier Transform (Fig. 6).
Fig. 6 Input and output of the system
Transforming all the equations previously obtained along with the boundary
conditions gave us these results:
In the plate:
⎛ 1 ⎞
⎛θ 0 ⎞ ⎜ cos pl sin pl ⎟⎛ θ l ⎞
⎜⎜ ⎟⎟ = ⎜ λ p ⎟⎜⎜ ⎟⎟ (4)
⎝ φ 0 ⎠ ⎜ − λ p sin p l φ
cos pl ⎟⎠⎝ l ⎠
⎝
In order to use the thermal quadrupoles method on the liquid, we proceed by
dividing the film into N layers having each one the same “Eq. 1”. So the matrix in
each layer is:
(5)
And finally we have the matrix of our system:
⎛ 1 ⎞ ⎛ δ 1 δ⎞
⎛θw ⎞ ⎜ cos pl sin pl⎟ N ⎜ coshKi N λK sinhKi N⎟⎛θδ ⎞
⎜⎜ ⎟⎟ =⎜ pl ⎟*∏⎜ ⎟⎜ ⎟ (6)
δ ⎟⎜⎝φδ ⎟⎠
i
⎝φw ⎠ ⎜−λ psin pl ⎟ i=1 ⎜ δ

⎝ cos pl ⎠ ⎜λKi sinhKi coshKi ⎟
⎝ N N ⎠
With the inverse Fourier transform, we find the expression of the surface tem-
perature of the falling liquid film which is simulated then to make the profile:
T∂ = Re(θ ∂ 0 + ∑ θ ∂* exp( pxm (ix ))) / L (7)
Fig. 7 Simulated water surface temperature evolution
The Figure (Fig. 7) shows the curve of the temperature profile at the surface of the
flow. We can see the increase in the temperature in the heating zone.
314 S. Mejri et al.
Fig. 8 Simulated ethanol surface temperature evolution
Note that the temperature rises from 22.8 ° C, the input channel temperature and
the beginning of the heating zone, at x = 0.018 m, reached its peak at x = 0.052 m
with a temperature of 27.4 ° C, and begins to fall back to its initial temperature. This
drop is due to the periodicity of the proper function (temperature), since the resolution
was made by the integral Fourier transform. Same thing for the ethanol graph
(Fig. 8), rising from a temperature close to 22°C, the liquid exceeded 28°C in the
heating zone and then come down to the cooling temperature.
We studied also the changes in temperature within the liquid along the y axis. We
note that there is not a great change throughout the y axis which means we can avoid
the division of our multilayer medium, that is to say we can keep a single mesh.
There is, also, a resemblance between the numerically determined temperature profile
and that determined experimentally. From the figure that the temperature profile in
the flow, we can conclude that the independence with respect to y means that the
temperature can be considered as a single unit with one average speed.
We managed to achieve a temperature gradient close to 3 ° C / cm; this number
was calculated by the laboratory of Nancy LRGP for diabatic distillation.
4 Conclusion
In this work, we conducted experiments of flowing liquid on our micro-structured

plate, then on a single channel, and using an infrared camera, we filmed this flow to
compare the experimental curves with digital ones.
In the second part, we have developed a computer code using Matlab software,
which allowed us to describe the temperature profile on the surface of the flow based
on the semi-analytical method of thermal quadrupoles.
The code developed will allow us to address several problems, such as the control
problem, which is essentially to determine the thermal conditions necessary to
achieve a desired temperature profile. That is to say, the flux or seek to impose the
temperature from the temperature profile at the surface.
Acknowledgements. This work is partially supported by l’Ecole des mines d’Albi-

Carmaux and the Research Center of Albi in Process Engineering, Divided Solids, Energy
and Environment. The authors also gratefully acknowledge the helpful comments and sug-
gestions of the reviewers, which have improved the presentation.
References
Zhang, F., Wub, Y.T., Gengb, J., Zhangb, Z.B.: An investigation of falling liquid films on a
vertical heated/cooled plate. International Journal of Multiphase Flow 34, 13–28 (2008),
doi:10.1016/j.ijmultiphaseflow.2007.08.003
Mhiri, N.: Etude d’un micro-absorbeur Gaz/Liquide à film tombant. Récents progrès en
génie des procédés, Dissertation, Institut national polytechnique de Lorraine (2006)
Fudym, O., Ladevie, B., Batsale, J.C.: A seminumerical approach for heat diffusion in hete-
rogeneous media: one extension of analytical quadrupole method. Numerical Heat
Transfer, Part B 42, 325–348 (2002), doi:10.1080/10407790190053978
Bu, X., Ma, W., Huang, Y.: Numerical study of heat and mass transfer of ammonia-water in
falling film evaporator. Heat Mass Transfer 48, 725–734 (2011), doi:10.1007/s00231-
011-0923-4
Perry, I., Jannot, Y., Maillet, D., Fiers, B.: Effect of velocity distribution on external wall
temperature field for a flat microchannel. Experimental Heat Transfer 23, 27–43 (2010),
doi:10.1080/08916150903402724
Baldassari, C., Marengo, M.: Flow boiling in micro-channels and micro-gravity. Progress in
Energy and Combustion Science 39, 1–36 (2013), doi:10.1016/j.pecs.2012.10.001
Hung, T.C., Huang, Y.X., Yan, W.M.: Thermal performance analysis of porous-
microchannel heat sink with different configuration designs. International Journal of Heat
and Mass Transfer 66, 235–243 (2013), doi:10.1016/j.ijheatmasstransfer.2013.07.019
Wiesegger, L.E., Knauss, R.P., Guntsching, G.E., Brandner, J.J.: Vapor-liquid phase sepa-
ration in micro-/ministructured devices. Chemical Engineering Science 93, 32–46
(2013), doi:10.1016/j.ces.2013.01.030
Conti, A., Lorenzini, G., Jeluria, Y.: Transient conjugate heat transfer in straight micro-
channels. International Journal of Heat and Mass Transfer 55, 7532–7543 (2012),
doi:10.1016/j.ijheatmasstransfer.2012.07.046
316 S. Mejri et al.
Verma, N., Mewes, D., Luke, A.: Lattice Boltzman study of velocity, temperature, and con-
centration in micro-reactors. International Journal of Heat and Mass Transfer 53,
3175–3185 (2010), doi:10.1016/j.ijheatmasstransfer.2010.03.009
Foerster, M., Lam, K.F., Sorensen, E., Gavriilidis, A.: In situ monitoring of microfluidic distil-
lation. Chemical Engineering Journal 227, 13–21 (2013), doi:10.1016/j.cej.2012.11.125
Marenzana, G., Perry, I., Maillet, D.: Mini and micro-channels: influence of axial conduc-
tion in the walls. International Journal of Heat and Mass Transfer 47, 3993–4004 (2004),
Buonomo, B., Manca, O.: Transient natural convection in a vertical micro-channel heated at
uniform heat flux. International Journal of Thermal Sciences 56, 35–47 (2012),
doi:10.1016/j.ijthermalsci.2012.01.013
Aiello, G., Ciafolo, M.: Natural convection cooling of a hot vertical wall wet by a falling
liquid film. International Journal of Heat and Mass Transfer 52, 5954–5961 (2009),
Interfacial Stresses in FRP-Plated RC Beams:
Effect of Adherend Shear Deformations
Abderezak Rabahi1, Belkacem Adim1, Selma Chargui2,

and Tahar Hassaine Daouadji1
1
Département de Génie Civil, Université Ibn Khaldoun, BP 78 Zaaroura, 14000 Tiaret, Algérie
master1513@hotmail.fr,
bem13@live.fr,
daouadjitah@yahoo.fr
2
Département de Génie Civil, Université Hassiba Benbouali, Chlef, Algérie
chersel826@gmail.com
Abstract. A recently popular method for retrofitting reinforced concrete (RC)

beams is to bond fibre reinforced polymer (FRP) plates to their tensile faces. An
important failure mode of such plated beams is the debonding of the FRP plates
from the concrete due to high level of stress concentration in the adhesive at the
ends of the FRP plate. This paper presents an improved solution for interfacial
stresses in a concrete beam bonded with the FRP plate by including the effect of
the adherend shear deformations. The analysis is based on the deformation compa-
tibility approach where both the shear and normal stresses are assumed to be inva-
riant across the adhesive layer thickness. In the present theoretical analysis, the
adherend shear deformations are taken into account by assuming a parabolic shear
stress through the thickness of both the concrete beam and the bonded plate. Nu-
merical results from the present analysis are presented both to demonstrate the ad-
vantages of the present solution over existing ones and to illustrate the main
characteristics of interfacial stress distributions.
Keywords: FRP composites, Interfacial stresses, Concrete Strengthening.
1 Introduction
The structural plating of reinforced concrete (RC) beams with fibre reinforced po-
lymer (FRP) laminates represents a new technology in the civil engineering field.
In fact, until today, the most common method in structural strengthening has been
the hermitian technique using the application of steel plates.
The new method has many advantages:
• High strength to weight ratio;
• Lightness;
• Corrosion resistance.

318 A. Rabahi et al.
The transferring of stresses from concrete to the FRP reinforcement is central to

the reinforcement effect of FRP-strengthened concrete structures. This is because
the stresses are susceptible to cause the undesirable premature and brittle failure.
In strengthening reinforced concrete beams with FRP strips, different failure
modes have been observed.
The importance of including the shear-lag effect of the adherends was shown
by several authors such as Adams and Wake, Jones and Callinan and Tsai et al. in
adhesive lap joints. Tounsi has extended this theory to study concrete beams
strengthened by FRP plates. The basic assumption in these two studies is a linear
distribution of shear stress across the thickness of the adherends. However, it is
well known that, in beam theory, this distribution is parabolic through the depth of
beam. The objective of the present investigation is to improve the method devel-
oped by Tounsi by assuming a parabolic shear stress across the depth of both FRP
plate and RC beams. In view of this, it is desirable that a solution methodology be
developed where the effect of adherend shear deformations can be included in a
better manner so that the accuracy of Tounsi’s solution can be properly assessed.
With this in mind, the objectives of this paper are first to present an improvement
to Tounsi’s solution to obtain a new closed-form solution which accounts for the
parabolic adherend shear deformation effect in both the beam and bonded plate
and second to compare quantitatively its solution against the new one developed in
this paper by numerical illustrations. Numerical examples and a parametric study
are presented to illustrate the governing parameters that control the stress concen-
trations at the edge of the FRP strip. Finally, the adopted improved model de-
scribes better the actual response of the FRP- RC hybrid beams and permits the
evaluation of the adhesive stresses, the knowledge of which is very important in
the design of such structures. It is believed that the present results will be of inter-
est to civil and structural engineers and researchers.
2 Method of Solution
2.1 Basic Assumptions

The following assumptions were made in the analytical study:
1. All materials considered are linear elastic.
2. The beam is simply supported and shallow, i.e., plane sections remain plane
in bending.
3. No slip is allowed at the interface of the bond (i.e., there is a perfect bond at
the adhesive steel or FRP plate interface).
4. Bending deformations of the adhesive are neglected.
5. Stresses in the adhesive layer do not change with the thickness.
6. The shear stress analysis assumes that the curvatures in the beam and plate
are equal (since this allows the shear stress and peel stress equations to be un-
coupled). However, this assumption is not made in the peel stress solution. When
Interfacial Stresses in FRP-Plaated RC Beams: Effect of Adherend Shear Deformations 3119
the beam is loaded, vertical separation occurs between RC beam and FRP platte.
This separation creates ann interfacial normal stress in the adhesive layer. We noote
that this assumption is ussed in several works, e.g., Tounsi et al., Smith and Tenng
and Stratford and Cadei.
7. Aparabolic shear sttress distribution through the depth of both the concreete
beam and the bonded platte is assumed.
8. The section propertiies of the RC beam were based on the uncracked section,
excluding the conventionaal steel reinforcement. It is known that for uncracked secc-
tion, the concrete can susstain tension. However, for cracked section, the concreete
cannot sustain tension andd this is why the effect of steel reinforcement in concreete
is not neglected.
Fig. 1 Simply supported beaam strengthened with bonded FRP plate
Fig. 2 Forces in infinitesimal element of a soffit-plated beam
2.2 Shear Stress Diistribution along the FRP–Concrete Interfacee

A differential section, dxx, can be cutout from the FRP- strengthened RC beam m
(Fig. 2), as shown in Fig. 3. The strains in the RC beam near the adhesive interfacce
and the external FRP reinforcement can be expressed, respectively as
du1 (x) y1
d N t dτ a (1)
ε1 (x) = = M1 (x) + 1 + 1
dx E1I1 E1 A1 4G1 dx
du2 (x) − y2 N 5t dτ a
ε 2 (x) = = M2 (x) + 2 − 2 (2)
dx E2 I 2 E2 A2 12G2 dx
Where N(x) is the axial force in each adherend and A is the cross- sectional area.
The shear stress in the adhesive can be expressed as follows:
τ a = τ ( x ) = K s [u 2 ( x ) − u1 ( x )] (3)
Moment equilibrium of the differential segment of the plated beam in Fig. 3

gives
MT (x) = M1(x) +M2(x) + N(x)(y1 + y2 +ta ) (4)
Where MT(x) is the total applied moment. The bending moment in each adherend,
expressed as a function of the total applied moment and the interfacial shear stress,
is given as
R ⎡ ⎤
x
M1 (x) = ⎢MT (x) − b2 ∫τ (x)( y1 + y2 + ta )dx⎥ (5)
R +1 ⎣ 0 ⎦
1 ⎡ ⎤
x
(6)
M ( x) =
2 ⎢
R +1 ⎣
T 2 ∫
M ( x ) − b τ ( x )( y + y + t )dx
0
1 2 a ⎥
⎦
The first derivative of the bending moment in each adherend gives:

dM1(x) R
= [VT (x) − b2τ (x)(y1 + y2 + ta )] (7)
dx R +1
dM1 (x) R
= [VT (x) − b2τ (x)(y1 + y2 + ta )] (8)
dx R +1
Substituting eqs. (1) and (2) into Eq. (3) and differentiating the resulting equa-
tion yields
d 2τ (x) ⎛ − y dM2 (x) 1 dN2 (x) y1 dM1(x) 1 dN1 (x) ⎞ ⎛ t 5t ⎞ d 2τ (x) (9)
2
= Ks ⎜⎜ 2 + − − ⎟⎟ − Ks ⎜⎜ 2 + 1 ⎟⎟
dx ⎝ E2 I2 dx E2 A2 dx E1I1 dx E1 A1 dx ⎠ ⎝ 4G2 12G1 ⎠ dx²
d 2τ (x) ⎛ ( y + y )( y + y + t ) 1 1 ⎞ ⎛ y +y ⎞ (10)
− K1b2 ⎜⎜ 1 2 1 2 a + + ⎟⎟τ ( x) + K1⎜⎜ 1 2 ⎟⎟VT ( x) = 0
dx2 ⎝ E I
1 1 + E I
2 2 E A
1 1 E2 2⎠
A ⎝ E1I1 + E1I1 ⎠
Where
1 (11)
K1 =
⎛ ta t 5t ⎞
⎜⎜ + 1 + 2 ⎟⎟
⎝ a
G 4G 1 12G2 ⎠
General solution to Eq. (10) is given by

τ ( x ) = B1 cosh( λ x ) + B 2 sinh( λ x ) + m1VT ( x ) (12)
Interfacial Stresses in FRP-Plated RC Beams: Effect of Adherend Shear Deformations 321
⎛ ( y1 + y2 )(y1 + y2 + ta ) 1 1 ⎞
λ2 = K1b2 ⎜⎜ + + ⎟ (13)
⎝ E1I1 + E2 I 2 E1 A1 E2 A2 ⎟⎠
K1 ⎛ y1 + y2 ⎞
m1 = ⎜ ⎟ (14)
λ2 ⎜⎝ E1 I1 + E2 I 2 ⎟⎠
The expression for the interfacial shear stress at any point
− λx
⎡ m2 a ⎤ qe ⎛L ⎞ (15)
τ ( x) = ⎢ ( L − a) − m1 ⎥ + m1q⎜ − a − x ⎟ 0 ≤ x ≤ LP
⎣ 2 ⎦ λ ⎝2 ⎠
Where q is the uniformly distributed load and x; a; L and Lp are defined in Fig. 1.
K1 y1
m2 = (16)
E1I1
2.3 Interfacial Normal Stress Distribution along the FRP–

Concrete Interface
The interfacial normal stress in the adhesive can be expressed as follows:
σn (x) = KnΔw(x) = Kn [w2 (x) − w1(x)] (17)
Where Kn is the normal stiffness of the adhesive per unit length and can be written
as
σ n ( x) σ n ( x) ⎛ 1 ⎞ E
Kn = = ⎜ ⎟= a (18)
Δ w( x ) Δw( x) / t a ⎜⎝ t a ⎟⎠ t a
W1(x) and W2(x) are the normal displacements of adherends 1 and 2, respectively.
Differentiating Eq. (17) twice results in
d 2σ n ( x ) ⎡ d 2 w 2 ( x ) d 2 w1 ( x ) ⎤ (19)
= n⎢ −
dx 2 ⎥⎦
K
dx 2 ⎣ dx
2
Adherend 1:
d 4 w1 ( x) 1 y dτ ( x) q
4
= b2σ ( x) + 1 b2 +
dx E1I1 E1 I1 dx E1 I1 (20)
Adherend 2:
d 4 w2 ( x) 1 y dτ ( x)
4
=− b2σ ( x) + 2 b2 (21)
dx E2 I 2 E2 I2 dx
Substitution of Eqs. (20) and (21) into the fourth derivation of the interfacial
normal stress obtainable from Eq. (19) gives the following governing differential
equation for the interfacial normal stress:
d 4σ n ( x) Eab2 ⎛ 1 1 ⎞ Eb ⎛ y y ⎞ dτ ( x) qEa
4
+ ⎜⎜ + ⎟⎟σ n ( x) + a 2 ⎜⎜ 1 − 2 ⎟⎟ + =0 (22)
dx ta ⎝ E1I1 E2 I 2 ⎠ ta ⎝ E1I1 E2 I 2 ⎠ dx ta E1I1
General solution to Eq. (22) is given by

dτ ( x)
σ n ( x) = e−βx [C1 cos(βx) + C2 sin(βx)] + eβx [C3 cos(βx) + C4 sin(βx)] − n1 − n2q (23)
dx
For large values of x it is assumed that the normal stress approaches zero and, as a
result, C3 = C4 = 0. The general solution therefore becomes
dτ ( x)
σ n ( x) = e−βx [C1 cos(βx) + C2 sin(βx)] − n1 − n2 q (24)
dx
Where:
E a b2 ⎛ 1 1 ⎞
β =4 ⎜⎜ + ⎟ (25)
4t a ⎝ E1 I1 E 2 I 2 ⎟⎠
⎛ y E I − y2 E1I1 ⎞
n1 = ⎜⎜ 1 2 2 ⎟⎟ (26)
⎝ E1I1 + E2 I 2 ⎠
E2 I 2
n2 = (27)
b2 ( E1 I1 + E 2 I 2 )
As is described by Tounsi 2006, the constants C1 and C2 in Eq.(24) are deter-
mined using the appropriate boundary conditions and they are written as follows:
Ea
C1 = [VT (0) + βMT (0)] − n3 3 τ (0) +
2β ta E1I1
3
2β (28)
n1 ⎛ d 4τ (0) d 3τ (0) ⎞
⎜
3 ⎜
+β ⎟
2β ⎝ dx 4
dx3 ⎟⎠
Ea n1 d 3τ (0)
C2 = M ( 0 ) − (29)
2β 2 t a E1 I1 2 β 2 dx 3
T
Where:
Eat2 ⎛ y1 y2 ⎞
n3 = ⎜⎜ − ⎟⎟ (30)
ta ⎝ E1 I 1 E 2 I 2 ⎠
The effect of the adherend shear deformations using the as sumption of parabolic
shear stress distribution across the thickness of the adherends on the maximum
shear and normal stress is examined by comparing the results obtained with the
present theory and those determined by two other theories [Smith S.T 2001 a,b,c].
This difference is due to the assumption used in the present theory which is in
agreement with the beam theory (parabolic distribution of shear stresses through
the depth of the beam). Hence, it is apparent that the adherend shear deformation
reduces the interfacial stresses concentration and thus renders the adhesive shear
distribution more uniform. The interfacial normal stress is seen to change sign at a
short distance away from the plate end.
Fig. 3 Comparison of interfacial shear/normal stress for CFRP-plated RC beam.
4 Parametric Study
4.1 Effect of Plate Thickness

The thickness of the FRP plate is an important design variable in practice. Fig. 4
shows the effect of the thickness of the FRP plate on the interfacial stresses. Here,
the values of the thickness, 2, 4, 6, 8, 10 and 12mm, are considered. It is shown
that the level and concentration of interfacial stress are influenced considerably by
the thickness of the FRP plate. The interfacial stresses increase as the thickness of
FRP plate increases.
Fig. 4 Effect of plate thickness on interfacial stresses in CFRP-strengthened beam
4.2 Effect of Plate Material on Interfacial Stresses in

Strengthened Beam
Fig. 5 gives interfacial normal and shear stresses for the RC beam bonded with a
CFRP plate and GFRP plate, respectively, which demonstrates the effect of plate
material properties on interfacial stresses. The length of the plate is Lp = 2400mm,
and the thickness of the plate and the adhesive layer are both 4mm. The results
show that, as the plate material becomes softer (from CFRP to GFRP), the interfa-
cial stresses become smaller, as expected. This is because, under the same load,
the tensile force developed in the plate is smaller, which leads to reduced interfa-
cial stresses. The position of the peak interfacial shear stress moves closer to the
free edge as the plate becomes less stiff.
Fig. 5 Effect of plate material on interfacial stresses in strengthened beam

4.3 Effect of Adherend Thickness

Fig. 6 shows the effect of the thickness of adherend on the interfacial stresses.
Here, the values of the thickness, 2, 4, 6, and 8 mm, are considered. It is shown
that the level and concentration of interfacial stress are influenced considerably by
the thickness of the FRP plate. The interfacial stresses reduce as the thickness of
adherend increases.
Fig. 6 Effect of Adherend thickness on interfacial stresses in strengthened beam
5 Conclusion
New theoretical interfacial stress analysis has been presented for simply supported
RC beams bonded with a thin composite or steel plate and subjected to a uniform-
ly distributed bending load. The salient features of the new analysis include the
consideration of the adherend shear deformations by assuming a parabolic shear
stress through the thickness of both the concrete beam and bonded plate. The solu-
tion methodology is general in nature and may be applicable to the analysis of
other types of composite structures. By comparing with experimental results, this
new solution provides satisfactory predictions to the interfacial stress in the plated
beams. The classical solutions which neglect the adherend shear deformations
over-estimate the non-uniformity of the adhesive stresses’ distributions and max-
imum interfacial stresses. In the final part of this paper, extensive parametric
studies were undertaken by using the new solution for strengthened beams with
various ratios of design parameters. Observations were made based on the numeri-
cal results concerning their possible implications to practical designs.
References
[Smith S.T 2001a] Smith, S.T., Teng, J.G.: FRP – strengthened RC beams. I: Review of
debonding strength models. Engineering Structures 24(4), 385–395 (2002)
[Smith S.T 2001b] Smith, S.T., Teng, J.G.: FRP – strengthened RC beams. II: Assessment
of debonding strength models. Engineering Structures 24(4), 397–417 (2002)
[Smith S.T 2001c] Smith, S.T., Teng, J.G.: Interfacial stresses in plated RC beams. Engi-
neering Structures 23(7), 857–871 (2001)
[Yu 1989] Yu, T.Q., Miao, X.S., Xiong, J.M., Jiang, H., Lee, H.: An orthotropic dam-age
model for concrete at different temperatures. Engng. Fract. Mech. 32, 775–786 (1989)
[Tounsi 2006] Tounsi, A.: Improved theoretical solution for interfacial stresses in con-crete
beams strengthened with FRP plate. Int. J. Solids Struct. 43, 4154–4174 (2006)
A Sensitivity Analysis of Multi-objective
Cooperative Planning Optimization Using
NSGA-II
Wafa Ben Yahia*, Omar Ayadi, and Faouzi Masmoudi
Mechanics, Modelling and Production Research Laboratory,

National Engineering School of Sfax (Route de Sokra B.P.1173 - 3038,
Sfax, Tunisia ENIS), Sfax University, Tunisia
wafa.benyahia@hotmail.fr, omar.ayadi@yahoo.fr,
faouzi.masmoudi@enis.rnu.tn
Abstract. The Non-dominated Sorting Algorithm II (NSGA-II) is one of the most

popular genetic algorithms (GA). It is characterized with a high optimization qual-
ity that is demonstrated for several multi-objective problems in various disciplines.
During the optimization, several genetic parameters are involved and influence the
solution quality. The purpose of this paper is to investigate the influence of the
NSGA-II parameters on the optimization process, while solving a multi-objective
planning model. Two cases, having opposite demand topology, are studied. Results
show a considerable impact of NSGA-II parameters, especially the population size
and the mutation operators, on the algorithm behaviour and the optimization
process. This investigation offers to the partners several optimal production plans
with different parameters combinations, and allows them to select the most
influential parameter that provide several good solutions.
Keywords: Multi-objective optimization, cooperative planning, NSGA-II, NSGA-

II parameters.
1 Introduction
The genetic algorithms (GA) have proven their robustness and powerful search
mechanism to solve a variety of single as well as multi-objective problems. It
features by the large search space and the multiple alternative trade-offs generated in
a single run. NSGA-II is one of the most popular multi-objectives GAs that shows
high performance and was successfully applied in various disciplines ((Murugan et
al. 2009), (Kanagarajan et al. 2007), (Kannan et al. 2009), (Agrawal et al. 2007),
(Bekele et al. 2007), (Atiquzzaman et al. 2006), (Hnaien et al. 2010), (Huang et al.
*

328 W. Ben Yahia, O. Ayadi, and F. Masmoudi
2010)). Several GA parameters, like the population size and the mutation and
crossover operators, are involved in the optimization process and interact in a
difficult way. Many studies tried to give some guidelines for the choice of the GA
parameters, like the population size or the probabilities operators ((Harik et al.
1999), (Goldberg et al. 1993), (Deb & Agrawal, 1999)...). However, most of these
works made general observations about the parameters setting. According to Hart &
Belew (1991) the parameters interaction mostly depends on the function being
optimized. The problem arises especially in the choice of the adequate population
size. Many researchers scruple about using small population sizes that may risk the
possibility of finding good solutions, or using large population sizes which require
more processing time, increase computational costs and may slow the performance.
That is why in most of works, repeated simulations are required to select the suitable
parameters values. Deb & Agrawal, (1999) investigated the effect of GA operators
for solving simple and complex problems. The considered problems are
Himmeblau's function, Rastrigin's function and deceptive function. They observed
that mutation operator plays a significant role as well the crossover operator, in the
case of simple problems. But, each operator has its working zone. In the case of
complex problem, the most significant operator is the crossover operator, but this
assertion necessitates the choice of the adequate population size. Pongcharoen et al.
(2002) developed a factorial experiment to identify the most efficient genetic
parameters (the probabilities of crossover and mutation, the population size and the
generations’ number) to solve a scheduling problem that aims to minimize a total
cost function. According to their study, the only significant parameters are the
population size, the generation number and the mutation probability. Tran (2005)
developed parameterless NSGA-II, to make NSGA-II easier. The developed
algorithm performs quite well than the original NSGA-II, while solving a
benchmark problem. Nevertheless, it is slower and has worse distribution of the
solutions. According to Zeng et al. (2010), the performance of multi-objective
evolutionary algorithms is limited by the suitability of their corresponding parameter
settings. Thus, they develop a self-adaptive mechanism adjustment of the
distribution index of the simulated binary crossover (SBX) operator to improve
NSGA-II performance.
Most of previous researches, consider benchmark functions in order to study the
GA parameters. This paper focuses on cooperative planning problem, and considers
the seven genetic parameters involved in the optimization process of NSGA-II.
In Ben Yahia et al. (2013), the NSGA-II is adapted to solve the proposed multi-
objective planning model. The model is a bi-objective one, which aims to minimize
the total production cost and the average of inventory level of a cooperative supply
chain (SC). While the proposed model is applied to experimental examples in order
to generate corresponding optimal Pareto solutions, optimization process failed to
generate Pareto of more than one solution and the sensitivity of the results to
NSGA-II parameters variations is not addressed. Our interest here is to investigate
and to analyze this sensitivity during the optimization process. The idea is to
generate optimal production plans while varying the NSGA-II parameters and to
explore their effects.
A Sensitivity Analysis of Multi-objective Cooperative Planning Optimization 329
The paper is organized as follows. A general description of the NSGA-II

algorithm and its parameters is presented in section 2. The cooperative scheme is
described and formulated in section 3, followed by the tests description in
section 4. The effect of the parameters variation is explored in section 5. Finally,
summary and potential research directions are pointed out in section 6.
2 The Non-dominated Sorting Algorithm II (NSGA-II)
The NSGA-II algorithm is detailed in (Deb & Agrawal, 2002). It is characterised

by the use of the elitism and non-domination concepts. A solution dominates
another if; first, it is not worse than the other in all the objective functions,
second at least it is strictly better than the other in one objective function.
Similarly, to any genetic algorithm, firstly, a random parent population is created.
After that, it is sorted to provide different fronts composed of feasible solutions
having the same rank. In fact, individuals are ranked based on the concept of
domination. In addition, the parameter crowding distance is calculated for each
individual to estimate the density of solutions surrounding a particular individual
in the population. Then, the solution located in a lesser crowded region is selected.
Tournament selection is used to select individuals for the next generation. In
every generation, “N” new individuals (offspring) are generated from “N” parents
using the Simulated Binary Crossover (SBX) and Polynomial mutation.
Mainly, seven parameters are involved in the optimization process of the NSGA-
II. First, the generation number which is the maximum number of optimization
iterations. The maximum number used here is equal to 1500 generations. Then, two
parameters influencing the treated solutions, first the population size, which is the
number of the treated solutions during the optimization, second the random seed
influencing the initial population. The population size used for experiments varies
between 25 and 400 and the random seed varies between 0 and 1. In addition, two
crossover operators are involved in the optimization process. First, the distribution
index for crossover “ηc” which relates the parent to their offspring by non-linear
relationship. This parameter varies between 0.5 and 100. Second, the crossover
probability “Pc” varies between 0.5 and 1. As well, two mutation operators are
considered. First, the mutation distribution index “ηm” having also a non-linear
relationship between parent and offspring, and it varies between 0.5 and 500.
Finally, the mutation probability “P m” which varies between 0 and 1/n; where n is
the number of decision variables.
3 Mathematical Formulation
The considered model is developed in Ben Yahia et al. (2013). It is a biobjective

model, which aims to minimize simultaneously the total production cost and the
average of inventory level, while considering a cooperative manufacturing supply
chain scheme. The partners cooperate together in order to generate a global
optimal production plan for a whole manufacturing supply chain, within a fixed
time horizon with a finite capacity of personnel and machines.
The multi-objective optimization problem must satisfy the following assump-
tions:
Several resources, with limited availabilities, can process several items. Raw
materials are always available.
The sequence of operations required to produce an item is fixed, and any
alternative routing is prohibited.
Inventories at the initial period are void.
Items can be only produced if all their predecessor components are available.
Periodic external demand of each item is known.
Overtime is allowed to extend fixed capacity availabilities. Backlogging is not
allowed.
Inventory is calculated at the end of a time period.
External demand has to be fully met in time and quantity.
Setup time is neglected.

The mathematical formulation considers the following notations.
Indexes
t planning period, t= l,..., T.
j operation, j = 1,. . . , J.
r resource, r = 1,. . . , R.
Set Indexes
T set of planning periods
J set of operations
R set of resources
Sj set of direct successors of operation j
Parameters
cvj unit cost of operation j
cfi setup cost of operation j
cor unit cost of overtime (capacity expansion) for resource r
Dj,t (external) demand for operation j in period t
Cr,t Capacity of resource r in period t
Lj,t Large constant
ar,j Unit requirement of resource r by operation j
rj,k Unit requirement of operation j by successor operation k
Decision variables
C total production cost
Imoy average of inventory level for all operations
xj,t output level of operation j in period t (lot size)
ij,t inventory level of operation j at the end of period t

yj,t setup variable of operation j in period t
(yj,t =1 if product j is set up in period t; yj,t =0 otherwise )
or,t overtime of resource r in period t
Formulation
The formulation of the model (the signals disappeared):
M in ( C , Im oy ) (1)
T T
St C = ∑ ∑ [( cv . x
t =1 j∈ J
j j, t ) + ( cf j. y j , t )] + ∑ ∑ co r .o r , t
t =1 r ∈ R
(2)
1 T
Im oy = ∑ ∑ ij , t
T t =1 j ∈ J
(3)
i j , t −1 + x j , t = D j , t + r j , k x k , t + i j , t ∀j ∈ J , ∀t (4)
∑ j a r , j. x j , t ≤ C r , t + o r , t ∀j ∈ J , ∀t, ∀r ∈ R (5)
x j , t ≤ L j , t. y j , t ∀j ∈ J , ∀t (6)
x j , t ≥ 0, ij , t ≥ 0 ∀j ∈ J , ∀t (7)
or , t ≥ 0 ∀t, ∀r ∈ R (8)
y j , t ∈ {0,1} ∀j ∈ J , ∀t (9)
The first objective function (2) consists of the minimization of operations, setup
and overtime costs. The second objective (3) aims to minimize the average level of
inventory with respect to the number of planning periods. The constraint (4)
represents the flow balance between output, inventory and consumption by external
demand or successor operations. The constraints (5) ensure the capacity restrictions
in using the resources to produce the different items. Lot-sizing relationships and the
choice of the items to be produced at each time-period and at each plant location are
expressed in (6). The constraints (7), (8) and (9) specify the domains of the different
variables.
4 Tests Description
The purpose of this study is to investigate the influence of the NSGA-II parameters
on the previously proposed model. Two cases will be studied, where only the
demand typology changes. In each test, customer demands are supposed to be
deterministic and e to be fulfilled during the planning horizon while facing finite
resources capacities. The first case is labelled 2PU.2P.ID and is represented by a
supply chain of two production units (2PU) planning its production activities over
two periods (2P) to satisfy the external increasing demand (ID). As shown in table 1,
the demand of the second period exceeds the main available capacity. The second
case (2PU.2P.DD) has the same configuration, but the external demand has a
decreasing trend (DD). This time, the demand of the first period exceeds the main
available capacity. These two tests have the same number of decision variables and
constraints.
The search strategies of the both studied cases are completely different.
First, the 2PU.2P.ID case has a larger solutions region and the conflict between
considered objectives occurs especially in this case. In fact, for the increasing
demand, in order minimize the total production cost the use of the overtime must
be avoided. But like that, the production must be made in advance, which will
increase the average of inventory level. However, in the 2PU.2P.DD case, the
demand has a decreasing pattern, which does not exceed the main available
capacity at the end of the planning period. So there is no need to produce or to
storage. Therefore, during the optimization the algorithm has to find a balance
between the use of the overtime and the inventory level. Ultimately, the choice
of the most suitable production plan is made by the partners according to their
preferences.
Table 1 Customer demand features in the studied cases
2PU.2P.ID 2PU.2P.DD
Demand of 1st period 2nd period 1st period 2nd period

Product 1 20 140 90 15
Product 2 15 70 40 5
Product 3 10 70 50 5
The parameters are examined separately, i.e. changing the value for the given
parameter while keeping the values of the rest of parameters at a constant level.
During the investigation of the effect of each parameter is examined separately.
In other words, when the considered parameter is varied, the other parameters
keep constant values. The adopted constant values of each parameter are shown in
table 2.
Table 2 Genetic parameters
NSGA-II (parameters) Parameter values

N, Population size 150
G, Generation number 2000
Pc, crossover probability 0.99
Pm, mutation probability 1/n (number of variables)
ηc, Crossover Index 50
ηm, Mutation Index 100
r, random seed 0.123
5 Result and Discussion
5.1 Generation Number Influence

Figure 1 shows the evolution of the optimization process while increasing the
generation number. Both studied cases need more than 75 generations to find
feasible solutions. One can note that for the 2PU.2P.ID case, the objective functions
are optimized simultaneously and the algorithm is giving a Pareto of two optimal
solutions only during the first 250 generations. However, for 2PU.2P.DD case, the
objective functions are optimized simultaneously until generation 550. In addition,
the Pareto of solutions contains up to five solutions. Besides, for both studied cases,
the algorithm converges to a Pareto of unique solution. At convergence, the Pareto
Generation number
generation number
1000 800 600 400 200 20
Inventory level
Inventory
30
40
50
60 000
40000
Total cost
Cost total
20 000
Fig. 1 Influence of the generation number on the objective functions

takes a regular pattern in other word the found solution does not improves while
increasing the generation number. During the convergence to the optimal solution,
the total production cost is minimized by 91% in the 2PU.2P.DD case and 96% in
the 2PU.2P.ID case; while, the average of inventory level is minimized by 52% in
the 2PU.2P.DD case and 15% in the 2PU.2P.ID case. Therefore, increasing the
generation number improves the solution quality.
5.2 Other Parameters Influence

Figure 2 includes the graphics showing the influence of the NSGA-II parameters
on optimal solutions, for the studied cases. Mostly, the solutions region swept by all
the parameters is the same in the two cases. For the 2PU.2P.DD case, the solutions
region swept for all the parameters is the same except some deviation while varying
the mutation operators. Moreover, the solutions region is concentrated and many
solutions are repeated with different NSGA-II parameters combinations. However,
the solutions region of the 2PU.2P.ID case is very large, except when varying the
mutation distribution index. Besides, only when varying the population size or the
mutation probability the dispersion of the solutions is wider, otherwise, the browsed
solutions region is the same. In addition, the lowest total production cost is found
several times with different parameters combinations. While the lowest averages of
inventory level are found only with the crossover and mutation distribution indexes.
Whereas, in the case of decreasing demand, mostly a global optimal solution is
found, except with the variation of the crossover probability where a Pareto of two
optimal solutions is found. The best solution is found while the mutation distribution
index is varied. Moreover, the optimal solutions found in each studied case do not
correspond to the same parameters combination; and the objective functions are very
sensitive to any variation of genetic parameters especially in the case of the
increasing demand.
For the 2PU.2P.ID case, although the lowest total production cost is found with
small and large population size, the lowest average of inventory level is found
only with large population size. While for the 2PU.2P.DD case, the best solution
is found with large population size. Therefore, increasing the number of the
treated individuals improves the solution quality. Nevertheless, the execution
time will increase. Mostly, NSGA-II can obtain optimal solution in a reasonable
amount of time. For instance, when varying the crossover operators and fixing the
other parameters, in several times, 500 generations are enough to converge.
Moreover, for our model, increasing the amount of crossover to its maximum
value is not necessary to converge into optimal solutions. The same assertion is
true for the mutation operators.
Inventory
1 00
Inventory
00
Inventory
Inventory 50
50
25 000
25000
100
0.2 20 000 20 000
0.4 15 000 200 15 000
Cost total population size Cost total

random seed
Random
0.6
10 000 Total cost Population 300
10 000 Total cost
seed 0.8 size 5000
5000
400
1.0
a. The random seed b. The population size
20
40
Inventory
Inventory 60
100
80
100
Inventory
50 20 000
15 000
Cost total
Total cost 50 10 000
0.6 10 000 Cost total
crossover distribution index
0.8
ηc Total cost
5000
crossover probability
Pm 1.0
0
c. The crossover probability d. The crossover distribution index
Inventory
80
60
100
40 Inventory
Inventory
20
50 Inventory
25000
0.06
20 000
0.04
15 000
15 000 mutation probability
totalCost
Total cost
Pm 0 10000
10 000 0.02
Cost total 200
Total cost distribution index mutation 5000
5000 ηm 400
e. The mutation probability f. The mutation distribution index
Fig. 2 Influence of the Genetic parameters on the objective functions

6 Conclusion
This paper proposes an investigation of NSGA-II parameters impact on the

optimization process of a bi-objective cooperative planning model. The considered
model aims at minimizing the total production cost and the average of inventory
level of the SC cooperating partners. Two cases are tested. For each studied case, the
NSGA-II parameters have a different behaviour, due to the randomization of the
GA. In addition, while changing the GA parameters, a diversity of solutions is
offered and the range of the solutions’ performance varies significantly. It is noticed
that, the population size and the mutation operators have the most important and
noticeable impact on the optimization process. Moreover, using large population
sizes, with reasonable amount of crossover and mutation improve the solution
quality.
During optimization, in the increasing demand case, the lowest total production
cost is found many times with different inventory level while using only the main
capacity and a null inventory at the last period; but the lowest average of inventory
level corresponds to a very high total cost because of the use of overtime. However,
in the decreasing demand case, the lowest total production cost corresponds to the
lowest average of inventory level with the use of the necessary overtime and a small
reserve in inventory at the end of the planning period. The algorithm is giving
various optimal solutions that are useful for multi-criteria group decision- making
purposes, especially for decentralized systems.
References
Agrawal, N., Rangaiah, G., Ray, A., Gupta, S.: Design stage optimization of an industrial
low-density polyethylene tubular reactor for multiple objectives using NSGA-II and its
jumping gene adaptations. Chemical Engineering Science 62(9), 2346–2365 (2007)
Atiquzzaman, M., Liong, S., Yu, X.: Alternative decision making in water distribution
network with NSGA-II. Journal of Water Resources Planning and Management 132(2),
122–126 (2006)
Bekele, E.G., Nicklow, J.W.: Multi-objective automatic calibration of SWAT using NSGA-
II. Journal of Hydrology 341(3-4), 165–176 (2007)
Ben Yahia, W., Cheikhrouhou, N., Ayadi, O., Masmoudi, F.: A Multi-objective
Optimization for Multi-period Planning in Multi-item Cooperative Manufacturing
Supply Chain. In: Haddar, M., Romdhane, L., Louati, J., Ben Amara, A. (eds.) Design
and Modelling of Mechanical System, pp. 635–643. Springer, Heidelberg (2013)
Deb, K., Agrawal, S.: Understanding interactions among genetic algorithm parameters. In:
Foundations of Genetic Algorithms V, pp. 265–286. Morgan Kaufmann, San Mateo
(1999)
Deb, K., Agrawal, S.: A fast elitist non-dominated sorting genetic algorithm for multi-
objective optimization: NSGA-II. IEEE Transactions on Evolutionary Computation 6(2),
182–197 (2002)
Harik, G., Cantú-Paz, E.: The gambler’s ruin problem, genetic algorithms, and the sizing of
populations. Evolutionary Computation 7, 231–253 (1999)
Hart, W.E., Belew, R.K.: Optimizing an Arbitrary Function is Hard for the Genetic
Algorithm. In: Proceedings of the Fourth International Conference on Genetic
Algorithms, pp. 190–195 (1991)
Hnaien, F., Delorme, X., Dolgui, A.: Multi-objective optimization for inventory control in
two-level assembly systems under uncertainty of lead times. Computers & Operations
Research 37(11), 1835–1843 (2010)
Huang, B., Buckley, B., Kechadi, T.-M.: Multi-objective feature selection by using NSGA-
II for customer churn prediction in telecommunications. Expert Systems with
Applications 37(5), 3638–3646 (2010)
Kanagarajan, D., Karthikeyan, R., Palanikumar, K., Davim, J.P.: Optimization of electrical
discharge machining characteristics of WC/Co composites using non-dominated sorting
genetic algorithm (NSGA-II). The International Journal of Advanced Manufacturing
Technology 36(11-12), 1124–1132 (2007)
Kannan, S., Baskar, S., McCalley, J.D., Murugan, P.: Application of NSGA-II Algorithm to
Generation Expansion Planning. IEEE Transactions on Power Systems 24(1), 454–461
(2009)
Murugan, P., Kannan, S., Baskar, S.: NSGA-II algorithm for multi- objective generation
expansion planning problem. Electric Power Systems Research 79(4), 622–628 (2009)
Pongcharoen, P., Hicks, C., Braiden, P.M., Stewardson, D.J.: Determining optimum
Genetic Algorithm parameters for scheduling the manufacturing and assembly of
complex products. International Journal of Production Economics 78(3), 311–322
(2002)
Tran, K.D.: Elitist Non-Dominated Sorting GA-II (NSGA-II) as a Parameter-Less Multi-
Objective Genetic Algorithm. In: Proceedings of the IEEE SoutheastCon 2005,
pp. 359–367 (2005), doi:10.1109/SECON.2005.1423273
Zeng, F., Low, M., Decraene, J.: Self-adaptive mechanism for multi- objective evolutionary
algorithms. In: Proceedings of the International MultiConference of Engineers and
Computer Scientists, IMECS 2010, vol. I, pp. 7–12 (2010)
Transient Flow Analysis through Centrifugal
Pumps
Issa Chalghoum, Sami Elaoud, Mohsen Akrout, and Ezzeddine Hadj-Taïeb
Research Laboratory "Applied Fluid Mechanics, Process Engineering and Environment"

Engineering School of Sfax, BP'W'3038 Sfax, Tunisia
{chalghoumissa,elaoudsa}@yahoo.fr
Abstract. A theoretical study has been carried out on the transient flow through a
centrifugal pump during the starting period. The problem is governed by two
hyperbolic partial differential equations which the continuity and the motion equa-
tions. The mathematical model is solved numerically by using the method of cha-
racteristics with specified time intervals. The comparison between the obtained
numerical results and those obtained by experiment has shown a good concor-
dance. In this study, the effect of the starting time, the impeller diameter, the
number and the height of blades on the pressure increase were analyzed. The nu-
merical results have shown that the pressure increase is inversely proportional to
the starting time. However, during the starting period, these results have shown
that the number of blades has no significant effect on this pressure increase.
Keywords: centrifugal pump, transient flow, valve, starting period, method of

characteristics.
1 Introduction
Studies of dynamic characteristics of centrifugal pumps have reached new impor-

tance in recent times In order to understand transient behavior of a centrifugal
flow pump during the starting period, experimental and theoretical study has been
performed (Tsukamoto et al. 1982). These experiments and analysis have shown
that the impulsive pressure and lag in circulation formation around the impeller
vanes is the main reasons for the difference between dynamic and quasi-steady
characteristics. In his studies of the dynamic characteristics of a centrifugal pump
during starting period, Saito (1986) has examined the transient characteristics of
the pump with reference to three major factors disappear in the process when
pump start up to normal steady-state performance, namely the mass of water in the
pipeline, valve opening and starting time. Through experimental test, the dynamic
behavior of a pump volute type during startup and stopping periods has been in-

340 I. Chalghoum et al.
vestigated (Thanapandi et al 1994).The obtained results were also compared to a

mathematical model solved numerically by the method of characteristics. Dazin, et
al. (2007) proposed a new method for predicting the transient behaviour of turbo-
machinery based on the angular momentum and energy equations. The results
show that the transient behaviour of a pump impeller depends not only on the ac-
celeration rate and flow rate but also on velocity profiles and their evolution dur-
ing transient periods.
In this paper we study transient flow through a centrifugal pump. The transient
regime is created by considering a linear variation of the rotational speed. The go-
verning partial equations are solved by the methods of characteristics. In this
study, the effect of the starting time and the geometric properties of the impeller,
on the pressure and the flow rate, are analyzed.
2 Mathematical Formulations
2.1 Governing Equations

The local continuity equation and the equation of motion in a relative movement
in a reference frame rotating with the impeller passage having the same origin as
the fixed reference are the following:
dρ JJG
+ ρ divW = 0 (1)
dt
G
dW JJJJJG JJJG G G G G G G G G
ρ
dt
= grad p − p + divτ + ρ F − ρ ⎡ 2Ω ∧ W + Ω
⎣⎢
(
∧ oM + Ω ∧ Ω ∧ oM ⎤
⎦⎥
) (2)
G
Where ρ the density of the fluid is, τ is the stress tensor of viscosity, F is the
G G
external body force, W is the relative velocity and Ω is the rotational speed.
Figure 1 shows the control volume defined by two successive blades limited by
suction and discharge sides of the impeller. The principle of the study of turbo
machinery is based on the velocity diagram for determining the absolute velocity
at any point of the blade. In particular, point 2 which represent the outlet of the
G
impeller, the absolute velocity V2 is given by the following equation:
G G G
V2 = W2 + U 2 (3)
G
where U 2 the entrainment velocity at that point.
Transient Flow Analysis through Centrifugal Pumps 341
In a centrifugal pump impeller, only the radial component of the absolute

velocity which permits the passage of fluid from the impeller to the volute.
According to the velocity diagram shown in Figure 1, this component is:
V2 r = V × cos α 2 = W × sin β 2 (4)
This relationship allows the passage from the rotating frame to the absolute
frame. The moving frame is used as reference for the flow study through the im-
peller. Taking into account the assumption of radial flow through the different
sections of thicknesses Δr of the impeller and the shear forces act in the radial
direction, the equations of motion (1) and (2) are simplified [Analogy to unidirec-
tional flow (Wylie et al, 1993; Thanapandi and Prasad, 1994)]:
∂H C 2 ∂Q
+ =0 (5)
∂t gA ∂r
1 ∂Q ∂H QQ
+g +λ − r Ω2 = 0 (6)
A ∂t ∂r 2 Dh A2
where A is the impeller passage section, Q is the fluid discharge, H is the

head, λ is the Darcy Weisbach friction coefficient, C is the pressure wave speed
and Dh = 4lb 2(l + b) is the hydraulic diameter of the section A where b is the
blade height and l is the width of the control volume.
2.2 Numerical Resolution by the Method of Characteristics

The method of characteristics (Wylie et al, 1993) is applied to the PDEs (5) and
(6), and the partial differential terms associated with the flow velocity and the
pressure are the reduced to ordinary differential ones compatible with two charac-
teristics lines C + and C + (Figure.2).The ordinary differential equation of (5) and
(6) are:
⎧ gA
⎪ dH + dQ + JAdt − r Ω Adt = 0
2
C+ ⎨ C (7)
⎪ dx = Cdt
⎩
⎧ gA
⎪−
− dH + dQ + JAdt − r Ω 2 Adt = 0
C ⎨ C (8)
⎪dx = −Cdt
⎩
where J = λQ Q 2 DA2 is the head loss by unit of the impeller passage length.
Consider L = R2 − R1 the impeller total length subdivided in N equal

reaches Δx = L N . If we start with known steady-state conditions at t = 0 , then we
know Q and H at the N + 1 sections of the pipe. If we specify the time inter-
val Δt defined as the Δt = Δx C , the characteristic lines from the sections i − 1
and i + 1 intersect at P , which is the section i (figure 2). In these conditions, the
integration of equations (7) and (8) can be written in the following forms:
gAi −1
QPi − Qi −1 + ( H Pi − Hi −1 ) + Ai −1 Ji −1Δt − ri −1Ω2 Ai −1Δt = 0 (9)
C
gAi +1
QPi − Qi +1 − ( H Pi − H i +1 ) + Ai +1 J i +1Δt − ri +1Ω 2 Ai +1Δt = 0 (10)
C
Fig. 1 Forces in rotating impeller in the Fig. 2 Characteristic lines

frame of reference
The calculation of the discharge and head at any section i is obtained if we

solve Eqs. (9) and (110) for the H Pi by eliminating QPi ,thus:
1
H Pi = ⎡C1H i −1 + C 2 H i +1 + ( Qi −1 − Qi +1 ) − G ⎤⎦ (11)
C1 + C 2 ⎣
where G = Δt (A i −1 J i −1 − Ai +1 J i +1 ) + Δt (ri −1 Ai −1Ω 2 − ri +1 Ai +1Ω 2 ) , C1 = gAi −1 C and

C 2 = gAi +1 C
By substituting equation (11) in equation (10), one can obtain the discharge at
section i as follows:
Q pi = ⎡⎣Qi +1 + C 2 ( H Pi − H i +1 ) − Ai +1ΔtJ i +1 − ri +1 Ω 2 Ai +1Δ t) ⎤⎦ (12)
These equations permit to determine QPi and H Pi at any interior point of the
pump. In order to get the solution until an elevated time level, it is necessary to in-
troduce the appropriate boundary conditions. These conditions will be introduced
in the next section.
3 Applications and Results
To study the transient flow through centrifugal pumps, we consider the hydraulic
facility (TE-83), available at ENIS, represented by the simplified schematic (Fig-
ure 3). This installation has two identical centrifugal pumps permitting the study
of a single pump, two pumps in parallel and two pumps in series. The test facility
represented by figure 3 contains two digital pressure gages for measuring the pres-
sure at the suction side and the discharge side of the pump. The rotational speed of
the motors is controlled by potentiometers. A mechanical device is used to deter-
mine the engine torque.
Figure 4 illustrates the centrifugal pump impeller. It is composed of 8 blades
with a base suction diameter of 42.8 mm and a discharge diameter of 128 mm.
Fig. 3 Schematic description the hydraulic Fig. 4 pump impeller TE-83

facility (TE-83)
3.1 Boundary Conditions

In this section, we are, mainly, interested in determining the boundary conditions
through the valve (Figure 3).
Fig. 5 Boundary conditions at the discharge valve
When the valve is placed between the upstream end and the downstream end of
the discharge pipe, the resolution by the method of characteristics requires that the
main discharge pipe must be treated as if it is formed by two conduits (figure.5).
Compatibility equations that solve this problem are: equation (9) for the pipe 1
and equation (10) for the pipe 2. These equations are written for the pipes 1 and 2
respectively as follows:
QP1, N 1+1 − Q1, N 1 +

gA
C
( H P1, N 1+1 − H1, N1 ) + AJ1, N 1Δt = 0 (13)
QP 2,1 − Q2,2 −
gA
C
( H P 2,1 − H 2,2 ) + AJ 2,2 Δt = 0 (14)
These equations can be written in the following form:
QP1, N 1+1 = − K1 H P1, N 1+1 + K 2 (15)
QP 2,1 = K1 H P 2,1 + K 3 (16)
gA gA
where K 2 = Q1, N 1 + H1, N 1 − AJ1, N 1Δt , K 3 = Q2,2 − H 2,2 − AJ 2,2 Δt and
C C
K1 = gA C
Based on the theory of the flows through the orifices, we get:
QP1, N 1+1 = QP 2,1 = Cd AG H P1, N 1+1 − H P 2,1 , (17)
if the flow rate is positive, and
QP1, N 1+1 = QP 2,1 = Cd AG H P 2,1 − H P1, N 1+1 (18)
if the flow rate is negative

In the above equations, AG is the open section of the valve and Cd = Cc × Cv
is the flow coefficient, Cc is the coefficient of contraction since the actual flow is
passing through a lower section AG and Cv = 1 1 + Kv is the velocity coeffi-
cient since the actual velocity is less than theoretical velocity due the singular
pressure loss where K v is the singular pressure drop coefficient of the valve which
depends of the valve opening.
Tables 1 and 2 represent the coefficients K v and Cv respectively for DN40
gate valve (Same type of valve used in TE-83 test facility).
Table 1 Contraction coefficient
r/R 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Cc 0.611 0.612 0.616 0.622 0.633 0.644 0.662 0.687 0.722 0.781
Table 2 Coefficient K v for gate valve DN 40
Percentage of opening 100 75 50 25 0

Coefficient K v 0.15 0.26 2.1 17 ∞
By substituting equations (15) and (16) in the sum of equations (11) and (12),
we obtain the following quadratic equations:
K1
QP21, N1+1 − 2QP1, N1+1 + ( K1 + K2 ) = 0 (19)
( Cd AG )
2
K1
QP21, N1+1 + 2QP1, N1+1 − ( K1 + K2 ) = 0 (20)
( Cd AG )
2
Equation (19) is used to determine the positive flow rate while equation (20) is
used to determine the negative flow rate.
3.2 Results by the Method of Characteristics

To numerically determine the characteristics of the hydraulic facility (TE-83)
shown in figure 2, the impeller geometry is simplified. In this case, we considered
an impeller having the same number of blade as the original one (8 blades) but a
straight curve shape of the blades was adopted as illustrated in figure 6. Since the
relative flow is assumed to be in the radial direction, this simplification of the
blades shape can be considered as logical.
r
G
W2 G
β 2 V2
Ω
α2
G
U2
Fig. 6 simplified impeller Fig. 7 modelling the pump impeller
Furthermore, we consider an average blade height of 5 mm while the actual

height of the blade is variable (8 mm at the inlet of the impeller and 3.2 mm at its
outlet). In Figure 6, the angle β 2 = 90D is constant throughout the blade and par-
ticularly in the outlet point 2 of the centrifugal impeller. In this case, based on
equation (4), the radial component of absolute velocity is equal to the relative ve-
locity:
V2r = V × cosα2 = W × sin β2 = W (21)
When establishing the equations of motion, we model the impeller shown in

figure 7 by 8 parallel pipes. By simulating the start-up phase, we assume that to
reach the steady state, the engine’s rotational speed varies linearly during a start-
up time of 0.15 s.
5 2,5
pressure (m)
flow rate (l/s)

closed valve 0,25 valve open
valve half open 0,75 valve open
4 2,0
closed valve 0,25 valve open open valve
0,5 valve open 0,75 valve open
3 open valve 1,5
2 1,0
startup time 0.15 s - N = 1450 rpm
startup time 0.15 s/ N =1450 rpm
1 0,5
0 0,0
0,0 0,1 0,2 0,3 0,4 0,5 0,6 t (s) 0,7 0,0 0,1 0,2 0,3 0,4 0,5 0,6 t (s) 0,7
Fig. 8 the pressure difference versus time Fig. 9 Variation of flow rate versus time for
for different valve opening positions different valve opening positions
Figure 9 shows the evolution of the pressure difference between inlet and outlet
sides of the pump as a function of time and for the different valve opening posi-
tions. These curves are considered for a rapid starting period (≈ 0.15 s ) and a
rotational speed of 1450 rpm. This figure shows that the pressure is inversely pro-
portional to the valve opening percentage. Indeed, when the valve is fully open,
the pressure drops to 0.2 m. This pressure drop is due to the reduction of the flow
resistance caused by the friction at the suction and discharge circuits and mainly at
the valve. This figure also shows an oscillating effect at the pressure curve when
the valve is in the closed position. This phenomenon is due to the wave propaga-
tion along the pipes and its reflection at both ends of the hydraulic circuit. It can
be seen also in figure 9 the presence of peak of pressure from the mid-open posi-
tion up to complete valve opening. This peak is due to the pump start-up time and
to the mass of water which increases when the resistance in the circuit decreases
with the valve opening.
3.3 Validation of the Theoretical Results

Figure 11 shows a comparison between the pump characteristic curves of the ex-
perimental and the theoretical model for two rotational speeds 1450 rpm and 2900
rpm. This curve shows a slight shift between the experimental and theoretical
curves. In fact, the characteristic curve of a pump depends on the following pa-
rameters: the rotational speed of the engine; the suction and discharge diameter of
the impeller; the blade curve; the blade height; pressure losses within the pump (at
the impeller and the volute) and the shape of the volute. This difference is mainly
due to the approximation of shape and the averaged height of the blades. Never-
theless, the results can be considered as acceptable. The curves of Figure 11 show
that when is increased the rotational speed, the pressure range and the flow rate
range is increased too. This result is logic since the centrifugal force which per-
mits the flow through the impeller is proportional to the square root of the rota-
tional speed. Figure 12 shows the effect of the diameter of the impeller on charac-
teristic curve of the pump. It can be noted that the head range of the characteristic
curve is proportional to the diameter of the impeller. This is due to the effect of
centrifugal force which increases with the radius. For a constant impeller diameter
and by changing the blades height, the flow rate range increases while the pressure
remains unchanged as shown in Figure 13.
6 20
pressure (m)
pressure (m)
startup time 0.15 s - N = 1450 rpm startup time 0.15 s
5 16 N = 2900 rpm
4
12
3 Pump TE-83
Pump TE-83
8
2 theoretical model
theoretical m odel
4
1
0 0
Q(l/s) Q(l/s)
0,00 0,25 0,50 0,75 1,00 1,25 1,50 1,75 0,00 0,50 1,00 1,50 2,00 2,50 3,00 3,50
Fig. 10 Comparison of theoretical and experimental curves for N = 1450 rpm and N = 2900
rpm
10 6
pressure (m)
pressure(m)
s tartup time 0.15 s - N =1450 rpm s tartup time 0.15 s - N =1450 rpm
8
R = 84 mm
R = 74 mm 4
6 h = 5 mm
R = 64 mm h = 7 mm
4
2
2
0 0
0 0,4 0,8 1,2 1,6 Q (l/s ) 2 0 1 2 3 4 Q (l/s) 5
Fig. 11 Effect of the impeller diameter on Fig. 12 Influence of the blades height on the
the pump characteristics curve pump characteristics curve
3.4 Parameters Which Influence the Pressure During the

Starting Phase
Saito (1982), in his research, has shown that during the starting period of the
pump, the change of its dynamic characteristics is due to three factors, namely: the
mass of water in the pipe; the opening position of the valve and the starting time.
In fact, figure.14 shows that the peak of pressure during the starting period is
largely reduced when the starting time is increased. Besides the statement of Saito
which are well confirmed in this section and in the previous section, other parame-
ters have also an effect during the starting period. In fact, it can be seen on figure
15 that the peak of pressure is reduced when the diameter of the impeller is
reduced.
3 4
pressure (m)
pressure (m)
L2 = 1.5 m - N =1450 rpm startup time 0.15 s - N =1450 rpm
startup time 0.15 s 3

2 startup time 0.3 s R = 0.64 mm
startup time 0.6 s
2 R = 0.74 mm
R = 0.84 mm
1
1
0 0
0,0 0,1 0,2 0,3 0,4 0,5 0,6 t (s ) 0,7 0,0 0,1 0,2 0,3 0,4 0,5 0,6 t (s) 0,7
Fig. 13 Effect of starting time on the Fig. 14 Effect of the impeller diameter on
pressure evolutions the pressure curve
4 Conclusion
In this paper, the transient flow inside a centrifugal pump has been conducted. The
transient is created as a result of a linear variation of pump speed from rest to the
permanent regime. This type of flow is governed by two partial differential equa-
tions: the momentum and the continuity equations. These equations are solved
numerically by the use of the method of characteristics. The pump impeller was
modelled as conduits connected in parallel. The obtained results have shown a
good concordance with those obtained by experiment. In this study the effect of,
the starting time, the geometric proprieties of the impeller and the valve openings,
on the pressure evolution has been also analysed. These results have shown that
the pressure increase during the starting period was influenced by the starting time
and the valve openings. However, these parameters have no influence on the
steady state pump characteristic curve. These results have shown that this curve is
well influenced by the radius and the blade height of the impeller.
References
Tsukamoto, H., Ohashi, H.: Transient characteristics of a centrifugal pump during starting
period. ASME Journal of Fluids Engineering 104, 6–13 (1982)
Dazin, A., Caignaert, G., Bois, G.: Transient behavior of turbomachineries: application to
radial flow pumps startups. ASME Journal of Fluids Engineering 129, 1436–1444
(2007), doi:10.1115/1.2776963
Saito, S.: The transient characteristics of a pump during startup. Bulletin of JSME 25,
372–379 (1986)
Thanapandi, P., Prasad, R.: Centrifugal pump transient characteristics and analysis using
the method of characteristics. Int. J. Mech. Sci. 37, 77–89 (1995)
Streeter, V.L., Wylie, E.B.: Hydraulic Transients. McGraw-Hill Book Company, New York
(1967)
Vibro-Acoustic Analysis of Laminated
Double-Wall: Finite Element Formulation
and Reduced-Order Model
Walid Larbi, Jean-François Deü, and Roger Ohayon
Conservatoire National des Arts et Métiers (Cnam),

Structural Mechanics and Coupled Systems Laboratory,
2 rue Conté, 75003 Paris, France
walid.larbi@cnam.fr
Abstract. This paper presents a finite element model for sound transmission anal-
ysis through a double sandwich panels with viscoelastic core inserted in an infinite
baffle. The proposed model is derived from a multi-field variational principle in-
volving structural displacement of the panels and acoustic pressure inside the fluid
cavity. To solve the vibro-acoustic problem, the plate displacements are expanded as
a modal summation of the plate’s real eigenfunctions in vacuo. Similarly, the cavity
pressure is expanded as a summation over the modes of the cavity with rigid bound-
aries. Then, an appropriate reduced-order model with mode acceleration method
by adding quasi-static corrections is introduced. The structure is excited by a plane
wave. The radiated sound power is calculated by means of a discrete solution of
the Rayleigh Integral. Fluid loading is neglected. Various results are presented in
order to validate and illustrate the efficiency of the proposed reduced finite element
formulation.
Keywords: double-wall, viscoelastic, vibroacoustic, finite element, modal reduc-

tion.
1 Introduction
Double-wall structures are widely used in noise control due to their superiority over
single-leaf structures in providing better acoustic insulation. Typical examples in-
clude double glazed windows, fuselage of airplanes and vehicles, etc. Various the-
oretical, experimental and numerical approaches have been investigated to predict
the sound transmission through double walls [1, 2, 3].
By introducing a thin viscoelastic interlayer within the panels, a better acous-
tic insulation is obtained. In fact, sandwich structures with viscoelastic layer are
commonly used in many systems for vibration damping and noise control. In such
structures, the main energy loss mechanism is due to the transverse shear of the
viscoelastic core [4, 5].
c Springer International Publishing Switzerland 2015 349

350 W. Larbi, J.-F. Deü, and R. Ohayon
In the first part of this paper, a non-symmetric finite element formulation of

double-wall sandwich panels with viscoelastic core is derived from a variational
principle involving structural displacement and acoustic pressure in the fluid cav-
ity. Since the elasticity modulus of the viscoelastic core is complex and frequency
dependent, this formulation is complex and nonlinear in frequency. Therefore, the
direct solution of this problem can be considered only for small size models. This
has severe limitations in attaining adequate accuracy and wider frequency ranges
of interest. An original reduced order-model is then proposed to solve the problem
at a lower cost. The proposed methodology, based on a normal mode expansion,
requires the computation of the uncoupled structural and acoustic modes. The un-
coupled structural modes are the real and undamped modes of the sandwich panels
without fluid pressure loading at fluid-structure interface, whereas the uncoupled
acoustic modes are the cavity modes with rigid wall boundary conditions at the
fluid-structure interface. It is shown that the projection of the full-order coupled fi-
nite element model on the uncoupled bases, leads to a reduced order model in which
the main parameters are the classical fluid-structure and residual stiffness complex
coupling factors. Moreover, the effects of the higher modes of each subsystem is
taken into account through an appropriate so-called static correction. Despite its re-
duced size, this model is proved to be very efficient for simulations of steady-state
and frequency analyses of the coupled structural-acoustic system with viscoelastic
damping and the computational effort is significantly reduced.
In the last part, numerical examples are presented in order to validate and analyse
results computed from the proposed formulation.
2 Finite Element Formulation of the Coupled Problem

2.1 Local Equations
Consider a double-wall structure coupled to an acoustic enclosure shown in Fig. 1.
Each wall occupies a domain ΩSi , i ∈ {1, 2} such that ΩS = (ΩS1 , ΩS2 ) is a
partition of the whole structure domain. A prescribed surface force density Fd is
applied to the external boundary Γt of ΩS and a prescribed displacement ud is
applied on a part Γu of ΩS . The acoustic enclosure is filled with a compressible and
ΩF
Γt
Fd Ωs2
Ωs1
u d
Γu
Sandwich panel with
ud viscoelastic core
Fig. 1 Double sandwich wall structure coupled to an acoustic enclosure

Vibro-Acoustic Analysis of Laminated Double-Wall 351
inviscid fluid occupying the domain ΩF . The cavity walls are rigid except those in
contact with the flexible wall structures noted Σ.
The harmonic local equations of this structural-acoustic coupled problem can be
written in terms of structure displacement u and fluid pressure field p [6]
divσ(u) + ρS ω 2 u = 0 in ΩS (1)
σ(u)nS = Fd on Γt (2)
σ(u)nS = pn on Σ (3)
u=u d
on Γu (4)
2
ω
Δp + p=0 in ΩF (5)
c2F
∇p · n = ρF ω 2 u · n on Σ (6)
where ω is the angular frequency, nS and n are the external unit normal to ΩS and
ΩF ; ρS and ρF are the structure and fluid mass densities; cF is the speed of sound
in the fluid; and σ is the structure stress tensor.
2.2 Constitutive Relation for Viscoelastic Core

In order to provide better acoustic insulation, damped sandwich panels with a thin
layer of viscoelastic core are used in this study (Fig. 1). When subjected to mechanical
vibrations, the viscoelastic layer absorbs part of the vibratory energy in the form of
heat. Another part of this energy is dissipated in the constrained core due to the shear
motion.
The constitutive relation for a viscoelastic material subjected to a sinusoidal
strain is written in the following form:
σ = C∗ (ω)ε (7)
where ε denotes the strain tensor and C∗ (ω) is the complex elasticity tensor. It is
generally complex and frequency dependent (∗ denotes complex quantities). It can
be written as:
C∗ (ω) = C (ω) + iC (ω) (8)
√
where i = −1.
Furthermore, for simplicity, a linear, homogeneous, viscoelastic constitutive
equation will be used in this work. In the isotropic case, the viscoelastic material
is defined by a complex and frequency dependence shear modulus in the form:

G∗ (ω) = G (ω) + iG (ω) (9)

where G (ω) is know as shear storage modulus, as it is related to storing energy and

G (ω) is the shear loss modulus, which represents the energy dissipation effects.
With these assumptions, the stress tensor of the sandwich structure is complex and
frequency dependent.
2.3 Finite Element Equation

After variational formulation using the test-function method [6, 7] and discretiza-
tion by the finite element method, we obtain the following matrix system of the
fluid/elastic structure with viscoelastic damping coupled problem:
∗
Ku (ω) −Cup 2 Mu 0 U F
−ω = (10)
0 Kp CTup Mp P 0
where U and P are the vectors of nodal values of u and p respectively; Mu and
K∗u (ω) are the mass and stiffness matrices of the structure; Mp and Kp are the mass
and stiffness matrices of the fluid; Cup is the fluid-structure coupled matrix; F is
the applied mechanical force vector. Note that since the elasticity modulus of the
viscoelastic core of the sandwich panels is complex and frequency dependent, the
stiffness matrix K∗u (ω) is also complex and frequency dependent.
3 Reduced Order Model

In this section, we introduce a reduced-order formulation based on a normal mode
expansion with an appropriate static correction.
3.1 Eigenmodes of the Structure in Vacuo

In a first phase, the first Ns eigenmodes of the structure in vacuo are obtained from
∗
Ku (ω) − ω 2 Mu U = 0 (11)
Due to the frequency dependent of the stiffness matrix, this eigenvalue problem is
complex and nonlinear. It is assumed that vibrations of the damped structure can be
represented in terms of the real modes of the associated undamped system if appro-
priate damping terms are inserted into the uncoupled modal equations of motion.
Thus, the complex stiffness matrix is decomposed in the sum of two matrices:
K∗u (ω) = Ku0 + δK∗u (ω) (12)
where Ku0 = Re[K∗u (0)] is the real and frequency-independent stiffness matrix
calculated with a constant Young module’s of the viscoelastic core and δK∗u (ω) is
the residual stiffness matrix.
The ith real eigenmode is obtained from the following equation
2

Ku0 − ωsi Mu Φsi = 0 for i ∈ {1, · · · , Ns } (13)
where (ωsi , Φsi ) are the natural frequency and eigenvector for the ith structural mode.
3.2 Eigenmodes of the Internal Acoustic Cavity with Rigid Walls

In this second phase, the first Nf eigenmodes of the acoustic cavity with rigid
boundary conditions are obtained from the following equation

Kp − ωf2 i Mp Φf i = 0 for i ∈ {1, · · · , Nf } (14)
where (ωf i , Φf i ) are the natural frequency and eigenvector for the ith acoustic
mode.
It is important
to note that the physical acoustic modes in a rigid fixed cavity
are such that ΩF p dv = 0 which excludes the ωf i = 0, p = constant solution
which is not physical but had to be introduced in the formulation of the coupled
problem. Thus, the variational formulation of this kind of system, in order to be
regularized for zero frequency situation, i.e. valid for a static problem,
has to be
modified by adding the following constraint ρF c2F Σ u.n ds + ΩF p dv = 0 (see
[6] for details). When doing this, on one hand the static pressure is defined precisely
by
ρF c2F
p =−
s
u.n ds (15)
|ΩF | Σ
and on the other hand, the reduced order formulation will be carried only by projec-
tion on the physical acoustic modes.
3.3 Modal Expansion of the General Problem

By
introducing the matrices Φs = [Φs1 · · · ΦsNs ] of size (Ms × Ns ) and Φf =
Φf 1 · · · Φf Nf of size (Mf × Nf ) corresponding to the uncoupled bases (Ms
and Mf are the total number of degrees of freedom in the finite elements model
associated to the structure and the acoustic domains respectively), the displacement
and pressure are sought as
U = Φs qs (t) and P = Φf qf (t) (16)

T
where the vectors qs = [qs1 · · · qsNs ]T and qf = qf 1 · · · qf Nf are the modal
amplitudes of the structure displacement and the fluid pressure respectively.
Substituting these relations into Eq. (10) and pre-multiplying the first row by ΦTs
and the second one by ΦTf , we obtain
• Ns mechanical equations
Ns
Nf
2 ∗ 2
−ω qsi + γik (ω)qsi + ωsi qsi − βij qf j = Fi (17)
k=1 j=1
• Nf acoustic equations
Ns
−ω 2 qf i + ωf2 i qf i − ω 2 βij qsj = 0 (18)
j=1
where Fi (t) = ΦTsi F is the mechanical excitation of the ith mode; βij = ΦTsi Cup Φf j
∗
is the fluid structure coupling coefficient and γik (ω) = ΦTsi δK∗u (ω)Φsk the reduced
residual stiffness complex coefficient.
At each frequency step, the reduced system (Eqs. (17) and (18)) is solved by
∗
updating γik (ω). After determining the complex amplitude vectors qsi and qf i ,
the displacement and pressure fields are reconstructed using the modal expansion
(Eqs. (16)).
3.4 Static Corrections

The process of mode truncation introduces some errors in the response that can
be controlled or minimized by a modal truncation augmentation method. In this
method, the effects of the truncated modes are considered by their static effect only.
First the applied loading vector F is composed as:
L
F= αi (t)F0i (19)
i=1
where F0i is the invariant spatial portion and αj (t) is the time varying portion. For
each invariant spatial load, the static modal eigenvector Ψsi is given by:
Ψsi = K−1
u0 F0i (20)
The truncated basis containing the real and undamped structure modes is enriched
by the static modal eigenvectors such that
Φ̄s = [Φs1 · · · ΦsNs , Ψsi , · · · Ψsl ] = [Φs Ψs ] (21)
The truncated fluid basis is enriched with the static pressure Ps computed from
Eq. (15):
Φ̄f = [Φf Ps ] (22)
Thus, the displacement and pressure are sought as
U = Φs qs (t) + Ψs q0s (t) and P = Φf qf (t) + Ps qf0 (t) (23)
where the vectors q0s and qf0 are the quasi-static modal amplitudes of the structure
displacement and the fluid pressure respectively. Similar coupled differential equa-
tions than Eqs. (17) and (18) can of course be obtained after modal projection on
the enriched bases Φ̄s and Φ̄f .
4 Numerical Examples
4.1 Sound Transmission through an Elastic Double-Panel System
In this section, the validation of the proposed coupled finite elements formulation for
sound radiation is presented. The problem under consideration is shown in Fig. 2. A
normal incidence plane wave excites a double-plate system filled with air (density
ρF = 1.21 kg/m3 and speed of sound cF = 340 m/s). The plane wave has a pressure
amplitude of 1 N/m2 and is applied to plate 1 as the only external force to the system.
The plates are identical and simply supported with thicknesses of 1 mm. The density
of the plates is 2814 kg/m3 , the Youngs modulus is 71 GPa, the Loss factor is 0.01
and Poisson ratio 0.33. The surrounding fluid is the air. This example was originally
proposed by Panneton in [2].
Rigid baffle
Air Cav
ity
b=0.22 m
Plate 2
Normal plane
wave Plate 1 m
4
76
a=0.35 m 0.0
c=
Fig. 2 Double-plate system filled with air: geometric data
When the excitation is applied to the first plate, the second one vibrates and
radiates sound caused by the coupling of air and plate 1. The normal incidence sound
transmission loss is then computed using the Rayleigh’s integral method [1] which
needs the finite element solution of surface velocities of plate 2. For this purpose,
the resolution of the coupled system is done with a modal reduction approach using
the first 10 in vacuo structural modes and the first 10 acoustic modes of the fluid in
rigid cavity. The truncated bases are enriched by the static modal eigenvectors.
Fig. 3 shows the normal incidence transmission loss through a simply supported
plate (dashed line). Due to the modal behavior of the plate, dips in the transmission
loss curve are observed at its resonance frequencies (modes (1, 1), (3, 1) and (1,
3)). When a second plate is used to form an airtight cavity (continuous line), an
increase in the transmission loss is achieved except in the region of the so-called
plate-cavity-plate resonance (mode (1, 1)*). At this frequency, the two plates move
80
Air−filled double wall system
70 Simple panel
60
50
nSTL (dB)
40
30
20
10
mode (1, 3)
mode (1, 1) mode (3, 1)
0
mode (1, 1)*
1 2 3
10 10 10
Frequency (Hz)
Fig. 3 Comparison of the normal incidence sound transmission (nSTL) through an air-filled
double panel and a simple panel
out of phase with each other and the effect of the cavity on the plates is mostly
one of added stiffness. This frequency is similar to the mass-air-mass resonance of
unbounded double panels.
In addition, the variation of the nSTL of an air-filled panels and a simple panel is
in very good agreement with the published date from [2].
4.2 Sound Transmission through a Double Laminated Glazing

Window
The proposed reduced order finite-elements formulation is applied now to calculate
the transmission loss factor of a double laminated glazing window. The system con-
sists of two identical clamped laminated panels of glass separated by an air cavity
of 12 mm thickness. Each laminated glass is composed of two glass plates bonded
together by a Polyvinyl Butyral (PVB) interlayer. The thickness of outer and inner
glass ply is h1 = h3 = 3 mm and those of the PVB interlayer is h2 = 1.14 mm.
The glass ply is modeled as linear elastic material (density 2500 kg/m3 , Youngs
modulus 72 GPa, and Poisson ratio 0.22). The material properties of the PVB are
both thermal and frequency dependent. From dynamic and thermal tests, Havrillak
and Negami have found an empirical law describing this dependence. The resulting
complex frequency dependent shear modulus of the PVB is given at 20◦ C as [8]:
−β0
G∗ (ω) = G∞ + (G0 − G∞ ) 1 + (iωτ0 )1−α0 (24)
where G∞ = 0.235 GPa, G0 = 0.479 Mpa, α0 = 0.46, β0 = 0.1946, τ0 =

0.3979. The Poisson ratio of the PVB is 0.4 and density is 999 kg/m3 . Concerning
the excitation and the finite element discretization, we used the same ones as in the
previous example.
100
−12
Radiated Sound Power (dB) − ref 10
80
60
40
20
Simple glass pane

0 Laminted glass pane (3×1.14×3 mm)
0 500 1000 1500 2000
Frequency (Hz)
Fig. 4 Comparison of radiated sound power from a simple glass pane and a laminated glass
with the same mass
A comparison between a simple glass and a laminated glass with PVB inter-
layer with an equivalent surface mass is shown in Fig. 4. Calculation was limited
to 2000 Hz maximum. This comparison shows that laminated glass has a much
lower acoustic radiation compared to conventional glass at resonance frequencies
du to the effect of the viscoelastic layer. The reduction of sound radiation power
is around 10 dB in lower frequencies and around 20 dB in higher frequencies. In
fact, at low frequencies, the viscoelastic material is soft and the damping is small.
At higher frequencies, the stiffness decreases rapidly and the damping is highest.
Moreover, flexural vibrations causes shear strain in the viscoelastic core which dis-
sipates energy and reduces vibration and noise radiation. Note that the thickness of
the viscoelastic layer has a significant influence in terms of attenuation.
120
Modal reduction
Direct method
100
80
nSTL (dB)
60
40
20
0
1 2 3
10 10 10
Frequency (Hz)
Fig. 5 nSTL through an air-filled double panel: comparison between the modal reduction
approach and the direct nodal method
Fig. 5 shows a comparison between the nSTL of the coupled problem, obtained
with the proposed accelerated modal reduction approach with a truncation on the
first twenty structural modes (Ns = 20) and first twenty acoustic modes (Nf = 20)
and the direct nodal method (Eq. (10)) where the displacement and pressure vectors
are calculated for each frequency step. The structural modes are calculated from
Eq. (13) using the constant shear storage modulus G∞ . As can be seen, a very good
agreement between the two methods is proved. In this respect, it should be noted
that the resulting reduction of the model size and the computational effort using the
reduced order method are very significant compared to those of the direct approach.
5 Conclusions
In this paper, a finite element formulation for sound transmission through double
wall sandwich panels with viscoelastic core is presented. A reduced-order model,
based on a normal mode expansion, is then developed. The proposed methodology
requires the computation of the eigenmodes of the undamped structure, and the
rigid acoustic cavity. Quasi-static corrections are introduced in order to accelerate
the convergence. Despite its reduced size, this model is proved to be very efficient
for simulations of steady-state analyses of structural-acoustic coupled systems with
viscoelastic interlayers when appropriate damping terms are inserted into the modal
equations of motion.
References
1. Fahy, F.: Sound and structural vibration, 1st edn. Academic Press, New York (1985)
2. Panneton, R.: Modélisation numérique tridimensionnelle par éléments finis des milieux
poroélastiques: application au problème couplélasto-poro-acoustique. Ph.D., Université
de Sherbrooke (1996)
3. Quirt, J.D.: Sound transmission through windows I. Single and double glazing. Journal of
the Acoustical Society of America 72(3), 834–844 (1982)
4. Akrout, A., Hammami, L., Ben Tahar, M., Haddar, M.: Vibro-acoustic behaviour of lam-
inated double glazing enclosing a viscothermal fluid cavity. Applied Acoustics 70(1),
82–96 (2009)
5. Basten, T.G.H.: Noise reduction by viscothermal acousto-elastic interaction in double wall
panels. PhD-thesis, University of Twente, Enschede, The Netherlands (2001)
6. Morand, H.J.-P., Ohayon, R.: Fluid-structure interaction. John Wiley & Sons, New York
(1995)
7. Larbi, W., Deü, J.-F., Ohayon, R.: Finite element formulation of smart piezoelectric com-
posite plates coupled with acoustic fluid. Composite Structures 94(2), 501–509 (2012)
8. Havriliak, S., Negami, S.: A complex plane analysis of?-dispersions in some polymer
systems. Journal of Polymer Science Part C: Polymer Symposia 14(1), 99–117 (1966)
Time Step Size Effect on the Hydrodynamic
Structure around a Water Darrieus Rotor
Ibrahim Mabrouki, Ahmed Samet, Zied Driss, and Mohamed Salah Abid
Laboratory of Electro-Mechanic Systems (LASEM),

National School of Engineers of Sfax (ENIS), University of Sfax (US),
B.P. 1173, Road Soukra km 3.5, 3038 Sfax, TUNISIA
mabroukiibrahim@hotmail.fr, zied_driss@yahoo.fr,
MohamedSalah.Abid@enis.rnu.tn
Abstract. Turbulent flows are encountered in many hydraulic and water resources
engineering problems. Their understanding is thus a critical prerequisite for de-
signing stream and river restoration projects and a broad range of hydrodynamic
structures. For this purpose, an unsteady Reynolds averaged Navier-Stokes
(URANS) equations with a two-equations turbulence closure model is employed.
The present paper aims to numerically explore the three-dimensional unsteady
flow over a conventional Darrieus type rotor and to study the time step size effect
on the hydrodynamic structure.
Keywords: URANS, hydrodynamic, Darrieus rotor.
1 Introduction
Darrieus turbine was developed in France by Georges Darrieus in the 1920's to
generate power from wind. It consists of a set of long, rectangular airfoils con-
nected to a central rotating shaft. The airfoils may be curved to directly connect to
the shaft or be straight and held parallel to the shaft by struts, arms or discs. These
blades are there by oriented transversely to the fluid flow and parallel to the axis
of rotation. Significant testing in the 1980's and 1990's demonstrated the utility of
this turbine design. Several different models were successfully tested in the la-
boratory and various waterways. This kind of turbine considered for diffrent ap-
plications, such as tidal power, run-of-the-river hydroelectricity, and wave energy
conversion (Shiono et al. (2000), Prabhu et al. (2013) and Maitre et al. (2013)).
Hwang et al. (2009) studied the effect of number Blade on the performance of
Darrieus Water Turbine. It’s one of the important parameter of Darrieus water
turbine. These authors studied the influence of chord length of Darrieus water tur-
bine. They show the evolution of maximum power coefficient and the TSR for a
fixed number of blades. Attempts to improve the performance of H-Blade Dar-
rieus rotor by using a duct placed outside the rotor have been documented. Shimo-
kawa et al. (2010) measured the influence of the stream section with and without
nozzle in the distribution of velocity vectors. In addition, Matsushita et al. (2008)
studied the effect of draft tube on the efficiency in the case with narrow intake Sin
/D=0.8. Malipeddi et al. (2012) studied the effect of changing position of the tur-

360 I. Mabrouki et al.
bine in the duct by performing simulation with the turbine at various locations in
the duct, for a tip-speed ratio of 2. Rao et al. (2010) studied the performance of
Darrieus rotor equipped with a symmetric two-foiled channeling device (Diffuser
Annular Ring). They changed the channeling profile and, demonstrate their effect
in the performance of the turbine. A ducted Darrieus turbine system as appropriate
one for extra-low head hydropower has been proposed and developed by different
authors. The design guidelines of Darrieus turbine and the characteristics have
been clarified to obtain higher performance As the results of still continued inves-
tigation to simplify the structure for making cost effectiveness higher, it was rec-
ognized for small sized runner that there was no need to install a draft tube and the
sidewalls of runner casing section in the case of runner with inlet nozzle. A nu-
merical investigation was developed in the unsteady state using computational flu-
id dynamic (CFD) in order to characterize the hydrodynamic comportment of the
flow around the rotor. We studied also the effect of time step size on the hydrody-
namics characteristics. The used CFD code "Fluent" is based on the Unsteady
Navier-Stokes equations by a finite volume discretization model. The numerical
approach used is the sliding mesh (SM) model. In these condition, the test section
equipped with 2 parts stationary domain and moving domain in which we have
impose a rotational speed equal to ω = 25.61 rad.s-1. The turbulent flow is defined
by the Reynolds number Re=3.8 106.
2 Geometrical System
The test section under investigation in this work is composed of a duct equipped
with water Darrieus rotor. The duct have a prismatic shape with a height equal to
b=0.2 m, a width equal to s=0.526 m, and a length equal to L=2.886 m .The Dar-
rieus turbine, is composed by three straight blades and is placed inside the duct.
The blades have profiles belonging to the symmetrical NACA 0018 four digit with
a maximal thickness equal to 18% of the chord, a height equal to b=0.2 m and a
diameter equal to D=0.39 m.
The flow over Darrieus rotor is periodic and a proper time step selection is im-
portant in order to ensure a good results. The time step corresponds to one degree
rotation of the rotor and is expressed as:
Ts = / (180 ω) (1)
Where:
ω: Vitesse angulaire (rad/s)
Time Step Size Effect on the Hydrodynamic Structure 361
In this study the used time steps correspond to 1.25° and 2.5° rotation and they
are shown in Table 1.
Table 1 Time steps used for time step effect

Time step (s) Corresponding degree of rotation (degree)
0.1 2.5
0.05 1.25
3 Numerical Model
In the present work, the commercial CFD code Fluent is used to perform 3D simu-
lations, solving the incompressible Unsteady Reynolds-Averaged Navier-Stokes
(U-RANS) equations discretized by means of a finite volume approach. The im-
plicit segregated version of the solver is employed. The pressure-velocity coupling
is achieved by means of the SIMPLE algorithm. The convective terms are discre-
tized using a second order accurate upwind scheme, and pressure and viscous
terms are discretized by a second-order-accurate centered scheme. A first order
implicit time formulation is also used. Turbulence closure is provided by shear
stress transport k-ω turbulence model. The model has been validated, by compar-
ing numerical results with available experimental data (Matsushita et al. 2008).
3.1 Meshing Size

A mesh with structured triangular grid has been created in the stationery area and,
a structured quadrilateral grid in the moving area using "GAMBIT” software.
Total number of nodes Total number of cells
239 538 956 054
Fig. 2 Overall grid (left) and near blade grid (right)

3.2 Boundary Conditions

Figure 3 illustrates all information about boundary conditions is required any-
where fluid enters or exits the system and can be set as a pressure, mass flow, vo-
lume flow or velocity. In the considered simulation, the velocity inlet, we will take
as a value Vi=1 m.s-1 , and for the pressure outlet a value of 101325 Pa will
be considered which means that at this opening the fluid exits the model to an area
of static atmospheric pressure. Figure 3 illustrate all information. The rotating
-1
domain is characteristic by an angular velocity equal to w = 25.61 rad.s .
Fig. 3 Boundary conditions
4 Numerical Results
In this study, we are interested to visualize the velocity field, the static pressure,
the dynamic pressure, and the turbulent kinetic energy. Indeed, the variation of
each result as function of time in different points located in the downstream of the
rotor (figure 4).
points x (m) y (m) z (m)
Point 1 0.22 0 0
Point 2 0.22 0.15 0
Point 3 0.22 -0.15 0
Fig. 4 Visualization planes
4.1 Velocity Fields

Figure 5 shows the distribution of the velocity fields presented in the longitudinal
plane containing the blade and defined by z=0 m. According to these results, it is
clear that the velocity field is weak in the inlet of duct. Indeed, it is governed by
the boundary condition value of the inlet velocity witch is equal to 1 m.s-1. While
approaching to the water Darrieus rotor, the velocity field is affected by the geo-
metric configuration. In fact, flow acceleration appears in the gap between the
rotor and the blades walls. In the downstream of the rotor, the velocity value keeps
decreasing until the out of the test section. Then, a sharp increase has been noted
through the divergent angle where the velocity values are recorded in the lateral
walls of the duct. Globally, similar results are obtained with the different time step
sizes. However, the difference is localized in the duct extremity and in the blades.
Fig. 5 Velocity fields distribution
4.2 Static Pressure

Figure 6 shows the distribution of the static pressure in the test section equipped
by the water Darrieus rotor. According to these results, a considerable increase of
the static pressure has been observed. In the blade upstream, a compression zone
has been appeared. However, in the blade downstream depression zones have been
observed. In the longitudinal section plane defined by z=0 m, it has been noted
that the pressure increases slightly in the first part of the test section and it de-
creases rapidly in the water Darrieus rotor downstream. Globally, similar results
are obtained with the different time step sizes. However, the difference is localized
in the rotating are a around the blades. In fact, the compression zones are larger
with Ts= 0.1 s. However, the depression zone are larger with Ts= 0.05 s.
Fig. 6 Static pressure distribution

Figure 7 shows the evolution of the static pressure in three different points (1),
(2) and (3) localized in the downstream of the rotor for a period of 2 seconds.
Overall, it has been noted a significant fluctuations of the static pressure curves.
Particularly, it has been observed that the evolution of the static pressure is less
important in point (1) than in the points (2) and (3). This fact due to the position of
this point located in the middle of the duct. For the points (2) and (3) located in the
lateral faces of duct, it has been noted a little difference between the results due to
the effect of time steps sizes.
Fig. 7 Static pressure variation

Figure 8 illustrates the distribution of the dynamic pressure in the test section
equipped by the water Darrieus rotor. According to these results, it has been noted
that the dynamic pressure depends essentially on the fluid speed. A first view
shows that the dynamic pressure is found weak in the inlet of duct. When it gets to
the water turbine, the dynamic pressure keeps increasing in the upstream of the ro-
tor and around it. In fact, flow acceleration appears in the gap between the rotor
and the blades walls. This fact is confirmed in the longitudinal plane defined by
z=0 m. In the rotor downstream of, the dynamic pressure value keeps decreasing
until the out of the test section. Then, a sharp increase has been noted through the
divergent angle where the velocity values are recorded in the lateral walls of the
duct. This fact is confirmed in the transverse section plane defined by y=0 m. The
maximum value is obtained in the gap between the rotor and the duct walls. Under
these conditions, the dynamic pressure is equal to 2.36103 Pa at Ts= 0.1 s and
1.99103 Pa at Ts= 0.05 s. globally, similar results are obtained with the different
time step sizes. However, the difference is localized in the extremity of the duct
exactly in the gaps between the rotor and the blades walls.
Fig. 8 Dynamic pressure distribution
Figure 9 shows the evolution of the dynamic pressure indifferent in three points
(1), (2) and (3) located in the downstream of the Darrieus rotor for a period of 2 s.
Overall, it has been noted a significant fluctuations of the dynamic pressure
curves. Indeed, it’s clear that the fluctuating of the dynamic pressure is less impor-
tant in point (1) than the point (2) and (3). This fact is due to the position of this
point located in the middle of the duct. Indeed, it has been noted a difference be-
tween the results due to the time steps effect.
Fig. 9 Dynamic pressure variation

Figure 10 shows the distribution of the turbulent kinetic energy in the test section
equipped by the water turbine. From these results, it is clear that the turbulent ki-
netic energy is found to be very weak in the test section except in the area sur-
rounding the rotor. The distribution of the turbulent kinetic energy shows the in-
crease of the energy in the interior zone of the rotor. From a first glance, a small
difference appears between the distributions of the turbulent kinetic energy for
each time steps size. This difference is located in the rotating area, lacted around
the blades. The maximum value of the turbulent kinetic energy is recorded around
the blades as shown in the longitudinal and transverse planes defined by z=0 m
and y=0 m. Under these conditions, the turbulent kinetic energy is equal to 0.222
m2.s-1withTs=0.1s and 0.196 m2.s-1 with Ts=0.05s.Globally, similar results are ob-
tained with the different time step sizes. However, the difference is localized in
the lateral faces of the duct exactly around the blades walls.
Fig. 10 Turbulent kinetic energy distribution
Figure 11 shows the evolution of the turbulent kinetic energy in different three
points (1), (2) and (3) located in the downstream of the turbine for a period of 2 s.
Overall, it has been noted a significant fluctuations of the turbulent kinetic energy
curves. According to these results, it has been noted that the turbulent kinetic
energy fluctuation is less important in point (1) than the point (2) and (3). This fact
is due to the position of this point located in the middle of the duct. Indeed, it has
been noted a difference between the two results, due to the effect of the time steps.
Fig. 11 Turbulent kinetic energy variation
To verify our numerical results, we have compared in figure 12 the dimensional

axial velocity Va/Va value as a function of the 2y/D position obtained by our nu-
merical model with the experimental value founded by Matsushita et al. (2008).
Based on the numerical results, we found that the tow curves are close to the expe-
rimental data. Indeed, it has been noted that the curves with Ts=0.05 s close to the
experimental. Thus, so we can conclude that when the time steps size decreases,
the degree of rotation decrease, so we approach to the reality. According to these
results, the relative gap calculated between the numerical and the experimental
axial velocity values is equal to 8%. The good matching between these results in-
dicates the validity of our computer method.
Fig. 12 Velocity profiles variation
6 Conclusion
In this paper, we have studied the numerical parameters effect in the unsteady
state. Particularly, we have studied the time step sizes effect and we concluded
that this parameter has an influence in the rotation degree of the water Darrius ro-
tor. Then, we have concluded that the sliding mesh (SM) method can be used to
model the water turbine characteristics. This study allows us to find the adequate
numerical parameters enabling more detailed analyses of the flow around the Dar-
rieus rotor.
References
Shiono, M.I., Katsuyuki, S.U., Seiji, K.I.: An Experimental Study of the Characteristics of a
Darrieus Turbine for Tidal Power Generation. Electrical Engineering in Japan 132(3)
(2000)
Prabhu, S.V., Vimal, P.A., Himanshu, C.H.: Performance Prediction of H-Type Darrieus
Turbine by Single Stream Tube Model for Hydro Dynamic Application. International
Journal of Engineering Research & Technology 2(3) (2013)
Maitre, T., Amet, E., Pellone, C.: Modeling of flow in Darrieus water turbine: wall grid re-
finement analysis and comparison with experiments. Renewable Energy 51, 497–512
(2013)
Huang, S.W., Tsai, Y.D., Liang, S.Y., Hsieh, C.H., Chen, S.J.: Wind-tunnel study on aero-
dynamic performance of small vertical-axis wind turbines (2009)
Shimokawa, K.A., Akinori, F.U., Kusuo, O.K., Daisuke, M.A.: Satoshi WA Side-wall ef-
fect of runner casing on performance of Darrieus-type hydro turbine with inlet nozzle
for extra-low head utilization. Technological Sciences 53, 93–99 (2010)
Matsushita, D.A., Okuma, K.U., Watanabe, S.A., Furukawa, A.K.: Simplified structure of
ducted Darrieus-type hydro turbine with narrow intake for extra-low head hydropower
utilization. Journal of Fluid Science and Technology 3, 387–397 (2008)
Malipeddi, A.R., Chatterjee, D.: Influence of duct geometry on the performance of Darrieus
hydro turbine. Renewable Energy 43, 292–300 (2012)
Roa, A.M., Aumelas, V., Maître, T., Pellone, C.: Numerical and experimental analysis of a
Darrieus-type cross flow water turbine in bare and shrouded configurations. In: 25th
IAHR Symposium on Hydraulic Machinery and System (2010)
Simulation of the Dynamic Behavior
of Multi-stage Geared Systems with Tooth
Shape Deviations and External Excitations
Hassen Fakhfakh1, Jérome Bruyère1, Philippe Velex1,*, and Samuel Becquerelle2
1
INSA de Lyon, LaMCoS, France
2
Hispano-Suiza, France
Philippe.Velex@insa-lyon.fr
Abstract. In this paper, a torsional dynamic model of multi-stage idler spur and
helical gears is presented which combines time-varying internal and external exci-
tations such as time-varying external torques. Each contact line in the various base
planes is discretized in elemental cells which are all attributed a time-varying
mesh stiffness element and an initial separation to account for tooth shape devia-
tions from ideal involute flanks. The corresponding non-linear differential system
is solved by combining a Newmark’s numerical scheme and a normal contact al-
gorithm. A number of simulation results are presented on the influence of the
combined effect of errors and shape deviations along with external excitation
sources on dynamic tooth loads.
Keywords: Multi-stage idler gears, Spur and helical gears, Time-varying torques,
Tooth shape deviations.
1 Introduction
In certain conditions, power circulation can influence vibrations as is the case for
the dynamic behavior of multi-stage geared systems with the various possible
combinations of resisting and driving torque positions. In this context, Kuria and
Kihiu (2012) developed a numerical model to study the effect of three design vari-
able on dynamic stress of multi-stage spur gear. Al-Shyyab and Karahman (2005)
used a Harmonic Balance Method in conjunction with continuation technique to
describe the periodical responses of multi mesh gear system. Velex and Saada
(1991), Velex and Raclot (1999), Kubur and al. (2004) proposed FE models of
double stage drives and investigated the influence of key design parameters on the
system dynamics. Carbonelli (2014) used a Finite Difference approach accounting
for both internal and external sources in the time domain such as clearance nonli-
*

370 H. Fakhfakh et al.
nearity. Yang (2013) studied the vibrational response in torsion of a two-stage

gear submitted to deterministic and random excitations in the presence of back-
lash.
This paper presents a torsional dynamic model of multi-stage idler spur or heli-
cal gears aimed at simulating the dynamic tooth loading in the presence of several
time-varying mesh stiffness whose relative phases are determined based on the
gear geometry and their relative positions. The model can moreover account for
tooth shape modifications and constant or time-varying external loads which can
strongly influence dynamic tooth forces.
2 Torsional Model of Multi-Stage Idler Gears
2.1 Mesh Stiffness Matrix
The normal deflection at any point of contact M mij (m is the stage number), is
equivalent to the normal approach δ mij with respect to rigid-body positions minus
the initial separation δ emij possibly induced by tooth shape modifications:
Δ mij = δ mij − δ emij (1)
In what follows, k mij represents the discrete stiffness element associated with
each discrete cell on the contact lines which depends on load and shape deviations
(Velex and Maatar 1996) and which is expressed as:
⎧ H = 0 if Δ
⎪ mij < 0
k mij = kˆmij H ( Δ mij ) with ⎨ (2)
⎪ H =1 if Δ mij > 0
⎩
where H ( Δ mij ) is the Heaviside function used to simulate total or partial contact
losses.
For a pinion-gear pair with one torsional degree of freedom each, the mesh stiff-
ness matrix can be expressed as:
JJJG JJJJGT
⎡ K (t )⎤ = ∑ k (3)
⎣ m ⎦ mij Vm Vm
mij
JJJG
where Vm = ζ Mm Rbm cos βb ζ Mm Rbm +1 cos βb is a structural vector which accounts for
gear geometry.
Simulation of the Dynamic Behavior of a Multi-stage Geared Systems 371
Extending the methodology to a system with one pinion, one idler gear and one
output member, the global stiffness matrix reads:
⎡ ⎤
⎢ k1 ( t ) Rb1 k1 ( t ) Rb1 Rb 2
2 0 0 0 ⎥
⎢ ⎥
⎢k ( t ) R R
⎢1 b1 b 2 ( )
k1 ( t ) + k2 ( t ) Rb22 k2 ( t ) Rb 2 Rb 3 0 0 ⎥
⎥
⎣ ⎦ ( )
⎡ K ( t )⎤ = cos 2 β ⎢
b ⎢ 0 k2 ( t ) Rb 2 Rb 3 ( )
k2 ( t ) + k3 ( t ) Rb 3 "
2 0
⎥
⎥
(4)
⎢ ⎥
⎢ 0 0 # ⎥
⎢ ⎥
⎢⎣ 0 0 k N −1 ( t ) RbN + k h ⎥⎦
2
2.2 Model and Equations of Motion

All the numerical results presented in this paper are related to the torsional model
of N idler spur and helical gear shown in figure 1. The pinions are modelled as
rigid discs connected by time-varying mesh stiffness functions (Kubo 1978, Gregory
and al 1963, etc.) whose relative phasing are determined analytically. Time-varying
loads are considered which can be caused by generators or pumps and fluctuating
power input. In this paper, a periodic torque of the form Ct = C o + C s sin(k Ω t) will
be considered where Ω is the angular frequency of the member submitted to
torque Ct. At a given time t, the equivalent deviations are discretized on the base
planes using the same elemental cells as those introduced for mesh stiffness ele-
ments. From an initial position of the contact lines, a time-step procedure repro-
duces the mesh stiffness and gives the instantaneous equivalent normal deviation
em ( M ) at each contact point M. In these conditions, the equations of motion read:
[ M ] {q} + [C ] {q} + ⎡⎣K ( t ,q )⎤⎦ {q} = {F0( t )} + {F1( t ,q,δ em )} + {F2 ( t )} (5)
where [M] and [K(t,q)] are the global mass and stiffness matrix, the damping ma-
trix is expressed using a Rayleigh model such that [C]=a.[M]+b.[Kav] (Velex
1988) , {F0 ( t )} = Ct1 ( t ) Ct2 ( t ) ... Ct N ( t ) is the external torque vector,
{F1 (t , q ,δ em )} embodies the contributions of shape deviations and errors and its com-
ponents are:
⎧ k1 ( t )δ e1 ( M )ζ M 1Rb1 cos βb ⎫
⎪ ⎪
⎪
⎪ ( )
k1 ( t )δ e1 ( M )ζ M 1 + k2 ( t )δ e2 ( M )ζ M 2 Rb2 cos βb ⎪
⎪
⎪ ⎪
( )
{F1(t ,q,δ em )} ⎪ k2 ( t )δ e2 ( M )ζ M 2 + k3 ( t )δ e3 ( M )ζ M 3 Rb3 cos βb ⎪
= ⎨ ⎬ (6)
⎪ # ⎪
⎪ ⎪
( )
⎪ k N − 2 ( t )δ eN − 2 ( M )ζ M N − 2 + k N −1 ( t )δ eN −1 ( M )ζ M N −1 RbN −1 cos βb ⎪
⎪ ⎪
⎪ k N −1 ( t )δ eN −1 ( M )ζ M N −1RbN cos βb ⎪
⎩ ⎭
and {F2 ( t )}=

− J1Ω 1

− J 2Ω 2

... − J N Ω N is the inertial torque vector possibly gener-
ated by unsteady rotational speeds.
The differential system is solved by a Newmark’s time step integration me-
thods. The contact condition and constraints are directly inserted in the integration
scheme so that the partial or total contact losses problem between the mating
flanks can be solved at any time step.
Ω1(t)
O1
e1(t): excitation due to geometrical errors
K1: excitation due to
mesh stiffness
c1 K2
O3 ∆l2
c2
O2 e2(t)
kh
∆l1: relative phase shift ON
Fig. 1 Multi-stage idler gear with N meshes
3 Dynamic Results
The numerical applications were conducted on the 4-stage idler spur gear shown
in Figure 2 where the system is represented with the pinion centres all aligned
such that γ1 = γ2 = γ3 = 180°. The corresponding gear data are listed in Table 1.
Fig. 2 Idler gear system

Table 1 Gear data
Pinion 1 Pinion 2 Pinion 3 Pinion 4 Pinion 5

Tooth number 50 39 47 51 28
Face width(mm) 25.4 25.4 25.4 46.5 25.4
Module (mm) 3.18
Pressure angle (°) 20
Helix angle (°) 0 (spur) or 20 (helical)
Addendum coefficient 1.0 1.0 1.0 1.0 1.0
Profile shift coeff. 0.0 0.0 0.0 0.0 0.0
Dedendum coefficient 1.4 1.4 1.4 1.4 1.4
3.1 Influence of an Oscillating Torque and Tooth Shape

Modifications
The influence of a periodically varying torque is illustrated by comparing the dy-
namic tooth force curves versus speed for constant torque and those obtained
when considering a fluctuating torque on pinion 1 (figure 3). In both cases, a
damping factor ξ = 0.04 was used.
Table 2 Tooth critical frequency
Tooth critical frequency (rad/s)

W1 73.790
W2 11814.744
W3 25139.066
W4 38266.122
W5 34321.088
The critical speeds of the four stages appear simultaneously on the four curves
which proves the significant inter-stages coupling so that each pinion is excited by
all the mesh stiffness functions. The response curves for a constant load (figure
3.a) exhibit peaks at 775 rpm ≌ (W2/50)/3, 1075 rpm ≌ (W2/50)/2 and 1950 rpm
corresponding to a natural frequency being excited by the mesh frequency and its
harmonics. The introduction of a time-varying torque of frequency 20Ω1 gene-
rates mostly an additional response peak at 5650 ≌ W2/20 rpm. The system also
exhibits contact losses between the teeth at some critical speeds on some of the re-
duction stages. The introduction of short profile corrections (linear relief of depth
Pf=60μm at tooth tips and extents corresponding to 20% of the active profile) on
all the pinions, leads to a substantial improvement of the dynamic behaviour be-
low the main critical speed (≌775 rpm) as shown in figure 3.c.
(a)
(b)
(c)
Fig. 3 Influence of (a) constant torque and (b) time-varying torque (c) profile corrections on
dynamic mesh forces
3.2 Influence of an Intermediate Oscillating Torque and Tooth

Shape Modifications
This section deals with the influence of the power circulation combined with tooth
profile relief on the dynamic behavior. To this end, the following loading condi-
tions are investigated: i) a constant driving torque Ct1= 1200 Nm is applied to pi-
nion 1, ii) a resisting torque either constant or periodic Ct3=-500 or -500+200.sin
(20Ω1t) is imposed on pinion 3 and iii) a resisting torque on the output gear 5 (ad-
justed so that the total power is conserved).
(a)
(b)
Fig. 4 Influence of (a) constant torque and (b) time-varying torque (c) profile corrections on
dynamic mesh forces
(c)
Fig. 4 (continued)
The results in figure 4 show that although the critical rotational speeds are the
same as in the previous case, the different power circulation modifies the dynamic
amplifications. The introduction of the resisting torque on pinion 3 unloads the
meshes downstream (between pinions 3, 4 and 5) over the whole speed range. When
the periodic component is superimposed (Ct3=-500+200.sin (20Ω1t)), an additional
tooth critical speed emerges for all the meshes amplifying contact losses especially
on stage 3. The response spectra (not shown in this paper) exhibit significant mod-
ulation side-bands related to the torque frequency 20Ω1. The response curves for
short profile corrections shown in figure 4.c illustrates tooth forces amplitude and
contact losses decrease below the main critical speed (≌775 rpm).
4 Conclusion
In this paper, a simplified torsional dynamic model of N stage idler spur and heli-
cal gears has been presented. The time-varying mesh stiffness functions attached
to every individual reduction stage are estimated from the classic formulae of We-
ber & Banaschek and their relative phases are determined analytically. The model
can simulate the influence of periodic torques and power circulations (several out-
puts, time-varying resisting loads, etc.) on tooth load amplification and critical
speeds. It has been shown that short profile corrections can improve the dynamic
behavior for a range of rotational speeds. Research is under way to incorporate
three-dimensional degrees of freedom, shafts, bearings and casings along with a
variety of loading conditions.
Acknowledgements. The authors would like to thank the support of Hispano-Suiza from
the Safran Group.
Nomenclature

Ωi ,Ω = angular velocity, acceleration of pinion i
i
Rbi = base radius of pinion i
Ji = polar moments of inertia of pinion i
βb = base helix angle
ζ Mm = sign of the driving torque of stage m
2
kh = torsional stiffness added to node N ( k h = 10 Nm / rad )
c0 , cs = average, oscillation amplitude of the periodic torque
a ,b = constants that can be adjusted to get the desired damping factor ξ .
−5
In this case a = 1, 12 ; b = 1, 2.10
⎡K ⎤ = average stiffness matrix
⎣ av ⎦
References
Kuria, J., Kihiu, J.: Effect of Gear Design Variables on the Dynamic Stress of Multi stage
Gears. Innovative Systems Design and Engineering (2012) ISSN 2222-1727
Al-Shyyab, A., Karahman, A.: Non-linear dynamic analysis of a multi-mesh gear train us-
ing multi-term harmonic balance method: sub-harmonic motions. J. of Sound and Vib.
(2005), doi:10.1016/j.jsv.2003.11.029
Velex, P., Saada, A.: A model for the dynamic behavior of multi-stage geared system. In:
Proc. of the 8th World Congress on Theory of Machines and Mechanisms, Prague,
vol. 2, pp. 621–624 (1991)
Raclot, J.-P., Velex, P.: Simulation of the dynamic behavior of single and multi-stage
geared systems with shape deviations and mounting errors by using a spectral method. J.
of Sound and Vib. 220(5), 861–903 (1999)
Kubur, M., Kahraman, A., Zini, D., Kienzle, K.: Dynamic analysis of a multi-shaft helical
gear transmission by finite elements – Model and experiments. J. Vib. Acous. 126(3),
398–406 (2004)
Carbonelli, A.: Hammering noise modelling-Nonlinera dynamic of a multi-stage gear train.
In: International Gear Conference, Lyon, pp. 447–456 (2014)
Yang, J.: Vibration analysis on multi-mesh gear-trains under combined deterministic and
random excitations. Mechanism and Machine Theory 59, 20–33 (2013),
doi:10.1016/j.mechmachtheory.2012.08.005
Velex, P., Maatar, M.: A Mathematical Model for Analysing the Influence of Shape Devia-
tions and Mounting Errors on Gear Dynamic Behaviour. J. of Sound and Vibra-
tion 191(5), 629–660 (1996), doi:10.1006/jsvi.1996.0148
Kubo, A.: Stress condition, vibrational exciting force, and contact pattern of helical gears
with manufacturing and alignment error. J. Mech. Des. 100(1), 77–84 (1978),
doi:10.1115/1.3453898
Gregory, R.W., Harris, S.L., Munro, R.G.: Dynamic Behavior of Spur Gears. Proc. of the In-
stitution of Mechanical Engineers 178(1), 207–218 (1963),
doi:10.1177/002034836317800130
Velex, P.: Contribution à l’analyse du comportement dynamique de réducteurs à engrenag-
es à axes parallèles. Thèse de Doctorat, INSA de Lyon.N°88 ISAL 0032 (1988)
Fuzzy Modeling and Control of Centrifugal
Compressor Used in Gas Pipelines Systems
Ahmed Hafaifa1, Guemana Mouloud1,2, and Belhadef Rachid3
1
Applied Automation and Industrial Diagnostic Laboratory,
University of Djelfa 17000 DZ, Algeria
hafaifa.ahmed.dz@ieee.org
2
Faculty of Science and Technology, University of Medea 26000 DZ, Algeria
Guemana_M@univ-medea.dz
3
Faculty of Science and Technology, University of Sedik Ben yahia of Jijel, Algeria
r.belhadef@univ-biskra.dz
Abstract. Respond to changing technology industrial installations, this work pro-

pose solutions to the modeling and control problems in industrial processes with
the use of new approaches. The objective of this work is the use of fuzzy tech-
niques in modeling and control in the study of gas compression system instability.
The obtained results show clearly how the main dynamic characteristics, in our
examined compression system, are reproduced using the proposed fuzzy model,
allowing better performance during its control synthesis operation.
Keywords: Gas compression system, surge phenomena, centrifugal compressor,

exploitation instability, fuzzy modeling, fuzzy control.
1 Introduction
The development emergence in new automation techniques led to significant

changes in the control systems design. Indeed, in many industrial applications, the
scientific vision takes in some technology areas, a particularly interesting turn in
the direction of developing control theory and its applications [1, 18, 23 and 25]. It
is therefore natural to seek a mechanism to show the face of intelligent features
extensive changes in a more complex environment. In this concept, we are trying
to improve some traditional techniques are unable to adapt to complex environ-
ments, uncertain or variants over time. To solve the problem of instability in com-
pressors and in order to improve system performance, we propose in this work the
use of new techniques. These new techniques may in some cases prevent the in-
stability phenomenon, and give a certain approach to solving the problem of the
permanent opening of the recirculation valve. Subject to integration of new algo-
rithms that provide real-time regulation, this will recover the lost production to a
certain extent while ensuring the functioning machine and protect the compressor
from all dangers.

380 A. Hafaifa, G. Mouloud, and B. Rachid
In this work, we propose to examine and illustrate the application ability of

fuzzy logic to the modeling of our examined gas compression system, and to in-
volve this fuzzy model in the synthesis of control strategies appropriate for this
type of system. Several studies have focused on the problem of its modeling [4,
10, 14], most of which have presented rigorous models, with a very complex ma-
thematical structure limiting, therefore, the possibility of direct exploitation by the
conventional control. These models are established in nominal conditions of work,
making it difficult to control the compressor suction flow. Due to the precision, it
uses control systems fairly robust, able to adapt to some degree of satisfaction to
possible variations of process parameters and conditions of its operation. In fact, it
seems more than necessary to be sufficiently accurate and complete knowledge of
the process in order to satisfy the requirements of robustness and performance. It
is in this subject we propose, a fuzzy model for our examined gas compression
system to describing the dynamics of this system. Fuzzy logic provides perfor-
mance benefits in modeling of the considered process.
2 Gas Compression System Modeling
In this work, we are interested by the compression system model developed by

Tommy GRAVDAHL [4, 19]. Taking into account the maximum losses, in order
to obtain a model, able to predict the actual characteristics of the examined gas
compression system. The overall performance characteristics are identified at the
point of maximum efficiency and normal operating conditions. The examined gas
compression system, shown in figure 1, is composed by a turbine, compressor,
pipes and an exhaust valve, the figure 2 show allows us to situate the different
areas in the examined compressor.
Yt
1.5
Anti Surge
Valve User 1
Zone de fonctionnement
Normal
normale
operating area Zone de Pompage
Instability area
0.5
Mass flow coefficient
Coefficient de débit
N 0
-0.5
-1
Compressor Turbine
Gas Input
-1.5
-0.4 -0.2 0 0.2 0.4 0.6 0.8 1
Pa Ta Load factor
Coefficient de Charge
Fig. 1 Examined gas compression system Fig. 2 Centrifugal compressors characteristics

Fuzzy Modeling and Control of Centrifugal Compressor Used in Gas Pipelines Systems 381
The model developed by GRAVDAHL [4, 19] is based on the concepts of

thermodynamics and fluid dynamics and it is written in the following form:
2
a
PP = 01 ( m - m t )
VP
A1
=
m ( P2 - PP )
LC (1)
1
N = (τ t -τ c )
2Jπ
Where m is the mass flow of the compressor inlet, mt the output mass flow
of the valve, Pp the output pressure, V p the output gas volume, P2 the pressure
at the compressor outlet, a01 the sound speed at the compressor inlet, Lc the
length of the compressor driving, A1 the goose surface of the impeller, J the
compressor moment of inertia, τ c the compressor torque and τ t is the turbine
torque.
This model is similar to FINK AL model [7], while the two first equations of
(1) are equivalent to Greitzer model [9, 10], where the speed is not included. The
Pp equation is derived from the mass balance assuming the trial isentropic trans-
fer. The third equation is derived from the second law of dynamics:
∑ M = J W ⇒ = 1 (τ -τ )
W
J
t C (2)
With M is the moments applied to the compression system and W is the an-
gular acceleration.
The compressors characteristics can be defined by different methods, in this
work we adopting the definitions suggested by Greitzer in [9, 10], namely:
w V
• The specific speed: η S = ,
Δ H 3/4
t/2
Vm
• The mass flow coefficient: Φ = , D 2 is the outlet impeller diameter,
U2
Δ H t/2
• The load factor: Ψ = .
U 22
The trial is assumed isentropic, so we can write from a mass balance:
dPP dVP dρ
=-k =k P (3)
PP VP ρP
1 dρ dV
We have ρ = which implies =- , where:
V ρ V
dP
P P = k P ρ P = k R TP ρ P (4)
PP
The mass balance is given by the following relationship:
1 T
ρP = ( m - m t ) ⇒ P P = k R P ( m - m t ) = ( m - m t ) (5)
VP VP
Where a s = K R TP is the speed of sound and the temperature ratio of the

PP P01
compression close to unity = R TP , then the amount will be too near . So
ρP ρ 01
we can conclude that:
P
PP = k 01 VP ( m - m t ) (6)
ρ 01
To complete the model (1) for give the dynamical model, we need to the output
mass flow rate ( mt = k t Pp - P01 ) with k t the proportional gain of the gas re-
circulation valve. The driving torque of the turbine is calculated by dividing the
power of the turbine by the angular velocity, is given by :
Pt η t m tur C p , t Δ Ttur
τt = = (7)
w 2π N
Where: mtur is the steam flow at the inlet of the turbine and η is the mechan-
ical efficiency of the turbine.
Finally, we have the complete dynamical model given by the equations 8.
⎧ 2
a 01 ⎛⎜ m - k
⎪ p
P = Pp - P01 ⎞⎟
⎪ V p
⎝ t
⎠
⎪
⎪ ⎡ k −1
⎤
⎪ A ⎢ ⎛ Δ h ideal ⎞ k
⎥
⎨ m = 1 ⎢ Pp ⎜ 1 + η i ( m , N ) ⎟ - Pp ⎥ (8)
⎪ Lc ⎢ ⎝ ⎜ C p T01 ⎟
⎠ ⎥
⎪ ⎣ ⎦
⎪
⎪
⎪ N
⎩
= 1 τ -τ
2Jπ
(
t c )
The previous study was conducted to centrifugal compressors in a single stage.

The model of a multistage compressor is given by the equations (9).
⎧ k P01 ⎛
⎪ P p = ⎜ m - kt Pp - P01 ⎞⎟
⎪ ρ 01 Vp ⎝ ⎠
⎪
⎪ ⎡ 4 ( k -1 )
⎤
⎪ A1 ⎢ ⎛ Δ h ideal ⎞ k
⎥
⎪ ⎜ ⎟
⎨
m= ⎢ P01 ⎜ 1 + η i (m, N) ⎟
- Pp ⎥ (9)
⎪ Lc ⎢ ⎝ C p T01 ⎠ ⎥
⎪ ⎣ ⎦
⎪
⎪ ⎛ η m C Δ Ttur ⎞
⎪ = 1 ⎜ t tur p , t
N - 2 r22 σ π N m ⎟
⎪⎩ 2 J π ⎜⎝ 2π N ⎟
⎠
The obtained model (9) is complex and difficult to exploit in control. Admitted-
ly, it better describes the main physical characteristics of the system, but it is often
necessary to simplify this complex model by linearization near a nominal operat-
ing point or by neglecting the less dominant dynamic process.
3 Fuzzy Modelling and Control of the Gas Compression

System
Fuzzy logic offers a benefits modeling performance in industrial systems [3, 6, 8,

13, 21 and 22]. Let the differential system (9). In practice, the structure and para-
meters of this complex model are established under several simplifying assump-
tions. The linearization of the model given by the equations (9) about an operating
point M ( Ppc 0 , mc 0 , u t 0 , ub 0 ) is given by :
⎡ Pˆ ⎤ ⎡ B m −B ⎤ ⎡
P ⎤ ⎡ 0 ⎤
x = ⎢ pC ⎥ = ⎢ 1 1 ⎥ ⎢ pC ⎥ + ⎢
− V / B ⎥⎦
û b (10)
⎢⎣ mˆ c ⎥⎦ ⎢ B ⎥ m
⎣ c ⎦ ⎣
⎣⎢ Bmte ⎦⎥
~
With t = twH and wH is the Helmholtz frequency defined by the following
AC B
equation: wH = a with a = γRTa and Bm = . The parameters B and
V P LC G
Ut L A
G are defined as following: B = and G = t C , B is the Greitzer
2wH LC LC At
stability parameter [9, 10]. It follows that the linear model assuming (9) achieving
status as follows:
⎧δx (t ) = A δx(t) + B δu(t)
⎪
⎨ (11)
⎪δy (t ) = C δx(t) + D δu(t)
⎩
With δx = x - x 0 , δu = u - u 0 , δy = y - y 0 and x 0 = [ 5.74 1.25 12.3] ,

T
y 0 = [ 2.23 6.6 0] , u 0 = [ 0.20 0.13 - 0.15 ] . The matrices A, B, C and D

T T
assessed in equilibrium are given by:
⎡ ⎤ ⎡ ⎤
⎢ ⎥ ⎢ ⎥
⎢ 0.2076 0.1323 0⎥ ⎢ 0.2076 1.2365 0 ⎥
A = ⎢⎢ ⎥, B = ⎢⎢ ⎥
3.7181 - 0.4892 0⎥ 0 - 3.6891 - 2.312⎥
⎢ ⎥ ⎢ ⎥
⎢ ⎥ ⎢ ⎥
⎢⎣- 9.6870 - 1.539 0⎥⎦ ⎢⎣ 0 - 3.8230 0 ⎥⎦
⎡ ⎤ ⎡ ⎤
⎢ ⎥ ⎢ ⎥
⎢1 0 0⎥ ⎢ 0 0 0⎥
C = ⎢⎢ ⎥, ⎢
D=⎢ ⎥.
0 1 0⎥ 0 0 0⎥
⎢ ⎥ ⎢ ⎥
⎢ ⎥ ⎢ ⎥
⎢⎣0 0 1⎥⎦ ⎢⎣12.10740 0.3582 1⎥⎦
The first goal in centrifugal compressors modeling is to estimate losses, the first
model called "Single-Zone Modeling" includes only global characteristics of the
compressor. The proposed fuzzy model for the examined compression system is
symbolically expressed by a set of IF-THEN rules of the form [5, 26]:
IF Z 1 (t ) est Fi1 et ... .. et Z n est Fin
THEN
⎧ xˆ (t ) = A i x̂(t) + B i u(t) + K i ( y (t ) - ŷ(t) )

⎪
⎨ (12)
⎪ ŷ (t ) = C x̂(t) , i = 1, 2 ...... , r
⎩ i i
With: Rule i, i = 1, 2 ..... r

We are interested by fuzzy Takagi-Sugeno modelling [2, 20, 26], defined by:
⎧
[ ]
r
⎪ x (t) = ∑ h i ( z(t) ) A i x(t) + B i u(t)

⎪ i=
⎨ (13)
⎪ r
⎪ y (t) = ∑ h i ( z(t) ) C i x (t)

⎩ i=
The obtained results using fuzzy controller, shown in figure 3 are better than
those obtained with conventional control algorithms. The proposed controller is a
system based on special knowledge, using a limited depth of reasoning in a for-
ward chaining procedure rules (rule firing by the premises).
Compression Compression
system output system input
Knowledge Base
Decision mechanism
Premises Inference Aggregation

treatment rules
Fuzzification Defuzzification
Normalization System control Denormalizatio
Compression
system
Fig. 3 Proposed fuzzy controller structure
Several stages in the processing rules are distinguishing in the proposed fuzzy
controller structure. The variables are the output quantities of gas compression
system and capture decisive action to the dynamic evolution in the examined sys-
tem. The knowledge base is composed of a database and a database of rules.
For the different cases studied, the symbolic representation of the dynamics of
the compression system, we have adopted consists of nine rules type If _Then.
The membership functions associated with various linguistic variables are

represented. Each partition consists of five fuzzy sets (N, NM, Z, PM, P) is un-
iformly distributed over an interval normalized [-1 1]. Matrices ( A j , B j , C j , D j ),
j = 1, ..., 9, are determined from the combination of the operating point considered
in the rule j. The dynamic behavior was studied for various cases, varying the
Greitzer stability parameter B for a variation of the flow and then to a pressure
variation (load factor and the mass flow coefficient). The figures 4, 5 illustrate the
fuzzy model of the mass flow coefficient for B = 1.25 and B = 1.
Comparison
Comparison
0.35
0
0.3
-0.05
Fuzzy model
Linearized model
-0.1 0.25
Complex model
Mass flow coefficient
M a ss flow co effic ien t

Débit en P om page
Débit d'as pération

-0.15
0.2
-0.2
0.15
Fuzzy model
Linearized model
-0.25 Complex model
0.1
-0.3
0.05
-0.35
0 1.6 3.2 4.8 6.4 8
0
time(s) 0 1.6 3.2 4.8 6.4 8
time(s)
Fig. 4 Fuzzy model of the mass flow coeffi- Fig. 5 Fuzzy model of the mass flow coef-
cient for B = 1.25 ficient for B = 1
From the obtund results it is clear that the proposed fuzzy model is more accu-
rate than the linearized model. In the various cases considered, the characteristics
of the compression system described by the model complex are perfectly repro-
duced by the proposed fuzzy model.
To validate the proposed model, we used a fuzzy controller that allows distri-
buted seven fuzzy sets on the universe of discourse of each input variable and out-
put (NG, NM, NP, EZ, PP, PM, PG). Increasing the number of fuzzy sets to seven
(NG, NM, NP, EZ, PP, PM, PG) requires the processing of 49 rules, so that the
distribution of these sets on the universe of discourse of each variable becomes a
choice. As regards the matrix inference, the sets of the output variable (increment)
are always derived from the Mac-Vicar matrix. By analyzing the responses of
fuzzy controllers we see the smooth continuation of fuzzy controllers to variations
of the operating. The figure 6 shows the evolution of load factor, we see that it is
properly oriented and stabilizes reference and the figure 7 shows the response of
proposed fuzzy controller applied to the load factor.
Pression sortieLoad factorflou

du régulateur
B =d’aspiration
1.25 / f 1 Résidu sur la B = 1 d’aspiration
pression 2.5
45 0
-0.1 2
40
-0.2
1.5
35
-0.3
1
30 -0.4
0 100 200 300 0 100 200 300
0.5
B =d’aspiration
Pression 0.25 / f B = 0.50
2 Résidu sur la pression d’aspiration
45 -0.1 0
-0.2
40 -0.5
-0.3
35 -1
-0.4
30 -0.5 -1.5
0 100 200 300 0 100 200 300 0 200 400 600 800 1000 1200 1400
temps
Time
Fig. 6 Evolution of load factor in the Fig. 7 Response of proposed fuzzy controller
examined system
The increase of the fuzzy sets increasing linguistic rules, therefore the computa-
tion time, the fuzzy controller with five sets in this investigated case represents a
good choice that satisfies the tradeoff between the computation time and the re-
quirement of performance specifications.
4 Conclusion
The obtund results confirm that the proposed fuzzy model is more accurate than
the linearized model. In the various cases considered, the characteristics of the
compression system described by the model complex are perfectly reproduced by
the proposed fuzzy model. Also, the fuzzy controller used appears to have mul-
tiple benefits. It does not require a mathematical programming is difficult and
easy. It should be noted that it requires a relatively slow computation time, while
traditional algorithms often result in very short programs. It may seem at first
glance that the fuzzy controller requires very little knowledge of the system to set-
tle. The controller is very sensitive to variations of system parameters, so its
rugged good looks. In reality should not conceal a certain number of disadvantag-
es of this type of controller.
References
Adamy, J., Schwung, A.: Qualitative modeling of dynamical systems employing conti-
nuous-time recurrent fuzzy systems. Fuzzy Sets and Systems 161(23), 3026–3043
(2010)
Fiordaliso, A.: A nonlinear forecasts combination method based on Takagi–Sugeno fuzzy

systems. International Journal of Forecasting 14(3), 367–379 (1998)
Esogbue, A.O.: Fuzzy sets modeling and optimization for disaster control systems planning.
Fuzzy Sets and Systems 81(1), 169–183 (1996)
Bøhagen, B., Gravdahl, J.T.: Active surge control of compression system using drive tor-
que. Automatica 44(4), 1135–1140 (2008)
Bortolet, P., Merlet, E., Boverie, S.: Fuzzy modeling and control of an engine air inlet with
exhaust gas recirculation. Control Engineering Practice 7(10), 1269–1277 (1999)
Wong, C.-C., Lin, N.-S.: Rule extraction for fuzzy modeling. Fuzzy Sets and Sys-
tems 88(1), 23–30 (1997)
Fink, et al.: Surge dynamics in free-spool centrifugal compressor system. Journal of Tur-
bomachinery 114, 321–332 (1992)
Panoutsos, G., Mahfouf, M.: A neural-fuzzy modelling framework based on granular com-
puting: Concepts and applications. Fuzzy Sets and Systems 161(21), 2808–2830 (2010)
Greitzer, E.M.: Surge and rotating stall in axial flow compressors, part I: Theoretical com-
pression system model. Journal of Engineering for Power 98, 190–198 (1976)
Greitzer, E.M., Moore, F.K.: A Theory of Post-Stall Transients in Axial Compression Sys-
tems: Part II – Applications. American Society of Mechanical Engineers, paper 85-GT-
172 (1985)
Ahmed, H., Daoudi, A., Guemana, M.: SCADA for Surge Control: Using a SCADA net-
work to handle surge control in gas suppression systems in pipelines. Control Global |
Process Automation Technologies Journal, ISA Transactions 24(3), 69–71 (2011)
Ahmed, H., Daoudi, A., Laroussi, K.: Application of fuzzy diagnosis in fault detection and
isolation to the compression system protection. Control and Intelligent Systems 39(3),
151–158 (2011)
Ahmed, H., Laaouad, F., Laroussi, K.: A Numerical structural approach to surge detection
and isolation in compression systems using fuzzy logic controller. International Journal
of Control, Automation, and Systems, IJCAS 9(1), 69–79 (2011)
Ahmed, H., Daoudi, A., Laroussi, K.: Modelling and control of surge in centrifugal com-
pression based on fuzzy rule system. Studies in Informatics and Control (SIC) 19(4),
347–356 (2010)
Ahmed, H., Laroussi, K., Laaouad, F.: Robust fuzzy fault detection and isolation approach
applied to surge in centrifugal compressor modeling and control. Fuzzy Information and
Engineering Journal 02(01), 49–73 (2010)
Ahmed, H., Laaouad, F., Laroussi, K.: Fuzzy logic approach applied to the surge detection
and isolation in centrifugal compressor. Automatic Control and Computer
Sciences 44(01), 53–59 (2010)
Ahmed, H., Guemana, M., Daoudi, A.: Spectral analysis approach applied to the vibrations
detection and isolation in centrifugal pump used in petroleum industry. In: 5th Interna-
tional Congress on Technical Diagnostics, Didactics Center, AGH Univeristy of Science
and Technology, Kraków, Poland, September 3-5 (2012)
Hafaifa, A., Laaouad, F., Laroussi, K.: Fuzzy modeling and control for detection and isola-
tion of surge in industrial centrifugal compressors. Automatic Control Journal of the
University of Belgrade 19(01), 19–26 (2009)
Gravdahl, J.T., Egeland, O., Vatland, S.O.: Drive torque actuation in active surge control of
centrifugal compressors. Automatica 38(11), 1881–1893 (2002)
Guelton, K., Bouarar, T., Manamanni, N.: Robust dynamic output feedback fuzzy Lyapu-
nov stabilization of Takagi–Sugeno systems—A descriptor redundancy approach. Fuzzy
Sets and Systems 160(19), 2796–2811 (2009)
Wang, L.-X.: Modeling and control of hierarchical systems with fuzzy systems. Automati-
ca 33(6), 1041–1053 (1997)
Mei, F., Man, Z., Nguyen, T.: Fuzzy modelling and tracking control of nonlinear systems.
Mathematical and Computer Modelling 33(6-7), 759–770 (2001)
Bernal, M., Guerra, T.M., Kruszewski, A.: A membership-function-dependent approach for
stability analysis and controller synthesis of Takagi–Sugeno models. Fuzzy Sets and
Systems 160(19), 2776–2795 (2009)
Guemana, M., Aissani, S., Hafaifa, A.: Use a new calibration method for gas pipelines: An
advanced method improves calibrating orifice flowmeters while reducing maintenance
costs. Hydrocarbon Processing Journal 90(8), 63–68 (2011)
Guemana, M., Aissani, S., Hafaifa, A.: Flow measurement and control in gas pipeline sys-
tem using intelligent sonic nozzle sensor. Studies in Informatics and Control
(SIC) 20(2), 85–96 (2011)
Sugeno, M., Kang, G.T.: Fuzzy modelling and control of multilayer incinerator. Fuzzy Sets
and Systems 18(3), 329–345 (1986)
Burst Test and J-Integral Crack Growth
Criterion in High Density Poly-Ethylene Pipe
Subjected to Internal Pressure
Mohamed Amine Guidara1,2, Mohamed Ali Bouaziz1,2, Christian Schmitt2,

Julien Capelle2, Ezzeddine Hadj-Taïeb1, Zitouni Azari2, and Said Hariri3
1
Laboratoire de Mécanique des Fluides Appliqués, Génie des Procédés et Environnement,
Ecole Nationale d’Ingénieurs de Sfax, Université de Sfax, BP 1173, 3038, Sfax, Tunisie
2
Laboratoire de Biomécanique, Polymères et Structures (LaBPS), Ecole Nationale
d’Ingénieurs de Metz, 57070 Metz, France
3
Département Technologie des Polymères et Composites and Ingénierie Mécanique, Ecole
des Mines de Douai, 59508 Douai, France
mohamed-amine.guidara@univ-lorraine.fr,
mohamedali.bouaziz@gmail.com
Abstract. In the present work we are interested on the analysis of the severity of
crack defects created by a disc cutter and to study the behavior of a high density
polyethylene pipe (HDPE pipe) when subjected to an internal pressure, either in
the absence or presence of a pre-crack. In order to do this, experimental tests was
performed to measure the toughness and to determine the mechanical behavior of
HDPE. These features were used to perform numerical simulations using
ABAQUS on pipe solicited by an increase in internal pressure. This allows to
compare with burst tests of cracked pipes and to determine the fracture energy that
will be compared to the toughness.
The results show that the crack is initiated in the radial direction. And the size
of the crack has a great influence on the energy of rupture and consequently on the
ultimate pressure. There is a good agreement between experimental and numerical
results.
Keywords: J integral, burst test, FEM model, crack propagation, HDPE.
1 Introduction
High density polyethylene (HDPE) pipe provides the lowest life cycle cost when
compared to other systems due to significantly reduced or no leakage, increased
billable dollars, water conservation, fewer new water-treatment plants, reduced
maintenance crews, reduced seasonal water-main breaks, and its viscoelastic
behavior makes it a good resistant to water hammer (CSIRO 2008).

392 M.A. Guidara et al.
In the present study, two part investigation will be made for HDPE pipes, with
superficial defect, using the well-known approach “J-integral” of the fracture
mechanics (Rice 1968).
In the first part, experimental tests will be made to determine mechanical
behavior (stress-strain curve) and failure analysis (JIc) of high density
polyethylene.
In the next part, FEM cracked tube model was performed on ABAQUS for
calculating J-integral around the crack type. Placed into the J integral curve JIc
allow to know the internal pressure which leads to the crack initiation.
To validate FEM Model, experimental burst tests are performed on cracked
HDPE pipes. The measurement results will be presented as a curve of pressure
versus volume of the added water.
2 Mechanical Behavior and Failure Analysis of High Density

Polyethylene
2.1 Tensile Testing

Tensile test were conducted on a standard traction-compression machine piloted in
displacement mode using an extensimeter to measure the deformation of the
specimen were carried out at three different test strain rates ( ε ) of 10 , 5.10 and
-3 -3
-2 -1
10 s . Stress- strain curves for each strain rate are represented in figure1.
30
25 V=0.001(1/s)
V=0.005(1/s)
20 V=0.01(1/s)
Stress (MPa)
15
10
0
0 100 200 300 400 500 600 700
Strain (%)
Fig. 1 Stress-strain curves for the HDPE

Burst Test and J-Integral Crack Growth Criterion 393
2.2 Determination of Fracture Toughness JIC

Several methods have been developed specifically for determining the fracture
toughness of polymeric materials.
ASTM D 6068 describes a method for measuring J-R curves (a measure of
elastic-plastic fracture toughness) for polymer specimens that are not large enough
to experience conditions of plane strain during loading.
Standard compact tension (CT) specimen according to the recommendations of
ASTM designation D6068 is used (figure 2). All tests were carried out at room
temperature using partial unloading compliance method on a servo-hydraulic
testing machine. The tests followed the guide-lines of the ASTM D 6068 standard
for the fracture toughness determination.
±0.2
W = 41.7 width
±0.2
L = 52.5 overall length (1,25 w ± 0,01 w)
±0.2
d1 = 50 transverse width (1,2 w ± 0,01 w)
R = 5+0.1 Radius (0,125 w ± 0,005 w)
b = 8+0.1 Thickness (0,4 w < b < 0,6 w or b < w/2)
±0.1
a = 18 crack length (0,45 w < a < 0,55 w)
±0.1
d2 = 11.5 distance between the center of two holes, (0,55 w ± 0,005 w)
balanced/Plans crack ± 0.005w
Fig. 2 CT specimen
The crack growth was followed by an optical microscope on specimen cracked

surfaces to calculate the average physical crack extension Δap.
The energy required to extend the crack, U, is used to calculate J. The total
energy, UT, determined from the area under the load versus load-point
displacement curve obtained for each specimen is the sum of U and Ui, the
indentation energy.
And J is calculated according to the following relation:
ηU
J= (1)
b(W − a0 )
Figure 3 shows a representative J-Δa curve obtained from the 8 mm thick CT

specimen.
The intersection point between the regression line and 0.2 mm offset line gives
a candidate value JQ which becomes JIC provided that the validity requirements are
satisfied. In the case of HDPE PE100, JIc=7.69 KJ/m².
100
90
80
70
60
data points
J (KJ/m 2 )
50 regression line
40
30
20
10
J0.2=7.69
0
0 0,2 0,4 0,6 0,8 1 1,2 1,4 1,6 1,8 2 2,2 2,4 2,6 2,8 3 3,2 3,4 3,6
Cra ck extention Δa (mm)
Fig. 3 Fracture Resistance vs Crack Growth
3 Burst Tests
3.1 Experimental Study

After the experimental determination of HDPE’s fracture toughness, the
experimental investigation on this material was carried on by burst testing. The
specimens for these tests were a 400 mm length of pipe section with different pre-
cracks depths (figure 4). These pre-cracks are machined by disc cutter that has a
plate thickness of 0.5 mm and an outer diameter of 80 mm. The depth of the pre-
crack (a) is due to the disc cutters penetration depth into the pipe wall.
a (pre-crack depth in mm)

2 4 6 8
Fig. 4 Pre-crack geometry
All tests were performed on a sample size of 3 specimens per design type of the
same age and service history (new pipes). And these tests were made according to
the following instructions
- Setting up the specimen and fill it with water until completely filled and
bled from air.
- All specimens have been loaded with a continuous pressure rate of 10 bar
per minute until failure (figures 5a and 5b).
60
50
Pressure (bar)
40
30
20
10
0
0 1 2 3 4 5 6 7
Time (min)
(a) (b)
Fig. 5 Burt test
Unlike metallic structure, the use of strain gauge on HDPE pipes to obtain the
strain-internal pressure curve needs a complicated and delicate pre-treatment and
results will not accurate enough. So a more reliable solution should be adopted
The measurement results will be presented as a curve of pressure versus volume of
water added.
Comparison between experimental data and numerical simulation results will
be made to validate the FEM model.
3.2 Finite Element Modeling

Finite element simulation was performed using the code Warp3D (non-mechanical
linear Out). The numerical simulation of crack propagation in three dimensions is
known tricky for reasons related to the mesh near the pre-crack, especially when it
comes to polymer materials.
Figure 6 shows a three dimensional representation (3-D FE) of the finite
element mesh. Taking into account of the symmetry of loading and geometry, only
a quarter of the model is studied in order to reduce the computation time. The
mesh was created using HyperMesh in three dimensions with quadratic elements
C3D20. It has been sufficiently refined in the vicinity of crack with special
elements to increase the accuracy in local area and predict correctly the fracture
energy into the crack. Overall, the FE model contains 150000 elements and
600000 nodes. The calculations are made with the geometrically nonlinear
ABAQUS software (large displacements) by taking into account of the variation
of the strain rate.
Fig. 6 3D Crack Mesh
Thermoplastic materials such as PE 100 have a viscoelastic behavior and a high

degree of ductility. Normally, the fracture behavior (the initiation and propagation
of cracks) must be analyzed using the concept of viscoelastic-plastic fracture
mechanics (Brostow et al. 1991) and (Favier et al. 2002). However, it has been
shown (Benhamena et al. 2010, 2011) that elastic–plastic approaches can
approximate with acceptable accuracy the critical J integral of the polyethylene.
3.3 Results and Discussions

The figure 7 shows the variation of the internal pressure as a function of volume
of water added to each specimen during burst test. The ultimate pressure is greatly
affected by the size of the initial crack.
60
Without crack
a=2mm
a=4mm
50
a=6mm
a=8mm
40
Pressure (bar)
30
20
10
0
0,0 0,2 0,4 0,6 0,8 1,0 1,2 1,4
Added volume(l)
Fig. 7 Internal pressure vs Added volume (Burst test)

58
56
55,44
54 53,8
52,46
Burst pressure (bar)
52
50
48
46,8
46
44
43,1
42
40
0 1 2 3 4 5 6 7 8 9
pre-crack depth (mm)
Fig. 8 Burst pressure variation with pre-crack depth
Burst pressures for different cracked pipes are deduced from figure 7 and
represented on figure 8 as a function of the pre-crack’s depth. These results show
a sharp drop in the burst pressure when the depth of crack is more than 4mm.
The figure 9 shows the evolution of the fracture energy in different contours in
the normal plane to the crack front. They converge to the same value.
The curve of the fracture energy versus pressure and the value of the critical
2
failure energy (JIC=7.69KJ/m ) allow to determine the internal pressure (Pa=46bar)
which lead to the crack initiation (figure 9). It is less than the ultimate pressure
(Pu=52.5 bar) (figure 10).
14
contour2
contour3
12
contour4
contour5
fracture resistance J (KJ/m2)
10
contour6
contour7
8
JIC=7.69
6
-2
0 10 20 30 40 Pa=46 50 60
Pressure (bar)
Fig. 9 Variation of the fracture toughness J versus Pressure (a=4mm)

In the numerical application, the behavior of PE 100 was considered elastic-

plastic and the damage criterion is not studied. This explains the error
found between the experimental and numerical calculations (figure 10). For
any pressure, the volume of water added during the burst test is greater than that
found experimentally. It can also be caused by experimental errors (measurement,
leak ...).
60
Pu=52.2
50
Pa=46
Numerical test
40
Experimental test
Pressure (bar)
30
20
10
0
0 0,05 0,1 0,15 0,2 0,25 0,3 0,35 0,4
Added Volume(l)
Fig. 10 Pressure vs Added volume (a=4mm)
In the case of a cracked pipe with a 4 mm crack’s depth, the ultimate pressure
and maximum admissible pressure are calculated from the burst test using fracture
toughness JIC experimentally obtained. The result is quite logic since Pa < Pu. This
proves the accuracy of the value of JIC.
Numerical calculation of admissible pressure for the other cracked pipe will be
made soon.
4 Conclusion and Perspectives
In this study, the mechanical propriety of high density polyethylene (PE100) pipe
such as behavior law and fracture toughness JIC were experimentally obtained.
These parameters were used to study the effect of the crack’s depth on the ultimate
pressure of a PE100 pipe.
Results show that the crack is initiated in the radial direction. And the size of
the crack has a great influence on the energy of rupture and consequently on the
ultimate pressure. There is a good agreement between experimental and numerical
results.
In the upcoming study some improvement will be made like:
- Use a concept of visco elastic-plastic fracture mechanics to analyze the
fraction behavior
- Introduce a damage criterion in the numerical simulation
- Perform a dynamic testing to study the effect of transient flow in cracked
pipes
References
CSIRO, Life Cycle Analysis of Water Networks. presented at Plastics Pipe XIV, Budapest
(2008)
Rice, J.R.: A Path Independent Integral and the Approximate Analysis of Strain
Concentration by Notches and Cracks. Journal of Applied Mechanics 35, 379–386
(1968)
ASTM D 6068 Standard Test Method for Determining J-R Curves of Plastic Materials
WARP3D. Release 15.0 manual, civil engineering, reports no UILU-ENG-95-2012.
Urbana, University of Illinois (2004)
Benhamena, A., Bouiadjra, B., Amrouche, A., Mesmacque, G., Benseddiq, N., Benguediab,
M.: Three finite element analysis of semi-elliptical crack in high density poly-ethylene
pipe subjected to internal pressure. Materials and Design 31, 3038–3043 (2010)
Benhamena, A., Aminallah, L., Bouiadjra, B., Benguediab, M., Amrouche, A., Benseddiq,
N.: J integral solution for semi-elliptical surface crack in high density poly-ethylene pipe
under bending. Materials and Design 32, 2561–2569 (2011)
Favier, V., Giroud, T., Strijko, E., Hiver, J.M., G’Sell, C., Hellinckx, S., et al.: Slow crack
propagation in polyethylene under fatigue at controlled stress intensity. Polymer 43,
1375–1382 (2002)
Brostow, W., Fleissner, M., Müller, W.: Slow crack propagation in polyethylene:
determination and prediction. Polymer 3, 419–425 (1999)
Solving the Three-Dimensional Time-Harmonic
Maxwell Equations by Discontinuous Galerkin
Methods Coupled to an Integral Representation
Nabil Gmati1, Stéphane Lanteri2, and Anis Mohamed3

1
National Engineering School of Tunis, ENIT-LAMSIN BP 37, 1002 Tunis,
LR 95–ES–20 Tunisia
nabil.gmati@enit.rnu.tn
2
INRIA Sophia Antipolis Méditerranée 2004 route des lucioles-BP 93 06902,
Sophia Antipolis Nice cedex France
Stephane.Lanteri@inria.fr
3
National Engineering School of Tunis, ENIT-LAMSIN BP 37, 1002 Tunis,
LR 95–ES–20 Tunisia
midani.anis@gmail.com
Abstract. In this paper, we present a numerical study of three-dimensional time-

harmonic Maxwell equations. We use a finite element discontinuous Galerkin
method coupled with an integral representation. This study was completed by
several numerical examples to test the efficiency of the proposed approach. The
numerical simulation was down by an iterative solver implemented in FORTRAN.
Keywords: time-harmonic, maxwell equations, integral representation, disconti-

nuous galerkin method.
1 Introduction
Electromagnetic phenomena are generally described by the electric and magnetic
fields E and H which are related by the following Maxwell equations:
-ε∂ t E + curl H = 0 (1)
-μ∂ t H + curl E = 0 (2)
Where ε and μ are the complex-valued relative dielectric permittivity and the
relative magnetic permeability, respectively. We are interested in time-harmonic
solutions which satisfy the following equations:
curl E - iωμ H =0 (3)
curl H - iωε E =0 (4)
These two equations are posed in an unlimited domain. Or approach consists of
limiting the obstacle D by a fictitious boundary Γ a where we impose the
absorbent of Silver-Muller condition. On the boundary Γ m of D, we take the
perfect conductor condition.

402 N. Gmati, S. Lanteri, and A. Mohamed
Fig. 1 Diffraction of an electromagnetic wave in the presence of an obstacle D
2 Formulation of the Problem
Our problem is the following:

Find E,H such that:
curl E - iωμ H =0 In Ω (5)
curl H - iωε E =0 In Ω (6)
E ∧ n = − E inc ∧ n On Γ m (7)
E ∧ n − Z ( H ∧ n ) ∧ n = ℜ( E ) ∧ n − Z ( ℜ( H ) ∧ n ) ∧ n On Γa (8)
Where:
• ω : The angular frequency of the problem, n is a normal vector to the
boundary Γ a .
μ
• Z= .
ε
• ℜ( E ) , ℜ( H ) are the expression of the electric and magnetic fields E and
H , respectively, on Γ a . They are given by the following integral
representation:
1
ℜ( E ) = curl x ∫ n( y ) ∧ E ( y )G ( x, y )∂σ y − curlx curlx ∫ n( y ) ∧ H ( y )G ( x, y )∂σ y
Γa iωμ Γa
1
ℜ( H ) = curlx ∫ n( y ) ∧ H ( y )G ( x, y )∂σ y − curlx curlx ∫ n( y ) ∧ E ( y )G ( x, y )∂σ y
Γa iωε Γa
Let ( e1 , e2 , e3 ) be the canonical basis of \ . We consider the matrix Gl de-

3
fined as follows:
Solving the Three-Dimensional Time-Harmonic Maxwell Equations 403
⎛ 03×3 N el ⎞
Gl = ⎜ t , for l ∈ {1, 2,3} ,
⎜ N e 03×3 ⎟⎟
⎝ l ⎠
⎛ v1 ⎞
⎜ ⎟
where for a given vector v = ⎜ v2 ⎟ , the matrix N is:
⎜v ⎟
⎝ 3⎠
⎛ 0 v3 −v2 ⎞
⎜ ⎟
N v = ⎜ −v3 0 v1 ⎟ .
⎜v ⎟
⎝ 2 −v1 0 ⎠
⎛E ⎞
It is easy to verify that Gn = G1n1 + G2 n2 + G3 n3 . Let W = ⎜ ⎟ , therefore the
⎝H ⎠
system of equations (5), (6), (7) and (8) can be rewritten in the following
conservative form:
iωQW + G1∂ xW + G2 ∂ yW + G3 ∂ zW = 0 In Ω (9)

( M Γm − Gn )(W − W inc ) = 0 On Γ m (10)
( M Γa − Gn )(W − ℜ(W )) = 0 On Γ a (11)
Where:
⎛ ℜ( E ) ⎞ ⎛ ε .I 3 03×3 ⎞ ⎛ 03×3 N n ⎞
• ℜ(W ) = ⎜ ⎟, Q =⎜ ⎟ , and M Γa = ⎜ t ⎟.
⎝ ℜ( H ) ⎠ ⎝ 03×3 μ .I 3 ⎠ ⎝ N n 03×3 ⎠
• M Γa = Gn = Gn+ − Gn− Where Gn+ and Gn− are the positive and negative
parts of Gn , respectively.
3 Discretization
We discretize the domain Ω into N tetrahedral cells such that:
N
 Ωh =
K∈
∪τ h
K
 P mi [τ i ] = {The space of polynomials of degree ≤ mi defined on τ i }.

 {
Vh = W ∈ ( L2 (Ω))3 / W\τ i ∈ P mi [τ i ]
3
}
Wh = ⎡⎣ E1h , E2h , E3h , H1h , H 2h , H 3h ⎤⎦ the projection of W on ( P mi [τ i ])6
t

Also we note:
i , Γm =
Γ0 = ∪ K ∩K
i ∈τ
K ,K
∪K ∩ Γ and Γ a =
K ∈τ h
m ∪K ∩Γ
K ∈τ h
a
h
If we multiply the equation (5) by a test function V and we integrate on Ω h ,

we then get:
∫
Ωh
(iωQWh )t .Vdx + ∫ (∇.( F (W ))h )t .Vdx = 0 ,
Ωh
where F (W ) = (G1W , G2W , G3W ) is a linear application from \6 to

(\ 6
×\ ×\ 6 6
)
Using Green formula we obtain:
∑τ ( ∫ ( ( F (W ) ) ) .∇Vdx + ∫ ( F (W ) .n ) .V ∂σ ) = 0
t
∫Ωh
(iωQWh )t .Vdx −
K∈ h
K h ∂K h
(12)
Using the same technics adopted by Ern and Guermond [1,2], and by adding the
terms of the integral representation , we obtain the following formulation:
∀V ∈ Vh × Vh and K ∈ τ h ; Find Wh ∈ Vh × Vh such that:
∑τ ∫ ( ( F (W )) ) .∇Vdx
t
∫ Ωh
(iωQWh )t .Vdx −
K∈ h
K h
+∑ ∫ ⎡⎢⎣( S .aW b) .cdeV fgh − (G .aW b) .{V }⎤⎥⎦ ∂σ

t t
F F h nF h
F ∈Γ0
⎡ ⎛1 ⎞
t
⎛1 ⎞
t
⎤
( )
+ ∑ ⎢ ∫ ⎜ M F , K − I FK GnF Wh ⎟ .V ∂σ − ∫ ⎜ M F , K − I FK GnF ℜ (W ) ⎟ .V ∂σ ⎥ ( )
⎢ ⎝2
F ∈Γa ⎣
F
⎠ F
⎝2 ⎠ ⎦⎥
t t
⎛1 ⎞ ⎛1 ⎞
m
F 2
( ⎠ m ⎝
F 2
)
+ ∑ ∫ ⎜ M F , K − I FK GnF Wh ⎟ .V ∂σ = ∑ ∫ ⎜ M F , K − I FK GnF Whinc ⎟ .V ∂σ
⎝ ⎠
( )
F ∈Γ F ∈Γ
Where:
ℜ (W )( x ) = ∫ m K ( x, y ) W ( y )dy such that K : \ 6 × \ 6 → M 6 ( ^ ) is a Green
Γ
kernel. We also define the jump and average of a vector V ∈ Vh × Vh ,

i:
respectively, on the face F shared with two elements K and K
aV b = I FK .V\
1
2 K
K
+ I F Ki .V\ Ki and {V } =
V\ + V\ Ki such as I FK represents the ( )
incidence matrix between oriented faces and elements and they are given by:
⎧1 if F ∈ K and their orientations match,

⎪
I FK = ⎨-1 if F ∈ K and their orientations do not match,
⎪0 if F ∉ K .
⎩
We mention here that for the terms on the boundary, we used two classical
numerical flux.
3.1 Centered Flux

In this case, for all the faces S F = 0 and for the boundary faces, we have:
⎧ ⎛ 03×3 N nF ⎞
⎪⎪ I FK ⎜ ⎟ if F ∈ Γ
m
= ⎨ ⎜⎝ − N nF 03×3 ⎟⎠
t
M F ,K
⎪
⎪⎩ GnF if F ∈ Γ
a
3.2 An Upwind Flux
In this case:
⎧ ⎛η F N nF N nt F I FK N nF ⎞
⎛ α N nF N ⎞ ⎪⎪ I FK ⎜⎜ I N t ⎟ if F ∈ Γ m
E t
03×3 ⎟
SF = ⎜ ⎟ , M F ,K
F nF
=⎨ ⎝ FK nF 03×3 ⎠
⎜ α F N nF N nF ⎟⎠
H t
⎝ 03×3 ⎪
⎪⎩ GnF if F ∈ Γ
a
1
For homogeneous media, η F = α FE = α FH =
2
4 Linear System
The problem is reduced to the linear algebraic system: ( A − C ) .X = b such that

A and C are square size matrices N = 6 × di × N c and where:
• d i : number of degrees of freedom.
• N c : number of cells.
 A is a block defined matrix of size 6.d i × 6.d i such that:
For i, j =1,… ,N c ; A ( i , j ) ∈ M 6.di ( ^ )
A ( i , i ) = Di1 − Di2 + DiΓ × δ ij + DiΓ × δ Γm + DiΓ × δ Γa
0 m a
•
i i
• A ( j , i ) = B × δ ij
Γ0
ij
δij : Kronecker symbol

 C is a block defined matrix of size 6.di × 6.d j such that:
For i, j =1,… ,N c ; C ( i , j ) ∈ M 6.di ×6.d j (  )
• C ( i , j) = −Cij × δ Γa × δ Γm
i j
Γ0 Γm a 0
Where D , D , D 1
i i
2
i , D i and DiΓ are diagonal blocks, BijΓ defines the
extra-diagonal block.
Cij The Coupling matrix defined as follows:
Cij =
1
2
( I3 ⊗ Ψ ia ) .Ki ij .( I 3 ⊗ Ψ im )
5 Numerical Results
The numerical implementation, we used the FORTRAN code developed by Hugo

Fol & Stéphane Lanteri with a simple modification based on the Bi-Conjugate
Gradient method.
We have illustrated the numerical solutions obtained with a transparent
provided CT (Discontinuous Galerkin coupled to an integral representation) and
the exact solution in the following figures: (in these two cases, the inner radius is
equal to 1)
We choose in these tests a single layer enter the boundary Γ m and the
fictitious boundary Γ a .
 Test No. 1
Table 1 The first test data
Parameters Value
Outer radius 1.18
Frequency(Ghz) 0.3
H max 0.30
Number of cells 3119
Fig. 2 The contour lines of the computed solution

Fig. 3 The contour lines of the Exact Solution
 Test No. 2
Table 2 The second test data
Parameters Value
Outer radius 1.1
Frequency(Ghz) 0.3
H max 0.16
Number of cells 10512
Fig. 4 The contour lines of the computed solution

Fig. 5 The contour lines of the Exact Solution
According to figures we can see that this method leads to a solution which is
very similar to the exact one.
Acknowledgements. The authors also gratefully acknowledge the helpful comments and
suggestions of the reviewers, which have improved the presentation.
References
[1] Ern, A., Guermond, J.-L.: Discontinuous Galerkin methods for Friedrichs systems I.
General theory. SIAM J. Numer. Anal. 44(2) (2006)
[2] Ern, A., Guermond, J.-L.: Discontinuous Galerkin methods for Friedrichs systems II.
Second-order elliptic PDEâs. SIAM J. Numer. Anal. 44(6) (2006)
[3] El Bouajaji, M., Gmati, N., Lanteri, S., Salhi, J.: Coupling of an exact transparent
boundary condition with a DG method for the solution of the time-harmonic Maxwell
equations, vol. 95, pp. 237–247. Springer (2014)
[4] Ben Belgacem, F., Fournie, M., Gmati, N., Jelassi, F.: On the Schwartz algorithms for
the elliptic exterior boundary value problems. ESAIM. Math. Model. Numer. Anal. 39,
639–714 (2005)
[5] Colton, D., Kress, R.: Inverse acoustic and electromagnetic scattering theory. Springer
(June 1997)
Iterative Methods for Steady State Looped
Network Analysis
Zahreddine Hafsi, Sami Elaoud, Mohsen Akrout, and Ezzeddine Hadj-Taïeb
Laboratory of Applied Fluids Mechanics Process and Environment Engineering,

ENIS P.O. Box, W, Sfax, 3038, Tunisia
hafsi.zahreddine@gmail.com,
elaoudsa@yahoo.fr
Abstract. The aim of this paper is to numerically study the steady state of water
flows in looped networks. This study will be performed by the use of Hardy Cross
Method and Newton-Raphson algorithm. The comparison of the numerically
obtained results by these two methods to those obtained by the use the commercial
software Pipe Flow Expert has shown a good concordance between them. The
numerical analysis has shown that the convergence of Newton-Raphson method is
more rapid than that of the classic Hardy Cross Method.
Keywords: steady flow, pipe network, Newton-Raphson method, Hardy-Cross

method.
1 Introduction
Fluid transport from production sites to the exploitation sites is a major concern
for manufacturers. Many studies have shown that the most effective method is to
transport via apparent or underground pipelines. These pipelines, connected
together, form what called a "hydraulic network" that can be either branched or
looped or even adopting a hybrid topology. To analyze the steady state, hydraulic
networks flows are to be balanced. For a branched network, by knowing the flow
demand at each terminal, it’s easy to determine the flow rate at each conduit of the
network since the flow direction is evident.
The problem arises when we start facing a purely looped or hybrid network
(containing loops and branches). Several methods and techniques are used to
balance flows in these types of network. Between these methods, one can cite: the
Hardy Cross Method (HCM), Node-Loop Method, Modified Node Method and
Andrijašev Method. A comparative study between the previous techniques has
been carried out (Dejan 2011). In this study, the number of iterations needed to
achieve the network equilibrium is taken as criterion to put in evidence the
efficiency of these methods. Dejan proved that, for complex networks, Node Loop
Method is more efficient as it does not require complex numerical scheme for
algebraic addition of corrections in each of iterations.

410 Z. Hafsi et al.
These techniques, and as are iterative methods, require an initial guess of flow
rates through pipelines. This guess has to satisfy the first Kirchhoff’s law: nodes
law. In many studies the initial guess is introduced manually.
In this paper, Hardy Cross and Newton-Raphson methods are used to solve a
water looped network. A comparative study between the two methods in terms of
efficiency, elapsed time, and precision will be conducted. In this study, the initial
guess imposed by these iterative methods will be generated automatically.
2 Mathematical Formulations
In an electric network, Kirchhoff’s second law reposes on relationship between

nodal variable (electric potential ) and traversing variable (current ) given by
Ohm’s law ( is the electrical resistance). In a hydraulic circuit the
traversing variable is the flow-rate and the nodal variable is the head . Basing
in theory of analogy, a relationship between and is written as follows:
(1)
where is the hydraulic resistance.

The values to be assigned to and depend on the equation being used.
In literature, for a single-phase fluid flow, several relations in the above form
can be found. A common used relation is that of Darcy-Weisbach (Giles 1962;
Fox 1992; Walski 2001) given by:
(2)
where is the pipe length, is its diameter, is the gravitational acceleration and
is the friction factor.
The friction factor is given by Colebrook-White equation (Giles 1962; Fox
1992; Shames 1989; White 1994):
.
2 (3)
√ . √
where e is the internal roughness of the pipe and is the Reynolds number given
by the following expression:
(4)
where is the kinematic viscosity of the fluid.

Equation (3) is an implicit form in term of which can be solved numerically
by the use of an iterative process.
Iterative Methods for Steady State Looped Network Analysis 411
3 Hardy Cross Method
The Hardy Cross method is used to balance flows through looped networks. In
this method, an initial guess of flow per conduits has to be assigned, and in that
way chosen values are to be used for first iteration. Then, an iterative process is
conducted with correction of the flow-rate for each loop. The following
assumptions must be satisfied for Hardy Cross calculation:
- Algebraic sum of flows per each node must be exactly equal to zero (first
Kirchhoff”s law: continuity of flow).
- Algebraic sum of head losses per each loop must be approximately zero at the
end of iterative procedure (second Kirchhoff’s law: continuity of potential).
The loop flow correction is given by the following equation:
∑
∑ ⁄
(5)
where ∑ is the algebraic head loss sum of all pipes in the considered loop, is
the flow-rate through each pipe of the loop and is depending on the used
formula for head loss expression.
In general, the iterative correction process is continued until the integrated head
loss around all loops in the network is reduced to a specified limiting value .
4 Newton Raphson Method
The Newton Raphson method is based on the resolution of non linear equation
0 by successive iterations using the following expression:
; 0,1, … (6)
where is an initial guess of the solution.

In each step we have to verify the convergence criterion of the algorithm, for
example | | , where is a tolerance value.
To solve non linear system of equations 0 written under matrix form,
same principle can be applied. In that case equation 6 is written:
; 0,1, … (7)
5 Applications and Results
The following assumptions will be considered: the variation of the potential

energy is neglected; one-dimensional flow is admitted; the pipe is supposed to be
rigid (its section ) and the head loss through each pipe of the
network is calculated using Darcy-Weisbach formula (equation 2).
412 Z. Hafsi et al.
5.1 Branched Network Study

In a branched network having pipes, the number of nodes is 1. Only
nodes law is needed to analyze such network. Figure 1 and Table 1 show
respectively the case study and the input data needed to simulate the network.
Fig. 1 Example of a branched network
Table 1 Data describing the network topology
Pipe number 1 2 3 4 5 6
Upstream node 1 2 3 3 2 2
Downstream node 2 3 4 5 6 7
For a branched network we have to solve a linear system formed by nodes

equations. These equations can be written under the following matrix form:
(8)
where is commonly known as incidence matrix, represents the vector of the

unknowns flow-rates in the conduits and is the consumptions (or demands)
vector.
The terms of the matrix are determined by respecting the following rules:
1
1 (9)
0
The dimensions of the matrix are . By respecting equation (9) and table
1, the obtained matrix is:
1 0 0 0 0 0
1 1 0 0 1 1
0 1 1 1 0 0
0 0 1 0 0 0 (10)
0 0 0 1 0 0
0 0 0 0 1 0
0 0 0 0 0 1
Vector C is written as follows:
300 ; 0 ; 0 ; 70 ; 150 ; 0 ; 80 (11)
In this vector, the minus sign preceding the first flow-rate indicates that the first
node is a supply node (the node that feeds the network). In a hydraulic network,
we can encounter one or more supply nodes. Respecting the mass continuity
principle, the sum of vector elements has to be equal to zero.
To solve the system of equations (8), matrix must be non singular. However,
its first row can be written
∑ (12)
This equality has a physical signification since the first row concerns the first node
(the supply node) and the other rows concern consumption nodes. To ensure that
matrix is non singular, its first row is removed and subsequently the first
element of the vector . By taking into account the previous considerations, the
resolution of system (8) permits to determine the vector :
⁄ 300 ; 220 ; 70 ; 150 ; 0 ; 80 (13)
All the described steps to solve the branched network equations, in this
section, will be integrated to balance the looped network flows.
5.2 Looped Network Study

For a looped network containing pipes and nodes, the flow in each pipe
should be determined. By applying Kirchhoff’s laws, independent equations are
obtained. Since we have nodes, nodes equations can be written. Nevertheless,
as mentioned in section 4.1, only 1 are available. It remains to find
1 other independent equations that correspond to the loops equations.
As an application, we consider a looped network composed by 12 pipes and 9
nodes as shown in figure 2.
Tables 2 and 3 show respectively the geometric description of the pipes and the
fluid properties of the looped network shown in figure 2. Table 4 resumes the
studied looped network topology.
414 Z. Hafsi et al.
Fig. 2 Example of a looped network Fig. 3 Branched generated network
Table 2 Tubular pipes material proper- Table 4 Studied looped network characteristics
ties: steel Sch 80 pipe (ANSI standards)
Designation Value Unit Pipe Upstream Downstream Length
number node node (m)
Internal roughness 0.0459 1 2 900
1
Nominal Pipe Size 500 2 2 5 1200
3 6 5 900
Internal diameter 455.625
4 1 6 1200
5 2 3 900
Table 3 Fluid properties: water 6 3 4 1200
7 5 4 900
Designation Value Unit
8 5 8 1200
Temperature 25° ° 9
9 8 900
Density 997 / 10 6 9 1200
Kinematic 11 4 7 1200
0.89 . .
viscosity 12 8 7 900
5.2.1 Network Balancing by the Use of Hardy Cross Method
The initial guess required by the HCM will be generated automatically. As a first
step, a branched network is derived from the initial looped network. As mentioned
in section 5.1 the number of pipes in a branched network is equal to 1. As
consequence, the number of pipes to be removed from a looped network
containing pipes and nodes is : 1 one from each loop. In our case
study, the number of pipes to be removed is equal to 4. Figure 3 shows the
obtained branched network.
Then, an incidence matrix 1 of the eliminated pipes and 2 of the branched
network are built respecting the principles of equation (9) and a vector
solution for the eliminated pipes is generated. This vector can be introduced
randomly or, simply, chosen as null vector.
For the flow-rate vector of the branched network, we have:
1 2 (14)
1 2 1 (15)
By combining and , a vector of flow-rates through the looped network

respecting the nodes law is obtained. This vector is the starting guess needed to
apply the HCM principle developed in section 3.
To apply the loops law properly, A Direction Matrix is introduced. For
each pipe from each loop the initial flow-rate is multiplied by the correspondent
element in (1 if the supposed flow direction is clockwise or 1 if it’s
anticlockwise). For this case study, the matrix is written as follows:
1 1 1 1
1 1 1 1
(16)
1 1 1 1
1 1 1 1
For pipes that belong to two loops, the correction factor to apply is where
is the correction factor of the concerned loop and is that of the other one.
In each iteration of HCM algorithm, the friction factor of each pipe has to be
evaluated by solving equation 3 using an iterative method.
Respecting the above steps, the following matrix of flow-rates through each
pipe in each loop is obtained:
208.3183 97.3525 83.1999 191.6817
/ 110.9658 70.9658 78.9862 97.3525 (17)
83.1999 81.5662 28.4818 108.4818
78.9862 49.9520 30.0480 81.5662
It is noticed that the elements of the flow-rates matrix have the same signs of their
correspondents in the direction matrix , so, the real flow direction through
each pipe is the same we had supposed. In case we have a difference of sign, the
considered flow is actually in the opposite direction than we supposed.
200
Pipe 2 belonging to the loop 1
Pipe 2 belonging to the loop 2
150
100
50
Flow -rate (l/s)
-50
-100
-150
-200
2 4 6 8 10 12 14 16 18 20 22
Iteration number
Fig. 4 Variation of flow-rate through pipe 2 Fig. 5 Pipe Flow Expert simulation result
416 Z. Hafsi et al.
Figure 4 illustrates the correction of flow-rate through pipe 2 from the initial
supposed value to the real one. This figure shows sign respect while applying the
correction procedure, so that the flow-rate value of a common pipe between two
loops is considered positive through one loop negative through the other.
5.2.2 Network Balancing by the Use of Newton-Raphson Method
Respecting the Newton-Raphson Method’s principle mentioned in section 4, the

studied looped network is analyzed by adopting a matrix representation.
The system of equations to solve is formed by independent loops and nodes
equations taking in consideration the matrix. The flow rate vector Q is the
unknown to evaluate. Starting from an initial guess and by fixing a predefined
value for , the solution is evaluated in each iteration until verifying the
convergence criterion | | :
As in HCM, in each iteration, the friction factor λ of each pipe has to be
calculated using an iterative process (Newton Raphson principle was used to
evaluate in both algorithms) and the initial guess can be assigned using a
generated branched network as detailed in section 5.2.1. The final solution is:
⁄ 208; 97; 83 ; 191 ; 110 ; 70 ; 78; 81 ; 28; 108; 49; 30 (18)
The obtained vector contains no negative values. So, the real flow direction is
the same initially supposed (from the upstream to the downstream node).
5.3 Comparison and Validation of Results

It’s important to mention that when conducting a comparative study between
iterative methods, we have to ensure that the same convergence criterion is
Table 5 Obtained flow-rates values / by the used methods
Pipe number Hardy Cross Newton-Raphson Pipe Flow Expert

1 208.3183 208.3191 208.325
2 97.3525 97.3523 97.375
3 83.1999 83.2003 83.671
4 191.6817 191.6809 191.675
5 110.9658 110.9669 110.950
6 70.9658 70.9669 70.950
7 78.9862 78.9866 79.308
8 81.5662 81.5659 81.738
9 28.4818 28.4806 28.004
10 108.4818 108.4806 108.004
11 49.9520 49.9535 50.258
12 30.0480 30.0465 29.742
considered. In our case, and as mentioned above for both HCM and Newton
Raphson Method, the solution is reached when the difference between two
successive obtained results ( and ) is less than the chosen limiting value
for HCM and for Newton Raphson Method.
To evaluate the efficiency of the used methods of analysis, a simulation of the
studied network was performed by the use of the commercial software Pipe Flow
Expert. The obtained results by that software are shown in figure 5. Table 5 shows
the flow distribution in each conduit of the network obtained by the three used
methods under the same limiting value ( =10 ).
The comparison of Hardy Cross and Newton-Raphson methods results show a
good concordance. Added to that, our numerical results compare very favorably
with those obtained by “Pipe Flow Expert” software.
5.4 Comparison of Resolution Methods in Term of Efficiency

Figure 6 illustrates the convergence of HCM and Newton-Raphson Methods.
400 220
Hardy Cross Method Pipe 4 Newton Raphson Method
200
350 Pipe 6
Pipe 9
180 Pipe 4
300 Pipe 12
Pipe 6
160 Pipe 9
Flow-rate (l/s)
250
Flow-rate (l/s)
Pipe 12
140
200 120
150 100
80
100
60
50
40
0
10 15 20 25 30 35 40 45 50 55 60 20
6 8 10 12 14 16 18 20 22
Time (s)
Time (s)
Fig. 6 Convergence of Hardy Cross and Newton-Raphson Methods
For 4 pipes of the network, this figure shows the variations of flow-rate during
the correction process from the initial guess to the final result. The correction
phase has required more than 53 seconds to achieve the actual solution for HCM
and just 14 seconds for Newton-Raphson Method. Added to that the total elapsed
time for the HCM is about 60 seconds while it’s about 21 seconds for the Newton-
Raphson Method. This difference in elapsed time is explained by the quick
correction phase for the second method. Indeed the flow rate through each pipe is
evaluated once in each iteration, while in the HCM method, we have to correct the
flow-rate value through each loop that means pipes belonging to two loops have to
be calculated twice (in clockwise and anticlockwise directions).
In term of iterations number required to achieve final result, figure 7 shows that
Newton-Raphson Method converges after about 9 iterations and reaches the
desired solution while HCM requires about 21 iterations to find that solution.
Both used methods are efficient to analyze a looped network but in term of
rapidity convergence and simplicity to implement the algorithm, Newton-Raphson
Method is more adequate mainly that the elapsed time will be so important to
reduce when analyzing more complex networks.
418 Z. Hafsi et al.
180 Hardy Cross Method

Newton Raphson Method
160
140
120
Flow-rate (l/s)
100
80 Pipe 2
Final flow- rate: 97.35 l/s
60
40
20
0
2 4 6 8 10 12 14 16 18 20 22
Iteration number
Fig. 7 Comparison of convergence in term of iterations number
6 Conclusion
A numerical study of looped water networks is presented in this paper. The

network flow equilibrium is performed by the use of Hardy Cross and Newton-
Raphson methods.
The obtained results of each method are compared to those of the other method
and confirmed by a “Pipe Flow Expert” simulation.
This study also shows that the introduction of Newton-Raphson Method to
analyze such problems is very useful mainly with the evolution of computing tools
as that method fit non linear problems and even in convergence efficiency it’s
very favorable comparing to traditional methods like HCM.
References
Cross, H.: Analysis of flow in networks of conduits or conductors (1936)
Brkic, D.: Iterative Methods for Looped Network Pipeline Calculation. Springer
Science+Business Media B.V. (2011)
Giles, R.V.: Theory and Problems of Fluid Mechanics and Hydraulics, p. 274. Schaum
Publishing Co., New York (1962)
Fox, R.W., McDonald, A.T.: Introduction to Fluid Mechanics. Wiley, New York (1992)
Shames, I.H.: Mechanics of Fluids, p. 692. McGraw-Hill Book Co., Singapore (1989)
Walski, T.M., Chase, D.V., Savic, D.A.: Water Distribution Modeling. Haestad Press,
Waterbury (2001)
White, F.M.: Fluid Mechanics, p. 736. McGraw-Hill Book Co., Singapore (1994)
Investigation and Modeling of Surface
Roughness of Hard Turned AISI 52100 Steel:
Tool Vibration Consideration
Ikhlas Meddour, Mohamed Athmane Yallese, and Hamdi Aouici
Mechanics and Structures Research Laboratory (LMS), May 8th 1945 University,
P.O. Box 401, Guelma 24000, Algeria
meddour26@yahoo.fr
Abstract. Since the great interest of the achieved parts quality, this experimental
study focuses on the modeling and the investigation of the surface roughness of
hard turned AISI 52100 steel. The tool vibration was taking into consideration
beside cutting speed, depth of cut, feed rate and tool nose radius. The response
surface methodology (RSM) was employed for modeling process. Models
reliability was established by conducting confirmation tests. Slight divergence
between the experimental and their corresponding predicted values were observed.
The significance of the different factors on surface roughness was established by
applying analysis of variance (ANOVA). The results revealed that the best surface
roughness is obtained by using small feed rate and large nose radius. Furthermore,
a correlation between the surface roughness and tool vibrations was established.
Keywords: Hard turning, ANOVA, RMS, Surface roughness, tool vibration.
1 Introduction
A numerous researches were carried out to study the effect of cutting conditions
on surface roughness. (Yallese et al. 2005) examined the effect of cutting
parameters, i.e. cutting speed, depth of cut, feed rate on surface roughness in hard
machining of X200Cr12 steel. The results showed that large feed rate deteriorates
surface roughness while depth of cut did not show a remarkable effect. So, they
advised to machine with a large depth of cut in order to increase the material
removal rate. The mixed ceramic tool was used by (Aslan et al. 2007) in their
experimental investigation regarding the hard turning of AISI 4140 steel. They
indicated that the surface roughness was mainly affected by cutting speed-feed
rate and feed rate-depth of cut interactions. They modeled the surface roughness
by multiple linear regressions. (Deruja et al. 2009) observed the best surface
roughness at a small feed rate and high cutting speed in hard turning of AISI H11
steel with coated mixed ceramic inserts. The effect of cutting parameters on the

420 I. Meddour, M.A. Yallese, and H. Aouici
surface roughness was also investigated by (Lalwani et al. 2008) in finish hard
turning of MDN250 steel using coated ceramic tool. The employment of ANOVA
demonstrated that the feed rate is the prevailing parameter whereas cutting speed
had no significant effect on surface roughness. In hard turning of AISI 52100 steel
with CBN tool, (Bartarya and Choudhuryb 2012) used the multiple linear regression
to develop a surface roughness model prediction based on cutting parameters.
Other experimental studies, as the work of (Horng et al. 2008), considered the
tool nose radius as variable along with the cutting parameters. The ANOVA
revealed that the cutting speed and the tool nose radius influenced significantly the
surface roughness of Hadfield steel hard turned with mixed ceramic insert. Also
they linked the inspected increase of surface roughness when using a large tool
nose radius to the excessive thermal crack and flank wear. In their point of view it
is due to the enlargement of the contact area. The same number of variables was
also adopted by (Saini S et al. 2012) in the case of hard turning of AISI H11 steel
with ceramic tool, the ANOVA results showed that the feed rate influences the
surface roughness most significantly followed by cutting speed and nose radius.
Thus, according to the author the best surface quality was attainable by decreasing
feed rate and increasing cutting speed. In hard turning AISI H11 with CBN tool,
(Aouici et al. 2012) observed that feed rate had a statistical significance on the
surface roughness, because its increase generates helicoids furrows the result of
tool shape and helicoids movement tool-workpiece. They developed mathematical
models for surface roughness by using response surface methodology (RSM).
Other experimental researches were interested by the influence of the vibration
generated during the process on the machined products. This vibration could result
in the instability of the system, known as chattering, if an inadequate selection of the
cutting parameters is done. (Tlusty 1999), consider that this phenomenon occurs
when the uncut chip thickness is great regarding the dynamic stiffness of the system.
(Thomas et al. 1996) related the difference between the measured and the theoretical
surface roughness to the vibrations. Based on this concept, (Jang et al. 1996)
developed a real-time monitoring algorithm to calculate the surface roughness. The
algorithm superpose the theoretical profile and the oscillatory profile
determined by the relative vibration between the cutting edge and the
workpiece. The radial acceleration of the tool was taking into account beside the
cutting parameters by (K.A Risbood et al. 2003), when they developed a neural net
work models to predict surface roughness and dimensional deviation. (Kirby et al.
2004) developed a prediction surface roughness model based on feed rate and the
measured vibration in three directions. (Abouelatta and Madl 2001) disclosed that
the mathematical models of surface roughness including both machining conditions
and tool vibrations were more efficient than those depending on machining
conditions only. (Hessainia et al 2013) correlated between the surface roughness and
tool nose displacement. They developed roughness models containing all cutting
parameters, tool-nose displacements, spindle and machine tool frame.
Investigation and Modeling of Surface Roughness of Hard Turned AISI 52100 Steel 421
The aim of this work is the investigation of the effects of cutting conditions
(cutting speed, depth of cut, feed rate and tool nose radius) and tool vibration on the
surface roughness of hard turned AISI 52100 steel. This, by applying the analysis of
variance (ANOVA). Also, the surface roughness is modeled by employing the
response surface methodology (RSM), first by using only cutting conditions, then,
the tool vibration is included in the model and both models are compared.
2 Experimental Equipments and Methods
2.1 Equipments
Figure 1 shows the experimental procedure and the different equipments used in
this study. The experimentation was performed under dry conditions using AISI
52100 steel as workpiece material. The hardness was raised by quenching and
tempering treatment followed by checking measurement with a digital Micron
Hardness Tester DM2-D390. The average of measured values was 59 HRC. The
workpiece is mounted on a universal lathe, model SN40C, spindle power 6.6 KW.
The ceramic insert tool is a mix of 70% Al2O3 and 30% TiC. This type of inserts is
commonly called CC650, and its ISO designation is SNGN 1204(r) T01020 where
Fig. 1 Experimental procedure

(r) is the tool nose radius ranged from 0.8 to 1.6 mm in this work, indicating a 20°
chamfer angle over 0.1mm width. Tool holders are codified as PSBNR2525M12
with a common active part tool geometry described by: cutting edge angle χr
=+75°, clearance angle α =+6°, rake angle γ =-6° and cutting edge inclination
angle λ =-6°. New cutting edge is used for each test in order to reduce the wear
influence on measured responses. The vibrometer (SmartBalancer2) is used for the
measurement of the acceleration amplitude of the tool vibration in both tangential
and radial directions. The sampling frequency is 12800 Hz. Regarding the
measurement of the surface roughness, a Mitutoyo Surftest–201 (cut-off = 0.8) is
applied after each test in three different positions by going around the workpiece
each 120°. Afterward, the average of the three measurements is estimated. In
addition, the three-dimensional topographic maps of the machined surfaces are
produced using the optical platform of metrology modular Altisurf 500.
2.2 Plan of Experiments

Four cutting parameters are investigated in this study, which are: cutting speed
(Vc), feed rate (f), cutting depth (ap) and tool noise radius (r). In order to minimize
the cost and the time of the experiments, the parameters are varied according to a
central composite design CCD, in particular the commonly called face-centred. It
counts thirty runs composed of sixteen factorial points, six center points, and eight
axial points located at cube faces which mean α = 1. Thus, each factor is varied at
three levels which are presented in Table 1. The responses are: acceleration
amplitude in radial direction (AMPy), acceleration amplitude in tangential
direction (AMPz), and the arithmetic mean of roughness (Ra).
The principal of the RMS is to model the experimental data. The central
composite design allows the evaluation of the linear effects, and also to the
eventual quadratic interaction effects between the different variables. We used the
quadratic model with interactions as follow:
∑ ∑ ∑
k k k
Y = a0 + bX + b X Xj + b X2 (1)
i =1 i i i , j ij i i =1 ii i
where i, j, k =1, 2... n

a0, bi, bij and bii are the regression coefficients of the model and Xi, Xj, are the
explicative variables. The analysis of variance ANOVA was employed in order to
determine the significance of the independent variables on the output responses.
The results of the experimental tests are presented in Table 2.
Table 1 Factors variation
level Cutting speed (m/min) Feed rate (mm/rev) Depth of cut (mm) Tool nose radius (mm)
-1 100 0.08 0.05 0.8
0 150 0.11 0.15 1.2
+1 200 0.14 0.25 1.6
Table 2 Experimental run and results
Input variables Output responses

run Vc ap f r AMPy AMPz Ra
m/min mm mm/rev mm m/s² m/s² ȝm
1 200 0.25 0.14 0.8 0.53 0.56 0.83
2 150 0.15 0.11 1.2 0.55 0.62 0.59
3 100 0.05 0.08 0.8 0.75 0.69 0.64
4 100 0.05 0.14 0.8 0.39 0.56 1.09
5 100 0.25 0.08 1.6 0.22 0.42 0.33
6 150 0.15 0.11 1.2 0.63 0.8 0.54
7 100 0.25 0.08 0.8 0.14 0.48 0.35
8 100 0.25 0.14 1.6 0.17 0.38 0.62
9 100 0.05 0.14 1.6 0.71 0.52 0.45
10 200 0.25 0.14 1.6 0.53 0.46 0.44
11 200 0.25 0.08 0.8 0.47 0.39 0.37
12 200 0.05 0.08 0.8 0.8 0.63 0.65
13 100 0.25 0.14 0.8 0.12 0.46 0.79
14 200 0.05 0.14 0.8 1.41 1.24 0.98
15 150 0.15 0.11 1.2 0.63 0.74 0.53
16 200 0.05 0.08 1.6 0.69 0.56 0.34
17 200 0.25 0.08 1.6 0.27 0.37 0.30
18 150 0.15 0.11 1.2 0.99 0.71 0.53
19 200 0.05 0.14 1.6 1.04 0.73 0.37
20 100 0.05 0.08 1.6 0.39 0.46 0.31
21 150 0.15 0.14 1.2 1.04 0.67 0.61
22 150 0.15 0.11 1.2 0.6 0.63 0.48
23 150 0.25 0.11 1.2 0.4 0.59 0.63
24 100 0.15 0.11 1.2 0.86 0.82 0.45
25 150 0.15 0.11 1.2 0.54 0.53 0.56
26 150 0.05 0.11 1.2 0.62 0.9 0.59
27 150 0.15 0.11 1.6 0.63 0.6 0.41
28 200 0.15 0.11 1.2 1.09 1.42 0.54
29 150 0.15 0.08 1.2 0.63 0.37 0.37
30 150 0.15 0.11 0.8 0.63 0.48 0.61
3.1 Statistical Analysis and Modelling

The analysis of variance is performed to establish the statistical significance of the
regression models, model coefficients and lack of fit. It is done by comparing
“Prob > F” to 0.05 or in other words at 95% of confidence. The proportion of
contribution of each model coefficient was calculated. The regression equations
are generated by the statistical software Minitab 16. Those equations describe the
statistical relationship between the factors and the responses and could predict
new runs
3.1.1 Surface Roughness
The adopted model for surface roughness including cutting parameters is the
linear with interaction. The results given by ANOVA presented in Table 3 shows
that the terms r and f are the most significant with the respective contribution
(41.98, 35.51) %. The term ap has a small contribution with 3.23 %, whereas
cutting speed does not affect the surface roughness. A similar result was obtained
by (Aslan et al. 2007) when turning AISI 4140 steel (63 HRC) with mixed
ceramic tool. The interactions which show significant effect on surface roughness
are ap×r and f×r with the respective contributions (9.67 and 7.34) %. The
determination coefficient model’s R² is 0.94. The final equation in terms of real
factors is given as follow:
Ra = −0.049 + 2.242 × 103Vc − 3.618ap + 12.479 f + 0.051r − 0.015Vc × f − 6.875 × 10 −4 Vc × r + 7.917 ap × f (2)
+ 1.938ap × r − 5.625 f × r
The ANOVA results for surface roughness model including cutting parameters
and tool vibrations in both radial and tangential directions are presented in Table
4. It shows that in addition to the terms cited above, the surface roughness is
influenced by the tool vibration in the radial direction AMPy with 3.71 % of
contribution. The determination coefficient model’s R² is improved to 97.10 %.
The final equation in terms of real factors is given as follow:
Ra = −0.895964 − 2.32358 ap + 16.7492 f + 0.528119 r + 1.19221AMPz + 0.937027 AMPy + 0.0128869
Vc × ap − 0.0190204Vc × f-7.89172ap × f + 0.841952ap × r1.64463ap × AMPz − 1.41237ap × AMPy (3)
− 4.99520 f × r + 0.589721 f × AMPz-4.94334 f × AMPy-0.538542r × AMPz-0.486322r × AMPy-0.988144
AMPz × AMPy
3.1.2 Vibration Analysis
According to the Table 5, the significant terms of acceleration amplitude model in

radial direction (AMPy) were Vc, ap with the respective contribution (24.79,
39.11) % and f with the lower contribution of 6.27 %. The term ap² is also
significant (14.97 %). The lack of fit P-value of 0.561 was greater than 0.05.
Therefore, the regression model does not fail to adequately explain the functional
relationship between the experimental factors and the response. Lack of fit may
occur if important terms from the model such as interactions or quadratic terms
are not included. The correlation coefficient R² of about 84.47 is considered as
good. The final equation in terms of real factors acceleration amplitude in radial
direction (AMPy) is given as follow:
AMPy = 1.5015 - 0.0224Vc + 7.6269 ap - 20.6830 f + 2.0932r - 0.0069Vc × ap + 0.0578Vc × f - 0.0024Vc × r (4)
- 13.9167ap × f + 0.7063 ap × r + 3.0625 f × r + 7.7846×10-5Vc2 - 27.0386 ap2 + 60.6823 f 2 - 0.9399r 2
The results given by the ANOVA analysis of the acceleration amplitude in

tangential direction (AMPz) presented in Table 6, show that the most significant
model terms are Vc and ap with the respective contributions (10.29 and 19.85) %,
whereas f and r do not affect the tool vibration in tangential direction. The
quadratic term f² is significant with 21.71 % of contribution. The coefficient R² of
about 81.22 is considered good. The final equation in terms of real factors is given
as follow:
AMPz = -1.6633 - 0.0428Vc + 1.2938ap + 60.0136 f + 3.5225r - 0.0111Vc × ap + 0.0488Vc × f

- 9.0625 × 10- 04Vc × r - 10.625ap × f + 0.9219ap × r - 1.8229 f × r + 0.0001Vc 2 - 2.5790ap 2 - 278.655 f 2
(5)
2
- 1.4424r
Table 3 ANOVA for Ra vs cutting parameters
Source DF Seq SS Adj SS Adj MS F-value P-value Cont.%

Regression 10 0.99358 0.99358 0.09936 29.09 0.000
Vc 1 0.00245 0.00245 0.00245 0.72 0.408 0.25
ap 1 0.03209 0.03209 0.03209 9.40 0.006 3.23
f 1 0.35280 0.35280 0.35280 103.30 0.000 35.51
r 1 0.41709 0.41709 0.41709 122.13 0.000 41.98
Vc×ap 1 0.00000 0.00000 0.00000 0.00 1.000 0.00
Vc×f 1 0.00810 0.00810 0.00810 2.37 0.140 0.82
Vc× r 1 0.00302 0.00303 0.00303 0.89 0.358 0.30
ap× f 1 0.00902 0.00903 0.00903 2.64 0.121 0.91
ap× r 1 0.09610 0.09610 0.09610 28.14 0.000 9.67
f× r 1 0.07290 0.07290 0.07290 21.35 0.000 7.34
Residual Error 19 0.06489 0.06489 0.00342
Lack-of-Fit 14 0.05821 0.05821 0.00416 3.11 0.108
Pure Error 5 0.00668 0.00668 0.00134
Total 29 1.05847
Table 4 ANOVA for Ra vs cutting parameters and tool vibrations

Regression 21 1.02779 1.02779 0.04894 12.76 0.000

Vc 1 0.00245 0.00479 0.00479 1.25 0.296 0.24
ap 1 0.03209 0.00017 0.00017 0.04 0.840 3.12
f 1 0.3528 0.00258 0.00258 0.67 0.436 34.33
r 1 0.41709 0.02122 0.02122 5.53 0.047 40.58
AMPz 1 0.00067 0.00179 0.00179 0.47 0.514 0.07
AMPy 1 0.03813 0.0105 0.01050 2.74 0.137 3.71
Vc×ap 1 0.00003 0.00634 0.00634 1.65 0.234 0.00
Vc×f 1 0.00371 0.00298 0.00298 0.78 0.404 0.36
Vc×r 1 0.00753 0.00049 0.00049 0.13 0.729 0.73
Vc×AMPz 1 0.00262 0.00000 0.00000 0.00 0.996 0.25
Vc×AMPy 1 0.00000 0.00607 0.00607 1.58 0.244 0.00
ap×f 1 0.00801 0.0009 0.00090 0.24 0.640 0.78
ap×r 1 0.09658 0.00158 0.00158 0.41 0.539 9.40
ap×AMPz 1 0.00064 0.00175 0.00175 0.46 0.518 0.06
ap×AMPy 1 0.00011 0.00253 0.00253 0.66 0.440 0.01
f×r 1 0.05331 0.01917 0.01917 5.00 0.056 5.19
f×AMPz 1 0.00005 0.00001 0.00001 0.00 0.961 0.00
f×AMPy 1 0.00166 0.00426 0.00426 1.11 0.323 0.16
r×AMPz 1 0.00173 0.00139 0.00139 0.36 0.564 0.17
r×AMPy 1 0.00017 0.00176 0.00176 0.46 0.517 0.02
AMPz×AMPy 1 0.00840 0.0084 0.00840 2.19 0.177 0.82
Residual Error 8 0.03068 0.03068 0.00384
Total 29 1.05847
Table 5 ANOVA for AMPy

Regression 14 2.21852 2.21852 0.15847 5.83 0.001
Vc 1 0.52771 0.52771 0.52771 19.4 0.001 23.79
ap 1 0.86768 0.86768 0.86768 31.91 0.000 39.11
f 1 0.13904 0.13904 0.13904 5.11 0.039 6.27
r 1 0.01921 0.01921 0.01921 0.71 0.414 0.87
Vc×Vc 1 0.03511 0.09813 0.09813 3.61 0.077 1.58
ap×ap 1 0.33218 0.18942 0.18942 6.97 0.019 14.97
f×f 1 0.00029 0.00773 0.00773 0.28 0.602 0.01
r×r 1 0.0586 0.0586 0.05860 2.15 0.163 2.64
Vc×ap 1 0.01877 0.01877 0.01877 0.69 0.419 0.85
Vc×f 1 0.12041 0.12041 0.12041 4.43 0.053 5.43
Vc×r 1 0.03725 0.03725 0.03725 1.37 0.260 1.68
ap×f 1 0.02789 0.02789 0.02789 1.03 0.327 1.26
ap×r 1 0.01277 0.01277 0.01277 0.47 0.504 0.58
f×r 1 0.02161 0.02161 0.02161 0.79 0.387 0.97
Residual Error 15 0.40792 0.40792 0.02720
Lack-of-Fit 10 0.26719 0.26719 0.02672 0.95 0.561
Pure Error 5 0.14073 0.14073 0.02815
Total 29 2.62644
R² 84.47
Table 6 ANOVA for AMPz

Regression 14 1.33035 1.33035 0.09503 4.63 0.003
Vc 1 0.13694 0.13694 0.13694 6.68 0.021 10.29
ap 1 0.26402 0.26402 0.26402 12.88 0.003 19.85
f 1 0.08134 0.08134 0.08134 3.97 0.065 6.11
r 1 0.05445 0.05445 0.05445 2.66 0.124 4.09
Vc×Vc 1 0.00214 0.31596 0.31596 15.41 0.001 0.16
ap×ap 1 0.17865 0.00172 0.00172 0.08 0.776 13.43
f×f 1 0.28883 0.16296 0.16296 7.95 0.013 21.71
r×r 1 0.13800 0.13800 0.13800 6.73 0.02 10.37
Vc×ap 1 0.04951 0.04951 0.04951 2.41 0.141 3.72
Vc×f 1 0.08556 0.08556 0.08556 4.17 0.059 6.43
Vc×r 1 0.00526 0.00526 0.00526 0.26 0.62 0.40
ap×f 1 0.01626 0.01626 0.01626 0.79 0.387 1.22
ap×r 1 0.02176 0.02176 0.02176 1.06 0.319 1.64
f×r 1 0.00766 0.00766 0.00766 0.37 0.55 0.58
Residual Error 15 0.30754 0.30754 0.02050
Lack-of-Fit 10 0.26046 0.26046 0.02605 2.77 0.136
Pure Error 5 0.04708 0.04708 0.00942
Total 29 1.63790
R² 81.22
3.2 Graphics Analysis
3.1.1 Surface Roughness
It is well seen in Figure 2(a) that the use of small feed rate and large nose radius
results in the lowest surface roughness. Whereas, the use of large feed rate and
small tool nose radius results in highest surface roughness. Because at large feed
rate the distance between peaks and valleys of the feed marks is much more
important. The ANOVA revealed that cutting speed does not influence the surface
roughness. The depth of cut is found to be statistically significant. The effect of
tool nose radius and depth of cut on surface roughness is depicted in Figure 3(b).
We can observe that using a small depth of cut gives worst surface roughness
especially with tool nose radius of 0.8 mm. We could attribute this to the material
side flow phenomenon, due to temperature elevation and specific cutting pressure
in the cutting zone when using small undeformed chip thickness. It consists of
ploughing and squeezing of the machined material by the tool causing its
deterioration (Kishawy and Elbestawi 1999). Figures 3(a) and 3(b) show the
topography of machined surfaces using the lower feed rate and with both depth of
cut (0.05 and 0.25) mm respectively. We can observe that the valleys of surface
(a) are distant above those of surface (b) due to the material accumulation. This
additional volume of squeezed material presses on the tool flank face which can
explain the increase of acceleration amplitude in the radial direction (AMPy) at
small depth of cut (Figure 5(a)). The same trend is observed for the acceleration
amplitude in tangential direction (AMPz) (Figure 5(b)). This figure shows that the
tangential vibration (AMPz) increases with the increase in tool nose radius. Then,
it decreases by using tool nose radius of 1.6 mm. We can explain this by the
stabilization of cutting process when using large nose radius leading to the
improvement of surface roughness.
a) b)
Fig. 2 Effect of cutting conditions on surface roughness
a) Vc = 200 m/min, ap = 0.05 mm, b) Vc=200 m/min, ap = 0.25 mm,

f = 0.08 mm/rev and r = 0.8 mm f = 0.08 mm/rev and r = 0.8 mm
Fig. 3 Topography of AISI 52100 steel hard turned with mixed ceramic insert
a) b)
Fig. 4 Effect of Effect of cutting conditions on the tool vibration
4 Models Confirmation
The reliability of the developed mathematical model is investigated by performing

three new runs. The confirmation runs conditions were set within the range of the
factors levels already defined. Table 6 shows the percentage error between the
experimental values and their corresponding predicted values obtained from
the Eqs. (2). slight values of divergence were observed surface roughness.
Consequently, the mathematical models can be used to effectively predicting
surface roughness values for any combination of cutting speed, feed rate, depth of
cut and tool nose radius within the limits of the actual experimentation.
Table 7 Comparison between experimental and predicted values
Vc ap f r
run exp Ra pred Ra error (%)
1 115.4 0.05 0.09 1.60 0.34 0.33 - 2.78

2 131.9 0.20 0.09 0.80 0.48 0.48 0.68
3 160.9 0.10 0.12 1.20 0.64 0.60 - 5.78
5 Conclusion
Through the results obtained above, the following conclusions could be drawn:
1. The best surface roughness is obtained by using small feed rate and large
nose radius. The use of depth of cut of 0.05 mm with tool nose radius of 0.8
mm gives the worst surface roughness due to the material side flow
phenomenon.
2. The ANOVA results show that depth of cut and cutting speed influence
considerably the tool vibrations on both tangential and radial direction,
whereas the feed rate has a significant effect only on acceleration amplitude
in radial direction (AMPy). It is also revealed that the tool nose radius has
only a slight statistical significance on acceleration amplitude in tangential

direction (AMPy).
3. When using a small undeformed chip thickness, the material side flow
phenomenon could be detected by the increase of the tool vibrations. It’s due
to the exerted pressure of the squeezed material on the tool flank face.
4. The optical platform of metrology modular was an important tool in the
investigation of the surface roughness, through the produced three-
dimensional topographic maps of the machined surfaces.
References
Ozel, T., Karpat, Y., Figueira, L.: Predictive modeling of surface roughness and tool wear
in hard turning using regression and neural networks. Int. J. Mach. Tools Manuf. 45,
467–479 (2005)
Yallese, M.A., Rigal, J.-F., Chaoui, K., Boulanouar, L.: The effects of cutting conditions on
mixed ceramic and cubic boron nitride tool wear and on surface roughness during
machining of X200Cr12 steel (60 HRC). Proc. IMechE, Part B: J. Eng. Manuf. 219
(2005)
Aslan, E., Camuscu, N., Birgoren, B.: Design optimization of cutting parameters when
turning hardened AISI 4140 steel (63 HRC) with Al2O3 + TiCN mixed ceramic tool.
Mater. Design 28, 1618–1622 (2007)
Dureja, J.S., Gupta, V.K., Sharma, V.S., Dogra, M.: Design optimization of cutting
conditions and analysis of their effect on tool wear and surface roughness during hard
turning of AISI-H11 steel with a coated–mixed ceramic tool. J. Eng. Manuf. 223(B),
1441–1453 (2009)
Lalwani, D.I., Mehta, N.K., Jain, P.K.: Experimental investigations of cutting parameters
influence on cutting forces and surface roughness in finish hard turning of MDN250
steel. J. Mater. Process. Technol. 206, 167–179 (2008)
Bartarya, G., Choudhuryb, S.K.: Effect of cutting parameters on cutting force and surface
roughness during finish hard turning AISI52100 grade steel. Procedia CIRP 1, 651–656
(2012)
Horng, J.-T., Liu, N.-M., Chiang, K.-T.: Investigating the machinability evaluation of
Hadfield steel in the hard turning with Al2O3/TiC mixed ceramic tool based on the
response surface methodology. J. Mater. Process. Technol. (2008)
Saini, S., Ahuja, I.S., Sharma, V.S.: Influence of Cutting Parameters on Tool Wear and
Surface Roughness in Hard Turning of AISI H11 Tool Steel using Ceramic Tools. Int. J.
Precision Eng. Manuf. 13(8), 1295–1302 (2012)
Aouici, H., Yallese, M.A., Chaoui, K., Mabrouki, T., Rigal, J.F.: Analysis of surface
roughness and cutting force components in hard turning with CBN tool: prediction
model and cutting conditions optimization. Measurement 45, 344–353 (2012)
Tlusty, J.: Manuf. Process. Equipment. Prentice Hall (1999)
Thomas, M., Beauchamp, Y., Youssef, A.Y., Masounave, J.: Effect of tool vibration on
surface roughness during lathe dry turning process. Comput. Ind. Eng. 31(3-4), 637–644
(1996)
Jang, D.Y., Choi, Y.G., Kim, H.G., Hsiao, A.: Study of the correlation between surface
roughness and cutting vibrations to develop an on-line roughness measuring technique
in hard turning. Int. J. Mach. Tools Manuf. 36, 453–464 (1996)
Risbood, K.A., Dixit, U.S., Sahasrabudhe, A.D.: Prediction of surface roughness and
dimensional deviation by measuring citting forces and vibrations in turning process. J.
Mater. Process. Technol. 132, 203–214 (2003)
Kirby, E.D., Zhang, Z., Chen, J.C.: Development of An Accelerometer based surface
roughness prediction system in turning operation using multiple regression techniques.
J. Ind. Technol. 4(20), 1–8 (2004)
Abouelatta, O.B., Madl, J.: Surface roughness prediction based on cutting parameters and
tool vibrations in turning operations. J. Mater. Process. Technol. 118, 269–277 (2001)
Hessainia, Z., Kribes, N., Yalles, M.A., Mabrouki, T., Ouelaa, N., Rigal, J.-F.: Turning
roughness model based on tool-nose displacements. Mechanika 19(1), 112–119 (2013)
ISSN 1392-1207
Kishawy, H.A., Elbestawi, M.A.: Effects of process parameters on material side flow
during hard turning. Int. J. Mach. Tools Manuf. 39, 1017–1030 (1999)
Finite Element Simulation of Fatigue Damage
Accumulation for Repaired Component by Cold
Expansion Method
Abdelkrim Aid1, Mostefa Bendouba1, Mohamed Benguediab2,

and Abdewahab Amrouche3
1
Laboratoire LPQ3M, B.P. 305, Universite de Mascara, Algeria
1aid_abdelkrim@yahoo.com, 2bendoubamos@yahoo.fr
2
Département de Génie Mécanique, Université de Djilali Liabès, Sidi Bel Abèss, Algeria
benguediab_m@yahoo.fr
3
Laboratoire de Génie Civil et Géo-Environnement LGCgE, EA 4515, Faculté des Sciences
Appliquées FSA Béthune, Université d’Artois, France
abdelwaheb.amrouche@univ-artois.fr
Abstract. This manuscript investigates the effectiveness of applying the cold ex-
pansion process to extend the fatigue life of mechanical structures. During the
cold expansion process compressive residual stresses around the expanded hole
are generated. The enhancement of fatigue life and the crack initiation and growth
behavior of a holed specimen were investigated by using the 6082 Aluminum al-
loy. The present study suggests a simple technical method for enhancement of
fatigue life by a cold expansion hole of pre-cracked specimen. This technique pro-
duces beneficial high compressive residual stresses which have been predicted by
means of finite element models, both 3D for proper assessment of thickness ef-
fects. Finite element models have been developed to increase their complexity, Fa-
tigue damage accumulation of cold expanded hole in aluminum alloy which is
widely used in transportation and in aeronautics was analyzed. Experimental tests
were carried out using pre-cracked SENT specimens. Tests were performed in two
and four block loading under constant amplitude. These tests were performed by
using two and four blocks under uniaxial constant amplitude loading. The experi-
mental results were compared to the damage calculated by the Miner’s rule and a
new simple fatigue damage indicator based on an energy criterion. This compari-
son shows that the ‘energy criterion model’, which takes into account the loading
history, yields a good estimation according to the experimental results.
Keywords: cold expansion, compressive residual stress, finite element method,

energies criterion, fatigue damage accumulation.
1 Introduction
Over the last 25 years, the cold expansion process has been commonly used to
improve the fatigue life of components containing fastener holes. Cold expansion

434 A. Aid et al.
is employed usually in components that are exposed only to service conditions at

ambient temperature. The cold-worked process introduces beneficial residual
circumferential stresses into an annular region around the hole, and the presence
of this compressive residual stress inhibits the growth and propagation of cracks
(Lacarac et al. 2001, Pasta 2007).
In order to increase the service life of components some methods have been de-
veloped. One of these methods, which were used in this study, is to form a con-
trolled compressive residual stress field around the hole (Chandawanich and
Sharpe 1979; Su and Yan 1986). However, the improvement in fatigue life is dif-
ficult to quantify. The residual stresses resulting from the cold expansion process
are not uniform through the thickness (Pavier et al. 1997). Earlier work on cold
expansion has used both analytical and numerical two-dimensional models to pre-
dict residual stress (Poussard et al. 1995). Furthermore, combining the residual
stress distribution with a fatigue crack growth rate calculation is difficult due to
the three-dimensional nature of the problem and complicating phenomena such as
crack closure (Lacarac et al. 2004).
In (Bernard et al. 1995; Ghfiri et al. 2000; Semari et al. 2013; Aid et al. 2013),
the authors show the sensitivity of the degree of expansion. In a first phase the re-
sidual life increases but beyond a critical value of remaining life decreases sharply
and growth is extremely detrimental.
In service conditions, the components or structures are subjected to random or
variable block loading. Different relationships (Lacarac et al. 2004) have been
proposed to calculate the effect of variable amplitude loading conditions. Howev-
er, these procedures need the identification of many parameters. In literature, in
the particular case of block loading, the analysis for this phenomenon is oriented
only to two loading steps.
Miner’s rule (Miner 1945) is very much used to evaluate the fatigue damage
accumulation when the components or structures are subjected to variable block
loading. In this work, the results of a study on fatigue damage accumulation of
cold expanded hole in aluminum alloys subjected to block loading are presented.
Tests were carried out using pre-cracked SENT specimens and inserting an ex-
panded hole at the crack tip. The degree of the cold expansion was chosen equal to
4.3%. Tests were performed in two and four block loading under constant
amplitude.
The experimental results were compared to the damage calculated by the Min-
er’s rule and a new simple fatigue damage indicator (Aid 2006a; Aid 2011;
Amrouche 2008; Aid 2006b) and a new version of this damage indicator based on
energy approach (Bendouba et al. 2011; Djebli et al. 2013).
2 General Principal of the Method
In this study, cold expansion is achieved by inserting an oversized rigid ball from
one side (entry face) of the holed plate and removing it from the other side (exit
face). The degree of cold expansion DCE is defined by the relation:
Finite Element Simulation of Fatigue Damage Accumulation 435
(D − d )
DCE % = × 100 (1)
d
A cold expansion of the hole introduces compressive residual stresses that are
beneficial in terms of lifetime until a critical degree of expansion where it
becomes not beneficial for structures. Both aspects are shown in Figure 1.
Fig. 1 The curve of crack initiation according to the numbers of the cycles
Where d is the diameter of the hole drilling and D is the diameter of the rigid
ball. Different technique can be used to repair a cracked component. The cold
working expansion process was realized by forcing a hard steel ball of 6 mm in-
side a predrilled hole (the initial diameters of the hole are: 5.9, 5.8, 5.75, 5.6 and
5.5 mm for aluminium alloy and 5.8, 5.75 for steel), this process is illustrated in
(Aid 2006) .
3 Materials and Specimens
The material used for this study was aluminum alloy Al 6082 T6, some of the me-
chanical properties are given in (Aid 2006). A difference gap of 10% between the
characteristics of the uncracked specimens and the batch of specimens with hole
was noticed.
The specimens used for this investigation were conforms to ASTM standards
(ASTM 1997). The geometry of the fatigue test specimen cut in the longitudinal
direction (Aid 2006). For getting specimens with an expanded hole of 6 mm in di-
ameter we drilled a hole of Ø5.75 mm at the pre-crack tip and then a cold-working
expansion process was conducted by forcing a steel ball of 6 mm diameter.
The fatigue tests were carried out using a 100 kN capacity Instron hydraulic
machine. The loading frequency was 30 Hz and a stress ratio R of 0.57. During
436 A. Aid et al.
fatigue testing, a video camera with scale of 0.1 mm was used to determine the
crack initiation in the entry and exit faces of specimen.
3.1 3D Finite Element Modeling

3-D FEM simulations using MSC.Marc have been used to evaluate the rigid ball
during the cold expansion process. In this case the axial movement of the rigid
ball through the specimen has been introduced in the model to obtain the distribu-
tion of the residual stresses through the thickness. Because of the symmetry, half
of the lug is representative of the useful of the specimen. The quadrilateral
element topology QUAD4 has been adopted for the analysis. Globally, the finite
element model consists of 3886 elements and 5721 nodes. (see figure 2).
Fig. 2 3D FEM mesh, with rigid ball for cold-expansion and fatigue critical section
The contact between the surfaces of the steel ball and the hole is simulated by
using the contact elements. Residual stress is obtained at the three surfaces (entry
face, middle and exit face) to analyze the influence of cold expansion direction on
the fatigue crack initiation life. During the simulation, the rigid ball was moved
through the hole incrementally.
The boundary conditions imposed are as following:
The displacements of the nodes located on the X–Y plane are constrained in the
z direction.
The displacements of the nodes located on the X–Z plane are constrained in the
y direction within a rectangular surface on the exit face.
4 Results and Discussions of the Simulation
The figure 3 illustrates the hoop stresses at the entry, middle and exit expansion
step. Portrays the finite element prediction of the tangential residual stress varia-
tion at different through thickness positions from the entrance face, this figure re-
veals the large variation in hoop residual stress in the region of influence.
entry face middle face xit face

Fig. 3 Circumferential residual stress distribution in the 3D model
Low hoop residual stresses on entrance and exit faces are caused by the shear
stress from the axial movement of the rigid ball. The compressive residual stresses
into the annular region around the hole reduce the origin of cracks. During fatigue
testing, this residual stresses act to change the effective stress intensity factor at
the crack tip, i.e. the crack growth rates are lower than those in absence of residual
stress. The residual stress profiles are in agreement with other studies presented in
previous literature (Poussard et al.1985; Pavier et al. 1999 ; De Matos et al. 2005).
Figure 4 shows the FE prediction of hoop residual stresses through the thickness
of the plate from the entrance face to the exit face. A significant difference in the
magnitude of tangential stress is recorded between the entry side and exit side.
The magnitude of hoop residual stresses on the entrance and the exit faces are
lower than the stresses on the mid-thickness. These results are in accordance with
recent research (Yan et al. 2012 ; Chakherlou 2012).
5 Damage Accumulation Based an Energy Parameters

An energy approach is presented in this paper, which forms the basis for a new
damage model.
The strain energy density is proposed as a parameter of the fatigue analysis.
The models do not include a division of the strain energy density into elastic
and plastic parts, like in all the cases of the parameters proposed by Smith–
Watson–Topper (SWT) (Smith et al. 1970), Hoffman and Seeger (Hoffman and
Seeger 1989), Bergman and Seeger (Bergman and Seeger 1979). In the elastic
range, energy is calculated from
1
W = σε (2)
2
For uniaxial fatigue loading, authors (Bendouba et al. 2011; Djebli et al. 2013)
introduce a new damage parameter, Di, defined as the ratio of the increment of en-
ergy due to stress damage over the difference between the energy due to
ultimate stress and the applied stress. The damage indicator is defined by:
438 A. Aid et al.
200
150
Hoop residual stress (MPa)
100
50
0
-50 0 5 10 15 20
-100 Entry…
-150 Exit face
-200
-250
-300
Distance from hole edge (mm)
Fig. 4 Finite element predictions of residual hoop stress at entry face and exit face
W edi −W i
D= (3)
W u −W i
where:
Wedi : energy is due to stress damage.
Wi : energy is due to the applied stress.
Wu : energy is due to the ultimate stress of the material.
6 Fatigue Testes
The experimental conditions are given in (Aid 2006). Eight tests were carried out
for increasing loading conditions and as much for decreasing loading. To evaluate
the effect between these loading conditions, the Miner’s Rule was considered for
damage accumulation.
Endurance curves are shown in (Aid 2006). These are based on constant ampli-
tude test and the failure was considered at the crack initiation (Aid 2006).
Two and four cyclic stress levels were considered and two different sequences
were applied (Aid 2006). The aim of this set of tests is to determine the influence
of increasing or decreasing loading conditions on lifetime and to prove that the
proposed model takes into account the history of blocks loading and the nonlinear-
ity of the accumulated damage unlike the Miner’s rule.
Figure 5 demonstrates the comparison between the prediction models (Miner’
Rule, DSM model and the energy approach) with experimental results for different
loading conditions (two and four blocks with increasing and decreasing loads) for
specimens pre cracked and repaired by the technical process of cold expansion.
This figure confirms that the estimates results by the DSM model and the energy
approach used in this investigation are in good agreement with the experimental
results (Aid 2006).
Figure 5 demonstrates the comparison between the prediction models (Miner’

Rule, DSM model and the energy approach) with experimental results for different
loading conditions (two and four blocks with increasing and decreasing loads) for
specimens pre cracked and repaired by the technical process of cold expansion.
This figure confirms that the estimates results by the DSM model and the energy
approach used in this investigation are in good agreement with the experimental
results.
Fig. 5 Comparison between theoretical results and experimental results for different loading
conditions
7 Conclusion
At first, a modelization study on cold-expansion process by finite element method

has been made. Different FE models have been defined to evaluate different char-
acteristics as plastic strain representation, contacts simulation and material charac-
terization has been used. Residual stresses after the cold-expansion process have
been evaluated by means of FEM models using MSC Marc software.
Secondly, this investigation is also set to study fatigue damage accumulation in
arresting crack holes with cold expansion process was studied. In total, 24 speci-
mens were tested, eight specimens were used to obtain the fatigue endurance
curve and 16 were subjected to increasing or decreasing blocks loading. Using the
Miner’s rule to calculate the cumulative life time. We found that in both cases,
increasing and decreasing blocks loading, the experimental results were below
prediction.
The results obtained by the energy approach and model of the damage stress are
compared with the experimental results and a good agreement has been found.
The experimental results show that the load sequence has no significant effect on
the crack reinitiating.
440 A. Aid et al.
It seems that in the case of drilling with a cold expansion, the combination of
the geometrical and mechanical effect attributed to the stress concentration factor
associated with the compressive residual stresses predominate in the life time. In
the other hand the load sequence has a minor effect. In this investigation, the com-
pressive residual stresses at the edge of the hole are around of the yielding stress.
The effective local applied stress is lower than the residual stress; this observation
can explain the raison why there is no significant influence of the sequence load-
ing. Currently, we achieve tests with more important loading in order to evaluate
the sequence loading effect.
References
Lacarac, V.D., Smith, D.J., Pavier, M.J.: The effect of cold expansion on fatigue crack
growth from open holes at room and high temperature. Int. J. Fatigue 23, 161–170
(2001), doi:10.1016/S0142-1123(01)00125-6
Pasta, S.: Fatigue crack propagation from a cold-worked hole. Engng. Frac. Mech. 74,
1525–1538 (2007), doi:10.1016/j.engfracmech.2006.08.006
Chandawanich, N., Sharpe Jr., W.N.: An experimental study of fatigue crack initiation and
growth from cold worked holes. Engng. Frac. Mech. 11, 609–620 (1979),
doi:10.1016/0013-7944(79)90122-X
Su, X., Gu, M., Yan, M.: A simplified residual stress model for predict-ing fatigue crack
growth behavior at cold worked fastener holes. Fat. Frac. Engn. Mat. & Str. 9(1), 57–64
(1986), doi:10.1016/0013-7944(79)90122-X
Pavier, M.J., Poussard, C.G.C., Smith, D.J.: A finite element simulation of the cold work-
ing process for fastener holes. J. Strain Analysis 32, 287–300 (1997)
Poussard, C., Pavier, M.J., Smith, D.J.: Analytical and finite element pre-dictions of resi-
dual stresses in cold-worked fastener holes. J. Strain Analysis 30(4), 291–304 (1995)
Lacarac, V.D., Garcia-Granada, A.A., Smith, D.J., Pavier, M.J.: Prediction of the growth
rate for fatigue cracks emanating from cold expanded holes. Int. J. Fatigue 26(6), 585–
595 (2004), doi:10.1016/j.ijfatigue.2003.10.015
Bernard, M., Bui-Quoc, T., Burlat, M.: Effect of re-cold working on fatigue life enhance-
ment of a fastener hole. Fat. Fra. Eng Mat. Structures 18(17-18), 765–775 (1995),
doi:10.1111/j.1460-2695.1995.tb00902.x
Ghfiri, R., Shi, H.J., Guo, R., Mesmacque, G.: Effect of Expanded and non expanded Hole
on the Delay of Arresting Crack Propagation for Aluminum Alloys. Mat. Sci. Engineer-
ing A 286, 244–249 (2000), doi:10.1016/S0921-5093(00)00805-4
Semari, Z., Aid, A., Benhamena, A., Amrouche, A., Benguediab, M., Sadok, A., Bensed-
diq, N.: Effect of residual stresses induced by cold expansion on the crack growth in
6082 aluminum alloy. Eng. Fra. Mechanics 99, 159–168 (2013),
doi:10.1016/j.engfracmech.2012.12.003
Aid, A., Semari, Z., Benguediab, M.: Cold expansion effect on the fatigue crack growth of
Al 6082: numerical investigation. Structural Engineering and Mechanics 49(2), 225–235
(2014), doi:http://dx.doi.org/10.12989/sem.2014.49.2.225
Miner, M.A.: Cumulative damage in fatigue. J. App. Mech. 67, A159–A164 (1945)
Aid, A., Semari, Z., Amrouche, A., Mesmacque, G., Benguediab, M.: Non linear damage
cumulative criterion on fatigue under random loading. Alg. J. Adv. Materials (2006)
Aid, A., Amrouche, A., Bachir Bouiadjra, B., Benguediab, M., Mesmacque, G.: Fatigue life
prediction under variable loading based on a new damage model. Mat. Design 32,
183–191 (2011), doi:10.1016/j.matdes.2010.06.010
Amrouche, A., Su, M., Aid, A., Mesmacque, G.: Numerical study of the op-timum degree
of cold expansion: Application for the pre-cracked specimen with the expanded
hole at the crack tip. J. Mat. Pro. Technology 197, 250–254 (2008),
doi:10.1016/j.jmatprotec.2007.06.030
Aid, A.: Cumul d’endommagement en fatigue multiaxiale sous sollicitations variables.
[Thesis], p. 195. Universitè de Sidi-Bel Abbes, Algérie (2006)
Bendouba, M., Aid, A., Zengah, S., Benguediab, M.: Modèle énergétique de cumul du
dommage et prevision de la dureé de vie en fatigue sous chargements aléatoires. In: Pro-
ceedings of the Cong. Alg. Mécanique, Guelma, Algeria (2011)
Djebli, A., Aid, A., Bendouba, M., Amrouche, A., Benguediab, M., Benseddiq, N.:
A non-linear energy model of fatigue damage accumulation and its verification for
Al-2024 aluminum alloy. Int. J. Non-Linear Mechanics (2013), doi:10.1016
/j.ijnonlinmec.2013.01.007
ASTM E1049-85, Standard practices for cycle counting in fatigue analysis. Annual Book of
ASTM Standards, Philadelphia, vol. 03(01), 614–620 (1997, 1999)
Poussard, G.C., Pavier, M.J., Smith, D.J.: Analytical and finite element predictions of resi-
dual stresses in cold-worked fastener holes. J. Strain Anal. 32(4), 287–300 (1985)
Pavier, M.J., Poussard, G.C., Smith, D.J.: Effect of residual stress around cold worked
holes on fracture under superimposed mechanical load. J. Strain Anal. 63, 751–773
(1999)
de Matos, P.F.P., Moreira, P.M.G.P., Camanho, P.P., de Castro, P.M.S.T.: Reconstitution
of fatigue crack growth in Al-alloy 2024-T3 open-hole specimens using microfracto-
graphic techniques. Engng. Frac. Mech. 72, 2232–2246 (2005),
Yan, W.Z., Wang, X.S., Gao, H.S., Yue, Z.F.: Effect of split sleeve cold ex-pansion on
cracking behaviors of titanium alloy TC4 holes. Eng. Fra. Mechanics 88, 79–89 (2012),
Chakherlou, T.N., Shakouri, M., Aghdam, A.B., Akbari, A.: Effect of cold expansion and
bolt clamping on fretting fatigue behavior of Al 2024-T3 in double shear lap joints.
Mat., Design 33, 185–196 (2012), doi:10.1016/j.engfailanal.2012.04.008
Smith, K.N., Waston, P., Topper, T.H.: A stress–strain functions for the fatigue on mate-
rials. J. Mater. 5, 767–776 (1970)
Hoffman, H., Seeger, T.: Stress-strain analysis and life predictions of a notched shaft under
multiaxial loading. In: Leese, G.E., Socie, D. (eds.) Multi-axial Fatigue: Analysis and
Experiments AE-14, pp. 81–99. Society of Automative Engineers, Inc., Warrendale
(1989)
Bergman, J., Seeger, T.: On the influence of cyclic stress–strain curves, damage parameters
and various evaluation concepts on the life prediction by the local approach. In: Pro-
ceedings of the 2nd European Conference on Fracture, Darmstadt, Germany. VDI-
Report of Progress, vol. 18(6) (1979)
Geometrically Non-linear Free Vibration of
Fully Clamped Symmetrically Laminated
Composite Skew Plates
Hanane Moulay Abdelali1, Bilal Harras2, and Rhali Benamar1
1
Université Mohammed V Agdal-Ecole Mohammadia D’ingénieurs, Avenue Ibn Sina,
B.P.765, Agdal, Rabat, Morocco
h.moulay1983@gmail.com, rbenamar@emi.ac.ma
2
FST B.P.2202- Route D’Imouzzer Fes, Morocco
harrasbilal@yahoo.fr
Abstract. The present work concerns the geometrically non-linear free vibration
of fully clamped symmetrically laminated composite skew plates. The theoretical
model based on Hamilton’s principle and spectral analysis previously applied to
obtain the non-linear mode shapes and resonance frequencies of thin straight
structures, such as beams, plates and shells is used. A convergence study has been
performed and has shown that 18 plate functions should be taken into account. Re-
sults are given for the linear and non-linear fundamental mode shape of fully
clamped symmetrically laminated composite skew plates, considering different
parameters such as the skew angle, the number of layers, the fiber orientation, the
vibration amplitudes and plate aspect ratio. It was found that the non-linear fre-
quencies increase with increasing the amplitude of vibration and increasing the
skew angle, which corresponds to the hardening type effect, expected in similar
cases, due to the membrane forces induced by the large vibration amplitudes.
Keywords: Non-linear vibrations, Composite laminated skew plate.
1 Introduction
Composite materials are used in many industrial fields. Resulting composite ma-
terial present the best qualities of their constituents and some properties can be
improved by forming a composite material including reduction in structural
weight, reduction in fatigue and corrosion problems, that’s why it’s present inter-
ested use in aircraft structures and space vehicles. For the potential advantages
listed above, the fundamental carbon fiber reinforced plastic (CFRP) materials
have been used as structural members, taking the place of traditional metals. The
use of composite materials requires complex analytical methods in order to predict
accurately their response to external loading, especially in severe environments,

444 H.M. Abdelali, B. Harras, and R. Benamar
which may induce geometrically non-linear behaviour. The geometrically non-

linear analysis of composite skew plates exhibits specific difficulties due to the
anisotropic material behavior, and to the higher non-linearity induced by a higher
stiffness, inducing tensile mid-plane forces in skew plates higher than that ob-
served with conventional homogeneous materials. The problem of non-linear dy-
namic response of plates at large vibration amplitudes was investigated by many
authors and subjected to some important works and methodological approaches
during the last few decades. In the monograph of (Leissa 1969), a summary of the
basic knowledge on linear dynamic isotropic plate problems has been presented.
In the series of studies carried out by Benamar and his co-workers (Benamar and
Bennouna 1990), a wide programme of experimental work was concentrated on
the understanding of the dynamic behaviour of homogeneous and CFRP beams
and plates. In reference (Chia 1988), a survey was presented of the literature on
the geometric non-linear analysis of laminated composite elastic plates. (Han and
Petyt 1996), (Ribeiro and Petyt 1999) have been presented dealing with the geo-
metrically non-linear dynamic behavior of symmetrically laminated plates by us-
ing the hierarchical finite element method (HFEM). (Harras and Benamar 2001,
2002) investigated theoretical and experimental of the non-linear behavior of vari-
ous fully clamped rectangular composite panels at large vibration amplitudes.
Good agreement has been found in each case with previous published works. (Ray
et al. 1995) have analytically studied the non-linear vibration behavior of isotropic
skew plates. (Han and Dickinson 1997) applied Ritz approach for analyzing the
vibration of symmetrically laminated composite skew plates. Using a differential
quadrature (DQ) method, large amplitude free vibration analysis of laminated
composite skew thin plates have been considered by (Malekzadeh 2007). Nonli-
near vibration of symmetrically laminated composite skew plates by Finite ele-
ment method have been considered by (Singha an Daripa 2007). Recently, large
deformation flexural behavior of laminated composite skew plates using an analyt-
ical approach have been considered by (Upadhyay and Shukla 2012). The aim of
this paper is to apply the theoretical model developed by (Benamar and Bennouna
1990) to analyze the geometrical non-linear free dynamic response of CFRP
symmetrically laminated plates in order to investigate the effect of non-linearity
on the non-linear resonance frequencies and the non-linear fundamental mode
shape at large vibration amplitudes. The general formulation of the model for non-
linear vibration of laminated plates at large vibration amplitudes is presented. Ful-
ly clamped boundaries have been considered here and Periodic displacement was
assumed. Results were compared to the previous ones in the literatures. The first
non-linear mode shape is examined. The relationships between the non-linear re-
sonance frequency ratio ωnl/ωl and the vibration amplitudes for the first non-linear
mode shape, for various plate skew angle, fiber orientation and aspect ratios are
discussed.
Geometrically Non-linear Free Vibration 445
2 Geometrically Non-linear Free Vibration of Fully Clamped

Symmetrically Laminated Composite Skew Plates
2.1 Constitutive Equation of a Laminated Skew Plate at Large

Deflections
Consider the skew plate with a skew angle θ shown in Fig.1. For the large vibra-
tion amplitudes formulation developed here, it is assumed that the material of the
plate is elastic, isotropic and homogeneous. The thickness of the plate is consi-
dered to be sufficiently small so as to avoid the effects of shear deformation. The
skew plate has the following characteristics: a, b, S: length, width and area of the
plate; x-y: plate co-ordinates in the length and the width directions; ξ-η, H: Skew
plate co-ordinates and plate thickness; E, ν: Young’s modulus and Poisson’s ratio;
D, ρ: plate bending stiffness and mass per unit volume.
y η
θ b
x, ξ
a
Fig. 1 Skew plate in x-y and ξ-η co-ordinate system
For the classical plate laminated theory, the strain-displacement relationship for
large deflections are given by:
λ (1)
In which , and λ are given by:

U W W
k λ
V W
ε ; k k ; λ λ W
(2)
Y
U V k W λ W W
U, V and W are displacements of the plate mid-plane, in the x, y and z direc-

tions respectively. For the laminated plate having n layers, the stress in the Kth
layer can be expressed in terms of the laminated middle surface strains and
curvatures as:
σ Q ε (3)
In which and terms of the matrix can be obtained by

the relationships given in reference (Timoshenko 1975). The in-plane forces and
bending moments in a plate are given by:
N A B ε λ
(4)
M B D k
A, B and D are the symmetric matrices given by the following Equation 5.

0 for symmetrically laminated plates (Mallick 1993).
H
A ,B ,D H Q 1, z, z dz (5)
Here the are the reduced stiffness coefficients of the kth layer in the plate
co-ordinates. The transverse displacement function W may be written as in refer-
ence (Ribeiro and Petyt 1999) in the form of a double series:
T
W A W sinkωt (6)
Where A T a ,a ,…,a is the matrix of coefficients corresponding

to the kth harmonic, W T w ,w ,…,w is the basic spatial functions
matrix, k is the number of harmonics taken in to account, and the usual summation
convention on the repeated index k is used. As in reference (Benamar, Bennouna
and White 1993), only the term corresponding to k=1 has been taken into account,
which has led to the displacement function series reduced, to only one harmonic:
i.e.,
W a w x, y sinωt (7)
Here the usual summation convention for the repeated indexes i is used. i is
summed over the range 1 to n, with n being the number of basic functions consi-
dered. The expression for the bending strain energy Vb, axial strain energy Va and
kinetic energy T are given in reference (Harras 2001) in the rectangular co-
ordinate (x,y). The skew co-ordinates are related to the rectangular co-ordinate
(ξ,η) by: ξ=x-y tanθ ; η=y/cosθ. So, the strain energy due to bending Vb, axial
strain energy Va and kinetic energy T are given in the ξ-η co-ordinate system. In
the above expressions, the assumption of neglecting the in plane displacements U
and V in the energy expressions has been made as for the fully clamped rectangu-
lar isotropic plates analysis considered in reference (Benamar, Bennouna and
White 1993). Discretization of the strain and kinetic energy expressions can be
carried out leading to:
V sin ωt a a k ; V sin ωt a a a a b ;T ω cos ωt a a m (8)
In which mij, kij and bijkl are the mass tensor, the rigidity tensor and the geome-
trical non-linearity tensor respectively. Non-dimensional formulation of the non-
linear vibration problem has been carried out as follows.
ξ η
w ξ, η Hw , Hw ξ , η (9)
ξ η
Where ξ and η are non-dimensional co-ordinates ξ and η one then
obtains:
H E H E
k k ;b b ;m ρH abm (10)
Where the non-dimensional tensors m*ij, k*ij and b*ijkl are given in terms of in-
tegrals of the non-dimensional basic function wi*, non-dimensionnal extensional
and bending stiffness coefficient A*ij and D*ij , skew angle θ and aspect ratio α.
Upon neglecting energy dissipation, the equation of motion derived from Ham-
ilton’s principle is:
δ V T 0 (11)
Where V=Va+Vb. Insertion of Equations 8 into Equation 11, and derivation
with respect to the unknown constants ai, leads to the following set of non-linear
algebraic equations:
2a k 3a a a b 2ω a m 0, r=1…n. (12)
These have to be solved numerically. To complete the formulation, the proce-

dure developed in (Harras 2001) is adopted to obtain the first non-linear mode. As
no dissipation is considered here, a supplementary equation can be obtained by

applying the principle of conservation of energy, leads to the equation:
/
ω (13)
This expression for ω*2 is substituted into Equation 12 to obtain a system of n

non-linear algebraic equations leading to the contribution coefficients ai, i=1 to n.
ω and ω* are the non-linear frequency and non-dimensional non-linear frequency
parameters related by:
D
ω ρ
ω (14)
To obtain the first non-linear mode shape of the skew plate considered, the con-
tribution of the first basic function is first fixed and the other basic functions con-
tributions are calculated via the numerical solutions of the remaining (n-1)
non-linear algebraic equations.
The aim of this section is to apply the theoretical model presented above to ana-
lyze the geometrical non-linear free dynamic response of skew fully clamped
symmetrically laminated plates in order to investigate the effect of non-linearity
on the non-linear resonance frequencies and non-linear fundamental mode shape
at large vibration amplitudes. Convergence studies are carried out, and the results
are compared with those available from the literature through a few examples of
laminated composite skew thin clamped plates with different fiber orientation and
aspect ratio. The material properties, used in the present analysis are: (1) Plate 1,
5-layers symmetrical angle-ply (45°,-45°, 45°,-45°, 45°); EL/ET = 40.0; GLT/ET =
0.6; νLT = 0.25. (2) Plate 2, 8-layers symmetrically laminated plate (90°,-
45°,45°,0°)sym ; EL=120.5 Gpa; ET=9.63 Gpa; GLT=3.58 Gpa; νLT = 0.32. (3) Plate
3, 16-layers symmetrically laminated plate (45°,-45°,0°,-45°,45°,-45°,0°,45°)sym ;
EL=173Gpa; ET=7.2 Gpa; GLT=3.76 Gpa; νLT = 0.29. Where E, G and ν are
Young’s modulus, shear modulus and Poisson’s ratio. Subscripts L and T
represent the longitudinal and transverse directions respectively with respect to the
fibers. All the layers are of equal thickness. Calculation was made by using 18
functions corresponding to three symmetric beam functions in the ξ direction and
three symmetric beam functions in the η direction, and three anti-symmetric beam
functions in the ξ direction and three anti-symmetric beam functions in the η di-
rection. Table 1 shows the linear results for a fully clamped laminated skew com-
posite plate 2 for skew angle θ=30° and aspect ratio α=0.6 obtained using only 18
well-chosen plate functions. It can be seen a good convergence with results pre-
sented in reference (Harras 2001). The variation of non-dimensional nonlinear
Table 1 First linear mode shape of a fully clamped laminated skew composite plate 2
(Skew angle =30°, α=0.6): typical numerical results obtained with 18 basic functions
ωl* 9.4343
1 a11 1.05 10 a42 -2.591686E-02
2 a13 -1.549667E-02 11 a44 -2.517736E-03
3 a15 -1.665776E-03 12 a46 -3.854766E-04
4 a22 -1.689070E-01 13 a51 -1.109130E-02
5 a24 -4.314628E-03 14 a53 1.227799E-02
6 a26 -3.644569E-04 15 a55 1.692331E-03
7 a31 -2.185097E-02 16 a62 1.826523E-03
8 a33 2.017160E-02 17 a64 -3.078395E-03
9 a35 1.995894E-03 18 a66 -1.112399E-03
Table 2 Comparison of non-linear frequency ratio ωnl*/ ωl* of plate 2 for θ= 0°, α=2/3,
between results obtained in references (Han and Petyt 1997) and (Harras 2001) associated
with the fundamental non-linear mode shape and present results
Am/H results (Han results (Harras 2001) Present Error (%)

and Petyt 1997) results
0.2 1.0058 1.0054 1.0054 0,020
0.4 1.0232 1.0214 1.0214 0.088
0.6 1.0516 1.0475 1.0466 0.281
0.8 1.0903 1.0826 1.0821 0.400
1.0 1.1382 1.1267 1.1271 0.472
1.2 1.1941 1.1759 1.1743 0.903
Table 3 Convergence and accuracy of the nonlinear frequency ratios (ωnl/ωl) of clamped
5-layered angle-ply [45°/-45°/45°/−45°/45°] skew plate 1 (b/a = 1)
Skew Wmax/h
Angle θ°
0.2 0.4 0.6
0° Present 1.0074 1.0286 1.0641
(Singha, Daripa 2007) 1.0068 1.0271 1.0599
Error% 0.06% 0.14% 0.39%
15° Present 1.0077 1.0299 1.0675
(Singha, Daripa 2007) 1.0076 1.0300 1.0663
Error% 0.01% 0.01% 0.11%
30° Present 1.0086 1.0342 1.0741
(Singha, Daripa 2007) 1.0086 1.0340 1.0749
Error% 0.0016% 0.01% 0.08%
45° Present 1.0093 1.0379 1.0814
(Singha, Daripa 2007) 1.0098 1.0387 1.0849
Error% 0.0519% 0.08% 0.32%
Table 4 (continued)
Skew Wmax/h
Angle θ°
0.8 1 1.2
0° Present 1.1076 1.1609 1.2189
(Singha, Daripa 2007) 1.1041 1.1582 1.2214
Error% 0.32% 0.24% 0.20%
15° Present 1.1139 1.1710 1.2372
(Singha, Daripa 2007) 1.1149 1.1744 1.2431
Error% 0.10% 0.29% 0.48%
30° Present 1.1275 1.1944 1.2607
(Singha, Daripa 2007) 1.1297 1.1963 1.2729
Error% 0.19% 0.15% 0.96%
45° Present 1.1401 1.2096 1.2935
(Singha, Daripa 2007) 1.1458 1.2215 1.3076
Error% 0.50% 0.98% 1.08%
frequency ratio ωnl/ωl with respect to non dimensional maximum amplitude wmax/h
is evaluated for clamped square plate and is shown in Tables 2, along with pub-
lished results. It is observed that present results are in close agreement with those
of assumed hierarchical finite element method (Han and Petyt 1997) and Energy
balance method (Harras 2001). Next, the nonlinear free vibration behaviors of ful-
ly clamped angle-ply [45°/-45°/45°/-45°/45°] thin square and skew plates are ex-
amined in Table 3. It is observed that the results obtained here closely matches
with those from (Singha, Daripa 2007) for the case of symmetrically angle ply la-
minated composite skew plates. It is observed that the nonlinear frequency ratio
increases with the increase in amplitude of vibration, indicating hardening type of
nonlinear behavior. Furthermore, for the chosen amplitude, it can be noted that,
with the increase in skew angle, the degree of nonlinearity is high.
Fig. 2 Comparison of backbone curve of plate 1, 2, 3 and isotropic skew plate for θ=30°
In figure 2 comparison is make between non-linear frequency of the skew

symmetrically angle ply laminated plate 1, the skew symmetrically laminate plate
2 and plate 3 and the skew isotropic plate for a skew angle θ=30°. It can be seen
that plate 1 has a hardening effect compared to the isotropic one. Also the plate 2
and plate 3 are less hard than the isotropic skew plate.
4 Conclusion
The theoretical model established and applied to beams, plates and shells ((Bena-
mar, Bennouna and White 1993) has been successfully applied to calculate the
first non-linear mode shape of fully clamped skew symmetrically laminated plates
for various aspect ratios. The present formulation has been verified with the avail-
able analytical and finite element results. Limited numerical studies are conducted
to examine the effect of skew angle, fiber orientation, and aspect ratio on the
large-amplitude frequency of composite skew plates. As has been shown in this
work, the fundamental non-linear mode shape of the fully clamped skew symmetr-
ically laminated plate can be obtained with sufficient accuracy by using 18 basic
plate functions. This provides a more accurate description of the non-linear mode
shape, compared with previous results based on the use of nine symmetric basic
plate functions, since it allows the non-symmetry induced by the fiber orientation
to be taken into account. Present study reveals hardening type of nonlinearity and
the nonlinear frequency ratio in general increases with the increase in skew angle.
Further work is needed to investigate the behaviour of higher modes, the effects of
fiber orientation, and forced response of fully clamped symmetrically laminated
skew plates.
References
Leissa, A.W.: Vibration of Plates. US Government Printing Office, NASA SP-160, Wash-
ington, DC (1969)
Benamar, R.: Non-linear dynamic behaviour of fully clamped beams and rectangular iso-
tropic and laminated plates. Ph.D. Thesis, Institute of Sound and Vibration Research
(1990)
Chia, C.Y.: Geometrically nonlinear behavior of composite plates: a review. Applied Me-
chanical Reviews 41, 439–451 (1988)
Han, W., Petyt, M.: Linear vibration analysis of laminated rectangular plates using the hie-
rarchical finite element method-I: free vibration analysis. Computers and Structures 61,
705–712 (1996)
Ribeiro, P., Petyt, M.: Multi-modal geometrical non-linear free vibration of fully clamped
composite laminated plates. Journal of Sound and Vibration 225, 127–152 (1999)
Harras, B.: Theoretical and Experimental Investigation of the non-Linear Behaviour of var-
ious Fully Clamped Rectangular Composite Panels at large Vibration Amplitudes. Ph.D.
thesis, University Mohammed V-Agdal (2001)
Harras, B., Benamar, R., White, R.G.: Investigation of non-linear free vibrations of fully
clamped symmetrically laminated carbon-fibre-reinforced PEEK (AS4/APC2) rectangu-
lar composite panels. Composites Sciences and Technology 62, 719–727 (2002)
Han, W., Dickinson, S.M.: Free vibration of symmetrically laminated skew plates. Journal
Malekzadeh, P.: A differential quadrature nonlinear free vibration analysis of laminated
composite skew thin plates. Journal of Thin-Walled Structures 45, 237–250 (2007)
Singha, M.K., Daripa, R.: Nonlinear vibration of symmetrically laminated composite skew
plates by finite element method. Int. J. Non-linear Mech. 42, 1144–1152 (2007)
Ray, A.K., Banerjee, B., Bhattacharjee, B.: Large amplitude free vibrations of skew plates
including transverse shear deformation and rotary inertia-a new approach. J. Sound
Vib. 180, 669–681 (1995)
Upadhyay, A.K., Shukla, K.K.: Large deformation flexural behaviour of laminated com-
posite skew plates: An analytical approach. Composite Structures 94, 3722–3735 (2012)
Benamar, R., Bennouna, M.M.K., White, R.G.: The effects of Large Vibration Amplitudes
on the Fundamental Mode Shape of a Fully clamped, Symmetrically laminated rectan-
gular plate. In: Fourth International Conference of Recent Advances in Structural Dy-
namics, pp. 749–760. Elsevier Applied Science (July 1990)
Mallick, P.K.: Fibre-reinforced composites: Materials, Manufacturing and Design, vol. 152.
Marcel Dekker, Inc., New York (1993); 2nd edn. revised and expanded
Benamar, R., Bennouna, M.M.K., White, R.G.: The effects of large vibration amplitudes on
the fundamental mode shape of thin elastic structures. Part II fully clamped rectangular
isotropic plates. Journal of Sound and Vibration 164, 295–316 (1993)
Han, W., Petyt, M.: Geometrically Nonlinear vibration analysis of thin, rectangular plates
using the hierarchical finite element method-II: 1st mode of laminated plates and higher
modes of isotropic and laminated plates. Computers and Structures 63, 309–318 (1997)
Timoshenko, S., Weinsowsky-Krieger, S., Jones, R.M.: Mechanics of Composite Materials.
International Student Edition, vol. 51. McGraw-Hill Kogakusha, Ltd., Tokyo (1975)
Scratch Adhesion Characteristics of PVD
Cr/CrN Deposited on High Speed Steel and
Silicon Substrates
Kaouthar Khlifi, Hafedh Dhifelaoui, and Ahmed Ben Cheikh Larbi
Ecole Nationale Supérieure des Ingénieurs de Tunis, Université de Tunis,

5 Rue Taha Hussein, 1008 Tunis, Tunisia
khlifi.kaouther@yahoo.fr,
{dhafedh,ahmed.cheikhlarbi}@gmail.com
Abstract. The substrate is one of the significant parameters of technology

controlling the mechanical properties of PVD coatings. The quality of the coating
is highly influenced by the adhesion. An improvement of these properties can
open new areas of applications. In this paper, the adhesion results for PVD Cr/CrN
on high speed steel and silicon substrates are presented. The adherence of the
coatings to the steel substrate was evaluated using a scratch test. In the scratch
test, the critical force as the beginning of spalling or delaminating of the coating
was determined. All the coatings show critical values higher than 7 N. The critical
load depends on the film thickness and the substrate composition.
Keywords: PVD thin films, CrN, scratch test, damage mechanisms.
1 Introduction
Today PVD thin films are widely used in order to increase the machining
performance because PVD coated tools have significantly contributed to higher
production rates and higher tool life's. Several parameters can be controlled to
improve mechanical behavior of thin films. The substrate material is one of the
significant parameters of technology controlling the mechanical properties of PVD
coatings. In order to develop new PVD coatings, it is important to characterize the
mechanical properties of the coating/substrate system and the deformation
mechanisms in a tribological contact. Several methods are available to examine
the efficiency of this system [1]. But the study of adhesion is still essential to
examine coated systems interface. In this context scratch test is probably the most
common and used test to understand coating failure modes of a hard coating
substrate system [2]. During the conventional scratch test a diamond indenter
Rockwell C flattened the surface of the coated material under an increasing
normal load until failure occurs or the damage of the film is noted. The
identification the coating failure mechanisms can be made by the following terms:

454 K. Khlifi, H. Dhifelaoui, and A. Ben Cheikh Larbi
LC1, the lower critical load, which is defined as the load where the first cracks
occurred (cohesive failure), LC2, the upper critical load at which the first
delaminating [3] and LC3 the critical load when the damage of the film exceeds
50%. Critical normal loads can be controlled by the coating thickness, coating
hardness and substrate hardness .The present study focuses on the adhesion of Cr
single layer and CrN multilayer deposited on 100C6 Steel substrate and Silicon
substrate by magnetron sputtering system.
2 Experimental Details
Substrates were used in this study is 100C6 steel (AISI 52100) and Si. Cr/Si single
layer, CrN/Cr/Si multilayer and CrN/Cr/100C6 coatings were deposited with a
magnetron sputter unit with a pure chromium target. Thicknesses of thin films
studied were illustrated in table 1. The morphological and microstructural
characterizations were evaluated using an atomic force microscopy (AFM) and a
Scanning electron microscopy (SEM). The scratching response of the coating–
substrate composites were evaluated using the CSM Instruments with Rockwell C
diamond styli with radius 100μm. A normal load range of 3–28N, a scratch length
of 3 mm and a scratching velocity of 6mm/min were used in the experiments.
Hardness of 100C6 steel and Silicon substrate are respectively 11GPa and 8GPa.
Young's modules of both substrates are respectively 200 GPa and 130 GPa.
Table 1 Thickness and roughness of Cr/Si, CrN/Cr/Si and CrN/Cr/100C6 coatings

Coatings Thickness (μm) Roughness (nm)
Cr/Si 1,6 10
CrN/Cr/Si 2 4
CrN/Cr/100C6 3,2 14
3.1 Morphological and Microstructured Examination

Results obtained by AFM observation are illustrated in Figure 1.
The AFM examination revealed the presence of domes and craters which are
uniformly distributed over the surface. However, we have noticed a difference in
the size of the domes and craters that characterize the morphology. The passage
from coating pure chromium based to another nitrogen rich, a less rough surface is
observed which was characterized by bigger and larger grains. With the
introducing of nitrogen in the plasma roughness pass from 10 for Cr to 4 nm for
CrN. Referring to the structure of zones described by Thornton [4], we go from
the T-zone (less dense Zone) to a zone of type II with peaks pyramidal columns.
Coating de-posited on steel appears rougher due to the high substrate surface
Scratch Adhesion Characteristics of PVD Cr/CrN Deposited on High Speed Steel 455
roughness when comparing with Si surface roughness. Cross-section SEM of all

coatings showed a columnar structure as shown in Figure 2 similar to that
proposed by Thornton [4] in his model of thin film growth.
Fig. 1 Three-dimensional AFM images of Cr/Si single layer, CrN/Cr/Si and CrN/Cr/100C6
multilayer
Fig. 2 SEM micrographs CrN/Cr/Si single layer
3.2 Scratch Test

Scratch methodology was used to evaluate the relative adhesive and cohesive
failure. Progression of scratching is accompanied by successive degradations de-
scribed by: cohesive failure signaled by micro cracks at the edge, local chipping of
the coating from its substrate and, finally, a total loss of adhesion between the two
constituents. In this study, the corresponding critical forces are summarized in
table 2.
Table 2 Critical loads and friction coefficients of CrN/Cr/100Cr6, CrN/Cr/Si and Cr/Si
coatings
Coatings CrN/Cr/100C6 CrN/Cr/Si Cr/Si
LC1 5,68 3,23 4,27
LC2 12,79 7,42 9,16
LC3 13,54 12,34 13,16
Figures 3-5 shows the tracks were corresponding to the Cr/Si, CrN/Cr/Si and
CrN/Cr/100C6 coating after scratch test.
Fig. 3 Optical micrographs of the scratch track of Cr/Si coating
Fig. 4 Optical micrographs of the scratch track of CrN/Cr/Si coating

Scratch Adhesion Characteristics of PVD Cr/CrN Deposited on High Speed Steel 457
Fig. 5 Optical micrographs of the scratch track of CrN/Cr/100C6 coating
Under the same loading conditions, the coatings behavior is different. In the
single layer structure, with the increase of the scratch load, cracks have been
developed throughout the cross section of the film and have spread rapidly. In this
type of coating, adhesive failure occurred for low load comparing with the other
coatings and we observed a total chipping of the film and exposure of substrate at
higher load.
The critical load is the normal load at the first coating failure as detected. This
load was determined with combining observations of scratches by optical
microscopy with measurements of acoustic emission, of normal and tangential
forces and of the residual depth during scratch testing.
Several authors [5,6] have linked the appearance of the first crack (cohesive
failure) at a first critical load noted by LC1 and the upper critical load at which the
first delaminating at the edge of the scratch track occurred (adhesion failure) by LC2.
The critical loads LC1, LC2 and LC3 values were presented in figure 3-5. Higher
critical loads of adhesion were obtained for CrN/Cr/100Cr6. In multilayer
structure coatings, cracks develop mostly in the vicinity of the upper surface, and
the layers interfaces can change the initial crack orientation.
4 Conclusion
In this work, PVD thin films were deposited on different substrate materials,
100C6 (AISI 52100) steel and Si by PVD magnetron sputtering system. The mor-
phological and microstructural proprieties were evaluated using an atomic force
microscopy (AFM) and a Scanning electron microscopy (SEM). Adhesion quality
was studied with scratch tests. Damages mechanisms that were occurred during
scratching were further discussed. The AFM examination revealed the presence of
domes and craters which are uniformly distributed over the surface. CrN/Cr/Si has
a less rough surface because of a small grains size. SEM cross section of all
coatings showed a columnar structure
The results obtained in the present study show that the coating/substrate
damage mechanisms during scratching and the critical normal loads obtained are
strongly dependent on the mechanical properties of substrate material. The results
obtained in the present study show that the coating/substrate damage mechanisms
during scratching and the critical normal loads obtained are strongly dependent on
the mechanical properties of substrate material.
Acknowledgements. This work is partially supported by BALZER FRANCE. The authors

also gratefully acknowledge the helpful comments and suggestions of the reviewers, which
have improved the presentation.
References
[1] Sveen, S., Andersson, J.M., M’Saoubi, R., Olsson, M.: Scratch adhesion characteristics
of PVD TiAlN deposited on high speed steel, cemented carbideand PCBN substrates.
Wear 308, 133–141 (2013)
[2] Holmberg, K., Laukkanen, A., Ronkainen, H., Wallin, K., Varjus, S.: A model for
stresses, crack generation and fracture toughness calculation in scratched TiN- coated
steel surfaces. Wear 254, 278–291 (2003)
[3] Khlifi, K., Ben Cheikh Larbi, A.: Investigation of the adhesion of PVD coatings using
various approaches. Surface Engineering 29, 555–560 (2013)
[4] Thornton, J.A.: J. Vac. Sci. Technol. 11, 666–670 (1974)
[5] Yıldız, F., Alsaran, A., Celik, A., Efeoglu, I.: Surf. Eng. 26(6), 578–583 (2010)
[6] Kataria, S., Kumar, N., Dash, S., Ramaseshan, R., Tyagi, A.K.: Surf. Coat.
Technol. 205, 922–927 (2010)
Load Sharing Behavior in Planetary Gear Set
Ahmed Hammami1,2, Miguel Iglesias Santamaria2, Alfonso Fernandez Del Rincon2,

Fakher Chaari1, Fernando Viadero Rueda2, and Mohamed Haddar1
1
Laboratory of Mechanics, Modeling and Production (LA2MP),
National School of Engineers of sfax, BP1173 – 3038 – Sfax – Tunisia
ahmed_hammami@voila.fr, fakher.chaari@gmail.com,
mohamed.haddar@enis.rnu.tn
2
Department of Structural and Mechanical Engineering,
Faculty of Industrial and Telecommunications Engineering, University of Cantabria,
Avda de los Castros s/n 39005 – Santander - Spain
miguel.iglesias@unican.es, alfonso.fernandez@unican.es,
fernando.viadero@unican.es
Abstract. The objective of this paper is to study the effects of meshing phase be-
tween planets, the effects of the gravity of carrier and the planet position error on
the load sharing behavior in planetary gear set. These effects will be studied nu-
merically under the stationary condition and will be validated experimentally by a
back-to-back planetary gear test bench. In this test bench, strain gauges are in-
stalled on each planet’s pin hole in order to compare strains in the pinhole of each
planet.
Keywords: Planetary gear, load sharing, meshing phase, gravity, planet position
error.
1 Introduction
Planetary gear can transmit higher power because they use multiple power paths
formed by each planet branches. This allows the input torque to be divided be-
tween the n planet paths, reducing the force transmitted by each gear mesh.
Under ideal conditions, each planet path carries an equal load. Nevertheless,
planetary gears have inevitable manufacturing and assembly errors. So, the load is
not equally shared amongst the different sun-planet and planet-ring paths, which
can be a problem in terms of both dynamic behavior and durability.
Many researchers have done significant works on the subject of planetary gear
load sharing. Their works were based on transmission modeling and assessed by
experimental tests. Kahraman (Kahraman, 1994) used a discrete model to study
the influence of carrier pin hole and planet run-out errors on planet load-sharing
characteristics of a four-planet system under dynamic conditions. He employed
later (Kahraman, 1999) a planet load-sharing model to determine the static planet

460 A. Hammami et al.
load sharing of four-planet systems and presented experimental data for validation
of the model predictions. Iglesias (Iglesias et al, 2013) studied the effect of planet
position error on the load sharing and transmission error. Singh (Singh, 2005)
found that the tangential pin position error has a greater effect on the load sharing
than the radial pin position error. Ligata (Ligata et al, 2008) proved experimentally
that for the same amount of error, the degree of inequality in the planet load-
sharing behavior increases with the number of planets in the system. Guo and Kel-
ler (Guo and Keller 2012) presented a three-dimensional dynamic model which
take into account to the addressing gravity, bending moments, fluctuating mesh
stiffness, nonlinear tooth contact, and bearing clearance. They validated this mod-
el against the experimental data.
In this paper, the effects of meshing phase, the gravity and the error position of
the pin hole of planets on the load sharing characteristics are studied numerically
and validated experimentally with back-to-back planetary gear.
2 Description of the Test Bench
The test bench is composed of two identical planetary gear sets with the same gear
ratio (Fig. 1). The first planetary gear is a “test gear set” and the second planetary
gear is a “reaction gear set” having a free ring. An arm is fixed on this ring and al-
lowing the introduction of external load ((Hammami et al 2015) and (Hammami
et al 2014a)).
Fig. 1 Back-to-back planetary gear test bench

Load Sharing Behavior in Planetary Gear Set 461
The two planetary gears are connected back-to-back in order to provide a me-
chanical power circulation: the sun gears of both planetary gear sets are connected
through a common shaft and the carriers of both planetary gear sets are connected
to each other through a rigid hollow shaft (Hammami et al 2014b).
In order to compare the load sharing between the tests planets, three strain
gauges are installed in the pin holes of each planet in the tangential direction of
the test carrier (Fig. 2).
Fig. 2 Strain gauges mounted in each pin hole of the test carrier planets
The wires from the strain gauges are connected to the acquisition system
through a hollow slip ring (HBM SK5/95) which is installed with the hollow shaft
that connects the carriers (Fig.3).
Strains gauges are used in quarter bridge configuration.
Fig. 3 Instrumentation layout

Strain signals registered by strain gauges will be acquired by Programmable

Quad Bridge Amplifier module (PQBA) of the acquisition system “LMS
SCADAS 316 system”. This module can support four channels of strain transduc-
ers, piezo-resistive or variable capacitor sensors. Every channel is connected by 6
poles LEMO connector.
The data will be processed with the software “LMS Test.Lab” to visualize
time history of strains.
Additionally, an optic tachometer (Compact VLS7) is combined with pulse
tapes along the axe in order to measure its instantaneous angular velocity. It was
placed along the hollow shaft of carriers.
3 Numerical Results
The effect of meshing phase, the gravity of planets carrier and the position error of
planets on the load sharing behavior is studied in this part.
First of all, we define the planet load sharing ratio (LSR) as the ratio of the
meshing torque due to sun-planet(i) and ring-planet(i) meshes of planet (i) by the
meshing torque of all planets.
Tmesh ( Pi )
L Pi = n
(1)
∑ Tmesh ( Pi )
i =1
3.1 Effect of Meshing Phase

For the case of equally spaced planets and in phase meshes gear (sun planets and
ring planets), the planet load sharing factor is equal to 1/N (N: number of planets).
In our case, planets are equally spaced and gear meshes functions are sequen-
tially phased.
Z jψ i N
≠ n and ∑ Z jψ i = m π ( j = r , s) (2)
2π i =1
Z j is the number of tooth of the gear (j). ψ i is the angle position of the planet
(i). n and m are integer.
The gear meshes stiffness of ring-planets gear pairs (Fig. 4) and sun-planets
gear pairs are modeled using the finite element model (Fernandez Del Rincon et al
2013).
The dynamic response is computed according to the procedure given in (Ka-
hraman, 1994) and the load sharing ratio is computed according to equation (1).
Fig.5 shows the planet load sharing ratio for the nominal position of planets (fault-
less system) for 100 N.m of input torque and for speed of motor 165 rpm.
8
x 10
7.5
ring-planet1
ring-planet2
7 ring-planet1
Gear mesh stiffness (N/m) 6.5
5.5
4.5
4
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
Time (s)
Fig. 4 Gear mesh stiffness ring-planets
0.355
Planet1
0.35 Planet2
Planet3
0.345
0.34
0.335
LSR
0.33
0.325
0.32
0.315
0.31
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
Time (s)
Fig. 5 Planet load sharing ratio for the nominal position of planets
In this case, the LSR for all planets fluctuates slightly around the 1/N value
(N=3: number of planets) because the number of tooth in contact changes. The
fluctuation of LSR of each planet is the same with a phase shift of 2π/3. This
phase is induced by the fact that gear mesh sun-planets and ring-planets are
sequentially phased.
3.2 Effect of Gravity of Carrier

The gravity of carriers has effect on the distances between ring-planets and sun-
planets so that they are variable during the running of gearbox. Then, the values of
gear mesh stiffness decrease as the distance between gears (sun-planets and
ring-planets) increase and vice versa.
Fig.6 shows the evolution of gear mesh stiffness ring-planet1 during one period
of rotation of carrier with effects of gravity of carrier. This period can be divided
into four subperiods:
- The first subperiod : where the planet 1 is in the upper position. The values of
gear mesh stiffness are low.
- The second subperiod where the planet 1 is in an intermediate position in (the
left). The values of gear mesh stiffness are medium.
- The third subperiod where the planet 1 is in a lower position. The values of
gear mesh stiffness are high.
- The fourth subperiod where the planet 1 is again in an intermediate position
(the right). The values of gear mesh stiffness are medium.
8
x 10
7
ring-planet1
6.5
Gear mesh stiffness (N/m)
5.5
4.5
4
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
Time (s)
Fig. 6 Evolution of gear mesh stiffness ring-planets with gravity effects
0.5
Planet1
Planet2
0.45 3 2 1 3 Planet3
0.4
LSR
0.35
0.3
0.25
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8

Time (s)
Fig. 7 Planet load sharing ratio with gravity effects
The gravity of carrier will lead to a periodic fluctuation of the maximum ampli-
tude of gear mesh stiffness with a period corresponding to that of the rotation of
the carrier Tc. The higher values of the gear mesh stiffness of ring-planets are suc-
cessively on the planet 3 then planet 2 then planet 1.
Fig.7 shows the evolution of LSR during one period of rotation of carrier. This
period is divided into three sub-periods of Tc/3. For each sub-period, the LSR of
each planet does not oscillate with the same amplitude; there is always a planet
that will include the maximum (planet1then planet2 then planet3).
3.3 Effect of Planet Position Error

If a planet has an error “e” on the position of its pin hole, and all other planets are
at their ideal position, then the force due to this error is given by (Singh, 2010):
Fe = K eff .e (4)
Keff is the cumulative stiffness due to meshing stiffness of the contact at the
sun–planet Kps and planet–ring Kpr, and the planet bearing stiffness Kb.
Keff is defined as (Ligata et al, 2009):
1 1 1
= + (5)
K eff K b K ps + K pr
In our case, the planet 1 has an error “e1=62μm” and the planet 2 has an error
“e2=-60μm”. The LSR in this case is represented in Fig.8.
0.7
0.6
Planet 1
0.5 Planet 2
Planet 3
0.4
LSR
0.3
0.2
0.1
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
Time (s)
Fig. 8 Planet load sharing ratio with planets position errors
The position errors of planets 1 and 2 have an important effect in the LSR. Pla-
net 1 which has a positive error anticipates the contact, being preloaded before
planets 2 and 3 begin to transmit load; whereas planet 3 which has a negative error
is preloaded after planet 2 and 1.This defects are in tangential direction and they
have an effect more important in the LSR than the effect of gravity of carrier
which is a variable radial error (Bodas and Kahraman 2004).
4 Correlation with Experimental Results

Strain-time histories for three planets system having errors “e1=62μm” and “e2=-
60μm” are shown in Fig.9. It is clear that the positioning error has an important ef-
fect in the strain of each pin hole of planet.
170.00 1.00
F Time Planet1
F Time Planet2
F Time Planet3
Amplitude
Amplitude
uE
10.00 0.00
0.00 s 1.84
Fig. 9 Variation of the measured planets strains for the speed motor 165 rpm
1
Planet1(Experiments)
0.9 Planet2(Experiments)
Planet3(Experiments)
0.8 Planet1(Numerical)
Planet2(Numerical)
0.7 Planet3(Numerical)
0.6
LSR
0.5
0.4
0.3
0.2
0.1
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
Time (s)
Fig. 10 Evolution of measured and calculated LSR
The computational results of the load sharing ratio are compared to the meas-
ured LSR (Fig. 10). In general, the calculated load sharing agrees with the meas-
ured data. In fact, the planet position errors have an important effect. In addition, a
slight deviation of the load sharing evolution due to the effect of gravity of carrier
is observed on the numerical and experimental results whereas the effect of mesh-
ing phase is observed only in the numerical results because the recorded signals
presents noise due to the contact between brushes and slip ring. So, a signal
processing was necessary.
5 Conclusion
In this paper, the planets load sharing behaviour is studied and correlated with
those obtained experimentally using strain gages on planets of a back-to-back
planetary gear test bench.
The effect that planets are sequentially phased is included to the model and as
results; the load sharing ratio for all planets fluctuates slightly around the 1/N val-
ue. The time evolution of LSR of each planet bends a little bit every Tc/3 by the
effect of gravity of carrier. However, the planets position errors have an important
effect on the LSR of each planet. The final numerical results with the three effects
agree with the measured strains on planets.
Future investigation will be focused mainly on the planets load sharing under
non-stationary condition (run-up and run-down regimes).
Acknowledgements. This paper was financially supported by the Tunisian-Spanish Joint
Project N° A1/037038/11.
The authors would like also to acknowledge project “Development of methodologies for
the simulation and improvement of the dynamic behavior of planetary transmissions
DPI2013-44860” funded by the Spanish Ministry of Science and Technology.
Acknowledgment to the University of Cantabria cooperation project for doctoral train-
ing of University of Sfax’s students.
References
Bodas, A., Kahraman, A.: Influence of carrier and manufacturing errors on the static lad
sharing behavior of planetary gear sets. Bulletin of the Japan Society of Mechanical En-
gineers 47(3), 908–915 (2004)
Fernandez del Rincon, A., Viadero, F., Iglesias, M., García, P., de-Juan, A., Sancibrian, R.:
A model for the study of meshing stiffness in spur gear transmissions. Mechanism and
Machine Theory 61(2013), 30–58 (2013)
Guo, Y., Keller, J.: Combined effects of gravity, bending moment, bearing clearance, and
input torque on wind turbine planetary gear load sharing. American Gear Manufacturers
Association technical paper, 12FTM05 (2012)
Hammami, A., Fernández, A., Viadero, F., Chaari, F., Haddar, M.: Modal analysis of back-
to-back planetary gear: experiments and correlation against parameter model. Journal of
Theoretical and Accepted Applied Mechanics 53(1) (2015)
Hammami, A., Fernández, A., Viadero, F., Chaari, F., Haddar, M.: Dynamic behaviour of
back-to-back planetary gear in run up and run down transient regimes. Journal of Me-
chanics (2014a), doi:10.1017/jmech.2014.95
Hammami, A., Fernández, A., Chaari, F., Viadero, F., Haddar, M.: Dynamic behaviour of
two stages planetary gearbox in non-stationary operations. Mechatronic System: Theory
and Application, 23–35 (2014b)
Iglesias, M., Fernandez, A., De-Juan, A., Sancibrian, R., Garcia, P.: Planet position errors
in planetary transmission: Effect on load sharing and transmission error. Frontiers Me-
chanical Engineering 8(1), 80–87 (2013)
Kahraman, A.: Load Sharing Characteristics of Planetary Transmissions. Mechanism and
Machine Theory 29, 1151–1165 (1994)
Kahraman, A.: Static Load Sharing Characteristics of Transmission Planetary Gear Sets:
Model and Experiment. SAE Paper No. 1999–01–1050 (1999)
Ligata, H., Kahraman, A., Singh, A.: An experimental study of the influence of manufactur-
ing errors on the planetary gear stresses and planet load sharing. Journal of Mechanical
Design 130 (2008)
Ligata, H., Kahraman, A., Singh, A.: A closed-form planet load sharing formulation for
planetary gear sets using a translational analogy. Journal of Mechanical Design 131,
021007 (2009)
Singh, A.: Application of a system level model to study the planetary load sharing behavior.
Journal of Mechanical Design 127, 469–476 (2005)
Singh, A.: Load sharing behavior in epicyclic gears: Physical explanation and generalized
formulation. Mechanism and Machine Theory 45, 511–530 (2010)
Low Velocity Impact Behavior of Glass Fibre-
Reinforced Polyamide
Jamel Mars1, Mondher Wali1, Remi Delille2, and Fakhreddine Dammak1
1
Mechanical Modelisation and Manufacturing Laboratory (LA2MP),
National Engineering School of Sfax, B.P W3038, Sfax, University of Sfax, Tunisia
{Jamelmars,mondherwali}@yahoo.fr, Fakhreddine.Dammak@enis.rnu.tn
2
Laboratory of Industrial and Human Automation control (LAMIH / C2S),
University of Valenciennes
remi.delille@univ-valenciennes.fr
Abstract. The low velocity impact behavior of composites made of polyamide

(PA) as matrix and glass fibre as reinforcement has been investigated. The as-
sessment of the impact behavior has driven the need to perform tensile tests to
determine the elasto-plastic behavior of the composites. The specimens were
manufactured by injection molding techniques for the experimental tensile testing.
ABAQUS/EXPLICIT for finite element modeling is employed in order to predict
the impact behavior of glass fibre-reinforced polyamide. The determinations of the
impact force history and elasto-plastic structure deflection are the most important
objectives in impact engineering structures design.
Keywords: Mechanical Composite, PA66, mechanical characteristics.
1 Introduction
Fibre-reinforced composites are widely used in many engineering applications

such as automobile and aerospace due to their light weight, high strength and
stiffness, resistance to chemical and environmental agents, design freedom and
manufacturing advantages. During their manufacture, maintenance and service
life, this Fibre-reinforced composites may suffer different damage, of which low
velocity impact is considered as one of the most important and dangerous, because
it can induce internal damage in the form of delamination, matrix cracking, local
permanent deformation, debonding and fiber breakage, leading to a significant
strength reduction of the structure.
Numerous investigations dedicated to the problem of the impact behavior of
composite materials are performed by several authors such as (Wali et al. 2011),
(Dhakal et al. 2012) and (Yang et al. 2013). The published results show that the
impact resistance of composite structures depends, in a complex way, on the prop-
erties of composite structures (material, thickness, laminate stacking sequence in

470 J. Mars et al.
case of laminated composites), properties of impactor (mass, velocity, energy,

shape of impacting head) and experimental setup (clamping conditions of the test
specimen) (Giangiacomo et al. 2012). In addition, research has shown that compo-
sites are capable of absorbing energy and dissipating it by various fracture and
elastic processes when subjected to a low velocity impact (Cantwell and Morton
1991).
A number of investigations were conducted to assess the impact on thermoplas-
tic composites reported in the works of (Xu et al. 2011) and (Simeoli et al. 2014).
Glass fiber-reinforced polyamide (PA66) is one of the most widely used thermop-
lastic composites in the engineering industry and the security imposes to perform
impact behavior. Therefore, it becomes necessary to know the behavior of this
composite material which is very sensitive to elasto-plastic impact promoting high
levels of strain, leading to failure and damage.
Despite increased use in engineering applications, few published work on the
impact behavior of polyamide short fiber-reinforced thermoplastic composites is
found. The fracture toughness and impact behavior of a range of glass-fibre-
reinforced polyamides of high fiber content, which were promoted as potential
metal replacement materials, are examined by (Akay et al. 1995). The compres-
sive behavior of Nylon 6 and Nylon 66, on wide range of strain-rates, was investi-
gated by (Benaceur et al. 2008) using Hopkinson bar tests. A localized low
velocity impact experiments and simulations were conducted by (Mouti et al.
2010, 2013) on glass fiber-reinforced polyamide automotive product to investigate
typical flying stones impact scenarios.
The purpose of this paper is to predict the behavior of an elasto-plastic glass fi-
bre-reinforced polyamide structures under low velocity impact. The mechanical
elasto-plastic model of glass fibre-reinforced polyamide materials for different
glass fibre volume fractions is then modeled basing on experimental tensile tests.
The model is implemented into the commercial finite element code
ABAQUS/Explicit to predict the Low velocity impact behavior of circular glass
fibre-reinforced polyamide. Numerical results, including contact force and dis-
placement of the composite plate subjected to low-velocity impact, are calculated.
2 Experimental
2.1 Materials
Two commercial polyamide 66 with 30 wt.% of short glass fibre (HERAMID A
NAT with 30 wt.% glass fibre, provided by RADICI GROUP) and natural polya-
mide 66 (A HERAMID NAT) were used to produce molded glass fibre-reinforced
composites with 00, 10, 20 and 30 wt.% .They were designed as PA66_00,
PA66_10, PA66_20 and PA66_30, respectively. The specimens were made by
injection molding (Figure 1).
Low Velocity Impact Behavior of Glass Fibre-Reinforced Polyamide 4771
Fig. 1 Molded samples
2.2 Tensile Tests

Mechanical testing consisted of tensile tests. The tensile tests have been carrieed
out using an Instron Electtropuls tensile testing machine with crosshead speed off 1
mm/min (Figure 2). Thee jaws used for attachment of the test specimens arre
Fig. 2 Instron Electropuls ten

nsile testing machine
Fig. 3 Shape and dimensionss of the specimen, according to ISO 527-2 standard
472 J. Mars et aal.
controlled pneumatically, 4 bars. Three specimens of each composition werre

tested and the average vallue reported. Strains were measured with a RUDLPH laa-
ser extensometer. Markerrs (white color and width 3 mm) are placed in each teest
specimen for measuring the
t deformation in the active area during the tensile tesst.
Specimens had the shapee and dimensions reported in Figure 3, according to ISO
527-2 standard. Tensile testing
t was carried out in air at 20°C and 50% relativve
humidity.
2.3 Analysis of fibrre Length Distributions

The fibre orientation distrribution in a material sample was examined by using thhe
X-Ray microtomography machine (Skyscan 1172). The local variation of the anni-
sotropy from the shell to the core of the molded tensile sample is captured and inn-
formation about the local average fibre orientation angle is observed.
Figure 4 shows that thhe orientation of fibers in assemblies is moderately influu-
enced by the processes ussed to manufacture the samples. Fibre content-images foor
specimens used in tensilee tests is examined and illustrated in figure 5 to evaluaate
the global anisotropy .Figgure 5 shows that, the amount of homogeneously distrri-
buted fibers is greater.
PA66_10 PA66_20 PA66_30

Fig. 4 Microtomography of reeinforced PA66 (sagittal scanning in the test specimen middle)
Fig. 5 Fiber content-images for specimens PA66_10, PA66_20, PA66_30

3.1 Quasi-static Characterization

Tensile test data was captured for different weight fractions: 0, 10, 20 and 30 wt%.
For each material tested, three specimens were used. The degree of repeatability
was found excellent in all cases. For example, in tensile tests of PA_00 at a strain
rate of 1 mm/min, this repeatability is illustrated in Figure 6. The medium curve of
three tests is chosen for all polyamides tested.
Tensile test data was then used to plot stress strain relations. Data captured
from the testing machine and the laser extensometer was plotted in Figure 7. Fig-
ure 7 shows Stress-strain curves of natural PA66_00, PA66_10, PA66_20 and
PA66_30. These results are in good agreement with literature (Ghorbel et al.
2011).
50
40
30
Stress (MPa)
Test 1
20 Test 2
10 Test 3
0
0 10 20 30 40
Strain %
Fig. 6 Repeatability of simple tensile test of PA66_00
120
PA66_00
100 PA66_10
PA66_20
80
PA66_30
Stress (MPa)
60
40
20
0
0 10 20 30 40
Strain %
Fig. 7 Uniaxial stress-strain curves of PA66_00, PA66_10 , PA66_20 and PA66_30
474 J. Mars et al.
The increasing of the fiber content has the effect:

• An improvement in elastic modulus
• An increase in tensile strength
• A reduction in fracture strain. The materials behavior changes from ductile to
brittle behavior.
3.2 Identification of the Hardening Law

The mechanical elasto-plastic model of glass fibre-reinforced polyamide compo-
sites for different glass fibre volume fractions must be modeled to accurately pre-
dict the low velocity impact responses. However, the least-squares method was
used to fit the Stress/strain curves to the experimental tensile data points and the
mechanical parameters of glass fibre-reinforced polyamide material are deter-
mined. The elastic-plastic material model was implemented in the commercial
nonlinear finite element code ABAQUS/EXPLICIT for studying the impact beha-
vior of polyamide.
The mechanical elasto-plastic model used during the identification phase is
defined by
− β .ε p
σ = σ e + Q (1 − e ) + K .ε p (1)
where σ e is the yield stress, Q, β and K are the mechanical parameters of the elas-
to-plastic model and ε p is the plastic deformation.
Finally, the comparison of the numerical approach with the experimental data
shows the accuracy of the mechanical elasto-plastic model. Figure 8 shows that
the numerical approach is in good argument with the experimental result for
PA66_20 material. This approached law is used in next section when studying the
impact behavior of glass fibre-reinforced polyamide.
Table 2 shows the values of the mechanical parameters for the glass fibre-
reinforced polyamide during the identification phase.
Table 1 Identified mechanical parameters of the reinforced polyamide
PA66_00 PA66_10 PA66_20 PA66_30

Yield stress e (MPa) 15 25 30 34
E (MPa) 1100 2350 4200 6000
Q (MPa) 23.5 35.1 47 72.6
β (MPa) 30 144 198 198
K (MPa) 72 83 103 135
80
60
Stress (MPa)
40 Experience
20 Num Approach
0
0 0.02 Strain0.04 0.06
Fig. 8 True Stress/strain curves for PA66_20
3.3 Finite Element Modeling and Validation of Impact Model

 Finite element modeling
Various procedures can be used in ABAQUS for modeling the elasto-plastic

impact phenomena. The accurate modeling requires an appropriate selection of
contact modeling, element type, resolution method and number of element in plate
thickness. The determinations of the impact force history and elasto-plastic struc-
ture deflection are the most important objectives in impact engineering structures
design. For studying the impact ABAQUS 6.10 version is used. A clamped circu-
lar plate impacted at its center by a cylindrical impactor with hemi-spherical nose
is considered. The plate with radius R=60mm and thickness h=1mm is made of
aluminum alloy. The initial velocity of impactor is V0 (Figure 9).
Proj
Rp
h V0
Fig. 9 Cylindrical projectile impacted circular plate
The presence of the rotational symmetry axis, can simplify the three-
dimensional impact problem in an axisymmetric problem. The problem can thus
be analysed using 2D axisymmetric model. The plate was meshed using the struc-
tured meshing technique and 6000 4-nodes per elements with reduced integration
(CAX4R in ABAQUS). The impactor mesh consisted of 433 4-nodes per elements
(CAX4R in ABAQUS).
476 J. Mars et al.
 Validation of the contact model
Given that the impact model needs reliable data for simulation, we need to va-
lidate the mechanical properties and the contact law introduced into numerical
model. For that, a comparative study between the numerical predictions and expe-
rimental response found in literature is established. The plate material is assumed
to be an isotropic and homogeneous, and the elastic, perfectly plastic material and
the Von Mises yield criterion, J2, is employed same as in the work of (Chen et al.
2007). In this section, the deformable impactor is assumed to be elastic. For the
validation, the mechanical properties of both impactor and plate are illustrated in
Table 2.
The impactor velocity predicted by the present numerical model and the result
obtained by experimental data tested through LDA technique (Chen et al. 2007)
are plotted in Figure 10. One can note from this figure that the impactor velocity is
in good agreement with the experimental test.
40
30
Velocity (m/s)
20
10
-10 Chen et al. 2007

Present Axi-model
-20
0 0.5 1
Time (ms)
Fig. 10 Validation: impactor velocity time responses (mp = 54.4 g, Rp=6.35mm, V0 = 29.9 m/s)
Table 2 Mechanical properties of impactor and circular plate
Parameters Material Young modulus Poisson’s Density Yield stress

E ( GPa) ratio ν (
ρ Kg / m3 ) σ e ( MPa)
Circular Plate Aluminum 69 0.3 2600 290
Impactor Steel 200 0.3 7800 -
3.4 Impact Behavior of Glass Fiber Reinforced Polyamide

The dynamic behavior of a clamped circular plate, of natural PA66_00, PA66_10,
PA66_20 and PA66_30, subjected to impact by a projectile is discussed in the fol-
lowing section. The target circular plate with radius R = 60mm and thickness
h = 4mm whose density, Young’s modulus, Poisson ratio, yield stress and σe are
already mentioned in the previous paragraph. The impactor has the same characte-
ristics described above except that the mass and speed become respectively 1.369
Kg and V0 = 3.418 m / s.
The evolution of plate center displacement according to the time is presented in
Figure 11 for incidental impact energy of 8J for PA66_00, PA66_10, PA66_20
and PA66_30.
Time (ms)
0 2 4 6 8 10 12 14
0.000
-0.002
Displacement (m)
-0.004
-0.006
PA66_00
-0.008
PA66_10
-0.010 PA66_20
-0.012 PA66_30
Fig. 11 Center plate displacement time responses for impact energy of 8J
3
PA66_00
PA66_10
Force MAX (KN)
2 PA66_20
PA66_30
0
0 2 4 6 8 10 12
Time (ms)
Fig. 12 Impact force versus time curves for an impact energy of 8J
Figure 12 presents examples of force-times curves obtained for impact energy

of 8 Joules. One can note that an increase in values of the maximal impact forces
when varying the glass fibre volume fractions introduced in polyamide, but the
impact time decrease greatly by increasing the glass fibre volume fractions.
478 J. Mars et al.
According to the displacement response and impact force as shown in Figures

11 and 12, it can be noted that the central deflection keeps decreasing until the
peak force appears. Also, it can be noticed that the circular plate will rebound
when the projectile is completely separated, then the circular plate will be in a free
vibration mode around the permanent deflection position.
4 Conclusion
Impact behaviors at low velocity of glass fibre-reinforced polyamide are investi-

gated for different glass fibre volume fractions. The assessment of the impact be-
havior has driven the need to perform tensile tests to determine the elasto-plastic
model of the polyamides. The X-Ray microtomography machine is used to check
the glass fibre distribution. The mechanical elasto-plastic model used can predict
the stress-strain behavior of the studied materials. The elastic-plastic model was
implemented in ABAQUS/EXPLICIT for studying the behavior of the composites
under low velocity impact. The results have shown that the impact force increase
with increasing fibre volume fractions.
References
Akay, M., O’Regan, D.F., Bailey, R.S.: Fracture Toughness and impact behaviour of glass-
fibre-reinforced polyamide 6,6 injection mouldings. Composites Science and Technolo-
gy 55, 109–118 (1995)
Benaceur, I., Othman, R., Guegan, P., Dhieb, A., Damek, F.: Sensitivity of the flow stress
of Nylon 6 and Nylon 66 to strain-rate. International Journal of Modern Physics B 22,
1249–1254 (2008)
Cantwell, W.J., Morton, J.: The impact resistance of composite materials-a review. Compo-
sites 22, 347–362 (1991)
Dhakal, H.N., Zhang, Z.Y., Bennett, N., Reis, P.N.B.: The low-velocity impact response of
non-woven hemp fibre reinforced unsaturated polyester composites: Influence of impac-
tor geometry and impact velocity. Composite Structures 94, 2756–2763 (2012)
Chen, L.B., Xi, F., Yang, J.L.: Elastic-plastic contact force history and response characte-
ristics of circular plate subjected to impact by a projectile. Acta Mech. Sin. 23, 415–425
(2007)
Ghorbel, A., Saintier, N., Dhiab, A., Dammak, F.: Etude du comportement mécanique d’un
polyamide 66 chargé de fibres de verre courtes. Mécanique & Industries 12, 333–342
(2011)
Minak, G., Kostic, S.C., Šoškic, Z., Radovic, N.: Equivalent stiffness as measure of low ve-
locity impact damage of complex composite structures. Facta Universitatis: Mechanical
Engineering 10, 91–104 (2012)
Mouti, Z., Westwood, K., Kayvantash, K., Njuguna, J.: Low Velocity Impact Behavior of
Glass Filled Fiber-Reinforced Thermoplastic Engine Components. Materials 3,
2463–2473 (2010)
Mouti, Z., Westwood, K., Long, D., Njuguna, J.: An experimental investigation into loca-
lised low-velocity impact loading on glass fibre-reinforced polyamide automotive prod-
uct. Composite Structures 104, 43–53 (2013)
Simeoli, G., Acierno, D., Meola, C., Sorrentino, L., Iannace, S., Russo, P.: The role of in-
terface strength on the low velocity impact behaviour of PP/glass fibre laminates. Com-
posites 62, 88–96 (2014)
Wali, M., Abdennadher, M., Fakhfakh, T., Haddar, M.: Dynamic analysis of an elasto-
plastic sandwich subjected to low velocity impact. Multidiscipline Modeling in Mate-
rials and Structures 7, 184–206 (2011)
Wali, M., Abdennadher, M., Fakhfakh, T., Haddar, M.: Vibration response of sandwich
plate under Low-velocity impact loading. WSEAS Transactions on Applied and Theo-
retical Mechanics 6, 27–36 (2011)
Xu, J., Li, Y., Ge, D., Liu, B., Zhu, M.: Experimental investigation on constitutive behavior
of PVB under impact loading. International Journal of Impact Engineering 38, 106–114
(2011)
Yang, L., Yan, Y., Kuang, N.: Experimental and numerical investigation of aramid fibre
reinforced laminates subjected to low velocity impact. Polymer Testing 32, 1163–1173
(2013)
Burst Pressure Estimation of Corroded Pipeline
Using Damage Mechanics
Djebbara Benzerga
Department of Mines and Metallurgy, LSCMI, Faculty of Mechanical, University of Science and
Technology of Oran Mohamed Boudiaf, B.P. 1505 El M’naouer 31000 Oran, Algeria
djeb_ben@yahoo.fr
Abstract. Pipelines are being widely employed worldwide as means of con-

veyance of crude oil and its derivatives. Especially in the south and in the north of
Algeria many pipelines connect oil fields to oil refineries. Nevertheless, a consi-
derable distance is covered crossing hills, in which landslides could change not
only the pipelines alignment but also the stresses. Besides, landslides may cause
cracks in the pipes. Furthermore, both the close contact with soil and the action of
weather can provide the corrosion of the pipes, which will reduce the cross section
area, allowing the formation of disturbed flow areas, and also will develop stress
concentrated regions on the pipe wall. Generally, the main cause of high-pressure
gas and oil pipeline ruptures is metal loss in a pipe wall from corrosion. Particular-
ly, SONATRACH company data show that corroded defects (general corrosion
and pitting corrosion) are the primary causes of accidents. Corrosion is one of the
most common causes of accidents involving pipelines. To avoid these undesirable
situations, computational models are playing an important role, as they are able to
predict the behaviour of pipelines in several ways. . The computational simulation
through Finite Element Method (FEM) is one of the most efficient tools to quanti-
fy reliably the remaining strength of corroded pipes.
This work presents a new method based on the concept of Continuum Damage
Mechanics (CDM) which currently has reached a stage of maturity enabling it to
model any type of degradation. The value of the critical pressure, i.e., the allowa-
ble operating pressure, of a corroded pipeline is obtained by using a post processor
based upon damage mechanics. This post processor allows the calculation of the
crack initiation conditions from the history of strain components taken as the out-
put of the ANSYS Software. The ANSYS code is used for physical and geome-
trical non-linear analysis to obtain the critical point where at any time the damage
equivalent stress is maximum. This method was validated by comparing the
results of numerical simulations with experimental and ASME/ B31G results
available in the literature.
Keywords: Burst pressure, pipe, corrosion, damage mechanics, damage, finite

element.

482 D. Benzerga
1 Introduction
In most applications, the damage is very localized in such a way that the damaged
material occupies a volume small in comparison to the macroscale of the structural
component and even to the mesoscale of the representative volume element RVE.
This is due to the high sensitivity of damage to stress concentrations at the ma-
croscale and to defects at the microscale. This allows us to consider that the effect
of the damage on the state of stress and strain occurs only in very small damaged
regions. In other words, the coupling between damage and strains may be neg-
lected everywhere in the structure except in the RVE(s) where the damage devel-
ops. This is the principle of the locally coupled analysis [Lemaitre and Doghri]
where the procedure may be split into the following two steps as shown in Fig. 1:
- a classical structure calculation in elasticity or elastoplasticity by the finite

element method (FEM) to obtain the fields of strain and stress;
- a local analysis at the critical point only dealing with the elasto-plastic constitu-
tive equations coupled with the kinetic law of damage evolution, that is a set of dif-
ferential equations. This method is much simpler and saves a lot of computer time in
comparison to the fully coupled analysis, which takes into account the coupling be-
tween damage and strain in the whole structure. The fully coupled method must be
used when the damage is not localized but diffused in a large region [Lemaitre and
Doghri]. Our main motivation in this study is to show that this method is a
contribution to the rehabilitation of corroded pipes. The study and analysis of the
damage at critical point where at any time the damage equivalent stress is maximum
in the vicinity of the corrosion defect of a steel pipe X65. The method developed
here is to construct a numerical tool to determine the new maximum pressure that
could support a corroded pipe. The Modeling of the damage phenomenon is based
Fig. 1 Locally coupled analysis of crack initiation

Burst Pressure Estimation of Corroded Pipeline Using Damage Mechanics 483
upon Damage Mechanics, which currently has reached a stage of maturity enabling
it to model any type of degradation. Using the programmable ANSYS, a subroutine
was developed and implemented in the main code for the determination of the
critical point (M*) in the vicinity of corrosion defect where potential crack may
occurs. In a second step, a post-processor based on the iterative Newton method was
applied to this critical point (M*) for the determination of the maximum internal
pressure that a corroded pipe could support in the case of a longitudinal corrosion
defect. The maximum pressure is the pressure value corresponding to the critical
value of the DC damage (crack initiation). This method was validated by comparing
the results of numerical simulations with experimental and ASME/ B31G results
available in the literature.
2 Modeling
To model the pipeline we used the ANSYS code with two scenarios, a model
without a cavity (no corrosion, Figure 2) and a model with cavity (corroded
pipeline, Figure 3). Taking into account the symmetry of the pipeline, we modeled
only one-fourth of cylinder. The dimensions of the structure are length, inner
radius and thickness. To do this, we have created a rectangular area, which has the
same length as the pipeline, and a width equivalent to the thickness of the pipeline.
By a 90° rotation of this surface around the axis X, a volume was obtained in the
form of a quarter cylinder. For meshing the structure, we used volume elements
with 20 nodes Solid95. These elements have compatible displacement shapes and
are well suited to model curved boundaries. The mesh of the volume is made to
obtain a concentration of elements around the side where the cavity is represented
(Figure 2). For boundary conditions, we have respected the conditions of
symmetry and the experience.
On the corroded pipe (model with cavity), we kept the same data used in the
model without cavity, and created a cavity that represents the corrosion defects on
the pipeline. To do this we created a parabolic surface to the thickness of one
quarter of the cylinder with a length equivalent to the defect length and width
equal to the depth of the defect. The volume of the cavity is obtained by a rotation
of 90 ° around the upper line of the pipeline wall. This volume is subtracted from
the volume of the non-corroded pipeline (Figure 2). The same meshing is used
with a mesh filtering around the cavity. We varied the values of the depth and
extension of the cavity (corrosion defects, see Figure 2). The burst pressure is
determined according to the geometrical characteristics of each cavity. The critical
pressure or the maximum pressure that will support a corroded pipe is that
corresponding to the critical value of the damage Dc (this value depends on the
material and temperature).
484 D. Benzerga
Fig. 2 Pipe without cavity (pipe not corroded)
Fig. 3 Model with cavity
Using the programmable language ANSYS, a subroutine was developed and

implemented in the main code for the determination of critical point (M *) in the
zone of corrosion defect (see figure below) where microcracks may grow.
MEF Calculations Post-Processor
Dangerous area Critical point (M*)
Next, the constitutive law of the critical point (M, *) where the equivalent stress
σ * is maximum, obtained using the ANSYS code, is implanted in the post-
processor based on the iterative Newton method [Benallal et al, 1988.]:
σ* (M ) = Sup⎛⎜ σ* ⎞⎟
⎝ ⎠
1
with σ* = σ eq Rv 2 (1)
2
2 ⎛ σH ⎞
RV = (1 + ν ) + 3(1 − 2ν )⎜⎜ ⎟
⎟
3 ⎝ σ eq ⎠
In most cases, this criterion is satisfied in areas of high stress concentration
with a high coefficient of triaxiality σH / σeq. The final step is to determine the
evolution of damage in solving the following constitutive laws:
e +Ep
E ij = E ij ij
e = 1 + ν σ ij υ σ kk
E ij − δ ij
E 1− D E 1− D
~D (2)
3 σ ij
p
E ij = P Si f = f = 0
2 σ eq
2
σ eq
D= Rv P Si p ≥ p0
2ES
Using a stepwise method simultaneously with the method (scheme) implicit
Newton, which ensures good convergence [Benallal et al, 1988].
486 D. Benzerga
The method developed above gives versus of the internal pressure of the pipe,
the value of the damage, the accumulated plastic strain and stress components at
each instant until a macroscopic crack initiation at the defect corrosion. This
allows determining the new maximum pressure that will support a pipe with a
corrosion defect.
3 Validation of the Methodology

In order to validate our model, we compared our results with ASME/B31G
standard and experimental results obtained by TA Netto [Netto, 2005] [3] (see
table 1). For the model without cavity, we took the analytical value P = 50.89
MPa. Figure 4 illustrates the results obtained by three methods, the experimental,
ASME / B31G and our proposed method.
Table 1 Maximum predicted pressure
N° d L Burst Burst Predicted Burst Erreur

(mm) (mm) Pressure Pressure(MPa) pressure(MPa) en %
tests ASME/B31G
(MPa)
T1 - - 57.33 50.89 53.4 6.85
T2 1.58 42 37.02 35.05 36 2.75
T3 1.59 21 44.65 42.53 44.7 0.11
T4 1.87 42 32.47 30.85 32.5 0.099
T5 1.91 21 41.28 38.93 42.5 2.95
T6 2.13 42 26.76 24.50 29.5 10.23
T7 2.14 21 34.55 34.86 36.8 6.51
60 60
Burst pressure versus
58
depth for L=21 mm Burst pressure versus
56 55
54 depth for L=41 mm
52
50
50
Burst pressure MPa
Burst pressure MPa
48
46 45
44 Simulation Simulation
42 ASME/B31G 40 ASME/B31G
40 Expérimental Experimental
38
35
36
34
30
32
30
28 25
0.0 0.5 1.0 1.5 2.0 2.5

0.0 0.5 1.0 1.5 2.0 2.5
corrosion defect depth mm
Corrosion defect depth mm
Fig. 4 Burst pressure depending on the depth of corrosion defects for two extensions
L = 21mm and L = 41 mm.
4 Analysis and Interpretation of Results

We note that the numerical results are more realistic (experimental results). These
numerical results do not exceed a 10% difference. Comparing simulation results
and ASME/B31G, we note that the average error is smaller in numerical case.
This is due to a factor of safety used by B31G standard. The ASME/B31G uses
the projection of corrosion defect, while the numerically exact corrosion surface
defect is used. The difference between the numerical and experimental results
depends on several factors:
It was considered a perfectly continuous material, while in reality there

are defects, discontinuities and micro-voids in the material, which can be
potential sources of damage.
There are always residual stresses after machining that we have ignored
in the laws of behavior considered in this study.
The shape of the cavity is not identical for all specimens.
It was considered perfect symmetry, which is not the case in reality
For the boundary conditions, it was considered a locking surface, there
are small displacements, which cause deformations, and the structure of
the material is changed.
The mesh is not fine enough or sufficiently homogeneous.
S values (damage coefficient) εpD (plastic deformation damage threshold)
and DC (critical damage) of X65 material taken approximately, true
values are determined by load-unload tests.
The yield condition was considered without hardening. We considered
that damage only occurs when strain hardening is saturated
5 Conclusion
In this paper, we presented a new method based upon damage mechanics, which is
now in its stage of maturity. Our main motivation in this work was to show that
this method is a contribution to the rehabilitation of corroded pipes. In spite of
some differences between proposed method results and the experimental results,
our method have had a good performance in predicting the failure pressure of a
pipeline containing corrosion defects. The ASME/B31G results were more con-
servative than proposed method; however, our method provided more precise re-
sults. Our models proved to be capable of simulating the corroded pipe burst tests
adequately.
488 D. Benzerga
References
Benallal, A., Billardon, R., Doghri, I.: An integration algorithm and the coresponding con-
sistent tangent operator for fully coupled elastoplastic and damage equations. Comm.
Appl. Numer. Methods 4, 731–740 (1988)
Lemaitre, J., Doghri, I.: Damage 90: a post processor for crack initiation. Comput. Methods
Appl. Mech. Eng. 115, 197–232 (1994)
Netto, T.A., Ferraz, U.S., Estefen, S.F.: The effect of corrosion defects on the burst pressure
of pipeline. J. Constr. Steel Res. 61, 1185–1204 (2005)
Numerical Simulation of Incremental Sheet
Metal Forming Process
Lotfi Ben Said, Mondher Wali, and Fakhreddine Dammak
Mechanical Modelisation and Manufacturing Laboratory (LA2MP), National Engineering

School of Sfax, B.P W3038, Sfax, University of Sfax, Tunisia
{bensaid_rmq,mondherwali}@yahoo.fr,
fakhreddine.dammak@enis.rnu.tn
Abstract. This paper presents a numerical simulation of the incremental sheet

metal forming (ISF) process, type single point incremental forming (SPIF). A fi-
nite element model (FEM) is developed by using the commercial FE code
ABAQUS. An elasto-plastic constitutive model with quadratic yield criterion of
Hill and isotropic hardening behavior has been adopted during ISF operation. Re-
sults including thickness variation of sheet metal and forming force along Z-axis
are presented.
Keywords: Sheet forming, Incremental forming, Numerical simulation.
1 Introduction
Incremental Sheet Forming (ISF) is an innovative process to manufacture sheet

metal with numeric increments on CNC machines through plastic deformation.
This process is relatively slow, allows the manufacture of prototype parts with
complex shapes and high precisions. Two ISF types can be distinguished: the Sin-
gle Point Incremental Forming (SPIF) and the Two Points Incremental Forming
(TPIF). These tow variants of ISF are presented in Figure 1.
Fig. 1 The two variants of ISF (Azaouzi and Lebaal 2012)

490 L. Ben Said, M. Wali, and F. Dammak
The experiences realized with the ISF process are concentrated in the SPIF
process using a CNC machines with a forming tool with specific shape. This tool
replaces the milling tool and it uses the same mounting elements in CN machine.
Many trajectories are used in incremental forming with numerical increments by
the use of CN program in 3 axis or 5 axis machines. The incremental forming tool
trajectories are specific to this process and materials of sheets manufactured in
ISF.
The literature on the subject of ISF: the integrity of surfaces during and after
the incremental forming, the geometry of the specimen manufactured, forming ef-
forts and optimization of forming time are investigated by the majority of re-
searches. The majority of manufactured parts, mentioned in the articles, are not
industrial parts but they are test pieces with cone or pyramidal shape where a he-
mispherical punch is used in the sheet forming process, with a manner close to the
operation of a pocket in a CN milling process. So the study of an industrial piece
will be significant with a good experiment results.
Several studies have been made to understand ISF process taking into account
different material behavior and various methods have been proposed for assessing
the formability in ISF. (Robert et al. 2012) made comparison between incremental
deformation theory and flow rule to simulate sheet metal forming processes. They
used the material user subroutine (VUMAT) to implement two material behavior
laws (incremental deformation theory and classical flow rule plasticity) in their
simulation. At the aim to analyze the stress state and thickness distribution of
sheet metal studied. But in this study they did not take into account kinematic har-
dening in the material behavior for a best of prediction of spring back in sheet
metal incremental forming.
(Ben Ayed et al. 2014) presented a simplified numerical approach to simulate
ISF with efficient CPU time. A shell element DKT12 was implemented and
coupled with an elasto-plastic model based on a classical flow rule, isotropic har-
dening has been considered in their simulations.
(Fang et al. 2014) build an investigation on deformation mechanism and frac-
ture behavior in single point incremental forming. In their analytic model they
make on consideration two parameters, bending effect and strain hardening, to de-
scribe the localized deformation mechanism. By numerical and experimental in-
vestigation they demonstrate that the deformation occurs not only in the contact
zone, but also in the neighboring wall which has been already formed in the vi-
cinity of the contact zone. And that the fracture tends to appear at the transitional
zone between the contact area and the formed wall. They conclude also that the
strain hardening has a positive effect on the material formability.
(Belchior et al. 2014) developed an approach applied in robotized ISF process,
this approach consists in coupling a finite element analysis of the ISF (using to
predict efforts in this process) with an elastic model of a robot structure where the
punch is mounted. The result of finite element simulation was used like an input
data for the elastic model at the aim to optimize the punch path and to make cor-
rection in robotized ISF when the punch makes deviation. The effect of the har-
dening law (Ludwick or Voce) has been evaluated.
Numerical Simulation of Incremental Sheet Metal Forming Process 491
(Kurra and Regalla 2014) analyzed the formability and thickness distribution in
ISF of extra-deep drawing steel sheet which is a widely used material in automo-
tive applications involving simple and complex parts requiring high formability.
(Seong et al. 2014) concluded after their research that the stress states in top,
middle and bottom surfaces exceed the forming limit curve at different times and
each layer has different stress state in terms of their deformation history to sup-
press necking.
(Behera et al. 2014) propose an analysis methodology using topological con-
ceptual graphs to capture the effects of different phenomena on the final accuracy
of a sheet metal part manufactured by SPIF. So they create an algorithm creating
partial tool paths when the shape of the sheet formed is to complex and demand
sum accuracy.
(Han et al. 2013) studied the springback at the aim to predict and control accu-
rately this factor to design an accurate tool path for ISF. A three-dimensional elas-
to-plastic finite element model was implemented to simulate ISF this model is
based on the particle swarm optimization neural network and it has shown good
results in the prediction of springback.
The present investigation is a contribution of work that started using an incre-
mental sheet metal forming (ISF) process by taking into account the anisotropic
effect presented in sheet metal combined with isotropic hardening behavior. Re-
sults including thickness distribution and forming forces can be estimated from the
numerical model.
2 Elasto-plastic Constitutive Equations
An extensive description for formulating constitutive elasto-plastic rate equations

in order to model anisotropic yielding, and non-linear isotropic hardening can be
found elsewhere, and we will not elaborate on these details; see, for example,
(Lemaitre and Chaboche 1990). A summary of the resulting set of equations con-
sidered in this paper
ε = ε +ε
e p
- Partition of the total strain (1)
σ = D: ε
e
- Hooke’s law (2)
3
- Yield function f = ϕ (σ ) −σ p ≤ 0 (3)
2
ϕ (σ ) = σ Pσ
T
(4)
σ p =σY + R , R =Q 1− e ( − βκ
) (5)
∂f 3 1
ε = γ = γ n n= Pσ
p
- Flow rule , (6)
∂σ 2 ϕ
- Isotropic hardening κ = γ (7)
where ε denotes the elastic strain tensor and ε is the plastic strain tensor, σ
e p
is the stress tensor, D is the general elastic operator, Q and β are material pa-
rameters, γ is the plastic multiplier and P is a fourth order tensor which define
the yield criterion. This yield function includes the classical J2 plasticity yield
condition and the quadratic Hill criterion as special cases. The Hill yield criterion,
in three-dimensional cases, is obtained by taking
⎡H + G −H −G 0 0 0 ⎤
⎢ H+F −F 0 0 0⎥
⎢ ⎥
2 ⎢ F +G 0 0 0⎥
P= H , [H ]= ⎢ ⎥ (8)
2N 0 0
3 ⎢ ⎥
⎢ Sym 2M 0⎥
⎢ ⎥
⎣ 2L⎦
where F , G , H , N , M and L are material constants obtained by tests of the ma-

terial in different orientations. The J2 plasticity yield criterion is recovered using
Eq. (9) and setting
F = G = H = 0.5 , N = M = L = 1.5 (9)
In this section, square box application of ISF was investigated and the results car-
ried out with the use of numerical treatment based on ABAQUS/Explicit code.
The material is an aluminum alloy (AA5754-O). The mechanical properties of the
AA5754-O material are given in Table 1. The isotropic hardening behavior is
modeled by the exponential law (Eq. 5), and the Hardening parameters of the
AA5754-O material are presented in Table 2.
The initial dimension of the sheet is 200 x 200 x 1.5 mm, the punch tool has a
diameter of 10 mm. The punch tool is assumed as a rigid body and no property is
assigned. The frustum of the square box has a depth of 10 mm and a base area of
80x80 mm. The tool path and the dimensions of the desired shape are shown in
Figure 2. The friction coefficient between punch tool and sheet metal is assumed
to be 0.1.
The sheet is meshed using 2728 triangular elements (S3R in Abaqus) with five
integration points through the thickness.
Table 1 Mechanical properties of the aluminum AA5754-O material
Elastic Prop. Hill 48 Coefficients

E(GPa) ν F G H N
70 0.33 0.748 0.572 0.403 1.467
Table 2 Hardening parameters of the AA5754-O material

Parameter Unit IH
σY MPa 95
Q MPa 159
β -- 9
Fig. 2 Tool path and the final shape of sheet incremental formed (Robert et al. 2012)
Numerical predictions carried out using Abaqus are shown in Figures 3, 4 and
5. The thickness variation along the Y = 0 cut is presented in Fig. 3. According to
the thickness evolution, significant deformations of square box are located princi-
pally in the vicinity of punch path. The evolution of the vertical forming force dur-
ing the ISF process of the studied sheet is illustrated in Figure 4. The vertical
forming force due to the contact pressure reaches the maximum value in the last
pass of the punch when incremental forming is in a depth of 10 mm. the value of
this effort is in the same range of efforts recognized in the majority of articles
established by many research group.
The distribution of the von Mises stress in the deformed zone under the action
of the punch during the ISF process is as shown in Figure 5. According to this
Figure the major Von Mises stress is localized under the punch tool, located prin-
cipally along the punch path.
1.55
1.5
Thickness (mm)
1.45
1.4
1.35
1.3
1.25
0 0.05 0.1 0.15 0.2
X direction (m)
Fig. 3 Thickness variation in the cut plane Y = 0
Fig. 4 Evolution of the forming force Fz (N)

Fig. 5 Distribution of Von Mises Stress
4 Conclusion
In this paper, finite element model is developed, based on Hill yield criterion and
isotropic hardening behavior, simulating an example of single point incremental
sheet forming. The material model is implemented in the commercial finite ele-
ment code ABAQUS/Explicit to predict ISF process. The results show significant
deformations have been located along the contour of the square box base but it
should be noted that there are also minor deformations on the borders near to the
clamped shape.
References
Azaouzi, M., Lebaal, N.: Tool path optimization for single point incremental sheet forming
using response surface method. Simulation Modelling Practice and Theory 24, 49–58
(2012)
Behera, A.K., Lauwers, B., Duflou, J.R.: Tool path generation framework for accurate
manufacture of complex 3D sheet metal parts using single point incremental forming.
Comput. Ind. 65, 563–584 (2014)
Belchior, J., Leotoing, L., Guines, D., Courteille, E., Maurine, P.: A Proc-ess/Machine
coupling approach: Application to Robotized Incremental Sheet Forming. J. Mater.
Process. Tech. 214, 1605–1616 (2014)
Ben Ayed, L., Robert, C., Delamézière, A., Nouari, M., Batoz, J.L.: Simplified nu-merical
approach for incremental sheet metal forming process. Eng. Struct. 62-63, 75–86 (2014)
Fang, Y., Lua, B., Chena, J., Xua, D.K., Oub, H.: Analytical and experimental inves-
tigations on deformation mechanism and fracture behavior in single point incremental
forming. J. Mater. Process. Tech. 214, 1503–1515 (2014)
Han, F., Mo, J., Qi, H., et al.: Springback prediction for incremental sheet forming based on
FEM-PSONN technology. Trans. Nonferrous Met. Soc. China 23, 1061–1071 (2013)
Kurra, S., Regalla, S.P.: Experimental and numerical studies on formability of extra-deep
drawing steel in incremental sheet metal forming. Journal of Materials Research and
Technology 3, 158–171 (2014)
Lemaitre, J., Chaboche, J.L.: Mechanics of Solid Materials. Cambridge University Press,
Cambridge (1990)
Robert, C., Delamézière, A., Dal Santoa, P., Batoz, J.L.: Comparison between in-cremental
deformation theory and flow rule to simulate sheet-metal forming processes. J. Mater.
Process. Tech. 212, 1123–1131 (2012)
Seong, D.Y., Haque, M.Z., Kimb, J.B., Stoughton, T.B., Yoon, J.W.: Suppression of neck-
ing in incremental sheet forming. Int. J. Solids Struct. 51, 2840–2849 (2014)
A Higher Order Shear Strain Enhanced Solid-
Shell Element for Laminated Composites
Structures Analysis
Abdessalem Hajlaoui1, Abdessalem Jarraya2, Mondher Wali1,

and Fakhreddine Dammak1
1
Mechanical Modelisation and Manufacturing Laboratory (LA2MP),
National Engineering School of Sfax, B.P W3038, Sfax, University of Sfax, Tunisia
2
Faculty of Engineering, King Abdulaziz University, North Jeddah, Saudi Arabia
abdhajlaoui@gmail.com, jarraya_abdessalem,mondherwali}@yahoo.fr,
fakhreddine.dammak@enis.rnu.tn
Abstract. This paper presents a free from locking higher order solid-shell element
based on the Enhanced Assumed Strain (EAS) for laminated composite structures
analysis. The transverse shear strain is divided into two parts: the first one is inde-
pendent of the thickness coordinate and formulated by the Assumed Natural Strain
(ANS) method; the second part is an enhancing part which ensures a quadratic dis-
tribution through the thickness. This permit to remove the shear correction factors
and improves the accuracy of transverse shear stresses. Also, volumetric locking is
completely avoided by using the optimal parameters in the EAS method. Compar-
isons of numerical results with those extracted from literature show the perfor-
mance of the developed finite element.
Keywords: Laminated composites, Solid-shell, Higher Order Shear Deformation.
1 Introduction
In the last decades the use of composite materials is becoming more widely used
since they provide many advantages to structural designers. The stress and the
strain field in these structures is very complex, therefore numerical methods have
been developed to give an accurate prediction. Several methods have been pro-
posed in the literature namely: The Assumed Natural Strain method (ANS) used to
solve the shear locking problems in the case of thin plate and shell structures, we
can cite (MacNeal 1982), (Bathe and Dvorkin 1985), (Bucalem and Bathe 1993)
among many others. (Hauptmann and Schweizerhof 2000) assumed a linear distri-
bution of the strain in thickness direction and developed a solid-shell element for-
mulation. (Sze and Yao 2000) used this method to obtain a locking-free solid-shell
element.

498 A. Hajlaoui et al.
The Enhanced Assumed Strain method (EAS) was proposed by (Simo and Rifai
1990). It allowed avoiding the thickness locking caused by coupling of the in-
plane and transverse normal stress and normal strain responses (Doll et al. 2000).
Based on this method, (Alves de Sousa et al. 2003) proposed a locking-free solid-
shell element formulation. The application of the EAS method in shell analysis
was presented in the work of (Andelfinger and Ramm 1993) with linear elastic
range. The extension to non-linear aspects was carried out by (Büchter et al. 1994)
and (Bischoff and Ramm 1997) among others.
The use of a single method cannot entirely solve the locking problems in solid-
shell elements. Therefore many studies have combined them for a better prediction
of the structure behavior. (Klinkel et al. 1999) and (Vu-Quoc and Tan 2003) de-
veloped three-dimensional locking free solid-shell elements by using the EAS and
ANS methods. (Hajlaoui et al. 2012) presented a solid-shell finite element based
on the classical ANS method to compute the shear locking effect and a nine pa-
rameters EAS method which avoided completely the volumetric locking to study
the bucking behavior of laminated composite plate with delamination.
In the literature, there is no solid-shell formulation with high-order transverse
shear enhancement. The only exception is the work of (Quy and Matzenmiller
2008), where the authors developed a finite element based on the theoretical foun-
dations of high-order shear deformation theories. In this works authors decom-
posed the transverse shear strains into two parts: the first one is compatible and
continuous, which can present a shear locking in thin limit structures, the second
part is based on the EAS with parabolic function in terms of natural thickness
coordinate. The membrane part in this work is not enhanced, which can present a
numerical locking in the case of in-plane bending and in the nearly incompressible
elasticity.
However, in the case of plates and classical shell, the higher order shear defor-
mation theory (HOSDT) has been widely used. (Liu and Reddy 1985), (Rastgaar
Aagaah et al. 2003) and (Reddy and Lee 2004), among others, used this theory to
improve the transverse shear strain and stress distribution. In a recent work, (Wali
et al. 2014) presents an efficiency three dimensional double directors shell element
for the functionally graded material shell structures analysis. The vanishing of
transverse shear strains on top and bottom faces is considered in a discrete form
inspired form the work of (Dammak et al. 2005). Thus, the third-order shear de-
formation plate theory (TSDT) is a particular case of the developed formulation.
Nevertheless, the treatment of delamination with this HOSDT is complicated
which is not the case with the solid-shell elements.
This paper is an improvement of the solid-shell formulation developed by (Haj-
laoui et al. 2012); the transverse shear strain is composed by two parts: a compati-
ble part based on the ANS formulation of (Bathe and Dvorkin 1985), and a second
part based on EAS method with parabolic function in terms of natural thickness
coordinate as proposed in (Quy and Matzenmiller 2008). In addition, the en-
hancements of membrane part and the transverse strain were used to avoid the
in-plane bending and volumetric locking.
A Higher Order Shear Strain Enhanced Solid-Shell Element 499
2 Solid-Shell Finite Element Formulation
The strain field is enhanced with the introduction of internal variables (Simo and
Rifai, 1990), as following:
E = E + E
c
(1)
Where E and E are respectively the compatible part and the enhanced part
c
of the Green-Lagrange strain tensor.
2.1 Compatible Strains

The compatible part is arranged in (6x1) column matrix as follows:
E =⎡
⎣ E11 , E22 , E33 , 2 E12 , 2 E13 , 2 E23 ⎤⎦
c c c c c c c T
(2)
c c
The compatible transverse shear strains E13 and E23 and the transverse thick-
c
ness strains E33 are developed on the base of ANS method proposed by (Bathe
and Dvorkin 1985). This formulation is detailed in the author’s previous work,
c c
Hajlaoui et al. (2012). The compatible transverse shear strains E13 and E23 are
evaluated at four midpoints of the reference element mid-surface A = (-1,0,0), B =
c
(0,-1,0), C = (1,0,0), D = (0,1,0) and the transverse thickness strains E33 is eva-
luated at the following four points of the reference element A1=(-1,-1,0), A2= (1,-
1,0), A3= (1,1,0), A4= (-1,1,0).
⎡ 2E13c ⎤ ⎡(1 − η ) E13B + (1 + η ) E13D ⎤

⎢2 E c ⎥ = ⎢ 1 − ξ E A + 1 + ξ E C ⎥ (3)
⎣ 23 ⎦ ⎣( ) 13 ( ) 13 ⎦
4
1
E33 =
c
∑ 4 (1 + ξξ )(1 + ηη ) E
A A 33
A
(4)
A =1
Then the compatible part of the Green-Lagrange strain tensor takes the
following form
⎡ 1
2
( g11 − G11 ) ⎤
⎢ 1
( g 22 − G22 ) ⎥
⎢ 2
⎥
⎢ ⎥
4 (
1 + ξ A ξ )(1 + η Aη ) 2 ( g 33 − G33 ) ⎥
4
E =T
c −T ⎢ ∑ 1 1 A A
(5)
⎢ A =1 ⎥
⎢ ( g12 − G12 ) ⎥
⎢ 1 ⎡(1 − η ) ( g B − G B ) + (1 + η ) ( g D − G D ) ⎤ ⎥
⎢2⎣ 13 13 13 13 ⎦
⎥
⎢⎣ 2 ⎡⎣(1 − ξ ) ( g 23 − G23 ) + (1 + ξ ) ( g 23 − G23C ) ⎤⎦ ⎥⎦
1 A A C
Where Gij = Gi .G j and g ij = gi .g j are the components of the metric tensors at

reference and actual configurations. T is the transformation matrix of the strain
tensor from parametric coordinates to the local Cartesian coordinates. The varia-
tion and the increment of the total compatible strain tensor can be written as
follows:
δ E = B δUn Δ E = B ΔU n
c c
, (6)
Where Un is the displacement nodal vector and B is the strain interpolation

matrix, associated with node ( I ) and denoted BI is given by:
⎡ T
g1 N I ,1 ⎤
⎢ T ⎥
g1 N I ,1
⎢ ⎥
⎢ ⎥
4 (
1 + ξ A ξ )(1 + η Aη ) g3 N I ,3
4
⎢ ∑ 1 T
⎥
BI = T ⎢ ⎥
−T A =1
(7)
⎢ g2 N I ,1 + g1 N I ,2
T T
⎥
⎢1 ⎥
( ) (
⎢ 2 ⎡⎣(1 − η ) g3 N I ,1 + g1 N I ,3 + (1 + η ) g3 N I ,1 + g1 N I ,3 ) ⎤⎥
T T T T
B B B B D D D D
⎦
⎢ ⎥
( ) (
⎢⎣ 12 ⎡⎣(1 − ξ ) g3A N IA,1 + g1A N IA,3 + (1 + ξ ) g3C N IC,1 + g1C N IC,3 ) ⎤⎥
T T T T
⎦⎦
2.2 Enhanced Strain

The enhanced part is related to the vector of the internal strain parameters α as
E = M
α , δE = M
δα , ΔE = M
Δα (8)
where M is the interpolation function matrix for the enhanced assumed strain
field. The interpolation is first defined in the natural coordinate and must be trans-
ferred to the global coordinates as follows
= det J 0 T − T M
M (9)
0 ξηζ
det J
Where the subscript ‘ 0 ’ means evaluation at the center of the element in the
natural coordinates, J = [G1 , G2 , G3 ] is the Jacobian matrix. The interpolation
, in equation (9), is expression in term of the parametric coordinates
matrix M ξηζ
( ξ ,η ,ζ ) .
Three choices of matrix M ξηζ
will be considered with, 7, 9 and 11
parameters.
⎛ξ 0 0 0 0 0 0⎞
⎜0 η 0 0 0 0 0
⎟
⎜ ⎟
⎜0 0 ζ 0 0 0 0 ⎟
⎜ ⎟
0 0 0 ξ η
=⎜
7 0 0 ⎟ (10.a)
M ξηζ
⎜ 1 ⎟
⎜0 0 0 0 0 −ζ 0 ⎟
2
⎜ 5 ⎟
⎜ 1 2 ⎟
⎜0 0 0 0 0 0 −ζ ⎟
⎝ 5 ⎠
⎛ξ 0 0 0 0 0 0 0 0 ⎞
⎜0 η 0 0 0 0 0 0 0
⎟
⎜ ⎟
⎜0 0 ζ 0 0 0 0 ξζ ηζ ⎟
⎜ ⎟
0 0 0 ξ η
=⎜
9 0 0 0 0 ⎟ (10.b)
M ξηζ
⎜ 1 ⎟
⎜0 0 0 0 0 −ζ 0 ⎟
2
0 0
⎜ 5 ⎟
⎜ 1 ⎟
⎜0 0 0 0 0 −ζ 0 ⎟
2
0 0
⎝ 5 ⎠
⎛ξ 0 0 0 0 0 0 0 0 ξη 0 ⎞
⎜0 η 0 0 0 0 0 0 0 0 ξη ⎟
⎜ ⎟
⎜0 0 ζ 0 0 0 0 ξζ ηζ 0 0 ⎟
⎜ ⎟
ξ η
=⎜ ⎟
11 0 0 0 0 0 0 0 0 0
M ξηζ (10.c)
⎜ 1 ⎟
⎜0 −ζ 0 ⎟
2
0 0 0 0 0 0 0 0
⎜ 5
⎟
⎜ 1 ⎟
⎜0 0 0 0 0 0 −ζ
2
0 0 0 0 ⎟
⎝ 5 ⎠
Since the shear strain interpolation in thickness direction is parabolic, at least
2x2x3 Gauss quadrature integration rule is used. After including interpolation
functions for enhanced strain fields, we obtain the solid-shell element enhanced
with 7, 9 and 11 incompatible modes.
3 Weak Form and Linearization
The weak form of the proposed solid-shell formulation is
G( u , E ) = ∫
V
(
S : δ E + δ E dV −
c
) ∫ F .δ udV − ∫
V
V
∂Vf
FS .δ udA = 0 (11)
where S is the second Piola–Kirchhoff stress tensors and FS and FV are surface
and volumetric distributed load respectively. This equation will be solved by the
Newton-Raphson method and must be linearized. The strain parameters are then
eliminated at the element level which leads to the following tangent stiffness ma-
trix and the residual vector
−1 −1
KT = K e + K G − L H L , Re = L H h + f −f
T T ext int
(12)
int ext
where K e , L , H , f ,f and h are given by the following expressions
Ke = ∫ B  B dVe , L = ∫ T B dV , H= ∫ T M dV
T
M e
M e
(13)
Ve Ve Ve
= ∫ = ∫ N FV dVe + ∫ N FS dA; h = ∫ T S dV
int T ext T T
f B S dVe ; f M e
(14)
Ve Ve ∂V fe Ve
where,  is the 6 × 6 three directional isotropic elasticity tensor and K G is

given in (Hajlaoui et al. 2012).
n Enhanced Solid-Shell Element
A Higher Order Shear Strain 5003
4 amples
Numerical Exa
In this section we presentt two numerical simulations in order to illustrate efficienn-

cy of the proposed higherr order shear strain enhanced solid-shell element. Resullts
are compared to other w well-known formulations. A listing of elements and thhe
ntify them is given in Table 1.
abbreviations used to iden
ments
Table 1 Listing of shell elem
Name Description
Q8A3E5
(Klinkel eet al. (1999), ANS with E33 , E13 E23 and 5 parameters
Q8A3E7
(Vu-Quocc and Tan 2003), ANS with E33 , E13 E23 and 7 parameters
Q8S12 (Quy and Matzenmiller 2008), with 12 parameters
C3D8C11 Present ellement with 11 parameters
4.1 der with End Diaphragms

Pinched Cylind
To investigate the capaability of the developed elements to model the inn-
extensional bending and complex membrane states of stress, we consider one oof
the more severe tests for sshell and solid-shell formulations.
1.8
1.6
Normalized displacement (x10-5)
1.4
1.2
1
Q8A3E5
0.8
Q8A3E7,C3D8C11
0.6
Q8S12
0.4
0.2
0
0 10 20 30
Number of Elements per Side
Fig. 1 Description and resultts of pinched cylinder with end diaphragms
A short cylinder, witth two pinching vertical forces at the middle section,
ms at the end, is modeled using one octant with approo-
and two rigid diaphragm
priate symmetry boundarry conditions, (figure 1). The length of the cylinder is
L=600mm, the radius is R=300mm, and the thickness is h=3mm. The material
properties are: E = 3.0 × 10 MPa, ν = 0.49 . The results are shown in Figure 1.
6
In comparison to other solid-shell formulations in the literature listed in Table 1,

our developed formulation and those obtained with (Vu-Quoc and Tan 2003)
shows superior convergence behavior. This proves the accuracy and efficiency of
the present solid-shell formulation for the static analysis of shells.
4.2 Buckling of Laminated Plate with Delamination

To study the influence of the delamination size on the buckling behavior of a
simply supported symmetric laminated plate, we consider a cross ply square lami-
nated plate [0/90/90//0]. It is subjected to a uniform in plan force with through-
the-width délamination between the two last layers, as shown in figure 2.
Fig. 2 Laminated plate with delamination
The material properties are: E11 = E22 = open , E22 = E33 , G23 = 0.5E22 ,
G12 = 0.5G13 = 0.6 E22 and ν 12 = ν 13 = ν 23 = 0.25 .
Figure 3 illustrates the variation of the buckling load versus the delamination
size for symmetric cross ply laminated square plates. It is shown that the largest
buckling load is obtained for the case of symmetric laminated [0/90/90//0], be-
cause the fiber orientation at the layer with the delamination is parallel to the
in-plane loads. From the plots it can be concluded that the influence of the dela-
mination is more important for the symmetric laminated [0/90/90//0].
180
a/b=1 et b/h=20 E11/E22=30
160
Critical buckling load(N/mm²)

140
[90/0/0//90]
120
[0/90/90//0]
100
80
60
40
20
0
0 0.2 0.4 0.6 0.8
D/a
Fig. 3 Description and results of Pinched cylinder with end diphragrams
5 Conclusions
In this study a new solid-shell finite element based on the partition of shear strain
is developed: a first part is independent of the thickness coordinate and formulated
by the ANS method which avoid the shear locking in thin limit structure; a second
part, enhancing part, ensures a quadratic distribution of the transverse shear strain.
This permit to remove shear correction factors and improves the accuracy of
transverse shear stresses. The finite element formulation is completely free from
the volumetric locking phenomena which occur in the treatment of nearly incom-
pressible elasticity. The developed finite element model is validated by comparing
numerical results with other results from literature. The buckling behavior of
composite laminates is investigated based on the use of solid-shell element.
References
Alves de Sousa, R.J., Natal Jorge, R.M., Valente, R.A.F., César de Sá, J.M.A.: A new vo-
lumetric and shear locking-free 3D enhanced strain element. Eng. Computation 20,
896–925 (2003)
Andelfinger, U., Ramm, E.: EAS-Elements for two-dimensional, three-dimensional, plate
and shell structures and their equivalence to HR-elements. Int. J. Numer. Meth. Eng. 36,
1311–1337 (1993)
Bathe, K.J., Dvorkin, E.N.: A four-node plate bending element based on Mindlin/Reissner
plate theory and a mixed interpolation. Int. J. Numer. Meth. Eng. 2, 367–383 (1985)
Bischoff, M., Ramm, E.: Shear Deformable shell elements for large strains and rotations.
Int. J. Numer. Meth. Eng. 40, 4427–4449 (1997)
Bucalem, M.L., Bathe, K.J.: Higher-order MITC general shell elements. Int. J. Numer.
Meth. Eng. 36, 3729–3754 (1993)
Büchter, N., Ramm, E., Roehl, D.: Three-dimensional extension of non-linear shell formu-
lation based on the enhanced assumed strain concept. Int. J. Numer. Meth. Eng. 37,
2551–2568 (1994)
Dammak, F., Abid, S., Gakwaya, A., Dhatt, G.: A formulation of the non linear dis-crete
Kirchhoff quadrilateral shell element with finite rotations and enhanced strains. Euro-
pean Journal of Computational Mech. 14, 1–26 (2005)
Doll, S., Schweizerhof, K., Hauptmann, R., Freischlager, C.: On volumetric locking of low-
order solid and solid-shell elements for finite elastoviscoplastic deformations and selec-
tive reduced integration. Eng. Computation 17, 874–902 (2000)
Hajlaoui, A., Jarraya, A., Kallel-Kamoun, I., Dammak, F.: Buckling analysis of a laminated
composite plate with delaminations using the enhanced assumed strain solid shell ele-
ment. J. Mech. Sci. Technol. 26, 3213–3221 (2012)
Hauptmann, R., Schweizerhof, K., Doll, S.: Extension of the ‘solid-shell’ concept for appli-
cation to large elastic and large elastoplastic deformations. Int. J. Numer. Meth. Eng. 49,
1121–1141 (2000)
Klinkel, S., Gruttmann, F., Wagner, W.: A continuum based three-dimensional shell ele-
ment for laminated structures. Comput. Struct. 71, 43–62 (1999)
Liu, L.F., Reddy, I.N.: A higher-order shear deformation theory of laminated elastic shells.
Int. J. Engng. Sci. 23, 319–330 (1985)
MacNeal, R.H.: Derivation of element stiffness matrices by assumed strain distributions.
Nucl. Eng. Des. 70, 3–12 (1982)
Noor, A.K.: Stability of multilayered composite plates. Fibre Sci. Technol. 8, 81–88 (1975)
Quy, N.D., Matzenmiller, A.: A solid-shell element with enhanced assumed strains for
higher order shear deformations in laminates. Tech. Mech. 28, 334–355 (2008)
Rastgaar Aagaah, M., Mahinfalah, M., Nakhaie Jazar, G.: Linear static analysis and fi-nite
element modeling for laminated composite plates using third order shear theory. Com-
pos. Struct. 62, 27–39 (2003)
Reddy, J.N., Lee, S.J.: Vibration suppression of laminated shell structures investigated us-
ing higher order shear deformation theory. Smart Mater. Struct. 13, 1176–1194 (2004)
Simo, J.C., Rifai, M.S.: A class of mixed assumed strain methods and the method of in-
compatible modes. Int. J. Numer. Meth. Eng. 29, 1595–1638 (1990)
Sze, K.Y., Yao, L.Q.: A hybrid stress ANS solid-shell element and its generalization for
smart structure modeling. Part I—solid-shell element formulation. Int. J. Numer. Meth.
Eng. 48, 545–564 (2000)
Vu-Quoc, L., Tan, X.G.: Optimal solid shells for non-linear analyses of multilayer compo-
sites. I. Statics. Comput. Methods Appl. Mech. Engrg. 192, 975–1016 (2003)
Wali, M., Hajlaoui, A., Dammak, F.: Discrete double directors shell element for the func-
tionally graded material shell structures analysis. Comput. Methods Appl. Mech.
Engrg. 278, 388–403 (2014)
New Approch of High Cycle Fatigue Behaviour
of Defective Material under Multiaxial Loading
in 1045 Steel
Hassine Wannes1, Anouar Nasr2, and Chokri Bouraoui1
1
LGM, université de Monastir ENIM, Avenue Ibn Eljazzar, Monastir 5019
hassine.wannes@gmail.com, chokri.bouraoui@enim.rnu.tn
2
LGM, EPEIM, Avenue Ibn Eljazzar, Monastir 5019
anouar.nasr @hotmail.fr
Abstract. This paper presents a new approach to predicting the HCF behaviour of
defective material submitted to Multiaxial loading .Defects resulting from the
casting processes are simplified to semi-spherical pore at surface of specimen.
Finite Element (FE) method was used to determine stress distribution around
defect. Papadoupolos criterion was used to evaluate stress equivalent around
defect. A definition of affected area was given, which is the area close to surface
defect where papadoupolos criterion was violated .The evolution of the affected
area, with the amplitude of loading and defect size, allows us to determine fatigue
limit for different defect sizes. Results of the new approach are in good agreement
with experimental results and show that affected area is a good parameter to
characterise the influence of a defect on fatigue behaviour.
1 Introduction
The fatigue limit of components containing inherent defect is of great importance

for the industrial application and always depends on morphology defect and the
loading mode. The designer needs to compromise between the fatigue resistance
of the component and the allowable defect size due to the process. Consequently,
the influence of the defect behaviour must be characterized to correlate the fatigue
limit into size of defects. The objective of this paper is to study the triaxiality
repartition of stress gradient around defect and correlate fatigue limit of defective
material into defect size.
Several ways are explored and different authors have proposed fatigue limit
calculation based on the size of defect .Murakami [1] proposed an interesting
approch based on experimental results .A geometrical parameter, √ , which
is the square root of the projected area of defect on the plane perpendicular to the
direction of the maximal principal stress, is used to model the defect size. This
approch suggests a practical relationship between the fatigue limit, hardness and
this parameter of size.

508 H. Wannes, A. Nasr, and C. Bouraoui
Endo [2] has extended Murakami’s criterion to proportional biaxial loading.

According to this criterion, the ratio between torsion and tension limits is
supposed constant whatever the size of the defect.
Using numerical tools other approach have been proposed as The CDM
(Critical Distance Method) criterion which is based on measurement of a criterion
at a given distance from the surface of defect[3 ]. The last criterion leads to very
good results in the case of small defects as shown by Leopold and Nadot [4 ].
Recently, Nadot and Billaudeau [5] proposed an approch to determine the
influence of defect size and geometry on fatigue behaviour .They use the gradient
of the maximal hydrostatic stress, which was taken into account to propose the
HCF criterion for defective material .
In the present paper we propose a new approch to correlate fatigue limit to
defect size. Triaxiality of stress gradient around defect was taken into account. In
the critical plane, perpendicular to the maximal principal stress, we define the
affected area, which is the area close to defect where the Papadoupolos criterion
was violated .The evolution of affected area versus amplitude of loading and
defect size helps us to calculate the fatigue limit for each defect size .The
approach was applied for fully reversed tension and torsion for1045 steel. Results
are in good agreement with experimental and show that affected area is a good
parameter to characterise the influence of a defect on the fatigue behaviour.
2 Material and Experimental Database
This study was carried on the 1045 carbon steel. It has long been used in aircraft
and automotive industry. An experimental database containing Multiaxial fatigue
results for 1045 steel with induced defects has been previously published [6].
3 The 3D Simulation of Stress Gradient around Defect
3.1 F E Analysis
Defects are always the preferential sites for crack initiation .So we need to
characterise the stress distribution around the defect. Finite Element simulations
have been conducted for each size of defect and loading mode. Symmetry and
boundary conditions are applied . Plasticity occurs around defect so that all
simulations are conducted with an elastic- plastic Prandtl-Reuss type model with a
linear isotropic hardening law. Meshing is refined and optimized in the critical
zone around the defect. The cyclic loading is included in the cyclic stabilised
experimental stress-strain curve used to identify the model .
3.2 Stress Distribution around Defect

The plane perpendicular to maximum stress direction is the highest loaded plane
(HLP) .Experimental works provides that the crack initiated always at the tip of
New Approch of High Cycle Fatigue Behaviour of Defective Material 5009
the defect in the maximum m shear plane [7] .And the macroscopic crack that leadds
to failure of the sample propagate
p in HLP. So that we will be interested in thhe
distribution of stress on th
his plane (Fig.5).
(a) (b)
ound defect on HLP (S22 maximal principal stress) (a) Tensioon
Fig. 1 Stress distribution aro
loading = 180MPa (b) to
orsion loading =150 MPa
A calculation of Papad
doupolos equivalent stress in the highest loadeed
plane HLP allows us to deeduce the following results:
− is almostt constant along the equator of defect (AB) with an erroor
that doesn’t exceed
d 4% .
− The distribution off Papadoupolos equivalent stress is independent of thhe
chosen direction.
As first approximation to
o interpolate the variation of in HPL we can propose thhe
following equations:
• Fully reversed tensio

on:
= * +1) (1)
/
• Fully reversed torsio
on
= *( +1) (22)
/
Where:
: Defect freee fatigue limit fully reversed tension.
: Defect freee fatigue limit fully reversed torsion.
: Amplitudee of tension applying load
: Amplitudee of torsion applying load
R : Radius of defect.
d
r : Distance a from defect centre to a considered point on HLP
These equations, (1) an
nd (2), give good results with error less than 5% .
4 New Approch of High Cycle Fatigue Behaviour of

Defective Material
4.1 Methodology
From the results of numerical simulations, we can conclude that the variation of
is independent of the chosen direction. In fact, on a centred outline at the
centre of defect is constant .Our task is to study the variation of
in HLP.
In this plane, there is an area, close to the defect, where the Papadoupolos
criterion violated even with a loading less than the endurance limit of the defective
device. Consequently we are led to introduce the following definition:
Affected area
The affected area by the gradient of stress for a given defect and an applied
load is the area where the Papadoupolos equivalent stress is greater than or equal
to ( = =169MPa)
Fig. 2 Affected area in HLP
The approach that we suggest consists in following the evolution of the affected
area in relation to the loading and the size of the defect. When the loading grows
up, the affected area evolves.
To the loading of endurance limit corresponds an affected area limit. Thus, to a
given defect under a given loading; if affected area affected area limit, there is
no rupture before 10 cycles.
At the endurance limit we have: affected area = affected area limit.
Using equations (1) and (2) the affected area as defined above is determined
by the following equations:
• Fully reversed tension
.
affected = ( D 1 √area (3)
• Fully reversed torsion:

Affected = ( D 1 √area (4)
Table 1 Experimental results [6]
Fully reversed tension

√area m σD MPa Affected area MPa.
170 195 15630 24378
400 152 25140 24100
900 130 35280 24418
Fully reversed torsion
√area τD Affected area
300 157 20642 31000
400 145 42105 29700
900 128 57857 30788
Using experimental data base table 2.we note that at fatigue limit we have a
relationship between affected area limit and the fatigue limit as bellow:
* = ct = 24000 MPaμm in case of reversed
fully tension.
* =ct = 30000 MPaμm in case of fully
reversed torsion.
Affected area limit was calculated by equations (3) and (4) using the fatigue
limit value in each cases
Hypothesis
For fully reversed tension and torsion the product of fatigue limit by the square
root of is almost constant and independent of defect size .
The proposed approach consists on the following points:
♦At first we plot curves : = f( equations (3) and (4)
for many defect sizes.
♦ Secondly, from a medium size of a defect we deuce or
♦Then at the same graphic we plot the curve: =
or
♦Finally, the crossing points allow as to determinate the fatigue limit for each
defect size.
To summarize necessary steps for the proposed approach, we can use the
algorithm fig.3.
Fig. 3 Algorithm to simulate Kitagawa diagram
4.2 Applications
4.2.1 Alternative Tension
In alternative tension case the affected area was determined by (3). From an
experimental fatigue limit of a medium defect size we calculate the parameter
.In this study, a defect size √ = 400μm corresponding to a fatigue
strength = 152 MPa. Using hypothesis 2 we deduce =24100 MPa.um.
In the same graphic we plot the curves:
♦ Affected area versus loading amplitude, for many defect sizes.
♦ Affected area limit versus loading amplitude: Affected area = .
We obtain the graphic Fig.4

Affectedareaum2 40000
Affectedarealimit
30000
size:50um
H size:100um
20000 size:200um
size:300um
10000 size:500um
size:700um
size:900um
0
120 140 160 180 200 220 240
Loadingamplitude(MPa)
Fig. 4 Application of proposed approach: fully reversed tension
From the crossing point H we deduce that a defect size √ = 300 μm

corresponding to fatigue limit 165 . Referring to experimental results
where =167MPa.Using all crossing points, for each defect size we can
deduce Kitagawa diagram in case of fully reversed tension for 1045 steel fig.6
240
Fatiguelimt(MPa)
Experimentalresults
220
Proposedapproach
200
180
160
140
120
100
0 200 400 600 800 1000
Defectsize(um)
Fig. 5 Kitagawa diagram tension loading = -1. Material 1045 steel
As shown, obtained results are in good agreement with experimental results and
the approach treats well the issue of small defects.
4.2.2 Alternative Torsion
In alternative torsion case the affected area was determinate by (4).From an

experimental fatigue limit of a medium defect size we calculate the parameter
.In this study, a defect size √ = 400μm corresponding to a fatigue
strength = 145 MPa. Using hypothesis 2 we deduce =29700 MPa.um.
In the same graphic we plot the curves:
♦ Affected area versus loading amplitude, for many defect sizes.
♦ Affected area limit versus loading amplitude: =
.
We obtain the graphic Fig.6
70000
Affectedareaum2
60000
50000 Affectedarealimit
size:170
40000
N size:200
30000 size:300
20000 size:400
size:600
10000
size:900
0
120 130 140 150 160 170 180
Loadingamplitude(Mpa)
Fig. 6 Application of proposed approach: fully reversed torsion
From the crossing point N we deduce that a defect size √ = 300 μm

corresponding to fatigue limit
155 . Referring to experimental results where =155MPa.
Using all crossing points, for each defect size we can deduce Kitagawa diagram
in case of fully reversed torsion for 1045 steel fig.14
Fatiguelimit(Mpa) 180
170 Experimentalresults
160
__ Proposedapproach
150
140
130
120
110
100
0 200 400 600 800 1000
dfectsize(um)
Fig. 7 Kitagawa diagram torsion loading = -1.Material 1045 steel
As shown, obtained results are in good agreement with experimental results but
the approach overestimates fatigue limit for small defects (less than 200μm).
5 Conclusions
An experimental database containing multiaxial fatigue results for 1045 carbon

steel [7] was used in this study. Defects are simplified as spherical pores at
surface of specimen. A simplified model was taken. Load and boundary
conditions are applied. Around defect the mesh is refined. After simulation in each
point the stress tensor is deduced. Papadoupolos criterion is used to calculate the
equivalent stress ( ).
The plane perpendicular to the maximum principal stress is the highest loaded
plane (HLP).The repartition of was studied in HLP.
From this study the following conclusions and remarks can be done:
♦ For tension and torsion cases the is constant along the equator
of defect and independent of the chosen direction in the HLP
♦ In each case an equation is proposed to interpolate in HLP.
♦ A definition of affeted area is introduced, which is the area close to
defect where Papadoupolos criterion is violated.
♦ At fatigue limit we obtain affected area .which depends on
defect size and loading mode.
♦ A parameter is introduced, which is the product of fatigue limit
by the square root of affected area and supposed constant for all defect
sizes (hypothesis 1)
♦ In each case of loading mode an experimental fatigue limit of
medium size is used to evaluate
♦ The following of affeted area evolution versus the amplitude of load

and defect size allows us to determine fatigue limit for each defect size.
♦ The proposed approch adopted in this study provides encouraging
result in alternative tension for 1045 steel and treats well the issue of small
defects .
♦ The representation of the equivalent stress gradient in HLP by a
quadratic form, equations (1) and (2) is better representative than using the
average gradient of hydrostatic pressure or equivalent stress.
References
[1] Murakami, Y., Endo, M.: The parameter model for small defects and non metallic
inclusion in fatigue strength: experimental evidences and application. In: Blom, A.F.,
Beevers, C.J. (eds.) Actes de “ Theoretical Concepts and Numerical Analysis of
Fatigue”, pp. 51–71. EMAS, Birmingham (1992)
[2] Endo, M.: Effects of small defects on the fatigue strength of steel and ductile iron
under combined axial/torsional loading. In: Ravichandran, K.S., Ritchie, R.O.,
Murakami, Y. (eds.) Proceeding of Small Fatigue Cracks: Mechanics, Mechanisms and
Applications, Hawaii, pp. 375–387 (1999)
[3] Susmel, L., Taylor, D.: Two methods for predicting the Multiaxial fatigue limits of
sharp notches. Far. Frac. Eng. Mater. Struct. 26, 821–833 (2003)
[4] Leopold, G., Nadot, Y.: Fatigue from an induced defect: experiments and application
of different Multiaxial fatigue approach. J. ASTM Int., 7 (2007)
[5] Nadot, Y., Billaudeau, T.: Multiaxial fatigue limit criterion for defective materials.
Engineering Fracture Mechanics (2005)
[6] Billaudeau, T., Nadot, Y., And Bezine, G.: Multiaxial fatigue limit for defective
materials: mechanisms and experiments. Acta Mater. 52, 3911–3920 (2004)
Determination of Stress Concentration for
Orthotropic and Isotropic Materials Using
Digital Image Correlation (DIC)
Mhalla Mohamed Makki and Bouraoui Chokri
Laboratory of Mechanical Engineering, University of Monastir Tunisia

mhallamaki@yahoo.fr,
chokri.bouraoui@enim.rnu.tn
Abstract. The objective of this study was to perform a tensile test on isotropic and
orthotropic plates with holes and measured deformations around their hole by us-
ing the method of digital image correlation (DIC). The results of this measure
were used to determine the local stress at the edge of the hole and measured the
net and global stress concentration factor for isotropic and orthotropic materials
and see the difference between the two types of materials.
This paper presents an analysis of the net and global stress concentration factor
for different hole diameters and their influence on the isotropic and orthotropic
material using a camera DIC.
Keywords: orthotropic, isotropic, plate with hole, stress concentration factor,

digital image correlation DIC.
1 Introduction
The stress concentration is a local phenomenon that increases the stresses in an

area having a geometrical modification of the part. This discontinuity appears in a
part of the structure or with the presence of a cut (hole, notch ...). The stress con-
centration area is often the site of initiation failure, but may also be the cause of a
sharp break in the case of a brittle material. Heywood [1] Peterson [2] and Pilkey
[3] studied several forms of stress concentration for isotropic materials with a
wide range of holes. Howland [4] determined the solution of the rectangular iso-
tropic plate problem with a hole in the center with a tensile load. Peterson and
Heywood introduced different equations for a finite plate with different aperture
shapes. Lekhnitskii [5] and Tan [6] introduced different formulations for the stress
concentration in orthotropic materials. Lekhnitskii derived an equation for an infi-
nite plate with circular holes. And for a finite orthotropic plate Tan has a various
equations.

518 M.M. Makki and B. Chokri
2 Digital Image Correlation (DIC)
The camera allows the use of in situ recording of a burst frame. These are used to
measure the displacement fields and local deformation. The principle of image
correlation is to compare local travel in the plane perpendicular to the camera be-
tween two different images. Comparing a said deformed image and the reference
image allows the measure of local movements (Figure 1). The calculations do not
take into account the movement of solid body. From these measurements, the local
deformations are deducted in the directions of the study design. To determine the
displacement at point of the specimen, it is divided into small images and is
squared. The grid represents the space between two thumbnail images, they may
intersect. The user can therefore play on two adjustable parameters are the size of
the thumbnail images and size of the grid. Although it appears to have the perfect
pattern thinnest and finest grid in order to gain accurate measurements, this confi-
guration is not always the most suitable. When the grid is too fine, the correlation
has a noise significant measurement.
Fig. 1 Principle of image correlation: determining an optimal field of displacement by im-

age analysis. Reference image and images solicited are devisees of series thumbnail images
(set of pixels) [7]
This method is implemented on plate specimens with hole than is covered with
a white paint before the test. Is then carried out by spraying an artificial marking
paint with a speckle pattern on the surface of the material, via a black spray. The-
reafter, the image correlation is performed using the standard software
ICASOFT® [Icasoft] developed by Fabrice Morestin (Mguil-Touchal and al.1996)
[7] to LaMCoS laboratory of INSA Lyon (Example Icasoft image with Figure 2).
The principle is used to match two (original and distorted) images by comparison
pixel by pixel with using a criterion based on the grid level in order to evaluate the
displacement field and the corresponding strain.
Determination of Stress Concentration for Orthotropic and Isotropic Materials 519
.
Fig. 2 Example of an image with camera DIC
The equipment used for the tests of uniaxial tension is: traction machine EZ20
with computer controlled device and image acquisition, including a lamp for light-
ing. The tests summers Films has captured with the camera positioned facing the
test, all using software (Icasoft, INSA Lyon, LAMCOS-MSE) to determine dis-
placement fields and strain. See Figure 3:
Traction
machine
Lamp
Camera DIC
PC
Software
Icasoft
Fig. 3 Diagram of the mechanical acquisition data and digital images with a uniaxial tensile
test.
Tensile tests were carried out in a universal testing range of the EASY TEST
Machine combine flexibility with exceptional usability, EZ20 model with trans-
verse hydraulically setting with constant velocity about 2 mm / min, with ambient
temperature.
Tensile tests were performed according to standard ASTM D3039 [8] (Figure
4), using a minimum of twelve specimens (The most current specimens to ASTM
D3039 has a constant rectangular cross-sectional dimensions: 250 mm long x 25
mm wide x 2 mm thick test .A typical speed for standard test samples of 2 mm /
min) for each family of laminate.
Fig. 4 Standard ASTM D3039 [20]
The specimens were made from carbon / epoxy with 0.22 mm thick. The tapes
were strength carbon fibers (T700) high strength unidirectional of M21 Hexcel
epoxy resin (35% resin). The plate contains eight unidirectional layers [0/90] 2s
laminate with a total thickness of 1.76 mm. These tacking sequences have helped
to get an overview of the evaluation with different types of damage in notched
specimens under uniaxial loading using DIC.
The specimens were prepared by bonding the end of glass fiber / epoxy lami-
nate legs. This procedure resulted in the available 150 mm2 of the sample for the
tensile test surface. Figure below shows the laminated specimens with different
diameters:
Fig. 5 specimens of plate
4 Results
With an ambient temperature, it was two families of tests: steel and composite
plates with different hole in the center with the same size and speed of loading
(Figure 6) .on notice the break in the steel is still in hole and perpendicular to the
loading, against composite the breakage is around the hole but the orientation of
crack always followed the direction of tissue .In our example we have a orienta-
tion tissue [0.90]2s. So the crack almost perpendicular to the load.
Fig. 6 Tensile test for a composite plate with hole (a=4mm)
Thereafter we can plot the stress-time curve for the composite plate (Figure7)
such that we can see the first damage and then the fiber break with the total damage:
Fiber
break
First dommage
Fig. 7 Stress-Time curve for a composite plate with hole (a=4mm)
400
ı d=3 400
ı d=4
300 d=5
300 d=5.5
d=7
200 200
d=9
d=7
100 100
0 İ 0 İ
0 0,01 0,02 0,03 0,04 0 0,005 0,01 0,015 0,02 0,025
Ͳ100
Fig. 8 Stress-Strain curve for a steel plate Fig. 9 Stress-Strain curve for a composite
with different diameters plate with different diameters
failure stress
400
300
200
100
0
3 4 5 6 7 8 9 a
10
Fig. 10 Variation of failure stress with a different hole diameter
Fig. 11 Image of damaged Steel plate Fig. 12 Image of damaged steel plate with
camera DIC
Fig. 13 Image of damaged composite plate Fig. 14 Image of damaged composite plate
with camera DIC
Before beginning the analysis of the diameters influence on the intensity of

stress concentration, it is necessary to present the general appearance of the stress
respect to strain (Figures 8-9).
In addition, it is important to know the failure stress of each hole diameter to
estimate its influence on the resistance plate (Figure 10).
Note that the failure stress down with increasing of notch diameter .We con-
clude that the hole size has a very important role in the resistance of plaque rup-
ture. For this we used the method of image correlation with DIC camera that gives
travel and local deformations to determine the local stresses at geometric disconti-
nuity. Thereafter we can see in Figures (11-18) images of holes in steel and com-
posite plates given by DIC camera:
Figure (15) shows the variation of the longitudinal strain (ε11) along the y axis
and passing through the center hole, this variation measured from DIC camera.
In These planar structures, the state of stress is fairly well represented by a plane
stress tensor, that:
İ0,12
11
0,1
0,08
0,06
Isotropic material 0,04
Orthotropic material
0,02
0 Y pixel
1,38E+01 1,58E+01 1,78E+01 1,98E+01 2,18E+01 2,38E+01
Ͳ0,02
Fig. 15 Variation of the longitudinal strain (ε11) along the y axis
Subsequently, the longitudinal stress around the hole can be calculated from the
deformation field (ε11 and ε22) obtained from DIC:
Finally we can plot the stress concentration (Kt = σ /σ avec σ σ )

with a different hole diameters for steel and composite (carbon / epoxy) materials.
524 M.M. Makki and B. Chokkri
5 Theoretical Co
oefficient of Global Stress Concentration
Theoretical coefficient off stress concentration is defined as the ratio of the actuual
maximum stress in the zo one of discontinuity (notch, hole, for example) to the noo-
minal stress in the section
n:
Kt = σ /σ
σ : Is calculable by numerical
n methods such as finite element method or bby
analytical methods for siimple geometries. It is also measured by the techniccal
analysis of experimental constraints
c such as photo-elasticity or DIC methods.
σ : Is calculable using the formulas of strength materials, considering thhe
specimen like a bar or plaate without account the geometric discontinuity.
σ =
With F: traction, uniforrm on edges of plate.

w: width of the plate.
h : plate thickness.
6 Evolution of th
he Global Stress Concentration with Hole
Diameter
6.1 Case of Isotrop

pic Material
For a rectangular plate wiith a circular hole in the middle and subjected to a tensiile
loading in one direction.. It occurs around the hole increasing the value of thhe
stress that is characterized
d by the stress concentration factor KTg Howland [9]:
6.2 Case of Orthotrropic Material (Composite)

For a thin composite platee (assumed orthotropic) infinite (w >> r) with a central
hole. The coefficient Kt iss obtained by Lekhnitskii [5]:
=
=1+ 2 / /
Determination of Stress Con
ncentration for Orthotropic and Isotropic Materials 5225
In our case, the study focuses on an infinite orthotropic plate 3D is made oof
carbon /epoxy (HR [0/90] 2s) .According of Table 1 Kt equal to 5.24 for an inffi-
nite orthotropic plate.
According to the inveestigation of Heywood [1] on a rectangular orthotroppic
plate with center hole und
der the action of unidirectional axial load, the factor is inn-
troduced in the form of th
he following equation:
With
7 The Theoretica
al Value of the Net Stress Concentration
Another way to determinee the stress concentration is the application of the averagge
n section instead of the hole width of the plate (global
normal stress across the net
surface).
σ =
This concentration is know wn as the net concentration of stresses K T [9, 1] and

can be related to the globaal stress concentration factor K T by:
K T = K T (1-a/w)
8 Validation by Comparison with the Results of the FE

Analysis
8.1 Modelisation with Finite Elements

To validate our work can be compared with the results found by finite elements
analysis and with literature.
A simulation under "ABAQUS 6.12" software was performed with a mesh-like
continuum Shell hexagonal, progressive and refined around the hole. Two systems
of different materials are considered:
Table 1 Material Properties orthotropic carbon / epoxy (T700 / M21 [0 90] 2s)
E22 G12 ν12 σ11T σ11c τ12

Table (GPa) (GPa) (MPa) (MPa) (MPa)
2 E11
(GPa)
148 7.8 4.5 0.35 2375 1465 95
Table 2 Properties of Isotropic Steel Type
Elastic modulus E(GPa) Poisson ratio v mass density ρ (kg / m3)

210 0.3 7800
A tensile load of 119 MPa at one and a lock on the opposite. The fineness of
the mesh is optimized in order to have a steady value of maximum stress. To re-
duce the computation time while maintaining good accuracy, only a quarter of the
plate is meshed with progressive mesh designed to produce a very fine mesh
around the hole and larger elsewhere (Figure 16).
Fig. 16 gradual meshing of the plate

8.2 Evolution of the Stress Concentration with Diameter Hole

(Isotropic Material):
The results found after an experiment method with camera DIC and with
ABAQUS simulation are presented in Table 3:
Table 3 Isotrope Material
a a/w Kt Ktg Ktg Ktn Ktn

infinite DIC FEA DIC FEA
0,2 0,02 3 3,0048 3,004 2,82 2,78
2 0,1 3 3,0582 3,04 2,61 2,53
4 0,2 3 3,1900 3,15 2,34 2,21
6 0,3 3 3,4097 3,3 2,22 2,07
8 0,4 3 3,7543 3,6 2,12 1,91
10 0,5 3 4,2937 4,1 2,06 1,81
For a given width, this ratio Kt equal to 3 for an infinitely thin plate with a cir-
cular hole and thereafter the global stress concentration factor Ktg increases if the
hole diameter increases against the net stress concentration factor Ktn decreas-
es.(Figure 17 and 18):
Ktg
11,5 DIC
10,5 FEA
9,5
Heywood
8,5
Howland
7,5
6,5
5,5
4,5
3,5
a/w a/w
2,5
0 0,2 0,4 0,6
Fig. 17 Variation Ktg isotropic with ratio a/w

Ktn
3,5 DIC
3 FEA
2,5 Heywood
2 Howland
1,5
0,5
0 a/w
0 0,2 0,4 0,6
Fig. 18 Variation Ktn isotropic with ratio a/w
For isotropic material we can notes that the results of finite element analysis are
very close to the results of DIC camera but for the stress concentration Ktg there is
a difference from a / w = 0.4 compared with Heywood and Howland.
8.3 Evolution of the Stress Concentration with Diameter Hole

(Orthotropic Material):
Table 4 compares the stress concentration between the finite element analysis and
image correlation method:
Table 4 Orthotropic Material
a a/w Kt Ktg Ktg Ktn Ktn

infinite (DIC) (FEM) (DIC) (FEM)
0,2 0,01 5,24 3,4 5,41 2,9 4,81
2 0,1 5,24 3,9 6,02 2,72 4,12
4 0,2 5,24 4,52 6,32 2,47 3,2
6 0,3 5,24 5,26 6,62 2,21 2,8
8 0,4 5,24 6,43 7,02 1,82 2,1
10 0,5 5,24 7,12 7,22 1,62 1,7
Ktg
14 DIC
12
FEA
10
Heywood
8
0 a/w
0 0,2 0,4 0,6
Fig. 19 Variation Ktg orthotropic with ratio a/w
On observed that a / w is increased when the edges of the plate is closer to the
hole. Consequently, the lines of force will be more compressed and Ktg increases.
6Ktn DIC
5
FEA
4
Heywood
3
0 a/w
0 0,2 0,4 0,6
Fig. 20 Variation Ktn orthotropic with ratio a/w

Note that the net stress concentration decreases if the ratio a / w increases and
unlike different with global stress concentration factor Ktg.
It has almost the same variation of KTn respect to a/ w for isotropic and ortho-
tropic material.
9 Conclusion
In this work we have used the image correlation method with DIC camera to
measure the tensile stress concentration and which gives consistent results with
the finite element analysis and thereafter it is found that the DIC method given a
best results for a planar structure isotropic or orthotropic.
The results obtained show that the presence of a geometrical discontinuity in a flat
plate has a significant effect on the stress concentration to an isotropic and ortho-
tropic material. In addition we see the evolution of global and net stress concentra-
tion respect to different hole diameters for composite and steel.
Finally, the DIC approach don't provide good results for composite structures
because this method can be show only the surface damage against the composite
material having many damage inside the structure (delamination ,cracking ma-
trix….).
Acknowledgements. I want to thank deeply Mr. Jury members who have done me the hon-
or to judge my work and have improved my presentation with helpful comments.
References
[1] Heywood, R.B.: Designing by photoelasticity. Chapman and Hall (1952)
[2] Peterson, R.E.: Stress concentration factors. John Wiley & Sons (1974)
[3] Pilkey, W.D.: Peterson’s stress concentration factors, 2nd edn. John Wiley & Sons
Inc., New York (1997)
[4] Howland, R.C.J.: On the stresses in the neighborhood of circular hole in a strip under
tension. Phil. Trans. Roj. Soc. 229, 49–86 (1929)
[5] Lekhnitskii, S.G.: Theory of Elasticity of an Anisotropic Elastic Body, 1st edn., p. 4.
Holden-Day, San Francisco (1963)
[6] Tan, S.C.: Finite-width correction factors for anisotropic plate containing a central
opening. J. of Composite Materials 22, 1080–1097 (1988)
[7] Mguil-Touchal, S.: Une technique de corrélation directe d’images numériques. Thèse
INSA Lyon (1996)
[8] ASTM D3039/D3039M-00. Standard test method for tensile properties of polymer ma-
trix composite materi-als. American Society for Testing Materials [CD-ROM] (2004)
[9] Howland, R.: On the stresses in the neighborhood of a circular hole in a strip under
tension. Philosophical Transactions of the Royal Society A 229, 67 (1929-1930)
The Extended Finite Element Method for
Cracked Incompressible Hyperelastic
Structures Analysis
Mehrez Zaafouri1, Mondher Wali1, Said Abid1, Mohammad Jamal2,

and Fakhreddine Dammak1
1
Mechanical Modelisation and Manufacturing Laboratory (LA2MP), National Engineering
School of Sfax, B.P W3038, Sfax, University of Sfax, Tunisia
{zaafourimehrez,mondherwali}@yahoo.fr, said.abid@issatgb.rnu.tn,
Fakhreddine.Dammak@enis.rnu.tn
2
Faculté des Sciences Ben M’Sik, Laboratoire dÊIngénierie et Matériaux (LIMAT),
Université Hassan II de Casablanca, B.P. 7955, Sidi Othman, Casablanca, Morocco
jamalmoham@gmail.com
Abstract. This paper aims to examine the contribution of the extended finite ele-
ment method (XFEM) in finite strain fracture mechanics problems. A generalized
neo-Hookean hyperelastic material is considered in an incompressible plane stress
approximation. The accuracy of the implementation is demonstrated by a series of
numerical tests.
Keywords: XFEM, Hyperelasticity, Crack Tip fields, Finite strain.
1 Introduction
The numerical simulation of the cracked structures represents an important issue

for many industrial sectors. Among the numerical tools developed to simulate
cracked structures one finds the finite element methods (FEM).
A number of approaches have been developed within the FM framework in
over the years, which makes it as a most suited method to investigate the asymp-
totic stress and displacement fields at the crack tip. However, FEM requires that
the crack surface coincide with the finite elements boundary and a conformal
mesh is needed. Therefore, frequent remeshing is necessary for crack growth
modeling.
To overcome these difficulties, several approaches were proposed, the most re-
cent is the extended finite element method (XFEM) introduced by (Moës et al.
1999) and ameliorated by (Dolbow et al. 2000). It is an extension of the FEM ded-
icated to the modeling of geometrical discontinuities The principal characteristic
of the XFEM is the decoupling between discontinuity and the grid: the crack is

532 M. Zaafouri et al.
represented independently of the grid by introducing enrichment functions into the

finite elements approximation base taking into account the displacement jump
along the crack as well as the asymptotic displacement field near the crack tip.
And since, several problems of fracture mechanics were solved by the mentioned
approach (Fries and Belytschko 2010) and (Mohammadi 2008). Although this al-
ternative is promising, it remains often held for the linear elastic problems of frac-
ture mechanics.
The extension of the XFEM method to non-linear fracture mechanics raises ma-
jor difficulties and a few works are developed for this subject. (Dolbow and Devan
2004) present a geometrically non-linear assumed strain method that allows for
the presence of arbitrary intra-finite element discontinuities. In (Legrain et al,
2005), the others used the XFEM method in incompressible plane stress problems
with classical Neo-Hook materials. Later, (Karoui et al. 2014) present an exten-
sion of XFEM method to large deformation of cracked hyperelastic bodies in
plane strain approximation.
The major problem in XFEM method in to large strain of cracked hyperelastic
bodies is the singular enrichment witch concern nodes whose support contains the
crack tip. In linear elasticity, the asymptotic displacement field near the crack tip
is used as singular enrichment. In case of hyperelasticity, some authors obtained
the analytical expression of the asymptotic fields within the simplified framework
of plane elasticity. Theoretical results were established by (Knowles and Sternberg
1973, 1974) for compressible Blatz–Ko materials in plane strain, by (Stephenson
1982) for incompressible materials in plane strain, by (Knowles and Sternberg
1983) for neo-hookean material in plane stress and by (Geubelle and Knauss
1994) for incompressible generalized neo-Hookean material in plane stress. (Long
et al. 2011) presents a new method to determine the asymptotic stress and defor-
mation fields near the tip of a Mode-I traction free plane stress crack. There analy-
sis is based on the fully nonlinear equilibrium theory of incompressible hyperelas-
tic solids. Two types of soft materials are used: generalized neo- Hookean solids
and a solid that hardens exponentially. For the generalized neo-Hookean solids,
this method is able to resolve a difficulty in the work by (Geubelle and Knauss
1994).
In this paper, a Neo-Hook material is considered in incompressible plane stress
approximation. This paper is organized as follows: in the next section, the consti-
tutive equations of the problem are presented. Weak form and linearization are
given in section three. In section four, technical aspects of the implementation of
the XFEM in non-linear elasticity are presented. Numerical examples of this im-
plementation, which show the performance of the method, are given in section
five. Finally, same concluding remarks and possible extensions to this work are
given.
The Extended Finite Element Method 533
2 Constitutive Equation
Let us consider the deformation movement defined by the function

φ ( X ,t ) : B × R → R and F ( X ,t ) is the deformation gradient with X ∈ R
3 3
indicate the position of a point in the reference configuration.
∂φ
F = , J = det ( F ) (1)
∂X
The symmetric right Cauchy-Green tensor C is defined by
C=F F
T
(2)
The hyperelasticity implies the existence of an energy function dependant on

the tensor C. In the case of isotropic hyperelasticity, the energy function is a func-
tion of C as follows
ψ = ψ ( I1 ,I 2 ,I3 ) (3)
where I1 , I 2 and I 3 are the three invariants of C tensor
(I )
1
I1 = tr ( C ) , I 2 = − trC I 3 = det ( C ) = J
2 2 2
1
, (4)
2
In the case of incompressibility constraint, the third invariant of the right

Cauchy-Green tensor is equal to unit. Therefore, the principal invariant I1 and I 2
are the only independent deformation variables.
ψ = ψ ( I1 ,I 2 ) , J =1 (5)
To determine the constitutive equations for isotropic hyperelastic materials in

term of strain invariants, we consider the differentiation of ψ with respect to ten-
sor C. By means of the chain rule of differentiation we find, (Dammak et al. 2007
and Jarraya et al. 2011), the most general form of the second Piola-Kirchhoff
stress tensor in the incompressible case
2
∂I i
S = 2∑ψ ,i
−1
= α I +α C +α C
1 2 3
(6)
i =1 ∂C
−1
Where I , C and C are in plane tensors. However, the most general form
of the elasticity tensor  in terms of the principal invariants is given by the fol-
lowing expression
∂S ∂ ψ ( I1 , I 2 )
2
 =2 =4 (7)
∂C ∂C ∂C
 = δ I ⊗ I +δ
1 2
(I ⊗C +C ⊗ I)+δ 3
(I ⊗C −1
+C ⊗I +
−1
)
(8)
δ C ⊗C +δ
4 5
(C ⊗ C −1
)
+ C ⊗ C + δ C ⊗ C + δ I C + δ I ijkl
−1 6 −1 −1 7
−1
8
where the following notations are used
(I ) (C )
1 1
C
−1
ijkl
=
−1
ik
C
−1
jl
+C
−1
il
C
−1
jk
, I ijkl = (δ ik
δ jl + δ il δ jk ) (9)
2 2
The terms α ( )i =1,3

i
and δ iso
i
( ) i =1,8
are given in (Dammak et al. 2007 and Jar-
raya et al. 2011).
3 Weak Form and Linearization
The numerical solution with the finite element method is based on the weak form
of equilibrium equations. The three dimensional form of the latter in the total La-
grangian formulation is given as
G= ∫ V
S : δ E dV − Gext = 0 (10)
where Gext is the virtual work of the external forces, S is the second Piola-
Kirchhoff stress tenser and δE represent the virtual Green-Lagrange strain
tensor.
1
E = (C − I ) (11)
2
( δF )
1 1
δE = δC = F + F δF
T T
(12)
2 2
If the following kinematics is adopted
φ= x = X +u (13)
then, in 2D, F and δE can be written as
F = [ x ,1 x ,2 ] , (14)
with
⎡ x ,1 ( .) ,1 ⎤
T
⎢ ⎥
δE = B δu B= x ,2 ( .) ,2
T
, (15)
⎢ ⎥
⎢⎣ x ,T1 ( .) ,2 + x ,T2 ( .) ,1 ⎥⎦
where ( .),i is a material derivative. With these matrix notations, the weak form
(10) can be written in the following form
∫ ( Bδ u ) S dV − Gext ( δ u ) = 0
T
G= (16)
V
In the optics of a resolution with the Newton method, its linearization is devel-
oped in two parts: material and geometrical
∫ ( B δ u )  ( B Δu ) dV + ∫ ∇ (δ u ) :∇ ( Δu ) SdV
T
DG ⋅ Δu = (17)
V V
where  is the elasticity material tensor given in equation (8).
4 Extended Finite Element Formulation
Within the Xfem framework, the crack is taken into account by two different types
of enrichment: discontinuous enrichment and singular enrichment (Fig. 1). The
energy function of generalized neo-Hookean materials is
μ ⎛⎡ ⎤ ⎞
n
b
ψ = ⎜ 1 + ( I − 3) − 1 ⎟ (18)
⎢
2b ⎝ ⎣ n ⎥
⎦ ⎠
As in (Legrain at al. 2005), the singular enrichment function that will be used is
• Discontinuous enrichment
 Singular enrichment
Fig. 1 Xfem enrichment
F= r { 1/ 2
sin (θ / 2 ) } (19)
The numerical implementation of the presented formulation is based upon a

four node non-linear plane stress element. Initial geometry and displacement
vectors are approximated by
X= ∑N X i
i
(20)
i∈I n
u= ∑ N u + ∑ N (H − H )a + ∑ N (F − F )b
i
i
i
i i
i
i i
(21)
i ∈I n i∈ I H i ∈I F
δ u = ∑ N δ ui + ∑ N ( H − H i ) δ ai + ∑ N ( F − Fi ) δ bi
i i i
(22)
i ∈I n i ∈I H i ∈I F
Matrix B , in equation (15), will be at node (i):
B = ⎡⎣ Bu Bb ⎦⎤
i i i i
Ba (23)
⎡ T
x ,1 N ,1
i
⎤ ⎡ T i
x ,1 M ,1 ⎤
⎢ ⎥ ⎢ ⎥
Bu = Ba =
i T i i T i
x ,2 N ,2 , x ,2 M ,2 (24)
⎢ ⎥ ⎢ ⎥
⎢⎣ x ,1T N ,i2 + x ,T2 N ,1i ⎥⎦ ⎢⎣ x ,1T M ,i2 + x ,T2 M ,i1 ⎥⎦
⎡ T i
x ,1 L,1 ⎤
⎢ ⎥
Bb =
i T i
x ,2 L,2 (25)
⎢ ⎥
⎢⎣ x ,1T Li,2 + x ,T2 Li,1 ⎥⎦
With M =N
i i
(H − H ) i
, L =N
i i
(F − F ) i
(26)
and ( ⋅) ,
, i = 1,3 are materials derivatives.
i
Finally, the element residual and material and geometric tangent operator are
expressed as:
∫ ∫ B  B dV
T T
Gin = ΚM =
i i ij i j
B SdV , (27)
V V
⎡ f ( Ni ,N j ) I (
f N ,M )I i j
( )
f N ,L I ⎤
i j
⎢ ⎥
KG = ∫
⎢ f (M ,N ) I f ( M ,M ) I f ( M , L ) I dS
ij i j i j i j
S ⎥ (28)
⎢⎣ f ( Li , N j ) I f ( L ,M ) I
i j
f ( L , L ) I ⎥⎦
i j
where:
(
f N ,M
i j
) = N, (S i
1 11
j j
)
M ,1 + S12 M ,2 + N ,2 S 21 M ,1 + S 22 M ,2
i
( j j
) (29)
5 Numerical Example
In order to illustrate the accuracy and versatility of the extended finite element
method X-FEM in nonlinear fracture mechanical problems for large displacements
whose behavior is isotropic hyper-elastic kind, a numerical example is presented
in this section. The example is solved using both FEM and X-FEM techniques,
and the results are compared. For the purpose to perform a real comparison, the
number of elements in FEM and X-FEM meshes are taken almost equal. For X-
FEM a uniform meshes elements are used independent of the shape of crack.
While in FEM analyses, the crack is aligned with element edges and refined mesh
is applied near the crack region. All examples are simulated with the plane stress
approximation.
The example shown in Figure 2 illustrate a plate subjected to an imposed
displacement on its top edge equal to 4mm and its bottom edge is fixed.
538 M. Zaafouri et aal.
(a) (b) (c)
Fig. 2 (a) Plate with lateral ccrack under imposed displacement

(b) X-FEM mesh, (c) FEM m mesh (Abaqus)
Fig. 3 (a) XFEM deformed configuration

c (b) Abaqus deformed configuration
Plate geometry and maaterial properties are adopted from (Legrain et al, 20055).
The width and height of this plate are respectively w = 2 mm and h = 6 mm. A
crack of length a = 1mm is inserted at mid-height. The material is assumed to bbe
Neo-Hookean with μ = 0.4225.
The deformed configurations obtained by X-FEM and FEM methods are pre-
sented in figure 3. We notes that the XFEM field displacement discontinuity near
the crack is well represented and conformed to Abaqus results.
3,5 U2-XFEM
U2-Abaqus
U2 : Displacement
3
2,5
1,5
0,5
0 0,5 1 1,5
Distance X
Fig. 4 Comparison of crack U2-displacement
In order to make a comparison between two techniques, the deformed shape of

the crack obtained by the vertical displacements of the nodes situated on the edges
of crack are plotted in figure 4. The results are in perfect agreement proving the
advantage of the X-FEM method versus FEM.
6 Conclusions
In this paper, the extended finite element method (XFEM) is used to analyze the
finite strain fracture mechanics problems with a neo-Hookean hyperelastic materi-
al with an incompressible plane stress approximation. In order to demonstrate the
efficiency of the X-FEM method, a numerical example was considered showing
displacement field near the crack and the deformed shape. A comparison of
XFEM and FEM results was carried showing the reliability of XFEM method to
nonlinear fracture mechanic.
Acknowledgment. The authors gratefully acknowledged and express their gratitude to the
CMPTM (Tunisian-Moroccan Joint Standing Committee) under reference 13/TM42 for
their support and technical assistance of this research.
References
Dammak, F., Regaieg, A., Kallel Kammoun, I., Dhiab, A.: Modeling the law of trans-
versely isotropic hyperelastic behavior of elastomers. European Journal of Computa-
tional Mechanics 16(1), 103–126 (2007)
Dolbow, J., Moes, N., Belytschko, T.: Modeling fracture in Mindlin-Reissner plates with
the eXtended finite element method. Int. J. Solids Structures 37, 7161–7183 (2000)
Dolbow, J.E., Devan, A.: Enrichment of enhanced assumed strain approximations for
representing strong discontinuities: Addressing volumetric incompressibility and the
discontinuous patch test. International Journal for Numerical Methods in Engineer-
ing 59, 47–67 (2004)
Fries, T.P., Belytschko, T.: The extended/generalized finite element method: An overview
of the method and its applications. Int. J. Numer. Meth. Engng. 84, 253–304 (2010)
Geubelle, P.H., Knauss, W.G.: Finite strains at the tip of a crack in a sheet of hyperelastic
material: I. Homogeneous case. J. Elasticity 35, 61–98 (1994)
Jarraya, A., Kallel-Kammoun, I., Dammak, F.: Theory and finite element implementa-tion
of orthotropic and transversely isotropic incompessible hyperelastic membrane. Multi-
discipline Modeling in Materials and Structures (MMMS) 7(4), 424–439 (2011)
Karoui, A., Mansouri, K., Renard, Y., Arfaoui, M.: The extended Finite Element Method
for cracked hyperelastic materials: a convergence study. Int. J. Numer. Meth.
Engng. 100(3), 222–242 (2014)
Knowles, J.K., Sternberg, E.: An asymptotic finite-deformation analysis of the elastostatic
field near the tip of a crack. Journal of Elasticity 3, 67–107 (1973)
Knowles, J.K., Sternberg, E.: Finite deformation analysis of the elastostatic field near the
tip of a crack: reconsideration of higher order results. Journal of Elasticity 4, 201–233
(1974)
Knowles, J.K., Sternberg, E.: Large deformation near the tip of an interface crack between
two neo-Hookean sheets. J. Elasticity 13, 257–293 (1983)
Legrain, G., Moës, N., Verron, E.: Stress analysis around crack tips in finite strain problems
using the eXtended finite element method. Int. J. Numer. Meth. Engng. 63, 290–314
(2005)
Long, R., Krishnan, V.R., Hui, C.Y.: Finite strain analysis of crack tip fields in incom-
pressible hyperelastic solids loaded in plane stress. Journal of the Mechanics and Phys-
ics of Solids 59, 672–695 (2011)
Möes, N., Dolbow, J., Belytschko, T.: A finite element method for crack growth without
remeshing. International Journal for Numerical Methods in Engineering 46, 131–150
(1999)
Mohammadi, S.: Extended finite element method. Blackwell Publishing (2008)
Stephenson, R.A.: The equilibrium field near the tip of a crack for finite plane strain of in-
compressible elastic materials. Journal of Elasticity 12, 65–99 (1982)
Displacement Influence on Frequencies
and Modal Deformations of a Sandwich Beam
Idris Chenini1, Youssef Abdelli2, and Rachid Nasri3
1
Assistant to the ESIER: elkef road, 5 km Medjez elbâb, 9070, Beja
chenini.idris@yahoo.fr
2
Aassociate professor at the military academy, Address: Fondek ejjdid Grombalia, Nabeul
abdelliyoussef@gmail.com
3
Professor at the ENIT, Address: 37 B.P. Le Belvédère, 1002 Tunis
rachid.nasri@enit.rnu.tn
1, 2,3
Research laboratory on Applied Mechanics and Engineering (LR-MAI-ENIT)
Abstract. Composite structures are used in the aerospace, stars and automotive es-
pecially the structures made of composite sandwich panels which are subject to vi-
bration harmful sources of noise and mechanical failures. In this paper, we propose
to analyze the influence of the shapes of displacement fields on the vibration beha-
vior of a sandwich beam. For this energy method based on the minimum energy is
used to achieve the equations frequencies and a sandwich beam modes. The kinetic
and potential energies of the skins are, in turn, derived from the classical laminate
theory. Several fields of polynomial movements are tested. Other parts will study the
effects of rotational inertia, taking into account the bending energy of the body of
sandwich NIDA. We analyzed the frequencies and modes based on different para-
meters. The experimental data are obtained for recessed-free conditions limits excit-
ing near underrun using an impact hammer. The vibrational response is measured
with a laser vibrometer. The natural frequencies are obtained experimentally by
modal analysis. Numerical simulations complete this work for two types of sand-
wich Nomex paper and aluminum. The natural frequencies obtained from the
theoretical formulation for numerical solution of the system are compared with ex-
perimental results and the results of numerical simulation. The very good agreement
between the results shows that the model is correct.
Keywords: Displacement, energy, sandwich, Eigen frequency, modal deforma-

tion.
1 Introduction
Energy methods are among the most important methods to study the vibration
behavior of beams. The problem with these methods is that they require the

542 I. Chenini, Y. Abdelli, and R. Nasri
introduction of a field of kinematic ally admissible displacement or statically ad-

missible stress or both at once.
The accuracy of the method will therefore depend on the right choice of these
fields. In the literature, there are several choices of the field selection. But it is dif-
ficult to decide on their preferences depending on their application.
Mindlin [1] proposed a model to study the transverse shear of a thick sandwich
plate isotope; this model is known by the theory of Mindlin.
Yarlagadda Lesieutre and [2] have developed an analytical method based on the
theories of Rayleigh-Ritz to determine the influence of the change of bends orien-
tation directions, temperature and thickness of the laminated plates on the Fre-
quencies and the damping of the bending vibration of a composite plate.
Wang [3] compared the classical theories of thin plates with the theories of
Mindlin thick plates then he studied the bending vibrations of a rectangular face
and core isotropic sandwich panel.
Meunier and Shenoi [4] introduced the mechanical properties of the elements
which form the sandwich plate in the analytical equations of elastic-viscoelastic
model and the results of the differential equations give their frequencies and the
factors influencing them.
Soula et al. [5] have studied the influence of the vibration on the kinematic be-
havior of laminated plates. They used three theories plates (Kirchhoff, Mindlin
and Reddy) to estimate the natural frequencies and modal damping of symmetric
and antisymmetric plates.
Maheri and Adams [6] extended the tests used for measuring the damping of stra-
tified monolithic laminated beams in bending vibration for NIDA sandwiches
beams. The contribution due to the damping components, in particular the fiber
orientation of the skins, was considered. They compared their experimental results
with theoretical study.
Banerjee [7] applied the Timoshenko theory on a beam using the coupling of
bending and torsion and taking into consideration the rotational inertia. The cha-
racteristics of the composite are taken experimentally as if the material is homo-
geneous.
Nilsson et al [8] used equivalent homogeneous characteristics to study the
bending vibrations of a sandwich beam from the Hamilton integral taking into
consideration the transverse shear, the rotational inertia and distortion. They write
the dispersion equation giving the wave number for the three modes according to
the frequency for some boundary conditions and then they find the modal defor-
mations and dynamic stiffness, the formulation is then validated by a test.
Blevins [9] presented a detailed list of formulas for determining the natural fre-
quencies and their modes corresponding to various structure shapes.
On this paper, we study the influence of the displacement fields shape under
different boundary conditions of a sandwich beam on the natural frequencies and
modal deformations.
This study which is applied on two different beams shall give us more ideas
about the influence of various fields on the modeling accuracy.
Displacement Influence on Frequencies and Modal Deformations 543
An experimental study of two sandwich beams also allows us to understand

more this field by comparing the results of tests and calculations on the one hand
and various tests according to the excitation and measuring points on the other.
2 Mathematical Formulation
We choose an n order polynomial transversal displacement depending of the con-

stants (1..n = i-1) and the kinetic and potential energies are determined as follows.
The potential energy of a sandwich beam is the sum of the bending elastic defor-
mation energy of and traction of both upper and lower faces as well as the shear
and bending energy of the body.
Z
Y
c
X
b
a
Fig. 1 Nida Sandwich beam
The elastic deformation of the lower side can be written as follows:

f l ∂ 2W c
ε xB = ( z + )( ) (1)
2 ∂x 2
f u ∂ 2W c
ε ' xB = ( z − c − )( ) (2)
2 ∂x 2
E 0 a
2(1 − ν 2 ) ∫− fl ∫0
VBF = dz (ε xB ) 2 dx (3)
With the same method we determine the energy of elastic bending deformation
of the upper layer:
E c+ fu a
2 ∫c
V ' BF = dz ∫ (ε ' xB ) 2 dx (4)
2(1 − ν ) 0
To determine the tensile energy of one of the faces we must first determine the
normal force to this surface and its extension in the parallel axis to the face.
N FL : The normal Force at the lower face and N FU : the normal force at the
upper face
c + fu
N FU = ∫
0
N FL = ∫ σ f dz (5) And σ f dz (6)
− fl c
The Formula of σf which is the stress in an arbitrary face of the fiber is as

follows:
c ∂ 2W c
σ f = − E ( z − )( ) (7)
2 ∂x 2
Substitute (7) in (5) and (6) we obtain:
0 c ∂ 2W c c+ fu c ∂ 2W
N FL = − ∫ E ( z − )( )dz (8) And N FU = − ∫ E ( z − )( 2 c )dz (9)
− fl 2 ∂x 2 c 2 ∂x
Using the integral we determine the normal forces at the both the upper and the
lower faces:
fl ∂ 2W c fu ∂ 2W c
N FL = E (c + f l )( ) (10) N FU = − E (c + f u )( ) (11)
2 ∂x 2 2 ∂x 2
VNF : is the tensile energy of the lower layer of the normal force of this face.
ε xL : The elastic deformation in the x direction of the lower face.
ε xU : The elastic deformation in the x direction of the upper face.
1 a 1 a
2 ∫0 2 ∫0
V NF = N FL ε xL dx (12) V ' NF = (− N FL ).(−ε xL )dx (13)
The bending elastic deformation energy of the bottom layer per unit width is as
the following:
∂ 2W c
(14) and ε xU = − 1 (c + f u )( ∂ W c )
2
1
ε xL = (c + f l )( ) (15)
2 ∂x 2 2 ∂x 2
Substituting (10) and (14) in (12) we obtain:
E. f l .(c + f l ) 2 a ∂ 2W c 2
VNF =
8 ∫
0
(
∂x 2
) dx (16)
Substituting (11) and (15) in (13) we obtain:

E. f u .(c + f u ) 2 a ∂ 2W c 2
V ' NF =
8 ∫
0
(
∂x 2
) dx (17)
The body Shear energy (Vc) par unit width is as following:

1 c
dz ∫ G XZ ⋅ (γ XZ ) dx
a
Vc = ∫
2
(18)
2 0 0
∂
With: γ XZ = 1 ⋅ τ XZ (19) and τ XZ dx = − N FL dx (20)
G XZ ∂x
The body bending energy is as following:
2
1 a ⎛ ∂ 2W ⎞ c 3 .b
VFC = ∫ Ec .I .⎜⎜ 2 c ⎟⎟ dx With I= (21)
2 0 ⎝ ∂x ⎠ 12
EC : Young Modulus of core

From the previous formula we determine the total potential energy which is:
V = VCC + VFC + V NF + V ' NF + VBF + V ' BF
(22)
The kinetic energy of the vibration of the sandwich panel is equal to the sum of
kinetic energy of translational and rotational kinetic energy.
T = ECT + ECR (23)
2
1 a ⎛ ∂ ⎛ ∂W ⎞⎞
2
Where: E = 1 ρ a ⎛⎜ ∂W ⎞⎟ dx
∫ 2 ∫0 ⎜⎝ ∂t ⎝ ∂x
(24) And ECR = I⎜ ⎜ ⎟ ⎟⎟ dx (25)
⎝ ∂t ⎠
CT
2 0
⎠⎠
The vibration system is considered as conservative where Ai (i =1..n-1) are
constants. The resolve of the obtained system allows us to determine (n-1) natural
frequencies and their shape mode.
3 The Beam Results

First, we presented the mechanical and geometrical characteristic of test
specimens analyzed
Table 1 Geometrical and mechanical characteristics of used specimen
NIDA in Al NIDA in NOMEX

Length (a) in mm 250 250
Width (b) in mm 53 53
Thickness core (c) in mm 5 8
Each layer thickness f u = fl in mm 0.75 0.5
Density ρ in Kg/m3 573 221
Ec in Pa 130.106 2.5 .106
E p in Pa 70.109 8.63 .106
G XZ in Pa 5600.106 70.109
ν 0.33 0.33
Second, we take a beam where field checking the no translation and not the
non-rotation on the fixed end section x=0 is:
We present only the results where the differences between fields are observa-
ble. We mention that the fields are the following:
Field 1: order 2 polynomial Field 2: order 3 polynomial
Field 3: order 4 polynomial Field 4: order 5 polynomial
Field 5: order 6 polynomial
377 4000
sans prise en compte de l’inertie rotationnelle et de la flexion dans le corps sans prise en compte de l’inertie rotationnelle et de la flexion dans le corps
Avec flexions corps Avec flexions corps

Avec effet d'inertie rotationnelles
375
Avec effet d'inertie rotationnelles
avec prise en compte de l’inertie rotationnelle et flexion dans le corps 3500
avec prise en compte de l’inertie rotationnelle et flexion dans le corps
373
3000
371
2500
369
2000
367
365
1500
Champs 2 Champs 3 Champs 4 Champs 5
Champs 2 Champs 3 Champs 4 Champs 5
Fileds 1 Filds 2 Filds 3 Filds 4 Fileds 1 Filds 2 Filds 3 Filds 4
Fig. 2 First natural frequency depending Fig. 3 Second natural frequency depend-
of several parameters ing of several parameters
The first pulse from the first displacement fields (order 2 polynomial) with only
one variable is widely superior to other values which tend to the same region limit
370 rad/s if we consider both the effect of of rotational inertia and the bending
body. The value of the 2nd natural frequency tends to a limit close to 2300 rad/s
without the effect of the rotational inertia and converges to 2200 rad / s from the
third field (order 4 polynomial). The value of the third natural frequency con-
verges only from the fourth field (order 5 polynomial).
∗ Filds 2 ------- Filds 3 + Filds 4

------ Fileds 1 ∗ Filds 2 ------- Filds 3
*Fileds 1 - - - Numerical
+ Filds 4 *Fileds 5 - - -
Fig. 4 first modal deformation comparison Fig. 5 second modal deformation compari-
between different fields son between different fields
------- Filds 3 + Filds 4 *Fileds 5 + Filds 4 *Fileds 5

- - - Numerical simulation - - - Numerical simulation
Fig. 6 Third modal deformation compari- Fig. 7 Fourth modal deformation compari-
son between different fields son between different fields
It can be concluded that the high order frequencies require more order to stabil-
ize and more the mode order rises more we need a higher order polynomial.
We notice that the first modal deformation obtained from the field (1) is not
close to other deformed because of the function that characterizes the displace-
ment field does not define the shear energy of the body (∂ 3W / ∂x 3 = 0) . Even the first
obtained natural frequency is clearly superior to the others results, therefore this
result will be eliminated from our comparative study.
When we increase the number of parameters that characterize the displacement
fields we notice that the natural frequencies and modal deformed curves converge
to a common solution.
Each time we increase the number of parameters that characterize the dis-
placement fields, the calculations become more and more difficult and sometimes
we can‘t solve the systems of the obtained equations where we find unusable
results (problems with the field (5)), that’ s why we will consider the results
obtained from the field (4) as a reference.
From figure 5, 6 and 7 we remark that the deformed reach their max and min in
the same x-axis and also get canceled at the same points.
4.1 Test Presentation

4.1.1 Used Material Presentation
The test rig, shown in Figure 8, consists of a massive steel structure for embedding
beams and insulation from external vibrations. The beams are excited by an in-
strumented hammer connected to the acquisition system. The vibrational response
of the beam is itself measured with a laser vibrometer.
Fig. 8 Test rig with vibrometer laser
Performing a test of static bending with considering the effect of shear force
(the thickness of the specimen is relatively large) allows us to determine the static
stiffness and EIz. In these calculations the specimen is assumed to be elastic
homogeneous isotropic and linear.
The last two assumptions are verified in the used loading area.
4.1.2 Experimental Device
The specimen which is fixed at one end and free on the other is suspended verti-
cally to avoid the effect of static deflection.
The excitation hammer (Figure 10) with rubber tip and incorporated force sen-
sor excites the bar horizontally with a transient load of small amplitude.
The bi- channels spectrum analyzer calculates the transfer function averaged
over several samples. The coherence function is checked at each test and is close
to one.
Fig. 9 Experimental device- Hammer Fig. 10 Experimental device- Hammer shock

shock sensor sensor
4.2 Test Results

During the tests we used a single excitation point (at x = a) is the free end and the
acceleration is measured in a sufficiently close to the fixed end (x = a / 6).
The results of the transfer functions in amplitude and phase are shown in the
following section.
The choice of the frequency band is imposed by the bandwidth of the accele-
rometer concerning the adhesive assembly.
Fig. 11 Transfer function magnitude and Fig. 12 Transfer function magnitude and
phase of the Al beam phase of the Nomex
5 Numerical Simulation
Fig. 13 First modal deformation of flexion in Fig. 14 First modal deformation of flexion
plan (o,x,y) in plan (o,x,z)
Fig. 15 second modal deformation of flexion Fig. 16 First modal deformation of torsion
in plan (o,x,y)
Fig. 17 Third modal deformation of flexion

Fig. 18 Fourth modal deformation of flex-
in plan (o,x,y)
ion in plan (o,x,z)
6 Comparative Analysis
Comparing results among them, we see a slight variation between the natural
frequencies and modal deformation determined digital simulation compared with
analytical and experimental results this variation may be due to the not taken into
account the effect of the glue between the layers has a significant effect known
vibration damping.
7 Conclusion
The shape of the displacement field has a great influence on the natural frequency,
the right choice is essential to reach a common solution. The second value of the
first and the third natural frequency converge respectively from a displacement
field of 3, 4 and 5 order polynomial.
For each form of the displacement field we notice the relationship between the
coefficients of the matrix which may be written as follows
For n order polynomial:
i = 1....(n − 1)
M ij = K ij − L ij Avec
j = 1....(n − 1)
Where Lij can be in a general form.
1 2
L ij = ρω 2 a ( 3 + i + j)
2 ( 3 + i + j)
K 1 j = 4 (1 + j) a j ⋅ C K 2 j = N 2 j + (12 ⋅ j) ⋅ a ( j+1) ⋅ C
With relations between the N ij coefficients.
We can suggest to increase the degree of polynomial that characterizes the dis-
placement field or to modify its shape in order to reach at a general relationship.
We can also choose a field of displacement according to the three direction
(x, y, z) wherein for better vibration behavior of our beam while taking account of
the torsion and bending in both horizontal and vertical plane.
References
[1] Mindlin, R.D.: Influence of rotatory inertia and shear on flexural motions of isotropic,
elastic plates. J. Appl. Mech., 1831–1838 (1951)
[2] Yarlagadda, S., Lesieutre, G.: Fiber contribution to modal damping of polymer matrix
composite panels. Journal of Spacecraft and Rockets 32(5), 825–831 (1995)
[3] Wang Deducing, C.M.: thick plate solutions from classical thin plate solutions. Struct.
Eng. Mech. 11(1), 89–104 (2001)
[4] Meunier, M., Shenoi, R.A.: Dynamic analysis of composite sandwich plates with
damping modeled using higher-order shear deformation theory. Composite Struc-
tures 54, 243–254 (2001)
[5] Soula, M., Nasri, R., Ghazel, M.A., Chevalier, Y.: The effects of kine-matic model ap-
proximations on natural frequencies and modal damping of laminated composite
plates. Journal of Sound and Vibration 297, 315–328 (2006)
[6] Maheri, M.R., Adams, R.D.: Steady-state flexural vibration damping of honeycomb
sandwich beams. Composites Science and Technology 52(3), 333–347 (1994)
[7] Banerjee Frequency, J.R.: equation and modal deformation formulae for composite
Timoshenko beams. Composite Structures 51, 381–388 (2001)
[8] Nilsson, E., Nilsson Prediction, A.C.: Measurement of some dynamic properties of
sandwich structures with honeycomb and foam cores. Journal of Sound and Vibra-
tion 251(3), 409–430 (2002)
[9] Blevins, R.D.: Formulas for Natural Frequency and Modal deformation. Krieger Pub-
lishing Company, Florida (2001)
Author Index
Abbes, Mohamed Slim 189 Ben Mohamed, Ahmed 219

Abdallah, Bouabidi 209 Bentaleb, Fayçal 161
Abdelali, Hanane Moulay 443 Ben Yahia, Wafa 327
Abdelli, Youssef 541 Benzerga, Djebbara 481
Abid, Mohamed Salah 91, 101, 111, 359 Berrah, S. 179
Abid, Said 531 Bolaers, Fabrice 69
Adim, Belkacem 317 Bouaziz, Mohamed Ali 391
Affi, Zouhaier 47 Boubakeur, Ahmed 269
Agred, Souhila 9 Boudjemai, Abdelmadjid 229
Ahmed Benyahia, Ali 1, 169 Boudjenane, Nasr-Eddine 27
Aid, Abdelkrim 433 Boukharouba, Taoufik 281
Ait-Sghir, Khalid 69 Boukortt, A. 179
Akrout, Mohsen 339, 409 Boulahia, Ramdane 281
Amara, Idriss 161 Bouraoui, Chokri 507
Amira, Bilel Ben 91 Bruyère, Jérome 369
Amouri, Ammar 141
Amrouche, Abdewahab 433 Capasso, Clemente 249
Aouici, Hamdi 419 Capelle, Julien 391
Ayadi, Omar 289, 327 Chaari, Fakher 459
Azari, Zitouni 81, 391 Chalghoum, Issa 339
Chargui, Selma 317
Bacetti, Abdelmoumen 229 Chaterbache, Omar 59
Baganna, Moez 219 Chehaibi, Kaouther 239
Becquerelle, Samuel 369 Chelbi, Tarek 111
Belhadri, Mansour 27 Chen, Mian 299
Belkadi, Abdelghani 1 Chenini, Idris 541
Bellagi, Ahmed 307 Chokri, Bouraoui 517
Ben Said, Lotfi 489 Choley, Jean-Yves 189
Bennacer, Hamza 179
Benamar, Rhali 443 Dalleli, Manel 81
Benarous, Abdallah 9 Dammak, Fakhreddine 469, 489, 497,
Bendouba, Mostefa 433 531
Benguediab, Mohamed 433 Daouadji, Tahar Hassaine 317
Benguesmia, Hani 269 Delille, Remi 469
554 Author Index
Del Rincon, Alfonso Fernandez 459 Larbi, Ahmed Ben Cheikh 453
Deü, Jean-François 349 Larbi, Walid 349
Dhifelaoui, Hafedh 453 Loukarfi, Larbi 9
Driss, Dorra 101
Driss, Slah 131 Maaloul, Makram 101
Driss, Zied 91, 101, 111, 131, 209, Mabrouki, Ibrahim 359
359 Mahfoudi, Chawki 141
Dron, Jean Paul 69 Makki, Mhalla Mohamed 517
Mars, Jamel 469
Elaoud, Sami 339, 409 Masmoudi, Faouzi 289, 327
Meddour, Ikhlas 419
Fakhfakh, Hassen 369
Mejri, Sabrine 307
Felfel, Houssem 289
Merzoug, Mustapha 69
Fudym, Olivier 307
Meziani, S. 19
Ghanmi, Samir 37 Miloudi, Abdelhamid 59, 69, 151
Gloaguen, Jean-Michel 281 Mlayeh, Olfa 101
Gmati, Nabil 401 Mohamed, Anis 401
Graa, Mortadha 47 Mouloud, Guemana 379
Guedri, Mohamed 37 Moussaoui, M. 19
Guidara, Mohamed Amine 391 Mrabet, Elyes 37
Guizani, Amir 189 Mrad, Charfeddine 239, 259
M’Ziou, Nassima 269
Haddar, Mohamed 189, 459
Hadj-Taïeb, Ezzeddine 81, 339, 391, 409 Nasr, Anouar 507
Hafaifa, Ahmed 379 Nasr, Aymen 259
Hafsi, Zahreddine 409 Nasri, Rachid 219, 239, 259, 541
Hajlaoui, Abdessalem 497 Neder, Mahmoud 151
Hammadi, L. 27 Nejlaoui, Mohamed 47
Hammadi, Moncef 189, 249
Hammami, Ahmed 459 Ohayon, Roger 349
Hammami, Omar 299 Ouali, Nourdine 1, 169
Hariri, Said 391
Harras, Bilal 443 Patalano, Stanislao 199, 249
Hocine, Rachida 229
Houidi, Ajmi 47 Rabahi, Abderezak 317
Rachid, Belhadef 379
Issam Ziane, Mohamed 179 Renaud, Ruixian 249
Ichchou, Mohamed 37 Romdhane, Lotfi 47
Rueda, Fernando Viadero 459
Jamel, Mohammed 531
Jarraya, Abdessalem 497
Saidi, Mohamed 121
Kaddeche, Slim 121 Salah, Abid Mohamed 209
Kaffel, Ahmed 111 Samet, Ahmed 359
Kammoun, Imen Kallel 131 Santamaria, Miguel Iglesias 459
Karmed, Djamel 9 Schmitt, Christian 81, 391
Khlifi, Kaouthar 453 Sghaier, Jalila 307
Skander, Achraf 121
Lanteri, Stéphane 401 Soriano, Thierry 189
Lanzotti, Antonio 199 Soula, Mohamed 37
Author Index 555
Velex, Philippe 369 Yallese, Mohamed Athmane 419

Veneri, Ottorino 249
Vitolo, Ferdinando 199 Zaafouri, Mehrez 531
Zaatri, Abdelouahab 141
Wali, Mondher 469, 489, 497, 531 Zafrane, Mohammed Amine 229
Wannes, Hassine 507 Znaidi, Amna 219

Multiphysics Modelling and Simulation For Systems Design and Monitoring

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Multiphysics Modelling and Simulation For Systems Design and Monitoring

Uploaded by

Copyright:

Available Formats

Applied Condition Monitoring

Mohamed Haddar · Mohamed Slim Abbes

More information about this series at http://www.springer.com/series/13418

ISSN 2363-698X ISSN 2363-6998 (electronic)

Springer Cham Heidelberg New York Dordrecht London

Printed on acid-free paper

Springer International Publishing AG Switzerland is part of Springer Science+Business Media

Multiphysics problems are usually encountered when the behaviour of a system is

December 2014 Mohamed Haddar (Sfax-Tunisia)

Influence of the Plies Orientation on the Rigidity of the Laminated

Early Detection of Gear Failure by Vibration Analysis . . . . . . . . . . . . . . . 69

First Principal Calculations of Optical Properties of InGaN2 Using in

Multidisciplinary Design Optimization in Small Satellite . . . . . . . . . . . . . . 229

A System Engineering Conception of Multi-objective Optimization

Fuzzy Modeling and Control of Centrifugal Compressor Used in Gas

Iterative Methods for Steady State Looped Network Analysis . . . . . . . . . . 409

The Extended Finite Element Method for Cracked Incompressible

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 553

Abdelghani Belkadi, Ali Ahmed Benyahia, and Nourdine Ouali

Laboratoire de Mécanique Avancée (LMA) – USTHB – Bab Ezzouar / Alger

Abstract. The overall properties of a laminated composite structure depend on de-

Keywords: Laminated composites, optimization, stiffness, plies orientation.

© Springer International Publishing Switzerland 2015 1

Where and k correspond to vectors of deformations and bends of the average

Where A is the membrane stiffness matrix, D flexural stiffness matrix and B

With the parameters of lamination:

Furthermore, for a design of a balanced and decoupled laminate of which the

4 Results and Discussions

Fig. 1 Influence de l’orientation des plis sur le module de rigidité longitudinal

This table includes, in addition to results concerning the laminate studied by

Table 1 Optimization results by algorithm proposed

[ ± 53.1224 ± 39.3220 ] 14.5268

08 [ ± 53.1239 ± 39.3366 ] 14.5216

[ ± 39.3249 ± 53.1152 ] 14.5262

[ ±53.9459 ±35.1962 ±45.9894 ±49.6245] 14.5225

16 [±35.4489 ±45.3720 ±54.2212 ±49.6980] 14.5276

[±58.3783 ±40.9708 ±42.2751 ± 43.5982] 14.5239

To check the performance of our algorithm, we will apply it to another laminate

Table 3 Optimization of the sequence of an 8 ply carbon/epoxy laminate

structures to find the maximum longitudinal modulus of rigidity (Ex ). Validation

Souhila Agred1, Abdallah Benarous2,*, Djamel Karmed3, and Larbi Loukarfi4

Keywords: smoke and heat evacuation, calculation methods, natural/mechanical

© Springer International Publishing Switzerland 2015 9

Although the calculation methods deal with analytical /empirical formulations,

2.1 Input Variables and Calculation Assumptions

Fig. 1 The main graphical user interface (GUI)

2.2 The European Method

Q c = (1 − α R )Q f = 0.8Q f (1)

This value is retained if the regularity condition Ys ≤ A f is fulfilled, so that

For large compartments, the empirical constant C e is set to 0.19.

Fig. 2 Graphical user interface for smoke calculation

Vs = ( M s .Ts ) /( ρ ∞ T∞ ) (5)

In the case of a mechanical ventilation, a set of N MV ventilators /jet fans are to

Fig. 3 SHEVS calculation GUI (mechanical ventilation case)

2.3 The American Method

Q c = (1 − α R )Q f ; α R = 0.2 or 0.5 (10)

For a mechanical ventilation case, a critical volume flow rate is imposed to

Fig. 4 Graphical user interface for SHEVS calculation

3 Results and Discussions

Fig. 5 Hot layer temperature (natural ventilation)

Figure 6 shows a relative deviation of 27% on the Ys values obtained by the

of N MV = 11 ventilators. This latter value seems more realistic since ventilator

Fig. 7 Hot layer temperature (mechanical ventilation)

Q c = (1 − α R )Q f = 0.8Q f (1)

Vs = ( M s .Ts ) /( ρ ∞ T∞ ) (5)

Q c = (1 − α R )Q f ; α R = 0.2 or 0.5 (10)

My(t ) + Cy (t ) + Ky (t ) = − Mryb (1)

Introducing the state space vector Z 0 = ( yT , y s , yT , y s )T , Eq.(1) can be ex-

y f + 2ξ f ω f y f + ω 2f y f = − w(t ) and yb = y f + w(t ) = − (2ξ f ω f y f + ω 2f y f ) (2)

written in the space state form (Yu et al. 2013) as Z = AZ + f where

e(t ) = c s y s2 (t ) = 2ξ s ω s ms y s2 (t ) and e(t ) = de(t ) dt = 4ξ s ω s ms y s (t ) ys (t ) (4)