You are on page 1of 266

Advances in Mechanical Engineering

New Developments in
Multibody System Dynamics
and Its Applications
Guest Editors: Xiaoting Rui, Chuanzeng Zhang, and Caishan Liu
New Developments in Multibody System
Dynamics and Its Applications
Advances in Mechanical Engineering

New Developments in Multibody System


Dynamics and Its Applications

Guest Editors: Xiaoting Rui, Chuanzeng Zhang,


and Caishan Liu
Copyright © 2014 Hindawi Publishing Corporation. All rights reserved.

This is a special issue published in “Advances in Mechanical Engineering.” All articles are open access articles distributed under the
Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the
original work is properly cited.
Editorial Board
Koshi Adachi, Japan Luı́s Godinho, Portugal David R. Salgado, Spain
Mehdi Ahmadian, USA Tian Han, China Mohammad R. Salimpour, Iran
Rehan Ahmed, UK Francisco J. Huera-Huarte, Spain Sunetra Sarkar, India
Muhammad T. Akhtar, Japan Davood Jalali-Vahid, Iran Pietro Scandura, Italy
Nacim Alilat, France Jiin Y. Jang, Taiwan A. S. Sekhar, India
M. Affan Badar, USA Zhongmin Jin, UK Liyuan Sheng, China
Luis Baeza, Spain Xiaodong Jing, China Xi Shi, China
R. Balachandran, UK S.-W. Kang, Republic of Korea Seiichi Shiga, Japan
Claude Bathias, France Xianwen Kong, UK Chow-Shing Shin, Taiwan
Adib Becker, UK Michal Kuciej, Poland Ray W. Snidle, UK
Leonardo Bertini, Italy Yaguo Lei, China Margaret M. Stack, UK
Liam A. Blunt, UK Zili Li, The Netherlands Neil Stephen, UK
Noël Brunetière, France Yangmin Li, Macau Kumar K. Tamma, USA
Marco Ceccarelli, Italy Jaw-Ren Lin, Taiwan Yaya Tan, China
Fakher Chaari, Tunisia Cheng-Xian Lin, USA Anand Thite, UK
Chin-Lung Chen, Taiwan Seyed N. Mahmoodi, USA Cho W. Solomon To, USA
Hyung H. Cho, Republic of Korea Oronzio Manca, Italy Yoshihiro Tomita, Japan
Seung-Bok Choi, Korea Ramiro Martins, Portugal Shan-Tung Tu, China
Ahmet S. Dalkilic, Turkey Aristide F. Massardo, Italy Sandra Velarde-Suárez, Spain
Kangyao Deng, China Francesco Massi, Italy Moran Wang, China
Francisco D. Denia, Spain T. H. New, Singapore Junwu Wang, China
T. S. Dhanasekaran, USA Kim C. Ng, Singapore Jia-Jang Wu, Taiwan
Nihad Dukhan, USA Cong T. Nguyen, Canada Fengfeng Xi, Canada
Farzad Ebrahimi, Iran Hirosi Noguchi, Japan Gongnan Xie, China
Ali El Wahed, UK Hakan F. Oztop, Turkey Wei Mon Yan, Taiwan
Bogdan I. Epureanu, USA Duc T. Pham, UK Jianqiao Ye, UK
Mohammad R. Eslami, Iran Jurij Prezelj, Slovenia Byeng D. Youn, USA
Ali Fatemi, USA Xiaotun Qiu, USA Bo Yu, China
Mario L. Ferrari, Italy Pascal Ray, France Jianbo Yu, China
Siegfried Fouvry, France Robert L. Reuben, UK Zhongrong Zhou, China
Ian Frigaard, Canada Pedro A. R. Rosa, Portugal
Mergen H. Ghayesh, Canada Elsa de Sá Caetano, Portugal
Contents
New Developments in Multibody System Dynamics and Its Applications, Xiaoting Rui,
Chuanzeng Zhang, and Caishan Liu
Volume 2014, Article ID 671604, 1 page

Automatic Deduction Theorem of Overall Transfer Equation of Multibody System, Xiaoting Rui,
Jianshu Zhang, and Qinbo Zhou
Volume 2014, Article ID 378047, 12 pages

Optimal Design of One-Folded Leaf Spring with High Fatigue Life Applied to Horizontally Vibrating
Linear Actuator in Smart Phone, Ki Bum Lee, Chang Hyun Park, and Jin Ho Kim
Volume 2014, Article ID 545126, 7 pages

Adaptive Terminal Sliding Mode Control of Electromagnetic Spacecraft Formation Flying in


Near-Earth Orbits, Jingrui Zhang, Changqin Yuan, Dongmei Jiang, and Dawei Jin
Volume 2014, Article ID 512583, 9 pages

Application of Krylov Reduction Technique for a Machine Tool Multibody Modelling, M. Sulitka,
J. Šindler, J. Sušeň, and J. Smolı́k
Volume 2014, Article ID 592628, 9 pages

Study on Launch Dynamics of Self-Propelled Artillery Based on Transfer Matrix Method of Multibody
System, Hailong Yu and Xiaoting Rui
Volume 2014, Article ID 308049, 9 pages

Higher Order Variational Integrators for Multibody System Dynamics with Constraints, Jieyu Ding and
Zhenkuan Pan
Volume 2014, Article ID 383680, 8 pages

Free Vibration Characteristic of Multilevel Beam Based on Transfer Matrix Method of Linear
Multibody Systems, Laith K. Abbas and Xiaoting Rui
Volume 2014, Article ID 792478, 16 pages

Case Study on Human Walking during Wearing a Powered Prosthetic Device: Effectiveness of the System
“Human-Robot”, Svetlana Grosu, Pierre Cherelle, Chris Verheul, Bram Vanderborght, and Dirk Lefeber
Volume 2014, Article ID 365265, 9 pages

Controller Parameters Tuning Based on Transfer Matrix Method for Multibody Systems,
Hossam Hendy, Xiaoting Rui, Qinbo Zhou, and Mostafa Khalil
Volume 2014, Article ID 957684, 8 pages

All Terrain Vehicle Flexible Multibody Dynamic Simulation for Fatigue Prediction, Jia-Shiun Chen and
Hsiu-Ying Hwang
Volume 2013, Article ID 725315, 6 pages

Projectile Impact Point Prediction Based on Self-Propelled Artillery Dynamics and Doppler Radar
Measurements, Mostafa Khalil, Xiaoting Rui, Qicheng Zha, Hailong Yu, and Hossam Hendy
Volume 2013, Article ID 153913, 12 pages
Study on the Dynamics of Laser Gyro Strapdown Inertial Measurement Unit System Based on Transfer
Matrix Method for Multibody System, Gangli Chen, Xiaoting Rui, Fufeng Yang, Jianshu Zhang,
and Qinbo Zhou
Volume 2013, Article ID 854583, 9 pages

Recursive Formulation for Dynamic Modeling and Simulation of Multilink Spatial Flexible Robotic
Manipulators, Zhenjie Qian, Dingguo Zhang, and Jun Liu
Volume 2013, Article ID 216014, 15 pages

Trivariate Isogeometric Analysis for Flexible Multibody Dynamics, Ting Pi


Volume 2013, Article ID 139498, 8 pages

Dynamic Control and Disturbance Estimation of 3D Path Following for the Observation Class
Underwater Remotely Operated Vehicle, Hai Huang, Qirong Tang, Yueming Li, Lei Wan, and Yongjie Pang
Volume 2013, Article ID 604393, 16 pages

Application of Finite Volume Method to Structural Stochastic Dynamics, Weidong Chen, Yanchun Yu,
Ping Jia, Xiande Wu, and Fengchao Zhang
Volume 2013, Article ID 391704, 10 pages

Numerical Simulation on Dense Packing of Granular Materials by Container Oscillation, Jun Liu and
Dongxu You
Volume 2013, Article ID 284693, 7 pages

Research on Flexible Joint Stiffness Test and Trajectory Planning of Space Manipulator, Yongtai Yang,
Jili Rong, Jian Li, and Ling Tang
Volume 2013, Article ID 280453, 7 pages

Case Study on Influence of Step Blast-Excavation on Support Systems of Existing Service Tunnel with
Small Interval, Shaorui Sun, Ling Yue, Jimin Wu, Jin Liu, and Jihong Wei
Volume 2013, Article ID 257457, 13 pages

The Effect of First-Order Bending Resonance of Wheelset at High Speed on Wheel-Rail Contact
Behavior, Shuoqiao Zhong, Xinbiao Xiao, Zefeng Wen, and Xuesong Jin
Volume 2013, Article ID 296106, 19 pages

Dynamic Modeling and Analysis of Hub-Tapered Beam System, Si-jia Chen and Ding-guo Zhang
Volume 2013, Article ID 313279, 12 pages

Analysis of Elastic-Plastic Responses of a New Single-Story Frame-Bent Structure during an Earthquake


Based on the Transfer Matrix Method of Multibody System, Jianguo Ding, Yuwei Dai, and Zhi Qiao
Volume 2013, Article ID 784525, 10 pages

Cooperative Search by Combining Simulated and Real Robots in a Swarm under the View of Multibody
System Dynamics, Qirong Tang and Peter Eberhard
Volume 2013, Article ID 284782, 11 pages
Global Impact Dynamic Modeling and Verification of a Flexible Beam with Large Overall Motion,
Yuechen Duan, Dingguo Zhang, and Jiazhen Hong
Volume 2013, Article ID 362317, 8 pages

Evaluating Energy Flux in Vibrofluidized Granular Bed, N. A. Sheikh, S. Manzoor, and U. Ghani
Volume 2013, Article ID 327379, 7 pages
Hindawi Publishing Corporation
Advances in Mechanical Engineering
Volume 2014, Article ID 671604, 1 page
http://dx.doi.org/10.1155/2014/671604

Editorial
New Developments in Multibody System Dynamics and
Its Applications

Xiaoting Rui,1 Chuanzeng Zhang,2 and Caishan Liu3


1
Institute of Launch Dynamics, Nanjing University of Science and Technology, Nanjing 210094, China
2
Department of Civil Engineering, University of Siegen, 57068 Siegen, Germany
3
College of Engineering, Peking University, Beijing 100871, China

Correspondence should be addressed to Xiaoting Rui; ruixt@163.net

Received 6 January 2014; Accepted 6 January 2014; Published 12 March 2014

Copyright © 2014 Xiaoting Rui et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Multibody system dynamics is nowadays a widely used tool- and precision, the model and modal reduction techniques
box in the development of modern engineering science and for multibody system dynamics, the optimization and control
technology. It provides powerful dynamic design theories, of a swarm of robots, the optimal design of smart systems,
computational and analytical methods, and test technology the dynamics and control of underwater vehicle and space-
for innovative engineering applications. In order to sys- craft formation, the perturbation-finite volume method for
tematically show the important effects of multibody system stochastic multibody system dynamics, the blast-excavation
dynamics on the development of modern science and tech- dynamics, the granular system dynamics, the vehicle-track
nology and to strengthen the academic exchanges in the coupling dynamics, and the human-robot dynamics. Thus,
research field of multibody system dynamics and its applica- the general fundamental theories and several key engineering
tions, we organized this special issue containing twenty-five applications are covered representing the state of the art of
papers selected from the contributions coming from Belgium, recent intensive and extensive research activities in the field
Czech Republic, Egypt, Germany, Iraq, Netherlands, Pak- of multibody system dynamics to a certain extent.
istan, China, Republic of Korea, United Kingdom, Taiwan, The guest editors hope that this special issue will pro-
and so forth. vide some valuable information and guidance for scientists
The twenty-five papers selected for this Special Issue are working on multibody system dynamics and its applications
devoted to the following topics: an automatic deduction the- in modern engineering science and technology.
orem of the overall transfer equations of the transfer matrix
method for multibody systems (MSTMM), the application
of MSTMM in dynamics of a multilevel beam system, the Acknowledgments
application of MSTMM in dynamics of an elastic-plastic Many thanks are due to Professor Edwin Kreuzer from
system subjected to earthquakes, the application of MSTMM Hamburg University of Technology and Professor Dieter
in dynamics of controlled cars, the application of MSTMM Bestle from Brandenburg University of Technology Cottbus-
in dynamics of a laser gyro strapdown inertial measurement Senftenberg for their great unselfish help in organizing the
unit system, the application of MSTMM in launch dynamics special issue, to reviewers and authors, and to the reviewers
of a self-propelled artillery system and flight dynamics of for their voluntary works and constructive comments. All
projectile, the modeling of trivariate isogeometrics for flexible of these promoted greatly the projected publication of this
multibody dynamics, the modeling of hub-tapered beam special issue.
system dynamics, the formulations and modeling approaches
of spatial flexible robotic manipulators, the impact in flexible Xiaoting Rui
systems, the modeling and test methods for flexible joint Chuanzeng Zhang
stiffness of a space manipulator, the computational stability Caishan Liu
Hindawi Publishing Corporation
Advances in Mechanical Engineering
Volume 2014, Article ID 378047, 12 pages
http://dx.doi.org/10.1155/2014/378047

Research Article
Automatic Deduction Theorem of Overall Transfer
Equation of Multibody System

Xiaoting Rui, Jianshu Zhang, and Qinbo Zhou


Institute of Launch Dynamics, Nanjing University of Science and Technology, Nanjing 210094, China

Correspondence should be addressed to Xiaoting Rui; ruixt@163.net

Received 25 August 2013; Accepted 30 December 2013; Published 10 March 2014

Academic Editor: Caishan Liu

Copyright © 2014 Xiaoting Rui et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Transfer matrix method for multibody System (MSTMM) is a new multibody dynamics method developed in recent 20 years. It
has been widely used in both science research and engineering for its special features as follows: without global dynamics equations
of the system, high programming, low order of system matrix, and high computational speed. Based on MSTMM and its above
features, a theorem to deduce automatically the overall transfer equations of multibody systems by handwriting or by computer
is proposed in this paper. The theorem is effective for multibody systems with various topological structures, including chain
systems, closed-loop systems, tree systems, general systems composed of one tree subsystem, and some closed-loop subsystems.
This theorem makes it possible to program large scale software of multibody system dynamics with much higher programming,
and much higher computational speed because of the above features of MSTMM. Formulations of the proposed method as well as
two examples are given to verify this method.

1. Introduction this method as follows: study MSD without global dynamics


equations of the system, keep low order of the system
Lots of methods dealing with multibody system dynamics matrix so very high computational speed, and avoid the
(MSD) have been studied by many authors since 1960s [1– difficulties in computation caused by high-order matrices and
18]. They are widely used in many engineering fields such as high programming. It has been proved by lots of theories
aeronautics, astronautics, spacecraft, vehicle, robot, precision and experiments that MSTMM is effective for linear time-
machinery, and biomechanics. It is well known that almost invariant multibody systems [22], nonlinear time-variant
all the previous ordinary methods for MSD have the same multibody systems, multi-rigid-body systems [19, 21], multi-
characteristics as follows: it is necessary and very complicated rigid-flexible-body systems [22–24], and controlled multi-
to develop the global dynamics equations of the system; the body systems [19, 29].
order of system matrix depends on the number of degrees of Generally speaking, various multibody systems may be
freedom of the system and hence it is rather high for complex considered as one of the following four cases in topology [24]:
multibody system. (1) chain system, (2) closed-loop system, (3) tree system, (4)
To avoid establishing the global dynamics equations of general systems composed of one tree subsystem, and some
the system, simplify the study procedure and especially keep closed-loop subsystems. A chain system can be considered
high computational efficiency independent of the number as a special example of a tree system at the case with only
of degree of freedom of system in studying MSD, new two boundary ends. By “cutting” at one connection point
analytical method for MSD, namely, transfer matrix method of a system, a closed-loop system can be considered as
for multibody system (MSTMM), is presented by Rui and a chain system, and a general system composed of one
his co-workers [19–21] and constantly developed in recent tree subsystem and some closed-loop subsystems can be
20 years [22–25]. Nowadays, MSTMM is widely applied in dealt with a tree system [24]. Based on MSTMM and its
science research and engineering [26–29] for the features of above features, a theorem to deduce automatically the overall
2 Advances in Mechanical Engineering

transfer equations of various multibody systems mentioned time derivatives of the generalized coordinates describing
above by handwriting and by computer is proposed in this the deformation of flexible bodies using modal method; the
paper. superscript 𝑛 is the highest order of the modal considered, r,̈
𝜃,̈m, and q are the column matrices of accelerations, second
2. General Theorems and Steps of MSTMM time derivatives of the space-three angles, internal torques,
and internal forces in 𝑜𝑥𝑦𝑧, respectively.
2.1. Basic Idea of MSTMM. The basic idea of MSTMM [19] is For body and hinge elements moving in a plane, a similar
to break up a multibody system into the elements contain- definition of the state vector can be introduced, which is a
ing bodies (including rigid bodies, flexible bodies, lumped special example of spatial motion.
masses, etc.) and hinges (including joints, ball-and-sockets, The transfer equations of the 𝑗th element can be obtained
pins, springs, rotary springs, dampers and rotary dampers, easily by rewriting its dynamics equations as follows [19, 21,
etc.) whose dynamics properties can be readily expressed 30]
in matrix forms. These matrices of elements are considered
as building blocks that provide the dynamics properties of z𝑗,𝑗+1 = U𝑗 (𝑡𝑖 ) z𝑗,𝑗−1 , (5)
the entire system when assembling them together according
to the topology of the system. Particularly, the positions where U𝑗 (𝑡𝑖 ) is the transfer matrix of the 𝑗th element, which
of bodies and hinges are considered equivalent in transfer expresses the relationship between the state vectors of its
equations and transfer matrices, which is totally different output end z𝑗,𝑗+1 and input end z𝑗,𝑗−1 and is already known
from ordinary methods for MSD [1–18] and results in the at time instant 𝑡𝑖 .
very low order of system matrix and very high computational It should be pointed out that there are general linear
speed in MSTMM. relations among accelerations, angular accelerations, forces,
and torques of any mechanics system in an inertial coordinate
2.2. State Vector, Transfer Equation, and Transfer Matrix of system 𝑜𝑥𝑦𝑧, according to Newton motion law and Euler
Element. The same coordinate systems and sign conventions theorem of moment of momentum. It is to say that there are
as are described in [19, 21, 30] will be used. The state vector strict linear relations between the state vectors of output end
of the connection point between any rigid body and hinge and input end of any element and among all state vectors
moving in space is defined as of a multibody system. Thus, the transfer equation (5) is a
general equation and effective for any mechanics element in
𝑇 the inertial coordinate system.
z = [𝑥̈𝑦̈𝑧̈𝜃𝑥̈ 𝜃𝑦̈ 𝜃𝑧̈ 𝑚𝑥 𝑚𝑦 𝑚𝑧 𝑞𝑥 𝑞𝑦 𝑞𝑧 1] , (1)

or 2.3. Overall Transfer Equation and Overall Transfer Matrix of


𝑇 the System. The overall transfer equations of a chain system
𝑇
z = [r𝑇̈ 𝜃̈ m𝑇 q𝑇 1] , (2) can be deduced automatically as [19, 24]

where z1,0 = U𝑛−1 z𝑛,0 , (6)

𝑥̈ 𝜃𝑥̈ where the overall transfer matrix of the system is


[ ]
r̈= [𝑦]̈, 𝜃̈= [𝜃𝑦̈] ,
𝑧̈ ̈ U𝑛−1 = U1 U2 U3 ⋅ ⋅ ⋅ U𝑛 .
[ ] [𝜃𝑧 ] (3)
(7)

𝑚𝑥 𝑞𝑥 From equations (6) and (7), the features of the overall


m = [𝑚𝑦 ] , q = [𝑞𝑦 ] . transfer equation for a chain system can be clearly seen;
[𝑚𝑧 ] [𝑞𝑧 ] overall transfer matrix of a chain system can be deduced auto-
matically by successive premultiplication of transfer matrices
The state vector of connection point between flexible of every element of the system along the transfer path from
body and hinge moving in space is defined as the one end to another end.

z = [𝑥,̈𝑦,̈𝑧,̈𝜃𝑥̈, 𝜃𝑦̈, 𝜃𝑧̈, 𝑚𝑥 , 𝑚𝑦 , 𝑚𝑧 ,


2.4. Solutions of the System Motion. Applying the boundary
(4) conditions of the system, z1,0 and z𝑛,0 , to the overall system
𝑇
𝑞𝑥 , 𝑞𝑦 , 𝑞𝑧 , 𝑞1̈, 𝑞2̈, . . . , 𝑞𝑛̈, 1] , transfer equation (6), the unknown state variables in the
boundary state vectors can be computed. Then, the state
where 𝑥,̈𝑦,̈and 𝑧̈are the accelerations of the connection point vectors of each element at time 𝑡𝑖 can be computed by the
with respect to the global inertial coordinate system 𝑜𝑥𝑦𝑧; 𝜃𝑥̈, repeated use of corresponding transfer equations of element
𝜃𝑦̈, and 𝜃𝑧̈are the second time derivatives of the orientation (5). Then, using the computed values of the last step as the
angles of the body using space-three-angles 1-2-3 𝜃𝑥 , 𝜃𝑦 , and initial conditions, the entire procedure can be repeated from
𝜃𝑧 defined in [18]; 𝑚𝑥 , 𝑚𝑦 , 𝑚𝑧 , 𝑞𝑥 , 𝑞𝑦 , and 𝑞𝑧 are the cor- time 𝑡𝑖+1 until the time required for complete analysis. It can
responding internal torques and internal forces in the same be seen clearly from equations (5)–(7) that MSTMM belongs
reference system, respectively; 𝑞1̈, 𝑞2̈, . . . , 𝑞𝑛̈are the second to one of strict analytical methods in principle.
Advances in Mechanical Engineering 3

3. Topology Figure of a Multibody System and 0 𝒛15,0 0 𝒛16,0 0


𝒛17,0 0 𝒛18,0
Sign Conventions 15 16 17 18
3.1. Topology Figure of a Multibody System. A dynamics
model of any complex multibody system can be constructed 11 12 13 14
with dynamics elements including bodies and hinges. In
order to describe the transfer relationship among the state 9 10
vectors of elements and the transfer directions in a system, the
topology figure of the model will be very useful for deduction 8
of overall transfer equation in MSTMM. And the topology 7 6
figure of the dynamics model of a system, for example, a tree 5
multibody system, as shown in Figure 1, can be got very easily 4
and directly from its dynamics model if using the following
sign conventions. 3
2
3.2. The Sign Conventions. Besides the sign conventions
1
introduced in [19, 21], the sign conventions as follows are used Root
in the paper. 0
𝒛1,0

(1) A circle I denotes a body element and the number Figure 1: Topology figure of a dynamics model of tree multibody
inside this circle is the sequence number of the body system.
element.

(2) An arrow → denotes a hinge element and the transfer 4. Automatic Deduction of Overall Transfer
direction of state vectors; the number beside the
arrow is the sequence number of the hinge element. Equations of System
4.1. Automatic Deduction of the Overall Transfer Equation of
(3) Each body element is dealt with single output end and
a Chain System. The topology figure of any chain system is
single input end if the body has two connection ends
shown in Figure 2.
with other elements; otherwise, it is dealt with single
It is clear that we can rewrite the overall transfer equation
output end and multiple input ends if the body has
(6) of the chain system as
more two connection ends.
𝑇
(4) For a nonboundary end, the first and second sub- Uall [z𝑇1,0 z𝑇𝑛,0 ] = 0, Uall = [−I U𝑛−1 ] . (8)
scripts, 𝑖 and 𝑗 (𝑖, 𝑗 ≠0), in a state vector z𝑖,𝑗 of the
end denote the sequence numbers of the adjacent From equations (6) or (7), it can be seen clearly that
body element and hinge element, respectively. For the overall transfer matrix of any chain system Uall can be
a boundary end, the second subscript 𝑗 = 0 in deduced automatically by handwriting and by computer. In
the state vector z𝑖,𝑗 ; that is, the second subscript 0 the overall transfer matrix, the coefficient matrix of the state
means boundary end; then the first subscript 𝑖 in the vector of root is a minus unit matrix I, while the coefficient
state vector z𝑖,0 of the boundary end stands for the matrix of the state vector of tip U𝑛−1 is the successive
sequence number of the element involved. premultiplication of the transfer matrices of all elements in
the transfer path from this tip to the root as shown in equation
(5) In a multibody system, only one boundary end is
(7).
considered as the root; the state vector of root is noted
The highest order of the overall transfer matrix is 13 for
as z𝑖,0 , where 𝑖 is the sequence number of the root
spatial chain multi-rigid-body system or (13 + 𝑛) for chain
element; all of other boundary ends are considered as
multi-rigid-flexible-body system, where 𝑛 is the highest order
the tips; the state vectors of tips are denoted as z𝑗,0 ,
of the modal considered.
where 𝑗 is the sequence number of the tip element.
The transfer directions of a system are always from its
tips to the root. 4.2. Automatic Deduction of the Overall Transfer Equations
of a Closed-Loop System. For any closed-loop system, whose
(6) The subscript 𝑖 in transfer matrix U𝑖 denotes the topology figure is shown in Figure 3, after “cutting” at the
sequence number of element 𝑖. The subscript (𝑖 − 𝑘) junction of any two adjacent elements such as body 1 and
in the transfer matrix U𝑖−𝑘 and the partitioned matrix hinge 𝑛 as shown in Figure 4, consider the couple of “cutting
u𝑖−𝑘 means from element 𝑖 to element 𝑘. U𝑖−𝑘 and u𝑖−𝑘 point” as the “boundary ends” with the same state vectors
mean the successive premultiplication of the transfer noted as z𝑛,0 and z1,0 . Then the original closed-loop system
matrices of all elements in the transfer path from the becomes a chain system with the same “boundary ends” state
element 𝑖 to element 𝑘 of the system. vectors, z𝑛,0 and z1,0 , as shown in Figure 4.
4 Advances in Mechanical Engineering

0 n − 1 ··· 7 5 3 1 0 Root from the state vector when it premultiplies z𝑗,𝐼𝑘 , acted on the
𝒛n,0 n 6 4 2 same point.
𝒛1,0
However, in equation(13), the number of unknown vari-
Figure 2: Topology figure of a chain system. ables is more than that of algebraic equations. Therefore,
geometrical equations of the body, which describes the
geometrical relationship between the first input end and
kth (𝑘 = 2, 3, 4, . . . , 𝐿) input end of the body, should be
7 ··· n−1 n 1 introduced for body elements with single output end and
multiple input ends. It is verified later that the geometrical
6 equation can be written in the form of
2

3 H𝑗 z𝑗,𝐼1 = H𝑗,𝐼𝑘 z𝑗,𝐼𝑘 (𝑘 = 2, 3, 4, . . . , 𝐿) , (14)


5
4
where H𝑗 is a constant matrix extracting displacement vari-
Figure 3: Topology figure of a closed-loop system. ables (including position coordinates and orientation angles)
from a state vector when premultiplying z𝑗,𝐼1 ; H𝑗,𝐼𝑘 is related
to the relative position between the first input end and kth
Thus, the transfer equation of the closed-loop system can input end of body. The number of the geometrical equations
be deduced automatically as is achieved for the chain system is (𝐿 − 1).
A spatial motion rigid body with more than two ends is
z1,0 = U𝑛−1 z𝑛,0 (9) shown in Figure 5; the concrete form of the transfer equations
and geometrical equations of body element will be exhibited.
according to the proposed sign conventions, where The state vectors of inboard ends and outboard ends of
the rigid body, as defined in Section 2.2, are
U𝑛−1 = U1 U2 U3 ⋅ ⋅ ⋅ U𝑛 . (10)
𝑇
Attention should be paid that the state vectors of a couple z𝐼𝑘 = [𝑥,̈𝑦,̈𝑧,̈𝜃𝑥̈, 𝜃𝑦̈, 𝜃𝑧̈, 𝑚𝑥 , 𝑚𝑦 , 𝑚𝑧 , 𝑞𝑥 , 𝑞𝑦 , 𝑞𝑧 , 1]𝐼
𝑘
of “cutting points” are the same, namely
(𝑘 = 1, 2, . . . , 𝐿) (15)
z𝑛,0 = z1,0 . (11)
𝑇
z𝑂 = [𝑥,̈𝑦,̈𝑧,̈𝜃𝑥̈, 𝜃𝑦̈, 𝜃𝑧̈, 𝑚𝑥 , 𝑚𝑦 , 𝑚𝑧 , 𝑞𝑥 , 𝑞𝑦 , 𝑞𝑧 , 1]𝑂.
Then, the transfer equation (9) of the closed-loop system
can be deduced automatically by handwriting and by com- As shown in Figure 5, 𝐼𝑘 (𝑘 = 1, 2, . . . , 𝐿), 𝑂, and 𝐶 denote
puter the inboard ends, outboard end, and mass center of the rigid
body, respectively; the coordinate system with subscript 2
Uall z1,0 = 0, Uall = (I − U𝑛−1 ) . (12) denotes the body-fixed coordinate system whose initial point
𝑜2 is on the first inboard end 𝐼1 of the rigid body. According
4.3. Automatic Deduction of the Overall Transfer Equations to the properties of a rigid body, the geometrical relationship
of a Tree System. For each body element with more than between the first inboard end and the outboard end of the
two ends, as is mentioned in the convention, only one of rigid body can be obtained as
the ends is considered as output end and all the other ends
are input ends. Moreover, its transfer equations should cover 𝜃𝑂 = 𝜃𝐼1 ,
the geometrical relationship between its first input end and (16)
output end and describe the mechanical principle for the r𝑂 = r𝐼1 + Al𝐼𝐼 𝑂,
forces and moments acting on the element. Thus, it can be
verified later that the transfer equations of a rigid body 𝑗 with where 𝜃𝐼1 and 𝜃𝑂 are the column matrices of the orientation
𝐿 input ends can be written in the following form coordinates of the first inboard end 𝐼1 and the outboard end
𝑂; r𝑂 and l𝐼𝐼 𝑂 are the column matrices of the position coordi-
z𝑗,𝑂 = U𝑗 z𝑗,𝐼1 + U𝑗,𝐼2 z𝑗,𝐼2 + ⋅ ⋅ ⋅ + U𝑗,𝐼𝐿 z𝑗,𝐼𝐿 , (13) nates of point 𝑂 with respect to the global inertial coordinate
system 𝑜𝑥𝑦𝑧 and the body-fixed coordinate system 𝑜2 𝑥2 𝑦2 𝑧2
where the subscript 𝑗 is the sequence number of the body; respectively; A is the direction cosine matrix that defines the
𝑂 and 𝐼1 , 𝐼2 , 𝐼3 ⋅ ⋅ ⋅ 𝐼𝐿 denote the output end and input ends orientation of the body-fixed coordinate system 𝑜2 𝑥2 𝑦2 𝑧2 in
respectively, and the first input end 𝐼1 is considered as the the global inertial coordinate system 𝑜𝑥𝑦𝑧.
dominant input end; z𝑗,𝑂 and z𝑗,𝐼𝑘 (𝑘 = 2, 3, 4, . . . , 𝐿) are Similarly, the geometrical equations between the first
the state vectors of the output end and the kth input end of inboard end 𝐼1 and the other inboard ends 𝐼𝑘 (𝑘 = 1, 2, . . . , 𝐿)
the body element respectively; U𝑗 is just the transfer matrix can be obtained easily
of element 𝑗 when point 𝐼1 and point 𝑂 are considered as
the only input end and output end of this element; U𝑗,𝐼𝑘 is 𝜃𝐼𝑘 = 𝜃𝐼1 (𝑘 = 2, 3, . . . , 𝐿)
the corresponding extraction matrix that extracts the force (17)
variables (including internal force and internal moment) r𝐼𝑘 = r𝐼1 + Al𝐼𝐼 𝐼𝑘 (𝑘 = 2, 3, . . . , 𝐿) .
Advances in Mechanical Engineering 5

𝒛n,0 𝒛1,0 By equations (16), (18), and rewriting them in the form of
7 ··· n−1 0 0 1
equation (13), transfer matrices can be obtained as
n
I3 E1 O3×3 O3×3 E2
6
2 [ O3×3 I3 O3×3 O3×3 O3×1 ]
[ ]
U𝑗 = [
[E6 E3 + E7 E6 E4 + E8 I3 E6 E6 E5 + E9 ]
],
5 3 [ E3 E4 O3×3 I3 E5 ]
4
[ O1×3 O1×3 O1×3 O1×3 1 ]
Figure 4: Topology figure of a closed-loop system after “cutting” the (19)
hinge 𝑛.
O3×3 O3×3 O3×3 O3×3 O3×1
[O3×3 O3×3 O3×3 O3×3 O3×1 ]
[ ]
U𝑗,𝐼𝑘 =[
[O3×3 O3×3 I3 ̃r𝐼1 𝐼𝑘 O3×1 ]
]
[O3×3 O3×3 O3×3 I3 O3×1 ] (20)
y2
O [O1×3 O1×3 O1×3 O1×3 1 ]
y
x2 (𝑘 = 2, 3, 4, . . . , 𝐿) ,
C
o2 IL E1 = −Ãl𝐼𝑂H, E2 = Ã
𝜔𝜔̃l𝐼𝑂 − Ãl𝐼𝑂Ḣ𝜃̇
𝐼,
I1 I2
E3 = −𝑚I3 , E4 = 𝑚Ãl𝐼𝐶H,
o z2
x
E5 = f𝐶 − 𝑚Ã
𝜔𝜔̃l𝐼𝐶 + 𝑚Ãl𝐼𝐶Ḣ𝜃̇
𝐼,
z
E6 = ̃r𝐼𝑂, E7 = 𝑚̃r𝐼𝐶, E8 = AJ𝐼 H,
Figure 5: A spatial motion rigid body with more than two ends.
𝜔J𝜔 + AJḢ𝜃̇
E9 = −m𝐶 − ̃r𝐼𝐶f𝐶 + Ã 𝐼,
(21)
̃ = A𝑇 A,̇
𝜔 𝜔 = H𝜃,̇

Using Newton’s laws of motion and considering the sign 𝑐𝑦 𝑐𝑧 𝑠𝑥 𝑠𝑦 𝑐𝑧 − 𝑐𝑥 𝑠𝑧 𝑐𝑥 𝑠𝑦 𝑐𝑧 + 𝑠𝑥 𝑠𝑧


conventions, the dynamics equations of the rigid body can be A = [𝑐𝑦 𝑠𝑧 𝑠𝑥 𝑠𝑦 𝑠𝑧 + 𝑐𝑥 𝑐𝑧 𝑐𝑥 𝑠𝑦 𝑠𝑧 − 𝑠𝑥 𝑐𝑧 ] ,
obtained in the global inertial reference frame as [30] [ −𝑠𝑦 𝑠𝑥 𝑐𝑦 𝑐𝑥 𝑐𝑦 ]
1 0 −𝑠𝑦
H = [0 𝑐𝑥 𝑠𝑥 𝑐𝑦 ] ,
𝐿
𝑚r𝐶̈= ∑ q𝐼𝑘 − q𝑂 + f𝐶, [0 −𝑠𝑥 𝑐𝑥 𝑐𝑦 ]
𝑘=1
where l𝐴𝐵 is the column matrix of the position vector from
𝐿 𝐿 (18) point 𝐴 to point 𝐵 in the body-fixed coordinate whose
Ġ𝐼1 = − ∑ m𝐼𝑘 + ∑̃r𝐼1 𝐼𝑘 q𝐼𝑘 + m𝑂 − ̃r𝐼1 𝑂q𝑂 original point is the first input end 𝐼1 ; ̃l𝐴𝐵 is the cross product
𝑘=1 𝑘=2
matrix of l𝐴𝐵 . The meanings of other signs are the same with
− m̃r𝐼1 𝐶r𝐼̈
1
+ m𝐶 + ̃r𝐼1 𝐶f𝐶, those in [19, 30].
By geometrical equations (17) and rewriting them in the
form of equation (14), the concrete form of matrices H𝑗 and
H𝑗,𝐼𝑘 can be acquired as
where 𝑚 and r𝐶̈are the mass and the column matrix of mass
center acceleration of the rigid body respectively; q𝐼𝑘 (𝑘 = H𝑗 = [I6 O6×7 ]
1, 2, . . . , 𝐿) and q𝑂 are the column matrices of internal forces
acting on the point 𝐼𝑘 (𝑘 = 1, 2, . . . , 𝐿) and 𝑂, respectively; H𝑗,𝐼𝑘
f𝐶 and m𝐶 are the column matrices of external force and the
external torque acting on the mass center of the rigid body; I3 Ãl𝐼1 𝐼𝑘 H O3×3 O3×3 Ãl𝐼1 𝐼𝑘 Ḣ𝜃̇− Ã
𝜔𝜔̃l𝐼1 𝐼𝑘
G𝐼1 is the column matrix of relative moment of momentum =[ ]
O3×3 I3 O3×3 O3×3 O3×1
with respect to the first input end 𝐼1 of the body element;
m𝐼𝑘 (𝑘 = 1, 2, . . . , 𝐿) and m𝑂 are the column matrices (𝑘 = 2, 3, 4, . . . , 𝐿) .
of internal torques acting on points 𝐼𝑘 (𝑘 = 1, 2, . . . , 𝐿) (22)
and 𝑂, respectively; r𝐼1 𝑃 represents the column matrix of
position vectors from 𝐼1 to point 𝑃 with respect to the inertial It can be clearly seen from equation (19) that the matrix is
coordinate system, where 𝑃 represents 𝐼𝑘 (𝑘 = 2, 3, . . . , 𝐿), 𝐶, exactly the same with the transfer matrix of rigid body with
or 𝑂. single input end and single output end, which in fact can be
6 Advances in Mechanical Engineering

regarded as a special case of rigid body with more than two Similarly, the geometrical equations corresponding to
ends (multiple input ends and single output end). body elements 10 and 4 can be deduced as
Based on the transfer equations and geometrical equa-
tions of elements, it is then easy to get the overall transfer G17−10 z17,0 + G18−10 z18,0 = 0,
equation of the system automatically.
G15−4 z15,0 + G16−4 z16,0 + G17−4 z17,0 (30)
According to the topology figure of the system shown in
Figure 1, the relations among the state vectors and transfer + G18−4 z18,0 = 0,
equations of elements can been described more intuitively
and directly using topology described by state vectors and where
transfer equations as shown in Figure 6. G17−10 = −H10 U13 U17 ,
Then the main transfer equations of the system in
Figure 6 can be easily deduced, that is, G18−10 = H10,14 U14 U18 ,

z1,0 = U1 z2,1 = U1 U2 U3 z4,3 = U1 U2 U3 U4 z4,5 G15−4 = − H4 U5 U7 U8 U9 U11 U15 ,


(31)
+ U1 U2 U3 U4,6 z4,6 G16−4 = −H4 U5 U7 U8 U9,12 U12 U16 ,

= U1 U2 U3 U4 U5 U7 U8 z8,9 + U1 U2 U3 U4,6 U6 z10,6 G17−4 = H4,6 U6 U10 U13 U17 ,

= U1 U2 U3 U4 U5 U7 U8 U9 z9,11 G18−4 = H4,6 U6 U10,14 U14 U18 .


(23)
+ U1 U2 U3 U4 U5 U7 U8 U9,12 z9,12 The overall transfer equation of the system can be
obtained by combining the main transfer equation (23) and
+ U1 U2 U3 U4,6 U6 U10 z10,13 all the geometrical equations (28) and (30) of the system:
+ U1 U2 U3 U4,6 U6 U10,14 z10,14 Uall zall = 0, (32)
= T15−1 z15,0 + T16−1 z16,0 + T17−1 z17,0 + T18−1 z18,0 , where
−I T15−1 T16−1 T17−1 T18−1
where [ O G15−9 G16−9
[ O O ] ],
Uall = [
T15−1 = U1 U2 U3 U4 U5 U7 U8 U9 U11 U15 , O O O G17−10 G18−10 ]
[ O G15−4 G16−4 G17−4 G18−4 ]
T16−1 = U1 U2 U3 U4 U5 U7 U8 U9,12 U12 U16 ,
(24) z1,0 (33)
T17−1 = U1 U2 U3 U4,6 U6 U10 U13 U17 , [z15,0 ]
[ ]
zall = [ ]
[z16,0 ] .
T18−1 = U1 U2 U3 U4,6 U6 U10,14 U14 U18 . [z17,0 ]
Further, the geometrical equation of the body element 9 is [z18,0 ]
According to the sign conventions and equations (32)
H9 z9,11 = H9,12 z9,12 . (25) and (33), it can be seen clearly that the overall transfer
equation of the tree system, such as Figure 1, can be deduced
Applying system topology described by state vectors and automatically by handwriting and by computer. For more
transfer equations in Figure 6, the state vectors z9,11 and z9,12 details, see Section 5 please.
can be expressed using state vectors of system boundaries as As a short conclusion, in the overall transfer equation, zall
consists of all the state vectors at the boundary ends of system.
z9,11 = U11 U15 z15,0 ,
(26) Besides the main transfer equation, there exists a geometrical
z9,12 = U12 U16 z16,0 . equation at multiple input body elements that finally leads to a
branch in a tree system. Moreover, the number of geometrical
Substituting equation (26) into equation (25), one obtains equations in the overall transfer equation of a system is equal
to the number of tips of a tree minus one. For the system
−H9 U11 U15 z15,0 + H9,12 U12 U16 z16,0 = 0 (27) shown in Figure 1, the number of the tips is 4; thus the
number of geometric equations of the system is 3.
which can be written as It can be seen clearly that overall transfer equation (8) of a
chain system is a special example of overall transfer equations
G15−9 z15,0 + G16−9 z16,0 = 0, (28) (32) and (33) at the case of the tree system with only two
boundary ends.
where
G15−9 = −H9 U11 U15 , 4.4. Automatic Deduction of the Overall Transfer Equations of
(29) a General System. As is shown in Figure 7, a general multi-
G16−9 = H9,12 U12 U16 . body system can be considered as a system which consists
Advances in Mechanical Engineering 7

𝒛15,0 𝒛16,0 𝒛17,0 𝒛18,0

𝒛15,11 = 𝑼15 𝒛15,0 𝒛16,12 = 𝑼16 𝒛16,0 𝒛17,13 = 𝑼17 𝒛17,0 𝒛18,14 = 𝑼18 𝒛18,0

𝒛9,11 = 𝑼11 𝒛15,11 𝒛9,12 = 𝑼12 𝒛16,12 𝒛10,13 = 𝑼13 𝒛17,13 𝒛10,14 = 𝑼14 𝒛18,14

𝒛8,9 = 𝑼9 𝒛9,11 + 𝑼9,12 𝒛9,12 𝒛10,6 = 𝑼10 𝒛10,13 + 𝑼10,14 𝒛10,14

𝒛7,8 = 𝑼8 𝒛8,9

𝒛7,5 = 𝑼7 𝒛7,8

𝒛4,5 = 𝑼5 𝒛7,5 𝒛4,6 = 𝑼6 𝒛10,6

𝒛4,3 = 𝑼4 𝒛4,5 + 𝑼4,6 𝒛4,6

𝒛2,3 = 𝑼3 𝒛4,3

𝒛2,1 = 𝑼2 𝒛2,3

𝒛1,0 = 𝑼1 𝒛2,1

Figure 6: System topology described by state vectors and transfer equations.

of one tree subsystem and some closed-loop subsystems. where


After “cutting” at the junction of any two adjacent elements
body 16 and hinge 19 in a closed-loop subsystem, a couple
of “new boundaries” noted as z16,0 and z19,0 , which are the −I T15−1 T16−1 T19−1 T18−1
same in nature, will emerge at the “cutting point.” Then, as [ O G15−9 G16−9 O O ]
[ ]
shown in Figure 8, the original nontree system shown in Uall = [
[ O O O G 19−10 G ]
18−10 ] ,
Figure 7 will become a tree system with the same two “new [ O G15−4 G16−4 G19−4 G18−4 ]
boundaries” state vectors z16,0 and z19,0 at the “cutting points.”
[O O C −I O ]
And the overall transfer equation of the general system (37)
with closed-loop subsystems can be deduced automatically z1,0
by handwriting and by computer according to the method [z15,0 ]
[ ]
proposed in Section 4.3. zall = [ ]
[z16,0 ] .
It should be pointed out that either z16,0 or z19,0 is the [z19,0 ]
input end state vector of elements 16 or 19 respectively. Thus, [z18,0 ]
the internal force and internal moment in z16,0 and z19,0 are
of the same quantities but with opposite direction due to the
sign conventions, that is,
Considering that z19,0 can be explicitly expressed by z16,0
z19,0 = Cz16,0 , (34) due to equation (34), the overall transfer equation, the overall
transfer matrix, and the overall state vectors equation (37) of
where the general system can be rewritten in the following form
𝐼𝑛 𝑂𝑛 𝑂𝑛×1
C = [ 𝑂𝑛 −𝐼𝑛 𝑂𝑛×1 ] (35)
[𝑂1×𝑛 0 1 ] Uall zall = 0

𝑛 is equal to 3 for planar motion or 6 for spatial motion. −I T15−1 T16−1 + T19−1 C T18−1
Considering the method proposed in Section 4.3 for a [ O G15−9 G16−9 O ]
[
Uall = [ ],
tree system, and regarding equation (34) concerned with the O O G19−10 C G18−10 ]
relationship between z16,0 and z19,0 , the overall transfer equa- [ O G15−4 G16−4 + G19−4 C G18−4 ] (38)
tions of a general system with a closed-loop subsystem shown
in Figure 8 can be obtained automatically by handwriting and z1,0
[z15,0 ]
by computer: zall = [ ]
[z16,0 ] ,
Uall zall = 0, (36) [z18,0 ]
8 Advances in Mechanical Engineering

0
𝒛15,0
0 𝒛18,0 0 𝒛15,0 0 𝒛 𝒛19,0 0 𝒛18,0
16,0
15 16 17 18 15 16 19 17 18
19

11 12 13 14 11 12 13 14

9 10 9 10

8 8
7 6 6
7
5 5
4 4

3 3
2 2

Root 1 1
0 Root 0
𝒛1,0 𝒛1,0

Figure 7: Topology figure of a nontree system. Figure 8: Topology figure of a nontree system becoming a tree
system after “cutting” the hinge 19.

where
(2) For a tree system, in the first line of the overall
T15−1 = U1 U2 U3 U4 U5 U7 U8 U9 U11 U15 , transfer matrix, the coefficient matrix of the state
vector of root is a minus unit matrix, while each
T16−1 = U1 U2 U3 U4 U5 U7 U8 U9,12 U12 U16 , coefficient matrix of the state vector of a tip is the
successive premultiplication of the transfer matrices
T19−1 = U1 U2 U3 U4,6 U6 U10 U13 U17 U19 ,
of all elements in the transfer path from this tip to
T18−1 = U1 U2 U3 U4,6 U6 U10,14 U14 U18 , the root; besides the first line in the overall transfer
matrix, all coefficient matrices of state vectors in
G15−9 = −H9 U11 U15 , the first column are zero matrices. Except the first
line, in each row, each nonzero partitioned matrix
G16−9 = H9,12 U12 U16 , corresponds to the coefficient matrix of the tip state
(39)
G19−10 = −H10 U13 U17 U19 , vector, from which there is a transfer path to the input
end of the element with multiple input ends. Each
G18−10 = H10,14 U14 U18 , nonzero coefficient matrix of the state vector of a
tip is the successive premultiplication of all transfer
G15−4 = −H4 U5 U7 U8 U9 U11 U15 , matrices of elements in the transfer path from this tip
to the 𝑘th input end 𝐼𝑘 of jth body element, which has
G16−4 = −H4 U5 U7 U8 U9,12 U12 U16 ,
more than two ends, then premultiplied by −H𝑗 for
G19−4 = H4,6 U6 U10 U13 U17 U19 , 𝑘 = 1 or premultiplied by H𝑗,𝐼𝑘 for 𝑘 = 2, 3, 4 ⋅ ⋅ ⋅ 𝐿.

G18−4 = H4,6 U6 U10,14 U14 U18 . (3) For a chain system, its overall transfer matrix is
deduced automatically by successive premultiplica-
tion of the transfer matrices of all elements in the
5. Automatic Deduction Theorem of Overall transfer path from the tip to the root of system. In
Transfer Equation fact, any chain system can be considered as a special
example of the tree system in the case with only two
The following features of the overall transfer equation of a boundary ends.
multibody system can be clearly concluded from equations
(4) For a closed-loop system, its overall transfer equation
(23), (24), (28), and (30). These features make up the theorem
is deduced automatically as the chain system, after
to deduce automatically the overall transfer equation as the
treating the original system as the chain system by
following for tree system (1, 2), for chain system (1, 3), for
“cutting” a junction of any two adjacent elements and
closed-loop system (1, 3, 4), and for general system (1, 2, 5).
letting the couple of “cutting points” as the tip and
(1) The state vectors involved in an overall transfer root with the same state vectors of the chain system.
equation are the column matrix comprising the state (5) For a general system composed of one tree subsystem
vectors of all boundary ends of the system. and some closed-loop subsystems, its overall transfer
Advances in Mechanical Engineering 9

0 𝐾𝑥,7 = 1000 N/m, 𝐾𝑦,6 = 𝐾𝑦,7 = 500 N/m, and 𝐾6󸀠 = 𝐾7󸀠 =
8 150 N ⋅ m/rad.

7 According to the theorem to deduce automatically the


3 2 1 0 overall transfer equation, the topology figure of the system
4
5 x3 can be got as in Figure 10, and the overall transfer equation of
y above system is deduced automatically by handwriting and by
6
o x computer as follows
z
0
Uall zall = 0, (40)
Figure 9: Tree multi-rigid-flexible-body system.
where the overall transfer matrix is
−I T6−1 T8−1
0 𝒛 Uall = [ ]. (41)
8,0 O G6−5 G8−5
𝒛6,0 8
0 State vectors of all boundary ends are
7
6
5
z1,0
4 zall = [z6,0 ] . (42)
3
[z8,0 ]
2 According to the proposed sign conventions, we know that

1 T6−1 = U1 U2 U3 U4 U5 U6 ,
𝒛1,0
Root 0 T8−1 = U1 U2 U3 U4 U5,7 U7 U8 ,
(43)
Figure 10: Topology figure of the system. G6−5 = −H5 U6 ,
G8−5 = H5,7 U7 U8 ,
equation is deduced automatically as the tree system, where z1,0 is state vector of root, z6,0 and z8,0 are the state
after treating the original system as the tree system by vectors of tips, and U1 , U2 , U3 , U4 , U5 , U5,7 , U6 , U7 , and U8
“cutting” the junctions of any two adjacent elements are the transfer matrices of body 1, fixed hinge 2, uniform
in every closed-loop subsystem and letting every beam 3, fixed hinge 4, body 5, elastic hinge 6, elastic hinge 7,
couple of “cutting points” as new “boundary ends” and body 8. All of the transfer matrices can be found directly
with the same state vectors of the tree system. from the library of transfer matrices in [19, 30].
There are boundary conditions of the system
The theorem above is effective for various multi-rigid-body
𝑇
systems, chain multi-rigid-flexible-body systems, and any z1,0 = [𝑥̈𝑦̈ 𝜃𝑧̈ 0 0 0 1]1,0 ,
closed-loop multi-rigid-flexible-body systems and is effective
for various tree multi-rigid-flexible-body systems and general 𝑇
z6,0 = [0 0 0 𝑚𝑧 𝑞𝑥 𝑞𝑦 1]6,0 , (44)
multi-rigid-flexible-body systems if the bodies with more
than two ends are rigid bodies. For more general systems 𝑇
z8,0 = [𝑥̈𝑦̈𝜃𝑧̈ 0 0 0 1]8,0 .
including the flexible bodies with more than two ends, the
theorem to deduce automatically overall transfer equation is
The system experiences a step upward force at the mass
undergoing study and will be discussed in another paper.
center of body element 1 at time instant zero, while the
initial displacement and velocity of the whole system are zero.
6. Numerical Examples The computational results of the system dynamics obtained
by the proposal method and by Newton-Euler method are
By comparison with Newton-Euler method, the numerical shown in Figure 11. It can be seen from Figure 11 that the
examples here are carried out to validate the proposed computational results obtained by the above two methods
method. have good agreements.
Example 1. A tree multi-rigid-flexible-body system moving Example 2. A multi-rigid-body system containing a close
in plane, as shown in Figure 9, consists of two fixed hinges, 2 loop moving in plane is shown in Figure 12. The planar rigid
and 4, two elastic hinges, 6 and 7, three rigid bodies, 1, 5, and 8, bodies (numbered 1, 3, 5, 7, 9, 11, and 13) are connected by pin
and one uniform beam element, 3, with three boundary ends. hinges (numbered 2, 4, 6, 8, 10, 12, and 14), and rigid body
The simulation parameters are given as follows: 𝑚1 = 𝑚5 = 1 is connected with the ground by a smooth pin. Each rigid
𝑚8 = 7.8 kg, 𝐽𝐶,1 = 𝐽𝐶,5 = 𝐽𝐶,8 = 0.013 kg ⋅ m2 , 𝑙3 = 3 m, body has the identical dynamics parameter as 𝑚 = 1 kg and
𝐸𝐴 3 = 1000 N, 𝑚3 = 0.78 kg/m, 𝐸𝐼3 = 166.67 N ⋅ m2 , 𝐾𝑥,6 = 𝐽𝐶 = (1/6) kg ⋅ m2 .
10 Advances in Mechanical Engineering

0
0.2

1
0.15
Rotation angle (rad)

3
0.1
4 10

5 11
0.05
6 12

7 13
0
0 1 2 3 4 5 8 14
9
t (s)
Figure 12: A general system moving in plane.
MSTMM
Newton-Euler method
0 𝒛9,0
Figure 11: Computational results of the angle of the right end of the 𝒛14,0
beam. 9 0
8 14

By “cutting” at the junction of body 9 and hinge 14 7 13


and following the proposed theorem, topology figure of the 6 12
system can be drawn in Figure 13. And the overall transfer
equation of the system can be derived automatically as 5 11

Uall zall = 0, 4 10
(45)
I1 3 I2
where the overall transfer matrix takes the form: 2

−I T9−1 T14−1 1
Uall = [ O G9−3 G14−3 ] . (52) 0 𝒛1,0
[O C −I ]
Figure 13: Topology figure of the system.
State vectors of all boundary ends are

z1,0
zall = [ z9,0 ] . (46) The initial angle of rigid body 1 is (𝜋/6) rad and the
relative angles of pin hinges (numbered 2, 4, 6, 8, 10,
[z14,0 ] 12, and 14) are all zero. The system moves from the rest
According to the proposed sign conventions, one can acquire under the effect of gravity. The computational results of the
system dynamics are obtained by the proposal method and
T9−1 = U1 U2 U3 U4 U5 U6 U7 U8 U9 , by Newton-Euler method. The time history of rigid body
1’s angle is exhibited in Figure 14, which shows that the
T14−1 = U1 U2 U3,10 U10 U11 U12 U13 U14 , computational results obtained by the above two methods
(47)
G9−3 = −H3 U4 U5 U6 U7 U8 U9 , have good agreements.

G14−3 = H3,10 U10 U11 U12 U13 U14 . 7. Conclusions


z1,0 is the state vector of the root; z9,0 and z14,0 are the state Based on MSTMM and the features of the overall transfer
vectors of tips emerging after the “cutting.” U1 , U3 , U3,10 , equations of multibody systems with various topological
U5 , U7 , U9 , U11 , and U13 are the transfer matrices of the structures, including chain systems, closed-loop systems, tree
corresponding planar rigid bodies. U2 , U4 , U6 , U8 , U10 , U12 , systems, general systems composed of one tree subsystem,
and U14 are transfer matrices of the corresponding planar pin and some closed-loop subsystems, the theorem to deduce
hinges. automatically the overall transfer equation by handwriting
There are boundary conditions of the system: and by computer is presented. Formulations of the proposed
𝑇 theorem as well as two numerical examples are given to verify
z1,0 = [0 0 𝜃𝑧̈ 0 𝑞𝑥 𝑞𝑦 1]1,0 . (48) the theorem. This makes it possible to program large-scale
Advances in Mechanical Engineering 11

[8] W. J. Book, “Recursive Lagrangian dynamics of flexible manip-


ulator arms,” The International Journal of Robotics Research, vol.
0.6
3, no. 3, pp. 87–101, 1984.
[9] R. L. Huston, “Multi-body dynamics including the effects of
0.4
Rotation angle (rad)

flexibility and compliance,” Computers and Structures, vol. 14,


no. 5-6, pp. 443–451, 1981.
0.2
[10] Y. A. Khulief and A. A. Shabana, “Dynamic analysis of con-
strained system of rigid and flexible bodies with intermittent
0
motion,” ASME Journal of Mechanisms, Transmissions and
Automation in Design, vol. 108, no. 1, pp. 38–45, 1986.
−0.2
[11] R. A. Laskin, P. W. Likins, and R. W. Longman, “Dynamical
equations of a free-free beam subject to large overall motions,”
−0.4 Journal of the Astronautical Sciences, vol. 31, no. 4, pp. 507–527,
1983.
0 2 4 6 8 10 [12] M. Pascal, “Dynamical analysis of a flexible manipulator arm,”
t (s)
Acta Astronautica, vol. 21, no. 3, pp. 161–169, 1990.
[13] J. Rauh and W. Schiehlen, “A unified approach for the modeling
MSTMM of flexible robot arms,” in Proceedings of the 6th CISM-IFToMM
Newton-Euler method Symposium on Theory and Practice of Robots and Manipulators,
Cracow, Poland, September 1986.
Figure 14: Computational results of the angle of rigid body 1. [14] R. E. Roberson, “A form of the translational dynamical equa-
tions for relative motion in systems of many non-rigid bodies,”
Acta Mechanica, vol. 14, no. 4, pp. 297–308, 1972.
software for MSD with much higher computational speed and [15] O. A. Bauchau, “Computational schemes for flexible, nonlinear
much higher programming. multi-body systems,” Multibody System Dynamics, vol. 2, no. 2,
pp. 169–225, 1998.
[16] S. C. Chu and K. C. Pan, “Dynamic response of a high speed
Conflict of Interests slider crank mechanism with an elastic connecting rod,” ASME
Journal of Engineering for Industry, vol. 97, no. 2, pp. 542–550,
The authors declare that there is no conflict of interests 1975.
regarding the publication of this paper. [17] E. Kreuzer and O. Kust, “Analysis of long torsional strings by
proper orthogonal decomposition,” Archive of Applied Mechan-
Acknowledgments ics, vol. 67, no. 1-2, pp. 68–80, 1996.
[18] T. R. Kane, P. W. Likins, and D. A. Levinson, Spacecraft
The research was supported by Research Fund for the Dynamics, McGraw-Hill, New York, NY, USA, 1983.
Doctoral Program of Higher Education of China (no. [19] X. Rui, L. Yun, Y. Lu et al., Transfer Matrix Method for Multibody
20113219110025). The authors owe special thanks to Professor System and Its Application, Science Press, Beijing, China, 2008.
Dieter Bestle for his important discussion with them during [20] X. Rui, K. Huang, and Y. Qin, “Transfer matrix method of
his five months visit in Nanjing University of Science and multibody system and its application in spacecraft dynamics,”
Technology. in Proceeding of the 3rd China-Russia-Ukraine Symposium on
Astronautical Science and Technology, pp. 933–935, Northwest-
ern Polytechnical University Press, Xi’an, China, 1994.
References [21] X. Rui, B. He, Y. Lu, W. Lu, and G. Wang, “Discrete time transfer
matrix method for multibody system dynamics,” Multibody
[1] W. Schiehlen, Multibody Systems Handbook, Springer, Berlin,
System Dynamics, vol. 14, no. 3-4, pp. 317–344, 2005.
Germany, 1990.
[22] X. Rui, G. Wang, Y. Lu, and L. Yun, “Transfer matrix method for
[2] A. A. Shabana, Dynamics of Multibody Systems, Cambridge linear multibody system,” Multibody System Dynamics, vol. 19,
University Press, New York, NY, USA, 2009. no. 3, pp. 179–207, 2008.
[3] D. Bestle and P. Eberhard, “Analyzing and optimizing multibody [23] X. Rui, B. He, B. Rong, G. Wang, and Y. Lu, “Discrete time
systems,” Mechanics of Structures and Machines, no. 20, pp. 67– transfer matrix method for dynamics of a multi-rigid- flexible-
92, 1992. body system moving in plane,” Proceedings of the Institution of
[4] K. Thoma, F. Schäfer, S. Hiermaier, and E. Schneider, “An Mechanical Engineers K, vol. 223, no. 1, pp. 23–42, 2009.
approach to achieve progress in spacecraft shielding,” Advances [24] X. Rui and J. Zhang, “Automatical transfer matrix method of
in Space Research, vol. 34, no. 5, pp. 1063–1075, 2004. multibody system,” in Proceedings of the 2nd Joint International
[5] J. Wittenburg, Dynamics of Systems of Rigid Bodies, B. G. Conference on Multibody System Dynamics, Stuttgart, Germany,
Teubner, Stuttgart, Germany, 1977. May-June 2012.
[6] O. A. Bauchau, Flexible Multibody Dynamics, Springer, [25] X. Rui and B. Rong, “Advance in transfer matrix method for
Dodrecht, The Netherlands, 2011. multibody system dynamics,” Advance in Mechanics, vol. 42, no.
[7] J. A. C. Ambrósio, M. A. Neto, and R. P. Leal, “Optimization of a 1, pp. 4–17, 2012.
complex flexible multibody systems with composite materials,” [26] X. Rui, B. Rong, G. Wang, and B. He, “Discrete time transfer
Multibody System Dynamics, vol. 18, no. 2, pp. 117–144, 2007. matrix method for dynamics analysis of complex weapon
12 Advances in Mechanical Engineering

systems,” Science China Technological Sciences, vol. 54, no. 5, pp.


1061–1071, 2011.
[27] X. Rui, “Launch dynamics of multibody system and its applica-
tion,” Engineering Science, vol. 13, no. 10, pp. 76–81, 2011.
[28] G. Wang, X. Rui, F. Yang, and H. Xu, “Dynamics analysis
for wheeled and tracked multiple launch rocket system,” Acta
Armamentarii, vol. 33, no. 11, pp. 1286–1290, 2012.
[29] B. Rong, X. Rui, G. Wang, and F. Yang, “Discrete time transfer
matrix method for dynamics of multibody system with real-
time control,” Journal of Sound and Vibration, vol. 329, no. 6,
pp. 627–643, 2010.
[30] X. Rui, D. Bestle, and J. Zhang, “A new form of the trans-
fer matrix method for multibody systems,” in Proceedings of
the ECCOMAS Thematic Conference on Multibody Dynamics,
Zagreb, Croatia, 2013.
Hindawi Publishing Corporation
Advances in Mechanical Engineering
Volume 2014, Article ID 545126, 7 pages
http://dx.doi.org/10.1155/2014/545126

Research Article
Optimal Design of One-Folded Leaf Spring with
High Fatigue Life Applied to Horizontally Vibrating
Linear Actuator in Smart Phone

Ki Bum Lee,1 Chang Hyun Park,2 and Jin Ho Kim1


1
Department of Mechanical Engineering, Yeungnam University, Dae-dong, Gyeongsan-si,
Gyeongsangbuk-do 712-160, Republic of Korea
2
PIDOTECH Inc., Republic of Korea

Correspondence should be addressed to Jin Ho Kim; jinho@ynu.ac.kr

Received 4 July 2013; Revised 12 November 2013; Accepted 25 November 2013; Published 24 February 2014

Academic Editor: Xiaoting Rui

Copyright © 2014 Ki Bum Lee et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Horizontally vibrating linear actuator (HVLA) instead of VVLA has been under study in a few past years and recently HVLA
with thickness of 2.5 mm was developed. The one-folded leaf spring to guide the moving part is newly designed and applied in
HVLA, but unfortunately it makes HVLA be wider. Accordingly, this paper presents the optimal design of one-folded leaf spring,
which results in reduction of HLVA width. The commercial design optimization tool “PIAnO” was utilized based on design of
experiments (DOE), approximation techniques, and optimization algorithm. In addition, for the vibration modal analysis and
harmonic response analysis to generate metamodeling, the software “ANSYS” is utilized. The optimal width of leaf spring was
reduced by 46% compared to the initial one, while all the design constraints were satisfied, which clearly showed the validity of the
proposed design approach.

1. Introduction part of the actuator and enable the actuator to vibrate elas-
tically. Accordingly, the various designs of guide spring had
The slimming of a smart phone has become a competitive been presented to achieve the high fatigue life [3, 4] and
issue in the smart phone industry because the display panel is among them the one-folded leaf spring introduced by Lee and
enlarged to show more information to the user. A smart Kim has the highest fatigue life [5]. Figures 2(a) and 2(b) show
phone is comprised of numerous components: the battery, the the one-folded spring and the schematic diagrams of the
camera module, the vibrating actuator, and so forth. One of HVLA which has the one-folded spring, respectively. Table 1
the thickest components is vibrating actuator which imposes shows the performance characteristics of the HVLA.
the restrictions on the slimming of smart phones. In currently This one-folded leaf spring enables the HVLA to improve
manufactured smart phone in the market, a vertical vibrating the vibration force and product life. Instead, the HVLA has
linear actuator (VVLA) is being widely used as a vibration been enlarged in width due to the wide structure of one-
motor, as shown in Figure 1. folded spring. Figure 3 shows the top view of the HVLA with
However, it imposes the thickness problem on a smart one-folded leaf spring.
phone, because it needs vibration space in the vertical direc- This one-folded leaf spring has occupied 45% of total
tion [1]. Therefore, a horizontally vibrating linear actuator width of HVLA. Therefore, the design of one-folded leaf
(HVLA) has been developed in recent years, but it has not spring needs be optimized to reduce its width. Several
been commercialized in the smart phone industry so far [2]. researches about the optimal design of the mechanical springs
The primary reason is that the fatigue fracture of guide spring such as helical, coil, and leaf spring had been reported in
due to the cyclic compression and tension loads has not been past decades. These springs are usually designed to avoid
resolved. The guide spring has the role to support the moving resonance, but one-folded leaf spring of HVLA needs to
2 Advances in Mechanical Engineering

size of the vibration actuator. This procedure is represented by


the following equation:

Minimize width of leaf spring. (1)

Case
2.1.2. Design Constraint on the Maximum Stress. A one-
Spring folded spring that is affected by cyclic compression and ten-
sion loads should have a yield stress of approximately 215 MPa
Yoke
in order to maintain its fatigue life [7]. The one-folded spring
is represented as follows:
Magnet
𝜎max < 𝜎yield . (2)
Coil
2.1.3. Design Constraint on the Mechanical Natural Frequency.
The mechanical natural frequency of HVLA ranges from
175 Hz to 185.5 Hz:
Figure 1: Schematic diagram of VVLA.
175 ≤ 𝑓 ≤ 185.5. (3)

2.2. Design Variables. We selected the width, length, thick-


Table 1: Specification of the HVLA. ness, height, and radius as design variables for the leaf spring.
Figure 4 shows the shape of the leaf spring, and Table 2
Requirements Values
shows the initial, lower, and upper bound values of the design
Natural frequency 183 Hz
variables.
Thickness of actuator 2.5 mm
Vibration acceleration 2.1 to 2.3 G
2.3. Design Problem Formulation. The design problem for
Fatigue life Over 1,000,000 cycles
determining the design variables that satisfy all the design
requirements can be mathematically formulated as

accomplish the best performance. Moreover, there had been Find 𝑥1 , 𝑥2 , 𝑥3 , 𝑥4 , 𝑥5 ,


no studies reported about the optimal design of one-folded
leaf spring. Therefore, this paper presents an optimization of to minimize Width of spring
the one-folded leaf spring. For optimal design process, we subject to 𝜎max ≤ 𝜎yield
used the commercial optimization software PIAnO (Process
Integration, Automation, and Optimization) which provides 175 ≤ 𝑓 ≤ 185.5
the users with various tools for the efficient optimization
design [6]. (Ver.3.5, PIDOTECH) The optimal design objec- 1.79 ≤ 𝑊 ≤ 3.6 (4)
tive is to minimize the width of one-folded spring to satisfy
the stiffness and maximum stress at cyclic loading condition. 6 ≤ 𝐿 ≤ 15.8
Five design variables were selected to derive the optimal 0.15 ≤ 𝑇 ≤ 0.35
design of one-folded leaf spring and an orthogonal array used.
Modal and harmonic response analyses were performed 0.5 ≤ 𝐻 ≤ 1.5
according to the design of experiments to obtain the natural
frequency of the first vibration mode due to stiffness and the 0.5 ≤ 𝑅 ≤ 0.75.
maximum stress at resonance using the commercial struc-
tural analysis program ANSYS. (Version 14.5, ANSYS) And 3. Analysis Procedures and Optimal Design
then, analyses results were saved as a text file and reimported
into the PIAnO for generating metamodel. Metamodels were 3.1. Analysis Procedures. Figure 5 shows our analysis proce-
generated based on the results of structural analysis using dures. First, we generated a design of experiments (DOE) and
a Kriging model with PIAnO. Finally, we derived a global then the modal and harmonic analyses were response per-
optimal point using evolution algorithms. formed according to sampling points. After, metamodels
were generated using the Kriging algorithm provided by the
PIAnO software. We used an optimization technique using an
2. Design Problem evolution algorithm (EA) to find the optimum solution.
2.1. Design Requirements
3.2. Design of Experiments. After determining the experi-
2.1.1. Minimization of the Width of Leaf Spring. The width of mental design using an orthogonal array, which is one of the
one-folded spring should be minimized in order to reduce the tools provided by 𝐿 98 (715 ), the leaf spring was designed
Advances in Mechanical Engineering 3

15 mm
m
m
.5
16

Case

Leaf spring

Housing

Weight

Magnets Yoke
Coil
PCB

Bracket

(a) (b)

Figure 2: (a) The one-folded spring of HVLA and (b) schematic diagram of HVLA.

Table 2: Initial, lower, and upper bound values of the selected design variables.

Design variables Lower bound Initial Upper bound


𝑥1 𝑊 (mm) 1.79 3.6 3.6
𝑥2 𝐿 (mm) 6 11.6 15.8
𝑥3 𝑇 (mm) 0.15 0.25 0.35
𝑥4 𝐻 (mm) 0.5 1 1.5
𝑥5 𝑅 (mm) 0.5 0.64 0.75

Width Width W
of spring of spring

Length L
of weight
R
H

Figure 4: Design variables of the one-folded leaf spring.

Figure 3: Top view of HVLA with the one-folded leaf spring. saturated points is the minimum required number of sam-
pling points to generate a full quadratic polynomial model.
An orthogonal array with the main effect and interaction of
according to each design variable. 𝐿 98 (715 ) was selected each factor was represented by a table, which allows the
by considering the number of saturated points and levels design of experiments to be easily established [8]. An orthog-
according to the number of design variables. The number of onal array, regardless of the lack of theoretical background,
4 Advances in Mechanical Engineering

(1) Design of
Extract experiments
experiments

(2) ANSYS

PIAnO
(modal and harmonic response)
(3) Approximation Generate meta models

(4) Design
optimization Derive the optimal

Applied CAE tools Embedded design tools Used for

Figure 5: Diagram of analysis procedures.

Figure 6: 3D model of the meshed FE model and constraints.


Figure 7: Fatigue fracture of the one-folded spring at folded area.

where 𝑁 is the number of coil windings, 𝐵𝑔 is the magnetic


has the advantage of being able to apply easily fractional repli- flux density in the airgap, 𝑖 is the input electric current, 𝐿 eff is
cation, decomposition method, and so on. Also, the experi- the effective coil length, and 𝑓 is the input frequency of the
ment can use many factors without enlarging the size of the electric current.
experiment, and the interpretation of the experiment data is Accordingly, the harmonic response analysis was per-
simple. formed to calculate the maximum stress near the folded area
at resonance frequency. The maximum stress must be less
3.3. Modal Analysis and Harmonic Response Analysis. 98 leaf than the yield stress. The harmonic responses were performed
spring FE models are created according to sampling points in with 30 intervals between ±10 natural frequencies [12].
the orthogonal array. And then the modal analysis was per- Figure 8 shows the result of harmonic response analysis.
formed to obtain the vibration natural frequency and the
harmonic response analysis was performed to calculate the 3.4. Metamodeling. Metamodeling builds a metamodel that
maximum mechanical stress at the resonance using ANSYS. approximates the relationship between performance indices
and design variables of a real model by using the analysis
(1) Modal Analysis. Figure 6 shows a 3-dimensional finite ele- results at the sampling points specified by a DOE. Figure 9
ment (FE) model of moving part with the one-folded spring shows the result of a parameter study of the nonlinearity of
and constraint. This moving part consists of permanent mag- 𝑥1 (𝑊). The simulation was performed such that 𝑥1 (𝑊) for
nets, yoke, weight, housing, and one-folded spring. Table 3 the upper and lower bounds was divided into five sections.
lists the material properties used in FE model. The modal Then, we modeled the one-folded spring according to a
analysis was conducted to calculate the natural vibration fre- divided 𝑥1 (𝑊) after a fixed initial value of 𝑥2 (𝐿), 𝑥3 (𝑇),
quency according to DOE [9]. 𝑥4 (𝐻), and 𝑥5 (𝑅).
We found that 𝑥3 (𝑇) was a sensitive value. In a previous
(2) Harmonic Response Analysis. Figure 7 shows the fatigue optimization of the leaf spring, we found a sensitive value of
fracture of leaf spring at folded area due to the cyclic loading. 𝑥3 (𝑇) because the first frequency and stress level were very
Cyclic loading is created by the magnetic force which is responsive, depending on the thickness of the spring. For
energized by sinusoidal electric current. It is expressed by example, the stiffness of cantilever beams is given by
[10, 11]
3𝐸𝐼 𝐸𝐻𝑇3 𝐻𝑇3
𝑘= = , (𝐼 = ). (6)
𝐹magnetic = 𝑁𝐵𝑔 𝑖𝐿 eff , 𝑖 = 𝑖0 sin (2𝜋𝑓𝑡) , (5) 𝐿3 4𝐿3 12
Advances in Mechanical Engineering 5

Table 3: Material properties of the moving part.

Component Material Density (g/cc) Modulus of elasticity (Gpa) Poisson’s ratio


Leaf spring SUS304 8.00 193–200 0.29
Weight Tungsten 19.3 400 0.28
Housing SUS304 8.00 193–200 0.29
Magnet NdFe35 7.50–7.80 150–160 0.24
Steel yoke Steel 1010 7.84–7.87 200–205 0.29

148
Amplitude (MPa)

117
93.1
58.8
46.7
37.1
23.4
174 175 178 180 183 185 188 190 193
Frequency (Hz)

Figure 8: Harmonic response of HVLA.

189 200

188
First frequency (Hz)

150
187
Stress (MPa)

186 100

185
50
184

183 0
1.79 2.2425 2.695 3.1475 3.6 1.79 2.2425 2.695 3.1475 3.6
x1 width x1 width

Figure 9: Nonlinearity for 𝑥1 (𝑊).

The stiffness of a cantilever beam is directly proportional response analyses. Metamodels were regenerated with respect
to the cube of the thickness. Therefore, the thickness of the to the width of the one-folded spring, maximum stress, and
leaf spring has a nonlinearity, so we selected the Kriging natural frequency of the first mode. At this point, we added
model, which is one of the metamodels provided in the only 39 types of sampling points to satisfy the 15 multiples
PIAnO software that was selected. The Kriging model, which of the number of design variables corresponding to the
a type of interpolation model, was mathematically estab- sampling points (75). We used augmented Latin hypercube
lished by Metheron in 1963 based on research conducted design (ALHD) that is not overlapped with the existing
by mining researcher Krige in 1951 [13, 14]. And it shows sampling points [16]. Also, it is widely known for excellent
superb predictive performance under many design variables performance of the space filling. Nevertheless, the accuracy of
2
and is in strongly nonlinear systems, provides a statistical 𝑅pred was only about 80%. Therefore, 98 sampling points were
estimation [15]. Also, there are no parameters that depend on reextracted in order to obtain higher accuracy over 90% using
the experience and intuition of customers when choosing the orthogonal array 𝐿 98 (715 ) at a time. Finally, in the evaluation
design parameters because the Kriging model can optimize results of the metamodel, the accuracy of 𝑅pred 2
is over 90%.
design parameters through maximum likelihood estimation
(MLE). Therefore, a recent trend is an increase in the use of
Kriging models in the field of engineering. Our first Kriging 3.5. Optimization Technique. We selected an evolution algo-
model was generated by using an orthogonal array 𝐿 36 (313 ), rithm that is provided in PIAnO. The evolution algorithm
which was based on the analytical results of the width of (EA) was proposed by Holland in 1975 and is a global
the one-folded spring, the maximum stress, and the natural optimization technique [17]. This algorithm describes the
frequency. However, an accuracy evaluation of the Kriging evolution of the biological genetic trait, so it searches for a
2
model using 𝑅pred , which corresponds to the maximum stress, global optimum through a process of selection, recombina-
was underestimated 55% as accuracy. Consequently, we tion, and mutation. On the downside, it requires consider-
added 39 types of sampling points to improve the accuracy of able computation time, depending on the analytical model,
the metamodel and then performed the modal and harmonic because more calculation functions are required compared to
6 Advances in Mechanical Engineering

210
Violation area
200
200
Violation area
Maximum stress (MPa)

1st frequency (Hz)


180
190 187.32
160
148.75
141.12 180 178.17 176.67
142.06
140

170
120 Violation area

100 160
Initial Opt meta Opt exact Initial Opt meta Opt exact
(a) Maximum stress at bending area on leaf spring (b) Natural frequency of first mode

Figure 10: Comparison of the design constraints.

Table 4: Initial and optimal design variable values.

Design variables Lower bound Initial Optimal Upper bound


𝑥1 𝑊 (mm) 1.79 3.60 1.9 3.60
𝑥2 𝐿 (mm) 6.00 11.64 12.45 15.80
𝑥3 𝑇 (mm) 0.15 0.25 0.26 0.25
𝑥4 𝐻 (mm) 0.50 1.00 1.04 1.24
𝑥5 𝑅 (mm) 0.28 0.64 0.63 0.88

other optimization algorithms [18]. But we made up for the 4


disadvantage of the evolution algorithm because we used a 3.6
metamodel with a short analysis time. 3.5

3
Width (mm)

3.6. Results. In our results for the optimal design using the
2.5
metamodel (Opt meta), the optimal design satisfied the con-
straints, including the maximum stress and natural frequency 1.9
2 1.9
of first mode, and the width of one-folded leaf spring
decreased by 47% compared to the initial width of 1.9 mm. 1.5
However, the optimal design results can be changed based on
using the metamodel instead of the actual analytical model in 1
this research. The accuracy of the optimization results should Initial Opt meta Opt exact
be verified by actual analysis using ANSYS. To do this, the Figure 11: Comparison of the objective function.
Kriging model results (Opt meta) of the optimal design vari-
ables and the analysis results from ANSYS (Opt exact) were
compared as shown in Figures 10(a) and 10(b). 15 mm 12.2 mm
The Kriging model results (Opt meta) and the ANSYS
model results (Opt exact) were very similar; therefore, we
confirmed the high accuracy of the Kriging model’s predic-
tion. The initial and optimal values of the design variables
16.5 mm

were compared as shown in Table 4. Figure 11 shows the com-


pared widths of the leaf spring. Figure 12 shows a comparison
of the existing actuator and the actuator with the optimal
width of one-folded spring.

4. Conclusion
We minimized the existing size of a HVLA (16.5 × 15 × 2.5) by
decreasing the width of the leaf spring. The following conclu- Figure 12: Comparison of HLVA according to the initial design and
sions were drawn from our results. optimal design.
Advances in Mechanical Engineering 7

(1) We formulated the design problem in order to mini- [7] http://www.matweb.com/.


mize the maximum stress in the bending area and to [8] A. S. Hedayat, N. J. A. Sloane, and J. Stufken, Orthogonal Arrays:
satisfy the required first mode natural frequency, Theory and Applications, Springer, New York, NY, USA, 1999.
based on the design requirements. [9] Y. Kang, G.-J. Sheen, M.-H. Tseng, S.-H. Tu, and H.-W. Chiang,
(2) We generated an orthogonal array using the com- “Modal analyses and experiments for engine crankshafts,” Jour-
mercial optimization software PIAnO. Then, each leaf nal of Sound and Vibration, vol. 214, no. 3, pp. 413–430, 1998.
spring designed by sampling points was analyzed by [10] K. I. Hwang and J. H. Kim, “Single-axis flat electro-magnetic
using the commercial structural analysis software actuator using shorted turn for fast initial response,” Journal of
Korean Medical Science, vol. 19, no. 6, pp. 222–226, 2009.
ANSYS.
[11] S.-M. Yoon, J.-H. Kim, and J.-H. Kim, “Mathematical modeling
(3) The Kriging model, provided PIAnO, was generated and analysis of vibration characteristics of smart-phone,” Inter-
based on simulation results according to the design of national Journal of Precision Engineering and Manufacturing,
experiments 𝐿 98 (715 ). Then, the optimal design was vol. 14, no. 3, pp. 505–508, 2013.
determined using the evolution algorithm of the [12] Z. Wu, C. Xu, J. Zhang, D. Yu, and P. Feng, “Modal and harmonic
global optimization technique. As a result, all con- reponse analysis and evaluation of machine tools,” in Proceed-
straints were satisfied, and the derived the optimal ings of the International Conference on Digital Manufacturing
width of one-folded spring was decreased by 47% and Automation (ICDMA ’10), vol. 1, pp. 929–933, Changsha,
compared to the initial values. China, December 2010.
[13] D. G. Krige, “A statistical approach to some basic mine valuation
problems on the witwatersrand,” Journal of the Chemical, Met-
Abbreviations allurgical and Mining Society of South Africa, vol. 52, no. 6, pp.
𝑇: Thickness of leaf spring 119–139, 1951.
𝐻: Height of leaf spring [14] G. Metheron, “Principles of geostatistics, economic geology,”
𝐿: Length of leaf spring Economic Geology, vol. 58, no. 8, pp. 1246–1266, 1963.
𝑊: Width of leaf spring [15] T. W. Simpson, J. J. Korte, T. M. Mauery, and F. Mistree, “Com-
𝑅: Radius of leaf spring. parisons of response surface and kriging models for multidis-
ciplinary design optimization,” in Proceedings of the 7th AIAA/
USAF/NASA/ISSMO Symposium on Multidisciplinary Analysis
Conflict of Interests & Oprimization, vol. 1, pp. 381–3391, 1998.
[16] M. Stein, “Large sample properties of simulations using latin
The authors declare that there is no conflict of interests
hypercube sampling,” Technometrics, vol. 29, no. 2, pp. 143–151,
regarding the publication of this paper. 1987.
[17] J. H. Holland, Adaptation in Natural and Artificial Systems, The
Acknowledgment University of Michinan Press, 1975.
[18] R. L. Haupt and S. E. Haupt, Practical Evolutionary Algorithms,
The authors express their gratitude to PIDOTECH, Inc., for Wiley, 1998.
providing their PIAnO software as a PIDO tool.

References
[1] J.-H. Kim, H.-K. Il, J.-W. Lee, and J.-H. Yoo, Horizontally Vibra-
tion Motor for Slimming Mobile Smart Phone, vol. 6, Korean
Society for Precision Engineering, 2011.
[2] J.-H. Kim and J.-H. Kim, “A horizontally vibration linear actua-
tor for slimming of smart phone,” in Proceedings of the Interna-
tional Conference on Computer, Electrical and Systems Confer-
ence (ICCESSE ’11), Phuket, Thailand, December 2011.
[3] JA HWA ELECTRONICS CO., LTD., Linear type vibration
motor of horizontal vibration capable of generating vibration to
horizontal direction by linearly reciprocating mover, KOEAR
Patent, 1007201970000 (20070514).
[4] SAMSUNG ELECTRO-MECHANICS CO., LTD, Linear vibra-
tion motor capable of minimizing a friction area with adja-
cent components when a motor is driven, KOREA Patent,
1010595990000 (20110819).
[5] K.-B. Lee and J.-H. Kim, “Design of leaf spring with high fatigue
life applied to horizontal linear vibrating actuator,” The Korea
Academia-Industrial Cooperation Society, vol. 13, no. 12, pp.
5684–5688, 1975.
[6] PIDOTECH Inc, Process Intergration, Automation, and Opti-
mization, PIAnO User’s Manual, 2010.
Hindawi Publishing Corporation
Advances in Mechanical Engineering
Volume 2014, Article ID 512583, 9 pages
http://dx.doi.org/10.1155/2014/512583

Research Article
Adaptive Terminal Sliding Mode Control of Electromagnetic
Spacecraft Formation Flying in Near-Earth Orbits

Jingrui Zhang,1 Changqin Yuan,2 Dongmei Jiang,2 and Dawei Jin2


1
School of Aerospace Engineering, Beijing Institute of Technology, Beijing 100081, China
2
Department of Aircraft and Dynamics, Aviation University of Air Force, Changchun 130022, China

Correspondence should be addressed to Jingrui Zhang; ruierchat@163.com

Received 25 August 2013; Revised 17 November 2013; Accepted 9 December 2013; Published 11 February 2014

Academic Editor: Xiaoting Rui

Copyright © 2014 Jingrui Zhang et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

An adaptive terminal sliding mode control for six-degree-of-freedom electromagnetic spacecraft formation flying (EMFF) in near-
Earth orbits is presented. By using terminal sliding mode (TSM) technique, the output tracking error can converge to zero in
finite time, and strong robustness with respect to disturbance forces can be guaranteed. Based on a rotated frame 𝐹𝑟 and the
adaptive TSM controller, the special magnetic moment of the steerable magnetic dipole is computed. The angular momentum
management strategy (AMM) is implemented in a periodically switching fashion, by which the angular momentum buildup was
limited. Illustrative simulations of EMFF are conducted to verify the effectiveness of the proposed controller.

1. Introduction forces on the order of 10-9 N/W may be generated using elec-
tromagnetic gradient forces or scattering forces; microwave
Spacecraft formation flying (SFF) represents the concept beam powers of 10 kW can thus produce restoring forces of
of distributing the functionality of large spacecraft among approximately 10 𝜇N, which are sufficient to correct a number
smaller, less-expensive, cooperative spacecraft [1, 2]. Specif- of orbital perturbations. Miller et al. [10, 11] address the
ically, NASA and the U.S. Air Force have identified space- novel concept of electromagnetic formation flying (EMFF) in
craft formation flying as an enabling technology for future which high temperature superconducting (HTS) wire tech-
missions. The practical implementation of the SFF concept nology is used to create magnetic dipoles on each spacecraft
relies on the accurate control of the relative positions and ori- that can be used to maintain and reconfigure the spacecraft
entations between the participating spacecraft for formation formation. However, Since magnetic force on each spacecraft
configuration. The conventional thruster-based schemes may in the formation can be applied in any arbitrary direction
require continuous expenditure of fuel to maintain formation which can be easily created by steerable magnetic dipoles,
geometry that can contaminate the sensitive sensors on board EMFF has advantages in terms of controllability [4].
and mission lifetime also becomes dependent on the fuel A critical component of EMFF is an effective formation
available [3, 4]. flying control. The dynamics and control problem associated
To alleviate these concerns, several propellant-free for- with EMFF become highly challenging, due to the nonlinear
mation flying methods have been proposed in the literature. nature of the magnetic forces. Ahsun and Miller [3] have
The propulsive conducting tethers and spin-stabilized tether presented a hybrid adaptive control scheme in which trans-
systems have been proposed in place of on-board propulsion lation control is implemented in a centralized fashion with
systems to form and maintain satellite formations [5, 6]. a decentralized attitude control. Elias et al. [12] designed a
King et al. [7] have presented Coulomb force approaches linear optimal controller based on the linearized dynamics.
to maintain a formation. The flux pinning technology has Kong et al. [13] addressed the use of electromagnetic dipoles
been applied to achieve passively stable configurations by for relative position and orientation maintenance as needed
HTS electromagnetics [8]. LaPointe [9] has presented the for the terrestrial planet finder (TPF). Reference [14] derived
microwave scattering formation flight method. Radiation the dynamics of an N-spacecraft EMFF (in 2D) for deep
2 Advances in Mechanical Engineering

jth spacecraft spacecraft with respect to the leader spacecraft. The 𝑥-


(follower) axis aligned in the radial (zenith) direction, the 𝑧-axis is
Z rij perpendicular to the orbital plane and points in the direction
Rj r0j of the angular momentum vector, and the 𝑦-axis completes
r0i the right-hand system. (3) The orbital frame 𝐹𝑂𝑖 is used in
R0 ith spacecraft the attitude control. The reference frame is rotating about the
Leader
(follower) 𝑦-axis with respect to ECI frame at orbital rate. The roll axis
Ri 𝑥 is in the fight direction, the pitch axis 𝑦 is perpendicular
to the orbital plane, and the yaw axis 𝑧 points toward the
O
Earth
X Earth. (4) A body frame 𝐹𝐵𝑖 attached to the body of 𝑖th
spacecraft with the center of mass of the spacecraft is also
defined to describe the orientation of each spacecraft in the
Y inertial space. (5) R𝑖 ∈ R3 denotes the position vector from
the origin of ECI frame to 𝑖th spacecraft. r𝑖𝑗 ∈ R3 denotes
Reference orbit

the relative position vector from 𝑖th spacecraft to the follower


Figure 1: Schematic representation of the EMFF system. 𝑇
𝑗th spacecraft. 𝜌𝑖 ≜ [𝑥𝑖 𝑦𝑖 𝑧𝑖 ] ∈ R3 is the expression of
the relative position vector r0𝑖 from the origin of the leader
spacecraft coordinates system to the follower spacecraft 𝑖 in
space missions and discussed a nonlinear control law using the Hill frame 𝐹𝐻. (6) The rotated frame 𝐹𝑟𝑖 is used in a
potential functions. simplified algebraic form of the magnetic force equation. The
The orientation of a dipole obviously depends on the ori- 𝑥-axis is aligned with vector r𝑖𝑗 [3].
entation of the body axes in the inertial space, and changing
the dipole on one spacecraft affects actuation on all other 2.1. Translational Dynamics. Designating spacecraft 0 as the
spacecraft in EMFF. Therefore, general asymptotical stability leader of the formation, the relative dynamics of 𝑖th spacecraft
may not deliver fast enough convergence to meet EMFF in the Hill reference frame 𝐹𝐻 can be represented as
control for high-precision situation. The recently developed
𝜌̈ ̂ ̂
terminal sliding mode (TSM) control enables convergence 𝑖 + C𝑖𝑇 (𝜔𝑂, 𝜌̇
𝑖 ) + N𝑖𝑇 (𝜌𝑖 , 𝜔𝑂, R0 ) = F𝑑𝑖 + F𝑐𝑖 , (1)
to the desired state in finite time [15–18]. This technique where 𝜔𝑂 is the orbital angular velocity of the leader. C𝑖𝑇 (⋅) ∈
has been used successfully in some control designs, such as R3 is a nonlinear term defined as
robotic manipulators [17] and mobile target tracking [18]. The
𝑇
physical interpretation of finite time convergence lies in the C𝑖𝑇 (𝜔𝑂, 𝜌̇𝑖 ) ≜ 2𝜔𝑂[−𝑦𝑖̇ 𝑥𝑖̇ 0] . (2)
fact that the convergence rate of TSM grows exponentially
when the state is near equilibrium. N𝑖𝑇 (⋅) ∈ R3 is nonlinear term defined as
In this Paper, we confine our attention to adaptive TSMC N𝑖𝑇 (𝜌𝑖 , 𝜔𝑂, R0 )
design for EMFF in near-Earth orbits. The remainder of
󵄩 󵄩
this paper is organized as follows. Section 2 presents a 𝑥𝑖 + 󵄩󵄩󵄩R0 󵄩󵄩󵄩 1 2
detailed 6-DOF dynamic modeling for EMFF in near-Earth [ 𝜇 ( 󵄩 󵄩 3
− 󵄩 󵄩2 ) − 𝜔𝑂 𝑥𝑖 − 𝜔̇𝑂𝑦𝑖 ]
[ 󵄩
󵄩 R + 𝜌 󵄩
󵄩 󵄩
󵄩 R 󵄩
󵄩 ]
orbits. In Section 3, a position/attitude tracking controller [ 󵄩 0 𝑖 󵄩 󵄩 󵄩 0 ] (3)
[ 𝑦 ]
based on TSM is developed for EMFF, which ensures error ≜[ [ 𝜇󵄩 𝑖
− 𝜔 2
𝑦 + 𝜔̇ 𝑥 ],
]
convergence in finite time and strong robustness of the [ 󵄩󵄩R0 + 𝜌 󵄩󵄩󵄩3 𝑂 𝑖 𝑂 𝑖
]
bounded disturbances. Based on a rotated frame 𝐹𝑟 and the [ 󵄩 𝑖󵄩 ]
[ 𝑧𝑖 ]
adaptive TSM controller, the special magnetic moments of 𝜇󵄩 󵄩󵄩3
󵄩
the Steerable Magnetic Dipole (SMD) are computed. The [ 󵄩󵄩R0 + 𝜌𝑖 󵄩󵄩 ]
AMM strategy is implemented in a periodically switching where 𝜇 is gravitational constant of earth.
fashion. Simulation results are presented in Section 4. Finally, Note that F̂ 𝑑𝑖 denotes the parametric uncertainty and
Section 5 concludes this paper. external disturbances, such as differential J2 and higher-order
terms, differential solar pressure, and differential drag. The
2. System Model specific relative magnetic force F ̂𝑐𝑖 provided by the coils needs
to be strong enough to cancel these disturbances and at the
Spacecraft formation constitutes an 𝑁-body mechanics prob- same time provide additional force for trajectory following.
lem. Figure 1 shows a typical formation system including 𝑁 Consider the following:
spacecrafts orbiting a central body (Earth). The spacecraft 0’s
̂ 1 1
designated as the “leader” spacecraft. We make the following F𝑐𝑖 = F − F , (4)
considerations. (1) The ECI frame 𝐹𝐼 is defined with its 𝑚𝑖 𝑐𝑚𝑖 𝑚0 𝑐𝑚0
origin at the center of Earth, its 𝑋-axis points toward vernal where
equinox, 𝑍-axis points toward celestial north pole, and 𝑌- 𝑁−1
axis completes a right-handed axis system. (2) The Hill frame F𝑐𝑚𝑖 = ∑ F𝑚
𝑖𝑗 . (5)
𝐹𝐻 is used to visualize the relative motion of each follower 𝑗=0, 𝑗 ≠
𝑖
Advances in Mechanical Engineering 3

By approximating the coils on each spacecraft to SMD, the where 𝜏𝑚𝑖𝑗 is the torque exerted on the SMD of 𝑖th spacecraft
magnetic field due to the 𝑗th spacecraft can be written as [3, due to that of 𝑗th spacecraft [3, 10],
10]
𝜇0 𝜇𝑖 3r𝑖𝑗 (𝜇𝑗 ⋅ r𝑖𝑗 ) 𝜇𝑗
𝜇 3𝜇𝑗 ⋅ r𝑖𝑗 𝜇𝑗 𝜏𝑚
𝑖𝑗 = ×( − 3) (13)
B𝑗 (r𝑖𝑗 ) = 0 ( 5
r𝑖𝑗 − 3 ) (6) 4𝜋 𝑟𝑖𝑗5 𝑟𝑖𝑗
4𝜋 𝑟𝑖𝑗 𝑟𝑖𝑗
𝜏𝑐𝑤𝑖 and 𝜏𝑑𝑖 are the control and disturbance torques, respec-
and the force between spacecrafts can be written as
tively. 𝜏𝑑𝑖 includes the gravity gradient torques 𝜏𝑔𝑖 and the
F𝑚 Earth magnetic torques 𝜏𝑒𝑖 . Consider the following:
𝑖𝑗 ≜ 𝑓 (𝜇𝑖 , 𝜇𝑗 , r𝑖𝑗 )
𝜏𝑑𝑖 = 𝜏𝑔𝑖 + 𝜏𝑒𝑖 . (14)
3𝜇 𝜇𝑖 ⋅ 𝜇𝑗 𝜇𝑖 ⋅ r𝑖𝑗 𝜇𝑖 ⋅ r𝑖𝑗
= 0 (− 5
r𝑖𝑗 − 5
𝜇𝑗 − 𝜇𝑖 A very important factor presented in the disturbance
4𝜋 𝑟𝑖𝑗 𝑟𝑖𝑗 𝑟𝑖𝑗5 (7) torque term is the torque that acts on the spacecraft due to
the Earth’s magnetic field. This torque can be written as [4]
(𝜇𝑖 ⋅ r𝑖𝑗 ) (𝜇𝑗 ⋅ r𝑖𝑗 )
+5 r𝑖𝑗 ) , 𝜏𝑒𝑖 = 𝜇𝑖 × B𝑒 (R𝑖 ) , (15)
𝑟𝑖𝑗7
where B𝑒 is Earth’s magnetic field strength at the location of
where 𝜇0 = 4𝜋 × 10−7 T⋅m/A is the permeability constant. 𝜇𝑖 the spacecraft.
and 𝜇𝑗 are the dipole strength of the 𝑖th satellite and the 𝑗th Substituting (10) into (11), the attitude dynamics can be
satellite, respectively. Note that (7) gives the force on dipole written compactly as
𝑖 (located on 𝑖th spacecraft) due to dipole 𝑗 (located on 𝑗th
spacecraft), which depends on the distance between the two M𝑖𝐴 (J𝑖 , q𝑖 ) 𝜀̈
𝑖 + C𝑖𝐴 (J𝑖 , Ω𝑖 , h) + N𝑖𝐴 (J𝑖 , q𝑖 , 𝜔𝑏𝑖 , 𝜔𝑂𝑖 , 𝜔̇𝑂𝑖 )
dipoles and the orientation of both dipoles in the inertial = 𝜏𝑐𝑤𝑖 + 𝜏𝑚𝑖 + 𝜏𝑑𝑖 ,
space. It is the dependence on the orientation of the dipoles (16)
that gives rise to the complexity of the expression for the force,
since the orientation of a dipole obviously depends on the where
orientation of the body axes in the inertial space [3].
M𝑖𝐴 (J𝑖 , q𝑖 ) = 2J𝑖 T𝑖 −1 ,
2.2. Attitude Dynamics. To avoid singular points, the Euler 𝑇
C𝑖𝐴 (J𝑖 , Ω𝑖 , h) = Ω×𝑖 (J𝑖 Ω𝑖 + h) , X𝑖 = [𝜌𝑇𝑖 , 𝜀𝑇𝑖 ] ,
parameter is chosen to describe the attitude of the spacecraft.
Let q𝑖 represent the Euler parameters corresponding to the
N𝑖𝐴 (J𝑖 , q𝑖 , 𝜔𝑏𝑖 , 𝜔𝑂𝑖 , 𝜔̇𝑂𝑖 ) = 2J𝑖 T𝑖 −1 𝑓𝑖 , 𝑓𝑖 = −𝜔×𝑏𝑖 𝜔𝑂𝑖 + 𝜔̇𝑂𝑖 .
attitude of the body frame relative to the orbital frame 𝐹𝑂𝑖 (17)
[19],
𝑇
q𝑖 = [𝜀𝑇𝑖 𝜂𝑖 ] ∈ R4 , 2.3. Combined Attitude and Translational Dynamics. Com-
(8) bining the attitude dynamics in (1) and the translational
𝑇
𝜀𝑖 ≡ [𝜀𝑖1 𝜀𝑖2 𝜀𝑖3 ] ∈ R3×1 , 𝜂𝑖 ∈ R1×1 . dynamics in (16), the following 6-DOF dynamics equation for
formation flying is obtained:
The Euler parameters, which are equivalent to the coeffi-
cients of unit quaternion, have unit norm by definition; hence M𝑖 Ẍ
𝑖 + C𝑖 + N𝑖 = u𝑖 + D𝑖 , (18)
where
𝜀𝑇𝑖 𝜀𝑖 + 𝜂𝑖2 = 1. (9)
I3×3 0 C
Attitude kinematics and dynamics of the spacecraft are M𝑖 = [ ] ∈ R6×6 , C𝑖 = [ 𝑖𝑇 ] ∈ R6×1 ,
0 M𝑖𝐴 C𝑖𝐴
governed by [19]
N𝑖𝑇 F̂
Ω𝑖 = 2T𝑖 −1 𝜀̇𝑖 + 𝜔𝑂𝑖 , (10) N𝑖 = [ ] ∈ R6×1 , u𝑖 = [ 𝑐𝑖 ] ,
(19)
N𝑖𝐴 𝜏𝑐𝑤𝑖
J𝑖 Ω̇𝑖 + Ω×𝑖 (J𝑖 Ω𝑖 + h) = 𝜏𝑚𝑖 + 𝜏𝑐𝑤𝑖 + 𝜏𝑑𝑖 , (11)
𝑇 F̂
D𝑖 ≜ [𝑑1𝑖 𝑑2𝑖 𝑑3𝑖 𝑑4𝑖 𝑑5𝑖 𝑑6𝑖 ] = [ 𝑑𝑖 ].
where T𝑖 ≡ 𝜂𝑖 I + 𝜀×𝑖 3×3
∈ R , the superscript “×” denotes 𝜏𝑚𝑖 + 𝜏𝑑𝑖
skew-symmetric matrix of a vector, and Ω𝑖 = 𝜔𝑂𝑖 + 𝜔𝑏𝑖 , 𝜔𝑏𝑖 is
angular velocity of the spacecraft relative to the orbital frame 3. Adaptive Terminal Sliding
𝐹𝑂𝑖 . 𝜔𝑂𝑖 is the orbital angular velocity of 𝑖th spacecraft. J𝑖 Mode Control Design
is the moment of inertia of the spacecraft. h is the angular
momentum of the flywheel. The magnetic torque term in (11) In this section, an adaptive terminal sliding mode controller
can be written as is designed for the follower spacecraft based on the dynamic
𝑁−1 model in (18) and the TSM technique. With this controller,
𝜏𝑚𝑖 = ∑ 𝜏𝑚
𝑖𝑗 , (12) the follower spacecraft can track the desired attitude and
𝑗=0, 𝑖 ≠
𝑗 relative position trajectories simultaneously.
4 Advances in Mechanical Engineering

3.1. Error Dynamics Equation. The trajectory tracking errors Substituting (24) and (25) into (11), the error attitude
of the follower spacecraft are defined as dynamics are derived as follows [19]:

𝜌𝑒𝑖 ≜ 𝜌𝑖 − 𝜌𝑑𝑖 , 𝜌̇𝑒𝑖 ≜ 𝜌̇𝑖 − 𝜌̇𝑑𝑖 , (20) M𝑒𝑖𝐴 (J𝑖 , q𝑖 ) 𝜀̈ 𝑒 𝑒 𝑑 𝑑


𝑖 + C𝑖𝐴 (J𝑖 , Ω𝑖 , h) + N𝑖𝐴 (J𝑖 , q𝑖 , 𝜔𝑏𝑖 , 𝜔𝑏𝑖 , 𝜔̇𝑏𝑖 )

where 𝜌𝑑𝑖 , 𝜌̇𝑑𝑖 ∈ R3×1 are the relative position/velocity of the = 𝜏𝑐𝑤𝑖 + 𝜏𝑚𝑖 + 𝜏𝑑𝑖 ,
desired trajectory with respect to the leader. Differentiating (27)
(20) and substituting it into (1), one can obtain that −1 −1
where N𝑒𝑖𝐴 = 2J𝑖 T𝑒𝑖 f𝑖𝑒 , M𝑒𝑖𝐴 = 2J𝑖 T𝑒𝑖 .
𝑇
𝑒
𝜌̈ 𝑒 𝑑 ̂ ̂ 𝑇 𝑇
𝑖 + C𝑖𝑇 (𝜔𝑂, 𝜌̇
𝑖 ) + N𝑖𝑇 (𝜌𝑖 , 𝜔𝑂, R0 , 𝜌̈
𝑖 ) = F𝑑𝑖 + F𝑐𝑖 , Let e𝑖 = [𝜌𝑒𝑖 𝜀𝑒𝑖 ] , and utilizing (18), (21), and (27), the
(21) error dynamics of the whole system can be obtained

where N𝑒𝑖𝑇 (𝜌𝑖 , 𝜔𝑂, R0 , 𝜌̈


𝑑 𝑑 M𝑒𝑖 ë 𝑒
𝑖 + C𝑖 + N𝑖 = u𝑖 + D𝑖 , (28)
𝑖 ) = N𝑖𝑇 + 𝜌̈
𝑖.
The error Euler parameter is chosen to describe the atti-
where
tude error of the spacecraft with respect to the reference atti-
tude. The error Euler parameter is determined as follow [19]: I3×3 0 N𝑒𝑖𝑇
M𝑒𝑖 = [ ] ∈ R6×6 , N𝑒𝑖 = [ ] ∈ R6×1 .
0 M𝑒𝑖𝐴 N𝑒𝑖𝐴
𝑇
𝜀𝑒𝑖 = U𝑑𝑖 q𝑖 , (22) (29)
3.2. Controller Design. The terminal switching planes can be
𝑇 describe as [16–18]
𝜂𝑖𝑒 = q𝑑𝑖 q𝑖 , (23)
𝛾
S𝑖 = ė𝑖 + 𝛼𝑖 e𝑖 + 𝛽𝑖 e𝑖 𝑖 , (30)
𝑇 𝑇
where U𝑑𝑖 ≡ [T𝑑𝑖 −𝜀𝑖 ] ∈ R
4×3
, T𝑑𝑖 ≡ 𝜂𝑖𝑑 I + 𝜀×𝑑
𝑖 ∈R 3×3
, and
𝑇
𝑇 𝑇 where S𝑖 ≜ [𝑠1𝑖 𝑠2𝑖 𝑠3𝑖 𝑠4𝑖 𝑠5𝑖 𝑠6𝑖 ] , 𝛼𝑖 = diag(𝑎1𝑖 , 𝑎2𝑖 , . . . ,
q𝑑𝑖 ≡ [𝜀𝑑𝑖 R4×1 is the desired attitude.
𝜂𝑖𝑑 ] ∈ 𝑎6𝑖 ), 𝛽𝑖 = diag(𝑏1𝑖 , 𝑏2𝑖 , . . . , 𝑏6𝑖 ), 𝑎𝑛 > 0, 𝑏𝑛 > 0 (𝑛 = 1, 2, . . . , 6),
Differentiating (22) with respect to time and utilizing (10), and 0 < 𝛾𝑖 < 1. The unknown disturbances D𝑖 in (18) are
the attitude kinematics equations with error Euler parameter ∗
bounded but unknown. 𝜎𝑛𝑖 is the upper bound of 𝑑𝑛𝑖 . Here, it
are given by can be estimated by
1 󵄨󵄨 󵄨󵄨
Ω𝑏𝑖 = 2T𝑖 𝑒−1 𝜀̇𝑒𝑖 + A𝑒𝑖 𝜔𝑑𝑏𝑖 + A𝑏𝐼𝑖 𝜔𝑂𝑖 , (24) ̂̇𝑛𝑖 =
𝜎 󵄨𝑠 󵄨 , (𝑛 = 1, 2, . . . , 6) , (31)
𝜗𝑛𝑖 󵄨 𝑛𝑖 󵄨
where A𝑒𝑖 is the transformation matrix from the reference
where 𝜗𝑛𝑖 is the adaptive gain.
attitude frame to the body frame.
Utilizing (28), (30), and (31), an adaptive terminal sliding
Differentiating (24), one can obtain that
mode control law is designed as follows:
−1
Ω̇𝑏𝑖 = 2T𝑒𝑖 𝜀̈
𝑒 𝑒
𝑖 + f𝑖 ,
(25) u𝑖𝑇
u𝑖 ≜ [ ] = C𝑖 + N𝑒𝑖
u𝑖𝐴
where

𝜀𝑒 + M𝑒𝑖 (−𝛼𝑖 ė𝑖 − 𝛾𝑖 𝛽𝑖 diag (e𝑖 𝛾𝑖 −1 ) ė𝑖 − diag (̂


𝜎𝑖 ) sign (S𝑖 )) ,
f𝑖𝑒 = f𝑖 + 𝑖𝑒2 (𝜔𝑏𝑖 − A𝑒𝑖 𝜔𝑑𝑏𝑖 ) (𝜔𝑏𝑖 − A𝑒𝑖 𝜔𝑑𝑏𝑖 ) (32)
2𝜂𝑖
(26)
𝑇
̂ 𝑖 ≜ [𝜎̂1𝑖 𝜎̂2𝑖 𝜎̂3𝑖 𝜎̂4𝑖 𝜎̂5𝑖 𝜎̂6𝑖 ] and the vector
where 𝜎
+ A𝑒𝑖 𝜔̇𝑑𝑏𝑖 − 𝜔×𝑏𝑖 A𝑒𝑖 𝜔𝑑𝑏𝑖 signum function sign(S𝑖 ) is a column of signum functions

𝑇
sign (S𝑖 ) ≜ [sign (𝑠1𝑖 ) sign (𝑠2𝑖 ) sign (𝑠3𝑖 ) sign (𝑠4𝑖 ) sign (𝑠5𝑖 ) sign (𝑠6𝑖 )] . (33)

Substituting (32) into (18) produces the closed-loop dynamics Consider a Lyapunov function as follows:

𝛾−1
𝑖 = −𝛼𝑖 ė
ë 𝑖 − 𝛾𝑖 𝛽𝑖 diag (e𝑖 ) ė𝑖 − diag (̂
𝜎𝑖 ) sign (S𝑖 ) + D𝑖 . 1 1 𝑇
𝑉 = S𝑇𝑖 S𝑖 + 𝜎
̃ diag (𝜗𝑖 ) 𝜎
̃𝑖, (35)
(34) 2 2 𝑖
Advances in Mechanical Engineering 5

𝑇
where 𝜗𝑖 ≜ [𝜗1𝑖 𝜗2𝑖 𝜗3𝑖 𝜗4𝑖 𝜗5𝑖 𝜗6𝑖 ] . The estima- It means that the switching planes 𝑠𝑛𝑖 (𝑛 = 1, 2, . . . , 6)
tion error is defined as 𝜎 ̃𝑖 = 𝜎 ̂ 𝑖 − 𝜎∗𝑖 , 𝜎∗𝑖 ≜ converge to zero [20]. On the other hand, in the TSM 𝑠𝑛𝑖 = 0
∗ ∗ ∗ ∗ ∗ ∗ 𝑇
[𝜎1𝑖 , 𝜎2𝑖 , 𝜎3𝑖 , 𝜎4𝑖 , 𝜎5𝑖 , 𝜎6𝑖 ] . (𝑛 = 1, 2, . . . , 6), the system state will reach zero in finite time
[17].
The first derivative of (30) can be expressed as
Remark 1. For the purpose of eliminating chattering, a
Ṡ𝑖 = ë
𝑖 + 𝛼𝑖 ė
𝑖 + 𝛾𝑖 𝛽𝑖 diag (e𝑖
𝛾−1
) ė𝑖 . (36) common practice is to replace the signum function of (32)
with a continuous saturation function

Differentiating (35) and utilizing (36), one can obtain that sat (𝑠𝑛𝑖 , 𝜉𝑛𝑖 )
𝑠𝑛𝑖 󵄨 󵄨
{
{ if 󵄨󵄨󵄨𝑠𝑛𝑖 󵄨󵄨󵄨 < 𝜉𝑛𝑖 (39)
𝑉̇= S𝑇𝑖 Ṡ𝑖 + ̃̇𝑖
̃ 𝑇𝑖 diag (𝜗𝑖 ) 𝜎
𝜎 𝜉𝑛𝑖
={ 𝑛 = 1, 2, . . . , 6,
(37) { 󵄨󵄨 󵄨󵄨
= S𝑇𝑖 (ë
𝑖 + 𝛼𝑖 ė
𝑖 + 𝛾𝑖 𝛽𝑖 diag (e𝑖
𝛾−1
̃ 𝑇𝑖 󵄩󵄩󵄩󵄩S𝑖 󵄩󵄩󵄩󵄩 ,
) ė𝑖 ) + 𝜎 { sign (𝑠𝑛𝑖 ) if 󵄨 𝑠 󵄨
󵄨 𝑛𝑖 󵄨 ≥ 𝜉𝑛𝑖 ,

where 𝜉𝑛𝑖 > 0 is the width of the boundary layer. Equation


𝑇 (32) can be written as
where ‖S𝑖 ‖ ≜ [|𝑠1𝑖 | |𝑠2𝑖 | |𝑠3𝑖 | |𝑠4𝑖 | |𝑠5𝑖 | |𝑠6𝑖 |] . Substituting
(31) and (34) into (37) produces that u𝑖 = C𝑖 + N𝑒𝑖
𝛾 −1
𝑉̇= −S𝑇𝑖 (diag (̂ 𝜎𝑇𝑖 − 𝜎∗𝑇
𝜎𝑖 ) sign (S𝑖 ) − D𝑖 ) + (̂
󵄩󵄩 󵄩󵄩 + M𝑒𝑖 (−𝛼𝑖 ė𝑖 − 𝛾𝑖 𝛽𝑖 diag (e𝑖 𝑖 ) ė𝑖 (40)
𝑖 )󵄩󵄩S𝑖 󵄩󵄩
6 (38) − diag (̂
𝜎𝑖 ) sat (S𝑖 , 𝜉𝑖 ) ) ,
󵄨 󵄨 ∗ 󵄨󵄨 󵄨󵄨
≤ − ∑ 󵄨󵄨󵄨𝑠𝑛𝑖 󵄨󵄨󵄨 (𝜎𝑛𝑖 − 󵄨󵄨𝑑𝑛𝑖 󵄨󵄨) < 0 for 𝑠𝑛𝑖 ≠
0.
𝑛=1 where

𝑇
sat (S𝑖 , 𝜉𝑖 ) = [sat(𝑠1𝑖 , 𝜉1𝑖 ) sat(𝑠2𝑖 , 𝜉2𝑖 ) sat(𝑠3𝑖 , 𝜉3𝑖 ) sat(𝑠4𝑖 , 𝜉4𝑖 ) sat(𝑠5𝑖 , 𝜉5𝑖 ) sat (𝑠6𝑖 , 𝜉6𝑖 )] . (41)

Remark 2. There exists a possible singularity in sliding mode Substituting (5) and (7) into (44) produces
controller as e𝑖 → 0. Since e𝑖 = 0 only approaches along a
sliding mode, we observe that for a general choice of 𝛾𝑖 𝑁−1 𝑚𝑖 𝑚𝑗
∑ 𝑓𝑖𝑗 (𝜇𝑖 , 𝜇𝑗 , r𝑖𝑗 ) = u𝑖𝑇 . (45)
𝑗=0, 𝑗 ≠
𝑖 𝑚𝑖 + 𝑚𝑗
𝛾
ė𝑖 = −𝛼𝑖 e𝑖 − 𝛽𝑖 diag (e𝑖 ) (42)
Note that (7) gives the force on the 𝑖th SMD due to the 𝑗th
while sliding and that the component in (32) SMD and it depends on the distance between the two SMDs
and the orientation of both SMDs in the inertial space. It rises
𝛾−1 𝛾 2𝛾𝑖 −1 to the complexity of the expression for the magnetic. Here, we
𝛾𝑖 𝛽𝑖 diag (e𝑖 ) ė𝑖 󳨀→ −𝛾𝑖 𝛽𝑖 (𝛼𝑖 diag (e𝑖 𝑖 ) + 𝛽𝑖 diag (e𝑖 )) .
consider a two-spacecraft electromagnetic formation flying
(43) (EMFF) array. By defining a rotated frame 𝐹𝑟 (see Section 2),
a simplified algebraic form of (7) is obtained
Consequently there will be a singularity in (32) unless 𝛾𝑖 is
chosen so that 2𝛾𝑖 > 1. To satisfy this requirement we set 𝛾𝑖 = 2𝜇𝑖𝑥 𝜇𝑗𝑥 − 𝜇𝑖𝑦 𝜇𝑗𝑦 − 𝜇𝑖𝑧 𝜇𝑗𝑧
3/5 for the examples to follow. 3𝜇0 [ ]
𝑓𝑖𝑗𝑟 (𝜇𝑖 , 𝜇𝑗 , r𝑖𝑗 ) = [ −𝜇𝑖𝑥 𝜇𝑗𝑦 − 𝜇𝑖𝑦 𝜇𝑗𝑥 ] . (46)
4𝜋𝑟𝑖𝑗4
[ −𝜇𝑖𝑥 𝜇𝑗𝑧 − 𝜇𝑖𝑧 𝜇𝑗𝑥 ]
3.3. Compute the Special Magnetic Moments of SMD. The
adaptive sliding mode controller u𝑖𝑇 gives the desired special Utilizing (45) and (46) produces that
forces that can be used as input for a thruster-based system.
However, for EMFF, the desired special forces are produced 2𝜇𝑖𝑥 𝜇𝑗𝑥 − 𝜇𝑖𝑦 𝜇𝑗𝑦 − 𝜇𝑖𝑧 𝜇𝑗𝑧 4
by SMD. The SMD of each spacecraft is a complicated func- [ −𝜇𝑖𝑥 𝜇𝑗𝑦 − 𝜇𝑖𝑦 𝜇𝑗𝑥 ] 4𝜋𝑟𝑖𝑗 𝑚𝑖 𝑚𝑗
[ ]= A A u ,
tion of current position and orientation [3, 10]. For EMFF, 3𝜇0 𝑚𝑖 + 𝑚𝑗 𝑟𝐼𝑖 𝐼𝑏𝑖 𝑖𝑇
the control variables consist of the dipoles of individual [ −𝜇𝑖𝑥 𝜇𝑗𝑧 − 𝜇𝑖𝑧 𝜇𝑗𝑥 ]
spacecraft. Utilizing (32) and (4), one can obtain that (47)

𝑚0 𝑚1 𝑚0 𝑚1 where A𝑟𝐼𝑖 is the transformation matrix from the inertial


F𝑐𝑚1 = u , F𝑐𝑚0 = − u . (44)
𝑚0 + 𝑚1 𝑖𝑇 𝑚0 + 𝑚1 𝑖𝑇 ECI reference frame 𝐹𝐼 to the rotated frame 𝐹𝑟 . A𝐼𝑏𝑖 is
6 Advances in Mechanical Engineering

Table 1: Orbital parameters.

Spacecraft Semimajor axis Inclination Right ascension of Argument of Eccentricity Mean anomaly
ascending node perigee
LS 6878 km 1.106538745764 rad 0.523598775598 rad 0 rad 0.003 0 rad
FS 6878 km 1.106538745764 rad 0.523598071512 rad 0.000000315261 rad 0.002999636522 0 rad

the transformation matrix from the body frame 𝐹𝐵𝑖 of 𝑖th on a unique nonlinearity of the magnetic torques [3, 4]. The
spacecraft to ECI frame. force acting between a pair of dipoles depends on the product
𝑟 𝑟 𝑟 𝑟 𝑇 4
Let ŭ𝑖𝑇 ≜ [𝑢̆ 𝑖𝑇𝑥 , 𝑢̆
𝑖𝑇𝑦 , 𝑢̆
𝑖𝑇𝑧 ] = (4𝜋𝑟12 /3𝜇0 )(𝑚1 𝑚2 /(𝑚1 +
of the individual dipole values. By switching the polarity of
𝑚2 ))A𝑟𝐼𝑖 A𝐼𝑏𝑖 u𝑖𝑇 , 𝜇𝑖 = −𝜇𝑗 , and utilizing (47), one can obtain all dipoles in the EMFF, the torque acting on the spacecraft
that due to Earth’s magnetic field changes sign, but the torques
and forces due to the other spacecraft in the system do not.
1 As can be seen from (15), it results in a net cancellation
𝜇𝑖𝑥 = √ √𝑢̆
𝑟2 𝑟2
𝑖𝑇𝑥 + 2𝑢̆
𝑟2
𝑖𝑇𝑦 + 2𝑢̆
𝑟2
𝑖𝑇𝑧 − 𝑢̆
𝑖𝑇𝑥 ,
2 of the effect of the Earth’s magnetic field on the average
𝑟2
sense.
𝑢̆
𝑖𝑇𝑦
𝜇𝑖𝑦 = ,
𝑟2
√ √𝑢̆ 𝑟2 𝑟2 𝑟2
𝑖𝑇𝑥 + 2𝑢̆
𝑖𝑇𝑦 + 2𝑢̆
𝑖𝑇𝑧 − 𝑢̆
𝑖𝑇𝑥
(48) 4. Numerical Simulation
𝑟2 The adaptive terminal sliding mode controller equation (40)
𝑢̆
𝑖𝑇𝑧
𝜇𝑖𝑧 = . was simulated for a two-spacecraft formation flying control.
𝑟2
√ √𝑢̆ 𝑟2 𝑟2 𝑟2 Considering the nonlinear dynamics with disturbance D1 ,
𝑖𝑇𝑥 + 2𝑢̆
𝑖𝑇𝑦 + 2𝑢̆
𝑖𝑇𝑧 − 𝑢̆
𝑖𝑇𝑥
the effectiveness of the proposed controller was verified.
The orbital parameters of leader and follower spacecraft
The special magnetic moments of SMD can be computed in the simulations are listed in Table 1. The parameters of
using (48), by which the follower spacecraft can track the spacecraft are summarized in Table 2. The dynamics includes
desired relative position trajectories. the second harmonic of the gravitational field as well as
the Earth’s magnetic field, of which the vector is calculated
3.4. Angular Momentum Management. For EMFF in near- using the World Magnetic (WMM2005)Model block in the
Earth orbits, a constant disturbance torque may act on each paper.
spacecraft due to the Earth’s magnetic field that causes the The follower was commanded to move around the leader
reaction wheels on each spacecraft to quickly become satu- in an elliptic orbit. The desired trajectory was generated by
rated. In order to avoid this situation, the angular momentum solving nonlinear equation (1) numerically (set F̂ ̂
𝑑1 and F𝑐1
management strategy herein can be utilized, which is based equal to zero) with the following initial condition:

𝜌𝑑1 (0) = [2.499998640217, 0.000000000217, 4.317138211336] [m] ,


(49)
𝜌̇𝑑1 (0) = [0.000000000000, −0.000005559083863, −0.000000000758] [m/s] .

The relative position/velocity of the follower was initialized to

𝜌1 (0) = [2.599999999855, 0.722841614669, 4.503332099822] [m] ,


(50)
𝜌̇1 (0) = [−0.000000000000, −0.005557083863, −0.000000000758] [m/s] .

3
The parameters of the adaptive sliding mode controller are 𝛾1 = , 𝜉 = [5, 5, 5, 1, 1, 1] × 10−3 ,
then 5

𝛼1 = diag (2, 2, 3, 250, 80, 80) × 10−4 , 𝜗1 = [15, 22, 22, 80, 125, 125] .
(51)
𝛽1 = diag (10, 15, 15, 1200, 800, 800) × 10−5 ,
Advances in Mechanical Engineering 7

Table 2: Parameter values of model. Error (𝜃1 )


0.2
Spacecraft Mass Moment of inertia

(deg)
0
Ls 100 kg Diag (19, 19, 32) kg⋅m2
FS 100 kg Diag (19, 19, 32) kg⋅m2 −0.2
0 1000 2000 3000 4000 5000 6000
(a)
Relative trajectory rou
Error (𝜃2 )
5 0.2

(deg)
0
z (m)

0 −0.2
∗ 0 1000 2000 3000 4000 5000 6000
(b)
−5
10 Error (𝜃3 )
5 4 0.5
0 2
0

(deg)
y (m −5 −2 ) 0
) −10 −4 x (m
−0.5
Figure 2: Actual trajectory 𝜌1 (𝑡) of the follower spacecraft (∗ the 0 1000 2000 3000 4000 5000 6000
leader spacecraft.) Time (s)
(c)
Error (x-direction)
0.1 Figure 4: Attitude tracking error.
(m)

0.05
0
0 1000 2000 3000 4000 5000
(a) 𝜇x (Am2 )
1000
Error (y-direction) 0
0.8 −1000
0.6 −2000
(m)

0.4 0 1000 2000 3000 4000 5000


0.2
0 (a)
0 1000 2000 3000 4000 5000
𝜇y (Am2 )
(b)
0
Error (z-direction)
−2000
0.2
−4000
(m)

0.1 0 1000 2000 3000 4000 5000


0
(b)
500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500
Time (s) 𝜇z (Am2 )
1000
(c) 0
−1000
−2000
Figure 3: Position tracking error.
0 1000 2000 3000 4000 5000
Time (s)
(c)
In this simulation, a switching time of 100 second was
Figure 5: Magnetic dipole strength in 𝐹𝐵1 frame.
used for AMM. Figures 2–6 show the simulation results,
which are obtained by simulating the adaptive terminal
sliding mode controller. The phase portrait of the trajectory
𝜌1 (𝑡) of the follower spacecraft relative to the leader spacecraft
is illustrated in Figure 3, where ∗ represents the leader
spacecraft at the origin. Figure 3 depicts the position tracking and it is clear from this figure that the angular momentum
error of the follower spacecraft. Figure 4 gives attitude error buildup was limited by AMM. In a stable tracking situation,
of the follower spacecraft. Figure 5 shows the magnetic dipole simulation results show that ‖𝜌𝑒1 ‖ is smaller than 6 × 10−3 m
strength of the follower spacecraft. Figure 6 gives the reaction and ‖𝜀𝑒1 ‖ is smaller than 0.075∘ which meets the required
wheel angular momentum buildup of the follower spacecraft, tracking accuracy.
8 Advances in Mechanical Engineering

hx Conflict of Interests
5
(Nms)

0 The authors declare that there is no conflict of interests


regarding the publication of this paper.
−5
0 1000 2000 3000 4000 5000 6000
(a) Acknowledgment
hy This work was supported by the Natural Science Foundation
2
of China under Grant Nos. 11372353 and 10902125.
(Nms)

−2 References
0 1000 2000 3000 4000 5000 6000
[1] J. Bristow, D. Folta, and K. Hartman, “A formation flying
(b)
technology vision,” in Proceedings of the AIAA Space Conference
hz and Exposition, Long Beach, Calif, USA, September 2000,
5
AIAA-2000-5194.
(Nms)

0 [2] D. Folta and A. Hawkins, “Results of NASA’s first autonomous


formation flying experiment: earth observing-1 (EO-1),” in Pro-
−5
0 1000 2000 3000 4000 5000 6000 ceedings of the AIAA/AAS Astrodynamics Specialist Conference
Time (s) and Exhibit, Monterey, Calif, USA, August 2002, AIAA-2002-
4743.
(c)
[3] U. Ahsun and D. W. Miller, Dynamics and control of electromag-
Figure 6: Reaction wheel angular momentum. netic satellite formations [Ph.D. thesis], Massachusetts Institute
of Technology, 2007.
[4] U. Ahsun, D. W. Miller, and J. L. Ramirez, “Control of electro-
magnetic satellite formations in near-Earth orbits,” Journal of
Guidance, Control, and Dynamics, vol. 33, no. 6, pp. 1883–1891,
5. Conclusions 2010.
[5] L. Johnson, B. Gilchrist, R. D. Estes, and E. Lorenzini, “Over-
In this paper, we consider the control problem for six- view of future NASA tether applications,” Tech. Rep. 1998-
degree-of-freedom electromagnetic spacecraft formation fly- 0237034, NASA, 1998.
ing (EMFF) in near-Earth orbits. Firstly, the combined atti- [6] D. Quinn and D. Folta, “A tethered formation flying concept for
tude and translational dynamics are presented. Secondly, we the SPECS missions,” Tech. Rep. 2000-0032752, NASA, 2000.
confine our attention mainly to the position/attitude tracking [7] L. B. King, G. G. Parker, S. Deshmukh, and J. Chong, “Spacecraft
control using the terminal sliding mode and adaptive control formation-flying using inter-vehicle coulomb forces,” NIAC
technique. Based on a rotated frame 𝐹𝑟 and the adaptive Phase I Final Report, 2002, http://www.niac.usra.edu/.
TSM controller, the special magnetic moments of SMD are [8] J. P. Shoer and M. A. Peck, “A flux-pinned magnet-supe-
computed. The AMM strategy is implemented in a peri- rconductor pair for close-proximity station keeping and self-
odically switching fashion. Through numerical simulation, assembly of spacecraft,” in Proceedings of the AIAA Guidance,
it is clear that the output tracking error can converge to Navigation, and Control Conference and Exhibit, pp. 611–628,
zero in finite time by using the proposed controller and that Hilton Head, SC, USA, August 2007, AIAA Paper 2007-6352.
the angular momentum buildup was limited by AMM. At [9] M. R. LaPointe, “Formation flight method,” NIAC Phase I Final
the same time, strong robustness with respect to bounded Report, 2001, http://www.niac.usra.edu/.
disturbances can be guaranteed (in the simulation, the 𝐽2 dis- [10] D. W. Miller, R. J. Sedwick, D. Jiang, D. Jin et al., “Electro-
turbance, the gravity gradient torques 𝜏𝑔1 , the Earth magnetic magnetic formation flight,” NIAC Phase I Final Report, Mas-
torques 𝜏𝑒1 , and the magnetic torque term 𝜏𝑚1 have been sachusetts Institute of Technology, 2002.
considered). [11] D. W. Miller, R. J. Sedwick, E. M. C. Kong, and S. Schweighart,
“Electromagnetic formation flight for sparse aperture tele-
scopes,” in Proceedings of the IEEE Aerospace Conference, Big
Nomenclature Sky, Mont, USA, March 2002.
[12] L. M. Elias, D. W. Kwon, R. J. Sedwick, and D. W. Miller, “Elec-
EMFF: Electromagnetic spacecraft formation flying tromagnetic formation flight dynamics including reaction
TSM: Terminal sliding mode wheel gyroscopic stiffening effects,” Journal of Guidance, Con-
AMM: Angular momentum management trol, and Dynamics, vol. 30, no. 2, pp. 499–511, 2007.
SFF: Spacecraft formation flying [13] E. M. C. Kong, D. W. Kwon, S. A. Schweighart, L. M. Elias, R.
HTS: High-temperature superconducting J. Sedwick, and D. W. Miller, “Electromagnetic formation flight
TSMC: Terminal sliding mode control for multisatellite arrays,” Journal of Spacecraft and Rockets, vol.
SMD: Steerable magnetic dipole 41, no. 4, pp. 659–666, 2004.
ECI: earth center inertial frame [14] U. Ahsun and D. W. Miller, “Dynamics and control of electro-
TPF: terrestrial planet finder. magnetic satellite formations,” in Proceedings of the American
Advances in Mechanical Engineering 9

Control Conference, pp. 1730–1735, Minneapolis, Minn, USA,


June 2006.
[15] Y. Feng, X. Yu, and Z. Man, “Non-singular terminal sliding
mode control of rigid manipulators,” Automatica, vol. 38, no.
12, pp. 2159–2167, 2002.
[16] X. Yu and Z. Man, “Variable structure systems with terminal
sliding modes,” in Variable Structure Systems: Towards the 21st
Century, vol. 274 of Lecture Notes in Control and Information
Sciences, pp. 109–128, Springer, Berlin, Germany, 2002.
[17] X. Yu, Y. Wu, and Z. Man, “On global stabilization of nonlinear
dynamical systems,” in Variable Structure Systems, Sliding Mode
and Nonlinear Control, vol. 274 of Lecture Notes in Control and
Information Sciences, pp. 109–122, Springer, Berlin, Germany,
1999.
[18] D. J. Stonier and R. J. Stonier, “Obstacle avoidance and finite-
time tracking of mobile targets,” in Proceedings of the 2nd
International Conference on Autonomous Robots and Agents, pp.
58–63, Palmerston North, New Zealand, December 2004.
[19] C.-Q. Yuan, J.-F. Li, T.-S. Wang, and H.-X. Baoyin, “Robust
attitude control for rapid multi-target tracking in spacecraft
formation flying,” Applied Mathematics and Mechanics, vol. 29,
no. 2, pp. 185–198, 2008.
[20] D. R. Merkin, Introduction to the Theory of Stability, Spring, New
York, NY, USA, 1996.
Hindawi Publishing Corporation
Advances in Mechanical Engineering
Volume 2014, Article ID 592628, 9 pages
http://dx.doi.org/10.1155/2014/592628

Research Article
Application of Krylov Reduction Technique for a Machine Tool
Multibody Modelling

M. Sulitka, J. Šindler, J. Suše^, and J. Smolík


Research Center of Manufacturing Technology, Czech Technical University in Prague, Horská 3, 128 00 Prague, Czech Republic

Correspondence should be addressed to M. Sulitka; m.sulitka@rcmt.cvut.cz

Received 26 August 2013; Accepted 22 November 2013; Published 5 February 2014

Academic Editor: Xiaoting Rui

Copyright © 2014 M. Sulitka et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Quick calculation of machine tool dynamic response represents one of the major requirements for machine tool virtual modelling
and virtual machining, aiming at simulating the machining process performance, quality, and precision of a workpiece. Enhanced
time effectiveness in machine tool dynamic simulations may be achieved by employing model order reduction (MOR) techniques
of the full finite element (FE) models. The paper provides a case study aimed at comparison of Krylov subspace base and mode
truncation technique. Application of both of the reduction techniques for creating a machine tool multibody model is evaluated.
The Krylov subspace reduction technique shows high quality in terms of both dynamic properties of the reduced multibody model
and very low time demands at the same time.

1. Introduction dynamics with respect to control remains not fully utilized.


In the case of ball screw driven motion axes, the above
In recent years, with increasing computational power of mentioned issue relates predominantly to active length of the
computers, machine tool virtual modeling has become one ball screw. Next to it, structural modal properties of large
of the intensively developed tools for simulating the machine machine tools may vary significantly also depending on the
tool behavior in interaction with machining process, aiming kinematical configuration and actual position of the motion
especially at predicting the machining performance, process axes.
quality, and stability [1–4]. A key element in machine tool
dynamic behavior represents the compliant machine tool Relevant capturing of the machine tool structural prop-
mechanical system, including both the structural parts and erties represents an important tasks in virtual modeling.
the components of feed drive systems as well. Excitation Coupled modeling techniques employing the reduced FE
of natural eigenfrequencies of the machine tool mechanics models of the machine tool structure have been developed for
stands in the foreground of problems and represents one of describing the ball screw feed drive and machine tool inter-
the limiting factors related to enhancing the machining per- action. Commonly the coupled models rely on one kinematic
formance, since it can lead to either deteriorated machined configuration and application of modal or CMS reduction
surface quality, workpiece precision, or even generation of techniques [6–9]. Importance of proper prediction of higher
instable machining process. frequency range dynamics of feed drive mechanical structure
The modern machine tool, as a typical mechatronic with respect to feed drive control parameters shows [10].
system, features a number of interactions between the control Using the coupled models, machining process stability and
and mechanics [5]. Achievable parameters of both the CNC performance may be predicted by the oriented directional
control and feed drive control as well relate directly to frequency response functions (FRFs) at tool center point
mechanical eigenfrequencies and damping of the mechanical (TCP) [11].
system. Commonly, the setting of the control parameters is For updating of machine tool structural modal prop-
defined by the most compliant configuration of the machine erties, a technique based on continuous redistribution of
tool mechanics, and thus the potential of the structural contact force along the motion axis linear guideway has been
2 Advances in Mechanical Engineering

developed in [12]. Another approach employing an addition- The Krylov subspace methods [26, 29–31] are very
ally created FE node moving between the end positions of the interesting because of their iterative nature which allows
motion axes stroke has been described in [13]. the reduction of large-scale problems. The computational
The finite element method (FEM) is used as a common efficiency of Krylov subspace based MOR has encouraged
way to simulate the dynamic behavior of a machine tool. wide interest in the method, and therefore wide knowledge
The solution of FE problems usually involves the solution in different fields is available. We will mention the most
of a large set of sparse algebraic equations. The number important observations to date. The passivity and stability
of unknowns in ordinary engineering FE model nowadays preservation has been achieved using the Krylov MOR
is commonly 106 -107 . The computational time needed to methods in [26, 32]. A Krylov algorithm preserving structure
solve static analysis of such problems on current hardware is of second order ordinary differential equations has been pre-
acceptably low (∼hours). The issue arises when the harmonic sented in [33]. Handling of nonlinear convection coefficient
or transient analysis is required. The solution then requires was studied in [34]. Reduction of coupled physics problems
10 s or 100 s of iterations and therefore requires days or even was studied in [35] for the case of a thermomechanical model
weeks to solve. Such long solution time effectively hinders of packages and in [36] for the case of structural-acoustic
this kind of simulations in machine tool virtual model coupled models. Krylov subspace MOR was also successfully
simulations. The model order reduction (MOR) methods are used in optimization of MEMS devices [37] and sensitivity
one of possible ways to speed up the solution and make analysis of structural frequency response [38]. One of the
it feasible in practice. The idea behind MOR is to reduce most important directions in development on the Krylov base
the number of unknowns while producing sufficiently good reductions is a parametric model order reduction (PMOR).
approximation to the input/output behavior. The PMOR allows preservation of parameters which the
The following text gives a basic review of model order system depends on [39, 40]. The dependence of parameters
reduction techniques. The idea of reduction of the number may be either linear or nonlinear.
of unknowns during the solution of the FE model is almost The comparison of different model order reduction meth-
as old as the FE method itself. The first MOR method was ods has been discussed in [17], where the Krylov subspace
MOR method was found to be one of the best methods. The
static reduction proposed by Guyan [14] and Irons [15]. This
comparison of Krylov, CMS, and balanced truncation can be
method was introduced for structural mechanics problems
found in [41].
but it is also valid for thermal analysis and other analyses
Based on the findings presented in previous works the
regardless of the underlying physics. However, this method
Krylov subspace based MOR is very robust and computation-
is of questionable quality when using it for dynamic analyses
ally efficient. The goal of this study is to show a new strategy
as was shown in [16, 17].
of employing the Krylov MOR technique for creating the
To remedy the insufficiencies in the static condensation coupled models of compliant systems. The idea is to reduce
method, the component mode synthesis (CMS, [18]) was each part of the machine tool structure separately and to
proposed by Bampton and Craig. The CMS has become produce the coupled model of the whole assembly. This way
widely used by the engineering community. The CMS was enables coupling of reduced components in any kinematic
used to efficiently conduct not only large-scale structural configuration. To assess the quality of Krylov MOR and to
eigenanalysis [19] but also transient heat conduction analysis show its soundness the following comparison properties will
[20] and heat conduction/convection analysis [21]. Another be evaluated:
field of application of the CMS is coupled physics simulations.
The weakly coupled thermomechanical models were studied (1) low error in approximation,
in [22]. There is still active research regarding improvement (2) fast computation.
of the CMS [23]. A low error in the approximation of a full FE model
Another method enhanced to approximate dynamic sys- is required, as the objective is to replace the full FE model
tems well is the improved reduced system (IRS) proposed by in simulations. Fast computations are required to meet the
O’Callahan in [24]. Later, Friswell et al. developed an iterated requirements in virtual machine tools simulations.
version of IRS in [25]. The static condensation, CMS, and The proposed procedure will be tested on industrial scale
IRS can be viewed as engineering approaches to reduce the (106 -107 DOFs) FE model of machine tool. The harmonic
number of equations. and transient simulations on such large models take a lot of
The global error bounds and the preservation of passivity time (days, weeks), and therefore some kind of acceleration is
and stability are important questions posed on the MOR usually used. Mode truncation [42] is frequently used for fast
methods in a more mathematical point of view. Two of simulations and will be compared with Krylov based MOR.
the MOR techniques proposed in accordance with these Both methods will be employed in reduction of individual
questions are Krylov subspace reduction [26] and Balanced parts of machine tool. The reduced part will be then coupled
truncation [26]. Balanced truncation methods [27] have a in any kinematic position. The purpose is to simulate flexible
great advantage because there exists a priori global error multibody systems accurately and very fast (with possible
bound. But it also has a great disadvantage in that the applications in real-time CNC control simulations). The
Lyapunov equation [28] needs to be solved in order to proposed method of coupling the Krylov based reduced
reduce the system. Thus the usage of balanced truncation in substructures has not yet been reported in the literature, to
reduction of large-scale systems is limited. the best the authors’ knowledge.
Advances in Mechanical Engineering 3

The paper is organized in the following way: Section 1 Here 𝑥𝑢 is a vector of unique DOFs. Because (5) indicates that
contains the introduction and the motivation of work; the substructures DOF are obtained from the unique set 𝑥𝑢 ,
approach to substructuring is considered in Section 2; the compatibility equation (3) is satisfied for any set 𝑥𝑢 :
Section 3 contains the description of Krylov subspace reduc-
tion; Section 4 introduces the used machine tool model; and B𝑥 = BL𝑥𝑢 = 0 ∀𝑥𝑢 . (6)
Section 5 presents the comparison of the results obtained This means that L is null space of B and vice versa. And it is
by the methods. Conclusion suggestions on future work are only necessary to define one of the matrices L and B:
given in Section 6.
L = null (B) ,
2. Substructuring (7)
B𝑇 = null (L𝑇 ) .
One way to couple dynamic systems is to connect their mass
M𝑠 , damping C𝑠 , and stiffness K𝑠 matrices; external force Substituting (5) into (1), one gets equation of motion
vectors 𝑓𝑠 , and internal force vectors 𝑔𝑠 by coupling equations ML𝑥𝑢̈ + CL𝑥𝑢̇ + KL𝑥𝑢 = 𝑓 + 𝑔. (8)
[43]; the subscript 𝑠 denotes the systems being coupled. This
approach is similar to the global matrix assembly in FE Further multiplication of (8) from the left-hand side by L𝑇
software. The equations of motion of 𝑛 coupled subsystems yields the final form of the equation of motion of the coupled
can be written as system
M𝑥̈+ C𝑥̇+ K𝑥 = 𝑓 + 𝑔. (1) M ̃𝑥𝑢̇ + K𝑥
̃𝑥𝑢̈ + C ̃
̃ 𝑢 = 𝑓, (9)

The matrices M, C, and K are diagonal matrices containing ̃ C,


where matrices M, ̃ and vector 𝑓̃ are expressed as
̃ and K
the subsystem matrices. Vectors 𝑓 and 𝑔 are column vectors
containing the subsystem internal and external force vectors, ̃ ≜ L𝑇 ML,
M
respectively.
Consider ̃ ≜ L𝑇 CL,
C
(10)
M = diag (M(1) , . . . , M(𝑛) ) , ̃ ≜ L𝑇 KL,
K

C = diag (C(1) , . . . , C(𝑛) ) , 𝑓̃ ≜ L𝑇 𝑓.


Coupling of Reduced Components. Let us consider the case of
K = diag (K(1) , . . . , K(𝑛) ) , (2) 𝑛 independent substructures, each reduced using projection
𝑥(1) 𝑓(1) 𝑔(1) matrix R(𝑖) :
[ .. ] [ .. ] [ .. ]
𝑥 = [ . ], 𝑓 = [ . ], 𝑔 = [ . ]. 𝑥 ≅ R𝜂, R ≜ diag (𝑅(1) ⋅ ⋅ ⋅ 𝑅(𝑛) ) , (11)
(𝑛) (𝑛) (𝑛)
[𝑥 ] [𝑓 ] [𝑔 ] where 𝑥 are physical coordinates of substructures and 𝜂 are
generalized coordinates of substructures. Substituting (11)
The compatibility condition (coupling equations) is ex-
into (1) we get
pressed as
M(𝑠) R𝜂(𝑠)
̈ + CR𝜂(𝑠)
̇ + K(𝑠) R𝜂(𝑠) = 𝑓(𝑠) + 𝑔(𝑠) + 𝑟(𝑠) , (12)
B𝑥 = 0, (3)
where 𝑟(𝑠) are residual forces due to model order reduction.
B is a Boolean if the interface degrees of freedom match 𝑇
Following Galerkin method, we enforce R(𝑠) 𝑟(𝑠) = 0. The
(interfaces are conforming). In this case, the coupling equa- equations of motion in generalized coordinates 𝜂 have the
tions are very simple 𝑥(𝑘) − 𝑥(𝑙) = 0. If the connected degrees form
of freedom do not coincide, the relations among them are
more complex and the matrix B is real. M𝑚 𝜂̈+ C𝑚 𝜂̇+ K𝜂𝑚 = 𝑓𝑚 + 𝑔𝑚 , (13)
Equilibrium condition is given by
where
𝑇
L 𝑔 = 0, (4) M𝑚 ≜ R𝑇 MR,
where L is a Boolean matrix describing the relations among C𝑚 ≜ R𝑇 CR,
the interface forces.
Equations (1), (2), (3), and (4) can be now used to couple K𝑚 ≜ R𝑇 KR, (14)
any number of subsystems. Before it can be done, however,
the unknown interface forces have to be eliminated: 𝑓𝑚 ≜ R𝑇 𝑓,

𝑥 = L𝑥𝑢 . (5) 𝑔𝑚 ≜ R𝑇 𝑔.
4 Advances in Mechanical Engineering

Then following, procedure for coupling full matrices (3)– Full model with n physical DOFs
(10), we get coupled reduced system. This procedure allows
reducing substructures independently and then coupling of
reduced systems. The coupling interface is defined using
matrix B (3), and therefore the specification of the coupling
interface is arbitrary.
Mode superposition

3. Model Order Reduction Methods of Second


Full model with n model DOFs
Order Systems

Modal coordinate reduction


n
The underlying physics of the problem is described by a m1 m2 m3 mn
system of linear second order ordinary differential equations
with constant coefficients k1 k2 k3 ··· kn
c1 c2 c3 cn
M𝑥̈(𝑡) + C𝑥̇(𝑡) + K𝑥 (𝑡) = F𝑢 (𝑡) ,

𝑦 (𝑡) = L𝑇 𝑥 (𝑡) , (15) Modal truncation

𝑥 (0) = 𝑥0 , 𝑥̇(0) = 𝑥1 , Reduced model with m modal DOFs


m
where 𝑥(𝑡) ∈ 𝑅𝑁 is displacement vector of state variables,
𝑢(𝑡) ∈ 𝑅𝑁 is input function, and 𝑦(𝑡) ∈ 𝑅𝑚 is output function. m1 m2 m3 mm
The matrices M, C, and K ∈ 𝑅𝑁×𝑁 are mass matrix, damping
k1 k2 k3 ··· km
matrix and stiffness matrix. F ∈ 𝑅𝑁×𝑙 , and L ∈ 𝑅𝑁×𝑚 are c1 c2 c3 cm
input and output matrices. 𝑥0 , 𝑥1 ∈ 𝑅𝑁 are initial conditions.
The system matrices M, C, and K are constant therefore the
Figure 1: Diagram of modal truncation [42].
system (15) is a linear time invariant system (LTI).

3.1. Mode Truncation. One of the less complex methods of


model order reduction is modal truncation method [42]. independent because the matrices E, C𝑞 , and Λ are diagonal.
The projection matrix T is in this case obtained from modal This means that the whole system of equations can be looked
analysis performed on the full model. The diagram of model at as a sum of many one-degree-of-freedom systems (Mode
reduction by modal truncation can be seen in Figure 1. superposition).
The equation of motion of the full model is The model order is reduced by including only a small
number of mode shapes and degrees of freedom in (19). The
M𝑥̈+ BV 𝑥̇+ K𝑥 = 𝑓, (16)
number of mode shapes included depends on the required
where M is the mass matrix, B is damping matrix, K is accuracy of the reduced model. It is recommended that the
stiffness matrix, 𝑓 is force vector and 𝑥 is the displacement natural frequency of the last included mode shape is at least
vector. twice as high as the highest working frequency of the resulting
The projection matrix T consists of the system’s mode reduced model. The number of included DOFs does not affect
shapes; each column of T corresponds to one mode shape 𝑉𝑖 the model accuracy so only the important ones are included.
and each row represents a degree of freedom: One disadvantage of this method is the fact that the
reduced model’s FRF never fits the full model’s FRF near 0 Hz.
T = [𝑉1 ⋅ ⋅ ⋅ 𝑉𝑛 ] . (17) Another drawback is the relatively long time needed to create
the reduced model (modal analysis with extraction of lots of
The vector 𝑥 can be transformed to vector of modal coordi- modes has to be performed). The method is on the other hand
nates using simple and accurate in chosen frequency range.

𝑥 = T𝑞. (18)
3.2. Krylov MOR. In this section only the basics behind
This equation can be then substituted to (16) and after Krylov reductions will be described. The reader is encouraged
multiplying the result by T𝑇 from the left-hand side, one to read an excellent mathematical description of Krylov based
obtains the equation of motion in the full system in the modal reductions in [26]. An overview of the reduction methods
coordinates of is given in [27]. Although optimal Krylov based reduction
algorithms are available [44], a simpler and a possibly more
E𝑞 ̈+ C𝑞 𝑞 ̇+ Λ𝑞 = T𝑇 𝑓, (19) computationally efficient method will be used in this work—
a block rational Krylov method [45]. The structure of second
where E is identity matrix, C𝑞 is damping ratio matrix, order ODEs in (15) will be preserved using Bai’s algorithm
and Λ is matrix of eigenvalues. The equations in (19) are [33].
Advances in Mechanical Engineering 5

3.2.1. The Treatment of Initial Conditions. Let us consider


using the following coordinate transformation in case of (15):
e
x
𝑥 (𝑡) = 𝑥̃ (𝑡) + 𝑥0 ,

̃̇(𝑡) + 𝑥1 ,
𝑥̇(𝑡) = 𝑥 (20)
q
̃̈(𝑡) .
𝑥̈(𝑡) = 𝑥
V
Substituting (20) into (15),
̃̈(𝑡) + C𝑥
M𝑥 ̃̇(𝑡) + K𝑥̃ (𝑡) = F𝑢 (𝑡) − C𝑥1 − K𝑥0 . (21)
Figure 2: Projection onto K𝑛 .
Right-hand side of (21) is enriched with the constant term
−C𝑥1 − K𝑥0 , and we may consider this term as additional
force. Then we may assume initial conditions (15) without loss where
of generality in form of
M𝑛 = V𝑇𝑛 MV𝑛 ,
𝑥 (0) = 0, 𝑥̇(0) = 0. (22)
This kind of treatment of a nonzero IC was introduced in [46]. K𝑛 = V𝑇𝑛 KV𝑛 ,
(30)
C𝑛 = V𝑇𝑛 CV𝑛 ,
3.2.2. Krylov Subspace Based MOR. The Laplace transform of
(15) has the form of F𝑛 = V𝑇𝑛 F,
−1
H (𝑠) = L𝑇 (𝑠2 M + 𝑠D + K) F. (23) where M𝑛 , Cn , and K𝑛 ∈ 𝑅𝑛×𝑛 , F ∈ 𝑅𝑛 , and L𝑛 ∈ 𝑅𝑛×𝑚 .
The transfer function of the reduced system (30) has the
And the McLaurin series of transfer function (23) has the
form
form
−1
∞ H𝑛 (𝑠) = L𝑇𝑛 (𝑠C𝑛 + K𝑛 ) Q𝑛 . (31)
H (𝑠) = ∑𝑚𝑙 𝑠𝑙 , (24)
𝑖=0 The above procedure assures that the first 𝑛 moments of the
where 𝑚𝑙 are the so-called moments of the transfer function transfer function (23) of the full system equal the first 𝑛
moments of the transfer function (31) of the reduced system
𝑚𝑙 = L𝑇 𝑟𝑙 , [26].
The error induced by the projection (28) in the output
𝑟0 = K−1 F, function 𝑦(𝑡) has the form
(25) 󵄨 󵄨
𝑟1 = −K−1 D𝑟0 , 𝜖 = max 󵄨󵄨󵄨𝑦 (𝑡) − 𝑦̃ (𝑡)󵄨󵄨󵄨 . (32)
𝑡>0

𝑟𝑙 = −K−1 (D𝑟𝑙−1 − M𝑟𝑙−2 ) . An a priori expression for error norm (32) is not known
although there exist algorithms minimizing the error [44, 47].
The first 𝑛 vectors 𝑟𝑙 span Krylov space
The algorithm used in this paper to produce the reduced
K𝑛 = span (𝑟0 , . . . , 𝑟𝑛−1 ) . (26) order systems is the block Arnoldi algorithm [45].
There exist wide possibilities to improve the computa-
Letting V𝑛 be the orthonormal basis of K𝑛 tional performance of Krylov methods. One of the most
obvious options is parallelization [48]. Another is to use an
K𝑛 = span (V𝑛 ) , V𝑇𝑛 V𝑛 = I, V𝑛 ∈ 𝑅𝑁×𝑛 . (27)
iterative algorithm to solve the system [49]. The presented
The projection of state coordinates 𝑥 onto K𝑛 using V𝑛 is case is of medium size, and it is therefore suitable to use the
called generalized state coordinates 𝑞 ∈ 𝑅𝑛 direct sparse solver [50].
The procedure is easily extended to a multi-input/multi-
𝑥 = V𝑞 + 𝜖. (28) output case where Q, F ∈ 𝑅𝑁×𝑙 and L ∈ 𝑅𝑁×𝑚 . The size of
𝑛
The error 𝜖 ∈ 𝑅 in the projection rises while performing the reduced system is determined by the size of Q and L.
projection of 𝑥 onto K𝑛 (Figure 2). However, it is possible to use the superposition property [51]
We obtain reduced system of (29) by substituting gener- to keep the matrices small.
alized coordinates 𝑞 into (15) and using the Galerkin method.
The reduced equations have the form of 4. Machine Tool Model for the Case Study
M𝑛 𝑞 ̈(𝑡) + C𝑛 𝑞 ̇(𝑡) + K𝑛 𝑞 (𝑡) = F𝑛 𝑢 (𝑡) , The dynamic performance of a machine tool may be evalu-
(29) ated at the TCP using the oriented directional FRFs, which
𝑦̃ (𝑡) = L𝑇𝑛 𝑞 (𝑡) , directly relate according to stability theory to achievable
6 Advances in Mechanical Engineering

Table 1: Comparison of computational times.

MOR FRF simulation Total time


Full harmonic — 111 hours 111 hours
Mode truncation 40 min <1 s 40 min
Krylov MOR 6,9 min <1 s 6,9 min

Tool tip the X-slide (face coupling)—or by force interaction in the case
of linear guides (symbols of springs in Figure 4).
Z Z
X Y Y
5. Results
The following case studies compare computational efficiency
Figure 3: Machine tool FE model. of full harmonic in ANSYS software package and model
order reduction using mode truncation and Krylov MOR. To
assess the quality of results obtained using MOR the error
Face coupling norm is evaluated. The study is performed in one kinematic
Face coupling
configuration, and it is assumed that properties of MOR
techniques will be similar in other kinematic configurations.
The model used in the study is a multibody system.

5.1. The Harmonic Simulation of Machine Tool. The frequency


response function at the tip of the tool has been computed
Face coupling using a full FEA model and MOR using mode truncation
Z Face coupling
Y with first 100 modes for each reduced component and Krylov
subspace MOR with first 100 moments. Additionally FRF
computed using modal reduction is provided for comparison.
The comparison is shown in Table 1. The error norm is
Figure 4: Multibody system of the machine tool structure.
concluded to be
󵄨 󵄨
𝜖Mode Truncation = max 󵄨󵄨󵄨𝑦 (𝑓) − 𝑦̃ (𝑓)󵄨󵄨󵄨 > 4𝑒 − 5,
𝑓∈(0,400)
chip thickness [52–54]. FRF can be obtained by harmonic 󵄨 󵄨 (33)
𝜖KRYLOV = max 󵄨󵄨󵄨𝑦 (𝑓) − 𝑦̃ (𝑓)󵄨󵄨󵄨 < 4𝑒 − 6.
simulation of either a full FE model or reduced models. 𝑓∈(0,400)
The purpose of the study is to perform mathematical
verification of the FRFs generated at the TCP using a full FE The Table 1 shows comparison of methods in terms of time
model, or reduced order models. Modal reduction technique needed to produce the reduced order model, time needed to
and Krylov subspace method are considered. At the same computed FRF, and total time. The time needed to produce
time, the study aims at testing the suitability of reduction coupled reduced components is added to the time required
methods for creating a machine tool multibody model, to compute FRF. The reduced models are computed only
assembled from reduced FE models of separate structural once and then it is possible to couple them in any kinematic
parts. In this way, a model allowing for quick update of configuration. The coupling is computationally very cheap.
machine tool dynamic properties according to actual kine- Figures 5 and 6 show FRFs in axes 𝑥 and 𝑦. The results
matic configuration is created. clearly show the quality of approximation obtained using
A model of a large portal milling machine tool is con- Krylov MOR as well as computational efficiency compared
sidered. The FE model (Figure 3) consists of the volume, to the full solution and modal reduction. The Krylov MOR
shell, spring, matrix, and mass elements. The mesh consists of is about 1000x more computationally efficient than the full
about 1,5⋅106 nodes; the total number of degrees of freedom solution, and the approximation error is almost negligible.
number is almost 6⋅106 . The model was built in ANSYS v 14.5. The mode truncation is computationally less efficient and also
Linear guideways are represented by linear spring elements; the approximation error on frequencies higher than 50 Hz is
similarly the rack and pinion feed drives are modeled using very high rendering this method almost unusable.
linear spring elements, which reflect the closed position
control. 6. Conclusions
Figure 4 shows the decomposition of the FE model into
separate structural parts with a description of the type of In the paper a novel application of Krylov subspace based
coupling chosen according to the character of connections model order reduction method for coupling of compliant
between the parts. The parts are connected either through the components, modelled as FE bodies, has been proposed and
faces—typically, for example, connection of the column with successfully tested. To the best of the authors’ knowledge,
Advances in Mechanical Engineering 7

×10−4 axes. Precise capturing of the varying dynamic properties


1.2
will improve the quality of tool oscillation simulation and
1.0 will help increasing the accuracy of machined surface quality
Compliance (mm/N)

and workpiece precision in virtual machining. Low time


0.8 requirements of the Krylov subspace reduction technique
set also a good prerequisite for possible future real-time
0.6 applications of machine tool dynamic virtual model as an
0.4
observer in a real CNC control.

0.2 Conflict of Interests


0 50 100 150 200 250 300 350 400 The authors declare that there is no conflict of interests
Frequency (Hz) regarding the publication of this paper.

Full FE model
Modal reduction Acknowledgment
Krylov MOR
This research has been supported by the Competence Center-
Figure 5: Comparison of FRF, 𝑋-axis. Manufacturing Technology Project TE01020075 funded by
the Technology Agency of the Czech Republic.

×10−5
8 References
7
[1] Y. Altintas, C. Brecher, M. Weck, and S. Witt, “Virtual machine
Compliance (mm/N)

6 tool,” CIRP Annals—Manufacturing Technology, vol. 54, no. 2,


5 pp. 115–138, 2005.
4 [2] Y. Altintas, “Virtual machining system,” in Proceedings of the
4th International Conference and Exhibition on Design and
3 Production of Machines and Dies/Molds, Cesme, Turkey, 2007.
2 [3] M. F. Zaeh, M. Hennauer, and A. Poehler, “The machine tool
1 of tomorrow: an optimised, complete system,” in Proceedings of
the 6th International Conference on High Speed Machining, San
0 50 100 150 200 250 300 350 400 Sebastian, Spain, 2007.
Frequency (Hz) [4] G. Pritschow and N. Croon, “Wege zur virtuellen werkzeugmas-
chine,” Wt Werkstattstechnik, vol. 92, no. 5, pp. 194–199, 2002.
Full FE model
[5] J. Veselý, M. Sulitka, and J. Hornych, “Simulated impact of
Modal reduction
Krylov MOR
machine tool CNC system and feed drive control on workpiece
quality,” in Proceedings of the 2nd International conference
Figure 6: Comparison of FRF, 𝑌-axis. Process Machine Interactions, Vancouver, Canada, 2010.
[6] M. Zatarain, E. Lejardi, and F. Egaña, “Modular synthesis of
machine tools,” CIRP Annals—Manufacturing Technology, vol.
47, no. 1, pp. 333–336, 1998.
coupling of arbitrary substructures reduced using Krylov
[7] P. De Fonseca, D. Vandepitte, H. Van Brussel, and P. Sas,
MOR has not yet been reported in the literature. The “Dynamic model reduction of a flexible three-axis milling
proposed method has been tested on a dynamic simulation machine,” in Proceedings of the International Seminar on Modal
of multibody system of the machine tool structure. Analysis (ISMA ’98), pp. 1197–1206, Leuven, Belgium, Septem-
The case study has shown superior quality of the Krylov ber 1998.
subspace based MOR in comparison to mode truncation [8] R. Maj and G. Bianchi, “Mechatronic analysis of machine tools,”
technique for creating the machine tool multibody model, in Proceedings of the 9th SAMTECH Users Conference, Paris,
in which each of the structural parts is reduced separately France, February 2005.
and coupled to each other using either faced, or force cou- [9] M. Sulitka and P. Strakoš, “Complex model of a real machine
pling. The multibody model employing the Krylov subspace tool feed drive axis with ball screw,” in Proceedings of the
base MOR features almost identical match of the dynamic Conference VIDA, Poznan, Poland, 2007.
properties with the full FE model. At the same time, Krylov [10] J. Vesely and M. Sulitka, “Machine tool virtual model,” in
subspace method is very time efficient and proves to provide Proceedings of the International Conference on Machine Tools,
a very effective tool for quick model updates related to vary- Automation, Robotics and Technology (MATAR ’08), Prague,
ing structural dynamics with respect to various kinematic Czech Republic, 2008.
configurations or testing of different design variants in the [11] J. J. Zulaika, F. J. Campa, and L. N. Lopez de Lacalle, “An
machine tool development. The Krylov subspace will be next integrated processmachine approach for designing productive
used for creating a multibody model with movable motion and lightweight milling machines,” International Journal of
8 Advances in Mechanical Engineering

Machine Tools and Manufacture, vol. 51, no. 7-8, pp. 591–604, [29] E. Grimme, Krylov projection methods for model reduction
2011. [Ph.D. thesis], ECE Department, University of Illinois, Urbana-
[12] M. Zaeh and D. Siedl, “A new method for simulation of machin- Champaign, Ill, USA, 1997.
ing performance by integrating finite element and multi-body [30] E. Grimme and K. Gallivan, “Krylov projection methods
simulation for machine tools,” CIRP Annals—Manufacturing for rational interpolation,” 1997 http://citeseerx.ist.psu.edu/
Technology, vol. 56, no. 1, pp. 383–386, 2007. viewdoc/download?doi=10.1.1.54.5011&rep=rep1&type=pdf.
[13] R. Neugebauer, M. Kolouch, M. Richter, V. Wittstock, and S. [31] Z. Bai, “Krylov subspace techniques for reduced-order mod-
N. Saha, “Model updating of machine tools,” in Proceedings of eling of large-scale dynamical systems,” Applied Numerical
the 9th International Conference on Machine Tools, Automation, Mathematics, vol. 43, no. 1-2, pp. 9–44, 2002.
Robotics and Technology (MATAR ’12), Prague, Czech Republic,
[32] A. Odabasioglu, M. Celik, and L. T. Pileggi, “PRIMA: pas-
September 2012, (MM Science Journal 2012, Special Issue).
sive reduced-order interconnect macromodeling algorithm,”
[14] R. J. Guyan, “Reduction of stiffness and mass matrices,” AIAA in Proceedings of the IEEE/ACM International Conference on
Journal, vol. 3, p. 380, 1965. Computer-Aided Design (ICCAD ’97), pp. 58–65, Washington,
[15] B. M. Irons, “Structural eigenvalue problems: elimination of DC, USA, November 1997.
unwanted variables,” AIAA Journal, vol. 3, no. 5, pp. 961–962,
[33] Z. Bai, K. Meerbergen, and Y. Su, “Arnoldi methods for
1965.
structure-preserving dimension reduction of second-order
[16] L. B. Bushard, “On the value of Guyan reduction in dynamic dynamical systems,” in Dimension Reduction of Large-Scale
thermal problems,” Computers & Structures, vol. 13, no. 4, pp. Systems, P. Benner, D. Sorensen, and V. Mehrmann, Eds., vol.
525–531, 1981. 45 of Lecture Notes in Computational Science and Engineering,
[17] P. Koutsovasilis and M. Beitelschmidt, “Comparison of model pp. 173–189, Springer, Berlin, Germany, 2005.
reduction techniques for large mechanical systems,” Multibody
[34] L. H. Feng, E. B. Rudnyi, J. G. Korvink, C. Bohm, and T. Hauck,
System Dynamics, vol. 20, no. 2, pp. 111–128, 2008.
“Compact electro-thermal model of semiconductor device with
[18] M. Bampton and J. R. Craig, “Coupling of substructures for nonlinear convection coefficient,” in Proceedings of the 6th
dynamic analyses,” AIAA Journal, vol. 6, no. 7, pp. 1313–1319, International Conference on Thermal, Mechanical and Multi-
1968. Physics Simulation and Experiments in Micro-Electronics and
[19] Y. Aoyama and G. Yagawa, “Component mode synthesis for Micro-Systems (EuroSimE ’05), pp. 372–375, Berlin, Germany,
large-scale structural eigenanalysis,” Computers & Structures, April 2005.
vol. 79, no. 6, pp. 605–615, 2001.
[35] E. Zukowski, J. Wilde, E.-B. Rudnyi, and J.-G. Korvink, “Model
[20] D. Botto, S. Zucca, M. Gola, E. Troncarelli, and G. Pasquero, reduction for thermo-mechanical simulation of packages,” in
“Component modes synthesis applied to a thermal transient Proceedings of the 11th International Workshop on Thermal
analysis of a turbine disc,” in Proceedings of the 3rd Worldwide Investigations of ICs and Systems (THERMINIC ’05), pp. 134–
Aerospace Conference & Technology Showcase, Toulouse, France, 138, Belgirate, Italy, 2006.
2002.
[36] R. S. Puri, D. Morrey, A. J. Bell, J. F. Durodola, E. B. Rudnyi, and
[21] D. Botto, S. Zucca, and M. M. Gola, “Reduced-order models
J. G. Korvink, “Reduced order fully coupled structural-acoustic
for the calculation of thermal transients of heat conduc-
analysis via implicit moment matching,” Applied Mathematical
tion/convection FE models,” Journal of Thermal Stresses, vol. 30,
Modelling, vol. 33, no. 11, pp. 4097–4119, 2009.
no. 8, pp. 819–839, 2007.
[37] J. S. Han, E. B. Rudnyi, and J. G. Korvink, “Efficient optimization
[22] P. Nachtergaele, D. J. Rixen, and A. M. Steenhoek, “Efficient
of transient dynamic problems in MEMS devices using model
weakly coupled projection basis for the reduction of thermo-
order reduction,” Journal of Micromechanics and Microengineer-
mechanical models,” Journal of Computational and Applied
ing, vol. 15, no. 4, pp. 822–832, 2005.
Mathematics, vol. 234, no. 7, pp. 2272–2278, 2010.
[23] Z. Bai and B.-S. Liao, “Towards an optimal substructuring [38] J. S. Han, “Efficient frequency response and its direct sensitivity
method for model reduction,” in Applied Parallel Computing. analyses for large-size finite element models using Krylov
State of the Art in Scientific Computing, J. Dongarra, K. Madsen, subspace-based model order reduction,” Journal of Mechanical
and J. Waśniewski, Eds., vol. 3732 of Lecture Notes in Computer Science and Technology, vol. 26, no. 4, pp. 1115–1126, 2012.
Science, pp. 276–285, Springer, Berlin, Germany, 2006. [39] U. Baur, C. Beattie, P. Benner, and S. Gugercin, “Interpolatory
[24] J. O’Callahan, “A procedure for an improved reduced system projection methods for parameterized model reduction,” SIAM
(IRS) model,” in Proceedings of the 7th International Modal Journal on Scientific Computing, vol. 33, no. 5, pp. 2489–2518,
Analysis Conference (IMAC ’89), Las Vegas, Nev, USA, 1989. 2011.
[25] M. I. Friswell, S. D. Garvey, and J. E. T. Penny, “Model reduction [40] J. Šindler, A. Suleng, T. Jelstad Olsen, and P. Bárta, “Krylov
using dynamic and iterated IRS techniques,” Journal of Sound model order reduction of a thermal Subsea model,” World
and Vibration, vol. 186, no. 2, pp. 311–323, 1995. Academy of Science, Engineering and Technology, vol. 77, pp. 217–
[26] A. C. Antoulas, Approximation of Large-Scale Dynamical Sys- 224, 2013.
tems, Advances in Design and Control, SIAM, Philadelphia, Pa, [41] W. Witteveen, “Comparison of CMS, Krylov and balanced trun-
USA, 2005. cation based model reduction from a mechanical application
[27] A. C. Antoulas and D. C. Sorensen, “Approximation of large- engineer’s perspective,” in Topics in Experimental Dynamics
scale dynamical systems: an overview,” International Journal of Substructuring and Wind Turbine Dynamics, R. Mayes, D.
Applied Mathematics and Computer Science, vol. 11, no. 5, pp. Rixen, D. Griffith et al., Eds., vol. 2 of Conference Proceedings
1093–1121, 2001. of the Society for Experimental Mechanics Series, pp. 319–331,
[28] T. Stykel, Analysis and numerical solution of generalized Lya- Springer, New York, NY, USA, 2012.
punov equations [Ph.D. thesis], Technical University of Berlin, [42] Z. Qu, Model Order Reduction Techniques: With Applications in
Berlin, Germany, 2002. Finite Element Analysis, Springer, New York, NY, USA, 2004.
Advances in Mechanical Engineering 9

[43] D. De Klerk, D. J. Rixen, and S. N. Voormeeren, “General


framework for dynamic substructuring: history, review, and
classification of techniques,” AIAA Journal, vol. 46, no. 5, pp.
1169–1181, 2008.
[44] S. Gugercin, A. C. Antoulas, and C. Beattie, “H2 model reduc-
tion for large-scale linear dynamical systems,” SIAM Journal on
Matrix Analysis and Applications, vol. 30, no. 2, pp. 609–638,
2008.
[45] I. M. Elfadel and D. D. Ling, “A block rational Arnoldi algorithm
for multipoint passive model-order reduction of multiport
RLC networks,” in Proceedings of the IEEE/ACM International
Conference on Computer-Aided Design (ICCAD ’97), pp. 66–71,
Washington, DC, USA, November 1997.
[46] L. F. L. Feng, D. Koziol, E. Rudnyi, and J. Korvink, “Model
order reduction for scanning electrochemical microscope: the
treatment of nonzero initial condition,” in Proceedings of IEEE
Sensors, vol. 3, pp. 1236–1239, 2004.
[47] G. Flagg, C. Beattie, and S. Gugercin, “Convergence of the
iterative rational Krylov algorithm,” Systems & Control Letters,
vol. 61, no. 6, pp. 688–691, 2012.
[48] E. F. Yetkin and H. Daǧ, “Parallel implementation of iterative
rational Krylov methods for model order reduction,” in Pro-
ceedings of the 5th International Conference on Soft Computing,
Computing with Words and Perceptions in System Analysis, Deci-
sion and Control (ICSCCW ’09), Famagusta, Cyprus, September
2009.
[49] C. A. Beattie and S. Gugercin, “Inexact solves in Krylov-based
model reduction,” in Proceedings of the 45th IEEE Conference
on Decision and Control (CDC ’06), pp. 3405–3411, San Diego,
Calif, USA, December 2006.
[50] T. Davis, Direct Methods for Sparse Linear Systems, Fundamen-
tals of Algorithms, SIAM, Philadelphia, Pa, USA, 2006.
[51] P. Benner, L. Feng, and E. B. Rudnyi, “Using the superposition
property for model reduction of linear systems with a large
number of inputs,” in Proceedings of the 18th International
Symposium on Mathematical Theory of Networks and Systems
(MTNS ’08), pp. 1–12, Blacksburg, Va, USA, 2008.
[52] Y. Altintas, Manufacturing Automation: Metal Cutting Mechan-
ics, Machine Tool Vibrations, and CNC Design, Cambridge
University Press, New York, NY, USA, 2000.
[53] Y. Altintas and E. Budak, “Analytical prediction of stability lobes
in milling,” CIRP Annals—Manufacturing Technology, vol. 44,
no. 1, pp. 357–362, 1995.
[54] S. D. Merdol and Y. Altintas, “Multi frequency solution of
chatter stability for low immersion milling,” Journal of Manu-
facturing Science and Engineering, vol. 126, no. 3, pp. 459–466,
2004.
Hindawi Publishing Corporation
Advances in Mechanical Engineering
Volume 2014, Article ID 308049, 9 pages
http://dx.doi.org/10.1155/2014/308049

Research Article
Study on Launch Dynamics of Self-Propelled Artillery Based on
Transfer Matrix Method of Multibody System

Hailong Yu and Xiaoting Rui


Institute of Launch Dynamics, Nanjing University of Science and Technology, Nanjing 210094, China

Correspondence should be addressed to Xiaoting Rui; ruixt@163.net

Received 26 August 2013; Accepted 18 December 2013; Published 30 January 2014

Academic Editor: Ch Zhang

Copyright © 2014 H. Yu and X. Rui. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Due to the importance of increasing the firing precision of self-propelled artillery system corresponding to the projectile impact
point, the launch dynamics of self-propelled artillery based on the transfer matrix method for multibody system (MSTMM) is
studied in this paper. By using MSTMM to study launch dynamics of self-propelled artillery, there is no need of global dynamical
equation of the system, the order of the system matrix is low, the computational speed is high, and it meets the requirements
of dynamics of self-propelled artillery multibody system. The dynamical model of multibody system and its topological figure,
the overall transfer equation, overall transfer matrix, characteristics equation and generalized coordinates equation, and launch
dynamical equation of self-propelled artillery are developed. The vibration characteristics, initial disturbance of projectile,
dynamical response, and firing dispersion for self-propelled artillery are simulated. The results of eigenfrequency and the time
history of system dynamics got by simulations and experiments have good agreements. The study results can be used to improve
the firing precision and design of self-propelled artillery.

1. Introduction At present, the methods to study vibration characteristics


of mechanical system are mainly finite element method,
The research results of launch dynamics provided the the- modal analysis method, and the structural modal syn-
oretical basis and technical means for design and test of thesis method. The multibody system dynamics methods,
various weapon systems [1–6]. The movements of projectile such as Wittenburg method, Schiehlen method, and Kane
and artillery are very complicated because of the complex method have been widely accepted and applied by engi-
mechanical structure of self-propelled artillery and the severe neer of mechanical system dynamics [7–9]. Finite element
mechanical environments, such as high temperature, high method and multibody system dynamics method have been
pressure, high speed, instantaneous state, multibody, and an important basis of weapon dynamics. However, when
mutation in launch process. Because of their great influence calculating natural vibration characteristics of weapons for
on firing precision of self-propelled artillery, the vibration complex multirigid-flexible body system containing rigid and
characteristics and dynamical response of the system are paid flexible bodies using these methods, it faces the difficulties
great attention in the study of launch dynamics. To increase that computational scale is large and it is easy to result in com-
firing precision of self-propelled artillery, it is necessary to putation singularity caused by computation ill-condition. To
compute accurately the frequencies distribution and establish meet the requirement of accurate dynamics modeling of
the quantitative relationship among the global structure launch dynamics and fast calculation, Rui et al. presented
parameters of the system and its vibration characteristics, transfer matrix method for multibody systems [2, 10–20].
dynamical response, and firing precision. By adjusting the Due to the important feature that order of the overall transfer
structure parameters to design system vibration frequency matrix is very low, MSTMM avoids eigenvalue computation
distribution, it is then able to improve the firing precision for ill-condition for complex multi-rigid-flexible body system,
self-propelled artillery system. which significantly improves the computational speed of
2 Advances in Mechanical Engineering

Figure 1: Self-propelled artillery.

vibration characteristics and is successfully applied to many Taking self-propelled artillery as an example, according
engineering design and test of various types of vehicles, to its system structure, from bottom to top in sequence, it can
airborne, and ship-borne weapons [1, 2, 4]. be divided into road wheels, hull, and revolving part which
In this paper, by using MSTMM and its automatic does not contain elevating part, elevating part which does
deduction method [2, 20], the dynamical model of multibody not contain recoil part, gun breech, and muzzle brake. Each
system for self-propelled artillery is developed; the overall component can be regarded as rigid body and elastic beam
transfer equation, overall transfer matrix, and characteristics according to its natural attribute. Rigid body and elastic beam
equation are deduced. Generalized coordinates equation of are called “body,” the connection between “body” and “body”
self-propelled artillery is obtained by using the orthogonality is called the “hinge,” which can be seen in [2].
of augmented eigenvector and body dynamical equation. The ground that supports the self-propelled artillery is
Combining with launch dynamical equation of projectile, regarded as an infinity rigid body, whose sequence number
launch dynamical equation of self-propelled artillery system is 0. The elastic and damping effect of each road wheel
is developed. The vibration characteristics, initial disturbance and the interaction between ground and each road wheel
of projectile, dynamical response, and firing dispersion are are, respectively, modeled as springs, rotary springs, and
simulated, and the factors of different ground conditions the accompanying dampers connected in parallel, which can
and connection stiffness among various components, which represent relative linear motion and relative angular motion
influence the firing precision and vibration characteristics, in 3 directions at the same time; their sequence numbers are
are analyzed. It provides a theoretical basis and simulation 42, 43, . . . , 53. Each road wheel can be regarded as a rigid
tool for improving the firing precision by adjusting structural body, whose sequence number is 30, 31, . . . , 41. The inter-
parameters of the system to change the vibration characteris- action between each road wheel and hull are, respectively,
tics of the self-propelled artillery. modeled as springs, rotary springs, and the accompanying
dampers connected in parallel, which can represent relative
2. Dynamical Model of Self-Propelled Artillery linear motion and relative angular motion in 3 directions
Multibody System at the same time; their sequence numbers are 18, 19, . . . , 29.
Hull, revolving part, elevating part, gun breech, and muzzle
Main components of self-propelled artillery are muzzle brake, brake can be regarded as a rigid body, respectively, whose
barrel, gun breech, recoil and counter-recoil mechanisms, sequence numbers are 17, 15, 13, 10, and 1. The effect of
cradle, elevating mechanism, equilibrator, turret, travers- traversing mechanism associating the elastic, and damping
ing mechanism, chassis, torsion bar, balance elbow, shock effects of hull, the effect of elevating mechanism and equi-
absorber, track chain, road wheel, and so on. According to the librator associating the elastic and damping effects between
motion state of each component, the firepower system of self- revolving part and elevating part are, respectively, modeled as
propelled artillery can be divided into recoil part, elevating springs and rotary springs accompanying dampers which can
part, revolving part, suspension part, walking part, and so represent relative linear motion and relative angular motion
on. The recoil part contains muzzle brake, barrel, gun breech, in 3 directions at the same time, whose sequence numbers
and recoil and counter-recoil mechanisms. The elevating are 16, 14. The interaction between barrel and elevating
part contains total recoil part, cradle, and the components par, is, respectively, modeled as springs and rotary springs
moving with cradle, which includes elevating mechanism, accompanying dampers which can represent relative linear
equilibrator, and so on. The revolving part contains total motion and relative angular motion in 3 directions at the
elevating part, turret, and the elements moving with turret same time, whose sequence numbers are 11, 12. According
that include traversing mechanism and so on. The walking to its structure characteristics, the barrel is divided into 6
part is used to support the weight of the self-propelled segments, each segment can be regarded as a beam with
artillery and drive self-propelled artillery to run placidly, equal sectional area, whose sequence numbers are 2, 3, 4, 5, 7,
which contains track chain and road wheels. The suspension and 9. The connection points among joints and beam can be
part is used to connect chassis to walking part, which contains regarded as massless rigid body, whose sequence numbers are
torsion bar, balance elbow, and shock absorber. The self- 6, 8. The dynamical model of self-propelled artillery multi-
propelled artillery is shown in Figure 1. rigid-flexible system is composed of 19 rigid bodies, 6 beams,
Advances in Mechanical Engineering 3

and 28 joints, as shown in Figure 2. The topology figure of where, 1 is the sequence number of body and 0 denotes the
dynamical model of the self-propelled system is shown in boundary.
Figure 3. The form of Z10,0 , Z11,13 , Z13,0 , Z42,0 , Z43,0 , Z44,0 , Z45,0 ,
Z46,0 , Z47,0 , Z48,0 , Z49,0 , Z50,0 , Z51,0 , Z52,0 , and Z53,0 is similar
3. Overall Transfer Equation of to Z1,0 .
Self-Propelled Artillery
3.2. Overall Transfer Equation of Self-Propelled Artillery Sys-
3.1. The State Vector of Self-Propelled Artillery. According to tem. According to MSTMM and the topology figure of
the dynamical model and its topology figure, the state vectors the dynamical model, the overall transfer equation of self-
at boundary points for self-propelled artillery are defined as propelled artillery system is automatically deduced as follows:
follows
𝑇 Uall Zall = 0, (2)
Z1,0 = [𝑋, 𝑌, 𝑍, Θ𝑥 , Θ𝑦 , Θ𝑧 , 𝑀𝑥 , 𝑀𝑦 , 𝑀𝑧 , 𝑄𝑥 , 𝑄𝑦 , 𝑄𝑧 ]1,0 ,
(1) where overall transfer matrix

−I12 T10−1 T42−1 T43−1 T44−1 T45−1 T46−1 T47−1 T48−1 T49−1 T50−1 T51−1 T52−1 T53−1 T13−1 + T11−1
[006×12 06×12 G42−17 G43−17 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 ]
[ 6×12 06×12 G42−17 06×12 G44−17 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 ]
[06×12 06×12 G42−17 06×12 06×12 G45−17 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 ]
[06×12 06×12 G42−17 06×12 06×12 06×12 G46−17 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 ]
[ ]
[06×12 06×12 G42−17 06×12 06×12 06×12 06×12 G47−17 06×12 06×12 06×12 06×12 06×12 06×12 06×12 ]
[06×12 06×12 G42−17 06×12 06×12 06×12 06×12 06×12 G48−17 06×12 06×12 06×12 06×12 06×12 06×12 ]
[ ]
Uall = [06×12 06×12 G42−17 06×12 06×12 06×12 06×12 06×12 06×12 G49−17 06×12 06×12 06×12 06×12 06×12 ].
[06×12 06×12 G42−17 06×12 06×12 06×12 06×12 06×12 06×12 06×12 G50−17 06×12 06×12 06×12 06×12 ]
[06×12 06×12 G42−17 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 G51−17 06×12 06×12 06×12 ]
[0 ]
[ 6×12 06×12 G42−17 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 G52−17 06×12 06×12 ]
[06×12 06×12 G42−17 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 06×12 G53−17 06×12 ]
[0 06×12 G42−13 G43−13 G44−13 G45−13 G46−13 G47−13 G48−13 G49−13 G50−13 G51−13 G52−13 G53−13 G13,𝐼2 ]
[ 6×12 ]
06×12 G10−8 G42−8 G43−8 G44−8 G45−8 G46−8 G47−8 G48−8 G49−8 G50−8 G51−8 G8−52,𝑏12 G53−8 G13−8
[06×12 G10−6 G42−6 G43−6 G44−6 G45−6 G46−6 G47−6 G48−6 G49−6 G50−6 G51−6 G52−6 G53−6 G13−6 + G11−6 ]
(3)

remnant variables after these 0 elements are removed from


The overall state vector of all boundary points is given by Z𝑖,0 (𝑖 = 42, 43, . . . , 53), which contains 6 variables. Z10,0 and
Z1,0 are the state vectors at the boundary points of gun breech
and muzzle. Each of them contains 12 variables, including
displacements, angle displacements, forces, and moments in 3
Zall = [Z1,0 , Z10,0 , Z42,0 , Z43,0 , Z44,0 , Z45,0 , Z46,0 , directions, respectively. Both gun breech and muzzle are free
boundary, therefore, the 6 variables representing the forces
Z47,0 , Z48,0 , Z49,0 , Z50,0 , Z51,0 , Z52,0 , Z53,0 , Z13,0 ] .
and moments are always equal to 0. In the same way, let the
(4) symbol Z10,0 and Z1,0 represent the remnant after these zeros
are removed from Z10,0 and Z1,0 , respectively, and each of
The elements of the overall transfer matrix Uall are deter- them has 6 variables. In this case, there are 84 variables always
mined in appendix. equal to 0 in Zall ; eliminating them from Zall , a new state
vector Zall can be obtained, which has 96 variables, given by
4. Characteristics Equation of Self-Propelled
Artillery System Zall = [Z1,0 , Z10,0 , Z42,0 , Z43,0 , Z44,0 , Z45,0 , Z46,0 , Z47,0 ,
(5)
In (4), Zall is composed of the state vectors Z1,0 , Z10,0 , Z13,0 , Z48,0 , Z49,0 , Z50,0 , Z51,0 , Z52,0 , Z53,0 , Z13,0 ] .
Z42,0 , Z43,0 , Z44,0 , Z45,0 , Z46,0 , Z47,0 , Z48,0 , Z49,0 , Z50,0 , Z51,0 ,
Z52,0 , and Z53,0 at the system boundary points. For the state Deleting the columns 7∼12, 19∼24, 25∼30, 37∼42, 49∼54, 61∼
vectors at the boundary points, half of its elements have 66, 73∼78, 85∼90, 97∼102, 109∼114, 121∼126, 133∼138, 145∼150,
been identified using boundary conditions. For self-propelled 157∼162 in Uall , a new matrix Uall with 96 orders can be
artillery system, each state vector Z𝑖,0 (𝑖 = 42, 43, . . . , 53) obtained; (2) can be written as
between 12 road wheels and the ground contains 12 variables,
including displacements, angle displacements, forces, and Uall Zall = 0. (6)
96×96 96×1
moments in 3 directions respectively, in which the 6 vari-
ables representing the displacements and angle displacements The components of Uall or Uall are only decided by parameters
are always equal to 0; therefore, note symbol Z𝑖,0 (𝑖 = and eigenfrequency of the system. When the system parame-
42, 43, . . . , 53) as the state vector that only includes the ters are determined, (6) should have nonzero solution for any
4 Advances in Mechanical Engineering

Elevating part 13
Projectile
Gun breech 10 14 11
12 Barrel 2∼5 Muzzle brake 1
0

16

Revolving part 15 Hull 17

18 19 20 21 22 23
30 31 32 33 34 35
42 43 44 45 46 47 Road wheel 30∼41

0 0

Figure 2: Dynamical model of the self-propelled artillery multibody system.

Z48.0 Z49.0 Z50.0 Z51.0 Z52.0 Z53.0

48 49 50 51 52 53

36 37 38 39 40 41
9 7 5-2
Z10.0 10 I1 8 I1 6 1 Z1.0
24 25 26 27 28 29 I2 I2
I7 I8 I9 I10 I11 I12 12 11
Z11.13
17 15 I1 13 Z13.0
16 14 I2
I1 I2 I3 I4 I5 I6

18 19 20 21 22 23

30 31 32 33 34 35

42 43 44 45 46 47

Z42.0 Z43.0 Z44.0 Z45.0 Z46.0 Z47.0

Figure 3: Topology figure of the self-propelled system dynamical model.

eigenvalue of the system. Hence, the system characteristics an arbitrary point on beam (barrel) can be obtained. Thus,
equation is obtained by the vibration characteristics of self-propelled artillery are
obtained, which include the eigenfrequency 𝜔𝑘 (𝑘 = 1, 2, . . .)
det (Uall ) = 0. (7) and the mode shape corresponding to each 𝜔𝑘 , and the datum
of mode shape is included in the variables of all state vec-
By solving (7), the eigenfrequency of self-propelled artillery, tors of system, so that the mode shape can be got after the
𝜔𝑘 (𝑘 = 1, 2, . . .), can be obtained. Solving (6), the state vector corresponding variables selected them from the state vec-
𝑘
Zall corresponding to 𝜔𝑘 can be obtained, in other words, Z𝑘all tors.
is obtained. That is to say, the state vectors Z1,0 , Z10,0 , Z13,0 , It can be seen from (6) that the order of the overall
Z42,0 , Z43,0 , Z44,0 , Z45,0 , Z46,0 , Z47,0 , Z48,0 , Z49,0 , Z50,0 , Z51,0 , transfer matrix of the self-propelled artillery system is only
Z52,0 , and Z53,0 corresponding to 𝜔𝑘 are obtained. Based on 96 and is much lower than that for other multibody system
these state vectors at the system boundaries, through using dynamics methods so computational speed is high and the
the transfer equation of each element one by one, all the computational ill-condition caused by high matrix order and
state vectors in system including each connection points and large stiffness gradient is avoided.
Advances in Mechanical Engineering 5

5. Launch Dynamical Equation of 5.3. Launch Dynamical Equation of the Projectile. Launch
Self-Propelled Artillery dynamical equation of a projectile is given by [1, 5]

5.1. Body Dynamical Equation of Self-Propelled Artillery. 𝑃𝑏 𝑆𝑏 𝜕2 𝑥󸀠


Body dynamical equation of self-propelled artillery can be 𝑎𝑝 = − 2
𝑚𝜑3 𝜕𝑡
written as
󸀠
M𝑗 k𝑗,𝑡𝑡 + C𝑗 k𝑗,𝑡 + K𝑗 k𝑗 = f𝑗 ̈ = − 𝑔 cos 𝜃1
𝑦𝑜𝑐
(8)
(𝑗 = 1, 2, 3, 4, 5, 7, 9, 10, 13, 15, 17, 30, 31, . . . , 41) , 𝐾 󸀠 󸀠
𝐹𝑦𝑠𝑓
− [𝑦𝑜𝑜 − 𝜇𝑧𝑜𝑜 sin 𝛼] + − 𝑦𝑜󸀠̈,
𝑚 𝑚
where 𝑗 is the sequence number of all body elements of
the self-propelled artillery and M𝑗 , C𝑗 , K𝑗 , k𝑗 , and f𝑗 are,
respectively, named as parameter matrix of mass, parameter 󸀠 −𝐾 󸀠 󸀠 𝐹𝑠𝑓 󸀠
matrix of damper, parameter matrix of stiffness, displacement 𝑧̈
𝑜𝑐 = [𝑧𝑜𝑜 + 𝜇𝑦𝑜𝑜 sin 𝛼] + 𝑧 − 𝑧̈
𝑜,
𝑚 𝑚
matrix, and external force matrix, which can be seen in [2].
𝐶 𝐶
𝛿1𝐼̈ = − 𝐼
𝛾̇(𝜓̇ 𝐼̇
2 + 𝛿2 ) + (1 − ) (𝛾2̇𝛽𝐷𝜂 + 𝛾𝛽
̈𝐷𝜉 )
5.2. Dynamic Response of Self-Propelled Artillery System. The 𝐴 𝐴
dynamic response in the physical coordinate system may be 𝐶𝛾̈ 𝐼 𝑃𝑏 𝑆𝑏 󸀠
− 𝛿 + 𝑦
expanded using augmented eigenvectors V𝑘1 up to a selected 𝐴 2 𝐴𝜑3 𝑜𝑐
modal order 𝑛,
𝐾ℎ2 𝐼 𝐾 (𝑙𝑅 + 𝑟𝑏 𝜇) 󸀠
+∞
𝑘 − (𝛿 − 𝛿2𝐼 𝜇 sin 𝛼) + 𝑦𝑜𝑜
k𝑗 = ∑ V𝑗 1 𝑞𝑘1 (𝑡) . (9) 12𝐴 1 𝐴
𝑘1 =1
𝐾𝑙𝑅 𝜇 sin 𝛼 󸀠 𝑙
− 𝑧𝑜𝑜 + 1 𝐹𝑦𝑠𝑓 − 𝜓̈
𝐼
1,
Substituting (9) into (8), thus we obtain 𝐴 𝐴
+∞ +∞ 𝐶 𝐶
𝑘
∑ (M𝑗 V𝑗 1 ) 𝑞𝑘̈1 (𝑡) + ∑ (C𝑗 V𝑗 1 ) 𝑞𝑘̇1 (𝑡)
𝑘 𝛿2𝐼̈ = 𝐼
𝛾̇(𝜓̇ 𝐼̇
1 + 𝛿1 ) + (1 − ) (𝛾2̇𝛽𝐷𝜉 − 𝛾𝛽
̈𝐷𝜂 )
𝐴 𝐴
𝑘1 =1 𝑘1 =1
(10) 𝐶𝛾̈ 𝐼 𝑃𝑏 𝑆𝑏 󸀠
+∞ + 𝛿 + 𝑧
𝑘 𝐴 1 𝐴𝜑3 𝑜𝑐
+ ∑ (K𝑗 V𝑗 1 ) 𝑞𝑘1 (𝑡) = f𝑗 .
𝑘1 =1
𝐾ℎ2 𝐼 𝐾 (𝑙𝑅 + 𝑟𝑏 𝜇) 󸀠
− (𝛿2 + 𝛿1𝐼 𝜇 sin 𝛼) + 𝑧𝑜𝑜
Taking the inner product of both sides of (10) with V𝑘𝑗 (𝑘 = 12𝐴 𝐴
1, 2, . . . , 𝑛), thus we obtain 𝐾 𝑙
+ 𝑙 𝜇𝑦󸀠 sin 𝛼 + 1 𝐹𝑧𝑠𝑓 − 𝜓̈
𝐼
2,
∑𝑗 ⟨∑+∞
𝑘
(C𝑗 V𝑗 1 ) 𝑞𝑘̇1 (𝑡) , V𝑘𝑗 ⟩ 𝐴 𝑅 𝑜𝑜 𝐴
𝑘1 =1
𝑞𝑘̈(𝑡) + + 𝜔𝑘2 𝑞𝑘 (𝑡) 2𝑡𝑔𝛼0 𝑘𝛼 2
𝑑𝑘 {
(11) { 𝑑 𝑥𝑞 + 𝑑 𝑥𝑞
{ (𝑥𝑞 < 𝑙𝛼 ) ,
0
𝛾 = { 2𝑡𝑔𝛼 0
∑𝑗 ⟨f𝑗 , V𝑘𝑗 ⟩ {
{ 𝑔 𝑘𝛼 2
= . 𝑥𝑞 − 𝑙𝛼 (𝑥𝑞 ≥ 𝑙𝛼 ) ,
𝑑𝑘 { 𝑑0 𝑑0
For proportional damping, (16)

C𝑗 = 𝛼M𝑗 + 𝛽K𝑗 . (12) where the specific meaning of all symbols is no longer given,
which can be seen in [1, 5].
There are
+∞
𝑘
∑ ⟨ ∑ (C𝑗 V𝑗 1 ) 𝑞𝑘̇1 (𝑡) , V𝑘𝑗 ⟩ = (𝛼 + 𝛽𝜔𝑘2 ) 𝑞𝑘̇(𝑡) 𝑑𝑘 . (13) 6. Numerical Simulation and
𝑗 𝑘1 =1 Experimental Validation
Generalized coordinate equation of self-propelled artillery is Launch dynamics of self-propelled artillery can be computed
obtained, by combining (14) and (16). Eigenfrequency, vertical target
dispersion, the time history of barrel recoil displacement,
𝑞𝑘̈(𝑡) + (𝛼 + 𝛽𝜔𝑘2 ) 𝑞𝑘̇(𝑡) + 𝜔𝑘2 𝑞𝑘 (𝑡) = 𝑝𝑘 (𝑡) , (14)
and the time history of muzzle displacement in the plumb
where direction in cement ground for a self-propelled artillery got
by simulation and test are shown, respectively, in Tables 1 and
𝑘
∑𝑗 ⟨f𝑗 , V𝑘𝑗 ⟩ 2 and Figures 4 and 5. The simulation parameters can be seen
𝑝 (𝑡) =
𝑑𝑘 (15) in [1].
It can be seen from Tables 1 and 2 and Figures 4 and 5
(𝑗 = 1, 2, 3, 4, 5, 7, 9, 10, 13, 15, 17, 30, 31, . . . , 41) . that results got by simulation and test have good agreements,
6 Advances in Mechanical Engineering

Table 1: Eigenfrequencies of self-propelled artillery got by simulation and test (rad/s).

Mode order 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Simulation 3.0 15.3 18.5 29.1 44.0 46.8 64.7 75.8 87.5 94.2 101.5 — 141.0 214.2 225.9 243.7
Test — 15.5 18.7 29.9 42.2 49.8 — — 82.1 — 105.5 128.4 136.3 202.9 229.2 244.4
Error (%) −1.3 −1.0 −2.7 4.3 6.0 6.2 −3.8 3.4 5.6 −1.4 −0.3

0 0.6

Muzzle displacement in plumb direction (m)


−0.2
Barrel recoil displacement (m)

0.4

−0.4
0.2
−0.6

0
−0.8

−1 −0.2
0 0.5 1 1.0 2 0 1 2 3
t (s) t (s)

Test Test
Simulation Simulation

Figure 4: The time history of barrel recoil displacement got by Figure 5: The time history of muzzle displacement in plumb
simulation and test. direction got by simulation and test.

Table 2: Vertical target dispersion of self-propelled artillery got by


simulation and test.
is significantly great and then affects the firing precision of
Vertical target dispersion self-propelled artillery. The simulation results explained this
Vertical dispersion Direction dispersion phenomenon that firing precision of self-propelled artillery is
Test 0.400 0.380 different due to different ground conditions.
Simulation 0.391 0.404 For self-propelled artillery, the vibration characteristics
Error (%) −2.3 6.3 and firing dispersion on the cement ground are simulated by
changing the connection stiffness between hull and revolving
part; eigenfrequencies and firing dispersion got by simulation
which shows that the study on launch dynamics of self- are shown, respectively, in Tables 5 and 6.
propelled artillery by using MSTMM is very effective. For a self-propelled artillery, the vibration characteristic
and firing dispersion on the cement ground are simulated by
changing the connection stiffness between revolving part and
7. Analysis of Factors Influenced Firing elevating part; eigenfrequencies and firing dispersion got by
Precision of Self-Propelled Artillery simulation are shown, respectively, in Tables 7 and 8.
It can be seen from Tables 5–8 that the connection
Many shooting tests have shown that the difference of firing
stiffness between hull and revolving part, revolving part, and
dispersion is big due to different ground conditions and
elevating part has great effect on vibration characteristic and
different connection stiffness among the various components.
firing dispersion of self-propelled artillery. The factors, such
In fact, different ground conditions mean that connection
as different ground hardness, connection stiffness among
stiffness between road wheel and ground is different; the
various components, have a great influence on vibration
different ground hardness can be simulated by changing
characteristics, and then greatly influence firing precision of
connection stiffness between road wheel and ground. The
self-propelled artillery.
vibration characteristics and firing dispersion on the different
ground are simulated for self-propelled artillery. Eigenfre-
quencies and firing dispersion of self-propelled artillery got 8. Conclusion
by simulation are shown, respectively, in Tables 3 and 4.
It can be seen from Tables 3 and 4 that the effect of differ- In this paper, the launch dynamics of self-propelled artillery is
ent ground conditions on the natural vibration characteristics studied using MSTMM and the automatic deduction method
Advances in Mechanical Engineering 7

Table 3: Eigenfrequencies of self-propelled artillery on the different ground got by simulation.

Different ground conditions Natural vibration frequency/(rad/s)


Mode order 1 2 3 4 5 6 7 8
Cement
Eigenfrequencies 3.0 15.3 18.5 29.1 44.0 46.8 64.7 75.8
Mode order 1 2 3 4 5 6 7 8
Soil
Eigenfrequencies 3.0 7.7 9.2 12.0 15.2 21.1 30.2 31.2

Table 4: Firing dispersion of self-propelled artillery on the different T44−1 = U1 U2 U3 U4 U5 U6 U7


ground got by simulation.
× U8,12 U12 U13 U14 U15 U16 U17,20 U20 U32 U44 ,
Vertical target dispersion
Different ground conditions Vertical Direction T45−1 = U1 U2 U3 U4 U5 U6 U7
dispersion/m dispersion/m
× U8,12 U12 U13 U14 U15 U16 U17,21 U21 U33 U45 ,
Cement 0.411 0.367
Soil 0.283 0.393 T46−1 = U1 U2 U3 U4 U5 U6 U7
× U8,12 U12 U13 U14 U15 U16 U17,22 U22 U34 U46 ,

of the overall transfer equation. Dynamical model of self- T47−1 = U1 U2 U3 U4 U5 U6 U7


propelled artillery multibody system and its topology figure × U8,12 U12 U13 U14 U15 U16 U17,23 U23 U35 U47 ,
are developed. According to MSTMM, dynamical model of
self-propelled artillery and its topology figure, overall transfer T48−1 = U1 U2 U3 U4 U5 U6 U7
equation, characteristics equation, and launch dynamical
equations of self-propelled artillery are developed. By solving × U8,12 U12 U13 U14 U15 U16 U17,24 U24 U36 U48 ,
launch dynamical equations of self-propelled artillery, eigen- T49−1 = U1 U2 U3 U4 U5 U6 U7 U8,12
frequency, firing dispersion, and dynamical response of self-
propelled artillery are obtained. Results got by simulation × U8,12 U12 U13 U14 U15 U16 U17,25 U25 U37 U49 ,
and experiment have good agreements, which show that
MSTMM and the automatic deduction method of overall T50−1 = U1 U2 U3 U4 U5 U6 U7
transfer equation are useful to study the launch dynamics of × U8,12 U12 U13 U14 U15 U16 U17,26 U26 U38 U50 ,
self-propelled artillery.
Different ground conditions and connection stiffness T51−1 = U1 U2 U3 U4 U5 U6 U7
among various components greatly influences vibration char-
acteristics and firing precision of self-propelled artillery. × U8,12 U12 U13 U14 U15 U16 U17,27 U27 U39 U51 ,
Decreasing connection stiffness between hull and revolving T52−1 = U1 U2 U3 U4 U5 U6 U7
part, revolving part, and elevating part makes firing disper-
sion of self-propelled change significantly. Firing precision × U8,12 U12 U13 U14 U15 U16 U17,28 U28 U40 U52 ,
of self-propelled artillery can be significantly improved by
reasonable adjustment of the system stiffness or connection T53−1 = U1 U2 U3 U4 U5 U6 U7
stiffness among various components.
× U8,12 U12 U13 U14 U15 U16 U17,29 U29 U41 U53 ,

Appendix T13−1 = U1 U2 U3 U4 U5 U6 U7 U8,12 U12 U13,0 ,


T11−1 = U1 U2 U3 U4 U5 U6,11 U11 C,
Overall Transfer Matrix Elements for
Self-Propelled Artillery G42−17 = −H17 U18 U30 U42 ,
All elements for overall transfer matrix mentioned in (3) are G43−17 = H17,19 U19 U31 U43 ,
expressed as
G44−17 = H17,20 U20 U32 U44 ,

T10−1 = U1 U2 U3 U4 U5 U6 U7 U8 U9 U10 , G45−17 = H17,21 U21 U33 U45 ,

T42−1 = U1 U2 U3 U4 U5 U6 U7 G46−17 = H17,22 U22 U34 U46 ,

× U8,12 U12 U13 U14 U15 U16 U17 U18 U30 U42 , G47−17 = H17,23 U23 U35 U47 ,

T43−1 = U1 U2 U3 U4 U5 U6 U7 G48−17 = H17,24 U24 U36 U48 ,

× U8,12 U12 U13 U14 U15 U16 U17,19 U19 U31 U43 , G49−17 = H17,25 U25 U37 U49 ,
8 Advances in Mechanical Engineering

Table 5: Eigenfrequencies of a self-propelled artillery got by simulation on the cement ground.

Different connection stiffness Eigenfrequencies/(rad/s)


(2.2 × 107 , 2.2 × 107 , 1.1 × 108 ) Mode order 1 2 3 4 5 6 7 8
(5.5 × 107 , 1.3 × 107 , 1.1 × 107 ) Eigenfrequencies 3.0 15.3 18.5 29.1 44.0 46.8 64.7 75.8
(2.2 × 106 , 2.2 × 106 , 1.1 × 107 ) Mode order 1 2 3 4 5 6 7 8
(5.5 × 106 , 1.3 × 106 , 1.1 × 106 ) Eigenfrequencies 1.6 3.0 10.9 19.7 25.6 39.1 45.1 52.0

Table 6: Firing dispersion of a self-propelled artillery got by G46−8 = H8,12 U12 U13 U14 U15 U16 U17,22 U22 U34 U46 ,
simulation on the cement ground.
G47−8 = H8,12 U12 U13 U14 U15 U16 U17,23 U23 U35 U47 ,
Vertical target dispersion
Different connection stiffness Vertical Direction G48−8 = H8,12 U12 U13 U14 U15 U16 U17,24 U24 U36 U48 ,
dispersion/m dispersion/m
(2.2 × 107 , 2.2 × 107 , 1.1 × 108 ) G49−8 = H8,12 U12 U13 U14 U15 U16 U17,25 U25 U37 U49 ,
0.411 0.367
(5.5 × 107 , 1.3 × 107 , 1.1 × 107 )
G50−8 = H8,12 U12 U13 U14 U15 U16 U17,26 U26 U38 U50 ,
(2.2 × 106 , 2.2 × 106 , 1.1 × 107 )
0.543 0.380
(5.5 × 106 , 1.3 × 106 , 1.1 × 106 ) G51−8 = H8,12 U12 U13 U14 U15 U16 U17,27 U27 U39 U51 ,

G52−8 = H8,12 U12 U13 U14 U15 U16 U17,28 U28 U40 U52 ,

G53−8 = H8,12 U12 U13 U14 U15 U16 U17,29 U29 U41 U53 ,
G50−17 = H17,26 U26 U38 U50 ,
G51−17 = H17,27 U27 U39 U51 , G13−8 = H8,12 U12 U13,0 ,

G52−17 = H17,28 U28 U40 U52 , G42−6 = H6 U7 U8,12 U12 U13 U14 U15 U16 U17 U18 U30 U42 ,

G53−17 = H17,29 U29 U41 U53 , G43−6 = H6 U7 U8,12 U12 U13 U14 U15 U16 U17,19 U19 U31 U43 ,

G42−13 = H13 U14 U15 U16 U17 U18 U30 U42 , G44−6 = H6 U7 U8,12 U12 U13 U14 U15 U16 U17,20 U20 U32 U44 ,

G43−13 = H13 U14 U15 U16 U17,19 U19 U31 U43 , G45−6 = H6 U7 U8,12 U12 U13 U14 U15 U16 U17,21 U21 U33 U45 ,
G44−13 = H13 U14 U15 U16 U17,20 U20 U32 U44 , G46−6 = H6 U7 U8,12 U12 U13 U14 U15 U16 U17,22 U22 U34 U46 ,
G45−13 = H13 U14 U15 U16 U17,21 U21 U33 U45 , G47−6 = H6 U7 U8,12 U12 U13 U14 U15 U16 U17,23 U23 U35 U47 ,
G46−13 = H13 U14 U15 U16 U17,22 U22 U34 U46 , G48−6 = H6 U7 U8,12 U12 U13 U14 U15 U16 U17,24 U24 U36 U48 ,
G47−13 = H13 U14 U15 U16 U17,23 U23 U35 U47 ,
G49−6 = H6 U7 U8,12 U12 U13 U14 U15 U16 U17,25 U25 U37 U49 ,
G48−13 = H13 U14 U15 U16 U17,24 U24 U36 U48 ,
G50−6 = H6 U7 U8,12 U12 U13 U14 U15 U16 U17,26 U26 U38 U50 ,
G49−13 = H13 U14 U15 U16 U17,25 U25 U37 U49 ,
G51−6 = H6 U7 U8,12 U12 U13 U14 U15 U16 U17,27 U27 U39 U51 ,
G50−13 = H13 U14 U15 U16 U17,26 U26 U38 U50 ,
G52−6 = H6 U7 U8,12 U12 U13 U14 U15 U16 U17,28 U28 U40 U52 ,
G51−13 = H13 U14 U15 U16 U17,27 U27 U39 U51 ,
G53−6 = H6 U7 U8,12 U12 U13 U14 U15 U16 U17,29 U29 U41 U53 ,
G52−13 = H13 U14 U15 U16 U17,28 U28 U40 U52 ,
G13−6 = H6 U7 U8,12 U12 U13,0 ,
G53−13 = H13 U14 U15 U16 U17,29 U29 U41 U53 ,
G11−6 = −H6,11 U11 C,
G13,I2 = −H13,0 ,
G10−6 = H6 U7 U8 U9 U10 ,
G10−8 = −H8 U9 U10 ,
I O6×6
G42−8 = H8,12 U12 U13 U14 U15 U16 U17 U18 U30 U42 , C=[ 6 ].
O6×6 −I6
G43−8 = H8,12 U12 U13 U14 U15 U16 U17,19 U19 U31 U43 , (A.1)
G44−8 = H8,12 U12 U13 U14 U15 U16 U17,20 U20 U32 U44 ,
All matrices is no longer given in detail, which can be seen in
G45−8 = H8,12 U12 U13 U14 U15 U16 U17,21 U21 U33 U45 , [1, 2, 20].
Advances in Mechanical Engineering 9

Table 7: Eigenfrequencies of self-propelled artillery got by simulation on the cement ground.

Different connection stiffness Natural vibration frequency/(rad/s)


(2.2 × 107 , 2.1 × 107 , 5.3 × 107 ) Mode order 1 2 3 4 5 6 7 8
(5.5 × 107 , 1.1 × 107 , 1.1 × 108 ) Eigenfrequencies 3.0 15.3 18.5 29.1 44.0 46.8 64.7 75.8
(2.2 × 106 , 2.1 × 106 , 5.3 × 106 ) Mode order 1 2 3 4 5 6 7 8
(5.5 × 106 , 1.1 × 106 , 1.1 × 107 ) Eigenfrequencies 2.1 2.9 7.9 16.6 27.9 36.5 42.9 50.1

Table 8: Firing dispersion of self-propelled artillery got by simula- [11] X. Rui and W. Schiehlen, “Multibody system dyanamics: pref-
tion on the cement ground. ace,” Multibody System Dynamics, vol. 18, no. 4, pp. 485–492,
2007.
Vertical target dispersion
[12] X. Rui, L. Yun, G. Wang et al., “Augmented eigenvector and its
Different connection stiffness Vertical Direction orthogonality of linear multi-rigid-flexible-body system,” Jour-
dispersion/m dispersion/m nal of China Ordnance, vol. 4, no. 2, pp. 100–105, 2008.
(2.2 × 107 , 2.1 × 107 , 5.3 × 107 ) [13] X. Rui, B. He, B. Rong et al., “Discrete time transfer matrix
0.411 0.367
(5.5 × 107 , 1.1 × 107 , 1.1 × 108 ) method for dynamics of multi-rigid-flexible-body system mov-
(2.2 × 106 , 2.1 × 106 , 5.3 × 106 ) ing in plane,” Journal of Multi-Body Dynamics, vol. 23, no. 1, pp.
0.458 0.371
(5.5 × 106 , 1.1 × 106 , 1.1 × 107 ) 23–42, 2009.
[14] X. Rui, G. Wang, Y. Lu, and L. Yun, “Transfer matrix method for
linear multibody system,” Multibody System Dynamics, vol. 19,
Conflict of Interests no. 3, pp. 179–207, 2008.
[15] X. Rui, E. Kreuzer, B. Rong, and B. He, “Discrete time transfer
The authors declare that there is no conflict of interests matrix method for dynamics of multibody system with flexible
regarding the publication of this paper. beams moving in space,” Acta Mechanica Sinica, vol. 28, pp.
490–504, 2012.
Acknowledgments [16] X. Rui, B. Rong et al., “Some new developments in discrete time
transfer matrix method of multibody systems,” in Proceedings
The research was supported by the Research Fund for of the International Conference on Dynamics Vibration and
the Doctoral Program of Higher Education of China Control, Hangzhou, China, 2010.
(20113219110025), the Natural Science Foundation of China [17] X. Rui, B. Rong, G. Wang, and B. He, “Discrete time transfer
Government (11102089), and the Program for New Century matrix method for dynamics analysis of complex weapon sys-
Excellent Talents in University (NCET-10-0075). tems,” Science China Technological Sciences, vol. 54, no. 5, pp.
1061–1071, 2011.
[18] X. Rui, B. He, Y. Lu, W. Lu, and G. Wang, “Discrete time transfer
References matrix method for multibody system dynamics,” Multibody
System Dynamics, vol. 14, no. 3-4, pp. 317–344, 2005.
[1] X. Rui, Y. Liu, and H. Yu, Launch Dynamics of Tank and Self-
Propelled Artillery, Science Press, Beijing, China, 2008. [19] B. He, X. Rui, and G. Wang, “Riccati discrete time transfer
matrix method for elastic beam undergoing large overall mo-
[2] X. Rui, L. Yun, Y. Lu et al., Transfer Matrix Method for Multibody
tion,” Multibody System Dynamics, vol. 18, no. 4, pp. 579–598,
System and its Application, Science Press, Beijing, China, 2008.
2007.
[3] X. Rui, L. Yun, G. Wang et al., Direction to Launch Safety of
[20] X. Rui and J. Zhang, “Automatical transfer matrix method of
Ammunition, National Defense Industry Press, Beijing, China,
multibody system,” in Proceedings of the 2nd Joint International
2009.
Conference on Multibody System Dynamics, Stuttgart, Germany,
[4] X. Rui, Y. Lu, G. Wang et al., Simulation and Test Methods of May-June 2012.
Launch Dynamics of Multiple Launch Rocket System, National
Defense Industry Press, Beijing, China, 2003.
[5] X. Rui, Launch Dynamics of Multibody System, National Defense
Industry Press, Beijing, China, 1995.
[6] X. Rui and Q. Yang, Launch Process of Projectile, South East
University Press, Nanjing, China, 1991.
[7] J. Wittenburg, Dynamics of Systems of Rigid Bodies, B. G.
Teubner, Stuttgart, Germany, 1977.
[8] W. Schiehlen, Multibody Systems Handbook, Springer, Berlin,
Germany, 1990.
[9] T. R. Kane, P. W. Likins, and D. A. Levinson, Spacecraft Dynam-
ics, McGraw-Hill, New York, NY, USA, 1983.
[10] X. Rui, D. Bestle, and J. Zhang, “A new form of the trans-
fer matrix method for multibody systems,” in Proceedings of
the ECCOMAS Thematic Conference on Multibody Dynamics,
Zagreb, Croatia, 2013.
Hindawi Publishing Corporation
Advances in Mechanical Engineering
Volume 2014, Article ID 383680, 8 pages
http://dx.doi.org/10.1155/2014/383680

Research Article
Higher Order Variational Integrators for Multibody System
Dynamics with Constraints

Jieyu Ding and Zhenkuan Pan


College of Information Engineering, Qingdao University, Qingdao 266071, China

Correspondence should be addressed to Jieyu Ding; djy@qdu.edu.cn

Received 25 August 2013; Revised 16 November 2013; Accepted 25 November 2013; Published 23 January 2014

Academic Editor: Xiaoting Rui

Copyright © 2014 J. Ding and Z. Pan. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

The continuous and discrete Euler-Lagrangian equations with holonomic constraints are presented based on continuous and
discrete Hamiltonian Principle. Using Lagrangian polynomial to interpolate state variables and Gauss quadrature formula to
approximate Hamiltonian action integral, the higher order variational Galerkin integrators for multibody system dynamics with
constraints and the computation procedure are given. Numerical examples are provided to show the long-time behavior of the
methods proposed in this paper via comparisons with traditional Runge-Kutta methods.

1. Introduction an algorithm that is symplectic and energy preserving at


the same time [6]. The variational integrators [7, 8] based
Dynamics of multibody systems [1] are usually described on discrete variational principle [9] provides an excellent
by a set of differential-algebraic equations (DAEs) [2, 3], framework for design of geometric numerical integrators
which are solved traditionally using numerical methods with many merits: it preserves momentum naturally with
of ordinary differential equations (ODEs) and combining good behavior of approximate energy conservation, even
some constraint stabilization techniques. In the community preserves strictly symplectic-momentum-energy [10]; it can
of numerical analysis, numerical methods for ODEs are be easily extended to a large class of problems [11, 12],
always designed in a small area with local stability analysis, such as the construction of geometric structure-preserving
which lead to a lot of difficulties for long-time simulation. numerical integrators for PDEs [13], nonsmooth collisions
Geometric numerical integration methods [3] originated [14, 15], stochastic systems [16, 17], nonholonomic systems
from long-term simulation of molecular dynamics and planet [18, 19], and constrained systems [20], dissipative systems
dynamics in solar system with global numerical stability due [21], optimal control [22, 23], parameter optimization [24]; it
to the property of structure-preserving provide alternatives can generate easily a large class of higher order methods based
to overcome these problems. These numerical methods are on polynomial interpolation and numerical quadrature for-
referred to as the ones preserving the invariants in their mulas systematically [25–29] and some Lie group variational
corresponding continuous systems, such as preserving sym- integrators [30–32] combining Lie group methods [33].
plectic, energy, momentum, symmetry, and orthogonality, For the general conservative systems without constraints,
in the original systems, thus revolutionizing the traditional [25, 26] propose the same framework of design higher order
concepts of numerical stability. It has been widely accepted variational Galerkin integrators using numerical interpola-
that the more invariants conserved in a numerical method, tion and quadrature techniques for state variables and action
the more stable it is. integral, respectively, during one integration step based on
Symplectic algorithms [4] and energy conservation meth- discrete variational principle. Motivated by these works, we
ods [5] are two frequently used methods for numerical will investigate higher order variational Galerkin integrators
integration of constrained Hamiltonian systems, but they are for multibody system dynamics with constraints for the long-
designed for conservative systems and it is difficult to design term simulation purpose. Due to the space limit, we will
2 Advances in Mechanical Engineering

focus on the cases with holonomic constraints, but the results where 𝐿 𝑑 (q𝑖 , q𝑖+1 , ℎ) and 𝑓𝑑 (q𝑖 , q𝑖+1 , 𝜆𝑖 , 𝑡𝑖 ) are defined as the
can be easily extended to the cases with nonholonomic discrete approximation of the integration of 𝐿(q, q,̇ 𝑡) and
constraints and the cases with nonconservative forces. 𝜆𝑇 Φ(q, 𝑡) on the interval 𝑡 ∈ [𝑡𝑖 , 𝑡𝑖+1 ],
The paper is organized as follows. In Section 2, the 𝑡𝑖+1
continuous and discrete Euler-Lagrangian equations with 𝐿 𝑑 (q𝑖 , q𝑖+1 , 𝑡𝑖 ) = ∫ 𝐿 (q, q,̇ 𝑡) 𝑑𝑡,
𝑡𝑖
holonomic constraints are presented based on continuous (5)
𝑡𝑖+1
and discrete Hamiltonian Principle, respectively; in Section 3, 𝑓𝑑 (q𝑖 , q𝑖+1 , 𝜆𝑖 , 𝑡𝑖 ) = ∫ 𝜆𝑇 Φ (q, 𝑡) 𝑑𝑡.
we derive the higher order variational Galerkin integra- 𝑡𝑖
tors for multibody system dynamics with constraints using
Lagrangian polynomial to interpolate state variables and Using standard variational method, the discrete Euler-
using Gauss quadrature formula to approximate Hamilto- Lagrangian (DEL) equations are derived as
nian action integral; the computation procedure is given in 𝐷1 𝐿 𝑑 (q𝑖 , q𝑖+1 , 𝑡𝑖 ) + 𝐷2 𝐿 𝑑 (q𝑖−1 , q𝑖 , 𝑡𝑖−1 )
Section 4; numerical examples are provided in Section 5 to
show the long-time behavior of the methods proposed in this − ℎΦ𝑇q (q𝑖 , 𝑡𝑖 ) 𝜆𝑖 = 0, (6)
paper via comparisons with traditional methods. The last one Φ (q𝑖+1 , 𝑡𝑖+1 ) = 0,
is concluding remarks including summary and future works.
where 𝐷𝑗 𝐿 𝑑 (𝑗 = 1, 2) is the partial derivative of 𝐿 𝑑 with
2. Discrete Euler-Lagrangian respect to the 𝑗th variable.
Equations with Constraints
3. Higher Order Variational Galerkin
The Hamilton Principle of a multibody system with holo- Integrators of Multibody
nomic constraints and conservative forces can be stated as
Systems with Constraints
𝑡𝑓
𝛿𝑆 = 𝛿 ∫ (𝐿 (q, q,̇ 𝑡) − 𝜆𝑇 Φ (q, 𝑡)) 𝑑𝑡 = 0, (1) The key technique to improve the accuracy of variational
0
integrators is the approximation of the discrete Hamilto-
where 𝑆 is Hamilton action integral, 𝐿 is Lagrangian, q, q̇are nian action integral, which can be realized through general
generalized displacement and velocity, respectively, and 𝜆 is Galerkin methods [24, 25]. Using the same methods in this
Lagrange multiplier vector corresponding to the constraint section, Hamiltonian action integral is approximated in a
subinterval [𝑡𝑖 , 𝑡𝑖+1 ] with an introduced local time parameter
Φ (q, 𝑡) = 0. (2) 𝜏 ∈ [0, 1], where
Via variational method, the following Euler-Lagrange 𝑡 − 𝑡𝑖
𝜏= . (7)
equations of a constrained mechanical system can be got: 𝑡𝑖+1 − 𝑡𝑖
𝑑 𝜕𝐿 (q, q,̇ 𝑡) 𝜕𝐿 (q, q,̇ 𝑡) Given 𝑠 + 1 control points 0 = 𝜏0 < 𝜏1 < ⋅ ⋅ ⋅ < 𝜏𝑠 = 1,
( )− + Φ𝑇 (q, 𝑡) 𝜆 = 0, for 𝑘 = 0, 1, . . . , 𝑠, Lagrange interpolation polynomial 𝑙𝑘 (𝜏) :
𝑑𝑡 𝜕q̇ 𝜕q (3) [0, 1] → 𝑅 is defined as
Φ (q, 𝑡) = 0. 𝜏 − 𝜏𝑗
𝑙𝑘 (𝜏) = ∏ . (8)
It is a typical differential-algebraic equation which can be 𝜏 −𝜏 𝑘 𝑘
0≤𝑗≤𝑠,𝑗 ≠ 𝑗
solved using different traditional numerical methods of DAEs
[2, 3]. The state variables q(𝑡) on [𝑡𝑖 , 𝑡𝑖+1 ] can be interpolated as
Following the variational integrators based on discrete follows:
𝑠
variational principle, the time interval 𝑡 ∈ [0, 𝑡𝑓 ] is divided q𝑑 (𝜏) = ∑ 𝑙𝑘 (𝜏) q𝑖,𝑘 , (9)
into 𝑁 subintervals with time step ℎ = 𝑡𝑓 /𝑁 equally along 𝑘=0
with 𝑁 + 1 grids, 0 = 𝑡0 , 𝑡1 , 𝑡2 , . . . , 𝑡𝑁−1 , 𝑡𝑁 = 𝑡𝑓 , 𝑡𝑖 = 𝑖ℎ,
𝑖 = 0, 1, 2, . . . , 𝑁. The discrete generalized displacement and where q𝑖,𝑘 = q(𝑡𝑖 + 𝑘 ⋅ (ℎ/𝑠)). The derivative of q𝑑 (𝜏) with
velocity and Lagrangian multiplier can be denoted as q𝑖 = ̇ on [𝑡𝑖 , 𝑡𝑖+1 ] as
respect to 𝑡 provides an approximation of q(𝑡)
q(𝑡𝑖 ), q̇𝑖 = q(𝑡
̇ 𝑖 ), 𝜆𝑖 = 𝜆(𝑡𝑖 ), 𝑖 = 0, 1, 2, . . . , 𝑁. Equation (1) can 1 𝑠 󸀠
be rewritten as the following discrete counterpart: q̇𝑑 (𝜏) = ∑ 𝑙 (𝜏) q𝑖,𝑘 , (10)
ℎ 𝑘=0 𝑘
𝛿𝑆𝑑 (q0 , q1 , . . . , q𝑁−1 , q𝑁, 𝜆0 , . . . , 𝜆𝑁−2 , 𝜆𝑁−1 )
where 𝑙𝑘󸀠 (𝜏) = (𝜕/𝜕𝜏)𝑙𝑘 (𝜏). Then, Hamiltonian action integral
𝑁−1 can be approximated on [𝑡𝑖 , 𝑡𝑖+1 ] as follows:
= 𝛿 ∑ 𝐿 𝑑 (q𝑖 , q𝑖+1 , 𝑡𝑖 ) 𝑡𝑖+1
𝑖=0
(4) ∫ (𝐿 (q (𝑡) , q̇(𝑡) , 𝑡) − 𝜆𝑇 Φ (q, 𝑡)) 𝑑𝑡
𝑁−1 𝑡𝑖
1
− 𝛿 ∑ 𝑓𝑑 (q𝑖 , q𝑖+1 , 𝜆𝑖 , 𝑡𝑖 ) (11)
𝑖=0
= ∫ ℎ (𝐿 (q𝑑 (𝜏) , q̇𝑑 (𝜏) , 𝑡𝑖 + 𝜏ℎ)
0

= 0, −𝜆𝑇𝑖 Φ (q𝑑 (𝜏) , 𝑡𝑖 + 𝜏ℎ)) 𝑑𝜏.


Advances in Mechanical Engineering 3

Table 1: Lagrange interpolation polynomials and comparable x2


derivatives with respect to 𝜏. x1
𝑠 𝑙𝑘 (𝜏) 𝑙𝑘󸀠 (𝜏)
𝑙0 (𝜏) = 1 − 𝜏 󸀠
1 𝑙0 (𝜏) = −1
𝑙1 (𝜏) = 𝜏 𝑙1󸀠 (𝜏) = 1
𝑙0 (𝜏) = (2𝜏 − 1) (𝜏 − 1) 𝑙0󸀠 (𝜏) = 4𝜏 − 3 𝜃1
󸀠 y1 l1
2 𝑙1 (𝜏) = −4𝜏 (𝜏 − 1) 𝑙1 (𝜏) = −4 (2𝜏 − 1)
𝑙2 (𝜏) = 𝜏 (2𝜏 − 1) 𝑙2󸀠 (𝜏) = 4𝜏 − 1
y2

Vertical reference line


Table 2: Quadrature points and weights of the Gauss quadrature m1
formula with respect to the interval [−1, 1].
𝜃2 l2
𝑛 𝜏̃𝑟 𝐴𝑟
2 ±√3/3 1 m2
±√15/5 5/9
3
0 8/9

Figure 1: Double pendulum.


Lagrange interpolation polynomials in (9) and comparable
derivatives with respect to 𝜏 in (10) are listed in Table 1.
Using Gauss quadrature formula, which has maximal Step 1. Given the order of Lagrange interpolation polynomial
degree of accuracy 2𝑛 − 1 for a fixed number 𝑛 of quadrature 𝑠, the number of Gauss quadrature points 𝑛, fixed time step ℎ,
points, we have initial state variables q0 , and velocity variables q̇0 .

𝑡𝑖+1 Step 2. Get q1 , q̇1 , and 𝜆1 by Euler method or Runge-Kutta


∫ (𝐿 (q (𝑡) , q̇(𝑡) , 𝑡) − 𝜆𝑇 Φ (q, 𝑡)) 𝑑𝑡 method. Use q0,0 = q0 , and q0,𝑠 = q1 to calculate q0,𝑘 , 𝑘 =
𝑡𝑖
1, . . . , 𝑠 − 1.
𝑛
𝜏𝑟 ) , q̇𝑑 (̃
= ∑𝐴 𝑟 ℎ (𝐿 (q𝑑 (̃ 𝜏𝑟 ) , 𝑡𝑖 + 𝜏̃𝑟 ℎ) Step 3. For 𝑖 = 2 to 𝑁 − 1, solve (13) by Newton iteration
𝑟=1
method to get q𝑖,1 , . . . , q𝑖,𝑠 and 𝜆𝑖 , thus q𝑖+1 = q𝑖,𝑠 , q̇𝑖+1 =
−𝜆𝑇𝑖 Φ (q𝑑 (̃
𝜏𝑟 ) , 𝑡𝑖 + 𝜏̃𝑟 ℎ)) q̇𝑑 (1) = (1/ℎ) ∑𝑠𝑘=0 𝑙𝑘󸀠 (1)q𝑖,𝑘 .
= 𝐿 𝑑 (q𝑖,0 , q𝑖,1 , . . . , q𝑖,𝑠 , 𝑡𝑖 )
Step 4. Use the solution q𝑖 , q̇𝑖+1 , 𝑖 = 0, 1, . . . , 𝑁 − 1 to get
− 𝑓𝑑 (q𝑖,0 , q𝑖,1 , . . . , q𝑖,𝑠 , 𝜆𝑖 , 𝑡𝑖 ) ,
the potential energy 𝑇, kinetic energy 𝑉, total energy of the
(12) system 𝐻 = 𝑇 + 𝑉, and constraints Φ.

where 𝐴 𝑟 is the weights and 𝜏̃𝑟 is the quadrature point.


Quadrature points and weights of the Gauss quadrature 5. Numerical Example
formula with respect to the interval [−1, 1] are listed in Consider a simple double pendulum system shown in
Table 2. Figure 1 the state variables are chosen as q = [𝑥1 𝑦1 𝑥2
Apply discrete Hamilton’s principle, the derivatives of
𝑦2 ]𝑇 , then the potential energy, kinetic energy of the system
the action with respect to q𝑖,𝑘 have to vanish for all 𝑖 =
are
0, 1, . . . , 𝑁 − 1 and 𝑘 = 0, 1, . . . , 𝑠. That leads to the discrete
Euler-Lagrange equations 1 1
𝑇 = 𝑚1 (𝑥1̇2 + 𝑦12̇) + 𝑚2 (𝑥2̇2 + 𝑦22̇) ,
2 2 (14)
𝐷1 𝐿 𝑑 (q𝑖,0 , q𝑖,1 , . . . , q𝑖,𝑠 , 𝑡𝑖 ) − 𝐷1 𝑓𝑑 (q𝑖,0 , q𝑖,1 , . . . , q𝑖,𝑠 , 𝜆𝑖 , 𝑡𝑖 )
𝑉 = 𝑚1 𝑔𝑦1 + 𝑚2 𝑔𝑦2 .
+ 𝐷𝑠+1 𝐿 𝑑 (q𝑖−1,0 , q𝑖−1,1 , . . . , q𝑖−1,𝑠 , 𝑡𝑖−1 )
− 𝐷𝑠+1 𝑓𝑑 (q𝑖−1,0 , q𝑖−1,1 , . . . , q𝑖−1,𝑠 , 𝜆𝑖−1 , 𝑡𝑖−1 ) = 0 The mass matrix is a constant matrix as follows:
𝐷𝑘 𝐿 𝑑 (q𝑖,0 , q𝑖,1 , . . . , q𝑖,𝑠 , 𝑡𝑖 ) M = diag [𝑚1 𝑚1 𝑚2 𝑚2 ] . (15)
− 𝐷𝑘 𝑓𝑑 (q𝑖,0 , q𝑖,1 , . . . , q𝑖,𝑠 , 𝜆𝑖 , 𝑡𝑖 ) = 0, 𝑘 = 1, . . . , 𝑠
The constraint equations are
Φ (q𝑖+1 , 𝑡𝑖+1 ) = 0.
(13) 𝑥12 + 𝑦12 − 𝑙12
Φ (q) = ( 2 2 ) = 0. (16)
(𝑥2 − 𝑥1 ) + (𝑦2 − 𝑦1 ) − 𝑙22
4. Computation Procedure
The following computation procedure can be used to get the Given 𝑚1 = 𝑚2 = 1 kg, 𝑙1 = 𝑙2 = 1 m, with time step ℎ =
solution of the discrete Euler-Lagrange equations (13). 0.01 s, the results of the method presented in this paper are
4 Advances in Mechanical Engineering

Table 3: Comparison of the methods mentioned with time step ℎ = 0.01.

Numerical method Run time (s) max|𝜀(𝐻)| max|𝜀(Φ)|


Runge-Kutta 15.0385 3.0045 0.2345
P1L-Q2G 16.0993 0.4049 1.9984 × 10−15
P2L-Q2G 24.8510 0.2270 8.8818 × 10−16
P2L-Q3G 31.2938 0.2273 1.3323 × 10−15

Table 4: Comparison of the methods mentioned with time step ℎ = 0.001.

Numerical method Run time (s) max|𝜀(𝐻)| max|𝜀(Φ)|


−4
Runge-Kutta 99.7938 4.7378 × 10 3.6633 × 10−5
P1L-Q2G 121.4468 0.0405 9.6040 × 10−11
P2L-Q2G 212.9882 0.0023 1.2912 × 10−9
P2L-Q3G 265.9505 0.0021 1.2912 × 10−9

Runge-Kutta P1L-Q2G

3 3
Total energies (N·m)

Total energies (N·m)

2 2

1 1

0 0

0 50 100 0 50 100
Time (s) Time (s)
(a) (b)
P2L-Q2G P2L-Q3G

3 3
Total energies (N·m)

Total energies (N·m)

2 2

1 1

0 0

0 50 100 0 50 100
Time (s) Time (s)
(c) (d)

Figure 2: Total energy of double pendulum by different method, ℎ = 0.01.


Advances in Mechanical Engineering 5

y
xj

yj A
xi i
x

B
yi

Oj

Oi

𝜃
O x

Figure 3: A rotary rod slider system.

P1L-Q2G P1L-Q2G
30 3

20 2

10 1
Total energies (N·m)
Energies (N·m)

0 0

−10 −1

−20 −2

−30 −3
0 20 40 60 80 100 0 20 40 60 80 100
Time (s) Time (s)
Kinetic energy
Potential energy
Total energy
(a) (b)

Figure 4: Energies of rotary rod slider system by method P1L-Q2G, ℎ = 0.01.

compared with the Runge-Kutta But using the discrete variational methods presented in this
method in Table 3 for the ter- paper, the time step can be chosen longer as 0.01 s. Table 3 sh-
minal time 100 s. Here, P1L-Q2G means that the order of Lag- ows that, with this longer time step, the discrete variational
range interpolation polynomial is 1 and the number of Gauss methods can keep the total energy and constraints in higher
quadrature points is 2, and the same as P2L-Q2G and P2L- accuracy than Runge-Kutta method, the cost instead is the
Q3G. 𝜀(Η), 𝜀(Φ) are the maximum errors of the total energy longer computation time because of the iterations during the
𝐻 and constraints Φ. nonlinear equations solving process, but it is still shorter than
Using the traditional method such as Runge-Kutta met- the computation time of Runge-Kutta method with smaller
hod, to keep the errors of the total energy 𝐻 and constraints Φ time step to reach the same accuracy of the results.
are all smaller, the time step should be chosen as about 0.001 s It is also shown in Tables 3 and 4 that when the order of
in this example, and the total computation time is 99.7938 s. Lagrange interpolation polynomial is higher, the total energy
6 Advances in Mechanical Engineering

×10−16 ×10−16
4 4

2 2
Constraints 1

Constraints 2
0 0

−2 −2

−4 −4
0 50 100 0 50 100
Time (s) Time (s)
(a) (b)
−16 −16
×10 ×10
4 4

2 2
Constraints 3

Constraints 4

0 0

−2 −2

−4 −4
0 50 100 0 50 100
Time (s) Time (s)
(c) (d)
Figure 5: Constraints of rotary rod slider system by method P1L-Q2G, ℎ = 0.01.

and constraints can be kept better, but there are no big In the case of nonconstant mass matrix, the methods of
differences between 2 or 3 points in Gauss quadrature higher order variational Galerkin integrators are also applica-
formula. ble. For another example, Figure 3 shows a rotary rod slider
Figure 2 shows the energies of the previously mentioned system. 𝑂𝐴 is a rigid rod with uniform mass 𝑚𝐴 and length 𝑙,
methods. With the time step ℎ = 0.01 s, the total energy is which rotates round 𝑂 in the plane 𝑂𝑋𝑌. 𝐵 is a slider with the
up and down around zero with little errors during the long-ti- mass 𝑚𝐵 , and the stiffness of the spring on it is 𝑘; the mass of
me simulation by the methods presented in this paper, wh- the spring is ignored. Suppose only gravity in the plane 𝑂𝑋𝑌
ile increases quickly during the simulation by Runge-Kutta is considered for the system.
method. The results of high order variational Galerkin inte- The state variables are chosen as q = [𝑅𝑥𝑖 𝑅𝑦𝑖 𝑅𝑥𝑗
grators are better than low order integrators. 𝑅𝑦𝑗 𝜃 𝑥𝑖 ]𝑇 . Then, the mass matrix is a nonconstant matrix

𝑚𝐵 0 0 0 −𝑚𝐵 𝑥𝑖 sin 𝜃 𝑚𝐵 cos 𝜃


[ 0 𝑚 0 0 𝑚𝐵 𝑥𝑖 cos 𝜃 𝑚𝐵 sin 𝜃 ]
[ 𝐵 ]
[ 0 0 𝑚𝐴 0 0 0 ]
[ ]
M=[ 0 0 0 𝑚𝐴 0 0 ]. (17)
[ 1 ]
[ ]
[−𝑚𝐵 𝑥𝑖 sin 𝜃 𝑚𝐵 𝑥𝑖 cos 𝜃 0 0 𝑚𝐴 𝑙2 + 𝑚𝐵 𝑥𝑖2 0 ]
3
[ 𝑚𝐵 cos 𝜃 𝑚𝐵 sin 𝜃 0 0 0 𝑚𝐵 ]
Advances in Mechanical Engineering 7

The constraint equations are [6] Z. Ge and J. E. Marsden, “Lie-Poisson Hamilton-Jacobi theory
and Lie-Poisson integrators,” Physics Letters A, vol. 133, no. 3, pp.
𝑙 134–139, 1988.
𝑅𝑥𝑗 − cos 𝜃
2 [7] J. M. Wendlandt and J. E. Marsden, “Mechanical integrators der-
( 𝑙 ) ived from a discrete variational principle,” Physica D, vol. 106,
𝑅𝑦𝑗 − sin 𝜃
Φ=(
( 2 ) = 0.
)
no. 3-4, pp. 223–246, 1997.
[8] J. E. Marsden and M. West, “Discrete Mechanics and Variational
(𝑅𝑥𝑗 − 𝑅𝑥𝑖 ) cos 𝜃 + (𝑅𝑦𝑗 − 𝑅𝑦𝑖 ) sin 𝜃 − 𝑥𝑖 Integrators,” Acta Numerica, vol. 10, pp. 357–514, 2001.
( (𝑅𝑥𝑗 − 𝑅𝑥𝑖 ) sin 𝜃 − (𝑅𝑦𝑗 − 𝑅𝑦𝑖 ) cos 𝜃 ) [9] J. Moser and A. P. Veselov, “Discrete versions of some classical
(18) integrable systems and factorization of matrix polynomials,”
Communications in Mathematical Physics, vol. 139, no. 2, pp.
Given 𝑙1 = 2𝑚, 𝑚𝐴 =2 kg, 𝑚𝐵 = 5 kg, 𝑘 = 1000 N/m. 217–243, 1991.
With the initial step length ℎ = 0.01 s and the terminal [10] C. Kane, J. E. Marsden, and M. Ortiz, “Symplectic-energy-mo-
time 100 s, using the method P1L-Q2G, where the order of mentum preserving variational integrators,” Journal of Mathe-
matical Physics, vol. 40, no. 7, pp. 3353–3371, 1999.
Lagrange interpolation polynomial is 1 and the number of
Gauss quadrature points is 2, the results are also satisfying. [11] A. Lew, J. E. Marsden, M. Ortiz, and M. West, “Variational time
integrators,” International Journal for Numerical Methods in
The total computer time is 86.0034 s; the maximum errors of
Engineering, vol. 60, no. 1, pp. 153–212, 2004.
the total energy 𝐻 and constraints Φ are 0.2611 and 3.8858 ×
[12] A. Lew, J. E. Marsden, M. Ortiz, and M. West, “An overview of
10−16 , which is illustrated in Figures 4 and 5. variational integrators,” in Finite Element Methods: 1970s and
Beyond, pp. 85–146, CIMNE, Barcelona, Spain, 2003.
6. Conclusions [13] A. Lew, J. E. Marsden, M. Ortiz, and M. West, “Asynchronous
variational integrators,” Archive for Rational Mechanics and
Based on Lagrange interpolation and Gauss quadrature, hig- Analysis, vol. 167, no. 2, pp. 85–146, 2003.
her order variational Galerkin integrators have been devel- [14] A. Pandolfi, C. Kane, J. E. Marsden, and M. Ortiz, “Time-dis-
oped for multibody systems with constraints. Numerical res- cretized variational formulation of non-smooth frictional con-
ults validate the accuracy of the method for longer time step tact,” International Journal for Numerical Methods in Engineer-
and long-time simulation. Nevertheless, as the computation ing, vol. 53, no. 8, pp. 1801–1829, 2002.
is more complex, the computation cost is higher. Further [15] R. C. Fetecau, J. E. Marsden, M. Ortiz, and M. West, “Nonsmo-
study is needed to have more efficient integrators. oth lagrangian mechanics and variational collision integrators,”
SIAM Journal on Applied Dynamical Systems, vol. 2, no. 3, pp.
381–416, 2003.
Conflict of Interests [16] N. Bou-Rabee and H. Owhadi, “Stochastic variational integra-
tors,” IMA Journal of Numerical Analysis, vol. 29, no. 2, pp. 421–
The authors declare that there is no conflict of interests regar-
443, 2009.
ding the publication of this paper.
[17] W. Hu, Z. Deng et al., “Generalized multi-symplectic integrators
for a class of Hamiltonian nonlinear wave PDEs,” Journal of
Acknowledgments Computational Physics, vol. 235, pp. 394–406, 2013.
[18] J. Cortés and S. Martı́nez, “Non-holonomic integrators,” Nonlin-
This research was supported by the National Natural Sci- earity, vol. 14, no. 5, pp. 1365–1392, 2001.
ence Foundation of China (Grants nos. 11272166, 11002075, [19] R. McLachlan and M. Perlmutter, “Integrators for nonholonom-
10972110). ic mechanical systems,” Journal of Nonlinear Science, vol. 16, no.
4, pp. 283–328, 2006.
References [20] S. Leyendecker, J. E. Marsden, and M. Ortiz, “Variational integr-
ators for constrained dynamical systems,” ZAMM Zeitschrift für
[1] A. A. Shabana, Dynamics of Multibody Systems, Cambridge Uni- Angewandte Mathematik und Mechanik, vol. 88, no. 9, pp. 677–
versity Press, 3rd edition, 2005. 708, 2008.
[2] A. Laulusa and O. A. Bauchau, “Review of classical approaches [21] C. Kane, J. E. Marsden, M. Ortiz, and M. West, “Variational
for constraint enforcement in multibody systems,” Journal of integrators and the newmark algorithm for conservative and
Computational and Nonlinear Dynamics, vol. 3, no. 1, Article ID dissipative mechanical systems,” International Journal for Num-
011004, 2008. erical Methods in Engineering, vol. 49, no. 10, pp. 1295–1325,
[3] O. A. Bauchau and A. Laulusa, “Review of contemporary appr- 2000.
oaches for constraint enforcement in multibody systems,” Jour- [22] S. Leyendecker, S. Ober-Blöbaum, J. E. Marsden, and M. Ortiz,
nal of Computational and Nonlinear Dynamics, vol. 3, no. 1, “Discrete mechanics and optimal control for constrained sys-
Article ID 011005, 2008. tems,” Optimal Control Applications and Methods, vol. 31, no. 6,
[4] L. Jay, “Symplectic partitioned Runge-Kutta methods for con- pp. 505–528, 2010.
strained Hamiltonian systems,” SIAM Journal on Numerical [23] S. Ober-Blöbaum, O. Junge, and J. E. Marsden, “Discrete me-
Analysis, vol. 33, no. 1, pp. 368–387, 1996. chanics and optimal control: an analysis,” Control, Optimisation
[5] O. Gonzalez, “Mechanical systems subject to holonomic con- and Calculus of Variations, vol. 17, no. 2, pp. 322–352, 2011.
straints: differential-algebraic formulations and conservative [24] K. Witting, S. Ober-Blöbaum, and M. Dellnitz, “A variational
integration,” Physica D, vol. 132, no. 1-2, pp. 165–174, 1999. approach to define robustness for parametric multi-objective
8 Advances in Mechanical Engineering

optimization problems,” Journal of Global Optimization, vol. 57,


no. 2, pp. 331–345, 2013.
[25] M. Leok and J. Zhang, “Discrete Hamiltonian variational int-
egrators,” IMA Journal of Numerical Analysis, vol. 31, no. 4, pp.
1497–1532, 2011.
[26] S. Ober-Blöbaum and N. Saake, “Construction and analysis of
higher order Galerkin variational integrators,” Numerical Anal-
ysis. In press. http://arxiv.org/abs/1304.1398 .
[27] M. Leok, T. Shingel, and T. P, “rolongation-collocation variatio-
nal integrators,” IMA Journal of Numerical Analysis, vol. 32, no.
3, pp. 1194–1216, 2012.
[28] S. Leyendecker and S. Ober-Blöbaum, “A variational approach
to multirate integration for constrained systems,” in Multibody
Dynamics, J.-C. Samin and P. Fisette, Eds., vol. 28 of Computa-
tional Methods in Applied Sciences, pp. 97–121, Springer, 2013.
[29] C. M. Campos, O. Junge, and S. Ober-Blöbaum, “Higher order
variational time discretization of optimal control problems,” in
Proceedings of the 20th International Symposium on Mathemat-
ical Theory of Networks and Systems,, pp. 9–13, Melbourne,
Australia, July 2012.
[30] M. Leok and T. Shingel, “General techniques for constructing
variational integrators,” Frontiers of Mathematics in China, vol.
7, no. 2, pp. 273–303, 2012.
[31] N. Bou-Rabee and J. E. Marsden, “Hamilton-Pontryagin inte-
grators on lie groups part I: introduction and structure-
preserving properties,” Foundations of Computational Mathe-
matics, vol. 9, no. 2, pp. 197–219, 2009.
[32] T. Lee, M. Leok, and N. H. McClamroch, “Lie group variational
integrators for the full body problem,” Computer Methods in
Applied Mechanics and Engineering, vol. 196, no. 29-30, pp.
2907–2924, 2007.
[33] A. Iserles, H. Z. Munthe-Kaas, S. P. Nørsett, and A. Zanna, “Lie-
Group methods,” Acta Numerica, vol. 9, pp. 215–365, 2000.
Hindawi Publishing Corporation
Advances in Mechanical Engineering
Volume 2014, Article ID 792478, 16 pages
http://dx.doi.org/10.1155/2014/792478

Research Article
Free Vibration Characteristic of Multilevel Beam Based on
Transfer Matrix Method of Linear Multibody Systems

Laith K. Abbas and Xiaoting Rui


Institute of Launch Dynamics, Nanjing University of Science and Technology, Nanjing 210094, China

Correspondence should be addressed to Laith K. Abbas; laithabbass@yahoo.com

Received 18 August 2013; Accepted 28 November 2013; Published 20 January 2014

Academic Editor: Caishan Liu

Copyright © 2014 L. K. Abbas and X. Rui. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.

In this paper, an approach based on transfer matrix method of linear multibody systems (MS-TMM) is developed to analyze the free
vibration of a multilevel beam, coupled by spring/dashpot systems attached to them in-span. The Euler-Bernoulli model is used for
the transverse vibration of the beams, and the spring/dashpot system represents a simplified model of a viscoelastic material. MS-
TMM reduces the dynamic problem to an overall transfer equation which only involves boundary state vectors. The state vectors at
the boundaries are composed of displacements, rotation angles, bending moments, and shear forces, which are partly known and
partly unknown, and end up with reduced overall transfer matrix. Nontrivial solution requires the coefficient matrix to be singular
to yield the required natural frequencies. This paper implements two novel algorithms based on the methodology by reducing the
zero search of the reduced overall transfer matrix’s determinate to a minimization problem and demonstrates a simple and robust
algorithm being much more efficient than direct enumeration. The proposal method is easy to formulate, systematic to apply, and
simple to code and can be extended to complex structures with any boundary conditions. Numerical results are presented to show
the validity of the proposal method against the published literature.

1. Introduction The vibration of systems composed of uniform double-beam


coupled by translational springs or elastic layers have been
The vibration problem of beam-type structures is of partic- studied extensively in the literature. Inceoğlu and Gürgöze
ular urgent issue in many branches of modern aerospace, [1] studied the bending vibrations of a combined system
mechanical, and civil engineering. Natural vibration frequen- consisting of two clamped-free beams carrying tip masses to
cies and modes are one of the most important dynamic which several double spring-mass systems are attached across
characteristics of these kinds of systems. For example, the the span. Using Green’s function method, the frequency
precision in manufacturing can be highly influenced by equation of the system is established. Kukla [2] solved the
vibrations. If the vibration characteristics cannot be solved or problem of free vibration of two axially loaded beams which
preestimated exactly when designing a mechanism system, are connected by many translational springs. The solution
it is often hard to obtain a good dynamic performance of contains possible combinations of the classical boundary
the mechanism system and consequently hard to control its conditions. The technique of the solution consists of devel-
vibration. oping a Green function. In [3], the vibrations of uniform
There are different types of beam models. One of the well- beams connected by homogeneous elastic layer are devoted.
known models is the Euler-Bernoulli beam theory that works Oniszczuk [4] discussed the free transverse vibrations of
well for slender beams. According to the Euler-Bernoulli two parallel simply supported beams continuously joined
beam theory, the length of each beam section is much greater by a Winkler elastic layer. The motion of the system was
than the height of each section and the shear and rotary described by a homogeneous set of two partial differential
inertia effects are ignored. equations, which was solved by using the classical Bernoulli-
The vibration theory of single-beam systems is well Fourier method. Oniszczuk [5] is devoted to analyze the
developed and studied in detail in hundreds of contributions. undamped forced transverse vibrations of an elastically
2 Advances in Mechanical Engineering

connected complex double-beam system in the case of simply and 𝐼 is the area moment of inertia. Beam mass per unit
supported beams. The classical modal expansion method was length 𝑚 = 𝜌𝐴, where 𝜌 is the material density and 𝐴 is the
applied to determine the dynamic responses of the beams beam cross-section area. Of course, any different materials,
due to arbitrarily distributed continuous loads. Vu et al. [6] different dimensions, or different boundary conditions could
presented an exact method for solving the vibration problem be considered in MS-TMM. The beams have the same length
of a damped double-beam system subjected to harmonic 𝐿 and are joined by the 𝑗-system of spring/dashpot located
excitation. The double-beam system consists of two identical at different positions (e.g., 𝑥1 and 𝑥𝑗 , where 𝑗 is the number
beams with the same boundary conditions on both sides. of spring/dashpot systems between two levels). 𝑘𝑦 and 𝑐 are
The beams are connected by a viscoelastic layer. Natural linear translational spring constant and damping coefficient,
frequencies and mode shapes of vibration of the system respectively, as shown in Figure 1.
are determined and the forced vibrations are investigated.
Gürgöze and Erol [7] determined the natural frequencies
of a clamped-free double-beam system carrying tip masses 3. MS-TMM Strategy in the Context of
to which several spring mass systems are attached across Free Vibration Characteristics
the span. However, there are only few contributions dealing
with the vibration of multibeam systems. That is, probably, 3.1. MS Topologies. According to the natural attribute of
the general vibration analyses of an elastically connected bodies, a complicated MS can be represented by various
multibeam system are complicated and laborious in view bodies (e.g., rigid bodies, elastic bodies, lumped masses,
of a large variety of possible combinations of boundary etc.) interconnected by hinges (e.g., spherical joints, sliding
conditions, and thus, the solution of the governing coupled joints, cylindrical joints, dampers, springs, etc.). In MS-
partial differential equations is difficult [8]. TMM, there are different topologies based on a certain set
Multibody system dynamics (MSD) has become an of modeling variables to formulate the dynamic equations of
important theoretical tool for wide engineering problems MS. Such topologies are chain, tree, closed loop (as illustrated
analysis in the world. Lots of methods of MSD have been in Figure 2 for reference), network, and so forth.
studied by many authors on theory and computational
method [9–15]. Professor Rui Xiaoting and his students have
been enlightened by the method of letting state vectors (SVs) 3.2. State Vector, State Variables, and Transfer Direction.
be transferred into classical transfer matrix method and The state vector (SV) at a connection point of MS is a
built up a new multibody dynamics method called “Transfer column vector denoting the mechanics state of this point.
Matrix Method of Linear Multibody Systems-MS-TMM” It includes the displacements of the point (including angu-
[16]. Using MS-TMM, the eigenvalue of linear multi-rigid- lar displacements) and the corresponding internal forces
flexible-body system is computed easily, the computational (including internal moments). Therefore, the SV is given
ill-condition is overcome, and the computational efficiency is by kinematics (displacements) and kinetics (internal forces)
increased. Over 20 years, MS-TMM has been developed and quantities, called state variables. For convention in this paper,
used widely in engineering applications. z with bold lowercase represents the SV in the physical
Motivated by the interesting study by Kukla [2] which coordinates and Z with bold capital represents the SV in
was published on the problem of the natural longitudinal the modal coordinates. Vibrations in space are described
vibrations of two rods coupled by many translational springs by displacement coordinates 𝑥, 𝑦, 𝑧 along the Cartesian axes
and by the two novel algorithms based on the new recursive and angular rotations 𝜃𝑥 , 𝜃𝑦 , 𝜃𝑧 about Cartesian axes. Cutting
scanning approach presented by Bestle et al. [17], this paper forces and moments are given by 𝑞𝑥 , 𝑞𝑦 , 𝑞𝑧 and 𝑚𝑥 , 𝑚𝑦 , 𝑚𝑧 ,
presents a unique yet simple scenario of obtaining the respectively. Positive directions at input points are shown
exact free vibration characteristics of undamped/damped in Figure 3(a). Positive directions of forces and moments at
multilevel beam coupled elastically. The scenario developed output points (Figure 3(b)) are opposite due to the principle
in this paper is based on MS-TMM and Euler-Bernoulli beam of action equals reaction. In 3D case with 𝑛𝑠 = 12 (𝑛𝑠 is the
theory. number of state variables in the SV), the SVs in physical and
The text is organized as follows. The problem statement modal coordinates at the connection point 𝑝𝑖,𝑘 (where the
is presented in Section 2. In Sections 3 and 4, the general first subscript 𝑖 is the serial number of element for boundary
theorem brief of MS-TMM and problem solution scenario are end and the second subscript 𝑘 is the serial number of the
shown. In Section 5, some results calculated by MS-TMM and hinge element and 𝑘 = 0 for boundary end) are summarized
the other method are given which can validate the proposed in a vector, receptively:
method. The conclusions are presented in Section 6.
󵄨
z𝑖,𝑘 󵄨󵄨󵄨physical coordinates

2. Problem Statement 𝑇
= [𝑥, 𝑦, 𝑧, 𝜃𝑥 , 𝜃𝑦 , 𝜃𝑧 , 𝑚𝑥 , 𝑚𝑦 , 𝑚𝑧 , 𝑞𝑥 , 𝑞𝑦 , 𝑞𝑧 ]𝑖,𝑘 ,
The transversely vibrating system in a plane under considera- (1)
󵄨
tion consists of multi-level parallel, elastic, and homogeneous Z𝑖,𝑘 󵄨󵄨󵄨modal coordinates
Euler-Bernoulli beam with general boundary conditions. 𝑇
Beam bending stiffness is 𝐸𝐼, where 𝐸 is the elastic modulus = [𝑋, 𝑌, 𝑍, Θ𝑥 , Θ𝑦 , Θ𝑧 , 𝑀𝑥 , 𝑀𝑦 , 𝑀𝑧 , 𝑄𝑥 , 𝑄𝑦 , 𝑄𝑧 ]𝑖,𝑘 .
Advances in Mechanical Engineering 3

y
General
boundary x Beam 1 m1 , (EI)1
conditions
ky c ··· ky c
Beam 2 m2 , (EI)2

ky c ··· ky c
..
x1 .
xj General
boundary
conditions
ky c ··· ky c
Beam m mm , (EI)m

Figure 1: A system of a multi-level beam with a distributed spring/dashpot system(s).

Hinges

Loop

Bodies
Loop
Base
(a) (b) (c)

Figure 2: Multibody systems (a) chain, (b) tree, and (c) closed loop topology.

In case of 1D or 2D applications, the SV will be reduced direction shown in Figure 3, the transfer equation between the
as shown later. Defining a boundary point of the MS as component input and output is
the transfer end, the direction from all other boundary points
to the transfer end is called transfer direction. Along the z𝑛,0 = U𝑛 z𝑛−1,𝑛
transfer direction, the nodes entering into elements are called
inputs denoted by 𝐼 and the nodes leaving from elements are z𝑛−1,𝑛 = U𝑛−1 z𝑛−2,𝑛−1
called outputs 𝑂. z𝑛−2,𝑛−1 = U𝑛−2 z𝑛−3,𝑛−2 (2)
..
3.3. Transfer Equation, Transfer Matrix, Overall System Trans- .
fer Matrix, and Overall System State Vector. A vibrating MS
comprised of 𝑛-components, see Figure 4, is used as an z1,2 = U1 z1,0 .
example to show how to deduce the overall transfer equation
The constant matrix U𝑘 is the transfer matrix of the
and overall transfer matrix of the system. In order to describe
𝑘th component. Transfer matrices of basic components are
conveniently the idea, the chain topology is considered
considered as building blocks, which can be assembled
in the following. This vibrating system is comprised of 𝑛
together to provide the transfer matrix of the whole system
components and 𝑛 + 1 connection points. The SVs of the
according to the chain MS-TMM topology as follows:
boundary right extremity and other boundary left extremity
of the system are expressed as z𝑛,0 and z1,0 , respectively. 𝑛−1
Transfer direction of the system is always from another z𝑛,0 = Tz1,0 , where T = ∏U𝑛−𝑘 . (3)
boundary end to may call it the root. Following the transfer 𝑘=0
4 Advances in Mechanical Engineering

yI , qyI yO , −qyO

𝜃yI , −myI −𝜃yO , myO

−𝜃xO , mxO
𝜃xI , −mxI

I xI , qxI O xO , −qxO

𝜃zI , −mzI −𝜃zO , mzO

zI , qzI zO , −qzO

(a) (b)

Figure 3: Sign convention at (a) input; (b) output.

Transfer direction

2 4 ··· n − 1 n 0 Root
0 1 3
Left end Right end
z1,0 zn,0

I 3 O
{

z2,3 z3,4

Figure 4: TMM-MS in the sense of chain topology. A vibrating system comprised of 𝑛-components with 𝑛 + 1 connections.

Rewrite (2) as from elimination of all columns of Uall (4c) associated zeros
󵄨 󵄨 in zall . For harmonic vibrations, solutions maybe written as
Uall 󵄨󵄨󵄨𝑛𝑠 ×(2×𝑛𝑠 ) zall 󵄨󵄨󵄨(2×𝑛𝑠 )×1 = 0, (4a) z𝑖,𝑗 = Z𝑖,𝑗 𝑒𝜆𝑡 where 𝜆 = −𝜆𝑟 ± 𝑖𝜆𝑖 , 𝜆𝑟 , 𝜆𝑖 ∈ R are the
eigenvalues. The real part (−𝜆𝑟 ) is related to the magnitude
where
of damping, where the imaginary part (𝜆𝑖 ) is related to the
𝑇 vibration frequency of the damped system. For undamped
z𝑇all = [z𝑇1,0 z𝑇𝑛,0 ] , (4b)
systems, 𝜆𝑟 = 0 and 𝜆𝑖 = 𝜔. Finally, Uall is only a function of
Uall = [T −I𝑛𝑠 ] . (4c) the unknown 𝜆 𝑖 of the system. For nontrivial solutions, the
Eigenfrequency equation
Herein, Uall is the overall system transfer matrix and zall is the
!
overall system state vector. Δ (𝜆) = det Uall = 0 (5)

3.4. Eigenfrequency Equation of the Whole MS. The overall has to be fulfilled. The natural frequencies of the system can
transfer equation (4a) only involves the boundary SVs, and now be computed.
the SVs at all other connection points do not appear. The
SVs at the boundary are composed of displacements, rotation 3.5. Beam Transfer Matrix. The full derivation of transfer
angles, moments, and shears, which are partly known and matrices for the Timoshenko and Euler-Bernoulli beams
partly unknown. For common boundary conditions, half of vibrating in a plane (with kinematics and kinetics’ SV defined
state variables of zall (4b) are zeros due to known constraints. as z = [𝑦, 𝜃𝑧 , 𝑚𝑧 , 𝑞𝑦 ]𝑇 ) may be found in [18], which is an
Thus, (4a) reduces to Uall zall = 0, where zall is composed of the open access article and the reader may download it from the
unknown state variables and Uall is a square matrix resulting Internet. However, for completeness, only the transfer matrix
Advances in Mechanical Engineering 5

for the Euler-Bernoulli beam will be presented, Figure 5. The where 𝑌 (𝑥) = 𝐴 1 cosh 𝛽𝑥 + 𝐴 2 sinh 𝛽𝑥
differential equation of a Euler-Bernoulli beam is
+ 𝐴 3 cos 𝛽𝑥 + 𝐴 4 sin 𝛽𝑥,
(6)
𝜕4 𝑦 𝜕2 𝑦
𝐸𝐼 4 + 𝑚 2
𝜕𝑥 𝜕𝑡 𝐴 1 , 𝐴 2 , 𝐴 3 , and 𝐴 4 are arbitrary constants, and 𝛽 =
√4 −𝑚𝜆2 /(𝐸𝐼). For the Euler-Bernoulli beam, the linearized
𝜕4 𝑌 (𝑥) 𝑚𝜆2 relations in modal coordinates Θ𝑧 = 𝑌󸀠 , 𝑀𝑧 = 𝐸𝐼𝑌󸀠󸀠 , and
= 0 𝑦 (𝑥, 𝑡) = 𝑌 (𝑥) 𝑒𝜆𝑡 4
+ 𝑌 (𝑥) = 0,
󳨀󳨀󳨀󳨀󳨀󳨀󳨀󳨀󳨀󳨀󳨀󳨀󳨀󳨀→ 𝜕𝑥 𝐸𝐼 𝑄𝑦 = 𝑀𝑧󸀠 maybe added to end up with the transfer relation:

𝑌 cosh 𝛽𝑥 sinh 𝛽𝑥 cos 𝛽𝑥 sin 𝛽𝑥 𝐴1


[ ] [ ][ ]
[ Θ𝑧 ] [ 𝛽 sinh 𝛽𝑥 𝜆 cos 𝛽𝑥 ]
[ ] [ 𝛽 cosh 𝛽𝑥 −𝛽 sin 𝛽𝑥 ][ 𝐴 2]
[ ] [ ][[ ]
]
[ ]
Z (𝑥) = B (𝑥) a or [
[𝑀 ]
] =[ ] [ ]. (7)
[ 𝑧] [𝐸𝐼𝛽2 cosh 𝛽𝑥 𝐸𝐼𝛽2 sinh 𝛽𝑥 −𝐸𝐼𝛽2 cos 𝛽𝑥 −𝐸𝐼𝛽2 sin 𝛽𝑥 ] [ ]
[𝐴 3 ]
[ ] [ ]
[ ] [ ][ ]
[
]
[ 3 3 3 3
]
𝑄𝑦 𝐸𝐼𝛽 sinh 𝛽𝑥 𝐸𝐼𝛽 cosh 𝛽𝑥 𝐸𝐼𝛽 sin 𝛽𝑥 −𝐸𝐼𝛽 cos 𝛽𝑥 𝐴4
[ ]𝑥 [ ][ ]

The coefficient vector a = [𝐴 1 , 𝐴 2 , 𝐴 3 , 𝐴 4 ]𝑇 summarizes (1) Break up the complicated MS into components with
the unknown constants to be adopted to boundary condi- simple dynamic properties, which can be expressed in matrix
tions. At input end Z𝐼 (𝑥 = 0), we get Z𝐼 = [B(0)]a. Thus, form and SVs (for each component, it is possible to obtain
the coefficient vector can be expressed as a = [B(0)]−1 Z𝑂 and the close form expression of the transfer matrix giving the
substituting it into (7) for the beam output end at 𝑥 = 𝑙, one displacements and the forces applied to one extremity to
gets the displacements and forces applied to the other extremity).
In other words, on the component level, the governing
Z𝑂 = [B (𝑙)] a = [B (𝑙)] [B (0)]−1 Z𝐼 = UZ𝐼 , (8a) partial and ordinary differential or algebraic equations are
transformed to algebraic transfer equations, where the output
where state results from a product of the input state and an
element specific transfer matrix. These component matrices
U = B (𝑙) B−1 (0) are considered as building blocks. In fact, the transfer matrix
of such components needs not to be rededuced but may be
𝑇 𝑈 𝑉
[ 𝑆
taken directly from a transfer matrix library.
𝛽 𝐸𝐼𝛽 2 𝐸𝐼𝛽3 ]
[ ] (2) Following the transfer direction that has been
[ 𝑇 𝑈 ]
[ ] designed already by the analyst and according to the topology
[ 𝛽𝑉 𝑆 ]
=[
[
𝐸𝐼𝛽 𝐸𝐼𝛽2 ] ,
]
of the MS, these component transfer matrices are then
[ 𝑇 ] assembled and end up with a system of linear algebraic
[𝐸𝐼𝛽2 𝑈 𝐸𝐼𝛽𝑉 𝑆 ]
[ 𝛽 ]
equations called the overall transfer equation.
[ ] (3) Substitute the boundary conditions into the overall
3 2
[ 𝐸𝐼𝛽 𝑇 𝐸𝐼𝛽 𝑈 𝛽𝑉 𝑆 ] transfer equation to construct the eigenfrequency equation.
Consequently, the vibration characteristics such as frequen-
𝑐ℎ + 𝑐 𝑠ℎ + 𝑠
𝑆= , 𝑇= , cies can be deduced as the roots of a transcendental equation.
2 2 Due to narrow couples of natural frequencies, the classical
𝑐ℎ − 𝑐 𝑠ℎ − 𝑠 zero search method is likely to fail. However, a new recursive
𝑈= , 𝑉= , scanning approach for minima of the absolute values of the
2 2
determinant shows much more efficiency and reliability than
𝑐ℎ = cosh (𝛽𝑙) , 𝑠ℎ = sinh (𝛽𝑙) , direct enumeration.
𝑐 = cos (𝛽𝑙) , 𝑠 = sin (𝛽𝑙)
4. Problem Solution Scenario
(8b)
Figure 6 illustrates the suggested scenario for the solution
is the transfer matrix of the Euler-Bernoulli beam compo- of the problem statement and as follows. There are (1 : 𝑚)
nent. multi-level beam. Each beam level is divided into (1 : n)
components, which have (0 : 𝑛 + 1) connection points
3.6. Summary. In the context of free vibration characteristic, and coupled with another beam level by a system (say 1 :
the general strategy of linear MS-TMM, in summary, is as 𝑗) of a viscoelastic material (modeled as a spring/dashpot
follows. system). The massless dummy body as shown in Figure 6
6 Advances in Mechanical Engineering

qyI
𝜃zO mzO
mzI 𝜃zI m, L, EI

qyO
yI yO

Figure 5: The direction of the state vector of Euler-Bernoulli beam component.

Massless dummy body

i−1 i n−1 n
0 1 2
A G
U1,1 U1,2 ··· U1,i ··· U1,n−1 U1,n 1

1, 1 j, 1

U2,1 U2,2 ··· U2,i ··· U2,n−1 U2,n 2


B C
.. ..
. .
.. 1, m − 1
.
..
.. j, m − 1 .
.
V m
Um,1 Um,2 ··· Um,i ··· Um,n−1 Um,n

Beam segment component

Spring/dashpot component

Figure 6: Structure of a multi-level beam with the distribution of spring/dashpot.

(e.g., 𝐴, 𝐵, 𝐶, and so on) is a connection point between Figure 7(a) sketches the first step. According to the continuity,
the two beam segment components and spring/dashpot 𝑗 has identical displacements, angles, and moments at input
system(s). The main key of MS-TMM is transferring the and output:
SV from one component to another following the general
transfer equation (2). For the beam segment component, 𝑌𝑂 = 𝑌𝐼 , Θ𝑧𝑂 = Θ𝑧𝐼 , 𝑀𝑧𝑂 = 𝑀𝑧𝐼 . (9a)
the transfer matrix is available (8b), while it is not for
spring/dashpot system at the connection point between two And from the force analysis shown in Figure 7(b), the
or multi-level beam in this paper. However, it is based on the spring/dashpot force 𝐹 changes the shear forces:
kinematics and kinetics of the spring/dashpot to formulate 𝑄𝑦𝑂 = 𝑄𝑦𝐼 + 𝐹. (9b)
the transfer matrix and it needs two steps. First, as an
example, let us consider a connection point 𝑗 between two Equations (9a) and (9b) in the view of the SV as a matrix form
beam segment components and system of spring/dashpot
may be grounded or connected to another connection point. Z𝑂 = Z𝐼 + e4 𝐹. (9c)
Advances in Mechanical Engineering 7

Massless dummy body

Beam segment I j Beam segment


O
Fsys yO
Qy,I F Qy,O

ky c
YI YO
ky c
I j O yI

Mz𝐼 Θz𝐼 Θz𝑂 Mz𝑂


Maybe grounded or connected
to another connection point ZI ZO Fsys

(a) (b) (c)

Z1,I 1 Z1,O
G Level
1

F
Fsys

ky𝑗,1

cj,1

Fsys

Z1,I 1 F
Z1,O Level
Level
A 1 C 2
Z2,I Z2,O
F F
Fsys
Fsys

Massless dummy body


ky1,1 ky1,𝑚−1

c1,1 c1,m−1

Fsys Fsys
F F
Level Level
B 2 V 3
Z2,I Z2,O Z3,I Z3,O

(d) (e)

Figure 7: (a) Two beam segments and spring/dashpot system connected at massless dummy body, (b) state vectors (SV) at the massless
dummy body 𝑗-connection point, (c) force analysis of a spring/dashpot system, (d) a system 1,1 connected two levels 1-2 through two
connection points 𝐴 and 𝐵 shown in Figure 6, and (e) two systems 𝑗, 1 and 1, 𝑚−1 connected with multi-level (1-2-3) through three connection
points 𝐺, 𝐶, and 𝑉 shown in Figure 6.
8 Advances in Mechanical Engineering

𝑇
e4 = [0 0 0 1] is a unit vector assigning 𝐹 to the points 𝐺, 𝐶, and 𝑉 shown in Figures 6 and 7(e) can be
transfer equation of the massless connection body. The obtained similarly:
spring compression and damping forces are given as (see
Figure 7(c))
U𝑑𝑗,1 U𝑐𝑗,1 0
Z [ ] Z
𝐹spring = 𝑘𝑦 (𝑌𝐼 − 𝑌𝑂) = 𝑘𝑦 Δ𝑌 = 𝑘𝑦 e𝑇1ΔZ { }
{ 1
} [ 𝑐 𝑑 𝑐 ] { 1}
{ }
{ Z2 } = [ 𝑗,1 𝑗,1/1,𝑚−1 1,𝑚−1 ]
[ U U U
] { Z2} ,
{ } [ ] { }
) 𝑦 = 𝑌𝑒𝜆𝑡 = 𝑐𝜆 (𝑌𝐼 − 𝑌𝑂) = 𝑐𝜆e𝑇1 ΔZ [ ]
𝐹dashpot = 𝑐 (𝑦𝐼̇
− 𝑦𝑂̇ {Z3 }𝑂 0 U𝑐1,𝑚−1 U𝑑1,𝑚−1 {Z3 }𝐼
󳨀󳨀󳨀󳨀󳨀󳨀󳨀󳨀→ [ ]
󳨀→ 𝐹sys = (𝑘𝑦 + 𝑐𝜆) e𝑇1ΔZ.
where U𝑑𝑗,1 ≡ I − D𝑗,1 , U𝑐𝑗,1 ≡ D𝑗,1
(10) (12d)

U𝑑1,𝑚−1 ≡ I − D1,𝑚−1 , U𝑐1,𝑚−1 ≡ D1,𝑚−1


Herein, e𝑇1 = [1 0 0 0] is a unit vector, which extracts 𝑌
from the transfer SV. Substitute (10) into (9c) and knowing
that 𝐹 = ± 𝐹sys , we ended the first step with U𝑑𝑗,1/1,𝑚−1 ≡ I − D𝑗,1/1,𝑚−1 ,

Z𝑂 = Z𝐼 ± DΔZ, D𝑗,1/1,𝑚−1 = ((𝑘𝑦 + 𝑐𝜆)𝑗,1 + (𝑘𝑦 + 𝑐𝜆)1,𝑚−1 ) e4 e𝑇1 .

0 0 0 0
[ 0 0 0 0] (11)
where D = (𝑘𝑦 + 𝑐𝜆) e4 e𝑇1 =[
[
]. Now we are in the position to end the solution scenario.
0 0 0 0] From the MS-TMM topology point of view, the system shown
[ 𝑘𝑦 + 𝑐𝜆 0 0 0] in Figure 1 or equivalently Figure 6 can be modeled as a mul-
tichain, following the similar procedures mentioned above
For system 1,1 connected two levels 1-2 through two in Section 3.3 to determine the overall transfer equation and
connection points 𝐴 and 𝐵 shown in Figure 6, the second step ending up with determinant solution of the reduced overall
is how to deduce the transfer equation between the input and transfer matrix.
output related to these two levels. For the connection point 𝐴
(see Figure 7(d)) and using (11) with 𝐹 = 𝐹sys ,
5. Numerical Examples
Z1,𝑂 = Z1,𝐼 + D1,1 (Z2,𝐼 − Z1,𝐼 ) Before the free vibration analysis of a multi-level elastic
beam coupled by a spring/dashpot system(s) is performed,
(I − D1,1 )Z1,𝐼 + D
= ⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟ ⏟⏟Z2,𝐼
⏟⏟⏟⏟⏟
1,1 the reliability of the suggested scenario and the Matlab
U𝑑1,1 U𝑐1,1
(12a)
computer program developed for this paper are confirmed
by comparing the present results with those obtained from
Z1,𝑂 = U𝑑1,1 Z1,𝐼 + U𝑐1,1 Z2,𝐼 . the existing literature.

Similarly, for connection point 𝐵 and using (11) with 𝐹 =


5.1. Reliability of the Suggested Scenario. Figure 8(a) consists
−𝐹sys ,
of two free-free and clamped-clamped undamped beams
connected to each other with two symmetrically distributed
Z2,𝑂 = Z2,𝐼 − D1,1 (Z2,𝐼 − Z1,𝐼 ) , linear translational springs. The hybrid system is vibrating
(12b) transversely in the 𝑥 − 𝑦 plane. The beams are supposed to
Z2,𝑂 = U𝑐1,1 Z1,𝐼 + U𝑑1,1 Z2,𝐼 . have the same length, material, and geometrical data, that is,
(𝐿 = 1 m, 𝑚1 = 𝑚2 = 𝑚 and (𝐸𝐼)1 = (𝐸𝐼)2 = 𝐸𝐼). It is worth
Combining (12a) and (12b) in one matrix form, we ended up mentioning here that the proposal scenario can be extended
the second step by writing the transfer equation and transfer to a system consisting of any number of uniform/nonuniform
matrix of a system located between two levels as beams coupled with any number of spring/dashpot systems.
However, the two springs 𝑘𝑦1,1 and 𝑘𝑦2,1 are located at distances
𝑥1 and 𝑥2 = 𝐿 − 𝑥1 , respectively, and are moving as a
U𝑑1,1 U𝑐1,1 function of 𝑥1 from 0 to 0.5 m as shown in Figure 8(a). This
Z1 ] {Z1 } .
{ } =[ 𝑐
Z2 𝑂 U1,1 U𝑑1,1 Z2 𝐼 (12c) example is presented by Kukla [2], which studied the dynam-
[ ]
⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟ ics of the longitudinal vibrations of two rods coupled by
transfer matrix several translational springs using Green’s functional method.
Figure 8(b) shows a vibrating two-level beam comprised of
The transfer equation for two systems 𝑗, 1 and 1, 𝑚 − 1 10 components with 4 boundaries, namely, Z1,0 , Z6,0 on the
connected with multi-level (1-2-3) through three connection system left side and Z5,0 , Z10,0 on the right side. Herein,Z𝑖,𝑘 =
Advances in Mechanical Engineering 9

y Transfer direction
x m1 , (EI)1
Beam 1 1 3 5
ky1,1 ky2,1 0 0 Level 1
Z1,0 Z1,2 Z2,3 Z3,4 Z4,5 Z5,0
m2 , (EI)2 Chain 1
Beam 2 ky1,1 ky2,1
2 4
6 7 8
x1 0 Level 2
0 Z Z6,2 Z2,7 Z7,4 Z4,8 Z8,0
x2 = 1 − x1 6,0 Chain 2
x1 L − 2x1 x1
L

Massless dummy body


Beam segement
(a) (b)

Figure 8: (a) Free-free and clamped-clamped beams connected with two linear translational springs, (b) chain TMM-MS: State vectors and
transfer direction of the suggested solution scenario.

[𝑌, Θ𝑧 , 𝑀𝑧 , 𝑄𝑦 ]𝑇𝑖,𝑘 . Chain MS-TMM topology is systematic to ↓


apply. Following the transfer direction from the left to right Z5,0
{ }
and applying (2) for two-level, the transfer equation can be Z8,0 𝑂
written as
U5 U4,𝑑 2,𝑑
2,1 U3 U1,1 U1 U5 U4,𝑑 2,𝑐
2,1 U3 U1,1 U6
[( ]
[ +U U4,𝑐 U U2,𝑐 U ) (+U U4,𝑐 U U2,𝑑 U )]
U5 0 [ 5 2,1 7 1,1 1 5 2,1 7 1,1 6 ]
Z5,0 Z4,5 [ ]
{ } =[ ]{ } =[ ]
Z8,0 𝑂 0 U8 Z4,8 𝐼 [ U8 U4,𝑐 U U 2,𝑑
U U U 4,𝑐
U U 2,𝑐
U ]
[ 2,1 3 1,1 1 8 2,1 3 1,1 6 ]
[( 4,𝑑 2,𝑐 ) ( 4,𝑑 2,𝑑 )]
↓ +U8 U2,1 U7 U1,1 U1 +U U U U
8 2,1 7 1,1 6 U
[ ]
U4,𝑑 U4,𝑐 Z1,0
Z4,5 [ 2,1 2,1 ] Z3,4
{ } = [ 4,𝑐 4,𝑑 ] { } ×{ }
Z4,8 𝑂 U2,1 U2,1 Z7,4 𝐼 Z6,0 𝐼
[ ] ↓

T1,1 T1,2
Z5,0 ] {Z1,0 } .
Z3,4 U3 0 Z2,3 { } =[
Z8,0 𝑂 T2,1 T2,1 Z6,0 𝐼
{ } =[ ]{ } [ ]
Z7,4 𝑂 0 U7 Z2,7 𝐼
(13)

U1 , U3 , U5 , U6 , U7 , and U8 represent the beam segment
U2,𝑑 U2,𝑐 transfer matrices (8b). U2,𝑑 2,𝑐 4,𝑑 4,𝑐
1,1 , U1,1 , and U2,1 , U2,1 are the
Z2,3 [ 1,1 1,1 ] Z1,2
{ } =[ 𝑐 ]{ } transfer matrices of a system between two connection points
Z2,7 𝑂 U1,1 U2,𝑑
1,1
Z6,2 𝐼
located between two beams (12c). The overall transfer equa-
[ ]
tion, overall SV, and overall transfer matrix are, respectively,

󵄨 󵄨
Uall 󵄨󵄨󵄨8×16 Zall 󵄨󵄨󵄨16×1 = 0, (14a)
Z1,2 U1 0 Z1,0
{ } =[ ]{ } 𝑇 𝑇 𝑇 𝑇 𝑇
Z6,2 𝑂 0 U6 Z6,0 𝐼 where Z𝑇all = [Z ⏟⏟⏟ ⏟ Z
⏟⏟⏟1,0 ⏟⏟⏟ ⏟ Z
⏟⏟⏟6,0 ⏟⏟⏟ ⏟ Z
⏟⏟⏟5,0 ⏟⏟⏟8,0
⏟⏟⏟ ⏟] , (14b)
4×1 4×1 4×1 4×1
⇓ 󵄨 󵄨
T1,1 󵄨󵄨󵄨4×4 T1,2 󵄨󵄨󵄨4×4 −I4×4 O4×4
U4,𝑑 U4,𝑐 Uall = [ ]. (14c)
Z5,0 U5 0 U 02,1 2,1 󵄨󵄨 󵄨󵄨
{ } = [ ][ ][ 3 ] T 󵄨 T
[ 2,1 󵄨4×4 2,2 󵄨4×4 󵄨 O 4×4 −I4×4 ]
Z8,0 𝑂 0 U8 4,𝑐 4,𝑑 0 U7
[U2,1 U2,1 ]
Applying the boundary conditions listed in Table 1, half
U2,𝑑 2,𝑐
1,1 U1,1
of state variables of Zall (14b) are zeros due to known
×[ 𝑐 ] [U1 0 ] {Z1,0 } constraints. Thus, the overall transfer equation reduced to
U1,1 U2,𝑑
1,1
0 U6 Z6,0 𝐼
Uall |8 × 8 Zall |8 × 1 = 0 that is ready for eigenproblem (5).
[ ]
10 Advances in Mechanical Engineering

Table 1: Common boundary conditions for a beam vibrating in a grids 𝑁𝑥0 = 500, absolute precision tolerance 𝜀 = 10−6 ).
plane. Figure 9 shows log10 |Δ| obtained from the fMin1D algorithm
versus the first 10 dimensionless frequency values (𝜔 =
Support type Zero terms Nonzero terms
Fixed 𝑌, Θ𝑧 𝑄𝑦 , 𝑀𝑧 √4 (𝜆𝑖 ≡ 𝜔)2 𝑚𝐿4 /𝐸𝐼) for 𝑘𝑦 = 𝑘𝑦 = 𝑘𝑦 = 100(𝐸𝐼/𝐿3 ) and
1,1 2,1

Pinned 𝑌, 𝑀𝑧 𝑄𝑦 , Θ𝑧 𝑥1 = 0.3 m. 𝜔 values are plotted versus the location of the


Free 𝑄𝑦 , 𝑀𝑧 𝑌, Θ𝑧 two springs in Figures 10(a) and 10(b) for 𝑘𝑦 = 100(𝐸𝐼/𝐿3 )
Guided 𝑄𝑦 , Θ𝑧 𝑌, 𝑀𝑧 and 𝑘𝑦 = 1000(𝐸𝐼/𝐿3 ), respectively. The first and second
springs are moving opposite to each other from 𝑥1 = 0 at
the beams support ends and coincide together at the middle
The natural frequencies of the system can now be com- of the beams (𝑥1 = 0.5 m). Reference [2] does not provide the
puted by zero search of the determinant, which is based on results as tables or specific data but only showed figures. The
sign change of Δ(𝜔) (in case of real numbers) or Δ(𝜆) (in reader may download the reference from the library or from
case of complex numbers) during a scanning of an interesting the Internet because the authors of the present paper cannot
frequency range. This procedure can be cumbersome for include the figures to maintain the journal publication rights.
several reasons. Therefore, reliable and efficient algorithms However, the MS-TMM results are in very good agreement
called recursive scanning approach are applied (see [17] for with Figures 2(a) and 2(b) [2, page 131].
more details of the proposed algorithm) by switching from
zero search for Δ to minimization of the absolute value |Δ| 5.2. Free Vibration of Damped Elastically Coupled Triple
of the determinant, which is equally well applicable to both Beams. The physical model of the transversely vibrating
the real and the complex cases. The algorithm general idea system under consideration is composed of three paral-
is as follows. In a first iteration step, it divides an interesting lel uniform rectangular Euler-Bernoulli beams of homoge-
band of frequencies into a number of sample points and nous properties. Each two-level beam is joined (connected)
searches for lower peaks. Each region having a lower peak is together by two spring/dashpot systems located at 𝑥1 and 𝑥2 ,
then divided again into small intervals to find more narrow respectively (see Figure 1). The beams have the same length
regions of lower peaks as second iteration step. The algorithm and are pinned at their ends. The small damped vibrations of
proceeds until the required precision of roots is achieved. MS- the system are considered. In the sense of the chain MS-TMM
TMM natural frequency results are obtained using fMin1D (see Figure 11), the overall transfer equation, overall transfer
[1, 3000] rad/sec, scanning
algorithm [17] (frequency ranges ⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟ matrix, and overall SV are, respectively,
𝜆𝑖 =𝜔

Uall Zall = 0, (15a)


󵄨 󵄨 󵄨
T1,1 󵄨󵄨󵄨4×4 T1,2 󵄨󵄨󵄨4×4 T1,3 󵄨󵄨󵄨4×4 −I4×4 O4×4 O4×4
[ ]
[ 󵄨 󵄨 󵄨 ]
where Uall = [ T2,1 󵄨󵄨󵄨4×4 T2,2 󵄨󵄨󵄨4×4 T2,3 󵄨󵄨󵄨4×4 O4×4 −I4×4 O4×4 ] , (15b)
[ ]
󵄨 󵄨 󵄨
[ T3,1 󵄨󵄨󵄨4×4 T3,2 󵄨󵄨󵄨4×4 T3,3 󵄨󵄨󵄨4×4 O4×4 O4×4 −I4×4 ]12 × 24
𝑇
Z𝑇all = [Z1,0 𝑇 Z6,0 𝑇 Z11,0 𝑇 Z5,0 𝑇 Z10,0 𝑇 Z13,0 𝑇 ]24 × 1 , (15c)

T1,1 = U5 U4,𝑑 2,𝑑 4,𝑐 2,𝑐


2,1 U3 U1,1 U1 + U5 U2,1 U8 U1,1 U1 ,

T1,2 = U5 U4,𝑑 2,𝑐 4,𝑐 2/7,𝑑


2,1 U3 U1,1 U6 + U5 U2,1 U8 U1,1/1,2 U6 ,

T1,3 = U5 U4,𝑐 7,𝑐


2,1 U8 U1,2 U11 ,

T2,1 = U10 U4,𝑐 2,𝑑 4/9,𝑑 2,𝑐


2,1 U3 U1,1 U1 + U10 U2,1/2,2 U8 U1,1 U1 ,

T2,2 = U10 U4,𝑐 2,𝑐 4/9,𝑑 2/7,𝑑 9,𝑐 7,𝑐


2,1 U3 U1,1 U6 + U10 U2,1/2,2 U8 U1,1/1,2 U6 + U10 U2,2 U12 U1,2 U6 ,
(15d)

T2,3 = U10 U4/9,𝑑 7,𝑐 9,𝑐 7,𝑑


2,1/2,2 U8 U1,2 U11 + U10 U2,2 U12 U1,2 U11 ,

T3,1 = U13 U9,𝑐 2,𝑐


2,2 U8 U1,1 U1 ,

T3,2 = U13 U9,𝑐 2/7,𝑑 9,𝑑 7,𝑐


2,2 U8 U1,1/1,2 U6 + U13 U2,2 U12 U1,2 U6 ,

T3,3 = U13 U9,𝑐 7,𝑐 9,𝑑 7,𝑑


2,2 U8 U1,2 U11 + U13 U2,2 U12 U1,2 U11 .
Advances in Mechanical Engineering 11

ky = 100(EI/L3 ) and x1 = 0.3 m


15
10

ky1,1 ky2,1

1010

x1
x2 = 1 − x1
5
10
log10 |Δ|

100

10−5

10−10
0 2 4 6 8 10 12 14 16
𝜔

ky = 100(EI/L3 ) and x1 = 0.3 m ky = 100(EI/L3 ) and x1 = 0.3 m


8 10
10 10

106 108
log10 |Δ|

log10 |Δ|

104 106

102
104

10.6 10.7 10.8 10.9 11 11.1 11.2 11.3 11.4 14 14.05 14.1 14.15 14.2
𝜔
𝜔

Figure 9: fMin1D function determinant of a system consisting of free-free and clamped-clamped beams coupled with two springs. 𝑘𝑦 =
100(𝐸𝐼/𝐿3 ) and 𝑥1 = 0.3 m.

Herein, U1 , U3 , U5 , U6 , U8 , U10 , U11 , U12 , and to ⏟⏟U


⏟⏟⏟⏟⏟ ⏟⏟Z⏟⏟⏟
all all⏟⏟ = 0. The geometric and material properties
U13 represent the beam segment transfer matrices; 12 × 1212 × 1
U2,𝑑 2,𝑐 4,𝑑 4,𝑐 7,𝑑 7,𝑐 9,𝑑
1,1 , U1,1 , U2,1 , U2,1 , U1,2 , U1,2 , U2,2 , and U9,𝑐2,2 are the (steel) of the multi-level beam system are given as follows:
transfer matrices of a system between two connection 𝐿 = 1 m, 𝑏1 = 𝑏2 = 𝑏3 = 0.02 m, ℎ1 = ℎ2 = ℎ3 = 0.01 m,
points located between two beams (12c). U2/7,𝑑 4/9,𝑑 𝐸1 = 𝐸2 = 𝐸3 = 𝐸 = 2.069 × 1011 N/m2 , and
1,1/1,2 and U2,1/2,2 𝜌1 = 𝜌2 =𝜌3 = 𝜌 = 7850 kg/m3 . Here, 𝑏𝑖 and ℎ𝑖 represent
are the transfer matrices for two systems connected with
three-level through three connection points (12d). Applying the width and height of the beam, respectively. The cross-
sectional area and the moment of inertial of the beam
the boundary conditions, that is, Z1,0 = [0, Θ𝑧 , 0, 𝑄𝑦 ]𝑇1,0 ;
cross-section are 𝐴 𝑖 = 𝑏𝑖 × ℎ𝑖 and 𝐼𝑖 = 𝑏𝑖 × ℎ𝑖3 /12, respectively.
Z6,0 = [0, Θ𝑧 , 0, 𝑄𝑦 ]𝑇6,0 ; Z11,0 = [0, Θ𝑧 , 0, 𝑄𝑦 ]𝑇11,0 , Z5,0 =
For comparison, the lowest three “exact” dimensionless
[0, Θ𝑧 , 0, 𝑄𝑦 ]𝑇5,0 , Z10,0 = [0, Θ𝑧 , 0, 𝑄𝑦 ]𝑇10,0 , and Z13,0 = frequency values 𝜔 for the single pinned-pinned beam
[0, Θ𝑧 , 0, 𝑄𝑦 ]𝑇13,0 , the overall transfer equation (15a) reduced that might be found in structural or vibration text books
12 Advances in Mechanical Engineering

16 16

14 14

12 12

10 10

8 8
𝜔

𝜔
6 6

4 4

2 2

0 0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
x1 (m) x1 (m)

(a) (b)

Figure 10: Dimensionless frequency parameter 𝜔 values as a function of the spring locations (𝑥1 = 0 → 0.5 m) for a system consisting of
free-fee and clamped-clamped beams coupled by two springs. (a) 𝑘𝑦 = 100(𝐸𝐼/𝐿3 ); (b) 𝑘𝑦 = 1000(𝐸𝐼/𝐿3 ).

Transfer direction

1 3 5
0 0 Level 1
Z1,0 Z1,2 Z2,3 Z3,4 Z4,5 Z5,0 Chain 1

ky1,1 2 ky2,1 4
c1,1 c2,1

6 8 10
0 0 Level 2
Z6,0 Z6,2 Z2,8 Z8,4 Z4,10 Z10,0 Chain 2
or or or or
Z6,7 Z7,8 Z8,9 Z9,10
ky1,2 7 ky2,2 9
c1,2 c2,2
11 12 13
0 0 Level 3
Z11,0 Z11,7 Z7,12 Z12,9 Z9,13 Z13,0 Chain 3
x1 L − 2x1 x1

Figure 11: Chain TMM-MS: state vectors and transfer direction of pinned-pinned multi-level beam connected by spring/dashpot systems.

1010

105

100

10−5
log10 |Δ|

10−10

L 10−15

10−20

10−25

10−30
0 1 2 3 4 5 6 7 8 9 10 11 12
𝜔

Figure 12: fMin1D function determinant to evaluate the lowest three 𝜔 for undamped, uncoupled pinned-pinned three-beam.
Advances in Mechanical Engineering 13

ky1,1 ky2,1
10
ky1,2 ky2,2 9
8

x1 7

x2 = 1 − x1 6
5

𝜔
L
4
3
9.54
2
9.52
1
9.5 0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
9.48
𝜔

x1 (m)
9.46 (a) ky = 25(EI/L3 )
9.44

9.42 1010
0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
x1 (m)
105
(a) ky = 25(EI/L3 )
log10 |Δ|

6.6
100
6.55
6.5
6.45 10−5

6.4
𝜔

6.35 10−10
0 2 4 6 8 10 12
6.3
𝜔
6.25
(b) ky = 25(EI/L3 ) and x1 = 0.25 m
6.2
0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
x1 (m)

(a) ky = 25(EI/L3 )
102

100 100

10−2 10−2
log10 |Δ|
log10 |Δ|

10−4 10−4

10−6 10−6

10−8 9.41 9.42 9.43 9.44 9.45 9.46 9.47 9.48


6.25 6.3 6.35 6.4 6.45 6.5 6.55 6.6 𝜔
𝜔
(b) ky = 25(EI/L3 ) and x1 = 0.25 m
(b) ky = 25(EI/L3 ) and x1 = 0.25 m

Figure 13: (a) The lowest three 𝜔 values as a function of the spring systems locations (𝑥1 = 0 → 0.5 m) for undamped, coupled pinned-pinned
three-beam; (b) fMin1D function determinant to evaluate the dimensionless frequency parameter 𝜔 for 𝑘𝑦 = 25(𝐸𝐼/𝐿3 ) and 𝑥1 = 0.25 m.
14 Advances in Mechanical Engineering

ky1,1 ky2,1
c1,1 c2,1

ky1,2 ky2,2
c1,2 c2,2

x1
x2 = 1 − x1

2000 1000

1500 800
600
1000
400
500
200
𝜆r = 𝛿

𝜆r = 𝛿

0 0

−500 −200
−400
−1000
−600
−1500
−800
−2000 −1000
−200 0 200 400 600 800 1000 1200 1400 1600 −200 0 200 400 600 800 1000 1200 1400 1600
𝜆i = 𝜔 (rad/s) 𝜆i = 𝜔 (rad/s)
(a) (b)
500 600
400
500
300
200 400

100
300
𝜆r = 𝛿
𝜆r = 𝛿

0
200
−100
−200 100
−300
0
−400
−500 −100
−200 0 200 400 600 800 1000 1200 1400 1600 −200 0 200 400 600 800 1000 1200 1400 1600
i i
𝜆 = 𝜔 (rad/s) 𝜆 = 𝜔 (rad/s)

(c) (d)

Figure 14: fMin2D first step: damping (𝜆𝑟 = 𝛿) range scanning of a specific frequency band (𝜆𝑖 = 𝜔 ≤ 1500 rad/sec) for damped, coupled
pinned-pinned three-beam, (a) [−2000, 2000], (b) [−1000, 1000], (c) [−500, 500], and (d) [−100, 600].
Advances in Mechanical Engineering 15

the order of the frequency increases, the shape of the


8 frequency vibration versus the spring systems location
resembles a half or full sine wave with insignificant variation
6 in the frequency ranges. The strong ability of fMin1D
4 algorithm is shown in Figure 13(b) by capturing the very
narrow regions of lower peaks.
log10 |Δ|

2 The effect of the dashpot is included in the system


0 making the model more complicated. The two spring/dashpot
systems coupled with three-beam are located at position 𝑥1 =
−2 0.25 m. The spring/dashpot system in parallel is similar to
−4 viscoelastic material model. The given values of damping are
600 𝑐1,1 = 𝑐2,1 = 𝑐1,2 = 𝑐2,2 = 𝑐 = 5 √𝐸𝐼 𝑚/𝐿2 . Next, it is
400 1500
200 1000 the turn to implement fMin2D algorithm [17] for complex
0 500
𝜆r= −200 −500 0 eigenproblem solution. For undamped given system as it is
𝛿 (rad/ s)
𝜆 =𝜔
i
computed above, 𝜔 values do not exceed more than 10, which
is equivalent to (𝜆𝑖 = 𝜔 ≤ 1500 rad/sec). However, the
damping (𝜆𝑟 = 𝛿) is unknown in range. Therefore, within
the required system frequency band, a couple scanning
6 grids are demanded to understand and explore the damping
4 ranges of the system. Figure 14 shows different scanning
2 grids for damping ranges as follows: (a) [−2000, 2000], (b)
0 [−1000, 1000], (c) [−500, 500], and (d) [−100, 600]. It is
log10 |Δ|

−2 obviously seen that there are 9 promising regions distributed


−4 randomly on the surface. These regions represent the complex
−6 eigenvalues of the system. In the following searching complex
−8 roots, the damping range (Figure 14(d)) is considered. MS-
−10 TMM for damped, coupled three-beam results are shown in
600 Figure 15 and Table 2 after using fMin2D algorithm coded
400 1500 under Matlab environment with the following input data:
2000 1000
𝜆r =𝛿 0 0
500
s) ([−1,
⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟ [−100, 600], 𝑁𝑥0 = 35, 𝑁𝑦0 = 20, 𝜀𝑥 = 10−8 ,
1500], ⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟
(rad/
𝜆 =𝜔
i
𝜆𝑖 𝜆𝑟
−8
𝜀𝑦 = 10 ).
Figure 15: fMin2D second step: function determinant to evaluate
Chain MS-TMM for undamped/damped coupled multi-
the eigenvalues (𝜆 = −𝜆𝑟 ± 𝑖𝜆𝑖 ) for a damped, coupled pinned-
pinned three-beam. level beam with the suggested scenario and the two novel
recursive scanning algorithms provide a closed-form solu-
tion, not only presents the principles of the vibration problem
but also shed light on practical applications. Since the
are 3.1416, 6.2832 and 9.4248. MS-TMM results for solution is almost exact, it allows a complete understanding
undamped, uncoupled pinned-pinned three-beam are of a problem.
3.141592, 6.28318, and 9.424777 as shown in Figure 12
using fMin1D algorithm. For undamped elastically coupled
three-beam, Figure 13 shows that 𝜔 values were plotted 6. Conclusions
versus the location movement of spring systems. Systems
Starting from the principle of mechanics and the elementary
2 and 4 (𝑘𝑦1,1 = 𝑘𝑦1,2 = 𝑘𝑦 = 25 (𝐸𝐼/𝐿3 )) and systems 7 and formulations for the flexible beam, the free vibration analysis
9 (𝑘𝑦1,2 = 𝑘𝑦2,2 = 𝑘𝑦 = 25 (𝐸𝐼/𝐿3 )) are moving opposite to of laterally vibrating system made up of a multi-level Euler-
each other from (𝑥1 = 0 → 0.5 m). The system shows Bernoulli beam to which spring/dashpot systems are attached
symmetric and antisymmetric vibrations. For symmetric across the span is performed using one of the Transfer Matrix
vibration, the frequency parameters 𝜔1,1 , 𝜔1,2 , and 𝜔1,3 Method of Linear Multibody Systems (MS-TMM) scenarios.
(where 𝑙 and 𝑝 are the beam number and vibration mode, Although the number of coupling springs or spring/dashpot
resp.) are constant throughout the spring systems movement systems considered in the examples given was limited to
as seen in Figure 13(a). The antisymmetric of the first mode three, there is no inherent difficulty in extending the current
vibration 𝜔2,1 and 𝜔3,1 started with equally values as 𝜔1,1 at method to solve the problems of vibration of systems con-
𝑥1 = 0 (spring systems are rigidly mounted at the support sisting of any number of uniform/nonuniform beams with
ends) and increased significantly at 𝑥1 = 0.5 m (both spring different boundary conditions and coupled with any number
systems are in the same position). However, it is evident of spring/dashpot systems. The numerical results obtained to
that 𝜔2,1 and 𝜔3,1 are sensitive to the variation of spring reveal that the eigenfrequencies calculated by this method are
positions. The results presented of the other antisymmetric in very good agreement with those obtained by the published
vibration modes, 𝜔2,2 , 𝜔3,2 , 𝜔2,3 , and 𝜔3,3 , indicate that as literature. Moreover, MS-TMM is encouraging for further
16 Advances in Mechanical Engineering

Table 2: Chain MS-TMM eigenvalues results of damped, coupled Journal of Sound and Vibration, vol. 269, no. 1-2, pp. 431–438,
pinned-pinned three-beam. 2004.
[8] M. Abu-Hilal and N. Beithou, “Free transverse vibrations of a
𝜆 = −𝜆𝑟 ± 𝑖𝜆𝑖 Chain MS-TMM results using fMin2D
triple-beam system,” Journal of Mechanical Engineering, vol. 58,
𝜆𝑖 (rad/sec)
no. 1, pp. 30–50, 2007.
𝜆1 −2.463512 × 102 + 4.010226 × 101 𝑖 [9] W. Schiehlen, Multibody Systems Handbook, Springer, Berlin,
𝜆2 −2.074369 × 10−9 + 1.462698 × 102 𝑖 Germany, 1990.
𝜆3 −7.409510 × 101 + 1.652322 × 102 𝑖 [10] W. Schiehlen, “Multibody system dynamics: roots and perspec-
𝜆4 −4.577982 × 102 + 4.615502 × 102 𝑖 tives,” Multibody System Dynamics, vol. 1, no. 2, pp. 149–188,
1997.
𝜆5 −2.074369 × 10−9 + 5.850790 × 102 𝑖
[11] J. Wittenburg, Dynamics of Systems of Rigid Bodies, Edited by:
𝜆6 −1.484833 × 102 + 5.861790 × 102 𝑖 B. G. Teubner, Stuttgart, Germany, 1977.
𝜆7 −2.159398 × 102 + 1.243120 × 103 𝑖 [12] J. Wittenburg, Dynamics of Multibody Systems, Springer, Berlin,
𝜆8 −7.448404 × 101 + 1.311750 × 103 𝑖 Germany, 2nd edition, 2008.
𝜆9 −2.074369 × 10−9 + 1.316428 × 103 𝑖 [13] A. A. Shabana, Dynamics of Multibody Systems, Cambridge
University Press, New York, NY, USA, 3rd edition, 2010.
[14] A. A. Shabana, Computational Dynamics, John Wiley & Sons,
investigations of more complex multibody systems of this New York, NY, USA, 3rd edition, 2010.
type with rigid bodies due to simplicity in the formulation [15] A. A. Shabana, “Flexible multibody dynamics: review of past
of the transfer equation, being systematic to apply, and being and recent developments,” Multibody System Dynamics, vol. 1,
easy to program. no. 2, pp. 189–222, 1997.
[16] X. Rui, L. Yun, Y. Lu, B. He, and G. Wang, Transfer Matrix
Method of Multibody System and Its Application, Science Press,
Conflict of Interests Beijing, China, 2008, (Chinese).
The authors declare that there is no conflict of interests [17] D. Bestle, L. K. Abbas, and X. Rui, “Recursive eigenvalue
search algorithm for transfer matrix method of linear flexible
regarding the publication of this paper.
multibody systems,” Multibody System Dynamics, 2013.
[18] L. K. Abbas, M. J. Li, and X. Rui, “Transfer matrix method for the
Acknowledgments determination of the natural vibration characteristics of realistic
thrusting launch vehicle—part I,” Mathematical Problems in
The research was supported by the Research Fund for Engineering, vol. 2013, Article ID 764673, 16 pages, 2013.
the Doctoral Program of Higher Education of China
(20113219110025), the Natural Science Foundation of China
Government (11102089), and the Program for New Century
Excellent Talents in University (NCET-10-0075).

References
[1] S. Inceoğlu and M. Gürgöze, “Bending vibrations of beams
coupled by several double spring-mass systems,” Journal of
Sound and Vibration, vol. 243, no. 2, pp. 370–379, 2001.
[2] S. Kukla, “Free vibration of the system of two beams connected
by many translational springs,” Journal of Sound and Vibration,
vol. 172, no. 1, pp. 130–135, 1994.
[3] S. Kukla and B. Skalmierski, “Free vibration of a system
composed of two beams separated by an elastic layer,” Journal
of Theoretical and Applied Mechanics, vol. 32, no. 3, pp. 581–590,
1994.
[4] Z. Oniszczuk, “Free transverse vibrations of elastically con-
nected simply supported double-beam complex system,” Jour-
nal of Sound and Vibration, vol. 232, no. 2, pp. 387–403, 2000.
[5] Z. Oniszczuk, “Forced transverse vibrations of an elastically
connected complex simply supported double-beam system,”
Journal of Sound and Vibration, vol. 264, no. 2, pp. 273–286,
2003.
[6] H. V. Vu, A. M. Ordóñez, and B. H. Karnopp, “Vibration of a
double-beam system,” Journal of Sound and Vibration, vol. 229,
no. 4, pp. 807–822, 2000.
[7] M. Gürgöze and H. Erol, “On laterally vibrating beams carrying
tip masses, coupled by several double spring-mass systems,”
Hindawi Publishing Corporation
Advances in Mechanical Engineering
Volume 2014, Article ID 365265, 9 pages
http://dx.doi.org/10.1155/2014/365265

Research Article
Case Study on Human Walking during Wearing a Powered
Prosthetic Device: Effectiveness of the System ‘‘Human-Robot’’

Svetlana Grosu,1 Pierre Cherelle,1 Chris Verheul,2 Bram Vanderborght,1 and Dirk Lefeber1
1
Department of Mechanical Engineering, Vrije Universiteit Brussel, 1050 Brussels, Belgium
2
SayField International, Broeksloot 10, 3474 HP Zegveld, The Netherlands

Correspondence should be addressed to Svetlana Grosu; sgrosu@vub.ac.be

Received 12 September 2013; Accepted 16 November 2013; Published 9 January 2014

Academic Editor: Xiaoting Rui

Copyright © 2014 Svetlana Grosu et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

It is known that force exchanges between a robotic assistive device and the end-user have a direct impact on the quality and
performance of a particular movement task. This knowledge finds a special reflective importance in prosthetic industry due to
the close human-robot collaboration. Although lower-extremity prostheses are currently better able to provide assistance as their
upper-extremity counterparts, specific locomotion problems still remain. In a framework of this contribution the authors introduce
the multibody dynamic modelling approach of the transtibial prosthesis wearing on a human body model. The obtained results are
based on multibody dynamic simulations against the real experimental data using AMP-Foot 2.0, an energy efficient powered
transtibial prosthesis for actively assisted walking of amputees.

1. Introduction the design of the mechanism, sensors selection, actuators, and


control architecture [2].
A definition of the functionalities/duties between a human Compared with healthy persons, walking amputees
and a robotic device, also the organization of their interac- require 10–60% more metabolic energy depending on walk-
tion, basically, includes a number of different criteria that ing speed, physical individual properties, cause of amputa-
influence the effectiveness of the “human-robot” system. tion, amputation level, and prosthetic intervention charac-
The hierarchy of criteria importance depends on a general teristics. Furthermore, amputees walk at 11–40% slower self-
approach in a certain domain application. Generally, the selected gait speed than do persons with intact limbs [3, 4].
requirements in a robotic device design should assure the To date, commercially available prostheses comprise spring
maximum economical effectiveness of the system in combi- structures that store and release elastic energy throughout
nation with a personal security of the end-user. each walking stance period [5]. Due to their passive nature,
Robots for physical assistance to humans are meant such prostheses are unable to generate more mechanical
to reduce fatigue and stress, increase human capabilities energy than what is stored during each walking step. Also,
in terms of force, speed, and precision, and improve in experiences in clinical environment indicate that transtibial
general the quality of life. In other words, the crucial goal (TT) amputees suffer from a nonsymmetrical gait while
of a robot for physical human-robot interactions (pHRI) is wearing a completely passive prosthesis [6]. In distinction,
a generation of supplementary forces to overcome human the human ankle performs positive net work and has a greater
physical limits. Moreover, the human can bring experience, peak power over the stance period, especially at moderate to
global knowledge, and understanding for a correct execution fast walking speeds [7, 8].
of movements [1]. In case of assistive devices, an improved In literature several engineering challenges still slowing
analysis of the problems related to the physical interaction down the further development of a powered ankle-foot
with robots becomes mandatory. Also, in a special perspec- prosthesis [9, 10] are discussed. In the field of prosthetic leg
tive for the interaction with humans should be considered design, a critical objective is to progress a powered ankle
2 Advances in Mechanical Engineering

prosthesis capable of mimicking the dynamics of the human Motor, gearbox


ankle. A study of TT prosthetics research contributions & ballscrew
shows that none of the commercially available devices are
skillful of mimicking an human ankle-foot complex. With
current actuator technology, it is challenging to build an PO spring
ankle-foot prosthesis that matches the size and weight of
the human ankle-foot structure but still provides sufficient Locking system
stance-period work and instantaneous power output to drive
an amputee. In 1998, Klute and colleagues [11] were the
first to build a powered ankle-foot prosthesis efficient in
performing net positive work. Their device employed a
pneumatic actuation strategy with an off-board power supply.
More recent work has focused on the design of energetically
autonomous powered systems [12–14]. In the growing field
of rehabilitation robotics, the use of compliant actuators is Lever arm
becoming a standard where accurate trajectory tracking is not PF spring
required. The advantages of such actuators are represented by
safely interaction with the patient and large forces absorption
Figure 1: Design representation of AMP-Foot 2.0.
during gait. In the particular case of trans-tibial prostheses,
compliance of the actuation provides even more advantages.
Besides shock absorption in case of collision with objects
during walking, energy provided by the actuator (e.g., electric
motor) can be stored into its elastic element (e.g., spring in
series). The current state-of-the-art in powered ankle-foot loading a “push-off (PO)” spring during the complete stance
prostheses with focus on devices using compliant actuators phase. The prosthesis includes a locking mechanism which
has been presented in [15]. provides the energy implementation into the PO spring and
In the next sections, the authors are focused on the can be delayed and released at push-off. This way, full torque
influences on normal human gait of the forces that are and power required for locomotion can be obtained with less
generated by a motor and forces that are stored in and then power from the actuator.
released by springs, also the reaction of the mechanism as a Structurally, the device consists of three bodies: a leg, a
whole in interaction with user. Then, the comparison between foot, and a lever arm, pivoting around the ankle axis; see
the real “human-robot” setup and virtual model is discussed. Figure 2. As mentioned before, the system comprises 2 spring
sets: a PF and a PO spring set. The PF spring is placed
between a fixed point 𝑝 on the foot and a cable that runs
2. The Ankle Mimicking Prosthetic- (AMP-) over a pulley 𝑎 to the lever arm at point 𝑏 and is attached
Foot 2.0: Background to the lever arm at point c, while the PO spring is placed
between the motor-ballscrew assembly and a fixed point 𝑑
This section is dedicated to a summary description of the
on the lever arm. A critical part of this device is the locking
prosthesis used in the study. The section includes presenta-
mechanism, that is meant to bear high forces while being as
tion of the AMP-Foot 2.0 mechanical design and validation
compact and lightweight as possible and is represented by a
part in the framework of real experiments, with an amputee
four bar linkage moving in and out of its singular position.
subject participation.
The working principle of such a system has been proposed in
[17].
2.1. Design of the AMP-Foot 2.0. The AMP-Foot 2.0, see The locking mechanism which provides a rigid connec-
Figure 1, claims a new energy concept, based on a principle tion between the leg and the lever arm when energy is injected
of optimal power distribution which is elaborated in [16]. into the system is not represented in Figure 2.
Basically, the working principle is similar to existing powered In order to illustrate the behaviour of the AMP-Foot 2.0
assistive devices, except that the actuator is working during prosthesis, the gait cycle is divided in 5 phases starting with
the complete stance phase. Gravitational potential energy is a controlled plantar flexion from heel strike (HS) to foot flat
gradually stored into a series elastic element, in time which (FF); see Figure 3. A step is initiated by touching the ground
the drive still has to provide the same torque and power with the heel. During this phase the foot rotates with respect
output. But there is approximately 3 times more time available to the leg, until 𝜃 (= 𝜙) reaches approximately −5∘ . This is
to generate the power by the electric motor. As a consequence, followed by a controlled dorsiflexion phase ending in push-
the overall power rating of the actuator can be divided by off at heel off (HO), during which a generation of propulsive
approximately the same amount, reducing the weight and size forces by Soleus and Gastrocnemius muscle groups will have
of the drive power considerably. place. In the late stance phase, the torque produced by the
The functional principle of the AMP-Foot 2.0 uses a ankle decreases until the leg enters the swing phase at toe off
“plantar flexion (PF)” spring, which stores energy from the (TO). Once the leg is engaged in the swing phase, the foot
controlled dorsiflexion phase of stance. An electric actuator is resets the locking mechanism.
Advances in Mechanical Engineering 3

Motor, gearbox Computational dynamics has grown in last year’s along


and ballscrew the need to develop simulations and analysis for mechanical
𝜃
systems that consist of interconnected bodies. Simulation
Leg is meant to describe and analyze the behaviour of various
system configurations, ask what-if questions about the real
system, and optimize the structural design.
Due to the high complexity of modern robotic systems,
PO spring
almost any research conducted in the area of robotics
k2
can benefit from a simulation of the system behaviour
L1 before experiments on a real platform take place [20]. The
Lever arm computational modelling including anatomic, physiologic,
L2 L3 A d
and engineering analyses serve to study various activities
c L4 in a normal and pathological condition of humans. Com-
puter simulations represent an effective, faster, and cheap-
PF spring a
k1
est approach than experiments, which necessarily consume
b
physical resources. Computer modelling is considered the
p 𝜙 most effective, when employed in combination with real
experiments.
Foot

Figure 2: Schematic representation of AMP-Foot 2.0.


3.1. Modelling Approach for AMP-Foot 2.0. A 3D model was
developed in conformity with the real mechanical properties
of AMP-Foot 2.0 device. In other words, all material and
The more detailed description of AMP-Foot 2.0 dynamic structural characteristics are preserved relative to the existing
behaviour during a complete gait cycle can be found under prototype. In order to model and simulate the AMP-Foot
contribution [16]. 2.0 prosthesis, MSC Adams 3D multibody dynamics software
was used [21]. The geometries of the AMP-Foot 2.0 compo-
2.2. Validation of the AMP-Foot 2.0 Device. The experiments nent parts were exported from Autodesk Inventor software
with participation of one disabled patient were effectuated in MSC Adams environment and were converted into a set
by the research group of Vrije Universiteit Brussel [18] of Adams/View geometry elements. This importing approach
and, in present paper, the authors are referring to results reduces the need to recreate geometry primitives within
obtained from those experiments. Briefly, in the performed Adams software and enhances the ability to realistically
experiments one transfemoral amputee subject of 75 kg was view the behavior of complicated mechanical systems. After
considered and the experiment was divided in three trials: (1) importing, the co-rrelated constraints between all geometry
walking at a self-selected speed on a treadmill; (2) walking parts were defined and applied to the model. Two spring
speed raised to a faster cadence; (3) an overground walking sets (PF and PO), which play a critical role in AMP-Foot
at self-selected speed; see Figure 4 [19]. 2.0 dynamic behaviour, have been modelled according to
Conform analysed data acquired during the experiments, the real design stiffness and damping characteristics. The
and compared to existing powered prosthetic devices, it was stiffness of the PF spring is about 300 N/mm and for every PO
observed that the AMP-Foot 2.0 prosthesis can improve an spring the value of stiffness is modeled as 60 N/mm. Damping
amputee gait, in conditions when little power is required values were determined experimentally, and were found to
for the actuation. Since the present contribution is focused be 10 Ns/mm for PF spring set and 1.2 Ns/mm for each PO
on modelling and simulation approach of human-robot spring respectively. The FF phase is considered as the initial
system, all additional details regarding the real experiment position of the model. In the beginning stage of this study the
design, prosthesis behaviour, power consumption, and torque model dynamic behaviour, see Figure 5, was analysed without
characteristics are given in [18]. human body model and was elaborated in [22]. The AMP-
Foot 2.0 model complexity can be appreciated and visualised
in Figure 5, where moving parts and constraints of the model
3. Modelling and Simulations in are presented. There are 11 degrees of freedom (DoF) in the
Virtual Environment AMP-Foot 2.0 system.
The model includes 2 actuation forces: one represents
In this section, the authors present modelling methods of the motor actuator, which is constantly pulling up the PO
the AMP-Foot 2.0 device. Then, modelling and simulation springs from HS to TO period of the gait cycle; the second
results of an amputee walking using the assistive device force is responsible for triggering the locking mechanism.
are described. Further, the comparison of a normal walking The control of the force actuation period is based on timing
(healthy person) and the same person wearing the AMP-Foot approach, implemented in the model in form of a STEP
2.0 is effectuated. The section is concluded with discussion on function.
obtained results from simulations and the real experiments, The STEP function approximates the Heaviside step func-
referred in Section 2.2. tion with a cubic polynomial and has the following format:
4 Advances in Mechanical Engineering

(HS) (FF) (Max DF) (HO) (TO)

Figure 3: Working principle of AMP-Foot 2.0 during a complete gait cycle.

150

Torque (Nm)
100 HO

50
FF
0 TO HC
−25 −20 −15 −10 −5 0 5 10 15
Angle (deg)
AMP-Foot 2 ankle angle data
AMP-Foot 2 ankle angle average
D. winter data
(a) (b)

Figure 4: Overground walking at self-selected speed (∼2.9 km/h).

The cable properties, such as density, Young’s modulus,


axial stiffness, and strain value, are defined with respect to
the ones used in real mechanical system and were determined
experimentally. These properties are presented in Figure 6(c)
and can be easily modified for further experimental purposes
within the framework of Graphical User Interface (GUI),
developed in MSC Adams software; see Figure 6(c); the mass
of the cable elements is assumed to be negligible.

Figure 5: The AMP-Foot 2.0 model composition and DoF. 3.2. Modelling and Simulation of Human Walking during
Wearing the AMP-Foot 2.0 Prosthesis. There are specialized
commercially available simulation tools that can be used for
analysis of human walking, such as AnyBody [23], OpenSim
STEP (𝑥, 𝑥0 , ℎ0 , 𝑥1 , ℎ1 ). It has continuous first derivatives. Its
[24], SIMM [25], and LifeMod [26]. However, these tools
second derivatives are discontinuous at 𝑥 = 𝑥0 and 𝑥 = 𝑥1 ,
cannot be used for analyzing human-robot interactions,
where 𝑥 is the independent variable (time, in present model),
which is becoming an essential requirement for modeling and
𝑥0 is a real variable that specifies the 𝑥 value at which the
simulation of robotic systems as their physical interaction
STEP function begins, 𝑥1 is a real variable that specifies the 𝑥
with humans becomes more complex. In general, the model-
value at which the STEP function ends, ℎ0 is the initial value
ing and simulation phase of the development of such systems
of the step, and ℎ1 is the final value of the step.
is becoming demanding. Since the robot and robot-human
The arrows, that can be observed in Figure 6, are repre-
interactions increase in complexity, the simulation with a
senting the actuation and tension forces during the simula-
single simulation tool is not effective anymore. Therefore,
tion.
in this research work the combination of MSC Adams with
Due to cables nonlinear geometric properties and a
LifeMode plug-in was used as the most appropriate.
complex static and dynamic behaviour, the challenge was to
model the cable segments which play an important role in
a force transmission system of the AMP-Foot 2.0. The cable 3.2.1. Methods. The creation of human models begins by
transmission system was modelled using TKC toolbox feature generating a base human segment set, followed by joints,
and the tension forces inside of the system were obtained [22]. soft tissues, and contact elements between the model and the
Advances in Mechanical Engineering 5

(FF) (Max DF) (HO) (TO) (Swing)


(a)

(b) (c)

Figure 6: (a, b) Simulation of AMP-Foot 2.0 behaviour. (c) Cable properties GUI.

(a) (b)

Figure 7: Joints representation and definition.

environment. In order to study the influences of AMP-Foot lower body (two legs and pelvis) which is rigidly attached
2.0 device on normal overground walking the mechanical to the right extremity AMP-Foot 2.0 prosthesis. Once the
model of the human body was built using the Lifemodeler segments of the lower body are established, joints are created
plug-in in framework of MSC Adams environment. This between the segments. Along an inverse-dynamics simu-
combination of tools supports the analysis of the “human- lation, joints learn angulation patterns while the model is
robot system” effectiveness and mutual interaction. The being driven by the motion capture data. The nominal joints
considered model, see Figure 8, includes a model of human’s stiffness for both legs at the hip, knee, and the ankle is
6 Advances in Mechanical Engineering

20.0

10.0

Angle (deg)
0.0

−10.0

−20.0

−30.0
0.0 0.5 1.0 1.5 2.0
Time (s)

Ankle angle during normal walking


AMP-Foot 2.0 ankle angle during normal walking

Figure 9: Ankle angle during normal walking of the same subject:


red graph with AMP-Foot 2.0; blue line using the healthy foot.
Figure 8: Model of human lower limbs cosimulated with AMP-Foot
2.0.

reference inputs for the joints’ PD controllers which are also


assumed to be 1e5 Ns/mm and the damping to be 1e4 Ns/mm. implemented in MSC Adams.
The joint consists of a triaxis hinge and passive or active For simulation, driving of the human body is based
forces acting on each of the three degrees of freedom. They on captured motion data obtained through marker-based
are implemented as an assembly of two “virtual” bodies motion capture system. Motion capture (MOCAP) systems
of negligible mass and inertia and three revolute joints; track the trajectories of markers attached at various locations
see Figure 7. For every joint there are three axes: sagittal, on the body. The marker trajectories are then used to train
transverse, and frontal. The Lifemode software offers the the human model. During the training, the response of the
possibility to define every axis in a way suitable to certain body is recorded and later used for a forward dynamics
application. The axes of lower limb joints in present human simulation. Marker trajectory data is used to drive elements
model are defined as in the table in Figure 7. called “motion agents,” which are massless parts fixed to the
The settings specified for every joint axis in the model are body segments using spring elements.
interpreted as follows. The right foot part was removed and replaced by the
AMP-Foot 2.0 model and connected to the right lower leg
Driven. Kinematically driven using data from a driv- by a fixed joint; see Figure 8. Then, contact forces between
ing spline. prosthesis toe, heel parts, and the ground were defined as an
Passive. A torsional spring force with user-specified IMPACT function model. The IMPACT function represents
stiffness, damping, angular limits, and limit stiffness a simple model for contacts. It evaluates a force that turns
values. These joints are used in an inverse dynamics on when a distance falls below a nominal free length (when
analysis to record the joint angulations while the two parts collide). The force has two components: a spring
model is being manipulated with motion agents. or stiffness component and a damping or viscous compo-
nent. The stiffness component opposes the penetration. The
Hybrid III. The Hybrid III strength model is created
damping component of the force is a function of the speed of
for the individual joint axis with a user-specified scale
penetration and opposes the direction of relative motion.
value. The Hybrid III strength model is based on
physical measurements of an actual crash dummy.
The strength model consists of nonlinear stiffness, 3.2.2. Results. The obtained simulation data were filtered
damping, and frictional values and also includes joint by low pass filter and processed accordingly. In Figure 9
limit stop stiffness with hysteresis. the ankle angle data of the human body model with the
Servo. This selection creates a trained PD-servo type same physical properties as in real experiment are illustrated
controller on the joint axis. The joint is commanded (weight 75 kg, height 175 cm, male) during normal over-
to track an angular history spline with a user-specified ground walking. In the first simulation the human was
gain on the error between the actual angle and the walking without AMP-Foot 2.0 prosthesis, using a normal
commanded error. A user-specified derivative gain is human foot model (see Figure 9, blue dot-line).
specified to control the derivative of the error. Afterwards, the normal foot was removed and replaced by
the AMP-Foot 2.0 model and, therefore, we can follow from
It is important to notice that the mass (75 kg) of the whole the graph (see red line representation) the consequences for
human body is considered in simulation, although just the this change in gait kinematics. Please notice that in the first
lower part is represented. The human body model generated slot of time the AMP-Foot 2.0 prosthesis is passive; there is
by the LifeMod plug-in stores the joint motion trajectories in no force acting on the motor part and pulling the spring sets.
MSC Adams. These trajectories are used during simulation as The single actuation is coming from the human leg during
Advances in Mechanical Engineering 7

Torque data right leg 30.0


80000 25.0
70000 20.0
60000
Torque (N-mm)

15.0
50000 10.0

Angle (deg)
40000 5.0
30000 0
20000 −5.0
10000 −10.0
0.0
0.0 0.5 1.0 1.5 2.0 −15.0
Time (s) −20.0
−25.0
Torque right Hip 0 10 20 30 40 50 60 70 80 90 100
Torque right Knee Stride (%)
Torque ankle AMP Foot 2.0
Figure 12: Ankle angle comparison: red line: simulation of human
Figure 10: Torque values for right limb. model during wearing the AMP-Foot 2.0; blue line: simulation of
human model without AMP-Foot 2.0; green line: real experiment
data of subject during wearing the AMP-Foot 2.0 device.
60.0
50.0
40.0
Angle (deg)

30.0 reaction forces are pushing up the AMP-Foot 2.0 prosthesis,


20.0
by this way creating an additional torque motion.
10.0
0.0 4. Discussion
−10.0
0.0 0.5 1.0 1.5 2.0 It is crucial for design and control developing of assistive
Time (s) devices to have a model of the robotic system interacting
with the human before the actual physical system is ready
Figure 11: Knee angle: red line, with AMP-Foot 2.0 model; grey line, for the use. Sometimes, performing the real experiments is
with human foot model. too exhausting, expensive, or almost impossible and in this
case modeling and simulation approach is considered very
effective. In the framework of this paper the authors have
walking only. In the grey zone, see Figure 9, the AMP-Foot 2.0 researched in virtual environment the differences between
becomes actuated and, as a result, we can observe an increase human walking during wearing an assistive device and
of the ankle angle, due to the provided push off propulsion normal walking of the healthy person. In both simulation and
force of the prosthetic device. real-life testing the same human model was considered. This
In Figure 10 are illustrated the obtained torque values experiment is impossible to be repeated in real life. It was
of the right lower limb with the AMP-Foot 2.0 prosthesis expected to have some differences in the ankle angle data,
attached. but the main conclusion is that the human walking is not
Also, a slight difference was observed at the knee level affected in negative way while wearing the AMP-Foot 2.0. The
joint. In case the AMP-Foot 2.0 is attached to the human powered device is providing the human leg with additional
leg model, the authors noticed a small knee angle increase; propulsion force that helps locomotion.
see Figure 11. This occurrence can be explained by the lower If we will refer to real experiments, performed with
compliancy in the TO moment. Then the locking mechanism an amputee person with comparison to results obtained
opens, providing the additional propulsion force to the leg. from simulations with human model during wearing the
As was discussed before, the authors were interested to AMP-Foot 2.0, we can notice many common similarities in
make a comparison between data from simulations with the walking pattern. Even, if the gait kinematics of people is
data, based on real experiments; see Figure 12. individual, (in simulations a similar subject with the same
Conform Figure 12, the time-based data on level ground physical characteristics was considered) the obtained results
walking, one can observe the similarities between the angle remain valid and can be interpreted. If one will do the
ankle data of the human model wearing the AMP-Foot 2.0 comparison between the normal walking simulation and
device and the walking pattern of real amputee subject with both experiments which uses the AMP-Foot 2.0 prosthesis
attached prosthesis. It can be noticed for the human model (Figure 12), one will observe some difference in TO moment.
with the AMP-Foot 2.0 that there is an increase of the ankle In case of normal walking without the device the transaction
angle around HO moment. This fact can be the result of the of the foot from TO to a swing moment is more compliant. In
influence of contact forces between the heel and the ground case of simulations there can be other external factors that can
part. In other words, when the heel is touching the ground the slightly influence obtained data results, such as the simplified
8 Advances in Mechanical Engineering

contact definition (which is not so complex as in case of real [6] H. Bateni and S. J. Olney, “Kinematic and kinetic variations of
foot-ground contact). below-knee amputee gait,” Journal of Prosthetics and Orthotics,
vol. 14, no. 1, pp. 2–10, 2002.
[7] D. A. Winter, “Biomechanical motor patterns in normal walk-
5. Conclusions ing,” Journal of Motor Behavior, vol. 15, no. 4, pp. 302–330, 1983.
The main focus in the framework of the presented study [8] A. H. Hansen, D. S. Childress, S. C. Miff, S. A. Gard, and K.
P. Mesplay, “The human ankle during walking: implications for
was to inspect how the attached trans-tibial prosthesis will
design of biomimetic ankle prostheses,” Journal of Biomechan-
influence normal healthy human walking kinematics and ics, vol. 37, no. 10, pp. 1467–1474, 2004.
how close the AMP-Foot 2.0 working behaviour comes to
[9] S. K. Au, P. Dilworth, and H. Herr, “An ankle-foot emulation
a real human foot subsystem. The effectiveness of human- system for the study of human walking biomechanics,” in
robot interaction was analyzed during the tests with a real Proceedings of the IEEE International Conference on Robotics and
subject and in virtual environment. The authors admit that it Automation (ICRA ’06), pp. 2939–2945, Orlando, Fla, USA, May
is difficult to compare individuals having a different walking 2006.
pattern, even if for simulation a subject with the same physical [10] D. A. Winter and S. E. Sienko, “Biomechanics of below-knee
parameters as the one tested in real experiment was consid- amputee gait,” Journal of Biomechanics, vol. 21, no. 5, pp. 361–
ered. Moreover, the simulation comparison (subject walking 367, 1988.
with and without AMP-Foot 2.0) is almost impossible to [11] G. K. Klute, J. Czerniecki, and B. Hannaford, “Development
arrange in real life. This fact that the simulation results more of powered prosthetic lower limb,” in Proceedings of the 1st
interesting and significant. National Meeting, Veterans Affairs Rehabilitation R&D Service,
Thanks to the built human-prosthesis model further Washington, DC, USA, October 1998.
improvements on prototype design parameters can be [12] S. K. Au, Powered ankle-foot prosthesis for the improvement of
achieved and the prototype can be adapted to certain subject’s amputee walking economy [Ph.D. thesis], MIT Department of
individual properties. The result data demonstrate that AMP- Mechanical Engineering, Cambridge, Mass, USA, 2007.
Foot 2.0 prosthesis can successfully replace the human ankle- [13] J. K. Hitt, R. Bellman, M. Holgate, T. G. Sugar, and K.
foot subsystem, provided some individual adjustments are W. Hollander, “The SPARKy (spring ankle with regenerative
kinetics) project: design and analysis of a robotic transtibial
made.
prosthesis with regenerative kinetics,” in Proceedings of the 2007
ASME International Design Engineering Technical Conferences
Conflict of Interests and Computers and Information in Engineering Conference
(IDETC/CIE ’07), pp. 1587–1596, Las Vegas, Nev, USA, Septem-
The authors, Grosu Svetlana, Pierre Cherelle, Chris Verheul, ber 2007.
Bram Vanderborght, and Dirk Lefeber, declare that there is no [14] R. Versluys, G. Lenaerts, M. Van Damme et al., “Successful
conflict of interests regarding the publication of this paper. preliminary walking experiments on a transtibial amputee
fitted with a powered prosthesis,” Prosthetics and Orthotics
International, vol. 33, no. 4, pp. 368–377, 2009.
Acknowledgments [15] R. Versluys, P. Beyl, M. Van Damme, A. Desomer, R. Van Ham,
and D. Lefeber, “Prosthetic feet: state-of-the-art review and
This work has been partially supported by the European
the importance of mimicking human anklefoot biomechanics,”
Commission 7th Framework Program as part of the CORBYS Disability and Rehabilitation: Assistive Technology, vol. 4, no. 2,
(Cognitive Control Framework for Robotic Systems) project pp. 65–75, 2009.
under Grant Agreement no. FP7 ICT-270219. [16] P. Cherelle, A. Matthys, V. Grosu, B. Vanderborght, and D.
Lefeber, “The AMP-Foot 2.0: Mimicking intact ankle behavior
References with a powered transtibial prosthesis,” in Proceedings of the
IEEE International Conference on Biomedical Robotics and
[1] O. Khatib, K. Yokoi, O. Brock, K. Chang, and A. Casal, “Robots Biomechatronics, pp. 544–549, June 2012.
in human environments: basic autonomous capabilities,” Inter- [17] G. van Oort, R. Carloni, D. J. Borgerink, and S. Stramigioli,
national Journal of Robotics Research, vol. 18, no. 7, pp. 684–696, “An energy efficient knee locking mechanism for a dynamically
1999. walking robot,” in Proceedings of the IEEE International Confer-
[2] A. de Santis, B. Siciliano, A. de Luca, and A. Bicchi, “An atlas ence on Robotics and Automation, pp. 2003–2008, May 2011.
of physical human-robot interaction,” Mechanism and Machine [18] P. Cherelle, V. Grosu, A. Matthys, B. Vanderborght, and D.
Theory, vol. 43, no. 3, pp. 253–270, 2008. Lefeber, “Design and validation of the ankle mimicking pros-
[3] M.-J. Hsu, D. H. Nielsen, S.-J. Lin-Chan, and D. Shurr, “The thetic (AMP-) foot 2.0,” Transaction on Neural Systems and
effects of prosthetic foot design on physiologic measurements, Rehabilitation Engineering, 2013.
self-selected walking velocity, and physical activity in people [19] http://www.youtube.com/watch?v=tXLQq9a5kyI.
with transtibial amputation,” Archives of Physical Medicine and [20] J. G. de Jalón and E. Bayo, Kinematic and Dynamic Simulation
Rehabilitation, vol. 87, no. 1, pp. 123–129, 2006. of Multibody Systems: The Real-Time Challenge, Springer, New
[4] A. Esquenazi and R. DiGiacomo, “Rehabilitation after amputa- York, NY, USA, 1994.
tion,” Journal of the American Podiatric Medical Association, vol. [21] http://www.mscsoftware.com/.
91, no. 1, pp. 13–22, 2001. [22] S. Grosu, P. Cherelle, C. Verheul, B. Vanderborght, and D.
[5] S. Ron, Prosthetics and Orthotics: Lower Limb and Spinal, Lefeber, “AMP-Foot 2.0 prosthesis dynamic behavior, prelim-
Lippincott Williams & Wilkins, Baltimore, Md, USA, 2002. inary computational multibody dynamics simulation results,”
Advances in Mechanical Engineering 9

in Proceedings of the ECCOMAS Thematic Conference on Multi-


body Dynamics, Zagreb, Croatia, July 2013.
[23] AnyBody, http://www.anybodytech.com.
[24] OpenSim, https://simtk.org.
[25] SIMM, http://www.musculographics.com.
[26] LifeMod, http://www.lifemodeler.com.
Hindawi Publishing Corporation
Advances in Mechanical Engineering
Volume 2014, Article ID 957684, 8 pages
http://dx.doi.org/10.1155/2014/957684

Research Article
Controller Parameters Tuning Based on Transfer Matrix Method
for Multibody Systems

Hossam Hendy, Xiaoting Rui, Qinbo Zhou, and Mostafa Khalil


Institute of Launch Dynamics, Nanjing University of Science and Technology, Nanjing 210094, China

Correspondence should be addressed to Hossam Hendy; hosam hendi@yahoo.com

Received 25 August 2013; Accepted 9 December 2013; Published 5 January 2014

Academic Editor: Caishan Liu

Copyright © 2014 Hossam Hendy et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Transfer matrix method for multibody systems (MS-TMM) is a rife method to multi-rigid-flexible-body systems dynamics model
deduction due to that there are no needs to establish the global dynamics equations of the system. Its basic idea is transferring a
state vector between the body input(s) and output(s); this idea is close to the linear theories in control analysis and design. In this
paper, three controllers’ parameters tuning techniques for the proposed system model using MS-TMM are utilized; one technique is
applied to get the stability regions via the frequency response of MS-TMM derived model. Another technique considers a classical
PID controller design through the analysis of step input response of the system, and the last technique can be applied in both time
and frequency domains if the model has a known mathematical model. A car suspension system is considered to represent modeling
and tuning problems. In-depth study of MS-TMM with control techniques and defining the controllers’ parameters stability regions
provide an opportunity to formulate a relationship between MS-TMM and control design for novel control applications due to the
powerful strength of MS-TMM dealing with more complex problems of the controlled multibody systems.

1. Introduction based on PID control and [5] indicated that more than 97%
of regulatory controllers utilize the PID algorithm. But the
With the increments of complexity of multibody systems tuning of the controller gains is a problem because many
and the development of their design and control methods, industrial models suffer some burdens such as nonlinearities,
the need for more elegant formulations of the equations higher order, and time delay [6]. The basic idea of MS-TMM
of motion becomes an issue of paramount importance. is transferring the state vector between the body input(s)
Many methods and theories for developing the model of and output(s); this idea is close to linear theories in control
the multibody system dynamics and control are presented analysis and design. The control and feedback variables could
for such reasons. In transfer matrix method for multibody be considered in the transfer equation of MS-TMM. The
systems (MS-TMM) there are two cases to deal with control tuning of the controller parameters is a great issue to get a
element, one is to express the control force with state of stabilized system. However Ziegler-Nichols technique is one
system of previous time such as the delay control, and the of the simplest techniques of PID tuning; there are several
second is that the control force is relative to present state of disadvantages, such as many trials are needed to find the
system, such as real-time control systems. Rui et al. have trials stability limits and also not all tuning values work efficiently
to develop new controlled systems depending on deriving for all situations [7–10]. Shamsuzzoha and Skogestad (2010)
the dynamics equations using MS-TMM [1, 2]. Bestle et al. developed a new procedure for PI/PID tuning method in
reformulated the car suspension system using MS-TMM as closed-loop mode, only one closed-loop step test is required
similar as classical control theory’s structural diagram [3]. to deduce PI controller values. In this method also it is simple
Proportional-Integration-Differentiation (PID) control- to obtain the PID tuning parameters in one step [11, 12]. This
ler is prevalent in industry applications. Reference [4] indi- paper is organized as follows: the proposed system is modeled
cated that more than 90% of feedback control loops are by MS-TMM including the control law in Section 2. The
2 Advances in Mechanical Engineering

0 Disturbance X4,0
𝒁4,0
m1
𝑼g1

K2 fc
𝑼m1
C2
𝑼g
m3
x fc + 𝒁2,1 Output X1,0
r=0 + 𝑼m
C(s) 𝑬c
+ 𝒁1,0
K4 − X1,0

𝑬m
0

Figure 1: Dynamic model of controlled car. Figure 2: The structural diagram of the controlled system [3].

1 0
where U𝑗 = [ 𝑚𝑗 Ω2 1 ] and Ω is the external excitation
design of PI controller values stability region that achieves a
specified gain and phase margins, in addition to tuning the frequency.
controller parameters using different techniques, is presented For the linear spring 4, the transfer equation is
in Section 3. Simulation and results of car active suspension
Z4,𝑂 = U4 Z4,𝐼 (2)
system model with the designed controllers are provided in
Section 4. Conclusions are offered in Section 5.
with transfer matrix U4 = [ 10 −1/𝐾
1
4 ].

For the controlled element 2, the transfer equation should


2. Problem Formulation be
A two-degree-of-freedom car dynamic model is established Z2,𝑂 = U2 Z2,𝐼 + E𝑐 𝐹𝑐 , (3)
to simulate the system as shown in Figure 1. It does not only
simplify the system design and analysis, but also represents
where U2 = [ 10 −(1/(𝐾21+𝑖𝐶2 Ω)) ], E𝑐 = [ 1/(𝐾2 +𝑖𝐶2 Ω) ], and 𝐹 is
𝑐
most of full car features [13]. In Figure 1, 𝑚1 is the sprung 0
the control force.
mass that represents the car body, 𝑚3 is the unsprung mass
Considering the negative PID feedback control and
which stands for the wheel, 𝐶2 is the damping coefficient,
regarding 𝑥1,0 as the measurement signal, the control force
and 𝐾2 and 𝐾4 are the springs stiffness. In this section,
takes the form
this model is deduced using MS-TMM. According to MS-
TMM, the system is portioned into many elements which are 𝑡 𝑑𝑥1,0
numbered as 1 to 4 from up to down, where body 1 and 3 𝑓𝑐 = − (𝑘𝑃 𝑥1,0 + 𝑘𝐼 ∫ 𝑥1,0 𝑑𝑡 + 𝑘𝐷 ). (4)
0 𝑑𝑡
are lumped masses, element 4 is a linear spring, while the
controlled element 2 includes spring and damper in parallel By transformation 𝑥 = 𝑋𝑒𝑖Ω𝑡 , 𝑓𝑐 = 𝐹𝑐 𝑒𝑖Ω𝑡 , it yields
connections. The boundaries at two ends are noted as 0. The
positive direction of 𝑥 axis is also shown in the figure. It 𝑘𝐼
should be mentioned that the controlled force 𝑓𝑐 serves as an 𝐹𝑐 = − (𝑘𝑃 + + 𝑖Ω𝑘𝐷) 𝑋1,0 =: −𝐶 (𝑖Ω) 𝑋1,0 . (5)
𝑖Ω
internal force of the system.
The transfer direction is stipulated as the positive direc- Meanwhile there should be
tion of 𝑥 axis. The state vector of an arbitrary connection
point of the system is defined as Z𝑖,𝑗 = [𝑋, 𝑄𝑥 ]𝑇𝑖,𝑗 , where 𝑖 and 𝑋1,0 = [1 0] [
𝑋
] =: E𝑚 Z1,0 . (6)
𝑗 stands for the sequence number of the body element and 𝑄𝑥 1,0
hinge element, respectively. Thereby, the transfer equation of
each element is discussed as follows: Regarding the transfer direction and (1)∼(3) and (5)∼(6), the
Considering the rise and fall of the road, the boundary structure diagram of the controlled system can be drawn in
end 𝑥4,0 is assumed to be in the form of 𝑥4,0 = 𝐴𝑒𝑖Ω𝑡 , where Figure 2.
𝐴 is the complex amplitude and Ω is the frequency of the From Figure 2, the overall transfer equation of the con-
movement. Thus, in a steady-state sense, the whole system trolled system can be written as
experiences the forced vibration in the sine form.
According to MS-TMM for the lumped mass 1 and 3, their Z1,0 = U1 Z2,1
transfer equations are = U1 (U2 U3 U4 Z4,0 + E𝑐 𝐹𝑐 ) (7)
Z𝑗,𝑂 = U𝑗 Z𝑗,𝐼 , (𝑗 = 1, 3) , (1) = U1 (U2 U3 U4 Z4,0 − E𝑐 𝐶 (𝑖Ω) E𝑚 Z1,0 ) .
Advances in Mechanical Engineering 3

Table 1: Elements values of controlled car. 20

Magnitude (dB)
0
Name Value Company −20
𝑚3 50 kg −40
𝑚1 250 kg −60
−80
𝐾2 16 kN/m −100
𝐾4 160 kN/m 0
𝐶2 −45

Phase (deg)
1500 N⋅s/m −90
−135
−180
which finally ends up with −225
−270
−1
Z4,0 = (U1 U2 U3 U4 ) [I2 + 𝐶 (𝑖Ω) U1 E𝑐 E𝑚 ] Z1,0 . (8) 100 101 102 103
Frequency (rad/s)
This can be rewritten considering the boundary condition of
Classical
the system Z1,0 = [ 𝑋0 ]1,0 , Z4,0 = [ 𝑄𝐴𝑥 ]4,0 as MS-TMM

𝐴 𝑈 𝑈 𝑋 Figure 3: Bode plots for MS-TMM and classical method.


[ ] = [ 11 12 ] [ ] . (9)
𝑄𝑥 4,0 𝑈21 𝑈22 0 1,0
Thus the frequency response function from the disturbance
to the output can be read as
3. Controller Parameters Tuning
𝑋1,0 1 This section illustrates different approaches for designing
= . (10) both PI and PID controllers and some methods for tuning
𝐴 𝑈11
their parameters. Considering the control scheme shown in
Setting 𝑘𝑃 = 𝑘𝐼 = 𝑘𝐷 = 0, one can acquire the formula for the Figure 2, at the beginning a PI controller is designed; from
uncontrolled system. (5) the designed PI controller will be at the form of
From the classical approaches for modeling the transfer
function of a passive suspension system using ordinary dif- 𝑘𝐼 𝑘𝑃 𝑠 + 𝑘𝐼
𝐶 (𝑠) = 𝑘𝑃 + = . (13)
ferential equations, the equations which describe the system 𝑠 𝑠
can be found in [14], and the transfer equation is From the previous literature reviews, instability is the disad-
𝐾4 𝐶2 𝐾 vantage of feedback control system. There is a risk that the
𝐺 (𝑠) = (𝑠 + 2 ) closed-loop system becomes unstable when using feedback.
𝑚3 𝑚1 𝐶2
Thus, analyzing the closed-loop stability of system is essential
𝐶2 𝐶2 3 requirement for feedback control system. The mean obstacle
× (𝑠4 + ( + )𝑠 is to indicate the controller parameters or gains stabilizing
𝑚3 𝑚1
region [16].
𝐾2 𝐾2 𝐾4 2 𝐾4 𝐶2 𝐾 𝐾 −1 Decomposing (12) numerator and denominator to their
+( + + )𝑠 + 𝑠+ 4 2) . even and odd parts, the equation can be rewritten as
𝑚3 𝑚1 𝑚3 𝑚3 𝑚1 𝑚3 𝑚1
(11)
𝑁𝑒 (−Ω2 ) + 𝑗Ω𝑁𝑜 (−Ω2 )
The described system elements values are given in Table 1 𝐺 (𝑗Ω) = . (14)
𝐷𝑒 (−Ω2 ) + 𝑗Ω𝐷𝑜 (−Ω2 )
[15].
For the uncontrolled case, Bode plots for both classical From the control scheme diagram shown in Figure 2, and
method and MS-TMM results are shown in Figure 3, consid- considering the control law in (13), the closed-loop character-
ering the elements values given in Table 1 and substituting istic polynomial derived in (10) can be separated and solved
into (10) after eliminating the controller gains, the transfer to zero; then the control parameters can be written as [17]
equation of MS-TMM system as a function of the excitation
𝑋 (Ω) 𝑈 (Ω) − 𝑌 (Ω) 𝑅 (Ω)
frequency Ω is 𝑘𝑃 = , (15)
𝑄 (Ω) 𝑈 (Ω) − 𝑅 (Ω) 𝑆 (Ω)
𝐺 (𝑗Ω) = (𝐾4 𝐾2 + (𝑗Ω) 𝐶2 𝐾4 )
𝑌 (Ω) 𝑄 (Ω) − 𝑋 (Ω) 𝑆 (Ω)
𝑘𝐼 = , (16)
× (𝑚3 𝑚1 Ω4 − 𝑗Ω3 𝐶2 (𝑚3 + 𝑚1 ) 𝑄 (Ω) 𝑈 (Ω) − 𝑅 (Ω) 𝑆 (Ω)
(12)
where
− Ω2 (𝐾2 𝑚3 + 𝐾2 𝑚1 + 𝐾4 𝑚3 )
−1
𝑌 (Ω) = −Ω𝐷𝑒 (−Ω2 ) , 𝑄 (Ω) = −Ω2 𝑁0 (−Ω2 ) ,
+ (𝑗Ω) 𝐾4 𝐶2 + 𝐾4 𝐾2 ) .
This equation represents the open loop uncontrolled part of 𝑋 (Ω) = Ω2 𝐷0 (−Ω2 ) , 𝑆 (Ω) = Ω𝑁𝑒 (−Ω2 ) , (17)
the system enclosed in the solid blue line in Figure 2; results
from Figure 3 indicate that both techniques are identical. 𝑈 (Ω) = Ω𝑁𝑜 (−Ω2 ) , 𝑅 (Ω) = 𝑁𝑒 (−Ω2 ) .
4 Advances in Mechanical Engineering

140 30
Stability boundary for 25
120
M = 1 and 𝜑 = 0
20
100 Stability boundary for
15
M = 1 and 𝜑 = 0

KI
80
10
KI

60 5 Stability boundary for


Stability boundary for M = 1 and 𝜑 = 60
0
40
M = 1 and 𝜑 = 45 w = 31.5 KI = 0 w = 41.5
−5
20 −1 0 1 2 3 4 5
KP
w = 57
0
KI = 0 w = 41.5 Figure 5: Stability region for two cases.
−20
−1 0 1 2 3 4 5 6
KP 10

Magnitude (dB)
5
Figure 4: Stability region for frequency range Ω → [0, 56.7]. 0
−5
−10
−15
−20
−25
To determine the missing expressions in (15), (16), and (17), 0
the transfer function in (12) can be divided into odd and
Phase (deg)

−45
even portions for both numerator and denominator, and −90
−135
after the comparison with (14), the following variables can be −180
substituted as −225
10−1 100 101
Frequency (rad/s)
𝑁𝑒 (−Ω2 ) = 𝐾1 𝐾3 , 𝑁𝑜 (−Ω2 ) = 𝐾1 𝐶3 ,
PI controller
𝐷𝑒 (−Ω2 ) = 𝑚2 𝑚4 Ω4 Uncontrolled
(18) Figure 6: Bode plot for the controlled and uncontrolled system.
− (𝐾1 𝑚4 + 𝐾3 𝑚2 + 𝐾3 𝑚4 ) Ω2 + 𝐾1 𝐾3 ,

𝐷𝑜 (−Ω2 ) = −𝐶3 (𝑚2 + 𝑚4 ) Ω2 + 𝐾1 𝐶3 .


Substituting (17) and (18) into (15) and (16) and consider-
ing (19) yields to
For the controller design problems which based on gain and
phase margins design, it is inevitable to estimate the stabi- 𝑁𝐴 cos [𝜑] + 𝑁𝐵 sin [𝜑]
𝐾𝑃 = ,
lizing region of the controller gains. The stability boundary −𝑀 (𝑁𝑒2 + Ω2 𝑁𝑜2 )
locus in proportional and integral gains plane can be obtained (20)
from (15) and (16) after substituting the missing terms from Ω𝑁𝐵 cos [𝜑] − Ω𝑁𝐴 sin [𝜑]
𝐾𝐼 = ,
(17) and (18), then verifying the presented model using the −𝑀 (𝑁𝑒2 + Ω2 𝑁𝑜2 )
table of contents in Table 1.
The region of stability of the proportional gain 𝑘𝑃 and where
the integral gain 𝑘𝐼 that cause the closed-loop polynomial
to satisfy Hurwitz stability test is shown in Figure 4, where 𝑁𝐴 = (Ω2 𝑁𝑜 𝐷𝑜 + 𝑁𝑒 𝐷𝑒 ) , 𝑁𝐵 = Ω (𝑁𝑜 𝐷𝑒 − 𝑁𝑒 𝐷𝑜 ) .
line 𝑘𝐼 = 0 divides the controller gains plane into stable and (21)
unstable regions.
For designing of controller to meet a certain demanded Using these formulae makes it easy to satisfy certain con-
gain and phase margins which are important frequency ditions and find the stabilization region for the controller
domain requirements for user specifications, consider a parameters. Assuming two cases, one as the phase margin is
control function: greater than 45∘ and gain margin is greater than 1, another
as the phase margin is greater than 60∘ and gain margin is
greater than 1.
𝐺𝑐 (𝑗Ω) = 𝑀𝑒−𝑗𝜑 , (19) Figure 5 shows two regions of stabilities for two different
conditions. The blue circles locus shows the stability region
for a demanded conditions gain margin =1 and phase margin
where 𝑀 and 𝜑 are gain and phase margins, respectively. 45∘ , the maximum frequency for this region is 41.5 rad/sec,
Advances in Mechanical Engineering 5

90 1.6
80 1.4
70
Phase margin (deg)

1.2
60
1

Amplitude
50
40 0.8
30 0.6
20 0.4
10 0.2
0
0 0.5 1 1.5 2 2.5 3 0
0 0.5 1 1.5 2 2.5 3 3.5
Damping ratio
Time (s)
Figure 7: Phase margin versus damping ratio.
KP = 1; KI = 7.5
KP = 1; KI = 25
Passive susp.
the red solid locus shows another design conditions with
gain margin =1 and phase margin 60∘ , and the maximum Figure 8: Step response of the system.
frequency for this region is 31.5 rad/sec.
It is an applicable method to achieve user specified
gain and phase margins to define the values of controller Returning back to (10) and substituting 𝑗Ω = 𝑠, another
parameters with the ability to justify the maximum frequency technique is applied in this section; a developed methodology
without the need of sweeping over the parameters and also by Shamsuzzoha and Skogestad requires only one closed-loop
without using programming for solving inequality problems. step test to obtain PI controller setting. In this method it is
This method can be expanded to estimate the stabilizing also simple to obtain the PID tuning parameters [11, 12].
region of PID controller gains as shown in [17]. Considering an ideal PID controller in the form, as shown
From Figure 6, the plotted line in red solid color indicates in (5)
the output response of the uncontrolled system and also
shows the stability region of frequencies shown in red points 1 𝐾
𝑐 (𝑠) = 𝐾𝑐 (1 + + 𝜏𝐷𝑠) ⇐⇒ 𝑐 (𝑠) = (𝐾𝑃 + 𝐼 + 𝐾𝐷𝑠) .
in the range Ω ∈ [10.2, 15.6] rad/sec, while the blue dashed 𝜏𝐼 𝑠 𝑠
locus indicates the system response after applying the PI (24)
controller which is chosen from the stability region shown in
Figure 6 as 𝐾𝑃 = 1 and 𝐾𝐼 = 2.5; it is obvious that the stability 𝐾𝑐 is the controller gain and 𝜏𝐼 , 𝜏𝐷 are the controller integral
region of frequencies shown in blue squares is expanded in and derivative time, respectively.
the range Ω ∈ [0, 15.5] rad/sec. The defined parameters can be calculated using the
The relationship between the phase margin Φ𝑀, the following basic formulae:
overshoot (OS), and the damping ratio 𝜁 can be expressed as:
2𝜏 + 𝜃 𝜃
𝐾𝑐 = , 𝜏𝐼 = min {(𝑟 + ) , 8𝜃} ,
√−2𝜁2 + √1 + 4𝜁4 3𝑘𝜃 2
Φ𝑀 = 90∘ − tan−1 , (22) (25)
2𝜁 𝑟𝜃
𝜏𝐷 = ,
2𝜏 + 𝜃
where
where 𝑘 is the gain, 𝜏 is the time lag constant, and 𝜃 is
√1−𝜁2 )
OS% = 𝑒−(𝜁𝜋/ × 100%. (23) the time delay. The controller gain is only depending on
the overshoot, while the integral and derivative times are
The relation between Φ𝑀 and 𝜁 is plotted in Figure 7. function of the system peak time. Applying this approach, the
So it is noticeable to get different overshoot values due to tuning results simulation for the designed problem assuming
the changes of the phase margins with the controller design different values for 𝜏/𝜃 is indicated in Table 2.
parameters; Figure 9 shows the effect of changing these values An improved definition of (25) parameters were con-
as follow. cluded in Shamsuzzoha et al. [12], a closed-loop test with
In Figure 8 the red dash-dot line indicates the uncon- P-controller was applied to adjust overshoot value of 0.3021
trolled system step response; the overshoot is about 46%, and for P-only control then assigned the other parameters which
choosing a test point from Figure 3 with values of 𝐾𝑃 = 1 leaded to gain 𝐾𝑐 = 2.5 with 𝜏𝐼 = 0.2040 and peak time 0.153
and 𝐾𝐼 = 25, it is clear that it locates outside the regions for the controller gains.
of stabilities from the used technique, so it is clear from the In spite of Zeigler-Nichols methods’ drawbacks which
figure that the green solid line response overshoot value is are mentioned in previous sections, it is still one of the
about 55%, and unlike the line in blue dot color it is clear that easiest methods for determining the proportional gain 𝐾𝑃 ,
the overshoot value is improved to be 7% when the test point integral time 𝑇𝑖 , and derivative time 𝑇𝑑 based on the transient
is chosen as 𝐾𝑃 = 1 and 𝐾𝐼 = 7.5. response of a given system. Zeigler-Nichols suggested the
6 Advances in Mechanical Engineering

Table 2: Controller tuning parameters. Step response


1.5
System: Z-N PID
Controller parameters 1.4 Time (s): 0.0789 System: passive susp.
𝜏/𝜃 Amplitude: 1.33 Time (s): 0.33

𝐾𝑃 𝐾𝐼 𝐾𝐷
Amplitude: 1.45
1.3
0.1 0.4000 0.6667 0.0333 1.2 System: PI

Amplitude
Time (s): 0.23
1.1 Amplitude: 1.08
0.5 0.6667 0.6667 0.1667 System: shass PID
Time (s): 0.158
1 Amplitude: 1.08
2.5 2.0000 0.6667 0.8333
0.9
10 7.0000 0.8750 3.3333
0.8
20 13.6667 1.7083 6.6667 0.7
50 33.6667 4.2083 16.6667 0.6
100 67.0000 8.3750 33.3333 0.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time (s)
Table 3: Zeigler-Nichols tuning rules. Passive susp. Shass PID
Z-N PID PI
Type of controller 𝐾𝑃 𝑇𝑖 𝑇𝑑
P 0.5𝐾cr ∞ 0 Figure 9: Step response of the system.
PI 0.45𝐾cr 0.83 𝑃cr 0
PID 0.6𝐾cr 0.5 𝑃cr 0.125 𝑃cr Table 5: System response analysis parameters.

Peak time Peak value Overshoot %


Marker
Table 4: Zeigler-Nichols tuning values. 𝑇𝑝 𝑀𝑝 O.S%
Passive
Type of controller 𝐾𝑃 𝑇𝑖 𝑇𝑑 0.336 sec 1.45 45%
suspension
P 2.55 ∞ 0 Z-N PID 0.0789 1.33 33% x x

PI 2.295 0.091976 0 Shamsuzzoha


0.158 1.08 8%
PID 3.06 0.036568 0.013852 PID
PI 0.23 1.08 8%

values of the controller parameters according to the shown response for a vehicle coming out of a pothole can be
formulae in Table 3. simulated as a step input with amplitude of 0.08 as [18]:
For the systems which have a mathematical model, root-
locus method can be used to find the critical gain 𝐾cr and the 0 󳨐⇒ 𝑡 ≤ 0,
sustained oscillation frequency 𝜔cr ; then get 𝑃cr = 2𝜋/𝜔cr . Amp = { (26)
0.08 󳨐⇒ 𝑡 ≥ 0.
Also the systems’ open loop system (passive suspension) and
closed-loop system (Active suspension) can be analyzed in The system output displacement is shown in Figure 10.
frequency domain to find the bandwidth, gain, and phase Figure 10 shows that the peak value of the passive suspen-
margins, and then the design parameters can be found easily. sion system in red solid line is nearly 0.116 m although the
Considering the previous controlled car problem, the input step amplitude is 0.08 m; that is, the overshoot is 45%,
results are found as follows: the critical gain is 5.1, and the and after using the estimated values from the described tech-
sustained oscillation frequency is 56.7 rad/sec. It is the same niques, the overshoot and response of the system improved
result shown in Figure 4, where the stability frequency for although the transient time is increased for some of them but
the case of assuming gain margin 1 and phase margin 0 is the displacement action is smoothed; Zeigler-Nichols results
Ω → [0, 56.7]. are plotted in dash-dot line, Shamsuzzoha results are plotted
From the previous values, the P, PI, and PID Zeigler- in dash lines, and PI controller is plotted in x marker line.
Nichols controller values can be tabulated in Table 4. Another simulation model for random roughness road is
Applying the controller parameters obtained from the applied where the road model is expressed as a differential
different tuning techniques, the system response is plotted in equation with the formula [19]:
Figure 9.
From Figure 9 and related to the different tuning meth- 𝑍̈
(𝑡) + 𝑎V𝑍 (𝑡) = 𝑊 (𝑡) , (27)
ods, the systems’ time of the first peak, the peak values, and ̈is the road random incentive and the constant “𝑎”
the overshoot values can be tabulated in Table 5. where 𝑍(𝑡)
depends on road grade; the values of “𝑎” are shown in Table 6.
𝑊(𝑡) is the white noise, with zero mean value and
4. Simulations and Results amplitude equals 1, and V is the vehicle speed.
Focusing for more details can be shown in Figure 12.
After defining the controller parameters, it is necessary to test The desired problem is validated using the results of
the system performance for some situations. The suspension road simulation of grade B and assuming the velocity is
Advances in Mechanical Engineering 7

0.12
0.1
0.1
0.05

Displacement (m)
0.08
Amplitude

0
0.06 −0.05

0.04 −0.1
−0.15
0.02
−0.2
0 0 5 10 15 20 25 30
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
Time (s)
Time (s)
Passive susp. Shass PID
Passive susp. Shass PID PI
Z-N PID
Z-N PID PI
Figure 11: Output displacement of the system.
Figure 10: The output displacement from the system.

Table 6: “𝑎” value under different road grades.


0.1
Road grade A B C D E
𝑎 0.132 0.1303 0.12 1.1007 0.09 Displacement (m) 0.05

0
50 km/hr; the results show the effectiveness of the used
controllers to reduce the displacement motion as shown in −0.05
Figure 11 and zoomed in Figure 12, also results illustrated
that different control techniques can be applied to MS-TMM, −0.1
results of the passive suspension response are plotted in blue 0 1 2 3 4 5 6 7 8 9 10
solid line, Zeigler-Nichols result is plotted in blue dot plot, Time (s)
Shamsuzzohas’ result is plotted in green dash-dot plot, and Passive susp. Shass PID
finally the PI controls’ result is plotted in red dash plot. Z-N PID PI

Figure 12: Output displacement zoomed in.


5. Conclusion and Future Work
As the increments of complexity of multibody systems and
the development of their design and control methods, the advantage of indicating inputs and outputs easily such as
need for more elegant formulations of the equations of power spectral density measurements for such kinds of active
motion becomes an issue of paramount importance. The vibration isolation techniques.
results showed identical values for MS-TMM results with the
ordinary classical methods. Thus, ongoing researches, MS-
TMM strategies, and its related topologies are highly recom- Conflict of Interests
mended for analyzing and solving the controlled systems due
to easy formulations, being systematic to apply, and simple The authors declare that there is no conflict of interests
program coding with less computational time. The results regarding the publication of the paper.
also showed that different controller design methods can be
impeded with MS-TMM which leads to more availability for Acknowledgments
other controllers applications; one depends on the frequency
response analysis, another considers the step input response The research was supported by the Research Fund for
of the system, and the last technique can be applied for the Doctoral Program of Higher Education of China
both time and frequency domains if the model has a known (20113219110025), the Natural Science Foundation of China
mathematical model. All controllers’ parameters values are Government (11102089), and the Program for New Century
applied to a model of controlled car system with brief analysis Excellent Talents in University (NCET-10-0075).
for the obtained results. The future work will concern with
deeply impeding different control techniques with MS-TMM
to develop new techniques for the more complex multibody
References
controlled systems considering nonstationary inputs by tak- [1] X.-T. Rui, B. He, B. Rong, G.-P. Wang, and Y.-Q. Lu, “Discrete
ing into consideration the system’s equations of motion with time transfer matrix method for dynamics of a multi-rigid-
studying practical systems in frequency domain due to its flexible-body system moving in plane,” Proceedings of the
8 Advances in Mechanical Engineering

Institution of Mechanical Engineers K, vol. 223, no. 1, pp. 23–42,


2009.
[2] B. Rong, X. Rui, G. Wang, and F. Yang, “Discrete time transfer
matrix method for dynamics of multibody system with real-
time control,” Journal of Sound and Vibration, vol. 329, no. 6,
pp. 627–643, 2010.
[3] D. Bestle, “Application of the transfer matrix method to control
problems,” in Proceedings of the Thematic Conference, Multibody
Dynamics, Zagreb, Croatia, 2013.
[4] S. Yamato and I. Hashimoto, “Present status and future needs
the view from Japanese industry,” in Proceedings of the 4th
International Conference on Chemical Process Control, Y. Arkun
and W. H. Ray, Eds., Padre Island, Tex, USA, 1991.
[5] C. C. Yu, Auto Tuning of PID Controllers, Springer, 2nd edition,
2006.
[6] G. J. Suva, A. Datta, and S. P. Bhattacharyya, “New results on the
synthesis of PID controllers,” IEEE Transactions on Automatic
Control, vol. 47, no. 2, pp. 241–252, 2002.
[7] J. G. Ziegler and N. B. Nichols, “Optimum settings for automatic
controllers,” Journal of Dynamic Systems, Measurement, and
Control, vol. 115, no. 2, pp. 220–222, 1993.
[8] T. S. Schei, “A method for closed loop automatic tuning of PID
controllers,” Automatica, vol. 28, no. 3, pp. 587–591, 1992.
[9] B. D. Tyreus and W. L. Luyben, “Tuning PI controllers for
integrator/dead time processes,” Industrial and Engineering
Chemistry Research, vol. 31, no. 11, pp. 2625–2628, 1992.
[10] K. J. Astrom and T. Hagglund, PID Controllers: Theory, Design
and Tuning, Instrument Society of America, Research Triangle
Park, NC, USA, 2nd edition, 1995.
[11] M. Shamsuzzoha and S. Skogestad, “The setpoint overshoot
method: a simple and fast closed-loop approach for PID
tuning,” Journal of Process Control, vol. 20, no. 10, pp. 1220–1234,
2010.
[12] M. Shamsuzzoha, M. Lee, and H. Seki, “Closed-loop PI/PID
controller tuning for stable and unstable processes,” in Pro-
ceedings of the American Control Conference, pp. 2368–2373,
Montreal, Canada, 2012.
[13] A. J. Barr and J. L. Ray, “Control of an active suspension
using fuzzy logic,” in Proceedings of the 5th IEEE International
Conference on Fuzzy Systems, pp. 42–48, September 1996.
[14] M. Cai, H. Liu, P. Du, J. Guo, and H. Chang, “Suspension
dynamics simulation based on simulink,” in Proceedings of the
International Conference on Electric Information and Control
Engineering, pp. 2715–2718, April 2011.
[15] N. Changizi and M. Rouhani, “Comparing PID and fuzzy
logic control a quarter-car suspension system,” The Journal of
Mathematics and Computer Science, vol. 2, no. 3, pp. 559–564,
2011.
[16] K. J. Åström and T. Hägglund, “The future of PID control,”
Control Engineering Practice, vol. 9, no. 11, pp. 1163–1175, 2001.
[17] N. Tan, I. Kaya, C. Yeroglu, and D. P. Atherton, “Computation of
stabilizing PI and PID controllers using the stability boundary
locus,” Energy Conversion and Management, vol. 47, no. 18-19,
pp. 3045–3058, 2006.
[18] J. Y. Wong, Theory of Ground Vehicles, John Wiley & Sons, New
York, NY, USA, 1988.
[19] J. Zhihong, E. Zhang, L. Zhonghua, J. Jing, and Z. Fan,
“Simulation and experiment research on human riding comfort
for vehicle vibration in dynamic environment,” in Proceedings of
the 9th International Conference on Computer-Aided Industrial
Design and Conceptual Design: Multicultural Creation and
Design, pp. 175–181, November 2008.
Hindawi Publishing Corporation
Advances in Mechanical Engineering
Volume 2013, Article ID 725315, 6 pages
http://dx.doi.org/10.1155/2013/725315

Research Article
All Terrain Vehicle Flexible Multibody Dynamic
Simulation for Fatigue Prediction

Jia-Shiun Chen and Hsiu-Ying Hwang


Department of Vehicle Engineering, National Taipei University of Technology, No. 1, Section 3,
Zhong-Xiao E. Road, Taipei 10608, Taiwan

Correspondence should be addressed to Jia-Shiun Chen; chenjs@mail.ntut.edu.tw

Received 26 August 2013; Accepted 12 December 2013

Academic Editor: Xiaoting Rui

Copyright © 2013 J.-S. Chen and H.-Y. Hwang. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.

This study presents the flexible-body dynamic analysis and simulated stress recovery of vehicle components to predict their lifetime
when maneuvering on an uneven random road. The subject of this study was an all-terrain vehicle (ATV). The body frame and
suspension system were modeled as flexible elements for multibody dynamic simulation. The simulations in this study revealed the
stress from the flexible elements and predicted the component fatigue life using the retrieved stresses. This approach considers the
interaction between dynamic forces and structure deformation and achieves more accurate structure stress prediction and fatigue
life prediction.

1. Introduction of structural deformation and can therefore predict more


accurate loads and stresses of the vehicle components than
Fatigue is one of the major concerns in automotive engineer- the general rigid-body simulation can. The good correlation
ing. The structure and components of a vehicle are constantly between simulation and experimental results was presented
under cyclic loading, and especially on rough roads. This as well in their study. Some studies have focused on vehicle
could cause the fatigue failure of the structure or components.
components by modeling them as flexible elements in multi-
Ohchida [1] showed that 60% of machine equipment fail-
body dynamic simulation. Moon et al. [5] presented a taper
ures are caused by component fatigue. Therefore, a reliable
leaf spring with hysteretic characteristics. This study devel-
method for predicting the potential fatigue failures of vehicle
components is highly desirable. In the design cycle of auto- oped a flexible multi-body dynamic simulation to inter-
motive vehicles, the majority of automotive manufacturers face the leaf spring finite-element model and computation
use empirical methods or dynamic simulations to predict the model and correlated simulation and experimental results.
component and structure loadings [2]. The FEA model was Shabana and Sany [6] presented a rolling contact theory
designed to predict the stress of components and fatigue with multi-body dynamics to simulate the effects because
failure [3]. However, the FEA analysis process does not of the structural flexibility of the vehicle component and
consider the effects of the dynamic loads caused by structural track. Zhu et al. [7] predicted the fatigue life of a truck
deformations. The dynamic loading that occurs when the frame. They modeled the frame as a flexible element and the
vehicle is running can cause structural deformation, which in vehicle system as rigid multi-body dynamic model. Yang et al.
turn may affect load conditions and further change the stress [8] predicted the fatigue life of a wheel by simulating the EMU
of components. Thus, dynamic analysis with a rigid body wheel as a flexible body in multi-body simulations. Rathod
system is insufficient. The dynamic loading changes caused et al. [9] presented a multi-body railroad vehicle system that
by structural deformation should be considered in structure accounted for the dynamic coupling between a 3D wheel and
and component stress analysis. Cuadrado et al. [4] presented the rail structure flexibility. However, these studies did not
flexible-body dynamic simulations considering the effects include the deformation of the entire vehicle structure.
2 Advances in Mechanical Engineering

This study uses ADAMS to perform flexible-body where 𝑚𝑖 and 𝐼𝑖 represent the modal mass and modal inertia
dynamic simulation and models all structural components as of moving body, respectively.
flexible elements. The stress history of the structure was then The mass matrix of flexible body, 𝑀, can be expressed as
retrieved from the dynamic simulation for fatigue calcula-
tion. 𝑀𝑡𝑡 𝑀𝑡𝑟 𝑀𝑡𝑚
The remaining sections are as follows: the theoretical [ 𝑇 ]
[ 𝑀𝑡𝑟 𝑀𝑟𝑟 𝑀𝑟𝑚 ]
background of the flexible-body dynamics and fatigue anal- [
𝑀 (𝜉) = [ ], (6)
]
ysis, the simulation model, the simulation results, and the [ 𝑇
]
𝑀𝑡𝑚 𝑀𝑟𝑚 𝑀𝑚𝑚
conclusion. [ ]
where 𝑡, 𝑟, and 𝑚 represent the translational, rotational, and
2. Theory Background modal degrees of freedom, respectively.
2.1. Equations of Flexible-Body Dynamics. In equations of The stiffness matrix of the flexible body, 𝐾, is relatively
motion, the linear deformation of a structure can be repre- simple because there is no rigid body contribution:
sented by the combination of mode shapes and mode coordi-
𝐾𝑡𝑡 𝐾𝑡𝑟 𝐾𝑡𝑚
nates as follows: [ 𝑇 ] 0 0 0
[ ]
𝑀 𝐾 (𝜉) = [ 𝐾𝑡𝑟 𝐾𝑟𝑟 𝐾𝑟𝑚 ] = [0 0 0 ] . (7)
𝑢 = ∑𝜙𝑖 𝑞𝑖 , [ ]
[0 0 𝐾𝑚𝑚 ]
(1) 𝑇 𝑇
𝑖=1 𝐾𝑡𝑚 𝐾𝑟𝑚 𝐾𝑚𝑚
[ ]
where 𝜙 is the mode shape and 𝑞 is the mode coordinate.
The equations of motion with a Lagrange multiplier can
With the Cartesian coordinate (𝑥, 𝑦, 𝑧), Euler’s angle
be written as follows:
(𝜓, 𝜃, 𝜙), mode coordinate 𝑞, and the generalized coordinates
of the flexible element can be expressed as follows: 1 𝜕𝑀 ̇𝑇 ̇ 𝜕𝜓 𝑇
𝑀𝜉 ̈
+ 𝑀̇𝜉 ̇− [ 𝜉] 𝜉 + 𝐾𝜉 + 𝑓𝑔 + 𝐷𝜉 ̇+ [ ] 𝜆 = 𝑄.
𝑥} 2 𝜕𝜉 𝜕𝜉
{
{ }
{
{ 𝑦 }
} (8)
{
{ }
}
𝑥 { }
{𝑧}
{ } The term 𝐷 is the modal damping matrix, and 𝐾𝜉 and 𝐷𝜉,̇
𝜉 = (𝜓) = { 𝜓 } , (𝑖 = 1, 2, 3, . . . , 𝑀) . (2)
𝑞 {𝜃}
{ } respectively, represent the structural internal force caused
{
{ }
}
{𝜙}
{ } by the elastic deformation and velocity. The term 𝜆 is the
{ }
{ }
𝑞 Lagrange multiplier, 𝑓𝑔 is the structure weight, and 𝑄 is the
{ 𝑖} external force.
The position on the deformed body is written as The strain and stress can be retrieved using the mode
𝑟𝑖 = 𝑥 + 𝐴 (𝑆𝑖 + 𝜙𝑖 𝑞) , (3) coordinate [10]. This study calculates the stress in the struc-
ture using the modal stress recovery technique:
where 𝐴 is the transformation matrix between global coordi-
nates and the body local coordinates. The term 𝑆𝑖 represents {𝜀} = [𝐻] {𝑥} ,
the position in the body before deformation, and 𝜙𝑖 is the (9)
mode shape matrix. {𝜎} = [𝐸] {𝜀} .
From the position, the velocity of the deformed body is The term {𝜎} is the stress vector, {𝜀} is the strain vector,
𝑑𝑟𝑖 𝑑𝑥 𝑑𝐴 𝑑 (𝑠𝑖 + 𝜙𝑖 𝑞) and [𝐻] is the finite-element geometric deformation-to-
V𝑖 = = + (𝑠 + 𝜙𝑖 𝑞) + 𝐴 strain matrix. The term [𝐸] is the material property matrix.
𝑑𝑡 𝑑𝑡 𝑑𝑡 𝑖 𝑑𝑡
= 𝑥̇− 𝐴 (̃𝑠𝑖 + 𝜙̃𝑖 𝑞) 𝜔 + 𝐴𝜙𝑖 𝑞 ̇ (4)
2.2. Fatigue Prediction. Metal material under cycling stress
󵄨 󵄨 can fail even when the stress is under the ultimate stress limit.
= 󵄨󵄨󵄨󵄨𝐹 − 𝐴 (̃𝑠𝑖 + 𝜙̃𝑖 𝑞) 𝐵 + 𝐴𝜙𝑖 󵄨󵄨󵄨󵄨 𝜉.̇ The relationship between fatigue stress and life cycle is usually
The term 𝜔 is the angular velocity vector of the body represented by a stress-cycles curve (𝑆-𝑁 diagram). For steel,
coordinate, and 𝐹 is the transfer matrix between the time a cycle life greater than 106 is treated as a lasting life, and the
derivative of Euler’s angle and the angular velocity. stress corresponding to a 106 cycle is the endurance limit 𝑆𝑒
The kinematic energy of the body is [11].
The 𝑆-𝑁 diagram is plotted in log scale, and the formula
1 of 𝑆-𝑁 curve can be simplified as
𝑇= ∫ 𝜌V𝑇 V𝑑𝑉
2
log 𝑆𝑓 = log 𝑎 + 𝑏 log 𝑁 󳨐⇒ 𝑆𝑓 = 𝑎𝑁𝑏 , (10)
1𝑁
≈ ∑ (𝑚𝑖 V𝑖𝑇 V𝑖 + 𝜔𝑖𝑇 𝐼𝑖 𝜔𝑖 ) (5)
2 𝑖=1 where 𝑏 is the slop of curve and 𝑁 is the life cycle.
During vehicle maneuvering, the structural stress is fluc-
1
= 𝜉𝑇̇𝑀 (𝜉) 𝜉,̇ tuating. The amplitude of stress variation varies with the driv-
2 ing and road conditions. The rain-flow counting technique
Advances in Mechanical Engineering 3

Revolute
joint

Bushing

Ball
joint

Spring and damper

y x
Figure 1: Vehicle structure and suspension frames. z

Figure 2: Assembled vehicle and connecting joints and forces.


can be applied to measure different stress variations during
driving. The Palmgrem-Miner cycle-ratio summation rule Table 1: Material property of vehicle structure.
can then be used to accumulate the damages of various loads
and cycles: Property Steel Steel
1015 1018
𝑖=𝑁
𝑛𝑖 Elastic modulus (MPa) 2.07𝐸05 2.07𝐸05
∑ = Damage, (11)
𝑁
𝑖=1 𝑓𝑖
Poisson ratio 0.29 0.29
Density (ton/mm3 ) 7.9𝐸 − 09 7.9𝐸 − 09
where 𝑛𝑖 is the number of cycles at the change of stress ampli- Ultimate strength (MPa) 414 455
tude (Δ𝜎𝑖 ) and 𝑁𝑓𝑡 is the number of cycles until failure at the
change of stress amplitude (Δ𝜎𝑖 ). The “Damage” parameter
usually ranges from 0.7 to 2.2. When the accumulated damage
[10], which were connected with a revolute joint. The upper
is larger than the “Damage” parameter, the component fails
torso, including the seat back, and the pelvis, including the
[12].
seat, were attached to the vehicle structure by springs and
dampers. Table 2 lists the spring stiffness and damping coef-
3. Vehicle Model ficient. The driveline was seated on the engine mounts, and
the engine mounts were modeled as rubber bushings. Table 2
The object of research in this study is a dune buggy. Figure 1 lists the parameters.
shows the geometric model based on the measurement on the
vehicle structure. For each component of vehicle structure,
MSC Nastran was used to perform modal analysis, and 3.2. Tire and Road Models. The simulations in this study
MSC Patran was used to process finite-element models and modeled tires as force elements, and these tire models
modal analysis data to create flexible elements for dynamic included the tire mass and inertia. The simulations in this
simulation [13]. These elements were imported to the vehicle study use the Fiala tire model, which was designed for vertical
model to assemble the whole vehicle. Then, we used ADAMS and longitudinal motion. Table 3 presents the tire parameters.
to perform the flexible-body dynamic simulations [14] and For simulation road condition, this study applies a ran-
determine the stress generated by random road maneuvers. dom 2D surface (Figure 3).

3.1. Components. The structure of vehicle model included 4. Simulation and Result
three primary parts: the body frame, the front suspension,
and the rear suspension (Figure 2). The powertrain and In the simulation, the vehicle maneuvered in straight runs
passenger weights were attached to the structure, and the total over a random uneven surface. The vehicle accelerated from
weight of vehicle was 423 kg. All the structure components standing still, and the data for analysis was collected when
were modeled as a flexible body. The tires were simulated as it reached a constant speed of 50 km/h. The duration of data
force elements in dynamic simulation [14]. Figure 2 shows collecting was 10 s. Several thicknesses of the structure frame
the connecting joint and forces applied between components tube were applied in the simulation: 1 mm, 1.5 mm, 2 mm, and
[14, 15], and Table 1 shows material properties of the frame 2.5 mm.
[16].
In addition to the weight of the vehicle structure and the 4.1. Dynamic Simulation. In flexible-body dynamic simula-
driveline, two dummies weighing 80 kg each were modeled tion, ADAMS can retrieve the structural stress which could be
and seated on the vehicle structure. The dummies were applied to failure or fatigue analysis. Figure 4 shows the high-
modeled as two parts (the upper torso; pelvis and legs) est stress locations during the dynamic simulation. Table 4
4 Advances in Mechanical Engineering

Table 2: Properties of dummies and driveline.

Mass (kg) Spring stiffness (N/mm) Damping coefficient (N-s/mm)


Upper torso (including seat back) 45 10 0.4
Hips and leg (including seat) 35 15 0.2
Powertrain 40 52 0.12

Table 3: Tire parameters.

Property Front tire Rear tire


Tire model Fiala tire type Fiala tire type
Mass (kg) 10 11 Force
direction
Moment of inertia (kg-mm ) 2 Ixx, Iyy: 8𝐸5; Ixx, Iyy: 8𝐸5;
Izz: 16𝐸5 Izz: 16𝐸5
Tire radius (mm) 254 280
Tread width (mm) 200 200 Flexible
body

10.0 y
z x
V-axis length (mm)

5.0
Figure 5: Joint locations.
0.0

−5.0
simulation cannot determine the structural stress, the max-
−10.0 imum rear suspension joint forces were used to present the
0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 90.0 100.0 difference between these two types of simulation (Table 5).
H-axis length (m) Figure 5 shows the joint locations. The frame tube thickness
of the model was 2.0 mm in this simulation.
Figure 3: 2D random road profile.
In addition to the suspension and tire elements, the
flexible element in the vehicle structure also absorbed part of
582079 dynamic energy during the uneven road simulation. Thus, the
585076 joint force of the flexible-body model was smaller than that of
von Mises stress (MPa) the rigid-body model during the simulation.
592.32
533.09
470.86
586151
582887 4.2. Fatigue Analysis. The FEM model in this study adopts
414.62 several thicknesses of the frame tube. The flexible elements
355.39 were built based on these models, and flexible-body dynamic
296.16 simulations were run with different tube thicknesses. The
236.93 tube thicknesses were 1 mm, 1.5 mm, 2 mm, and 2.5 mm. The
177.7 maximum stress was retrieved and plotted in a stress-time
118.46 curve. Figure 6 shows the stress history of the structure with
59.23 a 2.5 mm tube thickness in 10 s simulation at a constant speed
5.48E − 014 of 50 km/h.
The rainflow counting method was applied to the stress
history curve, and the stress cycle number of different stress
amplitudes was used for fatigue analysis. Figure 7 shows the
Figure 4: Highest stress locations at random road simulation. stress range and cumulative cycle number of different frame
tube thicknesses.
Based on the stress cycle and material 𝑆-𝑁 diagram, the
structural life can be predicted using the Miner rule. Using
shows the maximum von Misses stress of these locations in a 10 s simulation cycle, Table 6 presents the structural life
uneven random road simulations. of different frame tube thicknesses. This duration time and
To compare the differences in dynamic load running on the maximum stress position listed in Table 4 show that
an uneven road surface, this study created both rigid-body the structure connecting points of two or more structure
and flexible-body models. Because the rigid-body dynamic members are the most vulnerable to fatigue damage.
Advances in Mechanical Engineering 5

Table 4: Structure stress retrieved from flexible body dynamic simulation.

Frame tube thickness (mm) Road surface Max. stress grid point Max. stress (MPa)
Flat 582887 215.36
1
Random 585076 498.30
Flat 582887 163.08
1.5
Random 586151 430.28
Flat 582887 137.87
2
Random 586151 361.84
Flat 582887 105.33
2.5
Random 585079 347.21

Table 5: Joint force comparison between flexible-body and rigid-body simulation.

Frame tube thickness (mm) Right and left joints average force at rear Difference
suspension (N)
Flexible-body model 623.01 —
Rigid-body model 758.09 21.7%

350 Table 6: Fatigue analysis of different tube thicknesses.

300 Tube thickness 1 mm 1.5 mm 2 mm 2.5 mm


Cycle number Less than 1 𝑁𝑓 = 218 𝑁𝑓 = 1265 𝑁𝑓 = 2780
250
Amplitude (MPa)

200
5. Conclusions
150
Conventional analysis methods model the vehicle body as a
100
rigid body in dynamic simulations. In this case, the dynamic
50 energy passing from the tire running over an uneven road can
only be absorbed and damped by tires, springs, and dampers
0 in suspension, and bushings. After this rigid body dynamic
0.0 2.0 4.0 6.0 8.0 10.0 simulation, the component load was applied to a finite-
Time (s) element model to determine the structural stress of the
Figure 6: Stress history of the max. stress grid point (585079).
component for further analysis. The approach used in this
paper is different from the conventional one. The dynamic
model in this study uses flexible bodies. The advantage of
using flexible-body dynamic simulation is that it simulates
450
structural stresses in dynamic analysis and can be retrieved
400
through the simulation. The dynamic response considers the
350 effect of structural deformation. This provides more realistic
Stress range (MPa)

300 and better prediction, especially for not-so-rigid vehicle


250 structures. The structure can also absorb part of the energy.
200 Thus, the loads at joints were smaller compared with the loads
150 obtained in rigid-body simulation.
100 The stress received from flexible dynamic analysis can
50
be used in fatigue analysis. The structure connecting points
where two or more structure members were welded together
0
appeared to be the most vulnerable to fatigue damage. There-
1 10 100 1000
fore, these points should be reinforced during structure
Cumulative cycles
design and assembly to increase the expected lifetime.
2.5 mm (point 585079) Vehicle fatigue testing is costly and time consuming.
2 mm (point 586151)
1.5 mm (point 586151)
Flexible-body dynamic simulations can assist in the life pre-
diction of vehicle structures and components. The proposed
Figure 7: Cumulative cycle number of different frame tube thick- analysis tool can shorten the design cycle and can reduce
nesses. vehicle development cost and time.
6 Advances in Mechanical Engineering

Notations approach for modelling track flexibility,” Proceedings of the Insti-


tution of Mechanical Engineers K, vol. 223, no. 4, pp. 269–282,
𝐴: Transformation matrix between global 2010.
coordinates and the body local coordinates [10] T. M. Wasfy and A. K. Noor, “Computational strategies for flex-
𝐷: Modal damping matrix ible multibody systems,” Applied Mechanics Reviews, vol. 56, no.
𝐸: Material property matrix 6, pp. 553–613, 2003.
𝐹: Transfer matrix between the time [11] W. Yao, Structure Fatigue Analysis, Defence Inductry Publisher,
derivative of Euler’s angle and angular 2003.
velocity [12] J. E. Shigley and C. R. Mischke, Mechanical Engineering Design,
𝐻: Finite-element geometric McGraw Hill, 2010.
deformation-to-strain matrix [13] MSC Software, Nastran User’s Manual, MSC Software, Newport
𝐼𝑖 : Modal inertia Beach, Calif, USA, 2010.
𝐾: Stiffness matrix of the flexible body [14] MSC Software, MD Adams. ADAMS User’s Manual, MSC
𝑀: Mass matrix of flexible body Software, Newport Beach, Calif, USA, 2010.
𝑄: External force [15] M. Azadi, S. Behzadipour, and G. Faulkner, “Performance anal-
𝑆𝑖 : Position before deformation ysis of a semi-active mount made by a new variable stiffness
𝑓𝑔 : Structure weight spring,” Journal of Sound and Vibration, vol. 330, no. 12, pp.
𝑚𝑖 : Modal mass 2733–2746, 2011.
𝑞: Mode coordinate [16] R. C. Hibbeler, Mechanics of Materials, Prentice Hall, 2010.
𝜀: Strain vector
𝜆: Lagrange multiplier
𝜉: The generalized coordinate of flexible
elements
𝜎: Stress vector
𝜙: Mode shape
𝜙𝑖 : Mode shape matrix
𝜔: Angular velocity vector.

References
[1] H. Ohchida, “Analysis of service failures of Hitachi products
1970–1975,” Hitachi Company Report, 1979.
[2] S. Lu, Y. Li, and Y. Dong, “Dynamic modeling and virtual test
of a new all terrain off-road vehicle,” Applied Mechanics and
Materials, vol. 86, pp. 125–128, 2011.
[3] B. R. Miao, W. H. Zhang, L. M. Zhang, T. Zhu, and H. T. Yin,
“Research of modern railway vehicle structure fatigue design
method based on multibody dynamic simulation,” Advanced
Materials Research, vol. 118–120, pp. 327–331, 2010.
[4] J. Cuadrado, R. Gutiérrez, M. A. Naya, and M. González,
“Experimental validation of a flexible MBS dynamic formu-
lation through comparison between measured and calculated
stresses on a prototype car,” Multibody System Dynamics, vol. 11,
no. 2, pp. 147–166, 2004.
[5] I.-D. Moon, H.-S. Yoon, and C.-Y. Oh, “A flexible multi-body
dynamic model for analyzing the hysteretic characteristics and
the dynamic stress of a taper leaf spring,” Journal of Mechanical
Science and Technology, vol. 20, no. 10, pp. 1638–1645, 2006.
[6] A. A. Shabana and J. R. Sany, “A survey of rail vehicle track simu-
lations and flexible multibody dynamics,” Nonlinear Dynamics,
vol. 26, no. 2, pp. 179–210, 2001.
[7] S. Zhu, Z. Xiao, and X. Li, “Vehicle frame fatigue life prediction
based on finite element and multi-body dynamic,” Applied
Mechanics and Materials, vol. 141, no. 1, pp. 578–585, 2012.
[8] J. Yang, Q. Shi, G. Zhang, and J. Zhang, “The fatigue life simula-
tion of the wheel of CHR3 EMU in random loading,” Advanced
Materials Research, vol. 430–432, pp. 1424–1427, 2012.
[9] C. Rathod, R. Chamorro, J. L. Escalona, M. El-Sibaie, and A. A.
Shabana, “Validation of three-dimensional multi-body system
Hindawi Publishing Corporation
Advances in Mechanical Engineering
Volume 2013, Article ID 153913, 12 pages
http://dx.doi.org/10.1155/2013/153913

Research Article
Projectile Impact Point Prediction Based on Self-Propelled
Artillery Dynamics and Doppler Radar Measurements

Mostafa Khalil, Xiaoting Rui, Qicheng Zha, Hailong Yu, and Hossam Hendy
Institute of Launch Dynamics, Nanjing University of Science and Technology, 200 Xiao Ling Wei, Nanjing 210094, China

Correspondence should be addressed to Mostafa Khalil; mostafakhalil80@gmail.com

Received 24 August 2013; Accepted 26 November 2013

Academic Editor: Ch Zhang

Copyright © 2013 Mostafa Khalil et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Any trajectory calculation method has three primary sources of errors, which are model error, parameter error, and initial state
error. In this paper, based on initial projectile flight trajectory data measured using Doppler radar system; a new iterative method
is developed to estimate the projectile attitude and the corresponding impact point to improve the second shot hit probability. In
order to estimate the projectile initial state, the launch dynamics model of practical 155 mm self-propelled artillery is defined,
and hence, the vibration characteristics of the self-propelled artillery is obtained using the transfer matrix method of linear
multibody system MSTMM. A discrete time transfer matrix DTTM-4DOF is developed using the modified point mass equations
of motion to compute the projectile trajectory and set a direct algebraic relation between any two successive radar data. During
iterations, adjustments to the repose angle are made until an agreement with acceptable tolerance occurs between the Doppler
radar measurements and the estimated values. Simulated Doppler radar measurements are generated using the nonlinear six-
degree-of-freedom trajectory model using the resulted initial disturbance. Results demonstrate that the data estimated using the
proposed algorithm agrees well with the simulated Doppler radar data obtained numerically using the nonlinear six-degree-of-
freedom model.

1. Introduction During testing new weapons, an accurate instrumenta-


tion is needed to measure the in-flight projectile parameters.
There are different computation methods to estimate the Projectiles do not have enough space to install on-board
projectile ballistic trajectory and subsequent impact point. sensors to acquire these parameters. Further, considering the
These methods have three primary sources of errors, which severe environment faced by the projectile during flight such
are model error, parameter error, and initial state error. The as velocity and spinning rate, a low cost and accurate mea-
model error depends on how much the projectile equations surement method is required. The most promising method is
of motion are simplified. Increasing model complexity means using a Doppler radar system [4] due to its robustness, porta-
decreasing model error, but the parameter and initial state bility, accuracy, and providing nearly continuous velocity-
errors will increase [1]. In 1964, the projectile six-degree-of- time information on the projectile during the flight down
freedom (6-DOF) equations of motion in terms of direction range. A number of authors presented different algorithms
cosines [2] was developed by the ballistic research laboratory, based on Doppler radar measurements. In 1958, Shapiro [5]
which is the most accurate model that can simulate the rigid proposed three different estimation methods to compute the
body flight dynamics under the condition that all the aerody- six parameters that define the orbital parameters describing
namic forces and moments, and the initial state, are known the ballistic missile trajectory using single radar measure-
to a high degree of accuracy. Using this system of equa- ments which include range, range-rate, and the azimuth and
tions, a discrete time transfer matrix method DTTM-6DOF elevation angles. These methods are the maximum likelihood,
[3] has been developed to accurately compute the projectile the weighted least squares, and a deterministic method. In
ballistic trajectory and minimize the corresponding compu- 1985, a new iterative method had been developed [6] and
tational time. applied to a modified version of 6-DOF model to extract
2 Advances in Mechanical Engineering

actual thrust made during the flight of rocket-assisted projec- equation of self-propelled artillery was obtained to compute
tiles using Doppler radar measurements. In 2008, an iterative the corresponding vibration characteristics. In order to itera-
algorithm had been developed [7] using maximum likelihood tively compute the projectile attitude, a discrete time transfer
method to predict the drag characteristics of projectiles in matrix DTTM-4DOF is developed based on linearization
motion by processing data acquired by Doppler radar system, of the nonlinear modified point mass equations of motion
where the point mass model was used due to the projectile using second-order Taylor series [16]. The system state vector,
small angle of attack along the trajectory. which defines the projectile trajectory parameters during
The movements of artillery and projectile are very com- Doppler radar observations, is defined by 13 state variables,
plex in launch process because of the complex structure of which are the projectile position, velocity, spin rate, repose
self-propelled artillery and the severe environments, such angle, and wind velocity. During iterations, adjustments to
as, high temperature, high pressure, high speed, instanta- the repose angle are made until agreement with a certain tol-
neous state, multibody, and mutation, in launch process erance occurs between the Doppler radar velocity measure-
[8]. Therefore, gun launch dynamics has a primary concern ments and the value predicted using DTTM-4DOF. Finally,
in designing embedded navigation and guidance systems projectile impact point can be predicted precisely depending
in case of guided projectiles and evaluating the projectile on the length of radar data available, where the data estimated
initial disturbance in case of unguided projectiles. In 2000, at the end of radar observations is used as an initial state to
the launch dynamics of self-propelled artillery [9] based the 6-DOF trajectory model.
on transfer matrix method of multibody system MSTMM
was developed for studying the vibration characteristics and 2. Launch Dynamics Model for
dynamics of self-propelled artillery and the projectile’s initial Self-Propelled Artillery System
disturbance. In 2001, a finite element analysis [10] had been
done for the breech closure for the 155 mm Cannon M199, To accurately estimate the projectile initial state, launch pro-
which is normally mounted on the Towed Howitzer M198. cess has to be decided precisely by studying the artillery and
The analysis is for a 9 body problem with 13 contact surfaces projectile dynamics from the moment of firing to the state
and was solved for both static and dynamic load cases. In of muzzle point. Four interactions have to be considered, the
2004, a simplified model [11] had been done to predict joint interaction between projectile and artillery, the interaction
forces and accelerations along the length of a simplified between the artillery transverse and longitudinal motion, the
155 mm projectile using modal superposition. These pre- interaction between rigid and flexible body, and, finally, the
dicted loads used to locate all guidance equipment, sensors, interaction between space coordinate and time coordinate for
joints, and computers that must operate reliably after exiting describing motion parameters.
the gun. In 2005, a sophisticated 3D-finite-element model The launch dynamics model of self-propelled artillery is
[12] had been developed to investigate the survivability a multi-rigid-flexible system which has been studied in [8].
of embedded electrical systems by simulating the launch This system is composed of 51 elements (23 bodies and 28
dynamics of a surrogate Excalibur projectile. This study joints) as shown in Figure 1. It has 3 boundary ends including
determined free mounting locations trouble for sensitive ground, gun breech, and muzzle. The ground (0) is regarded
components and parametric investigation in identifying sen- as an infinity rigid body and the first input point, the gun
sitive factors affecting the muzzle-exit motion of projectile breech (44) is regarded as rigid body and the second input
substructures. In 2008, a new method [13] named as transfer point, and the muzzle (52) is the output point. The road
matrix method of linear multibody system MSTMM for lin- wheels (13–24) are regarded as lumped masses. The hull
ear hybrid multibody system dynamics was developed. This (37), revolving part (39), elevating part (41), and muzzle
method had been applied to a Multiple Rocket System as a brake (51) are regarded as rigid bodies; the barrel is divided
linear multi-rigid-flexible-body system. In 2011, the dynamics into 6 segments (45–50) and each is regarded as a beam
problem of a shipboard gun system had been solved [14] using with equal sectional area. The interactions (1–12) between
the discrete time transfer matrix method of multibody system ground and road wheel and the connections (25–36) between
MSDTTMM [15]. road wheel and the hull are, respectively, modeled with
In this paper, an iterative method is developed to deter- springs and dampers in 3 directions. Connections (38, 40,
mine the projectile kinematics including attitude and angular 42, and 43) are modeled with springs and rotary springs
motion using Doppler radar measurements. Doppler radar accompanying dampers to represent relative linear motion
measurements are only available during the first portion and relative angular motion in 3 directions at the same time.
of projectile trajectory including the range, the range-rate, All connections between barrel and gun breech, barrel and
and the azimuth and elevation angles (position and velocity muzzle brake, and each segment of barrel are regarded as
vectors), in order not to be recognized by the enemy coun- fixed. The masses of traversing mechanism, elevating mecha-
terattack systems. Simulated Doppler radar measurements nism, and equilibrator fall into elevating part, revolving parts,
are generated during flight time using 6-DOF trajectory and hull, respectively. According to the launch dynamics
model including the projectile initial disturbance problem. model of self-propelled artillery and the sequence number of
The projectile initial disturbance is computed by solving each element, there are 49 connection points which are 𝑃𝑖,𝑖−12 ,
the self-propelled artillery and projectile dynamics as a 𝑃𝑖,𝑖+12 (𝑖 = 13–24), 𝑃37,𝑖 (𝑖 = 25–36), 𝑃37,38 , 𝑃39,38 , 𝑃39,40 ,
multi-rigid-flexible system using Transfer Matrix Method of 𝑃41,40 , 𝑃41,42 , 𝑃41,43 , 𝑃44,45 , 𝑃45,46 , 𝑃46,47 , 𝑃47,48 , 𝑃48,49 , 𝑃49,50 ,
Multibody System MSTMM [8], where the overall transfer and 𝑃51,50 and 14 boundary points which are 𝑃0,𝑖 (𝑖 = 1–12),
Advances in Mechanical Engineering 3

Elevating part 41
Projectile
Breech 44 42 40 43 Beams 45–50 Muzzle brake 51
52

38
Revolving part 39
Hull 37
25 26 27 28 29 30
13 14 15 16 17 18
1 2 3 4 5 6 Road wheels 13–24
0

Figure 1: Launch dynamic model of self-propelled artillery [8].

𝑃44,0 , 𝑃51,52 , respectively. The overall transfer equation of the X2


practical 155 mm self-propelled artillery shown in Figure 1 𝒙
was obtained using the transfer matrix method of linear V2
Vt 𝛼t
multibody system MSTMM [8]. p x2
During launch process, all forces and moments acting
between barrel and elevating part, elevating part and revolv-
X3
ing part, revolving part and the hull, the hull and road wheels, h2
and road wheels and the ground are regarded as internal V3
forces. The forces that acted on the barrel by projectile [8, 17] X1 h1 x3

BT
and the forces that acted on the barrel and muzzle brake V1 h3
x1
by propellant gas [8] are regarded as external forces of the O
artillery. The recoil resistance between the barrel and elevat-
ing part produced by recoil and counter-recoil mechanism is Figure 2: Coordinate system and directions of ballistic target.
decomposed into external force and internal force, and the
internal force is proportional to the relative displacement
𝑝𝛼 𝑝
between barrel and elevating part. where 𝐶𝐷, 𝐶𝐿𝛼 , 𝐶𝑁 , and 𝐶𝑙 are the aerodynamic drag
force, lift force, Magnus force, and spin damping moment
3. Projectile Flight Trajectory Model coefficients, respectively; V is the projectile velocity vector in
earth fixed coordinate; Vt is the projectile total velocity vector
Due to the complexity of projectile initial disturbance deter- with respect to air; V is the projectile total aerodynamic speed;
mination, a modified point mass 4-DOF model for an 𝜌 is the air density; 𝑚, 𝐼𝑥 , 𝐼𝑦 , 𝑑, and 𝑆ref are the projectile mass,
unguided rigid projectile was stated in [18, 19]. These equa- axial and transverse moments of inertia, reference diameter,
tions assume that the epicyclic pitching and yawing motion and reference area, respectively; and g and Λ are the earth’s
of the projectile are small everywhere along the trajectory; gravitational and Coriolis accelerations vectors.
therefore, the yaw and pitch moments considered in 6-DOF The projectile repose angle 𝛼𝑅 is defined by [19]
trajectory model [19] are neglected. 󵄨 󵄨
The projectile modified point mass equations of motion 𝛼𝑅 = x − (cos 󵄨󵄨󵄨𝛼𝑡 󵄨󵄨󵄨) i, (4)
with respect to earth fixed coordinate system (𝑋1 , 𝑋2 , 𝑋3 ) as
shown in Figure 2 are given by [19] where x is the projectile’s axis of symmetry unit vector and i
is the unit vector in the direction of Vt .
𝑇
And the total aerodynamic angle of attack is [3, 18, 19]
Ẋ= V = [𝑉1 𝑉2 𝑉3 ] , (1)
Vt ⋅ x
𝜌V𝑆ref 𝐶𝐷 𝜌𝑆 𝐶𝛼 𝛼𝑡 = cos−1 . (5)
V̇= − Vt + ref 𝐿 V2 𝛼𝑅 V
2𝑚 2𝑚
𝑝𝛼
(2) The simplified form of the repose angle is given by [19]
𝜌𝑆ref 𝑑𝐶𝑁
+ 𝑝 [Vt × 𝛼𝑅 ] + g + Λ,
2𝑚 2𝐼𝑥 𝑝
𝛼𝑅 = [V̇× Vt ] , (6)
2 𝜌𝑆ref 𝑑V4 𝐶𝑚
𝛼
𝜌𝑆ref 𝑑 V 𝑝
𝑝̇= 𝑝𝐶𝑙 , (3)
2𝐼𝑥 𝛼
where 𝐶𝑚 is the aerodynamic pitching moment coefficient.
4 Advances in Mechanical Engineering

For discrete systems, a good approximation was done for The equations of motion matrix A, summarizing (1)–(9)
the projectile repose angle [20] as multiplied by the time step Δ𝑡, are given by
2𝐼𝑥 𝑝
𝛼𝑅 = [g × Vt ] . (7) Δ𝑡2
𝜌𝑆ref 𝑑V4 𝐶𝑚
𝛼
[G3×3 I3 ⋅ Δ𝑡 O3×1 O3×3 O3×3 ]
[ 2 ]
The Coriolis acceleration due to earth’s rotation and gravita- [G ⋅ Δ𝑡 VV ⋅ Δ𝑡 V𝑝 ⋅ Δ𝑡 V𝛼 ⋅ Δ𝑡 VW ⋅ Δ𝑡]
̇ ̇ ̇ ̇
[ 3×3 ]
tional acceleration are given by [3, 19] A=[
[ O1×3 O1×3 𝑝𝑝̇ ⋅ Δ𝑡 O1×3 O1×3 ]
],
[ ]
0 − cos 𝜇 sin 𝜆 − sin 𝜇 𝑉1 [𝛼 G 𝛼V̇V̇V 𝛼V̇V̇𝑝 𝛼V̇ V̇𝛼 𝛼V̇ V̇W ]
[ V̇3×3 ]
Λ = 2Ω [cos 𝜇 sin 𝜆 0 cos 𝜇 cos 𝜆] [𝑉2 ] ,
ẆV ⋅ Δ𝑡 O3×3
[ O3×3 O3×3 O3×3 ]
[ sin 𝜇 − cos 𝜇 cos 𝜆 0 ] [𝑉3 ] (14)
𝑇
𝑋1 2𝑋2
g = −𝑔 ⋅ [ 1− 0] , where I3 is the 3 × 3 identity matrix.
𝑅𝑒 𝑅𝑒
All elements of matrix A in (14) are stated as follows:
(8)

where 𝑔 = 𝑔𝑜 ⋅[1−0.0026 cos(2𝜇)], 𝑔𝑜 = 9.80665 m/s2 ; Ω is the −𝑔 0 0


earth’s angular velocity; 𝜇, 𝜆 are the corresponding latitude G3×3 = [ 0 2𝑔 − 𝑔𝑋2 0] , (15)
and longitude of the firing sight; and 𝑅𝑒 is the average radius [0 0 0]
of the earth (=6356766 m).
Due to spherical earth approximation, the instantaneous
projectile altitude 𝐻 is defined as [19] where 𝑔𝑋2 = 𝑔/𝑋2 , 𝑔 = 𝑔/𝑅𝑒 ,

𝑋12
𝐻 = 𝑋2 + . (9) V̇V = ΛV − 𝜌𝑎 1 𝐶𝐷 VI3 ,
⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟
2 ⋅ 𝑅𝑒 (16)

W

4. Discrete Time Transfer Matrix for Modified 0 − cos 𝜇⋅sin 𝜆 − sin 𝜇


Point Mass Equations of Motion where ΛV = 2 ⋅ Ω ⋅ [ cos 𝜇⋅sin 𝜆 0 cos 𝜇⋅cos 𝜆 ],
sin 𝜇 − cos 𝜇⋅cos 𝜆 0
To develop a discrete time transfer matrix DTTM-4DOF
using the modified point mass equations of motion men- 𝑝𝛼
V̇𝑝 = 𝜌𝑎2 𝐶𝑁 B1 ,
tioned above, a system state vector S is defined by 13 state
variables to define the projectile trajectory during flight time. V̇𝛼 = 𝜌𝑎1 𝐶𝐿𝛼 V2 I3 ,
These state variables are the projectile position, velocity, spin
rate, repose angle, and wind velocity, and are as follows: 𝑝 (17)
𝑝𝑝̇ = 𝜌𝑏1 V𝐶𝑙 ,
S13×1 −𝑏2 𝑝
𝛼V̇
= [Vt ]× ,
= [𝑋1 𝑋2 𝑋3 𝑉1 𝑉2 𝑉3 𝑝 𝛼1 𝛼2 𝛼3 𝑊1 𝑊2 𝑊3 ] .
𝑇 𝜌V4 𝐶𝑚𝛼

(10)
where 𝑎1 = (𝑆ref /2𝑚), 𝑎2 = ((𝑆ref 𝑑)/2𝑚), 𝑏1 = ((𝑆ref ⋅ 𝑑2 )/(2 ⋅
DTTM-4DOF is the relation between any two successive state 𝐼𝑥 )), 𝑏2 = (𝑑/𝑏1 ), and B1 =Vt × 𝛼𝑅 .
vectors, where
The change in wind velocity vector with the projectile
S𝑖+1 𝑖 position ẆV is given by
10×1 = DTTM-4DOF10×13 ⋅ S13×1 . (11)
The 2nd-order Taylor series is given by [16]
𝜕W/𝜕𝑡 𝜕W/𝜕𝑡 𝜕W
2 ẆV = = = . (18)
Δ𝑡 𝜕V 𝜕X/𝜕𝑡 𝜕X
S𝑖+1
10×1 = S𝑖10×1 + Ṡ
𝑖 ̈
𝑖
10×1 ⋅ Δ𝑡 + S10×1 ⋅ , (12)
2
where The Jacobian matrix J of the system, mentioned in (13b),
which represents discretization errors in matrix A, is given

10×1 ⋅ Δ𝑡 = I10×13 A13×13 ⋅ S13×1 , (13a) by
𝑖
̈ Δ𝑡2 𝜕Ṡ
𝑖
Δ𝑡2 𝜕Ṡ Δ𝑡2
𝑖
S10×1 ⋅ =[ ⋅ ] = [( ) ⋅ Ṡ
𝑖
13×1 ⋅ ] O3×3 I3 O3×1 O3×3 O3×3
2 𝜕𝑡 2 𝜕S 10×13 2 [O
[ j22 V̇𝑝 O3×3 j25 ]]
J = [ 3×3 ]. (19)
= [T10×10 J10×13 A13×13 S13×1 ] . [O1×3 j32 𝑝𝑝̇ O1×3 −j32 ]
(13b) [O3×3 j42 j43 O3×3 j45 ]
Advances in Mechanical Engineering 5

The partial derivatives of the projectile acceleration in (2) 5. Projectile Attitude


with respect to wind velocity and projectile velocity are given Determination Algorithm
by
The algorithm presented here is an iterative method to
V compute (1) the projectile muzzle velocity and line of sight
j25 = 𝜌𝑎1 𝐶𝐷VI3 + [𝜌𝑎1 ((𝐶𝐷 + 𝐶𝐷𝑀 𝑀) t by using the first projectile displacement observed by radar,
V
and (2) the projectile flight attitude and the corresponding
angular momentum by using the Doppler velocity observed
− (2𝐶𝐿𝛼 + 𝐶𝐿𝛼 𝑀 𝑀) 𝛼𝑅 ) by radar. At each time step, the difference between the data
𝑝𝛼
measured by radar and the estimated parameters by using
𝜌𝑎2 𝑝𝐶𝑁𝑀 𝑀 𝑝𝛼 DTTM-4DOF are linearly related as follows:
− B1 ] V𝑇t + 𝜌𝑎2 𝑝𝐶𝑁 [𝛼𝑅 ]× ,
V2
ΔX
(20)
ΔX m11 m12 m13 m14 [[ ΔV ]
].
[ ] =[ ]. (28)
where 𝐶𝑖 𝑀 = 𝜕𝐶𝑖 /𝜕𝑀 is the rate of change of any aerody- ΔV 𝑖+1 m21 m22 m23 m24 [ Δ𝑝 ]
namic coefficient with respect to Mach number [Δ𝛼𝑅 ]𝑖
j22 = ΛV − j25 . (21) At the beginning of flight time, the initial guess of the
projectile velocity V𝑜 is iteratively corrected by adding ΔV𝑜
Similarly, using (3), the rate of change of projectile spin rate using the first radar data available (𝑖 = 1) until the position
with respect to its velocity is given by difference ΔX1 between the estimated and observed value is
𝜌𝑏1 𝑝 𝑝 small enough (less than a given tolerance) as follows:
𝑝
j32 = (𝐶𝑙 + 𝐶𝑙 𝑀) Vt . (22)
V 𝑀
−1
ΔV𝑜 = [m12 ] ΔX(𝑖=1) . (29a)
The partial derivatives of the projectile yaw of repose in (7)
with respect to projectile velocity and spin rate are as follows: The projectile muzzle spin rate is corrected by adding Δ𝑝𝑜
𝛼 𝛼
using the velocity difference ΔV𝑜 as
𝑏2 𝑝 𝑏2 𝑝 (4𝐶𝑚 + 𝐶𝑚 𝑀)
j42 = 4 𝛼
[g]× + 2
𝑀
B2 V𝑇t , 󵄨 󵄨 󵄨 󵄨
2𝜋 ⋅ 󵄨󵄨󵄨V𝑜 󵄨󵄨󵄨 − 󵄨󵄨󵄨V𝑜 + ΔV𝑜 󵄨󵄨󵄨
𝜌V 𝐶𝑚 𝜌V6 (𝐶𝛼 ) 𝑚
(23) Δ𝑝𝑜 = , (29b)
𝑛⋅𝑑
𝑏2
j43 = B,
𝛼 2 where 𝑛 is the gun twist rate.
𝜌V4 𝐶𝑚
At the beginning of each iteration cycle, the DTTM-
where B2 = g × Vt . 4DOF matrix is computed and hence the projectile states
The time matrix T mentioned in (13b) is given by S𝑖+1
10×1 in (27) to obtain the projectile velocity difference ΔV𝑖+1
which will be reduced until reaching a predefined tolerance
Δ𝑡
I . O3×3 O3×1 O3×3 by adding Δ𝛼𝑅𝑖 to the initial state as
[3 2 ]
[ Δ𝑡 ]
[O I . O O ] −1
T10×10 [
=[ 3×3 3
2 3×1 3×3 ]. (24) Δ𝛼𝑅𝑖 = [m24 ] ΔV𝑖+1 , (30a)
Δ𝑡 ]
[ ]
[ O1×3 O1×3 O1×3 ] 𝛼𝑅𝑖 = 𝛼𝑅𝑖 + Δ𝛼𝑅𝑖 . (30b)
2
[ O3×3 O3×3 O3×1 I3 ]
Then, the estimated parameters are used as initials for the
By Applying (14), (19), and (25) into (12), the overall discrete next time step.
time transfer matrix is given by As shown in Figure 3, during each time step, the projectile
attitude and the corresponding angular momentum can be
S𝑖+1 𝑖
10×1 = [M10×13 + [I10×13 + T10×10 ⋅ J10×13 ] ⋅ A13×13 ] ⋅ S13×1 , estimated using the following procedure.
(25)
I O7×6
(1) The projectile total angle of attack can be computed
where, M10×13 = [ O7×7
3×7 O3×6 ], by
m11 m12 m13 m14 m15 󵄨 󵄨
𝛼𝑡 = 󵄨󵄨󵄨𝛼𝑅 󵄨󵄨󵄨 . (31)
[m21 m22 m23 m24 m25 ]
DTTM-4DOF = [
[m31
], (26)
m32 m33 m34 m35 ]
(2) The projectile’s axis of symmetry unit vector x is
[m41 m42 m43 m44 m45 ]
computed by [19]
S𝑖+1 𝑖
10×1 = [DTTM-4DOF] ⋅ S13×1 . (27)
x = 𝛼𝑅 + (cos 𝛼𝑡 ) i, (32)
All elements m𝑖𝑗 of the matrix DTTM-4DOF are listed in the
appendix. where i = Vt /|Vt |.
6 Advances in Mechanical Engineering

Radar data
Si[10×1] DTTM-4DOF [ X V p 𝛼R ]T Si+1 = DTTM ∗ Si −+
[Xi+1 Vi+1 ]T

Nominal data j=1


Yes

No
[Xo Vo 𝜃0 po ]T Calculate ΔVo No
ΔX (i=1) ≤ tolerance
Yes j=0
eq. (29a)
i = 0, j = 0 No
Calculate Δ𝛼Ri No ΔVi ≤ tolerance
Radar data length < i
eq. (30a)
Yes
No
i>0 Calculate 𝛼t , x 𝛼R
i= i+1
eq. (31)–(32)
Yes
Calculate h Calculate 𝛼̇ ̇
R , i, ẋ
eq. (36) eq. (33)–(35) Projectile impact point

No
Yes

Trajectory parameters Yes


6DOF model Altitude = 0
at the end of radar data

Figure 3: Projectile attitude and impact point determination flow chart.

(3) The rate of change of the projectile total velocity unit Table 1: Corresponding relation between the input point and the
vector is given by [19] sequence number.

𝜌𝑆ref V𝐶𝐿𝛼 1 Input point Sequence no.


i ̇= 𝛼𝑅 + [i × (g × i)] . (33) 1 Ground
2𝑚 V
2–13 𝑃13,1 , 𝑃14,2 , . . . , 𝑃24,12
(4) Due to the small variation in projectile repose angle 14 𝑃37,25
during flight [19], its rate of change is approximated 15 𝑃37,38
to be linear during each time step as 16 𝑃39,38
17 𝑃39,40
𝛼𝑅𝑖+1 − 𝛼𝑅𝑖
𝛼̇
𝑅 ≈ . (34) 18 𝑃41,40
Δ𝑡 19 𝑃41,42
20 𝑃44,0
(5) The rate of change of projectile’s axis of symmetry unit
vector ẋis computed by [19] 21 𝑃44,45
22–29 45 (beam 1)
ẋ= 𝛼̇ ̇
𝑅 + (cos 𝛼𝑡 ) i − (sin 𝛼𝑡 ) i. (35) 30 𝑃45,46
31–68 46 (beam 2)
(6) The projectile angular momentum divided by 𝐼𝑦 in 69 𝑃46,47
earth fixed coordinate is given by [19] 70–104 47 (beam 3)
105–126 48 (beam 4)
𝐼𝑥 ⋅ 𝑝 127–148 49 (beam 5)
h=( ) x + [x × x]̇. (36)
𝐼𝑦 149–229 50 (beam 6)
230 𝑃51,52
6. Simulated Trajectory Generation
In order to validate the proposed algorithm, a simulated 6.1. Natural Vibration Characteristics of the Self-Propelled
noiseless ballistic trajectory for 155 mm high explosive HE Artillery. To compute the vibration characteristics for
spin stabilized projectile is generated to simulate the Doppler 155 mm self-propelled artillery, the input points correspond-
radar velocity measurements data. The projectile mass, ing to sequence numbers of connection points and elements
length, center of mass measured from the nose, and the axial shown in Figure 1 are identified in Table 1.
and transverse mass moments of inertia are 46.5 kg, 0.902 m, All simulation and test results of natural frequencies for
0.593 m, 0.1585 kg/m2 , and 1.8816 kg/m2 , respectively. This the first sixteen ranks are listed in [8], where the simulation
trajectory is generated using the nonlinear 6-DOF trajectory results have a good agreement with that obtained by exper-
model [2, 19] including projectile initial disturbance problem iments. Figures 4 and 5 show part of the resulted system
with 52∘ firing elevating angle. mode shapes, where the 1st mode mainly represents the recoil
Advances in Mechanical Engineering 7

1 0.75
Mode shape

Lateral displacement (mm)


0.5 0.5

0 0.25

−0.5 0
0 40 80 120 160 200
Input point
−0.25
1st mode, X-direction
2nd mode, Z-direction −0.5
0 0.25 0.5 0.75 1
Figure 4: The 1st and 2nd mode shapes in 𝑋- and 𝑍-direction, xo /L b
respectively.
Y-direction
Z-direction
1
Figure 8: Projectile lateral displacement.
0.5
Y-direction

10
0

Roll angle (rad)


−0.5
5
−1
0 40 80 120 160 200
Input point
0
3rd mode 0 0.25 0.5 0.75 1
16th mode xo /L b

Figure 5: The 3rd and 16th mode shape in 𝑌-direction.


0.0015
Swing angle (rad)

8
0
xo (m)

4
−0.0015
0 0.25 0.5 0.75 1
0
0 0.25 0.5 0.75 1 xo /L b
Nondimensional launch time Plumb
Sidewise
Figure 6: Projectile longitudinal displacement.
Figure 9: Projectile angular displacement during launch.

800 motion of barrel, the 2nd mode represents the horizontal


revolving motions of turret, cradle, and barrel, the 3rd mode
Vx𝑜 (m/s)

400 mainly represents the elevating motions of cradle and barrel,


and the 16th mode mainly represents the transverse motion
of barrel and complex motions of other elements. The low
0 rank mode shapes do not represent the transverse elastic
0 0.25 0.5 0.75 1
deformation of the barrel that is only represented in high rank
xo /L b
[8].
6000
6.2. Computation of Projectile Initial Disturbance. The
A x𝑜 (gee)

3000 resulted projectile launch dynamics parameters with respect


to gun tube coordinate system are shown in Figures 6–10.
0
The gun tube coordinate system 𝑂3 𝑥𝑂𝑦𝑂𝑧𝑂 is defined as
0 0.25 0.5 0.75 1 follows: the origin 𝑂3 is the intersection point between
xo /L b axis of gun tube and the vertical plane of projectile’s axis of
symmetry, the 𝑥𝑂-axis is along gun tube axis of symmetry
Figure 7: Projectile longitudinal velocity and acceleration. and points to muzzle, the 𝑦𝑂-axis (plumb direction) is vertical
8 Advances in Mechanical Engineering

1500 0.6
Roll rate (rad/s)

1000 0.4

𝛼R|1 (rad)
0.2
500
0
0
0 0.25 0.5 0.75 1 −0.2
xo /L b
−0.4
0 0.1 0.2 0.3 0.4 0.5
6 Time (s)
Swing rate (rad/s)

0.3
3
0.15
0

𝛼R|2 (rad)
0
−3
−0.15
−6
0 0.25 0.5 0.75 1
−0.3
xo /L b 0 0.1 0.2 0.3 0.4 0.5
Plumb Time (s)
Sidewise 0.4

Figure 10: Projectile angular velocity during launch. 𝛼R|3 (rad) 0.2

−0.2
438
−0.4
433 0 0.1 0.2 0.3 0.4 0.5
V1 (m/s)

Time (s)
428
Estimated
6DOF-R
423
6DOF-N
418 Figure 12: Projectile total angle of attack vector versus flight time.
0 0.1 0.2 0.3 0.4 0.5
Time (s)
552 Table 2: Simulation results of initial disturbance of projectile.
547 Initial disturbance Plumb Sidewise
V2 (m/s)

542 Deflection angle/(10−3 rad) 0.71131 1.4600


Swing angle/(10−3 rad) 2.11265 −0.06758
537
Swing angle velocity/(rad/s) −2.92724 −3.31593
532
0 0.1 0.2 0.3 0.4 0.5
Time (s)
to 𝑥𝑂-axis and points up, the direction of 𝑧𝑂-axis (swing
3
direction) is determined by the right-hand role. The projectile
2 longitudinal displacement, velocity, and acceleration with
respect to gun tube axis are shown in Figures 6 and 7, where
V3 (m/s)

1
𝐿 𝑏 is the total barrel length. Figure 8 shows the projectile
0 center of gravity plumb and sidewise displacement with
−1 respect to gun tube axis as function of nondimensionalized
−2
longitudinal displacement.
0 0.1 0.2 0.3 0.4 0.5 The projectile axes orientation with respect to line of sight
Time (s) LOS, which includes the projectile roll, plumb and sidewise
swing components, angular displacement, and velocity, are
Estimated
6DOF-R
illustrated in Figures 9 and 10, respectively. The correspond-
6DOF-N ing projectile angular displacement and velocity components
at muzzle are shown in Table 2, where the deflection angle
Figure 11: Projectile velocity vector versus flight time. is the angle between the projectile velocity vector and LOS
Advances in Mechanical Engineering 9

Table 3: Simulation results of the projectile parameters at muzzle.

𝑉1 (m/s) 𝑉2 (m/s) 𝑉3 (m/s) 𝛼 (10−3 rad) 𝛽 (10−3 rad) 𝑝 (rad/s) 𝑞 (rad/s) 𝑟 (rad/s)
429.3847 551.9871 1.021021 1.40134 −1.52758 1279.72 −2.92724 −3.15983

1
30

𝛿𝛼R1 /𝛿t (rad/s)


0.8
15

0.6 0
x1

0.4 −15

−30
0.2 0 0.1 0.2 0.3 0.4 0.5
0 0.1 0.2 0.3 0.4 0.5
Time (s)
Time (s)
20
1

𝛿𝛼R2 /𝛿t (rad/s)


10
0.8
0
x2

0.6 −10

−20
0.4 0 0.1 0.2 0.3 0.4 0.5
0 0.1 0.2 0.3 0.4 0.5
Time (s)
Time (s)
30
0.4
𝛿𝛼R3 /𝛿t (rad/s)

15
0.2
0
0
x3

−15
−0.2
−30
−0.4 0 0.1 0.2 0.3 0.4 0.5
0 0.1 0.2 0.3 0.4 0.5
Time (s)
Time (s)
Estimated Estimated
6DOF-R 6DOF-R
6DOF-N 6DOF-N

Figure 13: The projectile’s axis of symmetry unit vector. Figure 14: Rate of change of the projectile AOA.

while projectile is just flying off from muzzle, and the swing projectile’s position and velocity vectors, using the initial
angle is the angle between projectile axes and LOS at the same flight conditions listed in Table 3. The Doppler radar output
moment. data frequency is assumed to be 100 Hz. The 6DOF-R data
The projectile initial flight conditions corresponding to is assumed to be the real projectile trajectory data. The
the self-propelled artillery launch dynamics, and hence, the corresponding impact point range, drift, maximum altitude,
projectile initial disturbance calculations proposed before are velocity, and flight time are 19689.2 m, 1027.85 m, 8180.9 m,
listed in Table 3, which include the projectile initial velocity 338.76 m/s, and 81.932 s, respectively.
𝑉1 , 𝑉2 , 𝑉3 , initial roll 𝑝, pitch 𝑞 and yaw 𝑟 rates, initial angle Secondly, a nondisturbed trajectory simulation has been
of attack 𝛼, and side-slip angle 𝛽. performed as nominal trajectory 6DOF-N, where the initial
flight conditions used are 700 m/s, 52∘ , and 1279.7 rad/s for
7. Estimation Results muzzle velocity, elevating angle, and spin rate, respectively.
The corresponding impact point range, drift, maximum
A computer program has been developed by using the non- altitude, velocity, and flight time are 19729 m, 995 m, 8162 m,
linear 6-DOF dynamic equations stated in [19]. All simula- 338.66 m/s, and 81.829 s, respectively.
tions were computed with fixed time step size of 0.001 s and Finally, the iterative algorithm shown in Figure 3 has been
integrated using Runge-Kutta-Gill method. applied to estimate the projectile attitude, using the generated
Firstly, a simulated noiseless Doppler radar velocity Doppler radar data 6DOF-R, where the initial flight condi-
measurements data 6DOF-R are generated, including the tions are as the data used to simulate the projectile nominal
10 Advances in Mechanical Engineering

90
30
80
𝛿x1 /𝛿t (s−1 )

15

h1 (rad/s)
0 70

−15 60
−30
0 0.1 0.2 0.3 0.4 0.5 50
0 0.1 0.2 0.3 0.4 0.5
Time (s)
Time (s)
20
100
10
𝛿x2 /𝛿t (s−1 )

90

h2 (rad/s)
0

−10 80

−20
0 0.1 0.2 0.3 0.4 0.5 70
Time (s) 0 0.1 0.2 0.3 0.4 0.5
Time (s)
30
20
15
𝛿x2 /𝛿t (s−1 )

10
0
h3 (rad/s)

0
−15
−10
−30
0 0.1 0.2 0.3 0.4 0.5
−20
Time (s) 0 0.1 0.2 0.3 0.4 0.5
Estimated Time (s)
6DOF-R Estimated
6DOF-N 6DOF-R
6DOF-N
Figure 15: Rate of change of the projectile’s axis of symmetry unit
vector. Figure 16: Projectile angular momentum per unit 𝐼𝑦 .

trajectory 6DOF-N. Figures 11, 12, and 13 show that the be −0.0205% and −0.0448%, respectively, as the radar data
estimated projectile velocity, AOA, and attitude, respectively, availability is 15% of total flight time.
are matching very well with the simulated Doppler radar
trajectory 6DOF-R, although the nominal (non-disturbed) 8. Conclusion
trajectory results 6DOF-N are mismatching the 6DOF-R.
Figures 14 and 15 show the rate of change of the projectile The main objective of this paper is determining the effect
AOA and attitude, respectively, during flight time, where the of self-propelled artillery dynamics on the projectile impact
estimated values are matched very well with 6DOF-R except point prediction accuracy, and hence, an iterative projectile
at some peaks/dips data points due to the large difference attitude determination algorithm was developed to accurately
between the data frequency obtained from radar and the compute the projectile ballistic trajectory and the corre-
simulation speed of the real trajectory data (1 : 10). Therefore, sponding impact point parameters by using a Doppler radar
the final estimated value shown in Figure 16, which rep- data. The projectile initial disturbance was computed by
resents the projectile angular momentum per-unit 𝐼𝑦 has solving the interaction between the self-propelled artillery
mismatched with 6DOF-R data during these peaks/dips data and the projectile dynamics during launch process. A discrete
points. time transfer matrix DTTM-4DOF was developed by lin-
For nominal trajectory data 6DOF-N, the impact range 𝜀𝑅 earizing the projectile nonlinear modified point mass equa-
and drift 𝜀𝐷 errors relative to 6DOF-R are 0.202% and −3.2%, tions using second-order Taylor series. The main advantages
respectively. All impact point parameters inaccuracies for the of using DTTM-4DOF are (1) linearly relating any two
proposed algorithm as function of simulated Doppler radar successive radar data, and (2) increasing the computation
data availability are shown in Figure 17, where these errors, speed due to solving a set of algebraic equations instead of
𝜀𝑅 and 𝜀𝐷, can be reduced by approximately 8 and 7 times, using traditional numerical methods. An iterative projectile
respectively, as the radar data is available for the first 10% attitude determination algorithm was developed to accurately
of total flight time and approximately by 10 and 70 times to compute the projectile ballistic trajectory and predict the
Advances in Mechanical Engineering 11

0
0.05

−0.01
0

𝜀R (%)
𝜀R (%)

−0.02

−0.05
−0.03

−0.1 −0.04
0 20 40 60 80 100 0 10 20 30 40 50 60 70 80 90 100
Data availability (%) Data availability (%)
10 1.2

5 0.6
𝜀D (%)

𝜀D (%)
0 0

−5 −0.6

−10 −1.2
0 20 40 60 80 100 0 10 20 30 40 50 60 70 80 90 100
Data availability (%) Data availability (%)

Figure 17: Impact range and drift error versus Doppler Radar data availability.

2
impact point parameters by using a Doppler radar data. In Δ𝑡
[m13 m14 m15 ] = [V̇𝑝 V̇𝛼 V̇W ] ,
order to validate the proposed algorithm, a simulated noise- 2
less trajectory data was generated as Doppler radar velocity Δ𝑡
measurements data 6DOF-R for 155 mm HE spin stabilized m21 = (I3 + j22 ) GΔ𝑡,
2
projectile using the nonlinear six-degree-of-freedom trajec-
2
tory model including projectile initial disturbance problem. Δ𝑡
m22 = I3 + V̇V Δ𝑡 + (j22 V̇V + j25 ẆV ) ,
A nominal ballistic trajectory 6DOF-N was generated with 2
52∘ firing elevating angle and without including the effect of 2
Δ𝑡
launch dynamics to show the importance of the projectile ini- m23 = V̇𝑝 Δ𝑡 + [j22 V̇𝑝 + V̇𝑝 𝑝𝑝̇] ,
tial disturbance. Based on the results obtained, the proposed 2
algorithm estimated the projectile velocity and attitude very Δ𝑡2
well compared to the 6DOF-R trajectory data. The impact m24 = V̇𝛼 Δ𝑡 + j22 V̇𝛼 ,
2
range and drift inaccuracies were reduced from 0.202% and 2
Δ𝑡
−3.2% in case of nominal trajectory 6DOF-N to be −0.0205% m25 = V̇W Δ𝑡 + j22 V̇W ,
and −0.0448% by applying the proposed algorithm as the 2
radar data availability is 15% of total flight time. As the projec- Δ𝑡2
tile dynamic stability increased, the radar data availability can m31 = j32 G ,
2
be decreased. Finally, the proposed algorithm can estimate
2
the projectile impact point accurately without computing the Δ𝑡
m32 = j32 [V̇V − ẆV ] ,
projectile initial disturbance, where different projectiles have 2
different initial disturbances based on many factors, such
Δ𝑡2
as, gun barrel wear and erosion, projectile initial position m33 = 1 + 𝑝𝑝̇Δ𝑡 + [j32 V̇𝑝 + 𝑝𝑝̇
2
] ,
and attitude inside barrel, and the propellant charge amount, 2
composition, and homogeneity. 2
Δ𝑡
[m34 m35 ] = j32 [V̇𝛼 V̇W ] ,
2
Appendix
m41 = [𝛼V̇
+ j42 Δ𝑡] G,
Modified Point Mass Discrete Time
+ j42 Δ𝑡] V̇V + j45 ẆV Δ𝑡,
m42 = [𝛼V̇
Transfer Matrix Elements
All DTTM-4DOF elements mentioned in (26) are expressed + j42 Δ𝑡] V̇𝑝 + j43 𝑝𝑝̇Δ𝑡,
m43 = [𝛼V̇
as follows:
+ j42 Δ𝑡] V̇𝛼 ,
m44 = [𝛼V̇
m11 = I3 + GΔ𝑡2 ,
2 + j42 Δ𝑡] V̇W .
m45 = [𝛼V̇
Δ𝑡
m12 = I3 ⋅ Δ𝑡 + V̇V , (A.1)
2
12 Advances in Mechanical Engineering

Acknowledgments [16] E. Miletics and G. Molnárka, “Taylor series method with


numerical derivatives for initial value problems,” Journal of
The research was supported by the Research Fund for Computational Methods in Sciences and Engineering, vol. 4, no.
the Doctoral Program of Higher Education of China 1, pp. 105–114, 2004.
(20113219110025), the Natural Science Foundation of China [17] B. Rong, X. Rui, and L. Tao, “Discrete time transfer matrix
Government (11102089), and the Program for New Century method for launch dynamics modeling and cosimulation of
Excellent Talents in University (NCET-10-0075). self-propelled artillery system,” Journal of Applied Mechanics,
vol. 80, no. 1, 2013.
[18] R. F. Lieske and R. L. McCoy, “Equations of Motion for a
References Modified Point Mass Trajectory,” Report R1314 (AD, 485869),
[1] F. Fresconi, G. Cooper, and M. Costello, “Practical assessment of U.S. Army Ballistic Research Laboratory, Aberdeen Proving
real-time impact point estimators for smart weapons,” Journal of Ground, Md, USA, 1966.
Aerospace Engineering, vol. 24, no. 1, pp. 1–11, 2011. [19] R. L. McCoy, Modern Exterior Ballistics: the Launch and Flight
[2] R. F. Lieske and R. L. McCoy, “Equations of Motion of a Dynamics of Symmetric Projectiles, Schiffer Publishing, 1999.
Rigid Projectile,” Report 1244 (AD, 441598), U.S. Army Ballistic [20] J. W. Bradley, “An Alternative form of the Modified Point-Mass
Research Laboratory, Aberdeen Proving Ground, Md, USA, Equation of Motion,” Report BRL-MR-3875 (AD A229514),
1964. U.S. Army Ballistic Research Laboratory, Aberdeen Proving
[3] M. Khalil, X. Rui, and H. Hendy, “Discrete time transfer Ground, Md, USA, 1990.
matrix method for projectile trajectory prediction,” Journal of
Aerospace Engineering, 2013.
[4] J. D. Pinezich, J. Heller, and T. Lu, “Ballistic projectile tracking
using CW doppler radar,” IEEE Transactions on Aerospace and
Electronic Systems, vol. 46, no. 3, pp. 1302–1311, 2010.
[5] I. I. Shapiro, The Prediction of Ballistic Missile Trajectories From
Radar Observations, McGraw-Hill, New York, NY, USA, 1958.
[6] E. M. Friedman and M. J. Amoruso, “Determining Thrust or
Drag of a Rocket-Assisted Projectile From Doppler Trajec-
tory Data,” Tech. Rep. ARAED-TR-85008, Army Armament
Research Development and Engineering Center, New Jersey, NJ,
USA, 1985.
[7] G. G. Dutta, A. Singhal, A. Kushari, and A. K. Ghosh, “Estima-
tion of drag coefficient from radar-tracked flight data of a cargo
shell,” Defence Science Journal, vol. 58, no. 3, pp. 377–389, 2008.
[8] X. Rui, L. Yun, Y. Lu, B. He, and G. Wang, Transfer Matrix
Method of Multibody System and Its Application, Science Press,
Beijing, China, 2008.
[9] X. Rui, Y. Lu, W. Lu, J. Qian, and G. Wang, “A study on the launch
dynamics of self-propelled artillery,” Acta Armamentarii, vol. 21,
supplement 1, pp. 38–40, 2000.
[10] G. P. O’Hara, “Dynamic analysis of a 155 mm cannon breech,”
Shock and Vibration, vol. 8, no. 3-4, pp. 215–221, 2001.
[11] J. A. Cordes, D. Carlucci, and R. Jafar, “Dynamics of a simplified
155-MM projectile,” in Proceedings of the 21st International Sym-
posium on Ballistics, vol. 2, pp. 1164–1170, Adelaide, Australia,
April 2004.
[12] M. Chowdhury, A. M. R. Frydman, J. Cordes, L. Reinhardt,
and D. Carlucci, “3-D finite-element gun launch simulation
of a surrogate Excalibur 155 mm guided artillery projectile-
modeling capabilities and its implications,” in Proceedings of
the 22nd International Symposium on Ballistics, pp. 259–267,
Vancouver, Canada, November 2005.
[13] X. Rui, G. Wang, Y. Lu, and L. Yun, “Transfer matrix method for
linear multibody system,” Multibody System Dynamics, vol. 19,
no. 3, pp. 179–207, 2008.
[14] X. Rui, B. Rong, G. Wang, and B. He, “Discrete time transfer
matrix method for dynamics analysis of complex weapon
systems,” Science China Technological Sciences, vol. 54, no. 5, pp.
1061–1071, 2011.
[15] X. Rui, B. He, Y. Lu, W. Lu, and G. Wang, “Discrete time transfer
matrix method for multibody system dynamics,” Multibody
System Dynamics, vol. 14, no. 3-4, pp. 317–344, 2005.
Hindawi Publishing Corporation
Advances in Mechanical Engineering
Volume 2013, Article ID 854583, 9 pages
http://dx.doi.org/10.1155/2013/854583

Research Article
Study on the Dynamics of Laser Gyro Strapdown Inertial
Measurement Unit System Based on Transfer Matrix Method for
Multibody System

Gangli Chen, Xiaoting Rui, Fufeng Yang, Jianshu Zhang, and Qinbo Zhou
Institute of Launch Dynamics, Nanjing University of Science and Technology, Nanjing 210094, China

Correspondence should be addressed to Xiaoting Rui; ruixt@163.net

Received 26 August 2013; Accepted 1 December 2013

Academic Editor: Caishan Liu

Copyright © 2013 Gangli Chen et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

The dynamic test precision of the strapdown inertial measurement unit (SIMU) is the basis of estimating accurate motion of various
vehicles such as warships, airplanes, spacecrafts, and missiles. So, it is paid great attention in the above fields to increase the dynamic
precision of SIMU by decreasing the vibration of the vehicles acting on the SIMU. In this paper, based on the transfer matrix
method for multibody system (MSTMM), the multibody system dynamics model of laser gyro strapdown inertial measurement
unit (LGSIMU) is developed; the overall transfer equation of the system is deduced automatically. The computational results
show that the frequency response function of the LGSIMU got by the proposed method and Newton-Euler method have good
agreements. Further, the vibration reduction performance and the attitude error responses under harmonic and random excitations
are analyzed. The proposed method provides a powerful technique for studying dynamics of LGSIMU because of using MSTMM
and its following features: without the global dynamics equations of the system, high programming, low order of system matrix,
and high computational speed.

1. Introduction Dynamics analysis is the basis of LGSIMU system design


and error compensation. The simplified model which has two
As there is no need to use external information for measure- degrees of freedom is often used for the dynamics analysis of
ments, LGSIMU is widely used in tactical missiles, chariots, LGSIMU [7, 8]. Due to the excessive simplification, the model
torpedoes, warships, airplanes, and so on [1–4]. Inertial can not accurately reflect the influence of some realistic
measurement unit (IMU) usually operates in bad dynamic factors on the system dynamics performance. To analyze the
environment such as vibration, shock, and overload. For dynamics performance of LGSIMU, attention is gradually
instance, broadband random excitations are produced by paid to more realistic dynamics model, which is established
missiles in powered phase and reentry phase flights as a result based on the theory of multibody system dynamics. Liu et
of its engine thrust, jet noise, and turbulent boundary layer al. [9] used the theory of multibody system dynamics to
pressure. LGSIMU is fixed on the vehicle and undergoes establish the dynamics model of LGSIMU and analyze its
the dynamic environments directly, which makes its opera- dynamic characteristics. The ordinary dynamics methods
tional environment very bad. Thus, its dynamic accuracy is are complicated due to the necessity to deduce the global
declined. dynamics equation and inconvenience for engineering appli-
In order to improve the dynamic accuracy, passive control cations. Rui et al. [10] put forward the transfer matrix method
method is usually adopted to restrain the dynamics response for multibody system (MSTMM) which does not need to
of LGSIMU system in the dynamic environment. The meth- establish the global dynamics equation and is highly stylized.
ods such as optimizing the bracket structure, matching the MSTMM has been widely used in the fields of multiple launch
stiffness of isolator, and improving the sustaining mode are rocket system, shipboard gun system, spacecraft, underwater
the main ways of passive control for the moment [5, 6]. towed system, and so on [11–15]. Yang et al. [16] used
2 Advances in Mechanical Engineering

MSTMM to establish the dynamics model of LGSIMU and 0


analyzed the effect of shock on the output precision. Ma et
1 m
al. [17] used MSTMM to simulate the output power spectral
density (PSD) of LGSIMU undergoing random excitation.
Their works are both based on the body dynamics equations 2 k d
in MSTMM and modal superposition. w

As a result of the bad dynamic environment, the shifting 3


of gyroscope centroid and the drift produced by structure
deformation greatly reduce the accuracy of strapdown IMU. Figure 1: One degree of freedom system used to introduce
However, due to a lack of the error analysis approach, ordi- MSTMM.
nary inertial instruments can hardly compensate the error
caused by dynamic environment. Savage [18] established the f1,0
technique for compensating the attitude, velocity, and posi-
tion errors under sinusoidal and random vibrations. Savage x1,2
also presented the simplified dynamics model of strapdown
f1,2
IMU system which had two degrees of freedom including
linear and angular motions. It provides an effective approach
Figure 2: Free body diagram of mass 1.
for error analysis and compensation, but the simplified model
cannot reflect the reality perfectly. So it has great engineering
significance to establish more realistic dynamics model for
error analysis of strapdown IMU. Applying Newton’s second law of motion to the lumped mass
In this paper, MSTMM is used to establish more realistic yields
dynamics model of LGSIMU system, elements transfer equa- ̈ = 𝑓1,2 − 𝑓1,0 .
𝑚𝑥1,2 (2)
tions, and transfer matrices. The overall transfer equation
and overall transfer matrix of the system are derived based For the system undergoing harmonic excitation 𝑤 =
on the MSTMM and its automatic deduction method [19]. 𝑊e , its steady state responses can be written as 𝑥1,0 =
iΩ𝑡
The method for computing frequency response functions
𝑋1,0 eiΩ𝑡 , 𝑥1,2 = 𝑋1,2 eiΩ𝑡 , 𝑓1,2 = 𝐹1,2 eiΩ𝑡 , 𝑓1,0 = 𝐹1,0 eiΩ𝑡
of LGSIMU system based on MSTMM is presented. The
[20]. Substituting these equations into (1) and (2), the transfer
responses of LGSIMU system undergoing various sinusoidal
equation of lumped mass 1 can be found as
and random excitations are obtained. Then the dynamics
performance is analyzed. Combining the formulae of errors Z1,0 = U1 Z1,2 , (3)
caused by vibration, the attitude errors of LGSIMU under
sinusoidal and random vibrations are discussed. The method- T
where Z = [𝑋 𝐹] is the state vector which consists of the
ology presented in this paper provides an effective technique complex amplitudes of displacement and internal force. The
for improving the dynamics performance of LGSIMU under transfer matrix is
dynamic environment.
1 0
U1 = [ ]. (4)
𝑚Ω2 1
2. Introduction to MSTMM: A Simple Example
For hinge 2, considering its constitutive relation, one can
In order to describe the proposed method presented in the obtain
following sections conveniently, the simple system shown in
Figure 1 is taken as an example. The system consists of lumped ̇ − 𝑥3,2
𝑓1,2 = 𝑓3,2 = −𝑘 (𝑥1,2 − 𝑥3,2 ) − 𝑑 (𝑥1,2 ̇ ). (5)
mass 1, spring-and-damper hinge 2, and base 3. The free
By substituting the steady state responses into (5), the
boundary is numbered as 0. The mass of the lumped mass
transfer equation of hinge 2 can be acquired as
is 𝑚, and the stiffness and viscous coefficients of hinge 2 are 𝑘
and 𝑑, respectively. The input and output of the system are the Z1,2 = U2 Z3,2 , (6)
displacements of base 3 and lamped mass 1, respectively. The
input of the system is assumed as harmonic excitation 𝑤 = T
where the state vector also has the form Z = [𝑋 𝐹] and the
𝑊eiΩ𝑡 . In the following of this section, the frequency response transfer matrix is
function of the system will be derived by using MSTMM.
1
Using the sign conventions in [10], the positive directions 1 −
of displacements and forces are shown in the free body U2 = [ 𝑘 + iΩ𝑑 ] . (7)
0 1
diagrams of Figures 2 and 3. For the lumped mass, the
displacements of output end and input end are equal, namely, Combining (3) and (6), the overall transfer equation of
the system can be obtained as

𝑥1,0 = 𝑥1,2 . (1) Uall Zall = 0, (8)


Advances in Mechanical Engineering 3

x1,2 3
6 5
f1,2

1 4 2
x3,2
7
10
f3,2 11
16
Figure 3: Free body diagram of spring-and-damper hinge 2.
15

8 12 y
9
where O
T 13 14 z x
Uall = [−I T] , Zall = [ZT1,0 ZT3,2 ] ,
Figure 4: Dynamics model of LGSIMU system.
1
1 − (9)
[ 𝑘 + iΩ𝑑 ]
T = U 1 U2 = [
[
].
2 𝑚Ω2 ] 1∼3, platform 7, and foundation 12 are considered as rigid
𝑚Ω 1 −
[ 𝑘 + iΩ𝑑 ] bodies, respectively; since the accelerometers are fixed on
the platform and their masses are relatively small, they
The boundary conditions of the system are are included in the platform; the connections 4∼6 between
T T platform and gyros are considered as spatial spring-and-
Z1,0 = [𝑋1,0 0] , Z3,2 = [𝑊 𝐹3,2 ] . (10) damper hinges; meanwhile, platform and foundation are
linked by isolators 8∼11, which are regarded as spring-and-
Substituting the boundary conditions into (8) yields damper hinges longitudinally vibrating in space. The motion
of the vehicle is the input of LGSIMU system. The isolators
Uall Zall = −B𝑊, (11) 13∼16 between foundation and platform are considered as
spring-and-damper hinges longitudinally vibrating in space.
where
The inertial coordinate system shown in Figure 4 is used
1 to describe the motion of elements. The state vectors of the
−1 −
[ 𝑘 + iΩ𝑑 ] input and output points of every element and the boundary
Uall = [ ],
T
Zall = [𝑋1,0 𝐹3,2 ] , [ 𝑚Ω2 ]
points take the form
0 1− (12)
[ 𝑘 + iΩ𝑑 ] T
Z = [𝑋, 𝑌, 𝑍, Θ𝑥 , Θ𝑦 , Θ𝑧 , 𝑀𝑥 , 𝑀𝑦 , 𝑀𝑧 , 𝑄𝑥 , 𝑄𝑦 , 𝑄𝑧 ] . (15)
1
B=[ ].
𝑚Ω2 The state variables in the state vector are complex
amplitudes of displacements, angular displacements, internal
Solving (11), the unknown elements in the state vectors of forces, and internal torques when the system undergoes
boundary points can be obtained as harmonic excitation, respectively.
−1 𝑊 (𝑘 + iΩ𝑑) 1
Zall = −Uall B𝑊 = [ 2] . (13) 4. Topology Figure of the Dynamics
−𝑚Ω2 + 𝑘 + iΩ𝑑 −𝑚Ω
Model of LGSIMU System
As a result, the frequency response function of the system
is The dynamics model of LGSIMU system is constructed with
dynamics elements including bodies and hinges. In order
𝑋1,0 𝑘 + iΩ𝑑 to describe the transfer relationship among the state vectors
𝐻 (Ω) = = . (14)
𝑊 −𝑚Ω2 + 𝑘 + iΩ𝑑 of elements in the system, the topology figure of the model
will be used for deducing the overall transfer equation of
Equation (14) can easily be verified using classical fre- LGSIMU system. Isolators 8∼11 and 13∼16 can be treated
quency response function analysis (i.e., Newton’s second law as one equivalent hinge element, respectively, which will be
of motion and Fourier transform). shown in Section 5.3 ; thus the dynamics model of LGSIMU
system can be dealt with a tree system. Its topology figure can
3. Dynamics Model of LGSIMU System be got readily as illustrated in Figure 5.
Besides the sign conventions introduced in [10], the sign
According to the elements dynamics properties of LGSIMU conventions used in this paper are introduced as follows.
system, the dynamics model shown in Figure 4 is established
as a multibody system. The system can be divided into (1) A circle I denotes a body element and the number
elements which can be modeled as follows: three gyros inside it is the sequence number of the body element.
4 Advances in Mechanical Engineering

0 0 0 0 where Z𝑖,I and Z𝑖,O denote the state vectors of input and
output ends of element 𝑖, respectively. The transfer matrices
13 14 15 16
are

I3 −̃lIO O3×3 O3×3


12 [ O
[ I O3×3 O3×3 ]
]
U𝑖 = [ 3×3 3
̃l ] ,
[𝑚Ω lCO −Ω (𝑚̃lIÕlIC + JI ) I3
2̃ 2
IO ] (17)
8 9 10 11 2
[ 𝑚Ω I3 −𝑚Ω2̃lIC O3×3 I3 ]

5 6 (𝑖 = 1, 2, 3, 12) ,
0 2 7 3 0
where C denotes the mass center and 𝑚 is the mass of rigid
4 body, JI is the inertia matrix relative to the input point I, ̃lIO
and ̃lIC are the cross product matrices from the input point I
Root
1 to the output point O and mass center C, respectively, and ̃lCO
0 is the cross product matrix from the mass center C to output
point O.
Figure 5: Topology figure of LGSIMU system.

5.2. Transfer Equations and Transfer Matrices of Spatial


Spring-and-Damper Hinges 4∼6. The transfer equations of
elements 4∼6 are
(2) An arrow → denotes a hinge element and the transfer
direction of state vectors, and the number beside it is Z𝑖,O = U𝑖 Z𝑖,I (𝑖 = 4, 5, 6) . (18)
the sequence number of the hinge element.
The transfer matrices are
I3 O3×3 O3×3 U14
(3) Body element 7 is dealt with three input ends and one [O3×3 I3 U23 O3×3 ]
output end and the other body elements are dealt with U𝑖 = [ ]
[O3×3 O3×3 I3 O3×3 ] (𝑖 = 4, 5, 6) , (19)
single input end and single output end.
[O3×3 O3×3 O3×3 I3 ]
(4) For a nonboundary end, the first and second subscript where
𝑖 and 𝑗 (𝑖, 𝑗 ≠0) in the state vector Z𝑖,𝑗 of the end 1
denote the sequence numbers of the adjacent body [− 𝑘 + iΩ𝑑 0 0 ]
element and hinge element, respectively. For a bound- [ 𝑥 𝑥 ]
[ 1 ]
ary end, the second subscript 𝑗 = 0 in the state vector U14 [
=[ 0 − 0 ],
𝑘 + iΩ𝑑 ]
Z𝑖,𝑗 , that is, the second subscript means boundary [ 𝑦 𝑦 ]
[ 1 ]
end, and the first subscript 𝑖 stands for the sequence 0 0 −
number of the element involved. [ 𝑘𝑧 + iΩ𝑑𝑧 ]
(20)
1
0 0
[ 𝑘𝑥󸀠 + iΩ𝑑𝑥󸀠 ]
(5) The boundary end corresponding to element 1 is [ ]
[ 1 ]
considered as the root, its state vector is denoted as U23 =[
[
0 󸀠 + iΩ𝑑󸀠
0 ].
]
Z1,0 , other boundary ends are considered as the tips, [ 𝑘𝑦 𝑦 ]
and their state vectors are denoted as Z2,0 , Z3,0 , and [ 1 ]
0 0 󸀠 󸀠
Z13∼16,0 , respectively. The transfer directions are from [ 𝑘𝑧 + iΩ𝑑𝑧 ]
the tips to the root.
𝑘𝑥 , 𝑘𝑦 , 𝑘𝑧 represent the stiffness coefficients of linear spring,
𝑘𝑥󸀠 , 𝑘𝑦󸀠 , 𝑘𝑧󸀠 denote the stiffness coefficients of rotary spring, 𝑑𝑥 ,
5. Transfer Equations and Transfer 𝑑𝑦 , 𝑑𝑧 stand for the damping coefficients of linear damper,
Matrices of LGSIMU Elements and 𝑑𝑥󸀠 , 𝑑𝑦󸀠 , 𝑑𝑧󸀠 denote the damping coefficients of rotary
damper, respectively.
5.1. Transfer Equations and Transfer Matrices of Gyros 1∼3 and
Foundation 12. The gyros 1∼3 and foundation 12 are rigid
bodies with single input end and single output end and their 5.3. Transfer Equations and Transfer Matrices of Isolators 8∼11
transfer equations are and 13∼16. As four isolators are sandwiched between two
rigid bodies, they can be treated as one new element, namely,
an equivalent spring-and-damper hinge with single input end
and single output end. The input and output points of the
Z𝑖,O = U𝑖 Z𝑖,I (𝑖 = 1, 2, 3, 12) , (16) new element can be selected at any positions on the two rigid
Advances in Mechanical Engineering 5

16 Foundation (𝑎𝑖 , 𝑏𝑖 , 𝑐𝑖 ) (𝑖 = 13 ∼ 16). The same goes for their output points
in 𝑂2 𝑥𝑦𝑧. The stiffness and damping coefficients of isolators
y 13∼16 are (𝑘𝑥𝑖 , 𝑘𝑦𝑖 , 𝑘𝑧𝑖 ) and (𝑑𝑥𝑖 , 𝑑𝑦𝑖 , 𝑑𝑧𝑖 ) (𝑖 = 13 ∼ 16),
x
O2 15 respectively. Simplify the system of forces acting on the input
z y
O1
ends of isolators 13∼16 into a system of forces only acting on
x the input point of the new element, and simplify the system
z
of forces acting on the output ends of isolators 13∼16 into a
13
Vehicle system of forces only acting on the output point of the new
14
element. Further, since the principle vectors and principle
Figure 6: Equivalent model of isolators. moments of the two new systems of forces should be equal,
the transfer equation of the equivalent spring-and-damper
hinge can be obtained as
bodies, respectively. Hereby, isolators 13∼16 are taken as an
example to deduce the transfer equation and transfer matrix Z13∼16,O = U13∼16 Z13∼16,I . (21)
of the new element. The input and output points of isolator
13 are selected as those of the new element, respectively. The transfer matrix is
Two coordinate systems 𝑂1 𝑥𝑦𝑧 and 𝑂2 𝑥𝑦𝑧 shown in Figure 6
are introduced for deduction, whose origins are at the −1
equilibrium positions of the input and output points of the U13∼16 = [ I6 (K13∼16 + iΩD13∼16 ) ] , (22)
new element. And their directions are the same with the O6×6 I6
coordinate system shown in Figure 4. The coordinates of
the input points of isolators 13∼16 in 𝑂1 𝑥𝑦𝑧 are noted as where Ω is the frequency of harmonic excitation,

K13∼16 =

16 16 16 16 16
[ 0 − ∑ 𝑘𝑦𝑖 𝑐𝑖 − ∑ 𝑘𝑧𝑖 𝑏𝑖 ∑ (𝑘𝑦𝑖 𝑐𝑖2 + 𝑘𝑧𝑖 𝑏𝑖2 ) − ∑ 𝑘𝑧𝑖 𝑎𝑖 𝑏𝑖 − ∑ 𝑘𝑦𝑖 𝑎𝑖 𝑐𝑖 ]
[ 𝑖=13 𝑖=13 𝑖=13 𝑖=13 𝑖=13 ]
[ ]
[ 16 16 16 16 16 ]
[ 2 2 ]
[ ∑ 𝑘𝑥𝑖 𝑐𝑖 0 − ∑ 𝑘 𝑎 − ∑ 𝑘 𝑏 𝑎 ∑ (𝑘 𝑐 + 𝑘 𝑎 ) − ∑ 𝑘 𝑏 𝑐 ]
[ 𝑧𝑖 𝑖 𝑧𝑖 𝑖 𝑖 𝑥𝑖 𝑖 𝑧𝑖 𝑖 𝑥𝑖 𝑖 𝑖 ]
[ 𝑖=13 𝑖=13 𝑖=13 𝑖=13 𝑖=13 ]
[ 16 ]
[ 16 16 16 16 ]
[− ∑ 𝑘 𝑏 ∑ 𝑘 𝑎 0 − ∑ 𝑘 𝑐 𝑎 − ∑ 𝑘 𝑐 𝑏 ∑ (𝑘 𝑏 2
+ 𝑘 𝑎2 ]
) (23)
[ 𝑥𝑖 𝑖 𝑦𝑖 𝑖 𝑦𝑖 𝑖 𝑖 𝑥𝑖 𝑖 𝑖 𝑥𝑖 𝑖 𝑦𝑖 𝑖 ]
[ 𝑖=13 𝑖=13 𝑖=13 𝑖=13 𝑖=13 ]
[ ].
[ 16 16 16 ]
[ ]
[ − ∑ 𝑘𝑥𝑖 0 0 0 − ∑ 𝑘𝑥𝑖 𝑐i ∑ 𝑘𝑥𝑖 𝑏i ]
[ 𝑖=13 ]
[ 𝑖=13 𝑖=13 ]
[ ]
[ 16 16 16 ]
[ 0 − ∑ 𝑘 0 ∑ 𝑘 𝑐 0 − ∑ 𝑘 𝑎 ]
[ 𝑦𝑖 𝑦𝑖 i 𝑦𝑖 i ]
[ 𝑖=13 𝑖=13 𝑖=13 ]
[ ]
[ 16 16 16 ]
[ ]
0 0 − ∑ 𝑘𝑧𝑖 − ∑ 𝑘𝑧𝑖 𝑏i ∑ 𝑘𝑧𝑖 𝑎i 0
[ 𝑖=13 𝑖=13 𝑖=13 ]

Replacing the stiffness coefficients (𝑘𝑥𝑖 , 𝑘𝑦𝑖 , 𝑘𝑧𝑖 ) (𝑖 = 13 ∼ 16) platform and gyros 2∼3, respectively. The output point is
in K13∼16 by damping coefficients (𝑑𝑥𝑖 , 𝑑𝑦𝑖 , 𝑑𝑧𝑖 ) (𝑖 = 13 ∼ 16), the connection point between the platform and gyro 1. The
one can immediately achieve D13∼16 . The transfer equation transfer equation of platform 7 is
and transfer matrix of the equivalent spring-and-damper
Z7,O = U7 Z7,I1 + U7,I2 Z7,I2 + U7,I3 Z7,I3 . (24)
hinge of elements 8∼11 have the same forms as (21) and (22).
The transfer matrices are
5.4. Transfer Equations and Transfer Matrices of Platform 7. I3 −̃lI1 O O3×3 O3×3
Platform 7 is considered as a rigid body with three input ends [ ]
[ O3×3 I3 O3×3 O3×3 ]
and single output end. The first input end is the connection U7 = [
[𝑚Ω2̃l 2 ̃ ̃ ̃l ]
],
point between platform 7 and isolator 8 and the second [ CO −Ω (𝑚lI1 O lI1 C + JI1 ) I3 I1 O ]
2 2̃
and third input points are the connection points between [ 𝑚Ω I3 −𝑚Ω lI1 C O3×3 I3 ]
6 Advances in Mechanical Engineering

O3×3 O3×3 O3×3 O3×3 theory of vibration [20], the steady state response can be
[O3×3 O3×3 O3×3 O3×3 ] expressed as 𝑦7,C = 𝑌7,C eiΩ𝑡 and the complex amplitude 𝑌7,C
U7,I𝑟 =[ ] (𝑟 = 2, 3) .
[O3×3 O3×3 I3 ̃lI𝑟 O ] is the frequency response of LGSIMU system.
[O3×3 O3×3 O3×3 I3 ] The boundary conditions of LGSIMU system are
(25) T
Z1,0 = [𝑋, 𝑌, 𝑍, Θ𝑥 , Θ𝑦 , Θ𝑧 , 0, 0, 0, 0, 0, 0]1,0 ,
The geometrical equations of platform 7 are
T
H7 Z7,I1 = H7,I𝑟 Z7,I𝑟 , 𝑟 = 2, 3 (26) Z13∼16,0 = [0, 1, 0, 0, 0, 0, 𝑀𝑥 , 𝑀𝑦 , 𝑀𝑧 , 𝑄𝑥 , 𝑄𝑦 , 𝑄𝑧 ]13∼16,0 ,
T
where Z2,0 = [𝑋, 𝑌, 𝑍, Θ𝑥 , Θ𝑦 , Θ𝑧 , 0, 0, 0, 0, 0, 0]2,0 ,
I3 O3×3 O3×3 O3×3
H7 = [ ], T
Z3,0 = [𝑋, 𝑌, 𝑍, Θ𝑥 , Θ𝑦 , Θ𝑧 , 0, 0, 0, 0, 0, 0]3,0 .
O3×3 I3 O3×3 O3×3
(27) (32)
I ̃lI1 I𝑟 O3×3 O3×3
H7,I𝑟 =[ 3 ] (𝑟 = 2, 3) .
O3×3 I3 O3×3 O3×3 Substituting these boundary conditions into (28) yields
The notations in these matrices are similar to those in Uall Zall = −B, (33)
Section 5.1.
where Uall is a 24 × 24 square matrix composed of the 1st∼
6. Automatic Deduction of the Overall 6th, 19th∼30th, and 37th∼42nd columns of Uall , Zall is a
Transfer Equation of LGSIMU System column matrix consisting of the unknown elements in Zall ,
and B is the 14th column of Uall . For a damped system, Uall
According to the topology figure and the automatic deduc- is a nonsingular square matrix. Solving (33), the unknown
tion method of MSTMM, one can obtain the overall transfer elements in the state vectors of boundary points can be
equation of LGSIMU system: acquired as
Uall Zall = 0. (28) −1
Zall = −Uall B. (34)
The overall transfer matrix is
Using the transfer equations of elements, one can easily
−I12 T13∼16−1 T2−1 T3−1 obtain the frequency response function:
Uall = [O6×12 G13∼16−7 G2−7 O6×12 ] , (29)
[O6×12 G13∼16−7 O6×12 G3−7 ] 𝐻 (Ω) = 𝑌7,C = E7,I1 C U8∼11 U12 U13∼16 Z13∼16,0 , (35)
where where E7,I1 C = [0, 1, 0, −𝑐C , 0, 𝑎C , 0, 0, 0, 0, 0, 0], and (𝑎C , 𝑏C ,
T13∼16−1 = U1 U4 U7 U8∼11 U12 U13∼16 , 𝑐C ) denotes the coordinate of the platform’s mass center in the
body fixed coordinate system.
T2−1 = U1 U4 U7,5 U5 U2 , Similarly, the frequency response functions of other
points can be got. Using the frequency response function, one
T3−1 = U1 U4 U7,6 U6 U3 , can easily obtain the responses under sinusoidal and random
(30)
G13∼16−7 = −H7 U8∼11 U12 U13∼16 , excitations.
If the input is 𝑤 = 𝑊 sin(Ω𝑡+𝛼), the steady state response
G2−7 = H7,5 U5 U2 , is
G3−7 = H7,6 U6 U3 . 𝑦7,C = 𝑊 |𝐻 (Ω)| sin (Ω𝑡 + 𝛼 + 𝜑 (Ω)) , (36)
Zall is a column matrix consisting of the state vectors of system where |𝐻(Ω)| is the amplitude of 𝐻(Ω) and 𝜑(Ω) is the phase
boundary points, that is angle of 𝐻(Ω).
T If the input is a stationary random excitation 𝑤(𝑡) with
Zall = [ZT1,0 ZT13∼16,0 ZT2,0 ZT3,0 ] . (31) PSD 𝑆𝑤 (Ω), the response is also a stationary random process
and its PSD can be expressed as
7. Dynamic Responses of LGSIMU System
𝑆𝑦7,C (Ω) = |𝐻 (Ω)|2 𝑆𝑤 (Ω) . (37)
In order to introduce the method based on MSTMM to
calculate the frequency response, the input and output are For the case that the input is a stationary acceleration
assumed as the displacements of the vehicle and the mass random excitation 𝑤(𝑡)̈ with PSD 𝑆𝑤̈ (Ω) and the output is
center of platform along y-axis, respectively. Setting the input ̈ (𝑡) whose PSD
the acceleration of platform’s mass center 𝑦7,C
to be a harmonic excitation 𝑤 = eiΩ𝑡 and according to the is denoted as 𝑆𝑦7,C
̈ (Ω), regarding the relationship between
Advances in Mechanical Engineering 7

acceleration PSD and displacement PSD 𝑆𝑤̈ (Ω) = Ω4 𝑆𝑤 (Ω), If the response PSD of LGSIMU system undergoing
4
𝑆𝑦7,C
̈ (Ω) = Ω 𝑆𝑦7,C (Ω), and (37), one can obtain random excitation is 𝑆O (Ω), the mean value of the attitude
error caused by random vibration is
2
̈(Ω) = |𝐻 (Ω)| 𝑆𝑤̈
𝑆𝑦7,C (Ω) . (38)
𝐸 (𝛿Φ̇Algo )
According to the Wiener-Khintchine theorem, the auto- ∞ 󵄨󵄨 1 sin ΩT𝑙 󵄨󵄨
󵄨 󵄨
correlation function of the response can be obtained as = ∫ Ω 󵄨󵄨󵄨[1 + (1 − cos ΩT𝑙 )] − 1󵄨󵄨󵄨 𝑆O (Ω) dΩ.
0 󵄨󵄨 3 ΩT𝑙 󵄨󵄨
(44)
1 ∞
𝑅𝑦7,C
̈(𝜏) = ∫ 𝑆 ̈(Ω) eiΩ𝜏 dΩ. (39)
2𝜋 −∞ 𝑦7,C
9. Numerical Example
Substituting (38) into (39) and setting 𝜏 = 0, the response
For the LGSIMU system shown in Figure 4, the mass and
mean square value can be written as
inertia matrix of foundation 12 are given as
1 ∞ 𝑚12 = 1.2 kg,
𝜙𝑦27,C
̈= 𝑅𝑦7,C
̈(0) = ∫ |𝐻 (Ω)|2 𝑆𝑤̈
(Ω) dΩ. (40)
2𝜋 −∞
0.0045 0 0 (45)
Similarly, the response PSDs and mean square values of J12,C = [ 0 0.0045 0 ] kg ⋅ m2 .
other points can be readily achieved. Then one can analyze [ 0 0 0.0045]
the dynamics performance and errors of LGSIMU system.
The mass and inertia matrix of platform 7 are
8. Error Analysis of LGSIMU 𝑚7 = 2 kg,
In order to improve the precision of LGSIMU in the dynamic 0.0075 0 0 (46)
environment, the errors must be compensated. According to J7,C = [ 0 0.0075 0 ] kg ⋅ m2 .
the error analysis method presented in [18], the attitude errors 0 0 0.0075]
are taken as examples to discuss the errors under sinusoidal [
and random vibrations in this section. The masses and inertia matrices of gyros 1∼3 are
Assuming that LGSIMU system is excited by a sinusoidal
excitation 𝑊 sin (Ω𝑡 + 𝛼), the angular motion response of the 𝑚𝑖 = 0.2 kg,
platform can be written as
0.0006 0 0
𝜃 (𝑡) = e𝑥 𝜃𝑥 (𝑡) + e𝑦 𝜃𝑦 (𝑡) + e𝑧 𝜃𝑧 (𝑡) , (41) J𝑖,C = [ 0 0.0006 0 ] kg ⋅ m2 (𝑖 = 1 ∼ 3) .
[ 0 0 0.0006]
where (47)

𝜃𝑥 (𝑡) = Θ𝑥 sin (Ω𝑡 + 𝜙𝑥 ) , The amplitude and phase frequency responses of the
LGSIMU system got by the proposed method and Newton-
𝜃𝑦 (𝑡) = Θ𝑦 sin (Ω𝑡 + 𝜙𝑦 ) , (42) Euler method are shown in Figures 7 and 8, respectively. It
can be seen that the results got by the two methods have good
𝜃𝑧 (𝑡) = Θ𝑧 sin (Ω𝑡 + 𝜙𝑧 ) . agreements, which verifies the effectiveness of the proposed
method. The attitude error rate under sinusoidal vibration
e𝑥 , e𝑦 , and e𝑧 are the unit vectors of the body-fixed coordinate is demonstrated in Figure 9, where the excitation frequency
frame of platform. Θ𝑥 , Θ𝑦 , Θ𝑧 and 𝜙𝑥 , 𝜙𝑦 , 𝜙𝑧 are the varies from 10 to 2000 Hz, the time interval is 20 ms, and the
amplitudes and phase angles of the angular motion steady amplitude of excitation is 0.2 mm. From this figure, it can be
state responses, respectively, which can be computed using concluded that when the excitation frequency is close to the
MSTMM as presented above. natural frequencies of the LGSIMU system (25 Hz, 46 Hz, and
The error caused by sinusoidal vibration is 85 Hz), attitude error rate reaches maximum values. So the
natural frequencies cannot be close to excitation frequency.
1 Currently, the vibration reduction system of LGSIMU is
𝛿Φ̇Algo = e𝑧 ΩΘ𝑥 Θ𝑦 sin (𝜙𝑥 − 𝜙𝑦 ) in the development phase. Typically, its vibration reduction
2
(43) index is the root mean square value of acceleration response
1 sin ΩT𝑙 should be less than 2 g, and the PSD curve comes down
× {[1 + (1 − cos ΩT𝑙 )] − 1} ,
3 ΩT𝑙 rapidly after 200 Hz [21]. In the following, the dynamics
response and the attitude error response of the LGSIMU
where 𝛿Φ̇Algo is the average attitude error rate, 𝑙 is high speed system under random excitation is computed and analyzed.
algorithm computation cycle index, and T𝑙 is time interval The PSD of the excitation is shown in Figure 10 and its
between 𝑙 cycles. root mean square value is 6.03 g. The PSD of the centroid
8 Advances in Mechanical Engineering

10 ×10−4
2
0
1
−10

Attitude error rate (∘ /h)


|H(Ω)| (dB)

0
−20
−1
−30
−2
−40
−3
−50
101 102 103 104 −4
0 500 1000 1500 2000
f (Hz)
f (Hz)
MSTMM
Newton-Euler method Figure 9: Attitude error rate under sinusoidal vibration.

Figure 7: Amplitude frequency response.

−50
PSD (g 2 /Hz)

0.02
𝜑(Ω) (∘ )

−100 −3 dB/dec
3 dB/dec

−150

20 80 1300 2000
−200 f (Hz)
101 102 103 104
f (Hz) Figure 10: Input PSD.
MSTMM
Newton-Euler method

Figure 8: Phase frequency response. In this paper, the method for calculating the frequency
response function of LGSIMU system is established by using
MSTMM. Based on this method, the responses of LGSIMU
acceleration of platform along y-axis is shown in Figure 11 system are obtained and the dynamics performance is dis-
and its root mean square value is 1.55 g. Since the root cussed. Furthermore, the attitude errors under sinusoidal
mean square value is less than 2 g and the PSD curve comes and random vibrations are calculated and the effect of
down rapidly after 200 Hz, the vibration reduction index damping coefficient on the errors is analyzed. The proposed
is achieved. The isolators 8∼11 are usually identical. The method can also be used to analyze the velocity and position
mean value of the attitude error rate versus the damping errors caused by vibration. The simulation results verify the
coefficient of isolators 8∼11 is shown in Figure 12. The mean effectiveness of the proposed method. Compared with the
value of the attitude error rate decreases with the increasing ordinary method, the proposed method does not need the
of the damping coefficient. Consequently, within the index global dynamics equation. It is also highly stylized, flexible
of vibration reduction, the damping coefficient of isolators for modeling, and easy to program and provides a powerful
should be as large as possible. technique for studying the dynamics of LGSIMU.

10. Conclusion Conflict of Interests


Improving the precision of LGSIMU in a dynamic environ- The authors declare that there is no conflict of interests
ment is an important problem over a long period of time. regarding the publication of this paper.
Advances in Mechanical Engineering 9

100 of the IEEE Position Location and Navigation Symposium, pp.


32–39, March 1992.
[5] J. I. Lahham, D. J. Wigent, and A. L. Coleman, “Tuned
10 −2 support structure for structure-borne noise reduction of inertial
navigator with dithered ring laser gyros (RLG),” in Proceedings
of the IEEE 2000 Position, Location and Navigation Symposium,
PSD (g 2 /Hz)

pp. 419–428, March 2000.


10−4 [6] J. F. Li, J. J. LIU, and X. L. Tang, “Aseismatic control of system
structure for laser strapdown IMU,” Journal of Chinese Inertial
Technology, vol. 14, no. 1, pp. 21–26, 2006.
10−6 [7] B. Wu, H. F. Wang, T. Qu et al., “Design of mounting plate
for a missile-bone laser gyro SINS,” Journal of Chinese Inertial
Technology, vol. 15, no. 2, pp. 132–135, 2007.
10−8 [8] Y. Xie, Y. J. Lei, J. P. Zhou et al., “Optimum parameter of INS
101 102 103 104 system absorber,” Structure and Environment Engineering, vol.
f (Hz) 32, no. 4, pp. 39–45, 2005.
[9] P. Liu, H. P. Liu, and C. Zhang, “Research on dynamical
Figure 11: Output PSD of platform along y-axis. characteristics of laser gyro strapdown IMU,” Optics and Opto-
×10−3 electronic Technology, vol. 9, no. 5, pp. 70–74, 2011.
3.5
[10] X. T. Rui, L. F. Yun, Y. Q. Lu et al., Transfer Matrix Method for
Multibody System and Its Applications, Science Press, Beijing,
China, 2008.
3 [11] X. Rui, G. Wang, Y. Lu, and L. Yun, “Transfer matrix method for
Attitude error rate (∘ /h)

linear multibody system,” Multibody System Dynamics, vol. 19,


no. 3, pp. 179–207, 2008.
2.5 [12] X. Rui, B. Rong, G. Wang, and B. He, “Discrete time transfer
matrix method for dynamics analysis of complex weapon
systems,” Science China Technological Sciences, vol. 54, no. 5, pp.
1061–1071, 2011.
2
[13] X. Rui, B. He, Y. Lu, W. Lu, and G. Wang, “Discrete time transfer
matrix method for multibody system dynamics,” Multibody
System Dynamics, vol. 14, no. 4, pp. 317–344, 2005.
1.5 [14] B. Rong, X. Rui, G. Wang, and H. Yu, “Discrete time transfer
0 500 1000 1500 2000
matrix method for dynamic modeling of complex spacecraft
d (N·s/m)
with flexible appendages,” Journal of Computational and Non-
Figure 12: The mean values of attitude error rates. linear Dynamics, vol. 6, no. 1, Article ID 011013, 2011.
[15] G. P. Wang, B. Rong, L. Tao et al., “Riccati discrete time transfer
matrix method for dynamic modeling and simulation of an
underwater towed system,” Journal of Applied Mechanics, vol.
Acknowledgments 79, no. 1, Article ID 041014, 2012.
The research was supported by Research Fund for the [16] F. F. Yang, X. T. Rui, and L. Ma, “Dynamical model and
Doctoral Program of Higher Education of China (no. simulation of laser gyro strapdown inertial navigation system,”
20113219110025), Natural Science Foundation of China Gov- Journal of Chinese Inertial Technology, vol. 16, no. 3, pp. 301–305,
2008.
ernment (no. 11102089), and Program for New Century
Excellent Talents in University (NCET-10-0075). [17] L. Ma, X. Rui, F. Yang, and B. Rong, “Power spectrum analysis
for laser gyro damping system of strapdown inertial navigation,”
Journal of Vibration Engineering, vol. 22, no. 6, pp. 603–607,
References 2009.
[18] P. G. Savage, Strapdown Analytics, Strapdown Associates, Maple
[1] J. M. Oelschlaeger and L. O. Thielman, “GG1308 ring laser gyro Plain, Minn, USA, 2000.
inertial measurement systems: Honeywell’s low-cost solution
[19] X. T. Rui and J. S. Zhang, “Automatical transfer matrix method
for tactical applications,” in Proceedings of the Position Location
of multibody system,” in Proceedings of the 2nd Joint Inter-
and Navigation Symposium, pp. 528–536, March 1990.
national Conference on Multibody System Dynamics, Stuttgart,
[2] A. T. Daniel, G. John, and R. E. Mark, “Ring laser gyroscope Germany, 2012.
dither drive system and method,” United States Patent, No.
[20] W. T. Thomson and M. D. Dahleh, Theory of Vibration with
4981359, January 1, 1991.
Applications, Prentice Hall, Englewood Cliffs, NJ, USA, 5th
[3] R. D. Curby, G. H. McCammon, and S. Vallev, “Dither controller edition, 1993.
for ring laser angular rotation sensor,” United States Patent, No.
[21] H. F. Wang, B. Wu, M. Zhang et al., “Research on technique of
4740084, April 26, 1988.
aseismatic control for laser strapdown IMU,” Aerospace Control,
[4] J. I. Lahham and J. R. Brazell, “Acoustic noise reduction in the vol. 25, no. 6, pp. 81–85, 2007.
MK49 ship’s inertial navigation system (SINS),” in Proceedings
Hindawi Publishing Corporation
Advances in Mechanical Engineering
Volume 2013, Article ID 216014, 15 pages
http://dx.doi.org/10.1155/2013/216014

Research Article
Recursive Formulation for Dynamic Modeling and Simulation of
Multilink Spatial Flexible Robotic Manipulators

Zhenjie Qian, Dingguo Zhang, and Jun Liu


School of Sciences, Nanjing University of Science and Technology, Nanjing, Jiangsu 210094, China

Correspondence should be addressed to Dingguo Zhang; zhangdg419@mail.njust.edu.cn

Received 25 August 2013; Accepted 11 October 2013

Academic Editor: Xiaoting Rui

Copyright © 2013 Zhenjie Qian et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

The dynamics for spatial manipulator arms consisting of n flexible links and n flexible joints is presented. All the transversal,
longitudinal, and torsional deformation of flexible links are considered. Within the total longitudinal deformation, the nonlinear
coupling term, also known as the longitudinal shortening caused by transversal deformation, also is considered here. Each flexible
joint is modeled as a linearly elastic torsional spring, and the mass of joint is considered. Lagrange’s equations are adopted to
derive the governing equations of motion of the system. The algorithmic procedure is based on recursive formulation using 4 × 4
homogenous transformation matrices where all the kinematical expressions as well as the final equations of motion are suited for
computation. A corresponding general-purpose C++ software package for dynamic simulation is developed. Several examples are
simulated to illustrate the performance of the algorithm.

1. Introduction for structural geometric non-linear effects. Xi and Fenton


[4] studied a manipulator consisting of one flexible link and
Flexible multibody dynamics has become a key methodology one flexible joint based on the assumption that the link is
for various engineering fields, such as robotic manipula- constrained to move only in a horizontal plane, whereas the
tor arms, large radar antennas, solar panels, transportation gravity and some coupling terms between the equations of
vehicles, and manufacturing equipment as well as flexible motions have been dropped. It has been demonstrated by
ligament in human musculoskeletal system [1], and so forth. Wallrapp and Schwertassek [5], that the so-called “geometric
In particular, the flexible robotic manipulators are typi- stiffening” problem can be solved by keeping second order
cally precise and complex operated at high speed. They are terms in the expression of the deformations of the material,
designed to be light with low inertia in order to achieve cost- and the amplitude of the flexible motions must remain small
reduction, energy-saving, and high-performance. Therefore, in those formalisms which are generally based on kinematic
the dynamic analysis of flexible manipulator arms is compli- restrictions as regards the flexibility.
cated due to the coupling between the large rigid body motion Low [6] presented vibration analysis of a rotating beam
and deformation. carrying a tip mass at its end by using Hamilton’s principle
A large number of the literature related to the so-called and the associated boundary conditions. Yoo et al. [7] used a
dynamic stiffening effects have been published, which is first non-Cartesian variable along with two Cartesian variables to
proposed by Kane et al. [2] due to link’s high-order coupling describe the elastic deformation and investigate the dynamic
flexibility. It was observed that an industrial link under high- stiffness effect. Ryu et al. [8] put forward a criterion on inclu-
speed rotational motion would exhibit instability problems sion of stiffening effects such that it clarifies the limit of the
when its angular velocity exceeds a certain limit. The incor- validity of the linear modeling method. El-Absy and Shabana
rect simulation solutions are attributed to neglecting the [9] introduced the effect of longitudinal deformation due to
high-order coupling deformation terms in the dynamic equa- bending and studied the influence of geometric stiffness on
tions. Wu and Haug [3] modeled a flexible multibody system instability problem of nonlinear elastic model. Al-Bedoor and
by means of substructure synthesis formulation to account Hamdan [10], based on the condition of inextensibility to
2 Advances in Mechanical Engineering

relate the axial and transverse deformations of the material


point, studied a rotating flexible arm deformation undergoing ith link
large planar motion. Hong et al. [11, 12] established higher- Y
ith joint
order rigid-flexible coupling dynamic model of rotating
flexible beam undergoing large overall motion. When the
deformation of flexible beam or the deformation rate is larger,
the first-order or zero-order approximation coupling model ith link deformation
will appear with divergent results, nevertheless the com-
plete high-order coupling model’s results are still converged.
Schwertassek et al. [13] presented the fundamental shape
functions choice in floating frame of reference formulation,
by separating the flexible body motion into a reference
motion and deformation. X
Book [14] used the 4 × 4 homogenous transformation O
matrices and assumed modes to describe the kinematics of Z
the rigid-joint and flexible-link robots, whereas the dynamic
Figure 1: Flexible robotic arms.
model cannot deal with the torsional deformation of links. A
recursive formulation for the spatial kinematic and dynamic
analysis of open chain mechanical systems containing inter- as well as the gravitational potential energy of system, and
connected deformable bodies is given by Changizi and then focuses on the derivation of the recursive rigid-flexible
Shabana [15], and Kim and Haug [16]. Jain and Rodriguez [17] coupling dynamic equations of the system. In the modeling,
developed new spatially recursive dynamics algorithms for the high-order coupling terms related to the non-linear
flexible multibody systems by using spatial operators, which geometry deformation are retained and the recursive strategy
is based on Newton-Euler factorization and innovations fac- for kinematics is adopted. In Section 4, several examples
torization of the system mass matrix. Hwang [18] developed a are simulated to illustrate the performance of the algorithm
recursive formulation for the flexible dynamic manufacturing proposed in the paper. Finally, Section 5 summarizes the
analysis of open-loop robotic systems with the generalized results and draws conclusions from them.
Newton-Euler equations. Zhang and Zhou [19, 20] did a
further work based on Book’s work in [14]. Both the bending 2. Kinematics of Flexible Robots
and torsional flexibility of links were taken into account.
Dynamic simulation of a spatial flexible manipulator arm was As a point of departure, the system considered here is an
given as an example to validate the algorithm. However, the assembly of n flexible links connected by n rotary joints, as
dynamic stiffening effect was not considered yet. shown in Figure 1.
When the system’s operating speed becomes high,
neglecting the flexibility of joints is quite devastating, which 2.1. Simplified Model of Flexible Joint. Figure 2 shows the
will usually give rise to the errors of position precision. A typi- flexible joint model. We can model the flexibility of joint 𝑖 as
cal dynamic model for taking into account the joint flexibility a linear torsional spring with stiffness 𝐾𝑡𝑖 ⋅ 𝐽𝑟𝑖 is the moment
was presented by Spong [21]. In his work, the joint flexibility of inertia of rotor 𝑖 about its spinning axis. And 𝜏𝑖 is the
is modeled as a torsional spring with more emphasis on torque exerted at joint 𝑖. For simplicity, we neglect friction
simplifying the equations of motion for control purposes. or damping in the flexible joint. Let 𝑞1𝑖 be the theoretical
Other dynamic analyses of robots with joint flexibilities can rotational angle of link 𝑖, 𝑞2𝑖 be the real rotational angle of
be seen in the work of Wasfy and Noor [22], Dwivedy and link 𝑖, 𝜀𝑖 be the torsional angle of joint 𝑖, 𝜑𝑖 be the angular
Eberhard [23], and Na and Kim [24]. displacement of rotor 𝑖, and 𝜙𝑖 be the gear ratio, respectively.
As mentioned above, there are a lot of research work The relationships among them are as follows:
in the dynamic modeling and simulation. But it is still very
difficult for us to deal with the dynamics of the complex 𝜀𝑖 = 𝑞2𝑖 − 𝑞1𝑖 , (1)
multibody systems, such as the spatial flexible-link and
flexible-joint robotic manipulators with consideration of the 𝜑𝑖 = 𝜙𝑖 𝑞1𝑖 . (2)
axial, bending, and torsional deformation for links, and
the flexibility and mass effects for joints, and the so-called 2.2. Simplified Model of Flexible Link. Assume that the links
“dynamic stiffening” effects. In this paper, we will present the are slender beams. Analysis here is based on the Euler-
dynamic modeling methodology to include all such terms. Bernoulli beam theory in the elastic small displacements
In the following section, the kinematics of the system are field.
presented, in which coordinate frames are established, and
4 × 4 homogeneous transformation matrices are used to 2.3. Coordinate Systems and Transformation Matrices. To
describe the kinematics of flexible links and flexible joints. express the transformation between different coordinate
The approach of assumed modes is employed to describe the systems clearly, we establish four coordinate systems for link
deformation of the flexible links. Section 3 firstly deals with 𝑖. Fix the coordinate system (𝑋𝑏 𝑌𝑏 𝑍𝑏 )𝑖 at the proximal end
the description of the kinetic and elastic potential energy of link 𝑖 (oriented so that the 𝑋𝑏 coincides with the neutral
Advances in Mechanical Engineering 3

Here, 𝑥𝑖𝑗 , 𝑦𝑖𝑗 , and 𝑧𝑖𝑗 are the 𝑥𝑏𝑖 , 𝑦𝑏𝑖 , and 𝑧𝑏𝑖 components of
the elastic linear displacement mode 𝑗 of link 𝑖 at the origin of
the coordinate (𝑋𝑑 𝑌𝑑 𝑍𝑑 )𝑖 , respectively. 𝜃𝑥𝑖𝑗 , 𝜃𝑦𝑖𝑗 , and 𝜃𝑧𝑖𝑗 are
ith link the 𝑥𝑏𝑖 , 𝑦𝑏𝑖 , and 𝑧𝑏𝑖 rotation components of the elastic angular
Kti displacement mode 𝑗 of link 𝑖 at the origin of the coordinate
(𝑋𝑑 𝑌𝑑 𝑍𝑑 )𝑖 , respectively. 𝛿𝑖𝑗 is the time-varying amplitude
q2i of mode 𝑗 of link 𝑖, and 𝑁𝑖 is the number of modes used to
Jri 𝜙i2 describe the deformation of link 𝑖.
Tran
smi
ssio q1i
Define 0 W𝑖 or W𝑖 to be the 4 × 4 homogeneous transfor-
n mation matrix from the base coordinate frame (𝑋𝑏 𝑌𝑏 𝑍𝑏 )0
𝜏i Roto to (𝑋𝑏 𝑌𝑏 𝑍𝑏 )𝑖 . Then, we have
r
(i − 0 ̂ 𝑖−1 A𝑖 ,
1)th
lin k
W𝑖 = W𝑖 = W𝑖−1 E𝑖−1 Ai = W (7)

̂ 𝑖−1 is the 4 × 4 homogeneous transformation matrix


where W
from the base coordinate frame (𝑋𝑏 𝑌𝑏 𝑍𝑏 )0 to the distal
coordinate system (𝑋𝑑 𝑌𝑑 𝑍𝑑 )𝑖−1 of link 𝑖 − 1.

Figure 2: Flexible joint model. 2.4. Velocity of a Point of Link 𝑖. Let 𝑖 h𝑖 (𝜂) be the homo-
geneous coordinates in the system (𝑋𝑏 𝑌𝑏 𝑍𝑏 )𝑖 of a point of
axis of link 𝑖 in undeformed shape). This will be referred the deformed link 𝑖 at position 𝜂 with the link under an
to as the base reference frame of link 𝑖. Fix the coordinate undeformed condition from the origin of (𝑋𝑏 𝑌𝑏 𝑍𝑏 )𝑖 . Then,
system (𝑋𝑑 𝑌𝑑 𝑍𝑑 )𝑖 at the distal end of link 𝑖. This is the 𝑖
h𝑖 (𝜂) can be approximated as
distal frame of link 𝑖. When link 𝑖 is in its undeformed
state, the distal frame can be located by a pure translation 𝑁𝑖
𝑖 T T
of the base reference frame (𝑋𝑏 𝑌𝑏 𝑍𝑏 )𝑖 along the length 𝐿 𝑖 h𝑖 (𝜂) = [1 𝜂 0 0] + ∑ 𝛿𝑖𝑗 [0 𝑥𝑖𝑗 (𝜂) 𝑦𝑖𝑗 (𝜂) 𝑧𝑖𝑗 (𝜂)]
of link 𝑖. Let (𝐻𝑥 𝐻𝑦 𝐻𝑧 )󸀠𝑖 and (𝐻𝑥 𝐻𝑦 𝐻𝑧 )𝑖 be two Denavit- 𝑗=1

Hartenberg (D-H) frames fixed at the proximal end (at joint 𝑁 𝑁


𝑖) and the distal end (at joint 𝑖 + 1) of link 𝑖, respectively. 1 𝑖 𝑖 T
− ∑ ∑ 𝛿𝑖𝑗 𝛿𝑖𝑘 [ 0 𝑥𝑖𝑗𝑘 (𝜂) 0 0 ] .
When joint 𝑖 is motionless, (𝐻𝑥 𝐻𝑦 𝐻𝑧 )󸀠𝑖 is coincident with 2 𝑗=1 𝑘=1
(𝐻𝑥 𝐻𝑦 𝐻𝑧 )𝑖−1 , and matrix HH𝑖−1 𝑖 , that is, the transformation
(8)
matrix between them, is the function of 𝑞2𝑖 . Matrix H𝑖 ,
Here, 𝑥𝑖𝑗𝑘 is the nonlinear strain coupling term, also known
the 4 × 4 homogeneous transformation matrix between
frames (𝑋𝑑 𝑌𝑑 𝑍𝑑 )𝑖 and (𝐻𝑥 𝐻𝑦 𝐻𝑧 )𝑖 , is a constant matrix. The as the axial shortening due to the bending deformations of
the link. When the flexible links are undergoing a high speed,
transformation matrix Hb𝑖 of (𝐻𝑥 𝐻𝑦 𝐻𝑧 )󸀠𝑖 and (𝑋𝑏 𝑌𝑏 𝑍𝑏 )𝑖 this term will bring the so-called dynamic stiffening effect
is also a constant matrix. Define the joint-transformation which will have a great influence on the dynamic behavior of
matrix A𝑖 of joint 𝑖 to be the transformation matrix from flexible arms. In the dynamic modeling presented here, the
(𝑋𝑑 𝑌𝑑 𝑍𝑑 )𝑖−1 to (𝑋𝑏 𝑌𝑏 𝑍𝑏 )𝑖 . Then, high-order terms related to the non-linear coupling term are
A𝑖 = dH𝑖−1 HH𝑖−1 retained, which are ignored in the first-order and zero-order
𝑖 Hb𝑖 . (3)
approximation coupling modeling.
Obviously, A𝑖 is a function of 𝑞2𝑖 . Define E𝑖 to be the In terms of the fixed inertial coordinates of the base
link-transformation matrix of link 𝑖 from (𝑋𝑏 𝑌𝑏 𝑍𝑏 )𝑖 to (𝑋𝑏 𝑌𝑏 𝑍𝑏 )0 , the position 0 h𝑖 or h𝑖 of the point is given as
(𝑋𝑑 𝑌𝑑 𝑍𝑑 )𝑖 . According to the assumption of small deforma-
tion of the links, the small deformable angles can be added h𝑖 = W 𝑖 𝑖 h𝑖 . (9)
vectorially. E𝑖 can be written as
𝑁𝑖 Taking the time derivative of the position h𝑖 , we have the
E𝑖 = H𝑖 + ∑𝛿𝑖𝑗 M𝑖𝑗 , (4) velocity of the point as
𝑗=1
𝑑h𝑖 𝑖
in which = ḣ ̇𝑖 ̇
𝑖 = W𝑖 h𝑖 +W𝑖 h𝑖 . (10)
1 0 0 0 𝑑𝑡
[𝐿 𝑖 1 0 0 ]
H𝑖 = [ ]
[ 0 0 1 0] , (5) To accelerate the computation of the matrices Ẇ𝑖 or Ẅ𝑖 , we
use the recursive kinematics method. By differentiating (7),
[ 0 0 0 1] one obtains
0 0 0 0
[𝑥𝑖𝑗 1 −𝜃𝑧𝑖𝑗 𝜃𝑦𝑖𝑗 ] ̂̇𝑖−1 A𝑖 + W
Ẇ𝑖 = W ̂ 𝑖−1 Ȧ𝑖 ,
M𝑖𝑗 = [
[𝑦𝑖𝑗 𝜃𝑧𝑖𝑗
]. (6)
0 −𝜃𝑥𝑖𝑗 ] (11)
[ 𝑧𝑖𝑗 −𝜃𝑦𝑖𝑗 𝜃𝑥𝑖𝑗 0 ] Ẅ ̂̈ ̂̇ ̇ ̂ ̈
𝑖 = W𝑖−1 A𝑖 + 2W𝑖−1 A𝑖 + W𝑖−1 A𝑖 ,
4 Advances in Mechanical Engineering

where kinetic energy accounting for the torsional deformation. Note


that the torsional angle of link 𝑖, 𝜃𝑥𝑖 , is expressed as
Ȧ𝑖 = U𝑖 𝑞2𝑖
̇,
𝑁𝑖
(12)
Ä 2
̇+ U𝑖 𝑞2𝑖̈ 𝜃𝑥𝑖 = ∑𝛿𝑖𝑗 𝜃𝑥𝑖𝑗 , (17)
𝑖 = U2𝑖 𝑞2𝑖 .
𝑗=1

Here, U𝑖 ≜ 𝜕A𝑖 /𝜕𝑞2𝑖 , U2𝑖 ≜ 𝜕2 A𝑖 /𝜕𝑞2𝑖


2
, and 𝑞2𝑖 is the joint vari- where 𝛿𝑖𝑗 and 𝜃𝑥𝑖𝑗 are mentioned in (4) and (6), respectively.
̇ ̈
able of joint 𝑖. Thus, W𝑖 and W𝑖 can be computed recursively 𝛿𝑖𝑗 is the function of time, whereas 𝜃𝑥𝑖𝑗 is the function of
from Ŵ 𝑖−1 and its derivatives. Here, one additionally needs position 𝜂.
̂ Substituting (10) and (17) into (16), expanding it, and then
W𝑖−1 and its derivatives. These can be computed recursively
summing over all 𝑛 links, one finds the links’ kinetic energy
from W𝑖−1 and its derivatives as follows:
to be
𝑛 𝑛
̂ 𝑖 = W𝑖 E𝑖 ,
W 𝐾𝑏 = ∑𝐾𝑖𝑏 = ∑ Tr {Ẇ𝑖 B3𝑖 WT𝑖 + 2Ẇ𝑖 B2𝑖 WT𝑖 + W𝑖 B1𝑖 WT𝑖 }
𝑖=1 𝑖=1
̂̇𝑖 = Ẇ𝑖 E𝑖 + W𝑖 Ė
W 𝑖, (13)
𝑛 𝑁𝑖 𝑁𝑖

̂̈ ̈ ̇ ̇ ̈ + ∑ ∑ ∑ 𝛿𝑖𝑗̇𝛿𝑖𝑘
̇𝑇 ,
𝑖𝑗𝑘
W𝑖 = W𝑖 E𝑖 + 2W𝑖 E𝑖 + W𝑖 E𝑖 , 𝑖=1 𝑗=1 𝑘=1
(18)
where
where
𝑁𝑖
Ė ̇ 1 𝐿𝑖
𝑖 = ∑ 𝛿𝑖𝑗 M𝑖𝑗 , 𝑇𝑖𝑗𝑘 = ∫ 𝐽 𝜃 𝜃 𝑑𝜂,
𝑗=1 2 0 𝑥𝑖 𝑥𝑖𝑗 𝑥𝑖𝑘
(14) 1 𝐿𝑖 𝑖 ̇𝑖 ̇
𝑁𝑖 B1𝑖 = ∫ 𝜇 h hT 𝑑𝜂,
Ë ̈
𝑖 = ∑ 𝛿𝑖𝑗 M𝑖𝑗 .
2 0 𝑖 𝑖 𝑖
(19)
𝑗=1 1 𝐿𝑖 𝑖 𝑖 ̇
B2𝑖 = ∫ 𝜇 h hT 𝑑𝜂,
2 0 𝑖 𝑖 𝑖
3. Dynamics of Flexible Robots 1 𝐿𝑖 𝑖 𝑖 T
B3𝑖 = ∫ 𝜇 h h 𝑑𝜂.
To use Lagrange’s equations, we need the kinetic and potential 2 0 𝑖 𝑖 𝑖
energy of the system. With the consideration of 𝑖 h𝑖 expressed in (8) and its
derivative, consider
3.1. The System Kinetic Energy. The system kinetic energy 𝐾 𝑁𝑖
contains two parts: the links kinetic energy 𝐾𝑏 and the joints 𝑖
ḣ ̇
𝑖 (𝜂) = ∑ 𝛿𝑖𝑗 [0 𝑥𝑖𝑗 (𝜂) 𝑦𝑖𝑗 (𝜂) 𝑧𝑖𝑗 (𝜂)]
T

kinetic energy 𝐾𝑟 as follows: 𝑗=1


(20)
𝑁𝑖 𝑁𝑖
𝐾 = 𝐾𝑏 + 𝐾 𝑟 . (15) − ∑ ∑ 𝛿𝑖𝑗̇𝛿𝑖𝑘 [0 𝑥𝑖𝑗𝑘 (𝜂) 0 0] ,
T

𝑗=1𝑘=1
Assume that the links are slender beams, so the rotary
inertia and shear effects can be neglected. Therefore, the the matrices B1𝑖 , B2𝑖 , and B3𝑖 can be written as
present analysis is based on the Euler-Bernoulli beam theory. 𝑁𝑖 𝑁𝑖
Also assume that the links can undergo a large overall rigid B1𝑖 = ∑ ∑ 𝛿𝑖𝑗̇𝛿𝑖𝑘
̇D
𝑖𝑗𝑘
motion, however the elastic displacements are small. The 𝑗=1 𝑘=1
kinetic energy of the 𝑖th link is 𝑁𝑖 𝑁𝑖 𝑁𝑖
+ ∑ ∑ ∑𝛿𝑖𝑗̇𝛿𝑖𝑘
̇𝛿 (−F − FT )
𝑖𝑙 𝑖𝑗𝑘𝑙
1 𝐿𝑖 𝑖𝑗𝑘𝑙
𝐾𝑖𝑏 = ∫ 𝜇 (𝜂) Tr {ḣ ̇
T
𝑖 h𝑖 } 𝑑𝜂
𝑗=1 𝑘=1 𝑙=1
2 0
(16) 𝑁𝑖 𝑁𝑖 𝑁𝑖 𝑁𝑖
1 𝐿𝑖 𝜕𝜃 (𝜂, 𝑡)
2
+ ∑ ∑ ∑ ∑𝛿𝑖𝑗̇𝛿𝑖𝑘 𝛿𝑖𝑙̇𝛿𝑖𝑠 E𝑖𝑗𝑘𝑙𝑠 ,
+ ∫ 𝐽𝑥𝑖 (𝜂) ( 𝑥𝑖 ) 𝑑𝜂, 𝑗=1 𝑘=1 𝑙=1 𝑠=1
2 0 𝜕𝑡
(21)
where Tr{⋅} is the trace operator; 𝜇(𝜂) and 𝐽𝑥𝑖 are the mass per
𝑁𝑖 𝑁𝑖 𝑁𝑖
unit length and the polar moment of inertia per unit length
of the link about the neutral axis 𝑥, respectively. For slender B2𝑖 = ∑ 𝛿𝑖𝑗̇D𝑖𝑗 + ∑ ∑ 𝛿𝑖𝑗̇𝛿𝑖𝑘 DT𝑖𝑗𝑘
𝑗=1 𝑗=1 𝑘=1
beams with uniform cross section area along 𝑥 axis, 𝜇(𝜂) = 𝜇𝑖 .
The first term in (16) is the kinetic energy of link 𝑖 accounting 𝑁𝑖 𝑁𝑖
for the rigid-body motion and the lateral and longitudinal + ∑ ∑ 𝛿𝑖𝑗̇𝛿𝑖𝑘 (−E𝑖𝑗𝑘 )
deformation due to flexibility, whereas the second term is the 𝑗=1 𝑘=1
Advances in Mechanical Engineering 5

𝑁𝑖 𝑁𝑖 𝑁𝑖
1 Case 2. Link 𝑖 consists of a flexible beam with the con-
+ ∑ ∑ ∑𝛿𝑖𝑗̇𝛿𝑖𝑘 𝛿𝑖𝑙 (− FT𝑖𝑗𝑘𝑙 − F𝑖𝑙𝑗𝑘 ) centrated mass 𝑚𝑖 at its proximal end. To account for the
𝑗=1 𝑘=1 𝑙=1 2
contribution of the concentrated mass to the kinetic energy
𝑁𝑖 𝑁𝑖 𝑁𝑖 𝑁𝑖 of link 𝑖, the extra term D󸀠𝑖 should be added to the matrix D𝑖 ,
1
+ ∑ ∑ ∑ ∑𝛿𝑖𝑗̇𝛿𝑖𝑘 𝛿𝑖𝑙 𝛿𝑖𝑠 ( ET𝑖𝑗𝑙𝑘𝑠 ), where
𝑗=1 𝑘=1 𝑙=1 𝑠=1 2
1
D󸀠𝑖 = 𝑚𝑖 [1 0 0 0] [1 0 0 0] .
T
(26)
(22) 2
𝑁𝑖 𝑁𝑖 𝑁𝑖
B3𝑖 = D𝑖 + ∑𝛿𝑖𝑗 (D𝑖𝑗 + DT𝑖𝑗 ) + ∑ ∑ 𝛿𝑖𝑗 𝛿𝑖𝑘 D𝑖𝑗𝑘 Should the concentrated mass 𝑚𝑖 locate at the position
𝑗=1 𝑗=1 𝑘=1
(𝑥0 , 𝑦0 , 𝑧0 ) near the proximal end, the extra matrix D󸀠𝑖 is
modified to be
𝑁𝑖 𝑁𝑖
1 1
+ ∑ ∑ 𝛿𝑖𝑗 𝛿𝑖𝑘 (− E𝑖𝑗𝑘 − ET𝑖𝑗𝑘 ) 1
D󸀠𝑖 = 𝑚𝑖 [1 𝑥0 𝑦0 𝑧0 ] [1 𝑥0 𝑦0 𝑧0 ] .
T
2 2 (27)
𝑗=1 𝑘=1 2
𝑁𝑖 𝑁𝑖 𝑁𝑖
(23)
1 1 Case 3. Link 𝑖 consists of a flexible beam with the concen-
+ ∑ ∑ ∑𝛿𝑖𝑗 𝛿𝑖𝑘 𝛿𝑖𝑙 (− F𝑖𝑗𝑘