See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/349285179

Optimum design of composite structures with ply drop-offs using response

surface methodology

Article in Engineering Computations · January 2021

DOI: 10.1108/EC-07-2020-0354

CITATIONS READS

5 607

5 authors, including:

Camila Diniz Fabricio Alves de Almeida

Universidade Federal de Itajubá (UNIFEI) Universidade Federal de Itajubá (UNIFEI)
13 PUBLICATIONS 192 CITATIONS 57 PUBLICATIONS 586 CITATIONS

SEE PROFILE SEE PROFILE

Sebastião S. Cunha Jr Guilherme Ferreira Gomes

Universidade Federal de Itajubá (UNIFEI) Universidade Federal de Itajubá (UNIFEI)
81 PUBLICATIONS 1,761 CITATIONS 163 PUBLICATIONS 2,385 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Guilherme Ferreira Gomes on 10 August 2021.

The user has requested enhancement of the downloaded file.

 The current issue and full text archive of this journal is available on Emerald Insight at:
https://www.emerald.com/insight/0264-4401.htm

Optimum
Optimum design of composite design of
structures with ply drop-offs using composite
structures
response surface methodology
Camila Aparecida Diniz and Yohan Méndez
Mechanical Engineering Institute, Federal University of Itajub
a, Itajuba, Brazil
Received 2 July 2020
Fabrício Alves de Almeida Revised 2 October 2020
17 December 2020
Institute of Industrial Engineering and Management, 18 December 2020
Federal University of Itajub
a, Itajuba, Brazil, and Accepted 19 December 2020

Sebastião Simões da Cunha Jr and G.F. Gomes

Mechanical Engineering Institute, Federal University of Itajub
a, Itajuba, Brazil

Abstract
Purpose – Many studies only take into account the ply stacking sequence as the design variable to
determine the optimal ply drop-off location; however, it is necessary to optimize other parameters that have a
direct influence on the ply drop-off site such as which plies should be dropped and in which longitudinal
direction. That way, the purpose of this study is to find the most significant design variables relative to the
drop-off location considering the transversal and longitudinal positions, seeking to achieve the optimal
combination of ply drop-off locations that provides excellent performance for the laminate plate.
Design/methodology/approach – This study aims to determine the optimal drop-off location in a
laminate plate using the finite element method and an approach statistical with design of experiments (DOE).
Findings – The optimization strategy using DOE revealed to be satisfactory for analyzing laminate structures with
ply drop-offs, demonstrating that not all design factors influence the response variability. The failure criterion
response variable revealed a poor fit, with an adjusted coefficient of determination lower than 60%, thus
demonstrating that the response did not vary with the ply drop-off location. Already the strain and natural frequency
response variables presented high significance. Finally, the optimization strategy revealed that the optimal drop-off
location that minimizes the strain and maximizes the natural frequency is the ply drop-off located of the end plate.
Originality/value – It was also noted that many researchers prefer evolutionary algorithms for optimizing
composite structures with ply drop-offs, being scarce to the literature studies involving optimization
strategies using response surface methodology. In addition, many studies only take into account the ply
stacking sequence as the design variable to determine the optimal ply drop-off location; however, in this
study, the authors investigated other important parameters that have direct influence on the ply drop-off site
such as which plies should be dropped and in which longitudinal direction.
Keywords Optimization, Failure, Composites, Response surface methodology, Ply drop-off
Paper type Research paper

Nomenclature
F = Strength parameters;
s1 = Longitudinal normal stress;

The authors would like to acknowledge the financial support from the Brazilian agency CNPq (Conselho
Nacional de Desenvolvimento Científico e Tecnol
ogico), CAPES (Coordenação de Aperfeiçoamento de Engineering Computations
Pessoal de Nível Superior), FAPEMIG (Fundação de Amparo à Pesquisa do Estado de Minas Gerais – © Emerald Publishing Limited
0264-4401
APQ-00385–18) and Tutorial Education Program (PET – Programa de Educação Tutorial). DOI 10.1108/EC-07-2020-0354
EC s2 = Transverse normal stress;
s 1T = Longitudinal tensile strength;
s 1C = Longitudinal compression strength;
s 2T = Transverse tensile strength;
s 2C = Transverse compression strength;
t 12 = Shear stress in the 1-2 plane;
t 12U = Shear strength;
r = Density;
E1 = Elasticity modulus in the longitudinal direction;
E2 = Elasticity modulus in the transverse direction;
G12 = Shear modulus in the 1-2 plane;
y 12 = Poisson ratio;
x1 = Design factor 1;
x2 = Design factor 2;
y = Response;
f = Function;
e = Random error term or noise;
b = Undetermined regression coefficient;
k = Number of design parameters;
RSM = Response surface methodology;
FEM = Finite element method;
DOE = Design of experiments;
SFO = Sunflower optimization;
CFRP = Carbon fiber reinforced polymer;
D = Composite desirability function;
MO = Multi-objective optimization;
d = Individual desirability function;
n = Number of responses;
Lp = Longitudinal position;
Pp = Ply position;
TW = Tsai–Wu failure criterion;
ANOVA = Analysis of variance;
R2adj = Adjusted coefficient of determination;
e = Strain; and
x = First natural frequency.

1. Introduction
Nowadays, many engineering applications have been developed considering laminate
composite structures such as wind turbine blades, ships and aircraft fuselages and wings
(Gomes et al., 2017). This kind of material has attracted much attention because of their
outstanding properties that are often influenced by important parameters such as the
number and thickness of plies and the stacking sequence, which are greatly responsible for
defining the structural behavior of the laminates.
To achieve the optimal design for laminate structures aiming at the reduction of the mass
and manufacturing cost, in the mid-1980s, the aircraft industry adopted the use of ply drop-
offs along the length of the structure, yielding a reduction in the laminate thickness. These
thickness variations are achieved by dropping-off the plies at different locations, creating
resin pockets (Barbero, 2014; Irisarri et al., 2014), as shown in Figure 1.
 Meanwhile, thickness variations because of ply drop-offs generally lead to structural Optimum
problems, given that high stress concentrations in the region near the ply drop-offs may design of
cause a reduction in the strength and even failure (Dhurvey and Mittal, 2013). To tackle this,
studies began to be developed to analyze the behavior of the ply drop-offs on the structures
composite
(Adams et al., 1984; Kemp and Johnson, 1985; Fish and Lee, 1989). Shim (2002) noted that the structures
main difficulties involving structures with ply drop-offs are a large number of possibilities
for ply drop-offs locations for a single structure and the influence that the ply drop-off site
has on the strength of the laminate composite.
Hence, optimization strategies aiming at optimal ply drop-off designs for laminates
structures considering mechanical and manufacturing issues have been proposed in many
studies. Irisarri et al. (2014) elaborated a simultaneous optimization of the fiber angle
distribution and, ply drop-off location and order. The authors used topology optimization,
gradient-based methods and evolutionary algorithms in this approach, which must be an
expensive process of high computational cost. Sudhagar et al. (2017) developed a structural
optimization for laminate composite plates with ply drop-off using the finite element method
(FEM) and the genetic algorithm (GA). The main aim was to maximize the first natural
frequency and damping factors to obtain an optimal ply sequence and orientation. Albanesi
et al. (2018) proposed a novel methodology using the GA with the inverse FEM to determine
the optimal ply drop-off configuration of laminate wind turbine blades. The design variables
were a set of ply layouts, considering as constraints the deflection, maximum stress, natural
frequencies and number of plies. Shrivastava et al. (2019) used a GA and the finite element
model for an optimization study of wing panels with ply drop-off. A multi-objective
optimization using the Tsai–Wu failure index and lateral deflection was considered to
achieve the optimal ply drop-off design that offers minimum weight. Many studies only take
into account the ply stacking sequence as the design variable to determine the optimal ply
drop-off location; however, it is necessary to optimize other parameters that have a direct
influence on the ply drop-off site such as which plies should be dropped and in which
longitudinal direction.
It was also noted that many researchers prefer evolutionary algorithms for optimizing
composite structures with ply drop-offs (Peeters and Abdalla, 2016; Muc, 2018; Vemuluri
et al., 2018; Zhao and Kapania, 2018; Sanz-Corretge, 2019). Nasab et al. (2018) state that the
discrete nature of GA does not allow for continuous variables; however, the ply drop-off
location may be a continuous variable, as considered in this study. Moreover, Sanz-Corretge
(2019) emphasizes that evolutionary algorithms such as GA must be run many times to
generate consolidated solutions, which entails a high computational cost. For this reason,
An et al. (2019) suggest the use of approximation concepts such as the response surface
methodology (RSM) to reduce the computational cost. The RSM is considered a somewhat
simple methodology that does not demand many experimental runs (Singh and Rao, 2007;
Skrypnyk et al., 2019).

Figure 1.
Typical laminate
with two ply drop-
offs
EC This study aims to determine the suitable location for ply drop-offs considering two design
variables relative to the longitudinal (panel) and transversal (plies) planes of a laminate
plate, aiming to minimize the strain and Tsai–Wu failure criterion and maximize the first
natural frequency. The meta-model created by the optimization strategy is based on a
statistical approach using a design of experiments (DOE) with full factorial design and the
RSM. To the best of our knowledge, there are considerably scarce studies in the literature
applying the RSM to optimize laminate composites with ply drop-offs.
The paper is arranged as follows. In Section 2, an approach explaining the design
guidelines for structures with ply drop-offs, the Tsai–Wu failure criterion and the statistical
method is considered. In Section 3, the methodology proposed in this paper using the FEM
and a statistical approach is discussed. In Section 4, the results are demonstrated. Finally, in
Section 5, the conclusions are presented.

2. Theoretical background
2.1 Main design guidelines for laminate composites with ply drop-offs
A typical construction of a composite structure with ply drop-off often suffers from a
variation in the thickness obtained by dropping-off the plies, starting with all plies (thick
panel) and finishing with those that were not dropped (thin panel), thereby creating a taper
angle at the taper section (Irisarrri et al., 2016), as shown in Figure 2.
One may observe in Figure 2 that the laminate is longitudinally fragmented into panels
or zones (x-axis) and transversely comprised of a ply stacking sequence (z-axis). When a ply
is dropped off, theoretically, voids are generated filled with resin, and, for this reason, the
region near the ply drop-off is known as resin pockets (Vidyashankar and Murty, 2001).
Dropping off plies on a structure is not trivial; indeed, it may be considered quite complex
because the discontinuity between plies may impose difficulties relative to the stress
concentration in the vicinity of the ply drop-off, as well as possible premature failures such
as delamination, matrix cracking and fiber fractures (Mukherjee and Varughese, 2001;
Weiss et al., 2010). To avoid these unwanted situations, there are guidelines considered the
basis for the creation of composite structures with ply drop-offs, elaborated based on
industrial knowledge (Irisarri et al., 2014):

The stacking sequence should be symmetric around the mid-plane, that is, the same
orientation upward relative to the plane must be repeated downward.

For symmetric laminates, no more than two plies with the same orientation are
allowed to be stacked together.

Figure 2.
Nomenclature used in
structures with ply
drop-offs

The stacking sequence should be balanced, that is, the same number of þu and –u Optimum
plies. design of

It is required that at least 10% of plies are oriented at 0°, 645° and 90° angles. composite

Lower and upper surface plies of the laminate should not be oriented at 0°. structures

The covering plies localized on the lower and upper surface should not be dropped.

It is not allowed that more than two plies stopped at the same increment in
thickness, that is, in the same panel.

The taper angle should not be over 7°.

Dropped plies between two vicinity panels with ply orientation disagreement are
not allowed.

All these guidelines were considered in this work, thus avoiding unwanted failure modes,
shear-extension and membrane bending coupled behavior. Other guidelines may be
consulted in the study by Irisarri et al. (2014).

2.2 Tsai–Wu failure criterion for composite materials

Failure criteria have been elaborated to detect failures through the estimation of effects
caused by efforts applied to a structure. Because of composite material anisotropy, failure
analysis may become quite complex because of the presence of some specific failure
mechanisms. The Tsai–Wu failure criterion is the most suitable because of its
approximation to experimental data (Tsai and Wu, 1971). According to this criterion, a
failure occurs in the structure if there is a disagreement in equation (1) (Voyiadjis and
Kattan, 2005), where F11, F22, F66, F1, F2 and F12 denote the strength parameters, s 1 and s 2
are the normal stresses and t 12 is the shear stress.

F11 s 21 þ F22 s 22 þ F66 t 212 þ F1 s 1 þ F2 s 2 þ F12 s 1 s 2 # 1 (1)

Most researchers have used failure criteria in their analyses when investigating laminate
composite structures seeking to diagnose the damage before it happens, thus avoiding a
possible collapse of the structures (Sjølund and Lund, 2018; Lund, 2018; Adluru et al., 2019;
Ghayoor et al., 2019; Diniz et al., 2019).

2.3 Design of experiments

DOE is a traditional statistical method widely used in many analyses. It refers to an
experiment planning process in which a suitable data set is collected and examined for
providing meaningful conclusions (Montgomery, 2017). This method comprises several
techniques, such as factorial designs, the RSM and the Taguchi method, with the choice of
strategy depending on the process characteristics and the responses that will be analyzed. In
this study, both a full factorial design and the RSM were chosen for analyzing the laminate
plate with ply drop-offs. The full factorial design was used to determine the design and most
significant response variables, while the RSM was used in the optimization strategy. Some
studies have used statistical methods for analyzing composite structures (Naresh et al., 2018;
Vedrtnam, 2019; Feito et al., 2019), yet it is hard to find optimization studies using statistical
methods that analyze composite laminates with ply drop-offs.
2.3.1 Full factorial design. Full factorial designs are a strategy capable of estimating the
main effects and their interactions, as well as allowing the verification of the presence of
quadratic terms. The effects are measured through the change in response caused by
EC modifying the levels of the factors, as depicted in Figure 3 (Montgomery, 2017). Thus, it
becomes possible to distinguish the relevance among the factors that influence the response
and define the experiments that will be done.
For instance, Oliveira et al. (2018) used the full factorial design to evaluate some
components of the mechanical properties of reinforced composites. Almeida et al. (2018) used
a fractional factorial design to collect the data and analyze the different characteristics of
mechanical structures joints.
The factorial design may be demonstrated through a linear regression model as shown in
equation (2) (Montgomery, 2017), where y is the response, b is the estimated coefficients, x1
and x2 are the design factors and e is the random error term.

y ¼ b 0 þ b 1 x1 þ b 2 x2 þ b 12 x1 x2 þ e (2)

2.3.2 Response surface methodology. The RSM is a mathematical and statistical technique
based on fitting a second-order equation that describes the behavior of a data set. First, one
must obtain a suitable first-order approximation function between design and response
variables; however, if the system reveals a curvature, it is necessary to use a second-order
model represented by the RSM, as shown in equation (3) (Montgomery, 2017), where k
represents the number of design parameters.

X
k X
k X X
y ¼ b0 þ b i xi þ b ii x2i þ 0 b ij xi xj þ e (3)
i¼1 i¼1 i<j

A typical RSM considers a complete quadratic model, which, in this case, is constituted of
two design factors related to the ply drop-off location considering the panel and ply
positions, as shown in Figure 4(a). The factors in an RSM are plotted depending on a fitted
response surface, as Figure 4(b) depicts with a three-dimensional plot.
The optimization strategy developed with the RSM applies the desirability function, in
which response values are restricted to one and indicated by individual desirability
functions (d) ranging from 0, for an undesirable response, to 1, for a fully desired response,
creating the composite desirability function (D) defined by equation (4) (Myers et al., 2016),

Figure 3.
Full factorial
experiment
 Optimum
design of
composite
structures

Figure 4.
Typical response
surface

where di represents the desirability considering each output response and n is the number of
responses.
!1=n
1=n
Y
n
D ¼ ðd1 xd2 xd3 . . . xdn Þ ¼ di (4)
i¼1

Optimization based on the RSM using the desirability function has been applied in many
studies (Mohammed et al., 2018; Majdi et al., 2019; Bakhtiari et al., 2020; Sandhu et al., 2020;
de Almeida et al., 2020).

3. Methodology
3.1 Modeling using the finite element method
In the full factorial design, we created an experiment with 42 runs and no replicates
considering the two design factors related to the ply drop-off location: ply position,
symbolizing the longitudinal distance where the ply should be dropped. The initial analysis
considered a laminate plate divided into 8 panels (longitudinal position) with 8 stacked plies,
generating a total of 16 stacked plies and 2 ply drop-offs on the structure because of
symmetry.
An analysis using the FEM is performed to determine the response variables for the
experiment. The numerical simulation is carried out using software AnsysV R considering

two types of analyses: a static one for the response to strain and failure criterion, and a
modal one in search of the first natural frequency response, keeping in mind that no type of
loading was applied in this analysis. The carbon/epoxy composite material was selected for
the analyses because of its high strength and low mass, and its properties were obtained
through experimental tests and are depicted in Table 1. This laminate composite material is
a unidirectional prepreg with a thickness of 0.175 mm represented in the simulation by a
shell type element with 8 nodes and 6 degrees of freedom per node. The laminate plate with
ply drop-offs was designed with 16 stacked plies and discretized considering each line
divided into 10 fragments, generating a total of 800 elements and 2,581 nodes in the entire
structure. This type of element is used in many studies involving laminate composites with
EC ply drop-offs (Nasab et al., 2018; Liu et al., 2018; Panettieri et al., 2019; Shrivastava et al.,
2019) limited to an overall structural evaluation.
According to An et al. (2019), there exists a feasible angle set based on industry
experiences, such as 00, 630°, 645°, 660° and 90°, that should be taken into account when
manufacturing structures with ply drop-offs. The design considered the stacking sequence
proposed by Irisarri et al. (2014) of [450/900/450/00/450/00/450/900]s, always respecting the
design guidelines. In a simulation, the laminate plate with ply drop-offs was subjected to a
constant force evenly distributed on the central line of the thin end, creating a compression
condition. The other end of the plate was clamped, restricting movements of translation and
rotation in all directions, as shown in Figure 5. The geometry and dimensions of the
structure were chosen according to several applications using this type of structure with ply
drop-offs (Peeters and Abdalla, 2016; Sudhagar et al., 2017; Muc, 2018; Panettieri et al. (2019).
It is known that thin structures subjected to axial compression loads may suffer
buckling, as is the case of this plate with ply drop-off. Hence, a linear buckling analysis was
carried out to verify if the force applied to the plate is below the maximum critical load that
causes buckling, aiming to guarantee structural stability. The maximum critical load
supported by the plate before buckling is 2.46 MPa, and as the force applied to the plate is

Table 1. Propriety Value Unit Propriety Value Unit

Properties of a
E1 101.86 GPa E2 3.41 GPa
carbon/epoxy G12 7.56 GPa v12 0.30 –
composite material r 1550.00 kg/m3 s 1T 1363.49 MPa
considering the s 1C 572.27 MPa s 2T 5.86 MPa
failure criterion s 2C 102.00 MPa t 12U 200.61 MPa

Figure 5.
Representation of
dimensions and
boundary conditions
for the laminate plate
with ply drop-offs
lower, one can state that the plate is secure and will not suffer buckling. The first buckling Optimum
mode is presented in Figure 6. design of
composite
3.2 Optimization approach using design of experiments structures
The full factorial design was created with two factors. Based on preliminary tests, it was
possible to define the experimental boundaries of the present study considering factorial
points (1 and þ1), axial points (a and þ a) and central points, as depicted in Table 2.
The analysis with the FEM allowed completing the full factorial design (Appendix 1) with
the response variables for each factor combination. Furthermore, the results revealed by the full
factorial design identified the significance of the design factors and response variables.
The optimization strategy was performed using the RSM with a desirability function,
seeking to achieve the optimal combination of ply drop-off locations that provides excellent
performance for the laminate plate. To satisfy the requirements of the RSM, the panel
position was changed to the longitudinal position in meters. The experimental matrix
established by the RSM considered axial points and generated a total of 14 runs, as shown in
Table 3, where TW, Lp and Pp represent, respectively, the Tsai–Wu failure criterion, the
longitudinal position in meters and the ply position, with the latter being based on the
number of plies, so the values from 1 to 4 were rounded to 3–7. The strategy used in this
work is fully detailed in Figure 7.

Figure 6.
First buckling mode
of the laminate plate

Levels
Design factors Symbol a 1 0 þ1 þa
Table 2.
Ply position () Pp 2 3 5 7 8 Design factors and
Longitudinal position (m) Lp 2 2.73 4.5 6.27 7 their levels (a = 2k/4)

Design factors Response variables

Lp (m) Pp (uni) TW Strain (mm/mm) Frequency (Hz)

1 0.05119 2.87867 0.45243 0.00140 100.27000

2 0.11730 2.87867 0.45243 0.00133 100.57000
3 0.05119 7.12132 0.45243 0.00140 98.48200
4 0.11730 7.12132 0.45243 0.00133 100.54000
5 0.08425 5.00000 0.45243 0.00137 101.06000
6 0.08425 5.00000 0.45243 0.00137 101.06000
7 0.08425 5.00000 0.45243 0.00137 101.06000
8 0.03750 5.00000 0.45252 0.00142 95.39500
9 0.13100 5.00000 0.45243 0.00132 98.73700
10 0.08425 2.00000 0.45243 0.00132 101.88000
11 0.08425 8.00000 0.45243 0.00132 100.83000
12 0.08425 5.00000 0.45243 0.00137 101.06000 Table 3.
13 0.08425 5.00000 0.45243 0.00137 101.06000 Experimental matrix
14 0.08425 5.00000 0.45243 0.00137 101.06000 for the RSM analysis
EC

Figure 7.
Flowchart of the
global optimization
strategy using DOE

4. Main results and discussion

4.1 Numerical ply drop-off modeling
The design factor combinations were inputted into the finite element analysis algorithm so
the obtained results could be used in a full factorial design as response variables. The Tsai–
Wu failure criterion response variable showed reliable results in analyses using the FEM,
with it being over one (TW  1) in all analyses. As one may observe in Table 3, many failure
criterion values demonstrated similarity; because only two plies are dropped, there was no
considerable change to the failure mechanisms.
Given that the main aim of this study is to determine the optimal ply drop-off location,
the problem formulated here considered a single stacking sequence as composite structures
are quite susceptible to imperfections during their manufacture, mainly with regard to ply
orientation, a topic that should be included in future studies. It is important to highlight that
the model adopted did not take into account the resin pockets in the manufacturing process.
Regarding the strain and first natural frequency response variables, all results were
feasible, as one may observe in Table A1 and A2 (Appendix). Figure 8 illustrates the results
for each response variables considering the ten combinations (Table A1 and A2 in
Appendix) generated in the full factorial design.

4.2 Analysis of variance for the full factorial design

In this study, the primary interest in using the full factorial design is to create variations in
the factor levels and verify which factors influence the response variables the most. The
results generated by the full factorial design were examined using the analysis of variance
 Optimum
design of
composite
structures

Figure 8.
Carbon fiber
reinforced polymer
plate with ply drop-
off responses
EC (ANOVA), which, by means of comparison of the variances, determine the difference in a set
of means and studies the quality of a fit, establishing a regression model for the system.
To measure the quality of a fit, the adjusted coefficient of determination R2adj was used.
This coefficient demonstrates how well the model fits the data and determines the predictive
capability of the regression model. The R2adj value ranges from 0 to 100 and a model with a
value above 80% is considered appropriate, revealing that it explains the variability of the
fitted data (Montgomery et al., 2012). Besides the adjusted coefficient of determination, the
ANOVA analyzes the factors through the p-value, with values lower than 0.05 representing
that there is a statistically significant association between the response variables and design
factors. Table 4 shows the ANOVA results relative to the strain, Tsai–Wu failure criterion
and first natural frequency response variables, while Table 5 presents the ANOVA results
considering the p-value for design factors.
In this case, the ANOVA revealed that the model using two design factors is only
appropriate for the strain and first natural frequency response variables. In contrast, the
failure criterion response value showed a poor fit, presenting a low adjusted coefficient of
determination, as shown in Table 4. Considering the analyses of the design factors using the
p-value in Table 5, it is possible to establish that only the ply position has a significant
influence on the strain response variable.

4.3 Optimization approach using the response surface methodology

Considering the more expressive parameters obtained using the full factorial design, the
RSM was applied with the intention of examining the quality of the fit, verifying the effect of
interactions and determining the optimal parameters for the ply drop-off location. The
ANOVA generated in the RSM again revealed that the strain and first natural frequency
response variables provide a good interpretation of the problem, while the failure criterion
response variable remains with a low fit quality, as shown in Table 6.
Values of R2adj higher than 90% highlight how well the model fits the data relative to the
strain and first natural frequency response variables. In turn, the failure criterion showed a
low predictive capability, with R2adj less than 60%. An R2(pred) value of 0% demonstrates
overfitting, and a possible interpretation for this is the approximation of the decimal places
of the failure criterion values.

Table 4.
Response variables R2adj (%)
Adjusted coefficient
of determination Strain 84.35
R2adj in the ANOVA Tsai–Wu failure criterion 25.41
model First natural frequency 82.10

Strain Tsai–Wu Frequency

Source p-value Source p-value Source p-value
Table 5. Plies 0.049 Plies 0.462 Plies 0.113
Model summary of Panels 0.084 Panels 0.159 Panels 0.064
the ANOVA Error 0.000 Error 0.000 Error 0.000
To predict the effect of interactions among the design factors and the contribution of each Optimum
one on the response variability, Pareto charts were generated, as shown in Figure 9. design of
Pareto charts allow clear interpretations about the effects of the factors and their
interactions, with the effects located above the dashed red line being the most significant.
composite
One may observe that the effects of the longitudinal position and ply position are significant structures
for the strain response variable. For the failure criterion response variable, only the
longitudinal position has a significant effect, and, finally, for the first natural frequency, all
effects are significant. Consequently, it is possible to state that the longitudinal position has
a considerable influence on the ply drop-off locations on the laminate plate.
Considering the effects of the factors and their interactions on each response variable,
main effect charts were elaborated aiming to provide the difference between the levels of the
design factors relative to the response variables, as depicted in Figure 10.
The main effect charts enhance different positions for ply drop-offs with the purpose of
meeting the objectives of each response variable. The optimal ply drop-off location
considering the strain response variable is after 0.12 meters in the longitudinal position,
dropping-off at ply position 2 or 8. Regarding the first natural frequency response variable,
the optimum is 0.09 meters in the longitudinal position, dropping-off at ply position 2, as
shown in Figure 10. The effect of the ply position factor proved not to be adequate for the
failure criterion response because of its low level of significance; therefore, only the
longitudinal position was presented in Figure 10.
To evaluate the numeric relation among the design factors and response variables,
regression equations were generated considering each response variable, shown in
equations (5)–(7).

Strain « ¼ 0:001338  0:001014Lp þ 0:000049Pp  0:000005Pp2 (5)

Tsai  Wu TW ¼ 0:452580  0:003150Lp þ 0:01584Pp2 (6)

frequency v ¼ 92:54 þ 274:3Lp  1:470Pp  1655L2p þ 0:0747Pp2 þ 6:27Lp Pp

(7)

It is worth highlighting that regression equations demonstrate the combination of levels for
each factor, and are often used as prediction models to optimize design factors. To check
how the fitted model relates to the design factors, surface plots were generated together with
contour plots, as shown in Figures 11 and 12. Because of the low level of significance, the
failure criterion was not plotted.
The surface plots showed in Figure 11 reveal a quadratic response surface in which the
maximum points for first natural frequency and the minimum point for strain are located in
the optimal region. The contour plots exposed in Figure 12 reveal the optimal region in a
two-dimensional view.

Response S R2 (%) R2adj (%) R2(pred) (%)

Strain 0.0000075 96.31 94.67 88.96 Table 6.

Tsai–Wu 0.0000172 56.51 48.60 0.00 Model summary of
First natural frequency 0.414654 96.67 93.81 70.87 the RSM
EC

Figure 9.
Pareto chart for the
effects
 Optimum
design of
composite
structures

Figure 10.
Main effects plot for
(a) strain, (b) failure
criterion and (c) first
natural frequency
EC The optimal region that minimizes the strain and maximizes the first natural frequency is
known; however, it is essential to find the optimal conditions for the ply drop-off locations
considering specific values for each design factor. Hence, in this study, an optimization
strategy was developed using the desirability function in both mono-objective and multi-
objective approaches. The mono-objective optimization is presented in Table 7 with the
individually optimized responses, while Table 8 shows a multi-objective optimization
considering the responses optimized simultaneously. One may observe that the meta-model
created using the RSM with a desirability function describes the structural behavior in

Figure 11.
Surface plot for (a)
strain and (b) first
natural frequency
considering the fitted
response

Figure 12.
Contour plot for (a)
strain and (b) first
natural frequency
considering the fitted
response

Objective Response Lp Pp Predicted response FEM response Desirability

Table 7.
Mono-objective 1 Minimize Strain 0.1310 8 0.0013 0.0013 1.0000
optimization 2 Maximize First natural frequency 0.086600 2 102.3260 101.8250 1.0000

Objective Response Lp Pp Predicted response FEM response Desirability

Table 8.
Multi-objective Minimize Strain 0.1008 2 0.0013 0.0013 1.0000
optimization Maximize First natural frequency 101.9964 101.3290
terms of the minimum strain with almost zero errors and less than 0.50% for the first Optimum
natural frequency. design of
In the multi-objective optimization, the optimal ply drop-off location relative to the
longitudinal position factor is 0.1008 meters, with this position being located between five
composite
and six panels, dropping-off at ply position 2. The dropping-off of plies oriented at a 90° structures
angle is most indicated in both optimization strategies, as one may observe in
Tables 7 and 8. This orientation is considered smoother, and when plies oriented at 90°
are dropped on the thin end of the laminate plate, the stress concentration is reduced
(Vidyashankar and Murty, 2001).
The optimum results were obtained through an optimization method considering a multi-
objective algorithm. Regarding the multi-objective algorithm, it was decided to use the
evolutionary multi-objective algorithm sunflower optimization (Gomes et al., 2018; Gomes
and Giovani, 2020). The sunflower algorithm is a recent metaheuristic developed especially
(but not limited) for problems in mechanical structures. The algorithm’s principle is the
phototropic behavior of sunflowers in search of the best source of solar radiation. Combined
with the search for additional pollination and plant mortality parameters, the metaheuristic
shows significant performance (Gomes and de Almeida, 2020).
For multi-objective optimization, a population of 40 individuals was considered as
parameters of control of the algorithm, with pollination rate pp = 0.80, mortality rate = 0.10
and survival = 0.10. The number of suns ns = 3 were considered, which is a metric to
increase the spread distance in multi-objective optimization. A total of 100 iterations were
adopted as the stopping criterion.
Figure 13 shows the final result of the optimization considering both static and dynamic
objectives (strain – first natural frequency). A significant amount of possible (non-
dominated) solutions is observed. As a decision-making criterion, it was decided to use the
“knee point” strategy. The knee point is considered the point with the greatest convexity
and is often located in the middle of the curve, representing the best combination of
responses variables (Zhang et al., 2014; Zou et al., 2017). The knee point showed the optimal
response variable with a strain of 0.001304 and a first natural frequency of 101.4 Hz, values
quite similar to those obtained through the validation using the FEM in Table 8.

Figure 13.
Optimal solution
found using a Pareto
front considering the
strain and first
natural frequency
response variables
EC The optimization strategy demonstrated very appropriate results, as one may observe by
comparing the predicted response and the response generated using the FEM. To prove the
accuracy of the results, merely replace the optimal design factors in the regression equations
[equations (4)–(6)], which will render response variables similar to the FEM responses. The
optimal solution is represented in Figure 14.
With knowledge of the optimal solution that satisfies the conditions imposed by the
response variables, Figure 15 illustrates the optimal ply drop-off location on a laminate plate
considering the optimized responses found by the optimization strategies.

Figure 14.
Optimal solution for
ply drop-off location
on a laminate plate
considering the ply
and longitudinal
positions

Figure 15.
Optimized laminate
plate considering the
ply drop-off location
 Optimum
design of
composite
structures

Figure 16.
Plot of the failure
criterion for each ply
considering the
optimal parameters
obtained from the
multi-objective
optimization
EC Regardless, the optimal parameters for a laminate plate with ply drop-off found in the multi-
objective optimization were used in a failure analysis considering the individualized plies, as
shown in Figure 16.
One may observe that, for the first eight plies, the failure criterion increased in the region
near the fixed end, then, in the symmetrical part of the plate, the failure criterion decreased
in the region near the fixed end and increased at the plate end. Hence, it may be stated that
the ply drop-off location does not influence the failure criterion, given that ply #2 was
dropped, and the failure criterion changed after ply position 8 of the laminate plate, as
shown in Figure 16. One reason for this is that the ply dropped is oriented at 90°, and when
this type of ply is dropped, it presents low inter-laminar stress, demonstrating a negligible
effect on the stress distribution that usually does not cause structural failure.

5. Conclusions
In this study, an optimization strategy was proposed using DOE to analyze the optimal ply
drop-off location in the laminate plate. The main objective of this study was to optimize the
ply drop-off location considering the performance of the laminate structure in relation to
failure, strain and stiffness.
First, a full factorial design with the possible combinations of ply drop-off locations was
created. These combinations were implemented using a finite element analysis, aiming at
the structural performance of the laminate plate, and then the full factorial design was
completed and executed. The ANOVA presented the full factorial design results
demonstrating the influence of each design factor on the response variables. The failure
criterion response variable did not reveal a statistical significance in relation to the design
factors. Finally, the RSM was used to analyze the process variability and optimize the ply
drop-off location, with the finite element analysis being used again for the structural
performance responses. The RSM confirmed the low fit quality for the failure criterion
response variable, which implies a lack of predictive capability.
The optimization strategy using DOE revealed to be satisfactory for analyzing laminate
structures with ply drop-offs, demonstrating that not all design factors influence the
response variability, as one may observe with the ply position. In turn, the longitudinal
position design factor was statistically significant for all response variables.
The failure criterion response variable revealed a poor fit, with an adjusted coefficient of
determination lower than 60%, thus demonstrating that the response did not vary with the ply
drop-off location. This fact may be justified by the reduced number of ply drop-offs in the
laminate structure that led to changes in the stress associated with failures. Another situation that
stimulates poor fit for the failure criterion is the dropping-off of plies oriented at 90°, which when
dropped in the thin section, reduces the stress concentration in the region near the ply drop-off.
Future works should consider a hybrid laminate structure with carbon/epoxy/glass and/or
metallic layers with an increased amount of ply drop-offs in the structure. In addition, the
orientation of the plies is a very promising design factor for future studies, given that it showed
a strong relevance for the optimal ply drop-off location relative to the response variables.

References
Adams, D.F., Ramkumar, R.L. and Walrath, D.E. (1984), Analysis of Porous Laminates in the Presence of
Ply Drop-Offs and Fastener Holes (N. NOR-84-113), Northrop Corp Hawthorne Ca Aircraft Div.
Adluru, H.K., Hoos, K.H., Iarve, E.V. and Ratcliffe, J.G. (2019), “Delamination initiation and migration
modeling in clamped tapered laminated beam specimens under static loading”, Composites Part
A: Applied Science and Manufacturing, Vol. 118, pp. 202-212.
Albanesi, A., Bre, F., Fachinotti, V. and Gebhardt, C. (2018), “Simultaneous ply-order, ply-number and Optimum
ply-drop optimization of laminate wind turbine blades using the inverse finite element method”,
Composite Structures, Vol. 184, pp. 894-903.
design of
Almeida, F.A., Gomes, G.F., Sabioni, R.C., Gomes, J.H.F., Paula, V.R., Paiva, A.P. and Costa, S.C. (2018),
composite
“A gage study applied in shear test to identify variation causes from a resistance spot welding structures
measurement system”, Strojniški vestnik – Journal of Mechanical Engineering, Vol. 63,
pp. 621-631.
An, H., Chen, S. and Huang, H. (2019), “Stacking sequence optimization and blending design of
laminated composite structures”, Structural and Multidisciplinary Optimization, Vol. 59 No. 1,
pp. 1-19.
Bakhtiari, A., Ashenai Ghasemi, F., Naderi, G. and Nakhaei, M.R. (2020), “An approach to the
optimization of mechanical properties of polypropylene/nitrile butadiene rubber/halloysite
nanotube/polypropylene-g-maleic anhydride nanocomposites using response surface
methodology”, Polymer Composites, Vol. 41 No. 6.
Barbero, E.J. (2014), Finite Element Analysis of Composite Materials using Ansys, CRC press, VT.
De Almeida, F.A., Santos, A.C.O., de Paiva, A.P., Gomes, G.F. and de Freitas Gomes, J.H. (2020),
“Multivariate Taguchi loss function optimization based on principal components analysis and
normal boundary intersection”, Engineering with Computers, pp. 1-17.
Dhurvey, P. and Mittal, N. (2013), “Study the effect of externally and internally ply drop-off in
composite laminate analysis”, ARPN Journal of Engineering and Applied Sciences, Vol. 8 No. 4,
pp. 595-605.
Diniz, C.A., Cunha, S.S., Gomes, G.F. and Ancelotti, A.C. (2019), “Optimization of the layers of composite
materials from neural networks with Tsai–Wu failure criterion”, Journal of Failure Analysis and
Prevention, Vol. 19 No. 3, pp. 709-715.
Feito, N., Muñoz-S
anchez, A., Díaz-Álvarez, A. and Miguelez, M.H. (2019), “Multi-objective optimization
analysis of cutting parameters when drilling composite materials with special geometry drills”,
Composite Structures, Vol. 225, p. 111187.
Fish, J.C. and Lee, S.W. (1989), “Delamination of tapered composite structures”, Engineering Fracture
Mechanics, Vol. 34 No. 1, pp. 43-54.
Ghayoor, H., Marsden, C.C., Hoa, S.V. and Melro, A.R. (2019), “Numerical analysis of resin-rich areas
and their effects on failure initiation of composites”, Composites Part A: Applied Science and
Manufacturing, Vol. 117, pp. 125-133.
Gomes, G.F. and de Almeida, F.A. (2020), “Tuning metaheuristic algorithms using mixture design:
application of sunflower optimization for structural damage identification”, Advances in
Engineering Software, Vol. 149, p. 102877.
Gomes, G.F. and Giovani, R.S. (2020), “An efficient two-step damage identification method using
sunflower optimization algorithm and mode shape curvature (MSDBI–SFO)”, Engineering with
Computers, pp. 1-20.
Gomes, G.F., da Cunha, S.S., Jr and Ancelotti, A.C. Jr (2018), “A sunflower optimization (SFO) algorithm
applied to damage identification on laminated composite plates”, Engineering with Computers,
Vol. 35 No. 2, pp. 619-626.
Gomes, G.F., Diniz, C.A., da Cunha, S.S. and Ancelotti, A.C. (2017), “Design optimization of composite
prosthetic tubes using GA-ANN algorithm considering Tsai-Wu failure criteria”, Journal of
Failure Analysis and Prevention, Vol. 17 No. 4, pp. 740-749.
Irisarrri, F.X., Peeters, D.M.J. and Abdalla, M.M. (2016), “Optimisation of ply drop order in variable
stiffness laminates”, Composite Structures, Vol. 152, pp. 791-799.
Irisarri, F.X., Lasseigne, A., Leroy, F.H. and Le Riche, R. (2014), “Optimal design of laminated composite
structures with ply drops using stacking sequence tables”, Composite Structures, Vol. 107,
pp. 559-569.
EC Kemp, B.L. and Johnson, E.R. (1985), “Response and failure analysis of a graphite/epoxy laminate
containing terminating internal plies”, 26th Structures, Structural Dynamics, and Materials
Conference, p. 608.
Liu, Q., Li, Y., Cao, L., Lei, F. and Wang, Q. (2018), “Structural design and global sensitivity analysis of
the composite B-pillar with ply drop-off”, Structural and Multidisciplinary Optimization, Vol. 57
No. 3, pp. 965-975.
Lund, E. (2018), “Discrete material and thickness optimization of laminated composite structures
including failure criteria”, Structural and Multidisciplinary Optimization, Vol. 57 No. 6,
pp. 2357-2375.
Majdi, H., Esfahani, J.A. and Mohebbi, M. (2019), “Optimization of convective drying by response
surface methodology”, Computers and electronics in Agriculture, Vol. 156, pp. 574-584.
Mohammed, B.S., Khed, V.C. and Nuruddin, M.F. (2018), “Rubbercrete mixture optimization using
response surface methodology”, Journal of Cleaner Production, Vol. 171, pp. 1605-1621.
Montgomery, D.C. (2017), Design and Analysis of experiments, John Wiley and Sons, AZ.
Montgomery, D.C., Peck, E.A. and Vining, G.G. (2012), Introduction to Linear Regression Analysis
(Vol. 821), John Wiley and Sons, NJ.
Muc, A. (2018), “Design of blended/tapered multilayered structures subjected to buckling constraints”,
Composite Structures, Vol. 186, pp. 256-266.
Mukherjee, B. and Varughese, B. (2001), “Design guidelines for ply drop-off in laminated composite
structures”, Composites Part B: Engineering, Vol. 32 No. 2, pp. 153-164.
Myers, R.H., Montgomery, D.C. and Anderson-Cook, C.M. (2016), Response Surface Methodology:
Process and Product Optimization Using Designed Experiments, John Wiley and Sons, NJ.
Naresh, K., Shankar, K., Velmurugan, R. and Gupta, N.K. (2018), “Statistical analysis of the tensile
strength of GFRP, CFRP and hybrid composites”, Thin-Walled Structures, Vol. 126, pp. 150-161.
Nasab, F.F., Geijselaers, H.J.M., Baran, I., Akkernan, R. and de Boer, A. (2018), “A leve-set-based
strategy for thickness optimization of blended composite structures”, Composite Structures,
Vol. 206, pp. 903-920.
Oliveira, L.Á., Santos, J.C., Panzera, T.H., Freire, R.T., Vieira, L.M. and Scarpa, F. (2018), “Evaluation of
hybrid-short-coir-fibre-reinforced composites via full factorial design”, Composite Structures,
Vol. 202, pp. 313-323.
Panettieri, E., Montemurro, M. and Catapano, A. (2019), “Blending constraints for composite laminates
in polar parameters space”, Composites Part B: Engineering, Vol. 168, pp. 448-457.
Peeters, D. and Abdalla, M. (2016), “Optimization of ply drop locations in variable-stiffness
composites”, AIAA Journal, Vol. 54 No. 5, pp. 1760-1768.
Sandhu, K.S., Sharma, L., Kaur, M. and Kaur, R. (2020), “Physical, structural and thermal properties of
composite edible films prepared from pearl millet starch and carrageenan gum: process
optimization using response surface methodology”, International Journal of Biological
Macromolecules, Vol. 143, pp. 704-713.
Sanz-Corretge, J. (2019), “A constraint satisfaction problem algorithm for large-scale multi-panel
composite structures”, Structural and Multidisciplinary Optimization, Vol. 60 No. 5, pp. 1-17.
Shim, D.J. (2002), “Role of delamination and interlaminar fatigue in the failure of laminates with ply
drop-offs”, Thesis (Doctor of Philosoply in Aeronautics and Astronautics), Department of
Aeronautics and Astronautics, Massachusetts Institute of Technology.
Shrivastava, S., Mohite, P.M. and Limaye, M.D. (2019), “Optimal design of fighter aircraft wing panels
laminates under multi-load case environment by ply-drop and ply-migrations”, Composite
Structures, Vol. 207, pp. 909-922.
Singh, D. and Rao, P.V. (2007), “A surface roughness prediction model for hard turning process”, The
International Journal of Advanced Manufacturing Technology, Vol. 32 Nos 11/12, pp. 1115-1124.
Sjølund, J.H. and Lund, E. (2018), “Structural gradient based sizing optimization of wind turbine blades Optimum
with fixed outer geometry”, Composite Structures, Vol. 203, pp. 725-739.
design of
Skrypnyk, R., Nielsen, J.C., Ekh, M. and Pålsson, B.A. (2019), “Metamodelling of wheel–rail normal
contact in railway crossings with elasto-plastic material behavior”, Engineering with Computers, composite
Vol. 35 No. 1, pp. 139-155. structures
Sudhagar, P.E., Babu, A.A., Rajamohan, V. and Jeyaraj, P. (2017), “Structural optimization of rotating
tapered laminated thick composite plates with ply drop-offs”, International Journal of Mechanics
and Materials in Design, Vol. 13 No. 1, pp. 85-124.
Tsai, S.W. and Wu, E.M. (1971), “A general theory of strength for anisotropic materials”, Journal of
composite Materials, Vol. 5 No. 1, pp. 58-80.
Vedrtnam, A. (2019), “Novel method for improving fatigue behavior of carbon fiber reinforced epoxy
composite”, Composites Part B: Engineering, Vol. 157, pp. 305-321.
Vemuluri, R.B., Rajamohan, V. and Sudhagar, P.E. (2018), “Structural optimization of tapered
composite sandwich plates partially treated with magnetorheological elastomers”, Composite
Structures, Vol. 200, pp. 258-276.
Vidyashankar, B.R. and Murty, A.V.K. (2001), “Analysis of laminates with ply drops”, Composites
Science and Technology, Vol. 61 No. 5, pp. 749-758.
Voyiadjis, G.Z. and Kattan, P.I. (2005), Mechanics of Composite Materials with Matlab, Springer Science
and Business Media, United Stated of American, USA.
Weiss, A., Trabelsi, W., Michel, L., Barrau, J.J. and Mahdi, S. (2010), “Influence of ply-drop location on
the fatigue behaviour of tapered composites laminates”, Procedia Engineering, Vol. 2 No. 1,
pp. 1105-1114.
Zhang, X., Tian, Y. and Jin, Y. (2014), “A knee point-driven evolutionary algorithm for many-objective
optimization”, IEEE Transactions on Evolutionary Computation, Vol. 19 No. 6, pp. 761-776.
Zhao, W. and Kapania, R.K. (2018), “BLP optimization of composite flying-wings with sparibs and
multiple control surfaces”, In: 2018 AIAA/ASCE/AHS/ASC Structures, Structural Dynamics,
and Materials Conference, p. 2150.
Zou, J., Li, Q., Yang, S., Bai, H. and Zheng, J. (2017), “A prediction strategy based on center points and
knee points for evolutionary dynamic multi-objective optimization”, Applied Soft Computing,
Vol. 61, pp. 806-818.
EC Appendix. General full factorial design

A full factorial design was generated first, aiming to recognize the influence of the design factors on
the ply drop-off locations, as shown in Tables A1 and A2.

Design factors Output responses

No. Plies Panels Tsai–Wu Strain (mm/mm) Frequency (Hz)

1 2 2 0.45215 0.00133 99.76500

2 2 3 0.45242 0.00132 101.28000
3 2 4 0.45243 0.00132 101.88000
Table A1. 4 2 5 0.45243 0.00131 101.63000
5 2 6 0.45243 0.00131 100.58000
General full factorial
6 2 7 0.45243 0.00130 98.73800
design considering 7 3 2 0.45252 0.00142 97.67300
two design factors 8 3 3 0.45243 0.00140 101.27000
and three response 9 3 4 0.45243 0.00137 101.50000
variables (Part I) 10 3 5 0.45243 0.00135 101.53000
 Design factors Output responses
Optimum
No. Plies Panels Tsai–Wu Strain (mm/mm) Frequency (Hz) design of
composite
11 3 6 0.45243 0.00133 100.57000
12 3 7 0.45243 0.00132 98.73800 structures
13 4 2 0.45212 0.00173 90.41700
14 4 3 0.45242 0.00166 96.46600
15 4 4 0.45243 0.00159 100.01000
16 4 5 0.45243 0.00151 101.14000
17 4 6 0.45243 0.00144 100.51000
18 4 7 0.45243 0.00137 98.73600
19 5 2 0.45252 0.00142 95.39500
20 5 3 0.45243 0.00140 99.11900
21 5 4 0.45243 0.00137 101.06000
22 5 5 0.45243 0.00135 101.42000
23 5 6 0.45243 0.00133 100.55000
24 5 7 0.45243 0.00132 98.73700
25 6 2 0.45212 0.00173 93.09600
26 6 3 0.45242 0.00166 97.91800
27 6 4 0.45243 0.00159 100.59000
28 6 5 0.45243 0.00151 101.29000
29 6 6 0.45243 0.00144 100.53000
30 6 7 0.45243 0.00137 98.73600
31 7 2 0.45252 0.00142 94.16200
32 7 3 0.45243 0.00140 98.48200
33 7 4 0.45243 0.00137 100.81000
34 7 5 0.45243 0.00135 101.35000
35 7 6 0.45243 0.00133 100.54000
36 7 7 0.45243 0.00132 98.73700 Table A2.
37 8 2 0.45215 0.00133 94.20900
General full factorial
38 8 3 0.45242 0.00132 98.51200
39 8 4 0.45243 0.00132 100.83000 design considering
40 8 5 0.45243 0.00131 101.36000 two design factors
41 8 6 0.45243 0.00131 100.54000 and three response
42 8 7 0.45243 0.00130 98.73700 variables (Part II)

Corresponding author
Camila Aparecida Diniz can be contacted at: camiladiniz@unifei.edu.br

For instructions on how to order reprints of this article, please visit our website:
www.emeraldgrouppublishing.com/licensing/reprints.htm
Or contact us for further details: permissions@emeraldinsight.com

View publication stats