The 2nd National Conference on Mechanical Engineering and Automation
October 7-8, 2016, Hanoi University of Science and Technology
Fuzzy Natural Frequencies Analysis of Geometrically
Nonlinear 2-D Steel Frame Structures with Fixity Factors and
Mass modeled as Triangular Fuzzy Numbers
Tran Thanh Viet * Vu Quoc Anh ** and Le Xuan Huynh ***
* Department of Civil Engineering, Ph.D. Student, DuyTan University, K7/25 Quang Trung, Da Nang, Vietnam
E-mail: ttviet05@gmail.com
** Department of Civil Engineering, Hanoi Architectural University, 10 Tran Phu, Hanoi, Vietnam
*** Department of Civil Engineering, National University of Civil Engineering, 55 Giai Phong, Hanoi, Vietnam
Abstract
This paper concerns on using algorithms for determining the
fuzzy natural frequencies of geometrically nonlinear planar steel
frame structures in which fixity factors of beam – column,
column – base connections, and mass per unit volume are
modeled as triangular fuzzy numbers. The finite element
method with elastic semi-rigid connection and Response
Surface Method (RSM) in mathematical statistic are applied in
the fuzzy structural analysis with symmetric triangular fuzzy
numbers. The Į – level optimization using the Differential
Evolution (DE) involving integrated finite element modeling is
proposed to apply in problems with any triangular fuzzy
numbers. The advantage of the proposed methodologies is
demonstrated on example problem relating to the twenty – story,
four – bay planar steel frame.
Keywords: Steel frame, Geometric nonlinearity, Natural
frequency, Fuzzy connection, Response surface method,
Differential evolution algorithm.
1. Introduction
In the dynamic analysis of steel frame structures with
the elastic semi-rigid connection, the fixity factor of
connection and the mass per unit volume have a significant
influence on the natural frequencies of a steel frame. In
practice, however, many parameters like worker skill,
quality of welds, and type of the connecting elements
affect the behavior of a connection, and this fixity factor is
difficult to determine exactly. Therefore, in a practical
analysis of structures, a systematic approach is needed to
include the uncertainty in the joints behavior and the fixity
factor of a connection modeled as a fuzzy number is
reasonable [1]. Moreover, the presence of an axial force
affects the stiffness of a beam-column and the geometric
nonlinearity (P - G effect) is also considered.
For the classical finite element method, Anh et al. [2]
considered the natural frequencies of the geometrically
nonlinear steel frame structures with different semi-rigid
connections. In recent years, the static analysis for planar
steel frame structure with fuzzy connections has been
reported [1]. However, the dynamic analysis for
determining the fuzzy natural frequencies of geometrically
nonlinear steel frames by using the exact approach with
209
the numerical method has been limited. In the fuzzy
structural analysis, for a rigid frame, Tuan et al. [5]
presented an approach by using Response Surface
Method (RSM) for fuzzy free vibration analysis of linear
elastic structure. In their study, response surfaces
(surrogate functions) in terms of complete quadratic
polynomials are presented for model quantities and all
fuzzy variables are standardized. The usage of the RSM
shows that this approach has effectiveness for the
complex structural problems with a large number of
fuzzy variables. However, the RSM is only suitable for
problems in which all fuzzy variables are modeled as
symmetric triangular fuzzy numbers. For the problems
with non-symmetric triangular fuzzy numbers, the fuzzy
structural analysis must use another approach. Anh et al.
[6] presented an optimization algorithm for fuzzy analysis
by combining the Differential Evolution (DE) with the Į
– level optimization. DE is a global optimization
technique, which combines the evolution strategy and the
Monte Carlo simulation, and is simple and easy to use.
In this paper, the fuzzy first two natural frequencies of
the geometrically nonlinear planar steel frame structure
with fuzzy fixity factor and fuzzy mass per unit volume
are determined by using two approaches for solutions.
The first approach is based on the classical finite element
method in combination with the response surface method
for fuzzy fixity factor input and obtained fuzzy natural
frequencies output. This is implemented similarly to the
approach reported in [5], however, the finite element is
extended with elastic semi-rigid connection presented by
Anh et al. [2]. The second approach is based on finite
element model by combining the Į – level optimization
with the Differential Evolution algorithm which is a
population-based optimizer. The DE is similar to a
genetic algorithm (GA), but it is simple, easy for
application and its global convergence and robustness are
better than most other GAs [6, 7]. Two solution
approaches are different and applied to problems with
various fuzzy inputs. In the first approach, the fuzzy
fixity factor modeled as the non-symmetric triangular
fuzzy number has not considered yet. This is
implemented in the second approach and that is the
advantage of DE. A comparison of the fuzzy natural
frequencies between the RSM and the DE is presented by
 Tran Thanh Viet, Vu Quoc Anh and Le Xuan Huynh

considering the twenty floor, four bay planar steel frame

12 EI s1 (3c)
structure in which the fixity factors are modeled as k33 2k63
symmetric triangular fuzzy numbers. The obtained results L  4  s1s2 
are not significantly different. Hence, the Į – level
optimization in combination with the Differential 6 EI s2  s1  2  (3d)
k62 k65
Evolution algorithm is applied to this analysis, in which L2  4  s1s2 
considering the fuzzy fixity factors at the boundary
constrains are modeled as non-symmetric triangular fuzzy 12 EI s2 (3e)
k66
numbers. In addition, the fuzzy natural frequencies of L  4  s1s2 
different cases are also solved by using DE with both
ª0 º
fuzzy fixity factors and symmetric triangular fuzzy mass «0 »
per unit volume. « k 22g symmetric »
«0 k32g k33g » (4)
2. Finite element with geometrically ª¬ K Gel º¼ c« »
nonlinear elastic semi-rigid connection «0 0 0 0 »
The natural frequencies are determined by solving the
«0 k52g k53g 0 k55g »
« g »
Eigenvalue equation. ¬« 0 k62g k63g 0 k65g k66 ¼»
with
det > K  K G @  Z 2 > M @ 0 (1) P
c (5a)
30 L  4  s1s2 
2

where [K] is the assembled stiffness matrix of the frame,

[KG] is the assembled geometric stiffness matrix of the k22g k55g k52g 12 s1s2  s2  34  
(5b)
frame, and [M] is the assembled mass matrix of the frame.  96  s  5   12  3s22  s2  8  s12
2
2
The frame element with elastic semi – rigid connection
k32g  k53g 12 Ls1s2  4 s2  7  
as shown in Fig. 1, with E - the elastic modulus, A – the (5c)
section area, I – the inertia moment, m - the mass per unit  3Ls12  s22  12 s2  32 
volume, and ki – the rotation resistance stiffness at
connections (i = 1,2). k33g 12 L2 s12  2s22  7 s2  8 (5d)

E, A, I, m k g
62 k g
65 96 Ls 2  3Ls2  s2  16  s 
2 2
1
1 2 (5e)
k1 k2  12 Ls1s2  3s2  7 
3 L2 s1 s2  28  16 s2  s1  7 s2  16  
L
k 63g (5f)
Fig. 1. Frame element with semi-rigid connections.
k g
66 12 L s  2s  7 s1  8
2 2
2
2
1
(5g)
The element stiffness matrix - [Kel], the element
geometric stiffness matrix - [KGel] of the frame, and the and
element mass matrix - [Mel] of the frame are given as
follows [2]
ª140d 2 º
« »
ª EA º « 0 m22 symmetric »
« L » mAL « 0 m32 m33 »
« » ª¬ M el º¼ « »
« 0 k22 symmetric » 420d 2 « 70d 2 0 0 140d 2 »
(2) « 0 »
« 0 k32 k33 » m52 m53 0 m55
« » « »
¬ª K ¼º
el

«  EA EA » ¬« 0 m62 m63 0 m65 m66 ¼»

« L 0 0 »
L
« »
« 0 k52 k53 0 k55 » (6)
«¬ 0 k62 k63 0 k65 k66 »¼
where
where
d 4  s1s2 (7a)
12 EI  s1  s2  s1s2 
k22 k55 k52 (3a)
L3  4  s1s2  m22 16 140  49 s2  8s22   4 s12  32  55s2  32 s22  
(7b)
 4 s1  224  328s2  50 s22 
6 EI s1  s2  2 
k32 k53 (3b)
L2  4  s1s2  m33 4 L2 s12  32  31s2  8s22  (7c)

210
 Fuzzy Natural Frequencies Analysis of Goemetrically Nonlinear 2-D Steel Frame Structures with Fixity Factors
and Mass modeled as Triangular Fuzzy Numbers

m55 64  35  14 s2  2 s22   4 s12  32  50 s2  32 s22   the response surface method (RSM) and the differential
(7d) evolution algorithm (DE).
 4 s1 196  328s2  55s22 

m66 4 L2 s22  32  31s2  8s12  (7e)

3. Two algorithms for fuzzy structural
 
analysis
m32 2 Ls1 32  7  5s2  s 2
2   s  64  86s
1 2  25s 2
2
(7f)

1120  56 s2  128s22  2 s1  28  184 s2  5s22  

3.1. Response Surface Method (RSM)
m52
(7g) In the statistical theory, surrogate models are often
 2 s12  64  5s2  41s22  used including polynomial regression model, Kriging
model, radial basis function. In this paper, to determine
m62 Ls2 128s2  392  2 s1  50  19 s2   s12  64  55 s2   (7h) the critical loads, a complete quadratic polynomial
regression model is used as surrogate model, in which all
m53 
Ls1 4  98  25s2  16 s22   s1 128  38s2  55s22   (7i) variables are standardized and assumed to be uncorrelated
[5]

m63 L2 s1s2  64 s2  124  s1  64  31s2   (7j) n n 1 n

y  X  a0  ¦ ai X i  ¦ a X X  ¦a X
ij i j ii i
2 (8)
i 1,i  j
2 Ls2  224  64 s2  s12  32  25s2   2 s1  80  43s2   (7k)
i 1 i 1
m65

where Xi are the standardized fuzzy variables; a0 =

y(X=0), and ai, aij are the unknown coef¿cients which
in which P is the axial force of each element,
determined by the method of least squares; y(X)
si=Lki/(3EI+Lki) denote the fixity factor of semi – rigid
represents the surrogate function of natural frequency and
connection at the boundaries (i=1,2). In steel structures, uncertain structural parameters are assumed as symmetric
the common connections can be defined by linguistic triangular fuzzy numbers, xi = (a,l,l)LR. The standardized
terms as shown in Fig. 2. Eleven linguistic terms are fuzzy variables Xi is de¿ned as
assigned numbers from 0 to 10. These include 0-Ideal
Hinged (Absolutely Hinged), 1-Very Hinged (e.g. single xi  a (9)
Xi
web angle), 2-Almost Hinged (e.g. single web plate), 3-
Fairly Hinged (e.g. double web angle), 4-More and Less
Hinged (e.g. header plate), 5-Half Rigid-Half Hinged
(e.g. top & seat angle), 6-More and Less Rigid (e.g. top
plate & seat angle), 7-Fairly Rigid (e.g. top & seat plate),
8-Almost Rigid (e.g. end plate), 9-Very Rigid (e.g. t-stub
& web angle), 10-Ideal Rigid (Absolutely Rigid) [1].
P (si )
1
si 0 1 2 3 4 5
0.15 0.25 0.35
0 0.1 0.2 0.3 0.4 0.5
Fig. 2. Memberships of the fuzzy fixity factors.
In the classical finite element method (FEM), the
values of Zj are determined by solving the eigenvalue
problem, which is easy to implement by using function
eig(K+KG,M) integrated to already-available analysis
tools in the MATLAB. In Eq. (1), when fixity factors of
connections and masses are given by fuzzy numbers, the
natural frequencies are also the fuzzy numbers. The fuzzy
natural frequencies are determined by the fuzzy finite
element method (FFEM) using the D-cut strategy with the
optimization approaches. FFEM is an extension of FEM
in the case that the input quantities in the FEM are
modeled as fuzzy numbers. In this study, two
optimization approaches are presented in the next section:
 l / 3
For the above de¿nition, the original fuzzy variables
xi = (a,l,l)LR are transformed to standardized fuzzy
variables Xi = (0,3,3)LR.
To complete the surrogate polynomial functions of
Eq. (8), all coef¿cients ai, aij shall be determined by a
¿tting procedure, which minimizes the difference (error)
between the outputs of surrogate function and the outputs
of classical ¿nite element model. Normally, some
experiments with deterministic input data are carried out
6 7 8 9 10 and the best fitting function can be obtained by
minimizing the sum of the square errors from the given
si
0.65 0.75 0.85
output data. In RSM, with the number of experiments not
0.6 0.7 0.8 0.9 1
too large, and in fact, maximum, minimum responses
usually occur on the surface of the cube, the
face-centered cube design, and the Box-Behnken designs
are often used [10].
To select the suitable design, the quality of the
response surface is assessed by error estimation. The
most prominent methods are split sample,
cross-validation and bootstrapping, in which the split
sample and the cross-validation are easy to use [11]. In
this study, the leave-one-out cross-validation is applied,
in which, each response point is tested once and trained k
í 2 times (since the center point has been used to
determine a0) with k is number of the Box-Behnken
designs. The error estimation of jth design (using X(j) as
the test set) is determined by the formulas
211
 Tran Thanh Viet, Vu Quoc Anh and Le Xuan Huynh

better objective function value then z becomes a member

GSE j y j  yˆ j 
( j) 2
o min
where GSEj - the square error of jth design; yj - output
value at X(j), determined by classical FEM; yˆ (j j ) -
estimated value at X(j) design of jth design.
3.2. Į – level optimization using Differential Evolution
(DE)
For fuzzy structural analysis, the Į-level optimization
is known as a general approach in which all the fuzzy
inputs are discretized by the intervals that are equal
Į-levels. The output intervals are then searched by the
optimization algorithms. The optimization process is
implemented directly by the finite element model and the
goal function is evaluated many times in order to reach to
an acceptable value. In this study, the solution procedure
is proposed by combining the Differential Evolution (DE)
with the Į-level optimization. DE, which is a
population-based optimizer, is suggested by Storn and
Price in 1995 [8]. The DE algorithm has shown to be
better than the genetic algorithm (GA) and is simple and
easy to use. The basic procedure of DE is described as
following [6-9].
For an objective function f(x), we want to seek the
global optima of f(x) over a continuous space domain:
x=^xi`, xi  [xi,min , xi,max], i = 1,2,…n.
For each generation G, a population of NP parameter
vector xk(G), k = 1,2,…,NP, is utilized. The initial
population is generated as
(10) of the next generation (G+1); otherwise, the old value
xk(G) is retained.
4. Numerical illustration
The example is considered by analysis a twenty –
story, four – bay planar steel frame structural system as
shown in Fig. 3. The elastic modulus E = 210E+06kN/m2,
the mass per unit volume of the columns is m1, the mass
per unit volume of the beams is m2 (included load dead
from slab), the fixity factor at column base is s1, and the
fixity factor at the ends of beams is s2. The axial force P
is calculated from the weight per unit volume. The section
properties used for analysis of the frame are shown in
Table 1. Five fuzzy cases for analysis of fuzzy natural
frequencies are considered as in Table 2. The fixity
factors in case 1 are symmetric triangular fuzzy numbers,
so the fuzzy natural frequencies are calculated with two
different techniques. Since the fixity factors are
non-symmetric triangular fuzzy numbers, the fuzzy
natural frequencies in other cases are solved by using
differential evolution algorithm.
W24x250
W24x306

xk , i  0  xi ,min  rand [0,1].  xi ,max  xi ,min  , i 1,...n (11)

where rand[0,1] is the uniformly distributed random real

value in the interval [0,1].
H = 3.6 x 20 = 72 m
W27x307

For each target vector in a population xk(G),

k=1,2,…NP, a mutant vector y is generated according to

y 
xr1  G   F . xr2  G   xr3  G   (12)

with r1 , r2 , r3 are randomly chosen integers and 1” r1 

r2  r3  k ” NP ; F is a real and constant factor usually
W30x326

chosen in the interval [0,1] to control the ampli¿cation of


the differential variation xr  G   xr  G  .
2 3

In order to increase the diversity of the perturbed
parameter vectors, the crossover is introduced. To this
end, the trial vector z with its elements determined by:
W30x391

­° y j if  rand [0,1] d Cr  or  r i 
zi ® (13)
°̄ xk ,i if  rand [0,1] ! Cr  and  r z i 
9.0 x 4 = 36 m
Here, r is randomly chosen integer in the interval [1,n];
Fig. 3. Twenty – story, four – bay elastic semi-rigid 2-D steel
Cr is use-defined crossover constant in the interval [0,1].
frame.
The new vector z is then compared to xk(G). If z yields

212
 Fuzzy Natural Frequencies Analysis of Goemetrically Nonlinear 2-D Steel Frame Structures with Fixity Factors
and Mass modeled as Triangular Fuzzy Numbers

Table 4. Coef¿cients of surrogate functions for the first two

natural frequencies
Table 1. Section properties used for analysis of steel frame
Coef¿cients Z1 Z2
Cross–section Moment of
Member Section a0 3.2474 9.8952
area, A (m2) inertia, I (m4)
a1 0.00282792 0.00976125
Col. (1st - 4th Fl.) W30x391 7.35E-02 8.616E-03
a2 0.09371500 0.27115556
th th
Col. (5 - 8 Fl.) W30x326 6.17E-02 6.993E-03
a12 0.00007611 0.00030583
Col.(9th -14th Fl.) W27x307 5.82E-02 5.453E-03
a11 -0.00008764 -0.00029986
Col.(15th-20th Fl.) W24x306 5.79E-02 4.454E-03
a22 0.00005194 0.00021472
Beam (1st-20th Fl.) W24x250 4.74E-02 3.534E-03

Table 5. The intervals of the natural frequencies by RSM-case 1

D-cut Z1(rad/s) Z2(rad/s)
Table 2. Cases for analysis of the fuzzy natural frequencies
D=1 [3.2474; 3.2474] [9.8952; 9.8952]
Fuzzy fixity Mass per unit volume
D=0.8 [3.1895; 3.3053] [9.7268; 10.0638]
factors (T/m3)
Case D=0.6 [3.1317; 3.3633] [9.5588; 10.2326]
~
s1 ~
s2 m 1 m 2 D=0.4 [3.0740; 3.4213] [9.3912; 10.4016]
D=0.2 [3.0163; 3.4793] [9.2240; 10.5707]
Case 1 8 7 (7.85, 0, 0) (50, 0, 0) D=0 [2.9588; 3.5374] [9.0571; 10.7399]
Case 2a 1 9 (7.85, 0, 0) (50, 0, 0)
Case 2b 1 9 (7.85,0.785,0.785) (50, 5, 5)
Case 3a 9 9 (7.85, 0, 0) (50, 0, 0) 4.2. Solving by DE
Case 3b 9 9 (7.85,0.785,0.785) (50, 5, 5) The output intervals of the first two natural
frequencies are calculated by using DE programmed in
MATLAB for all cases. The parameters for DE are: the
NP=50, F=0.5, Cr= 0.9. The optimization process is
4.1. Solving by RSM stopped after 30 iterations. The results of the natural
In case 1, the Box-Behnken designs with two input frequencies intervals are shown in Tables 6-10. It is found
fuzzy variables ( ~ s1 and ~s2 ) are presented in Table 3. In from the analysis results that the natural frequencies in
this table, the first two natural frequencies are calculated cases 2a and 2b (with ~ s1 is Very Hinged) are smaller than
by using the classical FEM programmed in MATLAB. the natural frequencies in cases 3a and 3b (with ~
s1 is Very
The results of the coefficients of the surrogate functions
for the natural frequencies are shown in Table 4. The Rigid). Fig.4 and Fig.5 show the comparison of the fuzzy
intervals of the natural frequencies are shown in Table 5. natural frequencies of case 1 by two differential
techniques. Figs.6-9 show the membership functions of
the fuzzy natural frequencies for different cases of
Table 3. The Box – Behnken designs and the first two natural
analysis.
frequencies
Table 6. The intervals of the natural frequencies by DE-case 1
Z1 Z2
No. x1=s1 X1 x2=s2 X2 D-cut Z1(rad/s) Z2(rad/s)
(rad/s) (rad/s)
D=1 [3.2474; 3.2474] [9.8952; 9.8952]
0 0.80 0 0.75 0 3.2474 9.8952
D=0.8 [3.1896; 3.3052] [9.7271; 10.0635]
1 0.80 0 0.85 3 3.5290 10.7107
D=0.6 [3.1318; 3.3631] [9.5591; 10.2320]
2 0.80 0 0.65 -3 2.9667 9.0835
D=0.4 [3.0740; 3.4210] [9.3910; 10.4009]
3 0.90 3 0.75 0 3.2551 9.9217
D=0.2 [3.0161; 3.4791] [9.2229; 10.5701]
4 0.90 3 0.85 3 3.5374 10.7399
D=0 [2.9581; 3.5374] [9.0545; 10.7399]
5 0.90 3 0.65 -3 2.9737 9.1076
6 0.70 -3 0.75 0 3.2381 9.8634
7 0.70 -3 0.85 3 3.5190 10.6758
8 0.70 -3 0.65 -3 2.9581 9.0545

213
 Tran Thanh Viet, Vu Quoc Anh and Le Xuan Huynh

Table 7. The intervals of the natural frequencies by DE-case 2a

D-cut Z1(rad/s) Z2(rad/s)
D=1 [3.6786; 3.6786] [11.0245; 11.0245]
D=0.8 [3.6219; 3.7015] [10.8653; 11.0940]
D=0.6 [3.5655; 3.7219] [10.7069; 11.1570]
D=0.4 [3.5095; 3.7403] [10.5491; 11.2143]
D=0.2 [3.4536; 3.7568] [10.3919; 11.2667]
D=0 [3.3980; 3.7717] [10.2352; 11.3148]
Fig. 5. The fuzzy 2nd natural frequencies solved by using RSM and
DE – case 1
Table 8. The intervals of the natural frequencies by DE-case 2b
D-cut Z1(rad/s) Z2(rad/s)
D=1 [3.6786; 3.6786] [11.0245; 11.0245]
D=0.8 [3.5856; 3.7397] [10.7567; 11.2081]
D=0.6 [3.4950; 3.8000] [10.4959; 11.3900]
D=0.4 [3.4067; 3.8597] [10.2415; 11.5711]
D=0.2 [3.3205; 3.9193] [9.9934; 11.7522]
D=0 [3.2364; 3.9790] [9.7511; 11.9341]

Fig. 6. The fuzzy 1st natural frequencies solved by using DE – case

Table 9. The intervals of the natural frequencies by DE-case 3a 2a and case 2b
D-cut Z1(rad/s) Z2(rad/s)
D=1 [3.9777; 3.9777] [12.0239; 12.0239]
D=0.8 [3.9176; 3.9777] [11.8479; 12.0239]
D=0.6 [3.8577; 3.9777] [11.6729; 12.0239]
D=0.4 [3.7982; 3.9777] [11.4989; 12.0239]
D=0.2 [3.7389; 3.9777] [11.3258; 12.0239]
D=0 [3.6799; 3.9777] [11.1535; 12.0239]

Fig. 7. The fuzzy 2nd natural frequencies solved by using DE –

Table 10. The intervals of the natural frequencies cases 2a and 2b
by DE-case 3b
D-cut Z1(rad/s) Z2(rad/s)
D=1 [3.9777; 3.9777] [12.0239; 12.0239]
D=0.8 [3.8784; 4.0187] [11.7299; 12.1472]
D=0.6 [3.7817; 4.0609] [11.4438; 12.2743]
D=0.4 [3.6874; 4.1045] [11.1650; 12.4054]
D=0.2 [3.5955; 4.1494] [10.8933; 12.5408]
D=0 [3.5058; 4.1958] [10.6283; 12.6806]
Fig. 8. The fuzzy 1st natural frequencies solved by using DE –
cases 3a and 3b

Fig. 9. The fuzzy 2nd natural frequencies solved by using DE –

cases 3a and 3b
Fig. 4. The fuzzy 1st natural frequencies solved by using RSM and
DE – case 1

214
 Fuzzy Natural Frequencies Analysis of Goemetrically Nonlinear 2-D Steel Frame Structures with Fixity Factors
and Mass modeled as Triangular Fuzzy Numbers
5. Conclusion
From the results of the numerical examples, the
following conclusions are obtained:
1. The fuzzy finite element analysis based on the
response surface method, the result is obtained the fuzzy
natural frequencies by using the response surface with the
surrogate function is the complete quadratic polynomial.
This approach is suitable for the fuzzy input variables
modeled as symmetric triangles. The benefit in the
application of this methodology is demonstrated through
an analysis of the twenty – story, four – bay geometrically
nonlinear planar steel frames with a lot of elements and
fuzzy variables. This case is also carried out by other
approach using the Differential Evolution (DE) in
combination with the Į – level optimization, and the
comparison of the fuzzy natural frequencies between two
solution approaches gives a good agreement.
2. From the accuracy of the result implemented by
using DE in the case 1, this paper is extended for the
other cases in which the fuzzy input variables modeled as
any triangles. The results are obtained by using this
approach show that the Į – level optimization algorithm
in combination with DE is more advantageous than the
RSM in combination with GA, in which the finite
element method is applied to geometrically nonlinear
elastic semi-rigid connection with
multi-degree-of-freedom systems and non-symmetric
triangular fuzzy variables.
215
References
[1] Ali Keyhani, Seyed Mohammad Reza Shahabi, Fuzzy
connections in structural analysis, ISSN 1392 – 1207,
MECHANIKA, Volume 18(4), (2012), pp. 380-386
[2] Vu Quoc Anh, and Nghiem Manh Hien, Geometric
nonlinear vibration analysis of steel frames with
semi-rigid connections and rigid-zones, Vietnam
Journal of Mechanics, VAST, Vol. 25, No. 2 (2003), pp.
122-128
[3] Amar Khennane, Introduction to Finite Element
Analysis Using MATLAB and Abaqus, CRS Press,
Taylor and Francis Group, New York, (2013)
[4] Angel L. Morales, Jam A. Rongong, and Neil D. Sims,
A fuzzy finite element method programmed in Matlab
for the analysis of uncertain control systems of
structures, MATLAB – A Ubiquitous Tool for Practice
Engineer, (2011), pp. 291-304
[5] Nguyen Hung Tuan, Le Xuan Huynh, Pham Hoang
Anh, A fuzzy finite element algorithm based on
response surface method for free vibration analysis of
structure, Vietnam Journal of Mechanics, VAST, Vol.
37, No. 1 (2015), pp. 17-27
[6] Anh Hoang Pham, Thanh Xuan Nguyen, and Hung Van
Nguyen, Fuzzy Structural Analysis Using Improved
Differential Evolution Optimization, International
Conference on Engineering Mechanic and Automation
(ICEMA 3) Hanoi, (October 15-16, 2014), pp. 492-498
[7] Efrén Mezura-Montes, Margarita Reyes-Sierra, and
Carlos A. Coello Coello, Multi-Objective
Optimization using Di erential Evolution: A Survey
of the State-of-the-Art, Soft Computing with
Applications (SCA), Volume 1, (Number 1, June 2013)
[8] Storn, R., and Price, K, Differential Evolution – A
Simple and Efficient Adaptive Scheme for Global
Optimization over Continuous Spaces, International
Computer Science Institute, Berkeley, (1995)
[9] Storn, R., and Price, K, Differential Evolution – A
Simple and Efficient Heuristic for Global Optimization
over Continuous Spaces, Journal of Global
Optimization 11, Netherlands, (1997), pp. 341-359
[10] R. L. Mason, R. F. Gunst, and J. L. Hess, Statistical
design and analysis of experiments: With applications
to engineering and science, Vol. 474. JohnWiley &
Sons, (2003)
[11] N. V. Queipo, R. T. Haftka, W. Shyy, T. Goel, R.
Vaidyanathan, and P. K. Tucker, Surrogate-based
analysis and optimization, Progress in Aerospace
Sciences, 41, (1), (2005), pp. 1-28.