Professional Documents
Culture Documents
TRUSS OPTIMIZATION
CONSIDERING WEIGHT AND
SAFETY MARGIN
5.1
Overview
In this chapter, the optimization problem in chapter 4 is reformulated from another standpoint
that if designers expect not only a minimum weight structural design but also a structural
design with maximum safety margin towards accidental loading conditions.
In section 5.2, the optimization problem in the section 4.3 of chapter 4 is reformulated as
multi-objective optimization problem.
In section 5.3, the newly formulated problem is applied to the analysis models solved in
chapter 4, and the results of two optimization problems are discussed.
87
5.2
In the introduction of chapter 1, we have learned that structures usually collapse because of
the lack of a safety margin against accidental load conditions such as catastrophic earthquake,
terrible typhoon, terrorist attack, mud-rock, or other accidental loads. Therefore, it is expected
that structures can be designed with less weight but higher resistance capactity against a
possibly accidental load on condition that the basic design requirements under ordinary load
conditions are satised. However, these two objectives conict with one another because less
weight may usually lead to lower structural performance, in other words, a lower safety margin.
In this chapter, a multi-objective optimization scheme for truss structures is proposed in an
eort to minimize their total mass from an economical standpoint while at the same time
maximizing their safety margin to withstand a possibly accidental load from a safety point
view. The safety margin of a structure to withstand an accidental load in this study is dened
by its excess collapse load factor in relation to the load factor of an accidental load. State
dierently, the larger the collapse load factor of the structure is, the higher the safety storage
is considered to be. Hence, the collapse load factor is directly dened as one of the objective
functions. The optimization problem for minimizing total structural weight and maximizing
collapse load factor with topology and sizing as design variables, subject to several practical
constraints, can be formulated as follows:
minimize
subject to
f (x, A) =
f1 (x, A) = W (x, A)
f2 (x, A) = 1/cr (x, A)
(5.1)
g(x, A) 0
where W (x, A) is the total weight; x is the vector of topology; A is the vector of cross section
area of members; cr (x, A)=collapse load factor under an accidental load; and g(x, A) is the
vector of functions of constraints. The total mass W (x, A) can be further specied as
W (x, A) =
m
i
ci li ai +
n
wj
(5.2)
where c=the compensate coecient of members due to the shrinking of cross section at both
their ends, as discribed in chapter 5; i = the density of material of member i; li =the calculated
length of member i; ai = the cross section area of member i; wj = globe mass of joint j; m= the
number of members; n= the number of nodes.
The same penalty function in chapter 4 is adopted:
P = (stress )c1 (disp )c2 (collapse )c3 angle (cross )k
(5.3)
89
where the terms stress , disp , collapse , angle and cross are penalty items associated with violation
of constraints of stress and displacement under ordinary loading conditions in the rst design,
collapse load factor under accidental loading conditions in the second design, interference between members at joint, and member-intersecting, respectively. These penalty items will be
explained explicitly in later sections. Here, c1 , c2 , and c3 , including the later explained c4 , and
c5 , are their corresponding weighting coecients. Among these constraints, the constraint of
interference between members at joint is easier to be satised according to authors experience.
Therefore, c4 is not set as a power of the term angle , only multiplied in it (later explained in
equation (5.12)). Also, as c5 is actually the bottom of (c5 )k (later explained equation (5.15)),
it is also not set to be a power of the value. All these weighting coecients need to be given
certain values before starting an optimization. Adjustment of their values can help to guide a
particular optimization problem towards convergence if it traps in a local solution. In the case
of an unstable structure, the penalty function will be given an innity value.
With this penalty function above, the original constrained optimization problem can be
transformed to an unconstrained optimization problem as follows:
minimize W (x, A) P
(5.4)
The optimization method used for solving this optimization problem is SPEA2 which is one of
the multi-objective genetic algorithms as introduced in table 2.2, chapter 2. For details, please
refer to appendix A.
5.2.1
Constraint of Stress
In the rst design stage, the elastic design allows no member stress to exceed the allowable
stress. According to the design code of AIJ64) , the allowable stresses for tension members and
compression members are dened as:
F
(5.5)
ft =
1.5
fc =
1 0.4(/)2
0.277
F
(/)2
if
(5.6)
if
>
where F = the characteristic value of strength of steel; = the slenderness ratio; = the critical
slenderness ratio; and 0.277 are the safety coecients. The penalty on stress for one truss
|i |
if |i | > lim
(5.7)
stressi = lim
1 otherwise
where i = the member stress under ordinary loading conditions in the rst desgin stage, and
lim represents the allowable stress, determined by the stress type of i . For the penalty of all
members, the stress in equation (5.3) therefore becomes
m
stress =
stressi
(5.8)
i=1
5.2.2
Constraint of Displacement
The maximum nodal displacement in the rst design stage needs to satisfy a demand for a
certain tolerance on the serviceability limit; the penalty on one node dispj is calculated as:
|dj |
1 otherwise
(5.9)
where dlim is the displacement tolerance set by the designer or related specications, and dj is
the nodal displacement in X, Y , or Z direction, calculated by structural analysis as shown in
gure 3.29. Therefore, the penalty iterm associated with displacement for the whole structure
in equation (5.3) is
n
disp =
dispj
(5.10)
j=1
5.2.3
To avoid collapse under accidental loading conditions in the second design stage, the collapse
load factor cr must be larger than the load factor a of an accidental load. The penalty item
collapse in equation (5.3) is dened as follows:
collapse =
cr
1
if cr < a
otherwise
(5.11)
5.2.4
91
For any two members connected to a joint, interference must be avoided from a construction
point of view. In this work, all prepared globe sizes for selection are tested for connection
without interference (as shown in gure 5.1), and the smallest one without interference will
be chosen as the nal choice. If the maximum globe size is equipped and two members still
collide with each other, the angle is to be penalized. The penalty item of a angle between two
members is
lim
c4
if t < lim
t
(5.12)
anglet =
1
otherwise
where lim is the smallest angle at which the maximum globe size is able to connect the two
members; t is the actual angle of two members at the joint and c4 is the weight coecient
aforementioned (must be larger than 1). For a joint connecting more than two members, the
smallest globe size among the nal choices is to be used. Therefore, the penalty term of angle
in equation (5.3) is dened as:
cj
n
angle =
anglet
(5.13)
j=1 t=1
where n is the number of nodes, and sj is the independent couples of member at node j
calculated by
(5.14)
cj = ej (ej 1)/2
where ej is the number of members connected at node j.
Slice face
Interference check
Globe
Wrapper
Bolt
Pipe
Slip
5.2.5
The existence of intersecting members is also considered infeasible from the construction standpoint. Thus, the distance between two members is required to be larger than the summation of
their radii. The penalty item regarding this constraint is dened as (cross )k in equation (5.3),
where k is the number of pairs of intersecting members and cross is a constant dened by the
following equation:
c5 if d < RA + RB
(5.15)
cross =
1 otherwise
where RA and RB are the radius of two members, respectively; d is the distance between the
two members as shown in gure 5.2; and c5 is the weighting coecient aforementioned (must
be larger than 1).
A
5.2.6
Optimization Method
The SPEA2 is utilized as the optimization method for solving this multi-objective optimization
problem. For detailed information about SPEA2, please refer to Appendix A. Figure 5.3 shows
the owchart of SPEA2 for handling present optimization problem, in which structural anlysis
is needed at the operations of Archive update and Fitness assignment. Structural analysis
is what aformentioned in gure 3.29, chapter 3.
93
t=t+1
Fitness assignment
Population (t+1)
Population(t)
Individual 1
Individual 2
Individual 1
(n-m)
Crossover
Individual n-m
Individual n-m+1
Mutation
Selection
Reproduction
Individual n
Archive Update
Individual n
Archive (t)
Individual 1
Individual 2
Archive (t+1)
Individual 1
Individual 2
Output
Individual m
Individual m
t=t+1
5.3
5.3.1
Numerical Analysis
Double Layer Truss Structure
The numerical example solves the same model in section 4.4.1, which is a double-layer truss
roof supported by four pinned supports at corners and two roller supports at middle span in
y direction (shown in gure 5.4). It is of 25 m in length, evenly divided into 5 units, 20 m
in width, evenly divided into 4 units, and 2.4 m in depth. The design constants used are,
modulus of elasticity, E = 205GP a, material density, = 78.5kN/m3 , characteristic value of
strength of steel, F = 235M P a, and critical slenderness ratio, = 120. The loading acting on
the model consists of a vertical uniform load of 0.5kN/m2 , which is considered as an ordinary
loading condition in the rst design stage, and a vertical seismic load of 20m/s2 in terms of
acceleration, which is considered as an accidental loading condition in the second design stage.
Actually, the seismic load is transformed into concentrated loads when performing elasto-plastic
analysis, and for an easy understanding its load factor is dened as 20.0 in accordance with
the acceleration value. Tetrahedron is chosen as the base stable unit in topology generation.
1
31
30
7
6
35
40
46
25
16
43
44
48
27
26
17
22
21
47
11
39
38
42
20
34
10
15
41
45
24
37
14
19
18
33
9
36
13
12
32
45 m=20
Pinned support
0
23
49
28
29
55 m=25 m
X
Top View
Big seismic load (20 m/s2 )
2.4 m
55 m=25 m
X
Side View
95
Table 5.2: Constraints
Stress
Allowable stress
Displacement(-Z)
125 mm
Collapse load factor 20
Member intersecting Not allowed
Member interference Not allowed
120
30
2000
0.80
0.05
No.
Diameter
(mm)
Thickness
(mm)
Area
(cm2 )
No.
Diameter
(mm)
Thickness
(mm)
Area
(cm2 )
No.
Radius
(mm)
Mass
(kg)
34.0
2.3
2.291
11
101.6
4.0
12.26
25.0
0.5
42.7
2.5
3.157
12
101.6
5.0
15.17
42.5
2.5
48.6
3.2
4.564
13
114.3
4.5
15.52
55.0
9.0
60.5
3.2
5.760
14
139.8
4.0
17.07
65.0
3.2
76.3
2.8
6.465
15
139.8
4.5
19.13
75.0
13.1
76.3
3.2
7.439
16
165.2
4.5
22.72
90.0
24.0
89.1
3.2
8.636
17
165.2
5.0
25.16
100.0
32.9
101.6
3.2
9.892
18
190.7
4.5
26.32
110.0
43.9
114.3
3.2
11.17
19
216.3
4.5
29.94
130.0
72.4
10
114.3
3.5
12.18
20
165.2
6.0
30.01
10
150.0
111.2
Analysis Result
Figure 5.6 shows the evolutionary history of the individuals in the archive. Clearly, these
individuals in archive move towards the lower left of the graphics as the generation grows. It
is noted that the non-dominated individuals only occupy the minority of the archive at the
beginning but completely ll it after around 1500 generations. The average weight and collapse
load factor of archive in several generations are presented in table 5.14, from which one can
learn that the average mass has been decreased whereas the average collapse load factor has
been increased. Subtracting the accidental load factor 20, their safety storage has increased
from 20 to 120.
Generation 0
Generation 20
Generation 40
Generation 80
Generation 200
Generation 300
Generation 400
Generation 500
Generation 1000
Generation 1500
Generation 1700
Generation 2000
97
Table 5.5: Average weight and collapse load factor of archive in dierent generations
Generation No.Average weightAverage collapseSeismic
(ton)
load factor
0
27.4
46.9
20.7
43.6
20
15.9
62.1
40
14.2
68.4
80
11.4
75.4
200
10.8
70.7
300
8.6
104.5
400
500
8.9
113.7
9.5
131.9
1000
8.2
140.2
1500
8.1
140.2
1700
8.1
140.4
2000
26.9
23.6
42.1
48.4
55.4
50.7
84.5
93.7
111.9
120.2
120.2
120.4
99
139.84.5
139.84.0
114.33.2
76.32.8
-57.6N/mm2
Compression
86.4N/mm2
Tension
-25.7mm
Displacement ( Z direction )
Figure 5.9: Stress distribution of solution No.0 Figure 5.10: Displacement distribution of sounder uniform load
lution No.0 under uniform load
STEP1 (c176)
STEP2 (c2=92)
Buckled member
Yielded member
STEP3 ( cr97)
Figure 5.11: Collapse process of solution No.0 (collapse load factor: 97)
101
165.25.0
139.84.5
114.34.5
89.13.2
-57.6N/mm2
Compression
86.4N/mm2
Tension
-25.7mm
Displacement ( Z direction )
Figure 5.13: Stress distribution of solution Figure 5.14: Displacement distribution of solution No.14 under uniform load
No.14 under uniform load
STEP1 ( c1=64)
STEP2 ( c2=109)
STEP4 (cr=146.9)
STEP3 ( c3=141.5)
Buckled member
Yielded member
Figure 5.15: Collapse process of solution No.14 (collapse load factor: 146.9)
103
165.26.0
165.2 4.5
114.34.5
89.13.2
-37.6N/mm2
Compression
63.7N/mm2
Tension
-18.7 mm
Displacement ( Z direction )
Figure 5.17: Stress distribution of solution Figure 5.18: Displacement distribution of soNo.29 under uniform load
lution No.29 under uniform load
STEP1 (c1=71)
STEP2 (c2=104.7)
STEP3 (c3=132.9)
STEP4 ( c4=134.0)
Buckled member
Yielded member
STEP5 (c5=160.0)
Figure 5.19: Collapse process of solution No.29 (collapse load factor: 160.0)
5.3.2
105
ax
is
2
Se
m
m
et
ry
ry
et
m
m
Se
4
is
ax
Semmetry axis 3
The same dome truss structure in section 4.4.2 is solved here. It is a double-layer dome structure
(shown in gure 5.20) with a diameter of 20 m at circular bottom, 3 m in height, and 0.5 m
in depth. Design variables are truss topology and cross-sectional areas of members. All the
topologies generated in the rst generation of genetic algorithm belong to a ground structure as
shown in gure 5.21, and four-face symmetry characteristic is considered. Four pinned support
nodes painted black in gure 5.20 as well as the peak node are nodes must be connected in
topology generation. Instead of tetrahedron, in the example, triangle is used as the base stable
unit for improving the diversity. The loading condition consists of a vertically concentrated
load at the peak as the ordinary loading condition in the rst design stage and a seismic load of
10 m/s2 in terms of acceleration as the accidental loading condition in the second design stage.
Similary, GA parameters, constraints, cross sections for selection, globes for connectiong, and
weight coecients are completely the same as that in single objective optimizatin problem.
The only dierence is the archive number in SPEA2 parameters, compared to the elite number
in GA parameters.
Seismic load
10.0 m / s 2
40 kN
Dead load
0.5 m
3m
18(19)
7
6
1
Y
0
13
12
15 17
16
14
11
10
9
5
8
3
Semmetry axis 1
Z
X
Side view
20 m
Top view
80
20
1000
0.80
0.05
Diameter
(mm)
34.0
42.7
42.7
48.6
48.6
48.6
48.6
60.5
60.5
60.5
76.3
76.3
76.3
89.1
89.1
101.6
101.6
101.6
114.3
114.3
114.3
Thickness
(mm)
2.3
2.3
2.5
2.3
2.5
2.8
3.2
2.3
3.2
4.0
2.8
3.2
4.0
2.8
3.2
3.2
4.0
5.0
3.2
3.5
4.5
No.
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
Diameter
(mm)
139.8
139.8
139.8
139.8
165.2
165.2
165.2
165.2
190.7
190.7
190.7
190.7
190.7
216.3
216.3
216.3
216.3
216.3
267.4
267.4
Thickness
(mm)
3.6
4.0
4.5
6.0
4.5
5.0
6.0
7.1
4.5
5.3
6.0
7.0
8.2
4.5
5.8
6.0
7.0
8.0
6.0
6.6
107
Analysis Result
Figure 5.22 shows the evolutionary history of the individuals in the archive. Clearly, these
individuals in archive move towards the lower left of the graphics as the generation grows. It
is noted that the non-dominated individuals only occupy the minority of the archive at the
beginning but completely ll it after around 200 generations. The average weight and collapse
load factor of archive in several generations are presented in table 5.14, from which one can learn
that the average mass has been decreased whereas the average collapse load factor has been
increased. It is found that the average collapse load factor did not gradually grow up during
the optimization process; it was uctuating up and down towards increase. This is probably
because that the dome truss model is of fewer unknowns than the double layer truss model and
slight change in topology would cause great change in collapse load factor. Nevertheless, the
average collapse load factor of archive has been increased from 143.4 to 503.0. Subtracting the
accidental load factor 10, their safety storage has increased from 133.4 to 493.0.
Generation 0
Generation 20
Generation 40
Generation 80
Generation 100
Generation 200
Generation 500
Generation 800
Generation 1000
133.4
27.8
25.0
167.3
490.6
542.5
516.1
508.4
493.0
109
are demonstrated from gures 5.24 to 5.33. It is found that these three solutions are completely
identical in topology but dier in cross sections distribution. Bigger cross sections can be
observed in solution with lager solution numbers, which exactly makes them higher collapse
load factors. Focusing on their collapse mechanisms, it is found that most of their collapse
mechanisms are comprised of buckled members except solution No.9.
Overall, since both objective functions signicantly improved both in solving a double layer
and dome truss structure, it can be said that the designed multi-objective optimization problem
is successfully implemented.
Table 5.15: Details of Pareto-optimal solutions in the Pareto-optimal front
Pareto
Fitness weightCollapsePenalty Penalty
Penalty of Penalty of Penalty of
solution No.
(ton) load
of
of
member
member collapse
factor stress displacementinterferenceintersectingload factor
0
0.425607 1.4
78.5
1
1
1
1
1
0.448214 1.6
103.3
1
1
1
1
1
1
0.469633 1.7
115.7
1
1
1
1
1
2
0.446677 1.8
229.1
1
1
1
1
1
3
0.41718 2.0
282.4
1
1
1
1
1
4
0.390563 2.4
298.4
1
1
1
1
1
5
6
0.380942 2.6
350.3
1
1
1
1
1
0.398031 2.8
371.8
1
1
1
1
1
7
0.423582 3.0
379.4
1
1
1
1
1
8
0.443681 3.3
518.4
1
1
1
1
1
9
0.423582 3.4
539.1
1
1
1
1
1
10
0.346061 4.2
677.2
1
1
1
1
1
11
0.350698 6.7
701.5
1
1
1
1
1
12
0.361167 6.8
708.1
1
1
1
1
1
13
0.369836 6.9
711.8
1
1
1
1
1
14
0.399767 7.6
746.7
1
1
1
1
1
15
0.435069 7.8
774.4
1
1
1
1
1
16
17
0.435069 8.1
792.1
1
1
1
1
1
0.413578 8.2
825.5
1
1
1
1
1
18
0.388207 8.4
856.6
1
1
1
1
1
19
18
15
7
-44.8 N/mm2
Compression
23.27 N/mm2
Tension
-5.5 mm
Displacement ( Z direction )
Figure 5.25: Stress distribution of solution Figure 5.26: Displacement distribution of soNo.0
lution No.0
111
Buckled member
Yielded member
STEP1 (c160.4)
STEP2 (c2=78.5)
18
15
7
Buckled member
Yielded member
STEP1 (c1400)
STEP3 (c3514.5)
STEP2 (c2=442.5)
STEP4 (c4=518.4)
113
267.46.6
267.46.0
114.33.2
18
15
7
-27.1 N/mm2
Compression
0.9N/mm2
Tension
-2.5 mm
Displacement ( Z direction )
Figure 5.31: Stress distribution of solution Figure 5.32: Displacement distribution of soNo.19
lution No.19
Buckled member
Yielded member
STEP1 (c1739.2)
STEP3 (c4850.5)
STEP2 (c2=841.3)
STEP4 (c4=856.6)
5.3.3
115
To compare the optimization results of single and multi-objective optimization problems respectively conducted in chapters 4 and 5, we plot the optimal solutions of single objective
optimization problem together with Pareto-optimal fronts in gure 5.34 and 5.35. It is found
that in solving double layer truss the optimal solution of single optimization problem is dominated by 4 Pareto-optimal solutions of the multi-objective problem, while in solving dome
truss structure the optimal solution of single optimization problem is not dominated any of the
Pareto-optimal solutions of the multi-objective problem. Nevertheless, in solving both optimization problems, the optimal solution of single objective optimization is close to the lightest
Pareto optimal solution of the Pareto optimal front of multi-objective optimization. Therefore,
it is unable to say whether GA has a superior search performance than SPEA2 or not, vice
versa. This is probably because SPEA2 adopts a special tness assignment scheme compared
the normal tness assignment in GA.
Figure 5.34: Comparison in double layer truss Figure 5.35: Comparison in dome truss structure
structure
5.4
Conclusion
In this chapter, a multi-objective optimization problem was formulated to minimize total structural weight while maximizing the collapse load factor, namely, the safety margin towards
an accidental loading condition. The same numerical examples solved in single optimization
problem was used to test the validity of designed multi-objective optimization problem. The
conclusion of the numerical analysis is that both objective functions were suciently improved
during the optimization process, i.e., the structural weight was decreased whereas the collapse
load factor was increased. This result indicates the successful realization of designed multiobjective optimization problem. And, the obtained Pareto-optimal solutions in Pareto-optimal
front are wildely distributed in the function space, providing a variety of selections of interest
for program user. In addition, we compared the optimal solutions in single and multi-objective
optimization problems. Under the same loading condition and GA parameters, solving double
layer truss concluded that the optimal solution of single optimization problem is dominated by
4 Pareto-optimal solutions of the multi-objective problem, while solving dome truss structure
concluded that the optimal solution of single optimization problem is not dominated any of
the Pareto-optimal solutions of the multi-objective problem. This conclusion indicates that
it is unable to say whether GA has a superior search performance than SPEA2 or not, vice
versa, because SPEA2 adopts a special tness assignment scheme completely dierent from the
normal tness assignment in GA.
Chapter 6
POST-BUCKLING INFLUENCE
6.1
Overview
This chapter is a special chapter, aiming to investigate inuence of the assumptions in the
application of hinge by hinge method to truss structures on collapse load factor. The commercial
software MIDAS is utilized to investigate this inuence.
In section 4.2, the commercial software MIDAS as well as its features in elasto-plastic
analysis is introduced.
In section 4.3, the post buckling behaviour of truss members is described and a new method
is proposed for considering the post buckling behaviour in MIDAS.
In section 4.4, dierent types of truss structure are investigated concerning the post-buckling
inuence.
117
118
6.2
Post-Buckling Hehaviour
In chapter 3, the hinge by hinge method was applied to truss structures based on the assumption
that truss structural members can maintain their buckling stresses until the overall collapse
of the structure. However, it is shown in the literature that the axial stress of an axialload component may decrease after buckling if the axial strain continues increasing.66) This
phenomemon is termed post buckling behaviour. Figure 6.1 shows the experimental behavior
of post-buckling, in which steel pipes with diverse slenderness ratio were axially imposed an
increasing load until their whole collapse. It is clear that the axial compression forces of test
specimens start to decrease after reaching their maximum values and the slopes of their decrease
are inversely proportional to their slenderness ratio; the smaller the slenderness ratio is, the
steeper the slope is. This behavior diers from our assumption in gure 3.18 in chapter 3. Does
this kind of post-buckling behavior have great impact on the collapse load factor? If it has a
great inuence, the collapse load factor calculated by our program will become very unreliable.
Therefore, it is necessary to investigate the inuence of post-buckling on the collapse load when
the hinge by hinge method is applied to truss structures for elasto-plastic analysis.
6.3
6.3.1
119
The commercial software MIDAS/Gen 7.3063) is utilized to investigate this kind of post-buckling
inuence. MIDAS/Gen 7.30 is general structural analysis software developed by the research
center of the Korean steel manufacturer Posco Group with cooperation from related industry. It
is built up based on the Windows operating environment and written in Visual C++ language.
Compared to other structural analysis software, the characteristic of MIDAS is that it divides
the process of modeling, structural analysis, results analysis, and group designating etc. into
a sequence of directory so that not only skillful users but also the beginners can eectively
operate without errors, providing a considerable convenience for users. Figure 6.2 shows part
of its operation interface.
6.3.2
The static incremental analysis in MIDAS is an analytical approach for evaluating buildings
earthquake resistance capability, capable of considering material nonlinearity. The static incremental analysis can be also called pushover analysis, functionally identical with the elastoplastic analysis as introduced in chapter 3. There are two control methods for static incremental
analysis in MIDAS.
120
Load control method
Load
Cs=30%
Cs=50%
Cs=0%
(Limit Point)
Cs=70%
Stable range
2STEP~N STEP:
Dividing the residual load by
arithmetric progression
Cs=80%
Force
Unstable
range
In load control method, the static incremental analysis is carried out by rst designating an approximate collapse load for the analysis model and then dividing the approximate collapse load
by arithmetic progression for load increment. To save computing times, the load is increased to
the load level of 90 of elastic limit at the rst step. Then, from the second step, residual load
are automatically divided into a number of load levels using arithmetic progression for sequent
load increment. A conceptual graph of load increment in this method is shown gure 6.3.
Cs=90%
Displacement
Displacement
STEP 1
Estimate the load level of elastic limit state of the structure, and dene the ratio of load
level to the designated approximate collapse load as load factor1 . The load increased at the
rst step is calculated by
P1 = c1 P
(6.1)
where, c1 is accumulative load factor at step 1; P1 is the load increment at the rst step;
P is the approximate collapse load designated by users.
STEP 2
Accumulative load factors after step 1 are then calculated by arithmetric progression as
1 c1
n1
121
(6.2)
where, i is the number of present step; i 1 is the step prior to present step; n is the number
of total steps; ci is acculumlative load factor at present step. Therefore, the accumulative load
at present step i is expressed as:
Pi = ci P
(6.3)
STEP n
Pn = cn P ; cn = 1.0
(6.4)
Stop criterion
1. Boundary rigidity rate (Cs):
The rate of present structural rigidity to the initial structural rigidity is calculated after
each step. Once the rate reaches boundary rigidity rate set by users or by default value, the
analysis will be stopped (shown in gure 6.4).
2. Boundary story deection angle:
If the story deection angle attains the boundary story deection angle specied by users,
the analysis will be halted.
3. Step number:
Arriving at the nal step for load increment is a stop criterion as well.
122
6.3.3
Element Varieties
In MIDAS, elements of 2D-beam, 3D beam-column, 3D wall and truss can be used in static
incremental analysis. As for space limitation, only elements of 2D-beam, 3D beam-column and
truss will be introduced next.
2D-beam element and 3D-beam-column element
Elements of 2D beam and 3D beam-column can be mathematized by the same method of
dening the relation between nodal forces and displacements, as shown in gure 6.5. Therefore,
the beam and beam-column elements have identical load-displacement relationship, expressed
in equation 6.5 and 6.6. In the case of 2D beam element, the vector of axial force in the
equations becomes 0.
{P }T = {Fx1 , Mx1 , Fy1 , My1 , Fz1 , Mz1 , Fx2 , Mx2 , Fy2 , My2 , Fz2 , Mz2 }
(6.5)
{U }T = {Ux1 , x1 , y1 , y1 , z1 , z1 , Ux2 , x2 , y2 , y2 , z2 , z2 }
(6.6)
Truss element
To model a plastic hinge that happens in a truss structure, a spring, able to bear axial tension
and compression force, is used, as shown in gure 6.6. In practice, signs of springs indicated
in elements are not the exact springs, such as that in gure 3.14 in chapter 3. They are only
ways of the expression of concepts.
6.4
123
6.4.1
Basically, in MIDAS/Gen 7.30, there are two types of plastic hinges available to model plastic
hinges in truss structures; one is named bilinear type hinge, and the other is named FEMA
type hinge. The denition of their load-displacement curve is shown in gures 6.7 and 6.8,
respectively, in which users need to input the yield and buckling strength values. It is fairly
easy to identify that the bilinear hinge type is the same as our assumption, while the FEMA
type is able to consider the axial force reduction after buckling. However, a diagonal denition
of the line between points b and c in gure 6.8 is ineective in MIDAS, which causes a
diculty in simulating the progressively reducing behaviour of post-buckling (shown in gure
6.1). To demonstrate the fact that a plastic hinge with denition of a diagonal line between
Force
Force
a
Ny
Ny : yield strength
Ncr : buckling strength
dy: yield displacement
db : buckling displacement
-db o
Disp
Ncr
dy
Ny : yield strength
Ncr : buckling strength
dy: yield displacement
db : buckling displacement
-c
-db o
dy
-a
Ny
-b
-a
Disp
Ncr
124
Force
a
dy
6dy
2m
48.63.2
Ny
-15db
-4db -db o
-c
Disp
Ncr
-b
-a
Buckling force
Expected variaton
6.4.2
Recognizing the situtation above, we proposed a new method to solve this problem, in which
we bind together a group of n elements with FEMA hinge to simulate one truss member and
give these elements dierent values at b and c (each line connecting points ci , bi , a, a
and b ( i i n) in gure 6.12). As a result, these elements may buckle one by one as external
force increases, resulting in a progressive reduction in the total axial force of these elements
(line connecting points C, B, A, B, and C in gure 6.12). Suppose the cross-sectional area
of original truss member is A0 , those elements have to satisfy the following conditions: (1) The
total cross-sectional area, yield and buckling forces of these elements must be equal to those of
125
Ai = A0 ;
i=1
n
Nyi = Ny0 ;
i=1
n
Ncri = Ncr0
(6.7)
i=1
where A0 = cross-sectional area of original truss member; Ny0 = yield force of original truss
member; Ncr0 = buckling force of original truss member; n= number of elements used for
simulation; Ai = cross-sectional area of element i; Nyi = input yield force of element i; and
Ncri = input buckling force of element i;(2) The elements have to yield or buckle simultaneously.
Therefore Nyi and Ncri input to each element are calculated by following equations.
Nyi =
Ai Ny0
A0
(6.8)
Ai Ncr0
= Ai cr0
(6.9)
A0
where fy = yield stress; E= modulus of elasticity; =boundary slenderness ratio; 0 = slenderness ratio of original truss member; and cr0 =buckling stress of original truss member,
calculated by
2E
0 (Euler s)
2
0
cr0 =
(6.10)
y 0
0
Ncri =
Force
N y0
N y0 /n
II
II
-ndb
-C -cn
An
-bn
A0
Change
-11db
-9db
-7db
-5db
-4db -db o
-10db -8db -6db
-c8 -c7 -c6 -c5 -c4 -c3 -c2 -c1
...
dy
A8 A7 A6 A5 A4 A3 A2 A1
Ncr0 /n
A1 A2
... An
A3
A 5 A4
II
-B
-A
Ncr0
C(c)
6dy Disp
126
6.5
6.5.1
Numerical Examples
Plane Truss
The rst analysis model is a 2m by 2m square plane truss, pinned supported to the ground
at two bottom nodes (shown in gure 6.13). All truss members have the same cross sectional
area, 48.6 3.2, with design constants, = 7850kg/m3 , = 235N/mm2 , E = 205GP a.
The yield and buckling forces input to bilinear type hinges for elements are listed in table 6.1.
An approximate collapse load of 100 kN is applied at the node C along the X direction for
load increment in static incremental analysis, and the structural selfweight is treated as initial
weight. Two analysis cases are considered for this model:
1) Static incremental analysis using bilinear type hinges (gure 6.7)
2) Static incremental analysis using proposed method (gure 6.12).
5
E
1
EA
EA
3
L=2 m
EA
3
EA
L=2 m
127
Figure 6.14: Accumulative load factor variation using bilinear type hinge
Analysis case 2
In this analysis case, each truss member is replaced by 10 elements with FEMA hinge using our
proposed methods in static incremental analysis. Figure 6.17 shows an image of the modied
structure. As a typical example, the denition of ten elements for replacing the member No.5
is shown in gure 6.18. The total cross-sectional area of the ten elements is the same as that
of No.5. They have the same yield and bucking forces but dierent ductility, namely, dierent
values at point -C. Displacement control method is chosen for load increment, with a step
number of 500. The reason why a larger step number is adopted than case 1 is to make the
128
load increase more slowly so that a group of elements would not collapse simultaneously but
collapse one by one. The larger number the step is given, the higher possibility of observing
a gradual decrease after buckling can be obtained. The accumulative load factor variation,
shown in gure 6.19, proves our expectation, in which it declines increasingly after reaching
exact collapse load factor.The collapse load facotr obtianed is 0.787 The collapse process is
shown from gure 6.20 to gure 6.21, which is identied the same as that in case 1.
Force
21~30
L=2 m
31~40
0
~6
-12db
-10db
-8db
-6db
-4db
-13db
-11db
-9db
-7db
-5db
-db
-c10 -c9 -c8 -c7 -c6 -c5 -c4 -c3 -c2 -c1
dy
6dy
51
~5
41
1~10
Ny=10725.4N
11~20
L=2 m
-b10 -b9 -b8 -b7 -b6 -b5 -b4 -b3 -b2 -b1
Ncr=5798.8N
-a
Disp
129
Comparison
The accumulative load factor variations in two cases are plotted together in gure 6.22. It
can be found that the collapse load factor obtained by new method is slightly lower than that
in case 1 due to the consideration of post-buckling inuence. However, it is found that their
dierence is not very signicant, only 2.6 dierent.
collapse load factor : 0.808
130
6.5.2
Frame Truss
A cube truss model with a 2m by 2m square on each face is considered in this section. The
truss structure is supported by 4 pinned supports at the bottom (shown in gure 6.13). All
truss members have the same cross sectional area, 48.6 3.2, with design constants, =
7850kg/m3 , = 235N/mm2 , E = 205GP a. Yield and buckling forces of a member are the same
as those in previous plane truss if they are the same length (table 6.1). An total approximate
collapse load of 200 kN is applied at the nodes No.2 and No.5, in the X direction. The structural
self weight is treated as initial weight. Like previous example, the same two analysis cases are
considered for the model:
1) Static incremental analysis using bilinear type hinges (gure 6.7)
2) Static incremental analysis using proposed method (gure 6.12).
8
5
2
6
10
9
7
2
16
15
17
18
2m
2m
14
11
13
12 7
6
4
1
3D- view
2m
Top view
2m
Side view
Analysis case 1
A step number of 100 is used, and displacement control method is adopted with a boundary
displacement of 40 mm at node No.2 in X direction. The accumulative load factor variation
in this case is shown in gure 6.24,and the corresponding collapse processs is shown in gure
6.25. The obtained collapse load factor is 0.97.
131
Figure 6.24: Accumulative load factor variation using bilinear type hinge
STEP 5
STEP 17
STEP 18
STEP 19
132
Analysis case 2
A step number of 200 is used, and displacement control method is chosen with a boundary
displacement of 40 mm at node No.2 in X direction. Like the previous example of plane truss,
each member is replaced by 10 elements, with a modied structure shown in gure 6.26. The
accumulative load factor variation is shown in gure 6.27; a collapse load factor of 0.928 is
obtained. It is interesting that the accumulative load factor does not gradually decrease after
reaching the collapse load factor. It undergoes a steep reduction at the displacement of around
30 mm. This is because that a group of elements with the same values at c (gure 4.19)
yielded and eventually collapsed at the same time. Figure 6.28 shows the axial force variation
of elements associated with the phenomenon of sudden reduction after buckling. Let us see
what has happened. Elements 31-40 reached their buckling force at step 25, and collapsed
one by one after during till step 85. And, elements 41-50 reached their yield force instead of
buckling force at step 38 and kept their yield force till step 155. However, after step 155, as
they have the same value in point c (gure 6.18), they suddenly collapsed at the same time,
therefore causing a sudden decrease in the accumulative load factor. Afterwards, the collapse
of the whole structure was reached. The collapse process is shown in gure 6.29. It is found
that the existence sequence of collapse members before reaching collapse is the same as that in
case 1.
Collapse load factor: 0.927
6170
111
1
0
16
20
141150 151
101
1
10
1120
2130
50
41
31
110
1
17
16
7180
0
1
14
13
91
90
10
5160
80
1
171
81
30
121
40
133
134
STEP 9
STEP 38
STEP 36
STEP 38
STEP 156
STEP 37
STEP 60
135
Comparison
The accumulative load factor variations in two cases are plotted together in gure 6.22. It can
be found that collapse load factor obtained by our new method is slightly lower than that in
case 1 due to the consideration of post-buckling inuence. However, it is also demonstrated
that the dierence is considerably low, only 4.3 .
6.5.3
Dome Truss
The commonly known Hangai truss is solved as an example of dome truss structures in this
section. The structure has 24 members, with each member having a cross-sectional area of
76.3 4.5. It is six-face symmetrical, and each face comprises two triangles shearing one side
with each other. It is supported by 6 pinned supports at the six bottom nodes, as shown in
gure 6.31. Design constants are = 7850kg/m3 , = 235N/mm2 , E = 205GP a. A uniform
vertical load of 10.68 kN/m2 , which is 20 times as large as standard snow load of Nagoya city,
is applied as an approximate collapse load. The structural selfweight is treated as initial load
in static incremental analysis. Also, the following two analysis cases are considered:
1) Static incremental analysis using bilinear type hinges (gure 6.7)
2) Static incremental analysis using new proposed method (gure 6.12).
136
Z
821.6 mm
X
9
Pinned support
11
13
X
12
8660 mm
cr=0.94
137
Analysis case 2
(141
-150
The same step number as that in case 1 is used in this analysis case, and each member is
replaced by 10 elements. The accumulative load factor variation is shown in gure 6.34, and
its corresponding collapse process is shown in gure 6.35. It is found that the collapse load
factor as well as the collapse mechanism is completely the same as that in case 1. This is
because both two cases underwent only one step to reach collapse. The very interesting thing
in both cases is that the outer layer 12 members did not buckle totally in hangai truss, only 8 of
them buckled, which were actually expected to mechanically buckle at the same time. This is
probably because that the hangai truss is a single-layer truss structure and has a possibility of
becoming non-symmetrical buckling. This can be proved by observing the form change before
and after collapse. Figure 6.36 shows the form variation of hangai truss in the analysis case
1. It is obvious that the structure had been keeping its six-face symmetry characteristic before
reaching collapse (before step 7), but it suddenly changed its shape when reaching collapse (after
step 7), without keeping 6 faces symmetrical again. The node displacement values indicated in
the graph further illustrate the phenomenon.
-80
(71
-20
(11
30
1-1
(12
(81-90)
(1-10)
cr=0.94
In addition, it is worth noting that the accumulative load factor in this case did not gradually
decrease after attaining collapse load factor. Instead, it increased for several steps after the
displacement of around 50 mm. After that, it continued decreasing till the whole collapse. This
phenomenon is due to members 2, 18, 24 changed their force type from compression force to
tension force during the analysis process. Figure 6.38 and 6.39 provides the force variation
of members of 1/6 part of the structure (corresponding to gure 6.35). It can be seen that
138
STEP 1
STEP 4
STEP 10
STEP 15
STEP 7
STEP 75
STEP 9
STEP 150
139
D
Member 17
Nyi=23852.5 N
Ncri=13270.2 N
Member 18
Nyi=23852.5 N
Ncri=13270.2 N
Member 7
Nyi=23852.5 N
Ncri=13270.2 N
140
A
D
Member 24
Nyi=23852.5 N
Ncri=17398 N
Member 23
Nyi=23852.5 N
Ncri=17398 N
Member 1
Nyi=23852.5 N
Ncri=174371 N
Member 2
Nyi=23852.5 N
Ncri=174371 N
6.5.4
141
Figure 6.40 shows the 8 by 8 m doulbe layer truss structure with 128 members of the same
cross sectional area of 48.6 3.2. The structure is supported by four pinned supports at four
corner nodes and by a roller support at mid span of each side. For each upper chord node, a
single load 50 kN is input as the approximate collpase load for static incremental analysis.
16
33
17
28
11
37
12
26
41
19
18
32
20
14
13
50kN
40
36
34
30
Pinned support
25
15
38
50kN
50kN
50kN
1m
29
24
22 m=8 m
23
22
21
22 m=8 m
2
4
27
31
3
6
35
5
10
Side view
39
7
22 m=8 m
Top view
Analysis case 1
Static incremental analysis using bilinear type hinge is performed in this case, with a step
number of 200. Displacement control method is selected to govern the analysis, with a maximum
vertical deection of 500 mm at node 13. The variation of accumulative load factor is shown in
gure 6.41, and the corresponding collapse mechanism is indicated in gure 6.42. The collapse
load factor obtained in this analysis case is 0.605.
Analysis case 2
A step number of 200 are used for static incremental analysis, and each member is also replaced by 10 elements. Displacement control method is selected to govern the analysis, with a
maximum vertical deection of 500 mm at node 13. The variation of accumulative load factor
142
is shown in gure 6.43; a collapse load factor of 0.599 is obtained. The corresonding collapse
mechanism is shown in gure 6.44. For space limitation, we do not discuss the details about
this example because the modied structure has as many as 1280 elements and our main goal
is to compare the collapse load factor of the two cases.
Figure 6.41: Accumulative load factor variation using bilinear type hinge
51
(6
50
-6
-6
41
60
(6
(271-280)
(491-500)
(471-480)
Comparison
The accumulative load factor variations in two cases are plotted together in gure 6.45. It is
also found that collapse load factor obtained by our new method is little smaller than that in
case 1, 0.9 dierent.
6.6. CONCLUSION
143
6.6
Conclusion
In this chapter, the inuence of post buckling behavior of truss members on the collapse load
factor of the whole structure was investigated by proposing a new method based on MIDAS
software. Four dierent types of truss structure, namely, plane truss, frame truss, dome truss
and double layer truss, were investigated regarding this inuence. The results shown in gures
6.22, 6.30, 6.37, and 6.45, indicates that the biggest dierent between collapse load factors
obtained by analyses with and without consideration post-buckling inuence is less than 5 ,
which means the post buckling inuence is quite smaller on collapse load factor, at least for
these solved truss structures in this study. Consequently, the hinge by hinge method applied
to truss structures based on some assumptions in section 3.32 of chapter 3 proves to be reliable
and valid when calculating the collapse load factor even though it could be slightly larger than
the exact collapse load factor.
Chapter 7
CONCLUSION
7.1
Summary
146
CHAPTER 7. CONCLUSION
a minimum weight structural design. And, conguration optimization is to consider the nodal
ordinate as design variable to nd an optimal conguration for trusses. The present research
can simultaneously deal with these three optimization problems for a truss structure, although
in the later numerical examples only optimization problems with topology and sizing as design
variables were demonstrated. In the end of this chapter, the concept of plastic design was
introduced. It is learned that plastic design is more economical than elastic design if the design
purpose is to prevent the structure from collapse.
Chapter 3
In this chapter, the method of genetic expression of truss structure was rst illustrated and
then the structural analysis for calculating the tness value of trusses was conducted. The
method of genetic expression for truss structures is previously developed by Kawamura. In
this respect, present study extended the technique of one-axis symmetry to that of multi-axis
symmetry in truss topology generation. After genetic expression of truss structure, present
study conducted an structural analysis for truss structures to calculate tness values, which
includes both elastic and plastic analysis. In elastic analysis, the most commonly used nite
element method (FEM) was employed. Concering plastic analysis, it contains a variety of
analysis types, such as elasto-plastic analysis, limit analysis, etc. In our study, since it was
intended to trace the collapse process for a truss, the elasto-plastic analysis was used, where
the hinge by hinge method is actually the critical calculation method. However, the hinge by
hinge method is a method generally used in elasto-plastic analysis for frames. To apply this
method to trusses, we assumed that a hinge happens in a truss structure when the stress in a
truss member reaches yield stress or buckling stress, and the member could keep its ultimate
stress until the overall collapse. Under this assumption, this method was successfully applied
to truss structures, and it calculation process was coded into program for present study. By
combining the FEM and Hinge by hing method, a structural analysis containing both elastic
and plastic analysis was established.
Chapter 4
vAt the beginning of this chapter, a comparison of elastic and plastic design in optimization
was rst made. It was found that under the same loading condition the optimal truss structure
obtained in optimization satisfying plastic design is lighter than that in optimization satisfying
elastic design, which nely corresponds to the concept of plastic design. Afterwards, a single
objective truss optimization problem satisfying both elastic and plastic design was established.
The objective was to minimize total truss structural weight with topology and sizing as design
variables, subject to constraints related to design requirements in both rst and second design
7.1. SUMMARY
147
148
CHAPTER 7. CONCLUSION
buckling. Therefore, it becomes necessary to investigate this behavior on the collapse load
factor of truss structures. To investigate this kind of inuence, commercial software MIDAS
was used. Unexpectedly, in MIDAS, there is no type of plastic hinge that can directly simulate
the phenomenon of gradual reduction in stress after buckling. Thus, we proposed a new method
that can consider the gradual stress reduction based on one type of plastic hinge in MIDAS,
named FEMA type hinge. Through our newly proposed method, we investigated dierent types
of truss structure about the post-buckling inuence. It was demonstrated that the biggest
dierence between our assumption and proposed method, reckoned close to real cases, is less
than 5 , indicating that the post-buckling behavior has quite small impact on the collapse
load factor in truss structures. This result could further support the reliability and validity of
optimization problems conducted in chapters 4 and 5.
7.2. EXPECTATION
7.2
149
Expectation
150
CHAPTER 7. CONCLUSION
Dynamic analysis
During the elasto-plastic analysis, it was considered that the structural model is always constant
till the overall collapse. However, with the gradual formation of plastic hinges in a truss
structure, its structural model may in fact vary. This could exert an inuence on the load
increment, which is to say that the load increment would become unproportional in elastoplastic analysis. Consequently, it is needed to incorporate dynamic analysis into the elastoplastic analysis for a higher computational accuracy in the future.
APPENDIX A
DETAILS OF MULTI-OBJECTIVE
GENETIC ALGORITHM
A.1
Overview
In this appendix, the research history about SPEA2 is reviewed and it algorithm is specically
described.
In section A.2, the Zitzler58) proposed SPEA2, one method of multi-objective genetic algorithm, is described.
In section A.3, the important schemes of multi-objective genetic algorithm are depicted.
151
152
A.2
SPEA2
The SPEA2 (Strength Pareto Evolutionary Algorithm 2, Improving the strength Pareto Evolutionary Algorithm) was proposed by Zitzler in 200158) as an improved version of SPEA. In
1999, SPEA was recognized as an algorithm with comparatively new exploration performance.
However, from 1999 to 2001, several potential weakness of SPEA were indentied after some
other important algorithms or mechanisms, such as the representative NSGA-II, were proposed.
Therefore, in order to pursue higher exploration eciency, the SPEA2 was proposed by making
some important modications to SPEA. Compared to SPEA, the advantages of SPEA2 are as
follows.
Advanced Fitness Assignment Scheme
For a given individual in a population, the improved tness assignment scheme can take into
account how many individuals it dominates as well as how many individuals dominate it. Also,
it takes into account the degree of density of the population.
New Truncation Method
The method is used for reducing the archive into an appropriate number. The new truncation
method can prevent individuals being the closest to the objective axes from being deleted.
Treatment of Archive and Population
Like SPEA, SPEA2 generally saves N non-dominated solutions so far discovered to archive.
However, simulating the scheme of NSGA-II, it further selects a group of individuals from
archive and performs GA operations on them so that more excellent individuals can be discovered.
The following describes the overall algorithm of SPEA2 as well as the archive update process
based on the improved tness assignment scheme and new truncation method.
A.2.1
Algorithm of SPEA2
(population size)
(archive size)
A.2. SPEA2
Output: A
153
(nondominated set)
Step 1 Initialization: Generate an initial population P0 and create the empty archive
P0 = Set t = 0
(A.1)
A.2.2
Fitness Assignment
To avoid the situation that individuals dominated by the same archive members have identical
tness values, in SPEA2, for each individual, both dominating and dominated solutions are
taken into account. In detail, each individual i in the archive Pt and the population Pt is
assigned a strength value S(i), representing the number of solutions it dominates:
r(i) =
jPt +Pt , ji
s(j)
(A.2)
154
where | | denotes the cardinality of a set, + stands for multiset unin and the symbol
corresponds to the Pareto dominance relation. On the basis of the S values, the raw tness
R(i) of an individual i is calculated
r(i) =
s(j)
(A.3)
That is, the raw tness is determined by the strengths of its dominators in both archive and
population, as opposed to SPEA where only archive members are considered in this context. It
is important to note that tness is to be minimized here, i.e., R(i) = 0 corresponds to a nondominated individual, while a high R(i) value means that i is dominated by many individuals
(which in turn dominate many individuals). This scheme is illustrated in gure A.1.
f2
f2
nondominated
2/6
nondominated
8/6
2
dominated
dominated
0
5/6
11/6
5
9
11/6
12
11/6
15/6
2/6
19
13/6
f1
0
14
f1
Figure A.1: Comparison of tness assignment schemes in SPEA and SPEA2 for a maximization
problem with two objectives f1 and f2 .
Although the raw tness assignment provides a sort of rich mechanism based on the concept
of Pareto dominance, it may fail when most individuals do not dominate each other. Therefore,
additional density information is incorporated to discriminate between individuals having identical raw tness values. The density estimation technique used in SPEA2 is an adaption of the
k th nearest neighbor method,67) where the density at any point is a (decreasing) function of
the distance to the k th nearest data point. Here, we simply take the inverse of the distance
to the k th nearest data neighbor as the density estimate. To be more precise, for each
individual i the distances (in objective space) to all individuals j in archive and population are
calculated and stored in a list. After sorting the list in increasing order, the k th element
gives the distance sought, denoted asik . As a common setting, we use k equal to the square
root of the sample size,67) thus, k =
A.2. SPEA2
155
i is dened by
D(i) =
1
ik + 2
(A.4)
In the denominator, two is added to ensure that its value is greater than zero and that D(i) < 1.
Finally, adding D(i) to the raw tness value R(i) of an individual i yields its tness F (i):
F (i) = r(i) + d(i)
(A.5)
The run time of the tness assignment procedure is dominated by the density estimator
(O(M2 log M )), while the calculation of the S and R values is of complexity O(M2 ), where
M = N + N.
A.2.3
Truncation Method
The archive update operation (step 3 in algorithm) in SPEA2 diers from the one in SPEA in
two respects: i) the number of individuals contained in the archive is constant over time, and
ii) the truncation method prevents boundary solutions being improved.
During environmental selection, the rst step is to copy all non-dominated individuals, i.e.,
those which have a tness value lower than one, from archive and population to the archive of
the next generation:
Pt+1 = {i|i Pt + Pt F (i) < 1}
(A.6)
If the nondominated front ts exactly into the archive (|P t+1 | = N ) the environmental selection
step is completed. Otherwise, there can be two situations: Either the archive is too small
(|P t+1 | < N ) or too large (|P t+1 | > N ). In the rst case, the best N |P t+1 | dominated
individuals in the previous archive and population are copied to the new archive. This can be
implemented by sorting the multi-set Pt + P t according to the tness values and copy the rst
N |P t+1 | individuals i with F (i) 1 from the resulting ordered list to P t+1 . In the second
case, when the size of the current non-dominated(multi)set exceeds N , an archive truncation
procedure is invoked which iteratively removes individuals from P t+1 until |P t+1 | = N . Here,
at each iteration that individuali is chosen for removal for which i d j for all j P t+1 with
i d j : 0 < k < |P t+1 | : ik = jk
0 < k < |P t+1 | : [( 0 < l < k : il = jl )] ik < jk
(A.7)
k
where i denotes the distance of i to its k th nearest neighbor in P t+1 . In other words, the
individual which has the minimum distance to another individual is chosen at each stage; if
there are several individuals with minimum distance the tie is broken by considering the second
156
smallest distances and so forth. How this truncation technique works is illustrated in gure
A.2.
Although, the worst run-time complexity of the truncation operator is O(M 3 ) (M = N +N ),
on average the complexity will be lower (O(M2 log M )) as individuals usually dier with regard
to the second or third nearest neighbor, and thus the sorting of the distances governs the overall
complexity.
f2
f2
f1
f1
Figure A.2: Illustration of the archive truncation method used in SPEA2. On the right, a
non-dominated set is shown. On the left, it is depicted which solutions are removed in which
order by the truncate operator (assuming that N = 5).
A.3
Until now, a lot of research regarding multi-objective genetic algorithm has been carried out.
Especially after 1999, very advanced methods like SPEA56) SPEA258) were proposed. This is
because the signicant mechanisms for plural exploration in multi-objective genetic algorithm
have been made clear, and most of the proposed methods are combinations of these mechanisms.
Therefore, although there are some dierences among those methods, their whole mechanisms
are not as much dierent.
The common mechanisms among these methods have a very important eect on the exploration using multi-objective genetic algorithm, and this eect has been veried through a
number of numerical analyses.5658, 68) A summary of the common mechanisms is presented in
the following.
A.3.1
157
A.3.2
In order to reect presently saved archive to population, it is necessary for archive to participate in GA operation of reproduction. The goal of this treatment is to accelerate the speed
of exploration through much better individuals. Methods like NSGA-II, SPEAS adopt this
technique and use archive to generate the population for the next generation.57, 58) In general,
selecting individuals form achive to create the next population is called mating selection.
t=t+1
Crossovermutation
2QRWNCVKQPPt
Pt + Pt
Population Pt+1
individual1
individual1
individual2
individual2
individual1
individual2
individual N
individual N
individual N
individualN + 1
Mating Seletion
individualN + N
#TEJKXG Pt
Archive Pt+1
individual1
individual1
individual2
Enviromental
Selection
individual N
individual2
Fitness
assignment
individual N
t=t+1
Output
158
A.3.3
Archive Truncation
The archive truncation is to reduce the number of non-dominated individuals to exactly t the
number of archive predened once the number of non-dominated individuals is larger than that
of archive. Depending on dierent problem, the number of non-dominated individuals may be
larger than that of archive. In this situation, because all the individuals for selection belong to
non-dominated individuals, it is necessary to perform truncation on archive according to the
information of non-dominated individuals, such as density, diversity, etc.
The greatest issue in archive truncation is that it may aect the convergence of optimization
towards the Pareto optimal front. As a matter of fact, each non-dominated individual is one of
the best individuals that exploration has found presently. However, due to archive truncation,
there may be existing non-dominated individuals that are worse than those have been deleted
by truncation. The worst result of this is that the archive will walk backwards the Pareto
optimal front. As a result, the exploration will lose the possibility of converging towards Pareto
optimal front due to the archive truncation. Therefore, it needs to be very prudent to perform
archive truncation.
The representative selection method according to diversity is the method using sharing.55)
However, in the selection using sharing, there is no guarantee that the best individual on each
objective axis can be saved, because parameters relevant to sharing are newly required. Losing
the best individuals on objective axes will narrow exploration domain, aecting the convergence
to a global solution.
On the other hand, some new methods, such as the new truncation method in SPEA2,
method of calculating disorder degree in NSGA-II, can consider the problems in using sharing.
These methods can free the parameters associated with sharing and of course save the best
individuals on objective axes. As a result, the distribution of non-dominated individuals will
not lose its width for each objective axis, i.e., the distribution of non-dominated individuals
will not shrink.
A.3.4
Fitness Assignment
As described previously, in multi-objective genetic algorithm, it is needed to explore highaccuracy solutions while maintaining the diversity in the population. Thereby, it is necessary
to assign tness values to individuals with consideration of their dominance relationship as well
as the magnitude of their access to optimal front in objective function space. The representative
methods of tness assignment are the combination method of Pareto ranking and sharing in
MOGA, the unique tness assignment method in SPEA2, and combination method of disorder
degree calculation and sorting of non-dominated solutions in NSGA-II, etc. In addition, a
special case is NPGA in which no concept of tness appears because a unique selection method
159
A.3.5
If the scale of each objective function value is not normalized, it may be dicult to precisely
measure the distances among individuals in the objective function space. This is not a problem
when the scales of all objective functions are the same. Nevertheless, when the scales of objective
functions are dierent, it becomes a critical problem.
An eective method dealing with this problem is the scaling method used in NPGA2, in
which the maximum and minimum values for each objective function in population is used and
objective function values of each individual are scaled according to the following equation after
tness assignment.
Oi Oi,min
(A.8)
Oi =
Oi,max Oi,min
where
Oi
Oi
Oi,max
Oi,min
The
The
The
The
REFERENCE
[1] A.Dominguez. Practical design optimization of truss structures using genetic algorithms.
Res Eng Design, 2006.
[2] Dorn W.S., R.E. Gomory, and H.J. Greenberg. Automatic design of optimal structures.
Journalde Mecanique, Vol. 3, pp. 2552, 1964.
[3] Dobbs M.W., and L.P. Felton. Optimization of truss geometry. Journal of Structural
Division, Proceeding of the American Society of Civil Engineers, Vol. 95, No. 19, pp.
21052118, 1969.
[4] Grieson D.E., and Pak W.H. Optimal sizing, geometrical and topological design using a
genetic algorithm. Structural Optimization, Vol. 6, pp. 151159, 1993.
[5] Hajela P. Genetic search - an approach to the nonconvex optimization problem. AIAA
Journal, Vol. 28, No. 7, pp. 12051210, 1990.
[6] Holland J. Adaptation in natural and articial systems: an introductory analysis with
applications to biology, control, and articial intelligence. Univesity of Michigan, 1975.
[7] Goldberg D. E. Genetic algorithms in search, optimization and machine learning.
Addition-Wesley Publishing Company, Inc., 1989.
[8] Ponslet E., and H. H. Curney. Optimal placement of tuning masses on truss structures
by genetic algorithms. 34thAIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Meterials Confernce AIAA/ASME Adaptive Structures Forum Lajolla, pp.
24482457, 1993.
[9] Rajan S. D. Sizing, shape, and topology design optimization of trusses using genetic
algorithm. Journal of Structural Engineering, Vol. 121, No. 10, pp. 14801487, 1995.
161
162
REFERENCE
[10] Rajeev S., and C. S. Krishnamoorthy. Genetic algorithm-based methdologies for design
optimization of trusses. Journal of Structural Engineering, Vol. 123, No. 3, pp. 350358,
1997.
[11] Ohsaki M. Genetic algorithm for topology optimization of trusses. Computers and Structures, Vol. 57, No. 2, pp. 219225, 1995.
[12] Hajela P., and E. Lee. Genetic algorithm in truss topological optimization. International
Journal of Solods and Structures, Vol. 32, No. 22, pp. 33413357, 1995.
[13] Matsumoto S., Kasuga Y., Fujii D., and Fujitani Y. Minimum cost analysis of semirigid skeleton steel structures considering connection cost. Journal of Structural and
Construction Engineering, No. 528.
[14] Ohsaki M. A genetic algorithm for structural optimization with a discontinuous cost
function. Journal of Structural and Construction Engineering, Vol. 3, No. 464.
[15] Onaka A., Arai H., Fujii D., and Fujitani Y. Minimum cost design of steel building and
reinforced concrete buildings. Journal of Computational Engineering and Science, Vol. 3,
No. 2.
[16] Arase M., Saitoh M., Okada A., and Abe H. Study on structural property of tower
consisted cables: Part 2: Mechanical characteristics and construction method. Summaries
of Technical Papers of Annual Meeting, AIJ (Structures B-1).
[17] Kanemitsu T., Hangai Y., and Mitsui K. Member arrangement of structures in construction process: Arrangement of temporary members by genetic algorithms. Journal of
Computational Engineering and Science, Vol. 2, No. 1.
[18] Kanemitsu T., and Hangai Y. Optimization analysis for initial and construction process
of truss structures. Journal of Structural and Construction Engineering, No. 510.
[19] Kanemitsu T., Hangai Y., Mitsui K. Member arrangement of structures in construction process: Optimization of dividing structures into blocks using genetic algorithms.
Summaries of Technical Papers of Annual Meeting, AIJ (Structures B-1).
[20] Kanemitsu T., Kawaguchi M., and Mitsui, K. Optimization of dividing structures into
blocks under construction. Journal of Structural and Construction Engineering.
[21] Kaneko Y., Mitusi K., Kanemitsu T., and Tosaka N. Construction process analysis for
truss structures by genetic algorithm. Journal of Computational Engineering and Science,
Vol. 2, No. 1.
REFERENCE
163
[22] Kaneko Y., Mitusi K., Kanemitsu T., and Tosaka N. Construction process optimization for truss structures by genetic algorithm. Journal of Structural and Construction
Engineering, No. 508.
[23] Kaneko Y., Mitusi K., Kanemitsu T., and Tosaka N. Multiobjective optimization of
construction process for truss structures. Proceedings of the Conference on Computational
Engineering and Science, Vol. 3, No. 2.
[24] Kawamura H. Computational morphogenesis of spatial structures using genetic algorithm.
Doctoral Dissertation of Nagoya University.
[25] Kunda M. Computational morphogenesis of truss structures considering homologous
deformation-structures for optical and infrared telescope. Master Thesis of Nagoya University.
[26] Sakamoto J., and J. Oda. A technique of optimal layout design for truss structures using genetic algorithm. 34th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Meterials Conference AIAA/ASME Adaptive Structures Forum Lajolla, pp.
24022408, 1993.
[27] Hajela P. Genetic algorithms in automated structural synthesis. Optimization and Articial
Intelligence in Civil and Structural Engineering, Vol. 1, pp. 639653, 1992.
[28] Kawamura T., Kito N., and Ohmori H. Structural optimization of truss topology by
genetic algorithms. Proceedings of the Conference on Computational Engineering and
Science, Vol. 3, No. 2, pp. 569572, 1998.
[29] Kawamura H., Kito N., and Ohmori H. Structural optimization of truss topology by
genetic algorithm: Part 1: Ga with triangle expression.
[30] Kawamura H., and Ohmori H. Structural creation of 3-dimentional trusses using genetic
algorithm. Journal of Structural and Construction Engineering, No. 538, pp. 115121,
2000.
[31] Kawamura H., and Ohmori H. Computational morphogenesis for discrete structures
with connection variables through usage of genetic algorithm. Journal of Structural and
Construction Engineering, No. 555, pp. 121128, 2002.
[32] Ohmori H., and Kito N. Form creation of truss structures by genetic algorithm. Journal
of Structural and Construction Engineering, No. 520, pp. 8592, 1999.
164
REFERENCE
REFERENCE
165
166
REFERENCE
[62] . II . , 1983.
[63] MIDAS Gen general structure design system. MIDAS Information Technology Co., Ltd,
2002.
[64] Kantou Branch of Architectural Institute of Japan. Design For Steel Structure. Gihodo
Press, 2005.
[65] Adeli H., and Cheng N.-T. Integrated genetic algorithm for optimization of space structures. J. Aerosp. Engrg., ASCE, Vol. 6, No. 4, pp. 315328, 1993.
[66] Stability problems in steel structures. Architectural Institute of Japan, 1992.
[67] B.W. Silverman. Density estimation for statistics and data analysis. Chapman and Hall,
1986.
[68] M. Erickson, A. Mayer, and J. Horn. The niched pareto genetic algorithm 2, applied to the
design of groundwater remediation systems. 1st International Conference on Evolutionary
Multi-Criterion Optimization, Springer-Verlag. Lecture Notes in Computer Science, No.
1993.
[69] Osaki H., and Hayashi M. Fairness metrics for shape optimization of shells. National
Congress of Theoretical and Applied Mechanics, Vol. 49, .
[70] Wasekura S., Kwamura H., Osada S., Ohmori H. Study on the possibility of creation of
new structural system using the new materials. Summaries of Technical Papers of Annual
Meeting, AIJ (Structures B-1).
[71] Dorn W. S., R. E. Gomory, and H. J. Greenberg. Automatic design of optimal structures.
Journal de Mecanique, Vol. 3, pp. 2552, 1964.
[72] Dobbs M. W., and L. P. Felton. Optimization of truss geometry. Journal of Structural
Division, Proceedings of the American Society of Civil Engineers, Vol. 95, No. 19, pp.
21052118, 1969.
[73] Fujii D., and Fujitani Y. Study on design optimization of a high rise building using thinwalled beam analysis. Journal of Structural and Construction Engineering, No. 510, pp.
99106, 1998.
[74] Fujii D., and Kikuchi N. Topology optimization using the homogenization design method:
Improvement of numerical instabilities in the optimization using slp method. Proceedings
of the Conference on Computational Engineering and Science, Vol. 4, No. 1, pp. 515518,
1999.
REFERENCE
167
[75] Takada T., Kohama Y., and Inoue T. Mathematical study on oor layout planning by
optimization method. Proceedings of the Conference on Computational Engineering and
Science, Vol. 6, No. 2, pp. 755758, 2001.
[76] Yagi T., Hagiwara N., Ohmori H., and Matui T. A new approach for cutting pattern
analysis of membrane structures by simultaneous consideration on both equilibrium condition and initial conguration. Journal of Structural and Construction Engineering, No.
508, pp. 7178, 1998.
[77] Land A. H., and A. G. Doig. An automatic method for solving descrete programming
problems. Econometrica, Vol. 28, pp. 497520, 1960.
[78] Takada T., Kohama Y., and Miyamura A. Study on optimal allocation of multi-story
shear walls in 3d rc frames. Journal of Structural and Construction Engineering, No. 522,
pp. 9398, 1999.
[79] Sakamoto J., and Oda J. Technique for determination of optimal truss layout using
genetic algorithm. The Japan Society of Mechanical Engineers, Vol. 59, No. 562, pp.
156161, 1993.
[80] Honma T., Tosaka N., and Sumi H. A computational method for inverse problems by using
autonomous decentralized approach. Journal of Structural and Construction Engineering,
No. 526, pp. 6976, 1999.
[81] Yamamoto K., Honma T., Minakawa Y., and Tosaka N. Minimum weight design of space
structure by using life-like approach : Genetic algorithms and autonomous decentralized
approach. Proceedings of the Conference on Computational Engineering and Science,
Vol. 4, No. 1, pp. 497500, 1999.
[82] Tasaka N., Honma T., Yamamoto K., and Minakawa Y. Minimum weight analysis of
space structures by using genetic algorithms and autonomous decentralized procedure.
Summaries of Technical Papers of Annual Meeting, AIJ (Structures B-1), pp. 381382,
1999.
[83] Imoto Y., and Tosaka N. Morphogenesis of truss structure using by decentralized nite element method. Proceedings of the Conference on Computational Engineering and
Science, Vol. 6, No. 2, pp. 743746, 2001.
[84] Matsuyama K., and Homa T. Initial pre-tension decision analysis on hybrid cable structure using autonomous decentralized nite element method. Proceedings of the Conference
on Computational Engineering and Science, Vol. 6, No. 2, pp. 763766, 2001.
168
REFERENCE
[85] Kakoi B., Honma T., Yamamoto K., and Minakawa Y. The optimum weight design analysis of truss structures with biological approach: Comparison and characteristic of solution
by genetic algorithms and autonomous decentralized nite element method. Summaries
of Technical Papers of Annual Meeting, AIJ (Structures B-1), 2000.
[86] Hangai Y., and Guan F. Structural shape analysis with the constraint conditions of
homologous deformation. Journal of Structural and Construction Engineering, No. 405,
pp. 97102, 1989.
[87] Mitsui K., and Tosaka N. An application of genetic algorithms to form nding analysis
of spatial structures. Journal of Structural and Construction Engineering, No. 484, pp.
7583, 1996.
[88] Oda J., and Matsumoto N. Design method of homologous structure using genetic algorithms (ga). The Japan Society of Mechanical Engineers, No. 59-568, pp. 248253,
1993.
[89] J. Holland. Adaptation in Natural and Articial Systems : an Introductory Analysis with
Applications to Biology, Control, and Articial Intelligence. University of Michigan, 1975.
[90] Ponslet E., and H. H. Curney. Optimal placement of tuning masses on truss structures
by genetic algorithms. 34thAIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Meterials Confernce AIAA/ASME Adaptive Structures Forum Lajolla, pp.
24482457, 1993.
[91] Rajan S. D. Sizing, shape, and topology design optimization of trusses using genetic
algorithm. Journal of Structural Engineering, Vol. 121, No. 10, pp. 14801487, 1995.
[92] Hajela P., and E. Lee. Genetic algorithm in truss topological optimization. International
Journal of Solods and Structures, Vol. 32, No. 22, pp. 33413357, 1995.
[93] Kurita M. Analysis of Telescope Supporting Structure and Its Application. Doctoral
Dissertation of Nagoya University, 2005.
[94] . . , 2005.
[95] Tanse R. Distributed genetic algorithms. Proc. of the Third ICGA, pp. 434439, 1989.
[96] University of California and California Institute of Technology. California Extremely
Large Telescope Conceptual Design for a Thirty-Meter Telescope. 2002.
REFERENCE
169
[97] Stefan J Medwadowski. The Structure supporting a Cluster of Segments of the Primary
Mirror of CELT. CELT Report No.25, 2001.
[98] Stefan J Medwadowski. CELT Tube Structure, Final Conceptual Design. CELT Report
No.28, 2001.
[99] Marcal P. V., and King I. P. Elastic-plastic analysis of twodimensionalstress systems by
the nite element method. Int. J. Mech.Sci., Vol. 9, pp. 143155, 1967.
[100] Yamada Y., and Yoshimura N. Plastic stress-strain matrix and its application for the solution of elastic-plastic problems by the nite element method. Int. J. Mech.Sci., Vol. 10,
pp. 343354, 1968.
[101] Yamada Y., and Yoshimura N. Elastoplastic solution of engineering problems, initial
stress nite element approach. Int. J. Num. Meth. Eng., Vol. 1, pp. 75100, 1969.
[102] Ueda Y., Yamakawa T., Akamatsu T., and Matsuishi M. A new theory on elastic-plastic
analysis of framed structures. Tech. Repts., Osaka Univ., Vol. 19, pp. 263275, 1969.
[103] Nigam N. C. Yielding in framed structures under dynamic loads. J. Eng. Mech. Div.,
ASCE, Vol. 96, pp. 687709, 1969.
[104] Livesley R. K. Matrix methods of structural analysis. Pergamon Press, London, England,
1964.
[105] Davies J. M. The response of plane frameworks to static and variable repeated loading
in the elastic-plastic range. St. Eng., Vol. 44, No. 8, pp. 277283, 1966.
[106] Majid K. I. Non-linear structures. London, UK. Butterworth Co.Ltd., 1972.
[107] Franchi A., and Cohn M. Z. Computer analysis of elastic-plastic structures. Comp. Meth.
Appl. Mech. Eng., Vol. 21, pp. 27194, 1980.
[108] Maier G., and Munro J. Mathematical programming applications to engineering plasticity
analysis. Appl. Mech. Reviews, Vol. 35, pp. 16311643, 1982.
[109] Tin-Loi F., and Pang J. Elastoplastic analysis of structures with nonlinear hardening:
A nonlinear complementarity approach. Comp.Meth. Appl. Mech. Eng., Vol. 107, pp.
299312, 1993.
[110] . . , 1995.
170
REFERENCE
[111] . . , 1994.
[112] . , 1 . , 1993.
[113] J. Bowman. On the relationship of the tchebyche norm and the ecient frontier of
multiple-criteria objectives. Multiple Criteria Decision Making, pp. 7686, 1976.
[114] A. Charnes, and W.W. Cooper. Management Models and Industrial Applications of
Linear Programming. Addition-Wesley Publishing Company, Inc., 1961.
[115] A.M. Georion, J.S. Dyer, and A. Feinberg. An interactive approach for multi-criterion
optimization, with an application to the operation of an academic department. Management Science, Vol. 19, No. 4, pp. 357368, 1972.
[116] D.E. Goldberg, and K. Deb. A comparison of selection schemes used in genetic algorithms.
Foundations of Genetic Algorithms 1 (FOGA-1), pp. 6993, 1991.
[117] Y.Y. Haimes, and W. Hall. Multiobjectives in water resources systems analysis: the
surrogate trade-o method. Water Resources Research, pp. 614624, 1974.
[118] Y.Y. Haimes, W. Halland, and H. Freedman. Multi-Objective Optimization in Water
Resources Systems The Surrogate Trade-O Method. Elsevier, 1975.
[119] A.P. Wierzbicki. The use of reference objectives in multiobjective optimization. Multiple
Criteria Decision Making Theory and Application, pp. 469486, 1980.
[120] P.L. Yu. A class of solutions for group decision problems. Management Science, Vol. 19,
pp. 936946, 1973.
[121] M. Zeleny. Compromise programming. Multiple Criteria Decision Making, pp. 262301,
1973.
[122] E. Zitzler, M. Laumanns, and L. Thiele. Spea2: Improving the strength pareto evolutionary algorithm for multiobjective optimization. Evolutionary Methods for Design,
Optimisation, and Control, pp. 95100, 2002.
[123] Kanai Y., and Hangai Y. Shape control truss structures subjected homologous deformation condition: Control by actuators. Summaries of Technical Papers of Annual Meeting,
AIJ (Structures B-1), Vol. 1997, pp. 429430, 1997.
[124] Kurita M., S. Sato, K. Morishima, H. Achiwa, H. Ito, T. Nagata, N. Noda, and N. Koiso.
Development of the ultralight medium-size telescope. Astronomical Structures and Mechanisms Technology, Vol. 5495, pp. 518525, 2004.
CITATION
k th nearest neighbor method, 155
closed set, 18
collapse load, 117
collapse load factor, 41
common load factor, 39
Complete optimal solution, 19
Computational Morphogenesis, 8
coordinate transformation matrix, 38
Crossover, 16
decimal point , 15
density estimator, 155
Displacement control method, 121
ecient solution, 19
elastic state, 24
elastic-perfectly plastic material, 23
elastoplastic
behavior, 22
environmental selection, 153
172
objective function, 19
objective-conict, 18
Pareto Approach, 21
Pareto dominance, 154
Pareto optimal front, 19
Pareto optimal solution, 19
Plastic Design, 26
Plastic design of steel structures, 22
Plastic Hinge, 25
plastic state, 24
plastication, 23
Post-buckling behaviour, 118
pure bending yield criterion, 42
raw tness, 154
remaining plastic moment capacity, 43
rounding-o, 15
scaling method, 159
Selection, 13
Single Objective Optimization Method, 12
Sizing optimization problem, 11
SPEA, 152
SPEA2, 152
statical indeterminacy, 27
story deection angle, 121
strain hardening, 23
Strength Pareto Evolutionary Algorithm 2,
152
Topology optimization problem, 8
truncation operator , 156
Vector
Evaluated Genetic Algorithm, 21
Weak Pareto optimal solution, 19
CITATION
PUBLICATIONS
Journals
1. H. Wang, H. Ohmori: Truss optimization Using Genetic Algorithm, Considering Construction Process, International Journal of Space Structures, Vol. 25, No. 4, pp. 205-215,
2010. 12.
2. H. Wang, H. Ohmori: Elasto-plastic Truss Optimization Using Genetic Algorithm, Journal of Structural and Construction Engineering, Vol. 77, No. 673. (To be published in
2012.3.)
3. H. Wang, H. Ohmori: Truss Optimization Using Genetic Algorithm Considering Ultimate
Resistance, Journal of the International Association for Shell and Spatial Structures, in
Press. (To be published in 2012.4)
4. H. Wang, H. Ohmori: Multi-objective Truss Optimization Considering Weight and Safety
Margin, Engineering Structures, Submitted. (On reviewing)
174
PUBLICATIONS
4. Huaguo Wang, Hiroshi Ohmori: Truss Optimization Using Genetic Algorithm Considering Ultimate Resistance, International Symposium on Algorithmic Design for Architecture
and Urban Design (Tokyo, Japan), paper ID: 53, 2011. 11.
5. Huaguo Wang, Hiroshi Ohmori: Elasto-plastic Truss Optimization Using Genetic Algorithm, 7th International Conference on Computational Mechanics for Spatial Structures
(IASS-IACM 2012) (Sarajevo, Bosnia and Herzegovina), 2012. 4.
Oral Presentations
1. WANG Huaguo KUNDA Masashi OHMORI Hiroshi: Computational Morphogenesis of
Truss Structures Considering Homologous Deformation-Structures for Optical and Infrared Telescope, Summaries of Technical Papers of Annual Meeting, AIJ (Structures I)
(Sendai, Japan), pp. 829-830, 2009. 8.
2. Wang Huaguo Ohmori Hiroshi: Practical Truss Optimization Using Genetic Algorithm,
The 59th National Congress of Theoretical and Applied Mechanics (Tokyo, Japan), pp.
49-50, 2010. 6.
3. Wang Huaguo Ohmori Hiroshi: Truss Optimization Using Genetic Algorithm Considering
Ultimate Resistance, The 5th Sub Committee of Analysis and Generation of Structural
Shapes and Systems (Tokyo, Japan), pp. 145-150, 2010. 10.
4. Wang Huaguo Ohmori Hiroshi: Practical Truss Optimization Using Genetic Algorithm,
Summaries of Technical Papers of Annual Meeting, AIJ (Structures I) (Toyama, Japan),
pp.377-378, 2010. 9.
5. Wang Huaguo Ohmori Hiroshi: Truss Optimization Using Genetic Algorithm Considering
Ultimate Resistance, Summaries of Technical Papers of Annual Meeting, AIJ (Structures
I) (Tokyo, Japan), pp.331-332, 2011. 8.
6. Wang Huaguo Ohmori Hiroshi: Multi-objective Truss Optimization Using Genetic Algorithm Considering Weight and Safety Margin, AIJ Tokai Chapter Architectural Research
Meeting (Nagoya, Japan), 2012.
ACKNOWLEDGEMENTS
I would like to express my appreciation to everyone who has made this thesis possible. The
special thank goes to my supervisor prof. Ohmori Hiroshi whose insightful guidance, kind
advice and encouragement I will never forget. He has been my inspiration as I hurdle all the
obstacles in the completion of my doctoral thesis in the past three years.
Thank you to Prof. Teshikawara, Prof. Furukawa, for their valuable opinions and comments
on my research work. To Dr. Jin Sang Wook for his very helpful discussion in my daily
study. To my predecessor, Dr. Kwamura Hiroaski, Mr. Kunda Masashi, for their kind help in
understanding the program of truss optimization using genetic algorithm. Thanks also go to all
the rest members in our laboratory who helped me from time to time during my study in Japan.
They are secretary Nakai Chiyo, research assistant Takatsuka Masao, doctoral course student
Yan Xingyu, graduated student Ishida Takayoshi, Fujita Kei, Yamazaki Kota, Kodama Sinichi,
Nakata Satoshi, and Maene Ayako, master course student in grade 2 Hayata Hiroshi, Yoshida
Hideki, Kawai Ryoji, and Kawasaki Masaomi, master course student in grade 1 Nakayi Yuki,
Nagono Mitsuaski and Yamada Hiroyuki, bachelor course student in grade 4 Ikeda Nahoko,
Hirano Noriyasu, and Fujishita Kazuhiro.
I owe my gratitude to prof. Fu Gongyi and Chen Wujun in Shanghai Jiaotong University
during my master study, who guided me to the research eld of steel structures. I am grateful to
Japans Ministry of Education (Monbukagakusho) for the substantial nancial support, without
which it would be very hard for me to nish my study. I wish to thank Preparatory School for
Chinese Students to Japan of Northeast Normal University, for the preparatory education of
Japanese language before coming to Japan.
Finally, I would like to express my great thanks to my beloved family, especially my wife
Yangyan for her unselsh understanding and assistance.
Wang Huaguo
February 2012
175