You are on page 1of 8

Engineering Structures 27 (2005) 17–24

www.elsevier.com/locate/engstruct

Optimum design of cold-formed steel channel beams using


micro Genetic Algorithm
Jaehong Leea,∗, Sun-Myung Kimb, Hyo-Seon Parkc, Byung-Hun Wood
a Department of Architectural Engineering, Sejong University, 98 Kunja Dong, Kwangjin Gu, Seoul 143-797, Republic of Korea
b Korea Institute of Construction Technology, 2311 Daewha Dong Ilsan Gu Goyang, Gyeonggi Do 411-712, Republic of Korea
c Department of Architectural Engineering, Yonsei University, 134 Sinchon Dong, Seodaemun Gu, Seoul 120-749, Republic of Korea
d LOTTE Engineering and Construction, 50-2 Jamwon Dong, Seocho Gu, Seoul 137-906, Republic of Korea

Received 24 February 2003; received in revised form 11 August 2004; accepted 16 August 2004
Available online 25 September 2004

Abstract

An important advantage of cold-formed steel is the great flexibility of cross-sectional profiles and sizes available to structural steel
designers. However, this flexibility makes the selection of the most economical section difficult for a particular situation. In this study, a
micro Genetic Algorithm (µ-GA) is used to find an optimum cross-section of cold-formed steel beams. The µ-GA is one of the improved
form of GAs, to reduce iteration and computing resources by using small populations. The design curves are generated for optimum values
of the thickness and the web flat-depth-to-thickness ratio for unbraced beams under uniformly distributed load. As numerical results, the
optimum design curves are presented for various load level.
© 2004 Elsevier Ltd. All rights reserved.
Keywords: Cold-formed steel channel beams; Optimization; Micro-genetic algorithm

1. Introduction optimization techniques for cold-formed steel sections have


been studied since the late 1990s. The optimization of cold-
Cold-formed steels are structural members that are made formed steels using a neural network has been studied by
from cold-forming a thin plate which is from 0.8 to 6 mm Karimm and Adeli [1], and El-Kassas et al. [2]. Adeli [1]
and has a section shape with the right purpose. One of the ad- investigated the optimum hat section of cold-formed steel
vantages of cold-formed steels is that the strength-to-weight beams under uniformly distributed loading. Mackie [2]
ratio is much higher than that of common hot-rolled shapes, studied the potential for using neural networks to optimize
thus it can reduce the total weight of the structures. They can a cold-formed steel channel, lipped channel and hat section.
also be easily manufactured and sized to meet any structural In this paper, an optimum design is presented for simply
requirement. Due to these advantages, cold-formed steel supported cold-formed steel channel beams under uniformly
members are considered economical for low-rise buildings. distributed load by using micro genetic algorithms (MGAs).
Different from hot-rolled shapes that must have been In the optimization process, all the possible structural effects
selected from precast goods, cold-formed steels have including moment, shear, deflection, and web crippling are
permanent choices, thus provide full advantages to considered as constraints, and the optimum design cross-
designers. However, this merit can be sometimes be a section is determined for various levels of loading.
disadvantage as designers must find the optimum section
size and shape for given conditions. For this reason, 2. Design of cold-formed steel channel beams

∗ Corresponding author. Tel.: +82 2 3408 3287; fax: +82 2 3408 3331. Cold-formed steel structures are susceptible to local
E-mail address: jhlee@sejong.ac.kr (J. Lee). buckling well below the global buckling load level due

0141-0296/$ - see front matter © 2004 Elsevier Ltd. All rights reserved.
doi:10.1016/j.engstruct.2004.08.008
18 J. Lee et al. / Engineering Structures 27 (2005) 17–24

Fig. 1. Consecutive stages of stress distribution in stiffened compression elements.

Fig. 3. Compression flange.

Fig. 2. Effective width (b) of stiffened compression elements.


where ρ = (1 − 0.22/λ)/λ, and λ is the plate slenderness
factor determined by
to high web flat-width-to-thickness (w/t) and flange flat- 
width-to-thickness ratios. Unlike one-dimensional structural f
λ= . (3)
members such as columns, stiffened compression elements Fcr
such as a compression web do not collapse when the
buckling stress is reached. Instead, an additional load can In Eq. (3), Fcr is the buckling stress of the plate given by
be carried by the element after buckling by means of a  2
π2E t
redistribution of stress as shown in Figs. 1 and 2. This Fcr = k (4)
12(1 − ν 2 ) w
phenomenon is known as postbuckling strength and is most
pronounced for elements with large w/t ratios. In Fig. 1, Fcr where k is the plate buckling coefficient, which becomes
is the local buckling stress and Fy is the yield stress of the 0.43 for an unstiffened compression element. E is the elastic
plate. modulus, ν is Poisson’s ratio of steel. In Fig. 3, f is the
Because the postbuckling strength is not treated enough average normal stress in the element.
in the structural design criteria of hot-rolled shapes, the
American Iron and Steel Institute (AISI) [4] estimates the
strength using a concept of “Effective width” which was first 2.2. Effective width of the compression web
introduced by von Kármán [5] in 1932 (Fig. 2) to consider
postbuckling strength as given in Eq. (1).
 w In the structural design process of cold-formed steel
members, the effective width in the compression portion of
f dx = b f max (1)
o the web is the sum of b1 and b2 as shown in Fig. 4. The
detailed procedure of the determination of the effective web
where b is called “effective width”.
depth is described in Fig. 5.
In what follows, the AISI specification [4] governing the
Fig. 5 illustrates the determination of the effective width
design of cold-formed steel channel beams used in this work
of the web using an iterative method. First, supposing
is summarized.
the web is fully effective, calculate the neutral axis, and
get the sum of b1 and b2 of the effective width of the
2.1. Effective width of the compression flange compression web using the plate buckling coefficient and
plate slenderness factor. If the sum of b1 and b2 is larger
than the length of the compression web, the web is fully
A flat compression element that is stiffened at only one effective. Otherwise, the neutral axis should be recalculated.
edge parallel to the direction of stress is an unstiffened This process should be repeated until the error becomes
compression element (u.c.e.) as shown in Fig. 3. According negligibly small.
to AISI [4], the effective width (b) under uniformly In Fig. 5, h 0 is the out-to-out depth of the web, b0 is
distributed load is calculated as the out-to-out width of the compression flange, h c is the

w (λ ≤ 0.673) compression portion of the web, and ε is the value of error
b= (2)
ρw (λ < 0.673) predefined by the user.
J. Lee et al. / Engineering Structures 27 (2005) 17–24 19

Fig. 4. Webs and other stiffened elements under a stress gradient.

2.3. Nominal bending strength of cold-formed steel channel where


beams
π 2 Es
σey = (11)
(K y L y /r y )2
The nominal moments of cold-formed steel channel  
beams can be calculated by considering the effective section 1 π 2 E s Cw
σt = GJ + (12)
modulus (Se ) which is calculated based on the effective Ar02 (K t L t )2
length of the section. The nominal moment, Mn , is given
by
A: Full unreduced cross-sectional area
Mn = Se Fy (5) Cb : Bending coefficient
Cw : Warping coefficient
where Fy is design yield stress of the steel. Sf: Section modulus of the full section
r0 : Polar radius of gyration
2.4. Nominal moments for lateral buckling strength of the cross-section about the shear center
K y , Kt : Effective length factor for bending
Cold-formed steel beams are prone to lateral buckling if about the y-axis and for twisting
braces are not adequately provided. In the design of cold- L y, Lt : Unbraced length of compression member for
formed steel beams, the moment capacity is governed by the bending about the y-axis and for twisting
lateral buckling strength of the member as well as the section
strength of the cross-section. AISI [4] suggests nominal
moments strength for lateral buckling as follows: 2.5. Shear strength of the cold-formed steel channel beams
Mn = Sc Fc (6)
According to AISI [4], the nominal shear strength of
where Sc is the elastic section modulus of the effective cold-formed steel channel beams, Vn , can be calculated as
section calculated at a stress Fc relative to the extreme follows:
compression fiber and Fc is determined as follows:
Vn = Aw Fv (13)
(a) For Fe ≥ 2.78Fy
where Aw is the area of the web (=ht), and Fv is the nominal
Fc = Fy . (7) shear stress calculated as follows:
(b) For 2.78Fy > Fe > 0.56Fy 
  (a) For h/t ≤ Ekv /Fy
10 10Fy
Fc = Fy 1 − . (8) Fv = 0.60Fy . (14)
9 36Fe
 
(c) For Fe ≤ 0.56Fy (b) For Ekv /Fy < h/t ≤ 1.51 Ekv /Fy
Fc = Fe (9) 
0.60 Ekv /Fy
Fv = . (15)
(h/t)

where Fe is the elastic critical lateral–torsional buckling (c) For h/t > 1.51 Ekv /Fy
stress calculated as follows:
Cb r0 A √ π 2 Ekv
Fe = σey σt (10) Fv = = 0.904Ekv /(h/t)2 (16)
Sf 12(1 − µ2 )(h/t)2
20 J. Lee et al. / Engineering Structures 27 (2005) 17–24

Fig. 5. Calculation of effective web depth.

where h is the depth of the flat portion of the web measured reaction. The nominal web crippling strength Pn for a single
along the plane of web, t is the thickness of the web, and kv unreinforced web can be determined as follows:



is the shear buckling coefficient, which is equal to 5.34 for R N
unreinforced webs. Pn = Ct Fy sin θ 1 − C R
2
1 + CN
t t
2.6. Combined bending and shear strength

h
× 1 − Ch (18)
The following interaction equation should be satisfied for t
combined bending and shear strength of cold-formed steel where N is the bearing length, R is the inside bend radius,
channel beams having unstiffened web: θ is the angle between the plane of the web and the bending
    surface, C is the web crippling coefficient, C R is the inside
Ωb M 2 Ωv V 2
+ ≤ 1.0 (17) bend radius coefficient, C N is the bearing length coefficient
Mnxo Vn
and Ch is the web slenderness coefficient. The coefficients
where M and V are the required allowable moment and and safety factors are given in Table 1.
shear strength resulting from uniformly distributed load,
respectively, Ωb and Ωv are the factor of safety for bending 2.8. Combined bending and web crippling strength
(1.67) and shear (1.60), Mnxo is the nominal flexural strength
about the centroidal x-axes determined for bending only, Unreinforced flat webs of shapes subjected to a
excluding lateral buckling, and Vn is the nominal shear combination of bending and concentrated load or reaction
strength when shear alone is considered. should be designed to meet the following requirements:
   
ΩW P Ωb M
2.7. Web crippling strength 1.2 + ≤ 1.5 (19)
Pn Mnxo
The webs of thin-walled cold-formed steel beams may where Ωb and ΩW are the factor of safety for bending
cripple due to the high local intensity of the load or (1.67) and web crippling (1.80), P is the required allowable
J. Lee et al. / Engineering Structures 27 (2005) 17–24 21

Table 1
Coefficients of single web channel and C-sections

C CR CN Ch Safety Factors Limits


ASD ΩW LRFD φW Canada LSD φW

4.00 0.40 0.60 0.03 1.80 0.85 0.7 R/t ≤ 2

strength for the concentrated load or reaction in the presence


of bending moment, and Pn is the nominal strength for
concentrated load or reaction in the absence of bending
moment determined in accordance with Section 2.7.

2.9. Deflection

The maximum deflection of a simply supported beam is


limited to L/300.

3. Optimization formulation for cold-formed steel chan- Fig. 6. Cross-section of a cold-formed steel channel.
nel beams
as the manufacturing section size changes.
The optimum design formulation of cold-formed steel
D L ≤ D ≤ DU (22)
channel beams is derived based on the AISI specification [4]
mentioned in the previous section. The objective function is B L ≤ B ≤ BU (23)
the cross-sectional area of beams consisting of four design R L ≤ R ≤ RU (24)
variables—depth (D), width (B), thickness (t) and radius t L ≤ t ≤ tU . (25)
(R). Eqs. (20) and (21) represent the objective function and
constraints of the present problem. Since the Genetic Algorithm is an unconstrained opti-
Minimize f (x) = (X 1 + 2X 2 − 0.86X 3 − 2.43X 4)X 4 mization technique, the optimization formulation is trans-
(20)
(X 1 = D, X 2 = B, X 3 = R, X 4 = t) formed by using the penalty function method as given in
Eq. (26). Penalty terms are added to the original objective
Mn
Subject to : M ≤ function when any constraint is violated. The amount of
Ωb penalty is decided by the ratio of violation. Each gene might
L have a higher fitness value in the fitness evaluation in the
δ≤
300 Genetic Algorithm as each gene has a lower objective func-
Vn tion value. The opportunity to be a parent gene is reduced
V ≤ as having lower fitness value; therefore, the gene with lower
Ωv
    fitness value would have lower survival probability on the
Ωb M 2 Ωv V 2
+ ≤ 1.0 next generation. Those genes would become extinct after
Mnxo Vn several generations and surviving genes with higher fitness
(21)
Pn value should produce better children on the next generation.
P≤
Ωw Minimize F(X, M) = f (X)
   
ΩW P Ωb M (Required Strength ≤ Allowable Strength)
1.2 + ≤ 1.5 (26)
Pn Mnxo F(X, M) = f (X) + penalty(X)
h (Required Strength > Allowable Strength)
≤ 200  
t Ωb M 300δ Ωv V Ωb M 2
penalty (X) = r + + +
b Mn L Vn Mnxo
≤ 60  2
t Ωv V 1 h 1 b Ωw P
+ + + +
where f (x) is the cross-sectional area of a cold-formed steel Vn 200 t 60 t Pn
channel beam.     2
1 ΩW P Ωb M
The upper and lower bounds of each design variable are + 1.2 + (27)
1.5 Pn Mnxo
considered for manufacturing, and used as side constraints
as Eqs. (22)–(25). The side constraints are subject to change where, r is a penalty constant.
22 J. Lee et al. / Engineering Structures 27 (2005) 17–24

Fig. 7. Micro Genetic Algorithm (MGA).

4. Micro Genetic Algorithms elitism and convergence checking with re-initialization to


obtain the optimal or near optimal solutions.
Genetic Algorithms (GAs) are search algorithms based In order to determine the appropriate population size in
on the mechanics of natural selection and natural genetics. this study, the various population sizes of 10, 12, 14, 16
They combine survival of the fittest among string structures and 18 individuals are tested, respectively. It appears that all
with a structured yet randomized information exchange to the five population sets show similar convergence rate with
form a search algorithm with some of the innovative flair different computational time. Accordingly, the population
of human search. While heuristic search methods such as size of 10 individuals is selected. The flowchart of MGAs
simulated annealing or taboo search use one solution on their used in this study is illustrated in Fig. 7.
process to find the optimum point, GAs use the population The modified MGA with multi point crossover shows
of solutions to find the optimum point. It is known that with better performance than the algorithms with one point
mathematical optimization methods which use a gradient crossover. In the MGAs, there are two stopping criteria
vector and Hessian it is difficult to find the optimum point called inner and outer criteria. If the solution does not evolve
if there are a lot of local optima around the optimum point within 10 inner loops, the inner loop is terminated, and the
and a steep gradient around the optimum point [7]. GAs do outer loop is performed until the total generation reaches 500
not use a gradient vector and Hessian, but use object function times. Therefore, the number of inner and outer loops varies,
value during their search. In its standard form, application of but the total number of generations is fixed. Moreover, the
a GA requires the representation of design variables in terms crossover rate is set as 1.0; therefore all populations must
of bit strings that are counterparts of natural chromosomes, perform a crossover operation at every generation. There is
made up of a string of genes. no use of mutation operation because the restart operation
It is known that more than 30 individuals should be used and 1.0 crossover rate provide adequate variability.
in Genetic Algorithms in order to prevent genetic drift [8].
As the population size increases, the algorithms find a better
solution. A bigger population size, however, requires more 5. Optimization of the cold-formed steel channel beam
computational time to find the optimum solution [6]. For this section
reason, Goldberg [6,7] proposed Serial Genetic Algorithms
(SGAs) which use a small population size comparing In this parametric study, a simply supported cold-formed
to conventional GAs. Based on SGAs, Krishnakumar [3] steel channel beam under uniformly distributed load (q) is
proposed Micro Genetic Algorithms (MGAs) in 1989. considered as shown in Fig. 8. The load is assumed to be
MGAs use a relatively smaller population size than SGAs applied at the shear center, and thus no flexural–torsional
resulting in less computational time. Moreover, MGAs use coupling is generated prior to lateral buckling.
J. Lee et al. / Engineering Structures 27 (2005) 17–24 23

Fig. 8. A simply supported cold-formed steel channel beam under uniformly


distributed loading.

Fig. 10. Optimum design thickness versus span length with no bearing
length.

Fig. 9. Convergence curve of the cross-sectional area.

The span length of the beam is assumed to be varied from


1 to 5 m. The yield stress and modulus of elasticity of steel
are 240 MPa and 210 GPa, respectively. In the optimum
design process of the cold-formed steel channel beam, the
depth of the web (D) and the thickness of the section (t)
are considered as design variables while the flange length Fig. 11. Optimum design web flat-depth-to-thickness ratio versus span
(B) and corner radius (R) are fixed to 4.0 and 0.5 cm, length with no bearing length.
respectively (Fig. 6). The upper and lower bounds of D
and t are assumed to be 10.0 cm ≤ D ≤ 30.0 cm and Range B: Lateral buckling governs the behavior of the
0.07 cm ≤ t ≤ 0.3 cm considering the manufacturing. The beams. While the depth of the web is fixed, the
flange length (B) and corner radius (R) are fixed because thickness increases remarkably to satisfy the lateral
it is shown from the preliminary studies that they are less buckling strength constraint. Accordingly, the flat-
sensitive than the depth of the web and thickness on the depth-to-thickness ratio of the web decreases
strength of cold-formed steel channel beams. greatly compared with range A and the web is still
The convergence curve of the cross-sectional area is fully effective.
illustrated with respect to the generation size in Fig. 9 for Range C: Only the web depth increases remarkably to
L = 4.0 m and q = 0.3 kN/m. Three different simulations satisfy the lateral buckling constraints because the
were performed with error less than 3.0%. thickness has already reached the upper bound.
The optimum thickness and the flat-depth-to-thickness The minimum cross-sectional area versus span length for
ratio of the web with respect to the span length are shown in cold-formed steel channel beams is illustrated in Fig. 12.
Figs. 10 and 11 for various intensity of the load. Among the Generally, the cross-sectional area increases as the span
structural behavior of beams (i.e. moment, shear, deflection, length increases. Especially, it increases greatly in range C
web crippling, etc.), cold-formed steel channel beams yield where the web depth increases remarkably to satisfy all the
to web crippling in the range A where the span length is constraints.
relatively short and to moment in the range of B and C. Each
range can be explained as follows.
6. Conclusion
Range A: While the web depth is fixed, the thickness
increases to satisfy the web crippling strength In this study, the shape optimization of cross-sections
constraint. Accordingly, the flat-depth-to-thickness of cold-formed steel channel beams under uniformly
ratio of the web decreases with increasing span distributed load was presented by using MGAs. The
length and the web is fully effective. optimum solution is obtained by using a small population
24 J. Lee et al. / Engineering Structures 27 (2005) 17–24

span. The study for more various sections of cold-formed


steel channels awaits further attention.

Acknowledgment

The first author would like to express his gratitude


for the financial support by the Korea Institute of
Industrial Technology Evaluation and Planning through
Grant 10002825.

References

[1] Karimm A, Adeli H. Global optimum design of cold-formed steel hat-


Fig. 12. Optimum design cross-sectional area versus span length. shape beams. Thin-Walled Structures 1999;35:275–88.
[2] El-Kassas EMA, Mackie RI, El-Sheikh AI. Using neural networks in
cold-formed steel design. 2001. p. 1687–96.
compared with SGAs, and the parameters used in genetic [3] Krishnakumar K. Micro-genetic algorithms for stationary and non-
operation can be predetermined. MGAs showed excellent stationary function optimization. SPIE, Intelligent Control and
performance on the minimum weight design of cold- Adaptive Systems, vol. 1196. 1989. p. 282–96.
[4] American Iron and Steel Institute. North American Specification for the
formed steel channel beams which use two design variables
design of cold-formed steel structural members. 2001.
(depth of the web and thickness). It is found that simply [5] von Kármán T, Sechler EE, Donnell LH. The strength of thin plates in
supported cold-formed steel channel beams under uniformly compression. Transactions ASME, vol. 54. APM; 1932. p. 54–5.
distributed load are governed by bending strength and [6] Goldberg DE. Sizing populations for serial and parallel genetic
web crippling rather than the other structural constraints algorithms. In: Proceeding 3rd international conference on genetic
algorithms and their applications. 1989. p. 70–9.
(i.e., shear and deflection). For relatively short span length,
[7] Goldberg DE. Genetic algorithms in search, Optimization and machine
web crippling is governing the constraints, whereas the learning. Addison-Wesley; 1989.
bending strength governs the behavior of beams with longer [8] Mitchell M. An introduction to genetic algorithms. MIT Press; 1996.

You might also like