Advances in Engineering Software 41 (2010) 442450

Contents lists available at ScienceDirect

Advances in Engineering Software

journal homepage: www.elsevier.com/locate/advengsoft

Cost optimization of industrial steel building structures

S. Kravanja *, T. Zula
University of Maribor, Faculty of Civil Engineering, Smetanova 17, 2000 Maribor, Slovenia

a r t i c l e

i n f o

a b s t r a c t
The paper presents the simultaneous cost, topology and standard cross-section optimization of singlestorey industrial steel building structures. The considered structures are consisted from main portal frames, which are mutually connected with purlins. The optimization is performed by the mixed-integer non-linear programming approach, MINLP. The MINLP superstructure of different structure/topology and standard cross-section alternatives has been generated and the MINLP optimization model of the structure has been developed. The dened cost objective function is subjected to the set of (in)equality constraints known from the structural analysis. Internal forces and deections are calculated by the elastic rst-order analysis constraints. The dimensioning constraints of steel members are dened in accordance with Eurocode 3. The modied outer-approximation/equality-relaxation (OA/ER) algorithm, a two-phase MINLP strategy and a special prescreening procedure of discrete alternatives are used for the optimization. A numerical example of the cost optimization of a single-storey industrial steel building is presented at the end of the paper. 2009 Elsevier Ltd. All rights reserved.

Article history: Available online 17 April 2009

Keywords: Cost optimization Structural optimization Topology optimization Cross-section optimization Mixed-integer non-linear programming MINLP Industrial building Steel structures

1. Introduction Structural engineers and designers are in the daily engineering praxis required to design the cheapest possible structures with the minimum amount of used material and technical equipment. The use of modern optimization methods thus becomes a great opportunity in the area of structural engineering. Single-storey industrial steel building structures are probably the most frequently built type of structures among various skeletal framed steel constructions. Many different optimization approaches have been proposed in the near past for the optimization of these structures. E.g. Lee and Knapton [1] have performed a constrained non-linear cost optimization of steel portal framed building. OBrien and Dixon [2] have proposed a linear programming approach for the optimal design of pitched roof frames. Gurlement et al. [3] have introduced a practical method for single-storey steel structures, based on a discrete minimum weight design and Eurocode design constraints. Saka [4] has considered an optimum design of pitched roof steel frames with haunched rafters by using a genetic algorithm. Kamal et al. [5] have carried out a weight optimization of two-hinged steel portal frames under multiple loadings. One of the latest researches reported in this eld is the work of Hernndez et al. [6], where authors have considered minimum weight design of steel portal frames with software developed for structural optimization.
* Corresponding author. Tel.: +386 2 2294 300; fax: +386 2 2524 179. E-mail addresses: stojan.kravanja@uni-mb.si (S. Kravanja), tomaz.zula@ uni-mb.si (T. Zula). 0965-9978/$ - see front matter 2009 Elsevier Ltd. All rights reserved. doi:10.1016/j.advengsoft.2009.03.005

This paper deals with the simultaneous cost, topology and standard cross-section optimization of single-storey industrial steel building structures. The considered building structures are consisted from main portal frames, which are mutually connected with purlins. The task of the optimization is to nd the minimal structures material and labour costs, the optimal topology with the optimal number of portal frames and purlins as well as the optimal standard cross-sections of steel members. The optimization is performed by the mixed-integer non-linear programming (MINLP). The MINLP is a combined discrete and continuous optimization technique. It handles with continuous and discrete binary 01 variables simultaneously. While continuous variables are dened for the continuous optimization of parameters (dimensions, stresses, strains, weights, costs, etc.), discrete variables are used to express discrete decisions (topology and standard cross-section alternatives). Since continuous and discrete optimizations are carried out simultaneously, the MINLP approach also nds optimal continuous parameters (e.g. structural costs), structural topology and discrete standard sizes simultaneously. The MINLP discrete/continuous optimization problems of such framed building structures are in most cases comprehensive, non-convex and highly non-linear. The optimization requires the generation of the buildings MINLP superstructure of different topology and standard cross-section alternatives and the development of the MINLP optimization model. Since the objective of the optimization is to minimize the structures self-manufacturing costs, the cost objective function has been dened. It comprises the material, fabrication and anti-corrosion protection painting costs as well as the assembling and erection costs of the structure.

 S. Kravanja, T. Zula / Advances in Engineering Software 41 (2010) 442450

443

The cost objective function is subjected to the set of equality and inequality constraints known from the structural analysis. Internal forces and deections are calculated by the elastic rst-order analysis constraints. The dimensioning constraints of steel members are dened in accordance with Eurocode 3 [7] for the conditions of both the ultimate and serviceability limit states. The modied outer-approximation/equality-relaxation algorithm is used to perform the optimization, see Kravanja and Grossmann [8], Kravanja et al. [911]. The two-phase MINLP optimization is proposed. It starts with the topology optimization, while the standard dimensions are temporarily relaxed into continuous parameters. When the optimal topology is found, the standard dimensions of the cross-sections are re-established and the simultaneous discrete topology and standard dimension optimization of the beams, columns and purlins is then continued until the optimal solution is found. In order to reduce a high number of structure alternatives and enable a normal solution of the MINLP, a special prescreening procedure has been developed, which automatically reduces the binary variables of alternatives into a reasonable number. The optimization at the second phase includes only those 01 variables which determine the topology and standard dimension alternatives close to the values, obtained at the rst MINLP optimization phase. 2. MINLP model formulation The MINLP optimization of the industrial steel building needs the generation of the buildings MINLP superstructure, which is composed of various topology and discrete design alternatives that are all candidates for a feasible and optimal solution. While topology alternatives represent different selections and interconnections of corresponding structural elements portal frames and purlins, discrete design alternatives include different standard cross-sections of columns, beams and purlins. The MINLP superstructure is modeled according to the MINLP model formulation. It is assumed that a general non-convex and non-linear discrete/continuous optimization problem can be formulated as an MINLP problem in the form:

deection, etc. constraints known from the structural analysis. Logical constraints that must be fullled for discrete decisions and structure congurations, which are selected from within the superstructure, are given by By + Cx 6 b. These constraints describe relations between binary variables, restore interconnection relations between currently selected or existing structural elements (corresponding y = 1) and cancel relations for currently rejected or nonexisting elements (corresponding y = 0), dene continuous design variables for each existing structural element and dene the structural topology and standard cross-sections of elements. It should be noted, that the comprehensive MINLP model formulation for mechanical structures may be found elsewhere, see Kravanja et al. [12].

3. Optimization model The single-storey industrial steel building structure is consisted from equal main portal frames, mutually connected with equal purlins, see Fig. 1. Each the portal frame is constructed from two columns and two beams. Purlins run continuously over the portal frames. Columns, beams and purlins are proposed to be built up from steel standard hot rolled European wide ange I sections (HEA sections), see Fig. 2. The global building geometry (including the frame span Lf, the building length LTOT, the column height HC and the overhight f) is proposed to be x through the optimization. The vertical and horizontal bracing systems as well as the wall sheeting rails are not included in this optimization. On the basis of the mentioned MINLP model formulation, the MINLP optimization model Single-Storey Industrial Steel Building OPTimization (SSISBOPT) has been developed for the cost optimization of the industrial steel building structures. As an interface for mathematical modeling and data inputs/outputs general algebraic modeling system (GAMS), a high level language by Brooke et al. [13] is used. The optimization model comprises input data, continuous and discrete binary variables, the structures cost objective function, structural analysis constraints and logical constraints. The cost objective function is subjected to the set of (non)linear structural analysis constraints and linear logical constraints. Input data comprises sets for the topology and cross-section alternatives, scalars and parameters. Dened are m, m2M, number of purlins; n, n2N, number of portal frames; i, i2I; j, j2J; and k, k2K, standard cross-section alternatives for columns, beams and purlins separately. Scalars in input data include the industrial building global geometry: the frame span Lf, the length of the industrial building LTOT, the height of the column HC and the overheight of the frame beam f. The yield strength of structural steel fy, the elastic modulus of steel E, the shear modulus of steel G, the density of steel q, the mass of the roof gr, snow s, the vertical wind wv, the horizontal wind wh, the partial safety factor for permanent load cg (1.35), the partial safety factor for variable load cq (1.50), the resistance partial safety factors cM0 (1.10) and cM1 (1.10), the price of the structural steel Cmat, the price of the anti-corrosion and re protection painting Cpaint, the erection price of the portal frame Cerect,frame, the erection price of the purlin Cerect,purlin, the coefcient for calculating the fabrication costs Cfabr, etc. are dened as input data. Parameters in input data comprise the vectors of different discrete alternative constants, e.g. qAC , a vector of i, i2I, discrete stani dard cross-section area alternatives for columns; qAB , a vector of j, j j2J, discrete standard cross-section area alternatives for beams; and qAP , a vector of k, k2K, discrete standard cross-section area k alternatives for purlins. Similarly are dened all other cross-section constants for heights, breadths, the web and ange thickness, the second moment of areas, the warping constants, etc.

min s:t:

z cT y f x hx 0 gx 0 By Cx b x xUP g MINLP

x 2 X fx 2 Rn : xLO y 2 Y f0; 1gm

where x is a vector of continuous variables specied in the compact set X and y is a vector of discrete, binary 01 variables. Functions f(x), h(x) and g(x) are non-linear functions involved in the objective function z, equality and inequality constraints, respectively. All functions f(x), h(x) and g(x) must be continuous and differentiable. In the context of structural optimization, continuous variables x dene structural parameters (actions, dimensions, stresses, deections, costs,. . .) and binary variables y represent the potential existence of structural elements within the dened superstructure. An extra binary variable y is assigned to each structural element. The element (the portal frame or purlin) is then selected to compose the structure if its subjected binary variable takes value one (y = 1), otherwise it is rejected (y = 0). Binary variables also dene the choice of discrete/standard cross-sections. The economical objective function z involves xed costs in the term cTy, while the dimension dependant costs are included in the function f(x). Non-linear equality and inequality constraints h(x) = 0, g(x) 6 0 and the bounds of the continuous variables represent the rigorous system of the design, loading, resistance, stress,

444

S. Kravanja, T. Zula / Advances in Engineering Software 41 (2010) 442450

f P P P

ef ef ef

ep
ef ef
Lf

ef

LT

OT

Fig. 1. Single-storey industrial steel building.

LB

q P AB IB AB I B f

HEA SECTION:
z HC H TOT

AC IC

A C IC

t f,c

tw,c

Lf
Fig. 2. Portal frame and the cross-sections of elements.

bC

Continuous variables include the self-manufacturing costs of the building structure COSTS, the number of portal frames NOFRAME, the number of purlins NOPURLIN, the intermediate distance between the portal frames ef, the intermediate distance between purlins ep, the length of the beam LB; the overall breadths of beam, column and purlin bB, bC, bP; the cross-sectional heights of beam, column and purlin hB, hC, hP; the ange thickness of beam, column and purlin tf,B, tf,C, tf,P; the web thickness of beam, column and purlin tw,B, tw,C, tw,P; the cross-section areas of beam, column and purlin AB, AC, AP; the torsional constants of beam, column and purlin It,B, It,C, It,P; the second moments of areas about yy and zz axes for beam, column and purlin Iy,B, Iz,B, Iy,C, Iz,C, Iy,P, Iz,P; the elastic section modules Wel,y,B, Wel,y,C, Wel,y,P, Wel,z,P; the warping constants for beam, column and purlin Ix,B, Ix,C, Ix,P; the design bending moments MSd,B, MSd,C, MSd,P, the design axial forces NSd,B, NSd,C, and the design shear forces VSd,B, VSd,C, VSd,P for beam, column and purlin separately; the uniformly distributed vertical surface variable loads q, qy and qz; the concentrated horizontal variable load P; the elastic critical moment for lateral-torsional buckling MCR; the vertical deections of beams dmax,B, d2,B, and purlins dmax,P, d2,P, as well as the horizontal deections of the portal frames dhoriz,F. Binary variables include binary variables yi of i, i2I, standard cross-section alternatives of columns; binary variables yj of j, j2J, standard cross-section alternatives of beams; binary variables yk of k, k2K, standard cross-section alternatives of purlins; binary variables ym assigned to m, m2M, topology alternatives of purlins; and binary variables yn assigned to n, n2N, topology alternatives of portal frames.

Cost objective function is dened by Eq. (1). It comprises the material costs, the fabrication costs, the anti-corrosion and re (R 30) protection painting costs and the erection costs of the structure:

min COST NOFRAME Volframe NOPURLIN Volpurlin q C mat 1 C fabr NOFRAME Aframe NOPURLIN Apurlin C paint NOFRAME C erect;frame NOPURLIN C erect;purlin 1

where COST [EUR] represents the self-manufacturing costs of the building structure, NOFRAME and NOPURLIN denote the numbers of the portal frames and purlins, Volframe [m3] and Volpurlin [m3] stand for the volumes of each frame and purlin, q denotes the steel density [kg/m3], Cmat is the price of the structural steel [EUR/kg], Aframe [m2] and Apurlin [m2] represent the surface areas of each frame and purlin, Cpaint [EUR/m2] is the price of the anti-corrosion and re protection painting, Cerect,frame [EUR] is the erection price for each portal frame and Cerect,purlin [EUR] represents the erection price for each purlin. The fabrication costs of steel elements are calculated with a factor Cfabr regarding the obtained material costs. Structural analysis constraints comprise the calculation of loads, internal forces and deections as well as the checking the ultimate end serviceability limit state constraints. Considered is a single load case only, where the partial safety factors and combination of actions are dened according to Eurocodes. The optimization of the structure is performed under the combined effects of:

hC

H TOT

HC

 S. Kravanja, T. Zula / Advances in Engineering Software 41 (2010) 442450

445

 the self-weight of the structure (the line uniform load of columns, beams and purlins) and the weight of the roof gr (the vertical surface load) plus  snow s and the vertical wind wv (the uniformly distributed vertical surface variable load) plus  the horizontal wind wh (the horizontal force at the top of the columns P). The horizontal concentrated load at the top of the columns P and the vertical uniformly distributed line load on beams and purlins, caused by the self-weight and the vertical wind, are calculated automatically through the optimization considering the calculated intermediate distance between the portal frames and purlins. The design bending moments Msd, axial forces Nsd and shear forces Vsd as well as the deections are calculated by the elastic rst-order method. The portal frames are classied as a non-sway steel portal frames. The ratio between the design value of the total vertical load NSd,C and the elastic critical value for failure in a sway mode Ncr is constrained: NSd,C/Ncr 6 0.1. The portal frame is calculated as a laterally supported frame. Hereby, the steel members are checked only for the in-plane instability. The dimensioning constraints are determined according to Eurocodes 3 (ultimate and serviceability limit states). When the ultimate limit state of structural members is considered, the elements are checked for axial resistance, shear resistance, bending moment resistance and the interaction between the bending moment and the axial force. Columns are additionally designed for the compression/buckling resistance plus the lateral-torsional buckling. Buckling lengths of columns are calculated as the in-plane buckling lengths for the non-sway mode. Beams are checked for the in-plane bending moment resistance, the compression axial force resistance and for the interaction. Purlins are checked for the interaction between bending moments about both yy and zz axes. The design bending moment resistances about the in-plane yy axis of beams, columns and purlins are checked with the inequality constraints, dened by Eq. (2).

NSd;C My;Sd;C

AC f y

cM1

vLT

W el;y;C fy

cM1

The interaction in column between the compression/buckling and the bending/lateral-torsional buckling is dened by Eq. (8). It is supposed that buckling effects in beams are not decisive due to the adequate resistance of the horizontal bracing systems in the roof. Beams are thus checked for the interaction between the axial compressed force and the bending moment, Eq. (9).

NSd;C
f v AcCM1y

kLT My;Sd;C

vLT W el;y;C fy cM1

W el;y;B fy

1:0 1:0

8 9

NSd;B
AB fy

ky M y;Sd;B
cM0

cM0

The shear resistances of beams, columns and purlins are dened by Eq. (10). VSd,B, VSd,C and VSd,P denote the design shear forces in beams, columns and purlins; Av,B, Av,C and Av,P are the effective shear areas of cross-sections (e.g. for beam: 1.04hBtw,B).

V Sd;B V Sd;P

fy 1 Av ;B p ; 3 cM0 fy 1 Av ;P p 3 cM0

V Sd;C

fy 1 Av ;C p ; 3 cM0 10

M y;Sd;B

W el;y;B fy

cM0

M y;Sd;C

W el;y;C fy

cM0

M y;Sd;P

W el;y;P fy

cM0

Purlins are in addition checked for the design bending moment resistance about the zz axis, Eq. (3), and for the interaction between both the bending moments, see Eq. (4).

Considering the serviceability limit state, the vertical deections of portal frames and purlins are calculated and checked. The total deections dmax subjected to the overall load and the deections d2 subjected to the variable imposed load are calculated to be smaller than limited maximum values: span/200 and span/250, respectively. The horizontal frame deections dhoriz,F are also checked for the recommended limits: the relative horizontal deections should be smaller than the portal height/150. In this way, Eqs. (11) and (12) respresent the inequality constraints for checking the vertical deections of portal frames (i.e. beams) dmax,B, d2,B, and purlins dmax,P, d2,P. The horizontal deections of portal frame dhoriz,F are veried by Eq. (13). Lf denotes the span of the portal frame, ef is the intermediate distance between the portal frames and HC stands for the height of the column.

dmax;B d2;B dhoriz;F

M z;Sd;P
W el;y;P fy

W el;z;P fy

cM0
kz M z;Sd;P
W el;z;P fy

3 1 4

ky My;Sd;P
cM0

Lf ; 200 Lf ; 250 HC 150

dmax;P d2;P

ef 200

11 12 13

ef 250

cM0

My,Sd,B, My,Sd,C, My,Sd,P and Mz,Sd,P stand for the design bending moments about the yy and zz axes for beams, columns and purlins; fy represents the yield strength of structural steel; Wel,y,B, Wel,y,C, Wel,y,P and Wel,z,P are the elastic section modules; coefcients ky and kz are equal 1 (no axial forces in purlins); and cM0 is the resistance partial safety coefcient. The resistances to the design axial forces of beams and columns are achieved by Eq. (5). NSd,B and NSd,C represent the design compression axial forces in beams and columns, and AB and AC are the cross-section areas of beams and columns.

NSd;B

AB fy

cM0

NSd;C

AC fy

Integer and mixed-integer logical constraints represent linear constraints, which determine the structure topology and standard cross-sections of steel elements. In this way, Eq. (14) denes the number of portal frames NOFRAME as a sum of binary variables yn, which are assigned to the portal frames. Only the frames with the calculated non-zero binary variables (yn = 1) are participated in the variable NOFRAME and in the building structure. Eq. (15) enables the determining the only one possible vector of binary variables for each frame topology. Eq. (16) calculates the even number of purlins, where binary variables ym are assigned to the purlins. Eq. (17) denes the only one possible vector of binary variables for each purlin topology.

cM0

NOFRAME yn yn1

X
n

yn X
m

14 15 ym 16 17

While the compression/buckling resistance of columns is checked by Eq. (6), the bending/lateral-torsional buckling is determined by Eq (7). v is the reduction factor due to the exural buckling, vLT is the reduction factor due to the lateral-torsional buckling, see Eurocode 3 [7], and cM1 is the resistance partial safety coefcient.

NOPURLIN 2 ym ym1

446

S. Kravanja, T. Zula / Advances in Engineering Software 41 (2010) 442450

Eqs. (18)(20) calculate the standard cross-section characteristics for columns. The cross-section area of column AC is determined as a scalar product between the vector of i, i2I, discrete standard cross-section area alternatives qAC and the vector of i associated bini ary variables yi, see Eq. (18). Only one discrete value is selected for each standard cross-section area since the sum of the binary variables yi has to be equal to 1, see Eq. (20). In the similar way is then calculated the elastic section modulus Wel,y,C, see Eq. (19), and all other cross-section dimensions and characteristics.

MINLP Superstructure Combined optimization Fixed binary variables

AC

X
i

qAC yi ; i X
i

i2I i2I

18 19

NLP Subproblem Continuous optimization

W el;y;C X
i

qW C y i ; i

yi 1

20
MILP Master problem Discrete optimization

New binary variables

Similarly, Eqs. (21)(23) determine the discrete values of the crosssectional characteristics for the frame beams and Eqs. (24)(26) for purlins.

AB

X
j

qAB yj ; j X
j

j2J j2J

21 22 23

W el;y;B X
j

qW B y j ; j

Convergence ? NO YES

yj 1 X
k

AP

qAP yk ; k X
k

k2K k2K

24
STOP

W el;y;P X
k

qW P y k ; k

25 26

Fig. 3. Steps of the OA/ER algorithm.

yk 1

4. MINLP optimization A general MINLP class of optimization problem can be solved in principle by the following algorithms and their extensions:  the Non-linear Branch and Bound, NBB, proposed and used by many authors, e.g. Beale [14], and Gupta and Ravindran [15];  the Sequential Linear Discrete Programming method, SLDP, by Olsen and Vanderplaats [16] and Bremicker et al. [17];  the Extended Cutting Plane method by Westerlund and Pettersson [18];  the Generalized Benders Decomposition, GBD, by Benders [19], Geoffrion [20];  the Outer-Approximation/Equality-Relaxation algorithm, OA/ER, by Kocis and Grossmann [21];  the Feasibility Technique by Mawengkang and Murtagh [22]; and  the LP/NLP based Branch and Bound algorithm by Quesada and Grossmann [23]. Since the optimization problems of the industrial steel building structures are comprehensive, non-convex and highly non-linear, the modied outer-approximation/equality-relaxation (OA/ER) algorithm by Kravanja and Grossmann [8] is used to perform the MINLP optimization. The OA/ER algorithm consists of solving an alternative sequence of non-linear programming optimization subproblems (NLP) and mixed-integer linear programming master problems (MILP), see Fig. 3. The former corresponds to the optimization of parameters for a framed structure with a xed topology and standard sizes and yields an upper bound to the objective to be minimized. The latter involves a global approximation to the superstructure of alternatives in which a new topology and standard sizes are identied so that its lower bound does not exceed

the current best upper bound. The search of a convex problem is terminated when the predicted lower bound exceeds the upper bound, otherwise it is terminated when the NLP solution can be improved no more. The OA/ER algorithm guarantees the global optimality of solutions for convex and quasi-convex optimization problems. The OA/ER algorithm as well as all other mentioned MINLP algorithms do not generally guarantee that the solution found is the global optimum. This is due to the presence of non-convex functions in the models that may cut off the global optimum. In order to reduce undesirable effects of non-convexities, the following modications are applied for the master problem of the OA/ER algorithm: the deactivation of linearizations, the decomposition and the deactivation of the objective function linearization, the use of the penalty function, the use of the upper bound on the objective function to be minimized as well as the global convexity test and the validation of the outer approximations. The optimal solution of a complex non-convex and non-linear MINLP problem with a high number of discrete decisions is in general very difcult to obtain. The optimization is thus proposed to be performed sequentially in two different phases to accelerate the convergence of the OA/ER algorithm. The optimization is proposed to start with the discrete topology optimization of the building, while the standard dimensions are temporarily relaxed into continuous parameters. Topology and continuous parameter optimization is soluble (a smaller combinatorial problem) and accumulates a good global linear approximation of the superstructure (a good starting point for the next phase overall optimization). When the optimal topology is found, the standard sizes of the cross-sections are re-established and the simultaneous discrete optimization of the topology and standard dimensions of the beams, columns and purlins is then continued until the optimal solution is found. The two-phase strategy requires that the binary variables should be dened in one uniform set. In the rst phase, only the

 S. Kravanja, T. Zula / Advances in Engineering Software 41 (2010) 442450

447

binary variables which are assigned to topology alternatives become active. Binary variables of standard dimension alternatives are temporarily excluded (set on value zero) until the beginning of the second phase, in which they participate in the simultaneous overall optimization. The same holds for standard dimension logical constraints. In the rst phase they are excluded, while the second phase includes them into the optimization. The data and variables are initializated only once in the beginning of the optimization. An advantage of this strategy is also in the fact that binary variables for topology and standard dimensions need not be initialized: after the rst NLP, the rst phase always starts in the subspace of the topological binary variables only, while the second phase starts with the MILP master subproblem which then predicts a full set of binary variables for the successive NLP. Under the convexity condition, the two-phase strategy guarantees a global optimality of the solution. The optimization model may contain up to thousand binary 01 variables of the alternatives. Most of them are subjected to standard dimensions. Since this number of 01 variables is too high for a normal solution of the MINLP, a reduction procedure was developed, which automatically reduces the binary variables of alternatives into a reasonable number. The optimization at the second phase includes only those 01 variables which determine the topology and standard dimension alternatives close to the values, obtained at the rst MINLP optimization phase. 5. Example The paper shows an example of the simultaneous cost, topology and standard dimension optimization of a single-storey industrial steel building. The building was 24 m wide, 82 m long and 5.5 m high, see Fig. 4. The overheight of the frame beam was 0.45 m. The portal frame was subjected to the self-weight of structure and roof g, to the uniformly distributed variable load of snow s and vertical wind wv as well as to the concentrated variable load P at the top of columns (caused by the horizontal wind wh). The weight of the roof was gr = 0.22 kN/m2. Snow s = 2.20 kN/m2, the vertical wind wv = 0.12 kN/m2 and the horizontal wind wh = 0.45 kN/m2 were dened in the model input data as the variable loads.

The material used was steel S 355. The yield strength of the steel fy is 35.5 kN/cm2, the density of steel q is 7850 kg/m3, the elastic modulus of steel E is 21000 kN/cm2 and the shear modulus G is 8076 kN/cm2. An industrial building superstructure was generated in which all possible constructional variations were embedded by 30 portal frame alternatives, 10 various purlin alternatives and a variation of different standard cross-sections. In this way, the superstructure consisted of n possible number of portal frames, n2N, N = {1,2,3,. . .,30}, and 10 various even (2m) number of purlins, m2M, M = {1,2,3,. . .,10}, which gave 30 10 = 300 different topology alternatives. In addition, the superstructure comprised also 24 different standard hot rolled European wide ange I sections, i.e. HEA sections (from HEA 100 to HEA 1000) for each column, beam and purlin separately. Three identical vectors of discrete cross-section area alternatives qAC ; qAB ; qAP were dened for coli j k umns, beams and purlins:

qAC qAB qAP i j k f21:2; 25:3; 31:4; 38:8; 45:3; 53:8; 64:3; 76:8; 86:8; 97:3; 113:0; 124:0; 133:0; 143:0; 159:0; 178:0; 198:0; 212:0; 226:0; 242:0; 260:0; 286:0; 321:0; 347:0g
Since i, j and k, i2I, j2J, k2K, I = J = K = {1,2,3,. . .,24}, different standard sections were dened for columns, beams and purlins separately, there existed n m i j k = 30 10 24 24 24 = 4147200 different discrete constructional alternatives altogether. The optimization was performed by the proposed MINLP optimization approach. The task of the optimization was to nd the minimal structures material and labour costs, the optimal topology with the optimal number of portal frames and purlins as well as the optimal standard cross-sections of members. The economical objective function included the material, anti-corrosion and re (R 30) protection painting as well as assembling and erection costs of the structure. The economic data for the optimization are presented in Table 1. The fabrication costs of steel elements were calculated to be equal to 40% of the obtained material costs (Cfabr = 0.40), see also Eq. (1). The optimization was carried out by a user-friendly version of the MINLP computer package MIPSYN [2426], the successor of

0.45 m
,0 82 m

24,0 m
Fig. 4. Global geometry of the single-storey industrial building.

5.05 m

5.5 m

448 Table 1 Economic data for the optimization. Cmat Cpain Cerect,frame Cerect,purlin

S. Kravanja, T. Zula / Advances in Engineering Software 41 (2010) 442450

Material costs for structural steel S 355 Anti-corrosion resistant painting costs (R30) Erection costs per 1 portal frame Erection costs per 1 purlin

1.05 22.50 450.00 250.00

EUR/kg EUR/m2 EUR/frame EUR/purlin

PROSYN [8] and TOP [911]. The Modied OA/ER algorithm and the two-phase optimization were applied, where GAMS/CONOPT2 (Generalized reduced-gradient method) [27] was used to solve the NLP subproblems and GAMS/Cplex 7.0 (Branch and Bound) [28] was used to solve the MILP master problems. At the initialization, i.e. at the rst NLP, the optimization model SSISBOPT contained 111 (in)equality constraints, 61 continuous and 112 binary variables (30 binary variables yn for the topology optimization of portal frames, 10 binary variables ym for the topology optimization of purlins, 24 binary variables yi for the standard

cross-section optimization of columns, 24 binary variables yj for cross-section optimization of beams and 24 binary variables yk for cross-section optimization of purlins). The two-phase MINLP optimization was applied. After the rst performed continuous NLP (the initialization), the rst phase started with the discrete topology optimization at the relaxed standard dimensions, see also the convergence of the Modied OA/ER algorithm in Table 2. At this level, only the binary variables yn and ym for topology optimization, the parameter structural nonlinear and linear constraints as well as the logical constraints for topology optimization were included. When the optimal topology was reached (219427 EUR at the 2nd MINLP iteration, all the following solutions were poorer), the optimization proceeded with a simultaneous discrete topology and standard dimension optimization at the second level. At this phase, the binary variables yi, yj and yk of standard sizes for columns, beams and purlins, as well as the logical constraints for standard cross-sections were added into the optimization. The nal

Table 2 Convergence of the Modied OA/ER algorithm. MINLP Iteration MINLP Subphaze Result costs () Topology Frames Phase 1: Topology optimization 1. Initialization 1. NLP 2. 1. MILP 2. NLP 3. 2. MILP 3. NLP Phase 2: Topology and standard cross-section optimization 4. 3. MILP 4. NLP* 5. 4. MILP 5. NLP* 6. 5. MILP 6. NLP 7. 6. MILP 7. NLP 8. 7. MILP 8. NLP 9. 8. MILP 9. NLP
*

Cross-sections (cm2) Purlins 11.60 12 12 Column 281.498 317.778 276.595 316.251 270.952 321.00 HEA 900 321.00 HEA 900 321.00 HEA 900 321.00 HEA 900 321.00 HEA 900 321.00 HEA 900 Beam 206.503 176.074 202.229 173.136 197.345 198.00 HEA 500 198.00 HEA 500 198.00 HEA 500 212.00 HEA 550 212.00 HEA 550 212.00 HEA 550 Purlin 40.015 35.049 35.544 32.464 31.691 38.80 HEA 160 38.80 HEA 160 38.80 HEA 160 38.80 HEA 160 38.80 HEA 160 38.80 HEA 160

217,551 215,788 219,427 221,762 221,263 240,657 240,965 229,431 229,431 241,090 240,757 252,407 252,291 235,619 235,619 246,946 246,946

13.20 14 15

15 14 13 13 14 14

12 12 14 16 12 14

Locally infeasible.

m .31 x6 m 13 2.0 8

24.0 m

Fig. 5. Optimal design of the single-storey industrial building.

5.5 m

 S. Kravanja, T. Zula / Advances in Engineering Software 41 (2010) 442450

449

HEA 160 HEA 550 HEA 550

HEA 900

HEA 900

5.05 m

0.45 m

10 x 2.4 m

24.0 m

Fig. 6. Optimal design of the portal steel frame.

optimal solution of 235619 EUR was obtained at the 8th main MINLP iteration (all the following solutions were not as good). At the second phase, where all the calculated dimensions were standard ones, a feasible optimal result was very difcult to be obtained. The optimization model contained the mentioned high number of 4147200 different discrete construction alternatives. In order to reduced the binary variables of alternatives into a reasonable number, the prescreening procedure of alternatives was thus automatically applied. The optimization at the second phase included only those 01 variables which determined the topology and standard dimension alternatives close to the values, obtained at the rst phase. In this way, only 4 binary variables (2 variables under and 2 over the calculated value from the previous phase) were used separately to determine the topology (the numbers of columns and purlins) and to calculate the standard cross-sections of columns, beams and purlins. Altogether only 20 binary variables were used in the second phase instead of all 112 binary variables. The number of 4147200 discrete construction alternatives was signicantly reduced to n m i j k = 4 4 4 4 4 = 1024 alternatives, which considerably improved the efciency of the search. The optimal result represented the obtained structures minimal material and labour costs of 235619 EUR. The selling price may be at least twice higher. The solution also comprised the calculated optimal topology of 14 portal frames and 12 purlins, see Fig. 5, and the calculated optimal standard sections of columns (HEA 900), beams (HEA 550) and purlins (HEA 160), see Fig. 6. The obtained structure mass was 121,558 kg. 6. Conclusions The paper presents the simultaneous cost, topology and standard cross-section optimization of single-storey industrial steel building structures. The considered structures are consisted from main portal frames, which are mutually connected with purlins. The optimization is performed by the mixed-integer non-linear programming approach, MINLP. The MINLP is a combined discrete and continuous optimization technique. It handles with continuous and discrete binary 01 variables simultaneously. While continuous variables are dened for the continuous optimization of parameters (dimensions, stresses, deections, weights, costs, etc.), discrete variables are used to express different structure/ topology and standard cross-section discrete decisions. An extra binary variable y is assigned to each structural element. The element (the portal frame or purlin) is then selected to compose the structure if its subjected binary variable takes value one (y = 1), otherwise it is rejected (y = 0). Binary variables also dene the choice of discrete/standard cross-sections.

The MINLP superstructure of different structure/topology and standard cross-section alternatives has been generated and the MINLP optimization model SSISBOPT for the structure optimization has been developed. Dened is a cost objective function, which comprises the material, fabrication, anti-corrosion protection painting and erection costs of the structure. The cost objective function is subjected to the set of (in)equality constraints known from the structural analysis. Internal forces and deections are calculated by the elastic rst-order analysis constraints. The dimensioning constraints of steel members are dened in accordance with Eurocode 3. The modied outer-approximation/equalityrelaxation (OA/ER) algorithm was used for the optimization. A two-phase MINLP strategy and a special prescreening procedure of discrete alternatives are developed to accelerate the convergence of the mentioned algorithm. A practical example of the simultaneous cost, topology and standard cross-section optimization of a 24 m wide and 82 m long single-storey industrial steel building structure is presented at the end of the paper. Beside the minimal structures manufacturing costs, the optimal topology with the optimal number of portal frames and purlins as well as all standard cross-sections of steel elements have been obtained. This example demonstrates the efciency of the introduced simultaneous MINLP approach. The use of such modern optimization techniques can essentially improve the economical competitive capabilities of engineers in the eld of structural engineering. Acknowledgements Funding for this research was provided by the Slovenian Research Agency and the Ministry of Higher Education, Science and Technology of the Republic of Slovenia, National Research Program P2-0129. References
[1] Lee BS, Knapton J. Optimum cost design of a steel framed building. Eng Optim 1975:13953. [2] OBrien EJ, Dixon AS. Optimal plastic design of pitched roof frames for multiple loading. Comput Struct 1997;64:73740. [3] Gurlement G, Targowski R, Gutkowski W, Zawidzka J, Zawidzki J. Discrete minimum weight design of steel structures using EC3 code. Struct Multidisc Optim 2001;22:3227. [4] Saka MP. Optimum design of pitched roof steel frames with haunched rafters by genetic algorithm. Comput Struct 2003;81:196778. [5] Kamal OA, El-Mahdy OO, El-Komy GA. Optimum design of one-bay portal steel frames. J Eng Appl Sci 2003:72341. [6] Hernndez S, Fontn AN, Perezzn JC, Loscos P. Design optimization of steel portal frames. Adv Eng Software 2005;36:62633. [7] Eurocode 3. Design of steel structures. European Committee for Standardization; 1992.

5.5 m

450

S. Kravanja, T. Zula / Advances in Engineering Software 41 (2010) 442450 [18] Westerlund T, Pettersson F. An extended cutting plane method for solving convex MINLP problems. In: European Symposium on Computer Aided Process Engineering, Supplement to Computers Chem Eng Bled Slovenia, vol. 5, 1995;S131S6. [19] Benders JF. Partitioning procedures for solving mixed-variables programming problems. Numer Math 1962;4:23852. [20] Geoffrion AM. Generalized benders decomposition. J Optim Theory Appl 1972;10(4):23760. [21] Kocis GR, Grossmann IE. Relaxation strategy for the structural optimization of process owsheets. Ind Eng Chem Res 1987;26:186980. [22] Mawengkang H, Murtagh BA. Solving nonlinear integer programs with largescale optimization software. Ann Operat Res 1986;5:42537. [23] Quesada I, Grossmann IE. An LP/NLP based branch and bound algorithm for convex MINLP optimization problems. Comput Chem Eng 1992;16:93747. [24] Kravanja S, Sorak A, Kravanja Z. Efcient multilevel MINLP strategies for solving large combinatorial problems in engineering. Optim Eng 2003;4:97151. [25] Iric Bedenik N, Pahor B, Kravanja Z. An integrated strategy for the hierarchical multilevel MINLP synthesis of overall process owsheets using the combined synthesis/analysis approach. Comput Chem Eng 2004;28:693706. [26] Iric Bedenik N, Ropotar M, Kravanja Z. MINLP synthesis of reactor networks in overall process schemes based on a concept of time-dependent economic regions. Comput Chem Eng 2007;31:65776. [27] Drudd AS. CONOPT a large-scale GRG code. ORSA J Comput 1994;6(2):20716. [28] CPLEX User Notes, ILOG Inc.

[8] Kravanja Z, Grossmann IE. New developments and capabilities in PROSYN An automated topology and parameter process synthesizer. Comput Chem Eng 1994;18(11/12):1097114. [9] Kravanja S, Kravanja Z, Bedenik BS. The MINLP optimization approach to structural synthesis. Part I: A general view on simultaneous topology and parameter optimization. Int J Num Meth Eng 1998;43:26392. [10] Kravanja S, Kravanja Z, Bedenik BS. The MINLP optimization approach to structural synthesis. Part II: Simultaneous topology, parameter and standard dimension optimization by the use of the Linked two-phase MINLP strategy. Int J Num Meth Eng 1998;43:293328. [11] Kravanja S, Kravanja Z, Bedenik BS. The MINLP optimization approach to structural synthesis. Part III: Synthesis of roller and sliding hydraulic steel gate structures. Int J Num Meth Eng 1998;43:32964. [12] Kravanja S, ilih S, Kravanja Z. The multilevel MINLP optimization approach to structural synthesis: the simultaneous topology, material, standard and rounded dimension optimization. Adv Eng Software 2005;36(9):56883. [13] Brooke A, Kendrick D, Meeraus A. GAMS A Users Guide. Redwood City, CA: Scientic Press; 1988. [14] Beale EML. Integer Programming. In: Jacobs D, editor. The state of the art in numerical analysis. London: Academic Press; 1977. p. 40948. [15] Gupta OK, Ravindran A. Branch and bound experiments in convex nonlinear integer programming. Manage Sci 1985;31(12):153346. [16] Olsen GR, Vanderplaats GN. Method for nonlinear optimization with discrete design variables. AIAA J 1989;27(11):15849. [17] Bremicker M, Papalambros PY, Loh HT. Solution of mixed-discrete structural optimization problems with a new sequential linearization method. Comput Struct 1990;37(4):45161.