Sunteți pe pagina 1din 6

Struct Multidisc Optim 22, 322327 Springer-Verlag 2001

Discrete minimum weight design of steel structures using EC3 code


G. Guerlement, R. Targowski, W. Gutkowski, J. Zawidzka and J. Zawidzki

Abstract The paper presents a practical method of the discrete minimum weight design of steel structures. It is based on a concept of removal of the redundant material by successive diminishing of the size of the least stressed member. It is assumed that the member sizes are available from the European Steel Proles Catalogues, and the design constraints are given by the rules of the code EC3 for nonsway buildings. Two numerical examples are presented: the classical benchmark problem of a ten-bar truss made of circular hollow sections (24 element catalogue, buckling taken into account) and a portal frame made of HEB sections with broad parallel anges (11 element catalogue). The proposed method gives, after a reasonable calculation time, a very good approach to the exact optimum solution. Key words discrete optimization, steel structures, EC3, minimum weight, design

1 Introduction A large number of steel trusses and frames are designed using prefabricated rolled proles listed in European Steel Proles Catalogues. The European Code 3, on the other hand, presents the requirements for analysis and design of steel structures. In this paper we present an approach to the minimum weight design of steel structures using the mentioned Eurocode (1992) or more explicitly its detailed explanation found in the design handbook prepared by Chantrain et al. (1996).
Received January 21, 2000 G. Guerlement1 , R. Targowski1 , W. Gutkowski2 , J. Zawidzka2 and J. Zawidzki2 Institute of Mechanics and Architecture, Polytechnic Faculty of Mons, B-7000 Mons, Belgium e-mail: dsgt@fpms.ac.be 2 Institute of Fundamental Technological Research, ul. Swietokrzyska 21, 00-049 Warsaw, Poland e-mail: jzawid@ippt.gov.pl
1

Conceptually, a design from a list of available crosssection areas (CSA) is very simple. It is necessary to verify all possible combinations arising from the number of structural members (SM) and available proles, taking the one with the smallest weight, and fullling imposed constraints. Practically, however, the problem is not as simple as it looks at a rst glance. This is because of a very large number of combinations occurring in practical problems. In such cases direct enumerations are impracticable. Consider a simple structure composed of 10 elements or linking groups. Assume moreover that a catalogue, from which we choose CSA for particular SM, contains a list of 10 rolled proles. This means that the number of all possible combinations is equal to 1010 . It is easy to check that if one analysis with verication of constraints lasted 0.1 s, the whole enumeration would last more than 30 years. There are several solution methods for discrete optimization problems. Some of them can be found in several review papers on discrete structural optimization (DSO) (Arora and Haug 1994; Bauer 1994; Thander and Vanderplaats 1995; Templeman 1997). However, most or even all exact methods do not work well with larger, occurring in DSO, numbers of combinations. The diculties in nding an eective and exact method are causing that emphasis in research is directed to approximated, but simpler and more eective methods. One of them, recently the most known, is the Genetic Algorithm (GA). The description of the method can be found in many papers and monographs. One of the rst attempts to apply GA to DSO is due to Goldberg and Samatini (1986). Without going into the detailed presentation of GA, it is important to mention two main disadvantages of the algorithm. The rst consists in an unspecied population number needed to nd the best solution. The second one is caused by coecients in a tting function which have to be found experimentally. The minimum weight design deserves much attention from engineers and scientists. Continuous optimization, when CSA must be chosen from a continuous set of values, is relatively well-recognized. Our knowledge of DSO is more limited. Due to its complexity, a trend can be observed searching for problem oriented methods, giv-

323 ing approximate but relatively eective solutions. Professional designers are looking for simple algorithms not requiring too much additional study. They agree to pay with exactness for getting solutions in reasonable time, and with reasonable accuracy. It is our aim to present such an algorithm for discrete minimum weight design. The idea of problem oriented, approximated methods for DSO is not a new one. It is worthy to mention some of already existing ones. A controlled enumeration method presented by Iwanow (1981) starts from continuous solution of minimum weight. It is considered as the lower bound for discrete solution. Then, investigating the graph of all possible combinations, a large number of combinations are eliminated from further investigation. A segmental method (Templeman 1997), applied to minimum weight design allows us to solve the problem by any standard linear programming algorithm. A conventional truss member is replaced by segmental member. The latter one is divided in a number of listed cross-section areas. The length of each part is a design variable of the problem. Recently Nha et al. (1998) proposed an approach in which removal of the redundant material is controlled by sensitivities of state variables with respect to design variables. 2 Problem statement The proposed method deals with minimum weight design of steel trusses and frames of given topologies. Structures are composed of prefabricated rolled proles, listed in European Steel Proles Catalogues. The design is based upon the sequential optimization algorithm proposed by Gutkowski and Zawidzka (1997), Guerlement et al. (1998). The main idea of the method consists in removing, at each iteration step, the redundant material from the structural member with the least stress intensity ratio. This way, the algorithm allows us to reduce the numj0 to k0 j0 , where ber of possible combinations from k0 k0 is a number of listed proles and j0 is the number of structural members. It is assumed that the considered structure is composed of j0 SMs. Each j th SM is assumed to have CSA Ak j , and other important parameters taken from the k th position of a catalogue containing a set of k0 values. As already mentioned, the direct enumeration of the possible number of structures arising from the number of SM and catalogue values is practically impossible. The proposed approach allows to nd minimum (near minimum) solution with no. more than j0 k0 analyses. At each iteration step of the algorithm, according to Chapter III of EC3, the global analyses of the structure are performed with the most unfavourable Load Cases (LC) for the Ultimate Limit States (ULS) and for the Serviceability Limit States (SLS). As results we get the ULS action eects Sd for each SM (axial force, bending moment, shear force) and the SLS action eects Ed for the whole structure (displacements). These values are then compared with the ULS resistances Rd and with the SLS limits Cd respectively, applying the EC3 check procedures. Then, for the purpose of our method, we determine the following ratios. Stress Intensity Ratio (SIR): w(j ) = max(Sd /Rd ) for each SM, Displacement Intensity Ratio (DIR): d = max(Ed /Cd ) for the whole structure. Note that the above notation is rather symbolic and the detailed form of all these checks is precisely dened by Code EC3. The SIR and DIR values control our optimization algorithm. Neither SIR and DIR should exceed 1 when the structure is to be adopted. The mathematical statement of the discussed problem can be presented as follows. Find CSA Aj of the j th SM from a catalogue containing k0 rolled proles, giving minimum (near minimum) of the structural weight,
j0

W =
j =1

j Aj

min ,

where is the material density and j is the length of the j th SM. The nal design must fulll constraints dened by EC3 code: for stresses from ULS checks SIR: w(j ) = max(Sd /Rd ) 1 for each SM, and for displacements, from SLS checks DIR: d = max(Ed /Cd ) 1 for the checked structural points.

3 Successive removal of the redundant material The algorithm applied in the method is as follows (the ow chart is shown in Fig. 1). (i) (ii) Read design data. Order all available CSA, from an assumed EC catalogue, with their decreasing values. For k = 1 the largest CSA is associated, and for k = k0 the smallest one. Assign to all SM the largest, available CSA (k = 1). It is assumed that such a structure fullls all constraints imposed on stresses and displacements. Use the ANALYSIS MODULE to nd ULS and SLS action eects according to EC3 code. Use the CHECK MODULE to verify SIR and DIR.

(iii)

(iv) (v)

324 (vi) Remove the material from the nonxed member with the smallest SIR by assigning to it next, smaller value of the available CSA. (vii) Check the obtained structure for constraints. If they are not violated go to (vi). (viii) If one or more constraints are violated, restore the previous CSA and x it. The CSA of this member will be not changed in next steps of the algorithm. (ix) Repeat (vi) and/or (viii) until last SM will have a xed CSA. The ANALYSIS and CHECK MODULES are elements of the DSO algorithm. They can be replaced with other modules in cases of applications of other codes.

4 Eectiveness and exactness of the proposed algorithm The rst eectiveness tests of the method were performed by Gutkowski and Zawidzka (1997). The exact continuous solution of a ten-bar truss, commonly used as a benchmark problem, was compared with the results obtained by the method. The numerical results giving only 7% of dierence with respect to exact solution was obtained with only 44 analysis. It must be pointed out that the number of possible combinations arising from the number of SM and rolled proles in the catalogue is, in this case, equal to 810 . The two following examples, in solution of which EC3 code is applied, are conrming the eectiveness of the method.

4.1 Ten-bar truss with constraints on displacements, and stresses including buckling, using EC3 for nonsway buildings As the rst example of the practical application of the proposed method, a ten-bar truss (for which dead weight is not considered as a load) is designed (Fig. 2). The following data dene the structure: L = 360 in (914 cm), P = 445.374 kN (it is assumed that a partial safety factor equal to 1.5 is already included in this value considered as a ultimate value of the variable loading P ), Youngs modulus E = 20 600 kN/cm2 and imposed displacement limit value for serviceability limit state max = 1.333 in (3.38 cm). The truss is assumed to be constructed with circular hollow sections made of S235 steel, and listed in a catalogue of 24 elements. The truss is divided into 5 linking groups. Bars no 1 and 2 are assumed to have A1 CSA; bars 3 and 4 A2 ; bars 5 and 6 A3 ; bars 8

Fig. 1 Flow chart of the algorithm

Fig. 2 Ten-bar truss

325 and 10 A4 ; and bars 7 and 9 A5 . The design variables are then, cross-section areas for each group Aq q = 1, . . . , 5. In order to enable crossing of bracings, bars 7 and 9 are assumed to be composed of double equal circular sections, each from the 24 elements catalogue. This means that the fth group has a separate catalogue which elements are double cross-sections taken from the rst catalogue. Table 1 shows these catalogues in an abbreviated manner. Only the rst, the last and 5 intermediate rows are shown. The design is performed according to the EC3 rules for class 3 cross-sections. The ANALYSIS MODULE yields as action eects 10 values of axial forces Nx,Sd (j ) in each structural member, and 8 values ui of horizontal and vertical displacements of 4 truss nodes. These values are transferred to the CHECK MODULE.

Fig. 3 Iteration history

Table 1 The CSA catalogues and the adopted SM sizes Catalogue of circular hollow sections row no. 1 ... 7 ... 16 ... 19 20 21 ... 24 diameter (mm) 32 ... 60.3 ... 168.3 ... 244.5 273.0 323.9 ... 457.2 Wall thickness (mm) 2 ... 4.5 ... 6.3 ... 8.0 8.0 8.0 ... 8.0 CSA (cm2 ) 1.89 ... 7.89 ... 32.1 ... 59.4 66.6 79.4 ... 113.0 adopted CSA

A3 A5 / 2 A1 A4 A2 Fig. 4 Single portal frame

The CHECK MODULE compares the displacement values with the limit value, and establishes DIR: d = max(ui / max). According to Chapters VI and VII of EC3 the checks for tension and compression forces are performed, and SIR for each structural member is calculated. The values of DIR and SIR are then transferred to the main structural optimization program. The minimum solution with A1 = 59.4 cm2 , A2 = 79.4 cm2 , A3 = 7.89 cm2 , A4 = 66.6 cm2 , A5 = 2 32.1 cm2 was obtained after 38 iterations only. Note, that the full enumeration procedure would require 245 combinations . The iteration history is presented in Fig. 3.

4.2 A portal frame with constraints on stresses and displacements, using EC3 for nonsway buildings Consider the minimum weight design of a hall skeleton consisting of a set of portal frames (Fig. 4). In this

test problem we assume only two design variables A1 for both columns, and A2 for the beam. The CSA are chosen from a catalogue containing 11 sections. Because of a relatively small number of possible combinations equal to 121, it is possible to nd an exact solution enumerating all of them. This way the results obtained by the proposed method can be compared with the exact ones. The frame height is 1 = 4.5 m, the span 2 = 7.5 m and the distance between frames 5.0 m. The structure should support vertical loads: permanent load 2.4 kN/m2 , imposed load 2.0 kN/m2 , snow 0.6 kN/m2 , and the lateral wind load 0.7 kN/m2 . With the assumed distance between frames, each frame should support the distributed loads applied to the beam: g = 12.0 kN/m, q = 10.0 kN/m, s = 3.0 kN/m, and the lateral wind load of the column w = 3.5 kN/m. The columns and the beam are designed of S235 steel proles assuming the 11 element catalogue of sections with broad parallel anges (European Sections, Euro norm 53-62) with the smallest size HEB 100 and the largest size HEB 300. Design is performed with elastic rst-order analysis (nonsway building) and verications are made for class 3 cross-sections independently of the real class of proles.

326 At each iteration step of the optimization procedure, ANALYSIS and CHECK MODULES are applied. The ANALYSIS MODULE yields the SLS and ULS action eects. According to Chapter III of the EC3 the following Load Cases (LC) for SLS are considered: LC1: g + max(q, s), LC2: g + 0.9 (q + s), LC3: w. The SLS eects are max maximum vertical deection of the beam, 2 maximum variation of the vertical deection of the beam due to variable loading, u the maximum horizontal displacement of the beam. For ULS the following load cases are considered: LC1: 1.35 g + 1.50 w, LC2: 1.35 g + 1.50 q , LC3: 1.35 g + 1.50 s, LC4: 1.35 (g + q + w + s). The ULS eects are the most unfavourable values of the axial forces N , bending moments M and shear forces V for columns and beam respectively: Nx.Sd (j ), My.Sd (j ), Vz.Sd (j ), j = 1, 2. The SLS and ULS eects are then transferred to the CHECK MODULE. The SLS checks consist in comparison of the eects with the appropriate constraint values. According to Chapters IV and VIII of EC3 the geometrical constraints are: max 2 /200, 2 2 /250, u 1 /150. The SLS eects are related to these limits values, and the value of DIR is determined. The ULS checks are performed according to Chapters VII, VIII and IX of the EC3. Because the shear level is low the axial force and bending moment only are taken into account. Table IX.1 of EC3 contains a list of 10 checks for member submitted to combined axial force and bending moment. In our case, checks 4, 5 and 6 are irrelevant, therefore only the following checks are to be performed for each frame member. Resistance of cross-section to Nx.Sd . In-plane (xz ) stability of member to Nx.Sd. Out-of-plane (xy ) stability of member to Nx.Sd. Out-of-plane (xy ) stability of member to My.Sd (potential lateral-torsional buckling). 5. Resistance of cross-section to (Nx.Sd , My.Sd). 6. In-plane (xz ) stability of a member to (Nx.Sd , My.Sd). 7. Out-of-plane (xy ) stability of a member to (Nx.Sd , My.Sd ) (potential lateral-torsional buckling). As a result of each check we get a value wm , where m is the check number. If the check is satised wm is less than or equal to 1, otherwise wk exceeds 1. For the j th structural member we determine SIR: w(j ) = maxm (wm ). It is worth noting that, practically, checks 6 and 7 determine the values of SIR. The output values of the CHECK MODULE are DIR: d, and SIR: w(j ), j = 1, 2. They are then used as constraints in the main program of the sequential discrete optimization. After 8 iterations we get HEB 220 for columns and HEB 240 for the beam (see Table 2), with the total weight of the structure equal to 1267.5 kg. The exact solution obtained from all 121 possible combinations gives HEB 200 for the columns and HEB 260 for the beam, with total structural weight equal to 1249.2 kg. This means that the 1. 2. 3. 4.
Table 2 The CSA catalogue and adopted SM sizes Catalogue of HEB proles size CSA (cm2 ) 26.0 34.0 43.0 54.3 65.3 78.1 91.0 106.0 118.4 131.4 149.1 CSA exact method CSA proposed method

100 120 140 160 180 200 220 240 260 280 300

COLUMNS COLUMNS BEAM BEAM

Total weight (kg)

1249.2

1267.5

weight obtained by the proposed method is only 1.46% larger than that obtained from enumeration of all possible combinations.

5 Conclusions An eective and very simple method of nding minimum (near minimum) weight of a structure, applying EC3 code, is presented. The test examples conrm previous investigations, showing that the results can be obtained with some k0 j0 analysis only, compared with j0 enumeration of all k0 combinations. The method is competitive with GA for the same kind of problems, requiring, as known, hundreds, or even thousands of analyses. The method shows that it is worthy to search for problem oriented algorithms, which may be conceptually and numerically simpler than general ones. However, it should be noted that, so far, no conclusions can be drawn saying that the algorithm assures the global minimum.
Acknowledgements This work was partly supported by Grant No. 8 T I I F00813 from the Polish Committee of Scientic Research and by nancial support from Fonds National de Recherche Scientique from Belgium. This support is gratefully acknowledged.

References
Arora, I.S.; Haug, M.W. 1994: Methods for optimization of nonlinear problems with discrete variables: a review. Struct. Optim. 8, 6985 Bauer, J. 1994: A survey of methods for discrete structural design. Comput. Assis. Mech. Eng. Sci. 1, 2738

327
Chantrain, P.; Schleich, J.B.; Conan, Y.; Mauer, T. 1996: Design handbook for braced or nonsway steel buildings according to Eurocode 3. Prol Arbed Recherches Luxembourg . ECCS Bruxelles, Belgium Eurocode 3 1992: Calcul des structures en acier Partie 1.1 R` egles g` en` erales et r` egles pour les b atiments. NBN ENV 1993-1-1 . IBN Bruxelles, Belgium Goldberg, D.E.; Samatini, M.P. 1986: Engineering optimization via genetic algorithm. Proc. 9th Conf. on Electronic Computation , pp. 471482 Guerlement, G.; Gutkowski, W.; Targowski, R.; Zawidzka, J. 1998: Application of the sequential algorithm to the discrete minimum weight design of steel structures. Abstracts of papers for 32nd Polish Solid Mechanics Conf. (held in Solmec, Zakopane, Poland), pp. 169170 Gutkowski, W.; Zawidzka, J. 1997: Sequential algorithm for discrete minimum weight design of structures. In: Gutkowski, W.; Mroz, Z. (eds) Proc. WCSMO-2, 2nd World Congress on Structural and Multidisciplinary Optimization (held in Zakopane, 1996), pp. 313318. Warsaw: Inst. Fund. Tech. Res. Iwanow, Z. 1981: The method of enumeration according to the increasing value of the objective function in the optimization of bar structures. Bull. de lAcad. Polon. des Sc. Ser. des Sc. Tech. 29, 914 Nha, C.D.; Xie, Y.M.; Steven, G.P. 1998: An evolutionary structural optimization method for sizing problems with discrete design variables. Comp. Struct. 68, 419431 Templeman, A.B. 1997: Heuristic methods in discrete structural optimization. In: Gutkowski, W. (ed.) Discrete structural optimization (CISM Courses and Lectures No 373, held in Udine), pp. 135165 Thander, P.B.; Vanderplaats, G.N. 1995: Survey of discrete variable optimization for structural design. J. Struct. Eng. 121, 301306.

S-ar putea să vă placă și