Jtam.dvi JOURNAL OF THEORETICAL AND APPLIED MECHANICS 47, 3, pp. 667-684, Warsaw 2009 MODELING OF INITIAL GEOMETRICAL IMPERFECTIONS IN STABILITY ANALYSIS OF THIN-WALLED STRUCTURES Katarzyna Rzeszut Andrzej Garstecki Poznań University of Technology, Institute of Structural Engineering, Poznań, Poland e-mail: katarzyna.rzeszut@put.poznan.pl Imperfections are modeled using actual values measured in situ. The method proposed in the paper is based on the concept of developing the imperfections in series of eigenmodes, using a limited number of most critical eigenmodes. Error minimization of this representation is performed. The method is applied to the nonlinear stability analysis of structures made of steel thin-walled cold-formed sigma profiles. FEM with shell elements and the Riks method are used. Numerical examples illustrate the influence of initial imperfections on post buckling behavior of structures. Keywords: stability analysis, initial geometric imperfections, thin-walled beams, stability of cold-formed bars 1. Introduction In the case of cold-formed steel members, initial geometric imperfections can significantly influence stability response,becauseusuallya local bucklingappe- ars closed to the global one. There are several studies which consider different types of geometric imperfections and ways of their introduction into a nume- rical model. One way of taking initial geometric imperfections into account, whichdominates indesign codes, is to induce thembyapplying an appropriate pattern of additional loading.However, itworkswell only for the global type of imperfections, e.g. in the case ofmultistory frameswith columns exhibiting de- viations fromthe vertical direction. In the case of local sectional imperfections, it seems reasonable to introduce perturbed geometry by measured values of imperfections. However, this procedure can be tedious when the finite element method is used.Moreover, it is not a general procedure since it can be applied 668 K. Rzeszut, A. Garstecki only to those members for which the imperfections have been measured. An alternativemethod of introduction of imperfections, widely discussed in the li- terature, is stochastic generation of the imperfection signal (Laubscher, 2004). Another approach to the stability analysis of imperfect structures is based on the concept of sensitivity analysis. The potential of sensitivity analysis of thin walled beams and columns accounting for nonlinear effects was discussed in Chróścielewski et al. (2006), Szymczak (2006). However, keeping in mind that eigenmodes represent the most dangerous shapes of imperfections, the introduction of imperfections in the form of eigenmodes (Dubina et al., 2001) can be considered as a classical approach. It would be reasonable to treat the perturbation in the geometry as a linear superposition of bucklingmodeswith scale factors computed frommeasurements (Fang and Pekoz, 2001; Garstecki et al., 2002; Lechner andPircher, 2005). Therefore, propermodeling of imper- fections, which correspondwith real imperfections, can play an important role in structural analysis and design. In the paper, the problem of modeling the initial geometrical imperfec- tions basing on actual imperfections measured in situ is discussed and the methods presented in Garstecki et al. (2002) and Kąkol et al. (2002) are fur- ther developed. The present approach is based on the concept of developing the imperfections in series of eigenmodes computed froma linear stability pro- blem. A limited number of most critical eigenmodes is used. The coefficients in the series are evaluated using actual imperfections (Garstecki et al., 2002), accounting forGauss probability factors and implementing errorminimization in the approximation.We start from simple examples illustrating themethod and demonstrating its accuracy. Next, the method is applied to stability ana- lysis of structures made of steel thin-walled cold-formed sigma profiles. The Riks method is used for solution of the nonlinear stability problem. 2. Method of modeling the imperfection Notation N – total number of degrees of freedom in FEMmodel r – consecutive number of displacement in FEMmodel, r=1, . . . ,N n – total number of eigenmodes in approximation i – consecutive number of eigenmode, i=1, . . . ,n m – total number of measurements of imperfections Modeling of initial geometrical imperfections... 669 k – consecutive number ofmeasurement of imperfection, k=1, . . . ,m u – N-dimensional displacement vector of actual imperfections (unk- nown) ui – N-dimensional displacement vector representing i-th eigenmode α – n-dimensional vector of scale factors v – m-dimensional displacement vector of measured imperfections vi – m-dimensional displacement vector similar to v but extracted from ui The initial geometric imperfections are introduced by perturbations in the ”perfect” geometry. In a continuous formulation, the imperfection can bewrit- ten in the form ũ(αi,x)= n∑ i=1 αiui(x) (2.1) where x is the coordinate vector, ui(x) are test functions and αi are scale factors. Assume the test functions in the form of buckling modes obtained from the linear stability problem and associated with n lowest eigenvalues. Since we apply FEM, the buckling modes ui(x) and similarly ũ(x) have the form of N-dimensional displacement vectors ui, where N is the number of DOF in the FEMmodel. In FEM ũ,ui ∈R N, hence Eq. (2.1) takes the form ũ= [ũr] = n∑ i=1 αiui = [ n∑ i=1 αiUir ]⊤ =U⊤α (2.2) where Uir denotes the displacement r of eigenmode i. Hence the dimensions of U are n×N. Note, that in the approximation (Eqs. (2.1) and (2.2)) we used only n eigenvectors, where n ≪ N. Usually, it is recommended to use those ones, which are associated with the smallest eigenvalues. However, the assumed set of eigenvalues and eigenmodesmust contain those local andglobal modes which are similar to the shape of real imperfections. Otherwise, linear combinations of the limited number of eigenmodes will not be able to capture the real imperfection pattern. Our task is to find the factors αi which minimize the error u− ũ=u−U⊤α→min α u, ũ∈RN (2.3) However, wedonotknow the real imperfections u in the space RN, becau- se themeasurements of initial geometry in situ provide only m displacements representing the imperfections. Usually, m≪N and m>n. Denote the con- secutive displacements specifying the measured imperfection by vk and the 670 K. Rzeszut, A. Garstecki imperfection displacement vector by v = [vk], v ∈R m. The error, Eq. (2.3), can be evaluated using these m displacements, only. Let the FEM mesh and the nodal displacements are introduced to stabi- lity analysis in such a form that the measured imperfection displacements vk coincide with respective nodal displacements ur, and hence there is a unique mapping r→ k r→ k (2.4) UsingEq. (2.4), we can extract the respective displacements k of the eigen- mode i, namely a component Vik of the N-dimensional matrix of eigenmo- des Uir. The rows i of the matrix Vik represent the vectors vi, which will be used as test functions in error minimization of the approximation. Now, the approximation of the measured imperfection displacement vectorv takes the form ṽ= [ṽk] = n∑ i=1 αivi = [ n∑ i=1 αiVik ]⊤ =V⊤α (2.5) The error of approximation can now be represented by the vector ε in the space Rm ε=v− n∑ i=1 αivi =v−V ⊤ α→min α ε,v∈Rm α∈Rn (2.6) Following the Galerkin concept, we assume optimal α, when it makes ε orthogonal to all test functions vi, namely εvi =0 (2.7) where dot denotes scalar product. Introducing (2.6) into (2.7), we obtain ( v− n∑ i=1 αivi ) vj =0 (2.8) Thematrix form of (2.8) is Vv−VV⊤α=0 or VV⊤α=Vv (2.9) Introducing A=VV⊤ and b=Vv (2.10) we obtain Aα= b (2.11) Modeling of initial geometrical imperfections... 671 hence α=A−1b Note that the above presented discrete Galerkin method of evaluation of optimal α also providesminimumof the quadratic error of ε in the Rm space I= ε⊤ε I= εε⊤ = [v−V⊤α]⊤[v−V⊤α] = [v⊤−α⊤V][v−V⊤α] (2.12) The stationary condition ∂I/∂α= 0, provides a set of n linear equations (2.9) and (2.11). 3. Verification of the method In engineering practice, the number of measurements of imperfections is limi- ted.Moreover, patterns of imperfections, which aremost important for future stability analyses, are not a priori known. Hence, we face a difficulty that the measurements of imperfections in situ are not only limited, but often not in optimal points of the structure. Therefore, the proper calibration of scale factors αi plays the important role. The proposed method makes it possible to represent the imperfections as a linear superposition of a limited number of such eigenmodes which play the crucial role in stability of the structure. It is particularly important in the class of stability problems when two or more eigenvalues, which correspond with local and global buckling, coincide or are close to each other. This interactive buckling is typical for thin-walled struc- tures under consideration and is usually connected with high sensitivity to imperfections. The method of developing the imperfections in series of eigenmodes com- puted from the linear stability problem is aimed at complex nonlinear stability analysis accounting for initial imperfections. However, for the sake of simpli- city, let us start the considerations from the Euler column (Fig.1). The buckling modes are ui =siniπx (3.1) where x= [0,1] is the non-dimensional coordinate measured along the length of the column. Through the examples, we will study, verify and validate the proposed algorithm. 672 K. Rzeszut, A. Garstecki Fig. 1. Simply supported, axially loaded column: (a) geometry, (b) buckling modes. Coordinates x1, . . . ,x6 indicate points of simulated imperfections 3.1. Example 1 In order to check the correctness of the proposed method, let us assu- me that the measured imperfection corresponds with the second and fourth buckling mode with amplitudes α0 equal to 0.03 and 0.07, respectively. Let the imperfections be measured in m = 6 points, placed along the column at xk =0.2, 0.3, 0.4, 0.6, 0.7, 0.9. Themeasured imperfection vector v is v= [u(xk)]= [0.03sin2πxk +0.07sin4πxk] = (3.2) = [6.97,−1.26,−4.89,4.89,1.26,−8.42] ·10−2 The imperfectionpatternwill beapproximatedusingonly four eigenmodes, n = 4. The discrete representation of the test function ui at k = 1, . . . ,m points of measurement of the imperfections can be calculated as V= [Vik] = [ui(xk)] = [siniπxk] = (3.3) =   58.78 80.90 95.11 95.11 80.90 30.90 95.11 95.11 58.78 −58.78 −95.11 −58.78 95.11 30.90 −58.78 −58.78 30.90 80.90 58.78 −58.78 −95.11 95.11 58.78 −95.11   ·10−2 Modeling of initial geometrical imperfections... 673 Matrices A and b are Aij =Aji = m=6∑ k=1 ui(xk)uj(xk) (3.4) A=   3.59017 0.37738 0.19098 0.01599 0.37738 3.75 0.42898 −1.11804 0.19098 0.42898 2.440983 −0.21041 0.05199 −1.11804 −0.21041 3.75   and bj = m=6∑ k=1 uj(xk)u(xk) b=   0.01493336 0.03423762 −0.00185886 0.22895898   (3.5) hence α=A−1b=   α1 α2 α3 α4   =   0 0.03 0 0.07   (3.6) As a result, the expected exact value of α is obtained. Several other testing examples were solved for different numbers ofmeasu- rement points and test functions. However, in order to check the correctness of the method, in all those examples, idealized patterns of imperfections were introduced. They took the form of linear combinations of such eigenmodes which were used as test functions in the approximation. No wonder that the examples demonstrated that the exact approximation was always obtained in cases when the number of measurements m was equal or greater than the number of test functions n. This is illustrated in Table 1. The error of approximation can be calculated in the space L2 ‖ũ(x)−u(x)‖L2 = √√√√√ 1∫ 0 ε2 dx= √√√√√ 1∫ 0 ∑ i [αiũi(x)−u(x)] 2 (3.7) Basing on the orthogonality of eignmodes, the error (Eq. (3.7)) takes the form ‖ũ(x)−u(x)‖L2 = √√√√√ 1∫ 0 [ 4∑ i=1 (αi−α 0 i) 2 sin2 iπx ] dx (3.8) 674 K. Rzeszut, A. Garstecki Table 1. Error of the approximation for different numbers and distributions of points of simulated measurements No. of mea- Points of measurements Approximated values α Error of No. surement approxim. points m in spaceL2 1. 6 xk =0.2, 0.3, 0.4, α1 =0.0, α2 =0.03, 0.0 0.6, 0.7, 0.8 α3 =0.0, α4 =0.07 2. 6 xk =0.1, 0.2, 0.3, α1 =0.0, α2 =0.03, 0.0 0.4, 0.5, 0.55 α3 =0.0, α4 =0.07 3. 4 xk =0.2, 0.3, 0.4, 0.6 α1 =0.0, α2 =0.03, 0.0 α3 =0.0, α4 =0.07, xk =0.5, 0.6, 0.7, 0.8 α1 =−5.54 ·10 −15 4. 4 α3 =−4.02 ·10 −15 0.0 α2 =0.03, α4 =0.07 α1 =0.057, α2 =−0.01, 5. 3 xk =0.1, 0.2, 0.3 α3 =6.09 ·10 −3, 3.82 ·10−2 α4 =0.068 6. 3 xk =0.2, 0.3, 0.4 α1 =−0.034, α2 =0.067, 1.05·10−2 α3 =−0.019, α4 =0.075 Note that in the above idealized example the distribution of simulated measurement points did not affect the result of the approximation, provided that m­n. In the next examples presented in the paper, a more general pattern of imperfections will be used. 3.2. Example 2 Following example 1, we use the first four buckling modes (3.1) in the series approximating the imperfection pattern. However, we assume now that the imperfections correspond with the second, fourth and seventh buckling mode with amplitudes α0 equal to 0.03, 0.07 and 0.01, respectively. It means that the seventh buckling mode with a multiplier 0.01 is superposed on the imperfection pattern from example 1. Let the imperfections be measured in m = 6 points, placed along the column at xk = 0.2, 0.3, 0.4, 0.6, 0.7, 0.9. Now, the assumed imperfection vector v is Modeling of initial geometrical imperfections... 675 v ⊤ = [u(xk)] = [0.03sin2πxk+0.07sin4πxk+0.01sin7πxk] = (3.9) = [6.0,−0.9523,−4.3,5.5,1.6,−7.6] ·10−2 Following the presented above algorithm, we obtain the approximated va- lues of α α=A−1b=   α1 α2 α3 α4   =   0.0045 0.025 0.0029 0.065   (3.10) Now, the approximated imperfection vector ṽ is ṽ ⊤ = [u(xk)] = [α2 sin2πxk+α4 sin4πxk+α7 sin7πxk] = (3.11) = [5.247,−1.134,−4.125,5.30,1.752,−6.842] ·10−2 The error of approximation in the space L2 is √√√√√ 1∫ 0 [ 4∑ i=1 (αi−α 0 i) 2 sin2 iπx+(α7−α 0 7) 2 sin27πx ] dx=4.99 ·10−2 (3.12) where according to the assumption α01 =α 0 3 =α+7 =0,α 0 2 =0.03,α 0 4 =0.07, α07 =0.01. The exactness of approximation can be checked using themean quadratic error in the R6 space ε2average = 1 m m∑ i=1 (ṽi−vi) 2 =2.2 ·10−5 (3.13) The effectiveness of approximation was analysed for different numbers m and various distributions xk of points of simulated measurements. It was demonstrated that the number and distribution of measurement points can strongly affect the error of approximation. This is illustrated in Table 2. In examples 1 and 2, we considered discrete representations of the imper- fection pattern. Note that in the case of a continuous representation of the imperfection pattern in the form of a sine series, the variation of imperfection by superposition of the seventh mode could not be captured using a series limited to four terms, because this variation is orthogonal to each of the four modes. 676 K. Rzeszut, A. Garstecki Table 2.Error of the approximation for different number and distribution of points of simulated measurements No. m Points of measurements xk Approximated values α Error of No. of mea- approximation surem. in L2 α2average xk =0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 0.95 α1 =0.0, α2 =0.03, α3 =0.0, α4 =0.07 1. 10 0.0 0.0 xk =0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9 α1 =0.0, α2 =0.03, α3 =0.0, α4 =0.07 2. 9 0.0 0.0 3. 8 xk =0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9 α1 =8.333·10 −4, α2 =0.028, α3 =−2.182·10 −3, α4 =0.067 4.96·10−2 6.096·10−6 4. 4 xk =0.2, 0.3, 0.4, 0.6 α1 =0.017, α2 =−2.064·10 −3, α3 =0.018, α4 =0.053 5.30·10−2 1.57·10−4 5. 3 xk =0.2, 0.3, 0.4 α1 =−0.017, α2 =0.04, α3 =−2.39 ·10 −3, α4 =0.058 1.64·10−2 1.93·10−4 3.3. Example 3 Assume now that the simulated measurements are contaminated with a white noise produced by m = 6 independent sources, each with a random output following a uniform distribution between −0.5 and 0.5. Now, we assume the basic imperfection vector v0 to be similar to v from example 1 v 0 = [u(xk)] = [0.03sin2πxk+0.07sin4πxk] = (3.14) = [6.97,−1.26,−4.89,4.89,1.26,−8.42] ·10−2 The random generator of the white noise produced the vector whiten(6)⊤ = [−0.499,−0.307,0.085,−0.15,0.323,−0.326] (3.15) Modeling of initial geometrical imperfections... 677 Assume a white noise amplitude c=0.2. The output vector v is v n =whiten(6)c (3.16) The imperfection vector with the white noise can be expressed as v ⊤ = [v0+vn]⊤ = [7.3,−2.0,−5.8,3.9,1.6,−7.6] ·10−2 (3.17) Approximated components of the vector α are α=A−1b=    α1 α2 α3 α4    =    0.004667 0.026 0.009102 0.066    (3.18) Now, the approximated imperfection vector ṽ is ṽ= [7.3,−1.6,−5.4,5.4,1.6,−8.9] ·10−2 (3.19) Themean quadratic error in the R6 space can be calculated according to the following formula 1 m m∑ i=1 (ṽi−vi) 2 =5.299 ·10−6 (3.20) The error of approximation for different values of the white noise amplitu- de c is presented in Fig.2. Fig. 2. The mean quadratic error for different values of the white noise amplitude c 678 K. Rzeszut, A. Garstecki 4. Measurement of actual imperfections In thisSection theactual imperfectionsof steel thin-walled cold-formedprofiles are discussed. The measurements were performed on sigma sections 300mm high with walls 1.5mm thick (Σ300×1.5), shown in Fig.3a. Fig. 3. Profile Σ300×1.5 (a) design dimensions, (b) measured dimensions Eighteenmemberswere examined and initial local-sectional geometric im- perfectionswere identified.Thickness andwidthof all walls of the cross-section andvariationsof thecontourweremeasured in5cross-sections along the length of themembers. A statistical data processing of themeasured values was per- formed. The Gauss distribution and the confidence level 95% were assumed. Thus, we arrived at the limit values alower/upper = amean ∓1.64σ (4.1) where σ is the standard deviation of a. The limit values of imperfections are δamin/max = alower/upper −adesign (4.2) The results are presented in Table 3. The rows a-h refer to width of the walls, whereas the next rows describe deformations of the contour. The limit values of imperfections are presented in columns 5 and 6. The design cross-sectional dimensions are shown in Fig.3a. All these di- mensions were compared with actual values measured in situ and thus the dimensional imperfections were evaluated. Figure 3b shows the next class of measured imperfections, namely shape imperfections. Table 3 presents the re- sults of statistical processing of all design and shape imperfections. Modeling of initial geometrical imperfections... 679 Table 3.Profile Σ300×1.5. Values of imperfections in mm Symbol Design Arithmetic Standard Maximum Minimum dimension mean deviation imperf. imperf. 1 2 3 4 5 6 a 25.00 28.61 1.98 6.86 0.37 b 80.00 80.89 1.09 2.67 −0.89 c 63.50 63.97 1.36 2.70 −1.75 d 19.00 17.24 1.84 1.26 −4.77 e 139.00 138.64 0.88 1.08 −1.80 f 19.00 17.55 1.59 1.16 −4.06 g 59.50 60.50 1.32 3.16 −1.17 h 70.00 70.39 1.31 2.55 −1.76 l 0.00 0.00 1.10 1.80 −1.80 m 0.00 0.17 1.58 2.77 −2.43 n 0.00 0.11 1.65 2.82 −2.59 o 0.00 −0.14 1.17 1.78 −2.07 p 0.00 −0.06 1.29 2.06 −2.19 r 0.00 −0.11 1.25 1.95 −2.16 s 0.00 0.01 0.73 1.20 −1.18 The shape imperfections can be classified as symmetric ”opening” (SO), symmetric ”closing” (SC) or asymmetric deformation (AS). They are presen- ted in Fig.4. These three forms of local imperfections appeared periodically throughout the length of elements, which suggests that they originated in the cold-forming process. Fig. 4. Demonstrative imperfection patterns along the Σmembers The average values of shape imperfections l-s in different cross sections x are presented in Fig.5. These values can be used as coordinates vk of the vector v to calculate scale factors αi in series (1) using equation (2.12). 680 K. Rzeszut, A. Garstecki Fig. 5. Spectrum of average shape imperfections in a bar 5. Eigenvalue problem The eigenvectors ui(x) representing the bucklingmode are computed fromthe linear eigenvalue problem (K0+λKG)U =0 (5.1) where K0 denotes the small-displacement stiffnessmatrix, KG is the geometric stiffness matrix, λ is the load multiplier and the eigenvectors U represent the buckling mode shapes. Numerical examples were solved using the general purpose finite element program ABAQUS. A simply supported axially loaded column was analysed. To consider local imperfections presented in Section 4, local buckling modes had to be captured. Therefore, four-node doubly curved shell elements with reduced integration were employed. In the FEM 2D model, the boundary conditionswere introduced so as to represent a spherical hinge support.Hence, in all nodes of the boundary cross-section, displacements in the direction x (longitudinal axis of the column) were unconstrained except for the center point at one support. The transversal displacements at the supports were assumed to be zero.Nodal forceswere applied at the boundarycross-section in the longitudinal direction xwith themagnitude σx representing the force P. Assume the length of the axially compressed column to be 4.0m. Five different eigenmodes were extracted from linear stability analysis. During the analysis, fromthe eigenvectors ui(x), thematrix Vik representing thebuckling mode at the points ofmeasurement was calculated. Figure 6 shows the shapes of buckling modes related to the lowest eigenvalues. The next eigenvalues were much higher. Since the first three eigenvalues are close to each other, at least these three eigenmodes should be used in the series approximating the Modeling of initial geometrical imperfections... 681 initial imperfection pattern. However, for better illustration of the interaction of global and local imperfections, in the following example we will use only the first two eigenmodes.Mode 1 is a global one andmode 2 represents a local buckling form. Fig. 6. Shapes of buckling modes for 2×Σ column Since λ1 is close to λ2, an interactive buckling can appear. Hence, we can expect unstable postbuckling behavior, reduction of load capacity and sensitivity to imperfections. To study this issue, we will carry out a nonlinear stability analysis allowing for different kinds of initial geometric imperfections, global, local and global/local. 6. Nonlinear stability analysis The imperfections were introduced into stability analysis by perturbing the initial geometry by imperfections in the form of Eqs. (2.1) and (2.2). Our ob- jective is to study the influence of actualmagnitudes of global and local imper- fections. Therefore, the proportionality factor associated with global buckling mode 1 inFig.6 has been assumed α1 =4, thusmodeling the global imperfec- tion amplitude to be in accordance with the design code provisions referring to allowable execution tolerances. The scale factor for local mode 2 has been 682 K. Rzeszut, A. Garstecki evaluated as α2 =2.04, using the procedure described in Section 2 and basing onmeasurements described in Section 4. Fig. 7. Load proportionality factor for the columnmade of 2Σwith different shapes of imperfections Figure 7 shows plots of the load proportionality factor λ/λcr versus the total arc length in the Riks algorithm, where λcr is the critical load factor for the ideal columnwith perfect geometry obtained from linear stability analysis. The plots in Fig.7 refer to global (g), local (l) and global-local (g+l) shapes of imperfections. As expected, for the global imperfections, the post buckling path is stable, however themaximum load capacity is reduced by 20% in rela- tion to λcr. Conversely, for local imperfections, themaximum load is reduced only by 10% but the post buckling is unstable. The interaction of global and local imperfections represents the worst case when the post buckling path is unstable and 20% reduction of maximal load is observed. 7. Concluding remarks In the paper, themodeling of initial geometrical imperfections in stability ana- lysis was presented. Local sectional imperfections of cold-formed thin-walled steel sigma cross-section hadbeenmeasured in situ.Themeasured valueswere then subjected to statistical processing. Modeling of initial geometrical imperfections... 683 Thepaper presents amethod ofmodeling initial imperfections. The imper- fections are modeled in the form of a displacement vector with the dimension and physical meaning adequate to the displacement vector in the FEMmodel of the structure. The vector of initial imperfections is assumed in the form of a series of a limited number of eigenfunctions obtained from linear stability analysis. Such representation of imperfections makes it easy to employ FEM in non-linear stability analysis. The scale coefficients in the series were evalu- ated basing on measurements and using the discrete Galerkin approach, thus providing theminimum of discrepancy between themeasured values of initial imperfections and their approximated representation.Themethodwas verified by making use of idealized examples and examples where idealized simulated measurements were contaminated with a white noise. The stability analysis of a columnmade of a cold-formed thin-walled steel sigma cross-section was carried out. Particular attention was paid to the inte- raction of global and local bucklingwhich can result in excessive sensitivity to imperfections and in unstable behavior. It was found that the initial sectional imperfections did not remarkably reduce the maximum bearing capacity of the column, but theymade the post buckling behavior unstable. References 1. ABAQUS/Standard, Hibbitt, Karlsson& Sorensen, Inc., 2001 2. Chrościelewski J., Lubowiecka I., Szymczak C., 2006,On some aspects of torsional buckling of thin-walled I-beamcolumns,Computers and Structures, 84, 1946-1957 3. Dubina D., Ungureanu V., Szabo I., 2001, Codification of imperfections for advanced finite analysis of cold-formed steel members, Proceedings of the 3rd ICTWS 2001, 179-186 4. Fang Y., Pekoz T., 2001, Design of cold-formed steel plain channels, School of Civil and Environmental Engineering Report, Cornel University, Ithaca 5. Garstecki A., Kakol W., Rzeszut K., 2002, Classification of local- sectional geometric imperfections of steel thin-walled cold-formed sigmamem- bers,Foundations of Civil Environmental Engineering, 45 6. Kąkol W., Rzeszut K., Garstecki A., 2002, Stability analysis of cold- formed thin-walled members, Fifth World Congress on Computational Mecha- nics, Vienna, Austria 684 K. Rzeszut, A. Garstecki 7. Laubscher R.F., 2004, An experimental and numerical investigation of the axial resistance of hot rolled 3CR12 columns,Proceedings of the Fourth Inter- national Conference on Coupled Instabilities in Metal Structures, Rome, Italy, 411-427 8. Lechner B., Pircher M., 2005, Analysis of imperfection measurements of structural members,Thin-Walled Structures, 43, 361-374 9. Szymczak C., 2006, The effect of nonlinear restraint on torsional buckling and post-buckling behaviour of thin-walled column, Foundation of Civil and Environmental Engineering, 333-342 Modelowanie początkowych imperfekcji geometrycznych w zagadnieniach stateczności konstrukcji cienkościennych Streszczenie W pracy przedstawiono metodę modelowania początkowych imperfekcji geome- trycznych na podstawie rzeczywistych imperfekcji pomierzonych „in situ”. Zapropo- nowana metoda polega na automatycznym tworzeniu sygnału imperfekcji w postaci serii funkcji własnychuzyskanych z liniowej analizy stateczności.Domodelowania im- perfekcji użyto niewielkiej liczby najbardziej niekorzystnych postaci własnych. Efek- tywnośćmetody analizowano ze względu na liczbę punktów pomiaru oraz sposób ich rozmieszczenia na długości pręta, minimalizując błąd aproksymacji.W dalszej części pracy, propopnowana metoda została zastosowana do nieliniowej analizy stateczno- ści z uwzględnieniem imperfekcji prętów cienkościennych typu „Σ”. Do rozwiązania problemu nieliniowej analizy stateczności zastosowanometodę Riks’a. Przykłady nu- meryczne ilustrują wpływ początkowych imperfekcji geometrycznych na pokrytyczne zachowanie konstrukcji. Manuscript received April 2, 2009; accepted for print April 22, 2009