Jtam.dvi JOURNAL OF THEORETICAL AND APPLIED MECHANICS 49, 4, pp. 1059-1078, Warsaw 2011 METHOD OF FUNDAMENTAL SOLUTION AND GENETIC ALGORITHMS FOR TORSION OF BARS WITH MULTIPLY CONNECTED CROSS SECTIONS Piotr Gorzelańczyk Higher Vocational State School Pila, Polytechnic Institute, Piła, Poland e-mail: piotr.gorzelanczyk@pwsz.pila.pl The torsion of bars with a multiply connected cross sections by means of the method of fundamental solutions (MFS) is considered herein. To determine the optimal parameters ofMFS, genetic algorithmswere used. Seven cases of cross sections are considered. The numerical results for different cross sectional shapes are presented to demonstrate the effi- ciency and accuracy of the method. Non-dimension torsional stiffness was calculated by means of numerical integration of the stress function for one of the cases. This stiffness is compared with the exact stiffness for the first case and with the stiffness resulting from Bredt’s formulae for thin walled cross sections. Key words:Bredt’s formulae,method of fundamental solutions,multiply connected sections, genetic algorithms 1. Introduction The solution of the torsion problem for a multiply connected cross section is more difficult than for the simply connected one. Probably, it is the reason why there are not too many papers considering this problem. For a doubly connected cross section, the exact solution exists for the annular cross section. Weinel (1932) proposed the solution for a doubly connected cross section with excentric circles. In the work (Polya and Weinstein, 1950) it was found that for the doubly connected domains with prescribed area of the hole and cross section the ring bounded by two concentric circles has the maximal torsional rigidity. In the book (Arutiunian and Abramian,1963) the description of the method of solving the torsion problem for a doubly connected cross section in which outer and inner contour are rectangles is presented. This method 1060 P. Gorzelańczyk was proposed by Russian authors (Szerman, Abramian) in series of papers in the 50s and it is based on expansion of the stress function in Fourier series. In order to obtain an effective solution, this method requires the solution to an infinite system of linear equations. Wang (1995) presented a method for torsion analysis of two connected cross sections in shape of a flattened tube consisting of two half annular pieces and two rectangular pieces. He adopted an approximate solution in the formof truncated series of functions of its own, and to satisfy the boundary conditions he used boundary element methods. Wang (1998) generalized this method to treat arbitrary two connected cross sections consisting of circular arcs and straight lines, all with a uniform thick- ness. A modified Fourier series method for the torsion analysis of bars with multiply connected cross sectionswas presented byKimandYoon (1997). The effectiveness of this method was presented for polygonal cross sections with polygonal holes.Mejak (2000) presented amethod for an optimal shapedesign of doubly connected bars in torsion.He solved the problemnumerically by the finite element method. In the work (Kolodziej and Fraska, 2005), the Trefftz method using special purpose T-functions was used to solve the problem of torsion. As examples, one-connected, multiply-connected and composite cross sections of bars in the shape of regular polygons were considered. The pro- posed Trefftz function not only satisfies the governing equation, but also the boundary conditions on some sides. In the article, for the stress function, the boundary collocation methods and themethod of smallest squares were used. Using the analytical integration, an analytical solution for the dimensionless stiffness of the bar was obtained. A special group of papers considers thin-walled cross sections. A simple formulation for torsion analysis of thin-walled hollow bars can be found in ele- mentary textbooks of strength of materials, as was proposed byBredt (1896). In the work (Morassi, 1999) the author proves that Bredt’s theory remains true for thin tubes with multicell cross sections more than doubly connec- ted. A closed form expression for the torsion constant and thin-walled typical multicell profiles is presented by Lubarda (2009). Generalization of Bredt’s method for moderate thick hollow tubes with polygonal shapes is given by Hematiyan and Doostfatemeh (2007). The purpose of this paper is application of Method of Fundamental Solu- tions and genetic algorithms for the torsion problemwith multiply connected cross sections. This method belongs to so-called meshless methods which ha- ve been more and more popular in the two last decades. The MFS was first proposed by theGeorgian researchers Kupradze andAleksidze (1964). Its nu- merical implementation was carried out byMathon and Johnston (1977). The Method of fundamental solution and genetic algorithms... 1061 mathematical analysis (convergence and stability) of thismethodwas conside- red inBogomolny (1985), Katsura (1990), Katsura andOkamoto (1988, 1996), Kitagawa (1988, 1991). The comprehensive reviews of theMFS for various ap- plications can be found in Fairweather and Karageorghis (1998), Fairweather et al. (2003), Goldberg and Chen (1998). However, as yet, themethod of fun- damental solutions has been applied basically for simply-connected regions. There are only few papers with application of MFS for annular region, e.g. Chen et al. (2006), Li (2009), Tsangaris et al. (2006). In theworks (Fairweather andKarageorghis, 1988a,b, 1989; Fairweather et al., 2003), the location of sources is determined by minimizing the functional of mean, where the optimization parameters are also the fundamental solu- tion weighing factors and coordinates of the location of sources. These latter issues are the evidence of nonlinearity. These authors applied the collocation method, where the number of sources and the number of collocation points is determined. Genetic algorithms for one-connected areas were applied in Ko- łodziej and Klekiel (2008), Nishimura et al. (2000, 2001, 2003) to determine the optimal positioning of sources. Last time, Karageorghis (2009) appeared in which the method of the golden mean was used to determine the optimal location of source points. In this case, the error is multidimensional, with many local minima, nu- merical methods of searching for optimal solution fail. The result is that there can be a case where the solution lies outside the area and thus the process of finding the optimum can not achieve the desired result. The methods consist of careful movement from point to point in a certain area of decision-making, in accordance with the selection rule determining the next point. This is not a safe way, because it allows the location of false minima in a multi-node space exploration. For this reason, genetic algorithms were used to determine the optimum parameters. According to Goldberg (1995), genetic algorithms are search algorithms based on the mechanisms of natural selection and heredi- ty, which were developed by Howland. Combining the evolutionary principle of survival of the fittest with a systematic, although randomized exchange of information, they create a method of finding, reluctantly giving it her pro- per degree of inventiveness of the human mind. Genetic algorithms can cope well where the optimized function is noisy, changes over time and has many local extremes. Using genetic algorithms in an expeditious manner, we can determine the optimal solution of the method. This paper presents the application of this method to multiply connected cross sections namely: (I) circular with circular centered hole, (II) squarewith circular centered hole, (III) squarewith square centered hole, (IV) squarewith 1062 P. Gorzelańczyk square centered hole with rounded corners with the radius r = E/2 (where E is the characteristic dimension of the hole), (V) squarewith square centered hole with rounded corners with the radius r = 3E/4 (VI) circular with two circular symmetrical placed holes, (VII) square with two circular symmetrical placed holes. 2. Formulation of the problem The problem of torsion of prismatic bars with a multiply connected cross section (see Fig.1) is formulated in terms of the stress function,which satisfies Poisson’s equation (Arutiunian and Abramian,1962) ∇2ψ =−2Gω in Ω (2.1) with the boundary condition on the outer contour ψ =0 on Γ0 (2.2) and boundary conditions on the inner contour ψ = ψi on Γi i =1,2, . . . ,n (2.3) where ψ(x,y) is the stress function, µ is the shearmodulusof thebarmaterial, ω is the angle of twist of the bar per unit length, ψi are unknown values of the stress function at inner contours, n – number of hollow areas. Fig. 1. Multiply connected cross sections of a bar For determination of the unknown constants ψi, the following integral relations (Bredt’s theorem) are given ∮ Γi ∂ψ ∂n ds =−2ΩiGω i =1,2, . . . ,n (2.4) where Ωi is the area bounded by Γi. Method of fundamental solution and genetic algorithms... 1063 After introducing the non-dimensional variables X = x a Y = y a Ψ(X,Y )= ψ(x,y) a2Gω (2.5) The considered boundary value problemhas the following dimensionless form: governing equation for the stress function ∂2Ψ ∂X2 + ∂2Ψ ∂Y 2 =−2 in Ω̃ (2.6) with the boundary condition at the outside contour Ψ =0 on Γ̃0 (2.7) and the boundary condition at the inner contour Ψ = Ψi on Γ̃i i =1,2, . . . ,n (2.8) and an integral relation in the form ∮ Γ̃i ∂Ψ ∂n ds =−2Ω̃i i =1,2, . . . ,n (2.9) where Ω̃i is the dimension area bounded by Γ̃i. 3. Method of solution In theMFS, the approximate solution to the problem is represented in formof linear superposition of source functions (fundamental solutions) with singular points that are located outside the domain of the problem. These points, cal- led source points, are located on a “pseudo-boundary” outside the region. The pseudo-boundary has no common points with the boundary of the region. Be- cause the fundamental solution satisfies the differential equation at any point except for the source point, it follows that this representation exactly satis- fies the governing equationwhereas the boundary conditions are only satisfied approximately. Therefore, the MFS belongs to the group of Trefftz methods for which it is essential that the governing equation is exactly satisfied. The weights of coefficients which occur in the approximate solution are determined by the satisfaction of the boundary condition, usually on a set of boundary 1064 P. Gorzelańczyk points (collocation points). Using MFS, the solution of boundary value pro- blem formulated by (2.6)-*2.9) can now be given as the sum of the particular solution and the homogeneous solution Ψ =− 1 2 (X2+Y 2)+ MZ∑ j=1 cj ln[(X −XSZj) 2+(Y −Y SZj) 2] + n∑ i=1 MWi∑ k=1 c (i) k ln[(X −XSW (i) k )2+(Y −Y SW (i) k )2] (3.1) where: XSZj, Y SZj are the coordinates of source points which are placed outside the region Ω̃ (Fig.2), XSW (i) k , Y SW (i) k are the coordinates of source points which are placed inside the inner contours Γ̃i, where i = 1,2, . . . ,n, MZ is the number of source points outside the region Ω̃, MWi is the num- ber of source points inside each inner contour Γ̃i, cj and c (i) k are unknown coefficients. Fig. 2. Arrangement of the source points on a similar contour The unknown coefficients cj, c (i) k and constants Ψi are determined by col- location of boundary conditions (2.7) and (2.8) and application of integral relations (2.9). In order to do this, NCZ collocation points on the outer contour with coordinates XCZl, Y CZl are chosen and NCW collocation points on each inner contour with the coordinates XCW (i) m , Y CW (i) m are chosen. Method of fundamental solution and genetic algorithms... 1065 Substituting solution (3.1) to the boundary condition (2.7) we have (l =1,2, . . . ,NCZ) MZ∑ j=1 cj ln[(XCZl−XSZj) 2+(Y CZl−Y SZj) 2] + n∑ i=1 MWi∑ k=1 c (i) k ln[(XCZl−XSW (i) k )2+(Y CZl−Y SW (i) k )2] = 1 2 (XCZ2l +Y CZ 2 l ) (3.2) Similarly, substitution of solutions (3.1) into boundary condition (2.8) leads to a system of linear equations (m =1,2, . . . ,NCW) MZ∑ j=1 cj ln[(XCW (i) m +XSZj) 2+(Y CW(i)m −Y SZj) 2] + n∑ i=1 MWi∑ k=1 c (i) k ln[(XCWm−XSW (i) k )2+(Y CWm−Y SW (i) k )2] = 1 2 (XCZ2l +Y CZ 2 l )+Ψi (3.3) Using Bredt’s conditions (2.9) on the inner contour Γ̃i, we have (i =1,2, . . . ,n) MZ∑ j=1 cj ∮ Γ̃i ∂ ∂n ln[(Xs+XSZj) 2+(Ys−Y SZj) 2] ds + n∑ i=1 MWi∑ k=1 c (i) k ∮ Γ̃i ∂ ∂n ln[(Xs−XSW (i) k )2+(Ys−Y SW (i) k )2] ds =0 (3.4) In this way, we obtain NCZ +nNCW +n equations with MZ +nMW +n unknowns. For further numerical calculations, the following assumptions: M = MZ +MWi and NC = NCZ +NCW were made. 1066 P. Gorzelańczyk 4. Torsional stiffness and stress Therelations between thenonzero components of the stress and stress function are given by following formulae τyz =− ∂ψ ∂x τxz = ∂ψ ∂y (4.1) The torsional moment is given by an integral of the shear stresses over the area, which gives SZ = ∫∫ (τyzx−τxzy) dxdy =− ∫∫ ∂Ψ ∂x x dxdy− ∫∫ ∂Ψ ∂y y dxdy+2 n∑ i=1 Ωiψi (4.2) After simple manipulations, we get SZ =2− ∫∫ Ψ dxdy +2 n∑ i=1 Ωiψi (4.3) Introducing the non-dimensional variables into (4.3), the torsional moment can be related to the non-dimensional stress function SZ = Gωa4 [ 2 ∫∫ Ψ(X,Y ) dX dY +2 n∑ i=1 Ω̃iΨi ] (4.4) Next, the non-dimensional torsional stiffness can by expressed as Ms = SZ Gωa4 =2 ∫∫ Ψ(X,Y ) dX dY +2 n∑ i=1 Ω̃iΨi (4.5) In elementary textbooks of strength of materials (Dylag et al., 1999), one can find the expression for torsional stiffness for thin-walled hollow bars (Fig.3), known as Bredt’s formula SZ = 4GA2 mid ω ∮ ds t (4.6) where Amid is the area bounded by the centerline of the wall cross section, t is thickness. Method of fundamental solution and genetic algorithms... 1067 Fig. 3. Thin-walled bar with closed cross-section 5. Test examples In order to demonstrate the exactness and the effectiveness of the proposed method, seven cases of cross sections are considered: (I) circular with circular centered hole, (II) square with circular centered hole, (III) square with square centered hole, (IV) square with square centered hole with rounded corners with the radius r = E/2 (where E is the characteristic dimension of the hole), (V) square with square centered hole with rounded corners with the radius r = 3E/4, (VI) circular with two circular symmetrical placed holes, (VII) square with two circular symmetrical placed holes. The formulation of boundary values problems for seven considered cross sections in terms of the non-dimensional stress function Ψ(X,Y ) is given in Figs.4-10. In this case, the thickness E is a geometrical parameter which changes in a permissible range, i.e. 0 < E < 0.5 for problems I-V and 0 < E < 0.25 for problems VI, VII. Fig. 4. Formulation of the boundary value problem for the circular with circular centered hole cross section of the bar. Problem I 1068 P. Gorzelańczyk Fig. 5. Formulation of the boundary value problem for the square with circular centered hole cross section of the bar. Problem II Fig. 6. Formulation of the boundary value problem for the square with square centered hole cross section of the bar. Problem III Fig. 7. Formulation of the boundary value problem for the square with square centered hole cross section of the bar with rounded corners with the radius r = E/2. Problem IV In order to validate the proposednumericalmethod, themaximumrelative error on the outer and the inner boundary can be evaluated by δMAXou = max |Ψouter| max(i)|Ψmid outeri| δMAXini = max |Ψouteri −Ψmid inneri| max(i)|Ψmid inneri| (5.1) Method of fundamental solution and genetic algorithms... 1069 Fig. 8. Formulation of the boundary value problem for the square with square centered hole cross section of the bar with rounded corners with the radius r =3E/4. ProblemV Fig. 9. Formulation of the boundary value problem for the circular with two circular symmetrically placed holes cross section of the bar. ProblemVI Fig. 10. Formulation of the boundary value problem for the square with two circular symmetrically placed holes section of the bar. ProblemVII where: δMAXou is the maximum error on the outer contour, δMAXini – maxi- mum error on the inner contour, where i =1 for the problem I-V and i =1,2 for the problem VI-VII, Ψouter – value of the stress function on the outer contour, Ψmid inneri – value of the stress function on the inner contour, where i =1 for the problem I-V and i =1,2 for the problemVI-VII. 1070 P. Gorzelańczyk 6. Discussion on the numerical results The MFS applied in this paper to the problem of torsion of a prismatic bar depends on ythe number of parameters. These parameters are as follows: the distance of the outer contour containing the source points from the bounda- ry – Sd, the distance of the inner contour containing the source points from the boundary – Sm (Fig.3), the number of source points – NC, the num- ber of collocation points – M and thickness of the elements – E. In the first method, for given M and NC equal to 100 for problem I-II and 120 for the case of III-VII, Sd in the range from 0.001 to 2 and Sm were searched through in the range from 0.001 to E –0.01. Out of thousands of errors, the minimumvalue was determinedwhich was the value of the errormethod. De- termination of theminimumvalue of bugwas becoming very time consuming. In this case, the error is multidimensional, with many local minima (Fig.11) and numerical methods of searching for the optimal solution fail. For this re- ason, genetic algorithms to the parameters set out in Table 1 were used to determine the optimal values of M, NC, Sd and Sm for a given thickness of E. After determining the optimal values M, NC, Sd and Sm, the values of the minimum error on the internal and external contour were determined, which are presented in Tables 2-8. Based on the numerical results presented in the Tables, it can be concluded that the smallest values or error were obta- ined for case I, and the largest for case III (square with square centered hole). The possible explanation of the large error in this case can be the existence of the corners of the inner boundary. In the remaining cases, when the inner contour or contours were smooth (circular), the values of error were much smaller. Fig. 11. Searching errors Method of fundamental solution and genetic algorithms... 1071 Table 1.Parameters used for genetic algorithm calculation Population size 40 Crossover probability 0.9 Probability of mutation 0.1 Parameter of random number generator 100 Number of generations 100 The lower limit for Sd 0.001 The upper limit for Sd 2 The lower limit for Sm 0.001 The upper limit for Sm 0.01 The lower limit for M 5 The upper limit for M 25 The lower limit for NC 5 The upper limit for NC 25 Table 2.Values of maximum local errors – case I M NC E A∗ δMAXou A ∗∗ δMAXini 26 26 0.05 1.75 6.00E-10 0.01 4.49E-16 46 46 0.1 0.37 1.91E-06 0.01 8.33E-15 46 46 0.15 0.39 1.35E-06 0.01 9.76E-16 46 46 0.2 0.41 1.19E-06 0.01 1.45E-15 46 46 0.25 0.83 6.56E-11 0.02 1.92E-15 46 46 0.3 0.83 7.41E-11 0.07 1.14E-14 46 46 0.35 1.00 2.88E-12 0.09 8.75E-14 46 46 0.4 0.97 1.12E-11 0.01 9.50E-13 46 46 0.45 1.00 1.69E-11 0.01 1.63E-11 A∗ – Optimal values outer contour Sd A∗∗ – Optimal values inner contour Sm For the case of circular with circular centered hole, the calculation of the dimensionless torsional stiffness was conducted. The calculation of stiffness based on the approximate solutionwas completed bydividing the cross section into triangular elements which were integrated with the seven point Gauss quadrature. The obtained results were compared with the exact formula for dimensionless stiffness of the annular region Ms = π 2 [(1 2 )4 −E4 ] (6.1) 1072 P. Gorzelańczyk Table 3.Values of maximum local errors – case II M NC E A∗ δMAXou A ∗∗ δMAXini 65 65 0.05 0.16 4.45E-02 0.04 2.08E-10 65 65 0.1 0.15 8.51E-03 0.09 1.67E-08 70 70 0.15 0.04 7.74E-02 0.12 3.61E-07 65 65 0.2 0.04 7.94E-02 0.18 4.45E-06 65 65 0.25 0.04 2.28E-01 0.24 2.83E-05 65 65 0.3 0.04 4.30E-01 0.33 1.21E-04 65 65 0.35 0.04 1.81E-01 0.39 2.74E-04 65 65 0.4 0.08 1.46E-01 0.51 1.00E-03 65 65 0.45 0.01 8.90E-01 0.41 3.18E-02 Table 4.Values of maximum local errors – case III M NC E A∗ δMAXou A ∗∗ δMAXini 164 164 0.05 0.10 1.67E-05 0.01 5.72E-12 164 164 0.1 0.21 1.32E-05 0.01 4.44E-08 164 164 0.15 0.26 1.77E-03 0.07 7.67E-04 164 164 0.2 0.21 2.32E-05 0.01 1.24E-03 164 164 0.25 0.15 4.36E-05 0.08 5.09E-03 164 164 0.3 0.08 6.39E-05 0.08 7.18E-03 164 164 0.35 0.09 1.23E-04 0.07 9.13E-03 164 164 0.4 0.17 4.40E-05 0.01 3.20E-02 164 164 0.45 0.29 1.47E-02 0.01 4.71E-02 Table 5.Values of maximum local errors – case IV M NC E A∗ δMAXou A ∗∗ δMAXini 96 96 0.05 0.08 7.96E-02 0.01 1.01E-05 96 96 0.1 0.52 1.42E-03 0.05 4.19E-06 96 96 0.19 0.06 3.35E-01 0.11 2.03E-05 96 96 0.2 0.07 1.05E-01 0.04 9.14E-05 96 96 0.25 0.18 2.74E-01 0.11 1.57E-05 96 96 0.4 0.04 3.98E-01 0.06 1.20E-03 224 224 0.45 0.07 5.12E-02 0.03 1.85E-03 96 96 0.49 0.01 7.48E+00 0.41 2.75E-02 A∗ – Optimal values outer contour Sd A∗∗ – Optimal values inner contour Sm Method of fundamental solution and genetic algorithms... 1073 Table 6.Values of maximum local errors – case V M NC E A∗ δMAXou A ∗∗ δMAXini 96 96 0.05 0.34 1.09E-03 0.03 1.89E-10 224 224 0.06 0.07 9.72E-03 0.01 1.50E-09 96 96 0.1 0.45 8.71E-05 0.08 3.70E-09 96 96 0.11 0.52 3.63E-04 0.08 1.58E-09 96 96 0.2 0.54 7.97E-03 0.12 1.20E-07 96 96 0.22 0.32 3.72E-02 0.11 1.08E-07 224 224 0.25 0.21 4.87E-04 0.04 1.14E-06 224 224 0.45 0.18 6.32E-01 0.04 7.05E-05 Table 7.Values of maximum local errors – case VI M NC E A∗ δMAXou A ∗∗ δMAXini 69 69 0.025 0.61 8.80E-06 0.02 1.24E-11 69 69 0.05 0.54 3.79E-05 0.03 2.23E-11 69 69 0.075 0.27 1.33E-04 0.05 6.52E-11 69 69 0.1 1.35 1.58E-04 0.13 7.29E-10 72 72 0.125 1.06 6.01E-05 0.09 8.76E-10 72 72 0.15 1.37 1.41E-04 0.10 9.78E-10 69 69 0.175 1.23 1.61E-03 0.23 2.56E-08 72 72 0.2 0.80 1.57E-03 0.12 4.05E-06 72 72 0.225 1.85 1.40E-02 0.10 4.27E-05 Table 8.Values of maximum local errors – case VII M NC E A∗ δMAXou A ∗∗ δMAXini 78 78 0.025 0.34 1.09E-02 0.02 3.90E-10 78 78 0.05 0.43 5.34E-03 0.04 2.81E-09 78 78 0.075 0.01 4.41E-01 0.08 2.56E-08 84 84 0.1 0.17 2.87E-02 0.07 1.43E-07 78 78 0.125 0.17 3.80E-02 0.12 6.08E-06 78 78 0.15 0.39 2.59E-01 0.14 3.97E-05 84 84 0.175 0.51 7.68E-03 0.10 3.97E-05 78 78 0.2 0.60 1.97E-02 0.12 1.15E-03 78 78 0.225 0.63 1.06E-01 0.09 4.21E-03 A∗ – Optimal values outer contour Sd A∗∗ – Optimal values inner contour Sm 1074 P. Gorzelańczyk Next, the resultwascomparedwith the resultsof thedimensionless stiffness resulting fromBrendt’s formula for an annular region Ms = π 4 (1 2 +E )3(1 2 −E ) (6.2) The results of comparison are presented in Fig.12. Fig. 12. Comparison of the results given by proposed here formula (4.6) – points, Bredt’s formula (6.2) – solid line and exact solution (6.1) – broken line for the case of circular with circular centered hole; M =46, NC =46 7. Conclusion Themethod of fundamental solutionswas successfully applied to solve the bo- undary problem of torsion of bars withmultiply connected cross sections. The paper considered seven caseswithmultiply connected cross sections,where the thickness E varied between 0 < E < 0.5 for problems I-V and 0 < E < 0.25 for problems VI and VII.With the application of this method, the maximum error for problems I, II, and IV-VII is smaller on the inner boundary than on the outer boundary. For problem III, the errors on the inner and outer boundary are comparable. The values of themaximum errors are really small. Furthermore, on the basis of the numerical results for the circle with cir- cular hole cross section of the bar, it was concluded that well knownBrendt’s method of calculating stiffness of thin-walled bars can be successfully applied to bars with the dimensionless thickness larger than E =0.4. In all the cases studied, it was observed that the boundary condition is satisfied. Stress func- tion values along the X axis converge to zero, reaching zero at the edge of the outer section. Method of fundamental solution and genetic algorithms... 1075 In the first method for given M and NC, Sd and Sm were searched through. Out of thousands of errors, theminimumvalue, which was the value of themethod error, was determined. Determination of theminimum value of bug was becoming very time consuming. For this reason, genetic algorithms were used to determine the optimal coefficients M, NC, Sd and Sm of the method of fundamental solutions in order to determine values of the error method. Application of genetic algorithms increased the speed of finding the minimum error in comparison with the method of searching. References 1. Arutiunian N.H., Abramian B.L., 1962, Torsion of Prismatic Bars, AN SSSR, Erevan 2. ArutiunianN.H., AbramianB.L., 1963,Torsion of Elastic Body, Gosudar- stvennoe Izdatelstvo Fiziko-Matematicheskǒı Literatury,Moskva 3. Bogomolny A., 1985, Fundamental solution method for elliptic boundary value problems, SIAM Journal on Numerical Analysis, 22, 644-669 4. Bredt R., 1896, Kritische Bemerkungen zur drehungselastizitat, Zeitschrift des Vereines Deutscher Ingenieure, 40, 785-790 5. Chen K.H., Kao J.H., Chen J.T., Young D.L., Lu M.C., 2006, Regula- rizedmeshlessmethod formultiply-connected-domainLaplace problems,Engi- neering Analysis with Boundary Elements, 30, 882-896 6. Dyląg Z., Jakubowicz A., Orłoś Z., 1999, Wytrzymałość materiałów, tom I,WydawnictwoNaukowo-Techniczne,Warszawa 7. FairweatherG.,Karageorghis A., 1988a,The almansi of fundamental so- lutions fornumerical solutionof the biharmonic equation, International Journal for Numerical Methods in Engineering, 26, 1668-1682 8. FairweatherG.,Karageorghis A., 1998b,Themethod of fundamental so- lutions for elliptic boundary value problems, Advances in Computational Ma- thematics, 9, 69-95 9. Fairwearther G., Karageorghis A., 1989, The simple layer potential me- thod of fundamental solutions for certain biharmonic equations, International Fluids for Numerical Methods in Fluids, 9, 1221-1234 10. Fairweather G., Karageorghis A., Martin P.A., 2003, The method of fundamental solutions for scattering and radiation problems,Engineering Ana- lysis with Boundary Elements, 27, 759-769 1076 P. Gorzelańczyk 11. Goldberg D.E., 1995,Algorytmy genetyczne i ich zastosowania, Wydawnic- two Naukowo-Techniczne,Warszawa 12. Golberg M.A., Chen C.S., 1998, The method of fundamental solutions for potential, Helmholtz and diffusion problems, [In:]Boundary Integral Methods – Numerical and Mathematical Aspects, Golberg M.A. (Edit.), Boston, Compu- tationalMechanics Publications, 103-176 13. Hematiyan M.R., Doostfatemeh A., 2007, Torsion of moderately thick hollow tubes with polygonal shapes,Mechanics Research Communications, 34, 528-537 14. Karageorghis A., 2009, A practical algorithm for determining the optimal pseudo-boundary in the method of fundamental solutions,The Advances in Applied Mathematics and Mechanics, 1, 4, 510-528 15. Katsurada M., 1990, Asymptotic error analysis of the charge simulationme- thod, Journal of the Faculty of Science, University of Tokyo, Section 1A, 37, 635-657 16. Katsurada M., Okamoto H., 1988, A mathematical study of the charge simulation method, Journal of the Faculty of Science, University of Tokyo, Section 1A, 35, 507-518 17. KatsuradaM.,OkamotoH., 1996,The collocationpoints of the fundamen- tal solutionmethod for thepotential problem,Computers andMathematics with Applications, 31, 123-137 18. KimY.Y.,YoonM.S., 1997,AmodifiedFourier seriesmethod for the torsion analysis of bars with multiply connected cross sections, International Journal Solids Structures, 34, 4327-4337 19. Kitagawa T., 1988, On the numerical stability of the method of fundamental solutions applied to the Dirichlet problem, Japan Journal of Industrial and Applied Mathematics, 35, 507-518 20. KitagawaT., 1991,Asymptotic stability of the fundamental solutionmethod, Journal of Computational and Applied Mathematics, 38, 263-69 21. Kołodziej J.A., Fraska A., 2005, Elastic torsion of bars possessing regular polygon in cross-section,Computers and Structures, 84, 78-91 22. Kołodziej J.A., Klekiel T., 2008, Optimal parameters of method of fun- damental solutions for Poisson problems in heat transfer by means of gene- tic algorithms, Computer-Assisted Mechanics and Engineering Sciences, 15, 99-112 Method of fundamental solution and genetic algorithms... 1077 23. KupradzeV.D.,AleksidzeM.A., 1964,Themethod of functional equations for the approximate solution of certain boundary-value problems, Zurnal Vy- chislennǒı Matematiki i Matematycheskǒı Fizyki, 4, 683-715 [in Rusian] 24. Li Z.-C., 2009, The method of fundamental solutions for annular shaped do- mains, Journal of Computational and Applied Mathematics, 1, 355-372 25. Lubarda V.A., 2009, On the torsion constant of multicell profiles and its maximization,Thin-Walled Structures, 47, 798-806 26. Mathon R., Johnston R.L., 1977, The approximate solution of elliptic boundary-value problems by fundamental solutions, SIAM Journal on Nume- rical Analysis, 14, 638-650 27. Mejak G., 2000, Optimization of cross-section of hollow prismatic bars in torsion,Communications in Numerical Methods in Engineering, 16, 687-695 28. Morassi A., 1999, Torsion of thin tubes with multicell cross-section, Mecca- nica, 34, 115-132 29. Nishimura R., Nishimori K., Ishihara N., 2000, Determining the arrange- ment of fictious charges in charge simulation method using genetic algorithm, Journal of Electrostatic, 49, 95-105 30. Nishimura R., Nishimori K., Ishihara N., 2001, Automatic arrangement of fictitious charges and contour points in charge simulation method for polar coordinate system, Journal of Electrostatics, 51/52, 618-624 31. Nishimura R., Nishihara M., Nishimori K., Ishihara N., 2003, Automa- tic arrangement of fictitious charges and contour points in charge simulation method for two spherical electrodes, Journal of Electrostatics, 57, 337-346 32. PolyaG.,Weinstein A., 1950,On the torsion rigidity ofmultiply connected cross-sections,Annals of Mathematics, 52, 154-163 33. Tsangaris Th., Smyrilis Y. S., Karageorghis A., 2006, Numerical ana- lysis of themethod of fundamental solutions for harmonic and biharmonic pro- blems inannulardomains,NumericalMethod forPartial Differential Equations, 22, 507-539 34. Wang C.Y., 1995, Torsion of a flattened tube,Meccanica, 30, 221-227 35. Wang C.Y., 1998, Torsion of tubes of arbitrary shape, International Journal Solids Structures, 35, 719-731 36. Weinel E., 1932, Das Torsionsproblem fr den exzentrischen Kreisring, Inge- nieur Archivs, 3, 67-75 1078 P. Gorzelańczyk Zastosowanie metody rozwiązań podstawowych oraz algorytmów genetycznych do zagadnienia skręcania prętów o przekroju wielospójnym Streszczenie Wartykule rozważano skręcanie pretów zwielospójnymprzekrojempoprzecznym za pomocą metody rozwiązań podstawowych (MRP). Do wyznaczenia optymalnych parametrów MRP wykorzystano algorytmy genetyczne. W pracy rozważano siedem problemów testowych. Bezwymiarowe sztywności skręcania liczono za pomocą nume- rycznego całkowania funkcji naprężeń dla jednego z przypadków. Te sztywności po- równywano ze ścisłą sztywnością dla pierwszego przypadku i ze sztywnością uzyskaną ze wzoru Bredta dla cieńkich przekrojów poprzecznych. Manuscript received December 6, 2010; accepted for print January 17, 2011