Jtam-A4.dvi JOURNAL OF THEORETICAL AND APPLIED MECHANICS 51, 3, pp. 543-552, Warsaw 2013 SOLVING DIRECT AND INVERSE PROBLEMS OF PLATE VIBRATION BY USING THE TREFFTZ FUNCTIONS Artur Maciąg, Anna Pawińska Kielce University of Technology, Faculty of Management and Computer Modelling, Kielce, Poland e-mail: maciag@tu.kielce.pl; a.pawinska@tu.kielce.pl The paper presents an approximate method of solving direct and inverse problems which are described by a non-homogenous plate vibration equation. The key idea of the presented approach is to use solving polynomials that satisfy the considered homogenous differential equation identically. Inhomogeneity is expanded into the Taylor series and then, for each monomial, the inverseoperator is calculated. In thepaper, the properties of solving functions are investigated – a theorem concerning their linear independence is formulated and proved. The method of identification of the load (source) is described. It belongs to the group of inverseproblems.Thepaper includes exampleswhich illustrate theusefulness of themethod. Key words: plate vibration, Trefftz method, inverse problem, source identification 1. Introduction The term Trefftz methods means a wide class of approaches of solving partial differential equ- ations. The solution is approximated by a linear combination of functions satisfying the equ- ation identically. In general, we can distinguish two classes of Trefftz functions. The first one are F-functions called fundamental solutions. These functions have singularity in certain points. Usually, these points are chosen outside the domain. The second one are H-functions (Herrera functions). Good example are here harmonic or heat polynomials. Plate polynomials (Trefftz functions for plate vibration equations) presented in this paper can be classified as H-functions. Themethod originates from the paper byTrefftz (1926) andwere developed by various authors, including:Herrera, Sabina,Kupradze, Jirousek, Leon, Zieliński andZienkiewicz, who considered mostly stationary problems.Non-stationary problemswere solved usually by timediscretization. The time as a continuous variable, first appeared in Rosenbloom and Widde (1956) (1D heat polynomials). Next this aspect of the method was developed for the wave equation and ther- moelasticity problems in Grysa and Maciąg (2011), Maciąg (2004, 2005, 2007, 2011a), Maciąg and Wauer (2005a,b). The Trefftz functions for the equation of beam vibration are presented in Al-Khatib et al. (2008). Also comprehensive monographs exist concerning the Trefftz func- tions method (Ciałkowski and Frąckowiak, 2000; Grysa, 2010, Kołodziej and Zieliński, 2009; Li et al., 2008; Maciąg, 2009; Qin, 2000). Problems of plates vibrations can be solved by means of the Trefftz method. For example, F-functions were used in Reutskiy and Yu (2007), Wu et al. (2011). H-functions were used in Blanc et al. (2007), Vanmaele et al. (2007). Each method has advantages and disadvantages. For example, in four papers mentioned above, the authors assumed harmonic form of the vibrations. This approach is very effective for analyzing natural frequencies and solving direct problems. Trefftz functions with time as a continuous variable se- ems to bemore effective for solving inverse problems. Plate polynomials used here are described inMaciąg (2011b). The presented article can be considered as a remarkable step forward in comparison with the work by Maciąg (2011b). Firstly, the properties of the plate polynomials are considered here. Secondly, the method of source identification (inverse problem) is described. As a rule, 544 A.Maciąg, A. Pawińska inverse problems are difficult to solve.Moreover, the solution can be sensitive to disturbances in input data. The examples presented in this paper showhigh effectiveness of theTrefftz functions method for solving direct and inverse problems. 2. Stating the problem Let us consider the inhomogeneous equation of vibrations of a plate in the dimensionless coor- dinates ∂4u ∂x4 +2 ∂4u ∂x2∂y2 + ∂4u ∂y4 + ∂2u ∂t2 =Q(x,y,t) (x,y)∈ (0,1)× (0,1) t> 0 (2.1) where Q(x,y,t) is the load of the plate. In the case of a direct problem, equation (2.1) should be completed by suitable initial and boundary conditions. Initial conditions describe the deflection and velocity at the time t = 0. Boundary conditions indicate the support of the plate. In the case of identification of the load, we additionally assume that deflections of the plate in chosen internal points are known (internal responses). 3. Linear independence of the Trefftz functions Trefftz functions for the equation of plate vibration(plate polynomials) are presented inMaciąg (2011b). In this work, recurrent formulas for the plate polynomials and their derivatives are described. Table 1 includes the solving polynomials from degree 0 to 4. Table 1.The solving polynomials from degree 0 to 4 Degree Number of Polynomialspolynomials 0 1 1 1 3 x,y,t 2 5 x 2 2 ,xy, y2 2 , tx,yt 3 7 x 2t 2 ,xyt, y2t 2 , x 3 6 , x2y 2 , xy2 2 , y3 6 4 9 x 3t 6 , x 2yt 2 , xy 2t 2 , y 3t 6 , x 4 24 − t 2 2 yx3 6 , x 2y2 4 − t2, xy 3 6 , y 4 24 − t 2 2 According toTable 1,we can infer that exactly 2n+1plate polynomials of degree n (n­ 0) exist.Aparticularly importantpropertyof them is their linear independence,which is the content of Theorem 1. Theorem 1. Accurate to the polynomial of third degree, 2n+1 linearly independent plate po- lynomials of degree n, n­ 0 exist. Proof Let us denote un as a linear combination of the monomials of degree n un =αn00x n+(α(n−1)10y+α(n−1)01t)x n−1+(α(n−2)20y 2+α(n−2)11yt+α(n−2)02t 2)xn−2 + . . .+(α1(n−1)0y n−1+α1(n−2)1y n−2t+ . . .+α10(n−1)t n−1)x +α0n0y n+α0(n−1)1y n−1t+ . . .+α00nt n+R (3.1) Solving direct and inverse problems of plate vibration ... 545 where αpqr – coefficients, R – polynomial of three variables of degree less than n. Obviously, coefficients αpqr might be expressed in the form αpqr = 1 p!q!r! ∂nun ∂xp∂yq∂tr (3.2) We will show that if un is a linear combination of plate polynomials, the coefficients αpqr for r­ 2 equal zero. Using formula (3.2) and the equation of homogenous plate vibrationwe obtain αpqr = 1 p!q!r! ∂n−2 ∂xp∂yq∂tr−2 ∂2un ∂t2 = 1 p!q!r! ∂n−2 ∂xp∂yq∂tr−2 ( − ∂4un ∂x4 −2 ∂4un ∂x2∂y2 − ∂4un ∂y4 ) =− 1 p!q!r! ( ∂n+2un ∂xp+4∂yq∂tr−2 −2 ∂n+2un ∂xp+2∂yq+2∂tr−2 − ∂n+2un ∂xp∂yq+4∂tr−2 ) =0 (3.3) Basing on equality (3.3) the linear combination takes the form un =αn00x n+(α(n−1)10y+α(n−1)01t)x n−1+(α(n−2)20y 2+α(n−2)11yt)x n−2 + . . .+(α1(n−1)0y n−1+α1(n−2)1y n−2t)x+α0n0y n+α0(n−1)1y n−1t+R (3.4) where R is a properly chosen polynomial. It is worth mentioning that in formula (3.4) there occur only coefficients of the type αpq0, αpq1, in the number correspondingly to n+1 and n. To simplify the notation, we denote w=un−R In order to obtain the polynomial R, the operator L= ∂4 ∂x4 +2 ∂4 ∂x2∂y2 + ∂4 ∂y4 + ∂2 ∂t2 (3.5) should be applied to equation (3.4). Thenwe get: L(w)+L(R)= 0, hence L(R)=−L(w), and finally R=L−1(−L(w))+L−1(0). It follows that there are exactly 2n+1 linearly independent plate polynomials of degree n. 4. The Trefftz function method The solution to equation (2.1) with suitable initial and boundary conditions is approximated by the linear combination: u(x,y,t) ≈ w(x,y,t) = ∑N n=1 cnVn+wp, where cn are coefficients, Vn –Trefftz functions, wp –particular solution. In order to obtain cn,weminimize the functional describing the fitting of the approximate solution to the given initial and boundary conditions. The approximation of the particular solution wp is determined by the inverse operator L −1(Q). To obtain approximation of the inverse operator for the function Q(x,y,t), the function should be expanded into the Taylor series. Thenwe calculate inverse operators for monomials by using recurrent formulas presented inMaciąg (2011b). 5. Examples The plate polynomials will be used for solving a direct and an inverse problem for plate vibra- tions. In both cases, the exact solution will be known. It allows one to check the quality of the approximation. Two kinds of the load are considered. The first has a form of a polynomial. The second is a trigonometric function. The approximation will be calculated for the two dimension- less time intervals: (0,1/16) and (0,1/4). Additionally, the sensitivity of the method according to the noisy data is checked. 546 A.Maciąg, A. Pawińska 5.1. The solution of a direct problem As the first example let us consider the non-homogeneous equation of plate vibrations in dimensionless coordinates: ∂4u ∂x4 +2 ∂4u ∂x2∂y2 + ∂4u ∂y4 + ∂2u ∂t2 = 24t2(y4−2y3+y)+2t2(12x2−12x)(12y2 −12y) 5 + 24t2(x4−2x3+x)+2(x4−2x3+x)(y4−2y3+y) 5 (x,y)∈ (0,1)× (0,1) t> 0 (5.1) Equation (5.1) has been completed by initial and boundary conditions u(x,y,0)= sin(πx)sin(πy) 1000 ∂u(x,y,0) ∂t =0 u(0,y,t) =u(1,y,t) =u(x,0, t) =u(x,1, t) = 0 ∂2u(0,y,t) ∂x2 = ∂2u(1,y,t) ∂x2 = ∂2u(x,0, t) ∂y2 = ∂2u(x,1, t) ∂y2 =0 (5.2) The exact solution to problem (5.1) and (5.2) is given by the following formula u(x,y,t) = sin(πx)sin(πy)cos(2π2t) 1000 + t2(x4−2x3+x)(y4−2y3+y) 5 In order to determine the approximation of the solution, a linear combination of polynomials was used, together with the function being a particular solution to (5.1), i.e. u(x,y,t)≈w(x,y,t) = N∑ n=1 cnVn+ t2(x4−2x3+x)(y4−2y3+y) 5 (5.3) The approximate solution is calculated in the time interval (0,∆t). The coefficients cn are chosen so that the functional described by formula (5.4) is minimized I = 1∫ 0 dx 1∫ 0 ( w(x,y,0)− sin(πx)sin(πy) 1000 )2 dy ︸ ︷︷ ︸ condition (5.2)1 + 1∫ 0 dx 1∫ 0 (∂w(x,y,0) ∂t )2 dy ︸ ︷︷ ︸ condition (5.2)1 + ∆t∫ 0 dt 1∫ 0 ( [w(0,y,t)]2 +[w(1,y,t)]2 + (∂2w(0,y,t) ∂x2 )2 + (∂2w(1,y,t) ∂x2 )2) dy ︸ ︷︷ ︸ conditions (5.2)2,3 + ∆t∫ 0 dt 1∫ 0 ( [w(x,0, t)]2 +[w(x,1, t)]2 + (∂2w(x,0, t) ∂y2 )2 + (∂2w(x,1, t) ∂y2 )2) dx ︸ ︷︷ ︸ conditions (5.2)2,3 (5.4) The necessary condition to minimize functional (5.4) has a form: ∂I/∂c1 = . . .= ∂I/∂cN =0. Figure 1 shows the exact solution for the point x = y = 0.5 and its approximation by polynomials from the order 0 to: (a) – 9, (b) – 11, (c) – 13 in the time interval t∈ (0,1/16). In order to check the quality of the approximation, two kinds of errors can be calculated. The first, given by formula: ε= √√√√√√√√ ∆t∫ 0 [ w ( 1 2 , 1 2 , t ) −u ( 1 2 , 1 2 , t )]2 dt ∆t∫ 0 [ u ( 1 2 , 1 2 , t )]2 dt ·100% (5.5) Solving direct and inverse problems of plate vibration ... 547 describes the error in the point x= y=0.5. The second δ= √√√√√√√√ ∆t∫ 0 dt 1∫ 0 dy 1∫ 0 [w(x,y,t)−u(x,y,t)]2dx ∆t∫ 0 dt 1∫ 0 dy 1∫ 0 [u(x,y,t)]2dx ·100% (5.6) describes the error in the entire time-space domain. Fig. 1. Exact solution and their approximation by polynomials from order 0 to: (a) – 9, (b) – 11, (c) – 13 Table 2 shows values of errors (5.5) and (5.6) depending on the degree and the number of polynomials. For instance, if the approximation contains all polynomials from the order 0 to 9, we take 100 polynomials and so on. It is seen that the error decreases if the approximation containsmoreTrefftz functions. The error at the level of 0.0374% is remarkably low. Comparing errors (5.5) and (5.6), we can see that the error in the entire domain is bigger than in the particular point. However, this error remains at a low level. Table 2.Error (5.5) and (5.6) of the approximation in dependance of the degree of the polyno- mials and the number of polynomials Degree 9 10 11 12 13 (No. of polynomials) (100) (121) (144) (169) (196) ε [%] 3.48 0.286 0.253 0.0683 0.0374 δ [%] 5.10 0.376 0.328 0.118 0.0667 The Trefftz functions can be used as base functions in the Finite Elements Method (see Maciąg, 2009, 2011a). When considering inverse problems, a particularly important question is how big the time-space element should be. Therefore, we examine the influence of the length of the time interval on the error. In the calculations presented above, the time interval (0,1/16) has been used. Now we take into consideration the time interval (0,1/4). Figure 2 shows the exact solution for the point x= y=0.5 and its approximations by polynomials from the order 0 to: (a) – 9, (b) – 11, (c) – 13. The errors according to formulas (5.5) and (5.6) have been calculated to check the quality of the approximation in the time interval (0,1/4). The values of both errors are presented in Table 3. Comparing Tables 2 and 3, it is seen that we get a better approximation in a shorter time interval. In a longer interval, the solution has more oscillations – this result has been expected. All tables show thatmorepolynomials lead to better results. Itmeans that if wewant to obtain a better approximation in a longer time interval, more polynomials have to be taken. For example for 225 polynomials, the error (5.5) is at the level of 3.59% for time interval (1,1/4). 548 A.Maciąg, A. Pawińska Fig. 2. The exact solution for the point x= y=0.5 and its approximations by polynomials from order 0 to: (a) – 9, (b) – 11, (c) – 13 Table 3. Error (5.5) and (5.6) of approximation depending on the degree of polynomials and the number of polynomials Degree 9 10 11 12 13 (No. of polynomials) (100) (121) (144) (169) (196) ε [%] 64.3 58.3 30.7 21.4 7.67 δ [%] 65.8 56.7 28.4 19.5 7.09 In the secondexample let us consider theproblemwhich isdescribedby thenon-homogeneous equation for (x,y)∈ (0,1)× (0,1), t> 0 ∂4u ∂x4 +2 ∂4u ∂x2∂y2 + ∂4u ∂y4 + ∂2u ∂t2 = 2t2π4 sin(πx)sin(πy)+sin(πx)sin(πy) 50 (5.7) and conditions (5.2). The exact solution has the form u(x,y,t) = sin(πx)sin(πy)cos(2π2t) 1000 + t2 sin(πx)sin(πy) 100 As an approximate solution we take a linear combination of polynomials completed by the particular solution u(x,y,t)≈w(x,y,t) = N∑ n=1 cnVn+ t2 sin(πx)sin(πy) 100 (5.8) The approximate solutionwill be calculated in the time interval (0,1/16). The coefficients cn of inear combination (5.8) are calculated by minimizing functional (5.4). The errors of the appro- ximation have been calculated according to formulas (5.5) and (5.6). The results are presented in Table 4. Table 4. Error (5.5) and (5.6) of the approximation depending on the degree of polynomials and number of polynomials Degree 9 10 11 12 13 (No. of polynomials) (100) (121) (144) (169) (196) ε [%] 3.52 0.289 0.256 0.0691 0.0379 δ [%] 5.16 0.381 0.332 0.120 0.0675 Similarly as before, it is seen that the error decreases if the approximation contains more Trefftz functions.The error remains at a low level and is bigger for the entire domain.Comparing Tables 2 and 4, it is seen that the error in both kinds of the non-homogeneity stays at the same level. Solving direct and inverse problems of plate vibration ... 549 5.2. The identification of the load imposed to the plate – an inverse problem Nowadays, there exist a lot of methods that can be applied to solving linear problems with partial differential equations. Unfortunately, most of them cannot be used for solving inverse problems.TheTrefftz FunctionsMethod seems to be a very convenient approach to solving such problems, which is its most important advantage. In general, different types of inverse problems exist, includingaboundary inverse problem(identification of theboundarycondition), geometric inverse problems (identification of the shape of the body) and many others. One of the kinds of inverse problems is also the source identification (identification of the load). In this paper, we propose to apply Trefftz functions for the identification of the load Q(x,y,t) applied to the plate. Let us consider the non-homogeneous equation in dimensionless coordinates L(u)=Q(x,y,t), (x,y)∈ (0,1)× (0,1) t> 0 (5.9) where function Q(x,y,t) is an unknown load and the operator Lhas the form in (3.5). Equation (5.9) is completed by initial and boundary conditions (5.2). Additionally, we assume that the values of deflection of the plate in the internal points are known (internal responses) uijk =u ( i 10 , j 10 , k∆t 50 ) i,j =1, . . . ,9 k=1, . . . ,50 (5.10) For numerical simulation, we take the known function Q. Next, we calculate the internal re- sponses and then identify the load. Because we know the function Q, we can calculate the error of the solution.We assume that the approximation of the solution u(x,y,t) has the form u(x,y,t)≈w(x,y,t) = N∑ n=1 cnVn+ K∑ k=1 αkvk (5.11) where Vn are plate polynomials, vk are values of the inverse operator L −1 for monomials {1,x,y,t,x2, t2,y2,xt,ty,xy,t3, t2y,t2x,ty2,xyt,tx2,y3,xy2,x2y,x3, t4, t3y, t3x,t2y2, t2xy,t2x2, ty3, txy2, tyx2, tx3,y4,xy3,x2y2,yx3,x4, . . .} (5.12) In means that Q(x,y,t)≈L ( K∑ k=1 αkvk ) (5.13) The recurrent formulas for the inverse operator L−1 for monomials are presented inMaciąg (2011b). In order to determine the coefficients cn and αk, a functional similar to (5.4) should be build. As thefirst example of an inverse problem, let us consider theproblemdescribedbyequations (5.1) and (5.2). However, now we assume that the load Q(x,y,t) = 24t2(y4−2y3+y)+2t2(12x2−12x)(12y2 −12y) 5 + 24t2(x4−2x3+x)+2(x4−2x3+x)(y4−2y3+y) 5 (5.14) is unknown (we use the exact solution only for numerical simulation). Let us denote Q̃(x,y,t) as the approximation of the exact load Q(x,y,t). Similarly as before, we can define two kinds of the error. The first describes the error in the point x= y=0.5 ε= √√√√√√√√ ∆t∫ 0 [ Q̃ ( 1 2 , 1 2 , t ) −Q ( 1 2 , 1 2 , t )]2 dt ∆t∫ 0 [ Q ( 1 2 , 1 2 , t )]2 dt ·100% (5.15) 550 A.Maciąg, A. Pawińska the second describes the error in the entire time space domain δ= √√√√√√√√ ∆t∫ 0 dt 1∫ 0 dy 1∫ 0 [Q̃(x,y,t)−Q(x,y,t)]2dx ∆t∫ 0 dt 1∫ 0 dy 1∫ 0 [Q(x,y,t)]2dx ·100% (5.16) The values of errors (5.15) and (5.16) for ∆t=1/16 are shown in Table 5. Table 5. Error (5.15) and (5.16) depending on the degree of polynomials and the number of polynomials (∆t=1/16) Degree 10 11 12 13 (No. of polynomials) (121) (144) (169) (196) ε [%] 2.97 2.18 0.635 0.447 δ [%] 6.86 3.00 2.47 2.66 Table 5 shows that more polynomials in the approximation decreases the error. Obviously, the error is bigger in the entire time-space domain than in a particular point. However, the error at the level of 2.66% is very small, considering that the problem is inverse. The values of errors (5.15) and (5.16) for the bigger time interval (∆t=1/4) are shown in Table 6. Table 6. Error (5.15) depending on the degree of polynomials and the number of polynomials (∆t=1/4) Degree 10 11 12 13 (No. of polynomials) (121) (144) (169) (196) ε [%] 20.9 14.5 3.87 5.65 δ [%] 43.6 23.8 11.4 12.6 Obviously, the approximation in abigger time interval isworse.Moreover, we observe a slight increase of the error in the last column. Thismay be caused by the Runge effect (waving of the polynomials of the high degree). Itmeans that if we intend to solve an inverse problem, the time interval cannot be too long. The second example of an inverse problem is described by equations (5.7) and conditions (5.2). Similarly as before we assume that the exact load is unknown (the exact solution is used to generate internal responses in numerical simulation). As before, in this case the exact load is known. Therefore, we can calculate the errors (5.15) and (5.16). The results for ∆t = 1/16 and for ∆t = 1/4 are presented in Tables 7 and 8 correspondingly. Table 7. The error (5.15) and (5.16) depending on the degree of polynomials and the number of polynomials (∆t=1/16) Degree 10 11 12 13 (No. of polynomials) (121) (144) (169) (196) ε [%] 5.66 4.33 1.45 0.999 δ [%] 14.4 6.79 4.81 5.48 Also in this case, more polynomials in the approximation decrease the error, and the error is bigger in the entire time-space domain than in a particular point. The approximation in a bigger time interval is worse also for the second kind of the load. It confirms the conclusion that if we solve an inverse problem, the time interval cannot be too long. Solving direct and inverse problems of plate vibration ... 551 Table 8. Error (5.15) and (5.16) depending on the degree of polynomials and the number of polynomials (∆t=1/4) Degree 10 11 12 13 (No. of polynomials) (121) (144) (169) (196) ε [%] 36.1 26.3 7.73 10.4 δ [%] 88,7 48,3 25,4 28,1 5.2.1. Noisy data The solution of the inverse problem can be very sensitive to disturbances of the input data. Therefore, it is very important to examine each new method according to sensitivity to noisy data. To this end, the internal responses have been randomlydisturbed according to the formula udisijk =uijk(1+ δijk) i,j=1, . . . ,9 k=1, . . . ,50 (5.17) where δijk has a normal distributionwith themean equal to 0 and standard deviation 0.02. The disturbed internal responseshave beenused to calculate theapproximation of the load Q(x,y,t). Table 9 shows error (5.15) of the approximation of the load in polynomial form (5.14). Table 9. Error (5.15) of the identification of the load in polynomial form for disturbed data (∆t=1/4) Degree 10 11 12 13 (No. of polynomials) (121) (144) (169) (196) ε [%] 20.8 14.2 3.94 5.18 We expect that the error for the noisy data shouldbe bigger.However, comparing the results presented inTables 6 and 9, we can observe that the noisy data does not cause any considerable increase in the error. It means that the presented approach is resistant to disturbance of the internal responses. This is a key quality of the method in terms of solving inverse problems. 6. Concluding remarks In the paper, a method of solving problems described by an inhomogeneous equation of plate vibration is presented. The approach is relatively simple and suitable for solving both direct and inverse problems.The theorem of linear independence of the plate polynomials has been proved. The greatest advantage of themethod is its usability for solving the ill-posed inverse problems. In the paper, the way of identification of the load (non-homogeneity) has been presented. The results described in the paper show a remarkable efficiency of the method for solving inverse problems.Moreover, the approach proposed here is relatively invulnerable to disturbance of the input data. References 1. Al-Khatib M.J., GrysaK., Maciąg A., 2008, Themethod of solving polynomials in the beam vibration problems, Journal of Theoretical and Applied Mechanics, 46, 2, 347-366 2. Blanc L., Blanze C., Rouch P., 2007, A multiscale ”Trefftz” computational method for medium-frequency vibrations of assemblies of heterogeneous plates with uncertainties,Computers and Structures, 85, 595-605 552 A.Maciąg, A. Pawińska 3. Ciałkowski M.J., Frąckowiak A., 2000, Heat Functions and Their Application for Solving Heat Transfer and Mechanical Problems, PoznańUniversity of Technology Publishers [in Polish] 4. Grysa K., 2010,Trefftz Functions and their Applications in Solving the Inverse Problems, Kielce University of Technology Publishers [in Polish] 5. Grysa K., Maciąg A., 2011, Solving direct and inverse thermoelasticity problems by means of Trefftz base functions for finite element method, Journal of Thermal Stresses, 34, 4, 378-393 6. Kołodziej J.A., Zieliński A.P., 2009,Boundary Collocation Techniques and their Application in Engineering, WIT Press, Southampton, Boston 7. Li Z.-C., Qiu Lu T.-T., Hu H.-Y., Cheng H.-D., 2008,The Trefftz and Collocation Methods, WIT Press, Southampton, Boston 8. Maciąg A., 2004, Solution of the three-dimensional wave equation by using wave polynomials, PAMM – Proceedings in Applied Mathematics and Mechanics, 4, 706-707 9. Maciąg A., 2005, Solution of the three-dimensional wave polynomials,Mathematical Problems in Engineering, 5, 583-598 10. Maciąg A., 2007, Wave polynomials in elasticity problems, Engineering Transactions, 55, 2, 129-153 11. Maciąg A., 2009, Trefftz Functions for Some Direct and Invers Problems of Mechanics, Kielce University of Technology Publishers [in Polish] 12. Maciąg A., 2011a, The usage of wave polynomials in solving direct and inverse problems for two-dimensional wave equation. International Journal for Numerical Methods in Biomedical Engi- neering, 27, 1107-1125 13. Maciąg A., 2011b, Trefftz function for a plate vibration problem, Journal of Theoretical and Applied Mechanics, 49, 1, 97-116 14. Maciąg A., Wauer J., 2005a, Solution of the two-dimensional wave equation by using wave poly-nomials, Journal of Engineering Mathematics, 51, 4, 339-350 15. Maciąg A., Wauer J., 2005b, Wave polynomials for solving different types of two-dimensional wave equations,Computer Assisted Mechanics and Engineering Sciences, 12, 87-102 16. Qin Q.H., 2000, The Trefftz Finite and Boundary Element Method, WIT Press, Southampton, Boston 17. Reutskiy S.Yu., 2007, The method of fundamental solutions for problems of free vibrations of plates,Engineering Analysis with Boundary Elements, 31, 10-21 18. RosenbloomP.C.,WidderD.V., 1956, Expansion in terms of heat polynomials and associated functions,Transactions of the American Mathematical Society, 92, 220-266 19. Trefftz E., 1926, EinGegenstuek zumRitz’schen Verfahren, [In:]Proceedings 2nd International Congres of Applied Mechanics, Zurich, 131-137 20. Vanmaele C., Vandepitte D.. Desmet W., 2007, An efficient wave based prediction techni- que for plate bending vibrations,Computer Methods in Applied Mechanics and Engineering, 196, 3178-3189 21. Wu C.S., Young D.L., Fan C.M., 2011, Frequency response analyses in vibroacoustics using the method of fundamental solutions,Computational Mechanics, 47, 519-533 Manuscript received April 15, 2012; accepted for print September 6, 2012