Jtam-A4.dvi JOURNAL OF THEORETICAL AND APPLIED MECHANICS 50, 3, pp. 743-753, Warsaw 2012 50th Anniversary of JTAM NONLINEAR DYNAMIC ANALYSIS OF VISCOELASTIC MEMBRANES DESCRIBED WITH FRACTIONAL DIFFERENTIAL MODELS John T. Katsikadelis School of Civil Engineering, National Technical University of Athens, Athens, Greece e-mail: jkats@central.ntua.gr The dynamic response of an initially flat viscoelastic membrane is investigated. The visco- elastic model is described with fractional order derivatives. The membrane is subjected to surface transverse and inplane dynamic loads. The governing equations are three coupled second order nonlinear partial FDEs (fractional differential equations) of hyperbolic type in terms of the displacement components. These equations are solved using the BEM for frac- tional partial differential equations developed recently by Katsikadelis. Without excluding other viscoelastic models, the herein employed material is the Kelvin-Voigt model with a fractional order derivative. Numerical examples are presented which not only demonstrate the efficiency of the solution procedure, but also give a better insight into this complicated but very interesting response of structural viscoelastic membranes. It is worth noting that in case of resonance, phenomena similar to those of the Duffing equation are observed. Key words: viscoelastic membranes, nonlinear vibrations, nonlinear fractional differential equations, analog equationmethod, boundary elements 1. Introduction Membranes made of linear viscoelastic materials are extensively used in modern engineering applications. Various models of the integer differential form have been proposed in order to describe themechanical behavior of suchmaterials (e.g.Maxwell,Voigt,Kelvin,Zener).Recently, many researchers have shown that viscoelastic models of the differential form with fractional derivatives are in better agreement with the experimental results than the integer derivative models (Stiassnie, 1979; Adolfsson et al., 2005; Meral et al., 2010). The dynamic response of pure elastic and viscoelastic membranes, using differential consti- tutive equations of an integer order derivative or hereditary integral type models, have been examined by many investigators (Plaut, 1990; Koivurova and Pramila, 1997; Jevine and Mac- kintosh, 2002;Wineman, 2007; Goncales et al., 2009). The use of fractional differentialmodels is very limited and is restricted to viscoelastic beams (Galucio et al., 2004) and three dimensional viscoelastic bodies (Schmidt andGaul, 2002). However, viscoelasticmembranes of the fractional type derivative have not been analyzed to date. The reason is plausible as the response of such membranes is described by a system of nonlinear partial FDEs for which an analytical solution is out of question, while numerical methods were not available until lately. Recently, Katsika- delis developed the AEM, which in conjunction with an integral equation approach provides an efficient computational tool for solving linear and nonlinear FDEs, ordinary (Katsikadelis, 2009) and partial (Katsikadelis, 2011). This method is general and paves the way to analyze com- plicated systems whose response is described by such equations. It has been already employed to solve several problems. Among them, the linear fractional diffusion wave equation in boun- ded inhomogeneous anisotropic bodies (Katsikadelis, 2008), the response of the inhomogeneous anisotropic viscoelastic bodies (Nerantzaki and Babouskos, 2011) and the nonlinear dynamic 744 J.T. Katsikadelis response of viscoelastic plates (Babouskos and Katsikadelis, 2009; Katsikadelis and Babouskos, 2010). Without excluding other fractional differential models (Nerantzaki and Babouskos, 2011), the employed herein viscoelastic material is described by the Kelvin-Voigt type model with a fractional order derivative      σx σy τxy      = E 1−ν2     1 ν 0 ν 1 0 0 0 1−ν 2            εx+ηD a cεx εy +ηD a cεy γxy+ηD a cγxy        (1.1) where E, ν are the elastic material constants, η the viscoelastic parameter and Dac the Caputo fractional derivative of the order α defined as Dαcu(t)=            1 Γ(m−α) t ∫ 0 u(m)(τ) (t− τ)α+1−m dτ for m−1<α 0 and b(i)(x, t), i=1,2,3 represent timedependentfictitious sources, unknown in the first instance. The solution to Eq. (3.1)1 is given in integral form (Katsikadelis, 2002) εu(x, t) = ∫ Ω u∗b dΩ− ∫ Γ (u∗q−q∗u) ds x∈Ω∪Γ (3.2) in which q = u,n; u ∗ = ℓnr/2π is the fundamental solution to Eq. (3.1) and q∗ = u,∗n its derivative normal to the boundary with r= ‖ξ−x‖, x∈Ω∪Γ and ξ ∈Γ ; ε is the free term coefficient (ε=1 if x∈Ω, ε=ω/2π if x∈Γ and ε=0 if x /∈Ω∪Γ) and ω is the interior angle between the tangents of the boundary at the point x; ε=1/2 for points where the boundary is smooth.Eq. (3.2) is solvednumericallyusing theBEM.Theboundary integrals areapproximated using N constant boundary elements, whereas the domain integrals are approximated using M linear triangular cells. The domain discretization is performedautomatically using theDelaunay triangulation. Since the fictitious source is not defined on the boundary, the nodal points of the triangles adjacent to the boundary are placed on their sides (Fig. 1). Thus, after discretization Fig. 1. Boundary and domain discretization and application Eq. (3.2) at the N boundary nodal points, we obtain Hu=Gq+Ab(1) (3.3) where H,G are N×N known coefficientmatrices originating from the integration of the kernel functions on theboundaryand A is an N×M coefficientmatrix originating fromthe integration of the kernel function on the domain cells; u, q are the vectors of the N nodal displacements Nonlinear dynamic analysis of viscoelastic membranes... 747 and slopes and b(1) the nodal values of the fictitious source at the M domain nodal points, all at time t. The derivatives of u(x, t) inside Ω are obtained by direct differentiation of Eq. (3.2) with ε=1. Further, by applying Eq. (3.2) and its derivatives at the domain nodal points andmaking use of Eq. (3.3) combined with the boundary conditions to eliminate the boundary quantities, we can express u(t) and its derivatives at the domain nodal points in terms of the fictitious source u,pq (t)=S,pqb (1)(t)+s(1),pq p,q=0,x,y (3.4) Similarly, we obtain v,pq (t)=S,pqb (2)(t)+s(2),pq p,q=0,x,y (3.5) and w,pq (t)=S,pqb (3)(t)+s(3),pq p,q=0,x,y (3.6) where S,pq are known M×M matrices and s(i),pq known vectors. For homogeneous boundary conditions, we have s(i),pq=0. The notation u,pq designates u,00=u,u,x0=u,x etc. Applying now Eqs. (2.6) at the M domain nodal points and substituting the involved derivatives from Eqs. (3.4)-(3.6), we obtain M (i) b̈ (1)+F(i)(b(i),Dαcb (i))= f(i) i=1,2,3 (3.7) where M(i) are M ×M consistent mass matrices, F(i) are M × 1 vectors whose elements depend nonlinearly on the arguments and f(i) are known load vectors. Using Eqs. (3.4)-(3.6), initial conditions (2.7) for b(i) become b (i)(0)=S−1(fi−s(i)) ḃ(i)(0)=S−1gi (3.8) Equations (3.7) constitute a system of 3M three-term nonlinear FDEs, which are solved using the time step numerical procedure developed by Katsikadelis (2009). The use, however, of all degrees of freedom, i.e. 3M, may be computationally costly and in some cases inefficient due to the large number of the time dependent variables b (i) k (t). To overcome this difficulty in this investigation, thenumber of degrees of freedom is reducedusing theRitz transformation, namely b(i) =Ψ(i)z(i) (3.9) where z (i) k (t), (k =1, . . . ,L