Abstract
Fourier-based homogenization schemes are useful to analyze heterogeneous microstructures represented by 2D or 3D image data. These iterative schemes involve discrete periodic convolutions with global ansatz functions (mostly fundamental solutions). The convolutions are efficiently computed using the fast Fourier transform. FANS operates on nodal variables on regular grids and converges to finite element solutions. Compared to established Fourier-based methods, the number of convolutions is reduced by FANS. Additionally, fast iterations are possible by assembling the stiffness matrix. Due to the related memory requirement, the method is best suited for medium-sized problems. A comparative study involving established Fourier-based homogenization schemes is conducted for a thermal benchmark problem with a closed-form solution. Detailed technical and algorithmic descriptions are given for all methods considered in the comparison. Furthermore, many numerical examples focusing on convergence properties for both thermal and mechanical problems, including also plasticity, are presented.
Similar content being viewed by others
Notes
The first node according to the node numbering of the reference element Z, i.e., if \(\underline{\alpha }\) is the zero vector.
Here, the fact is considered that \(n_1, \ldots , n_d\) are even (cf. Sect. 2.2).
The index \( i^*\in \left\{ 1,\ldots ,n\right\} \) depends on the exact choice of the array representation in Sect. 2.6 and is uniquely determined by the requirement \( \mathscr {F}\left( \left[ \,{\mathcal {I}}\,\right] \right) \equiv 1 \) in view of (20). The DFT is generally defined such that \( {\varvec{x}}^{\langle i^*\rangle } \) corresponds to the first entry (with respect to all dimensions) of the input array. E.g., in the array representation of our implementation, this first entry corresponds to the first corner of the RVE \( (i^*=c_1) \).
The confinement to isotropic (local) constitutive relations is made for simplicity but does not reflect a restriction of the FANS scheme.
An assumption of this kind corresponds to a necessary Dirichlet condition that renders the boundary value problem well-defined.
With the confinement to the local constitutive behavior specified by Fourier’s law (22), the effective constitutive relation is expressed by \( \bar{{\varvec{q}}} = - \bar{{\varvec{\kappa }}} \bar{{\varvec{g}}} \). The global conductivity \( \bar{{\varvec{\kappa }}}\in Sym({\mathbb {R}}^d) \) is a positive semidefinite second-order tensor, which is in general anisotropic despite the fact that the local conductivity is assumed to be isotropic at all positions \( {\varvec{x}}\in \varOmega \). Obviously, this type of global constitutive behavior is independent of the macroscopic temperature \( \bar{\theta } \), which can therefore assume arbitrary values in the present case. However, \(\bar{\theta }\) needs to be imposed to the microscopic problem when temperature-dependent conductivities are incorporated.
In Eq. (28) it is exploited that \( {\underline{\underline{{\mathcal {P}}}}}^\mathsf{T} \underline{f} = \underline{0} \) holds, which is due to the assumed absence of heat sources.
The FANS fundamental solution is always computed using full integration.
By exploiting the superposition principle, the number of DFTs can be reduced from \( 2^dd \) to \( 2^d \).
Aside from the characteristic length \(l_\mathrm{c}\), the Fourier space expression of the error measure from Sect. 2.4 in [31] has been corrected by the factor \( 2 \pi \sqrt{n} \) in Eq. (52). Without this correction, a fixed error tolerance would be hit earlier when the discretization is refined for a given problem.
The closed-form solution field is not only divergence-free, but also curl-free within both phases. Hence it is only applicable to heat conduction problems if both phases are isotropic. Note that the necessity to numerically compute the Jacobian elliptic delta function at each discretization point makes the closed-form solution computationally more demanding than FANS. To facilitate reproduction of our results, we mention a minor error in [32]: Both denominators in Eq. (3.5) must be replaced by their square roots.
A violation of the HS bounds would of course not contradict the validity of the numerical results, but would indicate that the considered medium has anisotropy which is not negligible.
Otherwise they loose their sparsity and, thus, efficiency.
References
Arbenz P, van Lenthe GH, Mennel U, Müller R, Sala M (2008) A scalable multi-level preconditioner for matrix-free \(\upmu \)-finite element analysis of human bone structures. Int J Numer Methods Eng 73(7):927–947
Bornert M, Bretheau T, Gilormini P, Jeulin D, Michel J-C, Moulinec H, Suquet P, Zaoui A (2001) Homogénéisation en mécanique des matériaux 1. Hermes Science Publications, Cardiff
Brisard S (2017) Reconstructing displacements from the solution to the periodic Lippmann–Schwinger equation discretized on a uniform grid. Int J Numer Methods Eng 109(4):459–486
Brisard S, Dormieux L (2010) FFT-based methods for the mechanics of composites: a general variational framework. Comput Mater Sci 49(3):663–671
Brisard S, Dormieux L (2012) Combining Galerkin approximation techniques with the principle of Hashin and Shtrikman to derive a new FFT-based numerical method for the homogenization of composites. Comput Methods Appl Mech Eng 217–220:197–212
Cooley JW, Tukey JW (1965) An algorithm for the machine calculation of complex Fourier series. Math Comput 19:297–301
Dumontet H (1983) Homogénéisation par développements en séries de Fourier. Comptes rendus de l’Académie des sciences. Série II, Mécanique, physique, chimie, astronomie 296:1625–1628
Dvorak G J (1992) Transformation field analysis of inelastic composite materials. Proc Math Phys Sci 437:311–327
Dvorak GJ, Benveniste Y (1992) On transformation strains and uniform fields in multiphase elastic media. Proc R Soc Lond A Math Phys Eng Sci 437:291–310
Eshelby J D (1957) The determination of the elastic field of an ellipsoidal inclusion, and related problems. Proc R Soc Lond A 241:376–396
Feyel F, Chaboche J-L (2000) \(\text{ FE }^{2}\) multiscale approach for modelling the elastoviscoplastic behaviour of long fibre SiC/Ti composite materials. Comput Methods Appl Mech Eng 183(3–4):309–330
Frigo M, Johnson S G (2005) The design and implementation of FFTW3. Proc IEEE 93(2):216–231 Special issue on “Program Generation, Optimization, and Platform Adaptation”
Fritzen F, Böhlke T (2011) Nonuniform transformation field analysis of materials with morphological anisotropy. Compos Sci Technol 71(4):433–442
Fritzen F, Leuschner M (2013) Reduced basis hybrid computational homogenization based on a mixed incremental formulation. Comput Methods Appl Mech Eng 260:143–154
Fritzen F, Leuschner M (2015) Nonlinear reduced order homogenization of materials including cohesive interfaces. Comput Mech 56(1):131–151
Fritzen F, Hodapp M, Leuschner M (2014) GPU accelerated computational homogenization based on a variational approach in a reduced basis framework. Comput Methods Appl Mech Eng 278:186–217
Gélébart L, Mondon-Cancel R (2013) Non-linear extension of FFT-based methods accelerated by conjugate gradients to evaluate the mechanical behavior of composite materials. Comput Mater Sci 77:430–439
Hashin Z, Shtrikman S (1963) A variational approach to the theory of the elastic behaviour of multiphase materials. J Mech Phys Solids 11(2):127–140
Kabel M, Böhlke T, Schneider M (2014) Efficient fixed point and Newton–Krylov solvers for FFT-based homogenization of elasticity at large deformations. Comput Mech 54(6):1497–1514
Kochmann, J., Wulfinghoff, S., Ehle, L., Mayer, J., Svendsen, B., Reese, S (2017) Efficient and accurate two-scale FE-FFT-based prediction of the effective material behavior of elasto-viscoplastic polycrystals. Comput Mech. https://doi.org/10.1007/s00466-017-1476-2
Leuschner M, Fritzen F (2017) Reduced order homogenization for viscoplastic composite materials including dissipative imperfect interfaces. Mech Mater 104:121–138
Michel J-C, Suquet P (2003) Nonuniform transformation field analysis. Int J Solids Struct 40(25):6937–6955 Special issue in Honor of George J. Dvorak
Michel J-C, Suquet P (2004) Computational analysis of nonlinear composite structures using the nonuniform transformation field analysis. Comput Methods Appl Mech Eng 193(48–51):5477–5502 Advances in Computational Plasticity
Michel J-C, Suquet P (2016a) A model-reduction approach to the micromechanical analysis of polycrystalline materials. Comput Mech 57(3):483–508
Michel J-C, Suquet P (2016b) A model-reduction approach in micromechanics of materials preserving the variational structure of constitutive relations. J Mech Phys Solids 90:254–285
Mishra N, Vondřejc J, Zeman J (2016) A comparative study on low-memory iterative solvers for FFT-based homogenization of periodic media. J Comput Phys 321:151–168
Moakher M (2006) On the averaging of symmetric positive-definite tensors. J Elast 82(3):273–296
Mori T, Tanaka K (1973) Average stress in matrix and average elastic energy of materials with misfitting inclusions. Acta Metall 21(5):571–574
Moulinec H, Suquet P (1994) A fast numerical method for computing the linear and nonlinear mechanical properties of composites. Comptes rendus de l’Académie des sciences. Série II, Mécanique, physique, chimie, astronomie 318(11):1417–1423
Moulinec H, Suquet P (1995) A FFT-based numerical method for computing the mechanical properties of composites from images of their microstructures. In: Pyrz R (ed) IUTAM symposium on microstructure-property interactions in composite materials, vol 37. Solid mechanics and its applications. Springer, Springer, pp 235–246
Moulinec H, Suquet P (1998) A numerical method for computing the overall response of nonlinear composites with complex microstructure. Comput Methods Appl Mech Eng 157(1–2):69–94
Obnosov YV (1999) Periodic heterogeneous structures: New explicit solutions and effective characteristics of refraction of an imposed field. SIAM J Appl Math 59(4):1267–1287
Ponte P (1991) Castañeda. The effective mechanical properties of nonlinear isotropic composites. J Mech Phys Solids 39(1):45–71
Reuss A (1929) Berechnung der Fließgrenze von Mischkristallen auf Grund der Plastizitätsbedingung für Einkristalle. ZAMM J Appl Math Mech/ Zeitschrift für Angewandte Mathematik und Mechanik 9(1):49–58
Schneider M, Ospald F, Kabel M (2016) Computational homogenization of elasticity on a staggered grid. Int J Numer Methods Eng 105(9):693–720
Schneider M, Merkert D, Kabel M, Schneider M, Merkert D, Kabel M (2017) FFT-based homogenization for microstructures discretized by linear hexahedral elements. Int J Numer Methods Eng 109(10):1461–1489 nme.5336
Suquet P (1990) Une méthode simplifiée pour le calcul des propriétés élastiques de matériaux hétérogènes à structure périodique. Comptes rendus de l’Académie des sciences. Série II, Mécanique, physique, chimie, astronomie 311:769–774
Talbot DRS, Willis JR (1985) Variational principles for inhomogeneous non-linear media. IMA J Appl Math 35(1):39–54
Terada K, Miura T, Kikuchi N (1997) Digital image-based modeling applied to the homogenization analysis of composite materials. Comput Mech 20(4):331–346
Voigt W (1910) Lehrbuch der Kristallphysik. Teubner, Berlin
Vondřejc J, Zeman J, Marek I (2014) An FFT-based Galerkin method for homogenization of periodic media. Comput Math Appl 68(3):156–173
Willot F (2015) Fourier-based schemes for computing the mechanical response of composites with accurate local fields. Comptes Rendus Mécanique 343(3):232–245
Willot F, Abdallah B, Pellegrini Y-P (2014) Fourier-based schemes with modified Green operator for computing the electrical response of heterogeneous media with accurate local fields. Int J Numer Meth Eng 98(7):518–533
Yvonnet J (2012) A fast method for solving microstructural problems defined by digital images: a space Lippmann–Schwinger scheme. Int J Numer Meth Eng 92(2):178–205
Zeman J, Vondřejc J, Novák J, Marek I (2010) Accelerating a FFT-based solver for numerical homogenization of periodic media by conjugate gradients. J Comput Phys 229(21):8065–8071
Zeman J, de Geus T W J, Vondřejc J, Peerlings R H J, Geers M G D (2017) A finite element perspective on nonlinear FFT-based micromechanical simulations. Int J Numer Methods Eng 111(10):903–926
Acknowledgements
Funding by the German Research Foundation (DFG) via grant FR-2702/3 and within the Emmy Noether Programme of the DFG via grant FR-2702/6 is highly acknowledged. The stimulating discussions within the Cluster of Excellence Simulation Technology (funded by DFG in the scope of the excellence initiative under grant EXC 310) at the University of Stuttgart are highly appreciated.
Author information
Authors and Affiliations
Corresponding author
Appendices
Appendix A: Memory requirement
Taking into account symmetry, the number of nonzero entries that has to be stored for the periodized stiffness matrix \( \underline{\underline{{K^\mathrm{f}}}} \), defined for thermal problems in (28), is
where n is the number of pixels/voxels and \( {\mathcal {D}} \) denotes the number of degrees of freedom per node, i.e., \( {\mathcal {D}}=1 \) for thermal problems and \( {\mathcal {D}}=d \) for general (non-) linear mechanical problems. When the storage demand is reduced by a matrix-free element-by-element (EBE) approach for thermal/mechanical linear problems with isotropic constituents, one/two stored scalar variables per element are sufficient. For more general problems, the EBE approach becomes less efficient in terms of memory requirement, and for nonlinear problems its profitability depends crucially on whether or not reduced integration can be used.
Appendix B: Construction of fundamental solutions for the SLS method
Algorithm 7 describes how the fields \( {\varvec{\psi }}^{(1)}, \ldots , {\varvec{\psi }}^{(d)} \), which constitute the fundamental solution for the original SLS, are constructed. For FA-SLS, these are re-organized in expanded fields \( {\varvec{\Psi }}^{(1)}, \ldots , {\varvec{\Psi }}^{(d)} \) that are defined on the full set of element centers \( {\mathcal {X}} \). To construct \( {\varvec{\Psi }}^{(j)} \), the field \( {\varvec{\psi }}^{(j)} \) is positioned in \( {\mathcal {X}} \) such that its central element is placed at the position \( {\varvec{x}}^{\left\{ i^*\right\} } \) characterizing the identity with respect to convolution (cf. Sect. 2.8), whereby periodic continuations across the boundaries are taken into account. Afterwards, the remainder of \( {\mathcal {X}} \) is padded with zeros. The procedure is illustrated for the two-dimensional case in Fig. 23.
Rights and permissions
About this article
Cite this article
Leuschner, M., Fritzen, F. Fourier-Accelerated Nodal Solvers (FANS) for homogenization problems. Comput Mech 62, 359–392 (2018). https://doi.org/10.1007/s00466-017-1501-5
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00466-017-1501-5