# Homogenization of a Cauchy continuum towards a

## Comments

## Transcription

Homogenization of a Cauchy continuum towards a

Article published in Journal of the Mechanics and Physics of Solids 99 (2017), 394408. The nal publication is available at www.sciencedirect.com, doi:10.1016/j.jmps.2016.09.010. The preprint comprises corrections of a corrigendum doi:10.1016/j.jmps.2017.07.013. Homogenization of a Cauchy continuum towards a micromorphic continuum Geralf Hütter The micromorphic theory of Eringen and Mindlin, including special cases like strain gradient theory or Cosserat theory, is widely used to model size eects and localization phenomena. The heuristic construction of such theories based on thermodynamic considerations is well-established. However, the identication of corresponding constitutive laws and of the large number of respective constitutive parameters limits the practical application of such theories. In the present contribution, a closed procedure for the homogenization of a Cauchy continuum at the microscale towards a fully micromorphic continuum is derived including explicit denitions of all involved generalized macroscopic stress and deformation measures. The boundary value problem to be solved on the microscale is formulated either for using static or kinematic boundary conditions. The procedure is demonstrated with an example. Keywords: micromorphic theory; homogenization; gradient theory; generalized continua 1. Introduction Classical theories of continuum mechanics can only be applied when the macroscopic wavelengths of relevant eld quantities are much larger than the characteristic microstructural dimensions, a limitation manifested already in the lack of an intrinsic length scale in such continuum theories. However, in many engineering problems this condition is not fullled, e. g. for micro and nanodevices or if material instabilities lead to a localization of deformations. In principle, generalized continuum theories can overcome these limitations. Among the generalized theories of continuum mechanics, micromorphic theory of Eringen and Mindlin [8, 9, 10, 31] has an outstanding role since it incorporates many others like the Cosserat theory or strain gradient theory as special cases. For a recent review and a comprehensive classication the reader is referred to [17, 30]. Phenomenologically, it is well-established how such theories can be constructed based on macroscopic thermodynamic considerations and/or the principle of virtual powers. Thereby, additional, generalized stresses occur which appear in additional balance equations [19, 9, 14, 7, 23]. This requires to formulate respective additional constitutive relations and to identify the corresponding constitutive parameters. Mostly, for this purpose classical constitutive laws are generalized heuristically by linear and reversible approaches for the generalized stress measures (i. e. quadratic ansatzes in the thermodynamic potentials) for simplicity, e. g. [14, 7]. In particular in extension of highly non-linear classical constitutive laws, this is questionable and leads to unrealistic predictions as pointed out recently in [24, 15]. Homogenization of the heterogeneous microstructure oers a solution to this problem. Regarding classical continuum mechanics, the homogenization procedure, whereby the macroscopic stresses and strains are dened as the volume averages of their microscopic counterparts and can be prescribed via 1 corresponding boundary conditions at the microscale, is established already for decades [21, 22]. For the so-called couple stress theory (constrained Cosserat theory) the additional boundary conditions of bending-type were specied intuitively as well [1, 11, 4]. However, for the homogenization from a Cauchy continuum at the microscale towards an unconstrained micromorphic continuum at the macroscale the denition of the macroscopic quantities and the formulation of respective boundary conditions at the microscale is not that trivial. Forest and Sab provided explicit integral expressions for the relations between microscopic and macroscopic kinematic quantities of micromorphic theories [13, 16]. However, it turned out to be a problem that the expressions for the generalized deformation measures could not be transformed to surface integrals and thus, in contrast to classical homogenization, not be prescribed by boundary conditions at the microscale. For that reason polynomial expressions were identied according to several strategies that fulll the integral expressions identically and attempts were made to characterize additional uctuation elds [16, 12, 6, 26]. The drawback of this approach is that no boundary value problem could strictly be formulated on the microscale. Alternatively, Jänicke and Steeb [27] proposed to prescribe Forest's integral denitions of the macroscopic micromorphic deformation measures as minimum constraints to the classical PDE on the microscale. Then, the generalized stress measures are obtained implicitly as respective work-conjugate quantities. However, it had to be postulated that these generalized stresses fulll the balance equations of micromorphic continua as derived by Eringen. In this context it is to be remarked that Eringen [8, 9] derived those balance equations via a spatial averaging procedure and provided explicit denitions of the generalized stresses (and all ux quantities) in terms of a surface operator. Unfortunately, at least to the author's knowledge, Eringen did neither give an expression for this surface operator nor for the relation of the macroscopic deformation measures to the microscopic eld quantities. The present contribution aims to close this gap by providing explicit expressions for the macroscopic generalized stress measures and their work-conjugate deformation measures and formulating the boundary value problem to be solved at the microscale for obtaining the macroscopic micromorphic constitutive relations. The present work is organized as follows: section 2 briey reviews Eringen's micromorphic theory, before the general homogenization procedure is presented in section 3. In particular, thermodynamic considerations in section 3.1 lead to the generalized Hill-Mandel lemma before section 3.2 deals with the micro-macro relations for the generalized measures of stresses and deformations. Based on these ndings, the special case of the second gradient theory is considered in section 4. Subsequently, section 5 demonstrates the proposed procedure for a simple uniaxial example. Finally, the present contribution closes with a discussion and some concluding remarks in sections 6 and 7, respectively. 2. Micromorphic continuum The notation within the present contribution is adopted from Forest and Sab [16], i. e. scalars, vectors and tensors of second, third and forth rank are denoted by and threefold contractions are written as ·, . . ., a, b, c, d and ≈e, e e respectively. Single, double : and respectively, and are computed from left to right, I and denote the second rank identity tensor and the permutation e T e tensor, respectively. The operator (◦) denotes the complete transposition of all indices of a tensor. For T a second rank tensor this is done by the forth rank transposing tensor IT as c = IT : c. Analogously, ≈ ≈ e e 1 S T a symmetrization tensor IS is introduced as c = (c + c ) = IS : c. The symbols x and X refer to the 2 ≈ ≈ e e e e location vector at the microscale and macroscale, respectively. The nabla operator is ∇ whose subscript (∇X or ∇x ) species, whether it is computed with respect to X or x. The material time derivate is i. e. . d..e = dijk eijk . ee denoted by a dot In particular, . (◦). 2 ∆V (X) ξ X x Figure 1: Micromorphic continuum 2.1. Average theorem for general balance laws Consider a general balance equation in the reference conguration D Dt Z I ρϕm dV = Ω with ϕm , ψm and of mass density ρ. ψa x∈Ω of type Z n · ψ a dS + ρψm dV (1) Ω ∂Ω and being the densities of storage, sources and ux, respectively, in a continuum The global balance (1) can be localized as usual to . ρϕm = ∇x · ψ a + ρψm in Ω. (2) According to Eringen [8, 9], and given in detail in appendix A.1, macroscopic counterparts to these Ω balance equations are obtained by dividing the domain into small but nite volumes ∆V (X) as sketched in Figure 1 for each of which (1) is valid. Summing them up and approximating the sum of those many elements as integrals over the macroscopic domain normal is N) X ∈ ΩX (with boundary ∂ΩX whose yields the macroscopic balance law D Dt Z I hρϕm iV dV = Ω D E N · ψa Z S dS + hρψm iV dV (3) ΩX ∂ΩX with an equivalent local version D E . hρϕm iV = ∇X · ψ a + hρψm iV . S (4) Therein, the volume averaging operator is denoted as h◦iV = 1 ∆V (X) Z ◦(x) dV . (5) ∆V (X) As mentioned in section 1, to the author's knowledge no attempt is documented in the literature yet to provide a direct denition of the surface operator h◦iS . 3 2.2. Microscopic balance laws of Cauchy continuum For a geometrically linear analysis to be performed in the following, the Cauchy continuum at the microscale is described by the following balance equations: . ρ=0 Mass: Energy: Entropy: ∇x · σ + ρf − ρv = 0 e :σ=0 e e Angular momentum: Φ and η (7) (8) . Linear momentum: Therein, (6) . 1 ρΦ + ρ (v · v) = ∇x · σ · v − ∇x · q + ρf · v 2 e . ρη + ∇x · h ≥ 0 . (9) (10) are the specic intrinsic energy and entropy, respectively. Furthermore, denote the velocity, stress and body force, respectively. The symbols q and h v, σ e and f refer to the uxes of heat and entropy, respectively. 2.3. Approximation of microscopic velocity field For a micromorphic continuum of degree one, the microscopic velocity eld v(x) is approximated by a polynomial of order one: (11) Lχ . Special cases of (11) are the e is a microrate of rotation or the Lχ = −Ωχ · so that Ωχ (X) e1 . e microdilatational continuum with Lχ = χI. In this context, the Cauchy continuum with Lχ = 0 can 3 e e e be seen as a micromorphic continuum of order zero. with the macroscopic velocity V ṽ = V(X) + Lχ (X) · (x − X) e and a rate of microdeformation Cosserat (micropolar) continuum with 2.4. Macroscopic balance laws According to section 2.1, the microscopic balance laws of the Cauchy continuum from section 2.2 yield the following macroscopic counterparts: 2.4.1. Mass Dening the macroscopic mass density as ρ = hρiV , the macroscopic counterpart to (6) reads (12) . ρ=0 (13) Furthermore, the balances of microinertia . . hρ(x − X)iV = 0 , hρ(x − X) ⊗ (x − X)iV = 0 will be needed. They are obtained by weighting (6) with the relative location x−X (14) and its square, respectively, each of which has the structure of a balance equation (2) so that (4) can be applied yielding (14). 4 2.4.2. Linear and Angular Momentum The microscopic balance of linear momentum (9) yields . . 0 = ∇X · σ S + hρf iV −ρV − Lχ · hρ(x − X)iV {z } | | e{z } e =:Σ e and allows to dene the macroscopic values (15) =:ρf Σ e and f of (extrinsic) stress and body force, respectively. In (15), the approximation (11) of the velocity eld was inserted for the inertia term. Furthermore, a macroscopic counterpart of the balance of linear momentum weighted with the distance ξ = x−X will be needed for the energy balance. With respect to the linear approximation (11) of the velocity eld, this might be interpreted as a Galerkin approach. The balance of linear momentum (9) weighted by x can be written as . ∇x · σ ⊗ x − σ T + ρf ⊗ x − ρv ⊗ x = 0 e e (16) and thus exhibits also the structure of a balance equation (2) so that the average theorem (4) can be applied to obtain a macroscopic counterpart. Upon subtraction of (15) weighted by the macroscopic location X (and written in a form as (16)) and inserting again (11) for the inertia term, one obtains T . . 0 = ∇X · σ ⊗ (x − X) S + σ S − σ T V + hρf ⊗ (x − X)iV −V ⊗ ρξ V − Lχ · ρξ ⊗ ξ V {z } |e {z } | e{z } | e{z } | e | {z } =:M f =ΣT e =:ρm e =:σ̄ T e (17) =:Gρ e Σ and intrinsic e higher order σ̄ , and higher order body forces m, are the sources of hyperstresses M and e e f inertia. The magnitude of the latter depends on the second moment of inertia Gρ . e Looking at eqs. (15) and (17) it would be appealing to dene the yet unspecied macroscopic location This equation can be interpreted that the dierence between extrinsic (Cauchy) stress (mean) stress X as centre of gravity terms ρξ V hρxiV /ρ (as done by Eringen and Mindlin in [31, 8] but not in [9]) so that the would vanish. However, an alternative but similar denition as geometric centre will be chosen as explained in section 3.2. symmetries within the volume element the cross terms ρξ V If the distribution of mass density ∆V (X), ρ(x) X = hxiV exhibits certain e. g. a central symmetry as it is mostly the case, then in the balances of energy and momenta vanish nevertheless. The balance of angular momentum (10) yields the symmetry of the macroscopic intrinsic stress The macroscopic counterpart to (10) 0 = : σ̄ . e e weighted with ξ does (18) not yield additional information. Eq. (18) can be also inserted in the skew-symmetric part of (17): . . 0 = ∇X · M : + ΣT : + ρm : − V × ρξ V − Lχ · Gρ : e e e e e e f e e (19) ΣT : on the right-hand side can be transformed to storage and divergence parts as in classical e e continuum mechanics so that , in absence of body forces and moments, (19) as the skew-symmetric part The term of (17) has conservation type and can, in analogy to classical continuum mechanics, also be interpreted as macroscopic balance of angular momentum (as usually done in Cosserat theory). 2.4.3. Energy and Entropy The averaged microscopic energy balance (7) becomes .E .E 1D ρ (v · v) = ∇X · σ · v S − ∇X · q S + hρf · viV ρΦ + V e | {z } | {z V} 2 D . =:Q =:ρΦ 5 (20) Φ and Q of specic intrinsic energy and heat ux, rev(x) by its approximation (11) and inserting the denitions and allows to introduce the macroscopic values spectively. Replacing the local velocity eld of the macroscopic stress measures yields . . . . 1 1 ρΦ + ρ (V · V) + Gρ : Lχ T · Lχ + V · Lχ · ρξ V 2 2e e e e = ∇X · Σ · V + M : Lχ − ∇X · Q + ρf · V + ρm : Lχ e e e f e (21) The balance of internal energy is obtained by applying the product rule and inserting the balances of momenta (15), (17) and (18) as . . ρΦ = Σ : ∇X V + σ̄ T − ΣT : Lχ + M..∇X Lχ − ∇X · Q e e e e e f . = Σ : ∇X V − Lχ T + σ̄ : Lχ S + M..∇X Lχ − ∇X · Q e e e e e f Thus, the work-conjugate deformation rates to the macroscopic stress measures (22) Σ, σ̄ e e and rates of microstrain, relative deformation and microdeformation gradient, respectively: . . . e = ∇X V − Lχ T e e E = Lχ S e e M f K = ∇X Lχ . e e are the (23) The microscopic entropy balance (8) yields its macroscopic counterpart of identical structure . hρηi +∇X · hhiS ≥ 0 | {z V} |{z} . :=H :=ρS dening the macroscopic values S and H (24) of specic entropy and ux of entropy. 3. Homogenization 3.1. Thermodynamic considerations Let us assume, that the continuum at the microlevel does not only obey balance equations (6)(10), but that it is furthermore a Coleman-Noll continuum [5] so that the specic internal energy potential for temperature θ and stress θ= ∂Φ , ∂η σ: e ∂Φ σ=ρ ∂ε e e Consequently, the microscopic energy balance (7) reduces to . ρD − ρθη = ∇x · q , wherein D Φ forms a (25) (26) denotes the specic dissipation due to a change of internal variables. This equation is again of balance type and thus leads to a macroscopic relation . hρDiV − hρθηiV = ∇X · Q , | {z } | {z } =:ρD (27) . =:ρΘS which implies the given denition of macroscopic values respectively. Eq. (27) can be used to eliminate the heatux D Q and Θ of dissipation and temperature, from the macroscopic balance of internal energy (22): . . . . . . ρΦ = Σ : e + σ̄ : E + M..K + ρΘS − ρD e e e e f e 6 (28) Furthermore, (25) has the consequence that the left-hand side of (28) can be written as D .E . . . ρΦ = ρΦ = hρθηiV + σ : ε V − hρDiV . V e e (29) Equating this relation to (28) and eliminating identical terms on both sides using (31) and the denitions in (27) leads to a generalized Hill-Mandel lemma . . . . . σ : ε V =Σ : e + σ̄ : E + M..K e e e e e e f e . =Σ : ∇X V + σ̄ T − ΣT : Lχ + M..∇X Lχ e e e e e f (30) However, in contrast to classical homogenization it is no ad-hoc requirement but a consequence of the employed denitions of macroscopic quantities and the fact that the continuum at the microscale is of Coleman-Noll type. Φ(E, e, K, S, h) of macroe e h as well as deformation measures E, e and Ke then (28) can be e e e fullled for all kinematically admissible elds V(X) Lχ (X) and S(X) (and thus also arbitrary values e of their gradients) if and only if If the macroscopic specic internal energy can be identied as a function scopic entropy S and intrinsic variables ∂Φ σ̄ = ρ , ∂E e e ∂Φ Σ=ρ , ∂e e e ∂Φ M=ρ , ∂K f e ∂Φ ∂S Θ= Thus, a homogenization can equivalently be performed with respect to Φ and D (and uxes (31) Q and H of heat and entropy in thermomechanical problems). 3.2. Micro-macro transition 3.2.1. Review of classical homogenization In classical homogenization, rstly the thermal and mechanical behavior is considered independently of each other. Furthermore, for the mechanical behavior, static conditions are assumed on the microscale and the macroscopic velocity gradient is postulated to correspond to the average of its microscopic counterpart: D E ∇X V = ∇x v (32) V The right hand side can be transformed to a surface integral so that the macroscopic velocity gradient ∇X V can be prescribed by kinematic boundary conditions. Alternatively, the concept of minimal loading conditions proposed by Jänicke and Steeb [27] is employed here. reversible process. Let us consider rstly a Then, the specic internal energy does not depend on internal variables and the concept of minimal loading conditions can be implemented for classical homogenization by requiring the Lagrange functional D .E D E L = ρΦ + λ∇V : ∇X V − ∇x v → Min. V V e to become a minimum. As Φ is a potential for the stress stationarity conditions is σ e according to (25), the corresponding D E . 0 = σ : δ ε V − λ∇V : ∇x δv . V e e e together with (32). (33) (34) The stationarity conditions (34) and (32) can be generalized to hold also for irreversible processes. In this case, it corresponds to the principle of virtual power on the microscale. 7 δv(x), including the real v(x), a comparison with the Hill-Mandel lemma (30) shows, after reinserting the kinematic In the light of the fact that it holds for all kinematically admissible elds velocity eld micro-macro relations (32), that the Lagrange multiplier correspond to the respective work-conjugate macroscopic stress measure λ∇V = Σ . e e (35) With this substitution, the Euler-Lagrange equations to (34) read ∇x · σ = 0 and σ = σ T e e e n·σ =n·Σ e e in on ∆V (X) (36) ∂∆V (X) (37) Thus, if no further essential boundary conditions are specied, the concept of minimal loading conditions results in the static approach of homogenization. Consequently, the loading to the microscopic volume element needs to be self-equilibrating. This is the case only if the macroscopic stress is symmetric T Σ= e Σ . Alternatively to (37), essential boundary conditions can be specied which then need to fulll the e kinematic constraint (32) a priori corresponding to the kinematic approach of classical homogenization λ∇V e (in this case the constraint term with in (33) is obsolete and can be dropped). homogenization theory, the macroscopic stress is dened as the volume average consistent with (36) and (37). In Σ = σ V e e classical which is 3.2.2. Extension to micromorphic theory For the envisaged micromorphic theory, the macroscopic stress was already dened in (15) by the surface Σ = σ S and the intrinsic stress was introduced as volume average σ̄ = σ V . For both stress e e e e measures to be able to dier, the surface operator needs to be dened in a consistent way to (36) and operator (37) to be valid for all shapes of ∆V but must not coincide with the volume average assured by dening the surface operator 1 as 1 Σ = σ S := ∆V e e h◦iV . This is Z ∂∆V ξ ⊗ n · σ dS . e (38) It can easily be veried that inserting (37) from classical homogenization to (38) yields an identity for Σ. Furthermore, denition (38) ensures that the homogenization procedure is e becomes innitesimally small (asymptotic self-consistency, see appendix A.2). ∆V denition (38) of the surface operator h◦iS available, it is possible to evaluate the the macroscopic stress exact in the limit that With the explicit additional stress measures of the micromorphic theory. In particular, the hyperstress becomes M = σξ e f S 1 = ∆V Z ∂∆V ξ ⊗ n · σ ⊗ ξ dS e (39) and, applying Gauss theorem, the dierence of intrinsic and extrinsic macroscopic stresses according to denition (17) can be written as D E Σ − σ̄ = ξ ⊗ ∇x · σ . e e e V For the classical theory from section 3.2.1, inserting (36) and (37) leads to the denition ξ V = 0, (40) M=0 f and Σ − σ̄ = 0 e e under i. e. that the macroscopic location corresponds to the geometric center of the volume element: X = hxiV . 1A (41) similar denition was chosen in [28, 29] for the classical macroscopic stress. However, the hyperstress in [28, 29] diers from (39) by a factor of 1/2. 8 This implies a denition of the macroscopic velocity as V = hviV . (42) λV M according thef Hill-Mandel Eq. (42) can be incorporated into the Lagrangian (33) by an additional Lagrangian multiplier through which (36) becomes ∇x · σ = λV . e Due to (41), λV does not contribute to σ̄ , Σ e e to (38),(39), (40). A comparison of the corresponding stationarity condition (34) with relation (30) shows that λV = 0 ∇x · σ = 0 e so that enforced. remains in classical homogenization even if (42) is M Thus, consistently denition (38) ensures that the additional stresses shapes of the domain ∆V or and Σ − σ̄ e e vanish for all f in the classical case, i. e. if no additional macroscopic deformations are enforced in (33). In this case the additional balance (17) is fullled identically and what remains is the classical theory of continuum mechanics. M f + Kkji )/2 Furthermore, it has to be remarked that according to (39), the hyperstress . respect to its rst and last index. the Hill-Mandel Thus, only the part K contributes to the e lemma (30) becomes microdeformation . K ijk = (Kijk . is symmetric with of the gradient of internal energy which has the same symmetry. Consequently, . . . σ : ε V =Σ : ∇X V + σ̄ T − ΣT : Lχ + M..K . e e e e e e f e (43) The next question is how (33) needs to be modied such that the additional stress measures M and f eqs. Σ − σ̄ become nonzero. Most easily, this problem is solved by going in inverse directions through e e (33)(40). Revisiting (39) and (40) in the light of (41), this means that the terms ∇x · σ and σ · n need e e to be extended by linear terms. Furthermore, the operator in (40) is insensitive to constant terms as mentioned already. Thus, ∇x · σ e and σ·n e are set to ∇x · σ = A · ξ + B − λV e e n·σ =n·Σ+n·C·ξ e e e in on ∆V (X) (44) ∂∆V (X) . (45) Inserting this ansatz to (39) and (40) allows to express the coecients in terms of the additional stress 2 measures : A = ΣT − σ̄ T · G−1 e e e e 1 1 Cijk = (Mijm + Mmji ) G−1 δik Mmjn G−1 mn mk − 4 2 (2 + δpp ) (46) (47) Therein, the geometric moment is dened as G= ξ⊗ξ V . e (48) Note, that the linear boundary term (47) is not self-equilibrating in (45) for all prescribed hyperstresses M. Consequently, for the boundary value problem (44),(45) at the microscale f non-equilibrating part needs to be compensated by the volume term h i 1 −1 B = ≈IT : C : I = I : M · G : I, T e 2+I:I ≈ e e e f e e 2 Due Bj = Ciji = to have a solution, the 1 Mmjn G−1 mn . 2 + δpp (49) to the symmetry of the hyperstress M, not all components of C contribute to M. Vice versa, (39) does not f only those componentseof C which havefthe respective symmetry and determine C uniquely. Eq. (47) incorporates e uniquely dened. e which are thus 9 Finally, the balance equation and natural boundary conditions at the microscale become i h 1 −1 · G : I − λV , σ T = σ ∇x · σ = ΣT − σ̄ T · G−1 · ξ + I : M T 2+I:I ≈ e e e e e e e e f e e h i 1 1 n · ξ ≈IT : M · G−1 : I n · σ = n · Σ + n · M + MT · G−1 · ξ − 2 2+I:I e e e 4 e f f f e e e in on ∆V (X) (50) ∂∆V (X) (51) 1 1 ni ξi Mnjp G−1 ni σij = ni Σij + ni (Mijp + Mpji )G−1 pm ξm − np 4 2(2 + δkk ) Equations (50) and (51) are the Euler-Lagrange equations of the stationarity condition E D . − λV · hδviV + ΣT − σ̄ T : δv ⊗ ξ V · G−1 0 = σ : δ ε V − Σ : ∇x δv V e e e e e e E E D D 1 .1 −1 −1 . − : ≈IT (52) − M. ≈IS : G · ξ ⊗ ∇x δv G ξ · ∇x δv 2+I:I e V V e f2 e e The actual microscopic velocity eld v(ξ) is among the admissible test elds so that (52) holds also if δv(ξ) is replaced by v(ξ). Then, a comparison with the Hill-Mandel lemma (43) shows that the macroscopic rates of deformation have to be identied as D E ∇X V = ∇x v V Lχ = v ⊗ ξ V · G−1 e. e D E D E 1 1 −1 −1 K = ≈IS : G · ξ ⊗ ∇x v − G ξ · ∇x v : ≈IT 2 2+I:I e V V e e e e . 1 1 1 1 −1 −1 −1 K ijk = hvj,i ξm iV Gmk + hvj,k ξm iV Gmi − hvj,m ξm iV Gik . 2 2 2 2 + δll Eq. (42) for the macroscopic velocity remains valid together with λV = 0. If the stresses variational potential, then the variational problem to (52) and (53) reads in (53) (54) σ e have a D .E D E L = ρΦ + Σ : ∇X V − ∇x v + λV · (V − hviV ) + σ̄ T − ΣT : Lχ − v ⊗ ξ V · G−1 V V e e e e . e D E D E 1 1 . + M.. K − ≈IS : G−1 · ξ ⊗ ∇x v − G−1 ξ · ∇x v : ≈IT → Min. (55) 2 2+I:I e V V e f e e e extension to the classical one (33). Therein, the macroscopic stresses Σ, σ̄ − Σ and M have the role e e e f of Lagrange multipliers to enforce the macroscopic measures of deformation (53). In (53), all deformation measures are insensitive to rigid translations. An ane mapping . v = F·ξ e K but leads to identical values ∇X V = Lχ = F. Consequently, the measures of rate . e e e S . T of deformation Lχ , e = ∇X V − Lχ and K dened in (23) are also insensitive to a rotation with e thus e objective and e so are their e F = · W and are work-conjugate stresses Σ, σ̄ and M. e e e e Note that the micro-macro relations for the deformation measures ∇X V and Lχ inf (53) are identical e to the denitions in [12, 18]. In particular, denitions (42) and (53) for the macroscopic values of V does not aect and Lχ e (together with (41) for the macroscopic location between the microscopic velocity eld D (ṽ − v) 2 E V X) are equivalent to a minimum average error → min (56) v and its macroscopic approximation ṽ according (11). Thus, the ṽ from (11) leads to kinematic micro-macro relations (53) are kinematically consistent in a sense that 10 x2 x2 x2 x1 x1 x1 (a) (b) Figure 2: Loading of the volume element (b) M = b1 b1 b1 − b1 b2 b1 , f V and Lχ . e Furthermore, it can be identities for (c) (c) ∆V (X) for several Σ = b1 b2 − b2 b1 e non-classical stresses: (a) M = b1 b1 b1 , f K diers to the ones given by Forest [12, 18]. theederivation of the deformation measures in However, the expression (53)3 for remarked that, in contrast to [12, 18], the present approach does not require that the macroscopic gradients of the geometry entities ∆V (X) and G(X) e vanish. The loading of the volume element according to (52) for natural boundary conditions is sketched in Figure 2 for several non-classical cases for a rectangular ∆V due to a non-vanishing spherical part of stress M111 = −M121 M bi (whereby the of the coordinate system). Figure 2a shows the eect of hyperstress M111 denote to the base vectors for which volume loads occur according to (49). In contrast, for a deviatoric hyper- f in Figure 2b the macroscopic stress results only in tractions at the microscopic boundary. In both cases, the tractions at opposite faces are identical, i. e. in the terminology of classical homogenization they are antimetric. Figure 2c shows the loading of stress Σ12 = −Σ21 . ∆V for a skew-symmetric extrinsic In this case the torque of the tractions is compensated by volume contributions of opposite direction which twist the Kinematic boundary conditions ∆V . can be specied alternatively to the static ones (51). However, as in classical homogenization they have to fulll the respective kinematic micro-macro relations ad hoc. For the present micromorphic theory, not all volume averages, which dene the macroscopic deformation measures in (53), can be transformed to surface integrals by a partial integration. In particular, the 3 part spherical . K ijk Gik = E i 1 hD (vj ξk ),k − δkk hvj iV 2 + δll V (57) of the symmetric part of the gradient of the rate of microdeformation contains the macroscopic velocity V = hviV as volume average and can thus not be prescribed as kinematic boundary condition. This is consistent with the fact that the corresponding part of the hyperstress appears as volume term in the . microscopic equilibrium condition (50). However, the deviatoric part of .d K ijk K e E E 1 . 1 1D 1D −1 := K ijk − K ljm Glm Gik = (vj ξm ),i G−1 (vj ξm ),k mk + δpp 2 2 2 V . V G−1 mi E 1 D −1 − (vj ξm ),m Gik δpp V (58) 3 For the practically most relevant shapes of the volume element ∆V of a cube or sphere (square and circle in 2D), the geometric moment G is a spherical tensor. e 11 (and the classical velocity gradient ∇X V = D E ∇x v ) can be transformed to pure surface terms by V application of Gauss' theorem and thus be prescribed by a polynomial kinematic boundary condition with additional quadratic term with coecient tensor going with ξ ): D (being symmetric with respect to those indices e v = V0 + ξ · ∇X V + ξ · D · ξ e on ∂∆V (X) (59) Inserting (59) to (57) and (58) leads to . K ijk Gik = Dljm Glm − .d K ijk = Dijk − 1 δkk (Vj − V0j ) 2 + δll (60) 1 Dljm Glm G−1 ik δpp (61) whereby denition (42) of the macroscopic velocity was inserted. The solution to this system of equations is . Dijk =K ijk + 1 (Vj − V0j ) G−1 ik 2 + δll (62) Thus, the kinematic boundary condition (59) becomes . 1 (V − V0 ) G−1 : ξ ⊗ ξ v = V0 + ξ · ∇X V + ξ · K · ξ + 2+I:I e e e e on ∂∆V (X) . (63) Inserting Eq. (63) to the Hill-Mandel lemma (43) shows that the additional constant term has to be identied as the macroscopic velocity V0 = V . Finally, the kinematic boundary condition (59) for micromorphic media to be specied in addition to the microscopic equilibrium condition (50) becomes . . v = V + ξ · ∇X V + ξ · K · ξ e on ∂∆V (X) (63b) . K in (63b) could be replaced equivalently by K as, due to double contraction with ξ , only e e hyperstress M. The boundary problem at those part enters which exhibits the same symmetry as the f stresses have a variational the microscale can also be specied equivalently (in the case the microscopic Note that potential) by the Lagrangian (55) together with the kinematic boundary condition (63b). However, since the boundary condition (63b) already fullls the kinematic constraints for (55) can be reduced to ∇X V and . K, the Lagrangian e D .E L = ρΦ + λ̃ V : (V − hviV ) + σ̄ T − ΣT : Lχ − v ⊗ ξ V · G−1 → Min. V e e e e whereby it has to be remarked that the Lagrange multiplier λ̃ V needs to be distinguished from (64) λV for the Euler-Lagrange equations of (64) to be consistent with the microscopic equilibrium conditions (50). The macroscopic stresses σ̄ , Σ e e and M f remain dened by (17), (38), and (39), respectively. 4. Second gradient theory 4.1. Macroscopic theory The special case of a second gradient theory is obtained if the rate of microdeformation is identied ad-hoc with the macroscopic velocity gradient Lχ := V ⊗ ∇X e 12 (65) v(x) becomes ṽ = V(X) + V ⊗ ∇X · (x − X) . so that the approximation (11) of velocity eld (66) Typically, (66) is motivated as a Taylor-expansion [20, 28, 29]. Under the kinematic constraint (65), the extrinsic stress internal energy, so that (22) becomes . . . Σ e drops out from the balances of total and . ρΦ = σ̄ : E + M..K − ∇X · Q e e f e (67) with the macroscopic rates of deformation being Consequently, Σ e . 1 E= ∇X V + V ⊗ ∇X 2 e . K = ∇X V ⊗ ∇X . e (68) can be eliminated from the balances of momenta by inserting (17) as . Σ = σ̄ − MT · ∇X − ρmT + Gρ · ∇X V e e e f e (69) h . i . 0 = ∇X · σ̄ − MT · ∇X − ρmT + Gρ · ∇X V + ρf − ρV e f e e (70) into (15) yielding whereby it was assumed for simplicity that the volume element at the microlevel exhibits a central symmetry so that ρξ V = 0. The balance of angular momentum σ̄ T = σ̄ e e (eq. (18)) remains valid. 4.2. Micro-macro transition The Hill-Mandel lemma (43) becomes . . . . σ : ε V =σ̄ : E + M..K . e e f e e e (71) Inserting the kinematic constraint (65) to the microscopic Lagrangian (55) yields D .E D E S L = ρΦ + Σ : G−1 · ξ ⊗ v V − ∇x v + λV : (V − hviV ) + σ̄ : ∇X V − v ⊗ ξ V · G−1 V V e e e . e D E D E 1 1 . − G−1 ξ · ∇x v : ≈IT → Min. + M.. K − ≈IS : G−1 · ξ ⊗ ∇x v (72) 2 2+I:I e V V e f e e e Thus, the equilibrium conditions at the microscale (50) and (51), or equivalently (52), of the unconstrained micromorphic theory remain valid for the second gradient theory. The only dierence to an unconstrained micromorphic theory is that the kinematic micro-macro relation (53) for the microdeformation Lχ e is to be replaced by D E v ⊗ ∇x = v ⊗ ξ V · G−1 V e Although the macroscopic extrinsic Σ stress does not contribute to the intrinsic work (67) e gradient theory, it remains the Lagrange multiplier to enforce (73) as microscopic pendant (73) in a second to (65). In this context, it should be remarked that in general a higher theory of mechanics can be reduced to a special lower order theory either by kinematic constraints or by relaxation of the respective kinetic quantity [31]. The kinematically constrained approach ensures full compatibility with the corresponding higher order theory. In particular, condition (73) ensures that the denition (15) of Σ remains valid (and e thus actually the complete procedure outlined in section 2 including a non-vanishing dierence between extrinsic and intrinsic stress Σ e and σ̄ ) e which can be prescribed as natural boundary conditions on the macroscale. In the kinetically relaxed approach, (73) is not enforced and thus Σ e (and the associated jump and boundary conditions) can only be dened implicitly on the macrolevel by (70) via (69) as in [28]. 13 X ρf ξ L L (a) Macroscopic problem (b) Microscopic volume element Figure 3: Homogenization for uniaxial two-phase laminate 5. Example: uniaxial case The developed homogenization procedure shall be demonstrated for the one-dimensional case with periodic microstructure as shown in Figure 3. In particular, for a volume element of length L, the geometric moment is computed as G = L2 /12 ∆V = {ξ ∈ [−L/2, L/2]} so that equilibrium conditions (50) for the volume element become σ 0 (ξ) = 12 4 (Σ − σ̄)ξ + 2 M L2 L for ξ ∈ [−L/2, L/2] (74) For the 1D case, equilibrium condition (74) allows to determine the microscopic stress eld σ(ξ) directly up to a constant of integration. Either static boundary conditions (51) σ(±L/2) = Σ ± 2 M L (75) or kinematic boundary conditions (63b) u(±L/2) = ± L dU L2 + K +U 2 dX 4 (76) complete the boundary value problem at the microscale. In the former case, the constant of integration is computed from (75) whereas for kinematic boundary conditions (76), the constant can be determined from denition (38) (39) for hyperstress Σ = [σ (L/2) + σ (−L/2)]/2. e e M is fullled identically and For both types of boundary conditions, the relation the microscopic stress eld becomes 3 ξ2 4 σ(ξ) = Σ + (σ̄ − Σ) 1 − 2 + 2 Mξ 2 L /4 L (77) Let us now consider the macroscopic problem sketched in Figure 3a of a volume loading f under static conditions so that the macroscopic balance equations (18) and (17) read dΣ + ρf dX dM 0= + Σ − σ̄ + ρm dX 0= In the special case (78) (79) f = 0, i. e. uniform macroscopic deformations, it is obvious that, if only trivial natural M = 0 and boundary conditions for the higher order terms are specied, the macroscopic stresses Σ − σ̄ = 0 fulll the higher order balance of momentum (79) identically. What remains is (78) together with the boundary value problem (74) on the microscale σ 0 (ξ) = 0, σ(±L/2) = Σ. This corresponds to the classical homogenization (which is in the 1D case furthermore the exact solution of the problem 14 with resolved microstructure everywhere) whatever the material law σ(ε) is on the microscale and which volume element is chosen (The situation changes when dynamic eect are considered since then nonvanishing hyperstresses M and a dierence stressmacrocomp − σ̄ are in this case necessary in general to compensate the higher order inertia which will lead, realistically, to dispersion of waves, cf. [31]). For non-vanishing f 6= 0, the higher order balance of momentum (79) is statically indeterminate so that the solution depends on the particular micromorphic constitutive law. The latter shall be derived for linear-elastic behavior σ = Y u0 of all constituents whereby Y denotes Young's modulus. By use of the stress eld (77), the microscopic material law can be solved for u0 (ξ). Inserting it to the denitions (53) of the macroscopic deformation measures yields for (a centro-symmetric unit cell Y (−ξ) = Y (ξ)) the macroscopic constitutive law 1 13 4ξ 2 , + (σ̄ − Σ) 1− 2 Y V Y 2 L V * 2 + 9 1 13 4ξ 2 4ξ 2 1 , + (σ̄ − Σ) χ = huξiV = Σ 1− 2 1− 2 G Y 2 L 4 Y L V V 4 1 4ξ 2 4 0 K = 2 hu ξiV = M 2 . L L Y L2 V dU = hu0 iV = Σ dX The simplest case which contains gradient eects is a constant volumetric loading case, the macroscopic volume loads become ρf = ρf and m=0 (80) ρf = const. In this according to (18) and (17). Thus, for M = const f Consequently, the the constitutive law (80) a particular solution of the macroscopic boundary value problem is and Σ = σ̄ together with the statically determined and classical solution of (78). microscopic stress eld (77) depends linearly on the location as in the exact solution with discretely resolved microstructure everywhere, a prediction which lies beyond the possibilities of classical homogenization. If macroscopic boundary conditions with respect to M or χ are specied which do not coincide with this particular solution, then additional exponentially decaying terms of M and σ̄ − Σ fmacroscopic occur at the macroscopic boundaries. According to (80)1 , such terms have an eect on the U (X), i. e. on the macroscopic stiness. This is reasonable as for non-homogeneous ρf = const the distribution of the microscopic stiness has indeed an eect on the macroscopic displacements loading stiness. Of course, this eect becomes negligible as the ratio of macroscopic length and intrinsic length L increases. For further discussion, the macroscopic constitutive law (80) shall be provided for homogeneous elastic properties i. e. Y = const.: dU σ̄ = Y , dX Σ − σ̄ = 5Y dU −χ , dX M= For a second gradient theory as described in section 4, the constraint 3Y L2 K. 4 (81) χ = dU/dX implies that on K = d2 U/dX 2 . The the macrolevel, the higher order deformation measure in (80)3 is identied as corresponding constraint (73) on the microlevel is implemented by equating (80)1 and (80)2 which allows to eliminate the corresponding Lagrange multiplier Σ from (80)1 . Thus, in the case of homogeneous linear-elastic material at the microscale, the macroscopic relations (78), (79) and (81) become d dM 0= σ̄ − , dX dX σ̄ = Y dU , dX M= 3Y L2 d2 U . 4 dX 2 (82) 6. Discussion In literature, there is a controversial debate whether a homogenization procedure for generalized continuum theory should yield vanishing higher order stresses in the case of homogeneous material at the 15 microscale ∆V or not, compare e. g. [29, 33]. Obviously, for a homogeneous ∆V neither with the present procedure the higher order terms do not vanish in (81), (82) nor was it the case in literature [20, 32, 25]. However, to the author's opinion this behavior is reasonable and the higher order stresses must not vanish in this case since, as argued by Mühlich et al. [33], a still have an inhomogeneous neighboring ∆V locally homogeneous ∆V can whose interaction is still described, in some average sense, by the additional higher order momentum equation (17). This neighbor might also be a macroscopic boundary at which non-vanishing higher order tractions are allowed to be prescribed due to (17). Consequently, from a mathematical point of view, the macroscopic boundary value problem would even be ill-posed if the macroscopic constitutive relation read M = 0. Whether such boundary conditions f are physically reasonable in the case of homogeneous material at the microlevel is of course another question (regarding (81), e. g. at the macroscopic boundary, the intrinsic length L could be correlated to the intrinsic length of an elastic foundation). The explicit denition of the involved generalized stress and deformation quantities derived in the present contribution provides a sound basis for addressing this question. A similar question is whether for heterogeneous material the obtained macroscopic constitutive relations may depend on the particular choice of the unit cell ∆V . It is obvious in (80), that with the present procedure, even in the uniaxial case the non-classical terms (with σ̄ − Σ and M) depend on location and size of the chosen unit cell. This is inevitably necessary since also potential generalized boundary conditions on χ or M require an interpretation with respect to the location of the correspond- ing macroscopic boundary relative to material heterogeneities, see e. g. [34]. However, it is recalled that when the non-classical stresses vanish, either through suitable boundary conditions as described above or in sucient distance to a boundary, then the classical theory is recovered whose solution does not depend on the particular choice of the unit cell. A further question, which was discussed in literature on homogenization of a Cauchy continuum on the microscale to a generalized continuum at the macroscale, was how the additional deformation measures of such theories can be prescribed via boundary conditions at the microscale. The kinematic micro-macro relations derived in the present contribution (as well as those in other papers, e. g. [28, 13, 11]) cannot be completely converted to surface integrals. That is why they are prescribed as integral constraints with corresponding Lagrange multipliers. To the author's opinion, generalized continuum theories like the micromorphic or second gradient theory investigated in the present contribution, should also work in the one-dimensional case, i. e. for rod theory (section 5). For a homogenization of a rod we can prescribe only two boundary conditions at the micro scale. However, in classical theory we have already two macroscopic kinematic quantities, namely strain and displacement. I. e., if we want to incorporate further macroscopic kinematic quantities in a generalized continuum theory, they cannot be linked to the microscale via boundary conditions only but some kind of volume averages must inevitably occur. The implementation of such volume average micro-macro links via Lagrange multipliers is straight forward, both in analytical investigations and numerical implementations. Regarding an FEM implementation, the volume average leads to a few full rows in the otherwise sparse system of equations for whose solution suitable algorithms are favorable. In classical homogenization, it is required that the volume element ∆V should be representative. The term was introduced by Hill [21] and associated with the conditions that the volume element (a) is structurally entirely typical of the whole mixture on average, and (b) contains a sucient number of inclusions for the apparent overall moduli to be eectively independent of the surface values of traction and displacement, so long as these values are `macroscopically uniform'. That is, they uctuate about a mean with a wavelength small compared with the dimensions of the sample, and the eects of such uctuations become insignicant within a few wavelengths of the surface. The contribution of this surface layer to any average can be made negligible by taking the sample large enough. classical homogenization, often In periodic boundary conditions are employed at the microscale to mimic the behavior of an innite and periodic arrangement of identical volume elements to fulll condition (b). For this purpose, periodic uctuations are allowed as deviations from the corresponding kinematic boundary 16 conditions. Regarding micromorphic continua, condition (a) remains reasonable whereas (b) contradicts the intention of most generalized continuum theories (for this reason, the author refrains from using the term representative in the context of micromorphic continua). Anyhow, several attempts were made to extend the concept of periodic boundary conditions to the homogenization of generalized continua in order to have the established classical homogenization contained as a strict special case, e. g.[16, 28, 26, 6, 3]. For this purpose, Hill's macroscopically uniform eld was taken as a polynomial eld ([6]: assigned eld, [26]: projected eld, [3]: inserted eld) which fullls the respective kinematic micro-macro relations ad hoc. Besides a certain arbitrariness in such a denition of a macroscopically uniform eld for micromorphic continua, the problem is that the kinematic micro-macro relations for such a generalized homogenization procedure can not completely be transformed to surface integrals. Thus, for performing the homogenization, the split into a macroscopically uniform eld and a uctuation eld does not lead to a simplication since volume average kinematic micro-macro relations then still need to be fullled, now with respect to the uctuation eld. In the procedure proposed in [3], the macroscopically uniform (inserted) eld actually serves to dene indirectly the macroscopic stress measures. In contrast, in the present contribution the macroscopic stress measures are dened directly and explicitly and corresponding work-conjugate deformation measures are derived. The boundary value problem at the microscale is completely dened either using static or kinematic boundary conditions, both including the classical and non-classical terms. Although the non-classical terms will in general prevent the total displacement eld from being periodic, it should be possible to modify the proposed kinematic boundary conditions as in [28] to include uctuations at the boundary of ∆V only in such a way that they reduce to classical periodic boundary conditions in the case the non-classical terms are absent. Generalized continuum models were often used phenomenologically to capture the experimentally observed size eects of porous media, see e. g. [34, 2]. In this context it has to be remarked that with the above-mentioned technique, the respective Lagrange multipliers (corresponding to the stress measures work-conjugate to the enforced macroscopic deformations) act like volume forces at the microscale. Thus, the boundary value problem at the microscale will only have a solution if the material can carry these volume forces. This issue is related to the fact that kinematic micro-macro links in form of volume averages require a uniquely dened displacement eld in the complete volume element ∆V (in classical homogenization with static boundary conditions, the same issue applies to material at the boundary). Thus, if porous media shall be homogenized to a micromorphic continuum, the volume element ∆V can encompass the matrix material only. This is plausible insofar as if the corresponding boundary value problem for the microstructure should be solved without homogenization (i. e. with discretely resolved microstructure in the complete domain), then the underlying fundamental equations of the Cauchy continuum (section 2.2) would as well apply only to the matrix material of such a porous medium. Furthermore, it is to be recalled that within the present theory, the hyperstress . its work-conjugate deformation measure K, e M, f and consequently are symmetric with respect to its rst and last index, eqs. (39) and (54), although this symmetry is not required per se in the macroscopic theories of Eringen and Mindlin. However, it should be mentioned that the same symmetry appears in the homogenization theories of Gologanu et al. [20], Kouznetsova et al. [28], Li [29] for strain gradient media which also employ quadratic relations between microscopic tractions and hyperstress (39) or for the gradient term in the kinematic boundary conditions (63b) so that the antimetric parts do not contribute. Within the present framework for the transition to the macroscopic continuum, as outlined in section 2.1, the symmetry is a consequence of the denition of the surface operator investigations whether alternative denitions of h◦iS h◦iS . It remains open for future can be found which do not induce this symmetry but which are nevertheless compatible with the classical theory of homogenization (section 3.2.1). 17 7. Summary and outlook In the present contribution, a consistent theory of homogenization of a classical Cauchy continuum at the microscale towards a micromorphic continuum at the macroscale was presented. Starting point was the average eld theory of Eringen whose up to then unspecied surface operator, which denes uxes like the macroscopic stress, was dened in a consistent way with classical homogenization. Thus having explicit denitions of all macroscopic stress-type quantities of the micromorphic theory, the (strong form of the) boundary value problem of classical homogenization, i. e. equilibrium condition and static boundary conditions, was modied by additional linear terms to yield non-vanishing values of the generalized macroscopic stresses. An equivalent variational formulation was derived in a standard way 2 (being, by the way, well-suited for FE techniques) from which the kinematic micro-macro relations for the macroscopic deformation measures could be derived in explicit form as well. Kinematic boundary conditions with additional quadratic terms could be identied which satisfy a part of the kinematic micro-macro relations ad hoc. Those macroscopic kinematic quantities which are not uniquely determined by the boundary conditions, i. e. the microdeformation and the macroscopic displacement and, if static boundary conditions are applied at the microscale, the classical strain and the microdeformation gradient, are determined by integral constraints over the microscopic domain. Subsequently, the special case of a second gradient theory was addressed before the procedure was demonstrated for the uniaxial case. The results of this simplest possible theory of continuum mechanics are taken as illustrative starting point for discussing certain contentious issues of homogenization for generalized continua like the role of boundary conditions and integral constraints at the microscale, the behavior for microscopically homogeneous material or the handling of pores. In general, homogenization techniques can be applied only if there is a scopic and microscopic length scales. separation of relevant macro- For classical homogenization, it is known that these scales have to be separated by about one order of magnitude. The homogenization towards generalized continua aims in reducing the necessary level of separation between these scales. Thus, it remains open to demonstrate the developed micromorphic homogenization technique on more elaborate problems in order to quantify the improvements compared to classical theory of homogenization. The results need also to be compared to existing strain gradient homogenization techniques from literature in order to identify those techniques which are suited best for certain purposes. In addition, in the future the present micromorphic homogenization needs to be extended towards large deformations. Acknowledgments Fruitful discussions with Jörg Brummund are gratefully acknowledged. References [1] Adomeit, G., 1968. Determination of elastic constants of a structured material. In: Kröner, E. (Ed.), Mechanics of Generalized Continua. Springer-Verlag, Berlin/Heidelberg, pp. 8082. [2] Anderson, W. B., Lakes, R. S., 1994. Size eects due to Cosserat elasticity and surface damage in closed-cell polymethacrylimide foam. J. Mater. Sci. 29 (24), 64136419. [3] Blanco, P. J., Sánchez, P. J., de Souza Neto, E. A., Feijóo, R. A., 2016. Variational foundations and generalized unied theory of RVE-based multiscale models. Arch. Comput. Method. E. 23 (2), 191253. [4] Bouyge, F., Jasiuk, I., Ostoja-Starzewski, M., 2001. A micromechanically based couple-stress model of an elastic two-phase composite. Int. J. Solids Struct. 38 (1013), 17211735. 18 [5] Coleman, B. D., Noll, W., 1963. The thermodynamics of elastic materials with heat conduction and viscosity. Arch. Ration. Mech. An. 13 (1), 167178. [6] De Bellis, M. L., Addessi, D., 2011. A Cosserat based multi-scale model for masonry structures. Int. J. Multiscale Com. 9 (5), 543563. [7] Dimitrijevic, B. J., Hackl, K., 2008. A method for gradient enhancement of continuum damage models. Tech. Mech. 28 (1), 4352. [8] Eringen, A. C., 1968. Mechanics of micromorphic continua. In: Mechanics of Generalized Continua: Proceedings of the IUTAM-Symposium on The Generalized Cosserat Continuum and the Continuum Theory of Dislocations with Applications, Freudenstadt and Stuttgart (Germany) 1967. Springer, Berlin, Heidelberg, pp. 1835. [9] Eringen, A. C., Kafadar, C. B., 1976. Polar and Nonlocal Field Theories. Vol. IV of Continuum Physics. Academic Press, New York. [10] Eringen, A. C., Suhubi, E. S., 1964. Nonlinear theory of simple micro-elastic solidsi. Int. J. Eng. Sci. 2 (2), 189203. [11] Forest, S., 1998. Mechanics of generalized continua: construction by homogenizaton. J. Phys. IV 08 (PR4), 3948. [12] Forest, S., 1999. Aufbau und Identikation von Stogleichungen für höhere Kontinua mittels Homogenisierungsmethoden. Tech. Mech. 19 (4), 297306. [13] Forest, S., 2002. Homogenization methods and the mechanics of generalized continua - part 2. Theor. Appl. Mech. 28-29, 113143. [14] Forest, S., 2009. Micromorphic approach for gradient elasticity, viscoplasticity, and damage. J. Eng. Mech. 135 (3), 117131. [15] Forest, S., 2016. Nonlinear regularization operators as derived from the micromorphic approach to gradient elasticity, viscoplasticity and damage. P. Roy. Soc. Lond. A. Math. 472 (2188), 20150755. [16] Forest, S., Sab, K., 1998. Cosserat overall modeling of heterogeneous materials. Mech. Res. Commun. 25 (4), 449454. [17] Forest, S., Sievert, R., 2006. Nonlinear microstrain theories. Int. J. Solids Struct. 43 (24), 7224 7245. [18] Forest, S., Trinh, D., 2011. Generalized continua and non-homogeneous boundary conditions in homogenisation methods. Z. Angew. Math. Mech. 91 (2), 90109. [19] Germain, P., 1973. The method of virtual power in continuum mechanics. part 2: Microstructure. SIAM J. Appl. Math. 25 (3), 556575. [20] Gologanu, M., Leblond, J. B., Perrin, G., Devaux, J., 1997. Recent extensions of Gurson's model for porous ductile metals part II: A Gurson-like model including the eect of strong gradients of the macroscopic eld. In: Suquet, P. (Ed.), Continuum micromechanics. No. 377 in CISM Courses And Lectures. Springer-Verlag, New York, pp. 97130. [21] Hill, R., 1963. Elastic properties of reinforced solids: Some theoretical principles. J. Mech. Phys. Solids 11 (5), 357372. [22] Hill, R., 1972. On constitutive macro-variables for heterogeneous solids at nite strain. P. Roy. Soc. Lond. A. Math. 326 (1565), 131147. 19 [23] Hütter, G., 2016. An extended Coleman-Noll procedure for generalized continuum theories. Contin. Mech. Therm. 28, 19351941. [24] Hütter, G., Linse, T., Mühlich, U., Kuna, M., 2013. Simulation of ductile crack initiation and propagation by means of a non-local GTN-model under small-scale yielding. Int. J. Solids Struct. 50, 662671. [25] Hütter, G., Mühlich, U., Kuna, M., 2015. Micromorphic homogenization of a porous medium: elastic behavior and quasi-brittle damage. Contin. Mech. Therm. 27, 10591072. [26] Jänicke, R., Diebels, S., Sehlhorst, H.-G., Düster, A., 2009. Two-scale modelling of micromorphic continua. Contin. Mech. Therm. 21 (4), 297315. [27] Jänicke, R., Steeb, H., 2012. Minimal loading conditions for higher order numerical homogenisation schemes. Arch. Appl. Mech. 82 (8), 10751088. [28] Kouznetsova, V., Geers, M. G. D., Brekelmans, W. A. M., 2002. Multi-scale constitutive modelling of heterogeneous materials with a gradient-enhanced computational homogenization scheme. Int. J. Numer. Meth. Engng. 54 (8), 12351260. [29] Li, J., 2011. A micromechanics-based strain gradient damage model for fracture prediction of brittle materials part I: Homogenization methodology and constitutive relations. Int. J. Solids Struct. 48 (24), 33363345. [30] Maugin, G. A., 2010. Generalized continuum mechanics: What do we mean by that? In: Maugin, G. A., Metrikine, A. V. (Eds.), Mechanics of Generalized Continua. Vol. 21 of Advances in Mechanics and Mathematics. Springer, New York, pp. 313. [31] Mindlin, R. D., 1964. Micro-structure in linear elasticity. Arch. Ration. Mech. An. 16 (1), 5178. [32] Mühlich, U., Zybell, L., Hütter, G., Kuna, M., 2013. A rst-order strain gradient damage model for simulating quasi-brittle failure in porous elastic solids. Arch. Appl. Mech. 83, 955967. [33] Mühlich, U., Zybell, L., Kuna, M., 2012. Estimation of material properties for linear elastic strain gradient eective media. Eur. J. Mech. A-Solid. 31 (1), 117130. [34] Wheel, M. A., Frame, J. C., Riches, P. E., 2015. Is smaller always stier? supposedly generalised continua. Int. J. Solids Struct. 67-68, 8492. 20 On size eects in A. Appendix A.1. Sum approximation of integrals According to [8, 9], the domain of subdomains ∆V (X) Z x∈Ω I ∇x · ψ a dV = Ω with its boundary Z is divided into a large but nite number l m n · ψ a dS n · ψ a dS + S || ∂Ω = ∂Ω so that the ux term of a general global balance law (1) becomes X K Z l m n · ψ a dS Z n · ψ a dS + ∆AK with ∆AK := ∂Ω ∩ ∆VK S || ∩∆VK Z Z l m X 1 1 n · ψ a dS ∆AK + n · ψ a dS ∆AK ∆AK ∆AK K K ∆AK S || ∩∆VK | {z } | {z } l m :=N·hψ i :=N· hψ i a S a S I Z I D E lD E m D E N · ψa dS + N · ψ a dS = ∇X · ψ a dV ≈ = X S ∂ΩX Therein, S || S S (83) ΩX S || refers to a potential surface of discontinuity with a jump d◦e of the eld quantities. An analogous procedure for the source terms yields Z ρϕm dV = Ω X Z ρϕm dV L ∆V L = X L Z Z 1 hρϕm iV dV ρϕm dV ∆VL ≈ ∆VL ∆VL ΩX {z } | (84) :=hρϕm iV The macroscopic balances (3) are thus obtained by inserting (83) and (84) into the global balance (1). A.2. Asymptotic self-consistency of homogenization procedures It is the nature of any homogenization that it yields only an approximate solution to the initial problem (1). The approximation character of (83) can be shown by comparing the volume average of the local balance (2) D E . hρϕm iV = ∇x · ψ a + hρψm iV (85) V with its macroscopic counterpart (4). For both to be equal would require that the residual of homogenization D E D E RΨ = ∇x · ψ a − ∇X · ψ a V (86) S as dierence between (4) and (85) would vanish (Further residuals may occur at the boundaries which are not considered here). In general, any homogenization procedure should be self-consistent in the sense that it reduces to the theory plugged in at the microscopic level in the limit of the volume element ∆V innitesimally small. Regarding the This requirement could be termed present approach, this requires that RΨ asymptotic self-consistency. becomes exactly zero for 21 ∆V → 0. becoming It is found that the classical Hill-Mandel homogenization with D E D E ψa := ψ a easily with a Taylor expansion of ψ a ). S is asymptotically self-consistent (as can be veried V Using the Gauss theorem, the denition (38) of the surface operator of the present contribution can be written (as in (40)) as D E D E D E ψa = ψa + ξ∇x · ψ a S V V . A Taylor expansion shows that this denition is asymptotically self-consistent as well. 22 (87)