Abstract
The computational analysis of fiber network fracture is an emerging field with application to paper, rubber-like materials, hydrogels, soft biological tissue, and composites. Fiber networks are often described as probabilistic structures of interacting one-dimensional elements, such as truss-bars and beams. Failure may then be modeled as strong discontinuities in the displacement field that are directly embedded within the structural finite elements. As for other strain-softening materials, the tangent stiffness matrix can be non-positive definite, which diminishes the robustness of the solution of the coupled (monolithic) two-field problem. Its uncoupling, and thus the use of a staggered solution method where the field variables are solved alternatingly, avoids such difficulties and results in a stable, but sub-optimally converging solution method. In the present work, we evaluate the staggered against the monolithic solution approach and assess their computational performance in the analysis of fiber network failure. We then propose a hybrid solution technique that optimizes the performance and robustness of the computational analysis. It represents a matrix regularization technique that retains a positive definite element stiffness matrix while approaching the tangent stiffness matrix of the monolithic problem. Given the problems investigated in this work, the hybrid solution approach is up to 30 times faster than the staggered approach, where its superiority is most pronounced at large loading increments. The approach is general and may also accelerate the computational analysis of other failure problems.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
Fiber network structures define the mechanics of a wide range of materials with applications towards paper, packaging materials, rubber-like materials, hydrogels, soft biological tissues, and composites [1,2,3,4,5,6,7,8]. In addition to the elastic properties of such materials, their failure properties determine industrial applicability. The mechanics of fiber network structures may be described through the interaction of one-dimensional elements, such as truss-bars and beams [2, 9,10,11,12,13,14,15], where especially the numerical description of material failure is challenging.
The theory of strong discontinuities (e.g., see [16,17,18,19,20,21,22,23,24,25,26,27,28,29,30]) has made a significant impact on the computational analysis of fracture and extended the functionality of many commercial finite element method (FEM) packages. Strong discontinuities enrich the displacement solution space with jumps across the fracture surfaces. The embedded discontinuity finite element method (ED-FEM or E-FEM) and the extended finite element method (X-FEM) have gained the most popularity (e.g., see [31, 32] for comparative studies). The E-FEM and the X-FEM represent strong discontinuities via elemental and nodal enrichments of the displacement solution space, respectively. It allows the approximation of fracture problems using coarse FEM meshes and requires no re-meshing. Most importantly, it avoids pathological mesh-dependency as known from the solution of strain softening materials within the classical non-polar continuum. Whilst the description of strain softening in non-polar continua results in an infinitesimally small (vanishing) localization volume [33,34,35], the failure in the discontinuity-based description is represented by an area;—the cross-section of the structural element in our applications. Since such an area remains finite, pathological mesh-dependency is not an issue in the discontinuity-based description, and it converges towards mesh-independent results. However, said approaches often require a crack-tracking algorithm (except for recent work [36] on E-FEM without crack tracking) and exhibit difficulties capturing crack branching and coalescence of multiple cracks, difficulties that are avoided with the newly emerged phase-field fracture method (e.g., see [37, 38]). Unlike discrete crack approaches, it is a smeared crack approach like damage mechanics [39,40,41,42] and therefore requires a localization limiter or a characteristic length scale parameter. Such measure is difficult to determine in practice [43] and can be difficult to implement upon unstructured meshes [44]. In addition, a sufficiently refined mesh is necessary to adequately describe the mechanics of the localization zone, leading to the high computational cost of smeared crack approaches. Thus, it prevents from an efficient analysis of failure in fiber networks where the fracture process zone is to be resolved in the individual fibers.
The tangent stiffness matrix of the two-field fracture problem can be indefinite or negative definite [45,46,47,48]. The energy potential to be minimized is therefore non-convex. Uncoupling the problem and using a staggered solution approach that solves for the displacement and the respective state of damage alternatingly, overcomes this issue and results in a convex alternating minimization problem. It is the preferred approach in phase-field modeling of fracture [49] and available in commercial software [50,51,52]. It results in a positive definite tangent stiffness matrix, but does not represent the linearized residuum, and the solution then converges much slower as compared to Newton–Raphson iterations. The efficiency of the consistently linearized coupled (monolithic) problem is therefore sacrificed towards the robustness of the staggered solution technique [53].
The E-FEM has received significant attention in the description of failure in truss-bars, beams, and structures thereof [54,55,56,57,58,59,60,61,62,63,64,65,66,67,68]. No crack-tracking algorithm is then needed [69]. The E-FEM framework allows static condensation of the additional DOF associated with the fracture kinematics [55] directly at the element level, resulting in an operator-splitting method [56] for the evaluation of these DOF. The implementation is based on the Hu-Washizu two-field variational principle, and given the discontinuity is positioned in the middle of the structural element, the orthogonality condition is a priori enforced [70, 71]. As with the phase-field method, a staggered solution method may be used, and the continuous and discontinuous fields are minimized alternatingly [61]. In the present work, we expand these ideas and propose a hybrid method to optimize the performance and robustness of the E-FEM models in the analysis of fiber network failure [61, 72, 73]. The hybrid solution approach is general and may also be applied to other fracture mechanics problems.
2 Enhanced finite element formulation
We consider a local Cartesian coordinate system \(\left\{{{\varvec{e}}}_{x},{{\varvec{e}}}_{y},{{\varvec{e}}}_{z}\right\}\) in the description of a 3D Timoshenko beam of the length \(L\) and the cross-section \(A\). The beam’s neutral axis \(x\in \left[0, L\right]\) is aligned with the \(x\)-coordinate along which distributed \({\varvec{f}}\) as well as concentrated \({\varvec{F}}\) loads are applied. The displacements \({\varvec{u}}\left(x\right)\) and the rotations \({\varvec{\theta}}\left(x\right)\) are collectively represented by the generalized displacement vector \(\overline{{\varvec{u}} }={\left[\begin{array}{cccccc}{u}_{x}& {u}_{y}& {u}_{z}& {\theta }_{x}& {\theta }_{y}& {\theta }_{z}\end{array}\right]}^{T}\) with the subscripts \(x\), \(y\) and \(z\) denoting the respective displacement and rotation components. The generalized strain measures of the 3D Timoshenko beam then read
where \(\varepsilon \) is the axial strain, \({\gamma }_{y}, {\gamma }_{z}\) denote the shear strains along the \(y\) and \(z\) directions, \({\kappa }_{x}\) is the change in the angle of twist \({\theta }_{x}\) around the beam’s neutral axis and \({\kappa }_{y}\), \({\kappa }_{z}\) represent bending curvatures. The stress resultants \({\varvec{\sigma}}={\left[\begin{array}{cccccc}N& {Q}_{y}& {Q}_{z}& {M}_{x}& {M}_{y}& {M}_{z}\end{array}\right]}^{T}\) are conjugate to the generalized strain measures \(\overline{{\varvec{\epsilon}} }={\left[\begin{array}{cccccc}\varepsilon & {\gamma }_{y}& {\gamma }_{z}& {\kappa }_{x}& {\kappa }_{y}& {\kappa }_{z}\end{array}\right]}^{T}\). Here, the axial force \(N\), the shear forces \({Q}_{y}, {Q}_{z}\) along the \(y\) and \(z\) directions, the torsional moment \({M}_{x}\), and the bending moments \({M}_{y},{M}_{z}\) along the \(y\) and \(z\) directions have been introduced.
The strong form of the local equilibrium equations [74] reads
where \({q}_{x},{q}_{y}, {q}_{z}\) and \({m}_{x}, {m}_{y}, {m}_{z}\) denote the components of the distributed forces and moments per unit length, respectively.
For simplicity, we consider a two-node beam element of the length \({l}_{e}\) with the linear shape functions \({N}_{1}=1-x/{l}_{e}\) and \({N}_{2}=x/{l}_{e}\), and a single Gauss point in the center is used to integrate the FEM equations. The interpolated generalized displacement field then takes the form
where \({\overline{{\varvec{u}}} }_{1}\) and \({\overline{{\varvec{u}}} }_{2}\) are the six-dimensional generalized nodal displacement vectors. In addition, a softening hinge allows for the formation of localized failure in the center of the Timoshenko beam. We therefore enrich (3) with a discrete displacement/rotation jump \({\varvec{\xi}}={\left[\begin{array}{cccccc}{\xi }_{{u}_{x}}& {\xi }_{{u}_{y}}& {\xi }_{{u}_{z}}& {\xi }_{{\theta }_{x}}& {\xi }_{{\theta }_{y}}& {\xi }_{{\theta }_{z}}\end{array}\right]}^{T}\) situated in the center \({x}_{c}\) of the element. Thus,
represents the total displacements, where
denotes the Heaviside step function centered at the middle of the element \({x}_{c}\). With the boundary conditions \({{\varvec{u}}}_{1}={\overline{{\varvec{u}}} }_{1}\) and \({{\varvec{u}}}_{2}={\overline{{\varvec{u}}} }_{2}+{\varvec{\xi}}\), where \({{\varvec{u}}}_{1}\) and \({{\varvec{u}}}_{2}\) are the total nodal displacement vectors, the displacement field (4) reads
where \({\varvec{d}}={\left[\begin{array}{cc}{{\varvec{u}}}_{1}& {{\varvec{u}}}_{2}\end{array}\right]}^{T}\) represents the element’s nodal displacement vector, and
denotes the corresponding interpolation matrix. The strain field then takes the form
and the matrix
interpolates the jump, where \(G=-d{N}_{2}/dx=-1/{l}_{e}\) and \({\varvec{I}}\) denotes the \(6\times 6\) identity matrix. Here, \({{\varvec{\epsilon}}}_{bulk}\) represents the strain in the bulk material at \(x\ne {x}_{c}\). In addition, \({\delta }_{{x}_{c}}=d{H}_{{x}_{c}}/dx\) is the Dirac delta function centered at \({x}_{c}\), and
represents the standard \(6\times 12\) strain–displacement interpolation matrix, where \({B}_{1}=d{N}_{1}/dx=-1/{l}_{e}\) and \({B}_{2}=d{N}_{2}/dx=1/{l}_{e}\).
Given the admissible variation \(\delta {\varvec{d}}\) of the element nodal displacements, \(\delta {\varvec{u}}\) of the displacement field and \(\delta{\varvec{\xi}}\) of the corresponding jump, the internal and respective external virtual work of the beam formulation reads
where \({\varvec{f}}={\left[\begin{array}{cccccc}{q}_{x}& {q}_{y}& {q}_{z}& {m}_{x}& {m}_{y}& {m}_{z}\end{array}\right]}^{T}\) represents a distributed load along the beam and \({\varvec{F}}\) a concentrated load applied at the position \({x}_{F}\) along the beam. In the derivation of \({\left(11\right)}_{2}\), we used the identity \({\int }_{{l}_{e}}\left({H}_{{x}_{c}}-{N}_{2}\right)dx=0\), a consequence of having the discontinuity in the middle of the beam, \({x}_{c}={l}_{e}/2\). In addition, the second term of \({\left(11\right)}_{2}\) vanishes if \({x}_{F}=0\) or \({x}_{F}={l}_{\mathrm{e}}\), which we apply here.
From the principle of virtual work, \(\delta {w}_{int}-\delta {w}_{ext}=0\), and the independence of the admissible variations \(\delta {\varvec{d}}\), \(\delta{\varvec{\xi}}\), we obtain the two variational statements [55, 56, 75]
where the condition \({\varvec{t}}={\int }_{{l}_{e}}{\delta }_{{x}_{c}}{\varvec{\sigma}}dx\) has been used. The condition \({\left(12\right)}_{2}\) represents the local residual at the discontinuity, \(x={x}_{c}\), and ensures the equilibrium between the vector of stress resultants \({\varvec{t}}\) at the discontinuity and the stress resultant \({\varvec{\sigma}}\) in the bulk material, \(-{\varvec{\sigma}}+{\varvec{t}}={\mathbf 0}\). To obtain the equilibrium condition \({\varvec{\sigma}}={\varvec{t}}\), we utilized the jump interpolation operator \({\varvec{G}}\) (9) and the fundamental property of the Dirac delta function \({\int }_{-\infty }^{\infty }f\left(x\right)\delta \left(x-{x}_{c}\right)dx=f\left({x}_{c}\right)\).
The implementation of (12) results in the set
of non-linear equations at the element level, where \({{\varvec{f}}}^{e\; int}={\int }_{{l}_{e}}{{\varvec{B}}}^{T}{\varvec{\sigma}}dx\) and \({{\varvec{f}}}^{e \; ext}={\int }_{{l}_{e}}{{\varvec{N}}}^{T}{\varvec{f}}dx+{{\varvec{N}}}^{T}\left({x}_{F}\right){\varvec{F}}\) are the internal and external element nodal force vectors, respectively.
Incremental formulations are used to express the constitutive relations of the bulk material and the fracture process zone, expressions that close the set of Eq. (13). We use
to describe the development of the stress resultants \({\varvec{\sigma}}\) and \({\varvec{t}}\), where \({\varvec{C}}\) and \({\varvec{H}}\) denote the respective tangent constitutive tensors. Given a beam of the cross-section \(A\) that is made of a linear-elastic material with Young’s modulus \(E\) and shear modulus \(\mu \), the tangent
determines the development of \({\varvec{\sigma}}\), where \(k\) is the shear correction factor, \(J\) is the polar moment of inertia, whilst \({I}_{11}\) and \({I}_{22}\) denote the area moments of inertia. Note that (15) is the simplest set of uncoupled linear elastic constitutive equations and is based on the assumption that the beam’s cross-section possesses appropriate symmetries [76].
With the strain \({{\varvec{\epsilon}}}_{bulk}\) in the bulk material (8), the increment of the stress resultant
and incremental local equilibrium across the discontinuity \(-\Delta{\varvec{\sigma}}+\Delta {\varvec{t}}={\bf 0}\) are to be enforced.
The linearization of the residual force equations (13) with respect to the unknown displacement \(\Delta {\varvec{d}}\) and the discontinuous displacement \(\Delta{\varvec{\xi}}\) yields the system
with the sub-matrices
The term \({{\varvec{C}}}^{*}{\varvec{B}}\) is explained in Sect. 4.
In the implementation of these equations, we distinguish between elastic and failure loading. We follow the framework of inelasticity [45, 77] and the decision is based on the introduction of a failure surface \(\Phi \), see Sect. 3. Given elastic loading, or unloading, the increment \(\Delta{\varvec{\xi}}={\mathbf {0}}\) and the system reduces to the standard Timoshenko beam FEM model \({{\varvec{K}}}_{dd}^{e}\Delta {\varvec{d}}=\boldsymbol{ }{{\varvec{f}}}^{e\; int}-{{\varvec{f}}}^{e\; ext}\). Different implementations concerning damage loading are discussed in the forthcoming sections.
3 Monolithic, staggered and hybrid FEM implementation
The monolithic implementation considers the consistently derived finite element stiffness as shown in (17), where off-diagonal terms couple the increment of the displacement \(\Delta {\varvec{d}}\) and the increment of the jump \(\Delta{\varvec{\xi}}\). As with other embedded approaches, our model allows for the static condensation of \(\Delta{\varvec{\xi}}\) directly at the element level, which then results in a displacement-based FEM implementation. The second equation in (17), the internal equilibrium, is then used to substitute \(\Delta{\varvec{\xi}}\) through
In the limit \({l}_{e}\to 0\), \(G\to {\infty }\) and \({{\varvec{K}}}_{\xi \xi }^{e} \left(18\right)\) therefore remains positive definite and invertible [21, 55]. In general, however, the condition (19) poses a limit on the size \({l}_{e}\) of the finite element in the analysis of strain-softening materials. The substitution of \(\Delta{\varvec{\xi}}\) by (19) in (17) results in the system of equations
at the element level, where
denotes the element stiffness. We emerged at a displacement-based model that may now be implemented through the standard user element interface of FEM packages. Whilst the aforementioned solution uses the consistent linearization of the residual forces, it results in a non-positive definite finite element stiffness \({{\varvec{K}}}_{mono}^{e}\), which then materializes through poor robustness of the monolithic implementation.
Towards reinforcing the robustness of the model, we may uncouple \({\varvec{d}},{\varvec{\xi}}\) and solve the problem in a staggered way. The internal equilibrium equation \({\left(13\right)}_{2}\) is then solved at the nodal displacement \({{\varvec{d}}}_{n}\) from the previous solution. It explicitly reads
and the embedded formulation again allows to solve it directly at the element level. The system
of FEM equations may then be assembled, and the solution of the global system yields the nodal displacements. Here,
determines the corresponding finite element stiffness, and given it represents an inconsistently linearized residuum, it results in poor convergence of the staggered approach.
Towards optimizing performance and robustness of the finite element model, we propose the hybrid definition
of the finite element stiffness, where \(\beta \) is a numerical parameter, chosen to ensure a positive definite finite element stiffness matrix \({{\varvec{K}}}_{hyb}^{e}\). It is set according to
where the condition for \({\beta }_{critical}\Rightarrow \mathrm{det}{{\varvec{K}}}_{hyb}^{e}=0\) is derived as follows. Excluding the six rigid body motion-related DOFs from the system, the effective stiffness matrices are of the dimension \(6\times 6\). The only physical root of \(\mathrm{det}{{\varvec{K}}}_{hyb}^{e}=0\) results in an expression for \({\beta }_{critical}\), and (26) then guarantees a positive definite finite element stiffness matrix. Although our stiffness matrices are sparsely populated, the direct solution of \(\mathrm{det}{{\varvec{K}}}_{hyb}^{e}=0\) requires the eigenvalue analysis of one \(12\times 12\) matrix at every Gauss point for each solution step; a faster and tailored implementation for beam rupture is discussed in Sect. 5.
4 Predictor–corrector implementation of beam rupture
Aiming at modeling the failure of soft fibers, we limit ourselves to the description of failure under tension. Therefore, only the component \({\xi }_{{u}_{x}}=\xi \) of the jump \({\varvec{\xi}}\) is allowed to evolve, whilst \({\xi }_{{u}_{y}}={\xi }_{{u}_{z}}={\xi }_{{\theta }_{x}}={\xi }_{{\theta }_{y}}={\xi }_{{\theta }_{z}}=0\). The development of the stress resultant according to \({\left(14\right)}_{2}\) is then determined by the tangent constitutive tensor component \({H}_{11}=H<0\), whilst all the other components of \({\varvec{H}}\) are 0. As a result, C* (18) takes the form
because we only allow failure in tension. In (16), \({{\varvec{C}}}^{{*}}{\varvec{B}}\) is the tangent when \(\Delta {\xi }_{{u}_{y}}=\Delta {\xi }_{{u}_{z}}=\Delta {\xi }_{{\theta }_{x}}=\Delta {\xi }_{{\theta }_{y}}=\Delta {\xi }_{{\theta }_{z}}=0\).
The failure criterion [56]
in the stress resultant space, determines the beam’s loading condition. Here, \({N}_{y}\) is the elastic limit resultant (force), whilst \(H<0\) and \(\alpha >0\) denote the softening modulus and an internal softening variable, respectively. Elastic deformation of the beam is then characterized by \(\Phi <0\), and a loading state that reaches the failure surface, \(\Phi =0,\) results in the accumulation of failure. At complete rupture, \(N=0\), the linear softening law (28) yields \({\alpha }_{max}={N}_{y}/|H|\) and determines the fracture energy of \({G}_{f}={\alpha }_{max}\left|H\right|{\alpha }_{max}/2={N}_{y}^{2}/(2\left|H\right|)\).
Towards closing the failure description, the evolution of the internal softening variable is to be linked to the evolution of the jump displacement, and
specifies said correspondence at the cases of elastic and failure loading, respectively.
With (16), the internal equilibrium \(-{\varvec{\sigma}}+{\varvec{t}}={\mathbf 0}\) for the only non-trivial loading mode reads \(-N+t= 0\), resulting in
Here, \({\epsilon }_{bulk}=\varepsilon +G\xi \) denotes the strain in the bulk material, and \(\xi \) is the jump displacement with the initial condition \(\xi =0\) at \(N={N}_{y}\). Given \({\xi }_{n}\) and \({\alpha }_{n}\) from the previous time point, we can iteratively derive \(\xi \) and \(\alpha \) at the current time point. We therefore expand (30) towards
where \({N}^{corr}\) eventually corrects the trial stress resultant \({N}^{trial}=EA\left(\varepsilon +G{\xi }_{n}\right)\). The implementation follows the classical concept of computational inelasticity [77], and once \(\Delta \gamma \), the consistency parameter that ensures (28), is given, \(\xi ={\xi }_{n}+\Delta \gamma \mathrm{sign}\left(N\right)\) updates the solution.
Alternatively, Eq. (31) may be expressed as \(\left(\left|N\right|-EAG\Delta \gamma \right)\mathrm{ sign}\left(N\right)=\left|{N}^{trial}\right|\mathrm{ sign}\left({N}^{trial}\right)\), and with \(\Delta \gamma >0,\) \(G<0\) and \(EA>0\), the term in the bracket to the left is always positive. Therefore, \(\mathrm{sign}\left(N\right)=\mathrm{sign}\left({N}^{trial}\right)\) and \(\left|N\right|=\left|{N}^{trial}\right|+EAG\Delta \gamma \) follows.
With (31), the failure surface (28) then reads
where \({\Phi }^{trial}={N}^{trial}-\left({N}_{y}+H{\alpha }_{n}\right)\) describes the failure surface assuming an elastic load increment. At failure loading, \(\Phi =0\), the algorithmic consistency parameter therefore reads
and allows us to update the solution. The algorithmic consistency parameter \(\Delta \gamma \) must be positive, a standard stability condition to be enforced to avoid snap-back behavior. Table 1 summarizes the predictor–corrector implementation.
5 Implementation
The description of beam rupture through failure under tension and the uncoupled constitutive model (15) result in an uncoupling of the condition (25). Consequently, the identification of the stability parameter (26) yields the single scalar equation
and avoids then the eigenvalue analysis of the stiffness matrices \({{\varvec{K}}}_{mono}^{e}\) and \({{\varvec{K}}}_{stagg}^{e}\). Here, \({K}_{min}={h}_{tol}EA/{l}_{e}>0\) determines a minimum stiffness, where the numerical tolerance level \({h}_{tol}\) is determined by the precision level of the hardware/software realization, whilst \({l}_{e}\) is the element length. To avoid ill-conditioning towards the development of complete fiber rupture, the corresponding elements in the global stiffness matrix are limited to be larger than the minimum stiffness \({K}_{min}\), where again said stiffness threshold has been used.
Given the finite element stiffness matrix (25) for the cases of a monolithic \((\beta =1)\), staggered \((\beta =0)\), or hybrid (\(0<\beta <1\)) implementation, we may now rotate the DOF vector \({\varvec{d}}\), the nodal force vectors \({{\varvec{f}}}^{e\; int}\), and the stiffens \({{\varvec{K}}}^{e}\) into the global Cartesian coordinate system \(\left\{{{\varvec{e}}}_{x}^{^{\prime}},{{\varvec{e}}}_{y}^{^{\prime}},{{\varvec{e}}}_{z}^{^{\prime}}\right\}\)[78]
Here,
denotes the corresponding transformation matrix, where \({\Lambda }_{ij}={{{\varvec{e}}}_{i}}^{T}{{\varvec{e}}}_{j}^{\boldsymbol{^{\prime}}};i,j=x,y,z\) are the directional cosines between the global \(\left\{{{\varvec{e}}}_{x}^{^{\prime}},{{\varvec{e}}}_{y}^{^{\prime}},{{\varvec{e}}}_{z}^{^{\prime}}\right\}\) and the local \(\left\{{{\varvec{e}}}_{x},{{\varvec{e}}}_{y},{{\varvec{e}}}_{z}\right\}\) systems, respectively. Table 2 summarizes the implementation.
6 Benchmark exercises
6.1 Mesh independence
To illustrate the behavior of the finite element, we consider a \(0.1 \; \mathrm{ mm}\) long cantilever beam that is loaded in tension. Young’s modulus \(E=1.0\; \mathrm{MPa}\), the cross-section area \(A=1.0 \; {\mathrm{mm}}^{2}\), and the ultimate tensile force \({N}_{y}=1.0 \; \mathrm{N}\) further describe the problem. The cantilever beam is discretized with one and ten evenly spaced finite elements, where the ultimate tensile force is lowered by 1% in the most left element. Otherwise, the homogenous stress field of the problem would have resulted in an ambiguous failure pattern, see Sect. 3 elsewhere [61]. Figure 1 presents the force–displacement response of our structural problem for three different fracture energies \({G}_{f}\), a property also expressed as the area under the curve (see Fig. 1). As expected from the fracture model, the force–displacement response is independent of the finite element discretization; it is identical between the one-element and ten-element discretization, respectively.
6.2 Tensile test of a fibrous tissue specimen
A planar and random network of interconnected 3D beams describes the mechanics of a fiber network of the densities \({\rho }_{s}=\left\{300;500;1000\right\}\) \(\mathrm{kg}/{\mathrm{m}}^{3}\) and with the fiber properties listed in Table 3. In-plane it covers the area \(18 \times 6 \; {\mathrm{mm}}^{2}\) and the displacement \({\delta }_{0}=9.0\; \mathrm{mm}\) is prescribed along one edge, while all six DOFs at the opposite edge are fixed. Out-of-plane displacements and rotations are prevented, and all information on how the mesh has been generated is reported elsewhere [12] together with the ANSYS input file [61]. A quasi-static failure analysis was computed through the incremental application of the prescribed displacement \({\delta }_{0}\).
Figure 2 shows the force-displacement response of the fiber networks. Computational results refer to the staggered solutions. As expected, we observe a more dissipative response of the fiber networks with the higher fracture energy of the fibers. As an example, Fig. 3 shows the evolution of the jump \(\xi \) for the densest network of fibers with the fracture energy \({G}_{f}=0.1 \; \mathrm{Nmm}\). The fiber network configurations refer to the points A, B and C in Fig. 2. Given said fracture energy, the linear softening law (28) provides the jump \({\xi }_{max}=0.85\; \mathrm{mm}\) at the state of complete rupture.
6.2.1 Numerical stability
The numerical stability of the \(1000 \, \mathrm{kg}/{\mathrm{m}}^{3}\) dense fibrous network made of fibers with the fracture energy of 0.2 \(\mathrm{Nmm}\) was investigated with respect to the size of the displacement increment. The prescribed displacement was applied through 500, 2000, 4000, and 8000 steps. In each sub-step a maximum of 500 equilibrium iterations were allowed (NEQIT = 500 in ANSYS), before the next step was processed. Whilst the staggered solution converged for all step sizes, the monolithic failed for some step sizes. Figure 4 reports the results from the stability analysis and indicates the points of failure of the monolithic solution. Failure is most likely linked to the inability to minimize the related non-convex problem, and a further investigation of ANSYS' internal algorithm was not feasible. Given the monolithic method did not fail, the monolithic and staggered approaches solve the same equilibrium equations (13) and therefore result in the same force-displacement response of the fibrous network.
6.3 Tensile test of a notched fibrous tissue specimen
The afore explored specimen geometry (see Sect. 6.2) is modified, and a sharp notch with an opening angle of \(20^\circ \) is introduced in the center of the tensile specimen. A \(1000 \, \mathrm{kg}/{\mathrm{m}}^{3}\) dense fibrous network with the fiber properties listed in Table 3 is considered. Figure 5 shows the force-displacement response of the notched tensile specimen. See Fig. 6 for the visualization of the jump \(\xi \) at the points G–K. Aligned with the previous problem, the monolithic approach is numerically unstable, given the prescribed displacement was applied through 2000 and 10,000 steps, respectively.
7 Performance and stability of the hybrid solution technique
Among all our simulations, the densest fibrous tissue specimens (\({\rho }_{s}=1000 \; \mathrm{kg}/{\mathrm{m}}^{3}\)) were computationally most demanding. These cases will therefore be considered to explore the performance of the proposed hybrid solution technique (25) and benchmarking it against the monolithic and staggered schemes. Towards the optimization of the computation time of large fiber networks with many DOFs, we limit the maximum number of steps to 500 in our benchmarking exercise. Table 4 lists the cumulative iterations, the number of all iterations needed to compute the solution until the prescribed displacement \({\delta }_{0}\) is reached. It corresponds to the displacement at the endpoints in Figs. 2 and 5, and the number of cumulative iterations represents a measure that is sensitive to the computational effort to solve the problem. The data reported in Table 4 is also shown in Fig. 7, where in addition a second-order polynomial fit visualizes the dependence of the cumulative iterations on the number of displacement increments. Whilst at large displacement increments the hybrid solution technique is clearly superior to the staggered approach, no such advantage is seen for small displacement increments. Searching for the solution in close vicinity of the previous solution, even a non-consistent linearization of the residuum, and thus the staggered solution approach, can find the solution within a low number (2–4) of iterations. We use the incremental constitutive implementation discussed in Sect. 4, and the result is therefore influenced by the size of the prescribed displacement increment. Figure 8 shows the resulting force-displacement curves for the hybrid solution technique as a function of the displacement increment and the numerical tolerance \({h}_{tol}\). Even the largest step size leads to results of practical use.
8 Summary and conclusion
We studied the computational analysis of failure in fibrous materials, where the individual fibers are modeled as Timoshenko beams with embedded strong discontinuities. Representative benchmark examples have been used, where the recently proposed staggered solution method [61] has been tested against the monolithic solution strategy. Whilst the staggered approach is numerically robust, it does not use the consistent linearization of the nodal forces and therefore suffers from a poor convergence rate. This is especially the case for large displacement increments. The monolithic approach, in contrary, follows from the consistent linearization but can result in a non-positive definite element stiffness matrix, that then requires the solution of unstable equilibria. It is therefore practically not applicable to solve the benchmark problems studied in this work; it erratically fails, and step-size refinement is not always successful. We therefore proposed a novel hybrid solution method that forms the element stiffness through an adaptive ‘mixing’ of the stiffness of the monolithic and staggered approaches. It may also be seen as a matrix regularization technique to retain a positive definite element stiffness matrix while approaching the tangent stiffness matrix of the monolithic problem. The hybrid method results in a robust and computational efficient solution technique with an up to 30-fold performance gain in the exploration of failure in fibrous materials. The approach is general and may also accelerate the computational analysis of other failure problems.
Change history
08 November 2022
A Correction to this paper has been published: https://doi.org/10.1007/s00466-022-02239-x
References
Picu CR (2021) Constitutive models for random fiber network materials: a review of current status and challenges. Mech Res Commun. https://doi.org/10.1016/j.mechrescom.2020.103605
Picu RC (2011) Mechanics of random fiber networks—a review. Soft Matter. https://doi.org/10.1039/c1sm05022b
Simon JW (2021) A review of recent trends and challenges in computational modeling of paper and paperboard at different scales. Archiv Comput Methods Eng. https://doi.org/10.1007/s11831-020-09460-y
Bai R, Yang J, Suo Z (2019) Fatigue of hydrogels. Eur J Mech A Solids. https://doi.org/10.1016/j.euromechsol.2018.12.001
Eichhorn SJ, Dufresne A, Aranguren M, Marcovich NE, Capadona JR, Rowan SJ, Weder C, Thielemans W, Roman M, Renneckar S, Gindl W, Veigel S, Keckes J, Yano H, Abe K, Nogi M, Nakagaito AN, Mangalam A, Simonsen J, Benight AS, Bismarck A, Berglund LA, Peijs T (2010) Review: current international research into cellulose nanofibres and nanocomposites. J Mater Sci. https://doi.org/10.1007/s10853-009-3874-0
Mansour R, Kulachenko A (2022) Stochastic constitutive model of thin fibre networks. Mech Fibrous Netw. https://doi.org/10.1016/b978-0-12-822207-2.00014-3
Bosco E, Peerlings RHJ, Schoenmakers NPT, Dave N, Geers MGD (2022) Hygro-mechanics of fibrous networks: a comparison between micro-scale modelling approaches. Mech Fibrous Netw. https://doi.org/10.1016/b978-0-12-822207-2.00009-x
Sozumert E, Silberschmidt VV (2022) Numerical models of random fibrous networks. In: Mechanics of fibrous networks. https://doi.org/10.1016/b978-0-12-822207-2.00012-x
Li Y, Yu Z, Reese S, Simon JW (2017) Evaluation of the out-of-plane response of fiber networks with a representative volume element model. Tappi J 1:325–334
Heyden S (2000) Network modelling for the evaluation of mechanical properties of cellulose fluff
Niskanen KJ, Alava MJ (1994) Planar random networks with flexible fibers. Phys Rev Lett. https://doi.org/10.1103/PhysRevLett.73.3475
Kulachenko A, Uesaka T (2012) Direct simulations of fiber network deformation and failure. Mech Mater. https://doi.org/10.1016/j.mechmat.2012.03.010
Berkache K, Deogekar S, Goda I, Picu RC, Ganghoffer JF (2019) Identification of equivalent couple-stress continuum models for planar random fibrous media. Continuum Mech Thermodyn. https://doi.org/10.1007/s00161-018-0710-2
Åslund PE, Isaksson P (2011) A note on the nonlinear mechanical behavior of planar random network structures subjected to in-plane compression. J Compos Mater. https://doi.org/10.1177/0021998311422749
Isaksson P, Dumont PJJ, Rolland du Roscoat S (2021) Statistical analysis of the crack sensitivity of fiber networks. Int J Solids Struct. https://doi.org/10.1016/j.ijsolstr.2020.10.021
Simo JC, Oliver J, Armero F (1993) An analysis of strong discontinuities induced by strain-softening in rate-independent inelastic solids. Comput Mech. https://doi.org/10.1007/BF00372173
Oliver J (1996) Modelling strong discontinuities in solid mechanics via strain softening constitutive equations. Part 2: numerical simulation. Int J Numer Methods Eng. https://doi.org/10.1002/(SICI)1097-0207(19961115)39:21%3c3601::AID-NME64%3e3.0.CO;2-4
Oliyer J (1995) Continuum modelling of strong discontinuities in solid mechanics using damage models. Comput Mech. https://doi.org/10.1007/BF00356478
Oliver J (1996) Modelling strong discontinuities in solid mechanics via strain softening constitutive equations. Part 1: fundamentals. Int J Numer Methods Eng. https://doi.org/10.1002/(SICI)1097-0207(19961115)39:21%3c3575::AID-NME65%3e3.0.CO;2-E
Armero F, Garikipati K (1996) An analysis of strong discontinuities in multiplicative finite strain plasticity and their relation with the numerical simulation of strain localization in solids. Int J Solids Struct. https://doi.org/10.1016/0020-7683(95)00257-X
Linder C, Armero F (2007) Finite elements with embedded strong discontinuities for the modeling of failure in solids. Int J Numer Methods Eng. https://doi.org/10.1002/nme.2042
Sukumar N, Moës N, Moran B, Belytschko T (2000) Extended finite element method for three-dimensional crack modelling. Int J Numer Methods Eng. https://doi.org/10.1002/1097-0207(20000820)48:11%3c1549::AID-NME955%3e3.0.CO;2-A
Moës N, Dolbow J, Belytschko T (1999) A finite element method for crack growth without remeshing. Int J Nume Methods Eng. https://doi.org/10.1002/(SICI)1097-0207(19990910)46:1%3c131::AID-NME726%3e3.0.CO;2-J
Wells GN, Sluys LJ (2001) A new method for modelling cohesive cracks using finite elements. Int J Numer Methods Eng. https://doi.org/10.1002/nme.143
Gasser TC, Holzapfel GA (2003) Geometrically non-linear and consistently linearized embedded strong discontinuity models for 3D problems with an application to the dissection analysis of soft biological tissues. Comput Methods Appl Mech Eng. https://doi.org/10.1016/j.cma.2003.06.001
Babuška I, Melenk JM (1997) The partition of unity method. Int J Numer Methods Eng. https://doi.org/10.1002/(SICI)1097-0207(19970228)40:4%3c727::AID-NME86%3e3.0.CO;2-N
Hansbo A, Hansbo P (2004) A finite element method for the simulation of strong and weak discontinuities in solid mechanics. Comput Methods Appl Mech Eng. https://doi.org/10.1016/j.cma.2003.12.041
Fagerström M, Larsson R (2006) Theory and numerics for finite deformation fracture modelling using strong discontinuities. Int J Numer Methods Eng. https://doi.org/10.1002/nme.1573
Daux C, Moës N, Dolbow J, Sukumar N, Belytschko T (2000) Arbitrary branched and intersecting cracks with the extended finite element method. Int J Numer Methods Eng. https://doi.org/10.1002/1097-0207(20000830)48:12%3c1741::AID-NME956%3e3.0.CO;2-L
Kaczmarczyk Ł, Nezhad MM, Pearce C (2014) Three-dimensional brittle fracture: configurational-force-driven crack propagation. Int J Numer Methods Eng. https://doi.org/10.1002/nme.4603
Oliver J, Huespe AE, Sánchez PJ (2006) A comparative study on finite elements for capturing strong discontinuities: E-FEM vs X-FEM. Comput Methods Appl Mech Eng. https://doi.org/10.1016/j.cma.2005.09.020
Cervera M, Barbat GB, Chiumenti M, Wu JY (2021) A comparative review of XFEM, mixed FEM and phase-field models for quasi-brittle cracking. Arch Comput Methods Eng. https://doi.org/10.1007/s11831-021-09604-8
Bažant ZP, Jirásek M (2002) Nonlocal integral formulations of plasticity and damage: survey of progress. J Eng Mech 128:1119–1149. https://doi.org/10.1061/(ASCE)0733-9399(2002)128:11(1119)
Pijaudier-Cabot G, Bažant ZP (1987) Nonlocal damage theory. J Eng Mech 113:1512–1533. https://doi.org/10.1061/(ASCE)0733-9399(1987)113:10(1512)
Phu Nguyen V, Lloberas Valls O, Stroeven M, Johannes Sluys L (2010) On the existence of representative volumes for softening quasi-brittle materials—a failure zone averaging scheme. Comput Methods Appl Mech Eng. https://doi.org/10.1016/j.cma.2010.06.018
Stanić A, Brank B, Ibrahimbegovic A, Matthies HG (2021) Crack propagation simulation without crack tracking algorithm: embedded discontinuity formulation with incompatible modes. Comput Methods Appl Mech Eng 1:21. https://doi.org/10.1016/j.cma.2021.114090
Miehe C, Welschinger F, Hofacker M (2010) Thermodynamically consistent phase-field models of fracture: variational principles and multi-field FE implementations. Int J Numer Methods Eng. https://doi.org/10.1002/nme.2861
Francfort GA, Bourdin B, Marigo JJ (2008) The variational approach to fracture. J Elast. https://doi.org/10.1007/s10659-007-9107-3
Peerlings RHJ, de Borst R, Brekelmans WAM, de Vree JHP (1996) Gradient enhanced damage for quasi-brittle materials. Int J Numer Methods Eng. https://doi.org/10.1002/(SICI)1097-0207(19961015)39:19%3c3391::AID-NME7%3e3.0.CO;2-D
Oliver J (1989) A consistent characteristic length for smeared cracking models. Int J Numer Methods Eng. https://doi.org/10.1002/nme.1620280214
Bažant ZP, Jirásek M (2003) Nonlocal integral formulations of plasticity and damage: survey of progress. In: Perspectives in civil engineering: commemorating the 150th anniversary of the American Society of Civil Engineers. https://doi.org/10.1061/(asce)0733-9399(2002)128:11(1119)
Bažant ZP, Oh BH (1983) Crack band theory for fracture of concrete. Matériaux et Constructions. https://doi.org/10.1007/BF02486267
Wu JY, Nguyen VP, Nguyen CT, Sutula D, Sinaie S, Bordas SPA (2020) Phase-field modeling of fracture. Adv Appl Mech. https://doi.org/10.1016/bs.aams.2019.08.001
Jirásek M, Bauer M (2012) Numerical aspects of the crack band approach. Comput Struct. https://doi.org/10.1016/j.compstruc.2012.06.006
Ibrahimbegovic A (2009) Nonlinear solid mechanics. Solid Mech Appl. https://doi.org/10.1007/978-90-481-2331-5_1
Hill R (1958) A general theory of uniqueness and stability in elastic-plastic solids. J Mech Phys Solids. https://doi.org/10.1016/0022-5096(58)90029-2
Drucker DC (1959) A definition of stable inelastic material. J Appl Mech. https://doi.org/10.1115/1.4011929
Kristensen PK, Martínez-Pañeda E (2020) Phase field fracture modelling using quasi-Newton methods and a new adaptive step scheme. Theoret Appl Fract Mech. https://doi.org/10.1016/j.tafmec.2019.102446
Miehe C, Hofacker M, Welschinger F (2010) A phase field model for rate-independent crack propagation: robust algorithmic implementation based on operator splits. Comput Methods Appl Mech Eng. https://doi.org/10.1016/j.cma.2010.04.011
Molnár G, Gravouil A (2017) 2D and 3D Abaqus implementation of a robust staggered phase-field solution for modeling brittle fracture. Finite Elem Anal Des. https://doi.org/10.1016/j.finel.2017.03.002
Khalil Z, Elghazouli AY, Martínez-Pañeda E (2022) A generalised phase field model for fatigue crack growth in elastic–plastic solids with an efficient monolithic solver. Comput Methods Appl Mech Eng. https://doi.org/10.1016/j.cma.2021.114286
Zhou S, Zhuang X, Zhu H, Rabczuk T (2018) Phase field modelling of crack propagation, branching and coalescence in rocks. Theoret Appl Fract Mech. https://doi.org/10.1016/j.tafmec.2018.04.011
Wu JY, Huang Y, Nguyen VP (2020) On the BFGS monolithic algorithm for the unified phase field damage theory. Comput Methods Appl Mech Eng. https://doi.org/10.1016/j.cma.2019.112704
Armero F, Ehrlich D (2006) Numerical modeling of softening hinges in thin Euler–Bernoulli beams. Comput Struct. https://doi.org/10.1016/j.compstruc.2005.11.010
Ehrlich D, Armero F (2005) Finite element methods for the analysis of softening plastic hinges in beams and frames. Comput Mech. https://doi.org/10.1007/s00466-004-0575-z
Ibrahimbegovic A, Melnyk S (2007) Embedded discontinuity finite element method for modeling of localized failure in heterogeneous materials with structured mesh: an alternative to extended finite element method. Comput Mech. https://doi.org/10.1007/s00466-006-0091-4
Jukić M, Brank B, Ibrahimbegović A (2013) Embedded discontinuity finite element formulation for failure analysis of planar reinforced concrete beams and frames. Eng Struct. https://doi.org/10.1016/j.engstruct.2012.07.028
Dujc J, Brank B, Ibrahimbegovic A (2010) Multi-scale computational model for failure analysis of metal frames that includes softening and local buckling. Comput Methods Appl Mech Eng. https://doi.org/10.1016/j.cma.2009.09.003
Tenorio-Montero E, Juárez-Luna G (2021) Beam-column finite element with embedded discontinuities for modelling damage in reinforced concrete prismatic elements. Structures. https://doi.org/10.1016/j.istruc.2020.12.055
Bitar I, Benkemoun N, Kotronis P, Grange S (2019) A multifiber Timoshenko beam with embedded discontinuities. Eng Fract Mech. https://doi.org/10.1016/j.engfracmech.2019.03.032
Tojaga V, Kulachenko A, Östlund S, Gasser TC (2021) Modeling multi-fracturing fibers in fiber networks using elastoplastic Timoshenko beam finite elements with embedded strong discontinuities—formulation and staggered algorithm. Comput Methods Appl Mech Eng. https://doi.org/10.1016/j.cma.2021.113964
Bitar I, Kotronis P, Benkemoun N, Grange S (2018) A generalized Timoshenko beam with embedded rotation discontinuity. Finite Elem Anal Des. https://doi.org/10.1016/j.finel.2018.07.002
Nikolić M, Karavelić E, Ibrahimbegovic A, Miščević P (2018) Lattice element models and their peculiarities. Arch Comput Methods Eng. https://doi.org/10.1007/s11831-017-9210-y
Nikolic M, Ibrahimbegovic A (2015) Rock mechanics model capable of representing initial heterogeneities and full set of 3D failure mechanisms. Comput Methods Appl Mech Eng. https://doi.org/10.1016/j.cma.2015.02.024
Karavelić E, Nikolić M, Ibrahimbegovic A, Kurtović A (2019) Concrete meso-scale model with full set of 3D failure modes with random distribution of aggregate and cement phase. Part I: Formulation and numerical implementation. Comput Methods Appl Mech Eng. https://doi.org/10.1016/j.cma.2017.09.013
Benkemoun N, Hautefeuille M, Colliat JB, Ibrahimbegovic A (2010) Failure of heterogeneous materials: 3D meso-scale FE models with embedded discontinuities. Int J Numer Methods Eng. https://doi.org/10.1002/nme.2816
Pham BH, Brancherie D, Davenne L, Ibrahimbegovic A (2013) Stress-resultant models for ultimate load design of reinforced concrete frames and multi-scale parameter estimates. Comput Mech. https://doi.org/10.1007/s00466-012-0734-6
Imamovic I, Ibrahimbegovic A, Mesic E (2017) Nonlinear kinematics Reissner’s beam with combined hardening/softening elastoplasticity. Comput Struct. https://doi.org/10.1016/j.compstruc.2017.04.011
Nikolić M, Do XN, Ibrahimbegovic A, Nikolić Ž (2018) Crack propagation in dynamics by embedded strong discontinuity approach: enhanced solid versus discrete lattice model. Comput Methods Appl Mech Eng. https://doi.org/10.1016/j.cma.2018.06.012
Simo JC, Rifai MS (1990) A class of mixed assumed strain methods and the method of incompatible modes. Int J Numer Methods Eng. https://doi.org/10.1002/nme.1620290802
Ibrahimbegovic A, Wilson EL (1991) Modified method of incompatible modes. Commun Appl Numer Methods. https://doi.org/10.1002/cnm.1630070303
Rocha FF, Blanco PJ, Sánchez PJ, de Souza Neto E, Feijóo RA (2021) Damage-driven strain localisation in networks of fibres: a computational homogenisation approach. Comput Struct. https://doi.org/10.1016/j.compstruc.2021.106635
Ghareeb A, Elbanna A (2020) An adaptive quasicontinuum approach for modeling fracture in networked materials: application to modeling of polymer networks. J Mech Phys Solids. https://doi.org/10.1016/j.jmps.2019.103819
Andersen L, Nielsen S (2008) Elastic beams in three dimensions, DCE Lecture Notes No. 23
Krenk S (2009) Non-linear modeling and analysis of solids and structures. Cambridge University Press, Cambridge. https://doi.org/10.1017/CBO9780511812163
Luo Y (2008) An efficient 3D Timoshenko beam element with consistent shape functions. Adv Theor Appl Mech 1:95–106
Simo JC, Hughes TJR (1998) Computational inelasticity, 1st edn. Springer, New York. https://doi.org/10.1007/b98904
Cook RD, Malkus DS, Plesha ME, Witt RJ (2001) Concepts and applications of finite element analysis, 4th edn. Wiley, New York
Kouko J, Turpeinen T, Kulachenko A, Hirn U, Retulainen E (2020) Understanding extensibility of paper: role of fiber elongation and fiber bonding. Tappi J. https://doi.org/10.32964/TJ19.3.125
Acknowledgements
V. Tojaga wishes to acknowledge Dr. August Brandberg for providing technical support in ANSYS and Professor Mijo Nikolic for insights into computational aspects. This project has received funding from the European Union’s Horizon 2020 research and innovation program under the Marie Skłodowska-Curie grant agreement No 764713, project FibreNet, and KTH Royal Institute of Technology. The computations were performed on resources provided by the Swedish National Infrastructure for Computing (SNIC) at HPC2N (projects SNIC 2020/5-428 and SNIC 2021/6-51).
Funding
Open access funding provided by Royal Institute of Technology.
Author information
Authors and Affiliations
Contributions
VT Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Data curation, Writing—original draft, Writing—review and editing, Visualization, Project administration. AK Funding acquisition, Writing—review and editing, Supervision. SÖ Funding acquisition, Writing—review and editing, Supervision. TCG Conceptualization, Methodology, Validation, Formal analysis, Investigation, Writing—original draft, Writing—review and editing, Visualization, Supervision.
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Tojaga, V., Kulachenko, A., Östlund, S. et al. Hybrid of monolithic and staggered solution techniques for the computational analysis of fracture, assessed on fibrous network mechanics. Comput Mech 71, 39–54 (2023). https://doi.org/10.1007/s00466-022-02197-4
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00466-022-02197-4