betaFeedback

WorksMathematics

A smoothed particle hydrodynamics approach for phase field modeling of brittle fracture

Abstract

Fracture is a very challenging and complicated problem with various applications in engineering and physics. Although it has been extensively studied within the context of mesh-based numerical techniques, such as the finite element method (FEM), the research activity within the Smoothed Particle Hydrodynamics (SPH) community remains scarce. SPH is a particle-based numerical method used to discretize equations of continuum media. Its meshfree nature makes it ideal to simulate fracture scenarios that involve extreme deformations. However, to model fracture, SPH researchers have mostly relied on ad-hoc empirical local damage models, cohesive zone approaches, or pseudo-spring models, which come with a set of drawbacks and limitations. On the other hand, phase field models of brittle fracture have recently gained popularity in academic circles and provide significant improvements compared to previous approaches. These improvements include the derivation from fundamental fracture theories, the introduction of non-locality, and the ability to model multiple crack initiation, propagation, branching, and coalescence, in situations where no prior knowledge of the crack paths is available. Nevertheless, phase field for fracture has not been studied within SPH. In this proof-of-concept paper we develop and implement a phase field model of brittle fracture within the context of SPH. Comprehensive mathematical and implementation details are provided, and several challenging numerical examples are computed and illustrate the proposed method's ability to accurately and efficiently simulate complex fracture scenarios.

Graphical abstract showing a smoothed particle hydrodynamics approach for phase field modeling of brittle fracture, with domains $\Omega$ and $\Omega_2$, interface $\Gamma$, coordinate labels $X^1,x^1$, $X^2,x^2$, and $X^3,x^3$, model and experiment comparison, and particle interaction schematic

Figure 1.

Highlights

A smoothed particle hydrodynamics approach for phase field modeling of brittle fracture

Mohammad Naqib Rahimi, Georgios Moutsanidis

  • A framework for modeling brittle fracture within SPH is proposed.

  • SPH is posed in a total Lagrangian formulation to eliminate tensile instability.

  • Cracks are regularized over a length scale using a hyperbolic phase field approach.

  • The hyperbolic phase field equation is compatible with explicit time integration.

  • Challenging problems with complex fracture paths are presented.

Introduction

The Smoothed Particle Hydrodynamics (SPH) is one of the oldest and well established particle methods, and has an outstanding application history in many fields. SPH was originally developed as an interpolation technique to study astrophysics related problems [1, 2], and was later extended to discretize fluid mechanics equations [3]. One of the major advantages of SPH over other existing particle methods, is that the inclusion of new physics is quite straightforward, and it can therefore be readily extended to new areas of application [4]. When it comes to solid mechanics, its ability to handle extreme material distortion makes it ideal for simulating large deformation problems, where conventional Lagrangian mesh-based techniques fail due to mesh entanglement and the need for frequent mesh updating or remeshing. Many interesting variants of the technique have been developed since it’s initial appearance, and the interested reader should consult [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13] and the references therein.

Although SPH has been successfully applied to some complex problems in solid mechanics, fracture is still a challenging and open research area within the SPH community, and has not been extensively studied. This is because SPH is traditionally used to discretize equations of continuum mechanics, thus modeling of fracture and distinct crack surfaces is difficult. SPH researchers have so far relied on ad-hoc empirical local damage models, “cracking particles” approaches, pseudo-spring and virtual link approaches, and cohesive zone models. All the aforementioned techniques have undoubtedly advanced the state-of-the-art in simulating fracture within SPH. However, they come with certain limitations and drawbacks. For example, local damage [14] models are based on empirical damage laws rather than comprehensive fracture theories, and lead to mesh dependency and non-convergent results under refinement. The “cracking particles” approach [15] resembles the extended finite element method (XFEM) [16, 17], therefore the fracture surfaces need to be tracked, whereas the local enrichment of the approximation space leads to increased computational costs. In the pseudo-spring approach [18, 19, 20] the damage evolution is based on rather simple linear damage models, the softening curve of the damage law might lead to instabilities, and past research has shown that they are prone to spurious damage patterns. Finally, when it comes to cohesive zone models, the authors in [21, 22, 23] developed a continuum constitutive model featuring a cohesive fracture process zone, with an intrinsic physical length scale that helps remove the mesh dependency, while at the same time fracture surfaces do not need to be explicitly tracked. Nonetheless, kinematic enrichment is performed (similar to XFEM), which might potentially lead to increased computational times and complex implementation. Even though the aforementioned crack simulation techniques proved sufficient for certain classes of problems, it is evident that there is significant room for improvement when it comes to modeling fracture within the SPH framework.

Recently, phase field models of brittle fracture gained popularity within academic circles. Many interesting variants of the method have emerged and several challenging applications have been addressed [24, 25, 26, 27, 28, 29, 30, 31, 32, 33]. In phase field for fracture, cracks are not explicitly introduced in the solid, but instead the fracture surface is approximated by a phase field parameter that diffuses the discontinuity over a small region. The phase field parameter represents the material integrity, and it is a continuous variable that describes the smooth transition from the fully intact to the fully damaged state. The evolution of phase field, and hence of the fracture surface, is governed by a partial differential equation (PDE). Thus, a degree of non-locality is introduced, since the damage state at one point depends (through derivative information) on the states of neighboring points. The non-locality of the method leads to a well-posed mathematical model that exhibits mesh independence and convergence under refinement. At the same time, it can easily handle complicated discontinuity scenarios, such as crack initiation, propagation, coalescence, and branching, without prior knowledge of the crack paths, and without the need to embed any evolving discontinuities in the displacement field. As a result, propagating cracks are tracked automatically through the solution of the phase field PDE, which makes the method very attractive over other numerical approaches that require the explicit or implicit tracking of the discontinuities. Several approaches to phase field modeling of brittle fracture have been independently developed within the physics and mechanics communities, and for a thorough review the reader is encouraged to consult [27]. In the mechanics community, the starting point for deriving the method is the variational formulation of brittle fracture [34], which was regularized in [35], and extended Griffith’s theory of fracture [36].

Despite its many advantages, phase field modeling of brittle fracture has been mostly applied within the context of mesh-based numerical techniques, such as the finite element method (FEM) and isogeometric analysis (IGA) [24, 25, 26, 27, 37, 38, 39]. Employing particle and meshfree methods to discretize phase field equations for fracture comes with its own set of challenges, and very few relevant works can be found in the literature [40, 41, 42, 43, 44, 45, 46, 47, 48]. To the best of the authors’ knowledge, phase field models of brittle fracture have not been developed and investigated within SPH. In this proof-of-concept paper we develop a phase field model of brittle fracture suitable for SPH, and we demonstrate that phase field is a viable option to model fracture without running into issues that many of the previously used methods exhibit.

This paper is outlined as follows. In § 3, we review the basics of SPH, with a particular emphasis on total Lagrangian SPH. In § 4 we provide the basics of phase field modeling of brittle fracture, with a focus on hyperbolic phase field models that are amenable to explicit time integration. In § 5 we outline the proposed SPH framework for coupling phase field for fracture with solid mechanics. § 6 contains several challenging numerical examples. § 7 draws conclusions and outlines future research directions.

SPH approximation and the Total Lagrangian SPH

Conventional SPH approximation

The SPH approach was initially proposed as a smoothed interpolation technique to deal with problems related to astrophysics [1, 2]. The fundamental principle of SPH requires the integral representation of functions. The SPH utilizes a weighted interpolation technique to approximate the value of any arbitrary function f≡f(x)f \equiv f(\mathbf{x}) at any arbitrary point x\mathbf{x} in Euclidean space. This is also known as the kernel approximation and can be written as [1, 2]

f(x)≈∫Ωxf(x′)W(r,h) dx′,(1)f(\mathbf{x}) \approx\int_{\Omega_{\mathbf{x}}} f(\mathbf{x}')W(r,h)\,d\mathbf{x}', \tag*{(1)}

where x′∈Ωx\mathbf{x}' \in\Omega_{\mathbf{x}} refers to the spatial coordinates of all the points located within the interpolation space Ωx\Omega_{\mathbf{x}} of point x\mathbf{x}. The term W≡W(r,h)W \equiv W(r,h), known as the kernel, represents the interpolation weights as a function of the Euclidean distance r=∣x−x′∣r = \lvert\mathbf{x} - \mathbf{x}' \rvert and the smoothing length h=1.33Δxh = 1.33\Delta x, where Δx\Delta x is the initial particle spacing. The interpolant in Eq. (1) reproduces the function ff exactly if the kernel is a delta function; that is W=1W = 1 if x′=x\mathbf{x}' = \mathbf{x}, and W=0W = 0 if x′≠x\mathbf{x}' \ne\mathbf{x}. In practice, the kernel is chosen to be a compactly supported function and approaches the delta function as h→0h \to0.

The essential advantage of the SPH interpolation is that it allows an exact differentiation of the interpolant to produce the derivatives of function ff. The spatial derivatives of ff can be computed using the exact differentiation of the kernel as [1, 2]

∇f≈∫Ωxf(x′) ∇W dx′.(2)\nabla f \approx\int_{\Omega_{\mathbf{x}}} f(\mathbf{x}')\,\nabla W\,d\mathbf{x}'. \tag*{(2)}

However, this kind of approximation does not vanish if the function is constant. To improve the accuracy of SPH for constant fields, Eq. (2) is rewritten by replacing ff with Φf\Phi f, allowing for the nabla operator to be written as ∇(Φf)=Φ∇f+f∇Φ\nabla(\Phi f) = \Phi\nabla f + f\nabla\Phi. Thus, the derivative of the function becomes

∇f=1Φ[∇(Φf)−f∇Φ],(3)\nabla f = \frac{1}{\Phi}\left[\nabla(\Phi f) - f\nabla\Phi\right], \tag*{(3)}

where Φ\Phi is any differentiable field. Generally, in continuum mechanics related applications, Φ\Phi is set to be the density ρ\rho since it represents a physical quantity in the continuity equation. By doing so, one arrives at the following form of SPH approximation which delivers a zero derivative for constant functions [8]

∇f(x)≈1ρ(x)∫Ωxρ(x′)[f(x′)−f(x)]∇W dx′.(4)\nabla f(\mathbf{x}) \approx\frac{1}{\rho(\mathbf{x})}\int_{\Omega_{\mathbf{x}}}\rho(\mathbf{x}')\left[f(\mathbf{x}')-f(\mathbf{x})\right]\nabla W\,d\mathbf{x}'. \tag*{(4)}

In a spatially discretized SPH particle domain Eq. (1) and Eq. (4) are written as

fi≈∑j=1NifjWijVj,(5)f_i \approx\sum_{j=1}^{N_i} f_j W_{ij} V_j, \tag*{(5)}
∂fi∂xis≈1ρi∑j=1Niρj(fj−fi)∂Wij∂xjsVj(6)\frac{\partial f_i}{\partial x_i^s} \approx\frac{1}{\rho_i}\sum_{j=1}^{N_i}\rho_j(f_j-f_i)\frac{\partial W_{ij}}{\partial x_j^s}V_j \tag*{(6)}

in which NiN_i is the total number of particles located within the interpolation space, also known as the neighborhood, influence, or support domain, of particle ii. VjV_j is the infinitesimal volume of particle jj. WijW_{ij} is the kernel function relating the particles ii and jj, xjsx_j^s is the ss component of the spatial (or Eulerian) coordinate of particle jj, and ρj\rho_j is the density of particle jj.

As noted in [7, 11, 49], the discretization scheme introduced in Eq. (5) and Eq. (6) suffers from the particle inconsistency arising from the non-homogeneous distribution of particles and truncation of the support domain near the boundaries, as seen in Fig. 1. This leads to a lack of conservation of mass, and linear and angular momentum. A number of correction strategies to restore the

Discretized SPH particle domain with the illustration of full and truncated supports

Figure 1 (PDF p. 7). Discretized SPH particle domain with the illustration of full and truncated supports.

particle consistency based on the kernel and gradient correction have been introduced [7, 12, 50]. For its simplicity, throughout this paper, we follow the gradient correction approach presented in [7], in which the gradient is corrected as ∇W~ij=Ci⋅∇Wij\widetilde{\nabla W}_{ij}=\mathbf{C}_i\cdot\nabla W_{ij}.

Ci=−(∑j=1Ni(∇Wij⊗rij)Vj)−1(7)\mathbf{C}_i=-\left(\sum_{j=1}^{N_i}(\nabla W_{ij}\otimes\mathbf{r}_{ij})V_j\right)^{-1} \tag*{(7)}

is a second rank correction tensor, and rij=xi−xj\mathbf{r}_{ij}=\mathbf{x}_i-\mathbf{x}_j is the relative position vector of particles ii and jj. The use of Ci\mathbf{C}_i ensures that the gradient of any linear velocity field is exactly evaluated. From this point forward, for brevity, we will omit the approximation sign and will use “∇Wij\nabla W_{ij}” or “∂W/∂x\partial W/\partial x” to refer to the corrected form of the kernel gradient ∇W~ij=Ci⋅∇Wij\widetilde{\nabla W}_{ij}=\mathbf{C}_i\cdot\nabla W_{ij}.

SPH in Total Lagrangian form

So far, the general procedure for SPH as an approximation technique was presented. In this section, we further explore its applicability for solid mechanics related applications. It is well established that, when SPH is applied to solve the governing equations of solid mechanics, the use of Eulerian kernels (kernels defined on Eulerian coordinates) leads to the so-called tensile instability [5, 51]. Such kind of instability is observed in the form of material distortion under tensile stress state. This phenomenon was first noticed in [5] by carrying out the Von Neumann stability analysis (see [5] for detailed explanation). The authors in [5] concluded that for a stable solution the following condition should be satisfied

∂2W∂x2σ≤0,(8)\frac{\partial^2 W}{\partial x^2}\boldsymbol{\sigma} \le0, \tag*{(8)}

where σ\boldsymbol{\sigma} is the stress. This implies that as long as the left hand side of Eq. (8) is greater than 0 for a particle, the SPH approximation delivers unstable solutions. Several remedies have been suggested to overcome such instability. For example, [52] suggested an approach based on “Mutating the kernel” (see Section 6.6 of [52]). Since the tensile instability has a close relation with the second derivative of the kernel, using a proper kernel will result in a stable solution. However, such an approach is only effective in special cases [9]. Instead, the authors in [6] suggested the use of “stress points” in which additional computational nodes are introduced away from the original SPH particles to carry the stress information separately. However, in [53] it was noted that the use of Eulerian kernels along with the stress points does not fully mitigate the innate instability of SPH. The authors in [53] proposed the use of Lagrangian kernels which deliver a more stable solution. The difference between Eulerian and Lagrangian kernels is that the former is a function of Eulerian (or spatial) coordinates, whereas the later is established based on the Lagrangian/reference (or material) coordinates.

In the Total Lagrangian SPH (TLSPH) formalism, the kernel function and its derivatives are written as functions of the Lagrangian coordinates, thus, leading to the TLSPH approximation of an arbitrary function and its derivative as

fi=∑j=1Nim0jρ0jfjW0ij,(9)f_i = \sum_{j=1}^{N_i} \frac{m_{0j}}{\rho_{0j}} f_j W_{0ij}, \tag*{(9)}
∂fi∂Xis=1ρ0i∑j=1Nim0j(fj−fi)∂W0ij∂Xis,(10)\frac{\partial f_i}{\partial X_i^s} = \frac{1}{\rho_{0i}} \sum_{j=1}^{N_i} m_{0j}(f_j-f_i)\frac{\partial W_{0ij}}{\partial X_i^s}, \tag*{(10)}

where

X=x−u.(11)\mathbf{X} = \mathbf{x} - \mathbf{u}. \tag*{(11)}

W0ijW_{0ij}, m0m_0, and ρ0\rho_0 are the kernel, mass, and density of the associated particle evaluated in the reference (initial) configuration, respectively. u\mathbf{u} is the displacement, and X\mathbf{X} is the Lagrangian coordinate, as depicted in Figure 2. One major advantage of TLSPH is that, unlike the conventional SPH, its support domain does not change and is evaluated only once at the beginning of the simulation. This decreases the computational cost considerably. However, it becomes problematic in situations where the domain distortion is such that it leads to significant changes in a particle’s neighbors. Cases like this are not considered in this paper and will be studied in subsequent works.

Initial and displaced configurations in TLSPH

Figure 2. Initial and displaced configurations in TLSPH.

Governing Equations in TLSPH framework

In this section we present the governing equation of elastic dynamics in TLSPH formalism. Our primary concern is the conservation of momentum, given that the mass is constant, and the total energy is naturally conserved2. In elastodynamics, the momentum balance equation in the reference configuration is written as [53]

dvdt=1ρ0∇0⋅P+b0 in Ω0,(12)\frac{\mathrm{d}\mathbf{v}}{\mathrm{d}t}=\frac{1}{\rho_0}\nabla_0\cdot\mathbf{P}+\mathbf{b}_0\ \text{in }\Omega_0, \tag*{(12)}

where Ω0\Omega_0 is the domain of the continuum body in the reference coordinates, and v\mathbf{v}, b0\mathbf{b}_0, and tt, are the velocity, external force, and time, respectively. ∇0⋅P\nabla_0\cdot\mathbf{P} is the divergence of the first Piola–Kirchhoff stress tensor with respect to the reference coordinates. The first Piola–Kirchhoff stress tensor, P\mathbf{P}, is computed through an appropriate constitutive model, in which the deformation gradient is the corresponding measure of deformation. Details of the specific constitutive models employed in this work will be presented in subsequent sections. The deformation gradient is given as

F=∂x∂X=∂(u+X)∂X=∂u∂X+I,(13)\mathbf{F} = \frac{\partial\mathbf{x}}{\partial\mathbf{X}} = \frac{\partial(\mathbf{u}+\mathbf{X})}{\partial\mathbf{X}} = \frac{\partial\mathbf{u}}{\partial\mathbf{X}} + \mathbf{I}, \tag*{(13)}

where I\mathbf{I} is the identity matrix. In order to calculate the deformation gradient for particle i we utilize the TLSPH approach as follows

Fi=I+1ρ0i∑j=1Nim0j(uj−ui)⊗∇0W0ij,(14)\mathbf{F}_i = \mathbf{I} + \frac{1}{\rho_{0i}}\sum_{j=1}^{N_i}m_{0j}(\mathbf{u}_j-\mathbf{u}_i)\otimes\nabla_0W_{0ij}, \tag*{(14)}

or in index notation as

Fiks=δks+1ρ0i∑j=1Nim0jujik∂W0ij∂Xjs,(15)F_i^{ks} = \delta^{ks} + \frac{1}{\rho_{0i}}\sum_{j=1}^{N_i}m_{0j}u_{ji}^{k}\frac{\partial W_{0ij}}{\partial X_j^s}, \tag*{(15)}

in which ujiku_{ji}^{k} is the kk component of the displacement difference vector uji=uj−ui\mathbf{u}_{ji}=\mathbf{u}_j-\mathbf{u}_i, and δks\delta^{ks} is the Kronecker delta. Recall that ∂W0ij/∂Xjs\partial W_{0ij}/\partial X_j^s is the corrected derivative of the Lagrangian kernel (see § 3.1). Similarly, Eq. (12) can be written for particle i as [55]

dvidt=∑j=1Nim0j(Piρ0i2+Pjρ0j2+Pvij)⋅∇0W0ij+b0i,(16)\frac{d\mathbf{v}_i}{dt} = \sum_{j=1}^{N_i}m_{0j}\left(\frac{\mathbf{P}_i}{\rho_{0i}^{2}}+\frac{\mathbf{P}_j}{\rho_{0j}^{2}}+\mathbf{P}_{vij}\right)\cdot\nabla_0W_{0ij}+\mathbf{b}_{0i}, \tag*{(16)}

or in index notation as

dvikdt=∑j=1Nim0j(Piksρ0i2+Pjksρ0j2+Pvijks)∂W0ij∂Xjs+b0ik,(17)\frac{dv_i^k}{dt} = \sum_{j=1}^{N_i}m_{0j}\left(\frac{P_i^{ks}}{\rho_{0i}^{2}}+\frac{P_j^{ks}}{\rho_{0j}^{2}}+P_{vij}^{ks}\right)\frac{\partial W_{0ij}}{\partial X_j^s}+b_{0i}^{k}, \tag*{(17)}

where Einstein’s summation rule is employed for the repeated index ss. Furthermore, the artificial viscosity term Pvij=det⁡(Fi) πijFi−1\mathbf{P}_{vij}=\det(\mathbf{F}_i)\,\pi_{ij}\mathbf{F}_i^{-1} is included to avoid numerical instabilities arising from zero-energy mode discrepancy in the form of a jump in the field variables or shock waves [13, 19, 56, 57]. Following the work of [57] we compute the coefficient πij\pi_{ij} as

πij=1ρ0i(β2Gij2−β1c0iGij),(18)\pi_{ij} = \frac{1}{\rho_{0i}}\left(\beta_2G_{ij}^{2}-\beta_1c_{0i}G_{ij}\right), \tag*{(18)}

in which

c0i=c0=κ+43μρ0(19)c_{0i}=c_0=\sqrt{\frac{\kappa+\frac{4}{3}\mu}{\rho_0}} \tag*{(19)}

is the speed of sound at particle i, κ=E/(3(1−2ν))\kappa=E/(3(1-2\nu)) is the bulk modulus, μ=E/(2(ν+1))\mu=E/(2(\nu+1)) is the shear modulus, with EE and ν\nu being the elastic modulus and Poisson’s ratio, respectively. β1\beta_1 and $\beta_2 are scalar factors, and GijG_{ij} is calculated as

Gij=h(vi−vj)⋅(Xi−Xj)r0ij2+0.001h2(20)G_{ij} = \frac{h\left(\mathbf{v}_i-\mathbf{v}_j\right)\cdot\left(\mathbf{X}_i-\mathbf{X}_j\right)}{r_{0ij}^{2}+0.001h^{2}} \tag*{(20)}

Here, r0ij=∣Xi−Xj∣r_{0ij}=\left|\mathbf{X}_i-\mathbf{X}_j\right| is the initial distance between particles i and j. In most of the numerical examples in this paper we set β1=0.04\beta_1=0.04 and β2=0\beta_2=0. Note that for larger values of time step and lower particle resolutions, higher β1\beta_1 and β2\beta_2 values may be necessary. It should be mentioned that the artificial viscosity introduces nonphysical dumping to the system. Thus, the smallest possible values of β1\beta_1 and β2\beta_2 are preferred.

Griffith’s theory of brittle fracture and the phase field approximation

In this section, to be self contained, we briefly present the basics of phase field modeling of brittle fracture. In the mechanics community, the prevailing models mostly originate from the variational formulation of brittle fracture [34] and the related regularized formulation [27, 35]. According to Griffith’s theory of brittle fracture [36], the fracture energy density GcG_c is the amount of energy required to open a unit area of crack surface. Then, the total potential energy of an elastic body Ω\Omega, being the sum of the elastic energy and the fracture energy, is given by the expression

Πp(u,Γ)=∫Ωψe(ε(u)) dx+∫ΓGc dΓ(21)\Pi_p(u,\Gamma)=\int_{\Omega}\psi_e(\varepsilon(u))\,\mathrm{d}x+\int_{\Gamma}G_c\,\mathrm{d}\Gamma \tag*{(21)}

where ψe\psi_e is the elastic strain energy density, the details of which will be presented in subsequent sections, ε\boldsymbol{\varepsilon} is the infinitesimal strain tensor, and Γ\Gamma is the evolving internal discontinuity boundary which represents a set of discrete cracks. Extending this approach to dynamic problems the kinetic energy is defined as

Πk(u˙)=12∫Ωρ u˙⋅u˙ dx,(22)\Pi_k(\dot{\mathbf{u}}) = \frac{1}{2}\int_{\Omega} \rho\, \dot{\mathbf{u}}\cdot\dot{\mathbf{u}} \, dx, \tag*{(22)}

where ρ\rho is the material density in the current configuration, and the superimposed dot denotes time differentiation. Combining the kinetic energy with the potential energy of Eq. (21) we arrive at the Lagrangian of the discrete fracture problem

L(u,u˙,Γ)=Πk(u˙)−Πp(u,Γ)=∫Ω(12ρ u˙⋅u˙−ψe(ε(u)))dx−∫ΓGcdΓ.(23)L(\mathbf{u},\dot{\mathbf{u}},\Gamma) = \Pi_k(\dot{\mathbf{u}}) - \Pi_p(\mathbf{u},\Gamma) = \int_{\Omega}\left(\frac{1}{2}\rho\, \dot{\mathbf{u}}\cdot\dot{\mathbf{u}}-\psi_e(\boldsymbol{\varepsilon}(\mathbf{u}))\right)dx-\int_{\Gamma}G_c d\Gamma. \tag*{(23)}

The Euler-Lagrange equations of this functional determine the equations of motion of the body, including the entire process of crack initiation, propagation, and branching of preexisting cracks. However, the numerical treatment is quite complex because tracking of the evolving discontinuity Γ\Gamma is required, which leads to complicated and expensive computations [24, 26]. Therefore, to circumvent the above-mentioned difficulties, the regularized expression for the fracture energy was instead proposed in [35] given as

∫ΓGcdΓ≈∫ΩGc((1−s)24ϵ0+ϵ0∣∇s∣2)dx,(24)\int_{\Gamma}G_c d\Gamma\approx\int_{\Omega}G_c\left(\frac{(1-s)^2}{4\epsilon_0}+\epsilon_0|\nabla s|^2\right)dx, \tag*{(24)}

where ss is the so-called phase field parameter (or damage parameter) which represents the fracture surface Γ\Gamma. It is a continuous variable that describes the smooth transition from the undamaged state (s=1s=1) to the fully damaged one (s=0s=0). ϵ0\epsilon_0 is a parameter that has dimension of a length and controls the width of the smooth approximation of the crack (see Figure 3). When ϵ0→0\epsilon_0 \to0 the phase field approximation converges to the discrete fracture surface. In practice, ϵ0\epsilon_0 should be sufficiently small so that the physics of the problem are not violated, but at the same time it should be greater or roughly equal to the spatial discretization size so that the crack is regularized over some finite width. To model the loss of material stiffness, the elastic strain energy density is defined as [25, 26]

A solid body with an internal discontinuity and its phase field approximation

Figure 3. (a) A solid body Ω\Omega with internal discontinuity Γ\Gamma. (b) Approximation of the internal discontinuity by the phase field s(x,t)s(\mathbf{x},t). ϵ0\epsilon_0 is the length scale parameter that controls the width of the discontinuity.

ψe(ε,s)=s2ψe++ψe−.(25)\psi_e(\boldsymbol{\varepsilon},s)=s^2\psi_e^+ + \psi_e^-. \tag*{(25)}

ψe+\psi_e^+ and ψe−\psi_e^- are the positive and negative parts of the elastic strain energy density, respectively, which will be defined in the next section according to the particular constitutive models employed. As can become evident from Eq. (25), crack propagation is only allowed in tension since the phase field parameter is applied only to the tensile part of the elastic strain energy. By substituting the phase field approximations for the fracture energy Eq. (24) and the elastic energy density Eq. (25) into the expression for the Lagrangian functional Eq. (23), and by deriving the corresponding Euler-Lagrange equations, we arrive at the strong form equations of motion that consist of the momentum balance presented in Eq. (12) and the phase field parameter evolution equation

(4ϵ0ψe+Gc+1)s−4ϵ02∇2s=1 in Ω0,(26)\left(\frac{4\epsilon_0\psi_e^+}{G_c}+1\right)s-4\epsilon_0^2\nabla^2s=1\text{ in }\Omega_0, \tag*{(26)}

where ∇2s\nabla^2s is the Laplacian of ss.

Remark 1. It can be seen that the kinetic energy term in the Lagrange energy functional is not affected by the phase field parameter ss, which leads to conservation of mass.

In order to model the irreversibility condition (cracks do not heal) a history functional is introduced as

H(X,t)=max⁡τ≤t(ψe+(X,τ)),(27)\mathcal{H}(\mathbf{X},t)=\max_{\tau\le t}\left(\psi_e^+(\mathbf{X},\tau)\right), \tag*{(27)}

and is used in place of the tensile elastic strain energy ψe+\psi_e^+ in the phase field’s governing equation. The updated strong form that governs the evolution of the phase field parameter is

(4ϵ0HGc+1)s−4ϵ02∇2s=1 in Ω0.(28)\left(\frac{4\epsilon_0\mathcal{H}}{G_c}+1\right)s-4\epsilon_0^2\nabla^2s=1\text{ in }\Omega_0. \tag*{(28)}

Finally, initial conditions need to be introduced for the history functional, i.e.

H(X,0)=H0(X).(29)\mathcal{H}(\mathbf{X},0)=\mathcal{H}_0(\mathbf{X}). \tag*{(29)}

The initial condition of the history functional H0(X)\mathcal{H}_0(\mathbf{X}) can be used to model preexisting cracks in the domain. Eq. (28) and the corresponding initial condition Eq. (29) need to be solved together with the equations of motion (e.g. momentum balance Eq. (12)) to compute both the displacement field u(X,t)\mathbf{u}(\mathbf{X},t) and the phase field parameter s(X,t)s(\mathbf{X},t).

Although the aforementioned approach and its variants have been successfully applied within the context of numerical methods based on implicit time integration, they become problematic when lumped-mass explicit dynamics schemes are employed. The reason is that Eq. (28) is an elliptic PDE, and therefore a linear system needs to be solved in every time step to determine the phase field parameter. Though it is possible from an implementation point of view, the solution of an elliptic equation in every explicit step would make the approach computationally expensive. To overcome this difficulty, phase field models which include time dependency have been proposed. In [24] the phase field parameter evolves according to a parabolic PDE as

1Ms˙+2sψe+−Gc(2ϵ0∇2s+1−s2ϵ0)=0 in Ω0,(30)\frac{1}{M}\dot{s}+2s\psi_e^+-G_c\left(2\epsilon_0\nabla^2s+\frac{1-s}{2\epsilon_0}\right)=0\text{ in }\Omega_0, \tag*{(30)}

where MM is a parameter controlling the rate at which local damage information diffuses into the bulk material. Clearly, when M→∞M \to\infty the model approaches the standard elliptic models like the one of Eq. (26). However, Eq. (30) is a heat-equation-like PDE, and is subjected to an unfavorable CFL condition when explicit time integration is employed [58]

Δt≤Δx2,(31)\Delta t \le\Delta x^{2}, \tag*{(31)}

where Δt\Delta t and Δx\Delta x are time and length scales associated with the discretization. Therefore, the authors in [58] proposed a novel model in which the phase field parameter evolves according to a hyperbolic PDE by adding a second-order time derivative to the model of Eq. (30). The resulting PDE is

2Gcϵ0c2s¨+1Ms˙+2sψe−Gc(2ϵ0∇2s+1−s2ϵ0)=0 in Ω0.(32)\frac{2G_{\mathrm{c}}\epsilon_{0}}{c^{2}}\ddot{s}+\frac{1}{M}\dot{s}+2s\psi_{\mathrm{e}}-G_{\mathrm{c}}\left(2\epsilon_{0}\nabla^{2}s+\frac{1-s}{2\epsilon_{0}}\right)=0\ \text{in }\Omega_{0}. \tag*{(32)}

cc is a speed limit on the propagation of the phase field parameter through the undamaged material and is taken to be equal to the sound speed given in Eq. (19). The model of Eq. (32) is subjected to the hyperbolic stability condition of

cΔt≤Δx,(33)c\Delta t \le\Delta x, \tag*{(33)}

which is much less restrictive than the parabolic one (Eq. (31)), especially in the limit of Δx→0\Delta x \to0. To avoid any wave-like behavior due to the hyperbolic nature of the governing equation, the authors in [58] proposed an upper bound on MM such that the system is overdamped and the behavior of the phase field parameter is monotonic. The upper bound is given as

M≤c24Gcϵ0ψe+Gc2.(34)M \le\frac{c}{2\sqrt{4G_{\mathrm{c}}\epsilon_{0}\psi_{\mathrm{e}}+G_{\mathrm{c}}^{2}}}. \tag*{(34)}

Finally, the irreversibility condition is enforced in a way similar to the elliptic problem (Eq. (27)) and the updated hyperbolic governing equation for the phase field damage parameter becomes

2Gcϵ0c2s¨+1Ms˙+2sH−Gc(2ϵ0∇2s+1−s2ϵ0)=0 in Ω0,(35)\frac{2G_{\mathrm{c}}\epsilon_{0}}{c^{2}}\ddot{s}+\frac{1}{M}\dot{s}+2s\mathcal{H}-G_{\mathrm{c}}\left(2\epsilon_{0}\nabla^{2}s+\frac{1-s}{2\epsilon_{0}}\right)=0\ \text{in }\Omega_{0}, \tag*{(35)}

with the initial condition of Eq. (29), whereas the upper bound on the parameter MM becomes

M≤c24Gcϵ0H+Gc2.(36)M \le\frac{c}{2\sqrt{4G_{\mathrm{c}}\epsilon_{0}\mathcal{H}+G_{\mathrm{c}}^{2}}}. \tag*{(36)}

Traditionally, explicit time integration is used in SPH. Although there have been some efforts to develop SPH variants based on implicit time integration (e.g. [59]), the process is rather tedious and computationally expensive due to the large number of neighbors associated with each particle, and requires the solution of large linear systems. Thus, the hyperbolic version of the phase field PDE Eq. (35) is employed in this work.

Proposed framework

Governing equations

With the two previous sections at hand (§ 3 and § 4), we present a total Lagrangian SPH framework for the solution of the coupled elastodynamics–phase field problem expressed through the following strong form governing equations,

dvdt=1ρ0∇0⋅P+b0in Ω02Gcℓ0c2s¨+1Ms˙+2sH−Gc(2ℓ0∇2s+1−s2ℓ0)=0in Ω0(37)\begin{aligned} \frac{d\mathbf{v}}{dt} &= \frac{1}{\rho_0}\nabla_0 \cdot\mathbf{P}+\mathbf{b}_0 && \text{in }\Omega_0 \\ \frac{2G_{\mathrm{c}}\ell_0}{c^2}\ddot{s}+\frac{1}{M}\dot{s}+2s\mathcal{H}-G_{\mathrm{c}}\left(2\ell_0\nabla^2s+\frac{1-s}{2\ell_0}\right)&=0 && \text{in }\Omega_0 \tag*{(37)} \end{aligned}

In the above-mentioned coupled problem, the first equation corresponds to the balance of linear momentum, that was briefly outlined in § 3, whereas the second equation corresponds to the phase field (damage) parameter evolution. The solution of the phase field PDE is based upon the basics of TLSPH, and the overall solution procedure is outlined in this section. Before proceeding to the solution procedure we provide the details of the constitutive model.

Constitutive modeling

In § 3 we presented the basics of TLSPH for discretizing solid mechanics equations in terms of the first Piola–Kirchhoff stress, whereas in § 4 we outlined the basics of the phase field approach for brittle fracture in terms of a general elastic strain energy density ψe\psi_e. Here, we provide the details of the elastic strain energy density and the corresponding constitutive equations based on two hyperelastic models; the isotropic Saint-Venant Kirchhoff and the neo-Hookean model. However, we would like to point out that other constitutive models can be used with the proposed framework.

Saint-Venant Kirchhoff

The Saint-Venant Kirchhoff model can be easily derived by extending the linear-elastic framework presented in [26] to large rotations. This is achieved by replacing the infinitesimal strain tensor with the Green–Lagrange strain tensor E=12(FTF−I)\mathbf{E}=\frac{1}{2}(\mathbf{F}^{\mathrm{T}}\mathbf{F}-\mathbf{I}). Then, the elastic strain energy density functional becomes

ψe=12λ(tr⁡E)2+μtr⁡(E2),(38)\psi_e=\frac{1}{2}\lambda(\operatorname{tr}\mathbf{E})^2+\mu\operatorname{tr}(\mathbf{E}^2), \tag*{(38)}

where λ=κ−23μ\lambda=\kappa-\frac{2}{3}\mu and μ=E/(2(ν+1))\mu=E/(2(\nu+1)) are the Lame parameters. We then define

ψe+=12λ{tr⁡E}+2+μtr⁡(E+E+),(39)\psi_e^+=\frac{1}{2}\lambda\{\operatorname{tr}\mathbf{E}\}_+^2+\mu\operatorname{tr}(\mathbf{E}^+\mathbf{E}^+), \tag*{(39)}
ψe−=12λ{tr⁡E}−2+μtr⁡(E−E−),(40)\psi_e^{-} = \frac{1}{2}\lambda\{\operatorname{tr}\mathbf{E}\}_{-}^{2} + \mu\operatorname{tr}(\mathbf{E}^{-}\mathbf{E}^{-}), \tag*{(40)}

where the following decomposition is employed

E+=QΛ+QT,(41)\mathbf{E}^{+} = \mathbf{Q}\mathbf{\Lambda}^{+}\mathbf{Q}^{T}, \tag*{(41)}
E−=QΛ−QT,(42)\mathbf{E}^{-} = \mathbf{Q}\mathbf{\Lambda}^{-}\mathbf{Q}^{T}, \tag*{(42)}
E=QΛQT.(43)\mathbf{E} = \mathbf{Q}\mathbf{\Lambda}\mathbf{Q}^{T}. \tag*{(43)}

Λ=diag⁡(λ1,λ2,λ3)\mathbf{\Lambda} = \operatorname{diag}(\lambda_1,\lambda_2,\lambda_3) has the eigenvalues of E\mathbf{E} on its diagonal, Q\mathbf{Q} has the corresponding eigenvectors as its columns, Λ±=diag⁡(λ1±,λ2±,λ3±)\mathbf{\Lambda}^{\pm} = \operatorname{diag}(\lambda_1^{\pm},\lambda_2^{\pm},\lambda_3^{\pm}), and {.}±\{.\}_{\pm} selects the ±\pm part of its argument, i.e.

{x}±={xx∈R±0otherwise.(44)\{x\}_{\pm} = \begin{cases} x & x \in\mathbb{R}^{\pm}\\ 0 & \text{otherwise}. \end{cases} \tag*{(44)}

The second Piola–Kirchhoff stress can then be computed by differentiating the strain energy density ψe\psi_e with respect to the Green–Lagrange strain tensor E\mathbf{E},

S±=∂ψe±∂E=λ{tr⁡E}±I+2μE±,(45)\mathbf{S}^{\pm} = \frac{\partial\psi_e^{\pm}}{\partial\mathbf{E}} = \lambda\{\operatorname{tr}\mathbf{E}\}_{\pm}\mathbf{I} + 2\mu\mathbf{E}^{\pm}, \tag*{(45)}

and

S=S++S−.(46)\mathbf{S} = \mathbf{S}^{+} + \mathbf{S}^{-}. \tag*{(46)}

Finally, the first Piola–Kirchhoff stress can be computed as

P=FS.(47)\mathbf{P} = \mathbf{F}\mathbf{S}. \tag*{(47)}

Neo-Hookean

The Saint-Venant Kirchhoff model is known to exhibit instabilities in the case of strong compression [60]. Thus, for the problems in this paper involving strong compression, we employ a Neo-Hookean material. For the Neo-Hookean model we follow the presentation in [42, 61]. The positive and negative parts of the elastic strain energy density are given as

ψe+={U(J)+ψ‾e(b‾)J≥1ψ‾e(b‾)J<1,(48)\psi_e^{+} = \begin{cases} U(J) + \overline{\psi}_e(\overline{\mathbf{b}}) & J \ge1\\ \overline{\psi}_e(\overline{\mathbf{b}}) & J < 1, \end{cases} \tag*{(48)}
ψe−={0J≥1U(J)J<1.(49)\psi_e^{-} = \begin{cases} 0 & J \ge1\\ U(J) & J < 1. \end{cases} \tag*{(49)}

where

U(J)=12κ(12(J2−1)−ln⁡J),(50)U(J)=\frac{1}{2}\kappa\left(\frac{1}{2}\left(J^2-1\right)-\ln J\right), \tag*{(50)}
ψ‾e(b‾)=12μ(tr⁡b‾−3),(51)\overline{\psi}_e(\overline{\mathbf{b}})=\frac{1}{2}\mu\left(\operatorname{tr}\overline{\mathbf{b}}-3\right), \tag*{(51)}
J=det⁡F,(52)J=\det\mathbf{F}, \tag*{(52)}
b=FFT,(53)\mathbf{b}=\mathbf{F}\mathbf{F}^{T}, \tag*{(53)}
b‾=J−2/3b.(54)\overline{\mathbf{b}}=J^{-2/3}\mathbf{b}. \tag*{(54)}

The second Piola—Kirchhoff stress can then be computed as

S=2∂ψ‾e∂b‾.(55)\mathbf{S}=2\frac{\partial\overline{\psi}_e}{\partial\overline{\mathbf{b}}}. \tag*{(55)}

For the given elastic strain energy density this results in

S=2{s2(U′(J)∂J∂b+∂ψ‾e(b‾)∂b)J≥1U′(J)∂J∂b+s2∂ψ‾e(b‾)∂bJ<1.(56)\mathbf{S}=2\begin{cases} s^2\left(U'(J)\frac{\partial J}{\partial\mathbf{b}}+\frac{\partial\overline{\psi}_e(\overline{\mathbf{b}})}{\partial\mathbf{b}}\right) & J\ge1 \\ U'(J)\frac{\partial J}{\partial\mathbf{b}}+s^2\frac{\partial\overline{\psi}_e(\overline{\mathbf{b}})}{\partial\mathbf{b}} & J<1. \end{cases} \tag*{(56)}

The derivatives in the above expression are computed as

U′(J)=12κ(J−J−1),(57)U'(J)=\frac{1}{2}\kappa\left(J-J^{-1}\right), \tag*{(57)}
∂J∂b=∂det⁡b∂b=12Jb−1,(58)\frac{\partial J}{\partial\mathbf{b}}=\frac{\partial\sqrt{\det\mathbf{b}}}{\partial\mathbf{b}}=\frac{1}{2}J\mathbf{b}^{-1}, \tag*{(58)}
∂ψ‾e∂b=∂ψ‾e∂b‾∂b‾∂b=J−2/32μ(I−13(tr⁡b)b−1).(59)\frac{\partial\overline{\psi}_e}{\partial\mathbf{b}}=\frac{\partial\overline{\psi}_e}{\partial\overline{\mathbf{b}}}\frac{\partial\overline{\mathbf{b}}}{\partial\mathbf{b}}=\frac{J^{-2/3}}{2}\mu\left(\mathbf{I}-\frac{1}{3}(\operatorname{tr}\mathbf{b})\mathbf{b}^{-1}\right). \tag*{(59)}

Substituting the above equations into Eq. (56) we get

S={s2(12κ(J2−1)b−1+J−2/3μ(I−13(tr⁡b)b−1))J≥112κ(J2−1)b−1+s2(J−2/3μ(I−13(tr⁡b)b−1))J<1.(60)\mathbf{S}=\begin{cases} s^2\left(\frac{1}{2}\kappa\left(J^2-1\right)\mathbf{b}^{-1}+J^{-2/3}\mu\left(\mathbf{I}-\frac{1}{3}(\operatorname{tr}\mathbf{b})\mathbf{b}^{-1}\right)\right) & J\ge1 \\ \frac{1}{2}\kappa\left(J^2-1\right)\mathbf{b}^{-1}+s^2\left(J^{-2/3}\mu\left(\mathbf{I}-\frac{1}{3}(\operatorname{tr}\mathbf{b})\mathbf{b}^{-1}\right)\right) & J<1. \end{cases} \tag*{(60)}

Finally, the first Piola—Kirchhoff stress is computed as in Eq. (47).

Solution procedure and numerical implementation

We first define two common terms that will be used regularly throughout this paper, the soft particle and the phase field limit.

Definition 1. Soft (damaged) particle refers to any SPH particle whose phase field parameter drops sufficiently so that its stress is fairly low, and thus the effect on neighboring particles is negligible. From a numerical implementation point of view, soft particles behave as rigid objects having inertia but no internal force.

Definition 2. The phase field limit refers to the threshold value of the phase field parameter below which the particle is assumed to be a soft particle.

In order to compute the first Piola–Kirchhoff stress through the constitutive model presented in the previous subsection, we redefine the deformation gradient as

Fi={∂ui∂Xi+Isi>slIotherwise(61)\mathbf{F}_i = \begin{cases} \dfrac{\partial\mathbf{u}_i}{\partial\mathbf{X}_i}+\mathbf{I} & s_i>s_l \\ \mathbf{I} & \text{otherwise} \end{cases} \tag*{(61)}

Here sls_l is the phase field limit that determines the particle’s stiffness state and ∂ui/∂Xi\partial\mathbf{u}_i/\partial\mathbf{X}_i is calculated utilizing the corrected form of the TLSPH approach (see § 3.3) as

∂uik∂Xis=1ρ0i∑j=1Nm0jujik∂W0ij∂Xjs,(62)\frac{\partial u_i^k}{\partial X_i^s}=\frac{1}{\rho_{0i}}\sum_{j=1}^{N}m_{0j}u_{ji}^k\frac{\partial W_{0ij}}{\partial X_j^s}, \tag*{(62)}

where ujiku_{ji}^k is the kk component of the displacement difference vector uji=uj−ui\mathbf{u}_{ji}=\mathbf{u}_j-\mathbf{u}_i between particles i\mathbf{i} and j\mathbf{j}. The kernel function in Eq. (62), and in any other calculation involving SPH interpolation, is chosen to be a cubic spline [62] defined in reference coordinates as

W0ij={0q>214C(2−q)31≤q≤2C(1−1.5q2+0.75q3)0≤q≤1(63)W_{0ij}= \begin{cases} 0 & q>2 \\ \dfrac{1}{4}C(2-q)^3 & 1\leq q\leq2 \\ C(1-1.5q^2+0.75q^3) & 0\leq q\leq1 \end{cases} \tag*{(63)}

in which q=∣Xi−Xj∣/hq=\lvert\mathbf{X}_i-\mathbf{X}_j\rvert/h and CC is a constant determined as

C={2/(3h)for one dimensional space10/(7πh2)for two dimensional space1/(πh3)for three dimensional space(64)C= \begin{cases} 2/(3h) & \text{for one dimensional space} \\ 10/(7\pi h^2) & \text{for two dimensional space} \\ 1/(\pi h^3) & \text{for three dimensional space} \end{cases} \tag*{(64)}

where hh is the smoothing length. We further rearrange Eq. (35) as

s˙=c22Gcϵ0[Gc(2ϵ0∇2s+1−s2ϵ0)−1Ms˙−2sH](65)\dot{s}=\frac{c^2}{2G_c\epsilon_0}\left[G_c\left(2\epsilon_0\nabla^2s+\frac{1-s}{2\epsilon_0}\right)-\frac{1}{M}\dot{s}-2s\mathcal{H}\right] \tag*{(65)}

in order to have an explicit definition of the phase field inertia s˙\dot{s}, in terms of the phase field parameter ss, its time derivative s˙\dot{s}, and its Laplacian ∇2s\nabla^2s. The Laplacian of the phase field is calculated utilizing the SPH Laplacian operator [63] in TLSPH formalism as

∇2si=2∑j=1N(si−sj)V0jr0ijs∣r0ij∣2∂W0ij∂Xjs,(66)\nabla^{2}s_i = 2\sum_{j=1}^{N}(s_i-s_j)V_{0j}\frac{r_{0ij}^{s}}{\lvert\mathbf{r}_{0ij}\rvert^{2}}\frac{\partial W_{0ij}}{\partial X_{j}^{s}}, \tag*{(66)}

where V0jV_{0j} is the initial volume of jj, ∣r0ij∣\lvert\mathbf{r}_{0ij}\rvert is the magnitude of the relative initial position vector r0ij=Xi−Xj\mathbf{r}_{0ij} = \mathbf{X}_i-\mathbf{X}_j, and r0ijsr_{0ij}^{s} is the ss-component of r0ij\mathbf{r}_{0ij}. Note that Einstein summation is used for the repeated index ss. After solving for the phase field inertia, the phase field and its first time derivative are evolved explicitly, and the results are used in updating the first Piola–Kirchhoff stress through the corresponding constitutive model. Finally, the momentum balance equation is solved for the other TLSPH field variables, e. g. v\mathbf{v}, v˙\dot{\mathbf{v}}, u\mathbf{u}.

Remark 2. It can be observed that for soft particles (si<sls_i < s_l) the deformation is assumed to be zero (Eq. (61)). Although in reality soft particles have obviously non-zero deformations, this is a convenient assumption that improves the numerical stability of the proposed algorithm, and does not affect the physics of the problem since the deformation gradient is only used in the stress computation which is close to zero for soft particles. From a numerical implementation perspective, soft particles stick to their neighboring undamaged particles and move with them as rigid bodies, with their only effect to the problem being through means of inertia. Obviously, a higher value of the soft particle limit would interfere with the physics of the fracture, and hence it should be selected as small as possible. Based on our experience, a value of 0.1 is enough to improve the numerical stability while preserving the physics of the problem.

The overall computational procedure of the proposed framework consists of two main modules; a pre-processing module and a time integration module. In the pre-processing module, the domain is first discretized into particles, the field variables are initialized, and any preexisting discontinuities are introduced. Then, a neighbor search is performed to determine the support domain of each particle. At this stage the conventional kernel and its derivatives are computed using Eq. (63), and then the gradient correction tensor for each particle is computed from Eq. (7). The corrected kernel gradients are subsequently computed and substituted utilizing the ∇~Wij=Ci⋅∇Wij\widetilde{\nabla}W_{ij} = \mathbf{C}_i \cdot\nabla W_{ij} relation (see § 3.1). The details of the prepossessing module are given in Algorithm 1.

Remark 3. Traditionally, in phase field for fracture, a preexisting crack is modeled either through a geometrical notch of finite width, or a prescribed value in the history functional, H0\mathcal{H}_0. Although both approaches have been extensively and successfully applied in the phase field literature, problems can potentially arise, especially in the case where the discontinuity is introduced as a physical notch. For example, in the case of meshfree and particle methods, care must be taken so that the physical discontinuity is large enough and the supports of the particles in either side of the crack do

Algorithm 1 Preprocessing module

Preprocess

  • Get input from file or user interface

  • Discretize the continuous domain into discrete particles

  • Introduce preexisting discontinuities

for i←1i \leftarrow1 to Total number of particles do

  • Search for j neighbor of particle i in its support domain. Add j to the neighbor list of i.

  • Compute kernel W0ijW_{0ij}, and its gradient ∇0W0ij\nabla_{0}W_{0ij}, between particles i and j using Eq. (63).

end for

for i←1i \leftarrow1 to Total number of particles do

  • Compute gradient correction tensor of particle i, CiC_i, using Eq. (7) or in index form as

Ciks=−(∑j=1Nir0ijs∂W0ij∂xjrV0j)−1,C_i^{ks}=-\left(\sum_{j=1}^{N_i}r_{0ij}^{s}\frac{\partial W_{0ij}}{\partial x_j^r}V_{0j}\right)^{-1},

where NiN_i is the number of neighbors of particle i.

end for

for i←1i \leftarrow1 to Total number of particles do

for j←1j \leftarrow1 to NiN_i do

  • Correct the kernel gradient as ∂W0ij∂xjk=Ciks∂W0ij∂xjs\frac{\partial W_{0ij}}{\partial x_j^k}=C_i^{ks}\frac{\partial W_{0ij}}{\partial x_j^s} (summation is done over index ss)

end for

end for

  • Initialize field variables

  • Apply static loads (if exist)

not overlap. Here, we propose a third way of modeling preexisting cracks, in which the neighbor search for particles that lie in either side of the discontinuity is restricted in that side of the crack only, as can be seen in Figure 4d. Simply put, we are blocking any communication among particles that lie in different sides of the preexisting crack surface Γ0\Gamma_0. This way the preexisting discontinuity is modeled easily and exactly without the need for a physical geometrical gap. One major advantage of this approach is that for arbitrarily shaped complex discontinuities a neighbor search restriction can be easily performed, whereas introducing geometrical gaps is fairly complicated. In this paper, we used either the physical gap or the neighbor search restriction approach and observed that both lead to similar results.

Four approaches to modeling a preexisting discontinuity

Figure 4. A description of a) preexisting discontinuity and its approximation by means of b) introducing a geometrical notch, c) assigning a prescribed value in the history functional, and d) restricting the neighbor search only among particles that lie in the same side of the discontinuity with the particle under consideration.

When it comes to time integration, a second-order Euler predictor-corrector time integration scheme is adopted with a CFL condition of

Δt≤min⁡(0.25hc0+∣vmax⁡∣,Δxc0)(67)\Delta t \leq\min\left(0.25\frac{h}{c_0+\lvert\mathbf{v}_{\max}\rvert},\frac{\Delta x}{c_0}\right) \tag*{(67)}

where the second term ensures the stability of the phase field solution, and vmax⁡\mathbf{v}_{\max} is the maximum possible velocity of the computational domain in space and time. At the beginning of each time step the field variables are updated utilizing the predictor scheme as

vi(t+1/2)=vi(t)+0.5Δt(dvidt)(t−1/2)(68)\mathbf{v}_i^{(t+1/2)}=\mathbf{v}_i^{(t)}+0.5\Delta t\left(\frac{d\mathbf{v}_i}{dt}\right)^{(t-1/2)} \tag*{(68)}
ui(t+1/2)=ui(t)+0.5Δtvi(t+1/2)(69)\mathbf{u}_i^{(t+1/2)}=\mathbf{u}_i^{(t)}+0.5\Delta t\mathbf{v}_i^{(t+1/2)} \tag*{(69)}
s˙i(t+1/2)=s˙i(t)+0.5Δts¨i(t−1/2)(70)\dot{\mathbf{s}}_i^{(t+1/2)}=\dot{\mathbf{s}}_i^{(t)}+0.5\Delta t\ddot{\mathbf{s}}_i^{(t-1/2)} \tag*{(70)}
s˙i(t+1/2)=s˙i(t)+0.5Δts¨i(t+1/2)(71)\dot{s}_i^{(t+1/2)}=\dot{s}_i^{(t)}+0.5\Delta t\ddot{s}_i^{(t+1/2)} \tag*{(71)}

where the superscripts (t+1/2)(t+1/2), (t−1/2)(t-1/2), and (t)(t) denote the values of the related field variable at half time step, previous half time step, and previous time step, respectively. Then, using the values of u(t+1/2)\mathbf{u}^{(t+1/2)}, the deformation gradient of the TLSPH particles is computed using Eqs. (61)-(62). At this stage, an appropriate constitutive model is utilized (see § 5.2.1 or § 5.2.2) to compute the first Piola–Kirchhoff stress tensor as a function of the computed deformation gradient and the predicted half time step values of phase field (damage). Then, Eq. (17), Eq. (66), and Eq. (65) are in turn solved using the values of the variables at half-time step to compute the acceleration (dv/dt)(t+1/2)(d\mathbf{v}/dt)^{(t+1/2)}, Laplacian of the phase field (∇2s)(t+1/2)(\nabla^2s)^{(t+1/2)}, and the phase field inertia s¨(t+1/2)\ddot{s}^{(t+1/2)}, respectively. Finally, the corrected values for the field variables are computed as

vi(t+1)=vi(t)+Δt(dvidt)(t+1/2),(72)\mathbf{v}_i^{(t+1)}=\mathbf{v}_i^{(t)}+\Delta t\left(\frac{d\mathbf{v}_i}{dt}\right)^{(t+1/2)}, \tag*{(72)}
ui(t+1)=ui(t)+Δtvi(t+1),(73)\mathbf{u}_i^{(t+1)}=\mathbf{u}_i^{(t)}+\Delta t\mathbf{v}_i^{(t+1)}, \tag*{(73)}
s˙i(t+1)=s˙i(t)+Δts¨i(t+1/2),(74)\dot{s}_i^{(t+1)}=\dot{s}_i^{(t)}+\Delta t\ddot{s}_i^{(t+1/2)}, \tag*{(74)}
si(t+1)=si(t)+Δts˙i(t+1),(75)s_i^{(t+1)}=s_i^{(t)}+\Delta t\dot{s}_i^{(t+1)}, \tag*{(75)}

where the (t+1)(t+1) superscript denotes the current time step. Note that the initial conditions for s˙\dot{s} and s¨\ddot{s} are set to zero. Algorithm 2 and Algorithm 3 present the implementation of the time integration module and the constitutive models for a single step calculation, respectively. It should be noted that for problems involving the contact of two separate bodies an additional interface force is included in the momentum equation (Eq. (17)). Hence, the momentum equation becomes

dvikdt=∑j=1Nim0j(Piksρ0i2+Pjksρ0j2+Pvijks)∂W0ij∂Xjs+b0ik+∑a=1N^ifairaik(76)\frac{d v_i^k}{dt}=\sum_{j=1}^{N_i}m_{0j}\left(\frac{P_i^{ks}}{\rho_{0i}^{2}}+\frac{P_j^{ks}}{\rho_{0j}^{2}}+P_{vij}^{ks}\right)\frac{\partial W_{0ij}}{\partial X_j^s}+b_{0i}^k+\sum_{a=1}^{\hat{N}_i}f_{ai}r_{ai}^k \tag*{(76)}

where N^i\hat{N}_i is the total number of particles located at a separate body within the contact distance, rcr_c, of particle i, raikr_{ai}^k is the k component of the relative position vector rai=xa−xi\mathbf{r}_{ai}=\mathbf{x}_a-\mathbf{x}_i between particles a and i calculated using the Eulerian coordinates x\mathbf{x}, as given in Figure 5, and faif_{ai} is a scalar multiplier whose details are given in Appendix § 9. For the cases involving contact interaction in this paper we set rc=2.5Δxr_c=2.5\Delta x unless stated otherwise.

Interface details of the contact domain with multiple body contacts in SPH

Figure 5. Interface details of the contact domain with multiple body contacts in SPH.

Numerical examples

In this section, we first apply the proposed framework to model various challenging problems from the literature. We then illustrate further capabilities of our model by addressing the failure

Algorithm 2 Single step calculations of time integration module

Time integration

▷ Apply dynamic force and/or displacement boundary conditions

for i←1i \leftarrow1 to Total number of particles do ▷ Compute half-time step values of field variables using Eqs. (68)-(71) end for

for i←1i \leftarrow1 to Total number of particles do ▷ Compute deformation gradient Fi\mathbf{F}_i as in Eq. (61), using half step displacement values ▷ Utilize Algorithm 3 to calculate Pi\mathbf{P}_i, ψei+\psi_{ei}^{+}, and ψei−\psi_{ei}^{-} using the appropriate constitutive model if ψei+>Hi\psi_{ei}^{+} > \mathcal{H}_i then &nbsp;&nbsp;▷ Hi=ψei+\mathcal{H}_i = \psi_{ei}^{+} (here we update the history functional) end if ▷ Compute MiM_i parameter as in Eq. (36) (note that we use the upper bound value of M) ▷ Using half-step values of the phase field parameter s(t+1/2)s^{(t+1/2)} compute the Laplacian ∇2si\nabla^2s_i as in Eq. (66) ▷ Using half step values of s(t+1/2)s^{(t+1/2)} and s˙(t+1/2)\dot{s}^{(t+1/2)} compute s¨(t+1/2)\ddot{s}^{(t+1/2)} as in Eq. (65) ▷ If contact exists, using half-step values of x(t+1/2)\mathbf{x}^{(t+1/2)}, compute contact force for particle i as in Appendix § 9 end for

for i←1i \leftarrow1 to Total number of particles do ▷ Using half step values of field variables compute acceleration (dvi/dt)(t+1/2)(d\mathbf{v}_i/dt)^{(t+1/2)} as in Eq. (17) (or Eq. (76) for problems involving contact interactions) ▷ Correct field variables using Eqs. (72)-(75) end for

▷ Write results (occasionally)

Algorithm 3 Implementation of the Saint-Venant Kirchhoff and neo-Hookean hyperelastic constitutive models

If constitutive model is Saint-Venant Kirchhoff:

  • Compute Green–Lagrange strain tensor as Eiks=12(FimkFims+δks)E_i^{ks} = \frac{1}{2}\left(F_i^{mk}F_i^{ms}+\delta^{ks}\right) (summation is done over index m)

  • Decompose EiE_i into Ei+E_i^{+} and Ei−E_i^{-} using Eqs. (41)-(42)

  • Compute ψei+\psi_{ei}^{+} and ψei−\psi_{ei}^{-} using Eqs. (39)-(40)

  • Compute Si+S_i^{+} and Si−S_i^{-} using Eq. (45)

  • Compute Si=(si(t+1/2))2Si++Si−S_i = \left(s_i^{(t+1/2)}\right)^2 S_i^{+} + S_i^{-}

  • Compute Piks=FikmSimsP_i^{ks} = F_i^{km}S_i^{ms} (summation is done over index m)

Else if constitutive model is neo-Hookean:

  • Compute JJ, b\mathbf{b}, and b‾\overline{\mathbf{b}} using Eq. (52), Eq. (53), and Eq. (54), respectively.

  • Compute U(J)U(J) and ψ‾e(b)\overline{\psi}_e(\mathbf{b}) using Eq. (50) and Eq. (51), respectively.

  • Compute ψei+\psi_{ei}^{+} and ψei−\psi_{ei}^{-} using Eq. (48) and Eq. (49), respectively.

  • Compute SiS_i from Eq. (60) using half-step values of phase field, si(t+1/2)s_i^{(t+1/2)}.

  • Compute Piks=FikmSimsP_i^{ks} = F_i^{km}S_i^{ms} (summation is done over index m)

Else:

  • Apply a different constitutive model here to calculate PiP_i, ψei+\psi_{ei}^{+}, and ψei−\psi_{ei}^{-}

occurrences in a three dimensional impact problem. All problems except the last one are simulated in two dimensions under plane strain assumption with a CPU parallel code. The simulations were carried out on a 40-core node on SeaWulf cluster located at the Institute for Advanced Computational Science (IACS) at Stony Brook University. The animated videos of the simulations are provided in Electronic Annex I (Appendix § 10).

Symmetric three point bending test

We first solve the crack propagation in a plate subjected to symmetric three point bending [25, 27]. The length and width of the plate are set to Lx=8.2L_x = 8.2 mm and Ly=2L_y = 2 mm, respectively. The material parameters λ=12\lambda= 12 GPa, μ=8\mu= 8 GPa, density ρ0=1190\rho_0 = 1190 kg/m3^3, and Gc=0.54G_c = 0.54 kJ/m2^2 are adopted from [25]. The contact parameters, length scale, and time step values, are set to Kp=9×1012K_p = 9 \times10^{12} and r0=2.5Δxr_0 = 2.5\Delta x, ϵ0=Δx\epsilon_0 = \Delta x, and Δt=0.1\Delta t = 0.1 ns, respectively. A preexisting crack with a length of 0.4 mm is located at the lower central portion of the plate and is modeled by restricting the neighbor search of the particles around the discontinuity region, as suggested in Section 4.3 and Figure 4d. We enforce the boundary conditions through contact with rigid circular bands with outer diameters of 0.4 mm at three regions, as shown in Figure 6. The contact force is applied in the X2X^2-direction only, and a velocity of −5-5 m/s in the X2X^2-direction is assigned to the upper band, whereas the two lower bands are fixed in the X1X^1- and X2X^2-directions. We set the artificial viscosity

Symmetric three point bending test geometry and loading conditions

Figure 6. Symmetric three point bending test. Geometry and loading conditions.

parameters (see § 3.3) to β1=0.2\beta_1 = 0.2 and β2=0\beta_2 = 0, and a phase field limit of sl=0.1s_l = 0.1 is chosen for soft particles. The domain is discretized into 423,665 particles with a particle spacing of Δx=6.25\Delta x = 6.25 μ\mum, out of which 3,825 particles belong to the contact bodies. Both the Saint-Venant Kirchhoff and Neo-Hookean constitutive models are employed. Figure 7 shows the snapshots of the phase field parameter at different stages of the propagation. As can be seen, the propagation starts from the preexisting crack tip, continues in a vertical direction, and stops near the upper boundary of the plate. The start and end time of the propagation correspond to a displacement of 0.045 mm and 0.1 mm of Point A (see Figure 6) in the X2X^2-direction, which agrees well with the ones reported in [25]. In Figure 8 we compare the propagation length predicted by Saint-Venant Kirchhoff and Neo-Hookean constitutive models, in which both models deliver similar propagation curves.

Contours of the phase field parameter for different values of the $X^{2}$-displacement

Figure 7. Symmetric three point bending test. Contours of the phase field parameter (crack path) for different values of the X2X^{2}-displacement of Point A. The snapshots correspond to computations performed with the Saint-Venant Kirchhoff constitutive model.

Evolution of the propagation length of the crack for the Saint-Venant Kirchhoff and Neo-Hookean constitutive models

Figure 8. Symmetric three point bending test. Evolution of the propagation length of the crack for the Saint-Venant Kirchhoff and Neo-Hookean constitutive models.

Asymmetric three point bending test

Here, we model the asymmetric three point bending test [25, 27, 64]. The boundary conditions are similar to the previous case and are given in Figure 9. The geometric properties and material parameters are adopted as Lx=20L_x = 20 mm, Ly=8L_y = 8 mm, λ=12\lambda= 12 GPa, μ=8\mu= 8 GPa, ρ0=1190\rho_0 = 1190 kg/m3^3, and Gc=1G_c = 1 kJ/m2^2. We model the preexisting crack by restricting the neighbor search of the

Asymmetric three point bending test geometry and loading conditions

Figure 9. Asymmetric three point bending test. Geometry and loading conditions.

particles around the discontinuity region. The contact force is applied in the X2X^2-direction with contact parameters of Kp=12×1012K_p = 12 \times10^{12} and r0=2.5Δxr_0 = 2.5\Delta x. Other simulation parameters are set to ϵ0=Δx\epsilon_0 = \Delta x, Δt=0.1\Delta t = 0.1 ns, β1=0.2\beta_1 = 0.2, β2=0\beta_2 = 0, and sl=0.1s_l = 0.1. We discretize the domain into approximately 400K particles with a particle spacing of Δx=0.02\Delta x = 0.02 mm. The Saint-Venant Kirchhoff constitutive model is employed. Figure 10 shows the snapshots of the phase field for the deformed plate at different stages of the propagation. As can be seen, the crack initiation occurs when point A has displaced by approximately 0.17 mm in the X2X^2-direction. Then, the crack joins the middle hole after a displacement of approximately 0.21 mm, and the complete rupture is recorded at a displacement of around 0.325 mm. The crack first propagates on a somewhat curved path from the preexisting crack tip to the middle hole, and then on an almost linear path until the full rupture. Figure 11 compares the experimentally [64] and numerically [25, 27] obtained crack paths with the one predicted by the present approach. As can be seen, the predicted crack path agrees very well with both the experimental and numerical studies found in the literature. There is a small deviation observed within a small region near the middle hole. One potential reason for this could be the imperfect representation of the circular holes in the discretized particle domain, which can alter the local strain energy concentrations. Nevertheless, the recorded deviation is negligible and does not change the overall physical behavior of the crack.

Asymmetric three point bending test

Figure 10. Asymmetric three point bending test. Contours of the phase field parameter (crack path) for different values of the X2X^2-displacement of Point A.

Asymmetric three point bending test

Figure 11. Asymmetric three point bending test. Crack path compared with experimental [64] and phase field results of Miehe et al. [25] and Ambati et al. [27].

Dynamic crack branching

The dynamic crack branching problem has been widely investigated for different boundary conditions and loading setups [26, 42, 58, 65, 66]. In the present effort we focus on the setup used in [26, 42, 58], where a plate with a notch is subjected to a tensile surface loading. We set the material properties of the plate to E=32E = 32 GPa, ρ=2450\rho= 2450 kg/m3^{3}, ν=0.2\nu= 0.2, and Gc=3G_{\mathrm{c}} = 3 J/m2^{2}, whereas the Saint-Venant Kirchhoff constitutive model is employed. The length and width of the plate are taken as Lx=100L_x = 100 mm and Ly=40L_y = 40 mm, respectively. The preexisting notch is introduced as a geometrical discontinuity with a width of 4Δx4\Delta x and length of 50 mm located at the center. The plate is assumed to be under a tensile load of 1 MPa applied to its upper and lower surfaces, as depicted in Figure 12. The length scale and time step values are set to ϵ0=0.09375\epsilon_0 = 0.09375 mm and Δt=2.5\Delta t = 2.5 ns, respectively. We set the artificial viscosity parameters (see § 3.3) to β1=0.04\beta_1 = 0.04 and β2=0\beta_2 = 0, and a phase field limit of sl=0.1s_l = 0.1 is chosen for soft particles. To demonstrate the resolution independence of the crack pattern in the phase field method, we consider four different particle resolutions3, R1-R4, with initial particle spacing of Δx=0.1\Delta x = 0.1 mm, Δx=0.08\Delta x = 0.08 mm, Δx=\Delta x = 0.06250.0625 mm, and Δx=0.05\Delta x = 0.05 mm, leading to a discretization of the domain into 398,000, 622,500, 1,020,800, and 1,596,000 particles, respectively.

Dynamic crack branching geometry and loading conditions

Figure 12. Dynamic crack branching. Geometry and loading conditions.

Figure 13 presents several snapshots of the phase field contours for R1-R4 resolutions at different stages of the crack propagation for the right half of the plate (the left half stays intact and we therefore omit that region). The crack propagation starts from the tip of the notch at approximately t=10t = 10 μ\mus and continues in a horizontal manner. At approximately t=50t = 50 μ\mus, branching occurs, and the fracture propagates at an angle of ≈37∘\approx37^\circ. At later stages, the propagation angles of the two branches tend to change back to horizontal. As expected, an identical crack pattern is predicted for all different resolutions, highlighting phase field’s feature of mesh independence, which is hard to achieve when local damage models are employed. At the same time, the predicted crack patterns are in excellent agreement with the ones reported in [26, 42, 58]. However, the crack in resolutions R1 and R2 fails to reach the end of the plate in the given time. This is a common behavior seen for low discretization resolutions and it is attributed to the fact that for lower resolutions the fracture zone is slightly wider and hence there is more fracture energy per unit length. Similar behavior was also reported in [26, 42, 58]. Figure 14 compares the crack propagation speed predicted by the current model with the ones of [26, 58]. A close agreement is observed between the present results and the ones of [58]. In [26], however, a slightly higher propagation speed is reported. This difference is attributed to the absence of the inertia term (s¨\ddot{s}) in the phase-field PDE utilized in [26] (see § 4). Obviously, the inclusion of the inertia term in the current effort (i.e. the use of a hyperbolic phase field PDE instead of an elliptic one, as described in § 4) introduces an extra amount of work to be carried out in order for the phase field to propagate, thereby causing problem.

Dynamic crack branching

Figure 13. Dynamic crack branching. Contours of the phase field parameter (crack path) at 55 μ\mus (left), 80 μ\mus (middle), and 105 μ\mus (right), for R1-R4 resolutions. The snapshots are presented for the right half of the plate.

Dynamic crack branching

Figure 14. Dynamic crack branching. X1X^{1}-component of crack velocity for resolutions R1-R4 compared with the phase field results of Kamensky et al. [58] and Borden et al. [26].

Additionally, in TLSPH, for lower resolutions and large time step values, higher amount of artificial viscosity is needed to overcome the zero-energy mode discrepancy [57], thus, leading to a higher amount of nonphysical forces in the domain. This can then alter the original speed of the crack propagation, and in some cases even the path.

a decrease in the propagation speed. However, it should be pointed out that no experimental data exist for this problem, therefore it is not clear which model’s crack propagation speed is more accurate. Nonetheless, as mentioned in [58], the elliptic models tend to overestimate the crack propagation speeds because they ignore the rate toughening effect in brittle fracture. It is also

worth mentioning that the strategy employed to track the crack tip, and hence the propagation speed, highly affects the recorded crack initiation time. Here, we consider the furthest particle in the +X1+X^{1}-direction with a phase field value of s<0.7s < 0.7 to be the tip of the crack. Therefore, as soon as the phase field value decreases to 0.7, the propagation is assumed to initiate. However, the choice of this value does not affect the computed propagation speed since the speed is computed between the two iso-curves of the same value.

In Figure 15 the post-processed final results of the phase field and the crack opening are given for resolution R4, where the particles with s<sl=0.1s < s_l = 0.1 are not shown, and the deformation is scaled by a factor of 50. As can be seen, after the complete failure, the domain separates into three regions, marking the end of the propagation.

Dynamic crack branching showing the opening of the crack and phase field distribution

Figure 15. Dynamic crack branching. Opening of the crack and phase field distribution at the failure moment (≈105 μs\approx105\,\mu\mathrm{s}) for resolution R4. Particles with s<sl=0.1s < s_l = 0.1 are not shown. The deformation is magnified by a factor of 50.

Kalthoff–Winkler experiment

Here we revisit the well-known Kalthoff–Winkler experiment where a plate is subjected to an impact shear loading [67]. The length and width of the plate are adopted from [26, 42, 58] as Lx=100L_x = 100 mm and Ly=100L_y = 100 mm, respectively. The material properties are set to E=190E = 190 GPa, ρ=8000\rho= 8000 kg/m3^{3}, ν=0.3\nu= 0.3, Gc=22.13G_c = 22.13 kJ/m2^{2}, and the Saint-Venant Kirchhoff constitutive model is employed. The domain is discretized into 638,400 particles with an initial particle spacing of Δx=0.125\Delta x = 0.125 mm. The length scale, time step, artificial viscosity parameters, and phase field limit for the soft particles, are chosen as ϵ0=0.125 mm\epsilon_0 = 0.125\ \mathrm{mm}, Δt=1.0 ns\Delta t = 1.0\ \mathrm{ns}, β1=0.04\beta_1 = 0.04 and β2=0\beta_2 = 0, and sl=0.1s_l = 0.1, respectively. A geometrical discontinuity with a width of 4Δx4\Delta x and length of 50 mm is introduced in the left side of the plate to model the preexisting notch, as shown in Figure 16. The plate is subjected

Kalthoff–Winkler experiment geometry and loading conditions

Figure 16. Kalthoff–Winkler experiment. Geometry and loading conditions.

to an impact loading in the horizontal direction, applied to the particles located at the lower portion of the left surface. The loading condition is applied as a time-dependent ramp-up velocity of

v={tt016.5 m/st≤t0,16.5 m/sotherwise,(77)v = \begin{cases} \frac{t}{t_0}16.5\ \mathrm{m/s} & t \le t_0, \\ 16.5\ \mathrm{m/s} & \text{otherwise}, \end{cases} \tag*{(77)}

in the X1X^{1}-direction, where t0=1 μst_0 = 1\ \mu\mathrm{s} is the ramp-up time. Upon impact (t=0t = 0), the displacement field starts propagating towards the right, eventually causing stress and strain energy concentrations at the notch tip, which are then dissipated by the propagating crack (or phase field). Figure 17 shows the snapshots of the results for different stages of the propagation. The crack propagation starts at approximately 25 μ\mus and continues diagonally until the full rupture at approximately 100 μ\mus. A local disturbance in the diagonal path of the crack is recorded between t≈63t \approx63 μ\mus and t≈75t \approx75 μ\mus (marked in Circle A in Figure 17) which implies a mixed mode I–II fracture (also reported in [26, 42, 58]). Figure 18 compares the propagation speed and crack orientation predicted by the

Figure 17. Kalthoff–Winkler experiment. a) Contours of the phase field parameter and b) opening of the crack at 50 μ\mus (left), 80 μ\mus (middle), and 101 μ\mus (right). In b), particles with s<sl=0.1s < s_l = 0.1 are not shown and the deformation is magnified by a factor of 5.

Propagation speed and orientation of crack compared to phase field results of Borden et al. and peridynamics results of Candas et al.

Figure 18. Kalthoff–Winkler experiment. Propagation speed and orientation of crack compared to phase field results of Borden et al. [26] and peridynamics results of Candas et al. [68].

current approach with the ones of the elliptic phase field [26] and peridynamic [68] models. Due to the explicit dynamics nature of the present computations that results in a somewhat oscillatory behavior, the graphs were smoothed for visualization purposes. The portion belonging to the disturbance region (Circle A) is marked with gray color. The predicted results of crack propagation and orientation, as well as the elastodynamic behavior of the plate, are in good agreement with the aforementioned computational efforts and the corresponding experimental work. Overall, the proposed framework is able to capture all the complex qualitative and quantitative characteristics of the crack propagation.

Notched plate with hole

In this example, we simulate the interaction of cracks with other structural defects in a porous plate with an eccentric hole. The geometry of the plate is provided in Figure 19. The material properties are set to λ=1.94\lambda= 1.94 GPa, μ=2.45\mu= 2.45 GPa, ρ=2000\rho= 2000 kg/m3^3, Gc=2.28G_c = 2.28 kJ/m2^2, and the Saint-Venant Kirchhoff constitutive model is employed. The domain is discretized into 732,852 particles with an initial particle spacing of Δx=0.1\Delta x = 0.1 mm. The length scale, time step, artificial viscosity parameters, and phase field limit for the soft particles, are chosen as ϵ0=0.1\epsilon_0 = 0.1 mm, Δt=4.0\Delta t = 4.0 ns, β1=0.04\beta_1 = 0.04 and β2=0\beta_2 = 0, and sl=0.1s_l = 0.1, respectively. The original experiment [27] was performed in a displacement-control manner, where a prescribed displacement rate of 0.1 mm/min was applied. The corresponding simulations [27, 29, 69] were carried out under static displacement loading condition. For both the experiment and the simulations, the prescribed displacement was applied to the upper hole (colored in blue in Figure 19) of the plate, while the lower hole (colored in red in Figure 19) was fixed in the horizontal and vertical directions. Here, due to the explicit nature of the SPH method, we conduct dynamic simulations, with three different cases of prescribed velocity in the vertical direction; v=0.5v = 0.5 m/s, 0.2 m/s, 0.1 m/s, applied on two layers of particles around the upper hole, whereas the two layers of particles around the lower hole are fixed in the X1X^1- and X2X^2-directions. A preexisting crack with a length of 10 mm is located in the left-central portion of the plate, as shown in Figure 19, and is modeled by restricting the neighbor search of the particles around the discontinuity region, as suggested in Section 4.3 and Figure 4d.

Notched plate with hole showing geometry and loading conditions

Figure 19. Notched plate with hole. Geometry and loading conditions. The particles around the lower hole (colored in red) are fixed in the X1X^1- and X2X^2-directions, and the prescribed velocity condition is applied to the particles around the upper hole (colored in blue) in the X2X^2-direction. The preexisting crack with a length of 10 mm can be seen in red 65 mm from the bottom edge of the specimen.

The crack initiation is observed after the upper hole has displaced by approximately 0.3 mm in the X2X^{2}-direction for all three cases of the prescribed velocity. The presence of the eccentric hole forms a weak-zone in the vicinity of the crack tip which generates the so-called accelerating effect on the propagating crack [65, 70]. As a result, the crack path is directed towards the hole, and eventually joins the hole after a displacement of approximately 0.4 mm. Upon joining the hole the crack dissipates its energy, leading to a significant delay in propagation, caused by the arresting effect of the hole. At a displacement of approximately 1.4 mm, a new crack initiates from the right side of the hole and causes a fast and complete rapture. The complex physics in terms of the accelerating and arresting effects of the hole on the crack are well-captured by the proposed approach. Identical accelerating and arresting effects were reported in [65, 70] using peridynamics. Figure 20 depicts the snapshots of the final crack pattern under different prescribed velocity loading conditions. As it is seen, the present approach is able to capture all the major qualitative features

Notched plate with hole: comparison of final crack paths with experimental results

Figure 20. Notched plate with hole. Comparison of the final crack path with the experimental results of Ambati et al. [27]. Particles with s<0.5s < 0.5 are not shown.

of fracture, such as the damaged region around the supports, as well as the tendency of the crack to branch in the vicinity of the eccentric hole (marked with circles in Figure 20), which demonstrates a good agreement with the experimental results of [27]. A small deviation of the crack path in terms of small branching is observed for higher prescribed velocity values. This is in agreement with the reported effects of the dynamic loading on crack branching [71] and the experimental observations in [27]. Figure 21 compares the phase field values for different loading conditions at various stages of the propagation. As can be seen, for lower values of the prescribed velocity, the branching effect close to the hole vanishes, which agrees well with the findings of [71]. Figure 22 presents the smoothed graph of the reaction force (at the lower hole) versus maximum displacement (displacement of the upper hole) for v=0.1v = 0.1 m/s. We would like to point out that although this problem has been investigated by several researchers in the past [27, 29, 69], the corresponding reaction force–displacement graphs are not in agreement among each other. Furthermore, all the

Notched plate with hole showing contours of the phase field parameter under different velocity loading conditions

Figure 21. Notched plate with hole. Contours of the phase field parameter (crack path) under different velocity loading conditions for upper hole displacement values of 0.3 mm (first column), 0.42 mm (second column), 1.42 mm (third column), and 1.48 mm (fourth column).

Reaction force–displacement graph for a notched plate with hole, with snapshots of the crack at various stages

Figure 22. Notched plate with hole. Reaction force–displacement graph for v=0.1v = 0.1 m/s. Snapshots of the crack at various stages are superimposed for a comparative view.

previous works, in contrast to the present effort, solved the problem in a static manner. Hence, some discrepancies are expected in the reaction force–displacement curve. However, the qualitative features of the graph are captured well. Moreover, the time at which the first crack initiation occurs (sudden drop in the curve at approximately 0.3 mm) is in excellent agreement with the one reported in [29, 69].

Impact of a spherical projectile on a notched circular plate

To further demonstrate the capabilities of our proposed framework we simulate the impact of a rigid spherical projectile on a notched circular plate. Figure 23 illustrates the geometric properties of the domain in which the diameter and thickness of the plate are given as 37 mm and 5Δx5\Delta x, respectively. A rigid spherical projectile with a diameter of 16Δx16\Delta x and a prescribed velocity of

Impact of a spherical projectile on a notched circular plate, showing the geometry and loading conditions

Figure 23. Impact of a spherical projectile on a notched circular plate. Geometry and loading conditions.

−10-10 m/s in the X3X^{3}-direction impacts the plate at time t=0t = 0. A through-the-thickness cross-shaped notch with side lengths of 18Δx18\Delta x is applied at the center of the plate and is modeled by restricting the neighbor search of the particles around the notch. The displacement is constrained at the outer edge of the plate. We set the material and simulation parameters to λ=4.75 GPa\lambda= 4.75\ \mathrm{GPa}, μ=0.9 GPa\mu= 0.9\ \mathrm{GPa}, ρ0=1200 kg/m3\rho_0 = 1200\ \mathrm{kg}/\mathrm{m}^3, Gc=703 J/m2G_c = 703\ \mathrm{J}/\mathrm{m}^2, ϵ0=Δx\epsilon_0 = \Delta x, Δt=2.5 ns\Delta t = 2.5\ \mathrm{ns}, β1=0.05\beta_1 = 0.05, β2=0\beta_2 = 0, sl=0.1s_l = 0.1, r0=2Δxr_0 = 2\Delta x, and Kp=1.2×1012K_p = 1.2 \times10^{12}. The domain is discretized into 800,774 particles and the particle spacing is Δx=0.165 mm\Delta x = 0.165\ \mathrm{mm}, whereas the Neo-Hookean constitutive model is employed. Figure 24 and Figure 25 show the three- and two-dimensional views of the phase field in the deformed plate, respectively. As can be seen, upon impact the plate starts to deform, and after approximately 80 μs\mu\mathrm{s}

Impact of a spherical projectile on a notched circular plate. Three dimensional view of the phase field parameter

Figure 24. Impact of a spherical projectile on a notched circular plate. Three dimensional view of the phase field parameter.

Impact of a spherical projectile on a notched circular plate. Two dimensional view of the phase field parameter

Figure 25. Impact of a spherical projectile on a notched circular plate. Two dimensional view of the phase field parameter.

the propagation initiates from the back side of the preexisting notch tips. Since the back side of the plate undergoes higher tension than the front, the propagation at the back starts earlier. At later stages of the propagation, the plate is forced to break from the boundaries of the weak regions under high tension which correspond to the regions under the impact of the projectile, thus, leading to an ultimate penetration. After the penetration, the plate bounces back and dissipates its accumulated strain energy marking the end of crack propagation. Figure 26 shows the displacements of Point A and Point B (see Figure 23) in the X3X^3-direction with respect to time. As can be seen, the proposed approach is able to efficiently capture the pre- and post-failure behaviors of the complex contact problem with complex fracture patterns. This problem is merely a demonstration of capabilities, and although neither experimental nor numerical results exist for comparison purposes, all the qualitative features that are expected from the underlying physics are produced.

Impact of a spherical projectile on a notched circular plate

Figure 26. Impact of a spherical projectile on a notched circular plate. Displacements of Point A and Point B in the X3X^{3}-direction.

Conclusions

We presented a robust, accurate, and convergent computational framework for modeling dynamic brittle fracture within the SPH method. The proposed approach employs the well-known SPH approximation technique to discretize a coupled system of a recently developed hyperbolic phase field model of brittle fracture and solid mechanics. The use of SPH allows for the simulation of large deformation problems that potentially involve multi-body interactions and fragmentation, whereas the use of a hyperbolic PDE for the phase field governing equation permits explicit integration in time, without having to solve expensive linear systems associated with elliptic models, or introduce severe time step restrictions associated with parabolic models.

The results of fracture simulations performed with the proposed formulation are in excellent agreement with solutions calculated using standard mesh-based numerical techniques, such as the finite element method and isogeometric analysis. At the same time, they are in very good agreement with results from experimental studies. The last numerical example, although it does not have any corresponding experimental or numerical results, is a demonstration-of-capabilities study, and produces all the qualitative features that are expected from the underlying physics.

This proof-of-concept work presents the core formulation for coupling solid mechanics with phase field for fracture within the SPH framework. Immediate future efforts will focus on large deformation fluid–structure interaction problems, functionally graded materials, as well as parallelization based on graphics processing units (GPU).

Acknowledgments

The authors would like to thank Stony Brook Research Computing and Cyberinfrastructure, and the Institute for Advanced Computational Science at Stony Brook University for access to the high-performance SeaWulf computing system, which was made possible by a $1.4M National Science Foundation grant (#1531492).

Interface force for contact problems in SPH

In a typical contact problem in SPH an additional term arising from the contact of two separate bodies is included in the governing equation. Thus, the resulting momentum equation (Eq. (17)) becomes

dvikdt=∑j=1Nim0j(Piksρ0i2+Pjksρ0j2+pijks)∂W0ij∂Xjs+b0ik+∑a=1N^ifairaik(78)\frac{dv_i^k}{dt}=\sum_{j=1}^{N_i}m_{0j}\left(\frac{P_i^{ks}}{\rho_{0i}^{2}}+\frac{P_j^{ks}}{\rho_{0j}^{2}}+p_{ij}^{ks}\right)\frac{\partial W_{0ij}}{\partial X_j^s}+b_{0i}^k+\sum_{a=1}^{\hat{N}_i}f_{ai}r_{ai}^k \tag*{(78)}

In the newly included term, N^i\hat{N}_i is the total number of particles located at a separate body within the contact distance of particle ii, raikr_{ai}^k is the k component of the relative position vector rai=xa−xi\mathbf{r}_{ai}=\mathbf{x}_a-\mathbf{x}_i between particles a and i calculated using the Eulerian coordinates x\mathbf{x} (see Figure 5), and faif_{ai} is a scalar multiplier. Borrowing mostly from [72] we calculate faif_{ai} as

fai=m0aKp∣rai∣2{BaiRai∣rai∣≤rc0otherwise(79)f_{ai}=\frac{m_{0a}K_p}{\lvert\mathbf{r}_{ai}\rvert^2}\begin{cases}B_{ai}R_{ai} & \lvert\mathbf{r}_{ai}\rvert\le r_c \\ 0 & \text{otherwise}\end{cases} \tag*{(79)}

where

Bai=1+∣rai∣rc1−e(e1+∣rai∣rc−1)(80)B_{ai}=\frac{1+\frac{\lvert\mathbf{r}_{ai}\rvert}{r_c}}{1-e}\left(e^{1+\frac{\lvert\mathbf{r}_{ai}\rvert}{r_c}}-1\right) \tag*{(80)}
Rai={(1+1.5q~)(2−q~)38d(∣rai∣)dt≤00otherwise(81)R_{ai}=\begin{cases}\frac{(1+1.5\tilde{q})(2-\tilde{q})^3}{8}\frac{d(\lvert\mathbf{r}_{ai}\rvert)}{dt} & \le0 \\ 0 & \text{otherwise}\end{cases} \tag*{(81)}
q~=∣rai∣1.33Δx.(82)\tilde{q}=\frac{\lvert\mathbf{r}_{ai}\rvert}{1.33\Delta x}. \tag*{(82)}

m0am_{0a} is the initial mass of particle a, rcr_c is the contact distance, and KpK_p is a constant determining the potential of the contact. The KpK_p and rcr_c constants need to be determined carefully through numerical experiments in order to preserve the physical nature of contact and to avoid any penetration.

Supplementary data

Supplementary material related to this article can be found online at https://doi.org/10.1016/j.cma.2022.115191.

References

  1. [1]Robert A. Gingold and Joseph J. Monaghan. Smoothed particle hydrodynamics: theory and application to non-spherical stars. Monthly Notices of the Royal Astronomical Society, 181(3):375–389, 1977.
  2. [2]Leon. B. Lucy. A numerical approach to the testing of the fission hypothesis. The Astronomical Journal, 82(12):1013–1024, 1977.DOI
  3. [3]Joseph J. Monaghan. Simulating free surface flows with sph. Journal of Computational Physics, 110(2):399–406, 1994.DOI
  4. [4]Larry Libersky and Albert Petschek. Smooth particle hydrodynamics with strength of materials’, in Advances in the Free-Lagrange Method, volume 395, pages 248–257. 01 2006.
  5. [5]Jeff W. Swelge, Darrell L. Hicks, and Steve W. Attaway. SPH Stability Analysis. Journal of Computational Physics, 116:123–134, 1995.
  6. [6]Carl T. Dyka and Robert P. Ingel. An approach for tension instability in smoothed particle hydrodynamics (sph). Computers & Structures, 57(4):573–580, 1995.DOI
  7. [7]Javier Bonet and T. S.L. Lok. Variational and momentum preservation aspects of Smooth Particle Hydrodynamic formulations. Computer Methods in Applied Mechanics and Engineering, 180(1-2):97–115, 1999.DOI
  8. [8]Joseph J. Monaghan. Smoothed particle hydrodynamics. Reports on Progress in Physics, 68(8):1703–1759, 2005.
  9. [9]Juan R. Reveles. Development of a total Lagrangian SPH code for the simulation of solids under dynamic loading. PhD thesis, Cranfield University, 2007.DOI
  10. [10]M.B. Liu and G.R. Liu. Smoothed Particle Hydrodynamics (SPH): an Overview and Recent Developments. Archives of Computational Methods in Engineering, 17(1):25–76, 2010.DOI
  11. [11]Leonardo Di G. Sigalotti, Jaime Klapp, Otto Rendón, Carlos A. Vargas, and Franklin Peña-Polo. On the kernel and particle consistency in smoothed particle hydrodynamics. Applied Numerical Mathematics, 108:242–255, 2016.
  12. [12]Shan Qun Chen, Bin Liao, and Tao Huang. Corrected SPH methods for solving shallow-water equations. Journal of Hydrodynamics, 28(3):389–399, 2016.
  13. [13]M. Naqib Rahimi, Deniz Can Kolukısa, Mehmet Yildiz, Murat Ozbulut, and Adnan Kefal. A generalized hybrid smoothed particle hydrodynamics–peridynamics algorithm with a novel lagrangian mapping for solution and failure analysis of fluid–structure interaction problems. Computer Methods in Applied Mechanics and Engineering, 389:114370, 2022.DOI
  14. [14]Timon Rabczuk and J Eibl. Simulation of high velocity concrete fragmentation using sph/mlsph. International Journal for Numerical Methods in Engineering, 56(10):1421–1444, 2003.DOI
  15. [15]Timon Rabczuk and Ted Belytschko. Cracking particles: a simplified meshfree method for arbitrary evolving cracks. International Journal for Numerical Methods in Engineering, 61(13):2316–2343, 2004.DOI
  16. [16]Nicolas Moës, John Dolbow, and Ted Belytschko. A finite element method for crack growth without remeshing. International Journal for Numerical Methods in Engineering, 46(1):131–150, 1999.
  17. [17]Nicolas Moës and Ted Belytschko. Extended finite element method for cohesive crack growth. Engineering Fracture Mechanics, 69(7):813–833, 2002.DOI
  18. [18]Sukanta Chakraborty and Amit Shaw. A pseudo-spring based fracture model for sph simulation of impact dynamics. International Journal of Impact Engineering, 58:84–95, 2013.DOI
  19. [19]Md Rushdie Ibne Islam and Chong Peng. A total lagrangian sph method for modelling damage and failure in solids. International Journal of Mechanical Sciences, 157:498–511, 2019.arxiv.org/abs/1901.08342
  20. [20]Md Rushdie Ibne Islam and Amit Shaw. Pseudo-spring sph simulations on the perforation of metal targets with different damage models. Engineering Analysis with Boundary Elements, 111:55–77, 2020.DOI
  21. [21]Yingnan Wang, Ha H. Bui, Giang D. Nguyen, and PG Ranjith. A new sph-based continuum framework with an embedded fracture process zone for modelling rock fracture. International Journal of Solids and Structures, 159:40–57, 2019.DOI
  22. [22]Yingnan Wang, Hieu T. Tran, Giang D. Nguyen, Pathegama G. Ranjith, and Ha H. Bui. Simulation of mixed-mode fracture using sph particles with an embedded fracture process zone. International Journal for Numerical and Analytical Methods in Geomechanics, 44(10):1417–1445, 2020.DOI
  23. [23]Ha H. Bui and Giang D. Nguyen. Smoothed particle hydrodynamics (SPH) and its applications in geomechanics: From solid fracture to granular behaviour and multiphase flows in porous media. Computers and Geotechnics, 138:104315, 2021.DOI
  24. [24]Charlotte Kuhn and Ralf Müller. A phase field model for fracture. In PAMM: Proceedings in Applied Mathematics and Mechanics, volume 8, pages 10223–10224. Wiley Online Library, 2008.DOI
  25. [25]Christian Miehe, Martina Hofacker, and Fabian Welschinger. A phase field model for rate-independent crack propagation: Robust algorithmic implementation based on operator splits. Computer Methods in Applied Mechanics and Engineering, 199(45-48):2765–2778, 2010.DOI
  26. [26]Michael J. Borden, Clemens V. Verhoosel, Michael A. Scott, Thomas J.R. Hughes, and Chad M. Landis. A phase-field description of dynamic brittle fracture. Computer Methods in Applied Mechanics and Engineering, 217-220:77–95, 2012.DOI
  27. [27]Marreddy Ambati, Tymofiy Gerasimov, and Laura De Lorenzis. A review on phase-field models of brittle fracture and a new fast hybrid formulation. Computational Mechanics, 55(2):383–405, 2015.DOI
  28. [28]Jian-Ying Wu and Vinh Phu Nguyen. A length scale insensitive phase-field damage model for brittle fracture. Journal of the Mechanics and Physics of Solids, 119:20–42, 2018.DOI
  29. [29]Adrian Egger, Udit Pillai, Konstantinos Agathos, Emmanouil Kakouris, Eleni Chatzi, Ian A. Aschroft, and Savvas P. Triantafyllou. Discrete and phase field methods for linear elastic fracture mechanics: A comparative study and state-of-the-art review. Applied Sciences, 9(12), 2019.DOI
  30. [30]Jian-Ying Wu, Vinh Phu Nguyen, Chi Thanh Nguyen, Danas Sutula, Sina Sinaie, and Stéphane PA Bordas. Phase-field modeling of fracture. In Advances in Applied Mechanics, volume 53, pages 1–183. Elsevier, 2020.DOI
  31. [31]Lampros Svolos, Curt A Bronkhorst, and Haim Waisman. Thermal-conductivity degradation across cracks in coupled thermo-mechanical systems modeled by the phase-field fracture method. Journal of the Mechanics and Physics of Solids, 137:103861, 2020.DOI
  32. [32]Tushar Kanti Mandal, Vinh Phu Nguyen, Jian-Ying Wu, Chi Nguyen-Thanh, and Alban de Vaucorbeil. Fracture of thermo-elastic solids: Phase-field modeling and new results with an efficient monolithic solver. Computer Methods in Applied Mechanics and Engineering, 376:113648, 2021.DOI
  33. [33]Lampros Svolos, Hashem M Mourad, Curt A Bronkhorst, and Haim Waisman. Anisotropic thermal-conductivity degradation in the phase-field method accounting for crack directionality. Engineering Fracture Mechanics, 245:107554, 2021.DOI
  34. [34]Gilles A Francfort and J-J Marigo. Revisiting brittle fracture as an energy minimization problem. Journal of the Mechanics and Physics of Solids, 46(8):1319–1342, 1998.DOI
  35. [35]Blaise Bourdin, Gilles A Francfort, and Jean-Jacques Marigo. Numerical experiments in revisited brittle fracture. Journal of the Mechanics and Physics of Solids, 48(4):797–826, 2000.DOI
  36. [36]Alan Arnold Griffith. Vi. the phenomena of rupture and flow in solids. Philosophical transactions of the royal society of London. Series A, containing papers of a mathematical or physical character, 221(582-593):163–198, 1921.DOI
  37. [37]Michael J Borden, Thomas JR Hughes, Chad M Landis, and Clemens V Verhoosel. A higher-order phase-field model for brittle fracture: Formulation and analysis within the isogeometric analysis framework. Computer Methods in Applied Mechanics and Engineering, 273:100–118, 2014.DOI
  38. [38]Lin Chen, Bin Li, and René de Borst. Adaptive isogeometric analysis for phase-field modeling of anisotropic brittle fracture. International Journal for Numerical Methods in Engineering, 121(20):4630–4648, 2020.DOI
  39. [39]Khuong D Nguyen, Cuong-Le Thanh, H Nguyen-Xuan, and Magd Abdel-Wahab. A hybrid phase-field isogeometric analysis to crack propagation in porous functionally graded structures. Engineering with Computers, pages 1–21, 2021.DOI
  40. [40]Fatemeh Amiri, Daniel Millán, Marino Arroyo, Mohammad Silani, and Timon Rabczuk. Fourth order phase-field model for local max-ent approximants applied to crack propagation. Computer Methods in Applied Mechanics and Engineering, 312:254–275, 2016.
  41. [41]Emmanouil G. Kakouris and Savvas P. Triantafyllou. Phase-field material point method for brittle fracture. International Journal for Numerical Methods in Engineering, 112(12):1750–1776, 2017.DOI
  42. [42]Georgios Moutsanidis, David Kamensky, JS Chen, and Yuri Bazilevs. Hyperbolic phase field modeling of brittle fracture: Part II—immersed IGA–RKPM coupling for air-blast–structure interaction. Journal of the Mechanics and Physics of Solids, 121:114–132, 2018.DOI
  43. [43]Emmanouil G. Kakouris and Savvas P. Triantafyllou. Phase-field material point method for dynamic brittle fracture with isotropic and anisotropic surface energy. Computer Methods in Applied Mechanics and Engineering, 357:112503, 2019.DOI
  44. [44]Ngoc-Hien Nguyen, Vinh Phu Nguyen, Jian-Ying Wu, Thi-Hong-Hieu Le, Yan Ding, et al. Mesh-based and meshfree reduced order phase-field models for brittle fracture: One dimensional problems. Materials, 12(11):1858, 2019.DOI
  45. [45]Yulong Shao, Qinglin Duan, and Shasha Qiu. Adaptive consistent element-free galerkin method for phase-field model of brittle fracture. Computational Mechanics, 64(3):741–767, 2019.DOI
  46. [46]Weidong Li, Nhon Nguyen-Thanh, and Kun Zhou. Phase-field modeling of brittle fracture in a 3d polycrystalline material via an adaptive isogeometric-meshfree approach. International Journal for Numerical Methods in Engineering, 121(22):5042–5065, 2020.
  47. [47]Junchao Wu, Dongdong Wang, Zeng Lin, and Dongliang Qi. An efficient gradient smoothing meshfree formulation for the fourth-order phase field modeling of brittle fracture. Computational Particle Mechanics, 7(2):193–207, 2020.DOI
  48. [48]Yulong Shao, Qinglin Duan, and Shasha Qiu. Adaptive analysis for phase-field model of brittle fracture of functionally graded materials. Engineering Fracture Mechanics, 251:107783, 2021.DOI
  49. [49]Yury Krongauz and Ted Belytschko. Consistent pseudo-derivatives in meshless methods. Computer Methods in Applied Mechanics and Engineering, 146(3-4):371–386, 1997.DOI
  50. [50]Jean Paul Vila. On particle weighted methods and smooth particle hydrodynamics. Mathematical Models and Methods in Applied Sciences, 9(2):161–209, 1999.DOI
  51. [51]Ted Belytschko, Yong Guo, Wing Kam Liu, and Shao Ping Xiao. A unified stability analysis of meshless particle methods. International Journal for Numerical Methods in Engineering, 48(9):1359–1400, 2000.doi.org/10.1002/1097-0207(20000730)48:9<1359::aid-nme829>3.0.co;2-u
  52. [52]Joseph Peter Morris. An overview of the method of smoothed particle hydrodynamics. 1995.
  53. [53]Timon Rabczuk, Ted Belytschko, and Shaoping Xiao. Stable particle methods based on Lagrangian kernels. Computer Methods in Applied Mechanics and Engineering, 193(12-14):1035–1063, 2004.
  54. [54]J.J. Monaghan. Particle methods for hydrodynamics. Computer Physics Reports, 3(2):71–124, 1985.DOI
  55. [55]Jun Lin, Hakim Naceur, Daniel Coutellier, and Abdel Laksimi. Geometrically nonlinear analysis of two-dimensional structures using an improved smoothed particle hydrodynamics method. Engineering Computations, 32(3):779–805, jan 2015.DOI
  56. [56]Jiandong He, Nima Tofighi, Mehmet Yildiz, Juanmian Lei, and Afzal Suleman. A coupled wc-tl sph method for simulation of hydroelastic problems. International Journal of Computational Fluid Dynamics, 31(3):174–187, 2017.DOI
  57. [57]Joseph J. Monaghan and Robert A. Gingold. Shock simulation by the particle method sph. Journal of Computational Physics, 52(2):374–389, 1983.DOI
  58. [58]David Kamensky, Georgios Moutsanidis, and Yuri Bazilevs. Hyperbolic phase field modeling of brittle fracture: Part I—Theory and simulations. Journal of the Mechanics and Physics of Solids, 121:81–98, 2018.DOI
  59. [59]Andreas Peer, Christoph Gissler, Stefan Band, and Matthias Teschner. An implicit sph formulation for incompressible linearly elastic solids. In Computer Graphics Forum, volume 37, pages 135–148. Wiley Online Library, 2018.
  60. [60]Gerhard A Holzapfel. Nonlinear solid mechanics: a continuum approach for engineering science. Meccanica, 37(4):489–490, 2002.DOI
  61. [61]Michael J. Borden, Thomas J.R. Hughes, Chad M. Landis, Amin Anvari, and Isaac J. Lee. A phase-field formulation for fracture in ductile materials: Finite deformation balance law derivation, plastic degradation, and stress triaxiality effects. Computer Methods in Applied Mechanics and Engineering, 312:130–166, 2016.DOI
  62. [62]Joseph J. Monaghan. Smoothed particle hydrodynamics. Annual Review of Astronomy and Astrophysics, 30(1):543–574, 1992.
  63. [63]Leigh Brookshaw. A method of calculating radiative heat diffusion in particle simulations. Publications of the Astronomical Society of Australia, 6(2):207–210, January 1985.DOI
  64. [64]Tulio N. Bittencourt, Paul A. Wawrzynек, Anthony R. Ingraffea, and J.L. Sousa. Quasi-automatic simulation of crack propagation for 2d lefm problems. Engineering Fracture Mechanics, 55(2):321–334, 1996.DOI
  65. [65]M. Naqib Rahimi, Adnan Kefal, Mehmet Yildiz, and Erkan Oterkus. An ordinary state-based peridynamic model for toughness enhancement of brittle materials through drilling stop-holes. International Journal of Mechanical Sciences, 182(May), 2020.DOI
  66. [66]M. Naqib Rahimi. Peridynamic modelling of internal features and interfaces for material toughening. Master’s thesis, Sabanci University, dec 2020.
  67. [67]J.F. Kalthoff and S. Winkler. Failure mode transition at high rates of shear loading. DGM Informationsgesellschaft mbH, Impact Loading and Dynamic Behavior of Materials, 1:185–195, 1988.
  68. [68]Adem Candaş, Erkan Oterkus, and Cevat Erdem İmrak. Dynamic Crack Propagation and Its Interaction With Micro-Cracks in an Impact Problem. Journal of Engineering Materials and Technology, 143(1), 08 2020.DOI
  69. [69]Yousef Navidtehrani, Covadonga Betegón, and Emilio Martínez-Pañeda. A unified abaqus implementation of the phase field fracture method using only a user material subroutine. Materials, 14(8):1–19, 2021.
  70. [70]M. Naqib Rahimi, Adnan Kefal, and Mehmet Yildiz. An improved ordinary-state based peridynamic formulation for modeling FGMs with sharp interface transitions. International Journal of Mechanical Sciences, 197(December 2020):106322, 2021.DOI
  71. [71]Florin Bobaru and GuanFeng Zhang. Why do cracks branch? A peridynamic investigation of dynamic brittle fracture. International Journal of Fracture, 196(1-2):59–98, 2015.DOI
  72. [72]Rui Yan, Yong qiang Bi, and Wei Jiang. Simulation of contact interface between elastic solids using smoothed particle hydrodynamics. Computational Particle Mechanics, 9(1):167–177, 2022.DOI

Paper details

vlnc:W_4qtfurs3
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 6

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents