Pan - 2018 - A Review of Lattice Type Model in Fracture Mechanics

Engineering Fracture Mechanics 190 (2018) 382–409
Contents lists available at ScienceDirect
Engineering Fracture Mechanics

journal homepage: www.elsevier.com/locate/engfracmech
A review of lattice type model in fracture mechanics: theory,

applications, and perspectives
Zichao Pan, Rujin Ma ⇑, Dalei Wang, Airong Chen
Department of Bridge Engineering, Tongji University, 200092 Shanghai, China
a r t i c l e i n f o a b s t r a c t
Article history: The lattice model is a discrete model that is typically used to simulate the fracture process
Received 14 August 2017 of brittle materials. This review summarizes the main achievements during the develop-
Received in revised form 12 December 2017 ment history of the state-of-the-art lattice model during the past 80 years from the theory
Accepted 22 December 2017
and application viewpoints. It is found that the classical lattice spring model (LSM) can
Available online 28 December 2017
only simulate a fixed Poisson’s ratio. This problem has been partially or fully solved in
developed versions of novel lattice models by using (1) extra nodal DOFs, (2) extra shear
Keywords:
springs, (3) extra nonlocal energy parameters, and (4) higher dimensional normal springs.
Review
Lattice model
The lattice model has already been successively applied in simulations of fracture pro-
Fracture process cesses of different materials and under different loads. However, the applications in frac-
Beam ture analyses of metals, dynamic load-induced fracture, and real large-scale structures,
Spring are not sufficient. Innovative areas of future research in reference to the use of the lattice
model in engineering practice include constitutive relations, failure criteria, anisotropic
material modeling and efficient computing techniques to expand its use and range of
applications.
Ó 2017 Elsevier Ltd. All rights reserved.
1. Introduction
Brittle materials, such as mortar and concrete, are vulnerable to cracking owing to their low tensile strength. To simulate
the fracture process of brittle materials, several discrete numerical models have been proposed and developed [1–5]. One of
them is the so-called ‘‘lattice model” which can be traced back to the 1940s. To solve the elasticity problem, Hrennikoff [6]
proposed a framework methodology that may be considered the prototype of the lattice model. In the 1970s and 1980s, the
lattice model was introduced into the field of theoretical physics to study the fracture process of disordered media [7–12]. In
the 1990s, the lattice model was further and considerably developed by various researchers [13–16]. With the rapid devel-
opment of computers and computational techniques at the beginning of this century, the previous 2D lattice models were
extended to 3D [17–19]. In addition, several new features were introduced. For example, Hou [20] proposed an improved
Abbreviations: FEM, finite element method; LSM, lattice spring model; LBM, lattice beam model; RPM, random particle model; CSL, confinement-shear
lattice model; LCM, lattice-cell model; DLSM, distinct lattice spring model; VCPM, volume-compensated particle model; DPM, discrete particle model; MD,
molecular dynamics; PD, PeriDynamics; DOF, degree of freedom; GB, generalized beam; ITZ, interfacial transition zone; LDPM, lattice discrete particle
model; DEM, discrete element method; AMR, adaptive mesh refinement; FA, fly ash; BFS, blast furnace slag; SF, silica fumes; HPC, high performance
concrete; UHPC, ultra high performance concrete; RC, reinforced concrete; FRC, fiber-reinforced concrete; SFRC, steel fiber-reinforced concrete; CUDA,
compute unified device architecture; GPU, graphics processing unit.
⇑ Corresponding author.
E-mail addresses: z.pan@tongji.edu.cn (Z. Pan), rjma@tongji.edu.cn (R. Ma), wangdalei@tongji.edu.cn (D. Wang), a.chen@tongji.edu.cn (A. Chen).
https://doi.org/10.1016/j.engfracmech.2017.12.037
0013-7944/Ó 2017 Elsevier Ltd. All rights reserved.
Z. Pan et al. / Engineering Fracture Mechanics 190 (2018) 382–409 383
Nomenclature
r nominal stress of element in failure criterion

½r maximum allowed nominal stress of element in failure criterion
rt tensile stress of element
aN , aM normal force and bending influence factors to modify tensile stress in element
F normal force of element
Ae cross-sectional area of element
We cross-sectional flexural modulus of element
Mi , Mj nodal bending moment of element
ft tensile strength
fc compressive strength
rn normal stress
s shear stress
c cohesion of material in Mohr-Coulomb criterion
/ friction angle of material in Mohr-Coulomb criterion
e nominal strain of element in failure criterion
½e maximum allowed nominal strain of element in failure criterion
et tensile strain of element
u difference of nodal displacements along normal direction
u difference of nodal bending angles
aS bending angle factor to modify tensile strain in element
he height of rectangular cross-section of element
de diameter of circular cross-section of element
E elastic modulus of material
v global poisson’s ratio
q density
n damage degree of element
kn , ks stiffnesses of normal and shear springs
ka ; kb ; kc stiffness of 4D spring in 4D LSM
k4D 4D stiffness ratio
lattice model to consider the effects of large strains. Guo et al. [21] used the lattice model to simulate the fatigue problem of
concrete. Liu et al. [22] incorporated fiber and rebar in the lattice model. Transport problems of moisture and chloride ions
have also been studied by the lattice model [23–26]. To overcome the fixed Poisson’s ratio issue in classical lattice models,
and correctly simulate the behavior of materials with an arbitrary Poisson’s ratio, numerous novel models have been
developed [27–32], especially after 2000. A timeline listing the important milestones during the historical evolution of
the lattice model is shown in Fig. 1. The full names of the abbreviations in the figure can be found in the abbreviation list
of this paper.
The general concept of the classical lattice model requires the discretization of a continuum into several connected
spring (truss) or beam elements, and conduction of a series of linear analyses on the discrete model. The fracture process
is simulated by removing the critical elements from the model, which are determined based on a failure criterion in a step-
by-step manner. Generally, the lattice model has two major advantages: (1) lattice models are based on a discontinuous
formulation, which avoids singularity-related issues in continuum-based numerical simulation methods, such as the finite
element method (FEM), and (2) the material heterogeneity can be easily implemented. Owing to the first advantage, the
lattice model is usually adopted to simulate the fracture process, especially for brittle materials, e.g., in uniaxial tensile
experiments of concrete [17,33], and rebar corrosion in concrete [34,35]. Owing to the second advantage, the lattice model
is especially preferred in research applications at the meso- or microscales where the effect of material heterogeneity
should be considered.
This study mainly reviews the evolution of the lattice model from 1990 onwards from the theory and application view-
points. The presented work is divided into three parts. In the first part, the general theoretical background of classical
lattice models and recent developments are briefly introduced. More detailed reviews on the general aspects of classical
lattice models can be found in other Refs. [36–39]. In the second part, the applications of the lattice model are reviewed
in accordance to three specific aspects: benchmarks of classic tests, types of materials, and types of loads. Finally, some
future challenges and perspectives of the lattice model are presented. Some possible solutions to the challenges are also
proposed. Through this review, it is expected that researchers can gain an understanding about the history and the
state-of-the-art status of the lattice model, and the possible development concerns and applications, or the hurdles for
its future applications.
384 Z. Pan et al. / Engineering Fracture Mechanics 190 (2018) 382–409
Fig. 1. Timeline of major lattice model developments including a listing of major milestones.
2. Theory
2.1. Definition of terms
The term ‘‘lattice” in mathematics usually refers to a group of points whose positions follow a predefined pattern. Based
on the pattern, a network that represents the connections of points can be obtained, as shown in Fig. 2. The lattice model can
be perceived as a model with a group of connected elements in a network. It should be pointed out that models based on the
rigid-body spring network (RBSN) are not included herein, even though they are referred to as lattice models in publications
[40–42], as these are based on different theoretical backgrounds.
During the history of the lattice model, several important concepts, e.g., network shape, mesh size, failed element,
and critical element, have been proposed and studied. However, these concepts may have different names or meanings
in different publications. To avoid possible misunderstandings, the names and definitions of these concepts are listed in
Table 1.
Fig. 2. Examples of regular lattice networks.

Table 1
Concepts and definitions of lattice model used in this study.
Concept Definition
Network shape Shape of cell in network
Mesh size Element length
Element type Spring (truss) or beam
Failed element Element that satisfies a failure criterion
Critical element Element that is removed from model
2.2. General philosophy and framework
The general philosophy of the lattice model is the discretization of a complex global problem into a series of simple local
problems. Thus, the first principle and the major advantage of the lattice model is the simplicity at a local level, which
requires that all issues of the local elements, such as the constitutive relation and the failure criterion, should be as simple
as possible. Based on this philosophy, the main framework of the lattice model is illustrated in Fig. 3.
2.2.1. Discretization
As shown in Fig. 3, the first step in the framework of the lattice model is to discretize the continuum. Generally, two
approaches can be adopted to conduct the discretization. The first one is to directly use the simulated or real material struc-
ture at the meso- or microscale to construct the network. In random particle model (RPM) [1] for example, each particle
(aggregate) in the matrix (cement paste) was considered as a single node in the lattice network. The network is then formed
using the Delaunay triangulation, based on the positions of the particles. Apparently, the mesh size in this approach is
strongly dependent on the particle spacing, and is fixed when the particle packing procedure is finished [1,27,43–45]. In
the second approach, a user-defined network is first prepared, which is not necessarily related to the material structure.
The continuum is then mapped to the network to form elements. Theoretically, the mesh size in the second approach can
be freely chosen. However, in a meso- or microscopic analysis where the material heterogeneity implementation is needed,
the mesh size is usually determined based on the size of the smallest particle in the structure of the material.
The lattice network used in the discretization can be either regular or irregular. As the crack in the lattice model can only
propagate along the orientations of the elements, two networks will be generated: (1) a regular network will result in accor-
dance to the dependence of the crack pattern on the network and (2) an over-regular crack pattern that is not realistic. The
random lattice model with an irregular network can eliminate this unfavorable dependence, but it also introduces extra dis-
orders to the lattice model at the same time [46]. In addition, an irregular lattice network can also result when nonuniform
stress conditions are imposed in a homogenous model under a uniform applied strain, which is unrealistic [14,47]. This prob-
lem, however, can be solved using the Voronoi scaling method [41].
2.2.2. Element type

In the prototype lattice model, normal springs, which only allow axial deformations, were used to describe the behavior of
the elements. However, it was found that the Poisson’s ratio of the prototype and other similar lattice models that are only
based on normal springs is always fixed in the case of a regular network [48–52]. To solve this problem, two approaches have
been previously adopted. The first approach requires the inclusion of extra shear springs into the lattice model, as shown in
Fig. 4, and this was further developed and referred to in the literature as the lattice spring model (LSM) [31,53,54]. The sec-
ond approach is to include more nodal DOFs, e.g., shear deformation and rotation into the lattice model. Owing to the extra
nodal DOFs, normal springs are effectively replaced by beam elements. Thus, the lattice model developed by the second
approach may be termed the lattice beam model (LBM) [9,15,17,33,55]. Owing to prior pioneering work [13,15,56], the
Fig. 3. Main framework of lattice model.

Fig. 4. Types of elements used in lattice model.
LBM was preferred over the LSM in the 1990s, and it is still extensively used today [19,34]. The LSM, however, has received
more attention recently and has been developed further [27,32,57,58] owing to its intrinsic simplicity compared to the
prototype.
2.2.3. Constitutive relation

The constitutive relation of elements describes the material’s behavior at the local level. The philosophy of lattice models
requires a simple constitutive relation. Thus, the perfect elastic–linear relation was extensively used in classical lattice mod-
els [3,14,15,59], and is still used nowadays [18,19]. To simulate the global tension softening of materials, multilinear rela-
tions, including local softening, was also adopted in publications [60–62], but this was not as extensively used as the
linear elastic relation. The reasons are because (1) the multilinear relation violates the principle of local simplicity in the lat-
tice model, and (2) the negative slope, which represents a negative tangent stiffness, may lead to numerical instability and
solution divergence. More details on the constitutive relations used in lattice models can be found in Section 3.2.
2.2.4. Failure criterion

To use the lattice model to simulate the fracture process of a material, a failure criterion should be defined to determine
the critical elements which will be removed from the model. Most of the failure criteria used thus far in lattice models have
been categorized into two types: (1) critical stress (force) criterion [3,17,22], and (2) critical strain (displacement) criterion
[33,62–67]. The general idea of these failure criteria and some specific and extensively used forms of these criteria are briefly
introduced below.
In the critical stress criterion, the failed element is defined as the element whose nominal stress exceeds a threshold,
which is usually the tensile or compressive strength.
r P ½r ð1Þ
where r is the nominal stress of the element, and ½r is the maximum allowed r. Dependent on the type of the nominal
stress, the general failure criterion in Eq. (1) can attain different specific forms. For example, in the maximum tensile stress
criterion, which is extensively used in LBMs, the nominal stress is replaced by the tensile stress, which is calculated by
F ðjMi j; jM j jÞmax
rt ¼ aN þ aM ð2Þ
Ae We
where F is the normal tensile force in the element, Ae and W e are the cross-sectional area and flexural modulus of the ele-
ment, respectively, M i and M j are the bending moments at two nodes of the element, and aN and aM are the normal force
influence factor and the bending influence factor, respectively. Typically, aN is usually defined to be equal to 1.0, while aM
needs to be determined based on experimental results, yet its elicited values differ in different publications. The commonly
used values for aM are 0.0 [14,15,17,18,68], 0.005 [3,16,17,55,61,69,70], 0.05 [19,34,71], 0.33 [3], and 1.0 [72,73].
Another example is the Mohr–Coulomb criterion [74], which was originally proposed in the field of soil mechanics. The
purpose of introducing the Mohr–Coulomb criterion into the lattice model is to simulate other types of failure, e.g., compres-
sive and shear failures. The Mohr–Coulomb criterion used in the lattice model was usually modified by a tension cut-off
[22,68,75,76]. Thus, a closed trapezoid can be defined as
8
< rn ¼ f t
>
jsj ¼ c rn tan/ ð3Þ
>
:
rn ¼ f c
where c and / are the cohesion and friction angle of the material, respectively, and rn and s are the normal and shear stres-
ses of the element, respectively. If the stress condition of an element ðrn ; sÞ lies outside the trapezoid, it is considered as a
failed element.
The critical strain criterion is similar to the critical stress criterion,
e P ½e ð4Þ
where e is the nominal strain, and ½e is the maximum allowed e. As a specific example in the LBM, e can be replaced by the
tensile strain, which is simply calculated based on the nodal displacements.
Du þ Du s aS
et ¼ ð5Þ
le
where u is the difference of the nodal displacements along the axial direction, u is the difference of the nodal bending angles
s ¼ he =2 [33] and s ¼ de =2 [64] for elements with a rectangular and circular cross-section where he and de are the height and
diameter, respectively. Similar to Eq. (2), a parameter aS is also introduced to consider the contribution of the bending angle
to the normal tensile strain. The values of aS are found to be equal to 0.005 in Karihaloo, Shao et al. [33], Ince, Arslan et al.
[62], and 1.0 in Mungule and Raghuprasad [64].
Both the critical stress and strain failure criteria have the disadvantage that the simulation results (especially the P-D
curve) are dependent on the mesh size of the lattice network [15,61,64]. To solve this problem, failure criteria based on frac-
ture energy have been studied recently [77,78]. The research, however, was not sufficient for extensive usage of these
criteria.
2.2.5. Element removal

The crack pattern in the lattice model is simulated by a sequence of element removals from the model. Thus, besides the
constitutive relation and failure criterion, the element removal method is also an important issue in the lattice model. Ele-
ment removal methods can be categorized into different types that are reviewed below.
Depending on whether the element stiffness can be regained or not, element removal methods can be categorized into
permanent removal and temporary removal methods. The permanent removal method is extensively used in classic lattice
models [3]. Effectively, once an element fails, its stiffness is lost forever, regardless of further deformations of the model.
However, in the temporary removal method, which is mainly used in recent lattice models [22,30], once an element fails,
its stiffness is only lost temporarily. When a certain criterion is satisfied, e.g., the nodal distance is smaller than a threshold
[79], the lost stiffness can be partially recovered [22].
Depending on the number of elements removed in each linear analysis, the element removal methods can be categorized
into ‘‘event-by-event” [80–82] (or ‘‘event-driven” [22]), and load-stepping methods. In the ‘‘event-by-event” method, the
load is increased gradually, and linear analyses are performed until the first failed element appears (which is called an
‘‘event”). Another linear analysis with an unchanged load is then conducted on the updated model to check whether another
element will fail. If no more elements fail, the entire fracture analysis is restarted again until the complete failure of the
model. Therefore, in the ‘‘event-by-event” method, only one element is removed in each linear analysis. In the load-
stepping method, the load is increased in a step-by-step manner. In each step, all failed elements (if they exist) will be
removed together from the model. The analysis then continues to the next step. The ‘‘event-by-event” and load-stepping
methods have their own advantages and disadvantages. The ‘‘event-by-event” method can produce a very realistic crack pat-
tern, but requires considerable computational work. The load-stepping method can significantly speed up the fracture anal-
ysis. However, a slow load-stepping process is required to obtain a realistic crack pattern, which means that the number of
removed elements in each step cannot be too large.
2.3. Recent developments
During 1990s, the LBM was preferred over the LSM owing to its flexibility in simulating complex fracture processes. After
the pioneering work by Schlangen [3], however, only a few minor improvements were implemented in the LBM, e.g., an
adaptive mesh scheme [83], extension to 3D [17], and a generalized beam (GB) element [22], using the spectral element
method (SEM) for modal analysis [84]. Compared with the LBM, the LSM has received increased attention and has undergone
extensive development, especially after 2000.
By discretizing a continuum into several particles, most of these newly developed models can allow the interaction
between pairs of particles, or the motion of particles. Thus, strictly speaking, although these models are still codified as LSMs
in publications, they should be more accurately considered as hybrid models combined with classical LSMs, and should be
based on the philosophy of other particle-based numerical models, such as the discrete particle model (DPM) [5], molecular
dynamics (MD) [85] and PeriDynamic (PD) [86–88] model, which may be viewed as a continuum version of MD. For exam-
ple, the confinement–shear lattice (CSL) model [27,43] discretizes a continuum into a lattice network of aggregates for
cementitious materials, which can be analogous to metal atoms, protein molecules, etc., in MD. Thus, the networks in CSL
and MD have physical meanings, e.g., they denote the real structures of the materials. Another example is the nonlocal
LSM [89,90] where the effects of all the surrounding particles in a finite zone are considered to determine the motion of
a single particle. This approach is similar to the concept of ‘‘horizon” in PD [86]. Despite the above similarities, the spring
stiffness is still the main concern in these newly developed models. Thus, it is reasonable to name these models as LSMs,
and some of the most representative ones are briefly reviewed in this section.
2.3.1. Confinement–shear lattice (CSL) model

The CSL was mainly proposed by Cusatis et al. [27] based on some previous research [1,57]. Different from a user-defined
network, the network of the CSL is directly determined by the simulated material mesostructure, i.e., aggregate packing. The
position of each lattice node coincides with the center of one aggregate. As shown in Fig. 5, each node has five DOFs: three for
displacements along the x, y, and z axes, and two for rotations around the y and z axes (rotation around the x axis is a torsion
which is not included in the CSL). The shear stress of the element is calculated by the relative shear deformations in the mid-
dle of the contact zone, whose width equals the particle spacing. The term ‘‘confinement” in the CSL refers to the transverse
confining stresses normal to the element between two large particles, which are induced by inclined elements connecting
the small aggregates (see [27] for more details). Shear interactions between particles are calculated based on the nodal rota-
tions and transversal displacements. In this sense, the philosophy of the CSL is somewhat similar to that of the LBM, whereby
more DOFs are included on the lattice nodes to increase the flexibility of the entire model. The difference is that the bending
displacements of beams in LBM are replaced by the rotations of particles in the CSL.
Owing to the allowed shear transfer between particles, the CSL can accurately simulate the correct Poisson’s ratio of the
material. Based on several benchmarks, it has been verified that the CSL can simulate the fracture process under complex
stress conditions, e.g., nonlinear uniaxial, biaxial, triaxial compressions, which cannot be properly simulated by classical lat-
tice models [27,43,44]. The success of the CSL indicates that shear springs can enable LSMs to simulate an arbitrary Poisson’s
ratio by modifying model parameters, such as the spring stiffness. However, it should be mentioned that by introducing
shear springs into classical LSMs, the rotational invariance may not be preserved [47,91,92]. The main reason, according
to Jagota and Scherer [93], is that shear springs cannot distinguish the differences in the tangential velocities (or displace-
ment) of two particles owing to a common rotation or shear, as shown in Fig. 6. Thus, a global, rigid body rotation may incor-
rectly result in an additional strain energy inside the shear springs, which is unrealistic.
The above problem of rotational variance in LSMs can be solved using two approaches. First, beam-type interactions
instead of shear springs can be introduced [39] that explain the inherent advantage of LBM on the issue of rotational invari-
ance. Second, a modification can be made to calculate the shear displacement, as proposed in the distinct LSM model (DLSM)
[30,94,95].
Based on CSL, Cusatis et al. [96] further developed the lattice discrete particle model (LDPM) [5,97–100], which can be
perceived as a synthesis of the CSL and the DPM methods. In addition to the similar characteristics to CSL, the LDPM has
inherited from DPM the long-range contact capability, which is one of major characteristics in the discrete element method
(DEM). The theoretical formulation of the LDPM is similar to that of the CSL. Correspondingly, herein, it will not be reviewed
again.
2.3.2. DLSM
DLSM was initially proposed by Zhao et al. [30] mainly to simulate the dynamic fracture process in geomechanics. In
DLSM, the continuous model is discretized into several particles which can be placed randomly or in accordance to a regular
pattern. To overcome the problem of a fixed Poisson’s ratio, normal and multibody shear-type springs are included for each
pair of particles. The spring is formed if the distance between the two particles is smaller than a given threshold, and
removed if the distance becomes larger than the threshold. Thus, the springs of the entire model should be updated based
on the calculated particle motion in each analysis step, which forms an analysis cycle, as shown in Fig. 7. Benchmarks
showed that DLSM can simulate a wide range of Poisson’s ratio values owing to the presence of multibody shear-type
springs. The major characteristics of DLSM are: (1) the spring stiffness recovery mechanism, (2) the least-squares method
for obtaining the local strain of each particle, and (3) the large-scale parallel computation.
In classical LBMs [3,13–15,56] and in some newly developed LSMs [31,90], the critical element is removed from the model
permanently, which means that the stiffness of the critical element will never recover. By contrast, the critical element in
DLSM is only removed temporarily. Once the distance between particles is smaller than a threshold owing to additional
motions of particles, the critical element with a partially recovered stiffness will be replaced again into the model. The
Fig. 5. Schematic of theoretical basis of CSL based on [27].

Fig. 6. Schematic showing incapability of shear spring to distinguish differences in tangential displacements of two particles owing to rigid rotation or
shear.
Fig. 7. DLSM analysis procedure [30].
flexibility of DLSM is increased by this stiffness recovery mechanism which also enables the simulations of complex failure
processes, especially caused by dynamic impacts.
In mathematics, the least-squares method is a general approach in regression analysis used to obtain the approximate
solution of an over-determined system. One of the basic requirements of the least-squares method is that more equations
than unknowns should exist. Thus, to use the least-squares method, it is assumed in DLSM that the local strain of one particle
is dependent on the displacement of the individual particle under consideration, and on the displacements of other particles
that have intact bonds with it, in a finite zone termed as the ‘‘cloud” [30]. The use of the least-squares method in association
with the proposed local strain-based method makes the model rotationally invariant, which eliminates the variance of the
calculated strain in the spring with respect to the rigid body rotation of the model, which is unrealistic.
In the analysis cycle shown in Fig. 7, the particle motion in the current step is solved by Newton’s 2nd law based on the
unbalanced particle force from the last step, which is only a local problem. Thus, in comparison to the classical lattice mod-
els, there is no need in DLSM to form the global stiffness matrix. Therefore, DLSM is especially suitable for large-scale parallel
computation on multicore computers or clusters [101,102]. The parallel computation enables the application of DLSM in
large-scale engineering problems, where a considerable number of particles has to be used to obtain a satisfactory simula-
tion outcome with an acceptable accuracy.
Thus far, DLSM has been already successfully applied in structural dynamic failure under external impacts [30], wave
propagation across joint rock masses [103], rock failure under dynamic loads [95], large deformation analysis [94], hetero-
geneous rock breakage [104], as well as in mechanism studies [79].
2.3.3. Nonlocal LSM

The nonlocal LSM developed by Chen et al. [31] is based on the volume-compensated particle model (VCPM) [32], which
can be seen as a special case of the nonlocal LSM. Regarding energy-based concepts, the classical LSM (also LBM) only con-
siders the strain energy in the springs, while the strain energy in the network cell surrounded by the springs is neglected.
This is one of the reasons for the fixed Poisson’s ratio of classical LSMs. Using a different approach, Grassl et al. [29] included
both the strain energy stored in the springs and in surrounded cells into the total energy of the lattice model, and let it
become equal to the strain energy in the continuum counterpart. With the so-called lattice–cell Model (LCM), it was found
that the additional strain energy can effectively remove the limitation of the Poisson’s ratio of the lattice model. VCPM is
mainly based on the idea of LCM in [29], but the model is extended to consider plastic deformations [32].
In VCPM, each particle is surrounded by several other particles referred to as ‘‘neighboring particles” depending on the
packing scheme. Based on the distance, neighboring particles can attain different levels, as shown in Fig. 8. In VCPM, only
the interactions from the 1st nearest neighboring particles are considered. When the interactions from the neighboring par-
ticles at higher levels are also considered, VCPM will be extended to the nonlocal LSM. In some sense, the idea of the nonlocal
LSM is similar to that of PD, which assumes that the behavior of a particle is affected by other particles in a finite zone
referred to as ‘‘horizon” in PD. The interactions between particles in PD only, however, have a mathematical meaning, while
the spring stiffness in the nonlocal LSM is closely related to the strain energy stored in the unit cell surrounded by springs,
which has a clear physical definition. The accuracy of the simulation result, and the amount of the computational work in the
nonlocal LSM are determined by the number of levels of neighboring particles that are considered.
The fracture process in VCPM and nonlocal LSM is simulated using an approach that is similar to classical lattice models.
In each analysis step, the critical springs are detected and removed from the model, and another analysis is then conducted
in the next step. This procedure continues until a crack pattern is obtained, or until the simulation stops. To detect the critical
springs, both the critical stress [31] and strain [90] criteria have been used in the 2D and 3D nonlocal LSMs, respectively.
However, it is not explained in prior publications why different criteria had been adopted for different dimensional models.
2.3.4. 4D LSM
A common characteristic of the new models is the noncentral or nonlocal interactions between particles. For example,
additional rotation DOFs are included on lattice nodes in CSL [27]. Extra shear springs are placed in DLSM [30]. Additional
energy parameters from the cell surrounded by springs are considered in LCM [29], VCPM [32], and nonlocal LSM [31]. As
pointed out in [28], the introduction of the aforementioned noncentral or nonlocal effects give rise to additional complex
mathematic calculations and incremental integration that might result in additional sources of cumulative errors. These
errors will invariably disrupt the simplicity and robustness of lattice models.
Therefore, Zhao [28] proposed the so-called ‘‘four-dimensional LSM” (4D LSM). In this model, only the axial stress of the
spring is considered, which preserves the pure central and local interactions between particles, as well as the simplicity of
the model. To overcome the fixed Poisson’s ratio, the original model is linked to its parallel mapped version in the fourth
dimension, which is plotted as a hypercube in Fig. 9. Extra 4D springs, e.g., A-A’, B-B’, A-B’, B-A’, are placed to link the original
and parallel models. The stiffness of the 4D springs can be obtained based on the spring stiffness of the original model,
ka ¼ kb ¼ kc ¼ k4D k ð6Þ
where k4D is the so-called ‘‘4D stiffness ratio” which is dependent on the Poisson’s ratio of the material, and ka ; kb ; kc are the
stiffness values of the three types of 4D springs, shown in Fig. 9. In the above equation it is assumed that ka ¼ kb ¼ kc , but this
equation is not necessarily required in 4D LSM.
The novelty of 4D LSM is mainly attributed to the introduction of the concept of the extra dimension from modern physics
into classical solid mechanics. The robustness and flexibility of LSM is ensured by the additional 4D springs, which can
remove the limitations of the fixed Poisson’s ratio in classical LSMs. Some benchmarks have already shown that 4D LSM
is capable of solving elastic problems with various Poisson’s ratios [28]. However, as one of the latest developed LSMs, 4D
LSM is still in the early stage of theoretical studies. Before any application in practical engineering, more research is needed
to test the feasibility of the model in complex mechanics problems.
2.4. Summary
In summary, the most important lattice models that have historically emerged, as well as their major novelties and con-
tributions, as reviewed herein, are listed in Table 2.
3. Application
The prototype lattice model was initially proposed mainly to solve the elasticity problems [6]. It is found, however, that
the lattice model can be efficiently used to simulate the fracture processes in different materials under various loadings. The
applications of lattice models in other fields, such as transport analysis [23–26] and model analysis [84], were also reported.
This research, however, will not be discussed here as it is irrelevant to the subject matter of this study. In summary, the
application of the lattice model in fracture analysis can be reviewed in accordance to three aspects: benchmarks of classic
tests, types of materials, and types of loadings.
3.1. Benchmarks of classic tests
In the history of engineering mechanics, several classic tests were proposed to study the fracture process of materials
under typical stress conditions, e.g., uniaxial tensile, shear, bending, and combined stress conditions, such as tensile and
shear in the Nooru–Mohamed test [105]. These classic tests were further used in numerical simulation research studies
to validate the feasibility and efficiency of proposed models and methods. Besides the validation, these benchmarks of classic
tests are also helpful in conducting the sensitivity analysis and optimization of the input parameters used in numerical
Fig. 8. Schematics of different orders of neighboring particles in 2D nonlocal LSM [31].
Fig. 9. A 4D spatial projection rendering of 4D LSM and different types of 4D springs [28].
Table 2
Summary of the most important lattice models.
Name Major novelties and contributions References

LBM Bending deformation of element [3]
RPM, CSL Material mesostructure-based network and DOF of nodal rotations [1,27,43]
LBM with GB Large generalized beam element [22,68]
LCM Strain energy item from cell surrounded by spring elements [29]
VCPM Extension of LCM to consider plastic deformation [32]
Nonlocal LSM Extension of VCPM to consider effects from multilevel of neighboring particles [31,90]
DLSM Multibody shear-type spring [30]
4D LSM Four-dimensional normal springs [28]
simulations by comparing the simulated fracture process with observations from tests. The major benchmarks adopted by
researchers in the historical evolution of the lattice model are listed in Table 3. Some of the most extensively used bench-
marks are briefly reviewed below.
3.1.1. Uniaxial tensile test

The uniaxial tensile test may be the simplest one to simulate the mode I (opening mode) fracture process. To ensure that
the fracture process starts from predefined positions, notches are often artificially included in specimens and numerical
models as initial defects. The notches can be placed on one side [3,107], or two sides, with symmetric positions [3,43]. In
both tests with single and double notches, cracks will initiate at the tips of the notches and propagate to the other side.
For the test with no notches [22,76], the position of crack initiation is randomly determined by the structure of the material
and lattice network used in the simulation. In the absence of the initial defects, microcracks will first appear in the most vul-
nerable part of the model, e.g., ITZ in cementitious materials, and propagate to form a connected crack pattern through the
surface.
Schlangen [3] validated the developed LBM using the uniaxial tensile test with single and double notches. The effects of
input parameters, such as the mesh size and ITZ tensile strength (bond strength), were also studied. It was found that the
number of microcracks outside the main crack decreased as the bond strength decreased, and the crack pattern became
straighter if higher bond strengths were used in simulations. Based on these comparisons, optimized bond strengths for
fracture analysis of cementitious materials can be proposed.
392
Table 3
Major benchmarks of classic tests based on the lattice model.
Test name Model Representative simulated crack pattern Type of lattice References
model
Uniaxial tensile test LBM [3,16,17,19,22,29,43,45,71,76,106–
(no notch and double notches) LCM 108]
LPM
Uniaxial tensile test with single LBM [3,31,57,83,109]

notch DLSM
Nonlocal LSM
Z. Pan et al. / Engineering Fracture Mechanics 190 (2018) 382–409

Three-point bending test LBM [22,32,45,102]
LPM
DLSM
VCPM
Shear cracking test Nonlocal LSM [31]
Four-point shear test LBM [3,13,44]

(single and double notches) CSL
Uniaxial compression LBM [22,43,45,68,79,107,110]

DLSM
Nooru–Mohamed test LBM [3,14,15,44,56,75]

LSM
CSL
Table 3 (continued)
Test name Model Representative simulated crack pattern Type of lattice References
model
Pull-out test LBM [13]
Diagonally loaded square plate test Nonlocal LSM [90]
Z. Pan et al. / Engineering Fracture Mechanics 190 (2018) 382–409

393
Cusatis et al. [43] used the uniaxial tensile test to validate their proposed CSL. As an additional step, they considered the
structural response under cyclic loads. It was found that the irreversible strains during cyclic loads were different in numer-
ical simulations and experiments. The experiment showed an increasingly irreversible strain after each cycle, while the
strain in numerical simulations was shown to be fully recovered if the external force was unloaded to zero, i.e., the irre-
versible strain was zero, which was unreasonable. The reason was not explained in [43], but it can be attributed to the fact
that the plastic behavior in the constitutive relation of lattice elements was ignored.
Liu et al. [22] tested their proposed LBM with GB to decrease the number of elements in the lattice model to accelerate the
simulation. The benchmark showed that although the new LBM can really improve the simulation efficiency and create a
reasonable crack pattern, the structure was too brittle, i.e., the P–D curve did not have any post-peak segment. This result
indicates that the mesh size used in the lattice model should still be small enough.
Chen et al. [31] used the single-notched uniaxial tensile test to study the effects of input parameters in their nonlocal LSM
of the simulated crack pattern. As reviewed in Section 2.3.3, the nonlocal LSM has two core concepts: neighboring particles,
and particle packing, which defines the network orientation (lattice rotation). Through the benchmark, it was found that the
simulated crack pattern was dependent on the particle packing scheme, which was unfavorable. However, this dependence
can be weakened or eliminated by introducing additional neighboring particles into the model at the cost of a higher com-
putational expense, as shown in Fig. 10.
3.1.2. Three-point bending test

Three-point bending tests are mainly used to simulate mode I (opening mode) fractures during flexural failure. Shear
stress exists in the critical zone in the middle of the specimen, but the fracture process is dominated by the flexural stress.
As an alternative, four-point bending tests can be used to create a critical zone with a pure flexural stress to eliminate the
effect of shear stress on the fracture process. However, the experimental instruments, are more complex.
Liu et al. [22] studied the effect of rebar and fiber on the simulated crack pattern in three-point bending tests on concrete
specimen. It was found that (1) the crack propagation can be prohibited by the rebar, which resulted in several distributed
small cracks instead of a main crack, and (2) the presence of the rebar can greatly increase the peak stress and ductility
depicted in the P –D curve. Both results coincided with experimental observations. However, the fiber in the concrete did
not show any obvious influence on the crack pattern, which was unreasonable and unrealistic. This result may be attributed
to the improper modeling of the fiber in the lattice model.
The three-point bending test was also used by Chen et al. [32] to validate the proposed VCPM, and by Zhao and Khalili
[102] who evaluated the possibility of using a graphics processing unit (GPU) to accelerate the analysis by the lattice model.
In the research [102], the GPU with a compute unified device architecture (CUDA) was used to parallelize the computer codes
Fig. 10. Effect of lattice rotation and number of neighboring particles on simulated crack pattern in single-notched uniaxial tensile test based on nonlocal
LSM based on [31].
of DLSM [30]. By using the three-point bending test as a benchmark, the efficiency of the parallelized codes, as well as the
speed-up algorithms, were evaluated and studied.
3.1.3. Four-point shear test

The four-point shear test was first proposed by Iosipescu [111] to test the ultimate strength of metals and welded joints
under pure shear load. The instruments for four-point shear test can create a critical zone with high-shear stress, while bend-
ing stress is not dominant, which is suitable to simulate mode II (shearing mode) fractures. Similar to the uniaxial tensile
test, artificial notches were also usually included in tests and simulations as initial defects to restrict the onset of cracks.
Schlangen [3] simulated the single- and double-notched four-point shear tests by LBM. In the simulation of single-
notched test, it was found that the crack started from the notch and grew towards the support at the opposite side of the
specimen along a curved path. In the simulation of the double-notched test, one crack started at each notch, and propagated
to the opposite side of the specimen in two separated curved paths. The phenomenon of crack face bridging observed in the
experiment was reproduced in the simulation of the double-notched test. However, crack face bridging was actually depen-
dent on the load spacing, which was further studied by Cusatis et al. [44]. As shown in Fig. 11, when the spacing between
loads was small, no crack face bridging was captured. The crack pattern was almost a straight line and connected the tips of
notches. However, as the spacing between loads increased, the inclined cracks initiated at the tips of the notches developed
in different paths, which resulted in the phenomenon of crack-face bridging. The reason for the difference is attributed to the
fact that as the load spacing increased, the critical zone in the middle of the specimen was gradually transformed from a zone
with pure shear stress into a zone with both shear and tensile stresses. Thus, the orientation of crack propagation will be
affected by the difference of the stress conditions, and can result in different principal stress directions according to the the-
ory of continuum mechanics.
3.1.4. Uniaxial compressive test

Although the lattice model is efficient in simulating the fracture process under tensile loads, compressive failure is rather
difficult to simulate. To overcome the difficulties, several improvements and novel models have been proposed and devel-
oped. For example, Abreu et al. [110] improved the Delft lattice [3,19] through the use of a special element removal mech-
anism. In the improved model, the element was completely removed only when it failed twice. In the first failure, only the
bending and shear stiffness values of the critical element were removed, while the axial stiffness remained. With this
improved model, the typical ‘‘X-shaped” crack pattern, which is observed in experiments with low friction at supports,
can be reproduced in the simulation. The uniaxial compressive test was also adopted in other models, e.g., LBM with GB
[22,68], CSL [43], DLSM [94], as the benchmark to check the feasibility of these models in the simulation of compressive
failure.
Fig. 11. Effect of spacing between loads in four-point shear test on crack pattern simulated by CSL [44].
3.1.5. Nooru–Mohamed test

The Nooru–Mohamed test was originally proposed by Nooru–Mohamed to study the nucleation and growth of cracks
under combined tension and shear loads [105]. The test instrument consists of two independent loading frames with two
independent control systems for deformation-controlled tests. By applying different external loads on frames, different
stress conditions can be simulated inside the specimen. Seven different load-paths were used in the original research
conducted by Nooru-Mohamed [105]. However, only two, i.e., load-paths 4 and 6, were typically used in the benchmarks.
Load-path 4 is a nonproportional path. The shear force was first increased up to a certain value, while the tensile force
was maintained at a zero value. This means that the specimen could freely deform in the vertical direction. The shear force
was then maintained constant, while the tensile force was applied and increased until the specimen completely failed.
Load-path 6 is a proportional path, where shear and tensile forces are applied and increased simultaneously. The ratio of
the prescribed tensile displacement to the shear displacement remained constant throughout the test.
The Nooru–Mohamed test was conducted by both Schlangen and Garboczi [15] and Cusatis, Bažant et al. [44], to validate
the developed LBM and CSL, respectively. The benchmarks showed that the crack patterns observed in the test could be well
reproduced by both models, especially CSL, as shown in Fig. 12. In addition, Schlangen and Garboczi [15] also illustrated the
respective advantages of irregular networks and beam elements against regular networks and spring elements in the lattice
model in accordance to the Nooru–Mohamed test.
3.2. Types of materials
Thus far, the lattice model has been successfully applied in simulations of fracture processes in various types of materials.
Most of them were brittle or quasibrittle materials, such as mortar and concrete. Since 1990, the lattice model has been
extensively used in the study of the mechanical behavior of concrete specimens under different loads at the meso- or the
micro-scale [3,19]. To improve the performance of pure concrete under tension and flexure, rebar and fibers were included
to reinforce the concrete. Corresponding simulations have also been implemented in numerous publications
[22,45,60,112,113]. Recently, owing to a series of research studies [30,79,95,101,103,104,114], novel LSMs, e.g., DLSM, 4D
DLSM, were used to simulate the fracture of rocks in the field of geomechanics. The fracture processes of wood [115–
122], plexiglass (polymethyl methacrylate) [90,123], paper [124], woven composites [125], and biomaterials [126], were also
successfully simulated using lattice models. However, the applications of lattice models in the fracture analysis of ductile
materials, such as metals, were seldom reported. A related research can be found in [31], where the benchmark of single-
notched uniaxial tensile tests with an Aluminum alloy specimen was conducted to validate the nonlocal LSM. The material
was not explicitly mentioned in this study, but this can be deduced based on the input parameters used in the simulations,
i.e., elastic modulus, Poisson’s ratio, density, and tensile strength. Another research conducted on the fracture process of an
Aluminum sheet was also recently published [36]. The key issue in the simulation of different materials is based on how well
they define the constitutive relation and failure criterion in lattice models. These matters are reviewed below.
Fig. 12. Comparisons of crack patterns estimated based on the Nooru–Mohamed test and numerical simulations based on CSL [44].
3.2.1. Concrete
Strictly speaking, concrete is a quasibrittle material owing to the phenomenon of tension softening after the peak of the
P–D curve [127–129]. To simplify the process, it was usually assumed in early research studies [3,14,15,56] that the element
was characterized by a perfect elastic–brittle constitutive relation based on the use of the lattice model, as shown in Fig. 13.
In accordance to the mesoscopic fracture analysis scheme where the concrete is divided into cement paste, aggregate, and
ITZ, it is also assumed that all these constituents have perfectly brittle behaviors. However, there was speculation and debate
on whether the ignorance of the local tension softening in the constitutive relation at the element level was reasonable. It
was claimed that global tension softening can be simulated using a lattice model with the proper input parameters [3]. In
other research studies, it was found that the simulated concrete behavior was too brittle [17,22,55,130], especially when
a coarse lattice network was used [22]. Thus, multilinear relations with local tension softening were adopted in prior pub-
lications [60–62], as shown in Fig. 13. However, the fact that these complex relations may violate the principle of the lattice
model, i.e., the principle of local simplicity, may also be debated.
Apart from the common constitutive relations mentioned above, additional damage-like relations were also adopted,
mainly for CSL [27,43,44]. Instead of constant element strength, as commonly used in lattice models, it was assumed in
CSL that the element strength can decay during the simulation of the fracture process, which results in a more complex con-
stitutive relation. However, with this constitutive relation, the global tension softening can be properly simulated by CSL.
Therefore, to simulate the fracture process of concrete, the constitutive relation at the element level can be either very
simple or complex. The perfect elastic–linear relation that was proposed and adopted in old lattice models is still preferred
nowadays owing to its simplicity [17,18,71,130]. Other, more complex relations, such as the multilinear relation, have their
own advantages and applications, but may violate the principle of the lattice model, i.e., the local simplicity principle.
The differences in the constitutive relation will also lead to the use of different failure criteria for element removal. In
accordance to the perfect elastic–linear relation, the critical elements are completely removed once they fail. However, in
multilinear relation, the failed elements can remain in the lattice model with decreased stiffness. Thus, they are ‘‘gradually”
removed from the model.
3.2.2. Reinforced concrete

The reinforced concrete (RC) structures have been extensively used in civil engineering since they were invented. The
implementation of the rebar in the lattice model has seldom been reported. The study of rebars is an inevitable topic of dis-
cussion when the lattice model is applied to solve real engineering problems. Liu et al. [22] considered the effect of rebar on
fracture simulation in a simple way, as shown in Fig. 14. A user-defined lattice network was mapped to a mesoscale model
with a rebar. The elements inside the rebar were then identified, and the material properties of rebar were correspondingly
assigned to these elements. Similar to concrete, the rebar was also assumed to exhibit an elastic–linear behavior, but with a
large tensile strength such that it would never fail during the entire fracture process. Thus, strictly speaking, the lattice
model in [22] can only simulate the fracture process where the ultimate failure is governed by the failure of concrete,
e.g., the uniaxial compressive test. In the case of the uniaxial tensile test and the bending test with too many rebars
(over-reinforced member), the ultimate failure of the specimen is dominated by the failure of the rebar, instead of concrete.
Thus, the model in these cases will result in an infinite ultimate strength, which is unrealistic, since the rebar is assumed as
unbreakable. With this consideration in mind, a trilinear or bilinear constitutive relation with a yield plateau for the steel
rebar should be used in the lattice model to simulate the crack pattern, especially near the ultimate failure of RC specimens.
3.2.3. Fiber-reinforced concrete

It is well known that fibers can improve the ductility of concrete [131–133]. In numerical simulations, fiber can be per-
ceived as a ‘‘tiny” rebar in regard to the geometry. Thus, the modeling of fiber in the lattice model is similar to that of the
rebar. Leite et al. [60] included the fiber by connecting distant mesh nodes in the cement paste after the generation of the
Fig. 13. Typical constitutive relations used in lattice model.

Fig. 14. Simulated crack patterns in three-point bending test on RC specimen [22].
concrete mesostructure. A bilinear constitutive relation shown in Fig. 13 was adopted to describe the behavior of the fiber
element. It was found that fibers did not affect the ultimate strength, but they could significantly increase the ductility of the
specimen.
Liu et al. [22] modeled fibers by mapping the concrete mesostructure to a regular lattice network. The elements inside the
fiber assigned the material properties of the fibers. Owing to the regular triangular lattice network used in the lattice model,
fibers were distributed along three directions at 60° orientations with respect to each other. The constitutive relation for
fibers was not explicitly mentioned in this study. However, based on the context of this publication, it seems that fibers were
assumed to have the same elastic properties as aggregates, which is questionable. Additionally, the benchmarks of uniaxial
tensile and compressive tests did not show any obvious effects of fibers on the ductility of the specimen. The P–D curves
obtained from these benchmarks indicated a very brittle material behavior with little or no tension softening. The reasons
for these unsatisfying results may be attributed to the constitutive relation of fibers, regularity of fiber orientation, and an
overly coarse mesh size, used in the simulation.
Montero-Chacon et al. [45] used LPM to study the fracture process of steel fiber-reinforced concrete (SFRC). The lattice
network used in the simulation was determined based on the concrete mesostructure, which is similar to RPM [1] and
CSL [27,43,44]. After the network was formed, several lines, which represented the fibers, were randomly placed into the
network. Special additional bond elements were formed to connect the fiber ends to nodes in the network, as shown in
Fig. 15. A similar failure criterion, shown in Eq. (2), characterized by the yield strength of fibers, was adopted to describe
the material behavior of fiber elements. The results showed that fibers could slightly increase the ultimate strength, and sig-
nificantly increase the ductility of the SFRC specimen; findings that are in good agreement with the experimental results.
3.2.4. Rock
The application of lattice models in fracture analysis of rock is mainly attributed to the research studies of Zhao et al. [30]
who developed DLSM [30,79,95] and 4D LSM [28] in the field of geomechanics. The fracture of rock is somewhat similar to
that of concrete in that both are brittle structures. Thus, the constitutive relation used in the fracture analysis of rock
[79,104,109] is similar to that for concrete, but with a few differences. In DLSM, the continuous model is discretized into
several particles, which are connected by normal and shear springs. The failure criterion of the springs is controlled by
the motions of the particles, i.e., the distance between the particles. Thus, the constitutive relation in DLSM is focused on
the ultimate strain, instead of the tensile strength. Furthermore, it is permitted in DLSM that the failed springs can regain
stiffness once the particles connected by the springs are close enough to each other owing to additional particle motions.
In other words, the failed springs still exist in the model, and can deform freely, but without the capacity to transfer forces,
Fig. 15. Modeling of fibers in lattice model and simulated model deformation in two-notched uniaxial tensile test on SFRC [45].
Fig. 16. Schematics of constitutive relations used in DLSM for fracture analysis of rock based on [79].
as illustrated in Fig. 16. A typical fracture analysis of a rock sample under a point-to-plane compressive load, based on DLSM,
is shown in Fig. 17.
3.2.5. Wood
Wood is a representative example of anisotropic materials owing to the tracheid and growth ring structures. The fracture
process perpendicular to the wood grain (longitudinal direction) is similar to the brittle fracture of concrete. Correspond-
ingly, the lattice model has also been used to study the fracture problem of wood.
Landis et al. [122] and Vasic et al. [120] proposed a lattice model containing both beam and spring elements to simulate
the fracture process of wood perpendicular to the longitudinal direction. The beam element refers to the bundles of tra-
cheid inside the wood, while truss elements connect different beam elements, as shown in Fig. 18. The presence of spring
elements between bundles of tracheid enables the simulation of a 2D orthotropic material. The material properties of beam
and spring elements are determined by a ‘‘trial-and-error” procedure until the simulated global material properties become
close to those obtained from experiments. The constitutive relations for beam and spring elements are not explicitly
mentioned in this publication. In a more recent and related research study [116], a bilinear constitutive relation was
adopted for all elements in the lattice model. For simulating the crack speed, the effects from the rate of deformation were
Fig. 17. Comparison between simulated failure modes of rock under point-to-plane compressive load and experimental results [104].
Fig. 18. Hybrid lattice modeling of wood using beam and spring elements [122].
incorporated—constituting a novelty in comparison to previous bilinear relations—resulting in a time-dependent constitu-

tive relation. A detailed explanation about the constitutive relation can be found in [120,122]. As a benchmark of single-
notched uniaxial tensile tests, the simulated crack pattern based on a perfect elastic–linear relation was compared with
that elicited by the novel time-dependent bilinear relations. It was found that the novel relations could predict the crack
pattern more reasonably.
Reichert and Ridley-Ellis [115] adopted three approaches, i.e., step-size control algorithms, and methods to account for
inelastic forces and hybrid models of lattice and solid elements, to speed up the simulation of wood fracture using the lattice
model. Each element in the lattice model was assumed to exhibit a bilinear constitutive relation, and it was claimed that the
trilinear relation shown in Fig. 13 should be used instead [134], if the model is preferred to be extended to account for plastic
compressive behavior. The anisotropic character of wood is simply simulated by adjusting the material properties, such as
the elastic modulus and tensile strength of individual elements, based on their positions in the growth-ring structure. More
details on this research work can be found in the doctoral thesis of Reichert [135].
3.2.6. Metal
The fracture process of metals is a highly complex problem, and has been studied for many years. However, the applica-
tion of the lattice model in fracture problems of metals has seldom been reported. In the research of Chen et al. [31], single-
notched uniaxial tensile tests were conducted based on a 2D non-local LSM. The material properties used in the simulation
were E ¼ 69 GPa, v ¼ 0:3, q ¼ 2700 kg=m3 , and f t ¼ 200 MPa. Based on these input parameters, it can be concluded that the
material is probably a type of Aluminum alloy. However, it is still assumed that the element has a perfect brittle behavior. As
a result, the simulated fracture process is similar to that of concrete.
At small (meso- or micro-) scales, the fracture process of metals can be more easily compared with the fracture of con-
crete. At the mesoscale where concrete is divided into aggregate, cement paste, and ITZ, the cracks usually propagate along
the surface of aggregates, as the ITZ is the most vulnerable part owing to its lowest tensile strength. However, in some other
cases, e.g., when light weight aggregate (LWA) is used, the cracks may also penetrate the aggregates [136]. Similarly, at the
microscale, where the crystal structure of metal is concerned, the cracks can also propagate along the interface between the
crystals or through the crystals, which results in intercrystalline (intergranular), or transcrystalline (transgranular) fractures,
respectively [137,138]. Both intercrystalline and transcrystalline fractures can be either brittle or ductile in typical condi-
tions [138–142]. Thus, the research in [31] only simulated the brittle fracture process of metal, which is essentially the same
as that for concrete. To simulate a ductile fracture, the yield plateau of metals may be accounted for in the constitutive rela-
tion to allow the ductile deformation of elements before failure.
3.3. Type of loads
3.3.1. Static loads

A pure static load is time-independent and constant during the entire part of the fracture analysis. Theoretically speaking,
pure static loads do not exist. They are often used in lattice models as ‘‘pseudo” loads to obtain the crack pattern [3]. In the
‘‘event-by-event” element removal method, which is reviewed in Section 2.2.5, the external load gradually increases from
zero to a certain level where the first failed element appears. Thus, the implementation of the removal method requires a
step-by-step analysis with an increased computational cost. To speed up the simulation, an arbitrary pseudo load, which
is not necessarily equal to the real load, can be used [3]. After the linear analysis of the entire lattice model, the element with
the maximum degree of damage (nm ), which is defined as the ratio of stress to strength, can be found. The variable nm can be
either larger or smaller than 1.0, depending on the pseudo load used in the analysis. The external load value at which the first
failed element appears is then simply calculated by multiplying the pseudo load by a scaling factor (1=nm ). The structural
responses, e.g., nodal displacement, element stress, are also multiplied by the scaling factor. In this approach, the external
load can be constant during the entire fracture analysis, and only one analysis is needed to identify each failed element,
which can effectively speed up the simulation. However, it should be mentioned that this simplified method can only be
adopted in a linear system.
3.3.2. Quasistatic loads

Thus far, most external loads used in lattice models for fracture analysis belong to the quasistatic type. In experiments
conducted to simulate the fracture process, external loads often gradually increase from zero to an ultimate level in a slow
manner, until the complete failure of the specimen. In numerical simulations, the load path is divided into several steps to
simplify the process. The load in each step is assumed to be constant. Based on these simplifications and assumptions, the
quasistatic loads in lattice models can be path-dependent or time-independent. Both the event-by-event and load-stepping
methods can be used for these types of loadings. The fracture analysis with the load-stepping method is much simpler than
the event-by-event method. Only one analysis in each step is required, and all failed elements (if they exist) are removed
from the lattice model. However, as reviewed in Section 2.2.5, the load-stepping method required the fine division of the
load path to obtain a realistic crack pattern. The fracture analysis conducted based on the event-by-event method is more
complex, and is briefly explained next. Suppose that the quasistatic load in the ith step is F i . An arbitrary load (F 0 ) is used
in the lattice model, and a linear analysis is conducted to identify the element with the degree of maximum damage (nm ).
The value of nm is then scaled by nm;i ¼ nm F 0 =F i . If nm;i < 1, no elements are removed, and the entire analysis procedure con-
tinues to the next step. If nm;i > 1, the element with nm is removed, and the structural responses are scaled by 1=nm;i . Subse-
quently, similar linear analyses are conducted until nm;i < 1.
Since quasistatic loads are time-independent, the simulated crack pattern is also time-independent. This means that the
cracking speed is impossible to simulate with these types of loads. Thus, the fracture analysis conducted based on lattice
models under static or quasi-static loads mainly focus on the final crack pattern at the complete failure of the model, or
on the intermediate crack pattern when a typical number of elements are removed from the model.
3.3.3. Dynamic loads

Most classical lattice models cannot simulate the fracture analysis under dynamic loads since the equilibrium and con-
stitutive relations used in these models are time-independent. Recently introduced lattice models, especially those based on
particle motions [30,79,95,102], can be used to simulate conveniently the fracture process under impact, blast, and other
types of dynamic loads. The several approaches used to consider the effect of dynamic loads in lattice models are briefly
reviewed below.
Guo et al. [21], Guo et al. [143] used classical LBM and LSM to simulate the fracture process of a concrete specimen under
fatigue loads. However, the fatigue loads were applied based on a simplified approach. Neither the time for each load cycle
nor the load form (triangular wave or sinusoidal wave) were explicitly considered. Instead, a fatigue damage function was
proposed as a function of the number of load cycles to reduce the stiffness of elements in the cement paste and ITZ (the
aggregate was assumed as unbreakable in this study) after each load cycle. When the strain of one element reached a thresh-
old, the element stiffness was suddenly reduced to 10% of the initial stiffness. There are several parameters in the fatigue
damage function. The effects of the characteristics of fatigue loads were considered by these parameters, but they needed
to be calibrated using experiments. Since the fatigue damage function was defined to be dependent on the load cycle, it
is possible to estimate how many cycles the model can endure before a typical fracture or complete failure occur, i.e., the
fatigue life in terms of number of load cycles can be predicted. The load used in the simulation is not actually a real dynamic
load, since it is still time-independent.
Zhao et al. [30] used DLSM to simulate various types of dynamic fracture problems. As reviewed in Section 2.3.2 (see
Fig. 7), the internal forces inside the springs are calculated based on the stiffnesses of the springs and distances between par-
ticles, while the distances are further determined by a time-dependent equilibrium equation of particle motions. Thus, real
dynamic loads can be fully considered in DLSM. As shown in Fig. 19, the fracture processes can be accurately simulated using
different solid strengths in the case of collisions. Other case studies, such as blasts inside tunnels [95], blasts inside hollow
spheres [28], and blasting waves [101,103], were also conducted.
4. Challenges and perspectives
Owing to the efforts of pioneers and other researchers, the lattice model has been extensively developed as an efficient
tool to simulate the fracture process of various types of materials under different types of loads. Based on the above review,
however, problems still exist both in theory and applications. To further develop the lattice model, several challenges should
be addressed. These challenges may also constitute the focus of future research.
4.1. Theory development
4.1.1. Interdisciplinary innovations

As mentioned previously, the newly developed lattice models, such as CSL, LDPM, and DLSM, have one thing in common:
they have benefitted from the advantages of other numerical models, such as DPM and PD, to eliminate the disadvantages of
classical lattice models, e.g., the fixed Poisson’s ratio. In the academic world where interdisciplinary cooperation and inno-
vation have become increasingly important, limitations of classical theory and models can be broken down by adopting new
ideas and concepts using other related theories and models, or by transforming the formulations to entirely different fields.
For example, 4D LSM [28] was developed based on the classical LSM, and the concept of extra dimensions from modern phy-
sics. Interdisciplinary research studies can lead to innovations that can be incorporated in the classical theory and models,
while simultaneously preserving their most important advantages.
4.1.2. New materials

During the past century, many innovations revolutionized construction and building materials. For example, in the con-
crete industry, the concept of ‘‘green concrete” has been proposed [144] to decrease the amount of carbon dioxide that is
released during the manufacturing of cement clinkers, thereby utilizing schemes where the cement is partially or fully
replaced by other materials, e.g., fly ash (FA) [145,146], blast furnace slag (BFS) [147], silica fume (SF) [148], and geopolymers
[149]. To increase the strength and decrease the permeability, high-performance concrete (HPC) [150] and ultra-high per-
formance concrete (UHPC) [151,152] have been invented and have already been used in engineering structures. To increase
the ductility, steel, and other types of fibers, have been added to concrete to form fiber-reinforced concrete (FRC) [153]. To
develop smart materials, the concept of self-healing has been introduced into concrete [154,155].
To deal with the material innovations, the lattice model should be further developed as well. The most important part is
the constitutive relation, which is essential for detecting the critical elements. A proper constitutive relation is based on a full
understanding of structural behaviors of different phases in materials, e.g., aggregate, cement paste, steel fiber, and also on
the interactions between them. Meanwhile, the constitutive relation should be as simple as possible to preserve the intrinsic
advantages of classical lattice models.
Failure criteria are also an important issue in lattice models. The failure criteria that are based on nominal stress (force) or
strain (displacement) are preferred owing to their simplicity. However, the use of the lattice model in association with an
oversimplified failure criterion may not be able to simulate the failure process under complex stress conditions. For example,
classical lattice models only adopted a critical tensile stress criterion [13,56] where the compressive strength of material was
assumed as infinite, which was unrealistic. Thus, these classical lattice models can only simulate the tensile fracture process.
That is why the Mohr–Coulomb criterion was adopted in research studies to consider the effect of shear stress in the fracture
Fig. 19. DLSM simulation of the dynamic failure process of a large body impacted by an intruder: (a) large body with a strength of 16 MPa, (b) large body
with a strength of 0.59 MPa, and a (c) large body with a strength of 0.16 MPa [30].
process. More advanced failure criteria, especially those based on fracture energy, may be needed in the future to enable the
lattice model to address complex cases and be applied in real structures.
4.1.3. From 2D to 3D
In the 20th century, the lattice model was mainly applied to solve 2D problems, i.e., plane stress and strain, due to the
limitations of computer ability. With more powerful computers available nowadays, 3D lattice models are already proposed
and developed [17,19,30,114], but related research is not sufficient yet. Typically, how to properly model the 3D materials
which are inherently anisotropic is still an open issue which needs more attention in further research. Two approaches can
be possibly adopted to correctly simulate the behaviors of 3D anisotropic materials such as wood and granular materials.
The first approach is to directly simulate the structures of the material which are responsible for the anisotropy. For
example, the anisotropy of rocks mainly comes from the specific features of material fabric such as bedding and layering
[156]. To properly simulate the behaviors of rocks, therefore, these specific features should be explicitly modeled and then
mapped to the network of the lattice model to attribute the material properties to the elements based on their positions. This
approach is based on the prerequisite that the reasons and specific features for the material anisotropy are clearly clarified. If
the reasons are unknown, or the specific features are too complex to model, the properties of the elements in the lattice
model needs to be properly calibrated to make the entire model behave like an anisotropic material. This approach is similar
to that in the simulation of an arbitrary Poisson’s ratio [37,90]. But how to effectively calibrate the properties of the elements
according to an arbitrary anisotropy of a material is still under investigation.
4.2. Applications in engineering practices
Thus far, the lattice models in publications were mainly applied in benchmarks of classical laboratory tests to validate
their accuracy and efficiency, or in mechanism studies at small scales to identify the relations between the micro- or
mesostructures of the material and its macroscopic properties. To the best of our knowledge, applications of lattice models
in real, large-scale structures have not been reported yet. Continuous models, such as FEM, are still popular in engineering
practice to predict the critical zones in the structure that are vulnerable to cracks. However, the crack propagation process is
usually not a concern owing to the disadvantages of continuous models in fracture analysis, which is not sufficient to prevent
maintenance during the life-cycle of structures.
4.2.1. Multiscale method

When the lattice model is applied into a large-scale structure, one of the most significant problems is the excessive com-
putational work, especially for lattice models (RPM, CSL, LDPM, etc.) that are directly based on the heterogeneous mesostruc-
ture of the material. Generally, it is impossible to simulate the mesostructure of the material in the entire structure owing to
the limitations of the computers. Alternatively, a feasible and a more efficient approach can be adopted, based on the mul-
tiscale concept, as shown in Fig. 20. First, the entire structure is analyzed by continuous models at the macroscale to identify
the critical zones that are vulnerable to cracks. Secondly, the material mesostructure as well as the lattice model is only
formed in critical zones. Boundary conditions used in the lattice model are determined based on the results of previous
macroscopic analyses. A local analysis is then conducted based on the use of the lattice model to identify the critical ele-
ments. With the removal of these elements, however, the material properties and boundary conditions of critical zones
may change correspondingly. Thus, a homogenization is needed to obtain the equivalent material properties of the critical
zones and transfer them to the macroscale. Another macroscopic analysis with an updated material and model is then per-
formed to identify new critical zones. The above procedure is repeated to simulate the crack propagation process. The effi-
ciency of this method can be further improved by lowering the frequency of information transfer between models at
different scales. For example, only when several critical elements—instead of just one—are removed from the mesoscale
model allows the analysis to proceed to the next step at the macroscale. Other multiscale methods, such as hierarchical
modeling can be also adopted [157].
Vulnerable
zone
Inner force
(bending moment)
Boundary Boundary
condition condition
Analysis at Analysis at
Analysis at
structure component
local level
level level
Homogeniztion Homogeniztion
Fig. 20. Schematic of multiscale method in a lattice model used to solve practical engineering problems.
4.2.2. Parallel computing

Another technique which can be used to realize the application of the lattice model in large-scale structures is parallel
computing. Some research on this topic has already been conducted [19,101]. The basic idea of parallel computing is to dis-
tribute the computational work to several processors (symmetric computing), or several computers (cluster computing), and
to execute the distributed works simultaneously. To use the technique of parallel computing, the lattice model should be
developed in such a manner that the programming codes can be parallelized. For example, the global stiffness matrix, which
is required by some lattice models, is not preferred in parallel programming, as it is usually obtained by assembling the ele-
ment stiffness matrices one-by-one. However, this is a serial procedure and is difficult to parallelize. Therefore, it is better to
replace a global problem by several local independent problems in the lattice model, such as the Delft lattice [19] and DLSM
[101], to increase the degree of parallelization of programming codes.
4.2.3. Adaptive mesh refinement

The technique of adaptive mesh refinement (AMR), which has been extensively used in continuous models [158–161],
can be also introduced into the lattice model. Based on the general idea of AMR, a coarse network can be used in the first
analysis step. When critical elements are identified, the network around these elements is refined. The updated network will
be used in the next analysis step. Two possible methods used to implement AMR in lattice models are proposed herein. For
the 2D random network formed using the Delaunay triangulation, shown in Fig. 21, a uniform grid with large cells is used to
obtain a coarse network. After the first analysis step, the critical element is identified. Subsequently, the cells in which the
nodes of the critical element are located are divided into several subcells. Finally, a local Delaunay triangulation is conducted
to refine the network around the critical element. For mesotructure-based lattice models, such as RPM and CSL, shown in
Fig. 22, large particles based on the gradation are first placed into the model to form the initial coarse network. When the
critical spring is identified, smaller particles are placed in a local region to refine the network. The above network refinement
does not require significant computational time, since both the Delaunay triangulation and particle packing are only con-
ducted in a small, local region, and since the other elements and particles remain unchanged.
5. Conclusions
This study reviewed the lattice model from the viewpoints of theory and applications. Owing to the contributions of the
researchers in the last 80 years, the lattice model has become an efficient tool in the study the fracture problems of various
types of materials. Based on this review, the following conclusions can be drawn:
(1) To correctly simulate the materials with an arbitrary Poisson’s ratio, classical LSMs should be modified using (1) extra
nodal DOFs, (2) extra shear springs, (3) extra nonlocal strain energy parameters, and (4) higher dimensional normal
springs.
(2) The lattice model has already been successively applied in simulations of fracture processes of brittle or quasibrittle
materials, such as concrete, rock, and wood, under static and quasistatic loads.
(3) The applications of lattice models in fracture analyses of metals, dynamic load-induced fracture, and real large-scale
structures are not sufficient.
(4) Further innovations regarding the lattice model will emerge from novel ideas and concepts in other related models,
and from applications to different research fields, to strengthen the theoretical basis, and eliminate its current
disadvantages.
Fig. 21. Schematic of AMR based on Delaunay triangulation in accordance to the lattice model.
Fig. 22. Schematic of AMR in mesostructure-based lattice models.
(5) The constitutive relation and failure criterion are the most important parts of the lattice model. Further improvements
in these parts are needed to deal with innovations in materials and complex stress conditions in real structures. The
energy-based failure criterion is of special importance and needs further attention.
(6) The method to properly simulate behaviors of 3D materials which are inherently anisotropic by lattice models needs
further research both in theory and application.
(7) Multiscale methods, parallel computing, AMR, and other techniques can be integrated into the lattice model to
decrease the computational efforts and realize its application in the fracture analyses of real structures where exces-
sive computational work is typically encountered.
Acknowledgments
The authors would like to thank the National Natural Science Foundation of China (No. 51608377) and the Fundamental
Research Funds for the Central Universities of China for the financial support. The suggestions from anonymous reviewers
are also appreciated.
Appendix A. Supplementary material
Supplementary data associated with this article can be found, in the online version, at https://doi.org/10.1016/j.en-
gfracmech.2017.12.037.
References
[1] Bazant ZP, MR Tabbara, MT Kazemi, Pijaudiercabot G. Random particle model for fracture of aggregate or fiber composites. J Eng Mech 1990;116
(8):1686–705.
[2] Kikuchi A, Kawai T, Suzuki N. The rigid bodies-spring models and their applications to three-dimensional crack problems. Comput Struct 1992;44
(1):469–80.
[3] Schlangen E. Experimental and numerical analysis of fracture processes in concrete. 1993, Fac Civil Eng Geosci.
[4] Bobet A, Fakhimi A, Johnson S, Morris J, Tonon F, Yeung MR. Numerical models in discontinuous media: review of advances for rock mechanics
applications. J Geotech Geoenviron Eng 2009;135(11):1547–61.
[5] Cusatis G, Pelessone D, Mencarelli A. Lattice discrete particle model (LDPM) for failure behavior of concrete. I: Theory Cem Concr Compos 2011;33
(9):881–90.
[6] Hrennikoff A. Solution of problems of elasticity by the framework method. J Appl Mech 1941;8(4):169–75.
[7] Ziman JM. Models of disorder: the theoretical physics of homogeneously disordered systems. Cambridge: Cambridge University Press; 1979.
[8] Feng S, Sen PN. Percolation on elastic networks: new exponent and threshold. Phys Rev Lett 1984;52(3):216–9.
[9] Roux S, Guyon E. Mechanical percolation: a small beam lattice study. J de Physique Lett 1985;46(21):999–1004.
[10] Sahimi M, Goddard JD. Elastic percolation models for cohesive mechanical failure in heterogeneous systems. Phys Rev B 1986;33(11):7848.
[11] Louis E, Guinea F. The fractal nature of fracture. EPL (Eur Lett) 1987;3(8):871–7.
[12] Herrmann HJ, Hansen A, Roux S. Fracture of disordered, elastic lattices in two dimensions. Phys Rev B: Condens Matter 1989;39(1):637–48.
[13] Schlangen E, Van Mier J. Simple lattice model for numerical simulation of fracture of concrete materials and structures. Mater Struct 1992;25
(9):534–42.
[14] Schlangen E, Garboczi EJ. New method for simulating fracture using an elastically uniform random geometry lattice. Int J Eng Sci 1996;34
(10):1131–44.
[15] Schlangen E, Garboczi EJ. Fracture simulations of concrete using lattice models: computational aspects. Eng Fract Mech 1997;57(2-3):319–32.
[16] Mier JGMV, Chiaia BM, Vervuurt A. Numerical simulation of chaotic and self-organizing damage in brittle disordered materials. Comput Methods Appl
Mech Eng 1997;142(1–2):189–201.
[17] Lilliu G, Mier JGMV. 3D lattice type fracture model for concrete. Eng Fract Mech 2003;70(7–8):927–41.
[18] Man HK, van Mier JGM. Damage distribution and size effect in numerical concrete from lattice analyses. Cem Concr Compos 2011;33(9):867–80.
[19] Qian Z. Multiscale modeling of fracture processes in cementitious materials; 2012.
[20] Hou P. Lattice model applied to the fracture of large strain composite. Theor Appl Fract Mech 2007;47(3):233–43.
[21] Guo L-P, Carpinteri A, Roncella R, Spagnoli A, Sun W, Vantadori S. Fatigue damage of high performance concrete through a 2D mesoscopic lattice
model. Comput Mater Sci 2009;44(4):1098–106.
[22] Liu JX, Zhao ZY, Deng SC, Liang NG. Modified generalized beam lattice model associated with fracture of reinforced fiber/particle composites. Theor
Appl Fract Mech 2008;50(2):132–41.
[23] Grassl P. A lattice approach to model flow in cracked concrete. Cem Concr Compos 2009;31(7):454–60.
[24] Sadouki H, van Mier JGM. Meso-level analysis of moisture flow in cement composites using a lattice-type approach. Mater Struct 1997;30
(10):579–87.
[25] Šavija B, Luković M, Schlangen E. Lattice modeling of rapid chloride migration in concrete. Cem Concr Res 2014;61–62:49–63.
[26] Šavija B, Pacheco J, Schlangen E. Lattice modeling of chloride diffusion in sound and cracked concrete. Cem Concr Compos 2013;42:30–40.
[27] Cusatis G, Zdeně, Baž P, Asce F, Cedolin L, Asce M. Confinement-shear lattice model for concrete damage in tension and compressioN: I. theory. J. Eng.
Mech. 2003;129(12):1439–48.
[28] Zhao G-F. Developing a four-dimensional lattice spring model for mechanical responses of solids. Comput Methods Appl Mech Eng 2017;315:881–95.
[29] Grassl P, ZP Bazant, and G Cusatis. Lattice-cell approach to quasibrittle fracture modeling. Comp. Mod. Concr. Struct., In: Meschke, de Borst, Mang,
Bicanic, editors. 2006: 263–268.
[30] Zhao G-F, Fang J, Zhao J. A 3D distinct lattice spring model for elasticity and dynamic failure. Int J Numer Anal Methods Geomech 2011;35(8):859–85.
[31] Chen H, Lin E, Jiao Y, Liu Y. A generalized 2D non-local lattice spring model for fracture simulation. Comput Mech 2014;54(6):1541–58.
[32] Chen H, Lin E, Liu Y. A novel volume-compensated particle method for 2D elasticity and plasticity analysis. Int J Solids Struct 2014;51(9):1819–33.
[33] Karihaloo BL, Shao PF, Xiao QZ. Lattice modelling of the failure of particle composites. Eng Fract Mech 2003;70(17):2385–406.
[34] Šavija B, Luković M, Pacheco J, Schlangen E. Cracking of the concrete cover due to reinforcement corrosion: a two-dimensional lattice model study.
Constr Build Mater 2013;44:626–38.
[35] Chen A, Pan Z, Ma R. Mesoscopic simulation of steel rebar corrosion process in concrete and its damage to concrete cover. Struct Infrastruct Eng
2017;13(4):478–93.
[36] Ostoja-Starzewski M, Sheng PY, Alzebdeh K. Spring network models in elasticity and fracture of composites and polycrystals. Comput Mater Sci
1996;7(1):82–93.
[37] Ostoja-Starzewski M. Lattice models in micromechanics. Appl Mech Rev 2008;55(1):2002.
[38] Voyiadjis GZ. Handbook of damage mechanics: Nano to macro scale for materials and structures. 2015: Springer.
[39] Ostoja-Starzewski M. Microstructural randomness and scaling in mechanics of materials. 2008: Chapman & Hall/CRC.
[40] Kim K, Lim YM. Simulation of rate dependent fracture in concrete using an irregular lattice model. Cem Concr Compos 2011;33(9):949–55.
[41] Bolander JE, Sukumar N. Irregular lattice model for quasistatic crack propagation. Phys. Rev B 2005;71(9):094106.
[42] Hwang YK, Lim YM. Validation of three-dimensional irregular lattice model for concrete failure mode simulations under impact loads. Eng Fract Mech
2017;169:109–27.
[43] Cusatis G, Zdeně, Baž P, Asce F, Cedolin L, Asce M. Confinement-shear lattice model for concrete damage in tension and compression: II. Computation
and validation. J Eng Mech 2003;129(12):1449–58.
[44] Cusatis G, Bažant ZP, Cedolin L. Confinement-shear lattice CSL model for fracture propagation in concrete. Comput Methods Appl Mech Eng 2006;195
(52):7154–71.
[45] Montero-Chacon F, Cifuentes H, Medina F. Mesoscale Characterization of Fracture Properties of Steel Fiber-Reinforced Concrete Using a Lattice-
Particle Model. Materials 2017;10(2): p. 19.
[46] Chiaia B, Vervuurt A, Mier JGMV. Lattice model evaluation of progressive failure in disordered particle composites. Eng Fract Mech 1997;57(2–
3):301–9.
[47] Jagota A, Bennison S. Spring-network and finite-element models for elasticity and fracture. In: Non-linearity and Breakdown in Soft Condensed
Matter. 1994, Springer. p. 186–201.
[48] Beale PD, Srolovitz DJ. Elastic fracture in random materials. Phys Rev B: Condens Matter 1988;37(10):5500.
[49] Srolovitz DJ, Beale PD. Computer simulation of failure in an elastic model with randomly distributed defects. J Am Ceram Soc 2010;71(5):362–9.
[50] Nayfeh AH, Hefzy MS. Continuum modeling of three-dimensional truss-like space structures. AIAA J 1978;16(8):779–87.
[51] Frédéric D, Magnier SA. Formulation of a 3-D numerical model of brittle behaviour. Geophy J R Astron Soc 2010;122(3):790–802.
[52] Pazdniakou A, Adler PM. Lattice spring models. Transp Porous Media 2012;93(2):243–62.
[53] Zhao S-F, Zhao G-F. Implementation of a high order lattice spring model for elasticity. Int J Solids Struct 2012;49(18):2568–81.
[54] Buxton GA, Care CM, Cleaver DJ. A lattice spring model of heterogeneous materials with plasticity. Modell Simul Mater Sci Eng 2001;9(6):485–97.
[55] Liu JX, Deng SC, Zhang J, Liang NG. Lattice type of fracture model for concrete. Theor Appl Fract Mech 2007;48(3):269–84.
[56] Schlangen E. Computational aspects of fracture simulations with lattice models, in Fracture mechanics of concrete structures, Proceedings FRAMCOS-
2. 1995: p. 913–28.
[57] Zubelewicz A, Bažant ZKP. Interface Element Modeling of Fracture in Aggregate Composites. J Eng Mech 1987;113(11):1619–30.
[58] Griffiths DV, Mustoe GGW. Modelling of elastic continua using a grillage of structural elements based on discrete element concepts. Int J Numer
Methods Eng 2001;50(7):1759–75.
[59] Ostoja-Starzewski M, Lee JD. Damage maps of disordered composites: a spring network approach. Int J Fract 1995;75(3):R51–7.
[60] Leite JPB, Slowik V, Mihashi H. Computer simulation of fracture processes of concrete using mesolevel models of lattice structures. Cem Concr Res
2004;34(6):1025–33.
[61] Arslan A, Ince R, Karihaloo B. Improved lattice model for concrete fracture. J Eng Mech 2002;128(1):57–65.
[62] Ince R, Arslan A, Karihaloo BL. Lattice modelling of size effect in concrete strength. Eng Fract Mech 2003;70(16):2307–20.
[63] Zhao G, Khalili N. A lattice spring model for coupled fluid flow and deformation problems in geomechanics. Rock Mech Rock Eng 2012;45(5):781–99.
[64] Mungule M, Raghuprasad BK. Meso-scale studies in fracture of concrete: a numerical simulation. Comput Struct 2011;89(11–12):912–20.
[65] Curtin WA, Scher H. Brittle fracture in disordered materials: A spring network model. J Mater Res 2011;5(3):535–53.
[66] Ostoja-Starzewski M, Sheng PY, Jasiuk I. Damage patterns and constitutive response of random matrix-inclusion composites. Eng Fract Mech 1997;58
(5–6):581–606.
[67] Alzebdeh K, Al-Ostaz A, Jasiuk I, Ostoja-Starzewski M. Fracture of random matrix-inclusion composites: scale effects and statistics. Int J Solids Struct
1998;35(19):2537–66.
[68] Liu JX, Zhao ZY, Deng SC, Liang NG. Numerical investigation of crack growth in concrete subjected to compression by the generalized beam lattice
model. Comput Mech 2008;43(2):277–95.
[69] Liu JX, Liang NG. Algorithm for simulating fracture processes in concrete by lattice modeling. Theor Appl Fract Mech 2009;52(1):26–39.
[70] Qiang Y, Yonggang C, Hao Z. Simulation of cracking processes of rock materials by lattice model. Eng Mech 2003;20(1):117–26.
[71] Van Mier JG, van Vliet MR, Wang TK. Fracture mechanisms in particle composites: statistical aspects in lattice type analysis. Mech Mater 2002;34
(11):705–24.
[72] Vidya Sagar R, Raghu Prasad B. Fracture analysis of concrete using singular fractal functions with lattice beam network and confirmation with
acoustic emission study. Theor Appl Fract Mech 2011;55(3):192–205.
[73] Vidya Sagar R, Raghu Prasad BK, Karihaloo BL. Verification of the applicability of lattice model to concrete fracture by AE study. Int J Fract 2009;161
(2):121–9.
[74] Labuz JF, Zang A. Mohr-Coulomb failure criterion. In: The ISRM Suggested Methods for Rock Characterization, Testing and Monitoring: 2007–2014.
2012, Springer; p. 227–31.
[75] Bolander JE, Saito S. Fracture analyses using spring networks with random geometry. Eng Fract Mech 1998;61(5–6):569–91.
[76] Liu JX, Zhao ZY, Deng SC, Liang NG. A simple method to simulate shrinkage-induced cracking in cement-based composites by lattice-type modeling.
Comput Mech 2008;43(4):477–92.
[77] Zhang Z, Ding J, Ghassemi A, Ge X. A hyperelastic-bilinear potential for lattice model with fracture energy conservation. Eng Fract Mech
2015;142:220–35.
[78] Kosteski L, Barrios D’Ambra R, Iturrioz I. Crack propagation in elastic solids using the truss-like discrete element method. Int J Fract 2012;174
(2):139–61.
[79] Zhao G-F, Russell AR, Zhao X, Khalili N. Strain rate dependency of uniaxial tensile strength in gosford sandstone by the distinct lattice spring model
with X-ray micro CT. Int J Solids Struct 2014;51(7–8):1587–600.
[80] Rots JG, Invernizzi S. Regularized sequentially linear saw-tooth softening model. Int J Numer Anal Methods Geomech 2004;28(7–8):821–56.
[81] Rots JG, Invernizzi S, Belletti B. Saw-tooth softening/stiffening-a stable computational procedure for RC structures. Comput Concr 2006;3(4):213–33.
[82] Rots JG, Belletti B, Invernizzi S. Robust modeling of RC structures with an ‘‘event-by-event” strategy. Eng Fract Mech 2008;75(3–4):590–614.
[83] Bolander J, Shiraishi T, Isogawa Y. An adaptive procedure for fracture simulation in extensive lattice networks. Eng Fract Mech 1996;54(3):325–34.
[84] Šavija B, Schlangen E. On the use of a lattice model for analyzing of in-plane vibration of thin plates. Comput Mater Contin 2015;48(3):181–202.
[85] Alder BJ, Wainwright TE. Studies in molecular dynamics. I. General Method. J Chem Phys 1959;31(2):459–66.
[86] Silling SA. Reformulation of elasticity theory for discontinuities and long-range forces. J Mech Phy Solids 2000;48(1):175–209.
[87] Silling SA, Bobaru F. Peridynamic modeling of membranes and fibers. Int J Non-Linear Mech 2005;40(2–3):395–409.
[88] Silling SA, Askari E. A meshfree method based on the peridynamic model of solid mechanics. Comput Struct 2005;83(17–18):1526–35.
[89] Chen H, Jiao Y, Liu Y. A nonlocal lattice particle model for fracture simulation of anisotropic materials. Compos B Eng 2016;90:141–51.
[90] Chen H, Liu Y. A non-local 3D lattice particle framework for elastic solids. Int J Solids Struct 2016;81:411–20.
[91] Lax M. The relation between microscopic and macroscopic theories of elasticity. Solid State Commun 1963;1(6). 195-195.
[92] Keating PN. Effect of invariance requirements on the elastic strain energy of crystals with application to the diamond structure. Phys Rev 1966;145
(2):637–45.
[93] Jagota A, Scherer GW. Viscosities and sintering rates of a two-dimensional granular composite. J Am Ceram Soc 1993;76(12):3123–35.
[94] Zhao GF. Development of the distinct lattice spring model for large deformation analyses. Int J Numer Anal Methods Geomech 2014;38
(10):1078–100.
[95] Zhao G-F, Khalili N, Fang J, Zhao J. A coupled distinct lattice spring model for rock failure under dynamic loads. Comput Geotech 2012;42:1–20.
[96] Cusatis G, Mencarelli A, Pelessone D, Baylot JT. Lattice discrete particle model (LDPM) for fracture dynamics and rate effect in concrete. In: Structures
Congress 2008: 18th Analysis and Computation Specialty Conference. 2008: p. 1–11.
[97] Cusatis G, Pelessone D, Baylot JT. Dynamic Pull-out Test Simulations Using the Lattice Discrete Particle Model (LDPM). In: Structures Congress 2008:
18th Analysis and Computation Specialty Conference. 2008.
[98] Alnaggar M, Cusatis G, Luzio GD. Lattice discrete particle modeling (LDPM) of alkali silica reaction (ASR) deterioration of concrete structures. Cem
Concr Compos 2013;41(8):45–59.
[99] Smith J, Cusatis G, Pelessone D, O’Daniel J, Baylot J. Calibration and Validation of the Lattice Discrete Particle Model for Ultra High- Performance Fiber-
Reinforced Concrete. In: 20th Analysis and Computation Specialty Conference. 2012: p. 394–405.
[100] Cusatis G, Mencarelli A, Pelessone D, Baylot J. Lattice discrete particle model (LDPM) for failure behavior of concrete. II: calibration and validation.
Cem Concr Compos 2011;33(9):891–905.
[101] Zhao G-F, Fang J, Sun L, Zhao J. Parallelization of the distinct lattice spring model. Int J Numer Anal Methods Geomech 2013;37(1):51–74.
[102] Zhao G, Khalili N. Graphics processing unit based parallelization of the distinct lattice spring model. Comput Geotech 2012;42:109–17.
[103] Zhu JB, Zhao GF, Zhao XB, Zhao J. Validation study of the distinct lattice spring model (DLSM) on P-wave propagation across multiple parallel joints.
Comput Geotech 2011;38(2):298–304.
[104] Ma J, Zhao G. Studying the influence of heterogeneity on particle breakage using distinct lattice spring model. Arab J Geosci 2015;8(9):6595–621.
[105] Nooru-Mohamed MB. Mixed-mode fracture of concrete: an experimental approach. 1992.
[106] Arslan A, Schlangen E, Van MJGM. Effect of Model Fracture Law and Porosity on Tensile Softening of Concrete. In: Int. Conf. on Fracture Mechanics of
Concrete and Concrete Structures, Framcos; 1995.
[107] Kozicki J, Tejchman J. Effect of aggregate structure on fracture process in concrete using 2D lattice model. Arch Mech 2007;59(4):365–84.
[108] Raghuprasad BK, Bhat DN, Bhattacharya GS. Simulation of fracture in a quasi-brittle material in direct tension – A lattice model. Eng Fract Mech
1998;61(3–4):445–60.
[109] Zhao G-F, Fang J, Zhao J. A MLS-based lattice spring model for simulating elasticity of materials. Int J Comput Methods 2012;9(3):1250037.
[110] Abreu M, Lemos J, Carmeliet J, Schlangen E. Modelling compressive cracking in concrete using a modified lattice model. In: Fracture mechanics of
concrete and concrete structures—new trends in fracture mechanics of concrete. Taylor & Francis Group, London; 2007.
[111] Iosipescu N. New accurate procedure for single shear testing of metals. J Mater 1967;2:537–66.
[112] Brighenti R, Carpinteri A, Spagnoli A, Scorza D. Continuous and lattice models to describe crack paths in brittle-matrix composites with random and
unidirectional fibres. Eng Fract Mech 2013;108:170–82.
[113] Brighenti R, Carpinteri A, Spagnoli A, Scorza D. Cracking behaviour of fibre-reinforced cementitious composites: a comparison between a continuous
and a discrete computational approach. Eng Fract Mech 2013;103:103–14.
[114] Zhao G-F. Modelling 3D jointed rock masses using a lattice spring model. Int J Rock Mech Min Sci 2015;78:79–90.
[115] Reichert T, Ridley-Ellis D. Computational issues regarding lattice models for wood. Trees & Timber Institute National Research Council; 2010.
[116] Vasić S, Ceccotti A, Smith I, Sandak J. Deformation rates effects in softwoods: Crack dynamics with lattice fracture modelling. Eng Fract Mech 2009;76
(9):1231–46.
[117] Sedighi-Gilani M, Navi P. Micromechanical approach to wood fracture by three-dimensional mixed lattice-continuum model at fiber level. Wood Sci
Technol 2007;41(7):619–34.
[118] Fournier Christopher R, Davids William G, Nagy E, Landis Eric N. Morphological lattice models for the simulation of softwood failure and fracture, in
Holzforschung. 2007; p. 360.
[119] Smith I, Snow M, Asiz A, Vasic S. Failure mechanisms in wood-based materials: A review of discrete, continuum, and hybrid finite-element
representations, in Holzforschung. 2007. p. 352.
[120] Vasic S, Smith I, Landis E. Finite element techniques and models for wood fracture mechanics. Wood Sci Technol 2005;39(1):3–17.
[121] Davids WG, Landis EN, Vasic S. Lattice models for the prediction of load-induced failure and damage in wood. Wood & Fiber Sci J Soc Wood Sci
Technol 2003;35(1):120–34.
[122] Landis EN, Vasic S, Davids WG, Parrod P. Coupled experiments and simulations of microstructural damage in wood. Exp Mech 2002;42(4):389–94.
[123] Ayatollahi MR, Aliha MRM. Analysis of a new specimen for mixed mode fracture tests on brittle materials. Eng Fract Mech 2009;76(11):1563–73.
[124] Ostoja-Starzewski M, Stahl DC. Random fiber networks and special elastic orthotropy of paper. J Elast Phys Sci Solids 2000;60(2):131–49.
[125] Boyina D, Kirubakaran T, Banerjee A, Velmurugan R. Mixed-mode translaminar fracture of woven composites using a heterogeneous spring network.
Mech Mater 2015;91(Part 1):64–75.
[126] Mayya A, Praveen P, Banerjee A, Rajesh R. Splitting fracture in bovine bone using a porosity-based spring network model. J R Soc Interface 2016;13
(124):20160809.
[127] Li VC, Chan C-M, Leung CKY. Experimental determination of the tension-softening relations for cementitious composites. Cem Concr Res 1987;17
(3):441–52.
[128] Nomura N, Mihashi H, Izumi M. Correlation of fracture process zone and tension softening behavior in concrete. Cem Concr Res 1991;21(4):545–50.
[129] Huang J, Li VC. A meso-mechanical model of the tensile behaviour of concrete. Part II: modelling of post-peak tension softening behaviour.
Composites 1989;20(4):370–8.
[130] Prado EP, van Mier JGM. Effect of particle structure on mode I fracture process in concrete. Eng Fract Mech 2003;70(14):1793–807.
[131] Yao W, Li J, Wu K. Mechanical properties of hybrid fiber-reinforced concrete at low fiber volume fraction. Cem Concr Res 2003;33(1):27–30.
[132] Zollo RF. Fiber-reinforced concrete: an overview after 30 years of development. Cem Concr Compos 1997;19(2):107–22.
[133] Lee MK, Barr BIG. Strength and fracture properties of industrially prepared steel fibre reinforced concrete. Cem Concr Compos 2003;25(3):321–32.
[134] Reichert T, Ridley-Ellis D. 3D Lattice Model for Post-Yield and Fracture Behaviour of Timber; 2008.
[135] Reichert T. Development of 3D lattice models for predicting nonlinear timber joint behaviour. 2009.
[136] Jonkers H. Bacteria-based self-healing concrete. Heron 2011;56(1/2):1–12.
[137] Srivatsan TS. Microstructure, tensile properties and fracture behaviour of aluminium alloy 7150. J Mater Sci 1992;27(17):4772–81.
[138] Pardoen T, Dumont D, Deschamps A, Brechet Y. Grain boundary versus transgranular ductile failure. J Mech Phys Solids 2003;51(4):637–65.
[139] Gräf M, Hornbogen E. Observation of ductile intercrystalline fracture of an Al-Zn-Mg-alloy. Acta Metall 1977;25(8):883–9.
[140] Panfilov P, Yermakov A. Brittle intercrystalline fracture in iridium. Platinum Met Rev 2001;45(4):179–83.
[141] Panfilov P, Yermakov A. Mechanisms of inherent and impurity-induced brittle intercrystalline fracture in pure FCC-metal iridium. Int J Fract 2004;128
(1):147–51.
[142] Liu S, Liu D, Liu S. Transgranular fracture in low temperature brittle fracture of high nitrogen austenitic steel. J Mater Sci 2007;42(17):7514–9.
[143] Guo L-P, Carpinteri A, Spagnoli A, Sun W. Experimental and numerical investigations on fatigue damage propagation and life prediction of high-
performance concrete containing reactive mineral admixtures. Int J Fatigue 2010;32(2):227–37.
[144] Imbabi MS, Carrigan C, McKenna S. Trends and developments in green cement and concrete technology. Int J Sustain Built Environ 2012;1
(2):194–216.
[145] Zhang MH. Canmet. Microstructure, crack propagation, and mechanical properties of cement pastes containing high volumes of fly ashes. Cem Concr
Res 1995;25(6):1165–78.
[146] Ammasi A, Santhi AS, Ganesh GM. Various utilization of fly ash and its properties on concrete-a review. Int J Emerg Trends Eng Dev 2012;2
(2):435–44.
[147] Özbay E, Erdemir M, Durmusß HI. _ Utilization and efficiency of ground granulated blast furnace slag on concrete properties – A review. Constr Build
Mater 2016;105:423–34.
[148] Chung DDL. Review: improving cement-based materials by using silica fume. J Mater Sci 2002;37(4):673–82.
[149] Singh B, Ishwarya G, Gupta M, Bhattacharyya SK. Geopolymer concrete: a review of some recent developments. Constr Build Mater 2015;85:78–90.
[150] AïTcin PC. The durability characteristics of high performance concrete: a review. Cem. Concr. Compos. 2003. 25(4–5): 409–420.
[151] Shi C, Wu Z, Xiao J, Wang D, Huang Z, Fang Z. A review on ultra high performance concrete: Part I. Raw materials and mixture design. Constr Build
Mater 2015;101:741–51.
[152] Wang D, Shi C, Wu Z, Xiao J, Huang Z, Fang Z. A review on ultra high performance concrete: Part II. Hydration, microstructure and properties. Constr
Build Mater 2015;96:368–77.
[153] Foster SJ, Attard MM. Strength and ductility of fiber-reinforced high-strength concrete columns. J Struct Eng 2001;127(1):28–34.
[154] Talaiekhozan A, Keyvanfar A, Shafaghat A, Andalib R, Majid A, Fulazzaky MA, et al. A review of self-healing concrete research development. J Environ
Treat Tech 2014;2(1):1–11.
[155] Mihashi H, Nishiwaki T. Development of engineered self-healing and self-repairing concrete-state-of-the-art report. ACT 2012;10(5):170–84.
[156] Shen WQ, Shao JF. A micromechanical model of inherently anisotropic rocks. Comput. Geotech. 2015; vol. 65(Supplement C): p. 73–9.
[157] Brely L, Bosia F, Pugno N. A hierarchical lattice spring model to simulate the mechanics of 2-D materials-based composites. Front Mater 2015;2:51.
[158] Bellenger E, Coorevits P. Adaptive mesh refinement for the control of cost and quality in finite element analysis. Finite Elem Anal Des 2005;41
(15):1413–40.
[159] Rong F, Xia M, Ke F, Bai Y. Adaptive mesh refinement FEM for damage evolution of heterogeneous brittle media. Modell Simul Mater Sci Eng 2005;13
(5):771–82.
[160] Rajasekaran S, Remeshan V. Adaptive mesh refinement in finite element analysis. Indian J Eng Mater Sci 1999;6(3):135–43.
[161] Rheinboldt WC. Adaptive mesh refinement processes for finite element solutions. Int J Numer Methods Eng 1981;17(5):649–62.

Pan - 2018 - A Review of Lattice Type Model in Fracture Mechanics

Încărcat de

Informații document

Titlu original

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Pan - 2018 - A Review of Lattice Type Model in Fracture Mechanics

Încărcat de

Drepturi de autor:

Formate disponibile

Engineering Fracture Mechanics 190 (2018) 382–409

Contents lists available at ScienceDirect

Engineering Fracture Mechanics

A review of lattice type model in fracture mechanics: theory,

r nominal stress of element in failure criterion

2.1. Definition of terms

Fig. 2. Examples of regular lattice networks.

2.2. General philosophy and framework

2.2.2. Element type

Fig. 3. Main framework of lattice model.

Fig. 4. Types of elements used in lattice model.

2.2.3. Constitutive relation

2.2.4. Failure criterion

2.2.5. Element removal

2.3. Recent developments

2.3.1. Confinement–shear lattice (CSL) model

Fig. 5. Schematic of theoretical basis of CSL based on [27].

Fig. 7. DLSM analysis procedure [30].

2.3.3. Nonlocal LSM

3.1. Benchmarks of classic tests

Fig. 8. Schematics of different orders of neighboring particles in 2D nonlocal LSM [31].

Name Major novelties and contributions References

3.1.1. Uniaxial tensile test

Uniaxial tensile test with single LBM [3,31,57,83,109]

Z. Pan et al. / Engineering Fracture Mechanics 190 (2018) 382–409

Shear cracking test Nonlocal LSM [31]

Four-point shear test LBM [3,13,44]

Uniaxial compression LBM [22,43,45,68,79,107,110]

Nooru–Mohamed test LBM [3,14,15,44,56,75]

Diagonally loaded square plate test Nonlocal LSM [90]

Z. Pan et al. / Engineering Fracture Mechanics 190 (2018) 382–409

3.1.2. Three-point bending test

3.1.3. Four-point shear test

3.1.4. Uniaxial compressive test

3.1.5. Nooru–Mohamed test

3.2. Types of materials

3.2.2. Reinforced concrete

3.2.3. Fiber-reinforced concrete

Fig. 13. Typical constitutive relations used in lattice model.

incorporated—constituting a novelty in comparison to previous bilinear relations—resulting in a time-dependent constitu-

3.3. Type of loads

3.3.1. Static loads

3.3.2. Quasistatic loads

3.3.3. Dynamic loads

4. Challenges and perspectives

4.1. Theory development

4.1.1. Interdisciplinary innovations

4.1.2. New materials

4.2. Applications in engineering practices

4.2.1. Multiscale method

4.2.2. Parallel computing

4.2.3. Adaptive mesh refinement

Fig. 22. Schematic of AMR in mesostructure-based lattice models.

Appendix A. Supplementary material

S-ar putea să vă placă și