Engineering Fracture Mechanics 283 (2023) 109196

Contents lists available at ScienceDirect

Engineering Fracture Mechanics

journal homepage: www.elsevier.com/locate/engfracmech

A reformulated non-ordinary state-based peridynamic method for

dynamic failure of ductile materials
Liwei Wu a, Han Wang b, Dan Huang a, *, Xin Cai a, Junbin Guo b
a
College of Mechanics and Materials, Hohai University, Nanjing 211100, China
b
Xi’an Research Institution of High Technology, Xi’an 710025, China

A R T I C L E I N F O A B S T R A C T

Keywords: A non-local correspondence peridynamic model is reformulated to describe the dynamic failure of
Ductile fracture ductile materials, under the framework of non-ordinary state-based peridynamics. To eliminate
Impact failure the zero-energy mode as well as to improve the numerical stability, the deformation gradient is
Crack propagation
updated on the bond level, which owns information including both the non-local deformation and
Adiabatic shear band
Peridynamics
the bond deviation. Peridynamic thermo-viscoplastic constitutive formulations are presented for
the failure analyses of ductile materials. Moreover, a mixed strain-stress criterion describing the
bond breaking and considering the temperature effect is presented on the bond level, in which the
bonds can keep intact in calculations. Several comprehensive cases, including a notched tension
test and Kalthoff-Winkler impact tests with different velocities, are investigated to verify the
applicability of the proposed model for ductile fracture and impact failure. Numerical results, in
terms of failure patterns, cracking switching and propagation, as well as adiabatic shear band
propagation, show a reasonable agreement with corresponding experimental observations.

1. Introduction

Ductile materials play a significant role in engineering and the failure mechanism of ductile materials has attracted the focus and
concern of researchers from various fields. One of the main characteristics of ductile failure is dissipation, that heat caused by the
plastic deformation can induce thermal softening, which has a nonnegligible effect on ductile crack propagation. Besides, on account of
the complicated failure mechanism, dynamic ductile failure has been one important topic of contemporary research, especially for
high-velocity machining, penetration, shear localization, etc.
Compared with specific experiments for ductile materials, numerical simulations have their own advantages being much more
flexible and economical, especially in cases where experimental tests are relatively difficult to be conducted. In the past decades, many
numerical models and approaches have been successfully proposed and used. The classical numerical methods [1], as an ideal choice
for solid mechanics with no discontinuities involved, are limited in the study of dynamic ductile failure. To have better solutions to
discontinuous problems, various achievements have been obtained, such as the cracking particle method (CPM) [2], mesh-less
Galerkin method (MSGM) [3], smoothed particle hydrodynamics method (SPH) [4], material point method (MPM) [5], and phase
field model [6]. A non-local theory, as a reformulation of traditional continuum mechanics, was proposed and referred to as peri­
dynamics (PD) [7,8].
Peridynamic theory was proposed [7] to overcome the limitations of classical methods when solving discontinuous problems. The

* Corresponding author.
E-mail address: [email protected] (D. Huang).

https://doi.org/10.1016/j.engfracmech.2023.109196
Received 29 October 2022; Received in revised form 3 February 2023; Accepted 10 March 2023
Available online 26 March 2023
0013-7944/© 2023 Elsevier Ltd. All rights reserved.
L. Wu et al. Engineering Fracture Mechanics 283 (2023) 109196

Fig. 1. Schematic diagram of the interaction between material points in peridynamics.

main advantage of peridynamic theory is that it replaces the differential equation with the spatial integral equation, which is greatly
significant for describing and handling discontinuities. Another key characteristic of peridynamics is the description of damage and
failure [9]. The damage is formed when the interaction between material points disappears, meaning that cracks could initiate and
propagate spontaneously without further assumptions. With foregoing inherent features, peridynamics exhibits its advantageous
capability and efficiency in handling various discontinuous problems with complicated failure morphologies, including fatigue
[10,11], corrosion [12,13], dynamic fracture [14,15], impact penetration [16,17], blast failure [18,19], etc.
The original model in peridynamics is called the bond-based peridynamic model, where an oversimplification is used that inter­
action between material points only depends on a central potential function instead of other local conditions. In consequence, the
bond-based model owns a limitation of Poisson ratio, and its application range and calculation accuracy are restricted as well. To
overcome the aforesaid restriction, peridynamic formulations were reformulated into new ones, referred to as state-based peridy­
namics [20]. Two types are defined as ordinary state-based (OSB) and non-ordinary state-based (NOSB) peridynamics. It is noteworthy
that the latter model is more generalized than the former one [21]. Moreover, non-ordinary state-based peridynamics could directly
establish the relationship with mature constitutive models from classical continuum mechanics, such as the Johnson-Cook (JC) model
[22], Holmquist-Johnson-Cook (HJC) model [23], Johnson-Holmquist II (JH-2) model [24], etc.
In non-ordinary state-based peridynamics, the zero-energy mode is common in many works and can cause calculating errors for
certain complex problems. It is of great significance to eliminate the zero-energy mode [25] on improving calculation stability. Besides,
failure in some works through the traditional non-ordinary peridynamic framework is depicted by removing the total damage point.
Once the damage reaches a critical value, the material point will no longer participate in the subsequent calculation. Nevertheless, this
way will neglect a series of bonds that should keep intact in the actual simulation. Recently, the bond-breaking criterion on the bond
level was used in some studies [26,27]. In this study, we aim at reformulating correspondence peridynamic formulations under the
framework of non-ordinary state-based peridynamics. On the one hand, the velocity gradient and deformation gradient in this study
are based on the bond level, which owns information about the non-local deformation as well as the bond deviation. On the other hand,
compared with the material point level, bond breaking on the bond level is undoubtedly more accurate and effective. A mixed strain-
stress criterion on the bond level is presented here, which considers the effect of temperature as well. The proposed peridynamic model
and method are then employed in typical ductile failure cases, including a notched tension test, and low-velocity and high-velocity
Kalthoff-Winkler impact experiments. The validation is established by comparing peridynamic predictions with corresponding
experimental data and other available numerical results.
This article is organized as follows: The details of the basic peridynamic theory and reformulated formulations of the correspon­
dence model are presented in Section 2. The damage model of correspondence peridynamics on the bond level is described in Section 3.
Numerical implementation and contact algorithm in peridynamics are introduced in Sections 4 and 5. In Section 6, numerical vali­
dation and demonstration are discussed through a notched tension test and Kalthoff-Winkler impact tests. Conclusions are summarized
in Section 7.

2. Peridynamic correspondence model

2.1. Basic formulations in non-ordinary state-based peridynamics

Peridynamics is a reformulation of continuum mechanics theory, which is based on the spatial integral equation other than the
spatial differential equation. As shown in Fig. 1, any material point interacts with others within a finite area. In state-based peridy­
namics [20], the equation of motion is given by:

2
L. Wu et al. Engineering Fracture Mechanics 283 (2023) 109196

Fig. 2. Schematic diagram of the semi-Lagrangian peridynamics.

∫
ρü(x, t) = {T[x, t]〈x’ − x〉 − T[x’ , t]〈x − x’ 〉 } dVx’ + b(x, t) (1)
Hx

where ρ and b denote the material mass density and external body force density, respectively. x’ is the relevant material point within
the compact horizon Hx of x, while the size of horizon is taken to be a sphere of radius δ, defined as Hx = {0 < |x − x’ |〈δ }, where δ > 0.
u(x, t) is the displacement field, and dVx’ is the volume associated with material point x .
′

Besides, the vector ξ between x and x’ is referred to as the bond, which is defined as ξ = x’ − x. The position of x in the deformed
configuration is denoted by y(x, t) = x + u(x, t), and the deformation vector state Y is expressed as:
Y[x, t]〈ξ〉 = y(x + ξ, t) − y(x, t) (2)
In the peridynamic correspondence model, the deformation gradient tensor is expressed as:
∫
F[x, t] = ω(|ξ|)(Y〈ξ〉 ⊗ ξ )dVx’ K− 1 (3)
Hx

where ω(|ξ| ) is a scalar influence function that is related to ‖ξ‖. As a non-local shape tensor, K is defined by:
∫
K[x, t] = ω(|ξ|)(ξ ⊗ ξ)dVx’ (4)
Hx

According to classical continuum mechanics, the first Piola-Kirchhoff stress can be obtained as follows:

P = (J σ )F− T
(5)

where σ denotes the rotated Cauchy stress.

Finally, the force vector state is calculated as:

T = ωPK− 1 ξ (6)

2.2. Peridynamic velocity gradient on bond level

With the motivation of the main idea in [28], this study is formulated through the semi-Lagrangian non-ordinary state-based
peridynamic framework, instead of the above traditional Lagrangian peridynamic theory. As shown in Fig. 2, the movement of ma­
terial points is still recorded during the deformation process. In other words, various mechanical properties of bonds are only related to
the current configuration. That is why it is called the semi-Lagrangian theory. More details of the semi-Lagrangian model are illustrated
in [28].
Similar to Eq. (1), the equation of motion here is rewritten as [28]:
Dẋ
ρ =f +b
Dt
∫ (7)
f= {T’ [x, t]〈η〉 − T’ [x + η, t]〈 − η〉 } dη
Hx

where f is the internal force density. η denotes the bond in the current configuration, and T ’ is the force vector state in the current
configuration as well. The difference between T and T ’ is that the former can be regarded as the first Piola-Kirchhoff stress (original
configuration) and the latter as the Cauchy stress (current configuration).
In [29], the spatial derivative operator in peridynamics is proposed as:

3
L. Wu et al. Engineering Fracture Mechanics 283 (2023) 109196

(∫ )
D{g} = M − 1
ω(η)ϛ(g) < η > ⊗ ηdη (8)
Hx

where g is a given function, and satisfied as:

ϛ(g) < η > = ϛ(g)[x] < η > = g(x + η) − g(x) (9)
The purpose of the shape tensor M is to reduce the surface effect in peridynamics, which is defined as:
∫
M= ω(η)(η ⊗ η)dη (10)
Hx

To compute the non-local velocity gradient L on the material point level, all associated bonds in its horizon region should be
considered to calculate an average over them. Substitute L in Eq. (9),
(∫ )
L = M− 1 ω(η)(η̇ ⊗ η)dη (11)
Hx

In the peridynamic correspondence model [21], the effect of zero energy modes is a non-ignorable issue. To enhance the numerical
stability, the velocity gradient in this work is on the bond level from the semi-Lagrangian peridynamic theory [28],
( )
L(x) + L(x + η) L(x) + L(x + η) η
L(x)〈η〉 = + η̇ − ⋅η ⊗ 2 (12)
2 2 |η|

where L and L is the velocity gradient on the bond and material point level, respectively. For L, there is a weak relationship between
the force density and bond’s individual deformation. This is why the velocity gradient on the bond level is chosen here. Moreover, it is
noted that L is symmetric, which means once a bond breaks, its counterpart bond breaks as well.

2.3. Peridynamic thermo-viscoplastic constitutive formulations

According to L in Eq. (11), the rate of deformation tensor D and spin tensor W are expressed by:
1( )
D= L + LT (13)
2

1( )
W= L − LT (14)
2
For thermo-viscoplastic deformation, D could be divided into:

D = De + Dvp + DT (15)

where De ,Dvp , and DT is the elastic, visco-plastic, and thermal deformation rate, respectively.
Heat is mainly transformed from the plastic work, and the energy equation of strong rate form is:
∂T ( )
ρCp = χτ : Dvp + ∇x JF− 1 ⋅ κ ⋅ F− T ⋅ ∇x T (16)
∂t

where Cp denotes the specific heat,χ is a Taylor-Quinney coefficient,τ is the second Kirchhoff stress, and κ denotes the heat conductivity
tensor. In the high-velocity impact process, the effect of heat conduction can be ignored, and only adiabatic heating needs to be
considered.
∂T
ρCp = χτ : Dvp (17)
∂t
In [28], the velocity gradient is considered in updating of stress, which has good accuracy and stability. Inspired by this work, we
use the unrotated rate of deformation tensor d to calculate the unrotated Cauchy stress Γ. Then, the Cauchy stress tensor σ can be
obtained by the rotation tensor R[30,31]. The right-handed rate of rotation (or angular velocity) of a material point is represented as:
Ω = eijk ω (18)

where eijk is the permutation tensor, and ω is the angular velocity vector,

ω = w + [trace(Vt )I − Vt ]− 1 z (19)

zi = eijk Djm Vmk (20)

w is a vector corresponding to spin tensor W,

4
L. Wu et al. Engineering Fracture Mechanics 283 (2023) 109196

w = − 0.5eijk W jk (21)

The deformation gradient can be described by the left stretch tensor and rotation tensor,
F = VR (22)

and V can be calculated as:

V n+1 = V n + (LV n − V n Ω)Δt (23)
R is an orthogonal tensor, which describes the rotation of a rigid body at a time,
sin(ΔtΩ) 1 − cos(ΔtΩ) 2
Rn+1 = [I + Ω− Ω ]Rn (24)
Ω Ω2
Then, the unrotated rate of deformation tensor can be expressed by rate of deformation tensor and rotation tensor,

d = de + dvp + dT = RT DR (25)
On account of the von-Mises plasticity theory and unrotated rate of deformation tensor, the strain increment can be calculated by:
Δeij = dij Δt (26)

Deviatoric strain increment is:

1
Δedev
ij = Δeij − tr(Δeij )δij (27)
3
The trial unrotated Cauchy stress is given by:

Γ trial
n+1 = Γ n + htr(Δe)I + 2μΔe
dev
(28)

where h and μ are Laḿ e’s constants.

The trial deviatoric stress tensor, based on Von-Mises plasticity theory, can be calculated as:
1
Γdev trial
ij = Γij − tr(Γtrial
ij )δij (29)
3
The trial von-Mises yield stress is expressed by:
3
σ2 = Γ trial : Γ trial (30)
2
When σ < σ0 , the step is in the elastic state, and trial unrotated Cauchy stress can be regarded as unrotated Cauchy stress. When
σ⩾σ0 , the yield stress σyield and Von-Mises equivalent stress are calculated, respectively. If σ ⩾σyield , the current step is in the plastic state,
and the plastic deformation should be considered. Otherwise, the step is still in the elastic state. The numerical calculation of σ yield
refers to Zhou’s work [32].
∂f
Dij = κ(σ, ε, T)
∂σij
[ ]m
σ
κ = ε̇0
g(ε, T) (31)
( )
( )
ε n( ( ))
T− T0
κ
g(ε, T) = σ 0 1 + 1 − δ0 e − 1
ε0
The equivalent plastic strain increment needs to be calculated and updated by Eq. (31):
⎛ √̅̅̅ ⎞
2 ε
⎜ σ0 (1 + )n (1− δ0 ⎟
⎜ 3 ε0 ⎟
1 ⎜ ⎟
Δe − Δλ − ⎜
⎜
⎛ ⎞1/m ⎟=0
⎟ (32)
2μ ⎜ T− T0 ε̇ ⎟
⎝ (e ( κ )
− 1)) ⎝ ⎠ dev
− Γn ⎠
ε̇0

√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
Γdev
n = Γ dev
n : Γn
dev
(33)

Δen+1 : Γ dev
Δe = n+1
(34)
Γn+1
By solving the implicit equation in Eq. (31) with the Bi-section method,Δλ can be obtained.

5
L. Wu et al. Engineering Fracture Mechanics 283 (2023) 109196

Fig. 3. Impact process of separated blocks: (a) contact model is not considered; (b) contact model is considered.

The Cauchy stress tensor σ can be expressed as:

σ = RT ΓR (35)
To establish the relationship between the Cauchy stress tensor and force vector state on the bond level, the force vector state is
written as follows [28]:

ω〈η〉 η
Ti = ∫ σ ij 〈η〉 j 2 +
H(x)
ω〈η 〉d η ‖η ‖ (36)

⎡ ( )⎤
1 1
∫ +∫
⎢∫ ω〈η〉dη ω〈η + ξ〉d η ⎥
1⎢⎢ H(x) H(x)
⎥
⎥ −1
⎢ ( ) ⎥Mlp ω〈η〉ηl
2 ⎢ H(x) ⎥
⎣ ξξ ⎦
ω〈ξ〉σ ij 〈ξ〉 δpj − p 2j dξ
|ξ|

3. Damage model of correspondence peridynamics on bond level

In original peridynamics, damage of a material point is given locally by the degraded weighted volume, that the ratio of broken
bonds to total bonds of that point in its horizon [9], expressed as:
∫
μ(x, ξ, t)dVx′
D(x, t) = 1 − Hx ∫ (37)
Hx
dVx′

where μ(x, ξ, t) is a Boolean function of the bond,

6
L. Wu et al. Engineering Fracture Mechanics 283 (2023) 109196

Fig. 4. The loading conditions and geometric parameters of the notched test specimen.

Table 1
Material parameters of notched specimen.
Parameters Value

E 67GPa
ν 0.33
ρ 2700 Kg/m3
σ0 225 MPa
ε0 0.01
ε̇ 0 0.001
m 50
n 0.1
T0 293 K
δ0 0.8
κ 925 K
Cp 90.6 J/(KgK)
α 2.4e− 5/K
χ 0.9
σcr 330 MPa

{
1 unbroken
μ(x, ξ, t) = (38)
0 broken

One damage model for the traditional correspondence peridynamic model was presented in [25]. In such a model, a material point
will no longer participate in the subsequent calculation and all associated bonds in its horizon would be broken once the damage
reaches a critical value. Nevertheless, this approach will lose a finite amount of bonds that should keep intact in the actual simulation.
Compared with the material point level, bond breaking on the bond level is undoubtedly more accurate and effective.
In Marchand and Duffy’s experimental tests [33] for dynamic ductile failure, there was an attractive phenomenon that the shear-

7
L. Wu et al. Engineering Fracture Mechanics 283 (2023) 109196

Fig. 5. Final failure pattern of notched ductile specimen, left: experimental observation [38], right: peridynamic prediction.

Fig. 6. Crack propagation process of ductile specimen in time sequence.

carrying capacity of ductile material disappears in a specific stage. This phenomenon is referred to as stress collapse, which is also
theoretically predicted in [34]. When stress collapse occurs, the corresponding bond is weakened, and the equivalent plastic strain
reaches the critical strain, given as [32]:

8
L. Wu et al. Engineering Fracture Mechanics 283 (2023) 109196

Fig. 7. The curve of load displacement by original peridynamic model and proposed model.

Fig. 8. Final failure patterns of notched ductile specimen, left: original correspondence peridynamic model, right: proposed peridynamic model in
this study.

ε̇r
εcr = ε1 + (ε2 − ε1 ) (39)
ε̇r + ε̇

where εcr denotes the critical strain and ε̇r is a rate-dependent parameter. Once the equivalent plastic strain ε reaches the critical strain,
the carrying capacity of bond includes two parts, hydrostatic pressure and viscous stress[32], so the Kirchhoff stress can be expressed
as:

9
L. Wu et al. Engineering Fracture Mechanics 283 (2023) 109196

Fig. 9. Crack pattern with different horizons in the δ-convergence study.

Fig. 10. The curve of load displacement with different horizons in the δ-convergence study.

γ[1 − J + α(T − T0 ) ] E
τ 〈η〉 = I + μd D (40)
J ν− 1

where γ is a stiffness parameter and α is a coefficient of thermal expansion.μd is a temperature-dependent viscosity parameter. When
the hydrostatic component of the stress is negative, the bond breaks,
τ 〈η〉 : I< 0 (41)

4. Numerical implementation

Through the standard spatial discretization methods, Eq. (1) can be discretized as:
∑{ [ ]〈 〉 [ ]〈 〉} ( )
ρüni = T xni , tn xnp − xni − T xnp , tn xni − xnp Vp + b xni , tn (42)
p∈Hx

where i and p denote the discretization number of material points, Vp = |Δx|3 denotes the volume of xp , and n is the time step number.

10
L. Wu et al. Engineering Fracture Mechanics 283 (2023) 109196

Fig. 11. Crack pattern with different horizon factors in the m-convergence study.

Fig. 12. The curve of load displacement with different horizon factors in the m-convergence study.

To meet the requirement of numerical cases for dynamic fracture and failure in this study, the explicit Verlet-Velocity method [35]
is used on account of its great stability and accuracy:
Δt n
u̇n+0.5
i = u̇ni + ü
2 i
Δt (43)
u̇n+1
i = u̇n+0.5
i + ün+1
2 i
ui = ui + Δtu̇n+0.5
n+1 n
i

where Δt is the unit time step, which should satisfy Eq. (46) according to [36]:
δ
Δt∝ ′ (44)
c
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
that c = (λ + 2μ)/ρ is the stress wave speed.
′

11
L. Wu et al. Engineering Fracture Mechanics 283 (2023) 109196

Fig. 13. Geometric model and boundary condition of Kalthoff-Winkler impact test.

Table 2
Material parameters of steel C300.
Parameters Value

E 200 GPa
ν 0.3
ρ 7830 Kg/m3
σ0 2000 MPa
ε0 0.01
ε̇0 0.001
m 70
N 0.01
T0 293 K
δ0 0.8
Κ 500 K
Cp 448 J/(KgK)
А 11.2e− 6/K
Х 0.9
ε1 4ε0
ε2 0.3
ε̇r 4.0e4/s
γ 0.002
μd 5.0 Pa⋅s
σcr 3σ0

5. Contact algorithm in peridynamics

To prevent the infiltration phenomenon occurs in different material points from separated blocks, one contact algorithm was
proposed in [37]. As shown in Fig. 3(a), if no contact force is considered among different material points, the impactor could penetrate
the target easily without any hindrance. Whereas, if the repulsive force from a peridynamic contact model is considered, the normal
contact behavior between the impactor and target can be described, as shown in Fig. 3(b).
The detailed contact force is presented as:
( ) csh ⃦ ⃦ yp − yi
f s yp , yi = min{0, (⃦yp − yi ⃦ − dpi )} ⃦
⃦yp − yi ⃦
⃦ (45)
δ
( )
where yp ,yi denote the position of material points. csh is a contact constant in peridynamics (Generally csh = 15c, c = (12E)/ πδ4 [37]).
dpi is a critical value to determine one point whether touches another point or not.

(46)
′
dpi = min{0.9‖x − x‖, 1.35|Δx| }

6. Numerical validation and demonstration

In this section, the numerical ability of the presented peridynamic model for ductile fracture and impact failure is demonstrated
through several typical examples. In the first case, one tensile test of a notched ductile specimen is simulated and predicted results are

12
L. Wu et al. Engineering Fracture Mechanics 283 (2023) 109196

Fig. 14. Comparisons between experimental observations and peridynamic results for damage patterns in the low-velocity case.

compared against experimental data. Moreover, Kalthoff-Winkler experiments, subjected to low-velocity and high-velocity impact
loadings, are numerically performed here, and different failure modes are analyzed and discussed.

6.1. Tensile test of a notched specimen

For a tensile test of a notched ductile plate in [38], geometry and boundary conditions in this study are the same as that in
experimental tests, which are shown in Fig. 4. There is a hollow notch at the center of the rectangular plate. The notch length of
specimen is 22 mm, inclination angle is 30◦ , and radius of notch is 2 mm. Same to the displacement-control conditions in experimental
tests, the boundary conditions in this study are implemented by displacement loading with a speed of 1 mm/min. The material pa­
rameters of ductile specimen are listed in Table 1. In this case, the loading condition can be regarded as a quasi-static tensile loading.
The bond breaking is determined by the maximum tensile stress, instead of the adiabatic shearing criterion for high-velocity impact
conditions. If the maximum principal stress is greater than σ cr , the failure occurs. In peridynamic simulation of this numerical case, the
horizon size δ is set to be 0.04 mm, and δ = 4Δx. The time step Δt = 1.0 × 10− 7 s.
The failure patterns obtained by peridynamic simulations are presented in Fig. 5, where the results measured from experimental
tests [38] are illustrated as well. Fig. 5 shows a satisfactory agreement on failure patterns from experimental observations and peri­
dynamic predictions. Fig. 6 illustrates the crack propagation process in time sequences, where cracks initiate from the notch tip, and
propagate symmetrically toward the edge of plate until the plate is completely damaged.
The load-displacement curve obtained by the proposed model is illustrated in Fig. 7 as the black line and blue area. The fluctuant
portion of the curve after the first peak means that plastic deformation occurs in the specimen. The cracks initiate at the moment when
load reaches the maximum peak point. Then, cracks gradually propagate from the notch tip, along with a falling tendency of the load-
displacement curve. Once the specimen is totally separated, the load-displacement curve rapidly drops to zero, indicating the specimen
loses its load capacity. The critical load predicted by peridynamic simulations is 43.5 KN, while the measured data from experimental

13
L. Wu et al. Engineering Fracture Mechanics 283 (2023) 109196

Fig. 15. Crack switching and propagation process in low-velocity test.

tests is 42.79 KN (an average value from three test results of 42,8 KN, 42,98 KN, and 42,6 KN respectively).
We also conduct numerical simulations for this case using the original correspondence visco-plastic model in peridynamics [36]. To
have a better comparison, the load-displacement curve obtained by this model is shown in Fig. 7 as well, which is depicted by the red
line and gray area. According to the comparison in Fig. 7, we can see that the maximum peak force of two curves is nearly identical.
Whereas, failure obtained by the original model is much earlier than that by the proposed model, and in the descent stage, the load-
displacement curve in former case is clearly rougher. The perhaps reason for this phenomenon is the numerical instabilities caused by
the zero-energy mode. Moreover, the final failure patterns of notched ductile specimen from two models are shown in Fig. 8, and main
failure characteristics can be captured by two models. Nevertheless, the specimen is full of numerous scattered damaged areas in
results from the original model, while this phenomenon is not observed in results from the proposed model in this study.
Further, we conduct a convergence study [39] for this tensile test. Figs. 9 and 10 show the results of δ-convergence study with three
different horizon sizes of 2 mm, 3 mm, and 4 mm (keeping the horizon factor fixed m = δ/Δx = 4). In Fig. 9, we can see that numerical
results present the same overall failure pattern with different horizon sizes, which have satisfactory agreements compared with
experimental measurements. Nevertheless, the roughness of the crack path is much more obvious for the smaller horizons. The load-
displacement curves with different horizons are shown in Fig. 10. It is obvious that all results for the δ-convergence study converge
well, just slightly different in the descent portion after the specimen is completely separated. Figs. 11 and 12 show the results of m-
convergence study with three different horizon factors 3, 4, and 5 (keeping the horizon size fixed δ = 2.4 mm). Identical conclusions
can be drawn that results show the same failure patterns with different horizon factors, and load-displacement curves of all cases
converge well.

6.2. Kalthoff-Winkler impact test

Through Kalthoff’s impact tests [40,41], a transition in the failure mode of ductile specimen under different impact velocities was

14
L. Wu et al. Engineering Fracture Mechanics 283 (2023) 109196

Fig. 16. Von Mises stress contours at various times in low-velocity test.

Fig. 17. Failure behavior of target in high-velocity test.

observed. In a lower velocity case, brittle fracture occurs with the initiation and propagation of tensile cracks. When at a higher
velocity case beyond the critical value, ductile failure occurs in the specimen in terms of the adiabatic shear bands.
The Kalthoff-Winkler impact test is numerically simulated and failure mode transitions with different impact velocities are
replicated and discussed in this section. The geometry and boundary conditions of the double-notched specimen and projectile are
shown in Fig. 13. The material parameters of 4340 steel are listed in Table 2. For the peridynamic modeling in this section, the horizon
size δ = 2 mm, δ = 4Δx, and the time step Δt = 6.0 × 10− 8 s is chosen here.

15
L. Wu et al. Engineering Fracture Mechanics 283 (2023) 109196

Fig. 18. Adiabatic shear band propagation in high-velocity test.

Fig. 19. Damage contour obtained from original model (left) and presented model (right).

6.2.1. Case I: Simulation of crack switching and propagation

In such a lower impact velocity of 25 m/s, tensile crack propagation was obtained in experimental tests [40]. As shown in Fig. 14
(a), tensile cracks initiate at an angle of about 70◦ , and typical roughness was observed on the fracture surface. Fig. 14(b) is the
numerical predictions from peridynamic simulations. The predicted behavior, i.e., tensile crack propagation and crack initiation angle
match well with the experimental measurements. From Fig. 14(b), it can be observed that the fracture surface is smooth in the
beginning, nevertheless, it is much rougher during the crack propagation process.
The crack propagation in time sequence is shown in Fig. 15. The tensile cracks occur at about 18 us, and the propagation path is
nearly linear in the beginning, where the vertical direction is about 69.34◦ . Due to the symmetry of geometry and boundary conditions,
two cracks are symmetrically distributed via the ductile specimen. As cracks close to the edge of specimen, a slight bending phe­
nomenon can be observed. The perhaps reason for it is the softening effect of boundaries in simulations. Tensile cracks continuously

16
L. Wu et al. Engineering Fracture Mechanics 283 (2023) 109196

Fig. 20. Von Mises stress contours at various times in high-velocity test.

propagate through the specimen and reach to edges at the time of 78 us.
The von Mises stress contours at various times are shown in Fig. 16. In the beginning, the stress wave is caused when the projectile
impacts the specimen and propagates toward the bottom surface. Then, a region of high stress is subsequently formed near the notch
tip. As shown in Fig. 16(b), the shear band has a propagation trend but arrests, while a tensile crack initiates and propagates from the
notch tip. The continuous propagation of tensile cracks is presented in Fig. 16(c-f), which keeps great agreement with experimental
observations.

6.2.2. Case II: Simulation on adiabatic shear band propagation

In this section, numerical simulations of Kalthoff’s impact test with an impact velocity of 45 m/s are presented. As revealed in
experimental tests [40], a completely different failure mode was observed subjected to high-velocity impact, as shown in Fig. 17(a).
Fig. 17(b) shows the predicted failure mode from peridynamic simulations. The shear band initiates from the notch tip and forms a
curved path, where a propagation angle of 70◦ is not observed.
The failure patterns for ductile specimen at various time steps are shown in Fig. 18. The proposed model and approach can suc­
cessfully capture the failure mode transition and entire adiabatic shear band propagation observed in experimental tests. Under a high-
velocity impact, the failure is controlled by a localized shear band process, no longer caused by the tensile fracture anymore. In this
case, the numerical result obtained by the original correspondence visco-plastic model with a velocity of 45 m/s, is illustrated in Fig. 19
(a). On the one hand, the fracture process predicted by that model is incomplete and numerical calculation is aborted due to its non-
convergence. On the other hand, tensile cracks still initiate and propagate when subjected to such a high velocity, markedly con­
tradicting experimental observations and aforementioned numerical predictions by the proposed peridynamic model. That is the
meaning of present work in this study.
Fig. 20 shows the evolution of von Mises stress in the shear band propagation process. In the beginning, stress localization occurs at
the tip of notch and leads to the initiation of shear band, shown in Fig. 20(a) and (b), indicating that the stress localization problem can

17
L. Wu et al. Engineering Fracture Mechanics 283 (2023) 109196

Fig. 21. Temperature contours at various times.

be accurately described by this proposed model. As the reflected stress wave assembles on the bottom surface, one high-stress region is
formed (Fig. 20(d) and Fig. 20(e)). The shear band propagates inwards in vertical direction and gradually closes to that region.
Moreover, it should be noted that the dispersion of waves is observed at the notch tip area, as shown in Fig. 20(f), especially after shear
band occurs. It is much more obvious during the shear band propagation than the tensile crack propagation in the previous section. The
relationship between the dispersion of waves and ductile failure under impact loading is not considered here, which will be studied and
discussed in our future work.
Fig. 21 shows the temperature distribution during shear band propagation process. A room temperature of 293 K is the threshold
value in this study. When a material point is completely damaged, the temperature state of that point will be attributed to room
temperature, and the stress state is set to 0. That is why the temperature of shear band is room temperature in Fig. 21. A temperature
reflection phenomenon [37], which was analyzed in a single-notch impact test, can be observed in Fig. 21(d). That is, before shear
band propagates to the specimen’s bottom surface, temperature at the bottom surface has already risen. The final shear band coincides
exactly with this local temperature rise area, as shown in Fig. 21(f).

7. Conclusions

In the present work, a non-local three-dimensional correspondence peridynamic model is reformulated under the framework of
non-ordinary state-based peridynamics. To eliminate the zero-energy mode and improve numerical stability, the velocity gradient and
deformation gradient are calculated and updated on the bond level, which owns information about the non-local deformation as well
as the bond deviation. Moreover, a mixed strain-stress criterion on the bond level is presented and used as well, and the effect of
temperature is considered as well. Bond breaking on the bond level, other than the material point level, is undoubtedly more accurate
and effective.
Several typical ductile failure problems, including a notched tension test, and low-velocity and high-velocity Kalthoff-Winkler
impact experiments, are investigated by the proposed peridynamic model. The validation is established though comparing available

18
L. Wu et al. Engineering Fracture Mechanics 283 (2023) 109196

Fig. A1. The flow chart of the whole solution procedure.

peridynamic results with corresponding experimental data and other numerical results, where a satisfactory agreement can be
inferred. Moreover, the observations and measurements in experimental tests, in terms of failure patterns, tensile crack initiation and
propagation, as well as adiabatic shear band propagation, are efficiently and naturally described, indicating that the proposed model
and approach can qualitatively and quantitatively describe the ductile fracture and impact failure.
Besides, we also conduct numerical simulations of Kalthoff-Winkler impact tests by the original correspondence visco-plastic model
and compare the results with that obtained by the proposed model in this study. By comparing failure models under a high-velocity
impact, we can conclude that the numerical results offer sufficient reasons for using the proposed model and approach to study dy­
namic failure problems of ductile materials.

CRediT authorship contribution statement

Liwei Wu: Writing – original draft, Validation, Methodology, Investigation, Data curation. Han Wang: Software, Methodology,
Investigation. Dan Huang: Writing – review & editing, Supervision, Project administration, Funding acquisition. Xin Cai: Supervision,
Resources. Junbin Guo: Writing – review & editing, Supervision.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to
influence the work reported in this paper.

Data availability

Data will be made available on request.

Acknowledgments

The authors would like to acknowledge the financial support from the National Natural Science Foundation of China (No.

19
L. Wu et al. Engineering Fracture Mechanics 283 (2023) 109196

12202136, 12072104), and the Jiangsu Funding Program for Excellent Postdoctoral Talent (No. 2022ZB161).

Appendix A

See Fig. A1.

References

[1] Hughes TJR. The finite element method: linear static and dynamic finite element analysis. Courier Corporation; 2012.
[2] Rabczuk T, Zi G, Bordas S, Nguyen-Xuan H. A simple and robust three-dimensional cracking-particle method without enrichment. Comput Methods Appl Mech
Eng 2010;199(37–40):2437–55.
[3] Li S, Liu WK, Rosakis AJ, Belytschko T, Hao W. Mesh-free Galerkin simulations of dynamic shear band propagation and failure mode transition. Int J Solids
Struct 2002;39(5):1213–40.
[4] Takaffoli M, Papini M. Material deformation and removal due to single particle impacts on ductile materials using smoothed particle hydrodynamics. Wear
2012;274:50–9.
[5] Wang S, Ding M, Gast TF, Zhu L, Gagniere S, Jiang C, et al. Simulation and visualization of ductile fracture with the material point method. Proc ACM Comput
Graph Interact Tech 2019;2(2):1–20.
[6] Ambati M, Gerasimov T, De Lorenzis L. Phase-field modeling of ductile fracture. Comput Mech 2015;55(5):1017–40.
[7] Silling SA. Reformulation of elasticity theory for discontinuities and long-range forces. J Mech Phys Solids 2000;48(1):175–209.
[8] Silling SA, Lehoucq RB. Convergence of peridynamics to classical elasticity theory. J Elast 2008;93(1):13–37.
[9] Silling SA, Askari E. A meshfree method based on the peridynamic model of solid mechanics. Comput Struct 2005;83(17–18):1526–35.
[10] Silling SA, Askari A. Peridynamic model for fatigue cracking. Rep SAND2014-18590, Sandia Natl Lab; 2014.
[11] Zhang G, Le Q, Loghin A, Subramaniyan A, Bobaru F. Validation of a peridynamic model for fatigue cracking. Eng Fract Mech 2016;162:76–94.
[12] Chen Z, Bobaru F. Peridynamic modeling of pitting corrosion damage. J Mech Phys Solids 2015;78:352–84.
[13] Jafarzadeh S, Chen Z, Zhao J, Bobaru F. Pitting, lacy covers, and pit merger in stainless steel: 3D peridynamic models. Corros Sci 2019;150:17–31.
[14] Gu X, Zhang Q, Madenci E. Non-ordinary state-based peridynamic simulation of elastoplastic deformation and dynamic cracking of polycrystal. Eng Fract Mech
2019;218:106568.
[15] Zhou X, Wang Y, Shou Y, Kou M. A novel conjugated bond linear elastic model in bond-based peridynamics for fracture problems under dynamic loads. Eng
Fract Mech 2018;188:151–83.
[16] Wu L, Wang L, Huang D, Xu Y. An ordinary state-based peridynamic modeling for dynamic fracture of laminated glass under low-velocity impact. Compos Struct
2020;234:111722.
[17] Wu L, Huang D, Bobaru F. A reformulated rate-dependent visco-elastic model for dynamic deformation and fracture of PMMA with peridynamics. Int J Impact
Eng 2021;149:103791.
[18] Fan H, Bergel GL, Li S. A hybrid peridynamics-SPH simulation of soil fragmentation by blast loads of buried explosive. Int J Impact Eng 2016;87:14–27.
[19] Zhu F, Zhao J. Peridynamic modelling of blasting induced rock fractures. J Mech Phys Solids 2021;153:104469.
[20] Silling SA, Epton M, Weckner O, Xu J, Askari E. Peridynamic states and constitutive modeling. J Elast 2007;88(2):151–84.
[21] Foster JT, Xu X. A generalized, ordinary, finite deformation constitutive correspondence model for peridynamics. Int J Solids Struct 2018;141:245–53.
[22] Amani J, Oterkus E, Areias P, Zi G, Nguyen-Thoi T, Rabczuk T. A non-ordinary state-based peridynamics formulation for thermoplastic fracture. Int J Impact Eng
2016;87:83–94.
[23] Wu L, Huang D, Xu Y, Wang L. A non-ordinary state-based peridynamic formulation for failure of concrete subjected to impacting loads. Comput Model Eng Sci
2019;118(3):561–81.
[24] Lai X, Liu L, Li S, Zeleke M, Liu Q, Wang Z. A non-ordinary state-based peridynamics modeling of fractures in quasi-brittle materials. Int J Impact Eng 2018;111:
130–46.
[25] Tupek MR, Rimoli JJ, Radovitzky R. An approach for incorporating classical continuum damage models in state-based peridynamics. Comput Methods Appl
Mech Eng 2013;263:20–6.
[26] Chen H, Spencer BW. Peridynamic bond-associated correspondence model: Stability and convergence properties. Int J Numer Methods Eng 2019;117(6):
713–27.
[27] Behera D, Roy P, Madenci E. Peridynamic correspondence model for finite elastic deformation and rupture in Neo-Hookean materials. Int J Non Linear Mech
2020;126:103564.
[28] Behzadinasab M, Foster JT. A semi-Lagrangian constitutive correspondence framework for peridynamics. J Mech Phys Solids 2020;137:103862.
[29] Bergel GL, Li S. The total and updated lagrangian formulations of state-based peridynamics. Comput Mech 2016;58(2):351–70.
[30] Flanagan DP, Taylor LM. An accurate numerical algorithm for stress integration with finite rotations. Comput Methods Appl Mech Eng 1987;62(3):305–20.
[31] Dienes JK. On the analysis of rotation and stress rate in deforming bodies. Acta Mech 1979;32(4):217–32.
[32] Zhou M, Ravichandran G, Rosakis AJ. Dynamically propagating shear bands in impact-loaded prenotched plates—II. Numerical simulations. J Mech Phys Solids
1996;44(6):1007–32.
[33] Marchand A, Duffy J. An experimental study of the formation process of adiabatic shear bands in a structural steel. J Mech Phys Solids 1988;36(3):251–83.
[34] Wright TW, Batra RC. The initiation and growth of adiabatic shear bands. Int J Plast 1985;1(3):205–12.
[35] Littlewood DJ. Roadmap for peridynamic software implementation. Rep SAND2015-9013. Sandia Natl Lab 2015.
[36] Warren TL, Silling SA, Askari A, Weckner O, Epton MA, Xu J. A non-ordinary state-based peridynamic method to model solid material deformation and fracture.
Int J Solids Struct 2009;46(5):1186–95.
[37] Macek RW, Silling SA. Peridynamics via finite element analysis. Finite Elem Anal Des 2007;43(15):1169–78.
[38] Torabi AR, Kamyab M. Notch ductile failure with significant strain-hardening: The modified equivalent material concept. Fatigue Fract Eng Mater Struct 2019;
42(2):439–53.
[39] Bobaru F, Yang M, Alves LF, Silling SA, Askari E, Xu J. Convergence, adaptive refinement, and scaling in 1D peridynamics. Int J Numer Methods Eng 2009;77
(6):852–77.
[40] Kalthoff JF. Modes of dynamic shear failure in solids. Int J Fract 2000;101(1):1–31.
[41] Kalthoff JF, Bürgel A. Influence of loading rate on shear fracture toughness for failure mode transition. Int J Impact Eng 2004;30(8–9):957–71.

20