Peridynamic Approach to Predict Ductile

and Mixed-Mode Failure

J.H. Conradiea, T.H. Beckerb, D.Z. Turnerc

Received 26 June 2018, in revised form 18 December 2018 and accepted 17 January 2019

Abstract: The peridynamic theory has been developed to Greek

address problems in solid mechanics regarding fracture δ radius [m]
through its integral non-local basis. It has been successful 𝜌𝜌 density [kg/m3]
in predicting brittle cracking, however, uncertainty still 𝝃𝝃 relative position of two particles [m]
remains with regards to mixed mode and ductile fracture. 𝜼𝜼 relative displacement of two particles [m]
This work presents a study in using peridynamics to 𝑣𝑣 Poisson’s ratio
simulate fracture in mixed mode or ductile type fractures. 𝜇𝜇 shear modulus [GPa]
The results are presented as a quantitative comparison 𝜅𝜅 bulk modulus[GPa]
between experimental tests and numerical simulations.
Standard compact tension tests were performed on 1 Introduction
polymethyl methacrylate (PMMA), stainless steel 304L and
Peridynamics has been successful in predicting crack
aluminium 1200H4 to obtain the respective JR-curves and
initiation, propagation, the crack path and subsequent failure
critical energy release rates, 𝑱𝑱𝑰𝑰𝑰𝑰 . In addition, digital image in a number of numerical applications [1-3]. It has mostly
correlation was employed to allow for qualitative been employed in linear-elastic applications [4, 5],
observation of the fracture process and choice in specifically in a crack opening mode (mode I) assuming a
peridynamic input parameters. An equivalent critical brittle fracture mechanism. The main advantages to using
stretch was determined for each material and applied to an peridynamics to model failure over other methods are that the
Arcan geometry. It is shown that the energy release rate for formulation is free of all spatial derivatives and is
mode I and mode II should be considered separately. Mixed computationally convenient to implement. Rather than
mode type failures cannot be simulated accurately by a computing spatial rates of change using traditional
single critical stretch criterion. Furthermore, ductile derivatives, nonlocal operators are employed which are
fracture requires careful consideration when selecting mathematically appropriate for discontinuous fields or fields
peridynamic input parameters. with low regularity. If the underlying problem is smooth the
peridynamic approach will converge to the finite element
Additional keywords: Peridynamics, critical energy solution. However, when the problem includes
release rate, digital image correlation discontinuities the peridynamic approach can capture these
non-smooth features whereas the finite element method
Nomenclature cannot without building in additional sophistication.
However, it is noted that monolithic peridynamic simulations
Roman
are rarely performed in practice.
𝒃𝒃 body force density [N/m3]
To date, literature shows limited peridynamic fracture
𝑐𝑐 micro-modulus [N/m]
studies in an in-plane shear opening mode (mode II) or a
D peridynamic damage
mixed mode (combination of mode I and II). It has been
E Young’s modulus of elasticity [GPa]
suggested that linear-elastic fracture mechanics (LEFM)
𝐻𝐻𝒙𝒙 horizon [m] parameters can be employed to model crack propagation in
𝐺𝐺𝐼𝐼𝐼𝐼 critical energy release rate [J/m2] materials that exhibit a ductile fracture mechanism [6].
𝑳𝑳𝒖𝒖 pairwise force function [N/m3] However, this can be contended when considering the current
𝑃𝑃𝒎𝒎𝒎𝒎𝒎𝒎 max load [kN] state of the art showcased in ductile failure x-prize. Here,
𝒖𝒖 displacement vector [m] peridynamics as well as other continuum theories, such as the
𝑡𝑡 force scalar state [N/m4] extended finite element method, were challenged to predict
𝒙𝒙 position vector [m] the fracture load of a mixed mode tension test [7]. All
𝒙𝒙′ position vector of neighbouring point[m] numerical simulations, including peridynamics, compared
𝑠𝑠 stretch poorly with experimental results.
sc critical stretch Recently it was shown that peridynamics can be
𝑻𝑻 piecewise force state [N/m3] employed to simulate the fracture initiation load and an
a. Department of Civil Engineering, University of estimate crack path of particular mixed-mode tests on brittle
Stellenbosch, Stellenbosch, South Africa polymers [8, 9]. Madenci and Otterkus [9] found good
b. Department of Mechanical and Mechatronic correlation from simulating mixed mode tests conducted by
Engineering, University of Stellenbosch, Ayatollahi and Aliha [10]. In another case Polymethyl
Stellenbosch, South Africa. Email: [email protected] methacrylate (PMMA) specimens were tested in mode I,
c. Centre for Computing Research, Sandia National mixed mode and mode II and simulated with a brittle elastic
Laboratories, Albuquerque, NM, USA
R & D Journal of the South African Institution of Mechanical Engineering 2019, 35, 1-8 1
http://dx.doi.org/10.17159/2309-8988/2019/v35a1
http://www.saimeche.org.za (open access) © SAIMechE All rights reserved.
 Peridynamic Approach to Predict Ductile and Mixed-Mode Failure

peridynamic model by Caimimi et al [8]. The simulations In the absence of body forces and in a quasi-static state,
were able to capture the fracture initiation load and the crack Equation 1 and 2 can be rewritten as
path position under the different modes. However, the pure
mode II results were overestimated compared to the tests. � {𝑻𝑻[𝒙𝒙, 𝑡𝑡]〈𝒙𝒙′ − 𝒙𝒙〉 − 𝑻𝑻[𝒙𝒙′ , 𝑡𝑡]〈𝒙𝒙 − 𝒙𝒙′〉} 𝑑𝑑𝑉𝑉𝑥𝑥 ′ = 0 (3)
The challenging aspect of simulating ductile fracture 𝐻𝐻𝒙𝒙
when using the peridynamic formulation lies with the single
critical stretch failure criterion that governs the local yield 𝑻𝑻 corresponds to a linear micro-elastic potential that has the
point. The critical stretch is often determined by using the form
total elongation in uniaxial tension tests, or with the 𝜼𝜼 + 𝝃𝝃
𝑐𝑐(𝜉𝜉)𝑠𝑠, 𝜉𝜉 ≤ 𝛿𝛿
calibration of numerical models to fit experimental data [6] 𝒇𝒇(𝜼𝜼, 𝝃𝝃) = �‖𝜼𝜼 + 𝝃𝝃‖ (4)
introducing a degree of unreliability. More recently, the 0, 𝜉𝜉 > 𝛿𝛿
critical stretch has been related to the fracture mechanics
where the relative position, 𝝃𝝃, and relative displacement, 𝜼𝜼,
based critical energy release rate by Foster et al. [11],
between 𝒙𝒙 and 𝒙𝒙′ with respect to reference configuration are
showing agreement with experimental data [9, 10]. Foster et
given by 𝝃𝝃 = 𝒙𝒙′ – 𝒙𝒙 and 𝜼𝜼 = 𝒖𝒖(𝒙𝒙′, 𝑡𝑡) − 𝒖𝒖(𝒙𝒙, 𝑡𝑡) respectivly.
al also proposed that the non-linear critical energy release rate
The stretch of bond 𝑠𝑠 is calculated by taking the magnitude
(more popularly known as the J-integral) can be used to
determine the critical stretch in ductile failure simulations [7]. of the position vector between two points such that 𝑠𝑠 =
This work presents a study in using the critical energy (‖𝜼𝜼 + 𝝃𝝃‖ − 𝜉𝜉)/𝜉𝜉 and 𝑐𝑐( 𝜉𝜉), where 𝜉𝜉 = ‖𝝃𝝃‖, is a micro-
release rate to simulate i) mixed mode and ii) ductile fracture modulus that has the meaning of the bond elastic stiffness.
loads. The results are presented as a quantitative comparison For a plane stress state, the micro-modulus may be
between experimental tests and numerical simulations. In obtained for the reference state using
addition, digital image correlation (DIC) obtained full field 6𝐸𝐸
displacement maps are imposed as boundary conditions onto 𝑐𝑐( 𝜉𝜉) = (5)
𝜋𝜋𝛿𝛿 3 (1− 𝑣𝑣)
the peridynamic formulation to demonstrate the concept of
the critical stretch parameter. It is shown that the critical where 𝐸𝐸 is the Young’s modulus and 𝑣𝑣 is the Poisson’s ratio.
stretch can be determined from a non-linear critical energy The bond stretch is calculated by taking the magnitude of the
release rate obtained from standard compact tension tests. It relative position vector between two points. Failure is
is also shown that the mode of failure and the corresponding governed by the critical stretch, 𝑠𝑠𝑐𝑐 , where subscript 𝑐𝑐
mode critical energy release rate should be taken into account designates the critical stretch. When a bond stretch exceeds
to obtain more accurate fracture loads. this quantity it is regarded as irreversibly broken and no
longer contributes to the force interactions.
2 Peridynamic Formulation |𝜼𝜼 + 𝝃𝝃| − |𝜼𝜼|
The peridynamic theory is a reformulation of classical , 𝑠𝑠 < 𝑠𝑠𝑐𝑐
𝑠𝑠 = � |𝜼𝜼| (6)
continuum mechanics and has many similarities to molecular
dynamics. In the state-based approach to peridynamics, 0, 𝑠𝑠 ≥ 𝑠𝑠𝑐𝑐
quantities of interested defined over local regions of the Peridynamic damage is defined as the reduction in the micro
domain or states are used to describe the interaction of forces modulus due to broken bonds and can be simply expressed as
between particles within a material radius called the horizon.
The integral formulation holds the advantage of allowing for number of broken bonds
𝐷𝐷 = (7)
a discontinuous continuum, as opposed to spatial partial number of initial bonds
derivatives that inherently cannot accommodate
where 𝐷𝐷 = 0 is virgin material and 𝐷𝐷 = 1 is complete
discontinuities.
disconnection of all bonds to the particle. The coalescence of
The governing formulation of peridynamics is based on
broken bonds represent a fracture surface and therefore a
the balance of forces between particle point 𝒙𝒙 and its
damage of 𝐷𝐷 = 0.5 and higher would indicate that a fracture
neighbours 𝒙𝒙′ within the horizon 𝐻𝐻𝒙𝒙 of radius 𝛿𝛿. The vectors
surface has formed. Silling and Askari [6] have related 𝑠𝑠𝑐𝑐 to
linking 𝒙𝒙 and 𝒙𝒙′ is called a bonds. A bond interaction is
the energy required fracture of a body per unit length, referred
expressed in terms of a piecewise force state 𝑻𝑻, defined at
to in the field of LEFM as the critical energy release rate, 𝐺𝐺𝐼𝐼𝐼𝐼 ,
position 𝒙𝒙 and time 𝑡𝑡. The basic equations of the peridynamic where subscript I indicates a mode I and subscript 𝑐𝑐 the
model include the equation of motion and is expressed as [12] critical energy release rate.
𝜌𝜌(𝒙𝒙)𝒖𝒖̈ (𝒙𝒙, 𝑡𝑡) = 𝑳𝑳𝒖𝒖 (𝒙𝒙, 𝑡𝑡) + 𝒃𝒃(𝒙𝒙, 𝑡𝑡) ∀𝑥𝑥 ∈ 𝐵𝐵, 𝑡𝑡 𝐺𝐺𝐼𝐼𝐼𝐼
(1) 𝑠𝑠𝑐𝑐 =
≥0 � 6 16 (8)
� 𝜇𝜇 + 2 (𝜅𝜅 − 2𝜇𝜇)� 𝛿𝛿
𝜋𝜋 9𝜋𝜋
𝑳𝑳𝒖𝒖 (𝒙𝒙, 𝑡𝑡) = � {𝑻𝑻[𝒙𝒙, 𝑡𝑡]〈𝒙𝒙′ − 𝒙𝒙〉 Here, 𝜇𝜇 is the shear modulus and 𝜅𝜅 the bulk modulus.
𝐻𝐻𝒙𝒙 (2)
Equation 7 can be generalised to determine the critical stretch
− 𝑻𝑻[𝒙𝒙′ , 𝑡𝑡]〈𝒙𝒙 − 𝒙𝒙′〉} 𝑑𝑑𝑉𝑉𝑥𝑥 ′ of a single bond in a two-dimensional domain as [9]. 𝐺𝐺𝐼𝐼𝐼𝐼 can
Here, 𝜌𝜌 is the density, 𝒖𝒖 is the displacement, and 𝒃𝒃 is the body be obtained experimentally by using standardised testing
force density. The integral 𝑳𝑳𝒖𝒖 (𝒙𝒙, 𝑡𝑡) is defined over 𝐻𝐻𝒙𝒙 , which methodologies, such as ASTM-E1820 [13].
is the compact supported domain of the pairwise force
function around point 𝒙𝒙.
R & D Journal of the South African Institution of Mechanical Engineering 2019, 35, 1-8 2
http://dx.doi.org/10.17159/2309-8988/2019/v35a1
http://www.saimeche.org.za (open access) © SAIMechE All rights reserved.
 Peridynamic Approach to Predict Ductile and Mixed-Mode Failure

3 Methodology shear) with configurations in 15-degree increments.

The study aims to conduct both mixed mode and ductile Originally developed by Arcan et al. [18] for mixed mode
fracture investigations to further establish the capabilities of fracture investigations of composites. The geometry was
the peridynamic methodology. As such, the experimental modified by Amstutz et al. [14, 15] to allow for a simpler
plan entails two separate investigations that focus on i) mixed butterfly-shape sample design. Figure 1a shows the Arcan
mode brittle fracture and ii) mode I ductile fracture. sample geometry and Figure 1b the test configuration of the
To obtain valid J-integral values, 𝐽𝐽𝐼𝐼𝐼𝐼 , a standardised Arcan setup. It is noted that the sample is clamped to the
Compact Tension (CT) test geometry as described in ASTM- fixture using 3 mm plates and therefore the sample was
E1820 was used, testing three different materials that exhibit considered as completely constrained.
distinctively different material behaviours. Subsequently, PMMA Arcan samples had a thickness of 𝐵𝐵 = 12 𝑚𝑚𝑚𝑚
experimental investigations were undertaken using a and in the crack plane width of 𝑊𝑊 = 40 𝑚𝑚𝑚𝑚. The steel
modified Arcan test geometry [14, 15] to obtain respective /aluminium sample had dimensions 𝐵𝐵 = 3 𝑚𝑚𝑚𝑚 and 𝑊𝑊 =
fracture loads. The study culminates in a numerical analysis 40 𝑚𝑚𝑚𝑚. The PMMA samples were pre-cracked using a
of the experimental test, comparing experimental and razorblade. As outlined for the CT tests, the stainless steel /
numerical load-displacement curves (plotted as crosshead aluminium samples were fatigue pre-cracked. Pre-crack
displacement vs. applied load) and the peak loads at fracture. lengths are summarised in Table 2.
In conjunction, DIC was utilised during the CT tests to Fracture tests were conducted in quasi-static loading of
allow for full field displacement measurements of the 0.1 mm/min. The PMMA samples were tested in mode I,
specimen surface during experimental testing. The DIC mode II and mixed mode at 45º crack plane inclination to the
obtained displacement maps were imposed as boundary loading direction. Steel / aluminium samples were tested in
conditions into the peridynamic framework to allow for mode I only. Recorded were load-displacement curves and
visual representation of damage using Equation 3. This fracture loads. All testing was conducted in ambient
technique was previously presented by the authors in [16]. conditions.
Table 1 Test data summary for CT tests.
3.1 Material selection
Polymethyl methacrylate (PMMA), stainless steel 304L and Test 𝑃𝑃𝑚𝑚𝑚𝑚𝑚𝑚 𝐽𝐽𝐼𝐼𝐼𝐼 𝐽𝐽𝐼𝐼𝐼𝐼ℎ
Material 𝑎𝑎/𝑊𝑊
aluminium 1200H4 were utilised to study both brittle and number (kN) (J/mm2) (J/mm2)
ductile fracture. PMMA exhibits a near perfect brittle 1 0.464 0.275
behaviour, stainless steel shows significant strain hardening PMMA 2 0.490 0.306 0.75 -
post yield and aluminium can be regarded as a near perfect
plastic material. All three materials are assumed to be 3 0.500 0.286
homogenous and isotropic. Elastic properties were taken as 1 0.444 10.0
Young Modulus 𝐸𝐸 = 1.7 𝐺𝐺𝐺𝐺𝐺𝐺 and Poisson’s ratio 𝜈𝜈 = Stainless
2 0.460 8.60 410 60
0.35 for PMMA, 𝐸𝐸 = 193 𝐺𝐺𝐺𝐺𝐺𝐺 and 𝐸𝐸 = 70 𝐺𝐺𝐺𝐺𝐺𝐺 [17] for steel 304L
3 0.536 9.60
stainless steel 304L and aluminium 1200H4 respectively,
with 𝜈𝜈 = 0.30. 1 0.540 4.10
Aluminium 2 0.544 2.40 120 18
1200H4
3.2 Standardised tests 3 0.545 4.20
PMMA CT samples had a thickness of 𝐵𝐵 = 12 𝑚𝑚𝑚𝑚 and a
width of 𝑊𝑊 = 50 𝑚𝑚𝑚𝑚 to allow for LEFM conditions. To Table 2 Test data summary for Arcan tests.
ensure ductile failure mechanisms, steel / aluminium samples Test 𝑃𝑃𝑚𝑚𝑚𝑚𝑚𝑚
were thinner with a thickness of 𝐵𝐵 = 6 𝑚𝑚𝑚𝑚. Samples were Material Mode 𝑎𝑎/𝑊𝑊
number (kN)
fatigue pre-cracked using an Intron servo-hydraulic
PMMA I 1 0.538 1.32
testing machine fitted with a 50 kN load cell at a cyclic
frequency of 𝑓𝑓 = 10 𝐻𝐻𝐻𝐻 and a load ratio 𝑅𝑅 ≈ 0.1. The pre- PMMA II 1 0.460 0.826
crack length, 𝑎𝑎, is given in Table 1. Three samples for each PMMA Mixed, 45º 1 0.530 1.43
material were tested. Stainless
J-integral tests were conducted in quasi-static load and steel I 1 0.288 34.0
unload rate of 0.1 mm/min and 1.5 mm/min respectively. For 304L
this, a Zwick® Z250 tensile machine fitted with a 250 kN Aluminium
load cell was used. Recorded were crack mouth opening I 1 0.300 9.70
1200H4
curves and fracture loads which were used to calculate JR-
curves and 𝐽𝐽𝐼𝐼𝐼𝐼 according to ASTM-E1820 for the respective 3.4 Damage mapping using digital image
materials. Tests were conducted in ambient conditions. correlation
A post-processing algorithm developed by the authors in
3.3 Arcan tests [16], uses the DIC obtained displacement maps to measure
An Arcan fixture was designed and built that allows for mode the extent of peridynamic damage, calculated using Equation
I (tension, crack plane perpendicular to load direction), mode 7, where a ratio of the number of broken bonds vs. the number
II (in-plane shear, crack plane parallel to load direction) and of initial bonds is considered for each node. Here, the
mixed mode loading (combination of tension and in-plane

R & D Journal of the South African Institution of Mechanical Engineering 2019, 35, 1-8 3
http://dx.doi.org/10.17159/2309-8988/2019/v35a1
http://www.saimeche.org.za (open access) © SAIMechE All rights reserved.
 Peridynamic Approach to Predict Ductile and Mixed-Mode Failure

technique is used to qualitatively demonstrate the choice in The noise floor value was measured at 0.68 × 10−4 𝑚𝑚𝑚𝑚,
critical stretch as a failure criterion. 2.60 × 10−4 𝑚𝑚𝑚𝑚 for the PMMA and steel / aluminium
respectively (measured as the standard deviation of the
displacements determined from two unloaded images). Rigid
body tests showed an average error of 3.10 × 10−4 mm.

4 Experimental results
The investigation into the energy release rate based critical
stretch considered two criteria for i) mixed mode and ii)
ductile fracture, where a designation of 𝐼𝐼 and 𝐼𝐼𝐼𝐼 designates
mode 1 or 2 respectively, and 𝑐𝑐 and 𝑡𝑡ℎ designates a critical or
threshold value. This is further outlined below.

4.1 Energy release rate and fracture load

Load-displacement curves for the CT samples are shown in
Figure 2, the resultant JR-curves are shown in Figure 3 and
the respective 𝐽𝐽𝐼𝐼𝐼𝐼 values summarised in Table 1. PMMA CT
𝐽𝐽𝐼𝐼𝐼𝐼 results correlated within 4% to that found by Ayatollahi
and Aliha [19]. It is worth noting that only a single load-
unloading compliance curve was achieved for the PMMA
specimens due to the brittle nature of PMMA (Figure 2a).
Similarly, Table 2 summarises the Arcan pre-crack lengths
and fracture loads. All data is reported to three significant
figures.

Figure 1 Arcan test setup showing (a) the specimen

geometry and (b) testing in 45º loading, with
LaVision® CCD cameras and lights to allow for full
field surface displacement measurements.
The DIC equipment used was a LaVision® 5-megapixel
system with 12-bit resolution as shown in Figure 1b. DaVis®
version 8.2, which assumes linear affine transformations and
a zero normalised sum-square difference (ZNSSD) algorithm
to map subsets, was used to compute the surface displacement
maps. A 6th order spline interpolation was employed for sub-
pixel determination. LaVision® claims a typical accuracy of
0.01 pixels for in-plane and 0.02 pixels for out-of-plane
deformation can be obtained. For specimen surface
preparation black and white spray paints were used to obtain
speckle patterns of five or more pixels in size. This allows for
three or more speckle features per subset. It is worth noting
that a trade-off between subset size and spatial resolution
exists: a larger subset results in more accurate displacement
maps at the cost of spatial resolution, i.e. accurately capturing
high strain gradients near the crack tip. To allow for a better-
defined crack tip field and crack path a ‘medium’ sized subset
(of 31×31 pixels) was chosen with a ‘small’ step size.
Analysis of the PMMA samples used a step size of 12
pixels, whereas the steel / aluminium samples utilised a step
size of 6 pixels. In addition, a dense displacement map allows
for a finer peridynamic discretization as recommended by the
authors in [16]. Figure 2 Load-unloading compliance curves of (a) PMMA,
(b) stainless steel and (c) aluminium
R & D Journal of the South African Institution of Mechanical Engineering 2019, 35, 1-8 4
http://dx.doi.org/10.17159/2309-8988/2019/v35a1
http://www.saimeche.org.za (open access) © SAIMechE All rights reserved.
 Peridynamic Approach to Predict Ductile and Mixed-Mode Failure

4.2 Peridynamic damage and critical stretch

Figure 4 depicts the DIC obtained peridynamic discretization
of a PMMA CT sample with the respective damage map
superimposed onto the image. Such analyses were
undertaken for the three considered materials. The effect of
choice in the critical stretch on the damage a head of the crack
tip is illustrated in Figure 5.
The critical stretch values obtained using Equation 8 for
the three considered materials are summarised in Table 3,
based on a horizon radius of 𝛿𝛿 = 3∆𝑥𝑥. Two critical stretch
values were considered: (i) 𝑠𝑠𝐼𝐼𝐼𝐼 based on the ASTM obtained
𝐽𝐽𝐼𝐼𝐼𝐼 , and 𝑠𝑠𝐼𝐼𝐼𝐼ℎ based on a 𝐽𝐽𝐼𝐼𝐼𝐼ℎ value at a crack extension of 0.015
mm, chosen as 10% of the 0.15 mm exclusion line (Figure 4).
It is argued that the energy dissipation at crack initiation
would be a more appropriate quantity for predicting fracture
loads when using the elastic peridynamic model. Moreover,
using 𝑠𝑠𝐼𝐼𝐼𝐼 corresponds to a state in which the material has
already undergone significant plastic deformation and hence
results in less sensitive damage mapping.
The peridynamic damage of the PMMA specimen
correlates closely to the crack position (approximately at
coordinates 4,0 in Figure 5), where the crack tip location is
interpreted at a damage of 𝐷𝐷 = 0.5. It is worth noting that the
inherent limitations of DIC, in resolving displacements near
discontinuities such as across the crack, introduces a degree
of inaccuracy. This results in smeared damage field over the
region of the crack as seen in Figure 3 for PMMA.
Figure 3 JR-curve of stainless steel (a) and aluminium (b).
Investigations into increasing the horizon radius showed a
more defuse damage field. Similar to the aforementioned DIC
limitations, a trade-off exists between the horizon radius.
Table 3 Test data summary for Arcan tests.

Material 𝑆𝑆𝐼𝐼𝐼𝐼 (𝛿𝛿 = 3∆𝑥𝑥) 𝑆𝑆𝐼𝐼𝐼𝐼ℎ (𝛿𝛿 = 3∆𝑥𝑥)

PMMA 0.0138 -
Stainless steel 0.0645 0.0260
Aluminium 0.0560 0.0225

Figure 4 DIC obtained peridynamic discretization of PMMA

CT specimen with respective peridynamic damage
𝐷𝐷 (Equation 7) map superimposed (white
rectangle). Crack propagation direction is from Figure 5 Damage fields at the onset of crack propagation of
right to left. PMMA, stainless steel and aluminium. Shown are
the damage fields obtained using the three
considered critical stretch values (na – not
applicable).
R & D Journal of the South African Institution of Mechanical Engineering 2019, 35, 1-8 5
http://dx.doi.org/10.17159/2309-8988/2019/v35a1
http://www.saimeche.org.za (open access) © SAIMechE All rights reserved.
 Peridynamic Approach to Predict Ductile and Mixed-Mode Failure

A larger horizon radius results in refinement of the

damage field, however at the cost in accuracy of the crack
location. The particle spacing used in Figure 5 coincided with
the DIC grid spacing, namely ∆𝑥𝑥 = 0.227 mm and ∆𝑥𝑥 = 0.113
mm for PMMA and steel / aluminium respectively.
It is seen from Figure 5 that 𝑠𝑠𝐼𝐼𝐼𝐼 gives a damage field in
the region where plastic deformation is occurring. The region
where the crack is interpreted as initiated is also revealed. A
lowered critical stretch, 𝑠𝑠𝐼𝐼𝐼𝐼ℎ , results in a more diffused
damage field: Implementing 𝑠𝑠𝐼𝐼𝐼𝐼 as failure criterion at the
onset of fracture shows a moderate plastic damage field,
however, for a 𝑠𝑠𝐼𝐼𝐼𝐼ℎ failure criterion the damage becomes
increasingly defused and shows extensive damage ahead of
the crack tip.

5 Numerical implementation
The Arcan geometry was modelled using an open source code
Mingus from Sandia Corporation. The geometry and mesh Figure 6 Arcan model, boundary conditions and notch
were generated in Cubit ver. 12.2. All models were simulated definition.
in two dimensions under a plane stress assumption. In line
with the experimental plan, the numerical study entails two 5.3 Numerical results
separate investigations that focus on i) brittle, mixed mode Figure 7 shows the load-displacement results for the mixed
fracture and ii) ductile fracture in mode I. The output of the mode investigations. Figure 8a summarizes the numerically
numerical simulation are load displacement curves and obtained fracture loads for PMMA. Mode I obtained fracture
respective fracture loads. loads correlated within 5.5%. This serves as further
quantitative validation in using the critical energy release rate
5.1 Boundary conditions and mesh to determine the critical stretch to predict fracture loads as
Figure 6 depicts the Arcan model, mesh, boundary conditions proposed by Foster et al [11]. The mode II and the 45-degree
and the pre-defined notch position. Peridynamic particles mixed mode results yielded less accurate fracture loads
were spaced ∆𝑥𝑥 = 1 mm apart in the middle section (blue) and (within 27%). It is argued that a mode I assumed critical
∆𝑥𝑥 = 4 mm apart at the respective left and right boundaries stretch does not incorporate the shear type failure
(white and red). The notch and pre-crack were defined by mechanisms. When using an equivalent mode II critical
removing the bond influence between particles over the stretch, the correlation between fracture loads improves to 9
location of the pre-defined crack. The left section (white) of %. It is therefore argued that the peridynamic formulation
the geometry was fully constrained in translation in both x does not capture the type of fracture mechanism, the increase
and y-directions. The right section (red) was assigned a in accuracy served as verification that the corresponding
displacement of 𝑢𝑢𝑥𝑥 = 0.01 mm for mode I, 𝑢𝑢𝑦𝑦 = -0.01 mm for energy release rate should be considered when determining
the critical stretch to analyse mode II loading. The mixed
mode II and 𝑢𝑢𝑥𝑥 = 0.00707 mm, 𝑢𝑢𝑦𝑦 = 0.00707 mm for 45º
mode simulations correlated poorly using both mode I and
mixed mode investigations.
mode II energy release rate based failure criterions. This
suggests that the corresponding mixed mode energy release
5.2 Input parameters
rate should be considered to determine the critical stretch.
The input parameters are summarised in Table 4. Note that
Figure 8b summarizes the numerically obtained fracture
for shear loading (mode II and mixed mode) the shear moduli
loads for steel and aluminium. It is evident that implementing
were reduced by 25%. It was found that this adjustment
𝑠𝑠𝐼𝐼𝐼𝐼ℎ yielded considerably more accurate fracture loads. It is
captured the shear behaviour more accurately, however,
argued for an elastic peridynamic material model, the
needs more verification. In addition, mode II and mixed mode
threshold approach should be considered for determining the
simulations also considered a critical stretch 𝑠𝑠𝐼𝐼𝐼𝐼𝐼𝐼 , derived
critical stretch. As opposed to implementing 𝐽𝐽𝐼𝐼𝐼𝐼 , which
from 𝐾𝐾𝐼𝐼𝐼𝐼𝐼𝐼 = 1.48 𝑀𝑀𝑀𝑀𝑀𝑀 √𝑚𝑚 fracture toughness by Ayatollahi corresponds to the energy dissipated after crack extension,
et al. [17] using the well-established relationship 𝐾𝐾 = �𝐽𝐽𝐽𝐽 the 𝐽𝐽𝐼𝐼𝐼𝐼ℎ corresponds to the fracture load at crack initiation..
[20].
Table 4 Peridynamic input parameters for Arcan simulations (na – not applicable).
Material Mode Shear modulus Bulk modulus ∆𝑥𝑥 𝛿𝛿 𝑠𝑠𝐼𝐼𝐼𝐼 𝑠𝑠𝐼𝐼𝐼𝐼𝐼𝐼 𝑠𝑠𝐼𝐼𝐼𝐼ℎ
(GPa) (GPa) (mm)
PMMA I 0.629 1.88 1 5∆𝑥𝑥 0.0106 na na
II 0.472 1.88 1 4∆𝑥𝑥 0.0132 0.0162 na
Mixed 0.472 1.88 1 4∆𝑥𝑥 0.0132 0.0162 na
Steel I 74.2 161 1 4∆𝑥𝑥 0.0266 na 0.0106
Aluminium I 26.9 58.3 1 4∆𝑥𝑥 0.0232 na 0.00926

R & D Journal of the South African Institution of Mechanical Engineering 2019, 35, 1-8 6
http://dx.doi.org/10.17159/2309-8988/2019/v35a1
http://www.saimeche.org.za (open access) © SAIMechE All rights reserved.
 Peridynamic Approach to Predict Ductile and Mixed-Mode Failure

Figure 8 Showing the % deviation between experimental

and numerical fracture loads for (a) PMMA in mixed mode and
for (b) steel and aluminium. The table below the figure
Figure 7 Comparison of analysing corresponding energy provides actual fracture loads.
release rates to Arcan test results for predicting (a)
mode I, (b) mode II and (c) 45-degree mixed mode • The JR-curve was identified as the essential means to
fracture loads. determine the critical stretch to predict fracture loads.
It is proposed that if the JR-curve is implemented as a However, it was shown that the energy release rate for
function the entire progression of ductile failure can be mode I and mode II should be considered separately.
captured. With other words, incorporating the JR-curve as a Mixed mode type failures cannot be simulated
non-linear failure criterion can provide the essential means to accurately by a single critical stretch criterion.
capture the material ductile fracture mechanism • The results suggest that the classical peridynamic
framework cannot accurately predict failure loads
in mixed mode type loading.
6 Conclusion
• Similarly, it was shown that ductile fracture load
• This study used the LEFM critical energy release rate
cannot be accurately simulated when using the
to simulate mixed mode and ductile fracture loads.
standardised 𝐽𝐽𝐼𝐼𝐼𝐼 , non-linear critical energy release
The results were presented as a quantitative
rate. Considerations need to be made with respect to a
comparison between experimental tests and numerical
non-linear crack initiation energy release rate.
peridynamic simulations.
• DIC obtained surface displacement maps were
imposed as boundary conditions onto the peridynamic 7 References
framework to investigate the critical stretch criterion. 1. S. A. Silling. Reformulation of elasticity theory for
The methodology served as a qualitative tool to discontinuities and long-range forces. Journal of the
demonstrate the damage mechanisms in Mechanics and Physics of Solids, 48(1):175-209, 2000.
peridynamics.
R & D Journal of the South African Institution of Mechanical Engineering 2019, 35, 1-8 7
http://dx.doi.org/10.17159/2309-8988/2019/v35a1
http://www.saimeche.org.za (open access) © SAIMechE All rights reserved.
 Peridynamic Approach to Predict Ductile and Mixed-Mode Failure

2. W. Gerstle, N. Sau and S. A. Silling. Peridynamic fibre-reinforced materials. Experimental Mechanics,

modeling of concrete structures. Nuclear Engineering 18(4):141–146, 1978.
and Design, 237(12-13):1250-1258, 2007. 19. M. R. Ayatollahi, M. R. M. Aliha and H. Saghafi. An
3. W. Gerstle, N. Sau and E. Aguilera. Micropolar improved semi-circular bend specimen for investigating
peridynamic constitutive model for concrete. 19th mixed mode brittle fracture. Engineering Fracture
International Conference on Structural Mechanics in Mechanics, 78(1):110–123, 2011.
Reactor Technology (SMiRT 19), B02/1-2, 2007. 20. T. L. Anderson. Fracture Mechanics: Fundamentals and
4. Y. D. Ha and F. Bobaru. Characteristics of dynamic Applications. 2nd edition, CRC Press, 1994.
brittle fracture captured with peridynamics. Engineering
Fracture Mechanics, 78(6):1156–1168, 2011.
5. E. Askari, J. Xu and S. A. Silling. Peridynamic analysis
of damage and failure in composites. 44th AIAA
Aerospace Sciences Meeting and Exhibit, 2006.
6. S. A. Silling and E. A. Askari. Meshfree method based
on the peridynamic model of solid mechanics.
Computers & Structures, 83(17-18):1526–1535, 2005.
7. B. L. Boyce, J. E. Bishop, T. Cordova, J. V. Cox, T. B.
Crenshaw, K. Dion, K. Dowding, J. M. Emery, J. T.
Foster, J. W. Foulk III, D. J. Littlewood, A. Mota, J.
Ostien, J. Robbins, S. A. Silling, B. W. Spencer and G.
W. Wellman. Ductile failure x-prize. Technical report
SAND2011-6801, Sandia National Laboratories, 2011.
8. F. Caimmi, E. Haddadi, N. Choupani, C. Marano and L.
Andena. Modeling mixed-mode fracture in poly
(methylmethacrylate) using peridynamics, Procedia
Structural Integrity, 2, 166-173, 2016.
9. E. Madenci and E. Oterkus. Peridynamic Theory and Its
Applications. Springer, New York, 2014.
10. M. R. Ayatollahi and M. R. M. Aliha. Analysis of a new
specimen for mixed mode fracture tests on brittle
materials. Engineering Fracture Mechanics,
76(11):1563–1573, 2009.
11. J. T. Foster, S. A. Silling and W. Chen. An energy based
failure criterion for use with peridynamic states. Journal
for Multiscale Computational Engineering, 9(6):675–
687, 2011.
12. S. A. Silling, M. Epton, O. Weckner, J. Xu and E. Askari.
Peridynamic states and constitutive modeling. Journal of
Elasticity, 88(2):151-184, 2007.
13. ASTM-E1820. Standard test method for measurement of
fracture toughness. Standard, American Society for
Testing Materials, West Conshohocken, 2011.
14. B. E. Amstutz, M. A. Sutton and D. S. Dawicke. An
experimental study of CTOD for mode I/II stable crack
growth in thin aluminium specimens. ASTM STP 1256
on Fracture Mechanics, 26:256–273, 1995.
15. B. E. Amstutz, M. A. Sutton, D. S. Dawicke and M. L.
Boone. Effects of mixed mode I/II loading and grain
orientation on crack initiation and stable tearing in 2024-
T3 aluminum. Fatigue and Fracture Mechanics,
27:217–224, 1997.
16. J. H. Conradie, D. Z. Turner and T. H. Becker.
Characterising damage in structural materials using
digital image correlation and peridynamics. 9th
Conference on Computational and Applied Mechanics,
South Africa, 2014.
17. N. Hallback and F. Nilsson. Mixed-Mode I/II fracture
behaviour of an aluminium alloy. Journal of Mechanics
and Physics of Solids, 42(9):1345–1374, 1994.
18. M. Arcan, Z. Hashin and A. A. Voloshin. Method to
produce uniform plane-stress states with applications to
R & D Journal of the South African Institution of Mechanical Engineering 2019, 35, 1-8 8
http://dx.doi.org/10.17159/2309-8988/2019/v35a1
http://www.saimeche.org.za (open access) © SAIMechE All rights reserved.