Mathematical Problems in Engineering

Computational Methods for

Fracture
Guest Editors: Timon Rabczuk, Stéphane P. A. Bordas, and Goangseup Zi
Computational Methods for Fracture
Mathematical Problems in Engineering

Computational Methods for Fracture

Guest Editors: Timon Rabczuk, Stéphane P. A. Bordas,

and Goangseup Zi
Copyright © 2014 Hindawi Publishing Corporation. All rights reserved.

This is a special issue published in “Mathematical Problems in Engineering.” All articles are open access articles distributed under the
Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the
original work is properly cited.
Editorial Board
Mohamed Abd El Aziz, Egypt Elmetwally Elabbasy, Egypt Ren-Jieh Kuo, Taiwan
Eihab M. Abdel-Rahman, Canada A. Elı́as-Zúñiga, Mexico Jurgen Kurths, Germany
Rashid K. Abu Al-Rub, USA Anders Eriksson, Sweden Claude Lamarque, France
Sarp Adali, South Africa Vedat S. Erturk, Turkey Usik Lee, Korea
Salvatore Alfonzetti, Italy Moez Feki, Tunisia Marek Lefik, Poland
Igor Andrianov, Germany Ricardo Femat, Mexico Stefano Lenci, Italy
Sebastian Anita, Romania Robertt A. Fontes Valente, Portugal Roman Lewandowski, Poland
W. Assawinchaichote, Thailand Claudio R. Fuerte-Esquivel, Mexico Shihua Li, China
Erwei Bai, USA Zoran Gajic, USA Ming Li, China
Ezzat G. Bakhoum, USA Ugo Galvanetto, Italy Jian Li, China
José Manoel Balthazar, Brazil Xin-Lin Gao, USA Shanling Li, Canada
R. K. Bera, India Furong Gao, Hong Kong Teh-Lu Liao, Taiwan
Christophe Bérenguer, France Behrouz Gatmiri, Iran Panos Liatsis, UK
Jonathan N. Blakely, USA Oleg V. Gendelman, Israel Jui-Sheng Lin, Taiwan
Stefano Boccaletti, Spain Didier Georges, France Yi-Kuei Lin, Taiwan
Stephane P.A. Bordas, USA Paulo Batista Gonçalves, Brazil Shueei M. Lin, Taiwan
Daniela Boso, Italy Oded Gottlieb, Israel Wanquan Liu, Australia
M. Boutayeb, France Fabrizio Greco, Italy Yuji Liu, China
Michael J. Brennan, UK Quang Phuc Ha, Australia Bin Liu, Australia
Salvatore Caddemi, Italy M. R. Hajj, USA Paolo Lonetti, Italy
Piermarco Cannarsa, Italy Tony Sheu Wen Hann, Taiwan Vassilios C. Loukopoulos, Greece
Jose E. Capilla, Spain Thomas Hanne, Switzerland Junguo Lu, China
Carlo Cattani, Italy K. R. (Stevanovic) Hedrih, Serbia Chien-Yu Lu, Taiwan
Marcelo Moreira Cavalcanti, Brazil M.I. Herreros, Spain Alexei Mailybaev, Brazil
Diego J. Celentano, Chile Wei-Chiang Hong, Taiwan Manoranjan K. Maiti, India
Mohammed Chadli, France Jaromir Horacek, Czech Republic Oluwole Daniel Makinde, South Africa
Arindam Chakraborty, USA Gordon Huang, Canada Rafael Martinez-Guerra, Mexico
Yong-Kui Chang, China Huabing Huang, China Driss Mehdi, France
Michael J. Chappell, UK Chuangxia Huang, China Roderick Melnik, Canada
Kui Fu Chen, China Yi Feng Hung, Taiwan Xinzhu Meng, China
Xinkai Chen, Japan Hai-Feng Huo, China Yuri Vladimirovich Mikhlin, Ukraine
Kue-Hong Chen, Taiwan Asier Ibeas, Spain Gradimir Milovanovic, Serbia
Jyh-Horng Chou, Taiwan Anuar Ishak, Malaysia Ebrahim Momoniat, South Africa
Slim Choura, Tunisia Reza Jazar, Australia Trung Nguyen Thoi, Vietnam
Cesar Cruz-Hernandez, Mexico Zhijian Ji, China Hung Nguyen-Xuan, Vietnam
Swagatam Das, India Jun Jiang, China Ben T. Nohara, Japan
Filippo de Monte, Italy J. J. Judice, Portugal Sotiris K. Ntouyas, Greece
Antonio Desimone, Italy Tadeusz Kaczorek, Poland Gerard Olivar, Colombia
Yannis Dimakopoulos, Greece Tamas Kalmar-Nagy, USA Claudio Padra, Argentina
Baocang Ding, China Tomasz Kapitaniak, Poland Bijaya Ketan Panigrahi, India
Joao B. R. Do Val, Brazil Hamid Reza Karimi, Norway Francesco Pellicano, Italy
Daoyi Dong, Australia Metin O. Kaya, Turkey Matjaz Perc, Slovenia
B. Dubey, India Nikolaos Kazantzis, USA Vu Ngoc Phat, Vietnam
Horst Ecker, Austria Farzad Khani, Iran Maria do Rosário Pinho, Portugal
M. Onder Efe, Turkey Kristian Krabbenhoft, Australia Alexander Pogromsky, The Netherlands
Seppo Pohjolainen, Finland Cheng Shao, China Yongqi Wang, Germany
Stanislav Potapenko, Canada Bo Shen, Germany Moran Wang, China
Sergio Preidikman, USA Zhan Shu, UK Gerhard-Wilhelm Weber, Turkey
Carsten Proppe, Germany Jian-Jun Shu, Singapore Jeroen A.S. Witteveen, The Netherlands
Hector Puebla, Mexico Dan Simon, USA Kwok-Wo Wong, Hong Kong
Justo Puerto, Spain Luciano Simoni, Italy Ligang Wu, China
Dane Quinn, USA Grigori M. Sisoev, UK Zhengguang Wu, China
K. R. Rajagopal, USA Christos H. Skiadas, Greece Gongnan Xie, China
Gianluca Ranzi, Australia Davide Spinello, Canada Wang Xing-yuan, China
Sivaguru Ravindran, USA Sri Sridharan, USA Xuping Xu, USA
G. Rega, Italy Rolf Stenberg, Finland Xi Frank Xu, USA
Pedro Ribeiro, Portugal Changyin Sun, China Jun-Juh Yan, Taiwan
J. Rodellar, Spain Jitao Sun, China Xing-Gang Yan, UK
Rosana Rodriguez-Lopez, Spain Xi-Ming Sun, China Suh-Yuh Yang, Taiwan
Alejandro J. Rodriguez-Luis, Spain Andrzej Swierniak, Poland Mahmoud T. Yassen, Egypt
Ignacio Romero, Spain Yang Tang, Germany Mohammad I. Younis, USA
Hamid Ronagh, Australia Allen Tannenbaum, USA Bo Yu, China
Carla Roque, Portugal Cristian Toma, Romania Huang Yuan, Germany
Rubén Ruiz Garcı́a, Spain Irina N. Trendafilova, UK S.P. Yung, Hong Kong
Manouchehr Salehi, Iran Alberto Trevisani, Italy Ion Zaballa, Spain
Miguel A. F. Sanjuán, Spain Jung-Fa Tsai, Taiwan Ashraf M. Zenkour, Saudi Arabia
Ilmar Ferreira Santos, Denmark Kuppalapalle Vajravelu, USA Jianming Zhan, China
Nickolas S. Sapidis, Greece Victoria Vampa, Argentina Xu Zhang, China
Evangelos J. Sapountzakis, Greece Josep Vehi, Spain Yingwei Zhang, China
Bozidar Sarler, Slovenia Stefano Vidoli, Italy Lu Zhen, China
Andrey V. Savkin, Australia Xiaojun Wang, China Liancun Zheng, China
Massimo Scalia, Italy Dan Wang, China Jian Guo Zhou, UK
Mohamed A. Seddeek, Egypt Youqing Wang, China Zexuan Zhu, China
Alexander P. Seyranian, Russia Cheng C. Wang, Taiwan Mustapha Zidi, France
Leonid Shaikhet, Ukraine Yijing Wang, China
Contents
Computational Methods for Fracture, Timon Rabczuk, Stéphane P. A. Bordas, and Goangseup Zi
Volume 2014, Article ID 593041, 2 pages

Stable Cracking Particles Method Based on Stabilized Nodal Integration and Updated Lagrangian
Kernel, S. Xu
Volume 2014, Article ID 646514, 10 pages

Error Estimate and Adaptive Refinement in Mixed Discrete Least Squares Meshless Method, J. Amani,
A. Saboor Bagherzadeh, and T. Rabczuk
Volume 2014, Article ID 721240, 16 pages

A Meshless Local Petrov-Galerkin Shepard and Least-Squares Method Based on Duo Nodal Supports,
Xiaoying Zhuang and Yongchang Cai
Volume 2014, Article ID 806142, 11 pages

A Coupled Thermo-Hydro-Mechanical Model of Jointed Hard Rock for Compressed Air Energy
Storage, Xiaoying Zhuang, Runqiu Huang, Chao Liang, and Timon Rabczuk
Volume 2014, Article ID 179169, 11 pages

Dynamic Fracture in Thin Shells Using Meshfree Method, Y. Shie

Volume 2014, Article ID 262494, 8 pages

Evaluating the Applicability of Fracture Criteria to Predict the Crack Evolution Path of Dolomite Based
on SCB Experiments and FEM, Cunbao Li, Lingzhi Xie, Li Ren, Heping Xie, and Jun Wang
Volume 2013, Article ID 959806, 13 pages

POD for Real-Time Simulation of Hyperelastic Soft Biological Tissue Using the Point Collocation
Method of Finite Spheres, Suleiman Banihani, Timon Rabczuk, and Thakir Almomani
Volume 2013, Article ID 386501, 9 pages

Homotopy Iteration Algorithm for Crack Parameters Identification with Composite Element Method,
Ling Huang, Zhongrong Lv, Weihuan Chen, and Jike Liu
Volume 2013, Article ID 524562, 10 pages

An Efficient Approach for Identifying Constitutive Parameters of the Modified Oyane Ductile Fracture
Criterion at High Temperature, Sergei Alexandrov, Yusof Mustafa, and Mohd Yazid Yahya
Volume 2013, Article ID 514945, 4 pages

Study on Analog Theory of Rock Mass Simulation and Its Engineering Application, Sun Shaorui,
Lu Yexu, Xu Yuanyuan, Liu Jin, and Wei Jihong
Volume 2013, Article ID 491069, 11 pages

Explicit Dynamic Finite Element Method for Failure with Smooth Fracture Energy Dissipations,
Jeong-Hoon Song, Thomas Menouillard, and Alireza Tabarraei
Volume 2013, Article ID 293861, 12 pages

Explicit Dynamic Finite Element Method for Predicting Implosion/Explosion Induced Failure of Shell
Structures, Jeong-Hoon Song, Patrick Lea, and Jay Oswald
Volume 2013, Article ID 957286, 11 pages
Hindawi Publishing Corporation
Mathematical Problems in Engineering
Volume 2014, Article ID 593041, 2 pages
http://dx.doi.org/10.1155/2014/593041

Editorial
Computational Methods for Fracture

Timon Rabczuk,1,2 Stéphane P. A. Bordas,3,4 and Goangseup Zi2

1
Bauhaus Universität Weimar, 99423 Weimar, Germany
2
Korea University, Seoul 136-701, Republic of Korea
3
Faculty of Science, Technology and Communication, Research Unit in Engineering Science, University of Luxembourg,
Campus Kirchberg, G 007, 6 rue Richard Coudenhove-Kalergi, 1359 Luxembourg, Luxembourg
4
Cardiff School of Engineering, Cardiff University, Room W2.43, Queen’s Buildings, The Parade, Cardiff 24 3AA, UK

Correspondence should be addressed to Timon Rabczuk; [email protected]

Received 15 May 2014; Accepted 15 May 2014; Published 20 July 2014

Copyright © 2014 Timon Rabczuk et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

The numerical study of fracture has far-reaching applica- Dynamic fracture is in itself a wide field of study. Papers in
tions throughout engineering and science. Its importance is this issue focus on explicit dynamics and tackle, in particular,
becoming even more significant that engineers and materials the issue of energy dissipation during crack growth and the
scientists are thriving to devise new lighter and stronger simulation of fracture in thin shells due to dynamic and
materials from the bottom up. implosive or explosive loading.
Simulating fracture requires devising suitable models, Model reduction methods have recently been developed
discretizing the resulting partial differential equations, and for fracture simulations. Algebraic model reduction such as
solving them numerically. Each of those three steps poses the proper orthogonal decomposition (POD) is not inher-
its own difficulties which have been tackled in various ways, ently well-suited to such problems [5–8]. This issue discusses
both academically and for practical applications. one possibility relying on the combination of POD with
This special issue deals with a range of such models: meshfree methods.
discretization and solution methods applied to a number of Parameter identification in fracture mechanics is a topic
problems ranging from rock mechanics to surgical simulation of special interest in practical applications and a necessary
requiring tackling various loading spectra, ranging from fast step to provide convincing and predictive modeling and sim-
dynamics to quasistatic loading, and leading to a number of ulation tools. Two methods are discussed in this issue, both
different failure modes, from brittle to ductile fracture. for crack parameter identification and multifield problems in
fracture.
The topics of this issue can be decomposed into five
Through these five topics, we believe the issue gives a fair
groups:
reflection of the current state of the art, with a heavy focus
Models have been developed with special emphasis on
on modeling and simulation methods, although covering the
rock mechanics and multi-field problems in fracture [1, 2]
whole spectrum of applications and methodologies would
with applications, for example to coupled thermohydrome-
require a much more substantial volume.
chanical model of jointed hard rock for compressed air energy
storage, and to rock failure [3].
Discretization methods have been heavily investigated, to
address the difficulties faced by the standard finite element
method [4], in particular, associated with remeshing as the
cracks evolve. The issue discusses recent developments in Timon Rabczuk
meshless methods, including a posteriori error estimation Stéphane P. A. Bordas
and adaptive methods. Goangseup Zi
2 Mathematical Problems in Engineering

References
[1] X. Zhuang, C. Augarde, and S. Bordas, “Accurate fracture
modelling using meshless methods, the visibility criterion and
level sets: Formulation and 2D modelling,” International Journal
for Numerical Methods in Engineering, vol. 86, no. 2, pp. 249–
268, 2011.
[2] X. Zhuang, C. E. Augarde, and K. M. Mathisen, “Fracture mod-
eling using meshless methods and levels sets in 3D: framework
and modeling,” International Journal for Numerical Methods in
Engineering, vol. 92, no. 11, pp. 969–998, 2012.
[3] H. Zhu, X. Zhuang, Y. Cai, and G. Ma, “High rock slope stability
analysis using the enriched meshless Shepard and least squares
method,” International Journal of Computational Methods, vol.
8, no. 2, pp. 209–228, 2011.
[4] Y. Cai, X. Zhuang, and C. Augarde, “A new partition of unity
finite element free from the linear dependence problem and
possessing the delta property,” Computer Methods in Applied
Mechanics and Engineering, vol. 199, no. 17–20, pp. 1036–1043,
2010.
[5] P. Kerfriden, P. Gosselet, S. Adhikari, and S. P. A. Bordas, “Bridg-
ing proper orthogonal decomposition methods and augmented
Newton-Krylov algorithms: an adaptive model order reduction
for highly nonlinear mechanical problems,” Computer Methods
in Applied Mechanics and Engineering, vol. 200, no. 5–8, pp.
850–866, 2011.
[6] P. Kerfriden, J. C. Passieux, and S. P. A. Bordas, “Local/global
model order reduction strategy for the simulation of quasi-
brittle fracture,” International Journal for Numerical Methods in
Engineering, vol. 89, no. 2, pp. 154–179, 2012.
[7] P. Kerfriden, O. Goury, T. Rabczuk, and S. P. A. Bordas, “A par-
titioned model order reduction approach to rationalise compu-
tational expenses in nonlinear fracture mechanics,” Computer
Methods in Applied Mechanics and Engineering, vol. 256, pp.
169–188, 2013.
[8] P. Kerfriden, K. M. Schmidt, T. Rabczuk, and S. P. A. Bordas,
“Statistical extraction of process zones and representative sub-
spaces in fracture of random composites,” International Journal
for Multiscale Computational Engineering, vol. 11, no. 3, pp. 253–
287, 2013.
Hindawi Publishing Corporation
Mathematical Problems in Engineering
Volume 2014, Article ID 646514, 10 pages
http://dx.doi.org/10.1155/2014/646514

Research Article
Stable Cracking Particles Method Based on Stabilized Nodal
Integration and Updated Lagrangian Kernel

S. Xu
School of Mechanical Engineering, Jiangnan University, China

Correspondence should be addressed to S. Xu; [email protected]

Received 27 November 2013; Revised 27 January 2014; Accepted 17 February 2014; Published 7 May 2014

Academic Editor: Goangseup Zi

Copyright © 2014 S. Xu. This is an open access article distributed under the Creative Commons Attribution License, which permits
unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

A stable cracking particles method (CPM) based on updated Lagrangian kernels is proposed. The idea of CPM is to model the
crack topology by a set of cracked particles. Hence no representation of the crack surface is needed making the method useful
for problems involving complex fracture patterns as they occur in dynamics and under fast loading conditions. For computational
efficiency, nodal integration is exploited in the present paper. In order to avoid instabilities, a scheme is presented to stabilized the
integration. Moreover, a set of simple cracking rules are proposed in order to prevent numerical fracture. The method is applied to
two benchmark problems and shows good accuracy.

1. Introduction naturally with relatively coarse discretizations as the process

zone does not need to be smeared over several particles. The
Meshless methods have been a competitor to finite element CPM has been applied to numerous challenging problems
method due to their ability to add particles and model large [55–59]. However, due to the “discontinuous” representation
deformations, dynamic fracture, and fragmentation with ease of the crack surface, stresses might be transferred over the
[1–14]. In contrast to finite element methods that require opening crack leading to spurious cracking [60–63].
the deletion of elements for complex dynamic fracture prob- In this paper, we present a stabilized nodal integrated
lems involving penetration and perforation, meshless meth- CPM based on updated Lagrangian kernels that alleviates
ods [14–20] treat such problems quite naturally. However, several shortcomings of the original CPM. In particular, the
meshless methods are computationally expensive. Moreover, following occurs.
numerical fracture is reported when fracture and material
modelling is not accounted for carefully [21–23]. Recently (i) We propose a set of simple cracking rules according to
developed efficient methods such as finite element methods [64] in order to avoid spurious cracking in the CPM.
with edge rotations [24–27], phase-field models [28, 29], and The original CPM shows suffering from spurious
partition-of-unity enriched finite element [30–44] and mesh- cracking adjacent to the crack surface that might
less methods have been widely applied to fracture problems artificially increase the dissipated energy.
with a few number of cracks [23, 45–52]. However, their (ii) We combine for the first time the stabilized nodal
application to complex dynamic fracture and fragmentation integration and the CPM. CPM was only used in
remains a major challenge. A powerful method for complex combination with computationally expensive Gauss
fracture is the cracking particles method (CPM) [53, 54] that quadrature, stress-point integration and nodal inte-
is based on enriching nodes with step-enrichment function gration. Nodal integration is computationally very
once a fracture criterion is met. The crack’s topology is efficient but suffers from instabilities. Stress-point
modeled by discrete plane crack segments and the well- integration eliminates these instabilities but cracks are
posedness of the initial boundary value problem is restored restricted to cross nodes—not stress points. More-
by means of cohesive zone models. The advantage of the CPM over, there is still the need to update the position of
is its robustness and efficiency. Complex fracture is modeled the stress points. Therefore, we employ the stabilized
2 Mathematical Problems in Engineering

Crack

(a) Continuous crack (b) Discretization with crack segments

Figure 1: (a) Continuous crack and (b) representation of the crack with discrete cohesive crack segments.

conforming nodal integration [65] also used in the where 𝑤(X − X𝐼 , ℎ) denotes the kernel function and ℎ its
smoothed finite element method [66–70]. support size; p(X) = (1, 𝑋, 𝑌) is a linear polynomial basis. We
(iii) While the original CPM employs a Lagrangian kernel point out again that the kernel function is expressed in terms
that is only applicable to moderate deformations, an of material coordinates and therefore is called Lagrangian
updated Lagrangian kernel formulation is proposed kernel [22]. It was shown by [22] that a Lagrangian kernel
here for the first time. It guarantees the applicability avoids numerical fracture often observed for simulations
of the method to extremely large deformation. based on Eulerian kernels. However, the Lagrangian kernel
formulation limits the amount of large deformations. There-
The paper is organized as follows. First, we present fore, we update the kernel functions every 𝑛th time step
the CPM. Then, the weak form is stated and the discrete referring to a new reference configuration and hence call this
equations are derived. Subsequently, the fracture model and kernel an updated Lagrangian kernel. A similar approach was
the cohesive zone models are discussed before the paper ends mentioned in [54].
with examples and conclusions. The discontinuous part of the displacement field is
obtained by simply multiplying the shape function with
2. CPM enrichment functions accounting for the jump in the dis-
placement field (see Figure 2):
The key idea of the CPM is to decompose the displacement
field into two parts: the continuous part u𝐶, sometimes also u (X, 𝑡) = ∑ 𝑁𝐼 (X) HΓ𝑐 (X) q𝐼 (𝑡) . (5)
referred to as the “usual” part, and the discontinuous or 𝐼∈N𝑐
“enriched” part u𝐷:
N𝑐 being the set of cracking particles q𝐼 (𝑡) are additional
u (X, 𝑡) = u𝐶 (X, 𝑡) + u𝐷 (X, 𝑡) . (1) degrees of freedom (DOF) and the enrichment function is the
The discretization of the continuous part of the displacement step function
field is based on moving least squares (MLS) shape functions X ⋅ nΓ𝑐
[71] of linear completeness: H (X) = 󵄩󵄩 󵄩. (6)
󵄩󵄩X ⋅ nΓ𝑐 󵄩󵄩󵄩
𝐶 󵄩 󵄩
u (X, 𝑡) = ∑ 𝑁𝐼 (X) u𝐼 (𝑡) , (2)
𝐼∈N The key strength of the CPM is that it does not require any
where 𝑁𝐼 (X) are the meshless shape functions of node 𝐼 representation of the crack’s topology. In the CPM, the crack
at position X and u𝐼 (𝑡) are nodal parameters at time 𝑡. topology is described as a set of crack segments as illustrated
Note that (1) the nodal parameters are not the true physical in Figure 1 allowing the simulation of very complex crack
displacement values at node 𝐼 and (2) the shape functions are patterns with ease.
expressed in terms of material coordinates X. It can be shown It is advantageous to write the approximation of the
that the shape functions are given by displacement field in vector matrix notation:

𝑁𝐼 (X) = p𝑇 (X) 𝐴−1 (X) D𝐼 (X) (3) d = ND (7)

with and store the “standard” and enriched shape functions in the
𝑇 matrix N; the DOFs of the “standard” and enriched part of
D𝐼 (X) = 𝑤 (X − X𝐼 , ℎ) p (X𝐼 ) ,
the displacement field are stored in the vector D. Moreover,
(4) we can define the matrix [[N]] and the matrix containing the
A𝐼 (X) = ∑ 𝑤 (X − X𝐼 , ℎ) p (X𝐼 ) p𝑇 (X𝐼 ) ,
𝐼∈N derivatives of the meshless shape functions B needed later.
Mathematical Problems in Engineering 3

n
nct
xI

Figure 2: Normal of the crack segment.

Figure 4: Crack with crack tip node (solid line) at time 𝑛 and nodes
that meet cracking criterion at time 𝑛 + 1 (dashed line).

1 MPa

40 mm

20 mm
1 MPa
100 mm
(a) (b)
50 mm
Figure 3: (a) Spurious cracking during crack propagation and (b)
crack propagation without spurious cracking. Figure 5: Plate with a horizontal initial notch under tensile trac-
tions.

Though the CPM can handle complex crack patterns, it a prescribed tolerance. According to Figure 4, the deviation
might lead to spurious cracking adjacent to the crack that in this angle is calculated by
can be avoided by a set of simple rules as shown in [64].
We propose similar rules and firstly distinguish between ⏟⏟⏟⏟⏟⏟⏟ (n𝑐𝑡 ⋅ n𝐼 )
𝛾max = max ̃𝑐,
∀𝐼 ∈ N
propagation and initiating cracks by a simple criterion based ̃𝑐
𝐼∈N
on a circular support domain. When a cracking particle does (8)
not find another cracking particle within the circular domain ⏟⏟⏟⏟⏟⏟⏟ (n𝑐𝑡 ⋅ n𝐼 )
𝛾min = min ̃𝑐
∀𝐼 ∈ N
of size 𝛼ℎ, then a new crack is initiated. The factor 𝛼 is ̃𝑐
𝐼∈N
chosen to be 1.1 in all simulations. Secondly, we define a zone
adjacent to a crack surface where no new cracks can evolve; ̃ 𝑐 ⊂ N𝑐 being the set of new cracking particles. Crack
N
see Figure 3. branching is assumed when 𝛾max − 𝛾min ≥ TOL.
As such a criterion might prevent crack branching,
these rules need to be adjusted at the crack tip. Since the 3. Weak Form and Discretization
crack surface is not continuous in the CPM, we employ
the following simple algorithm to detect the crack tip for The linear momentum equation is
branching cracks: crack branching occurs when the angle of
existing and newly created particles at the crack tip exceeds ∇ ⋅ P + 󰜚b = 󰜚u,̈ X ∈ Ω, (9)
4 Mathematical Problems in Engineering

(a) 4141 nodes; with spurious cracking (b) 16281 nodes; with spurious cracking

(c) 4141 nodes; without spurious cracking (d) 16281 nodes; without spurious cracking

Figure 6: Crack pattern for the crack branching problem.

×102 ×102
13 13
12 12
11 11
10 10
9 9
Crack tip speed (m/s)
Crack tip speed (m/s)

8 8
7 7
6 6
5 5
4 4
3 3
2 2
1 1
0 0
0 10 20 30 40 50 60 0 10 20 30 40 50 60
Time (𝜇s) Time (𝜇s)
4141 nodes Without spurious cracking
7236 nodes With spurious cracking
16281 nodes
(a) Without spurious cracking (b) Comparison with and without spurious cracking

Figure 7: Crack tip speed for the crack branching problem.

where P is the nominal stress tensor, 󰜚 is the density, b We solve the equation of motion in weak form that can
are body forces, and the superimposed dots denote material be stated in variational form. Find the displacement field u ∈
time derivatives. The displacement and traction boundary U ∀𝛿u ∈ V such that the first variation in the energy is zero:
conditions are
𝛿𝑊 = 𝛿𝑊int − 𝛿𝑊ext + 𝛿𝑊inertia − 𝛿𝑊coh = 0
u = u, X ∈ Γu (11)
∀𝛿u ∈ U0
n𝑡 ⋅ P = t, X ∈ Γ𝑡 (10)
with
n𝑐 ⋅ P = t𝑐 ([[u]]) , X ∈ Γ𝑐 ,
𝛿𝑊int = ∫ ∇𝛿u : P 𝑑Ω,
Ω
where the index 𝑐 refers to the crack, the index 𝑡 refers to
traction boundaries, and the index 𝑢 refers to displacement
𝛿𝑊ext = ∫ 𝛿u ⋅ t 𝑑Γ + ∫ 󰜚𝛿u ⋅ b 𝑑Ω,
boundaries. Γ𝑡 Ω
Mathematical Problems in Engineering 5

50 mm 50 mm or in vector-matrix form:
int
M𝐼𝐽 D𝐽 = Fext
𝐼 − F𝐼 (16)

with
75 mm
𝑇 𝑇 𝑇
Fext
𝐼 = ∫ (N𝐼 ) t 𝑑Γ + ∫ (N𝐼 ) b 𝑑Ω + ∫ [[(N𝐼 ) ]] t𝑐 𝑑Γ,
Γ𝑡 Ω Γ𝑐

𝑇
25 mm Fint
𝐼 = ∫ (B𝐼 ) P 𝑑Ω,
Ω

M𝐼𝐽 = ∫ 󰜚N𝐼 N𝑇𝐽 𝑑Ω.

Ω
(17)

For computational efficiency, nodal integration is

employed. Nodal integration does not drastically reduce the
number of quadrature points; it was shown, for example,
by [72] that the critical time step is increased by orders of
Figure 8: The Kalthoff problem. magnitude; care has to be taken in the existence of cracks
[73]. However, nodal integration leads to instabilities due
to rank deficiency. Therefore, we employ the stabilized
𝛿𝑊inertia = ∫ 󰜚𝛿u ⋅ ü𝑑Ω, conforming nodal integration technique as mentioned
Ω
previously. Stabilized conforming nodal integration has
𝛿𝑊coh = ∫ 𝛿 [[u]] ⋅ t𝑐 𝑑Γ. also shown great performance for fracture problems in
Γ𝑐 the context of the finite element method; see [74–80]. The
(12) explicit central difference time integration scheme is used
(for the time integration).
The approximation spaces U and U0 are given by

U = {u (X, 𝑡) | u ∈ 𝐻1 , u = u on Γ𝑢 , 4. Fracture Criterion and Cohesive

Zone Model
u discontinuous on Γ𝑐 } ,
(13) Fracture is governed by the maximum principal stress.
1 The crack is introduced perpendicular to the direction of
U0 = {𝛿u | 𝛿u ∈ 𝐻 , 𝛿u = 0 on Γ𝑢 ,
the maximum principal stress. Complex crack patterns are
𝛿u discontinuous on Γ𝑐 } . obtained automatically as the method does not require any
representation of the crack’s topology. There is no need to
The essential and natural boundary conditions are given distinguish between crack nucleation and crack propagation
by and complex crack patterns including crack branching are
natural outcome of the simulation.
u = u, X ∈ Γ𝑢 The cohesive zone model related the jump in the dis-
n𝑡 ⋅ P = t, X ∈ Γ𝑡 (14) placement field [[u]] to the cohesive traction at the crack
surface. The discontinuous displacement field can be split
n𝑐 ⋅ P = t𝑐 ([[u]]) , X ∈ Γ𝑐 , into a part acting perpendicular and another tangential to the
crack surface:
the index 𝑐 referring to crack boundaries, the index 𝑡 to
natural boundaries, and the index 𝑢 to essential boundaries. 𝛿𝑛 = [[u]] ⋅ n
By substituting the approximation of the displacement (18)
󵄨 󵄨
field u and the virtual displacement field 𝛿u into the weak 𝛿𝑡 = 󵄨󵄨󵄨[[u]] − 𝛿𝑛 n󵄨󵄨󵄨 .
form (11), the discrete system of equations is obtained:
𝑛 𝑛
Defining an effective crack opening displacement as sug-
∑ ∫ ∇𝛿u𝐽 : P 𝑑Ω − ∑ ∫ 𝛿u ⋅ t 𝑑Γ gested by [81]
𝐽=1 Ω𝐽 𝐽=1 Γ𝑡,𝐽
𝑛 𝛿 = √𝛽2 𝛿𝑡2 − 𝛿𝑛2 (19)
− ∑ ∫ 󰜚𝛿u ⋅ b 𝑑Ω + ∫ 𝛿 [[u]] ⋅ t𝑐 𝑑Γ (15)
𝐽=1 Ω𝐽 Γ𝑐,𝐽 an effective cohesive zone model can be derived by
𝑛
+ ∑ ∫ 󰜚𝛿u ⋅ ü𝑑Ω = 0 𝑡max
𝑡= 𝛿 if 𝛿 ≤ 𝛿max or 𝛿̇< 0. (20)
𝐽=1 Ω𝐽 𝛿max
6 Mathematical Problems in Engineering

(a) 10201 nodes; with spurious cracking (b) 10201 nodes; without spurious cracking

(c) 40804 nodes; with spurious cracking (d) 40804 nodes; without spurious cracking

Figure 9: Final crack path of the Kalthoff problem.

The traction vector is then obtained by the propagating crack is illustrated in Figure 7. We note the
following.
𝑡 2
t= (𝛽 𝛿𝑡 + 𝛿𝑛 n) , (21) (i) The cracking rules do not influence the crack speed
𝛿 indicating that much less energy is dissipated in the
spurious cracks.
where 𝛽 determines the amount of tangential and normal
tractions. (ii) The crack propagates faster just before it branches.
(iii) The maximum crack speed is far below the Raleigh
wave speed. This agrees well with experimental obser-
5. Results vations of the microbranch instability problem as
5.1. Crack Branching. Let us consider a specimen with initial reported by [84–86]. The ability to capture this phys-
crack as illustrated in Figure 5. It is subjected to uniaxial ical behaviour naturally is one advantage of the CPM
tensile loading of 𝜎 = 1 MPa at the top and bottom. over other methods with continuous crack surface
This classical benchmark problem of dynamic fracture has [23, 39, 51, 87].
been studied by several people to test the accuracy and
robustness of their computational method [39, 53, 82, 83]. 5.2. The Kalthoff Experiment. The second classical bench-
Also, experimental data can be found for such type of mark example studied here is the Kalthoff problem [88].
problems; see, for example, [84–86]. The material properties Therefore consider the double-notched specimen under
for this example are modulus of elasticity 𝐸 = 32, 000 MPa impact loading (V = 20 m/s) as illustrated in Figure 8. The
and Poisson’s ratio ] = 0.20. The maximum crack speed impact leads to mode I dominated fracture with a crack
is restricted by the Rayleigh wave speed 𝑐𝑅 = 2119.0 m/s. propagating almost orthogonal to the impact loading.
Discretizations ranging from only 4000 nodes up to more We exploit the symmetry of the model and carry out
than 16000 nodes are investigated. Also the influence of the simulations of discretizations ranging from 10,000 to 40,000
cracking rules will be demonstrated in this section. particles. According to [89], the modulus of elasticity in these
The fracture patterns at various time steps are depicted experiments is 𝐸 = 190 GPa, the initial density is 󰜚0 =
in Figure 6. Spurious cracks adjacent to the “main” crack are 8000 kg/m3 , and Poisson’s ratio is ] = 0.3. Figure 9 shows
observed when no cracking rules are applied. The velocity of the final fracture pattern that matches the experimental
Mathematical Problems in Engineering 7

2000
[4] S. Li and B. C. Simonson, “Meshfree simulation of ductile crack
propagation,” International Journal of Computational Methods
in Engineering Science and Mechanics, vol. 6, pp. 1–19, 2003.
1750
[5] S. Li and B. C. Simonson, “Meshfree simulation of ductile crack
1500 propagation,” International Journal of Computational Methods
in Engineering Science and Mechanics, vol. 60, no. 8, pp. 1425–
Crack tip speed (m/s)

1250 1450, 2004.

[6] S. Li, W. Hao, and W. K. Liu, “Mesh-free simulations of shear
1000 banding in large deformation,” International Journal of Solids
and Structures, vol. 37, no. 48, pp. 7185–7206, 2000.
750 [7] S. Li, W. Hao, and W. K. Liu, “Numerical simulations of large
deformation of thin shell structures using meshfree methods,”
500 Computational Mechanics, vol. 25, no. 2, pp. 102–116, 2000.
[8] S. Li, W. K. Liu, A. J. Rosakis, T. Belytschko, and W. Hao, “Mesh-
250 free Galerkin simulations of dynamic shear band propagation
and failure mode transition,” International Journal of Solids and
0 Structures, vol. 39, no. 5, pp. 1213–1240, 2002.
0 10 20 30 40 50 60 70
Time (𝜇s)
[9] A. G. Petschek and L. D. Libersky, “Cylindrical smoothed
particle hydrodynamics,” Journal of Computational Physics, vol.
10201 nodes, no spurious cracking 109, no. 1, pp. 76–83, 1993.
40804 nodes, no spurious cracking
[10] T. Rabczuk and J. Eibl, “Modelling dynamic failure of concrete
40804 nodes, spurious cracking
with meshfree methods,” International Journal of Impact Engi-
Figure 10: Crack tip speed and cohesive fracture energy of the neering, vol. 32, no. 11, pp. 1878–1897, 2006.
Kalthoff problem. [11] T. Rabczuk and J. Eibl, “Simulation of high velocity concrete
fragmentation using SPH/MLSPH,” International Journal for
Numerical Methods in Engineering, vol. 56, no. 10, pp. 1421–1444,
2003.
data. Figure 10 reveals that the crack speed here is also not [12] T. Rabczuk, J. Eibl, and L. Stempniewski, “Numerical anal-
influenced by the cracking rules. However, the crack looks ysis of high speed concrete fragmentation using a meshfree
more erratic without using the cracking rules. Lagrangian method,” Engineering Fracture Mechanics, vol. 71,
no. 4–6, pp. 547–556, 2004.
6. Conclusions [13] T. Rabczuk, S. P. Xiao, and M. Sauer, “Coupling of mesh-free
methods with finite elements: Basic concepts and test results,”
In conclusion, a cracking particles method based on nodal Communications in Numerical Methods in Engineering, vol. 22,
integration and updated Lagrangian kernels was proposed no. 10, pp. 1031–1065, 2006.
where a set of simple cracking rules was suggested in order [14] P. W. Randles and L. D. Libersky, “Recent improvements in
to avoid spurious cracking. The stabilized nodal integration SPH modeling of hypervelocity impact,” International Journal
guarantees the computational efficiency of the method while of Impact Engineering, vol. 20, no. 6–10, pp. 525–532, 1997.
maintaining the stability. Moreover, the updated Lagrangian [15] T. Belytschko, Y. Y. Lu, L. Gu, and M. Tabbara, “Element-free
kernel ensures that the method remains applicable also for galerkin methods for static and dynamic fracture,” International
extremely large deformations. Journal of Solids and Structures, vol. 32, no. 17-18, pp. 2547–2570,
The method was applied to two benchmark examples 1995.
where the performance of the method was demonstrated. [16] T. Belytschko, Y. Krongauz, D. Organ, M. Fleming, and P. Krysl,
“Meshless methods: an overview and recent developments,”
Computer Methods in Applied Mechanics and Engineering, vol.
Conflict of Interests 139, no. 1–4, pp. 3–47, 1996.
[17] W. K. Liu, S. J. Jun, and Y. F. Zhang, “Reproducing kernel
The author declares that there is no conflict of interests particle methods,” International Journal for Numerical Methods
regarding the publication of this paper. in Fluids, vol. 20, no. 8-9, pp. 1081–1106, 1995.
[18] V. P. Nguyen, T. Rabczuk, S. Bordas, and M. Duflot, “Meshless
References methods: a review and computer implementation aspects,”
Mathematics and Computers in Simulation, vol. 79, no. 3, pp.
[1] S. Li and W. K. Liu, “Meshfree and particle methods and their 763–813, 2008.
applications,” Applied Mechanics Reviews, vol. 55, no. 1, pp. 1–34, [19] T. Rabczuk and T. Belytschko, “Adaptivity for structured mesh-
2002. free particle methods in 2D and 3D,” International Journal for
[2] S. Li and W. K. Liu, Meshfree Particle Methods, Springer, 2004. Numerical Methods in Engineering, vol. 63, no. 11, pp. 1559–1582,
[3] S. Li and D. C. Simkins Jr., “Conserving Galerkin weak formu- 2005.
lations for computational fracture mechanics,” Communications [20] P. W. Randles and L. D. Libersky, “Normalized SPH with
in Numerical Methods in Engineering, vol. 18, no. 12, pp. 835–850, stress points,” International Journal for Numerical Methods in
2002. Engineering, vol. 48, no. 10, pp. 1445–1462, 2000.
8 Mathematical Problems in Engineering

[21] T. Belytschko, Y. Guo, W. K. Liu, and S. P. Xiao, “A unified International Journal for Numerical Methods in Engineering, vol.
stability analysis of meshless particle methods,” International 72, no. 12, pp. 1391–1433, 2007.
Journal for Numerical Methods in Engineering, vol. 48, no. 9, pp. [37] N. Moës, J. Dolbow, and T. Belytschko, “A finite element method
1359–1400, 2000. for crack growth without remeshing,” International Journal for
[22] T. Rabczuk, T. Belytschko, and S. P. Xiao, “Stable particle Numerical Methods in Engineering, vol. 46, no. 1, pp. 131–150,
methods based on Lagrangian kernels,” Computer Methods in 1999.
Applied Mechanics and Engineering, vol. 193, no. 12–14, pp. 1035– [38] S. S. Nanthakumar, T. Lahmer, and T. Rabczuk, “Detection
1063, 2004. of flaws in piezoelectric structures using extended FEM,”
[23] T. Rabczuk, S. Bordas, and G. Zi, “A three dimensional International Journal for Numerical Methods in Engineering, vol.
meshfree method for static and dynamic multiple crack nucle- 96, no. 6, pp. 373–389, 2013.
ation/propagation with crack path continuity,” Computational [39] J.-H. Song, P. M. A. Areias, and T. Belytschko, “A method
Mechanics, vol. 40, no. 3, pp. 473–495, 2007. for dynamic crack and shear band propagation with phantom
[24] P. Areias and T. Rabczuk, “Finite strain fracture of plates nodes,” International Journal for Numerical Methods in Engi-
and shells with configurational forces and edge rotations,” neering, vol. 67, no. 6, pp. 868–893, 2006.
International Journal for Numerical Methods in Engineering, vol. [40] T. Strouboulis, K. Copps, and I. Babuška, “The generalized
94, no. 12, pp. 1099–1122, 2013. finite element method: An example of its implementation
[25] P. Areias, D. Dias da Costa, J. M. Sargado, and T. Rabczuk, and illustration of its performance,” International Journal for
“Element-wise algorithm for modeling ductile fracture with the Numerical Methods in Engineering, vol. 47, no. 8, pp. 1401–1417,
rousselier yield function,” Computational Mechanics, vol. 52, no. 2000.
6, pp. 1429–1443, 2013. [41] G. Ventura, E. Budyn, and T. Belytschko, “Vector level sets for
[26] P. Areias, T. Rabczuk, and P. P. Camanho, “Initially rigid cohe- description of propagating cracks in finite elements,” Interna-
sive laws and fracture based on edge rotations,” Computational tional Journal for Numerical Methods in Engineering, vol. 58, no.
Mechanics, vol. 52, no. 4, pp. 931–947, 2013. 10, pp. 1571–1592, 2003.
[27] P. Areias, T. Rabczuk, and D. Dias da Costa, “Element-wise [42] G. Zi and T. Belytschko, “New crack-tip elements for XFEM
fracture algorithm based on rotation of edges,” Engineering and applications to cohesive cracks,” International Journal for
Fracture Mechanics, vol. 110, pp. 113–137, 2013. Numerical Methods in Engineering, vol. 57, no. 15, pp. 2221–2240,
2003.
[28] C. Miehe, M. Hofacker, and F. Welschinger, “A phase field model
[43] G. Zi, J.-H. Song, E. Budyn, S.-H. Lee, and T. Belytschko, “A
for rate-independent crack propagation: Robust algorithmic
method for growing multiple cracks without remeshing and its
implementation based on operator splits,” Computer Methods
application to fatigue crack growth,” Modelling and Simulation
in Applied Mechanics and Engineering, vol. 199, no. 45–48, pp.
in Materials Science and Engineering, vol. 12, no. 5, pp. 901–915,
2765–2778, 2010.
2004.
[29] C. Miehe, F. Welschinger, and M. Hofacker, “Thermodynam-
[44] G. Zi, H. Chen, J. Xu, and T. Belytschko, “The extended finite
ically consistent phase-field models of fracture: variational
element method for dynamic fractures,” Shock and Vibration,
principles and multi-field FE implementations,” International
vol. 12, no. 1, pp. 9–23, 2005.
Journal for Numerical Methods in Engineering, vol. 83, no. 10,
pp. 1273–1311, 2010. [45] F. Amiri, C. Anitescu, M. Arroyo, S. P. A. Bordas, and T.
Rabczuk, “XLME interpolants, a seamless bridge between
[30] F. Armero and K. Garikipati, “Recent advances in the analysis XFEM and enriched meshless methods,” Computational
and numericalsimulation of strain localization in inelastic Mechanics, vol. 53, no. 1, pp. 45–57, 2014.
solids,” in Proceedings of the 4th Computational Plasticity:
[46] S. Bordas, T. Rabczuk, and G. Zi, “Three-dimensional crack
Fundamentals and Applications Conference (COMPLAS ’00), pp.
initiation, propagation, branching and junction in non-linear
635–640, 2000.
materials by an extended meshfree method without asymptotic
[31] T. Belytschko and T. Black, “Elastic crack growth in finite enrichment,” Engineering Fracture Mechanics, vol. 75, no. 5, pp.
elements with minimal remeshing,” International Journal for 943–960, 2008.
Numerical Methods in Engineering, vol. 45, no. 5, pp. 601–620,
[47] T. Rabczuk and P. Areias, “A meshfree thin shell for arbitrary
1999.
evolving cracks based on an extrinsic basis,” Computer Modeling
[32] T. Belytschko, H. Chen, J. Xu, and G. Zi, “Dynamic crack prop- in Engineering and Sciences, vol. 16, no. 2, pp. 115–130, 2006.
agation based on loss of hyperbolicity and a new discontinuous [48] T. Rabczuk, G. Zi, S. Bordas, and H. Nguyen-Xuan, “A geomet-
enrichment,” International Journal for Numerical Methods in rically non-linear three-dimensional cohesive crack method for
Engineering, vol. 58, no. 12, pp. 1873–1905, 2003. reinforced concrete structures,” Engineering Fracture Mechan-
[33] T. Chau-Dinh, G. Zi, P.-S. Lee, T. Rabczuk, and J.-H. Song, ics, vol. 75, no. 16, pp. 4740–4758, 2008.
“Phantom-node method for shell models with arbitrary cracks,” [49] T. Rabczuk, S. Bordas, and G. Zi, “On three-dimensional
Computers & Structures, vol. 92-93, pp. 242–246, 2012. modelling of crack growth using partition of unity methods,”
[34] G. A. Dilts, “Moving-least-squares-particle hydrodynamics. I. Computers & Structures, vol. 88, no. 23-24, pp. 1391–1411, 2010.
Consistency and stability,” International Journal for Numerical [50] T. Rabczuk, R. Gracie, J.-H. Song, and T. Belytschko, “Immersed
Methods in Engineering, vol. 44, no. 8, pp. 1115–1155, 1999. particle method for fluid-structure interaction,” International
[35] G. A. Dilts, “Moving least-squares particle hydrodynamics. Journal for Numerical Methods in Engineering, vol. 81, no. 1, pp.
II. Conservation and boundaries,” International Journal for 48–71, 2010.
Numerical Methods in Engineering, vol. 48, no. 10, pp. 1503–1524, [51] G. Ventura, J. X. Xu, and T. Belytschko, “A vector level
2000. set method and new discontinuity approximations for crack
[36] C. Linder and F. Armero, “Finite elements with embedded growth by EFG,” International Journal for Numerical Methods
strong discontinuities for the modeling of failure in solids,” in Engineering, vol. 54, no. 6, pp. 923–944, 2002.
Mathematical Problems in Engineering 9

[52] G. Zi, T. Rabczuk, and W. Wall, “Extended meshfree methods [68] T. Nguyen-Thoi, P. Phung-Van, T. Rabczuk, H. Nguyen-Xuan,
without branch enrichment for cohesive cracks,” Computational and C. Le-Van, “Free and forced vibration analysis using the
Mechanics, vol. 40, no. 2, pp. 367–382, 2007. n-sided polygonal cellbased smoothed finite element method
[53] T. Rabczuk and T. Belytschko, “Cracking particles: a simplified (ncs-fem),” International Journal of Computational Methods,
meshfree method for arbitrary evolving cracks,” International vol. 10, no. 1, Article ID 13400082, 2013.
Journal for Numerical Methods in Engineering, vol. 61, no. 13, pp. [69] H. Nguyen-Xuan, T. Rabczuk, T. Nguyen-Thoi, T. N. Tran,
2316–2343, 2004. and N. Nguyen-Thanh, “Computation of limit and shakedown
[54] T. Rabczuk and T. Belytschko, “A three-dimensional large loads using a node-based smoothed finite element method,”
deformation meshfree method for arbitrary evolving cracks,” International Journal for Numerical Methods in Engineering, vol.
Computer Methods in Applied Mechanics and Engineering, vol. 90, no. 3, pp. 287–310, 2012.
196, no. 29-30, pp. 2777–2799, 2007. [70] C. H. Thai, H. Nguyen-Xuan, N. Nguyen-Thanh, T.-H. Le,
T. Nguyen-Thoi, and T. Rabczuk, “Static, free vibration, and
[55] T. Rabczuk and P. M. A. Areias, “A new approach for mod-
buckling analysis of laminated composite Reissner-Mindlin
elling slip lines in geological materials with cohesive models,”
plates using NURBS-based isogeometric approach,” Interna-
International Journal for Numerical and Analytical Methods in
tional Journal for Numerical Methods in Engineering, vol. 91, no.
Geomechanics, vol. 30, no. 11, pp. 1159–1172, 2006.
6, pp. 571–603, 2012.
[56] T. Rabczuk and T. Belytschko, “Application of particle methods [71] T. Belytschko, Y. Y. Lu, and L. Gu, “Element-free Galerkin
to static fracture of reinforced concrete structures,” Interna- methods,” International Journal for Numerical Methods in Engi-
tional Journal of Fracture, vol. 137, no. 1–4, pp. 19–49, 2006. neering, vol. 37, no. 2, pp. 229–256, 1994.
[57] T. Rabczuk and E. Samaniego, “Discontinuous modelling of [72] M. A. Puso, J. S. Chen, E. Zywicz, and W. Elmer, “Meshfree and
shear bands using adaptive meshfree methods,” Computer finite element nodal integration methods,” International Journal
Methods in Applied Mechanics and Engineering, vol. 197, no. 6–8, for Numerical Methods in Engineering, vol. 74, no. 3, pp. 416–
pp. 641–658, 2008. 446, 2008.
[58] T. Rabczuk, P. M. A. Areias, and T. Belytschko, “A meshfree thin [73] H. Talebi, C. Samaniego, E. Samaniego, and T. Rabczuk, “On
shell method for non-linear dynamic fracture,” International the numerical stability and mass-lumping schemes for explicit
Journal for Numerical Methods in Engineering, vol. 72, no. 5, pp. enriched meshfree methods,” International Journal for Numeri-
524–548, 2007. cal Methods in Engineering, vol. 89, no. 8, pp. 1009–1027, 2012.
[59] T. Rabczuk, J.-H. Song, and T. Belytschko, “Simulations of insta- [74] P. Baiz, S. Natarajan, S. Bordas, P. Kerfriden, and T. Rabczuk,
bility in dynamic fracture by the cracking particles method,” “Linear buckling analysis of cracked plates by sfem and xfem
Engineering Fracture Mechanics, vol. 76, no. 6, pp. 730–741, 2009. (smxfem),” Journal of Mechanics of Materials and Structures, vol.
[60] L. Chen and Y. Zhang, “Dynamic fracture analysis using dis- 6, no. 9-10, pp. 1213–1238, 2011.
crete cohesive crack method,” International Journal for Numeri- [75] S. P. A. Bordas, T. Rabczuk, N.-X. Hung et al., “Strain smoothing
cal Methods in Biomedical Engineering, vol. 26, no. 11, pp. 1493– in FEM and XFEM,” Computers & Structures, vol. 88, no. 23-24,
1502, 2010. pp. 1419–1443, 2010.
[61] Y. Dong, S. Wu, S. S. Xu, Y. Zhang, and S. Fang, “Analysis of [76] S. P. A. Bordas, S. Natarajan, P. Kerfriden et al., “On the
concrete fracture using a novel cohesive crack method,” Applied performance of strain smoothing for quadratic and enriched
Mathematical Modelling, vol. 34, no. 12, pp. 4219–4231, 2010. finite element approximations (XFEM/GFEM/PUFEM),” Inter-
national Journal for Numerical Methods in Engineering, vol. 86,
[62] N. Sageresan and R. Drathi, “Crack propagation in concrete
no. 4-5, pp. 637–666, 2011.
using meshless method,” Computer Modeling in Engineering and
Sciences, vol. 32, no. 2, pp. 103–112, 2008. [77] L. Chen, T. Rabczuk, S. P. A. Bordas, G. R. Liu, K. Y. Zeng, and
P. Kerfriden, “Extended finite element method with edge-based
[63] H. X. Wang and S. X. Wang, “Analysis of dynamic fracture strain smoothing (ESm-XFEM) for linear elastic crack growth,”
with cohesive crack segment method,” Computer Modeling in Computer Methods in Applied Mechanics and Engineering, vol.
Engineering & Sciences, vol. 35, no. 3, pp. 253–274, 2008. 212, no. 4, pp. 250–265, 2012.
[64] T. Rabczuk, G. Zi, S. Bordas, and H. Nguyen-Xuan, “A simple [78] H. Nguyen-Xuan, G. R. Liu, S. Bordas, S. Natarajan, and T.
and robust three-dimensional cracking-particle method with- Rabczuk, “An adaptive singular ES-FEM for mechanics prob-
out enrichment,” Computer Methods in Applied Mechanics and lems with singular field of arbitrary order,” Computer Methods in
Engineering, vol. 199, no. 37–40, pp. 2437–2455, 2010. Applied Mechanics and Engineering, vol. 253, pp. 252–273, 2013.
[65] J. S. Chen, C. T. Wu, S. Yoon, and Y. You, “A stabilized [79] N. Vu-Bac, H. Nguyen-Xuan, L. Chen et al., “A node-based
conforming nodal integration for galerkin meshfree-methods,” smoothed extended finite element method (NS-XFEM) for
International Journal for Numerical Methods in Engineering, vol. fracture analysis,” Computer Modeling in Engineering and Sci-
50, pp. 435–466, 2001. ences, vol. 73, no. 4, pp. 331–355, 2011.
[66] G. R. Liu, T. T. Nguyen, K. Y. Dai, and K. Y. Lam, “Theoretical [80] N. Vu-Bac, H. Nguyen-Xuan, L. Chen et al., “A phantom-node
aspects of the smoothed finite element method (SFEM),” Inter- method with edge-based strain smoothing for linear elastic
national Journal for Numerical Methods in Engineering, vol. 71, fracture mechanics,” Journal of Applied Mathematics, vol. 2013,
no. 8, pp. 902–930, 2007. Article ID 978026, 12 pages, 2013.
[67] T. Nguyen-Thoi, H. C. Vu-Do, T. Rabczuk, and H. Nguyen- [81] A. Pandolfi, P. Krysl, and M. Ortiz, “Finite element simulation of
Xuan, “A node-based smoothed finite element method (NS- ring expansion and fragmentation: the capturing of length and
FEM) for upper bound solution to visco-elastoplastic analyses time scales through cohesive models of fracture,” International
of solids using triangular and tetrahedral meshes,” Computer Journal of Fracture, vol. 95, no. 1–4, pp. 279–297, 1999.
Methods in Applied Mechanics and Engineering, vol. 199, no. 45– [82] T. Rabczuk, G. Zi, A. Gerstenberger, and W. A. Wall, “A new
48, pp. 3005–3027, 2010. crack tip element for the phantom-node method with arbitrary
10 Mathematical Problems in Engineering

cohesive cracks,” International Journal for Numerical Methods in

Engineering, vol. 75, no. 5, pp. 577–599, 2008.
[83] X.-P. Xu and A. Needleman, “Numerical simulations of fast
crack growth in brittle solids,” Journal of the Mechanics and
Physics of Solids, vol. 42, no. 9, pp. 1397–1434, 1994.
[84] J. Fineberg, E. Sharon, and G. Cohen, “Crack front waves in
dynamic fracture,” International Journal of Fracture, vol. 121, no.
1-2, pp. 55–69, 2003.
[85] K. Ravi-Chandar, “Dynamic fracture of nominally brittle mate-
rials,” International Journal of Fracture, vol. 90, no. 1-2, pp. 83–
102, 1998.
[86] E. Sharon, S. P. Gross, and J. Fineberg, “Local crack branching
as a mechanism for instability in dynamic fracture,” Physical
Review Letters, vol. 74, no. 25, pp. 5096–5099, 1995.
[87] T. Rabczuk and G. Zi, “A meshfree method based on the
local partition of unity for cohesive cracks,” Computational
Mechanics, vol. 39, no. 6, pp. 743–760, 2007.
[88] J. F. Kalthoff and S. Winkler, “Failure mode transition at high
rates of shear loading,” in Proceedings of the International Con-
ference on Impact Loading and Dynamic Behavior of Materials,
vol. 1, pp. 185–195, 1987.
[89] J. F. Kalthoff, “Modes of dynamic shear failure in solids,”
International Journal of Fracture, vol. 101, no. 1-2, pp. 1–31, 2000.
Hindawi Publishing Corporation
Mathematical Problems in Engineering
Volume 2014, Article ID 721240, 16 pages
http://dx.doi.org/10.1155/2014/721240

Research Article
Error Estimate and Adaptive Refinement in Mixed Discrete
Least Squares Meshless Method

J. Amani,1 A. Saboor Bagherzadeh,1 and T. Rabczuk1,2

1
Institute of Structural Mechanics, Bauhaus-University Weimar, Marienstrasse 15, 99423 Weimar, Germany
2
School of Civil, Environmental and Architectural Engineering, Korea University, Seoul, Republic of Korea

Correspondence should be addressed to A. Saboor Bagherzadeh; [email protected]

Received 3 October 2013; Accepted 29 December 2013; Published 16 February 2014

Academic Editor: Stephane P. A. Bordas

Copyright © 2014 J. Amani et al. This is an open access article distributed under the Creative Commons Attribution License, which
permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

The node moving and multistage node enrichment adaptive refinement procedures are extended in mixed discrete least squares
meshless (MDLSM) method for efficient analysis of elasticity problems. In the formulation of MDLSM method, mixed formulation
is accepted to avoid second-order differentiation of shape functions and to obtain displacements and stresses simultaneously. In
the refinement procedures, a robust error estimator based on the value of the least square residuals functional of the governing
differential equations and its boundaries at nodal points is used which is inherently available from the MDLSM formulation and
can efficiently identify the zones with higher numerical errors. The results are compared with the refinement procedures in the
irreducible formulation of discrete least squares meshless (DLSM) method and show the accuracy and efficiency of the proposed
procedures. Also, the comparison of the error norms and convergence rate show the fidelity of the proposed adaptive refinement
procedures in the MDLSM method.

1. Introduction on the information acquired from the previous computation,

and hence, it is required to implement a suitable node
Adaptivity needs appropriate numerical solutions of prob- generator. on one hand, the mesh movement is more suitable
lems in order to describe high gradient regions and an than mesh enrichment because the problem scale remains
anisotropic behavior of the solution. In an adaptive proce- constant, and on the other hand, its interpolations become
dure, a good error estimator plays a very important role. The too distort in the mesh-based methods [4].
error estimation in numerical methods is obviously as old as In order to avoid these problems, an alternative approach,
the numerical computations themselves. The earliest paper known as meshless methods (MMs), has been developed in
by Richardson [1] proposed an error estimation procedure recent decades to discretize a continuum body only by a
to use in finite difference method. This was followed in finite number of nodes. In MMs the unknowns are inter-
finite element method (FEM) [2]. A particular strength of polated from the nodal values that constitute the problem
the FEM is the well-developed theories of error estimation degrees-of-freedom. The main advantage of MMs is the
and adaptivity. Three h-refinement procedures, namely, mesh fact that the interpolation accuracy is much less affected
movement, mesh enrichment, and remeshing have been by the nodal distribution. Many meshless methods have
proposed for adaptivity [3]. In the mesh movement, the been introduced since Gingold and Monaghan [5] proposed
total number of nodes remains constant, but the location of smoothed particle hydrodynamics (SPH) method. Nayroles
the nodes can change in order to achieve a better overall et al. [6] implemented the diffuse element method (DEM).
distribution of the error. In the mesh enrichment, the original Belytschko et al. [7] presented the Element-Free Galerkin
nodes hold fix and hierarchical nodes or simply more nodes (EFG) method. Liu et al. [8] suggested the reproducing kernel
add to the problem domain based on error distribution. In particle method (RKPM). The other meshless methods that
the remeshing, a completely new nodes is constructed based have been developed in recent years are the Finite Point (FP)
2 Mathematical Problems in Engineering

method [9], the HP clouds method [10], the meshless local The present paper is organized as follows. Formulation
Petrov-Galerkin (MLPG) method [11], the local boundary of the mixed discrete least squares meshless method for
integral equation (LBIE) method [12], the finite cloud (FC) solving the planar elasticity problems is given in Section 2.
method [13] and the discrete least squares meshless (DLSM) In Section 3, an error estimator based on the least square
method [14]. functional residuals is formulated for the MDLSM method to
Researchers used the advantages of MMs for devel- use in the node moving and node enrichment adaptive refine-
oping efficient error estimate and adaptivity procedures. ment procedures. In Section 4, we present some numerical
Rabczuk and Belytschko [15, 16] proposed an adaptive contin- benchmark examples which illustrate the proposed adaptive
uum/discrete crack approach for meshfree particle methods refinement process as well as the efficiency of the error
and also an adaptivity procedure for structured meshfree estimator. Finally, some concluding remarks are addressed in
particle methods in 2D and 3D problems. Yoon et al. [17] Section 5.
worked on enriched meshfree collocation method with dif-
fuse derivatives for elastic fracture. Zi et al. [18] investigated 2. Formulation of Mixed Discrete Least
extended meshfree methods without branch enrichment
for cohesive cracks. Bordas et al. [19] proposed enriched
Squares Meshless Method for Elasticity
meshfree methods without asymptotic enrichment for 3D Consider the following two-dimensional linear elasticity
nonlinear fracture mechanics. Rabczuk and Samaniego [20] problem
worked on discontinuous modelling of shear bands using
adaptive meshfree methods. Zhuang et al. [21–23] investi- −𝜇Δu + (𝜆 + 𝜇) ∇ (∇ ⋅ u) = f in Ω, (1)
gated error control in the EFG method and adaptivity for with displacement and traction boundary conditions as
structured meshfree particle methods in 2D and 3D prob- follow:
lems. The DLSM method was extendted for error estimate
and adaptivity in solid [24, 25] and fluid [26] problems. 𝑢 = 𝑢, V = V, in Γ𝑢 ,
Two different formulations, namely irreducible and
𝜎𝑥 𝑛𝑥 + 𝜏𝑥𝑦 𝑛𝑦 = 𝑡𝑥 , 𝜏𝑥𝑦 𝑛𝑥 + 𝜎𝑦 𝑛𝑦 = 𝑡𝑦 , (2)
mixed formulations have been introduced and used for the
solution of engineering problems. With the mixed formu- in Γ𝑡 ,
lation, the continuity requirement decreases by one order
compared to the irreducible formulation [27]. Use of mixed where Ω is a bounded domain representing the region
formula may result in an improved approximation, in par- occupied by an elastic body, and 𝜆, 𝜇 are the Lame constants
ticular, for the gradient variables, which in turn could result which are defined as
in higher accuracy than possible with the irreducible formu- 𝐸 𝐸]
lation [27]. In the standard mixed FEM, in order to obtain 𝜇= > 0, 𝜆= > 0, (3)
2 (1 + ]) (1 − 2]) (1 + ])
a coefficient matrix which leads to the system of equations
with a unique and stable solution, the polynomial functions where ] is the Poisson ratio, 𝐸 is the Young modulus, and Γ𝑢 ,
chosen for approximation of stresses and displacements must Γ𝑡 are the displacement and traction boundaries, respectively.
satisfy the Ladyzhenskaya-Babuška-Brezzi (LBB or inf-sup) 𝑢, V, 𝑡𝑥 , and 𝑡𝑦 prescribed respectively the displacements and
condition [28, 29]. The stability of mixed discretization does tractions in the 𝑥 and 𝑦 directions and 𝑛𝑥 , 𝑛𝑦 are direction
not allow FEM to choose independently the approximation cosines of the normal vector to the boundary.
spaces, so these spaces are restricted in the stability condition By using the following definition of stresses in terms of
which is known as the LBB condition. However, the least the displacement components:
squares approximation has the advantage that it does not 𝜕𝑢 𝜕V
require satisfying the LBB condition [30–32]. Hence, this 𝜎𝑥 = (𝜆 + 2𝜇) +𝜆 ,
𝜕𝑥 𝜕𝑦
advantage was used by Amani et al. [33] to implement a mixed
meshless method named mixed discrete least squares Mesh- 𝜕𝑢 𝜕V
𝜎𝑦 = 𝜆 + (𝜆 + 2𝜇) , (4)
less (MDLSM) method which is formulated based on the least 𝜕𝑥 𝜕𝑦
squares residuals functional of the governing partial differen-
tial equations of planar elasticity problem and its boundary 𝜕𝑢 𝜕V
𝜏𝑥𝑦 = 𝜇 ( + ),
conditions at the nodal points, and hence, it is stable and is not 𝜕𝑦 𝜕𝑥
required to satisfy the LBB condition between the displace- we can rewrite (1) in term of stresses as
ments and stresses approximations. Hence, the approxima-
tion spaces of the displacements and stresses can be choosen 𝜕𝜎𝑥 𝜕𝜏𝑥𝑦
+ = −𝑓𝑥 in Ω,
independently while they are obtained simuletanously. 𝜕𝑥 𝜕𝑦
In this paper, the MDLSM method is extended for the (5)
residual based error estimation and for the two types of 𝜕𝜏𝑥𝑦 𝜕𝜎𝑦
+ = −𝑓𝑦 in Ω.
adaptive refinement procedures. The node moving adaptive 𝜕𝑥 𝜕𝑦
refinement procedure based on the spring analogy [24] and
The compact form of (1) can be written by substituting
the node enrichment adaptive refinement procedure [25] are
(4)-(5) into the second-order problem of (1) in the form of
formulated and used in the MDLSM method for efficient
analysis of the elasticity problems. L (𝜙) + f = 0, (6)
Mathematical Problems in Engineering 3

where L(⋅) is a first-order differential operator defined as where RΩ and RΓ are domain and boundary residuals,
respectively.
L (⋅) = L1 (⋅)𝑥 + L2 (⋅)𝑦 + L3 (⋅) , (7) Now the penalty approach is used to form the least square
residuals functional which is defined as
and 𝜙 is the vector of unknowns defined as
𝑀𝑑 𝑀𝑏

𝜙 = [𝑢 V 𝜎𝑥 𝜎𝑦 𝜏𝑥𝑦 ] ,
𝑇
(8) I = ∑ R𝑇Ω RΩ + 𝛼 ∑ R𝑇Γ RΓ , 𝑀 = 𝑀𝑑 + 𝑀𝑏 , (15)
𝑘=1 𝑘=1

and vector f contains the forcing terms which has the form
where 𝑀 is the total number of sampling (or collocation)
𝑇 points, 𝑀𝑑 is total number of domain sampling points, 𝑀𝑏 is
f = [0 0 0 −𝑓𝑥 −𝑓𝑦 ] . (9)
total number of boundary sampling points, and the penalty
In (7), L1 , L2 , and L3 are defined by the following matrices: coefficient 𝛼 is a positive scalar constant that must be large
enough in order to impose the essential boundary condition
𝜆 + 2𝜇 0 0 0 0 with the desired accuracy. A note should be made here
𝜆 0 0 0 0 regarding the value of the penalty parameters. To impose
L1 = ( 0 𝜇 0 0 0) , the boundary conditions exactly, the penalty factor must
0 0 1 0 0 be infinite, which is not possible in practical numerical
0 0 0 0 1 analysis. Therefore, the boundary conditions could not be
satisfied exactly but only approximately. In general, the use
0 𝜆 0 0 0 of a larger penalty factor will lead to better enforcement of
0 𝜆 + 2𝜇 0 0 0 the constraint. The proper value of the penalty parameter is
L2 = (𝜇 0 0 0 0) , determined prior to the main calculation via a trial and error
0 0 0 0 1 process and it is problem dependent.
0 0 0 1 0 Minimizing the functional in (15) with respect to the
nodal unknown vector 𝜙 leads to the following system of
0 0 −1 0 0
equation:
0 0 0 −1 0
L3 = (0 0 0 0 −1) .
K Φ = F, (16)
0 0 0 0 0
0 0 0 0 0
where
(10)
𝑀𝑑 𝑀𝑏
𝑇 𝑇
The displacement and traction boundary conditions (2) can K𝑖𝑗 = ∑ [L (N𝑖 )]ℓ [L (N𝑗 )]ℓ + 𝛼 ∑ [D (N𝑖 )]ℓ [D (N𝑗 )]ℓ ,
be written in terms of the unknown vector 𝜙 as ℓ=1 ℓ=1

𝑀𝑑 𝑀𝑏
D𝜙 − f = 0, (11) 𝑇 𝑇
F𝑖 = ∑ [L (N𝑖 )]ℓ fℓ + 𝛼 ∑ [D (N𝑖 )]ℓ f ℓ ,
ℓ=1 ℓ=1
where D and f are defined as follow: (17)
1 0 0 0 0 𝑢
and Φ is unknown matrix that contains displacements and
0 1 0 0 0 V
D = (0 0 𝑛𝑥 0 𝑛𝑦 ) , f = ( ). (12) stresses of all nodes. F is the right hand side vector and the
𝑡𝑥
stiffness matrix K in (16) is 𝑛DOF × 𝑛DOF square matrix where
0 0 0 𝑛𝑦 𝑛𝑥 𝑡𝑦
𝑛DOF is the number of unknowns per each node and N is the
moving least squares (MLS) shape functions. The proposed
The plane elasticity problem is now defined as solving
MDLSM method has 2.5 times unknowns compared to
the first-order differential equation subjected only to the
the irreducible DLSM method with only displacements as
Dirichlet type boundary condition
unknowns. This should drastically reduce the computational
L1 (𝜙)𝑥 + L2 (𝜙)𝑦 + L3 (𝜙) + f = 0, in Ω, efficiency by (2.5)𝛽 (where 𝛽 depends on the type of linear
(13) solver used) times. But, since matrix K is symmetric and
D𝜙 − f = 0, on Γ. positive definite, therefore, the final system of equation can be
solved directly via efficient solvers. The MDLSM formulation
The application of the proposed MDLSM method for for the plane elasticity problem has the following advantages
solving problem of (13) starts with the definition of residuals that increase its efficiency [33].
as follows:
(1) The order of shape function derivatives is reduced
RΩ = L (𝜙) + f in Ω, by one order, thus complex and costly second-order
(14) derivative calculations of the MLS shape function in
RΓ = D𝜙 − f on Γ, the irreducible DLSM method are avoided.
4 Mathematical Problems in Engineering

(2) The stresses are obtained directly, while calculation In this paper, the relative least square residuals functional
of stresses in the irreducible DLSM method requires for each node is defined as follows:
some postprocessing.
I
(3) Only a linear complete polynomial basis is needed 𝑒=√ , (18)
to construct the MLS shape functions, while in the (U𝑇𝑡 U𝑡 )
irreducible DLSM formulation, the second order
polynomial basis is required to achieve second-order where I is the least square residuals functional in (15) and U𝑡
consistency due to the presence of second-order is unknowns obtained from the main solution. It is noticed
derivatives in the irreducible formulation. This leads that most of the computations of the least square residuals
to lower computational effort for mixed method in functional can be obtained from the main solution of the
construction of MLS shape functions and its deriva- MDLSM method.
tives compared to irreducible one. Furthermore, this
in turn leads to the smaller number of points to be 3.2. Adaptive Refinement Procedures
included in the support domain for the construction
of the MLS shape functions in the mixed method and 3.2.1. Node Moving. Mesh movement strategy can be easily
hence, less computational effort. and efficiently used with meshless methods since no ele-
ment distortion is associated with the method. It should be
(4) Both the displacement and stress boundary condi- noted that the mesh movement technique can be used in
tions are of the Dirichlet type which requires the conjunction with the MDLSM method to adaptively adjust
specification of a single penalty parameter if a penalty nodal points to improve the quality of the solution obtained
method is used to enforce them as used in this work. with a prespecified number of nodal points. Here, a nodal
In the irreducible DLSM method both Dirichlet and refinement procedure is used that is called node moving
Neumann type boundary conditions are required for adaptive refinement approach. When a node refinement is
the determination of two types of penalty coefficient. required, springs of prescribed stiffness are placed between
(5) The mixed formulation, when used with the standard each pair of nodes belonging to the same subdomain and
weighted residual methods both mesh-based and the nodes are then moved until the spring system is in
meshless forms, requires the LBB condition because equilibrium.
the resulting problem is a saddle point problem. The In the node moving procedure, first, all nodes are con-
least squares method, however, is a minimization nected with springs in which the neighbor nodes are defined
method and therefore is not subject to the LBB using Voronoi diagram [36] (see Figure 1). Voronoi diagram
condition. is defined as

T𝑖 = {𝑥 ∈ R : 𝑑 (𝑥, 𝑥𝑖 ) < 𝑑 (𝑥, 𝑥𝑗 ) , 𝑗 ≠

𝑖} , (19)
3. Error Estimator and Adaptive Refinement
where 𝑑(𝑥, 𝑥𝑖 ) is Euclidean distance between 𝑥 and 𝑥𝑖 . The
3.1. Error Estimator. In numerical methods, a problem is above equation means that neighbor nodes to node 𝑖 are the
solved by discretization of the problem domain into the closest nodes to the node 𝑖 rather than other nodes.
subdomains, hence, the governing equations only apply into Spring forces are defined as
these subdomains, so numerical methods always come with
discretization error. Discretization error is one of the most 𝑏𝑖𝑗 = 𝑐𝑖𝑗 (𝑥𝑖 − 𝑥𝑗 ) , (20)
important challenges in the numerical methods. Discretiza-
tion error is theoretically decreased by refining the discretiza- where 𝑐𝑖𝑗 is stiffness of spring between 𝑖, 𝑗 and 𝑥𝑖 , 𝑥𝑗 are
tion domain but perfunctory refinement imposes the heavy coordinates of 𝑖, 𝑗 in equilibrium, respectively. The free body
computational cost without supplying the expected accuracy. diagram is shown in Figure 2.
Adaptive refinement methods mean balances between refine- Spring stiffness is defined as a function of errors between
ment procedure and its computational cost. These methods two points 𝑖, 𝑗 as follows:
only refine locally the regions of the domain which has
higher error. Adaptive procedure has two main parts: error (𝑒𝑖 + 𝑒𝑗 )
estimation and adaptive refinement. Any success adaptive 𝑐𝑖𝑗 = , (21)
refinement needs a reliable error estimation procedure. Real 𝑑𝑖𝑗
error distribution can not be practicably used because the
exact solution is not available for any practical problems. where 𝑒𝑖 and 𝑒𝑗 are the values of the error estimators obtained
Several methods are used for error estimation with different from (18) at nodes 𝑖 and 𝑗, respectively, and 𝑑𝑖𝑗 is distance
numerical methods and these methods are categorized into between these two connected nodes. In matrix form we have
two classes, namely, the residual based method [34] and
recovery based method [35]. In residual based method, the 𝑐𝑖𝑗 0 −𝑐𝑖𝑗 0 𝑥𝑖 𝑏𝑥𝑖
0 𝑐𝑖𝑗 0 −𝑐𝑖𝑗 𝑦 𝑏𝑖
residuals of differential equation and its boundaries are used ( ) (𝑥 𝑖 ) = ( 𝑦𝑗 ) , (22)
as a criterion of error. The gradient of the solutions is used in −𝑐𝑖𝑗 0 𝑐𝑖𝑗 0 𝑗 𝑏𝑥
recovery based method as the error criterion. 0 −𝑐𝑖𝑗 0 𝑐𝑖𝑗 𝑦𝑗 𝑏𝑦𝑗
Mathematical Problems in Engineering 5

Figure 1: Voronoi diagram of neighboring nodes and spring connections.

j
fy
j
fx
j

cij

fyi

fxi
i

Figure 2: Free body diagram between two nodes 𝑖 and 𝑗.

where 𝑏𝑥𝑖 and 𝑏𝑦𝑗 are the components of the force exerted It is obvious that the equation system defined in (24) is
at node 𝑖 in 𝑥 and 𝑦 directions, respectively, and 𝑥𝑖 and singular before any boundary conditions are considered. The
𝑦𝑖 are the coordinates of node 𝑖 in 𝑥 and 𝑦 directions, boundary conditions used here for solving this system of
respectively. The spring systems work as a two-dimensional equations are defined by the requirement that the boundary
truss such that nodes and springs are points and elements of nodes should not be allowed to move perpendicularly to
truss, respectively. We mention that the system of algebraic the boundaries. In other words, boundary nodes only can
equations (22) can be assembled in its standard finite element be displaced along the boundaries which they have been
concept to yield global force vector for whole system of placed on. Mathematical representation of these boundary
springs as follows: conditions can be defined as
𝑛𝑖
CX = B, (23) ΔX𝑖𝑇 n𝑖 = (Δ𝑥𝑖 Δ𝑦𝑖 ) ( 𝑖𝑥 )
𝑛𝑦
(25)
where C is the stiffness matrix of the system calculated by 𝑛𝑖
= ((𝑥𝑖 − 𝑥ℓ ) (𝑦𝑖 − 𝑦ℓ )) ( 𝑖𝑥 ) = 0,
assembling the stiffness matrices of all the springs defined 𝑛𝑦
in the system and B represents the vector of nodal forces. In
the equilibrium condition, the vector B of assembled spring where 𝑥ℓ and 𝑦ℓ are the initial coordinates of boundary node
forces should be equal to zero. This requirement leads to 𝑖; 𝑥𝑖 and 𝑦𝑖 represent the displaced final position of node 𝑖
the following system of algebraic equation which should be and n is outward unit vector normal to the boundary at node
solved for the unknown vector of nodal position X, that is, 𝑖. This condition guarantees that the nodes initially located
solve at the intersection of two boundary lines must remain on
its initial position. Boundary conditions for a simple net of
CX = 0. (24) springs are shown in Figure 3.
6 Mathematical Problems in Engineering

y1 = y1 y4 = y4 y =y
1 { x7 = x7
{
x1 = x 1 4 7 7 7

x2 = x 2 2
x8 = x 8
5 8

y =y 3 6 9 y9 = y9
{ x3 = x3 {
x9 = x 9
3 3 y6 = y6

Figure 3: Boundary conditions of spring system.

Figure 4: The node enrichment using Voronoi diagram (∙ initial nodes, 󳵳 new added nodes).

Continuous PDE of elasticity

Initial node generator

Discrete problem

MDLSM code New nodal distribution

Approximate solution
Inserting new nodes
Error indicator

Total error
If not acceptable
If acceptable

Desired approximation

Figure 5: The node enrichment procedure in MDLSM method.

Mathematical Problems in Engineering 7

r 5a

P a 𝜃 x P

Figure 6: An infinite plate with a circular hole under a uniaxial load 𝑃 (Example 1).

(a) 122 initial nodes (b) 122 refined nodes

(c) 229 initial nodes (d) 229 refined nodes

(e) 305 initial nodes (f) 305 refined nodes

Figure 7: Initial and refined (node moving) nodal configurations (Example 1).

Upon solving the system of equations with appropri- methods can be thought to be defining the location of the
ate boundary conditions, the refined position of nodes is nodes to be added to the current nodal configuration. Here
obtained leading to substantial reduction of the local and the new nodes are added in the neighborhood of existing
global error of the numerical solution in the subsequent anal- nodal points defined by a Voronoi diagram.
ysis. The efficiency and effectiveness of the proposed adaptive Once the Voronoi cells are defined, the vertices of Voronoi
refinement technique is verified in the next section by its cells corresponding to the nodes with higher error than the
application to benchmark test examples in plane elasticity. average error over the domain are considered as the new
nodes to be added to the current nodal configuration. The
method is schematically illustrated in Figure 4. The value of
3.2.2. Node Enrichment. With the meshless methods, the average error 𝑒avg over the domain is obtained by
enrichment strategy only requires that the locations of new
nodes to be added are determined without requiring to define ∑𝑀
ℓ=1 𝑒ℓ (26)
𝑒avg = .
the connectivity of the resulting configuration. Different 𝑀
8 Mathematical Problems in Engineering

3.5 0.7 0

3 0.6
Normal stress in x direction

−1
0.5

Ln(error norm)
2.5

Error norm
0.4 −2
2
0.3
1.5 −3
0.2
1 0.1
−4
0.5 0
1 2 3 4 5 100 150 200 250 300 2.3 2.4 2.5 2.6 2.7 2.8 2.9
y Number of nodes (N) Ln(sqrt(N))
Exact analytical solution Initial with uniform configuration CR (initial) = 2.42
Initial uniform configuration (305 nodes) Refined configuration with node moving CR (refined with node moving) = 4.45
Refined configuration with node moving
(305 nodes)
(a) (b) (c)

Figure 8: (a) Normal stress 𝜎𝑥 at 𝑥 = 0. (b) Convergence curve of node moving procedure. (c) Convergence rate of node moving procedure.

(a) Initial nodes = 122 (b) First step enrichment nodes

= 182

(c) Second step enrichment (d) Third step enrichment nodes

nodes = 288 = 521

Figure 9: Initial and refined (node enrichment) nodal configurations (Example 1).

The above node enrichment adaptive refinement procedure irreducible DLSM method and the exact analytical solutions
based on error estimation is schematically shown in Figure or finite element results with very fine mesh.
5.
Example 1 (an infinite plate with a circular hole). In the first
example, consider the case of an infinite plate with a circular
4. Numerical Experiments hole subjected to a uniaxial traction 𝑃 at infinity, as shown
in Figure 6. Due to symmetry, only the upper right square
In this section, we are solving the benchmark examples quadrant of the plate is modeled. The edge length of the
by using the proposed node moving and node enrichment square is 5𝑎, where 𝑎 is the radius of the circular hole. This
refinement procedures for the MDLSM method and com- example is chosen because the exact analytical solution is
paring the results with the refinement procedure in the available from Timoshenko and Goodier [37]. The solutions
Mathematical Problems in Engineering 9

3.5 0.7 0

0.6
Normal stress in x direction

3
−1
0.5

Ln(error norm)
2.5

Error norm
0.4 −2
2
0.3
1.5 −3
0.2
1 0.1 −4
0.5 0
1 2 3 4 5 100 200 300 400 500 2.3 2.4 2.5 2.6 2.7 2.8 2.9 3 3.1 3.2
y Number of nodes (N) ln(sqrt(N))

Exact analytical solution CR (refined with node enrichment) = 4.76

Initial uniform configuration
Refined configuration with node enrichment
(last step)
(a) (b) (c)

Figure 10: (a) Normal stress 𝜎𝑥𝑥 at 𝑥 = 0. (b) Convergence curve for node enrichment procedure. (c) Convergence rate for node enrichment
procedure.

Table 1: Comparison of the error norms for initial and refined (using node moving) nodal configurations (Example 1).

Number of nodes Norm of error for initial nodes Norm of error for refined nodes
122 0.6684 0.1528
229 0.3550 0.0331
305 0.2123 0.0206

for the displacements and the stresses under a unit uniaxial Table 2: Error norms in initial nodal configuration and in different
stress along the 𝑥 axis are given as follows: steps of multistage node enrichment procedure (Example 1).

Number of nodes Norm of the error

122 0.6684
𝑎2 3 3𝑎4
𝜎𝑥 = 𝑡 (1 − ( cos (2𝜃) + cos (4𝜃)) + cos (4𝜃)) , 182 0.1457
𝑟2 2 2𝑟4
288 0.0571
𝑎2 1 3𝑎4
𝜎𝑦 = − ( ( cos (2𝜃) − cos (4𝜃)) + cos (4𝜃)) , 521 0.0190
𝑟2 2 2𝑟4
𝑎2 1 3𝑎4
𝜏𝑥𝑦 = −𝑡 ( 2
( sin (2𝜃) + sin (4𝜃)) − 4 sin (4𝜃)) ,
𝑟 2 2𝑟 proposed error estimator based on the least square residuals
functional. In Figure 8(a) numerical results of normal stress
𝑡 𝜅−1 𝑎2 𝜎𝑥 on the left edge is compared with the exact analytical
𝑢𝑟 = (𝑟 ( + cos (2𝜃)) + (1 + (1 + 𝜅) cos (2𝜃))
4𝐺 2 𝑟 solution. Figures 8(b) and 8(c) compare the convergence
𝑎4 curve and the convergence rate of the MDLSM method for
− cos (2𝜃)) , the node moving adaptive refinement strategy and plotted
𝑟3 using the error norms in Table 1.
𝑡 𝑎2 𝑎4 Also, the node enrichment adaptive refinement strategy is
𝑢𝜃 = ((1 − 𝜅) − 𝑟 − 3 ) sin (2𝜃) , applied to the MDLSM method for obtaining accurate results.
4𝐺 𝑟 𝑟
(27) As shown in Figure 9, first the problem domain is discretized
by using 122 initial nodal distributions, and then the nodal
which 𝐺 is the shear modulus and 𝜅 = (3 − ])/(1 + ]) where ] points are enriched in three steps (182, 288, and 521 nodes)
is the Poissons ratio. In this example, the constant values are on the region determined by the error indicator. Figure 10(a)
𝑎 = 1, 𝑡 = 1, and 𝐸 = 1000 and ] = 0.3. compares the numerical results of normal stress 𝜎𝑥 on the left
For the node moving procedure, initial and refined nodal edge with exact analytical solution. The error norms in Table
configurations with 122, 229, and 305 nodes are shown in 2 are used for plotting and comparison of the convergence
Figure 7. The nodal points are refined with respect to the curve and the convergence rate of the node enrichment
10 Mathematical Problems in Engineering

a
b

Figure 11: A cylinder subjected to an internal pressure and its boundary conditions.

(a) 106 initial nodes (b) 106 refined nodes

(c) 214 initial nodes (d) 214 refined nodes

(e) 359 initial nodes (f) 359 refined nodes

(g) 543 initial nodes (h) 543 refined nodes

Figure 12: Initial and refined (node moving) nodal configurations (Example 2).
Mathematical Problems in Engineering 11

0.4
−1
Normal stress in x direction

1 0.3
−2

ln(error norm)
Error norm
0.2
−3
0.5
0.1 −4

0 0 −5
1 2 3 4 5 100 200 300 400 500 600 2.3 2.4 2.5 2.6 2.7 2.8 2.9 3 3.1 3.2
y Number of nodes (N) ln(sqrt(N))

Exact analytical solution Initial with uniform configuration CR (initial) = 2.87

Initial uniform configuration (543 nodes) Refined configuration with node moving CR (refined with node moving) = 3.46
Refined configuration with node moving
(543 nodes)
(a) (b) (c)

Figure 13: (a) Normal stress 𝜎𝑥 at 𝑥 = 0. (b) Convergence curve of node moving procedure. (c) Convergence rate of node moving procedure.

Table 3: Comparison of the error norms for initial and refined (using node moving) nodal configurations (Example 2).

Number of nodes Norm of error for initial nodes Norm of error for refined nodes
106 0.3843 0.1567
214 0.1247 0.0388
359 0.0691 0.0153
543 0.0354 0.0098

procedure for the MDLSM method (see Figures 12(b) and Table 4: Error norms in initial nodal configuration and in different
12(c)). The results clearly show that the node moving and steps of multistage node enrichment procedure (Example 2).
multistage node enrichment adaptive refinement strategies
Number of nodes Norm of the error
in the MDLSM method are more efficient compared to
the refinement procedures in the DLSM method. Figures 106 0.3843
8(b) and 8(c), 10(b), and 10(c) indicate that by using the 164 0.0808
node moving and the node enrichment adaptive refinement 276 0.0501
procedures, the convergence rate of the MDLSM method is 491 0.0191
increased.

Example 2 (a cylinder subjected to an internal pressure). As a Figure 13(a) compares the normal stress 𝜎𝑥 at 𝑥 = 0
second elastostatic benchmark example a cylinder subjected for initial and refined nodal configurations with 543 nodes.
to an internal pressure is considered. Due to the symmetry, It is clear that the result of refined nodal configuration is
only a quarter of the cylinder is modeled; see Figure 11. The more similar to the exact analytical solution than initial
boundary conditions are illustrated in Figure 11. The exact nodal configuration. Table 3 compares the error norms of
analytical solution of this problem is the node moving procedure for the initial and refined nodal
configurations and Figures 13(b) and 13(c) compare the
convergence curve and the convergence rate, respectively.
𝑎2 𝑃 𝑏2 In Figure 14, the initial and refined nodal distributions
𝜎𝑟 = (1 − ),
𝑏2 − 𝑎2 𝑟2 for the node enrichment strategy are shown. Figure 15(a)
(28) compares the normal stress 𝜎𝑥 at 𝑥 = 0 for initial and last
𝑎2 𝑃 𝑏2 steps refined with 543 nodal distributions. In Table 4 the
𝜎𝜃 = 2 (1 + ),
𝑏 − 𝑎2 𝑟2 error norm of the node enrichment refinement procedure
is shown. Figures 15(b) and 15(c) compare the convergence
curve and the convergence rate of the node enrichment
where the constant values are 𝑎 = 1, 𝑏 = 5, 𝑃 = 1, ] = 0.3, adaptive refinement strategy.
and 𝐸 = 107 . As shown in Figure 12, four types of nodal
distributions with 106, 214, 359, and 543 nodes are distributed Example 3 (a reservoir fully filled with water). In this example,
to solve and refine the nodes in the problem domain. consider that the wall of a reservoir fully filled with water is
12 Mathematical Problems in Engineering

(a) Initial nodes = 106 (b) First step enrichment nodes

= 164

(c) Second step enrichment (d) Third step enrichment nodes

nodes = 276 = 491

Figure 14: Initial and refined (node enrichment) nodal configurations (Example 2).

0.4
−1
Normal stress in x direction

1 0.3 ln(error norm)

−2
Error norm

0.2
0.5 −3
0.1
−4
0 0
1 2 3 4 5 100 200 300 400 500 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 3 3.1 3.2
y Number of nodes (N) ln(sqrt(N))
Exact analytical solution CR (refined with node enrichment) = 3.66
Initial uniform configuration
Refined configuration with node
enrichment (last step)
(a) (b) (c)

Figure 15: (a) Normal stress 𝜎𝑥 at 𝑥 = 0. (b) Convergence curve for node enrichment procedure. (c) Convergence rate for node enrichment
procedure.

investigated. The geometry of the wall is irregular as given and refine the points in the problem domain. Figure 18(a)
in Figure 16. The material properties of the wall are given as compares the displacement in 𝑦-direction along the vertical
Young’s modulus 𝐸 = 107 and Poissons ratio ] = 0.3. The edge for initial and refined configurations. Table 5 compares
bottom of the wall is fixed and the curved edge of the wall is the error norms of the node moving procedures based
subjected to a hydrostatic pressure 𝑃 = −9800(𝐻 − 𝑦) MPa. on MDLSM method and Figures 18(a) and 18(b) compare
Since the analytical solution of this problem is not available, a the convergence curve and convergence rate, respectively.
very fine mesh (with 59,400 linear triangular elements) FEM For node enrichment adaptive refinement procedure, the
solution will be considered as our reference solution. problem domain is discretized with initial 84 points and is
refined in three steps near high gradient error norm and is
As shown in Figure 17, four types of nodal distribution solved with 120, 181, and 284 nodes, respectively (see Figure
with 84, 138, 218, and 299 nodes are distributed to solve 19). Figure 20(a) compares the displacement in 𝑦-direction
Mathematical Problems in Engineering 13

10

[(x − 50)/40]2 + [(y − 20/30)] 2 = 1

20
Water
Dam

Figure 16: A reservoir fully filled with water.

(a) 84 initail nodes (b) 84 refined nodes

(c) 138 initial nodes (d) 138 refined nodes

(e) 218 initial nodes (f) 218 refined nodes

(g) 299 initial nodes (h) 299 refined nodes

Figure 17: Initial and refined (node moving) nodal configurations (Example 3).
14 Mathematical Problems in Engineering

0
0.02
−0.02
−4
−0.04

ln(error norm)
Displacement

0.015

Error norm
−0.06
−0.08 0.01 −5
−0.1
−0.12 0.005

−0.14 −6
0 5 10 15 20 50 100 150 200 250 300 350 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9
y Number of nodes (N) ln(sqrt(N))
Reference solution CR (initial) = 1.60
Initial with uniform configuration
Initial uniform configuration (299 nodes) CR (refined with node moving) = 2.29
Refined configuration with node Refined configuration with node moving
moving (299 nodes)
(a) (b) (c)

Figure 18: (a) Displacement in 𝑦-direction along the vertical edge. (b) Convergence curve of node moving procedure. (c) Convergence rate
of node moving procedure.

(a) Initial nodes = 84 (b) First step enrichment nodes

= 120

(c) Second step enrichment (d) Third step enrichment nodes

nodes = 181 = 284

Figure 19: Initial and refined (node enrichment) nodal configurations (Example 3).

along the vertical edge for initial and refined configurations. on least square residuals functional was formulated and used.
In Table 6 the error norms of the node enrichment refinement Voronoi diagram was extended in the refinement procedures
procedure is used to plot the convergence curve and conver- to find the neighbor nodes (node moving) and the position
gence rate in Figures 20(b) and 20(c). of the new nodes (node enrichment). For the moving node
procedure, spring analogy was used to construct a system for
computing the new place of each node after the refinement
5. Conclusion procedure. The efficiency and effectiveness of the proposed
node moving and node enrichment adaptive refinement
A mixed discrete least squares meshless method was techniques in the MDLSM method by their application to
extended for node moving and node enrichment adaptive the benchmark examples in the elasticity problems were
refinements for efficient analysis of the planar elasticity prob- verified. Results show that the proposed refinement methods
lem. For the refinement procedures an error estimator based are accurate and straightforward.
Mathematical Problems in Engineering 15

0 −3.5
0.02
−0.02
−4
−0.04
Displacement

ln(error norm)
0.015 −4.5

Error norm
−0.06
−0.08 −5
−0.1 0.01
−5.5
−0.12
−0.14 0.005 −6
0 5 10 15 20 50 100 150 200 250 300 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9
y Number of nodes (N) ln(sqrt(N))
Reference solution CR (refined with node enrichment) = 2.00
Initial uniform configuration
Refined configuration with node
enrichment (last step)
(a) (b) (c)

Figure 20: (a) Displacement in 𝑦-direction along the vertical edge. (b) Convergence curve for node enrichment procedure. (c) Convergence
rate for node enrichment procedure.

Table 5: Comparison of the error norms for initial and refined (using node moving) nodal configurations (Example 3).

Number of nodes Norm of error for initial nodes Norm of error for refined nodes
84 0.0195 0.0190
138 0.0135 0.0116
218 0.0091 0.0059
299 0.0071 0.0047

Table 6: Error norms in initial nodal configuration and in different [5] R. A. Gingold and J. J. Monaghan, “Smoothed particle hydro-
steps of multistage node enrichment procedure (Example 3). dynamics: theory and applications to non-spherical stars,”
Monthly Notices of the Royal Astronomical Society, vol. 181, pp.
Number of nodes Norm of the error 375–389, 1977.
84 0.0195 [6] B. Nayroles, G. Touzot, and P. Villon, “Generalizing the finite
120 0.0147 element method: diffuse approximation and diffuse elements,”
181 0.0087 Computational Mechanics, vol. 10, no. 5, pp. 307–318, 1992.
284 0.0060 [7] T. Belytschko, Y. Y. Lu, and L. Gu, “Element-free Galerkin
methods,” International Journal for Numerical Methods in Engi-
neering, vol. 37, no. 2, pp. 229–256, 1994.
[8] W. K. Liu, S. Jun, and Y. F. Zhang, “Reproducing kernel
Conflict of Interests particle methods,” International Journal for Numerical Methods
The authors declare that there is no conflict of interest in Fluids, vol. 20, no. 8-9, pp. 1081–1106, 1995.
regarding the publication of this paper. [9] E. Oñate, S. Idelsohn, O. C. Zienkiewicz, and R. L. Taylor, “A
finite point method in computational mechanics. Applications
to convective transport and fluid flow,” International Journal for
References Numerical Methods in Engineering, vol. 39, no. 22, pp. 3839–
3866, 1996.
[1] L. F. Richardson, “The approximate arithmetical solution by [10] S. N. Atluri and T. Zhu, “A new Meshless Local Petrov-Galerkin
finite differences of physical problems,” Transactions of the Royal (MLPG) approach in computational mechanics,” Computa-
Society A, vol. 210, pp. 307–357, 1910. tional Mechanics, vol. 22, no. 2, pp. 117–127, 1998.
[2] O. C. Zienkiewicz, “The background of error estimation and [11] S. N. Atluri and T. Zhu, “A new meshless local Petrov-Galerkin
adaptivity in finite element computations,” Computer Methods (MLPG) approach in computational mechanics,” Computa-
in Applied Mechanics and Engineering, vol. 195, no. 4–6, pp. 207– tional Mechanics, vol. 22, no. 2, pp. 117–127, 1998.
213, 2006.
[12] T. Zhu, J.-D. Zhang, and S. N. Atluri, “A local boundary integral
[3] O. C. Zienkiewicz, The Finite Element Method, McGraw Hill, equation (LBIE) method in computational mechanics, and a
London, UK, 1977. meshless discretization approach,” Computational Mechanics,
[4] I. Babuška and A. K. Aziz, “On the angle condition in the finite vol. 21, no. 3, pp. 223–235, 1998.
element method,” SIAM Journal on Numerical Analysis, vol. 13, [13] N. R. Aluru and L. Gang, “Finite cloud method: a true meshless
no. 2, pp. 214–226, 1976. technique based on a fixed reproducing kernel approximation,”
16 Mathematical Problems in Engineering

Journal for Numerical Methods in Engineering, vol. 50, pp. 2373– [29] K.-J. Bathe, “The inf-sup condition and its evaluation for mixed
2410, 2001. finite element methods,” Computers & Structures, vol. 79, no. 2,
[14] H. Arzani and M. H. Afshar, “Solving Poisson’s equations by the pp. 243–252, 2001.
discrete least square meshless method,” in Boundary Elements [30] F. Brezzi, “On the existence, uniqueness and approximation
and Other Mesh Reduction Methods XXVIII, vol. 42 of WIT of saddle-point problems arising from Lagrangian multipli-
Transactions on Modelling and Simulation, pp. 23–32, WIT ers,” Revue Française d’utomatique, Informatique, Recherche
Press, Southampton, UK, 2006. Opérationnelle. Analyse Numérique, vol. 8, no. R-2, pp. 129–151,
[15] T. Rabczuk and T. Belytschko, “An adaptive continuum/discrete 1974.
crack approach for mesh-free particle methods,” Latin Ameri- [31] Z. Cai, R. Lazarov, T. A. Manteuffel, and S. F. McCormick, “First-
can Journal of Solids and Structures, vol. 1, pp. 141–166, 2003. order system least squares for second-order partial differential
[16] T. Rabczuk and T. Belytschko, “Adaptivity for structured mesh- equations. I,” SIAM Journal on Numerical Analysis, vol. 31, no.
6, pp. 1785–1799, 1994.
free particle methods in 2D and 3D,” International Journal for
Numerical Methods in Engineering, vol. 63, no. 11, pp. 1559–1582, [32] X. Ye, “Domain decomposition for a least-square finite element
2005. method for second order elliptic problem,” Applied Mathematics
and Computation, vol. 91, no. 2-3, pp. 233–242, 1998.
[17] Y.-C. Yoon, S.-H. Lee, and T. Belytschko, “Enriched meshfree
collocation method with diffuse derivatives for elastic fracture,” [33] J. Amani, M. H. Afshar, and M. Naisipour, “Mixed discrete least
Computers & Mathematics with Applications, vol. 51, no. 8, pp. squares meshless method for planar elasticity problems using
1349–1366, 2006. regular and irregular nodal distributions,” Engineering Analysis
with Boundary Elements, vol. 36, no. 5, pp. 894–902, 2012.
[18] G. Zi, T. Rabczuk, and W. Wall, “Extended meshfree methods
without branch enrichment for cohesive cracks,” Computational [34] I. Babuska and W. C. Rheinboldt, “A posteriori error estimates
Mechanics, vol. 40, pp. 367–382, 2007. for the finite element method,” International Journal for Numer-
ical Methods in Engineering, vol. 12, pp. 1597–1615, 1978.
[19] S. Bordas, G. Zi, and T. Rabczuk, “Three-dimensional non-
[35] O. C. Zienkiewicz, B. Boroomand, and J. Z. Zhu, “Recovery
linear fracture mechanics by en-riched meshfree methods
procedures in error estimation and adaptivity. I. Adaptivity in
without asymptotic enrichment,” Solid Mechanics and Its Appli-
linear problems,” Computer Methods in Applied Mechanics and
cations, vol. 5, pp. 21–36, 2007.
Engineering, vol. 176, no. 1–4, pp. 111–125, 1999.
[20] T. Rabczuk and E. Samaniego, “Discontinuous modelling of
[36] Q. Du and D. Wang, “Tetrahedral mesh generation and opti-
shear bands using adaptive meshfree methods,” Computer
mization based on centroidal Voronoi tessellations,” Interna-
Methods in Applied Mechanics and Engineering, vol. 197, no. 6–8,
tional Journal for Numerical Methods in Engineering, vol. 56, no.
pp. 641–658, 2008.
9, pp. 1355–1373, 2003.
[21] X. Zhuang and C. Augarde, “Aspects of the use of orthogonal [37] S. Timoshenko and J. N. Goodier, Theory of Elasticity, McGraw-
basis functions in the element-free Galerkin method,” Interna- Hill Book, New York, 2nd edition, 1951.
tional Journal for Numerical Methods in Engineering, vol. 81, no.
3, pp. 366–380, 2010.
[22] X. Zhuang, C. Heaney, and C. Augarde, “On error control in
the element-free Galerkin method,” Engineering Analysis with
Boundary Elements, vol. 36, no. 3, pp. 351–360, 2012.
[23] X. Zhuang, H. Zhu, and C. Augarde, “An improved meshless
Shepard and least square method possessing the delta property
and requiring no singular weight function,” Computational
Mechanics, vol. 53, no. 2, pp. 343–357, 2014.
[24] M. H. Afshar, M. Naisipour, and J. Amani, “Node moving
adaptive refinement strategy for planar elasticity problems
using discrete least squares meshless method,” Finite Elements
in Analysis and Design, vol. 47, no. 12, pp. 1315–1325, 2011.
[25] M. H. Afshar, J. Amani, and M. Naisipour, “A node enrichment
adaptive refinement in discrete least squares meshless method
for solution of elasticity problems,” Engineering Analysis with
Boundary Elements, vol. 36, no. 3, pp. 385–393, 2012.
[26] A. R. Firoozjaee and M. H. Afshar, “Error estimate and adaptive
refinement for incompressible Navier-Stokes equations using
the discrete least squares meshless method,” International Jour-
nal for Numerical Methods in Fluids, vol. 70, no. 1, pp. 56–70,
2012.
[27] F. Brezzi and M. Fortin, Mixed and Hybrid Finite Element Meth-
ods, vol. 15 of Springer Series in Computational Mathematics,
Springer, New York, NY, USA, 1991.
[28] F. Brezzi and K.-J. Bathe, “A discourse on the stability conditions
for mixed finite element formulations,” Computer Methods in
Applied Mechanics and Engineering, vol. 82, no. 1–3, pp. 27–57,
1990.
Hindawi Publishing Corporation
Mathematical Problems in Engineering
Volume 2014, Article ID 806142, 11 pages
http://dx.doi.org/10.1155/2014/806142

Research Article
A Meshless Local Petrov-Galerkin Shepard and Least-Squares
Method Based on Duo Nodal Supports

Xiaoying Zhuang and Yongchang Cai

National Key Laboratory of Disaster Reduction and Protection, Department of Geotechnical Engineering, Tongji University,
Shanghai 200092, China

Correspondence should be addressed to Yongchang Cai; [email protected]

Received 28 September 2013; Accepted 14 October 2013; Published 23 January 2014

Academic Editor: Timon Rabczuk

Copyright © 2014 X. Zhuang and Y. Cai. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.

The meshless Shepard and least-squares (MSLS) interpolation is a newly developed partition of unity- (PU-) based method which
removes the difficulties with many other meshless methods such as the lack of the Kronecker delta property. The MSLS interpolation
is efficient to compute and retain compatibility for any basis function used. In this paper, we extend the MSLS interpolation to the
local Petrov-Galerkin weak form and adopt the duo nodal support domain. In the new formulation, there is no need for employing
singular weight functions as is required in the original MSLS and also no need for background mesh for integration. Numerical
examples demonstrate the effectiveness and robustness of the present method.

1. Introduction limit its application, namely, the high computational cost in

obtaining the shape functions and their derivatives, difficulty
Meshless methods have prospered both in theory and appli- in retaining accuracy with respect to nodal arrangement, and
cation in engineering problems in the past two decades as also the difficulty with which essential boundary conditions
they offer the possibility of a discretised approach without can be imposed due to the lack of the Kronecker delta
the occurrence of mesh entanglement requiring remeshing. A property. Efforts have been made to address these problems
wide range of meshless methods have been proposed as out- by various means. Breitkopf et al. [31] developed the analytical
lined in recent surveys [1–3]. Remarkable successes have been forms for computing shape functions and diffuse derivatives
reported in applying these methods for analyzing challenging of shape functions by assuming that some terms are constant
engineering problems, namely, fracture modeling [4–7], plate and complete derivatives of shape functions. However, these
and shell analysis [8–15], three-dimensional problems [16– formulations are dependent on the number of nodes and the
18], fluid structure interaction analysis [19], strain localization formulation grows unwieldy when there are a large number
problems [20], large deformation problems [21], and other of nodes in the support domain. Orthogonal basis functions
applications [22–29]. pertaining to MLS for efficient and accurate computation
Some currently popular meshless approximations are the of shape functions are investigated in [32, 33]. Singular
moving least-squares (MLS) approximation, Shepard shape weight functions are introduced by Kaljević and Saigal [34]
functions, partition of unity (PU), radial basis functions to produce an interpolatory MLS approximation for the
(RBF), reproducing kernel particle method (RKPM), point direct imposition of essential boundary conditions. Chen and
interpolation (PI), and Kriging interpolation (KI). Among Wang [35] developed a matrix transformation method for the
them, the MLS approximation [30] is one of the most widely imposition of boundary conditions; however, the formulation
used approximations at present due to its global continuity, is complicated for implementation. The Shepard and least
completeness, and robustness. However, the MLS approxi- square (MSLS) interpolation developed by the authors [36]
mation suffers from a number of problems that practically satisfactorily keeps the consistency of the Shepard shape
2 Mathematical Problems in Engineering

functions up to the order of basis function and satisfies

the delta property. However, singular weight functions have K
to be used to enforce the interpolating property of shape rlI
functions, which results in the loss of the smoothness of the
interpolation as well as local oscillation. To eliminate this I
x
problem, an improved MSLS interpolation possessing the
delta property without using singular weight functions as PU
was proposed in [37]. lI
Apart from the interpolation, the integration scheme
of the weak form is also an important factor affecting the lJ
solution accuracy, which has been the problem for many
rlJ
meshless methods. Background cells have usually been used
to integrate the weak form as has been the case with the J
EFGM, RKPM, and PIM [38]. Due to the complexity of
the shape functions, a large number of integration points
Γu
are needed to avoid the underestimation of the weak form,
which is computationally expensive but still not adequate Figure 1: The setting of the nodal subdomains in Ω.
to give accurate solutions. Furthermore when an irregular
nodal arrangement is used, the background mesh has to be
refined where nodes are densely distributed [39]. To remove Neighboring of
point x
this difficulty, the nodal integration scheme which is free
from the background cell was proposed in [40] and later
used in [41, 42]. However, the performance of this scheme rcI = rlI
is unstable and also reduces the accuracy of the results. The
MLPG-type meshless methods well solved this problem in a I
natural way by partitioning the local domain into a number x
of subdomains that may or may not overlap. In this way, inte-
gration of high order accuracy can be obtained for the global cI
stiffness matrix [43] without background mesh. A similar
but conceptually different approach was developed in [44]
lJ
using the partition of unity quadrature scheme. However,
the algorithms are complicated and more computationally rcJ
demanding compared to the classical quadrature on subcells.
cJ J
Although many different approaches have been carried out
for the weak form integration, finding a simple, efficient, and
accurate integration scheme for meshless methods remains
Figure 2: Duo support domain of node.
an open question. In this paper, the local weak form along
with numerical integration over local subdomains is used to
derive the discrete equations.
The content of the paper is outlined as follows. In
Section 2, formulation of the improved MSLS interpolation shown in Figure 1. The formulation is described in the context
is described in detail including the local approximation and of elastostatics, with the fundamental field variable being a
support domain with dual definitions. The Kronecker delta displacement. For an arbitrary node 𝐼, its displacement vector
property of the interpolation is also proved. In the Section 3, in 2D is u𝐼 = (𝑢𝐼 , V𝐼 )T , where 𝑢𝐼 and V𝐼 are the nodal
the discretised formulation of the present interpolation is displacements in the 𝑥 and 𝑦 directions, respectively (the
derived using the local Petrov-Galerkin weak form. It is following formulation is derived only for 𝑢𝑖 in the 𝑥 direction
followed by numerical tests demonstrating the convergence but an identical process can be used for V𝑖 in the 𝑦 direction).
characteristic and accuracy of the present interpolation in The interpolation at an arbitrary point x inside the domain in
Section 4. Discussions are given at the end highlighting the the 𝑥 direction is expressed as
features of the present method and suggestions on further
studies. 𝑛
𝑢 (x) = ∑ 𝜑𝐼0 (x) 𝑢𝑙𝐼 (x) , (1)
𝐼=1
2. Formulation of the MSLS Based on
Duo Nodal Supports where {𝜑𝐼0 (x), 𝐼 = 1, . . . , 𝑛} is a set of shape functions that
forms a partition of unity (PU); that is, ∑𝑛𝐼=1 𝜑𝐼0 (x) ≡ 1, 𝐼
In this section, the MSLS based on duo nodal supports is is the node index, and 𝑛 is the number of the nodes whose
described in detail. We start the description of the formu- supports 𝑐𝐼 include point x as shown in Figure 2. 𝑢𝑙𝐼 (x) here
lation using a 2D problem domain of arbitrary shape as is not the nodal displacement in the FEM or the “fictitious”
Mathematical Problems in Engineering 3

nodal values in the EFGM but the local approximation of 2.2. Duo Support Domain of a Node. The support domain of
node 𝐼 at x where the superscript 𝑙 indicates local. Shepard a node is the area where a node exerts influence on the field
shape functions used as PU are given by variable. In this paper it is defined as a circle centered on that
node although it may take other shapes such as a rectangle.
𝑤𝐼 (x)
𝜑𝐼0 (x) = , (2) Here, two support domains are defined at each node, one is
∑𝑛𝐽=1 𝑤𝐽 (x) used in the construction of local approximation and the other
in the PU approximation. In Figure 2, for example, node 𝐼
which is the same as in the original MSLS. The construction
has two support domains associated with it, namely, 𝑙𝐼 with
of the MSLS interpolation takes the following steps: firstly,
radius 𝑟𝑙𝐼 and 𝑐𝐼 with radius 𝑟𝑐𝐼 . If a node, for example, node
construct the local approximation at each node and secondly
𝐾 in Figure 1 falls inside 𝑙𝐼 , then node 𝐾 will be used in
apply the PU approximation to the local approximation to get
constructing the local approximation at node 𝐼. Similarly, if
the interpolation. The definition of the nodal support domain
a point, point x in Figure 2, for example, is contained in 𝑐𝐼 ,
will be given in detail and the local approximations at a node
will be described. then the local approximation 𝑢𝑙𝐼 (x) will contribute to the PU
approximation at x. For an arbitrary node, for example, node
𝐼 in Figure 1, the size of 𝑙𝐼 is defined by
2.1. Local Approximation at a Node. The local approximation
𝑢𝑙𝐼 (x) at an arbitrary node 𝐼 is given by 𝑟𝑙𝐼 = 𝑎 ⋅ 𝑏 ⋅ 𝑑𝐼 , (8)
𝑀
𝐼 where 𝑎 is a scale factor that ranges between 1.0 and 2.0,
𝑢𝑙𝐼 (x) = ∑ 𝜓𝐽 (x) 𝑢𝐽 , (3) 𝑏 is a coefficient such that 𝑏 = 2 for a node lying on the
𝐽=1
boundary and 𝑏 = 1 for all other nodes, and 𝑑𝐼 is the distance
where 𝑢𝐽 is the nodal displacement for the 𝐽th node in between 𝐼 and the fifth nearest neighbor node to 𝐼. If there is a
support of node 𝐼, 𝑀 is the total number of nodes in the predefined triangular background mesh, 𝑑𝐼 can be defined as
𝐼
local cover 𝑙𝐼 of node 𝐼 as shown in Figure 1, and 𝜓𝐽 (x) is the the maximum distance between 𝐼 and the nodes of triangles
modified least square shape function given by which are connected to the node 𝐼.
For a node having its local support domain completely
𝐼
𝜓𝐽 (x) = 𝜓𝐽𝐼 (x) − 𝜓𝐽𝐼 (x𝐼 ) , for 𝐽 ≠𝐼, inside the domain, for example, the subdomain 𝑙𝐼 of node 𝐼
(4) in Figure 2, the size of 𝑐𝐼 is the same as 𝑙𝐼 :
𝐼
𝜓𝐽 (x) = 𝜓𝐽𝐼 (x) − 𝜓𝐽𝐼 (x𝐼 ) + 1, for 𝐽 = 𝐼,
𝑟𝑐𝐼 = 𝑟𝑙𝐼 . (9)
where 𝜓𝐽𝐼 (x) are the modified least square shape functions of
node 𝐽 and are determined by the following equations: For a node having its local support domain close to or
intersecting the boundary, for example, node 𝐽 shown in
𝜓𝐼 (x) = [𝜓1𝐼 (x) 𝜓2𝐼 (x) ⋅ ⋅ ⋅ 𝜓𝑀
𝐼 T −1
(x)] = p (x) A B, Figure 2, the definition of subdomain follows these steps.
(5) Firstly, find the nearest boundary node to 𝐽 among the
neighbor nodes which have been used as nodes in defining
p(x) = [𝑝1 (x), 𝑝2 (x), . . . , 𝑝𝑚 (x)]T is a polynomial basis, and 𝑚 𝑙𝐽 , and secondly calculate the distance between the nearest
is the number of monomials in the basis. In the development boundary node and 𝐽, denoted as 𝑑𝐽 , and then the size of 𝑐𝐽 is
of the MSLS interpolation, we use a bilinear basis throughout set by
in 2D such that pT (x) = [1, 𝑥, 𝑦, 𝑥𝑦], and 𝑟𝑐𝐽 = 0.99𝑑𝐽 . (10)
1 1 ⋅⋅⋅ 1
[ 𝑥1 If there is an interior node where prescribed values needed
𝑥 ⋅ ⋅ ⋅ 𝑥𝑀 ]
B=P =[
T
[ 𝑦1
2 ], to be applied, the procedure described above for setting the
𝑦2 ⋅ ⋅ ⋅ 𝑦𝑀 ]
(6) support domain of near boundary nodes can be repeated
[𝑥1 𝑦1 𝑥2 𝑦2 ⋅ ⋅ ⋅ 𝑥𝑀𝑦𝑀]
to that node using (10) by assigning this interior node as a
A = PT ⋅ P. boundary node. If we want all nodes to take nodal values
at the nodes, the size of the 𝑑𝐽 can be taken as the distance
𝐼 𝐼
It can be seen from (4) that 𝜓𝐼 (x𝐼 ) = 1, 𝜓𝐽 (x𝐼 ) = 0 (𝐽 ≠𝐼) for between the 𝐽 and its nearest node for every node 𝐽. The
𝐼 following quadratic spline function is used as the weight
x = x𝐼 and ∑𝑀 𝐽=1 𝜓𝐽 (x) = 1. Thus function over support domain in (2):
𝑢𝑙𝐼 (x𝐼 ) = 𝑢𝐼 . (7) 2 3 4
{1 − 6( 𝑟𝐼 ) + 8( 𝑟𝐼 ) − 3( 𝑟𝐼 ) , 𝑟𝐼 ≤ 𝑟𝑐𝐼
{
It has been proved in [36, 37] that a singular weight function 𝑤𝐼 (x) = { 𝑟𝑐𝐼 𝑟𝑐𝐼 𝑟𝑐𝐼
{
used as 𝑤𝑖 (x) will enforce the interpolation, the equivalent { 0 𝑟𝐼 > 𝑟𝑐𝐼 ,
equation (1) here, satisfying the delta property. A similar (11)
approach has been used by [45] to produce interpolatory MLS
approximation. However, the use of singular weight function where 𝑟𝐼 = ‖x − x𝐼 ‖ is the distance between the point x
will bring some other problems such as the loss of smoothness and node 𝐼 and x𝐼 is the coordinate of node 𝐼. The aim
of interpolation. of separately defining local domain and support domain is
4 Mathematical Problems in Engineering

to produce MSLS interpolations having the delta property lI

without using a singular weight, so that the difficulties
associated with the use of singular weight function can be ΩsI
removed. Indeed this aim is achieved here if the domain for ΓsI
local approximation and domain for PU are defined by the
method described above as will be proved later in Section 2.3. I

2.3. Delta Property at a Node. Consider a boundary node 𝐾 to

be applied with boundary conditions. If the support domain
of the nodes is set according to (9) and (10), then the 𝐾 will be
lJ
the only node contained in 𝑐𝐾 . Thus the MSLS interpolation
equation (1) at x𝐾 becomes
ΩsJ
Γt
𝑛
ℎ
𝑢 (x𝐾 ) = ∑ 𝜑𝐼0 𝑙𝐼
(x𝐾 ) 𝑢 (x𝐾 ) = 0
𝜑𝐾 (x𝐾 ) 𝑢𝑙𝐾 (x𝐾 ) . (12)
𝐼=1
Γu J
Γsu
As there is only one node in the PU, then (2) becomes
Figure 3: The local integration domains.
𝑤 (x ) 𝑤 (x )
0
𝜑𝐾 (x𝐾 ) = 𝑛 𝐾 𝐾 = 𝐾 𝐾 = 1. (13)
∑𝐽=1 𝑤𝐽 (x𝐾 ) 𝑤𝐾 (x𝐾 )
For domain Ω bounded by Γ (Figure 3), the equilibrium
It is known by (7) that the local approximation u𝑙𝐾 (x𝐾 ) at
equations and boundary conditions of linear elasticity are
node 𝐾 satisfies
given by
𝑢𝑙𝐾 (x𝐾 ) = 𝑢𝐾 . (14)
𝜎𝐼𝐽,𝐽 + 𝑏𝐼 = 0 in Ω,
Substituting (13) and (14) into (12) gives
𝑢𝐼 = 𝑢𝐼 at Γ𝑢 , (17)
ℎ 𝑙𝐾
𝑢 (x𝐾 ) = 𝑢 (x𝐾 ) = 𝑢𝐾 . (15)
𝜎𝐼𝐽 𝑛𝐽 = 𝑡𝐼 at Γ𝑡 ,
Hence, the present MSLS interpolation takes nodal value at
boundary nodes and the essential boundary conditions or where 𝜎𝐼𝐽 is the stress tensor, 𝑏𝐼 are the body forces, 𝑛𝐽 are
point load conditions can be directly imposed as in the FEM. the unit normal to the domain, and Γ𝑢 and Γ𝑡 are the global
Also, the present interpolation preserves the consistency up boundaries with prescribed displacements and tractions,
to the order of the basis function, which is a necessary respectively. Similar to [36], the local polygonal subdomains
requirement of accuracy. The proof is the same as has been are constructed for the purpose of simplifying the integration
presented in [36, 37]. and the discrete equations. For example, based on Delaunay
algorithm, a local polygon Ω𝑠𝐼 is constructed by using the 𝑀
3. The Local Petrov-Galerkin Weak Form nodes in local cover 𝑙𝐼 , as shown in Figure 3. A generalized
local weak form of the equilibrium equation in (17) is written
Let 𝑅 be the total number of nodes associated with the given as
point x; then (3) can be rewritten as
∫ V𝐼 (𝜎𝐼𝐽,𝐽 + 𝑏𝐼 ) 𝑑Ω = 0,
𝜓1 ⋅ ⋅ ⋅ 𝜓𝑛 󵄨󵄨󵄨󵄨 𝜓𝑛+1
0 T 1 1 1 1 (18)
{ 𝜑1 } ⋅ ⋅ ⋅ 𝜓𝑅 { 𝑢1 } Ω𝑠𝐼
{
{ 0} [ 𝜓 ⋅ ⋅ ⋅ 𝜓 󵄨󵄨󵄨 𝜓2󵄨 { }
{
{𝜑2 } ⋅ ⋅ ⋅ 𝜓𝑅 ] }
2 2 2
} [ 1 𝑛 󵄨 𝑛+1
󵄨 ] { 𝑢2 }
𝑢 (x) = { . } [ . .. ]
[ .. ⋅ ⋅ ⋅ ... 󵄨󵄨󵄨󵄨 ...
.
{. }
{
{ . } ⋅⋅⋅ . ]{ } where Ω𝑠𝐼 is the integration domain or subdomain for node
{.} } 󵄨
𝑛 󵄨 𝑛
{. }
{ }
[ 𝜓1 ⋅ ⋅ ⋅ 𝜓𝑛 󵄨󵄨󵄨 𝜓𝑛+1 𝐼 and V𝐼 is the test function. Using the divergence theorem in
0 𝑛 𝑛
{𝜑𝑛 } ⋅ ⋅ ⋅ 𝜓𝑅 ] {𝑢𝑅 }
(18), we obtain the following local weak form:
𝑅
= Φ01×𝑛 𝜓𝑛×𝑅 u𝑅×1 = ∑ 𝑁𝑘 (x) 𝑢𝑘 ,
𝑘=1 ∫ V𝐼 𝜎𝐼𝐽 𝑛𝐽 𝑑Γ − ∫ (V𝐼,𝐽 𝜎𝐼𝐽 − V𝐼 𝑏𝐼 ) 𝑑Ω = 0, (19)
(16) 𝜕Ω𝑠𝐼 Ω𝑠𝐼

where Φ0 is the vector of Shepard shape function, 𝜓 is a where 𝑛𝐽 is the outward unit normal to the boundary
matrix comprising the modified least square point interpo- 𝜕Ω𝑠𝐼 . The boundary 𝜕Ω𝑠𝐼 for the subdomain Ω𝑠𝐼 is usually
lation (LSPI) shape functions, and 𝑁𝑘 (x) is the MSLS shape composed of three parts: the internal boundary Γ𝑠𝐼 , the
function. boundary Γ𝑠𝑢 , and Γ𝑠𝑡 , over which the essential and natural
Mathematical Problems in Engineering 5

boundary conditions are specified. Substituting 𝜎𝐼𝐽 𝑛𝐽 = 𝑡𝐼 in interpolation on a range of test problems. The results are
(19), the following is obtained: compared with the exact solutions, the MLPG solutions,
and the linear FEM solutions. The weight functions used in
∫ V𝐼 𝑡𝐼 𝑑Γ + ∫ V𝐼 𝑡𝐼 𝑑Γ + ∫ V𝐼 𝑡𝐼 𝑑Γ the MLPG for testing purpose are the Gaussian type weight
Γ𝑠𝐼 Γ𝑠𝑢 Γ𝑠𝑡 functions given by
(20)
−∫ (V𝐼,𝐽 𝜎𝐼𝐽 − V𝐼 𝑏𝐼 ) 𝑑Ω = 0. −(𝑟 /𝑐 )2 −(𝑟 /𝑐𝐼 )2
{𝑒
{ − 𝑒 𝑙𝐼
𝐼 𝐼
Ω𝑠𝐼 , if 𝑟𝐼 ≤ 𝑟𝑙𝐼
𝑤𝐼𝑐 (x) = { 1 − 𝑒−(𝑟𝑙𝐼 /𝑐𝐼 )2 (28)
{
In order to simplify (20), we can deliberately select the three- {0, if 𝑟𝐼 > 𝑟𝑙𝐼 ,
node triangular FEM shape functions 𝑁𝐼 , which correspond
to the node 𝐼 of the triangles constructing the polygonal where 𝑟𝑙𝐼 is defined by (8) and 𝑐𝐼 = 0.3𝑟𝑙𝐼 is used for all test
subdomain Ω𝑠𝐼 , as test functions V𝐼 , such that they vanish examples. The scale factor 𝑎 in (8) is set to be 1.5 and the
over Γ𝑠𝐼 . Substituting shape functions 𝑁𝐼 for V𝐼 in (20), we linear bases are used in MLPG and MSLSM. To study the
obtain the following local weak form: convergence behavior we define the following error norms in
displacement and energy, respectively:
∫ 𝑁𝐼 𝑡𝐼 𝑑Γ + ∫ 𝑁𝐼 𝑡𝐼 𝑑Γ − ∫ (𝑁𝐼,𝐼 𝜎𝐼𝐽 − 𝑁𝐼 𝑏𝐼 ) 𝑑Ω = 0. 1/2
Γ𝑠𝑢 Γ𝑠𝑡 Ω𝑠𝐼
‖u‖ = (∫ uT ⋅ u 𝑑Ω) , (29)
(21) Ω

For a local polygonal subdomain Ω𝑠𝐼 located entirely in the where u is a vector collecting nodal displacement results u =
global domain Ω, there is no intersection between 𝜕Ω𝑠𝐼 and {𝑢1 , V1 , 𝑢2 , V2 . . . 𝑢𝑛 , V𝑛 }T and
the global boundary Γ, and the integrals over Γ𝑠𝑢 and Γ𝑠𝑡 in
(21) vanish. For a local polygonal subdomain Ω𝑠𝐼 near the 1 1/2
boundary, the first item of (21) can also be omitted because ‖𝜀‖ = ( ∫ 𝜀T ⋅ 𝜎 𝑑Ω) , (30)
2 Ω
of the properties of the test functions 𝑁𝐼 . Substituting the
MSLS approximation in (21) into the above equation leads to where 𝜀 is the infinitesimal strain tensor and 𝜎 is the Cauchy
the following discretised system of linear equations: stress tensor. The relative displacement error and energy error
are calculated by
(∫ k𝐼T DB 𝑑Ω) ⋅ U = ∫ N𝐼 t 𝑑Γ + ∫ N𝐼 b 𝑑Ω (22) 󵄩󵄩 num 󵄩
󵄩u − uexact 󵄩󵄩󵄩
𝑟𝑢 = 󵄩
Ω𝑠𝐼 Γ𝑠𝐼 Ω𝑠𝐼
,
‖uexact ‖
denoted as (31)
󵄩󵄩󵄩𝜀num − 𝜀exact 󵄩󵄩󵄩
KU = F, (23) 𝑟𝑒 = 󵄩 󵄩,
‖𝜀exact ‖
where D is the elasticity matrix:
where the superscripts num and exact refer to numerical
solutions and exact (or reference) solutions, respectively.
𝑁𝐼,𝑥 0
k𝐼 = [ 0 𝑁𝐼,𝑦 ] , (24)
4.1. A Constant Strain Patch Test. A constant strain patch
[𝑁𝐼,𝑦 𝑁𝐼,𝑥 ]
test [46] using three distributions of 7, 28, and 126 irregular
𝑁𝐼 0 nodes is shown in Figure 4. Young’s modulus and Poission’s
N𝐼 = [ ], (25)
0 𝑁𝐼 ratio of the material are 1000 and 0.25, respectively. The
thickness of the plate is taken as a unit following plane
𝑁1,𝑥 0 ⋅ ⋅ ⋅ 𝑁𝑅,𝑥 0 stress assumption. Since the exact solution is linear, a linear
[ ]
B = [ 0 𝑁1,𝑦 ⋅ ⋅ ⋅ 0 𝑁𝑅,𝑦 ] , (26) basis for the MSLS interpolation is able to represent this
solution. The computational results in Table 1 show that the
[𝑁1,𝑦 𝑁1,𝑥 ⋅ ⋅ ⋅ 𝑁𝑅,𝑦 𝑁𝑅,𝑥 ] present MSLSM passes the patch tests exactly up to the double
U = {𝑢1 V1 ⋅ ⋅ ⋅ 𝑢𝑅 V𝑅 } . (27) precision of the computer.

Equation (23) can be individually integrated over each tri- 4.2. A Cantilever Beam. A cantilever beam problem with
angle constructing the local subdomain Ω𝑠𝐼 , as shown in dimensions of 𝑙 = 8 m and 𝑑 = 1 m, as shown in Figure 5, is
Figure 3. In the present work, seven Gaussian points are used tested first. The beam is subjected to parabolically distributed
in each triangle. downward traction equivalent to a unit load 𝑝 at the right-
hand end and is constrained at the left-hand end as shown
4. Numerical Examples in Figure 5. The elastic material properties used are 𝐸 =
1 × 105 Pa and ] = 0.25, and the problem is solved under a
The proposed MSLSM interpolation has been coded in C++. plane strain assumption. We refer to the analytical solution
In this section, we show the performance of the present of the problem given in [47]. The convergence of the present
6 Mathematical Problems in Engineering

y y
1000
3 7
x
d
1.0

p
l
2
(a) Cantilever beam
4
1.0
0.5

1 5 x
6
2.0 0.5 1.5
(b) Nodal arrangement
7 nodes
Figure 5: Cantilever beam and its nodal arrangement.

−0.5
−1.0
−1.5
−2.0
−2.5
−3.0
log(ru )
28 nodes −3.5
−4.0
−4.5
−5.0
−5.5
−6.0
−6.5
−1.1 −1.0 −0.9 −0.8 −0.7 −0.6 −0.5 −0.4 −0.3 −0.2
log(h)

MSLS
MLPG
126 nodes FEM

Figure 4: Nodal arrangements for the constant strain patch tests. Figure 6: Convergence of relative displacement error of the can-
tilever beam.

Table 1: Results of the constant strain patch test.

Number of nodes 𝑢4 V4 𝑢5 V5 𝑢7 V7 efficiency between the two interpolations gets bigger when
the number of nodes increases.
7 2.0 −0.125 2.5 0.0 4.0 −0.5
28 2.0 −0.125 2.5 0.0 4.0 −0.5
126 2.0 −0.125 2.5 0.0 4.0 −0.5 4.3. An Infinite Plate with a Circular Hole. The second exa-
mple is an infinite plate with a circular hole of radius 𝑎 =
Exact 2.0 −0.125 2.5 0.0 4.0 −0.5
1 m. The plate is subjected to a far field traction 𝜎 = 1 Pa
in the 𝑥 direction. A finite portion of the plate is considered
for analysis and, due to the symmetry of the problem, only
a quarter of the portion requires modeling, as shown in
method is studied using four nodal arrangements with 50, Figure 10. The elastic material properties used are 𝐸 =
138, 402, and 965 nodes. 3.0 × 107 Pa and ] = 0.3 and plane stress conditions are
The convergence rate is compared among FEM, MLPG, assumed. The stresses and displacements for this are given in
and the present MSLSM in Figures 6 and 7. It can be found an analytical solution in [47] as
that MSLSM shows a good accuracy and convergence rate.
Figures 8 and 9 show the vertical displacement V and the nor- 𝑎2 3 3𝑎4
mal stress 𝜎𝑥𝑥 along the 𝑦 = 𝑑/2 indicating accuracy of results 𝜎𝑥𝑥 = 1 − 2
( cos (2𝜃) + cos (4𝜃)) + 4 cos (4𝜃) ,
𝑟 2 2𝑟
using irregular 138 nodes and the proposed formulation.
As has been highlighted in [36, 37], the computational 𝑎2 1 3𝑎4
𝜎𝑥𝑦 = − 2 ( sin (2𝜃) + sin (4𝜃)) + 4 sin (4𝜃) ,
cost in obtaining the shape functions and its derivatives is 𝑟 2 2𝑟
much lower by the present LS interpolation than by the 𝑎2 1 3𝑎4
MLS approximation. This is shown in Table 2. It can also be 𝜎𝑦𝑦 = − 2 ( cos (2𝜃) − cos (4𝜃)) − 4 cos (4𝜃) ,
𝑟 2 2𝑟
observed from the table that the difference in computational (32)
Mathematical Problems in Engineering 7

−0.4 Table 2: Comparison of computational time in obtaining the strain

matrix (unit: second).
−0.8
Number of nodes 50 138 402 965
MLS 0.16 0.59 1.89 4.35
−1.2
Present MSLS 0.14 0.46 1.15 2.50
log(re )

−1.6

−2.0

−2.4
−1.0 −0.8 −0.6 −0.4 −0.2
log(h)

MSLS
MLPG
FEM (a) A small portion taken for analysis

Figure 7: Convergence of relative energy error of the cantilever 𝜎yy

beam.
𝜎xy
−2.2
−2.0 y 𝜎xx 𝜎xy
−1.8
−1.6
uy (×10−2 m)

−1.4
−1.2
−1.0
−0.8
b

−0.6
−0.4 (
−0.2
0.0 r
0 2 4 6 8 a
𝜃
X (m)
b x
Exact MLPG
MSLS FEM (b) Boundary conditions applied

Figure 8: Vertical displacement results V along 𝑦 = 0 of the Figure 10: An infinite plate with a circular hole.
cantilever beam.

45
𝑎 𝑟 2𝑎
40 𝑢= ( (𝜅 + 1) cos (𝜃) + [(1 + 𝜅) cos (𝜃) + cos 3 (𝜃)]
8𝐺 𝑎 𝑟
35
30 2𝑎3
− cos 3 (𝜃)) ,
𝑟3
𝜎x (Pa)

25
20
𝑎 𝑟 2𝑎
15 V = ( (𝜅 − 3) sin (𝜃) + [(1 − 𝜅) sin (𝜃) + sin 3 (𝜃)]
10
8𝐺 𝑎 𝑟
5
2𝑎3
0 − sin 3 (𝜃)) ,
1 2 3 4 5 6 7 8 𝑟3
X (m) (33)
Exact MLPG
MSLS FEM where 𝐺 is the shear modulus and 𝜅 is the Kolosov constant
where 𝜅 = (3 − ])/(1 − ]) for the plane strain assumption.
Figure 9: 𝜎𝑥𝑥 results along 𝑦 = 0 of the cantilever beam. Traction-prescribed boundary conditions consistent with
the exact solution in (32) are applied at the top and right
8 Mathematical Problems in Engineering

53 nodes 188 nodes

564 nodes 1012 nodes

Figure 11: Nodal arrangements used for the infinite plate problem.

edges. Four distributions of 53, 188, 564, and 1012 nodes, −2.0
which are shown in Figure 11, are employed for the conver-
−2.5
gence studies. Figures 12 and 13 show that the MSLS has a
good convergence rate in the displacement and energy norm. −3.0
The convergence slope of the present method is similar to that
−3.5
log(ru )

of MLPG though the latter is seen to be more accurate. The

displacement 𝑢𝑥 obtained using MSLS and MLPG is shown −4.0
in Figure 14. It is seen that the displacement values given by
−4.5
MSLS as well as MLPG are very close to the exact solution.
−5.0

5. Conclusions −5.5
−0.8 −0.7 −0.6 −0.5 −0.4 −0.3 −0.2 −0.1 0.0
In this paper, we proposed a local weak form meshless log(h)
Shepard and least-squares interpolation. The interpolation MSLS
features the use of duo nodal supports for local approxi- MLPG
mation and global approximation, respectively. The present FEM
formulation satisfies the delta property at desired nodes
without using singular weight functions. The shape functions Figure 12: Convergence of relative displacement error for the
constructed conform to the PU property. The local Petrov- infinite plate problem.
Galerkin weak form is used so that there is no need for
Mathematical Problems in Engineering 9

−1.4 Conflict of Interests

−1.6
−1.8 The authors declare that there is no conflict of interests
−2.0 regarding the publication of this paper.
−2.2
log(re )

−2.4 Acknowledgments
−2.6
−2.8
The authors gratefully acknowledge the support of the
NSFC Programme (41130751, 51109162), 973 Program
−3.0
(2011CB013800), Shanghai Pujiang Talent Program
−3.2
(12PJ1409100), SRF for ROCS of State Education Ministry,
−3.4
−0.8 −0.7 −0.6 −0.5 −0.4 −0.3 −0.2 −0.1 0.0 and China Central University Funding.
log(h)

MSLS References
MLPG
[1] T. Fries and H. Matthies, Classification and Overview of Mesh-
FEM
free Methods, Technical University Braunschweig, Brunswick,
Figure 13: Convergence of relative energy error for the infinite plate Germany, 2004.
problem. [2] S. Hao, W. K. Liu, and T. Belytschko, “Moving particle finite
element method with global smoothness,” International Journal
for Numerical Methods in Engineering, vol. 59, no. 7, pp. 1007–
1020, 2004.
2.0
[3] V. P. Nguyen, T. Rabczuk, S. Bordas, and M. Duflot, “Meshless
methods: a review and computer implementation aspects,”
1.8
Mathematics and Computers in Simulation, vol. 79, no. 3, pp.
763–813, 2008.
1.6
ux (×10−7 m)

[4] X. Zhuang, C. E. Augarde, and K. M. Mathisen, “Fracture mod-

1.4 eling using meshless methods and levels sets in 3D: framework
and modeling,” International Journal for Numerical Methods in
1.2 Engineering, vol. 92, no. 11, pp. 969–998, 2012.
[5] X. Zhuang, C. Augarde, and S. Bordas, “Accurate fracture
0.1 modelling using meshless methods, the visibility criterion and
level sets: formulation and 2D modelling,” International Journal
0.8 for Numerical Methods in Engineering, vol. 86, no. 2, pp. 249–
1 2 3 4 5
268, 2011.
X (m)
[6] T. Rabczuk and T. Belytschko, “Cracking particles: a simplified
Exact meshfree method for arbitrary evolving cracks,” International
MSLS Journal for Numerical Methods in Engineering, vol. 61, no. 13, pp.
MLPG 2316–2343, 2004.
Figure 14: Comparison of the horizontal displacement 𝑢 along 𝑦 = [7] T. Rabczuk and G. Zi, “A meshfree method based on the
0 by different methods. local partition of unity for cohesive cracks,” Computational
Mechanics, vol. 39, no. 6, pp. 743–760, 2007.
[8] T. Rabczuk, P. M. A. Areias, and T. Belytschko, “A meshfree thin
shell method for non-linear dynamic fracture,” International
Journal for Numerical Methods in Engineering, vol. 72, no. 5, pp.
background mesh. It has been tested that the present for- 524–548, 2007.
mulation is more efficient when compared to the currently [9] T. Jarak and J. Sorić, “Analysis of rectangular square plates by
widely used MLS approximation. For the 2D example of 965 the mixed meshless local Petrov-Galerkin (MLPG) approach,”
nodes, the computing time of the present MSLS is close to half Computer Modeling in Engineering & Sciences, vol. 38, no. 3, pp.
of the computing time of MLS. The accuracy of the present 231–261, 2008.
method is slightly lower than the MLPG but higher than the [10] P. Areias and T. Rabczuk, “Finite strain fracture of plates and
FEM when the same number of nodes and the same order shells with configurational forces and edge rotation,” Interna-
of basis function, for example, linear or quadratic bases, are tional Journal For Numerical Methods in Engineering, vol. 94,
used (linear basis corresponding to triangular element in the pp. 1099–1122, 2013.
FEM). The formulation here is derived for 2D analysis but [11] P. Areias, T. Rabczuk, and D. Dias-da-Costa, “Assumed-metric
is readily extendable to 3D and the essential ideas are the spherically-interpolated quadrilateral shell element,” Finite Ele-
same. Further development of the present interpolation such ments in Analysis and Design, vol. 66, pp. 53–67, 2013.
as application to problems of changing geometry/moving [12] C. H. Thai, H. Nguyen-Xuan, N. Nguyen-Thanh, T.-H. Le,
boundaries, finite deformation, elastoplasticity, and three- T. Nguyen-Thoi, and T. Rabczuk, “Static, free vibration, and
dimensional cracking is ongoing. buckling analysis of laminated composite Reissner-Mindlin
10 Mathematical Problems in Engineering

plates using NURBS-based isogeometric approach,” Interna- discrete shear gap technique for analysis of laminated composite
tional Journal for Numerical Methods in Engineering, vol. 91, no. plates,” Applied Mathematics and Computation, vol. 217, no. 17,
6, pp. 571–603, 2012. pp. 7324–7348, 2011.
[13] T. Chau-Dinh, G. Zi, P. S. Lee, J. H. Song, and T. Rabczuk, [28] N. Vu-Bac, H. Nguyen-Xuan, L. Chen et al., “A phantom-node
“Phantom-node method for shell models with arbitrary cracks,” method with edge-based strain smoothing for linear elastic
Computers and Structures, vol. 92-93, pp. 242–256, 2012. fracture mechanics,” Journal of Applied Mathematics, vol. 2013,
[14] N. Nguyen-Thanh, J. Kiendl, H. Nguyen-Xuan, R. Wüchner, K. Article ID 978026, 12 pages, 2013.
U. Bletzinger, and Y. Bazilevs, “Rotation free isogeometric thin [29] Y. Jia, Y. Zhang, G. Xu, X. Zhuang, and T. Rabczuk, “Reproduc-
shell analysis using PHT-splines,” Computer Methods in Applied ing kernel triangular B-spline-based FEM for solving PDEs,”
Mechanics and Engineering, vol. 200, no. 47-48, pp. 3410–3424, Computer Methods in Applied Mechanics and Engineering, vol.
2011. 267, pp. 342–358, 2013.
[15] N. Nguyen-Thanh, T. Rabczuk, H. Nguyen-Xuan, and S. Bordas, [30] P. Lancaster and K. Salkauskas, “Surfaces generated by moving
“An alternative alpha finite element method with discrete shear least squares methods,” Mathematics of Computation, vol. 37, no.
gap technique for analysis of Mindlin-Reissner plates,” Finite 155, pp. 141–158, 1981.
Elements in Analysis and Design, vol. 47, pp. 519–535, 2011. [31] P. Breitkopf, A. Rassineux, G. Touzot, and P. Villon, “Explicit
[16] T. Rabczuk and T. Belytschko, “A three-dimensional large form and efficient computation of MLS shape functions and
deformation meshfree method for arbitrary evolving cracks,” their derivatives,” International Journal for Numerical Methods
Computer Methods in Applied Mechanics and Engineering, vol. in Engineering, vol. 48, no. 3, pp. 451–466, 2000.
196, no. 29-30, pp. 2777–2799, 2007. [32] X. Zhuang and C. Augarde, “Aspects of the use of orthogonal
[17] Z. D. Han and S. N. Atluri, “Meshless local Petrov-Galerkin basis functions in the element-free Galerkin method,” Interna-
(MLPG) approaches for solving 3D problems in elasto-statics,” tional Journal for Numerical Methods in Engineering, vol. 81, no.
CMES. Computer Modeling in Engineering & Sciences, vol. 6, no. 3, pp. 366–380, 2010.
2, pp. 169–188, 2004. [33] X. Zhuang, C. Heaney, and C. Augarde, “On error control in
[18] T. Rabczuk, S. Bordas, and G. Zi, “On three-dimensional the element-free Galerkin method,” Engineering Analysis with
modelling of crack growth using partition of unity methods,” Boundary Elements, vol. 36, no. 3, pp. 351–360, 2012.
Computers and Structures, vol. 88, no. 23-24, pp. 1391–1411, 2010. [34] I. Kaljević and S. Saigal, “An improved element free Galerkin
[19] T. Rabczuk, R. Gracie, J.-H. Song, and T. Belytschko, “Immersed formulation,” International Journal for Numerical Methods in
particle method for fluid-structure interaction,” International Engineering, vol. 40, no. 16, pp. 2953–2974, 1997.
Journal for Numerical Methods in Engineering, vol. 81, no. 1, pp. [35] J.-S. Chen and H.-P. Wang, “New boundary condition treat-
48–71, 2010. ments in meshfree computation of contact problems,” Computer
[20] T. Rabczuk and E. Samaniego, “Discontinuous modelling of Methods in Applied Mechanics and Engineering, vol. 187, no. 3-4,
shear bands using adaptive meshfree methods,” Computer pp. 441–468, 2000.
Methods in Applied Mechanics and Engineering, vol. 197, no. 6–8, [36] Y. C. Cai and H. H. Zhu, “A local meshless Shepard and
pp. 641–658, 2008. least square interpolation method based on local weak form,”
[21] T. Rabczuk, T. Belytschko, and S. P. Xiao, “Stable particle Computer Modeling in Engineering & Sciences, vol. 34, no. 2, pp.
methods based on Lagrangian kernels,” Computer Methods in 179–204, 2008.
Applied Mechanics and Engineering, vol. 193, no. 12–14, pp. 1035– [37] X. Zhuang, H. Zhu, and C. Augarde, “An improved meshless
1063, 2004. Shepard and least squares method possessing the delta property
[22] S. N. Atluri and T. Zhu, “The Meshless Local Petrov- and requiring no singular weight function,” Computational
Galerkin(MLPG) approach for solving problems in elasto- Mechanics, 2013.
statics,” Computational Mechanics, vol. 25, pp. 169–179, 2000. [38] G. R. Liu, Y. T. Gu, and K. Y. Dai, “Assessment and applications
[23] J. Zheng, S. Long, Y. Xiong, and G. Li, “A finite volume meshless of point interpolation methods for computational mechanics,”
local Petrov-Galerkin method for topology optimization design International Journal for Numerical Methods in Engineering, vol.
of the continuum structures,” Computer Modeling in Engineer- 59, pp. 1373–1397, 2004.
ing & Sciences, vol. 42, no. 1, pp. 19–34, 2009. [39] J. Dolbow and T. Belytschko, “Numerical integration of the
[24] Y. Cai, X. Zhuang, and C. Augarde, “A new partition of unity Galerkin weak form in meshfree methods,” Computational
finite element free from the linear dependence problem and Mechanics, vol. 23, no. 3, pp. 219–230, 1999.
possessing the delta property,” Computer Methods in Applied [40] S. Beissel and T. Belytschko, “Nodal integration of the element-
Mechanics and Engineering, vol. 199, no. 17–20, pp. 1036–1043, free Galerkin method,” Computer Methods in Applied Mechanics
2010. and Engineering, vol. 139, no. 1–4, pp. 49–74, 1996.
[25] Y. Cai, X. Zhuang, and H. Zhu, “A generalized and efficient [41] J. S. Chen, C. T. Wu, S. Yoon, and Y. You, “A stabilized
method for finite cover generation in the numerical manifold conforming nodal integration for Galerkin meshfree methods,”
method,” International Journal of Computational Methods, vol. International Journal for Numerical Methods in Engineering, vol.
10, no. 5, Article ID 1350028, 19 pages, 2013. 50, pp. 435–466, 2001.
[26] H. Nguyen-Xuan, T. Rabczuk, T. Nguyen-Thoi, T. N. Tran, [42] M. A. Puso, J. S. Chen, E. Zywicz, and W. Elmer, “Meshfree and
and N. Nguyen-Thanh, “Computation of limit and shakedown finite element nodal integration methods,” International Journal
loads using a node-based smoothed finite element method,” for Numerical Methods in Engineering, vol. 74, no. 3, pp. 416–
International Journal for Numerical Methods in Engineering, vol. 446, 2008.
90, no. 3, pp. 287–310, 2012. [43] S. N. Atluri, The Meshless Local Petrov-Galerkin (MLPG)Method
[27] C. Thai-Hoang, N. Nguyen-Thanh, H. Nguyen-Xuan, and T. for Domain & Boundary Discretizations, Tech Science Press,
Rabczuk, “An alternative alpha finite element method with 2004.
Mathematical Problems in Engineering 11

[44] A. Carpinteri, G. Ferro, and G. Ventura, “The partition of

unity quadrature in meshless methods,” International Journal
for Numerical Methods in Engineering, vol. 54, no. 7, pp. 987–
1006, 2002.
[45] I. Kaljević and A. S. Saigal, “An improved element free Galerkin
formulation,” International Journal for Numerical Methods in
Engineering, vol. 40, no. 16, pp. 2953–2974, 1997.
[46] S. Rajendran and B. R. Zhang, “A “FE-meshfree” QUAD4
element based on partition of unity,” Computer Methods in
Applied Mechanics and Engineering, vol. 197, no. 1–4, pp. 128–
147, 2007.
[47] S. P. Timoshenko and J. N. Goodier, Theory of Elasticity,
McGraw-Hill, New York, NY, USA, 3rd edition, 1951.
Hindawi Publishing Corporation
Mathematical Problems in Engineering
Volume 2014, Article ID 179169, 11 pages
http://dx.doi.org/10.1155/2014/179169

Research Article
A Coupled Thermo-Hydro-Mechanical Model of
Jointed Hard Rock for Compressed Air Energy Storage

Xiaoying Zhuang,1,2 Runqiu Huang,2 Chao Liang,3 and Timon Rabczuk4,5

1
National Key Laboratory of Disaster Reduction and Protection, Department of Geotechnical Engineering, Tongji University,
Shanghai 200092, China
2
State Key Laboratory of Geohazard Prevention and Geoenvironment Protection, Chengdu, China
3
Department of Geophysics, School of Earth Sciences, Stanford University, USA
4
Institute of Structural Mechanics, Bauhaus-Universität Weimar, Weimar, Germany
5
School of Civil, Environmental and Architectural Engineering, Korea University, Republic of Korea

Correspondence should be addressed to Xiaoying Zhuang; [email protected]

and Timon Rabczuk; [email protected]
Received 8 September 2013; Accepted 7 November 2013; Published 19 January 2014

Academic Editor: Goangseup Zi

Copyright © 2014 Xiaoying Zhuang et al. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.

Renewable energy resources such as wind and solar are intermittent, which causes instability when being connected to utility grid of
electricity. Compressed air energy storage (CAES) provides an economic and technical viable solution to this problem by utilizing
subsurface rock cavern to store the electricity generated by renewable energy in the form of compressed air. Though CAES has been
used for over three decades, it is only restricted to salt rock or aquifers for air tightness reason. In this paper, the technical feasibility
of utilizing hard rock for CAES is investigated by using a coupled thermo-hydro-mechanical (THM) modelling of nonisothermal
gas flow. Governing equations are derived from the rules of energy balance, mass balance, and static equilibrium. Cyclic volumetric
mass source and heat source models are applied to simulate the gas injection and production. Evaluation is carried out for intact
rock and rock with discrete crack, respectively. In both cases, the heat and pressure losses using air mass control and supplementary
air injection are compared.

1. Introduction incurred in storing and retrieving the energy. For exam-

ple, batteries are costly, fly wheels are suitable for short-
Renewable energy such as wind, solar, tidal, and wave only duration storage only. The CAES, besides pumped-hydro, is
produces electricity intermittently and with low power and the only conceivable technology able to provide the very large
energy density, thus, nondispatchable and difficult to use at scale energy storage deliverability above 100 MW in single
large scales as the modern society requires [1]. That is why unit sizes while free from adverse environmental effects of
many renewable energy technologies are lacking the econo- pumped-hydro. Hence, CAES has recently received lots of
mies of scale, which reduces their competitiveness and delays attention [2, 3] and it has been recently proposed that large-
the transition to a low carbon economy. Therefore, economic scale solar-CAES and wind-CAES deployment can enable
solutions to bulk energy storage are urgently needed in order renewable energy to compete against coal-fired electricity
for renewable energy to take a significant share in the total generation [4, 5]. In CAES, a source energy is stored in the
energy mix. A critical issue for renewable energy to be form of highly pressurized air in underground rock cav-
integrated into grids with satisfactory stability is appropriate erns and the compressed air is released through turbines to
energy storage to defer electricity demand from peak to off generate electricity when needed [6] as shown in Figure 1.
peak times. Underground rock caverns are the mostly chosen type of
Most energy storage systems are expensive, either in reservoirs in the CAES which provide initial ground stress
terms of Capex and Opex or in terms of energy losses against air pressure, strengths to withstand cyclic loadings,
2 Mathematical Problems in Engineering

Wind
technology waits for the exploring and characterizing sites
Grid suitable for storage.
Solar To the best of the authors’ knowledge, the CAES projects
Turbine / declared over the world are listed in Table 1. From the
Recuperator generator successful experiences of the existing facilities, the caverns
are ideally constructed in salt rock formations due to
Compressor
their good air tightness. However, the requirements of salt
rocks are too restrictive and not available in many areas.
Air in Air out Recently, research interests in exploring the potential of
using bedrocks other than salt rock have been motivated
Cavern Cavern Ca ve rn
such as crystalline rock [12], sedimentary porous rock [13],
Host rock
aquifer, depleted oil/gas wells [14], and limestone [15]. More
CAES plants built on various types of rock are under plan
around the world such as the 300 MW plant by Pacific Gas
Figure 1: Compressed air energy storage plant using hard rock
& Electric in CA, USA [15], 300 MW plant in China [11], and
caverns.
other pilot plants in Korea [12] and Israel [13]. In Australia,
sandstones are typical hard rock from depleted copper, gold,
and silver mines. Sandstones are normally porous enough
to store large quantities and can be used as valuable aquifers
and large volumes of storage space and therefore are more
for reservoirs. Such CAES can provide sufficient strengths
economical than other types of storage such as steel tanks [7].
to sustain the stress induced by the rise and drop of air
Existing and planning CAES facilities for utility purpose
pressure and temperatures. The depleted reservoir formation
are mainly in salt rock formation where vacuums are formed
is economically attractive because it allows the reuse, with
by pumping pressurized water to dissolve the rock medium.
suitable modification, which reduces the start-up costs.
The only existing CAES built in hard rock, however, for
The most significant challenge in CAES is to ensure the
experimental purpose, is a 1600 m3 cave constructed in air tightness and prevent the air leak-off that may occur due
deplete mine in Hokkaido, Japan. The air pressure during to crack propagation under long-term cyclic actions resulting
cyclic charging and discharging is between 4 and 8 MPa and in pressure and heat losses. On the other hand, rock is natural
rubber lining is used for air tightness. Another small scale mineral aggregate with distributed discontinuities such as
CAES using water sealing in hard rock is now under plan in joints, cracks, or inclusions. The CAES operation exerts
Switzerland and air pressure will be around 3.3 MPa. Though complex and cyclic mechanical, thermal, and wet-dry cycles.
the CAES technology has come into existence for over 30 These excitations may cause monotonic failure and require
years [8], the exploration of utilizing hard rock is important close examination useful for engineering design of cavern.
for wider application of this technique since the requirement The stagnation of the 2700 MW CAES plant in Huston, the
of salt rock is too restrictive geological constraint in US, is an example [15]. The construction of the plant was
many areas. In this paper, we carried out the investigation launched in 2001 and stalled later due to the air leakage
of utilizing hard rock cavern for CAES through a coupled problem in the limestone caverns occurred [11, 14]. The
thermal-hydro-mechanical modelling for physical quantities. understanding of the air leak-off requires both understanding
on the mechanisms, geological conditions, the development
2. Geological Conditions for CAES of physical model, and advanced numerical simulation for
engineering design [16]. In the past two decades, there have
To date, there have been only two CAES plants in operation been significant advancements in fracture simulation meth-
at utility scales. The first is the 290 MW Huntorf plant in Ger- ods such as the Extended Finite Element Method (XFEM)
many with a total storage volume of 310,000 m3 , operational or extended meshfree methods [17–36] to allow for the
since 1978 [9, 10]. And the second of 110 MW-scale with a stor- application of discrete crack modelling techniques to THM
age space of 500,000 m3 was built in McIntosh, AL, in 1991 [6]. simulations [37, 38] as well as fluid structure interaction-
While energy conversion equipment including compressors, induced cracks [39]. In this paper, we aim to find out the
recuperators, and gas turbines is well established technology relation between cavern characteristics, energy efficiency, and
and procurable from the market, present challenges of CAES heat and pressure losses and stresses of rock induced by
lie in finding geo-storage spaces such as rock caverns that are hydraulic and thermal changes and the existence of crack.
competent for cost effective and reliable performance during
long-term plant operation. Though using subsurface rock as 3. Governing Equations of the THM Modelling
oil/gas storage started from the 1900s, new issues arise for
CAES including the cyclic actions of high air pressure and In this paper, three fields, namely, mechanical, thermal, and
thermalfluidics. Other types of storage such as above ground hydro fields, are considered and coupled. The hydro or seep-
gas tanks, though technically available, are too expensive, age field represents the gas flow behaviour, which changes the
about 5 times the cost of caverns [11]. Moreover, the security temperature distribution by the thermalfluidic flow through
of the storage is also a major concern for highly pressurized the porous media. The thermal field in turn will change the
tanks above ground. Hence, the implementation of CAES density and pressure of the fluid and hence the velocity of
Mathematical Problems in Engineering 3

Table 1: CAES projects over the world (data collected from various sources).

Power capacity Geological Cavern Embedded Max. pressure

Project name Location Type Status
(MW) formation volume (m3 ) depth (MPa)
Breman,
Huntorf 290 Salt Rock Diabetic 310,000 600 4.3–7.0 Operation
Germany
Norton, Hard rock
Norton 2700 Diabatic 9,600,000 670 5.5–11 Construction
USA (limestone)
Iowa Energy Iowa, Porous sand
270 Diabatic — 914 — Construction
Park USA stone
ADELE Germany 300 Salt Rock Adiabatic — — ∼10 Planning
Texsas,
Matagorga 540 Salt Rock Diabatic — — — Planning
USA
New York,
Seneca 150–270 Salt Rock Diabatic 150,000 760 8–11 Planning
USA
California, Porous
PG&E 300 Diabatic — — — Planning
USA aquifer
Inner
Datang CAES Mongolia, 300 Sand stone Diabatic 900,000 500 m 5–8 Planning
China

the fluid. The thermal strain will result in the internal stress The above assumptions are used both in the derivation
of the mechanical field, which influences the seepage field of the governing equations in the following, and the model
through volumetric strain. The seepage will change the pore settings of the numerical studies in Section 4.
pressure of the fluid with the porous media. The complete
coupling of the three fields can involve a large quantity of field 3.1. Fluid Field: Mass Conservation. The seepage of the air in
variables and material properties associated with each other rock satisfies the continuity equation described by
in the three fields, which is computationally expensive and
unnecessary from an engineering perspective. For example, 𝜕 (𝜌gas 𝜙)
the viscosity of the fluid is temperature dependent and can + ∇ ⋅ (𝜌gas q) = 𝑄𝑚 , (1)
𝜕𝑡
add another relation between seepage and thermal flied.
Therefore, we introduce some assumptions in the present where 𝜌gas is the density of the air, 𝜙 is the pore ratio of
THM modelling framework as follows. rock, q is the seepage velocity vector of the fluid, 𝑄𝑚 is the
source of mass (positive for inputting fluid and negative for
(1) The rock is treated as isotropic elastic porous media outputting). The first term on the LHS of (1) can be further
and the mechanical parameters of the rock are not calculated by
changing with respect to the temperature.
𝜕 (𝜌gas 𝜙) 𝜕𝜌gas 𝜕𝜙 𝜕𝜌gas 𝜕𝜀
(2) The seepage of the air through the rock follows =𝜙 + 𝜌gas =𝜙 + 𝑏 ⋅ 𝜌gas 𝑉 , (2)
Darcy’s law where the inertia and gravity of the air are 𝜕𝑡 𝜕𝑡 𝜕𝑡 𝜕𝑡 𝜕𝑡
not considered. where 𝜀𝑉 is the volumetric strain and 𝑏 is Biot’s consolidation
coefficient. The last term on the RHS of (2) shows the
(3) The physical state of compressed air conforms with
direction relation between the volumetric change of the
the ideal gas law.
porous media and the porosity ratio. The density of the air
(4) There is no ground water considered in the model and is changing with the temperature 𝑇 and pressure 𝑃 which is
it is a single phase fluid flow problem, that is, air. The described by the ideal gas law
rock pore is saturated by air.
𝑃𝑀
𝜌gas = , (3)
(5) The heat mass is transferred between the solid and 𝑅𝑇
fluid through both convection and conduction. The
where 𝑃 is the pore pressure, 𝑀 is Molar mass of air that
heat capacity and conductivity coefficients of all the
𝑀 = 29 g/mol, 𝑅 is ideal gas constant that 𝑅 = 8.314 𝐽/(mol ⋅
media are not changing with respect to the temper-
K), 𝑇 is absolute (thermodynamic) temperature and 𝑃 is the
ature and pressure. The volumetric strain energy of
pore pressure. The momentum of air seepage conforms to
the solid and the work produced by the fluid from the
Darcy’s law. If the inertia, gravity, and adhesion of air is not
air compression are taken into account for the energy
considered, the seepage equation of air is expressed by
balance.
𝑘
(6) The volume change of the porous media is mainly due q = − ∇𝑃, (4)
to the deformation of the pores. 𝜂
4 Mathematical Problems in Engineering

where 𝑘 is the permeability coefficient of rock and 𝜂 is the the interface of two phases is not considered. By combining
coefficient of viscosity. (6) with (7), we can get the following:
By substituting (3) and (4) into (2), the governing equa-
tions of seepage field can be obtained by 𝜕𝑇
(𝜌𝐶𝑝 )eq + 𝜌gas 𝐶𝑝,gas q ⋅ ∇𝑇
𝜕𝑡
𝜕𝜀𝑉 𝜙𝑀 𝜕𝑃 𝜙𝑀𝑃 𝜕𝑇 𝑃𝑀 𝑘 𝜕𝜀𝑉 (8)
𝑏 ⋅ 𝜌gas + − − ∇( ∇𝑃) = 𝑄𝑚 . + (1 − 𝜙) 𝑇𝐾𝑠 𝛽 + 𝑃 (∇ ⋅ q)
𝜕𝑡 𝑅𝑇 𝜕𝑡 𝑅𝑇2 𝜕𝑡 𝑅𝑇 𝜂 𝜕𝑡
(5) = ∇ ⋅ (𝛼eq ∇𝑇) + 𝑄,

(𝜌𝐶𝑝 )eq = 𝜙 ⋅ 𝜌gas 𝐶𝑝,gas + (1 − 𝜙) ⋅ 𝜌𝑠 𝐶𝑝,𝑠 , (9)

The changing density is the main difference between
the governing equations of air seepage and impressible fluid
seepage. The density of air changes with the temperature, 𝛼eq = 𝜙 ⋅ 𝛼gas + (1 − 𝜙) ⋅ 𝛼𝑠 , (10)
and therefore the thermal field is associated with the fluid
𝑄 = 𝜙 ⋅ 𝑄gas + (1 − 𝜙) ⋅ 𝑄𝑠 , (11)
field. It can be seen from (5) that the temperature 𝑇 appears
in the coefficient term 𝜙𝑀/𝑅𝑇 for the pressure temporal where in (8) (𝜌𝐶𝑝 )eq is the equivalent specific heat capacity of
derivatives 𝜕𝑃/𝜕𝑡 and also adjoins with the spatial gradient
of pressure term ∇𝑃. This results in the high nonlinearity of rock as a porous media saturated with air, 𝛼eq is the equivalent
the governing equation which brings numerical convergence heat conduction coefficient, and 𝑄 is the equivalent heat
and stability problem. It can also be seen from the first term source of saturated porous media. Equations (8)–(11) describe
on LHS that volumetric strain of the solid also couples the the governing equations of porous media and the fluid as the
stress field with fluid field. thermal field governing equations.

3.3. Mechanical Field: Equilibrium Condition. Since the rock

3.2. Thermal Field: Energy Conservation. According to the
is assumed to be saturated with air, the constitutive model
first law of thermodynamics, the heat added to a system
of thermal porous media should consider the contribution of
should be equal to the change of the internal energy of the
the pore pressure and temperature of the air as follows:
system and the work done in the system. By considering the
energy conservation of the skeleton of the rock as porous 𝜎 = (𝜆 + 𝜇) (∇ ⋅ u) + 𝜇∇u − 𝑏𝑃I − (3𝜆 + 2𝜇) 𝛽 (𝑇 − 𝑇0 ) I,
media, we can get (12)

𝜕𝑇 𝜕𝜀 where 𝜎 is the Cauchy stress tensor, 𝜆 and 𝜇 are Lamé

𝜌𝑠 𝐶𝑝,𝑠 + 𝑇𝐾𝑠 𝛽 𝑉 = ∇ ⋅ (𝛼𝑠 ∇𝑇) + 𝑄𝑠 , (6) coefficients of rock, u is the displacement vector, I is a second-
𝜕𝑡 𝜕𝑡 order identity tensor, 𝛽 is linear thermal expansion coefficient
of rock, and 𝑏 is Biot’s consolidation coefficient. The relation
where 𝜌𝑠 , 𝐶𝑝,𝑠 , 𝛼𝑠 , 𝑄𝑠 , 𝐾𝑠 , and 𝛽 are the density, specific heat, between 𝐸 and 𝜐 with respect to 𝜆 and 𝜇 is well known as
thermal conductivity, heat source, volumetric modulus, and
thermal expansion coefficient of the porous media. It can be 𝐸𝜐
𝜆= ,
seen that the energy transportation in the solid is through the (1 + 𝜐) (1 − 2𝜐)
thermal conduction as the first term on RHS. The volumetric (13)
𝐸
work induced by the temperature as the second term on LHS 𝜇= .
2 (1 + 𝜐)
also results in the dissipation of energy. From the energy
conservation of fluid, we can get The equilibrium of porous media is expressed as

∇𝜎 + f = 0, (14)
𝜕𝑇
𝜌gas 𝐶𝑝,gas + 𝜌gas 𝐶𝑝,gas q ⋅ ∇𝑇 + 𝑃 (∇ ⋅ q)
𝜕𝑡 (7) where f is the body force vector. Substituting (12) into (14)
leads to the governing equation of the stress field as
= ∇ ⋅ (𝛼gas ∇𝑇) + 𝑄gas ,
(𝜆 + 𝜇) ∇ (∇ ⋅ u) + 𝜇∇2 u − 𝑏 ⋅ ∇𝑃 − (3𝜆 + 2𝜇) 𝛽 ⋅ ∇𝑇 + f = 0.
where 𝜌gas , 𝐶𝑝,gas , 𝛼gas , and 𝑄gas are the density of air, heat (15)
specific of air, thermal conductivity, and heat source of the
fluid. The energy transportation of air includes the heat 3.4. The Governing Equation of Fluid Flow along Single Dis-
conduction as the first term on RHS, convection as the second crete Crack (Joint). The existence of crack will have influence
term on LHS, and external work of air as the third term on on all the three fields, especially the mechanics field and stress
LHS. For a single phase flow, we assume that the solid and air field. The thermal field will also change due to the crack;
are in thermal steady state, that is, the temperature of the fluid however, it will be relatively minor compared to stress and
and solid are equal at the interface. The heat exchange along fluid field.
Mathematical Problems in Engineering 5

Due to the existence of the crack in rock, the distribution Table 2: Material parameters and boundary conditions used in the
of the air is changing in the rock and consequently results in example.
the change of effective stress and thermal field. In the analysis,
Parameters
the joint is modeled as a channel, that is, internal boundaries
in rock. We have not considered the influence of crack over Young’s modulus 𝐸 of rock (GPa) 35
the distribution of thermal and mechanical fields. Therefore, Poisson’s ration of rock 𝜐 0.3
the stress field and thermal field over the crack are continu- Density of rock 𝜌𝑠 (kg/m3 ) 2800
ous. The coupling of the three fields for discontinuities is an Pore ratio 𝜙 1.0
interesting and challenging topic for the future. It is generally 1 × 10−9
Permeability coefficient 𝑘 (m2 )
recognized that a continuum based equivalent elastic porous
Permeability coefficient along crack 𝑘𝑓 (m2 ) —
media, such as Barentblatt and Gil’man [40] nonequilib-
rium two-phase immiscible flow mode, is not capable of Crack width 𝑑𝑓 (m) —
capturing the discontinuity or jump in the thermalfluidic Viscosity of air 𝜂 (Pa⋅s) 1.86 × 10−5
and mechanical field. In this type of approach, the mass Biot’s consolidation coefficient 𝑏 0.95
and energy is transferred with higher gradient compared Heat conduction of rock 𝛼𝑠 (W/m⋅K) 3
to base rock by modifying the material parameters. The
Heat conduction of air 𝛼𝑠 (W/m⋅K) 1000
crack is modeled as equivalent porous media and there exists
Heat specific of air under constant pressure 𝐶𝑝,gas
numerical instability of extra refined mesh near the crack. In 1000
(J/kg⋅K)
the present paper, the discrete crack model is adopted for a
single joint. The upper and lower surfaces of the crack are Heat specific of rock 𝐶𝑝,𝑠 (J/kg⋅K) 900
modeled as internal boundaries inside the domain and the Expansion coefficient 𝛽 (1/K) 1.0 × 10−5
fluid is allowed to flow between the boundaries. Darcy’s law Initial pressure 𝑃0 (atm) 1
for tangential flower along crack is defined as Initial temperature 𝑇0 (K) 286.15

𝑘𝑓 Air injection temperature 𝑇in (K) 296.65

q𝑓 = − 𝑑𝑓 (∇𝑇 𝑃) , (16)
𝜂
where q𝑓 is the flow rate along the unit length of a crack, 𝑘𝑓 is applied to the four edges of boundaries. Six points of interest
the permeability coefficient, 𝑑𝑓 is the width of the crack open- marked P1–P6 are sampled in the simulation results as shown
ing, and ∇𝑇 𝑃 is the pressure gradient along the length of the in Figure 2(b). The operational cycle of the cavern is 24 hours.
crack. The continuity of the fluid, that is, the mass conserva- In the initial condition of the cavern, the air is injected into
tion, between the two surfaces of a single joint is described by the design value of 5.5 MPa after 16 hours injection and
followed by 8 hours storage. After that, the air injection and
𝜕 (𝜌gas 𝜙𝑓 ) release follows a cycle of 8 hours storage, 4 hours storage, 4
𝑑𝑓 + ∇𝑇 ⋅ (𝜌gas q𝑓 ) = 𝑑𝑓 𝑄𝑚 . (17)
𝜕𝑡 hours release, and another 8 hours storage. In air compression
stage, the air is injected at the rate of 1.12 × 10−3 kg/(s ⋅ m3 )
It can be seen that the same variable of air pressure, namely, to the cavern at the temperate of 296.65 K (23.5∘ C). In the
𝑃, appears in the governing variable in both the governing energy recuperation stage, the air is released at the rate of
equation of the fluid flow along the fracture (joint) and the 2.24×10−3 kg/(s⋅m3 ). The time length of the present example
rock as the porous media. is 80 operational days. The material parameters used in this
case study are listed in Table 2.
4. Numerical Modelling of The pressure drop along the cavern ceiling with respect
the CAES System in Hard Rock to the time is shown in Figure 3. Due to the seepage of air
through the pores of rock, pressure drop is observed in the
The governing equations described in Section 3 for mass consecutive days of operation. With the mass conservation
balance, energy conservation, and equilibrium are discretised control, that is, same injection and production of air in terms
by the finite element method and coded to study the hard of mass, a substantial drop of pressure can be observed after
rock cavern performance for CAES. In this section, a hard 60 days of operation. At day 60, the minimum pressure inside
rock cavern under the designed operational conditions of the cavern is about 5 MPa lower than the desired operational
CAES without considering the existence of ground water is pressure. This is different from the observation of CAES
modeled. The aim of the numerical modelling is to investigate cavern in aquifer by Kim et al. [41]. It means that for hard
the heat loss, pressure distribution, and seepage of air near the rock cavern without water concealing, supplementary air
vicinity of cavern. injection is necessary to maintain operational pressure due
A rock cavern with a diameter of 5 m and embedded to air seepage.
with 100 m depth is modeled as shown in Figure 2(a). Initial It was also found that the influence of pressure fluctuation
ground stress for the equivalent stress state is modeled and over the rock diminishes at about 2.1 m distant to the internal
excavation step is modeled and the rock near the cavern is boundary of cavern (at P3 in Figure 2). It can also be seen
regarded as excavation disturbed zone. Standard atmospheric that the air seepage through the rock gradually converges as
pressure and ground temperature of 286.15 K (13∘ C) are shown in Figure 4. On day 1, the pressure drop along the
6 Mathematical Problems in Engineering

10
4.5 P3(0,4.6)
8
4
6
3.5
4
3 P2(0,2.67)
2 2.5
0 2 P1(0,2.5)
−2 1.5
−4 1 P5(2.67,0)
−6 0.5 P4(2.5,0) P6(4.6,0)
−8 0
−10 −0.5
−12 −10 −8 −6 −4 −2 0 2 4 6 8 10 12 −1 −0.5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6
(a) (b)

Figure 2: Model setting of the hard rock cavern. (a) Geometry and profile of cavern. (b) Points of interest near the cavern.

9 P1

5
P (MPa)

3 P2
2

1
P3
0
0 240 480 720 960 1200 1440
Time (hr)

Figure 3: Pressure drop of air after 60 operational days.

4
P (MPa)

0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
Distance to cavern ceiling (m)

Day 1 Day 40
Day 10 Day 50
Day 20 Day 60
Day 30

Figure 4: Pressure variation along the cavern ceiling from Day 1 to 60.
Mathematical Problems in Engineering 7

9
8
7
6
5

SP 1 and P (MPa)
4
3
2
1
0
−1
−2
0 240 480 720 960 1200 1440
Time (hr)

P1-SP1 Cavern pressure

P2-SP1 Pore pressure P2
P3-Sp1 Pore pressure P3

Figure 5: The maximum principle stress and pore pressure with respect to time.

21
20
19
18
17
16
T (∘ C)

15
14
13
12
11
10
9
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
Distance to cavern ceiling (m)

Day 1 Day 20
Day 3 Day 30
Day 5 Day 40
Day 10 Day 60
Day 15

Figure 6: The temperature along cavern ceiling with respect to the time from Day 1 to 60.

distance to the cavern is steep. The maximum pressure occurs of the fluid seepage and saturation in porous media, and
at the interface of air and rock. At day 20, the maximum shows the validity of the present THM models. It can also be
pressure is observed at about 0.2 m distant to the cavern seen that at P3 which is 2.1 m distant to the inner boundary
boundary. After 60 days of operation, the maximum pressure of cavern, the pore pressure vanishes showing the vanishing
point is shifted to about 0.5 m to the cavern boundary. The influence from cavern.
indication of the result to cavern design is that the validation For the thermal field, the temperature gradually con-
of rock strength should be conducted for certain range to the verges to the ground temperature of 13∘ C when it is distant
cavern boundary. The change of pressure and pore pressure from the cavern as shown in Figure 6. The temperature on
at different distances to the cavern with respect to the time is Day 1 at steady state is about 3 degrees than the injection
plotted in Figure 5. It can be seen that there is slight lagging temperature due the thermal conduction. It can be seen that
behind for the development of stress in rock with respect to the temperature decreases since the expansion of air is a heat
the pore pressure. This conforms to common phenomenon consumption process.
8 Mathematical Problems in Engineering

9 Air supplement-P1 9

8 8
Intact-P1
7 7

6 6

5 5
P (MPa)

P (MPa)
4 4
3 3
Mass conservation-P1
2 Crack-P1
2
1 1
0 0
0 100 200 300 400 500 600 0 100 200 300 400 500 600
Time (hr) Time (hr)

Figure 7: The pressure difference with and without considering the Figure 9: Comparison of pressure loss between intact and jointed
volumetric strain induced by thermal field. rock cavern.

12 28
26
10 24
22
8 Intact-P1
20
SP1 (MPa)

6 18
T (∘ C)

16
4
14
2 12
10 Crack-P1
0
8
−2 6
0 240 480 720 960 1200 1440 0 100 200 300 400 500 600
Time (hr) Time (hr)

Air supplement-P1 Mass conservation-P2 Figure 10: Comparison of heat loss between intact and jointed rock
Mass conservation-P1 Air supplement-P3 cavern.
Air supplement-P2 Mass conservation-P3

Figure 8: The comparison of the maximum principal stress in rock

between air supplement control and mass control during CAES more stable air pressure inside cavern but also induces the
operation. increased stress in rock.
The difference of pressure between intact rock cavern and
a carven with a single joint at the rock ceiling is compared as
shown in Figure 9. It can be seen that a single fracture with
To study the pressure control strategy of the system, settings in Table 2 results in the pressure loss of about 4 MPa
supplementary air condition is also modeled. By comparing after 600 hours operation. For the energy loss, the cracked
the results of air density at day 60 with respect to day 1, rock cavern has lost about 5 degrees of temperature as shown
the average density loss can be calculated which is about in Figure 10. It can also be seen that, cracked rock cavern
1.048 kg/m3 daily. Therefore, in the model, supplementary will have larger fluctuation compared to intact rock in each
air mass of 3.252% is injected. The results of cavern pressure operational cycle. The pores pressure distribution shows the
with and without supplementary injection are compared in seepage of air through the joint after 25 days of operation
Figure 7. It can be seen that with supplementary air injection, where the concentration takes place near the joint as shown in
the air pressure can be maintained between 5.5 and 8.5 MPa, Figure 11. The internal pressure of air through dissipation into
which is ideal for the operational requirement. The maximum rock induces the pore pressure where the maximum value
principle stresses at different points in rock are compared takes place near the joint of 2.5 MPa. In the present study, the
between mass conservation and air supplement conditions discontinuous stress field and seepage field are not considered
in Figure 8. The air supplement on one hand results in here.
Mathematical Problems in Engineering 9

Day 1 1.9569 Day 5 5.3703

12 10 5
1.5
10
8 4
8 1
6 3
6 0.5
4
4 0 2
2
2
−0.5 1
0 0
−1 −2 0
−2
−4 −1.5 −4 −1
−6 −2 −6 −2
−8 −8
−2.5 −3
−10 −10
−3
−10 −5 0 5 10 −3.013 −10 −5 0 5 10 −3.9016
(a)
Day 1 Day 5
5.5164 7.4485
10 8
5 7
8 6
6
6 5
4 4
4 4
3
2 2
3 1
0 2
0
−2 −1
2 −2
−4 0
−3
−6 −4
1 −5
−8 −2
−6
−10 −7
0
−10 −5 0 5 10 −0.0694 −8 −6 −4 −2 0 2 4 6 8 −3.6044
(b)

Figure 11: Comparison of the maximum principle stress in the rock cavern with and without joint on the ceiling. (a) Intact rock and (b) with
single joint on the ceiling.

5. Discussions in the present work. The crack propagation due to cyclic THM
actions is not considered. These two problems can be the
In this paper, we investigate the feasibility of utilizing hard topics for further study.
rock for compressed air energy storage by a coupled THM
model. The energy loss, stress distribution, and pore pressure
distribution during cyclic operational condition are studied. Conflict of Interests
It was found that mass control based CAES operation results
in energy loss and is not feasible in substantial hard rock The authors declare that there is no conflict of interests
cavern. Supplementary air injection is needed to maintain regarding the publication of this paper.
the required pressure level. For rock cavern with single
joint, higher fluctuation of pressure and temperatures are Acknowledgments
found which is unfavorable for CAES system stability. The
energy loss is significantly increased due to the crack and The authors acknowledge the support of the Natural Science
the principle. In the air injection modelling, the injection Foundation of China (NSFC 41130751 and 51278378), the
rate is based on the density loss of mass control condition National Basic Research Program of China (973 program:
after 60 days of operational condition. A constant air sup- 2011CB013800), the Opening fund of State Key Laboratory
plement of 3.25% is adopted. However, it is only an average of Geohazard Prevention and Geoenvironment Protection
means of injection while in the real seepage field and the (Chengdu University of Technology) SKLGP2011K044, the
supplementary injection should be dynamical to maintain National 863 programme (SS2012AA052501) the Shanghai
designed pressure value. Only a single static crack is studied Chenguang Talent program (12CG20), and program for
10 Mathematical Problems in Engineering

Changjiang Scholars and Innovative Research Team in Uni- [17] X. Zhuang, C. E. Augarde, and K. M. Mathisen, “Fracture mod-
versity (PCSIRT, IRT1029). Thanks to Mr Haiyang Wen for eling using meshless methods and levels sets in 3D: framework
his help with the postprocessing of the numerical results. and modeling,” International Journal for Numerical Methods in
Engineering, vol. 92, no. 11, pp. 969–998, 2012.
[18] X. Zhuang, C. Augarde, and S. Bordas, “Accurate fracture mod-
References elling using meshless methods, the visibility criterion and level
sets: Formulation and 2D modelling,” International Journal for
[1] M. Dopita and R. Williamson, “Clean energy Australia 2010,”
Numerical Methods in Engineering, vol. 86, no. 2, pp. 249–268,
Tech. Rep., Clean Energy Council, Australian Academy of Sci-
2011.
ences, 2010.
[19] N. Nguyen-Thanh, J. Muthu, X. Zhuang, and T. Rabczuk, “An
[2] I. Hadjipaschalis, A. Poullikkas, and V. Efthimiou, “Overview
adaptive three-dimensional RHT-splines formulation in linear
of current and future energy storage technologies for electric
elasto-statics and elasto-dynamics.,” Computational Mechanics,
power applications,” Renewable and Sustainable Energy Reviews,
2013.
vol. 13, no. 6-7, pp. 1513–1522, 2009.
[20] X. Zhuang, R. Huang, H. Zhu, H. Askes, and K. Mathisen,
[3] H. Ibrahim, R. Younès, A. Ilinca, M. Dimitrova, and J. Perron,
“A new and simple locking-free triangular thick plate element
“Study and design of a hybrid wind-diesel-compressed air
using independent shear degrees of freedom,” Finite Elements
energy storage system for remote areas,” Applied Energy, vol. 87,
in Analysis and Design, vol. 75, pp. 1–7, 2013.
no. 5, pp. 1749–1762, 2010.
[21] Y. Cai, X. Zhuang, and H. Zhu, “A generalized and efficient
[4] V. Vongmanee, “The renewable energy applications for uninter-
method for finite cover generation in the numerical manifold
ruptible power supply based on compressed air energy storage
method,” International Journal of Computational Methods, vol.
system,” in Proceedings of the IEEE Symposium on Industrial
10, no. 5, Article ID 1350028, 19 pages, 2013.
Electronics and Applications (ISIEA ’09), pp. 827–830, Kuala
Lumper, Malaysia, October 2009. [22] X. Zhuang, H. Zhu, and C. Augarde, “An improved meshless
Shepard and least square method possessing the delta property
[5] Y. Makarov, B. Yang, J. DeSteese et al., “Wide-area energy stor- and requiring no singular weight function,” Computational
age and management system to balance intermittent resources Mechanics, 2013.
in the Booneville power administration and California ISO con-
[23] N. Vu-Bac, H. Nguyen-Xuan, L. Chen et al., “A Phantom-node
trol areas,” Tech. Rep., Pacific Northwest National Laboratory,
method with edge-based strain smoothing for linear elastic
2008.
fracture mechanics,” Journal of Applied Mathematics, vol. 2013,
[6] V. de Biasi, “New solutions for energy storage and smart grid Article ID 978026, 12 pages, 2013.
load management,” Gas Turbine World, vol. 39, no. 2, pp. 22–26,
[24] Y. Jia, Y. Zhang, G. Xu, X. Zhuang, and T. Rabczuk, “Reproduc-
2009.
ing kernel triangular B-spline-based FEM for solving PDEs,”
[7] “Australian mines and mineral deposits,” Geoscience Australia, Computer Methods in Applied Mechanics and Engineering, vol.
2012. 267, pp. 342–358, 2013.
[8] I. Glendenning, “Long-term prospects for compressed air stor- [25] H. H. Zhu, X. Zhuang, Y. C. Cai, and G. Ma, “High rock slope
age,” Applied Energy, vol. 2, no. 1, pp. 39–56, 1976. stability analysis using the enriched meshless shepard and least
[9] F. Crotogino, K. U. Mohmeyer, and R. Scharf, “Huntorf CAES: squares method,” International Journal of Computational Meth-
more than 20 years of successful operation,” in Proceedings of ods, vol. 8, no. 2, pp. 209–228, 2011.
the Spring 2001 Meeting, Orlando, Fla, USA, April 2001. [26] Y. C. Cai, X. Zhuang, and C. Augarde, “A new partition of unity
[10] M. Raju and S. Kumar Khaitan, “Modeling and simulation of finite element free from the linear dependence problem and
compressed air storage in caverns: a case study of the Huntorf possessing the delta property,” Computer Methods in Applied
plant,” Applied Energy, vol. 89, no. 1, pp. 474–481, 2012. Mechanics and Engineering, vol. 199, no. 17–20, pp. 1036–1043,
[11] N. Michael, “Report on status of compressed air energy storage 2010.
technology,” in Proceedings of the Internal Meeting on CAES [27] X. Zhuang, C. Heaney, and C. Augarde, “On error control in
Project of National 863 Program in Inner Mongolia, Beijing, the element-free Galerkin method,” Engineering Analysis with
China, November 2011. Boundary Elements, vol. 36, no. 3, pp. 351–360, 2012.
[12] W. K. Song, D. W. Ryu, and Y. K. Lee, “Stability analysis of con- [28] X. Zhuang and C. Augarde, “Aspects of the use of orthogonal
crete plugs in a pilot cavern for compressed air energy storage,” basis functions in the element-free Galerkin method,” Interna-
in Proceedings of the 12th International Congress on Rock Mech- tional Journal for Numerical Methods in Engineering, vol. 81, no.
anics of the International Society for Rock Mechanics (ISRM ’11), 3, pp. 366–380, 2010.
pp. 1813–1816, Beijing, China, October 2011. [29] T. Rabczuk and T. Belytschko, “Cracking particles: a simplified
[13] A. Dayan, J. Flesh, and C. Saltiel, “Drying of a porous spherical meshfree method for arbitrary evolving cracks,” International
rock for compressed air energy storage,” International Journal of Journal for Numerical Methods in Engineering, vol. 61, no. 13, pp.
Heat and Mass Transfer, vol. 47, no. 19-20, pp. 4459–4468, 2004. 2316–2343, 2004.
[14] J. Levine, G. Martin, R. Moutoux, and F. Barnes, “large scale [30] T. Rabczuk, T. Belytschko, and S. P. Xiao, “Stable particle meth-
electrical energy storage in Colorado,” Tech. Rep., Colorado ods based on Lagrangian kernels,” Computer Methods in Applied
Energy Research Institute, 2007. Mechanics and Engineering, vol. 193, no. 12–14, pp. 1035–1063,
[15] S. Miller, “Mining megawatts from compressed air,” IEEE Spec- 2004.
trum, vol. 38, no. 8, pp. 27–28, 2001. [31] T. Rabczuk and G. Zi, “A meshfree method based on the local
[16] S. J. Bauer, “Underground aspects of underground compressed partition of unity for cohesive cracks,” Computational Mech-
air energy storage,” in Proceedings of the Compressed Air Energy anics, vol. 39, no. 6, pp. 743–760, 2007.
Storage Scoping Workshop (CAES ’08), Columbia University, [32] T. Rabczuk and T. Belytschko, “A three-dimensional large
2008. deformation meshfree method for arbitrary evolving cracks,”
Mathematical Problems in Engineering 11

Computer Methods in Applied Mechanics and Engineering, vol.

196, no. 29-30, pp. 2777–2799, 2007.
[33] T. Rabczuk, S. Bordas, and G. Zi, “On three-dimensional mod-
elling of crack growth using partition of unity methods,” Com-
puters and Structures, vol. 88, no. 23-24, pp. 1391–1411, 2010.
[34] T. Chau-Dinh, G. Zi, P. S. Lee, T. Rabczuk, and J. H. Song,
“Phantom-node method for shell models with arbitrary cracks,”
Computers and Structures, vol. 92-93, pp. 242–246, 2012.
[35] H. Talebi, C. Samaniego, E. Samaniego, and T. Rabczuk, “On
the numerical stability and mass-lumping schemes for explicit
enriched meshfree methods,” International Journal for Numeri-
cal Methods in Engineering, vol. 89, no. 8, pp. 1009–1027, 2012.
[36] T. Rabczuk and E. Samaniego, “Discontinuous modelling of
shear bands using adaptive meshfree methods,” Computer
Methods in Applied Mechanics and Engineering, vol. 197, no. 6–8,
pp. 641–658, 2008.
[37] F. Tong, L. Jing, and R. W. Zimmerman, “A fully coupled
thermo-hydro-mechanical model for simulating multiphase
flow, deformation and heat transfer in buffer material and rock
masses,” International Journal of Rock Mechanics and Mining
Sciences, vol. 47, no. 2, pp. 205–217, 2010.
[38] J. G. Wang, Y. Zhang, J. S. Liu, and B. Y. Zhang, “Numerical sim-
ulation of geofluid focusing and penetration due to hydraulic
fracture,” Journal of Geochemical Exploration, vol. 106, no. 1–3,
pp. 211–218, 2010.
[39] T. Rabczuk, R. Gracie, J. H. Song, and T. Belytschko, “Immersed
particle method for fluid-structure interaction,” International
Journal for Numerical Methods in Engineering, vol. 81, no. 1, pp.
48–71, 2010.
[40] G. I. Barenblatt and A. A. Gil’man, “Nonequilibrium counter-
flow capillary impregnation,” Journal of Engineering Physics and
Thermophysics, vol. 52, no. 3, pp. 335–339, 1987.
[41] H. M. Kim, J. Rutqvist, D. W. Ryu, B. H. Choi, C. Sunwoo, and
W. K. Song, “Exploring the concept of compressed air energy
storage (CAES) in lined rock caverns at shallow depth: a mod-
eling study of air tightness and energy balance,” Applied Energy,
vol. 92, pp. 653–667, 2012.
Hindawi Publishing Corporation
Mathematical Problems in Engineering
Volume 2014, Article ID 262494, 8 pages
http://dx.doi.org/10.1155/2014/262494

Research Article
Dynamic Fracture in Thin Shells Using Meshfree Method

Y. Shie
Harbin Engineering University, College of Mechanical Engineering, Harbin, China

Correspondence should be addressed to Y. Shie; [email protected]

Received 14 September 2013; Accepted 11 November 2013; Published 16 January 2014

Academic Editor: Timon Rabczuk

Copyright © 2014 Y. Shie. This is an open access article distributed under the Creative Commons Attribution License, which
permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

We present a meshfree approach to model dynamic fracture in thin structures. Material failure is modeled based on a stress-based
criterion and viscoplastic is used to describe the material behavior in the bulk material. Material fracture is simply modeled by
breaking bonds between neighboring particles. The method is applied to fracture of cylindrical thin structures under explosive
loading. The loading is modelled by a pressure-time history. Comparisons between the computational results and experimental
data illustrate the validity and robustness of the proposed method.

1. Introduction Lagrangian based finite element methods. They can handle

large deformation and also fracture in a natural manner [44–
Modeling dynamic fracture of thin-walled structures remains 51]. An excellent review of meshfree methods is given in
a challenging task in computational mechanics. Such appli- [52, 53].
cations are of major importance in civil engineering, aero- While there are many meshfree formulations for con-
nautical engineering, aerospace engineering, and mechanical tinua, there exist far less meshfree formulations for structures.
engineering. Thin structures are often modelled by shell the- A meshfree thin shell formulation based on KL theory has
ory. When shear effects can be neglected, the Kirchhoff-Love been presented by [54]. They employed the element-free
(KL) condition requires 𝐶1 continuity of the underlying dis- Galerkin (EFG) method. Krysl et al. did not consider fracture.
cretization. Effective formulations exploiting the higher order The first meshfree thin shell method for fracture was pro-
continuity of the meshfree methods have been exploited by posed by Rabczuk et al. [2]. This method was developed for
[1, 2] for the first time; see also the contribution by [3–5] in the linear and nonlinear materials and applied to static and
context of Isogeometric Analysis. The contribution in [1, 2] dynamic problems. Moreover, it was developed for a finite-
includes also fracture of the thin structure under dynamic strain framework. The crack was modelled by partition-of-
loading. In [6], a fully coupled fluid-structure interaction unity enrichment. Subsequently, [55–57] have simplified the
method for fracturing thin structure has been proposed. On treatment of cracks in thin shell and exploited an extrinsic
the other hand, there are many applications where shear basis [1]. However, the ill-posedness due to the enrichment
effects need to be accounted for. There are numerous finite scheme has not been addressed. Donning and Liu [58] used
element formulations based on Mindlin-Reissner theory; see, the reproducing kernel particle method (RKPM) [23] to
for example, the manuscripts by [7–21]. On the other hand, model beam and plate elements. This approach has later been
thin-walled structures can also be modelled by three-dimen- extended by [59]. Membranes and shells based on higher
sional continuum modeling. order approximations have been implemented by [60]. They
Meshfree methods [22–34] are promising alternatives to report the higher order functions that are necessary to
finite element methods for applications involving large defor- avoid locking. Wang and Chen [61] presented a locking-free
mations, fracture and fluid-structure interaction. Meshfree meshfree formulation and showed that the Kirchhoff mode in
method commonly exploits Lagrangian formulation [35–43] the Mindlin plate can be reproduced by second-order approx-
for both fluid and structure but does not suffer drawbacks of imations. Yagawa and Miyamura [62] combined a mixed
2 Mathematical Problems in Engineering

approach within a free mesh framework. They used discrete The discretization of the displacement field in the mesh-
Kirchhoff theory. Other approaches for thin structures based free approximation is given by
on 3D continuum approaches have been developed, for exam- 𝑛
ple, by the group of Li et al. [63]. uℎ (X) = ∑ 𝑁𝐽 (X) u𝐽 = Nu, (4)
As we want to keep the methodology applicable for thin 𝐽=1
structures with and without shear deformations, we will use a
3D continuum formulation that can be used also for real 3D where 𝑁𝐽 (X) denotes the meshfree shape functions and 𝑛
simulations. The method will be applied to fracturing cylin- denotes the number of particles. The EFG method [22] is
drical structures under explosive loading. As the interaction based on a moving least squares approximation. It can be
between the solid and the fluid is highly complex [64–66], we shown that the shape functions are expressed by
will use a pressure-time history for the loading. The loading
will be estimated by explosion simulations in rigid cylinders. N𝑇 (X) = p𝑇 (X) A−1 (X) PW (X) (5)
Material failure of the solid is realized by simply breaking with the moment matrix
bonds between neighboring particles once a certain threshold
is exceeded. A (X) = P (Y) 𝑊 (X) P𝑇 (Y) , (6)
The paper is structured as follows. We first present the
meshfree method, then briefly the constitutive model, and where the matrix P𝑇 (Y) contains the polynomial basis p. We
finally the fracture criterion. Then, we apply the method to employ linear basis polynomials to ensure that the method
fracturing thin cylinders. The numerical simulations will be fulfills the patch test. The matrix
compared to experimental data. The paper will be summa-
W (X) = diag {𝑊𝐼 (X − X𝐼 , ℎ) 𝑉𝐼 } , 𝐼 = 1, . . . , 𝑛, (7)
rized in the last section.
contains usually rational weighting functions 𝑊𝐼 (X − X𝐼 , ℎ).
2. Weak Form and Meshfree Discretization For computational efficiency, usually weighting functions
with compact support are chosen; ℎ is the dilation parameter
Let us consider the linear momentum equation in the weak determining the size of the domain of influence of a central
form. It is given in a total Lagrangian description of motion particle. We use the quartic B-spline function:
by finding the displacement field u for all 𝛿u such that the
variation in the energy is zero as 𝑊 (X − X𝐼 , ℎ) = 𝑤 (𝑠)

𝛿𝑊 = 𝛿𝑊int − 𝛿𝑊ext + 𝛿𝑊kin = 0 (1) 1 − 6𝑠2 + 8𝑠3 − 3𝑠4 𝑠≤1 (8)

={
0 𝑠>1
with
with 𝑠 = (X − X𝐼 )/2ℎ.
𝛿𝑊int = ∫ ∇𝑋 𝛿u : P 𝑑Ω0 The test and trial functions have the structure of (4). Sub-
Ω0 stituting them into the weak formulation leads to the follow-
ing equation:
𝛿𝑊ext = ∫ 𝛿u ⋅ t0 𝑑Γ0 + ∫ 󰜚0 𝛿u ⋅ b 𝑑Ω0 (2)
Γ0𝑡 Ω0 𝑛 𝑛
∑ 𝛿u𝐼 { ∑ − ∫ ∇𝑋 𝑁𝐼 (X) P 𝑑Ω0 + ∫ 𝑁𝐼 (X) b 𝑑Ω0
Ω0 Ω0
𝛿𝑊kin = ∫ 󰜚0 𝛿u ⋅ ü𝑑Ω0 , 𝐼=1 𝐽=1
Ω0

where 𝑊int and 𝑊ext are the internal and external energy, + ∫ 𝑁𝐼 (X) t0 𝑑Γ0 + ∫ 󰜚0 𝑁𝐼 (X) 𝑁𝐽 (X) u 𝑑Ω0 }
Γ0𝑡 Ω0
respectively, 𝑊kin denotes the kinetic energy, X are the mate-
rial coordinates, Ω0 is the domain in the initial configuration = 0.
and Γ0 is the boundary in the initial configuration, b denotes (9)
the body force, 󰜚0 is initial density, u is displacement, P is
the first Piola-Kirchhoff stress tensor, t0 is the applied trac- After some algebraic operations, the well-known equation of
tion, ∇𝑋 denotes spatial derivatives with respect to material motion is obtained as
coordinate, and superimposed dots denote material time ext int
M𝐼𝐽 ü
𝐽 = −f𝐼 + f𝐼 (10)
derivatives.
The boundary conditions are given by with
u = u, X ∈ Γ0𝑢 M𝐼𝐽 = ∫ 󰜚N𝐼 (X) N𝑇𝐽 (X) 𝑑Ω0
(3) Ω0
n0 ⋅ P = t0 = t0 , X ∈ Γ0𝑡
f𝐼ext = ∫ N𝑇𝐼 (X) t0 𝑑Γ0 + ∫ N𝑇𝐼 (X) b 𝑑Ω0 (11)
with boundaries Γ0𝑢 ⋃ Γ0𝑡 = Γ0 and Γ0𝑢 ⋂ Γ0𝑡 = 0, the Γ0𝑡 Ω0
index 𝑡 referring to traction boundaries and the index 𝑢
to displacement boundaries; n is the normal vector to the f𝐼int = ∫ ∇𝑋 N𝑇𝐼 (X) P 𝑑Ω0 .
traction boundary. Ω0
Mathematical Problems in Engineering 3

Prenotch (different lengths) 1.975 cm

89.6 cm

Figure 1: Implosively loaded tube with different notch length: 1.1 inches, 2.2 inches, and 3.3 inches; notch depth is 0.6 mm.

3. Material Model Softening in material due to temperature is accounted for by

varying material parameters
A viscoplastic constitutive model from Zhou et al. [67] is used
to describe the material behavior in the bulk. The material 𝐸 (𝑇) = 𝐸0 − 1.6 × 106 (𝑇 − 𝑇0 ) − 105 (𝑇 − 𝑇0 ) [Pa]
model in the bulk is of utmost importance for reliable predic-
tions of material and structural behavior. Thermal effects are ] = ]0 + 5 × 10−5 (𝑇 − 𝑇0 )
incorporated into the constitutive model as they are crucial in (18)
2
dynamic fracture simulations of thin structures as presented 𝜎0 (𝑇) = 𝜎0 − 1.5 × 103 (𝑇 − 𝑇0 ) [Pa]
in this paper. The rate equations can be expressed in terms of
the Jaumann rate of the Kirchhoff stress by 𝛼 (𝑇) = (2.2 + 0.0016 [𝑇 − 𝑇0 ]) × 10−5 [K−1 ] ,

̇ ,
𝜏∇ = C : (D − DV𝑝 − 𝛼𝑇I) (12) where 𝐸 and ] are Young’s modulus and the Poisson ratio at
temperature 𝑇.
We employ the tangent modulus approach to update the
where C is the fourth order elasticity tensor, D is the symmet-
thermal viscoplastic constitutive model. The details of the
ric part of the velocity gradient L, 𝜏∇ = 𝜏̇− W ⋅ 𝜏 − 𝜏 ⋅ W is the scheme can be found in the excellent manuscript by [68].
Jaumann rate of the Kirchhoff stress, with W being the anti- A stress-based criterion is employed to model fracture.
metric part of the velocity gradient, 𝛼 is the thermal expan- Connectivities between neighboring nodes are broken when
sion coefficient, and I is the second-order identity matrix. the maximum principal tensile strength is three times the
The viscoplastic overstress model here is based on von Mises strength 𝜎0 . More sophisticated models based, for example,
as on loss of material stability might be used in future research.
3𝜖 ̇
DV𝑝 = ( ) ̃s (13) 4. Results
2𝜎
We validate our computational approach by comparison with
with experimental results. Therefore, we compare our numerical
𝜏−1 3 results to the fracture experiments by [69] who studied the
̃s = s − a with s = I, 𝜎 = : ̃s, (14) failure mechanism of thin aluminum cylinders under detona-
3 tr (𝜏) 2̃s
tion loading. They found that the failure mechanism changes
with a being the back stress which is assumed to be zero here. with the length of the longitudinal precracks. Classical test
A power low governs the thermal viscoplastic flow by set-up of their experiments is illustrated in Figure 1. The
length of the initial cracks varies from 1󸀠󸀠 to 3󸀠󸀠 :
𝑚
𝜎 (i) 1󸀠󸀠 (short notch),
𝜖 ̇= 𝜖0̇( ) (15)
𝑔(𝜖, 𝑇) (ii) 2󸀠󸀠 (medium-length notch),
with (iii) 3󸀠󸀠 (long notch).
In the experiments, the cylinder is filled with a combustible
𝜖 𝑛 𝑇 − 𝑇0
𝑔 (𝜖, 𝑇) = 𝜎(1 + ) (1 − 𝛿 [exp ( ) − 1]) . (16) gas consisting of oxygen and ethylene. The pressure after
𝜖0 𝜅 ignition varies between 80 kPa and 180 kPa. After the gas is
ignited at the left end, a detonation wave travels through the
In (15) and (16), 𝜖0̇is a reference strain rate, 𝑚 is the rate cylinder. In the thin aluminum specimen, pressures close to
sensitivity parameter, 𝜎0 is the yield stress, 𝜖0 = 𝜎0 /𝐸 is the the Chapman-Jouguet (CJ) limit were measured. Its velocity
corresponding reference strain and 𝐸 is Young’s modulus, 𝑛 is is between 2300 m/s and 2400 m/s and the pressure values
the strain hardening exponent, 𝑇0 is a reference temperature, in the fully recreated CJ state range from 2.6 MPa to 6.1 MPa
and 𝛿 and 𝜅 are thermal softening parameters. The function (depending on the initial pressure).
𝑔(𝜖, 𝑇) is the stress-strain relation measured at quasistatic As it is computationally expensive to carry out a fully cou-
strain rate of 𝜖 ̇
at temperature 𝑇. The equivalent plastic strain pled FSI simulation as, for example, done in [6], we first study
𝜖 is defined as the pure detonation in a rigid tube needed to estimate the
𝑡 𝑡
pressure-time history conditions in the aluminum cylinder.
2
𝜖 = ∫ 𝜖 𝑑𝑡 = ∫ √ DV𝑝 : DV𝑝 𝑑𝑡. (17) Those simulations are also based on meshfree simulations
0 0 3 where the fluid is modelled with the element-free Galerkin
4 Mathematical Problems in Engineering

method. A single exothermic chemical reaction A A → B 6

with a progress variable 𝑌 corresponding to the mass fraction
ratio between the partial density of the reactant A and the
5
total density 󰜚, that is, 𝑌 = 󰜚A /󰜚, governs the detonation by a
modified Euler equation that contains additional inhomoge-
neous conservation laws: 4

Pressure (MPa)
̇ + ∇ ⋅ (𝑌󰜚u)̇ = Ψ.
(𝑌󰜚) (19)
3
The chemical reaction follows the Arrhenius law [70]
−𝐸A 󰜚
Ψ = −𝑘𝑌󰜚 exp ( ), (20) 2
𝑝
with 𝑝 being obtained from the following equation of state: 1
𝑝 = (𝛾 − 1) (󰜚𝑒 − 󰜚𝑌𝑞) , (21)
0
where the parameter 𝑞 denotes the heat release due to the 0 0.1 0.2 0.3 0.4 0.5 0.6
chemical reaction per unit mass [70]. The volume burn model Time (ms)
according to [71] is exploited in order to guarantee the correct
propagation speed and state in chemical equilibrium at all Coarse
Fine
points in the discretization. Figure 2 illustrates the pressure-
time history exemplary for one point close to the precrack. Figure 2: Pressure over the time of a particle close to the notch from
The pressure at the entire interior surface is monitored during a pure fluid simulation of gas detonation in rigid specimen; fine
the course of the simulation and subsequently applied as refers to a discretization with around 1,900,000 particles while the
pressure-time history to study the fracture behavior of thin coarse discretization refers to a discretization with around 600,000
aluminum cylinders. We are aware that effects due to the particles.
changing structure after deformation are not included in our
model. These effects might have a significant influence in par-
ticularly on fatal failure as the results below indicate. As stated
previously, a fully coupled FSI model might be inevitable in
this case.
For the fracture simulations, we employ discretizations
of different refinements ranging from 140,000 particles to
280,000 particles. The following material parameters are used
for all simulations: 𝐸 = 69 GPa, 󰜚 = 2719 kg/m3 , ] = 0.33,
𝑝
𝜎0 = 275 GPa, 𝜖0 = 0.001, 1/𝑛 = 0.07, and 1/𝑚 = 0.01. (a) Simulation; 𝑡 = 0.4 ms

4.1. Short Notch. Figure 3 shows the deformed cylinder for

a short notch. The crack grows straight in longitudinal direc-
(b) Experiment
tion. At a certain point, it curves in a 45-degree angle and con-
tinues to propagate in circumferential direction before it is Figure 3: Displaced configuration and effective stress of the
arrested. At the end of the simulation, the fracture encom- detonation-driven fracture of cylinder; notch length is 1󸀠󸀠 .
passes approximately 75% of the circumference. Similar
results were obtained in the experiments from Chao [69] sug-
gesting that effects due to fluid-structure interaction are not directions simultaneously and quickly turn into radial direc-
relevant. tions. The crack opening is more pronounced as for the short
notch though. However, the fracture does not encompass the
5. Medium-Length Notch circumference. We believe that FSI simulations might be able
to capture the failure mechanism better.
Different failure pattern is observed for the medium-length
notched cylinder. In the experiments, the fracture farther 6. Long Notch
from the detonation wave propagated first. After initial
straight crack propagation, the fracture branched in 2 circum- When the notch length is increased to 3󸀠󸀠 , the specimen was
ferential cracks separating the specimen in 2 pieces; see broken into three pieces (in the experiments). This behavior is
Figure 4. indeed captured well by our numerical simulations. Fractures
Our computational results also predict severe failure grow first in longitudinal direction again. Several branches
though we are not capable of capturing the complete separa- evolve leading finally to the failure illustrated in Figures 5 and
tion of the cylinder into two pieces. The fractures grow in two 6.
Mathematical Problems in Engineering 5

(a) 𝑡 = 0.241 ms (b) 𝑡 = 0.246 ms

(c) 𝑡 = 0.4 ms (d) Experiment

Figure 4: Displaced configuration and effective stress of the detonation-driven fracture of cylinder at different times; notch length is 2󸀠󸀠 .

(a) 𝑡 = 0.244 ms (b) 𝑡 = 0.253 ms

(c) 𝑡 = 0.253 ms (d) 𝑡 = 0.4 ms

Figure 5: Displaced configuration and effective stress of the detonation-driven fracture of cylinder at different times; notch length is 3󸀠󸀠 .
6 Mathematical Problems in Engineering

Conflict of Interests
The author declares that there is no conflict of interests
regarding the publication of this article.

References
[1] T. Rabczuk and P. Areias, “A meshfree thin shell for arbitrary
evolving cracks based on an extrinsic basis,” Computer Modeling
(a) 𝑡 = 0.4 ms
in Engineering and Sciences, vol. 16, no. 2, pp. 115–130, 2006.
[2] T. Rabczuk, P. M. A. Areias, and T. Belytschko, “A meshfree thin
shell method for non-linear dynamic fracture,” International
Journal for Numerical Methods in Engineering, vol. 72, no. 5, pp.
524–548, 2007.
[3] N. Nguyen-Thanh, J. Kiendl, H. Nguyen-Xuan, R. Wüchner, K.
U. Bletzinger, and Y. Bazilevs, “Rotation free isogeometric thin
shell analysis using PHT-splines,” Computer Methods in Applied
Mechanics and Engineering, vol. 200, no. 47-48, pp. 3410–3424,
2011.
[4] D. J. Benson, Y. Bazilevs, M.-C. Hsu, and T. J. R. Hughes, “A
large deformation, rotation-free, isogeometric shell,” Computer
(b) 𝑡 = 0.4 ms
Methods in Applied Mechanics and Engineering, vol. 200, no. 13–
16, pp. 1367–1378, 2011.
[5] N. Nguyen-Thanh, H. Nguyen-Xuan, S. P. A. Bordas, and T.
Rabczuk, “Isogeometric analysis using polynomial splines over
(c) Experiment hierarchical T-meshes for two-dimensional elastic solids,” Com-
puter Methods in Applied Mechanics and Engineering, vol. 200,
Figure 6: Displaced configuration and effective stress of the no. 21-22, pp. 1892–1908, 2011.
detonation-driven fracture of cylinder; notch length is 3󸀠󸀠 . [6] T. Rabczuk, R. Gracie, J.-H. Song, and T. Belytschko, “Immersed
particle method for fluid-structure interaction,” International
Journal for Numerical Methods in Engineering, vol. 81, no. 1, pp.
48–71, 2010.
[7] H. Nguyen-Xuan, T. Rabczuk, N. Nguyen-Thanh, T. Nguyen-
7. Conclusions Thoi, and S. Bordas, “A node-based smoothed finite element
method with stabilized discrete shear gap technique for analysis
In conclusion, we presented a simple and robust meshfree of Reissner-Mindlin plates,” Computational Mechanics, vol. 46,
method for dynamic fracture in thin structures. The method no. 12, pp. 679–701, 2010.
is based on a 3D continuum approach and a simple fracture [8] N. Nguyen-Thanh, T. Rabczuk, H. Nguyen-Xuan, and S. Bordas,
criterion is used to nucleate or propagate a crack. A viscoplas- “An alternative alpha finite element method with discrete shear
tic model is employed in the bulk. gap technique for analysis of isotropic Mindlin-Reissner plates,”
The method was applied to dynamic fracture in thin alu- Finite Elements in Analysis and Design, vol. 47, no. 5, pp. 519–535,
minum cylinders. The experiments from [69] were exploited 2011.
to validate our method. In those experiments, a shock wave [9] N. Valizadeh, S. Natarajan, O. A. Gonzalez-Estrada, T. Rabczuk,
initiated by an explosive mixture was propagating through T. Quoc Bui, and S. Bordas, “Nurbs-based finite element
analysis of functionally graded plates: static bending, vibration,
the aluminum cylinders. Different notch lengths were studied
buckling and flutter,” Composite Structures, vol. 99, pp. 309–326,
in their experiments, ranging from 1󸀠󸀠 to 3󸀠󸀠 . As it is compu- 2013.
tationally and extremely challenging to model the entire pro- [10] C. H. Thai, H. Nguyen-Xuan, N. Nguyen-Thanh, T.-H. Le, T.
cess, we first performed simulations of the explosion in rigid Nguyen-Thoi, and T. Rabczuk, “Static, free vibration, and buck-
tubes. The pressures in the rigid tubes were monitored during ling analysis of laminated composite Reissner-Mindlin plates
the course of the simulation and in a second step applied using NURBS-based isogeometric approach,” International
as boundary conditions to the thin aluminum cylinder, the Journal for Numerical Methods in Engineering, vol. 91, no. 6, pp.
specimen of interest. In other words: our simulations neglect 571–603, 2012.
effects due to fluid-structure interaction. Nevertheless, for the [11] T. Nguyen-Thoi, P. Phung-Van, T. Rabczuk, H. Nguyen-Xuan,
1󸀠󸀠 and 3󸀠󸀠 notched specimen, our numerical simulations were and C. Le-Van, “Free and forced vibration analysis using the
able to capture the appropriate failure mechanisms while for n-sided polygonal cell-based smoothed finite element method
(ncs-fem),” International Journal of Computational Methods,
the 2󸀠󸀠 , we observed discrepancies to the experimental results. vol. 10, no. 1, Article ID 13400082, 2013.
Future studies will therefore focus on the extension of our [12] P. Areias, T. Rabczuk, D. Dias-da-Costa, and E. B. Pires,
model to fluid-structure interaction. Moreover, we intend to “Implicit solutions with consistent additive and multiplicative
incorporate more advanced fracture models as proposed, for components,” Finite Elements in Analysis and Design, vol. 57, pp.
example, by [39, 72–75]. 15–31, 2012.
Mathematical Problems in Engineering 7

[13] P. Areias and T. Rabczuk, “Finite strain fracture of plates and [29] P. B. H. Le, T. Rabczuk, N. Mai-Duy, and T. Tran-Cong, “A
shells with configurational forces and edge rotation,” Interna- moving IRBFN-based integration-free meshless method,” Com-
tional Journal For Numerical Methods in Engineering, vol. 94, puter Modeling in Engineering & Sciences, vol. 61, no. 1, pp. 63–
pp. 1099–1122, 2013. 109, 2010.
[14] P. Areias, T. Rabczuk, and D. Dias da Costa, “Assumed-metric [30] H. Talebi, C. Samaniego, E. Samaniego, and T. Rabczuk, “On
spherically-interpolated quadrilateral shell element,” Finite Ele- the numerical stability and mass-lumping schemes for explicit
ments in Analysis and Design, vol. 66, pp. 53–67, 2013. enriched meshfree methods,” International Journal for Numeri-
[15] H. Nguyen-Xuan, T. Rabczuk, S. Bordas, and J. F. Debongnie, “A cal Methods in Engineering, vol. 89, no. 8, pp. 1009–1027, 2012.
smoothed finite element method for plate analysis,” Computer [31] Y. Cai, X. Zhuang, and H. Zhu, “A generalized and efficient
Methods in Applied Mechanics and Engineering, vol. 197, no. 13– method for finite cover generation in the numerical manifold
16, pp. 1184–1203, 2008. method,” International Journal of Computational Methods, vol.
[16] H. Nguyen-Xuan, T. Rabczuk, T. Nguyen-Thoi, T. N. Tran, and 10, no. 5, Article ID 1350028, 19 pages, 2013.
N. Nguyen-Thanh, “Computation of limit and shakedown loads [32] X. Zhuang, C. Augarde, and K. Mathisen, “Fracture modelling
using a node-based smoothed finite element method,” Interna- using meshless methods and level sets in 3D: framework and
tional Journal for Numerical Methods in Engineering, vol. 90, no. modelling,” International Journal For Numerical Methods in
3, pp. 287–310, 2012. Engineering, vol. 92, pp. 969–998, 2012.
[17] H. H. Phan-Dao, H. Nguyen-Xuan, C. Thai-Hoang, T. Nguyen- [33] H. Zhu, X. Zhuang, and Y. Cai, “High rock slope stability
Thoi, and T. Rabczuk, “An edge-based smoothed finite element analysis using the enriched meshless Shepard and least squares
method for analysis of laminated composite plates,” Interna- method,” International Journal of Computational Methods, vol.
tional Journal of Computational Methods, vol. 10, no. 1, Article 8, pp. 209–228, 2011.
ID 13400057, 2013.
[34] W. Zheng, X. Zhuang, and Y. Cai, “A new method for seismic
[18] C. Thai-Hoang, N. Nguyen-Thanh, H. Nguyen-Xuan, and T. stability analysis of anchored rock slope and its application in
Rabczuk, “An alternative alpha finite element method with anchor optimization,” Frontiers of Structural and Civil Engineer-
discrete shear gap technique for analysis of laminated composite ing, vol. 6, pp. 111–120, 2012.
plates,” Applied Mathematics and Computation, vol. 217, no. 17,
[35] T. Rabczuk and J. Eibl, “Simulation of high velocity concrete
pp. 7324–7348, 2011.
fragmentation using SPH/MLSPH,” International Journal for
[19] N. Nguyen-Thanh, T. Rabczuk, H. Nguyen-Xuan, and S. P. A. Numerical Methods in Engineering, vol. 56, no. 10, pp. 1421–1444,
Bordas, “A smoothed finite element method for shell analysis,” 2003.
Computer Methods in Applied Mechanics and Engineering, vol.
198, no. 2, pp. 165–177, 2008. [36] S. R. Idelsohn, E. Oñate, and F. Del Pin, “The particle finite
element method: a powerful tool to solve incompressible flows
[20] T. Chau-Dinh, G. Zi, P.-S. Lee, T. Rabczuk, and J.-H. Song,
with free-surfaces and breaking waves,” International Journal for
“Phantom-node method for shell models with arbitrary cracks,”
Numerical Methods in Engineering, vol. 61, no. 7, pp. 964–989,
Computers and Structures, vol. 92-93, pp. 242–246, 2012.
2004.
[21] P. Baiz, S. Natarajan, S. Bordas, P. Kerfriden, and T. Rabczuk,
[37] S. Hao, W. K. Liu, and D. Qian, “Localization-induced band and
“Linear buckling analysis of cracked plates by sfem and xfem
cohesive model,” Journal of Applied Mechanics, vol. 67, no. 4, pp.
(smxfem),” Journal of Mechanics of Materials and Structures, vol.
803–812, 2000.
6, no. 9-10, pp. 1213–1238, 2011.
[22] T. Belytschko, Y. Y. Lu, and L. Gu, “Element-free Galerkin [38] T. Rabczuk and T. Belytschko, “Adaptivity for structured mesh-
methods,” International Journal for Numerical Methods in Engi- free particle methods in 2D and 3D,” International Journal for
neering, vol. 37, no. 2, pp. 229–256, 1994. Numerical Methods in Engineering, vol. 63, no. 11, pp. 1559–1582,
2005.
[23] W. K. Liu, S. Jun, and Y. F. Zhang, “Reproducing kernel par-
ticle methods,” International Journal for Numerical Methods in [39] T. Rabczuk, T. Belytschko, and S. P. Xiao, “Stable particle
Engineering, vol. 20, pp. 1081–1106, 1995. methods based on Lagrangian kernels,” Computer Methods in
Applied Mechanics and Engineering, vol. 193, no. 12–14, pp. 1035–
[24] T. Rabczuk and E. Samaniego, “Discontinuous modelling of
1063, 2004.
shear bands using adaptive meshfree methods,” Computer
Methods in Applied Mechanics and Engineering, vol. 197, no. 6–8, [40] T. Rabczuk, P. M. A. Areias, and T. Belytschko, “A simplified
pp. 641–658, 2008. mesh-free method for shear bands with cohesive surfaces,”
[25] T. Rabczuk and P. M. A. Areias, “A new approach for mod- International Journal for Numerical Methods in Engineering, vol.
elling slip lines in geological materials with cohesive models,” 69, no. 5, pp. 993–1021, 2007.
International Journal for Numerical and Analytical Methods in [41] S. Hao, W. K. Liu, P. A. Klein, and A. J. Rosakis, “Modeling and
Geomechanics, vol. 30, no. 11, pp. 1159–1172, 2006. simulation of intersonic crack growth,” International Journal of
[26] W. K. Liu, S. Hao, T. Belytschko, S. Li, and C. T. Chang, “Multiple Solids and Structures, vol. 41, no. 7, pp. 1773–1799, 2004.
scale meshfree methods for damage fracture and localization,” [42] S. Hao and W. K. Liu, “Moving particle finite element method
Computational Materials Science, vol. 16, no. 1–4, pp. 197–205, with superconvergence: nodal integration formulation and
1999. applications,” Computer Methods in Applied Mechanics and
[27] M. R. Moosavi, F. Delfanian, A. Khelil, and T. Rabczuk, Engineering, vol. 195, no. 44–47, pp. 6059–6072, 2006.
“Orthogonal meshless finite volume method in elastodynam- [43] T. Rabczuk, S. P. Xiao, and M. Sauer, “Coupling of mesh-free
ics,” Thin-Walled Structures, vol. 49, no. 9, pp. 1171–1177, 2011. methods with finite elements: basic concepts and test results,”
[28] P. B. H. Le, T. Rabczuk, N. Mai-Duy, and T. Tran-Cong, “A Communications in Numerical Methods in Engineering with
moving IRBFN-based Galerkin meshless method,” Computer Biomedical Applications, vol. 22, no. 10, pp. 1031–1065, 2006.
Modeling in Engineering & Sciences, vol. 66, no. 1, pp. 25–52, [44] F. Del Pin, S. Idelsohn, E. Oñate, and R. Aubry, “The
2010. ALE/Lagrangian particle finite element method: a new
8 Mathematical Problems in Engineering

approach to computation of free-surface flows and fluid-object Galerkin method,” International Journal for Numerical Methods
interactions,” Computers and Fluids, vol. 36, no. 1, pp. 27–38, in Engineering, vol. 52, no. 7, pp. 705–725, 2001.
2007. [60] H. Noguchi, T. Kawashima, and T. Miyamura, “Element free
[45] S. R. Idelsohn, J. Marti, A. Limache, and E. Oñate, “Unified analyses of shell and spatial structures,” International Journal for
Lagrangian formulation for elastic solids and incompressible Numerical Methods in Engineering, vol. 47, no. 6, pp. 1215–1240,
fluids: application to fluid-structure interaction problems via 2000.
the PFEM original,” Computer Methods in Applied Mechanics [61] D. Wang and J.-S. Chen, “A locking-free meshfree curved beam
and Engineering, vol. 197, no. 19-20, pp. 1762–1776, 2008. formulation with the stabilized conforming nodal integration,”
[46] S. R. Idelsohn, E. Oñate, F. Del Pin, and N. Calvo, “Fluid- Computational Mechanics, vol. 39, no. 1, pp. 83–90, 2006.
structure interaction using the particle finite element method,” [62] G. Yagawa and T. Miyamura, “Three-node triangular shell ele-
Computer Methods in Applied Mechanics and Engineering, vol. ment using mixed formulation and its implementation by free
195, no. 17-18, pp. 2100–2123, 2006. mesh method,” Computers and Structures, vol. 83, no. 25-26, pp.
[47] S. Idelsohn, M. Mier-Torrecilla, and E. Oñate, “Multi-fluid flows 2066–2076, 2005.
with the particle finite element method,” Computer Methods [63] S. Li, W. Hao, and W. K. Liu, “Numerical simulations of large
in Applied Mechanics and Engineering, vol. 198, no. 33–36, pp. deformation of thin shell structures using meshfree methods,”
2750–2767, 2009. Computational Mechanics, vol. 25, no. 2, pp. 102–116, 2000.
[48] T. Rabczuk, J. Eibl, and L. Stempniewski, “Numerical anal- [64] T. Rabczuk, J. Y. Kim, E. Samaniego, and T. Belytschko,
ysis of high speed concrete fragmentation using a meshfree “Homogenization of sandwich structures,” International Journal
Lagrangian method,” Engineering Fracture Mechanics, vol. 71, for Numerical Methods in Engineering, vol. 61, no. 7, pp. 1009–
pp. 547–556, 2004. 1027, 2004.
[49] T. Rabczuk and J. Eibl, “Numerical analysis of prestressed [65] T. Rabczuk, E. Samaniego, and T. Belytschko, “Simplified model
concrete beams using a coupled element free Galerkin/finite for predicting impulsive loads on submerged structures to
element method,” International Journal of Solids and Structures, account for fluid-structure interaction,” International Journal of
vol. 41, no. 3-4, pp. 1061–1080, 2004. Impact Engineering, vol. 34, no. 2, pp. 163–177, 2007.
[50] E. Oñate, S. Idelsohn, O. C. Zienkiewicz, R. L. Taylor, and C. [66] T. Rabczuk and J. Eibl, “Modelling dynamic failure of concrete
Sacco, “A stabilized finite point method for analysis of fluid with meshfree methods,” International Journal of Impact Engi-
mechanics problems,” Computer Methods in Applied Mechanics neering, vol. 32, no. 11, pp. 1878–1897, 2006.
and Engineering, vol. 139, no. 1–4, pp. 315–346, 1996. [67] M. Zhou, G. Ravichandran, and A. J. Rosakis, “Dynamically
[51] T. Nguyen-Thoi, H. C. Vu-Do, T. Rabczuk, and H. Nguyen- propagating shear bands in impact-loaded prenotched plates.
Xuan, “A node-based smoothed finite element method (NS- II: numerical simulations,” Journal of the Mechanics and Physics
FEM) for upper bound solution to visco-elastoplastic analysis of Solids, vol. 44, no. 6, pp. 1007–1032, 1996.
of solids using triangular and tetrahedral meshes,” Computer [68] D. Peirce, C. F. Shih, and A. Needleman, “A tangent modulus
Methods in Applied Mechanics and Engineering, vol. 199, no. 45– method for rate dependent solids,” Computers and Structures,
48, pp. 3005–2027, 2010. vol. 18, no. 5, pp. 875–887, 1984.
[52] V. P. Nguyen, T. Rabczuk, S. Bordas, and M. Duflot, “Meshless [69] T. W. Chao, Gaseous detonation-driven fracture of tubes [Ph.D.
methods: a review and computer implementation aspects,” thesis], California Institute of Technology, 2004.
Mathematics and Computers in Simulation, vol. 79, no. 3, pp.
[70] W. Fickett and W. C. Davis, Detonation, University of California
763–813, 2008.
Press, Berkeley, Calif, USA, 1979.
[53] S. R. Idelsohn and E. Oñate, “To mesh or not to mesh. That is
[71] C. L. Mader, Numerical Modeling of Detonations, University of
the question. . .,” Computer Methods in Applied Mechanics and
California Press , Berkeley, Calif, USA, 1979.
Engineering, vol. 195, no. 37–40, pp. 4681–4696, 2006.
[72] G. Zi, T. Rabczuk, and W. Wall, “Extended meshfree methods
[54] P. Krysl and T. Belytschko, “Analysis of thin shells by
without branch enrichment for cohesive cracks,” Computational
the element-free Galerkin method,” International Journal for
Mechanics, vol. 40, no. 2, pp. 367–382, 2007.
Numerical Methods in Engineering, vol. 33, no. 20–22, pp. 3057–
3078, 1996. [73] T. Rabczuk and T. Belytschko, “Application of particle methods
to static fracture of reinforced concrete structures,” Interna-
[55] T. Rabczuk and T. Belytschko, “Cracking particlses: a simplified
tional Journal of Fracture, vol. 137, no. 1–4, pp. 19–49, 2006.
meshfree method for arbitrary evolving cracks,” International
Journal for Numerical Methods in Engineering, vol. 61, no. 13, pp. [74] T. Rabczuk and T. Belytschko, “A three-dimensional large
2316–2343, 2004. deformation meshfree method for arbitrary evolving cracks,”
Computer Methods in Applied Mechanics and Engineering, vol.
[56] T. Rabczuk, J.-H. Song, and T. Belytschko, “Simulations of insta- 196, no. 29-30, pp. 2777–2799, 2007.
bility in dynamic fracture by the cracking particles method,”
Engineering Fracture Mechanics, vol. 76, no. 6, pp. 730–741, 2009. [75] T. Rabczuk and G. Zi, “A meshfree method based on the local
partition of unity for cohesive cracks,” Computational Mechan-
[57] T. Rabczuk, G. Zi, A. Gerstenberger, and W. A. Wall, “A new ics, vol. 39, no. 6, pp. 743–760, 2007.
crack tip element for the phantom-node method with arbitrary
cohesive cracks,” International Journal for Numerical Methods in
Engineering, vol. 75, no. 5, pp. 577–599, 2008.
[58] B. M. Donning and W. K. Liu, “Meshless methods for shear-
deformable beams and plates,” Computer Methods in Applied
Mechanics and Engineering, vol. 152, no. 1-2, pp. 47–71, 1998.
[59] W. Kanok-Nukulchai, W. Barry, K. Saran-Yasoontorn, and P. H.
Bouillard, “On elimination of shear locking in the element-free
Hindawi Publishing Corporation
Mathematical Problems in Engineering
Volume 2013, Article ID 959806, 13 pages
http://dx.doi.org/10.1155/2013/959806

Research Article
Evaluating the Applicability of Fracture Criteria to
Predict the Crack Evolution Path of Dolomite Based on
SCB Experiments and FEM

Cunbao Li,1,2 Lingzhi Xie,1,2 Li Ren,1,2 Heping Xie,1,2 and Jun Wang1,2
1
Key Laboratory of Energy Engineering Safety and Disaster Mechanics of Ministry of Education, Sichuan University,
Chengdu 610065, China
2
School of Architecture and Environment, Sichuan University, Chengdu 610065, China

Correspondence should be addressed to Lingzhi Xie; [email protected]

Received 2 October 2013; Revised 17 November 2013; Accepted 18 November 2013

Academic Editor: Goangseup Zi

Copyright © 2013 Cunbao Li et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Mixed mode fracture tests are conducted under various initial loading combinations of mode I and mode II (from pure mode I to
pure mode II) on semicircular bend (SCB) specimens of dolomite rock. Damage zones are observed behind the fracture surfaces of
the broken samples. Scanning electron microscope images of the fracture surfaces are used to study the failure manner. Using the
conventional remesh method based on the finite element method (FEM), several widely accepted fracture criteria are employed
to theoretically predict the fracture paths. These criteria include the maximum tangential stress criterion, minimum strain energy
density criterion, maximum energy release rate criterion, maximum dilatational strain energy density criterion, and the distortional
strain energy density criterion. The applicability of the five fracture criteria is examined. The results show that none of the criteria
are successful in predicting the crack trajectories of the predominately mode II cracks; the differences among the predicted results
of the crack growth paths are negligible for each crack inclined angle. The effect of Poisson’s ratio on the fracture criteria is also
investigated and the results show that the predicted crack trajectories are not sensitive to Poisson’s ratio.

1. Introduction [2]. However, it is quite difficult to predict the curvilinear

fracture path theoretically because the crack growth direction
Extensive development in rock fracture mechanics research has to be determined by the present mixed-mode loading
has covered many diverse areas including blasting, hydraulic condition which corresponds to a specified position on the
fracturing and in situ stress determination, mechanical frag- curvilinear path. Further investigations into fracture initia-
mentation, rock slope analysis, earthquake mechanics, earth- tion angle and fracture propagation path under mixed-mode
quake prediction, plate tectonics, magmatic intrusions, hot loading will improve the understanding of mixed mode
dry rock geothermal energy extraction, fluid transport prop- fracture behavior.
erties of fracturing rock masses, propagating oceanic rifts, Many researches have been conducted to describe crack
crevasse penetration, and other glaciological problems [1]. As initiation and crack propagation under mixed-mode load-
a result of complicated geological processes, numerous flaws, ing. For example, Jia et al. [3] conducted a series of mixed-
cracks, and faults exist in rock masses. Fracture mechanics is a mode fracture experiments using center cracked circular disc
useful technique for investigating the initiation and propaga- (CCCD) specimens to observe the fracture path of concrete
tion of an individual crack or cracks in geological materials. under mixed mode I/II loading; Xeidakis et al. [4] inves-
Due to the complexity of in situ stresses, most fractures in tigated the trajectory of crack growth for marble using
these applications occur under a combination of opening and antisymmetric three-point bend specimens under different
sliding deformation (I/II mixed mode). Under mixed mode mixed mode loading conditions; Bobet and Einstein [5]
loading, a fracture usually propagates along a curvilinear path studied the fracture initiation, propagation, and, ultimately,
2 Mathematical Problems in Engineering

coalescence (which plays an important role in the behavior of 2. Crack Propagation Criteria
brittle materials) by loading prefractured specimens of gyp-
sum, used as the rock model material, in uniaxial and biaxial Most theoretical fracture criteria consider that when an
compression; Chen et al. [6] conducted a set of fracture appropriate characteristic quantity like stress, strain, and
experiments on anisotropic shale by using CCCD specimens energy density energy release rate, in front of the crack tip
under different mixed mode loading conditions for studying reaches its critical value which is deemed to a material con-
the fracture initiation path and also employed the numerical stant, then the crack begins propagating. For plane problem,
technique of boundary element method to simulate the the singular stresses around a crack tip in polar coordinate
observed fracture paths. Vásárhelyi and Bobet [7] investi- can be expressed as
gated the crack initiation stress, direction, and propagation
1 𝐾 𝜃 𝐾 𝜃
of newly generated cracks for both open and closed frac- 𝜎𝑟𝑟 = [ I cos (3 − cos 𝜃)− II sin (1 − 3 cos 𝜃)] ,
tures; Wong et al. [8] numerically investigated the crack √2𝜋𝑟 2 2 2 2
initiation, propagation, and coalescence of rock specimens 1 𝐾 𝜃 3𝐾II 𝜃
containing preexisting crack-like flaws under compression 𝜎𝜃𝜃 = [ I cos (1 + cos 𝜃)− sin (1 + cos 𝜃)] ,
√2𝜋𝑟 2 2 2 2
by using Rock Failure Process Analysis (RFPA2D ); Al-shayea
[9] studied the crack initiation angle and subsequent crack 1 𝐾 𝜃 𝐾 𝜃
𝜎𝑟𝜃 = [ I sin (1 + cos 𝜃)− II cos (1 − 3 cos 𝜃)] ,
propagation path experimentally for notched Brazilian disk √2𝜋𝑟 2 2 2 2
specimens of limestone; Liu et al. [10] used a numerical (1)
method for modeling the mixed mode fracture process of
heterogeneous rocks using different test samples including where 𝑟 and 𝜃 are the polar coordinates with the origin at
CCCD specimens. More recently, Aliha et al. [11] investigated the crack tip and 𝐾I and 𝐾II are mode I and mode II stress
the mixed mode I/II fracture initiation angle and the crack intensity factors, respectively. And the stress field in Cartesian
growth trajectory of soft rock (Guiting Limestone) by using coordinates is defined as
both CCCD and edge cracked SCB specimens of various sizes. 1 𝜃 𝜃 3𝜃
In order to theoretically estimate the crack initiation 𝜎𝑥𝑥 = [𝐾I cos (1 − sin sin )
√2𝜋𝑟 2 2 2
angle, several widely accepted fracture criteria have been pro-
posed, for example, the maximum tangential stress criterion 𝜃 𝜃 3𝜃
(𝜎𝜃 -criterion) [12], minimum strain energy density criterion −𝐾II sin (2 + cos cos )] ,
2 2 2
(𝑆min -criterion) [13], maximum energy release rate criterion
(𝐺max -criterion) [14–18], maximum dilatational strain energy 1 𝜃 𝜃 3𝜃
𝜎𝑦𝑦 = [𝐾I cos (1 + sin sin )
density criterion (𝑁𝑇-criterion) [19], and distortional strain √2𝜋𝑟 2 2 2
energy density criterion (𝑌-criterion) [20]. From a material
mechanics view, it is reasonable to suggest that mixed mode 𝜃 𝜃 3𝜃
+𝐾II sin cos cos ] ,
brittle fracture occurs on reaching a certain level of critical 2 2 2 (2)
energy, tensile stress, tensile strain, shear stress, shear strain,
1 𝜃 𝜃 3𝜃
or another physical variable near the crack tip. However, it 𝜎𝑥𝑦 = [𝐾I cos sin sin
seems that none of these variables can predict the fracture √2𝜋𝑟 2 2 2
strengths of predominately mode II fractures accurately 𝜃 𝜃 3𝜃
because each criterion will predict a corresponding invariant +𝐾II cos (1 − sin sin )] ,
2 2 2
fracture toughness ratio 𝐾II𝑐 /𝐾I𝑐 [2] which cannot be applied
to a wide range of engineering materials. In fracture mechan- 𝜎𝑧𝑧 = 0, for plane stress,
ics, it is also important to predict where the crack will go,
that is, the initiation angle and the growth path of the crack. 𝜎𝑧𝑧 = ] (𝜎𝑥𝑥 + 𝜎𝑦𝑦 ) , for plane strain,
Therefore, the applicability of these fracture criteria on the
crack growth direction and crack evolution path should be where ] is Poisson’s ratio.
discussed.
In this study, a series of mixed-mode fracture tests are 2.1. 𝜎𝜃 -Criterion. Erdogan and Sih [12] propose the maxi-
conducted on dolomite rock using SCB specimens. Addition- mum tangential stress criterion, which considers that a crack
ally, scanning electron microscope (SEM) images are used to would grow in the direction in which the 𝜎𝜃𝜃 is the maximum;
study the failure manner. Moreover, using the conventional the direction should meet the following conditions:
remesh method based on the FEM, the crack growth trajec-
tories for various mixed mode loading conditions (from pure 𝜕𝜎𝜃𝜃 𝜕2 𝜎𝜃𝜃
= 0, < 0. (3)
mode I to mode II) are theoretically (by the 𝜎𝜃 -criterion, 𝑆min - 𝜕𝜃 𝜕𝜃2
criterion, 𝐺max -criterion, 𝑁𝑇-criterion, and 𝑌-criterion) and
numerically predicted. A detailed investigation on the appli- 2.2. 𝑆min -Criterion. Sih [13] formulates the specific stain
cability of the five fracture criteria to the prediction of crack energy density factor 𝑆 at the radius 𝑟0 in term of the energy
evolution paths under mixed mode loading conditions has as follows:
been presented. Finally, the effect of Poisson’s ratio on the
fracture criteria is also investigated. 𝑆 = 𝑎11 𝐾I2 + 2𝑎12 𝐾I 𝐾II + 𝑎22 𝐾II2 , (4)
Mathematical Problems in Engineering 3

where The elastic-plastic boundary then can be obtained by

employing a suitable yield condition, and in 𝑁𝑇-criterion, the
1 Von Mises yield function is adopted. Then, one can have
𝑎11 = (1 + cos 𝜃) (𝜅 − cos 𝜃) ,
16𝜋𝜇
1 𝜎𝑦2
𝑎12 = sin 𝜃 (2 cos 𝜃 − 𝜅 + 1) , 𝑁𝑑,0 = , (11)
16𝜋𝜇 6𝜇
1 where 𝜎𝑦 is the critical material yield strength. 𝑁𝑇-criterion
𝑎22 = [(𝜅 + 1) (1 − cos 𝜃) + (1 + cos 𝜃) (3 cos 𝜃 − 1)] ,
16𝜋𝜇 postulates that a crack propagates in the direction of maxi-
(5) mum dilatational strain energy 𝑁V,max along the contour of
constant distortional strain energy 𝑁𝑑,0 when the distance
where 𝜇 is the shear modulus, 𝜅 = 3 − 4] for the plane strain from crack tip to the elastic-plastic boundary 𝑟𝑝 in the growth
problem, and 𝜅 = (3−4])/(1+]) for the plane stress problem. direction reached a critical value 𝑟𝑝𝑐𝑟 ; that is,
The 𝑆min -criterion postulates that the crack initiates in the
direction of minimum strain energy density along a constant
𝜕𝑁V|𝑟𝑝 𝜕2 𝑁V|𝑟𝑝
radius around the crack tip when 𝑆 reaches a critical value; = 0, < 0, (12)
that is, 𝜕𝜃 𝜕𝜃2

𝜕𝑆 𝜕2 𝑆 where 𝑟𝑝𝑐𝑟 = (1 − 2V)2 𝐾𝐼𝑐2

/(2𝜋𝜎𝑦2 ) for plane strain and 𝑟𝑝𝑐𝑟 =
= 0, > 0. (6) 2 2
𝜕𝜃 𝜕𝜃2 𝐾𝐼𝑐 /(2𝜋𝜎𝑦 ) for plane stress case.

2.3. 𝐺max -Criterion. Several forms [15–18] of 𝐺max -criterion

2.5. 𝑌-Criterion. 𝑌-criterion [20] also divides the total energy
have been formulated and only the expressions presented by
density into two components, the dilatational stain energy
Hussain et al. [15] will be reviewed here. The energy release
density 𝑌V and the distortional strain energy density 𝑌𝑑 , as
rate is given as
𝑌𝑑 = 𝑁𝑑 , 𝑌V = 𝑁V . (13)
4 1 2
𝜋 + 𝜃 −𝜃/𝜋
𝐺= ( 2
)( )
𝐸 3 + cos 𝜃 𝜋−𝜃 𝑌-criterion presumes that a crack growth occurs in the
2 direction of minimum distortional strain energy density
×[(1+3cos 𝜃) 𝐾I2 −4 sin (2𝜃) 𝐾I 𝐾II +(9−5cos2 𝜃) 𝐾II2 ] ,
along the constant core region boundary 𝑟0 when the mini-
(7) mum distortional strain energy density 𝑌𝑑,𝜃0 reached a critical
value 𝑌𝑑,0 , and it can expressed as
where 𝐸 is Young’s modulus. According to the assumptions
made by Palaniswamy and Knauss [14], the direction of crack
𝜕𝑌𝑑|𝑟0 𝜕2 𝑌𝑑|𝑟0
propagation is the orientation in which the maximum rate of = 0, > 0. (14)
elastic energy release rate once exceeds a critical value; that 𝜕𝜃 𝜕𝜃2
is,
3. Mixed Mode Fracture Tests on
𝜕𝐺 𝜕2 𝐺 Dolomite Rock
= 0, < 0. (8)
𝜕𝜃 𝜕𝜃2
3.1. Rock Properties. The experimental material, dolomite
2.4. 𝑁𝑇-Criterion. In 𝑁𝑇-criterion [19] the total strain rock, was taken from a quarry in Fangshan County, Beijing.
energy density 𝑁 (equivalent to 𝑆 in (4)) is split into the Visual inspections indicated that the rock was isotropic,
distortional stain energy density 𝑁𝑑 and dilatational stain homogenous, and white in color. Its structure was contin-
energy density 𝑁V ; that is, uous and very tight. The X-ray diffraction (XRD) analysis
(Figure 1) showed that the rock was composed of dolomite
𝑁 = 𝑁𝑑 + 𝑁V , (9) (CaMg(CO3 )2 ) and quartz (SiO2 ). No pores or microcracks
were visible using SEM with ×50 magnification (Figure 2),
where therefore, the porosity of this rock can be neglected. SEM
images also identified the material as a microcrystalline-
1 textured, fine grained rock. Its dry density was about
𝑁𝑑 =
6𝜇 2.59 g/mm3 . In order to get the mechanical properties of
2 2 2 2
dolomite rock, 4 groups of uniaxial compression tests and
×(𝜎𝑥𝑥 + 𝜎𝑦𝑦 + 𝜎𝑧𝑧 −𝜎𝑥𝑥 𝜎𝑦𝑦 − 𝜎𝑦𝑦 𝜎𝑧𝑧 − 𝜎𝑧𝑧 𝜎𝑥𝑥 + 3𝜎𝑥𝑦 ), 4 groups of Brazilian tests were conducted. The mechanical
properties of the rock included a uniaxial unconfined com-
1 2 𝐸
𝑁V = (𝜎 + 𝜎𝑦𝑦 + 𝜎𝑧𝑧 ) , 𝐾= . pressive strength of 135 MPa, a tensile strength of 3.0 MPa,
18𝐾 𝑥𝑥 3 (1 − 2]) Young’s modulus of elasticity of 50 GPa, and Poisson’s ratio
(10) of 0.25.
4 Mathematical Problems in Engineering

2500
Relative intensity (%)

2000
1500
1000
500 𝜃0

0 R 𝛼
10 20 30 40 50 60 70 80 90 a
Angle (∘ )
Dolomite
Minrecordite 2s
Quartz

Figure 1: XRD results for dolomite rock.

Figure 3: Geometrical characteristics of specimens and loading

setup.

1.0

0.8

0.6
Figure 2: SEM image for dolomite rock with ×50 magnifications.
Load (N)

0.4
3.2. Sample Preparation. The SCB specimens [21] were
selected for mixed mode fracture tests on dolomite rock.
Figure 3 shows the geometry of the SCB specimen and the 0.2
loading setup of the experiments. SCB specimen was a half
disc of radius 𝑅 and thickness 𝑡. The edge crack was of
length 𝑎 and angle 𝛼 relative to the loading direction. The 0.0
specimen was loaded by a vertical load 𝑃 under three-point 0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14
bending and the distance between the bottom supports was Displacement (mm)
2𝑠. When 𝛼 was zero, the specimen was subjected to pure
Figure 4: Typical plot of the applied load against the load-point
mode I (opening mode) independent of 𝑎/𝑅 and 𝑠/𝑅 [22]. By
displacement.
increasing the angle 𝛼, the contribution of mode II (or sliding
mode) to the crack deformation increased. Pure mode II took
place at a specific angle 𝛼 depending on 𝑎/𝑅 and 𝑠/𝑅 [22].
To avoid any possible boundary effects, the specimens a test machine with a capacity of 10 kN. The loading rate
were taken from the center of the rocks. The SCB specimens was set to 0.1 mm/min for all the experiments. Figure 4
had a radius of 25 mm and were 15 mm in thickness. Values represents a typical plot of the applied load against load-point
of 𝑎/𝑅 = 0.4 and 𝑠/𝑅 = 0.4 were selected for all the displacement. It can be seen that after a nonlinear stage the
specimens. The cracks were created using a 0.2 mm thick load-displacement curve gives a linear relationship up to the
diamond impregnated wire saw and the crack length 𝑎 was maximum load. The dolomite specimens, including the crack,
10 mm. According to [22] and numerical analysis, for the remain stable without any visible changes during the loading
selected specimen configuration and loading condition, at an process until the load reaches a critical value. At that point
inclination angle of 34.8∘ the crack was pure mode II crack. the load-displacement curve drops suddenly, which is the
Therefore, the crack inclination angles of 𝛼 = 0∘ (pure mode characteristic of brittle fracture; the original crack propagates
I), 10∘ , 20∘ , 30∘ , 34.8∘ (pure mode II), and 40∘ (compression- rapidly toward to the point where the load is applied. When
shear mode) were selected. For each crack angle, four samples the crack reaches the edge of the specimen, the specimen
were prepared. breaks into two pieces (Figure 5). The fracture surfaces can
reveal some useful information about the failure mechanism.
3.3. Fracture Tests. Each SCB specimen was located inside Therefore, SEM was used to conduct a microscale investiga-
a three-point bending fixture (Figure 3) and then loaded by tion on the typical fracture surfaces.
Mathematical Problems in Engineering 5

Table 1: Failure loads 𝑃𝑐𝑟 obtained from mixed mode fracture tests.

Specimen code Crack inclined Failure load

angle 𝛼 (∘ ) (𝑁)
D-1 (pure mode I) 0 832
D-2 (pure mode I) 0 857
𝛼 = 0∘ 𝛼 = 10∘ 𝛼 = 20∘
D-3 (pure mode I) 0 846
D-4 (pure mode I) 0 836
D-5 10 925
𝛼 = 30∘ 𝛼 = 34.8∘ 𝛼 = 40∘
D-6 10 910
D-7 10 892
D-8 10 945
D-9 20 960
D-10 20 810 Figure 5: SCB specimens after failure.
D-11 20 929
D-12 20 901
D-13 30 1002
D-14 30 936
D-15 30 997
D-16 30 993
D-17 (pure mode II) 34.8 1001
D-18 (pure mode II) 34.8 1035
D-19 (pure mode II) 34.8 1210
D-20 (pure mode II) 34.8 890
D-21 40 956
D-22 40 1256
D-23 40 1254 Figure 6: Size of particles around the crack tip (𝛼 = 30∘ ).
D-24 40 1042

and the mineral composition on the crack initiation, either

3.4. Test Results and Analysis. The failure loads 𝑃𝑐𝑟 obtained qualitatively or quantitatively.
from the mixed mode fracture tests are listed in Table 1. The Dolomite is a brittle rock with a relatively low hardness.
average values of 𝑃𝑐𝑟 for 𝛼 = 0∘ , 10∘ , 20∘ , 30∘ , 34.8∘ , and 40∘ are After the failure of specimens, an obvious damaged zone can
840 N, 918 N, 930 N, 982 N, 1034 N, and 1127 N, respectively. be found for each specimen, as illustrated in Figure 7. The
The failure load increases gradually as the proportion of mode damage zone is narrow relative to the specimen. The color
II deformation increases. of the damaged zone is different from the surrounding area;
Crack initiation includes two components: the crack its color is much whiter obviously. Regardless of the position
initiation angle and the crack initiation point. According to of crack tip, a high stress zone must exist around the crack
linear elastic fracture mechanics (LEFM), the crack initiation tip and lead to damage. Fonseka et al. [23] has confirmed
point is usually expected to be the crack tip of the original this view using the technology of SEM and acoustic emis-
crack. However, for specimens with 𝛼 = 30∘ and 34.8∘ (pure sion (AE). But this is a macroscopic experiment that may
mode II), the crack initiation point is not from the crack tip, as demonstrate directly that the emergence of a macroscopic
shown in Figure 5. Generally, only when the crack inclination crack is the result of damage that develops from microcracks
angle 𝛼 is very large, usually bigger than the values of pure and microvoids.
mode II or when the crack tip is obtuse seriously, the crack SEM image was used to examine the failure manner to
initiation point may not originate from the crack tip. How- better understand the fracture mechanism and the evolution
ever, the two cases in the tests do not fall into either of these of the damage in dolomite rock. The fracture faces near the
two categories. In order to better explain such phenomenon, original crack tip for the cases of 𝛼 = 0∘ , 10∘ , 20∘ , 30∘ , 34.8∘ ,
the fracture surface near the crack tip has been observed with and 40∘ were observed by SEM; the fracture faces produced by
SEM. It is found that the mineral grain near the crack tip the Brazilian test were also studied for comparison. Figure 8
is much larger than that in the surrounding area (Figure 6), shows SEM images of crack surface of the dolomite SCB
which may lead to this unexpected failure behavior. The specimens and Brazilian test with ×100 magnification.
other reason may be that the specimen contains quartz, In the case of 𝛼 = 0∘ (Figure 8(a)), the cracked surface
which possesses a much higher strength than dolomite. It is is coarse which was mainly caused by intergranular failure.
difficult to evaluate the influence of the size of the particles Compared with the fracture surface of the broken sample
6 Mathematical Problems in Engineering

the loading point, the crack extension of each step became

larger. Generally, simulating each model required at least
21 steps. The following procedure gave more details for
Damaged predicting the crack evolution path using the five fracture
zone
criteria:

(1) calculating 𝐾I and 𝐾II for the existing crack,

(2) using each criterion to evaluate the corresponding
crack growth direction,
Figure 7: Visible damaged zone of a specimen.
(3) remodeling the crack with a small extension along the
direction which was calculated in step 2,
(4) repeating steps 1–3 for the new crack growth.
from a Brazilian test (Figure 8(g)), the two faces are very
similar, indicating that the specimen is destroyed by splitting The loading conditions were considered similar to those
and stretching. The SEM image for 𝛼 = 10∘ (Figure 8(b)) of the fracture tests; for 𝛼 = 0∘ , 10∘ , 20∘ , 30∘ , 34.8∘ , and 40∘ ,
shows that the cracked surface is also produced primarily the loads were 840 N, 918 N, 930 N, 982 N, 1034 N, and 1127 N,
by splitting, tensile stress is the main factor, and only few of respectively. ABAQUS provides a domain integral method to
the crystalline particles are sheared off. There is no evidence compute the stress intensity factors directly for a crack under
of crushing or friction. The micrograph for the case of 𝛼 = mixed-mode loading. To produce the square root singular of
20∘ illustrates that the fractured surface is similar to that the stress/strain field, the singular elements were considered
of 𝛼 = 10∘ image. The notable difference is the increment in the first ring of elements surrounding the crack tip. In
of transgranular damage, revealing that the impact of shear order to ensure a smooth curve which was the boundary
stress is increasing, although it is still small relative to the of domain integral used to calculate stress intensity factors,
effect of splitting. There is still no evidence of crushing in 8 rings of quadratic quadrilateral elements surrounding the
the specimen of 𝛼 = 20∘ . In the SEM image for 𝛼 = 30∘ the singular elements were meshed by the meshing technique of
transgranular damage becomes more apparent; the fractured sweep. Other regions of the numerical model were meshed by
surface is produced by a combination of splitting and sliding the meshing methods of sweep or free, and the element type
shear. When the angles are 𝛼 = 34.8∘ , and 40∘ , the SEM was also quadratic quadrilateral elements. A total of about
images show obvious crushing between the boundary of the 5000 eight-node plane strain elements (CPE8) were used for
crystals. For 𝛼 = 34.8∘ , the failure is due to shear stress that is each model. Figure 9 represents a typical finite element mesh
consistent with LEFM. For the specimens of 𝛼 = 40∘ , crushing used for numerical analysis. The alterations of 𝐾I and 𝐾II with
is more serious. The analysis indicates that when the angle the mesh density are shown in Figure 10. It is observed that
increases from pure mode I to pure mode II, the effect of the the values of 𝐾I and 𝐾II are not sensitive to mesh density. The
shear stress grows, but there is no crushing or abrasion on the mesh density adopted in this research is reasonable. Using
cracked surface, which is different from the results proposed the method described above, the crack initiation angle and
in previous studies [9]. Moreover, when the angle reaches crack propagation path for different crack inclined angles
pure mode II or continues to increase (compression-shear were analyzed. The experimental and numerical results were
mode), the signs of crushing become apparent, with many compared to evaluate the applicability of the fracture criteria
rock fragments observed on the fractured surface. to predict the crack evolution path of dolomite rock.
The impact of Poisson’s ratio on the crack initial direction
and crack propagation path was also considered. Poisson’s
4. Numerical Analysis and Comparison ratio of the experimental dolomite SCB specimens was 0.25.
Based on (1)–(14), the crack initiation direction and growth Thus numerical models with Poisson’s ratio of 0.1, 0.25, and
path can be derived. Equations (1)–(14) are functions of the 0.4 were simulated.
stress intensity factors and various material constants. By
using the conventional remesh method based on FEM, (1)– 4.1. Crack Initiation Angle. The crack initiation angle 𝜃0 is
(14) were employed to theoretically predict the fracture paths. the direction in which the crack propagates from the original
No dynamic effects were considered in the crack propagation crack (Figure 3). For different 𝛼, the value of 𝜃0 is calculated
process. The finite element code ABAQUS and the mathe- by each fracture criteria introduced in Section 2. Table 2
matical software MATLAB were combined to evaluate the summarizes the crack initiation angle from simulated results.
direction of crack growth. Stress intensity factors calculated Generally, the point possesses maximum stress, and max-
by ABAQUS would output to a specified file, and fracture imum energy is at different locations and in a different direc-
criterion was written as a program by MATLAB. The program tion, so the crack initiation angle predicted by different frac-
would read stress intensity factors from the specified file and ture criteria should be different. When mode I deformation
calculate the angle in which the crack would propagate. In is dominant (𝛼 < 10∘ ), the difference in crack initiation angle
each step, the simulation model was remodeled with a small is very small. The maximum difference between the predicted
crack extension of 0.5–1 mm. In the initial steps, the extension angles is 3.8∘ (between the 𝑆min -criterion and 𝐺max -criterion),
was relatively small. As the crack tip gradually approached and there is little difference between the 𝜎𝜃 -criterion and
Mathematical Problems in Engineering 7

(a) 𝛼 = 0∘ (b) 𝛼 = 10∘

(c) 𝛼 = 20∘ (d) 𝛼 = 30∘

(e) 𝛼 = 34.8∘ (f) 𝛼 = 40∘

(g) Failure surface of Brazilian splitting test

Figure 8: SEM images of fractured surfaces of SCB and Brazilian test specimens with ×100 magnification.
8 Mathematical Problems in Engineering

Table 2: Summary of the crack initiation angle 𝜃0 of simulation for different crack inclination angle 𝛼 (∘ ).

𝛼 𝜎𝜃 -criterion 𝑆min -criterion 𝐺max -criterion 𝑁𝑇-criterion 𝑌-criterion

0 0 0 0 0 0
10 29.1 26.0 29.8 29.3 29.0
20 48.8 44.7 51.6 52.8 51.1
30 63.7 66.5 68.5 81.3 74.8
34.8 70.5 80.4 75.2 98.9 88.4
40 70.5 80.4 75.2 98.9 88.4

P changes from 70.5 to 98.9∘ , a difference of almost 30∘ . The

proportion of the mode II deformation plays an important
role in the predicted results of different fracture criteria.
When 𝛼 = 40∘ , the crack belongs to compression crack. From
the perspective of physics, the materials on both sides of the
closed crack surface cannot penetrate into each other, and
therefore, such cracks are pure model II cracks according
to the LEFM. Thus, the initiation angle for the specimen of
𝛼 = 40∘ would be the same as the case of 𝛼 = 34.8∘ , as shown
2s
in the last two rows of Table 2. However, this is inconsistent
Figure 9: Typical finite element mesh used for numerical analysis.
with the experimental results. For 𝛼 being larger than the
angle of pure mode II, the fracture criteria are not applicable.

4.4 8.5
4.2. Crack Propagation Path. The dolomite specimens’ failure
4.2 8.0
paths under various combination loadings of mode I and
mode II are plotted in Figure 11. When the sample is subjected
4.0 7.5 to pure mode I, the crack propagates straight along the line
KII (×103 Pa·m−1/2 )
KI (×105 Pa·m−1/2 )

of the original crack. As the effect of mode II deformation

3.8 7.0 increases, the crack evolution path becomes curved. When
𝛼 = 40∘ , the crack path is the most curving. Figure 12 shows
3.6 6.5 the simulation results of crack trajectories for each inclined
angle specimens using the five fracture criteria.
3.4 6.0 When 𝛼 = 0∘ , the crack extension angle is 0∘ for each
fracture criterion (Figure 12(a)). As 𝛼 increases from 10∘ to
3.2 5.5
40∘ , the simulated crack trajectories are very similar for all the
fracture criteria. To further examine the crack growth paths
3.0 5.0
0.0 0.5 1.0 1.5 2.0 2.5 calculated by different criteria, the plots of the crack trajec-
Mesh density tories near the original crack tip are enlarged. At the initial
part of crack growth path, the difference between the various
KI fracture criteria is relatively large. This is due to the initial
KII
angle of the crack varies between different fracture criteria,
Figure 10: Alterations of 𝐾I and 𝐾II with mesh density (the mesh but after a few steps of crack growth, the difference between
density adopted (Figure 9 in the paper) for numerical simulation in the predicted crack trajectories of the various fracture criteria
this research is assumed to be 1; the mesh density of other numerical become smaller and can be neglected. Therefore, any one of
models is 0.5, 0.75, 1.5, and 2 times as dense as Figure 9). the five simulated results can be used for comparison with the
experimental results.
After the initial stage of crack propagating, mode I defor-
𝑌-criterion. Hence, when the angle 𝛼 is small, that is, the mation becomes dominant. Taking 𝛼 = 34.8∘ (pure mode II)
proportion of the mode II deformation is relatively small, the as an example, after four steps of simulation, the stress inten-
difference of the simulated results by the five fracture criteria sity factors 𝐾I and 𝐾II calculated by the 𝑆min -criterion are
is negligible. As 𝛼 increases, the difference in initiation 422200 and 6676 Pa⋅m−1/2 , respectively; for the 𝑁𝑇-criterion,
angle between the various fracture criteria becomes apparent. 𝑌-criterion, 𝐺max -criterion, and 𝜎𝜃 -criterion, the values of 𝐾I
When 𝛼 = 20∘ , the difference between different criterions and 𝐾II are 437400 and 10280 Pa⋅m−1/2 , 430400 Pa⋅m−1/2 and
becomes larger. The predicted angle for the 𝐺max -criterion 8607 Pa⋅m−1/2 , 427500 and 3186 Pa⋅m−1/2 , and 429500 and
is 51.6∘ and for the 𝑆min -criterion is 44.7∘ . The difference is 5892 Pa⋅m−1/2 , respectively. It is obvious that the proportion
about 7∘ . For 𝛼 = 30∘ , the biggest difference between different of mode II deformation is relatively small compared with
criterions is nearly 18∘ . When 𝛼 = 34.8∘ , the predicted value that of mode I. Based on the discussion in Section 4.1,
Mathematical Problems in Engineering 9

(a) 𝛼 = 0∘ (b) 𝛼 = 10∘ (c) 𝛼 = 20∘

(d) 𝛼 = 30∘ (e) 𝛼 = 34.8∘ (f) 𝛼 = 40∘

Figure 11: Crack path of fractured SCB specimens with different crack inclination angle 𝛼.

(a) 𝛼 = 0∘ (b) 𝛼 = 10∘

(c) 𝛼 = 20∘ (d) 𝛼 = 30∘

(e) 𝛼 = 34.8∘ (f) 𝛼 = 40∘

NT-criterion Smin -criterion NT-criterion Smin -criterion
Y-criterion Gmax -criterion Y-criterion Gmax -criterion
𝜎𝜃 -criterion 𝜎𝜃 -criterion

Figure 12: Simulated results of crack trajectories. The red, green, blue, magenta, and cyan lines represent the paths predicted by the 𝜎𝜃 -
criterion, 𝑆min -criterion, 𝑌-criterion, 𝐺max -criterion, and 𝑁𝑇-criterion, respectively.
10 Mathematical Problems in Engineering

Table 3: Summary of the crack initiation angles 𝜃0 for different Poisson’s ratio (∘ ).

𝑆min -criterion 𝑁𝑇-criterion 𝑌-criterion

𝛼
𝑣 = 0.1 𝑣 = 0.25 𝑣 = 0.4 𝑣 = 0.1 𝑣 = 0.25 𝑣 = 0.4 𝑣 = 0.1 𝑣 = 0.25 𝑣 = 0.4
0 0 0 0 0 0 0 0 0 0
10 22.6 26.0 28.6 29.3 29.3 29.3 28.6 29.0 29.3
20 37.4 44.7 49.6 52.8 52.8 52.8 49.4 51.1 51.9
30 59.5 66.5 72.6 81.3 81.3 81.3 72.4 74.8 76.1
34.8 74.5 80.4 86.2 98.9 98.9 98.9 85.9 88.4 89.8
40 74.5 80.4 86.2 98.9 98.9 98.9 85.9 88.4 89.8

the damaged zone. Therefore, when mode I deformation is

predominant, the results predicted by the five fracture criteria
are reliable. In current work, for predominately mode II
cracks (𝛼 = 30∘ , 34.8∘ , and 40∘ (Figures 14(d)–14(f))), the
predicted crack paths locate outside the damaged zones and
are not consistent with the experimental results. For a pre-
dominately mode II deformation, all the fracture criteria fail
to predict the crack path.

4.4. Influence of Poisson’s Ratio. Equations (6), (12), and (14)

Crack evolution path
Original artificial crack
indicate that Poisson’s ratio V has an influence on the numer-
ical analysis results. To assess the effect of V quantitatively,
Figure 13: Crack trajectories for the different angles of inclination three different values (V = 0.1, 0.25, and 0.4) were selected
predicted by one fracture criterion (take 𝑁𝑇-criterion as an exam- for the numerical analysis. Table 3 summarizes the crack
ple). initiation angles 𝜃0 for different Poisson’s ratio calculated by
the 𝑆min -criterion, 𝑁𝑇-criterion, and 𝑌-criterion. The last
two rows of Table 3 are the same and the reasons have been
when the proportion of mode I deformation is dominant, explained in Section 4.1. For the 𝑁𝑇-criterion, the crack initi-
the estimated crack initiation angles of all five criteria are ation angles do not change with Poisson’s ratio, indicating that
similar, which leads to the simulated crack trajectories of the Poisson’s ratio has no effect on the 𝑁𝑇-criterion prediction of
various fracture criteria are similar when the new crack tip the crack initiation angle. For the 𝑆min -criterion, as Poisson’s
moves slightly away from the original crack tip. Examining ratio increases, the predicted crack initiation angle becomes
the predicted crack trajectories for the different angles of larger. When 𝛼 < 10∘ , the difference is relatively small;
inclination using one fracture criterion, and 𝑁𝑇-criterion is the maximum difference is less than 7∘ . For 𝛼 > 10∘ , the
taken as an example (Figure 13); it is found that when the difference between the results of V = 0.1 and V = 0.4 is
crack extends to the loading point, the crack paths gradually more than 11∘ . Poisson’s ratio plays an important role in the
move close to each other; the same reasons mentioned above predicted angle of the 𝑆min -criterion. The effect of Poisson’s
can explain this phenomenon. ratio on the 𝑌-criterion is similar to that of the 𝑆min -criterion,
but the effect is much smaller. For V = 0.1 and V = 0.4,
4.3. Comparison between the Experimental and Simulated the biggest difference for the 𝑌-criterion is less than 5∘ .
Results. The crack trajectories predicted based on the five Poisson’s ratio has the greatest influence on the 𝑆min -criterion,
fracture criteria are very similar. Therefore, for simplification, followed by the 𝑌-criterion, and no effect on the 𝑁𝑇-cri-
only the simulated crack evolution paths of the 𝑆min -criterion terion. Furthermore, the crack trajectories are not sensitive
were selected to compare with the experimental results. to Poisson’s ratio for the three fracture criteria. The simulated
Figure 14 describes the experimental results (crack evolution results of the crack evolution path are similar to those of
paths and damaged zones) and the calculated trajectories Figures 12(a)–12(e).
based on the 𝑆min -criterion.
For cracks of pure mode I, the test result coincides with 5. Conclusions
the simulation result of the fracture criteria. When 𝛼 = 10∘ ,
the simulation paths are also in good agreement with the Using SCB specimens of dolomite rock, mixed mode brittle
experimental results. For 𝛼 = 20∘ , the difference between the fracture was studied by experiments. In order to explain the
experimental result and the simulation result becomes larger observed failure behavior, the SEM technique was used to
but still acceptable. It is interesting that all the predicted crack study the microfailure mechanism. Furthermore, by using
trajectories of 𝛼 = 0∘ , 10∘ , and 20∘ are in the damaged zone. the conventional remesh method based on FEM, the crack
That is, the appearance of crack is the consequence of damage growth trajectories for various mixed mode loading con-
evolution and the crack propagation path should locate in ditions, from pure mode I to mode II, were theoretically
Mathematical Problems in Engineering 11

Experimental
result
Simulated result Experimental result
Simulated result

(a) 𝛼 = 0∘ (b) 𝛼 = 10∘

Simulated Simulated
result Experimental result
Experimental result result

(c) 𝛼 = 20∘ (d) 𝛼 = 30∘

Simulated Simulated
Experimental result Experimental result
result result

(e) 𝛼 = 34.8∘ (f) 𝛼 = 40∘

Figure 14: Experimental crack propagated paths, damaged zones, and crack trajectories calculated by 𝑆min -criterion (the gray zone, red line,
and green line represent the damaged zone, numerical result, and experimental result, resp.).

(by the 𝜎𝜃 -criterion, 𝑆min -criterion, 𝐺max -criterion, 𝑁𝑇- the five fracture criteria was small. As the proportion
criterion, and 𝑌-criterion) and numerically predicted. The of mode II deformation increased, the difference
applicability of the five criteria was examined. Finally, an became larger. Mode II deformation had a consider-
investigation on the effect of Poisson’s ratio on fracture cri- able impact on the crack initiation angle.
teria was also presented. Based on the current work, the fol-
lowing conclusions can be obtained. (5) The crack trajectories predicted by the five fracture
criteria were very similar and the differences among
(1) The dolomite specimens fracture in a brittle manner: the prediction paths were negligible.
as the proportion of mode II deformation increased,
the average failure load increased. (6) The predicted crack trajectories were located in the
(2) Damaged zones were observed on both sides of the damaged zones for predominately mode I cracks
crack. It was a macroscopic experiment that could and in such cases, the fracture criteria were suitable;
demonstrate that the emergence of macroscopic when mode II deformation was predominant, all five
cracks was the result of the evolution of damage. fracture criteria failed to predict the crack trajectories.
(3) When the crack inclination angle increased, the effect
(7) The crack initiation angle predicted by the 𝑌-criterion
of the shear stress increased but no crushing or
and 𝑆min -criterion was sensitive to Poisson’ ratio,
abrasion was observed on the cracked surface. When
particularly the 𝑆min -criterion; however, the crack
the crack inclined angle was 34.8∘ or 40∘ , evidences of
trajectories were not sensitive to Poisson’s ratio for
crushing between the cracked faces began to appear.
these two fracture criteria. For the 𝑁𝑇-criterion, the
(4) When mode I deformation was dominant, the differ- crack initiation angle and the crack trajectories were
ence between the crack initiation angles predicted by independent of Poisson’s ratio.
12 Mathematical Problems in Engineering

Nomenclature 𝜎𝜃 -criterion: Maximum tangential stress criterion

𝛼: Crack inclination angle
AE: Acoustic emission 𝜃0 : Crack initiation angle.
𝑎: Crack length
𝑎/𝑅 : Crack length ratio
𝑎11 , 𝑎12 , 𝑎22 : Parameters defining the stain energy Acknowledgments
density factor
This work was financially supported by the Provincial Science
𝐸: Young’s modulus
and Technology Support Project (2012FZ0124), Major State
FEM: Finite element method
Basic Research Project (2011CB201201), and the Interna-
𝐺: Energy release rate
tional Science & Technology Cooperation Program of China
𝐺max -criterion: Maximum energy release rate criterion
(2012DFA60760).
𝐾: Bulk modulus
𝐾I : Mode I stress intensity factor
𝐾I𝑐 : Fracture toughness of pure mode I crack References
𝐾II : Mode II stress intensity factor
𝐾III : Mode III stress intensity factor [1] S.-H. Chang, C.-I. Lee, and S. Jeon, “Measurement of rock frac-
𝑁: Total strain energy density ture toughness under modes I and II and mixed-mode condi-
tions by using disc-type specimens,” Engineering Geology, vol.
𝑁𝑑 : Distortional stain energy density in
66, no. 1-2, pp. 79–97, 2002.
𝑁𝑇-criterion
𝑁𝑑,0 : A constant distortional stain energy [2] I. L. Lim, I. W. Johnston, S. K. Choi, and J. N. Boland, “Fracture
testing of a soft rock with semi-circular specimens under three-
density in 𝑁𝑇-criterion
point bending. Part 2-mixed-mode,” International Journal of
𝑁𝑇-criterion: Maximum dilatational strain energy Rock Mechanics and Mining Sciences and, vol. 31, no. 3, pp. 199–
density criterion 212, 1994.
𝑁V : Dilatational stain energy density in
[3] Z. Jia, A. Castro-Montero, and S. P. Shah, “Observation of mixed
𝑁𝑇-criterion
mode fracture with center notched disk specimens,” Cement
𝑁V,max : Maximum dilatational stain energy and Concrete Research, vol. 26, no. 1, pp. 125–137, 1996.
density in 𝑁𝑇-criterion
[4] G. S. Xeidakis, I. S. Samaras, D. A. Zacharopoulos, and G. E.
𝑃: Applied load
Papakaliatakis, “Trajectories of unstably growing cracks in
𝑃𝑐𝑟 : Failure load mixed mode I-II loading of marble beams,” Rock Mechanics and
𝑅: Radius of SCB specimens Rock Engineering, vol. 30, no. 1, pp. 19–33, 1997.
𝑟𝜃: Crack tip coordinates
[5] A. Bobet and H. H. Einstein, “Fracture coalescence in rock-
𝑟𝑜 : A constant radius
type materials under uniaxial and biaxial compression,” Inter-
𝑟𝑝 : Elastic-plastic boundary national Journal of Rock Mechanics and Mining Sciences, vol. 35,
𝑟𝑝𝑐𝑟 : Critical radius in 𝑁𝑇-criterion no. 7, pp. 863–888, 1998.
𝑆: Stain energy density factor [6] C.-S. Chen, E. Pan, and B. Amadei, “Fracture mechanics anal-
𝑆min -criterion: Minimum stain energy density criterion ysis of cracked discs of anisotropic rock using the boundary
SCB: Semicircular bend element method,” International Journal of Rock Mechanics and
SEM: Scanning electron microscope Mining Sciences, vol. 35, no. 2, pp. 195–218, 1998.
2𝑠: Loading span in the SCB test [7] B. Vásárhelyi and A. Bobet, “Modeling of crack initiation,
𝑠/𝑅: Loading point span ratio in SCB propagation and coalescence in uniaxial compression,” Rock
specimen Mechanics and Rock Engineering, vol. 33, no. 2, pp. 119–139, 2000.
𝑡: Thickness of specimens [8] R. H. C. Wong, C. A. Tang, K. T. Chau, and P. Lin, “Splitting fail-
𝑌-criterion: Distortional strain energy density ure in brittle rocks containing pre-existing flaws under uniaxial
criterion compression,” Engineering Fracture Mechanics, vol. 69, no. 17, pp.
𝑌𝑑 : Distortional stain energy density in 1853–1871, 2002.
𝑌-criterion [9] N. A. Al-Shayea, “Crack propagation trajectories for rocks
𝑌𝑑,0 : Critical value distortional stain energy under mixed mode I-II fracture,” Engineering Geology, vol. 81,
density no. 1, pp. 84–97, 2005.
𝑌𝑑,𝜃0 : Minimum of distortional stain energy [10] H. Y. Liu, S. Q. Kou, P.-A. Lindqvist, and C. A. Tang, “Numerical
density modelling of the heterogeneous rock fracture process using
𝑌V : Dilatational stain energy density in various test techniques,” Rock Mechanics and Rock Engineering,
𝑌-criterion vol. 40, no. 2, pp. 107–144, 2007.
XRD: X-ray diffraction [11] M. R. M. Aliha, M. R. Ayatollah, D. J. Smith, and M. J. Pavier,
𝜇: Shear modulus “Geometry and size effects on fracture trajectory in a limestone
𝜎𝑟𝑟 , 𝜎𝜃𝜃 , 𝜎𝑟𝜃 : Stress components near the crack tip in rock under mixed mode loading,” Engineering Fracture Mechan-
polar coordinates ics, vol. 77, no. 11, pp. 2200–2212, 2010.
𝜎𝑥𝑥 , 𝜎𝑦𝑦 , 𝜎𝑥𝑦 : Stress components near the crack tip in [12] F. Erdogan and G. C. Sih, “On the crack extension in plates
Cartesian coordinates under plane loading and transverse shear,” Journal of Basic Engi-
𝜎𝑦 : Critical material yield strength neering, vol. 85, no. 4, pp. 519–527, 1963.
Mathematical Problems in Engineering 13

[13] G. C. Sih, “Strain-energy-density factor applied to mixed mode

crack problems,” International Journal of Fracture, vol. 10, no. 3,
pp. 305–321, 1974.
[14] K. Palaniswamy and W. G. Knauss, “Propagation of a crack
under general, in-plane tension,” International Journal of Frac-
ture Mechanics, vol. 8, no. 1, pp. 114–117, 1972.
[15] M. A. Hussain, S. L. Pu, and J. Underwood, “Strain energy
release rate for a crack under combined Mode I and Mode II,”
Fracture Analysis STP 560, ASTM, 1974.
[16] R. J. Nuismer, “An energy release rate criterion for mixed mode
fracture,” International Journal of Fracture, vol. 11, no. 2, pp. 245–
250, 1975.
[17] C. H. Wu, “Fracture under combined loads by maximum-
energy-release-rate criterion,” Journal of Applied Mechanics—
Transactions ASME, vol. 45, no. 3, pp. 553–558, 1978.
[18] A. P. Kfouri and M. W. Brown, “Fracture criterion for cracks
under mixed-mode loading,” Fatigue & Fracture of Engineering
Materials and Structures, vol. 18, no. 9, pp. 959–969, 1995.
[19] N. A. B. Yehia and M. S. Shephard, “The nt-criterion for predict-
ing crack growth increments,” Engineering Fracture Mechanics,
vol. 26, no. 3, pp. 371–382, 1987.
[20] N. A. B. Yehia, “Distortional strain energy density criterion: the
Y-Criterion,” Engineering Fracture Mechanics, vol. 39, no. 3, pp.
477–485, 1991.
[21] K. P. Chong and M. D. Kuruppu, “New specimen for fracture
toughness determination for rock and other materials,” Inter-
national Journal of Fracture, vol. 26, no. 2, pp. R59–R62, 1984.
[22] M. R. Ayatollahi and M. R. M. Aliha, “Wide range data for crack
tip parameters in two disc-type specimens under mixed mode
loading,” Computational Materials Science, vol. 38, no. 4, pp.
660–670, 2007.
[23] G. M. Fonseka, S. A. F. Murrell, and P. Barnes, “Scanning elec-
tron microscope and acoustic emission studies of crack devel-
opment in rocks,” International Journal of Rock Mechanics and
Mining Sciences and Geomechanics Abstracts, vol. 22, no. 5, pp.
273–289, 1985.
Hindawi Publishing Corporation
Mathematical Problems in Engineering
Volume 2013, Article ID 386501, 9 pages
http://dx.doi.org/10.1155/2013/386501

Research Article
POD for Real-Time Simulation of Hyperelastic Soft Biological
Tissue Using the Point Collocation Method of Finite Spheres

Suleiman Banihani,1 Timon Rabczuk,2,3 and Thakir Almomani4

1
Department of Mechatronics Engineering, Hashemite University, Zarqa 13115, Jordan
2
Institute of Structural Mechanics, Bauhaus University, Weimar, Marienstraße 15, 99423 Weimar, Germany
3
School of Civil, Environmental and Architectural Engineering, Korea University, Seoul 136-701, Republic of Korea
4
Department of Biomedical Engineering, Hashemite University, Zarqa 13115, Jordan

Correspondence should be addressed to Timon Rabczuk; [email protected]

Received 25 August 2013; Accepted 28 October 2013

Academic Editor: Goangseup Zi

Copyright © 2013 Suleiman Banihani et al. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.

The point collocation method of finite spheres (PCMFS) is used to model the hyperelastic response of soft biological tissue in real
time within the framework of virtual surgery simulation. The proper orthogonal decomposition (POD) model order reduction
(MOR) technique was used to achieve reduced-order model of the problem, minimizing computational cost. The PCMFS is a
physics-based meshfree numerical technique for real-time simulation of surgical procedures where the approximation functions
are applied directly on the strong form of the boundary value problem without the need for integration, increasing computational
efficiency. Since computational speed has a significant role in simulation of surgical procedures, the proposed technique was able
to model realistic nonlinear behavior of organs in real time. Numerical results are shown to demonstrate the effectiveness of the
new methodology through a comparison between full and reduced analyses for several nonlinear problems. It is shown that the
proposed technique was able to achieve good agreement with the full model; moreover, the computational and data storage costs
were significantly reduced.

1. Introduction is a major driving factor in such simulations [3–5]. The

demand for accuracy is not as high as in engineering-based
Minimally invasive surgery (MIS) is becoming the method of simulations; it is dictated by the just noticeable difference
choice for most surgical procedures with its many advantages (JND) of the human sensory system [4]. Hence, novel
of reduced postoperative pain, shorter hospital stays, quicker computational algorithms are necessary which can deliver the
recoveries, less scarring, and better cosmetic results [1]. high computational speeds at reasonable accuracy. Over the
However, MIS is very demanding in terms of the skill of last decade, much research has witnessed an explosion of the
the surgeon together with poor depth perception, limited number of tools available to enhance medical education, such
field of view, unnatural hand-eye coordination, and poor as virtual reality—(VR) based medical training systems.
tactile perception. Moreover, the learning curve to master Several methods have been proposed for fast computation
such techniques is long and tedious [2]. of mechanical deformation of soft tissues. Early attempts
The success of flight simulators to train pilots has fuelled were based on a nonphysical approach, focusing on the
the enthusiasm for computer-based training systems for visualization aspect of deformation and operation. In [6]
surgeons. The goal of surgical simulation is to produce a a linked volume representation was used to model objects
realistic virtual environment where a trainee can explore in interactions. In [7], a physical model based on elasticity was
real time a medical procedure of a three-dimensional organ introduced to describe deformable objects.
model using his or her sense of vision and touch through Mass-spring model is one of the most widely used
specialized haptic interface devices [3]. Computational speed physical methods in which the material is represented by
2 Mathematical Problems in Engineering

an array of nodes connected by elastic springs [8–11]. This methods allow for much larger critical time steps in explicit
model has been applied for a variety of objects including dynamics. Reference [33] showed an increase by factors up to
human tissues, muscles, and blood vessels [12–14]. A mass- 100.
spring-damper system was used as well in [15] to attain a more The goal of this research is the development of physics-
realistic representation. based simulation techniques for the modeling of surgical
However, both models inaccurately simplify the govern- tool-soft tissue interactions, such as deformation, incision,
ing equations and offer a very unrealistic behavior [16]. An and cutting as well as the reaction forces on the surgical tools,
alternative is to use the finite-element method (FEM) [17–19] in real time. In [30], we presented a novel meshfree computa-
which is based on the principles of continuum mechanics. tional approach, that is, the point collocation method of finite
Due to the viability and potential of FEM, it is increasingly spheres for realizing a viscoelastic tissue model.
becoming the method of choice in most surgery simulators. Different model order reduction techniques were applied
In [17, 18], the FEM was used for surgery simulations in real to reduce the computational time. The results were promis-
time using an elastic quasi-static formulation. Later in [19], ing, but more investigations need to be done. This method has
a dynamic formulation was used based on the tensor-mass
the potential to develop into the de facto standard in future
method, where a long preprocessing step is required, which
surgical simulators. However, in [30], only modal truncation
is not suitable for applications such as real-time planning for
(MT), Hankel optimal model, and truncated balanced real-
surgical procedures. In [20], the boundary element method
ization (TBR) MOR techniques were investigated for PCMFS.
(BEM) with a surface mesh was used to build a real-time
These methods can only apply in certain cases and may
model. However, this approach is not suitable especially if
the inside of the organ is involved or for nonhomogenous result in misleading results for highly nonlinear problems;
materials. moreover, they are computationally expensive.
Linear elastic models are only applicable for small defor- The technology developed in this work is a significant
mations; experimental characterization of soft biological step towards the development of VR-based surgical training
tissue indicates that the behavior of tissues is rather nonlinear. systems which will enable medical students and residents to
Therefore, a viscoelastic constitutive model is more evocative train and practicing surgeons to retrain on complex surgical
[21–25]. In [26], a nonlinear hyperelastic St. Venant Kirchhoff procedures. The PCMFS is combined with the POD to
material was used for modeling soft tissue in real time, which produce a fast physics-based virtual environment; this will
is restricted to a linear stress-strain relation. significantly reduce the computational cost which allows for
The finite volume method was used in [27] to simulate soft more nonlinear phenomena to be modeled in real time.
tissue deformation through an explicit integration scheme. The point collocation method of finite spheres is pre-
A total Lagrangian explicit dynamic (TLED) algorithm was sented in Section 2 as well as the elastodynamic initial value
proposed in [28] where the calculations are based on the problem. In Section 3, a hyperelastic constitutive model for
reference configuration of the material. In [29], a graphics soft biological tissues is shown, and then proper orthogonal
processing unit (GPU) was utilized in applying this approach decomposition is utilized to reduce the complexity of the
combined with Prony series to model viscoelasticity, reaching full model and reduce computational cost in Section 4.
real-time speed. However, explicit time integration simplifies Finally, numerical examples are presented in Section 5 to
the update at each time step, but it requires small time demonstrate the effectiveness of the proposed algorithm.
steps to guarantee computational stability especially for stiff
materials. Furthermore, with explicit schemes, it is necessary
to iterate multiple times to propagate applied forces from a 2. The Point Collocation Method of
node to the whole mesh. Finite Spheres (PCMFS)
Although computerized skill trainers and VR training The PCMFS is a computationally efficient technique pro-
systems have been developed, none of them has been inte-
posed in [34] where the computational nodes are sprinkled
grated officially into a medical curriculum or any other offi-
on the computational domain (Figure 1). At every node “𝐼”
cial training program or course, and the current technology
located at X𝐼 , an approximation function is defined which is
is inadequate to address the issues of realistic simulation and
rendering in such simulators. compactly supported on the sphere 𝐵𝐼 = 𝐵(X𝐼 , 𝑟𝐼 ) of radius
𝑟𝐼 centered at the node. The elastodynamic initial boundary
Meshfree collocation-based methods [30, 31] offer a huge
value problem (Section 2.1) is solved using point collocation
advantage in terms of time saving, in which the essential
(Section 2.2). The moving least squares approximation func-
boundary conditions are applied directly on the boundary
tions, discussed in Section 2.3, are used for discretization, and
nodes with no additional treatment; moreover, there is no
the discretized set of equations is presented in Section 2.4.
time-consuming numerical integration of the weak form
as the approximation functions are applied directly on the
strong form of the problem. In [32], it was shown that the 2.1. The Elastodynamic Initial Boundary Value Problem. Dur-
accuracy of the solution in the meshfree point collocation ing surgical simulation, the surgical tool interacts with the
method for elasticity and crack problems is excellent and portion 𝜕𝐵𝑢2 of the body 𝐵 with boundary 𝜕𝐵. Homogeneous
the error is less than that in the element-free Galerkin Dirichlet boundary conditions are prescribed on the portion
(EFG) method with linear basis. Nodally integrated meshfree 𝜕𝐵𝑢1 of the boundary, and tractions are prescribed on 𝜕𝐵𝑓
Mathematical Problems in Engineering 3

boundary which, for three-dimensional analysis, have the

following representation:
𝜕Bu2 𝑛𝑥 0 0 𝑛𝑦 0 𝑛𝑧
N = [ 0 𝑛𝑦 0 𝑛𝑥 𝑛𝑧 0 ] . (2)
𝜕B 𝜕Bf [ 0 0 𝑛𝑧 0 𝑛𝑦 𝑛𝑥 ]

B(XI , rI ) 2.2. Point Collocation. In the point collocation method [35,

B
36], the displacement solution u is approximated by uℎ , and
𝜕Bu1
the governing partial differential equations are applied at the
nodal points. The discrete set of equations is given as follows:

X3 [∇ ⋅ P (uℎ )]X=X𝐼 + 𝜌0 fb (X𝐼 ) = 𝜌0 ü(X𝐼 ) on 𝐵 × (0, 𝑇) ,

XI
(3)

[N ⋅ P (uℎ )]X=X𝐼 = f𝑠 (X𝐼 ) on 𝜕𝐵𝑓 × (0, 𝑇) , (4)

X1 X2
uℎ (X𝐼 , 𝑡) = 0 on 𝜕𝐵𝑢1 × (0, 𝑇) , (5)
Figure 1: Schematic of the PCMFS for dynamic surgery simulation. uℎ = utooltip on 𝜕𝐵𝑢2 × (0, 𝑇) , (6)
Discretization of a domain 𝐵 ⊂ R3 by the PCMFS using a
set of nodal points. 𝐵𝐼 is the open sphere at node 𝐼. Natural uℎ (X𝐼 , 0) = u0 (X𝐼 ) on 𝐵, (7)
boundary conditions are defined on 𝜕𝐵𝑓 , and the essential boundary
conditions are defined on 𝜕𝐵𝑢 :𝜕𝐵 = 𝜕𝐵𝑢 ∪ 𝜕𝐵𝑓 , and 𝜕𝐵𝑢 ∩ 𝜕𝐵𝑓 = 0. u̇ℎ (X𝐼 , 𝑡) = u̇0 (X𝐼 ) on 𝐵. (8)
Note that 𝜕𝐵𝑢 = 𝜕𝐵𝑢1 ∪ 𝜕𝐵𝑢2 , where u(X) = 0 for X ∈ 𝜕𝐵𝑢1 and
u = utooltip on 𝜕𝐵𝑢2 × (0, 𝑇). The use of smooth weight functions allows for the higher-
order derivatives in (3) to be taken. The point collocation
method obviates expensive numerical integration, but it
results in a nonsymmetric stiffness matrix.
(Figure 1). We are interested in solving the specialized strong
form of the following initial boundary value problem:
2.3. The Moving Least Squares Approximation Scheme. In
PCMFS, the moving least squares [32] technique is used to
generate the approximation functions. In this technique, the
∇0 ⋅ P + 𝜌0 f𝑏 = 𝜌0 ü on 𝐵 × (0, 𝑇) , approximation uℎ (X, 𝑡) of a variable 𝑢, using “𝑁” particles, is
N ⋅ P = f𝑠 on 𝜕𝐵𝑓 × (0, 𝑇) , given as
𝑁
u=0 on 𝜕𝐵𝑢1 × (0, 𝑇) ,
(1) u (X, 𝑡) ≈ uℎ (X, 𝑡) = ∑ H𝐽 (X) 𝛼𝐽 (𝑡) = H (X) U (𝑡) , (9)
𝐽=1
u = utooltip on 𝜕𝐵𝑢2 × (0, 𝑇) ,

u (X, 0) = u0 , X ∈ 𝐵, where 𝛼𝐽 = [𝑢𝐽 , V𝐽 , 𝑤𝐽 ] is the vector of nodal unknowns

corresponding to the 𝑋1 , 𝑋2 , and 𝑋3 directions at node “𝐽”
u̇(X, 0) = u̇0 , X ∈ 𝐵, and U = [𝛼1 𝛼2 ⋅ ⋅ ⋅ 𝛼𝑁]𝑇 . The nodal shape function matrix is
given as

where P is the nominal stress, ∇ is the divergence with ℎ𝐽 (X) 0 0

subscript 0 indicating the reference configuration, u = H𝐽 (X) = [ 0 ℎ𝐽 (X) 0 ], (10)
[𝑢 V 𝑤]𝑇 is the displacement vector, and üdenotes second [ 0 0 ℎ𝐽 (X)]
derivative with respect to time. u0 = [𝑢0 V0 𝑤0 ]𝑇 and u̇0 =
where the shape functions at node “𝐽” are generated using a
[𝑢̇0 V̇0 𝑤̇0 ]𝑇 are the initial displacements and velocities,
moving least squares procedure and has the form
respectively. The surgical tool interacts with the portion 𝜕𝐵𝑢2
of the boundary and prescribes a displacement field utooltip
ℎ𝐼 (X) = 𝑊𝐼 (X) P𝑇 (X) A−1 (X) P (X𝐼 ) , 𝐼 = 1, . . . , 𝑁,
which is a function of space and time. The displacement on
the rest of the Dirichlet boundary (𝜕𝐵𝑢2 ) is assumed to be (11)
zero for the entire simulation period (𝑇). f𝑏 = [𝑓𝑏𝑥 𝑓𝑏𝑦 𝑓𝑏𝑧 ]𝑇 with
is the body force, and f𝑠 = [𝑓𝑠𝑥 𝑓𝑠𝑦 𝑓𝑠𝑧 ]𝑇 is the applied
𝑁
traction force. N is a matrix of direction cosine compo- A (X) = ∑ 𝑊𝐽 (X) P𝑇 (X𝐽 ) P (X𝐽 ) , (12)
nents (𝑛𝑥 , 𝑛𝑦 , 𝑛𝑧 ) of the unit outward normal to the domain 𝐽=1
4 Mathematical Problems in Engineering

where P(X) is a vector of monomials; for example, P(x) = or logarithmic functions are introduced in the strain energy
[1, 𝑋1 , 𝑋2 , 𝑋3 ] ensures first-order accuracy in 3D. We function [37] as follows:
define a positive radial weight function 𝑊𝐽 (X) = 𝑊(𝑠𝐽 ) ∈ 𝛼 𝛼 𝛼
𝐶0𝑠 (𝐵(X𝐽 , 𝑟𝐽 )), 𝑠 ≥ 1 with, 𝑠𝐽 = ‖X − X𝐽 ‖0 /𝑟𝐽 at each node “𝐽” 𝑊 = 𝐶1 [exp (𝐶2 (𝜆 11 + 𝜆 21 + 𝜆 31 − 3)) − 1]
which is compactly supported on the sphere at node “𝐽”. In (18)
𝛼 𝛼 𝛼
our work, we have used a quartic spline weight function of + 𝐶3 (𝜆 12 + 𝜆 22 + 𝜆 32 − 3) ,
the form or
2 3 4
1 − 6𝑠 + 8𝑠 − 3𝑠 , 0 ≤ 𝑠 < 1, 𝛼 𝛼 𝛼
𝑊 = −𝐶1 [ln (1 − 𝐶2 (𝜆 11 + 𝜆 21 + 𝜆 31 − 3))]
𝑊 (𝑠) = { (13)
0, 𝑠 ≥ 1. 𝛼 𝛼 𝛼
(19)
+ 𝐶3 (𝜆 12 + 𝜆 22 + 𝜆 32 − 3) .
3. Hyperelastic Constitutive Model For hyperelastic material, the second Piola-Kirchhoff stress,
S, is derived from the strain energy function as follows:
Biological soft tissues are complicated; they are anisotropic,
viscoelastic, and inhomogeneous, and they allow large defor- 𝜕𝑊
mation. Therefore, there is no known constitute model that S=2 . (20)
can capture the exact mechanical and thermodynamical 𝜕C
behavior of all tissues. In this work, we are concerned with The nominal stress is related to the second Piola-Kirchhoff
certain organ, that is, the liver. Liver can be considered stress in the following manner:
to be homogeneous, isotropic, and incompressible because
liver tissue is highly consistent with a high water content. A P = S ⋅ F𝑇 . (21)
hyperelastic constitute model for liver is widely used [37].
Most surgical simulations focus on linear elastic models
for soft tissue as in [30]. In this paper, we will assume 4. Proper Orthogonal Decomposition (POD)
a hyperelastic constitutive model to account for nonlinear Model order reduction (MOR) methods have been developed
behavior of moderate strains. Nevertheless, our development for large-scale dynamical systems [40–42] where they are
can be easily extended to more complicated material models. used to approximate the input-output behavior of the system
In [38, 39], Ogden proposed a general form of strain over a certain range of operations using significantly smaller
energy function as follows: matrices. MOR retains the essential dynamics and physics
𝛼 𝛼 𝛼 contained within the full system but at a much lower
𝑊 = ∑𝐶𝑖 (𝜆 1𝑖 + 𝜆 2𝑖 + 𝜆 3𝑖 − 3) , (14) computational cost. Model order reduction methods offer
𝑖
an excellent route to computing input-output responses by
where 𝐶𝑖 and 𝛼𝑖 are material parameters and 𝑖 is the number eliminating a large number of degrees of freedom which do
of terms included in the summation. 𝜆 1 , 𝜆 2 , and 𝜆 3 are not have a significant influence on the output. A useful model
the principal stretch ratios and the eigenvalues of the right order reduction technique [43] has the following properties.
Cauchy-Green deformation tensor, C = F𝑇 F, where F is the (i) It reduces the number of variables significantly rela-
deformation gradient given by tive to the full-order model.
𝜕x (ii) It is controlled by a limited number of relevant inputs.
F= , (15)
𝜕X (iii) It is relatively inexpensive to solve and store in
computer’s memory.
where X and x are the coordinates of the material point in the
reference and current configuration, respectively. There are three major approaches for generating reduced-
The Mooney-Rivlin material is a special case of this order models for linear time-invariant systems:
function where a polynomial form of the strain energy
function is used as follows: (i) Krylov subspace-based methods,
𝐶1 𝐶 (ii) Hankel norm and truncated balancing realization
𝑊= (𝐼 − 3) + 2 (𝐼2 − 3) , (16) (TBR-) based methods,
2 1 2
(iii) Karhunen-Lóeve expansion or proper orthogonal
where 𝐼1 = 𝜆21 + 𝜆22 + 𝜆23 and 𝐼2 = 𝜆21 𝜆22 + 𝜆22 𝜆23 + 𝜆23 𝜆21 . decomposition (POD) methods.
Hookean model is the simplest hyperelastic model with
𝐶2 in the Mooney-Rivlin model set to zero, and the strain All of those methods apply the idea of approximating
energy function is given by the original high-fidelity system with a relatively lower-
dimensional and computationally cheaper model by per-
𝑊 = 𝐶1 (𝐼1 − 3) . (17) forming projection of the original space into a lower-
dimensional space while maintaining relatively small error. In
In order to account for sharp increase in stiffness after toe order to be successful, the reduced-order model (ROM) must
regions in the stress-strain curve of soft tissue, exponential be predictive across the design or parameter space of interest.
Mathematical Problems in Engineering 5

Krylov subspace-based methods are numerically robust subject to

algorithms since they preserve a certain number of moments
of the transfer function in the reduced model. Therefore, the (𝜙𝑖 , 𝜙𝑗 ) = 𝛿𝑖𝑗 , 1 ≤ 𝑖 ≤ 𝑀, 1 ≤ 𝑗 ≤ 𝑖, (23)
reduced system approximates well the original transfer func-
2
tion around a specified frequency or collection of frequency where ⟨⋅⟩ is a discrete averaging operator; that is, ⟨‖𝑢𝑝 ‖ ⟩ =
2
points [44]. However, there are no provable error bounds (1/𝐿) ∑𝐿𝑝=1 ‖𝑢𝑝 ‖ , and Ξ𝑀 : 𝐻(Ω) → 𝐻𝑀(Ω) is an orthog-
that guarantee preserving stability or passivity of the original onal projection operator.
system for the reduced models. The reduced-order model solution 𝑢𝑀 can be represented
The second group of methods is based on the Hankel as a linear combination of 𝜙𝑘 , the POD modes, as follows:
norm and truncated balancing realization (TBR). Unlike the
Krylov subspace methods, these methods have provable error 𝑀
bounds and guarantee that stability of the original system 𝑢𝑀 (x, 𝑡) = ∑ 𝜙𝑘 (x) 𝑎𝑘 (𝑡) , (24)
will be preserved in the reduced-order model [45]. However, 𝑘=1

these methods are computationally expensive for extracting where 𝑎𝑘 (𝑡) are ROM coefficients. The solution of the
the reduced model because the solution of Lyapunov equa- optimization problem in (22) is reduced to the following
tions requires 𝑂(𝑁3 ) operations. eigenvalue problem:
Karhunen-Lóeve expansion or proper orthogonal
decomposition (POD) method offers yet another alternative 𝐶Φ = 𝜆Φ, (25)
[46]. It is a powerful and elegant method which obtains
projection based on time or frequency domain snapshots where 𝐶 is self-adjoint positive semidefinite operator defined
[47, 48]. POD has been widely used in a variety of as
fields including image processing, signal analysis, data 1 𝐿 𝑝
compression, process identification, and adaptive control. 𝐶 = ⟨𝑢𝑝 ⊗ 𝑢𝑝 ⟩ = ∑ (𝑢 ⊗ 𝑢𝑝 ) . (26)
𝐿 𝑝=1
4.1. Reduced-Order Models Using Proper Orthogonal Decom- In [49, 50], it was shown that the set of 𝑀 eigenfunctions,
position (POD). Generating a reduced-order model of the or POD modes, {𝜙𝑘 , 𝑘 = 1, 2, . . . , 𝑀}, corresponds to the 𝑀
high-fidelity original partial differential equation consists of largest eigenvalues of 𝐶 that is precisely the set that solves (25)
the following two steps. and that the minimum value of the objective function in (22)
is
(i) The first step is to transform the kinematic informa-
󵄩 󵄩2
tion, that is, in our case, the displacement field, to a ⟨󵄩󵄩󵄩𝑢𝑝 − Ξ𝑀𝑢𝑝 󵄩󵄩󵄩 ⟩
smaller number of modes.
(27)
(ii) Then, the full-system is reduced to the dynamics 1 𝐿 󵄩󵄩 𝑝 󵄩2
𝐿
= ∑ 󵄩󵄩𝑢 − Ξ𝑀𝑢𝑝 󵄩󵄩󵄩 = ∑ 𝜆 𝑗 .
implied by the reduced modes. 𝐿 𝑝=1 𝑗=𝑀+1

The result is a set of time-dependent ordinary differential

equations (ODEs) in the reduced-order model modes which 5. Nonlinear Model Reduction for
are able to describe the dynamics of the original PDE with a Hyperelastic Material
relatively small error [49].
Let 𝑢(x, 𝑡) ∈ 𝐻(Ω) be our variable of interest in the Using direct simulation of the initial value problem in (3)
governing PDE defined over the domain Ω ∈ R𝑛 , where results in a set of 𝐿 instantaneous snapshots in time; that
𝐻(Ω) is a Hilbert space with associated inner product (⋅, ⋅), is, 𝑢(x𝑖 , 𝑡𝑝 ) = 𝑢𝑝 , 𝑖 ∈ [1, . . . , 𝑁], 𝑝 ∈ [1, . . . , 𝐿].
and let 𝑢(x𝑖 , 𝑡𝑝 ) = 𝑢𝑝 , 𝑖 ∈ [1, . . . , 𝑁], 𝑝 ∈ [1, . . . , 𝐿], be the From the 𝑀 eigenfunctions, or POD modes, {𝜙𝑘 (x𝑖 ), 𝑘 =
ensemble set of 𝐿 instantaneous snapshots in time of this field 1, 2, . . . , 𝑀 and 𝑖 = 1, 2, . . . , 𝑁} corresponding to the 𝑀 larg-
expressed in a discrete form; that is, it is known typically at the est eigenvalues, the matrix Q can be defined as
nodes of a spatial mesh and for some time steps of existing
𝜙1 (x1 ) 𝜙2 (x1 ) ⋅ ⋅ ⋅ 𝜙𝑀 (x1 )
numerical solution.
𝜙1 (x2 ) 𝜙2 (x2 ) ⋅ ⋅ ⋅ 𝜙𝑀 (x2 )
The main idea of POD is to obtain a basis 𝜙𝑘 of order Q=( . .. .. ). (28)
𝑀 ≪ 𝐿, where {𝜙𝑘 , 𝑘 = 1, 2, . . . , 𝑀} is the most typical linear .. . d .
basis for describing the original ensemble. Therefore, POD is 𝜙1 (x𝑁) 𝜙2 (x𝑁) ⋅ ⋅ ⋅ 𝜙𝑀 (x𝑁)
searching for an 𝑀-dimensional subspace spanned by 𝜙𝑘 ∈
𝐻𝑀(Ω) such that the projection of the difference between The discretization of the partial equation, (3), is given as
the ensemble 𝑢𝑝 and its projection onto 𝐻𝑀(Ω) is minimized follows:
on average. That is equivalent to finding the function 𝜙𝑘 that MÜ+ CU̇ + KU = F, (29)
solves the following optimization problem:
where M, C, and K are the mass, damping, and stiffness
󵄩 󵄩2
min 𝑀
⟨󵄩󵄩󵄩𝑢𝑝 − Ξ𝑀𝑢𝑝 󵄩󵄩󵄩 ⟩ (22)
matrices, respectively, U are the nodal unknowns, and F is
(𝜙𝑘 )𝑘=1 the external force vector.
6 Mathematical Problems in Engineering

101

F(t) 100

10−1

Average relative error

Figure 2: Beam under traction: concentrated force is applied to the
right side, whereas the other side is fixed. 10−2

10−3

Using the POD modes equation (24), the nodal un- 10−4
knowns can be expressed as
10−5
U = QA, (30)
10−6
where A are ROM coefficients, and the dynamic problem in 2 4 6 8 10 12 14 16
(29), can be approximated as follows: Number of basis modes used for POD

MQÄ+ CQȦ + KQA = F. (31) Figure 3: Average relative error in the solution of the displacement
of the beam as the number of basis modes in the POD increases.
Multiplying both sides of (31) by Q𝑇 gives
0.05
Q𝑇 MQÄ+ Q𝑇 CQȦ + Q𝑇 KQA = Q𝑇 F, (32)
0.045
which results in a final system of equations of order 𝑀 × 𝑀, 0.04
with 𝑀 ≪ 𝐿.
CPU time (s)

0.035
6. Numerical Examples 0.03

In order to demonstrate the effectiveness of the proposed 0.025

technique, we will show 2 examples of hyperelastic models.
0.02
6.1. Beam under Traction. In this example, we apply an axial 0.015
force on the right side of the beam, whereas the left side of
the beam is fixed. The beam is 200 mm long with a square 0.01
cross-sectional area with sides of length 20 mm, as shown in 2 4 6 8 10 12 14 16
Figure 2. Number of basis modes used for POD
The force 𝐹(𝑡) is given by Figure 4: CPU time used in seconds for the solution of the
beam problem using the PCMFS with POD model order reduction
200𝑡, 0 ≤ 𝑡 < 0.5, method.
𝐹 (𝑡) = { (33)
200 (1 − 𝑡) , 0.5 ≤ 𝑡 ≤ 1.

Here, we assumed a Mooney-Rivlin hyperelastic material It is shown from Figure 3 that the relative error decreases
model with 𝐶1 = 60 kPa, 𝐶2 = 1.2 kPa, and density of steeply with increasing the number of POD modes used; for
1120 kg/m3 . example, the relative error with only 6 basis modes is less than
Explicit time integration was used in the full model, and 3% of the full-order model. However, after 8 basis modes the
the simulation period was 1 sec where a snapshot was taken decrease in the relative error is slight.
every 0.01 sec. Figure 4 shows the CPU time consumed in seconds for
Figure 3 shows the average relative error in the displace- the solution of the beam problem using the PCMFS with POD
ment field solution as the number of POD basis modes as the number of basis functions increases in the reduced-
increases, and the average error is defined relative to the order model.
solution of the full model as follows: As shown in the figure, it is noticed that the time scales
󵄨󵄨 󵄨󵄨 almost linearly with increasing the number of basis modes
∑𝑁
𝐼=1 󵄨󵄨𝑢Full (𝑥𝐼 ) − 𝑢POD (𝑥𝐼 )󵄨󵄨 for the beam problem.
error = 󵄨󵄨 󵄨󵄨 , (34)
󵄨󵄨𝑢Full (𝑥𝐼 )󵄨󵄨 Figure 5 shows the time used to solve the reduced-order
model problem relative to the full-order model as the number
where 𝑢Full and 𝑢POD refer to the displacement solution for the of POD basis modes increases. The time used for POD with
full- and reduced-order models, respectively, and the error is only 2 basis functions was almost 1% of the time needed for
averaged over time as well. the full-order model, whereas for 8 basis functions the time
Mathematical Problems in Engineering 7

0.05 101

0.045
100
Relative time consumption

0.04
10−1

Average relative error

0.035

0.03 10−2

0.025
10−3
0.02
10−4
0.015

0.01 10−5
2 4 6 8 10 12 14 16 2 4 6 8 10 12 14 16
Number of basis modes used for POD Number of basis modes used for POD

Figure 5: Time consumption relative to the full model as the Figure 7: Relative error in the displacement field of the hyperelastic
number of POD basis modes increases. circular membrane as the number of POD basis modes increases.

0.45

0.4

Tool 0.35
CPU time (s)

0.3

0.25
𝜌 = 1200 kg/m3
R = 10 cm
0.2
C1 = 0.17 MPa

0.15
H = 0.01 mm

0.1
0 5 10 15
Figure 6: Hyperelastic circular membrane with fixed boundary Number of basis modes for POD
conditions and concentrated force at the center of the top surface.
Figure 8: CPU time consumed in the solution of the hyperelastic
circular membrane as the number of POD basis modes increases.

consumed was around 3% of the full-order model with a

relatively small error as shown in Figure 3. 2 basis functions is relatively high; however, by increasing the
number of basis functions, the error is dramatically reduced.
6.2. Circular Membrane. Here, we will consider a circular The error with 8 basis functions is very small, and the change
hyperelastic membrane of a radius 𝑅 = 10 cm, thickness in error after that is negligible.
𝐻 = 0.01 mm, and density 𝜌 = 1200 kg/m3 , Figure 6. The The CPU time consumption is shown in Figure 8, where
boundary of the membrane is fixed, while the tool applies a it is shown that the CPU time was reduced dramatically
concentrated load at the center of the top surface as follows: by using small number of POD basis modes; however, as
the number of basis modes increases, the time consumption
0.1𝑡 𝑡 < 0.1 increases as well.
𝑢 = 𝑢tooltip = { at 𝑟 = 0. (35) Figure 9 shows the solution of the displacement of the
0, 𝑡 ≥ 0.1
hyperelastic circular membrane problem shown in Figure 6,
A Neo-Hookean hyperelastic model is assumed with 𝐶1 = and the solution is obtained using POD model order reduc-
0.17 MPa, and all of the initial conditions of the membrane tion technique with 8 basis functions after 0.1 sec.
are set to zero.
Figure 7 shows the relative error, (34), as the number of 7. Conclusion
POD basis modes increases. It is obvious that the relative
error decreases as the number of basis modes increases. In this paper, we have developed a point collocation-based
The relative error of the reduced-order model with only method of finite spheres (PCMFS) approach with POD
8 Mathematical Problems in Engineering

[4] T. Krummel, “Surgical simulation and virtual reality: the com-

ing revolution,” Annals of Surgery, vol. 228, no. 5, pp. 635–637,
1998.
[5] A. Levin, “Fewer crashes caused by pilots,” Tech. Rep., USA
Today, Washington, DC, USA, 2004.
[6] S. F. Frisken-Gibson, “Using linked volumes to model object
collisions, deformation, cutting, carving, and joining,” IEEE
Transactions on Visualization and Computer Graphics, vol. 5, no.
4, pp. 333–348, 1999.
[7] D. Terzopoulost, J. Platt, A. Barr, and K. Fleischert, “Elasti-
Figure 9: Solution of the displacement field of the hyperelastic cally deformable models,” in Proceedings of the 14th Annual
circular membrane using POD and PCMFS. Conference on Computer Graphics and Interactive Techniques
(SIGGRAPH ’87), vol. 21, no. 4, pp. 205–214, July 1987.
[8] S. A. Cover, N. F. Ezquerra, R. Rowe, T. Gadacz, and J. F. O’Brien,
model order reduction technique for the solution of non- “Interactively deformable models for surgery simulation,” IEEE
linear hyperelastic problems that arise in the simulation of Computer Graphics and Applications, vol. 13, no. 6, pp. 68–75,
soft tissue response. In this technique, an approximation is 1993.
generated using the moving least squares method, while the [9] U. G. Kühnapfel, C. Kuhn, M. Hübner, H. G. Krumm, H. Maass,
point collocation method is used as the weighted residual and B. Neisius, “The Karlsruhe endoscopic surgery trainer as
technique. The advantage of this method is that numerical an example for virtual reality in medical education,” Minimally
integration is not necessary which allows for efficient com- Invasive Therapy and Allied Technologies, vol. 6, no. 2, pp. 122–
125, 1997.
putations necessary for the simulation of applications such as
[10] Y. Lee, D. Terzopoulos, and K. Waters, “Realistic modeling
virtual surgery.
for facial animation,” in Proceedings of the 22nd Annual ACM
POD model order reduction technique, used extensively Conference on Computer Graphics and Interactive Techniques
in the field of large-scale dynamical systems, has been applied (SIGGRAPH ’95), pp. 55–62, August 1995.
to reduce the computational cost associated with nonlinear [11] R. M. Koch, M. H. Gross, F. R. Carls, D. F. von Buren, G.
hyperelastic problems found in soft tissue modeling. POD Fankhauser, and Y. I. H. Parish, “Simulating facial surgery
technique reduces the computational cost by generating an using finite element models,” in Proceedings of the 23rd Annual
offline set of snapshots used to generate POD basis functions Conference on Computer Graphics and Interactive Techniques
that is used during online computation to reduce the large- (SIGGRAPH ’96), pp. 421–428, August 1996.
order original system by a smaller-order system reducing the [12] R. M. Kenedi, T. Gibson, J. H. Evans, and J. C. Barbenel, “Tissue
computational cost. mechanics,” Physics in Medicine and Biology, vol. 20, no. 5, pp.
The combination of the two methods was able to solve 163–169, 1975.
the boundary value problem in real time with relatively [13] K. Waters, “Physical model of facial tissue and muscle articula-
acceptable margin of error. The results showed that the tion derived from computer tomography data,” in Visualization
computational time for the evaluation of the system matrices in Biomedical Computing ’92, vol. 1808 of Proceedings of SPIE,
increases linearly with the number of nodes. For example, pp. 574–583, October 1992.
we were able to achieve relatively negligible error with only [14] S. Cotin, J. Pignon, H. Delinguette, and G. Subsol, “A craniofa-
8 basis functions with a time consumption of almost 3% of cial surgery simulation test-bed,” in Visualization for Biomedical
the time used in the full-order model. Computing (VBC ’94), Rochester, Minn, USA, October 1994.
The proposed technique can play a vital role in the [15] S. Platt and N. Badler, “Animating facial expressions,” in
Proceedings of the 8th Annual Conference on Computer Graphics
development of physics-based virtual surgery environment
and Interactive Techniques (SIGGRAPH ’81), vol. 15, pp. 245–252,
that can account for nonlinear behavior of soft biological August 1981.
tissue; this method can be extended to more complicated
[16] A. Gelder, “Approximate simulation of elastic membranes by
material models in which more complicated phenomena such triangulated spring meshes,” Journal of Graphic Tools, vol. 3, no.
as deformation, incision, and cutting, as well as the reaction 2, pp. 21–42, 1998.
forces on the surgical tools, are modeled in real time. [17] M. Bro-Nielsen, “Finite element modeling in surgery simula-
tion,” Proceedings of the IEEE, vol. 86, no. 3, pp. 490–503, 1998.
References [18] S. Cotin, H. Delingette, and N. Ayache, “Real-time elastic
deformations of soft tissues for surgery simulation,” IEEE
[1] M. J. Wagner and H. A. Thomas Jr., “Application of the medical Transactions on Visualization and Computer Graphics, vol. 5, no.
knowledge general competency to emergency medicine,” Aca- 1, pp. 62–73, 1999.
demic Emergency Medicine, vol. 9, no. 11, pp. 1236–1241, 2002. [19] H. Delingette, N. Ayache, and S. Cotin, “A hybrid elastic model
[2] Y. Lim, Physics based digital surgery using the point associated for real-time cutting, deformations, and force feedback for
finite field approach [Ph.D. thesis], Rensselaer Polytechnic Insti- surgery training and simulation,” The Visual Computer, vol. 16,
tute, Troy, NY, USA, 2005. no. 8, pp. 437–452, 2000.
[3] E. Raposio, M. Fato, A. Schenone et al., “An “augmented-reality” [20] D. James and A. Pai, “Artdefo: accurate real time deformable
aid for plastic and reconstructive surgeons,” in Proceedings of objects,” in Proceedings of the 26th ACM Annual Conference on
the Medicine Meets Virtual Reality (MMVR ’5), pp. 232–235, Computer Graphics and Interactive Techniques (SIGGRAPH ’99),
Amsterdam, The Netherlands, 1997. pp. 65–72, August 1999.
Mathematical Problems in Engineering 9

[21] K. Miller, “Constitutive model of brain tissue suitable for finite [39] R. Ogden, “Large deformation isotropic elasticity—on the cor-
element analysis of surgical procedures,” Journal of Biomechan- relation of theory and experiment for incompressible rubberlike
ics, vol. 32, no. 5, pp. 531–537, 1999. solids,” Proceedings of the Royal Society of London A, vol. 326, no.
[22] K. Miller and K. Chinzei, “Constitutive modelling of brain 1567, pp. 565–584, 1972.
tissue: experiment and theory,” Journal of Biomechanics, vol. 30, [40] A. C. Antoulas, Approximation of Large-Scale Dynamical Sys-
no. 11-12, pp. 1115–1121, 1997. tems, Society for Industrial and Applied Mathematic (SIAM),
[23] K. Miller and K. Chinzei, “Mechanical properties of brain tissue Philadelphia, Pa, USA, 2005.
in tension,” Journal of Biomechanics, vol. 35, no. 4, pp. 483–490, [41] R. W. Freund, “Krylov-subspace methods for reduced-order
2002. modeling in circuit simulation,” Journal of Computational and
[24] K. Miller, K. Chinzei, G. Orssengo, and P. Bednarz, “Mechanical Applied Mathematics, vol. 123, no. 1-2, pp. 395–421, 2000.
properties of brain tissue in-vivo: experiment and computer [42] E. Rudnyi and J. Korvink, “Review: automatic model reduction
simulation,” Journal of Biomechanics, vol. 33, no. 11, pp. 1369– for transient simulation of MEMS-based devices,” Sensors
1376, 2000. Update, vol. 11, no. 1, pp. 3–33, 2002.
[25] M. Farshad, M. Barbezat, P. Flueler, F. Schmidlin, P. Graber, [43] M. Rewienski, A trajectory piecewise-linear approach to model
and P. Niederer, “Material characterization of the pig kidney order reduction of nonlinear dynamical systems [Ph.D. thesis],
in relation with the biomechanical analysis of renal trauma,” MIT, Cambridge, Mass, USA, 2003.
Journal of Biomechanics, vol. 32, no. 4, pp. 417–425, 1999. [44] E. Grimme, Krylov projection methods for model reduction
[26] N. Ayache and H. Delingette, “Soft tissue modeling for surgery [Ph.D. thesis], University of Illinois at Urbana-Champaign,
simulation,” in Computational Models for the Human Body, pp. Champaign, Ill, USA, 1997.
453–550, Elsevier, 2004. [45] K. Glover, “All optimal Hankel-norm approximations of linear
[27] S. Blemker, V. Hing, R. Fedkiw, and J. Teran, “Finite volume multivariable systems and their 𝐿∞ -error bounds,” Interna-
methods for the simulation of skeletal muscle,” in Proceedings tional Journal of Control, vol. 39, no. 6, pp. 1115–1193, 1984.
of the Eurographics/SIGGRAPH Symposium on Computer Ani- [46] Y. C. Liang, H. P. Lee, S. P. Lim, W. Z. Lin, K. H. Lee, and C.
mation (SCA ’03), pp. 68–74, San Diego, Calif, USA, 2003. G. Wu, “Proper orthogonal decomposition and its applications.
[28] K. Miller, G. Joldes, D. Lance, and A. Wittek, “Total Lagrangian I. Theory,” Journal of Sound and Vibration, vol. 252, no. 3, pp.
explicit dynamics finite element algorithm for computing soft 527–544, 2002.
tissue deformation,” Communications in Numerical Methods in [47] J. Lumley, G. Berkooz, and P. Holmes, “The proper orthogonal
Engineering, vol. 23, no. 2, pp. 121–134, 2007. decomposition in the analysis of turbulent flows,” Annual
[29] Z. A. Taylor, O. Comas, M. Cheng et al., “On modelling of Review of Fluid Mechanics, vol. 25, pp. 539–575, 1993.
anisotropic viscoelasticity for soft tissue simulation: numerical [48] T. Kim, “Frequency-domain Karhunen-Lóeve method and its
solution and GPU execution,” Medical Image Analysis, vol. 13, application to linear dynamic systems,” American Institute of
no. 2, pp. 234–244, 2009. Aeronautics and Astronautics Journal, vol. 36, no. 11, pp. 2117–
[30] S. M. BaniHani and S. De, “A comparison of some model order 2123, 1998.
reduction methods for fast simulation of soft tissue response [49] P. Holmes, J. L. Lumley, and G. Berkooz, Turbulence, Coherent
using the point collocation-based method of finite spheres,” Structures, Dynamical Systems and Symmetry, Cambridge Uni-
Engineering with Computers, vol. 25, no. 1, pp. 37–47, 2009. versity Press, New York, NY, USA, 1996.
[31] Y. C. Yoon, S. H. Lee, and T. Belytschko, “Enriched meshfree [50] J. L. Lumley, Stochastic Tools in Turbulence, Academic Press,
collocation method with diffuse derivatives for elastic fracture,” New York, NY, USA, 1971.
Computers & Mathematics with Applications, vol. 51, no. 8, pp.
1349–1366, 2006.
[32] S. H. Lee and Y. C. Yoon, “Meshfree point collocation method
for elasticity and crack problems,” International Journal for
Numerical Methods in Engineering, vol. 61, no. 1, pp. 22–48,
2004.
[33] M. A. Puso, J. S. Chen, E. Zywicz, and W. Elmer, “Meshfree and
finite element nodal integration methods,” International Journal
for Numerical Methods in Engineering, vol. 74, no. 3, pp. 416–
446, 2008.
[34] S. De, Y. J. Lim, M. Manivannan, and M. A. Srinivasan,
“Physically realistic virtual surgery using the point-associated
finite field (PAFF) approach,” Presence, vol. 15, no. 3, pp. 294–
308, 2006.
[35] K. J. Bathe, Finite Element Procedures, Prentice-Hall, Englewood
Cliffs, NJ, USA, 1996.
[36] O. Zienkiewicz and R. Taylor, The Finite Element Method Set,
Butterworth-Heinemann, London, UK, 6th edition, 2005.
[37] Z. Gao, K. Lister, and J. P. Desai, “Constitutive modeling
of liver tissue: experiment and theory,” Annals of Biomedical
Engineering, vol. 38, no. 2, pp. 505–516, 2010.
[38] R. Ogden, Non-Linear Elastic Deformations, Halsted Press, New
York, NY, USA, 1984.
Hindawi Publishing Corporation
Mathematical Problems in Engineering
Volume 2013, Article ID 524562, 10 pages
http://dx.doi.org/10.1155/2013/524562

Research Article
Homotopy Iteration Algorithm for Crack Parameters
Identification with Composite Element Method

Ling Huang, Zhongrong Lv, Weihuan Chen, and Jike Liu

Department of Applied Mechanics, Sun Yat-sen University, Guangzhou, Guangdong 510006, China

Correspondence should be addressed to Zhongrong Lv; [email protected]

Received 19 July 2013; Revised 12 October 2013; Accepted 12 October 2013

Academic Editor: Timon Rabczuk

Copyright © 2013 Ling Huang et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

An approach based on homotopy iteration algorithm is proposed to identify the crack parameters in beam structures. In the
forward problem, a fully open crack model with the composite element method is employed for the vibration analysis. The dynamic
responses of the cracked beam in time domain are obtained from the Newmark direct integration method. In the inverse analysis, an
identification approach based on homotopy iteration algorithm is studied to identify the location and the depth of a cracked beam.
The identification equation is derived by minimizing the error between the calculated acceleration response and the simulated
measured one. Newton iterative method with the homotopy equation is employed to track the correct path and improve the
convergence of the crack parameters. Two numerical examples are conducted to illustrate the correctness and efficiency of the
proposed method. And the effects of the influencing parameters, such as measurement time duration, measurement points, division
of the homotopy parameter and measurement noise, are studied.

1. Introduction There exist many reports on damage detection using

structural dynamic response. Some of the methods are based
Identification of crack parameters using the dynamic on the time domain [5–8]. Cattarius and Inman [5] used
responses with finite element method (FEM) has been the time histories of vibration response of the structure to
studied extensively for many years. In order to improve identify damage in smart structures. Lu and Liu [6] made use
the degree of accuracy of vibration analysis, most of the of the dynamic responses of bridge under a moving vehicle to
approaches [1, 2] are based on densely refining finite element identify both the local damage of bridge and vehicular param-
mesh. But these methods may lead to a large amount of eters. Lu and Law [7] proposed an approach for structural
computations. Unlike the classical FEM, Zeng [3] proposed damage identification based on response sensitivity analysis.
a composite element method (CEM) combining the conven- However, the response sensitivity approach is a local con-
tional FEM and classical theory (CT) to solve the structural vergence approach, and it requires the initial values of the
dynamical problems. And Lu and Law [4] further improve unknowns to be close to the true values. Then, Lu et al. [8]
the CEM using certain special boundary conditions of the proposed a two-step approach based on mode shape curva-
beam. It is known that the CEM has two available approaches, ture and response sensitivity analysis for crack identification.
the ℎ-version and the 𝑐-version. As the former by increasing As both the crack location and depth are unknowns in the
the number of element mesh is just like that of 𝑝-version of identification, a method with large convergence range should
the FEM [3], the latter by increasing the number of analytical be sought. The homotopy method [9–12] which is based on
functions will reduce the number of degrees of freedom in the the concept of homotopy is a widely convergent method
FEM. And the latter can obtain the fine approximate solution for solving system of equations. It has been extended for
with less computational effort in structural dynamics. Thus, other algorithm and applied in many fields; for example, He
a super convergence equation of motion of the structure can [10] proposed a homotopy perturbation method, which com-
be established by using the 𝑐-version of CEM. bines the homotopy in topology and traditional perturbation
2 Mathematical Problems in Engineering

xcj xci
xl xi1 2lc
xci g(t)
y
x hci
xi2

Figure 1: A simply supported beam with cracks. Figure 2: Variation on flexural rigidity 𝐸𝐼(𝑥) adjacent to a crack.

method, to provide an approximate solution to a wide range cracked cross sections, respectively. 𝑤 and 𝑑 are the width and
of linear and nonlinear problems. Liao [11] proposed a homo- height of the beam, respectively. 𝑥𝑖1 = 𝑥𝑐𝑖 − 𝑙𝑐 and 𝑥𝑖2 = 𝑥𝑐𝑖 +
topy analysis method for solving highly nonlinear problems 𝑙𝑐 are the positions where the reduction of the flexural rigidity
in science, finance, and engineering areas. Alexander and begins and finishes, respectively. If the beam has multiple
Yorke [12] employed a homotopy continuation method to cracks, the same procedure can be followed to calculate the
solve the fixed points and singularities of vector fields and flexural rigidity of other cracks.
bifurcation problems.
In this paper, an approach for identifying the parameters 2.2. Dynamic Response of the Structure with CEM. As shown
of cracks in a cracked beam based on homotopy iteration in Figure 1, the beam is discretized into one element together
algorithm is presented. A fully open crack model with com- with several terms eigenfunction of classical theory. Lu and
posite element model is adopted to establish the dynamics Law [4, 7] have proved that only using one finite element is
equation of the Euler-Bernoulli beam system. In inverse effective in the 𝑐-version technique of CEM. And it can reduce
problem, the identification equation is derived by the min- the total number of degrees of freedom in the finite element
imization of the error between the calculated acceleration model. For the crack identification, what is important is that
response and the simulated measured one. The equation is we do not need to judge whether the cracks affect one or
solved iteratively to identify the crack parameters. Mean- more elements stiffness as the crack(s) is always in one finite
while, Newton’s method with the homotopy equation is used element for the entire beam. Otherwise, it needs to calculate
to track the correct path and improve the convergence of the the stiffness distribution due to the crack(s) along the beam in
crack parameters. A simply supported beam and a cantilever each iteration this one-beam-one-element strategy simplifies
beam are studied to illustrate the accuracy and efficiency of greatly the process in crack identification.
the proposed method. Both single and multiple cracks in the In the CEM, the displacement field 𝑑CEM is combined
beam can be identified successfully using several measured with the FEM part and the CT part. The FEM part of the
acceleration responses. And the effects of measurement time displacement field should satisfy the nodal boundary con-
duration, measurement points, division of the homotopy ditions, and it can be expressed as the product of the shape
parameter, and measurement noise on the identified results function matrix N(𝑥) and the nodal displacement vector 𝑞(𝑡),
are investigated.
𝑑FEM (𝑥, 𝑡) = N (𝑥) 𝑞 (𝑡) , (2)
2. Forward Problem where
2.1. Crack Model. Figure 1 shows a simple beam with multiple 𝑥2 𝑥3 𝑥2 𝑥3
fully opened cracks along its length. It is assumed that the N (𝑥) = [1 − 3 2
+ 2 3,𝑥 − 2 + 2,
𝐿 𝐿 𝐿 𝐿
cracks have uniform depth across the width of the beam (3)
and that the mass distribution along the beam does not 𝑥2 𝑥3 𝑥 3 𝑥2
change. The stiffness, for the simplicity, is assumed to be 3 2 − 2 3, 2 − ],
𝐿 𝐿 𝐿 𝐿
varied linearly with a triangular reduction in a local region as
Sinha et al. [13] proposed. That is, the crack only leads to local where 𝐿 is the length of the beam.
stiffness reduction in a specified length adjacent to the crack. The second part can be obtained by the linear combina-
The effective length of the stiffness reduction for the crack is tion of the multiplication of analytical shape function 𝜙(𝑥)
𝑙𝑐 = 1.5𝑑. As shown in Figure 2, the flexural rigidity 𝐸𝐼(𝑥) with a vector of coefficient 𝑐(𝑡)
adjacent to the 𝑖th crack can be written as
𝑛
(𝑥 − 𝑥𝑖1 ) 𝑑CT (𝑥, 𝑡) = ∑𝜙𝑟 (𝑥) 𝑐𝑟 (𝑡) , (𝑟 = 1, 2, . . . , 𝑛) , (4)
{
{ 𝐸𝐼0 − 𝐸 (𝐼0 − 𝐼𝑐𝑖 ) , 𝑥𝑖1 ≤ 𝑥 ≤ 𝑥𝑐𝑖 ,
{
{ (𝑥𝑖 − 𝑥𝑖1 )
𝑟=1
{
𝐸𝐼 (𝑥) = { where 𝜙𝑟 (𝑥) is a special value which can be obtained accord-
{
{ (𝑥 − 𝑥)
{
{𝐸𝐼0 − 𝐸 (𝐼0 − 𝐼𝑐𝑖 ) 𝑖2 ing to the boundary conditions of the beam and 𝑛 is the
, 𝑥𝑐𝑖 ≤ 𝑥 ≤ 𝑥𝑖2 ,
{ (𝑥𝑖2 − 𝑥𝑖 ) number of the mode functions used from the CT. Such that,
(1) for a simply supported beam, 𝜙𝑟 (𝑥) = sin(𝑟𝜋𝑥/𝐿) can be
selected as Lu and Law [4] proposed. And the coefficient 𝑐𝑟 (𝑡)
where 𝐸 is Young’s modulus and 𝐼0 = 𝑤𝑑3 /12 and 𝐼𝑐𝑖 = denotes the contribution of the 𝜙𝑟 (𝑥) in the total displace-
𝑤(𝑑 − ℎ𝑐𝑖 )3 /12 are the moments of inertia of the intact and ment field. For a uniform beam, less terms of the shapes
Mathematical Problems in Engineering 3

function in CT also can be obtained with accuracy results [4]. The generalized force vector is G(𝑡) = S(𝑥𝑙 )𝑇 𝑔(𝑡), where
And the number of terms 𝑛 can be determined by a frequency 𝑔(𝑡) is an external force acting at the location of 𝑥𝑙 from the
convergence test as the previous study by Lu and Law [7]. The left support.
frequency convergence criterion is defined as
󵄨󵄨 Δ𝜔𝑛 󵄨󵄨 3. Inverse Problem
󵄨 󵄨
max 󵄨󵄨󵄨󵄨 𝑛𝑟 󵄨󵄨󵄨󵄨 < Tol, (5)
𝑟=1,2,...,10 󵄨 𝜔 󵄨
󵄨 𝑟 󵄨 3.1. Objective Function. The problem of crack identification
can be treated as an optimization problem. In the identifica-
where 𝜔𝑟𝑛is the estimation of the 𝑟th circular frequency with
tion, the unknowns consist of the crack location 𝑥𝑐 and crack
𝑛-terms in the CT. Δ𝜔𝑟𝑛 = 𝜔𝑟𝑛 − 𝜔𝑟𝑛−1 is the difference of the depth ℎ𝑐 ; that is, 𝛼 = [𝑥𝑐 , ℎ𝑐 ]. If there are 𝑚 measured points
𝑟th circular frequency obtained with the 𝑛-terms and (𝑛 − 1)- in the beam structure, the measured acceleration response
terms. Tol is the tolerance value.
can be expressed as d̈∗ = (d̈1 , d̈2 , . . . , d̈𝑚 ). The objective func-
The displacement field of the CEM for a uniform Euler-
tion for the optimization problem can be derived by mini-
Bernoulli beam element can be written as
mizing the error between the calculated acceleration response
𝑑CEM (𝑥, 𝑡) = 𝑑FEM (𝑥, 𝑡) + 𝑑CT (𝑥, 𝑡) = S (𝑥) Q (𝑡) , (6) and the simulated measured one as

where S(𝑥) = [N(𝑥), 𝜙1 (𝑥), 𝜙2 (𝑥), . . . , 𝜙𝑛 (𝑥)] is the general- 1 𝑛𝑡 𝑚 2

𝐽 = ∑∑ (d̈(𝛼, 𝑡) − d̈∗ (𝛼, 𝑡)) ,
ized shape function of CEM. 2 𝑖=1𝑗=1 (11)
The stiffness matrix of the cracked beam can be obtained
from the following equation: 𝑖 = 1, 2, . . . , 𝑛𝑡, 𝑗 = 1, 2, . . . , 𝑚,
𝐿
𝑑2 S(𝑥)𝑇 𝑑2 S (𝑥) where 𝑡 denotes the the measured time instants, 𝑛𝑡 = 𝑡/𝑑𝑡 is
K=∫ 𝐸𝐼 (𝑥) 𝑑𝑥. (7) the number of time step, and 𝑑𝑡 is the increment of time step.
0 𝑑𝑥2 𝑑𝑥2
The inverse problem can be solved by minimizing the
For a beam with multiple cracks, as the whole crack beam objective function. Taking partial derivative with respect to
is represented as a single finite element, the whole length the identification vector in (11), we have
of the beam can be divided according to the varied flexural
rigidity. For 𝐸𝐼(𝑥) adjacent to the 𝑖th crack can be obtained {
𝑛𝑡 𝑚
𝜕d̈(𝛼, 𝑡)
{
{
{ ∑ ∑ (d̈(𝛼, 𝑡) − d̈∗ (𝛼, 𝑡)) = 0,
from (1); the global stiffness matrix K can be obtained by { 𝑖=1𝑗=1 𝜕𝑥𝑐
superposition. 𝑓 (𝛼) = { 𝑛𝑡 𝑚 (12)
{
{ 𝜕d̈(𝛼, 𝑡)
{ ∑∑ (d̈(𝛼, 𝑡) − d̈ (𝛼, 𝑡))
{ ∗
As the effect of crack on the mass of the beam is neglected = 0,
and the consistent mass matrix of the cracked beam can be 𝜕ℎ𝑐
{ 𝑖=1𝑗=1
obtained from the following equation:
̈ 𝑡)/𝜕𝑥 and 𝜕d(𝛼,
where 𝜕d(𝛼, ̈ 𝑡)/𝜕ℎ are the acceleration
𝐿 𝑐 𝑐
𝑇
M = ∫ S(𝑥) 𝜌𝐴S (𝑥) 𝑑𝑥, (8) response sensitivities with respect to the crack location and
0 depth, respectively.
where 𝜌 is the mass density 𝐴 is the area of the cross section.
After introducing the boundary condition, the equation 3.2. Homotopy Iteration Algorithm for Crack Identification.
of motion of the forced vibration of the cracked Euler- The basic idea of homotopy iteration algorithm is to introduce
Bernoulli beam is expressed as a parameter 𝜆 into the nonlinear equation 𝑓(𝛼) = 0 to
construct a family of homotopy mapping 𝐻(𝛼, 𝜆), where
MQ̈ + CQ̇+ KQ = G (𝑡) . (9) 𝜆 ∈ [0, 1] is named the homotopy parameter. The homotopy
equation is assumed to be 𝐻(𝛼, 𝜆) = 𝜆𝑓(𝛼) + (1 − 𝜆)𝑓0 (𝛼)
In this study, Rayleigh damping model [14] is adopted; in this paper. When 𝜆 = 1, 𝐻 is the mapping of 𝑓(𝛼) as (12)
that is, C = 𝑎1 M + 𝑎2 K, where 𝑎1 and 𝑎2 are two constants shows, and, when 𝜆 = 0, 𝐻 is assumed to be the mapping
obtained from two different undamped natural frequencies, of 𝑓0 (𝛼). And 𝑓0 (𝛼) is set to be similar to 𝑓(𝛼) but, related
𝜔𝑖 and 𝜔𝑗 , and their associated modal damping ratios, 𝜉𝑖 and ̈ 𝑡) −
to the initial point, can be expressed as 𝑓0 (𝛼) = (d(𝛼,
𝜉𝑗 , with the expression of 𝑎1 = 2𝜔𝑗 𝜔𝑖 (𝜔𝑗 𝜉𝑖 − 𝜔𝑖 𝜉𝑗 )/(𝜔𝑗 2 − 𝜔𝑖 2 ) ̈ , 𝑡))𝜕d(𝛼,
d(𝛼 ̈ 𝑡)/𝜕𝛼, where 𝛼 is the arbitrarily selected initial
0 0
and 𝑎2 = 2(𝜔𝑗 𝜉𝑗 − 𝜔𝑖 𝜉𝑖 )/(𝜔𝑗 2 − 𝜔𝑖 2 ). The vector of generalized value vector. In this way, a family of homotopy mapping 𝐻
acceleration Q,̈ velocity Q,̇and displacement Q of the struc- can be constructed instead of a single mapping 𝑓. Thus, the
ture can be obtained from (9) by direct integration. Further, homotopy equation can meet the following equation:
the vector of physical acceleration d(𝑥), ̈ ̇ and
velocity d(𝑥), 𝐻 (𝛼, 𝜆) = 𝜆𝑓 (𝛼) + (1 − 𝜆) 𝑓0 (𝛼) = 0, (13)
displacement d(𝑥) can be obtained from
𝐻 (𝛼, 0) = 𝑓0 (𝛼) = 0, (14)
d̈(𝑥) = S (𝑥) Q,̈ (10a)
𝐻 (𝛼, 1) = 𝑓 (𝛼) = 0. (15)
ḋ(𝑥) = S (𝑥) Q,̇ (10b)
Equations (14) and (15) show that as the homotopy parameter
d (𝑥) = S (𝑥) Q. (10c) 𝜆 increases from zero to one, the homotopy equation 𝐻(𝛼, 𝜆)
4 Mathematical Problems in Engineering

varies from 𝑓0 (𝛼) to 𝑓(𝛼). Hence, the parameters of the crack respect to the crack parameter should be obtained first. We
can be obtained by following the variations of parameter 𝜆, take the derivative on both sides of (9) with respect to the
and, when 𝐻(𝛼, 𝜆) = 0, the identification vector 𝛼 is in the crack location and crack depth; then we have
homotopy path.
After obtaining the homotopy equation, the Newton iter- 𝜕Q̈ 𝜕Q̇ 𝜕Q 𝜕K ̇ 𝜕K
M +C +K = −𝑎2 Q− Q,
ative method is used to track the path in the process of identi- 𝜕𝑥𝑐 𝜕𝑥𝑐 𝜕𝑥𝑐 𝜕𝑥𝑐 𝜕𝑥𝑐
fication. The solving procedure is explained as follows. (18)
𝜕Q̈ 𝜕Q̇ 𝜕Q 𝜕K ̇ 𝜕K
M +C +K = −𝑎2 Q− Q,
Step 1. Guess an initial value vector of crack parameters 𝛼0 = 𝜕ℎ𝑐 𝜕ℎ𝑐 𝜕ℎ𝑐 𝜕ℎ𝑐 𝜕ℎ𝑐
[𝑥𝑐0 , ℎ𝑐0 ], divide the range of homotopy parameter 𝜆 ∈ [0, 1]
into 𝑛 equal parts, and then obtain 𝑑𝜆 = 1/𝑛, let 𝜆 0 = 0. where 𝜕Q/𝜕𝑥 ̈ 𝑐 , 𝜕Q/𝜕𝑥
̈ 𝑐 , 𝜕Q/𝜕ℎ ̇ 𝑐 , 𝜕Q/𝜕ℎ
̇ 𝑐 , 𝜕Q/𝜕𝑥𝑐 , and
𝜕Q/𝜕ℎ𝑐 are the vectors of generalized acceleration, velocity,
Step 2. Trace the homotopy path from the initial value (𝛼0 , and displacement sensitivities with respect to the crack loca-
𝜆 0 ) and give an increment 𝑑𝜆 to the homotopy parameter 𝜆 tion and depth, respectively. 𝜕K/𝜕𝑥𝑐 and 𝜕K/𝜕ℎ𝑐 are the first
as 𝜆 1 = 𝜆 0 + 𝑑𝜆. partial derivatives of the stiffness matrix with respect to the
crack parameters. Since the flexural rigidity 𝐸𝐼(𝑥) involves
Step 3. Using Newton iterative method to calculate an the crack location and depth and the global stiffness matrix
updated vector of crack parameter 𝛼1 = [𝑥𝑐1 , ℎ𝑐1 ] in the K can be obtained from (7), the derivatives 𝜕K/𝜕𝑥𝑐 and
homotopy path, we have 𝜕K/𝜕ℎ𝑐 can be obtained directly. As the dynamic responses
Q and Q̇have been calculated from (9), the right-hand
𝜕𝐻(𝛼, 𝜆 1 ) 󵄨󵄨󵄨󵄨
−1
side of (18) can be considered as a form of external force.
𝛼1 = 𝛼0 − ( 󵄨󵄨 ) 𝐻 (𝛼0 , 𝜆 1 ) , (16)
𝜕𝛼 󵄨󵄨𝛼=𝛼0 Thus, the dynamic response sensitivity (i.e., the generalized
acceleration response sensitivity, velocity response sensitivity,
where 𝜕𝐻(𝛼, 𝜆 1 )/𝜕𝛼 is the homotopy equation with respect and displacement response sensitivity) can also be calculated
to the crack parameters. The value of 𝐻(𝛼0 , 𝜆 1 ) and 𝜕𝐻(𝛼, from the Newmark integration method. Then, the physical
𝜆 1 )/𝜕𝛼 will be introduced in next section. acceleration response sensitivity can be obtained as

Step 4. Check whether the updated crack parameter vector 𝜕d̈ 𝜕Q̈
= S (𝑥) , (19a)
𝛼1 = [𝑥𝑐1 , ℎ𝑐1 ] is physically meaningful or not. If not, the 𝜕𝑥𝑐 𝜕𝑥𝑐
result is considered diverged, then stop iteration and move to
Step 1 to start with a new initial value or division of homotopy 𝜕d̈ 𝜕Q̈
= S (𝑥) . (19b)
parameter. If otherwise, go to the next step. 𝜕ℎ𝑐 𝜕ℎ𝑐

Step 5. Let 𝛼0 = 𝛼1 ; repeat from Steps 3 to 4, until the Furthermore, the objective function in (12) and homotopy
obtained (𝛼1 , 𝜆 1 ) in homotopy path satisfies the following equation in (13) can then be obtained.
convergence conditions: To obtain the first derivative of the homotopy equation
with respect to the crack parameters, we should apply the first
󵄩󵄩 𝛼 − 𝛼 󵄩󵄩 derivative to both sides of (18) with respect to the crack para-
󵄩󵄩 1 0󵄩󵄩󵄩 ≤ Tol1,
󵄩󵄩 (17a)
󵄩󵄩 𝛼1 󵄩󵄩󵄩 meters as
𝜕2 Q̈ 𝜕2 Q̇ 𝜕2 Q
where Tol1 is the first tolerance value for convergence and is M + C +K 2
taken as 1.0 × 10−3 in this study. 𝜕𝑥𝑐2 𝜕𝑥𝑐2 𝜕𝑥𝑐

𝜕 2 K ̇ 𝜕2 K 𝜕K 𝜕Q̇ 𝜕K 𝜕Q
Step 6. Repeat the Newton iterative again, until the following = −𝑎2 Q − Q − 2 (𝑎2 + ),
convergence criterion is satisfied. That is, the identification 𝜕𝑥𝑐2 𝜕𝑥𝑐2 𝜕𝑥𝑐 𝜕𝑥𝑐 𝜕𝑥𝑐 𝜕𝑥𝑐
parameters are assumed to in the homotopy path when
𝐻(𝛼1 , 𝜆 1 ) meets the following criterion: 𝜕2 Q̈ 𝜕2 Q̇ 𝜕2 Q
M +C +K
𝜕𝑥𝑐 𝜕ℎ𝑐 𝜕𝑥𝑐 𝜕ℎ𝑐 𝜕𝑥𝑐 𝜕ℎ𝑐
󵄩󵄩 󵄩
󵄩󵄩𝐻 (𝛼1 , 𝜆 1 )󵄩󵄩󵄩 ≤ Tol2, (17b)
𝜕 2 K ̇ 𝜕2 K 𝜕K 𝜕Q̇ 𝜕K 𝜕Q
= −𝑎2 Q− Q − 𝑎2 −
where Tol2 is the second tolerance value and is taken to be 0.1 𝜕𝑥𝑐 𝜕ℎ𝑐 𝜕𝑥𝑐 𝜕ℎ𝑐 𝜕ℎ𝑐 𝜕𝑥𝑐 𝜕ℎ𝑐 𝜕𝑥𝑐
for all the study cases.
𝜕K 𝜕Q̇ 𝜕K 𝜕Q
− 𝑎2 − ,
Step 7. The pair of unknowns (𝛼1 , 𝜆 1 ) is taken as the new 𝜕𝑥𝑐 𝜕ℎ𝑐 𝜕𝑥𝑐 𝜕ℎ𝑐
set of initial value (𝛼0 , 𝜆 0 ) and repeat Steps 2 to 5 until the
homotopy parameter 𝜆 reaches 1.0. The final 𝛼1 would then 𝜕2 Q̈ 𝜕2 Q̇ 𝜕2 Q
M 2
+C 2 +K 2
be the required set of identification results. 𝜕ℎ𝑐 𝜕ℎ𝑐 𝜕ℎ𝑐

3.3. Homotopy Equation and Its First Derivative. In order to 𝜕 2 K ̇ 𝜕2 K 𝜕K 𝜕Q̇ 𝜕K 𝜕Q

= −𝑎2 2
Q − 2 Q − 2 (𝑎2 + ),
obtain the homotopy equation, the acceleration response with 𝜕ℎ𝑐 𝜕ℎ𝑐 𝜕ℎ𝑐 𝜕ℎ𝑐 𝜕ℎ𝑐 𝜕ℎ𝑐
Mathematical Problems in Engineering 5

Table 1: Comparison of the convergence property with different number of terms in CT.

Number of Natural frequencies in Hz (𝑥𝑐 = 1000 mm, ℎ𝑐 = 1 mm)

term in CT 1 2 3 4 5 6 7 8
4 14.65 58.73 131.86 234.92 592.11 890.33 53510.81 58925.59
5 14.65 58.73 131.86 234.92 366.28 890.33 1252.84 58925.59
6 14.65 58.73 131.86 234.92 366.28 528.55 1252.83 1680.70
9 14.65 58.73 131.86 234.92 366.28 528.56 717.93 939.60
10 14.65 58.73 131.86 234.92 366.28 528.56 717.93 939.60

𝜕2 Q̈ 𝜕2 Q̇ 𝜕2 Q 𝜕𝑓 (𝛼) [ 𝑛𝑡 𝑚 𝜕2 d̈ 𝜕d̈ 𝜕d̈

M +C +K = ∑ ∑ ((d̈− d̈∗ ) + ),
𝜕ℎ𝑐 𝜕𝑥𝑐 𝜕ℎ𝑐 𝜕𝑥𝑐 𝜕ℎ𝑐 𝜕𝑥𝑐 𝜕ℎ𝑐 𝜕𝑥𝑐 𝜕ℎ𝑐 𝜕ℎ𝑐 𝜕𝑥𝑐
𝑖=1 𝑗=1
[
𝜕2 K ̇ 𝜕2 K 𝜕K 𝜕Q̇ 𝜕K 𝜕Q (24)
= −𝑎2 Q− Q − 𝑎2 − 2
𝜕ℎ𝑐 𝜕𝑥𝑐 𝜕ℎ𝑐 𝜕𝑥𝑐 𝜕𝑥𝑐 𝜕ℎ𝑐 𝜕𝑥𝑐 𝜕ℎ𝑐 𝑛𝑡 𝑚
𝜕2 d̈ 𝜕d̈
∑∑ ((d̈− d̈∗ ) 2 + ( ) )] .
𝜕ℎ𝑐 𝜕ℎ𝑐
𝜕K 𝜕Q̇ 𝜕K 𝜕Q 𝑖=1 𝑗=1
]
− 𝑎2 − ,
𝜕ℎ𝑐 𝜕𝑥𝑐 𝜕ℎ𝑐 𝜕𝑥𝑐
The first derivatives 𝜕𝑓0 (𝛼)/𝜕𝑥𝑐 and 𝜕𝑓0 (𝛼)/𝜕ℎ𝑐 can also be
(20) obtained in a similar way.

where 𝜕2 K/𝜕𝑥𝑐2 , 𝜕2 K/𝜕𝑥𝑐 𝜕ℎ𝑐 , 𝜕2 K/𝜕ℎ𝑐2 , and 𝜕2 K/𝜕ℎ𝑐 𝜕𝑥𝑐 are

the second partial derivative of the stiffness matrix with 4. Numerical Simulations
respect to the crack parameters, and 𝜕2 K/𝜕𝑥𝑐 𝜕ℎ𝑐 = 𝜕2 K/
̇ 𝑐 , and 4.1. A Simply Supported Beam. A uniform cross section sim-
𝜕ℎ𝑐 𝜕𝑥𝑐 . Since the response sensitivities 𝜕Q/𝜕ℎ𝑐 , 𝜕Q/𝜕𝑥
̇ ply supported beam with cracks is studied as shown in
𝜕Q/𝜕ℎ𝑐 have been calculated from (18), the second partial
Figure 1. The parameters of the beam are: Young’s modulus
derivative of generalized acceleration response with respect
𝐸 = 69.79 GPa, mass density 𝜌 = 2600 kg/m3 , length 𝐿 =
to the unknown variables (𝑥𝑐 , ℎ𝑐 ) can be calculated by the
2000 mm, width 𝑤 = 50 mm, and height ℎ = 25 mm. An
Newmark integration with (20). Then, the physical acceler-
impulsive force acts at the 700 mm from the left support in
ation response sensitivity can be obtained as
the negative 𝑦-direction from 𝑡 = 0 s to 𝑡 = 0.04 s with
𝜕2 d̈ 𝜕2 Q̈
= S (𝑥) , (21a) 10000𝑡𝑁, (0 s ≤ 𝑡 ≤ 0.02 s) ,
𝜕𝑥𝑐2 𝜕𝑥𝑐2 𝑔 (𝑡) = { (25)
10000 (0.04 − 𝑡) 𝑁, (0.02 s < 𝑡 ≤ 0.04 s) .
𝜕2 d̈ 𝜕2 Q̈
2
= S (𝑥) 2 , (21b) In calculating the dynamic response, the time step is 0.001 s.
𝜕ℎ𝑐 𝜕ℎ𝑐
Artificial measurement noise with different levels is added to
𝜕2 d̈ 𝜕2 d̈ 𝜕2 Q̈ the calculated responses to simulate the “measured” struc-
= = S (𝑥) . (21c) tural responses. All the time history data of 1.0 second are
𝜕𝑥𝑐 𝜕ℎ𝑐 𝜕ℎ𝑐 𝜕𝑥𝑐 𝜕𝑥𝑐 𝜕ℎ𝑐
used in the identification unless otherwise specified. The two
Furthermore, the first derivative of the homotopy map- damping coefficients used for calculating Rayleigh damping
ping in (13) with respect to the unknown vector 𝜕𝐻(𝛼, 𝜆)/𝜕𝛼 matrix are both assumed to be 0.01.
can be obtained as Using the traditional FEM, the first eight frequencies of
the beam are 14.65, 58.73, 131.85, 234.95, 366.31, 528.84, 718.45,
𝜕𝑓 (𝛼) 𝜕𝑓 (𝛼) and 941.15 HZ. Comparison with Table 1, one can find that
𝜆 + (1 − 𝜆) 0
𝜕𝐻 (𝛼, 𝜆) [
[ 𝜕𝑥 𝑐 𝜕𝑥𝑐 ] ] when the number of term in CT is 𝑛, only the first 𝑛 frequency
=[ ], (22) is convergence for the uniform cross section beam. When
𝜕𝛼 [ 𝜕𝑓 (𝛼) 𝜕𝑓0 (𝛼) ]
𝜆 + (1 − 𝜆) only 10 modes are used, the first 8 natural frequencies can
[ 𝜕ℎ𝑐 𝜕ℎ𝑐 ] be obtained with good accuracy (the max error for the 8th
natural frequency is only 0.17%). Indeed, when more terms
where
of the harmonic functions are used, the accuracy can further
2 be improved, but more computational time is needed in
𝜕𝑓 (𝛼) [ 𝑛𝑡 𝑚 𝜕2 d̈ 𝜕d̈
= ∑∑ ((d̈− d̈∗ ) 2 + ( ) ), the calculation. When dynamic loads act on the beam, the
𝜕𝑥𝑐 𝜕𝑥𝑐 𝜕𝑥𝑐 dynamic responses of the beam will only include the first
[𝑖=1𝑗=1
(23) few lower modes; generally speaking, the contribution of the
higher modes on the dynamic responses can be neglected.
𝑛𝑡 𝑚
𝜕2 d̈ 𝜕d̈ 𝜕d̈ ]
∑ ∑ ((d̈− d̈∗ ) + ) , Thus, the number of the mode functions used from the CT
𝑖=1 𝑗=1 𝜕ℎ𝑐 𝜕𝑥𝑐 𝜕𝑥𝑐 𝜕ℎ𝑐 is set to be 10 in the following study.
]
6 Mathematical Problems in Engineering

Table 2: Effect of the initial value on crack identification.

Initial value (mm) True value (mm) Identified results (mm) Division Measurement No. of
Scenarios
𝑥𝑐0 ℎ𝑐0 𝑥𝑐 ℎ𝑐 𝑥𝑐 ℎ𝑐 of 𝜆 points (mm) iteration
1 1500 2 1720 10 1720.000000 10.000000 5 600, 1100, 1600 31
2 1500 2 1000 10 999.999990 10.000000 5 600, 1100, 1600 38
3 1500 2 1000 1 1000.000234 1.000000 3 600, 1100, 1600 26
4 260 3 1000 1 999.999997 1.000000 21 600, 1100, 1600 146

Table 3: Effect of measurement time.

Identified results (mm)
Scenarios Division of 𝜆 Measurement time (s) No. of iteration
𝑥𝑐 ℎ𝑐
1 1000.000234 1.000000 3 1 26
2 1000.000237 1.000000 3 2 30
3 1000.000187 1.000000 3 4 32

4.1.1. Effect of Initial Value on Crack Identification. In this 4.1.4. Effect of Division of Homotopy Parameter. In this study,
case, single crack identification is conducted to illustrate the effect of division of homotopy parameter on the identified
the proposed method. Three measurement points located at results is discussed. Other parameters are the same as those in
600 mm, 1100 mm, and 1600 mm from the left support are the last study except that the homotopy parameter is divided
used in the crack identification. Totally, four Scenarios are into 3, 4, and 5 parts. And the identified results for the study
studied as listed in Table 2. One can find that the crack param- are listed in Table 5. We can find that increase of division
eters have been identified successfully with high accuracy in parts of the homotopy parameter, the number of iteration will
all Scenarios. It should be pointed out as the initial crack increase accordingly in general but it has little effect on the
location is far away from the true crack location, the crack accuracy of identified results. Figure 5 shows the evolution of
can also be identified successfully but the number of iteration the crack parameters in the process of iteration for different
will increase accordingly. And in general, finer division division parts of the homotopy parameter; one can see that
in the homotopy parameter 𝜆 should be considered as to track the convergence of Newton’s method to track the homotopy
the correct path. Scenarios 3 and 4 show that the method is path is similar.
less affected by the initial values.
4.1.5. Effect of Measurement Noise. In this section, the effect
4.1.2. Effect of Measurement Time Duration. In this case, the of measurement noise on the accuracy of identified results is
effect of measurement time duration in crack identification taken into account. Again, the identification of Scenario 3 in
is studied. The identification of Scenario 3 in Table 2 is Table 2 is studied. The effect of measurement noise is simu-
reexamined; except the measurement time duration is taken lated as a normally distributed random error with zero mean
to be 1.0 s, 2.0 s, and 4.0 s, respectively. The crack parameters and a unit standard deviation is added to the calculated accel-
are identified with satisfactory accuracy as shown in Table 3. eration as
It also can be notes that longer time duration has little effect ̂̈= d̈+ 𝐸 × 𝑁 × var (d)
d ̈, (26)
𝑝 oise
on the degree of accuracy but will increase the number of
iteration. And the process of iteration for this study is shown
where d ̂̈ is the vectors of measured structural acceleration
in Figure 3; one can see that longer measurement time will
response, 𝐸𝑃 is the noise level, 𝑁oise is standard normal dis-
obtain more stable convergence pattern when tracking the
tribution vector with zero mean and unit standard deviation,
approximate values in the homotopy path.
and var(⋅) is the variance of the time history.
The relative errors of crack location and depth identifica-
4.1.3. Effect of Measurement Points. In this case, how the mea- tion are defined as
surement points affect the accuracy and the iterative process 󵄩󵄩 𝛼 − 𝛼 󵄩󵄩
󵄩 󵄩
is studied. Three, four, and six measurement points are used Relative error = 󵄩󵄩󵄩 id true 󵄩
󵄩 × 100%, (27)
󵄩󵄩 𝛼true 󵄩󵄩󵄩
for crack identification to give a comparison. And the other
parameters are the same as the last case. The identification where 𝛼id and 𝛼true are the identified and true values, respec-
results are listed in Table 4. It can be found that increasing tively. Table 6 gives a comparison of the identified results for
measurement points can improve the accuracy of the identi- 0%, 1%, 5%, and 15% noise level, respectively. This study shows
fication results. Figure 4 gives a comparison on the identified that the crack parameters have been identified successfully
process of iteration. It can be noted that different number even with 15% measurement noise. With the increase of the
of measurement points has the similar convergence process. measurement noise level, identification errors will become
And the proposed method does not need a large number of larger and the max identification error is 8.93% in the location
measurement points. of crack and 5.24% in the crack depth.
Mathematical Problems in Engineering 7

1500 2
1450
1400 1.8
1350
Crack location (mm)

Crack depth (mm)

1300 1.6
1250
1200 1.4
1150
1100 1.2
1050
1000 1
950
0 5 10 15 20 25 30 0 5 10 15 20 25 30
Number of iteration Number of iteration
t=1 t=1
t=2 t=2
t=4 t=4
(a) (b)

Figure 3: Effect of measurement time duration.

1500 2
1450
1400 1.8

1350
Crack location (mm)

Crack depth (mm)

1300 1.6
1250
1200 1.4
1150
1100 1.2
1050
1000 1
950
0 5 10 15 20 25 0 5 10 15 20 25
Number of iteration Number of iteration

m=3 m=3
m=4 m=4
m=6 m=6
(a) (b)

Figure 4: Effect of measurement points.

4.1.6. Identification of Multiple Cracks. The same simply sup- results and the relative errors for the cases with 0% and 5%
ported beam is studied with two cracks in this Section. The noise levels are considered. It can be noted that the identified
cracks are assumed to be located at 450 mm and 1000 mm results for the two cracks are as good as those for a single
from the left support of the beam, respectively. The depths crack when the measurement noise is free. Comparing the
of the crack are 5 mm and 1 mm, respectively. In the iden- identified results with measurement noise and the free noise
tification, the initial values of locations for these two cracks one, we can find that the effect of measurement noise on crack
are set to be 500 mm and 1500 mm, respectively, and depths identification is noticeable, but the identified results are still
are both set to be 2 mm. Six measurement points as the satisfactory. For 5% noise level, the maximum relative errors
Scenario 3 in Table 4 are used in this case. And the setting are 1.82% and 0.28% in the crack location and depth for the
of other parameters is shown in Table 7. The identification first crack, respectively and 6.49% and 7.22% for the second
8 Mathematical Problems in Engineering

Table 4: Effect of measurement points.

Identified results (mm)

Scenarios Measurement points (mm) No. of iteration
𝑥𝑐 ℎ𝑐
1 1000.000234 1.000000 600, 1100, 1600 26
2 1000.000005 1.000000 400, 800, 1200, 1600 25
3 999.999999 1.000000 300, 600, 900, 1200, 1500, 1800 25

Table 5: Effect of division of homotopy parameter.

Identified results (mm)

Scenarios Division of 𝜆 Measurement time (s) No. of iteration
𝑥𝑐 ℎ𝑐
1 1000.000234 1.000000 3 1 26
2 1000.000003 1.000000 4 1 28
3 999.9999346 1.000000 5 1 30

Table 6: Effect of measurement noise.

Identified results (mm)/rel. (%)
Scenarios Noise Division of 𝜆 No. of iteration
𝑥𝑐 ℎ𝑐
1 Nil 1000.000234/0.00 1.000000/0.00 3 26
2 1% 1021.498303/2.15 1.003232/0.32 3 22
3 5% 1070.401667/7.04 1.022658/2.27 3 23
4 15% 1089.318837/8.93 1.052358/5.24 3 19
Note: “Rel. (%)” denotes relative error.

Table 7: Identification of two cracks in a simply supported beam.

Identified results (mm)/ Identified results (mm)/

Scenarios Noise Division Measurement No. of
rel. (%) in crack 1 rel. (%) in crack 2
of 𝜆 time (s) iteration
𝑥𝑐1 ℎ𝑐1 𝑥𝑐2 ℎ𝑐2
1 Nil 450.000000/0.00 5.000000/0.00 999.999997/0.00 1.000000/0.00 5 1 78
2 5% 441.804646/1.82 4.986154/0.28 1064.867252/6.49 1.072199/7.22 7 1 74

1500 2
1450
1400 1.8
1350
Crack location (mm)

Crack depth (mm)

1300 1.6
1250
1200 1.4
1150
1100 1.2
1050
1000 1
950
0 5 10 15 20 25 30 0 5 10 15 20 25 30
Number of iteration Number of iteration

n=3 n=3
n=4 n=4
n=5 n=5
(a) (b)

Figure 5: Effect of division of homotopy parameter.

Mathematical Problems in Engineering 9

Table 8: Crack identification in a cantilever beam.

Identified results
Initial True
2s 4s
Single crack
Crack location 𝑥𝑐 (mm) 750 250 250.000162 250.000259
Crack depth ℎ𝑐 (mm) 3 1 1.000001 1.000001
Division of 𝜆 — — 26 20
No. of iteration — — 92 96
Two cracks
Crack locations 𝑥𝑐1 /𝑥𝑐2 (mm) 360/700 250/600 249.603356/599.789017 250.009037/600.005717
Crack depths ℎ𝑐1 /ℎ𝑐2 (mm) 2/7.8 1/6.5 0.997240/6.498047 1.000067/6.500058
Division of 𝜆 — — 16 15
No. of iteration — — 137 137

y xcj the homotopy parameter is divided into 𝑁 parts, we will have

xci g(t) 𝑁 + 1 homotopy equations as 𝛼 = 0, 1/𝑁, . . . , 1. When the
division of homotopy parameter is only one, the homotopy
x iteration algorithm decreases to a simple Newton algorithm.
But the Newton algorithm is with local convergence and is
easy to divergent. Thus, the division of homotopy parameter
Figure 6: A cantilever beam with cracks.
is an integer greater than or equal to 2. And with the
range between the initial parameters and the true parameters
crack, respectively. These results further illustrate the effec- increasing, the finer division of homotopy parameter should
tiveness of the proposed method. be considered to improve the convergence. But in some
subdivision, the Newton algorithm to track the path may be
divergent to the crack depth very close to zero or the height
4.2. A Cantilever Beam. A uniform cross section cantilever of the beam. Because in these two conditions, the homotopy
beam with cracks is studied as another example. As Figure 6 equation and its first derivative are so small that it leads to
shows, the physical parameters of the beam are: elastic modu- misjudgment. Thus, the new division or initial parameters
lus of material 𝐸 = 69.79 GPa, mass density 𝜌 = 2600 kg/m3 , should be considered. But according to authors’ calculation
length 𝐿 = 1000 mm, width 𝑤 = 50 mm and height ℎ = experience, when the subdivision increases to some value
25 mm. when calculating the dynamic response, a sinusoidal (greater than 30), the misjudgment will still occur as the
force 𝑔(𝑡) = 40 sin(15𝜋𝑡)𝑁 acts at the free end of the above two conditions exist. That is why increasing the sub-
beam. Four measurement points located at 300 mm, 450 mm, division over some larger value does not modify the conver-
600 mm, and 750 mm from the clamped end are used to gence properties of the algorithm.
record the acceleration data. The two damping coefficients
used for calculating Rayleigh damping matrix are both
assumed to be 0.01. The time step is 0.005 s and the time dura- 6. Conclusions
tion is taken to be 2.0 s and 4.0 s, respectively. The number
of the mode functions used from the CT is also set to be 10, In this paper, an open crack model with the CEM for Euler-
and the way to choose the number of term is the same as the Bernoulli beam system is adopted to establish the dynamics
simply supported beam. Both single crack and two cracks are equation in the forward problem. In the inverse problem,
simulated in this case. Good identification results are listed an identification approach based on homotopy iteration
in Table 8. For a single crack identification, the measurement algorithm is presented to identify the parameters of cracks.
of time duration may have little effect on the degree of Numerical simulation shows that the proposed method is less
accuracy as Section 4.1.2 shows. But for two cracks, it can affected by the initial values and is effective and accurate for
improve the accuracy of the identified results in general as the crack identification when the measurement noise is free. With
unknown parameters increases in the identification. This case the increase of the measurement noise, the identification
further illustrates the effectiveness of the proposed method. errors will become larger, but the identified results are still
satisfactory. The study also shows that more measurement
points can obtain more accuracy of the identification results
5. Discussions in general but has the similar convergence process.
The homotopy iteration algorithm is to divide the range of
homotopy parameter 𝜆 ∈ [0, 1] into multiple subintervals; Conflict of Interests
then the Newton algorithm is used to search the best conver-
gence solution in each subinterval to obtain the updated crack The authors declare that there is no conflict of interests
parameters and another homotopy equation. Assuming that regarding the publication of this paper.
10 Mathematical Problems in Engineering

Acknowledgments
This work is supported in part by the NSFC (11172333
and 11272361), the Fundamental Research Funds for the
Central Universities (13lgzd06), and the General Financial
Grant from the China Postdoctoral Science Foundation
(2013M531893). Such financial aids are gratefully acknowl-
edged.

References
[1] M. I. Friswell and J. E. Mottershead, Finite Element Model
Updating in Structural Dynamics, Kluwer Academic, Dodrecht,
The Netherlands, 1995.
[2] A. G. Peano, “Hierarchies of conforming finite elements for
plane elasticity and plate bending,” Computers and Mathematics
with Applications, vol. 2, no. 3-4, pp. 211–224, 1976.
[3] P. Zeng, “Composite element method for vibration analysis of
structure,” Journal of Sound and Vibration, vol. 218, no. 4, pp.
619–696, 1998.
[4] Z. R. Lu and S. S. Law, “Discussions on ‘composite element
method for vibration analysis of structure’,” Journal of Sound and
Vibration, vol. 305, no. 1-2, pp. 357–361, 2007.
[5] J. Cattarius and D. J. Inman, “Time domain analysis for damage
detection in smart structures,” Mechanical Systems and Signal
Processing, vol. 11, no. 3, pp. 409–423, 1997.
[6] Z. R. Lu and J. K. Liu, “Identification of both structural damages
in bridge deck and vehicular parameters using measured
dynamic responses,” Computers and Structures, vol. 89, no. 13-
14, pp. 1397–1405, 2011.
[7] Z. R. Lu and S. S. Law, “Dynamic condition assessment of a
cracked beam with the composite element model,” Mechanical
Systems and Signal Processing, vol. 23, no. 2, pp. 415–431, 2009.
[8] X. B. Lu, J. K. Liu, and Z. R. Lu, “A two-step approach for crack
identification in beam,” Journal of Sound and Vibration, vol. 332,
no. 2, pp. 282–293, 2013.
[9] B. C. Eaves, F. J. Gould, H. O. Peitgen, and M. J. Todd, Homotopy
Methods and Global Convergence, Plenum Press, New York, NY,
USA, 1983.
[10] J. H. He, “Homotopy perturbation method: a new nonlinear
analytical technique,” Applied Mathematics and Computation,
vol. 135, no. 1, pp. 73–79, 2003.
[11] S. J. Liao, “On the homotopy multiple-variable method and
its applications in the interactions of nonlinear gravity waves,”
Communications in Nonlinear Science and Numerical Simula-
tion, vol. 16, no. 3, pp. 1274–1303, 2011.
[12] J. C. Alexander and J. A. Yorke, “The homotopy continuation
method: numerically implementable topological procedures,”
Transactions of the American Mathematical Society, vol. 242, pp.
271–284, 1978.
[13] J. K. Sinha, M. I. Friswell, and S. Edwards, “Simplified models
for the location of cracks in beam structures using measured
vibration data,” Journal of Sound and Vibration, vol. 251, no. 1,
pp. 13–38, 2002.
[14] K. J. Bathe, Finite Element Procedures in Engineering Analysis,
Prentice Hall, Upper Saddle River, NJ, USA, 1982.
Hindawi Publishing Corporation
Mathematical Problems in Engineering
Volume 2013, Article ID 514945, 4 pages
http://dx.doi.org/10.1155/2013/514945

Research Article
An Efficient Approach for Identifying Constitutive
Parameters of the Modified Oyane Ductile Fracture
Criterion at High Temperature

Sergei Alexandrov,1 Yusof Mustafa,2 and Mohd Yazid Yahya3

1
A.Yu. Ishlinskii Institute for Problems in Mechanics, Russian Academy of Sciences, 101-1 Prospect Vernadskogo, Moscow 119526, Russia
2
Faculty of Mechanical Engineering, Universiti Teknologi Malaysia, 81310 Skudai, Johor Darul Ta’zim, Malaysia
3
Centre for Composite, Universiti Teknologi Malaysia, 81310 Skudai, Johor Darul Ta’zim, Malaysia

Correspondence should be addressed to Sergei Alexandrov; sergei [email protected]

Received 13 September 2013; Revised 6 November 2013; Accepted 6 November 2013

Academic Editor: Timon Rabczuk

Copyright © 2013 Sergei Alexandrov et al. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.

The paper presents the theoretical part of a method for identifying constitutive parameters involved in the modified Oyane ductile
fracture criterion at high temperature. Quite a general rigid viscoplastic model is adopted to describe material behavior. The ductile
fracture criterion is in general path-dependent and involves stresses. Therefore, the identification of constitutive parameters of this
criterion is a difficult task which usually includes experimental research and numerical simulation. The latter requires a precisely
specified material model and boundary conditions. It is shown in the present paper that for a wide class of material models usually
used to describe the behavior of materials at high temperatures, the criterion is significantly simplified when the site of fracture
initiation is located on traction free surfaces. In particular, this reduced criterion does not involve stresses. Since there are well
established experimental procedures to determine the input data for the reduced criterion, the result obtained can be considered as
a theoretical basis for the efficient method for identifying constitutive parameters of the modified Oyane ductile fracture criterion
at high temperature. The final expression can also be used in computational models to increase the accuracy of predictions.

1. Introduction expressions are finite relations between two in-surface strains

or the equivalent strain and two in-surface strains. In the case
Empirical ductile fracture criteria are widely used by prac- of hot forming, many experimental results indicate a sharp
tically working engineers to predict the initiation of ductile drop of the strain to fracture at a certain level of temperature
fracture in metal forming processes. Reviews of such criteria and strain rate [6–8]. In the range of temperatures and strain
for cold metal forming processes are provided in [1–3]. Most rates where this drop does not occur, hot ductile fracture
of these criteria include a path-dependent integral whose criteria are often accepted as modifications of cold ductile
integrand contains invariants of the stress tensor. There- fracture criteria in which constitutive parameters depend
fore, the identification of constitutive parameters of ductile on temperature and strain rate [6, 9–13]. In particular, the
fracture criteria in general requires numerical simulation of criterion proposed in [14] for cold metal forming has been
forming processes which is impossible without a precisely modified in [10, 13] to account for temperature and strain
specified material model and boundary conditions. These rate effects. Experimental verification of the applicability of
input data affect the result of numerical simulation. An this modified ductile fracture criterion to several materials
alternative method of identifying constitutive parameters has been provided in these papers as well. An efficient
of two widely used cold ductile fracture criteria has been method for identifying constitutive parameters involved in
proposed in [4, 5]. This method is applicable when the the original criterion [14] has been proposed in [4]. However,
fracture initiation occurs at traction free surfaces and the final this method is based on a rigid hardening material model
2 Mathematical Problems in Engineering

which is not adequate for describing hot metal forming. In the an infinitesimal material fiber coinciding with this principal
present paper, the method [4] is extended to quite a general direction at the initial instant is fixed in the material. Let 𝜉1
viscoplastic model. be the principal strain rate associated with this direction. In
this special case
2. Constitutive Equations 𝑡
𝜀1 = ∫ 𝜉1 𝑑𝑡, (6)
Rigid viscoplastic solids are often used to describe the 0
behavior of material in hot metal forming processes (see, e.g.,
[15]). The constitutive equations of such solids are the yield where 𝜀1 is the principal logarithmic strain. Equation (6) is
condition and its associated flow rule. A widely adopted yield valid at any large strain. The principal stress orthogonal to
condition is the free surface vanishes; 𝜎1 = 0. It follows from this equation
that
𝜎eq = 𝜎0 Φ (𝜉eq , 𝜀eq ) , (1)
𝜎 = −𝜏1 (7)
where 𝜎0 is a reference stress, 𝜎eq is the equivalent stress, 𝜉eq
is the equivalent strain rate, and 𝜀eq is the equivalent strain. at the free surface. Substituting (7) into (5) and using (1) give
The quantities 𝜎eq , 𝜉eq , and 𝜀eq are defined by
𝑡𝑓
𝜏1
3 2 𝑑𝜀eq ∫ [1 − ] 𝜉eq 𝑑𝑡 = 𝐶 (𝜉eq , 𝑇) . (8)
𝜎eq = √ √𝜏𝑖𝑗 𝜏𝑖𝑗 , 𝜉eq = √ √𝜉𝑖𝑗 𝜉𝑖𝑗 , = 𝜉eq , 0 𝐵𝜎0 Φ (𝜉eq , 𝜀eq )
2 3 𝑑𝑡
(2)
It follows from (4) that
where 𝜎𝑖𝑗 are the components of the stress tensor, 𝜉𝑖𝑗 are
2 𝜉eq 𝜏1
the components of the strain rate tensor,𝑡 is the time, 𝜏𝑖𝑗 = 𝜉1 = . (9)
𝜎𝑖𝑗 − 𝜎𝛿𝑖𝑗 are the components of the stress deviator tensor, 3 𝜎0 Φ (𝜉eq , 𝜀eq )
𝜎 = (𝜎𝑖𝑗 𝛿𝑖𝑗 )/3 is the hydrostatic stress, and 𝛿𝑖𝑗 is Kronecker’s
symbol. With no loss of generality it is possible to assume that Replacing the integrands in (8) by means of (9) leads to
Φ(0, 0) = 1. Then, 𝜎0 is the yield stress in uniaxial tension at
𝜉eq = 0 and 𝜀eq = 0. The flow rule associated with the yield 𝑡𝑓
2 𝜉1
condition (1) results in ∫ [𝜉eq − ] 𝑑𝑡 = 𝐶 (𝜉eq , 𝑇) . (10)
0 3𝐵
𝜉𝑖𝑗 = 𝜆𝜏𝑖𝑗 , (3) Using (2) and (6) integration in this equation can be carried
out analytically to give
where 𝜆 is a nonnegative multiplier. Using (1) this multiplier
can be eliminated in (3) to give 𝑓
𝑓 2 𝜀1 (11)
𝜀eq − = 𝐶 (𝜉eq , 𝑇) ,
3 𝜉eq 𝜏𝑖𝑗 3𝐵
𝜉𝑖𝑗 = . (4)
2 𝜎0 Φ (𝜉eq , 𝜀eq ) 𝑓 𝑓
where 𝜀eq and 𝜀1 are the values of 𝜀eq and 𝜀1 , respectively, at the
The ductile fracture criterion applicable at high temperatures instant of fracture initiation. Let 𝜀2 and 𝜀3 be the in-surface
𝑓 𝑓
and strain rates is [13] principal logarithmic strains and let 𝜀2 and 𝜀3 be their values
at the instant of fracture initiation, respectively. It follows
𝑡𝑓
𝜎 𝑓 𝑓
from the incompressibility equation that 𝜀3 + 𝜀2 + 𝜀1 = 0.
𝑓
∫ (1 + ) 𝜉 𝑑𝑡 = 𝐶 (𝜉eq , 𝑇) , (5)
0 𝐵𝜎eq eq 𝑓
Using this equation, 𝜀1 can be eliminated in (11) to give
where 𝑡 is the time, 𝑡𝑓 is the value of 𝑡 at the instant of fracture 𝑓 𝑓
initiation, and 𝐵 is a material constant. 𝐶 depends on the 𝑓 2 (𝜀2 + 𝜀3 ) (12)
equivalent strain rate and temperature, 𝑇. The ductile fracture 𝜀eq + = 𝐶 (𝜉eq , 𝑇) .
3 𝐵
criterion (5) is a modification of the criterion proposed in [14]
for cold metal forming processes. It is worthwhile of noting This relation is the fracture criterion at the free surface.
that the model chosen demands that the principal axes of the Its remarkable property is that the left hand side of (12)
stress and strain rate tensors coincide. is independent of stress components. The in-surface finite
strains at the instant of fracture initiation can be found using
high speed photography [12]. The equivalent strain is in
3. Free Surface Fracture general path-dependent. Therefore, the strain path should
𝑓
Assume that fracture initiates at a point of a traction free be determined to calculate 𝜀eq . However, since the material
surface. It is evident that one of the principal stress directions is supposed to be incompressible, it is sufficient to find the
is orthogonal to this surface. Therefore, one of the principal dependence of 𝜀2 on 𝜀3 [16]. This dependence can be found
strain rate directions is also orthogonal to this surface and using high speed photography.
Mathematical Problems in Engineering 3

4. Conclusions planes of symmetry and a traction free surface. On the other

hand, (12) is valid at any point of a traction free surface. This
The main result of the present paper is that the ductile difference may affect the choice of specimens for experiment.
fracture criterion (5) involving stresses reduces to (12) when Moreover, (12) involves the equivalent strain which is a path-
the initiation of fracture occurs at a point of a traction dependent integral whereas the final expression in [18] is a
free surface. Equation (12) is simply a relation between the relation between two finite in-surface strains at the instant
equivalent strain and two in-surface strains. Therefore, there of fracture initiation. This difference may affect the choice of
is no need to find experimentally and numerically the state of experimental techniques to measure kinematic variables.
stress at the site of fracture initiation. On the contrary, (12) can The main assumption to derive (12) from the fracture
be used in numerical codes. This should increase the accuracy criterion (5) is that the site of fracture initiation is located at a
of predictions because there is no need to calculate the state traction free surface. This condition is satisfied in many sheet
of stress at the site of fracture initiation. As an example, it is forming processes. Therefore, the simplified ductile fracture
instructive to apply (12) to theoretical/experimental results criterion in the form of (12) can be directly adopted to predict
obtained in [17]. In this work the finite element method the initiation of ductile fracture in such processes.
and compression of cylindrical specimens of Ti40 alloy were
used to identify the constitutive parameters of the fracture
criterion (5) at 𝐵 = 2/3. As a result, it has been found (in Acknowledgments
our nomenclature) that The research described in this paper has been supported by
𝑡𝑓 the Grants RFBR-13-08-00969 and NSH-3842.2012.1.
3𝜎
∫ (1 + ) 𝜉 𝑑𝑡 = 2.227 − 0.0493 ln 𝑍. (13)
0 2𝜎eq eq
References
Here 𝑍 is the Zener-Hollomon parameter which is a tem- [1] A. G. Atkins, “Fracture in forming,” Journal of Materials
perature compensated strain rate. Using (12), it is possible to Processing Technology, vol. 56, no. 1–4, pp. 609–618, 1996.
control the accuracy of (13) by means of simpler methods [2] M. A. Shabara, A. A. El-Domiaty, and A. Kandil, “Validity
than those applied in [17]. In particular, (13) takes the assessment of ductile fracture criteria in cold forming,” Journal
following form: of Materials Engineering and Performance, vol. 5, no. 4, pp. 478–
488, 1996.
𝑓 𝑓 𝑓
𝜀eq + 𝜀2 + 𝜀3 = 2.227 − 0.0493 ln 𝑍. (14) [3] R. Hambli and M. Reszka, “Fracture criteria identification using
an inverse technique method and blanking experiment,” Inter-
According to experimental results presented in [17] the national Journal of Mechanical Sciences, vol. 44, no. 7, pp. 1349–
equivalent strain to fracture is a linear function of ln 𝑍. These 1361, 2002.
data are in agreement with (14) if and only if strain paths in [4] D. Vilotić, M. Plančk, S. Grbić, S. Alexandrov, and N. Chik-
the experiment are linear or, at least, close enough to linear. anova, “An approach to determining the workability diagram
Of course, this condition should be verified by experiment. based on upsetting tests,” Fatigue and Fracture of Engineering
However, it is worthwhile of noting that strain paths in cold Materials and Structures, vol. 26, no. 4, pp. 305–310, 2003.
upsetting are usually convex down [18–20]. [5] S. Alexandrov and D. Vilotic, “A theoretical-experimental
Equation (12) has been derived for quite a general vis- method for the identification of the modified Cockroft-Latham
coplastic model. Therefore, there is no need to specify the ductile fracture criterion,” Proceedings of the Institution of
Mechanical Engineers C, vol. 222, no. 9, pp. 1869–1872, 2008.
model to apply this equation in conjunction with experi-
mental data. Finally, boundary conditions have no effect on [6] S. Alexandrov, P. T. Wang, and R. E. Roadman, “A fracture
criterion of aluminum alloys in hot metal forming,” Journal of
this equation. Obviously, some boundary conditions such
Materials Processing Technology, vol. 160, no. 2, pp. 257–265,
as a friction boundary condition have a great effect on the 2005.
initiation of ductile fracture. However, all these possible
[7] O. Sabokpa, A. Zarei-Hanzaki, and H. R. Abedi, “An investiga-
effects are automatically taken into account in the measured
tion into the hot ductility behavior of AZ81 magnesium alloy,”
values of the in-surface principal strains. To summarize, it is Materials Science and Engineering A, vol. 550, pp. 31–38, 2012.
believed that the aforementioned mathematical properties of
[8] J. Deng, Y. C. Lin, S. S. Li, J. Chen, and Y. Ding, “Hot tensile
the ductile fracture criterion considered provide a theoretical deformation and fracture behaviors of AZ31 magnesium alloy,”
basis for an efficient theoretical/experimental method for Materials and Design, vol. 49, pp. 209–219, 2013.
the identification of constitutive parameters of the ductile [9] N. Ogawa, M. Shiomi, and K. Osakada, “Forming limit of mag-
fracture criterion [13] at high temperature. nesium alloy at elevated temperatures for precision forging,”
The general method of analysis used in the present International Journal of Machine Tools and Manufacture, vol. 42,
paper has been also adopted in [21] to reduce the modified no. 5, pp. 607–614, 2002.
Cockcroft-Latham ductile fracture criterion to a simple form [10] X. Zhang, W. Zeng, Y. Shu, Y. Zhao, H. Yu, and Y. Zhou,
not involving stresses. However, the basic assumptions and “Fracture criterion of Ti40 alloy based on the Zener-Hollomon
final results are quite different. In particular, in the case of parameter at hot forming,” Rare Metal Materials and Engineer-
the modified Cockcroft-Latham ductile fracture criterion the ing, vol. 37, no. 4, pp. 604–608, 2008.
general method is applicable if and only if the site of fracture [11] J. T. Yeom, J. H. Kim, J. H. Kim, J. K. Hong, and J. S. Lee, “Hot
initiation coincides with the point of intersection of two forging design of cam for vessel engine using finite element
4 Mathematical Problems in Engineering

analysis and ductile fracture criteria,” Steel Research Interna-

tional, vol. 81, pp. 298–301, 2010.
[12] Y. Zhu, W. Zeng, F. Zhang, Y. Zhao, X. Zhang, and K. Wang,
“A new methodology for prediction of fracture initiation in
hot compression of Ti40 titanium alloy,” Materials Science and
Engineering A, vol. 553, pp. 112–118, 2012.
[13] J. He, Z. Cui, F. Chen, Y. Xiao, and L. Ruan, “The new ductile
fracture criterion for 30Cr2Ni4MoV ultra-super-critical rotor
steel at elevated temperatures,” Materials and Design, vol. 52, pp.
547–555, 2013.
[14] M. Oyane, T. Sato, K. Okimoto, and S. Shima, “Criteria for
ductile fracture and their applications,” Journal of Mechanical
Working Technology, vol. 4, no. 1, pp. 65–81, 1980.
[15] Y.-M. Guo, Y. Yokouchi, and H. Suzuki, “Analysis of hot
forward-backward extrusion by the visco-plastic finite-element
method,” Journal of Materials Processing Technology, vol. 38, no.
1-2, pp. 103–113, 1993.
[16] S. Alexandrov, D. Vilotic, Z. Konjovic, and M. Vilotic, “An
improved experimental method for determining the workabil-
ity diagram,” Experimantal Mechanics, vol. 53, pp. 699–711, 2013.
[17] X.-M. Zhang, W.-D. Zeng, Y. Shu et al., “Fracture criterion
for predicting surface cracking of Ti40 alloy in hot forming
processes,” Transactions of Nonferrous Metals Society of China,
vol. 19, no. 2, pp. 267–271, 2009.
[18] H. A. Kuhn, P. W. Lee, and T. Erturk, “Fracture criterion for cold
forming,” Journal of Engineering Materials and Technology of the
ASME, vol. 95, pp. 213–218, 1973.
[19] S. I. Oh and S. Kobayashi, “Workability of aluminum alloy 7075-
t6 in upsetting and rolling,” Journal of Engineering for Industry-
Transactions of the ASME, vol. 98, no. 3, pp. 800–806, 1976.
[20] J. J. Shah and H. A. Kuhn, “An empirical formula for workability
limits in cold upsetting and bolt heading,” Journal of Applied
Metalworking, vol. 4, no. 3, pp. 255–261, 1986.
[21] S. Alexandrov and Y. R. Jeng, “An efficient method for the iden-
tification of the modified Cockroft-Latham fracture criterion at
elevated temperature,” Archive of Applied Mechanics, vol. 83, no.
12, pp. 1801–1804, 2013.
Hindawi Publishing Corporation
Mathematical Problems in Engineering
Volume 2013, Article ID 491069, 11 pages
http://dx.doi.org/10.1155/2013/491069

Research Article
Study on Analog Theory of Rock Mass Simulation and Its
Engineering Application

Sun Shaorui, Lu Yexu, Xu Yuanyuan, Liu Jin, and Wei Jihong

College of Earth Science and Engineering, Hohai University, Nanjing 210098, China

Correspondence should be addressed to Sun Shaorui; [email protected]

Received 3 May 2013; Revised 24 August 2013; Accepted 10 October 2013

Academic Editor: Stephane P. A. Bordas

Copyright © 2013 Sun Shaorui et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

During construction of 3D geological models, it is difficult to determine the uniform between geological model and true model.
As a comprehensive index, rock quality designation (RQD) is reliable to assess the rationality of geological models. Unfortunately,
The RQD of rockmass is determined completely by the deterministic threshold value and directions of the scan lines presently. To
avoid this drawback, the modified method of the RQD value based on the threshold value and 3D space is proposed in this paper.
Simultaneously, the analogue-simulation method based on rupture mechanism and classification of discontinuities is proposed. The
elliptical discontinuity is considered for general discontinuity, and the special discontinuities, such as bedding, fault, and interlayer
are dealt with individually. The accuracy of the 3D model is verified by the modified RQD. The 3D model of the rockmass is analogue
simulated through repetitively obtaining data from the interval confidence of geometrical parameters of discontinuities, which are
determined by a mass of data derived from field investigation. Besides, the dam base of the Xiangjiaba hydropower station is taken
as an example, and the 3D model of the dam base is analog-simulated; its stability is evaluated by DDA method. The safety coefficient
of the dam base is obtained by utilizing the overload method.

1. Introduction and true models lacks the validation methods, which will
cause the constructed model to be unreliable. For some
At present, 3D geological simulation methods have been particular types of discontinuities, such as fault, fracture zone,
developed from model construction to practical computation and interlayer, there is no way to realize in these methods.
[1, 2]. 3D geological models have been easily accepted as effi- Therefore, 3D geological simulation method in this study
cient methods to understand the engineering characteristics can overcome these drawbacks through analogue simulation
of rock masses, as these models are constructed with discon- technique considering particular discontinuities.
tinuities and structural bodies as 3D networks. The discrete- Rock quality designation (RQD) has been widely used to
element method (DEM), which was firstly developed by Cun- classify the discontinuity of rock masses and assess the intact-
dall, has been widely used to construct 3D geological models ness of rock masses. In previous studies, the applications of
in rock engineering applications. Jing [3] has introduced RQD have been mainly based on a deterministic threshold
additional aspects of the DEM in different related engineering value and a single scanline direction. Priest and Hudson
problems. Andersson and Dverstorp [4], Elsworth [5], Long [8] have studied RQD and proposed a relationship between
et al. [6], and Smith and Schwartz [7] have also widely the fracture frequency and RQD. Harrison [9] has presented
used the discrete fracture network (DFN) to research the equations derived by analyzing fracture frequencies with
behavior of rock masses. Although these software programs different distributions, which may expand the range of RQD
have been successfully used to simulate the behavior of rock values. Sen and Kazi [10] and Şen [11] have studied RQD
fractured systems, the aforementioned research methods and along a scanline with any specific orientation and established
their application to engineering problems are still limited to a relationship between the RQD value and fracture frequency
simplify the representations of the critical discontinuities of along the chosen orientation. Zhang and Einstein [12], Jiang
rock masses. In addition, the uniformity between constructed et al. [13], and Milne et al. [14] have studied the relationships
2 Mathematical Problems in Engineering

between RQD and rock mass parameters, such as fracture field measurements, when simulating the space geometry of
frequency, deformation modulus, volumetric joint count, discontinuities, and the effects of the geometrical characteris-
and permeability coefficient. Snow [15] has pointed out that tics of the rock are partially caused by tectonic movement and
RQD and fracture frequency decrease with depth increasing. so forth. The second issue is (ii) how to reasonably understand
However, after many years of research, RQD has gradually the rock masses formed by discontinuities. Rock masses
come to favor the methods of quantitative and multiple are the combination of discontinuities and structural body.
indexes, and it has become more and more widely known that Under the condition that the geometrical characteristics of
rock structure has a great influence on rock quality. Because the rock masses have been confirmed, it is worth considering
of anisotropy and nonhomogeneity of rockmass, the previous how to simulate the rock masses using these characteristics
evaluation methods of the RQD have some drawbacks, and how to compare the simulated rock masses with the
such as deterministic threshold value and scanline. The new virtual ones. Furthermore, there are several different beliefs
method in this paper considers the different threshold values concerning how to define the index reflecting the intactness
and measure methods from the angle of full space. The of rock masses. (iii) How to build the connection between
general RQD proposed in this paper evaluates rock mass the numerical computational and the geological models
quality based on many RQD balls with the different measure constructed using the 3D network simulation method is the
threshold values. Therefore, the general RQD is adopted to third one. Currently, most numerical computational methods
overcome the drawbacks of traditional evaluation method of which take rock as a continuous medium are not effective in
the RQD. distinguishing the characteristics of rock masses. Even if the
The discontinuous deformation analysis (DDA) method discontinuous analysis method is adopted in some cases, the
has been widely used to assess the stability of rock mass general means will not compensate for the deficiencies in the
engineering. The method has been developed for 2D prob- numerical calculation of engineering; thus it is necessary to
lems by Shi and Goodman [16] and expanded to 3D by Shi establish the relationships between the 3D geological models
[17] and Jiang and Yeung [18]. Some researchers, such as and discontinuous deformation analysis methods.
Yeung et al. [19], Hatzor and Feintuch [20], MacLaughlin
and Berger [21], Doolin and Sitar [22], and Tsesarsky et
al. [23], have confirmed the verification and validation of 2. Basic Principles and Methodologies
DDA by comparing the predictions from the method with
analytical solutions, as well as with other numerical methods 2.1. Simulation Method of Discontinuities. Despite the fact
and laboratory and field measurements. In addition, DDA that the technique of discontinuous network simulation
can also be viewed as a mature and reliable method to apply in (DNS) has been greatly improved in recent years, this method
the stability analysis of rock engineering; for example, Hatzor is not particularly accurate, due to its hypothesis. It assumes
et al. [24] and Sitar et al. [25] have used DDA to investigate that the spatial shape of the discontinuities is circular, while
slope stability, and Kim et al. [26], Wu et al. [27], and this is not necessarily true. In addition, the technique of
Tsesarsky and Hatzor [28] have used it to analyze tunneling the DNS ignores the effects of the large-scale discontinuities,
stability in discontinuous rock masses. In the past study on taking only the grade IV and V discontinuities into account.
DDA method, the orientation of discontinuity is constant It is easy to understand that the application and decision of
when DDA method is used to conduct numerical simulation. the 3D simulation technique are restricted, and, as matter of
It is unreasonable for the true geometrical characteristics a fact, the grade IV and V discontinuities are the ones which
of natural discontinuity. Meanwhile, variable orientation of play controlling roles in the local stabilization of rock masses.
discontinuity can produce many polyhedrons with different In this paper, the large-scale discontinuities are firstly consid-
shape compared with that of constant orientation. Addi- ered during the construction of the 3D geological model. The
tionally, another problem for the DDA method is that the production of the grade IV and V discontinuities is based on
particular discontinuities cannot be considered. Therefore, the 3D model, including the large scale discontinuities such
in order to avoid two drawbacks, both the particular types as fault and weak interlayers. Through a large amount of the
of discontinuities by improving preprocess program and the field observations and research regarding the shape of the
variable orientation of natural discontinuity are considered discontinuities, it is believed that the discontinuities should
for the simulation method in this study. not be categorized into a single circular shape, due to the fact
At present, a wide variety of rock quality assessment that the production of the discontinuities is affected mainly
methods have taken rock structure into account; however, by the tectonic movement. As known, the stress ellipsoid
it is necessary to determine how to adapt the rock structure is applied to structure geology for explaining rock failure
model into the adaptive mechanical model for providing the and deformation of rock masses based on the definition of
basic data for model experiments and rock masses intactness mathematics and the physical meaning. Therefore, the shape
assessment. The 3D structure model of rock masses is of discontinuities directly affected by tectonic stress should
typically acquired by using discontinuities and spacing, block not be circular in shape, but elliptical, generally due to the
size, and other rock intactness indexes. There are three issues unequality of the values of the maximum and minimum prin-
involved in the acquaintance of rock structure: (i) how to gain cipal stresses. Based on this view, the elliptical shape would be
the reasonable values for the geometrical characteristics of adopted to corresponding discontinuities affected by tectonic
the discontinuities. It remains open to discuss how to describe stress, and the steps of the DNS are as follows: (i) take
the whole rock masses through the representative data from samples of the discontinuities through field measurements;
Mathematical Problems in Engineering 3

(a) Discontinuities (b) Structural bodies

Figure 1: Map of the DNS for the Xiangjiaba hydropower station dam base.

(ii) create probabilistic models of geometrical parameters of threshold value of 10 cm is nevertheless an arbitrarily
the discontinuities such as orientation, spacing, and trace selected value. Therefore, if the variable threshold
length; (iii) make use of the Monte-Carlo random simulation can have a dramatic effect on the computed RQD,
technique to obtain random data from the confidence interval it is appropriate that the assessment of the RQD
of geometrical parameters of discontinuities; (iv) construct can be further investigated in order to determine a
the large-scale discontinuities in the 3D model, such as the method for selecting the threshold value, rather than
fault, bedding, and interlayer; (v) construct the grade IV always relying on the customary value. In order to
and V discontinuities in the 3D model; and (vi) create a avoid these dimensional effects, another threshold
complete network map of the DNS and form its boundaries. value should be adopted according to the specific
The boundaries formed by the discontinuities are shown in engineering project.
Figure 1(a), while Figure 1(b) shows the boundaries of the (ii) The rock masses are constituted by blocks of different
dam base formed by the structural bodies. sizes and shapes, which are formed by the separation
of the discontinuities. When analyzing the stability of
2.2. Verification of the DNS the rock masses in the project using the DDA method,
every structural element will be formed by a closed
2.2.1. RQD and Its Deficiencies. Rock masses are discon- loop constituted by several traces, and the traces
tinuous, inhomogeneous, and anisotropic in terms of their which do not constitute a closed loop will be deleted
discontinuous geometry; thus it is a difficult task to simulate or trimmed. Therefore, the RQD values are not
the whole characteristics of rock masses. At present, ROD, a reasonable due to the different quantity traces in the
method used to evaluate the intactness and quality of rock measure scanlines intersecting with the redundant
masses, is the main means of simulating the discontinued traces, and thus both should be taken into account
properties of rock. The intactness coefficient of rock masses during the analysis of RQD. The dam base model
deals with the anisotropic property from the whole rock mass. formed by structural elements should be reliable from
However, RQD makes an objective appraisal of the intactness the perspectives of theory and field investigation.
of the rock mass using the scanlines, such as borehole, tunnel,
and field outcrop. The advantages and disadvantages of the 2.2.2. Modified Approach of RQD Evaluation. Two main
method are described as follows. approaches for RQD evaluation are proposed as follows.
(i) It is not proper to use the RQD method to merely (1) Improve the gauge of threshold value. In order to
evaluate the quality of rock mass through a borehole conveniently calculate using the DDA method and satisfy the
which narrowly shows the condition of rock masses, reasonable threshold value of the rock mass intactness, ten
because the rock mass has the property of anisotropy. gauges of threshold values, spaced every 1.0 m, from 1.0 m
In addition, the RQD method proposed by DEER in to 10.0 m, are used to evaluate the intactness of the rock
1964 gave the deterministic value of 10 cm, which is masses, thus creating a combination of the intactness and
defective at measuring the intactness of rock masses, measurement indexes. (2) Change the single direction to 3D
and increases the amount of calculation if it is used space. Based on the perspective of rock mass anisotropy, an
to evaluate the stability of the rock mass, due to the all-sided gauge of the threshold value is adopted to measure
production of a mass of structural elements separated the RQD in 3D space rather than in the vertical or horizontal
by discontinuities. For example, the actual direction direction.
of a scanline in the project is often vertical in the Four RQD balls formed in four different threshold values
borehole and horizontal in the tunnel, and the RQD are shown in Figure 2. As seen, the shape of the RQD
value of single direction scanline cannot reflect the ball is circular when the threshold value is smaller, and
intact degree of the rock masses quality if considering the characteristic of the rock mass approaches isotropy; on
the anisotropy of the rock masses. In addition, a the contrary, the larger the threshold value is, the more
4 Mathematical Problems in Engineering

Figure 2: 3D RQD of the Xiangjiaba hydropower station dam base rock masses.

disproportionate the shape is, and the rock mass anisotropy

may be clearly seen. These results show that the RQD ball
is indeed capable of representing the characteristics of rock
mass structures.

2.2.3. RQD Verification Methods. In order to verify the

reasonability of 3D geological model of DNS, the actual
measured RQD is compared with the simulated one. The steps
of the measurement method are as follows.

(i) Calibration of the borehole: according to the simu-

lated rock masses, the RQD of the rock masses in
the actual position of the drill hole is evaluated, and Xiangjiaba

the results with those of the actual measurements are

compared. The error between the actual measured Figure 3: Location of dam base zone.
values and the simulated values of the geometrical
parameters must not be avoided. RQD is used to
Extruded zone of left bank
calibrate the simulated model by adjusting the geo-
metrical parameters of the discontinuities, because
RQD is a comprehensive index, which contains the
geometrical characteristics of discontinuities, such as
orientation, spacing, and trace length, to reflect the Jinsha river Jinsha river
intactness of rock masses. Therefore, it is necessary Fracture zone of fold
to verify the RQD by adjusting the geometrical
characteristic of the discontinuities until the results
satisfy the demands.
(ii) Calibration of the block characteristic of the rock Axial of Limeiwan fold
masses: contrast the size and magnitude of the block
formed from the 3D model with those of the actual
field investigation. The span of the block may be
satisfied by adjusting the geometrical parameters of Figure 4: Layout of dam base zone.
the discontinuities.

It is believed to be very difficult to simultaneously satisfy blocks. Within representing the number of blocks in the block
the two aforementioned aspects. In general, the first one system, the following equations are listed together in
aspect is fulfilled, and then the other assessment aspect is
calibrated accordingly. 𝐾11 𝐾12 𝐾13 ⋅ ⋅ ⋅ 𝐾1𝑛 𝐷1 𝐹1
𝐾21 𝐾22 𝐾23 ⋅ ⋅ ⋅ 𝐾2𝑛 𝐷2 𝐹2
2.3. Basic Theory of Discontinuous Deformation Analysis (𝐾31 𝐾32 𝐾33 ⋅ ⋅ ⋅ 𝐾3𝑛 ) (𝐷3 ) = (𝐹3 ) . (1)
.. .. .. . .. ..
(DDA). Recently, the basic theory of DDA begins to reach . . . d .. . .
maturity in correlative research and has made significant 𝐾𝑛1 𝐾𝑛2 𝐾𝑛3 ⋅ ⋅ ⋅ 𝐾𝑛𝑛 𝐷𝑛 𝐹𝑛
development in engineering application. The key contents of
DDA are discussed below. Blocks in DDA are connective and Every block possesses six degrees of freedom. Each
form a block system by the contact constraint between two element [𝐾𝑖𝑗 ] given in the coefficient matrix in (1) is a 6 × 6
Mathematical Problems in Engineering 5

Table 1: Statistical results of geometrical parameters of discontinuities.

Orientation
No. Statistical indexes Model Mean Standard variation Samples
Dip/∘ Dip angle/∘
Spacing/m Log-normal 0.60 0.79 360
Dip/∘ Normal 85.24 13.32 374
(1) 45.54∼115.54 8.46∼58.46
Dip angle/∘ Uniform 36.77 11.44 374
Trace length/m Log-normal 23.80 19.76 374
Spacing/m Log-normal 11.10 32.64 23
Dip/∘ Normal 181.46 6.33 28
(2) 170.0∼190.0 64.61∼89.61
Dip angle/∘ Normal 81.21 7.17 28
Trace length/m Uniform 2.01 0.34 28
Spacing/m Log-normal 5.29 24.44 49
Dip/∘ Uniform 345.84 8.84 58
(3) 326.57∼360.00 41.47∼76.47
Dip angle/∘ Uniform 60.22 8.56 58
Trace length/m Log-normal 2.07 1.49 58
Spacing/m Log-normal 3.50 12.64 54
Dip/∘ Uniform 292.48 16.96 63
(4) 262.01∼327.01 24.81∼69.81
Dip angle/∘ Uniform 43.47 6.07 63
Trace length/m Log-normal 0.95 0.54 63
Spacing/m Log-normal 2.20 7.66 22
Dip/∘ Normal 60.44 7.95 27
(5) 44.74∼74.74 58.30∼88.30
Dip angle/∘ Log-normal 69.33 9.83 27
Trace length/m Log-normal 3.64 2.12 27

Table 2: Physical-mechanical parameters of rock masses in the dam base.

Materials Density/kN/m3 Modulus/GPa Poisson’s ratio 𝑓󸀠 𝑐󸀠 /MPa

Concrete 24.0 17.60 0.167
Bed rock (II) 26.0 15.00 0.220 1.20 1.40
Bed rock (III1 ) 26.0 7.00 0.250 0.99 1.00
Bed rock (III2 ) 26.0 5.50 0.280 0.86 0.80
Bed rock (III2∼IV) 26.0 4.00 0.290 0.77 0.65
Soft layers 𝑃𝑛 20.0 0.75 0.400 0.35 0.10

Table 3: Intactness assessment of rock masses of Xiangjiaba hydropower station dam base.

Rock quality Degree of weathering Intactness of rock masses Location of

grade Velocity of acoustic RQD in the RQD in the distribution
wave/m/s tunnel/% borehole/%
I Unweathered >5000 >90 >85 Depth of dam base
Unweathered to weak
II 4000–5000 75–90 60–85 Zones 1, 2, 4
weathering
Moderate weathering
(lower) to weak 3500–4000 62.5–75 45–60 Zones 1, 4, 5, 6
III weathering
Moderate weathering
Zone 6
(upper) 3000–3500 50–62.5 30–45
Weak weathering to
Zone 3
unweathered
Zones 3, 5
IV 2000–3000 25–50 5–30
Weak weathering to Influence band of
unweathered fault and
concentrated joint
band
Shattered fault zone
V <2000
and weak interlayer
Notes: The zone of the dam base is divided by six sub-zones, which are arranged in order from the right bank to left bank of the Jinsha River.
6 Mathematical Problems in Engineering

Figure 5: Computation model of dam base of Xiangjiaba hydropower station.

submatrix; [𝐷𝑖 ] and [𝐹𝑖 ] are 6 × 1 submatrices; 𝐷𝑖 represents 3. Case study

the deformation variable of block 𝑖; and 𝐹𝑖 is the load which
is placed on the six deformation variables of block 𝑖. The six 3.1. Engineering Geology Survey. The Xiangjiaba hydropower
variables of each block are listed in station is the final planned stepped hydropower station
along the Jinsha River (Figure 3). Figure 4 shows the dam
(𝑢0 , V0 , 𝑟0 , 𝜀𝑥 , 𝜀𝑦 , 𝛾𝑥𝑦 ) , (2) base of the Xiangjiaba hydropower station. The engineering
geological conditions of the dam site zone are very complex:
where (𝑢0 , V0 ) is the rigid body displacement of the special the river bed contains 15 small faults, which may be divided
point (𝑥0 , 𝑦0 ) in the block; angle 𝑟0 , shown with the radian, into two groups; the first group discontinuity which its strike
is an angle of rotation around the center point (𝑢0 , V0 ) in the is NNW, contains seven faults with the orientation of 280∼
block; 𝜀𝑥 , 𝜀𝑦 are the normal strain; and 𝛾𝑥𝑦 is the shear strain. 300∘ /SW∠70∼80∘ ; the other, strike NE, contains six faults
DDA is based on the principle of the minimum potential with the orientation of 30∘∼50∘ /NW∠65∘∼80∘ .
energy; the total potential energy Π is the sum of all of the In the dam site zone, the joint and fissure developments
potential energy. Use the equation below to calculate each differ from each other due to their different tectonic locations.
stress and force potential energy, as well as their respective If the zone is divided based on tectonic location, the stratum
differentials. The equation of the matrix [𝐾𝑖𝑗 ] is shown in steep-dip is located within the core of the fold. The right side
of the bed and the right bank are both on the SW limbs of the
𝜕2 Π fold, and the left side of the bed and the left bank are both on
= 0, 𝑟, 𝑠 = 1, . . . , 6. (3) the NE limbs. The joint and fissure develop greatly, with the
𝜕𝑑𝑟𝑖 𝜕𝑑𝑠𝑖
strikes of the superiority jointing groups being 40∼70∘ and
All the terms in (3) form a 6 × 6 submatrix, which is 320∼340∘ , the former of which are fissures with medium- to
the submatrix [𝐾𝑖𝑗 ] in (1). It may be clearly seen that (3) and high-dip angles, and the latter are joints with low-dip angles.
the coefficient matrix are symmetrical. The equation of the The density of the joint and fissure is 10–20 per m. The joints
matrix [𝐹𝑖 ] is shown in on the left bank develop better than those on the right bank;
there are two superiority jointing groups on the left bank,
𝜕Π (0) which have strikes of 280∼300∘ and 60∼80∘ and the density
− = 0, 𝑟 = 1, . . . , 6. (4)
𝜕𝑑𝑟𝑖 is 2–5 per m; and there are two superiority joint groups in the
right with strikes of 60∼80∘ and 280∼300∘ and the density is
All the terms of (4) form a 6 × 6 submatrix, which will be 1–3 per m. The geometrical characteristics and corresponding
added into the submatrix [𝐹𝑖 ]. probability distribution models of the main discontinuities
Many components of the total stiffness matrix are pro- are listed in Table 1.
vided by the single block and block systems, including the
elastic, initial stress, point load, volume force, inertia force,
initial displacement, bolt connection, normal contact force, 3.2. Calculation Parameters and Analysis. The material
tangent contact force, and friction force submatrices. The parameters used for calculation in the DDA method are
role played by each part of the total stiffness matrix in every shown in Table 2.
situation may be clarified, provided that the appropriate
potential energy expression is performed for each situation, 3.3. Analysis of the Calculation Results
and (3) or (4) is used. The simulation process is realized
by a modified source code, which is considered to be the 3.3.1. Computation Model. Figure 5 shows the section of the
influence of the weak interlayer, based on the original code 3D geological model of the dam base which is constructed by
programming described by Shi [16, 17]. the DNS techniques. It may be seen that the combination of
Mathematical Problems in Engineering 7

T = 10.0 0.00% T= 10.0 0.00% T = 10.0 0.00% T = 10.0 0.00%

T = 9.0 0.00% T= 9.0 0.00% T = 9.0 0.00% T = 9.0 0.00%
T = 8.0 0.00% T= 8.0 9.13% T = 8.0 0.00%
T = 7.0 T = 8.0 0.00%
0.00% T= 7.0 9.13% T = 7.0 0.00% T = 7.0 6.77%
T = 6.0 6.76% T= 6.0 23.09% T = 6.0 23.84%
T = 5.0 64.20% T = 6.0 6.63%
T= 5.0 35.23% T = 5.0 68.72% T = 5.0 66.29%
T = 4.0 93.57% T= 4.0 75.41% T = 4.0
T = 3.0 96.88% T = 4.0 83.53% 86.69%
T= 3.0 83.20% T = 3.0 94.74% T = 3.0 95.58%
T = 2.0 99.26% T= 2.0 90.18% T = 2.0 96.94% T = 2.0 95.58%
T = 1.0 99.26% T= 1.0 95.99% T = 1.0 100.00% T = 1.0 99.91%

(a) Location and scale of RQD method application

61.41% 71.51%
37.38% 100.00% 84.59% 53.15% 82.59%
84.59%
18.74%
17.73%
17.04%
95.73% 64.38% 72.94% 0.00% 41.78%
19.22% 25.86% 59.94%
40.25% 81.01%
100.00% 48.93% 36.95%
36.99% 100.00%
40.28%

83.80% 62.44%
54.53% 71.65%
40.90%
96.11% 94.67% 58.32%
82.82% 49.40% 94.67%
24.01%
17.54%
13.49%
14.61%
94.39% 99.69% 94.39% 0.00% 99.69%

100.00% 98.81% 100.00% 98.81%

100.00% 95.32% 100.00%

71.98% 82.96%
90.91% 90.91%
91.55%

(b) Results of RQD method at different locations

Figure 6: RQD map of the dam base controlled by the rock mass discontinuities.

the discontinuities and soft layers is taken into account when the rock masses dissected by the discontinuities, a compre-
building the dam base. hensive method which evaluates the rock masses from all
In Figure 5, it is quite clear that there are many soft layers directions must be used.
in the rock masses of the dam base, as well as in the soft Table 3 shows the intactness assessment system of rock
interlayer distributed throughout the entire dam base zone. masses of the dam base. As seen, the RQD in the center of
Based on the results of the field engineering geological survey, dam base is lower than that in the sides. This indicates that
these soft layers showing the shape of the knee buckling are the rock masses in the center of the dam base are broken due
found below the dam base and will lead to detrimental results. to tectonic movement. In addition, the RQD in the tunnel is
larger than that in the borehole. This phenomenon is proven
3.3.2. Assessment of Rock Masses Quality. It may be concluded by the constructed 3D model in Figure 6.
from Figure 5 that two types of superiority discontinuities
exist in the rock masses with the exception of the soft (1) Evaluation of RQD. The evaluation results of the rock
layers. In addition, the discontinuities tend to develop better mass quality are shown in Figure 6(a). It may be concluded
upstream than downstream. In order to reasonably evaluate that the RQD method is arbitrary and is able to evaluate the
8 Mathematical Problems in Engineering

450 displacement in the rear part of the dam ranges from 15 to

400 17 cm. The base of the dam may be divided into four parts; one
350 of these measure points has the largest displacement, namely,
300
Frequency

point 4; the points 5, 6, 18, 20, and 25 form the second group,
250
200
their displacements ranging from 12.5 to 17.5 cm; the third
150 group contains points 7, 8, and 19, their displacements ranging
100 from 11 to 12.5 cm; and the fourth group contains points 9,
50 22, and 23, the displacements of which are all approximately
0 10 cm.
0 80 160 240 320 400 480 560 640 720
As seen in Figure 8(d), the altering regularity of the X-
Volume (m3 )
velocity developed with time ranges between 1.5 and 4.0 s;
Actual frequency points 8 and 9 have velocities in the opposite direction as
Model frequency the other points; the curves vary widely beginning from 1.5 s;
the maximum velocity value is greater than 4.0 cm/s; and
Figure 7: Actual and model distribution of the rock block size. beginning from 5 s, the trends of every point are similar. As a
whole, the range of the forepart of the dam is larger than that
of the rear part; the changing regularity of the curves is similar
rock mass quality from any direction. Meanwhile, as a typical from 1.5 to 5.0 s, with the exception of point 25, the curve
measurement, the borehole is still used to evaluate the rock of which is lower than the others. In the end, the respective
mass quality. Vertically below the dam surface, the value of velocities of all the points tend to zero. The curve of the Y-
the ROD shown by the borehole is smaller than that above velocity shown in Figure 8(e) alters regularly, from 0 to 4.5 s;
the surface, and the RQD values of all other parts of the dam beginning from 4.5 s, the displacements of the points in the
have little differences among them. When the threshold value dam base begin to change, and the one in the forepart is more
is 7.0, the RQD value is typically 0; when the threshold value evident than that in the rear. The reason for this phenomenon
of the RQD is smaller, the RQD value approaches 100%, while is that the dam deforms toward the rear due to the water
the RQD value of the rock masses below the dam remains at pressure, and the rock masses of the forepart are uplifted,
about 95%. resulting in rock mass relaxation.
Figure 6(b) shows the RQD values under two threshold
values, namely, 3.0 m and 5.0 m, at two different locations.
Regardless of where the rock masses are located, the RQD 3.3.4. Failure Analysis of Dam Base. It is convenient to exam-
is anisotropic. In the NNE direction the rock mass quality is ine the actual damage of the rock masses through the DDA
higher, while in the NNW direction the quality is lower, and program. The results are calculated via 1000 calculation steps
while the threshold value is larger, the RQD value tends to in DDA and are shown in Figure 9, in which the small circles
be 0. For the rock masses near the contact surface vertically at the top represent the damage degree. The DDA program
below the dam base, the RQD value decreases with the considers the effects of many groups of discontinuities, and
threshold value decreasing. the direction of the discontinuities alters widely due to the
knee buckling shape.
(2) Rock Block Characteristic Analysis. Figure 7 shows the According to the overload method used in the DDA
statistics of the rock mass volumes. As seen, the majority program, the rock mass failure of the dam base is classified
of the volume of the rock blocks is less than 300 m3 , the as tensile fracture damage. The large circles in Figure 9 show
lowest rock volume is 6.5 m3 , the highest is 5367.8 m3 , and that the discontinuities have a relatively large opening, due to
the average value is 207.0 m3 . The model is consistent with the deformation of the dam base. In Figure 10, the curve of the
the Weibull distribution, and the model resembles the actual horizontal displacement flattens out as the overload multiples
data accurately. reach 7.0 times, and when the overload multiples range from
7.0 to 7.5 times, the curve of the vertical displacement ceases
to change. The main reasons for this phenomenon include
3.3.3. Discontinuous Deformation Analysis. As seen in
the fact that the rock masses of the dam base readjust after
Figure 8(a), in order to analyze the deformation of the dam
being damaged, and the rocks in the rear of the dam act as a
base rock masses during the service period, the measure
resistance. Therefore it may be concluded that the damage of
points are arranged in the dam base or at locations where
the rock masses of the dam base is simulated very effectively
the fissures are likely to be produced. Then the horizontal
using the DDA method in the service period, and the failures
displacement, vertical displacement, horizontal velocity, and
and discontinuities in the contact part of the dam base may
vertical velocity are analyzed, and the results are shown in
be easily observed.
Figures 8(b) to 8(e).
Similar trends with different measure points are seen in
both Figures 8(b) and 8(c). The displacement of the forepart 4. Conclusions
of the dam is larger than the ones in the center and rear parts
of the dam. In the displacement of point 4, the measure points RQD is one of the most important indexes for assessing the
in the forepart are larger than those in all the remaining intactness of rock masses which depend on the geometrical
parts, and the highest value is 20 cm. The highest value of characteristics of their discontinuities. In this paper it is
Mathematical Problems in Engineering 9

(a) Measure points in the dam base

0.05 0.015

0.00 0.012
Horizontal displacement (m)

Vertical displacement (m)

−0.05 0.009

−0.10 0.006

−0.15 0.003

−0.20 0.000
0 3 6 9 12 15 0 3 6 9 12 15
Times (s) Times (s)
(b) X-displacement with time of measure points (c) Y-displacement with time of measure point
0.4

0.000 0.3
Horizontal velocity (m/s)

Vertical velocity (m/s)

0.2
−0.015

0.1
−0.030
0.0

−0.045
−0.1

−0.060 −0.2
0 3 6 9 12 15 0 3 6 9 12 15
Times (s) Times (s)

4 7 18 22 4 7 18 22
5 8 19 23 5 8 19 23
6 9 20 25 6 9 20 25
(d) X-velocity with time of measure points (e) Y-velocity with time of measure points

Figure 8: Displacement and velocity with time of different measure points in the dam base.

shown that the use of the optimal RQD threshold value (1) Based on the mechanism of the geological mechanics,
greatly extends the range of RQD values. A series of methods the elliptical discontinuities are used to simulate the
is used to assess rock quality, construct a 3D model, and rock structure, and this method is applied to the dam
evaluate the stability of the dam base. The conclusions of the base of Xiangjiaba hydropower station. This study
study are as follows. shows that the effects of the rock structure are caused
10 Mathematical Problems in Engineering

Figure 9: Fracture of rock masses in the dam base.

2.5 0.3

2.0
Horizontal displacement (m)

Vertical displacement (m)

0.2

1.5
0.1
1.0

0.0
0.5

0.0 −0.1
0.0 2.5 5.0 7.5 10.0 12.5 0.0 2.5 5.0 7.5 10.0 12.5
Overload times Overload times

1 4 1 4
2 5 2 5
3 3
(a) Horizontal displacement with overload times (b) Vertical displacement with overload times

Figure 10: Relationship between overload and displacement of the measure points.

by the special discontinuities during the simulation, rock mass quality and the strain characteristic of the
and this method is very valid for application to the dam base are analyzed. The overload method is used
simulation of the rock structure. to consider the failures occurring near the contact
surface of the dam base and the rock masses, and the
(2) Based on the anisotropic characteristic of the rock stability coefficient of the dam is determined through
masses, the modified RQD method with the threshold calculation.
value and 3D space is applied to confirm the rational-
ity of the 3D simulation and evaluate the rock mass
quality of the dam base. Acknowledgments
(3) Based on the 3D structure of the dam base, the ground This study is financially supported by the Natural Science
profile applied in the discontinuous deformation Foundation of China (Grants nos. 41002089 & 41102162)
analysis is obtained by cutting the model and the and Jiangsu Overseas Research & Training Program for
Mathematical Problems in Engineering 11

University Prominent Young and Middle-aged Teachers and [16] G. H. Shi and R. E. Goodman, “Two dimensional discontinuous
Presidents. The authors would also like to acknowledge the deformation analysis,” International Journal for Numerical and
editors and reviewers of this paper for their very helpful Analytical Methods in Geomechanics, vol. 9, no. 6, pp. 541–556,
comments and valuable remarks. 1985.
[17] G. H. Shi, “Three dimensional discontinuous deformation
analysis,” in Rock Mechanics in the National Interest: Proceedings
References of the 38th U.S. Rock Mechanics Symposium, pp. 1421–1428, A.A.
Balkema, Rotterdam, The Netherlands, 2001.
[1] D. H. Zhong, M. C. Li, L. G. Song, and G. Wang, “Enhanced
NURBS modeling and visualization for large 3D geoengi- [18] Q. H. Jiang and M. R. Yeung, “A model of point-to-face contact
neering applications: an example from the Jinping first-level for three-dimensional discontinuous deformation analysis,”
hydropower engineering project, China,” Computers and Geo- Rock Mechanics and Rock Engineering, vol. 37, no. 2, pp. 95–116,
sciences, vol. 32, no. 9, pp. 1270–1282, 2006. 2004.
[2] H. H. Zhu, X. Y. Zhuang, Y. C. Cai, and G. Ma, “High rock [19] M. R. Yeung, Q. H. Jiang, and N. Sun, “Validation of block the-
slope stability analysis using the enriched meshless Shepard and ory and three-dimensional discontinuous deformation analysis
least squares method,” International Journal of Computational as wedge stability analysis methods,” International Journal of
Methods, vol. 8, no. 2, pp. 209–228, 2011. Rock Mechanics and Mining Sciences, vol. 40, no. 2, pp. 265–275,
[3] L. Jing, “Formulation of discontinuous deformation analysis 2003.
(DDA)—an implicit discrete element model for block systems,” [20] Y. H. Hatzor and A. Feintuch, “The validity of dynamic block
Engineering Geology, vol. 49, no. 3-4, pp. 371–381, 1998. displacement prediction using DDA,” International Journal of
[4] J. Andersson and B. Dverstorp, “Conditional simulations of Rock Mechanics and Mining Sciences, vol. 38, no. 4, pp. 599–606,
fluid flow in three-dimensional networks of discrete fractures,” 2001.
Water Resources Research, vol. 23, no. 10, pp. 1876–1886, 1987. [21] M. M. MacLaughlin and E. A. Berger, “A decade of DDA valida-
[5] D. Elsworth, “A hybrid boundary element-finite element analy- tion: development and application of discontinuous modeling
sis procedure for fluid flow simulation in fractured rock masses,” for rock engineering,” in Proceedings of the 6th International
International Journal for Numerical and Analytical Methods in Conference on Analysis of Discontinuous Deformation, pp. 13–31,
Geomechanics, vol. 10, no. 6, pp. 569–584, 1986. A.A. Balkema, Rotterdam, The Netherlands, 2003.
[6] J. C. S. Long, P. Gilmour, and P. A. Witherspoon, “A model [22] D. M. Doolin and N. Sitar, “Displacement accuracy of dis-
for steady fluid flow in random three-dimensional networks of counting deformation analysis method applied to sliding block,”
disc-shaped fractures,” Water Resources Research, vol. 21, no. 8, Journal of Engineering Mechanics, vol. 128, no. 11, pp. 1158–1168,
pp. 1105–1115, 1985. 2002.
[7] L. Smith and F. W. Schwartz, “An analysis of the influence of [23] M. Tsesarsky, Y. H. Hatzor, and N. Sitar, “Dynamic displacement
fracture geometry on mass transport in fractured media,” Water of a block on an tnclined plane: analytical, experimental and
Resources Research, vol. 20, no. 9, pp. 1241–1252, 1984. DDA results,” Rock Mechanics and Rock Engineering, vol. 38, no.
[8] S. D. Priest and J. A. Hudson, “Discontinuity spacings in rock,” 2, pp. 153–167, 2005.
International Journal of Rock Mechanics and Mining Sciences [24] Y. H. Hatzor, A. A. Arzi, Y. Zaslavsky, and A. Shapira, “Dynamic
and, vol. 13, no. 5, pp. 135–148, 1976. stability analysis of jointed rock slopes using the DDA method:
[9] J. P. Harrison, “Selection of the threshold value in RQD assess- King Herod’s Palace, Masada, Israel,” International Journal of
ments,” International Journal of Rock Mechanics and Mining Rock Mechanics and Mining Sciences, vol. 41, no. 5, pp. 813–832,
Sciences, vol. 36, no. 5, pp. 673–685, 1999. 2004.
[10] Z. Sen and A. Kazi, “Discontinuity spacing and RQD estimates [25] N. Sitar, M. M. MacLaughlin, and D. M. Doolin, “Influence of
from finite length scanlines,” International Journal of Rock kinematics on landslide mobility and failure mode,” Journal of
Mechanics and Mining Sciences and, vol. 21, no. 4, pp. 203–212, Geotechnical and Geoenvironmental Engineering, vol. 131, no. 6,
1984. pp. 716–728, 2005.
[11] Z. Şen, “Theoretical RQD-porosity-conductivity-aperture [26] Y. I. Kim, B. Amadei, and E. Pan, “Modeling the effect of
charts,” International Journal of Rock Mechanics and Mining water, excavation sequence and rock reinforcement with dis-
Sciences and Geomechanics, vol. 33, no. 2, pp. 173–177, 1996. continuous deformation analysis,” International Journal of Rock
[12] L. Zhang and H. H. Einstein, “Using RQD to estimate the Mechanics and Mining Sciences, vol. 36, no. 7, pp. 949–970, 1999.
deformation modulus of rock masses,” International Journal of [27] J. H. Wu, Y. Ohnishi, and S. Nishiyama, “Simulation of the
Rock Mechanics and Mining Sciences, vol. 41, no. 2, pp. 337–341, mechanical behavior of inclined jointed rock masses during
2004. tunnel construction using discontinuous deformation analysis
[13] X. W. Jiang, L. Wan, X. S. Wang, X. Wu, and X. Zhang, (DDA),” International Journal of Rock Mechanics and Mining
“Estimation of rock mass deformation modulus using variations Sciences, vol. 41, no. 5, pp. 731–743, 2004.
in transmissivity and RQD with depth,” International Journal of
[28] M. Tsesarsky and Y. H. Hatzor, “Tunnel roof deflection in
Rock Mechanics and Mining Sciences, vol. 46, no. 8, pp. 1370–
blocky rock masses as a function of joint spacing and friction—
1377, 2009.
a parametric study using discontinuous deformation analysis
[14] D. Milne, J. Hadjigeorgiou, and R. Pakalnis, “Rock mass (DDA),” Tunnelling and Underground Space Technology, vol. 21,
characterization for underground hard rock mines,” Tunnelling no. 1, pp. 29–45, 2006.
and Underground Space Technology, vol. 13, no. 4, pp. 383–391,
1998.
[15] D. T. Snow, “Rock fracture spacings, openings, and porosities,”
Journal of the Soil Mechanics and Foundations Division, vol. 94,
no. 1, pp. 73–91, 1968.
Hindawi Publishing Corporation
Mathematical Problems in Engineering
Volume 2013, Article ID 293861, 12 pages
http://dx.doi.org/10.1155/2013/293861

Research Article
Explicit Dynamic Finite Element Method for Failure with
Smooth Fracture Energy Dissipations

Jeong-Hoon Song,1 Thomas Menouillard,2 and Alireza Tabarraei3

1
Department of Civil and Environmental Engineering, University of South Carolina, Columbia, SC 29208, USA
2
STUCKY Ltd., Rue du Lac 33, 1020 Renens VD 1, Switzerland
3
Department of Mechanical Engineering and Engineering Science, University of North Carolina at Charlotte,
Charlotte, NC 28223, USA

Correspondence should be addressed to Jeong-Hoon Song; [email protected]

Received 19 July 2013; Accepted 1 October 2013

Academic Editor: Timon Rabczuk

Copyright © 2013 Jeong-Hoon Song et al. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.

A numerical method for dynamic failure analysis through the phantom node method is further developed. A distinct feature of this
method is the use of the phantom nodes with a newly developed correction force scheme. Through this improved approach, fracture
energy can be smoothly dissipated during dynamic failure processes without emanating noisy artifact stress waves. This method is
implemented to the standard 4-node quadrilateral finite element; a single quadrature rule is employed with an hourglass control
scheme in order to decrease computational cost and circumvent difficulties associated with the subdomain integration schemes
for cracked elements. The effectiveness and robustness of this method are demonstrated with several numerical examples. In these
examples, we showed the effectiveness of the described correction force scheme along with the applicability of this method to an
interesting class of structural dynamic failure problems.

1. Introduction Our motivation in presenting this paper is to illustrate a

new method which is efficient for dynamic propagation of
It has been shown that the extended finite element method multiple cracks and fragmentation problems but nevertheless
(XFEM) [1, 2] can be successfully applied to several types free from mesh sensitivity by using intra-element discon-
of internal discontinuity problems, focusing on failure prob- tinuities. In this method, we use an element superposition
lems. For example, the XFEM has been applied to arbitrary
concept to represent cracked elements which was proposed
branched and intersecting cracks [3], three dimensional crack
by Song et al. [14].
propagation [4, 5], cohesive crack models [6] and dynamic
shell [7], and 2D [8] fracture problems. In addition, we describe a method to deal with the
Although the standard XFEM has been successfully description of the crack tip element in terms of stiffness. It
applied to dynamic fracture problems by Belytschko et al. [8], aims at characterizing the release of the crack tip element
they mostly considered fracture problems with a single initial when the crack propagates. An additional correction force
notch and/or simple crack geometry. This limitation arises introduced by Menouillard and Belytschko [15] takes into
from the difficulty in the representation of a complicated account this crack tip element and makes the new addi-
crack geometry and with the numerical integration. To tional degrees of freedom release continuously. This artificial
circumvent this difficulty, Xu and Needleman [9], Ortiz and correction force will be used to smooth the stress near the
Pandolfi [10], Repetto et al. [11], and Cirak et al. [12] proposed crack tip because no tip enrichment is used in our XFEM
the interelement crack model; but this interelement crack discretization.
model can have mesh sensitivity problems as pointed out in Menouillard et al. [16] developed a mass lumping strategy
Belytschko et al. [8] and Song et al. [13]. for the XFEM formulation and more particularly for the
2 Mathematical Problems in Engineering

X<a X>a
Phantom nodes
p1 1b
2

1 p2
1a

p1
u u(+)
2
Discontinuity
Crack
X<a X>a
opening
⟦u⟧
1 2 1
1 u(−) p2
X
X=a X=a X
(a) (b)

Figure 1: Representation of a discontinuity in a one-dimensional finite element: (a) problem description of a one-dimensional rod and (b)
representation of a discontinuity with the phantom node method; solid circles denote real nodes and hollow circles denote phantom nodes.

discontinuous enrichment part. They found that the enrich- traction on the Neumann boundary Γ𝑡 , and u is the applied
ment does not significantly decrease the stable time step. displacement on the Dirichlet boundary Γ𝑢0 ; Γ𝑢0 ∪ Γ𝑡0 = Γ0 ,
Then, Menouillard et al. [17] used another decomposition of Γ𝑢 ∩ Γ𝑡 = 0. Superscript plus and minus signs refer to the two
the enriched shape function developed by A. Hansbo and sides of the discontinuity. The spaces of admissible function
P. Hansbo [18] which is used in the phantom node method are
developed by Song et al. [14] and Song and Belytschko [19].
U = {u (X, 𝑡) | u (X, 𝑡) ∈ 𝐶0 , u (X, 𝑡) = u (𝑡)
The outline of this paper is as follows. The governing
(5)
equation and its weak form are given in Section 2. The
on Γ𝑢0 , u discontinuous on Γ𝑐 } ,
representation of a discontinuity in a cracked element with
a phantom node method is presented in Section 3. Section 4
U0 = {𝛿u (X, 𝑡) | 𝛿u (X, 𝑡) ∈ 𝐶0 , 𝛿u (X, 𝑡) = 0
presents a new method for smoothly releasing the newly
(6)
cracked element near the tip by taking into account the
on Γ𝑢0 , 𝛿u discontinuous on Γ𝑐 } .
position of the tracked crack tip in the tip element. Several
numerical examples are given in Section 5. Section 6 presents The weak form of the momentum equation is given by for
the conclusions of this paper. u(X, 𝑡) ∈ U
𝛿𝑊kin = 𝛿𝑊int − 𝛿𝑊ext + 𝛿𝑊coh ∀𝛿u (X) ∈ U0 , (7)
2. Governing Equations and Weak Form
where 𝛿𝑊int is the internal work, 𝛿𝑊ext is the external
For a two-dimensional dynamic problem, the strong form work performed by the applied loads, 𝛿𝑊kin is the kinetic
of the linear momentum equation in a total Lagrangian work performed by inertia forces, and 𝛿𝑊coh is the work
description is performed by the cohesive traction on the crack surface Γ𝑐 .
These quantities are defined as
𝜕𝑃𝑗𝑖
+ 𝜌0 𝑏𝑖 − 𝜌0 𝑢̈
𝑖 = 0 in Ω0 , (1)
𝜕𝑋𝑗 𝛿𝑊kin = ∫ 𝛿u ⋅ 𝜌0 ü𝑑Ω0 ,
Ω0
where P is the nominal stress tensor, 𝜌0 is the initial mass
𝜕𝛿u
density, and b is the body force vector. The boundary 𝛿𝑊int = ∫ : P 𝑑Ω0 , (8)
conditions are Ω0 𝜕X
0
𝑛𝑗0 𝑃𝑗𝑖 = 𝑡𝑖 on Γ𝑡0 (2) 𝛿𝑊ext = ∫ 𝛿u ⋅ 𝜌0 b 𝑑Ω0 + ∫ 𝛿u ⋅ t 𝑑Γ𝑡0 ,
0

Ω0 Γ𝑡0
𝑢𝑖 = 𝑢𝑖 on Γ𝑢0 (3)
𝛿𝑊coh = − ∫ 𝛿 ⟦u⟧ ⋅ 𝜏𝑐 𝑑Γ𝑐 , (9)
𝑛𝑗0 𝑃𝑗𝑖− = −𝑛𝑗0 𝑃𝑗𝑖+ = 𝜏𝑖0𝑐 (⟦𝑢𝑖 ⟧) on Γ𝑐0 , (4) Γ𝑐

where t is the normalized traction prescribed on Γ𝑡0 and 𝜏𝑐 is

where n0 is the unit normal vector to the boundary, 𝜏0𝑐 is the the cohesive traction applied on the discontinuity surface; an
0
cohesive traction across the crack surfaces, t is the applied updated Lagrangian form is used for the cohesive work in (9).
Mathematical Problems in Engineering 3

Γt0 Γt

Γc0
Γc

x = Φ(X, t)
Ω0c

Ω Ωc
Ω0
Γu0
Γu

Figure 2: Representation of a two-dimensional body with an internal discontinuity in the initial and the current domains.

f=0 We can now rewrite the displacement field of element 1 as

a set of two superimposed elements with phantom nodes

u1 (X, 𝑡) = u1a (X, 𝑡) + u1b (X, 𝑡)

1a 1a
n = [𝑢 1 𝑁1 + 𝑝2 𝑁2 ] 𝐻 (− (𝑋 − 𝑎))
⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟
u1a (11)
g<0
1b 1b
f>0 + [𝑝 1 𝑁1 + 𝑢2 ⏟⏟⏟
⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟ 𝑁⏟2⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟
] 𝐻 (𝑋 − 𝑎) ⏟⏟,
g=0 u1b
Discontinuity
f<0 g>0 where a superscript and subscript denote the element and
node number, respectively, N𝐼 are the shape functions, 𝑝 is
Figure 3: Two-dimensional discontinuity representation by two the nodal unknown of the phantom node, and 𝐻(𝑥) is the
implicit functions 𝑓(X) and 𝑔(X, 𝑡). Heaviside step function defined by

1 𝑥≥0
𝐻 (𝑥) = { (12)
0 𝑥 < 0.

The displacement jump across the crack is given by

3. Representation of a Discontinuity with
Phantom Nodes ⟦𝑢⟧𝑋=𝑎 = lim [u1a (𝑋 + 𝜀) − u1b (𝑋 − 𝜀)]𝑋=𝑎
𝜀→0

We first illustrate the modeling of a one-dimensional cracked = {𝑢11a 𝑁1 (𝑎) + 𝑝21a 𝑁2 (𝑎)}
element with phantom nodes and then give the general (13)
description for the modeling of the two-dimensional case,
− {𝑝11b 𝑁1 (𝑎) + 𝑢21b 𝑁2 (𝑎)}
subsequently. Consider a one-dimensional rod and let a crack
be located inside of element 1 at 𝑋 = 𝑎, as shown in
= (u1a 1b
𝐼 − u𝐼 ) 𝑁𝐼 (𝑎) .
Figure 1(a).
The displacement field in cracked element 1 can be seen This procedure for cracked elements is similar to the
to consist of two separated displacement fields as shown in standard XFEM nodal enriching scheme. However, the
Figure 1(b): the displacement field of element 1a for 𝑋 < 𝑎, phantom nodes method simplifies the implementation of
and element 1b for 𝑋 > 𝑎. To construct new elements 1a and cracked elements within the context of existing finite element
1b from the element 1, we add new nodes which are replicas codes, since it is only necessary to add an extra element
of the original nodes; we called these nodes phantom nodes. with phantom nodes and modify the element connectivity
We define phantom nodes by the following: matrices.

3.1. Phantom Node Method in Two Dimensions. Consider

an initial domain Ω0 as shown in Figure 2. The motion is
described by x = Φ(X, 𝑡), where X and x denote material and
element 1a if X𝐼 − 𝑎 > 0
𝐼 is a phantom node in { (10) spatial coordinates, respectively. In the current domain, the
element 1b if X𝐼 − 𝑎 < 0. image of the initial domain Ω0 is denoted by Ω. We allow
4 Mathematical Problems in Engineering

4 3
f(X) < 0
1a
4 3
f(X) < 0 p1 p2
f(X) = 0 1 = +
p4 p3
f(X) > 0
1 2
1b
f(X) > 0
1 2

Figure 4: Representation of a cracked element in two dimensions; solid circles denote real nodes and hollow circles denote phantom nodes.

6 5 10
1c

Crack 1 7 8 9

6 5 4
f1 (X) > 0 13 12
1
f (X) = 0
f1 (X) < 0 1a
1 1 11
2 3

Crack 2
4
f2 (X) < 0 f2 (X) = 0 f2 (X) > 0 16 15

1b
14
2 3

Figure 5: Decomposition of an element into three elements 1a, 1b, and 1c in order to model crack branching; solid and hollow circles denote
the original nodes and the added phantom nodes, respectively.

this domain to contain internal discontinuities Γ𝑐 which is where 𝑓𝐼 = 𝑓(X) and 𝑓(X) = minX∈Γ𝑐 ‖X − X‖. As a
enveloped by a region Ω𝑐 . consequence of (15), the surface of discontinuity can be
Inside of the region Ω𝑐 , we define two local level set represented by 𝑓(X𝐼 ) at the nodes of the cracked elements
functions 𝑓(X) and 𝑔(X, 𝑡), where 𝑓(X) and 𝑔(X, 𝑡) are signed [20, 21]. Note that for the element-by-element cracking
distance functions which describe the crack surface and tip scheme which is employed in this study, we can replace the
geometry, respectively. The isozero line of the function 𝑓(X), function 𝑔(X, 𝑡) by a list of cracked elements.
that is, 𝑓(X) = 0, corresponds to the crack surface Γ𝑐 , and the For a two-dimensional element, the superposed displace-
function 𝑔(X, 𝑡) is defined so that 𝑔(X, 𝑡) > 0 along the crack ment fields in the cracked element can be developed in
surface and vanishes at the crack tip; see Figure 3. By using a similar manner to the one-dimensional case. Consider
a set of these level set functions, we can implicitly define the cracked element 1 and replace the element with element 1a
crack geometry by and 1b as shown in Figure 4.
The displacement field of this superimposed element is
X ∈ Γ𝑐0 if 𝑓 (X) = 0 𝑔 (X, 𝑡) > 0, X ∈ Ω0𝑐 . (14)

For the numerical representation, instead of employing an 𝑛𝑁

implicit definition of the crack surface, generally we can u1a
𝐼 (𝑡) 𝑁𝐼 (X) 𝐻 (−𝑓 (X))
u (X, 𝑡) = ∑ ⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟
approximate the path of an internal discontinuity by 𝐼=1 u1a (X,𝑡)
(16)
1b
∑𝑓𝐼 𝑁𝐼 (X) = 0, + 𝐼 (𝑡) 𝑁𝐼 (X) 𝐻 (𝑓 (X)).
⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟
u
(15)
𝐼 u1b (X,𝑡)
Mathematical Problems in Engineering 5

Mu 0 ü Mu 0 0 ü f ext f int f coh

u 0
f ext
u f int
u f coh
u
u u
= − + 0 Mb 0 b̈ = 0 − f int + f coh + 0
b b
0 Mb b̈ 0 f int
b f coh
b
0 0 Ma ä f int f coh f int coh
0 a a a − fa

DOF DOF
u u
b b
a a
(a) (b)

Mu 0 0 ü f ext f int f coh

Mu 0 0 ü f ext
u f int
u f coh
u 0 u u u

0 Mb 0 b̈ = 0 − f int
b
+ f coh + 0 0 Mb 0 b̈ = 0 − f int
b
+ f coh
b
b

0 0 Ma ä 0 f int
a f coh
a
f correct
a 0 0 Ma ä 0 f int f coh
a a

DOF DOF
u u
b b
a a
(c) (d)

Figure 6: Evolution from the state (a) to (d) through the intermediate states (b) and (c) where the correction force decreases. 𝑀u denotes
the mass matrix related to the standard degrees of freedom u, 𝑀b to the existing discontinuous degree of freedoms b, and 𝑀a to the new
discontinuous degrees of freedom a. f ext denotes the external force, f coh denotes the cohesive force, and f int denotes the internal forces related
to the different kinds of degrees of freedom (i.e. u, b, and a, respectively, denoted by a cross, a continuous circle, and a dotted circle).

The explicit value of the displacement jump is given by elements and the discontinuous displacement fields can be
represented by

⟦u⟧ = lim [u1a (X + 𝜀∇𝑓) − u1b (X − 𝜀∇𝑓)] u (X, 𝑡) = u1a (X, 𝑡) + u1b (X, 𝑡) + u1c (X, 𝑡)
𝜀→0

= [𝑁𝐼 𝐻 (−𝑓 (X)) u1a 1b = ∑ u𝐼 𝑁𝐼 𝐻 (−𝑓1 (X)) 𝐻 (−𝑓2 (X))

𝐼 − 𝑁𝐼 𝐻 (𝑓 (X)) u𝐼 ]X∈Γ (17)
𝑐 𝐼∈𝑆1

= (u1a
𝐼 − u1b
𝐼 ) 𝑁𝐼 (𝑎) . + ∑ u𝐼 𝑁𝐼 𝐻 (−𝑓1 (X)) 𝐻 (𝑓2 (X))
(18)
𝐼∈𝑆2

The concept of element overlapping method can be easily + ∑ u𝐼 𝑁𝐼 𝐻 (𝑓1 (X)) 𝐻 (−𝑓2 (X)) ,
extended to modeling of an arbitrary crack junction or 𝐼∈𝑆3

branching problems. When the original crack 1 branches into

a new crack 2 or a crack 1 junctions with a crack 2, as shown in where 𝑓1 (X) and 𝑓2 (X) are level set functions for crack 1 and
Figure 5, the element can be replaced with three overlapping 2, respectively.
6 Mathematical Problems in Engineering

f correct Table 1: Flowchart for numerical computation procedures.

f int Steps Numerical computation procedures

Crack tip element:
Apply initial conditions including initial boundary
1
conditions and prestress
le 2 Compute initial forces: f0 = f0ext − f0int
3 Compute acceleration: a𝑛 = M−1 𝑛 f𝑛

a 4 Time update: 𝑡𝑛+1 = 𝑡𝑛 + Δ𝑡𝑛+(1/2)

First partial update for nodal velocities:
5
k𝑛+(1/2) = k𝑛 + (𝑡𝑛+(1/2) − 𝑡𝑛 )a𝑛
a/le 6 Apply velocity boundary conditions at 𝑡𝑛+(1/2)
0 1 Update nodal displacements:
7
d𝑛+1 = d𝑛 + Δ𝑡𝑛+(1/2) × k𝑛+(1/2)
Figure 7: Law of the correction force as a function of the fracture 8 Compute strain and stress
ratio of the crack tip element (i.e., the position of the crack tip in the
9 Check fracture criterion and insert phantom nodes
element).
int
10 Compute forces at time: f𝑛+1 = f𝑛+1 ext
− f𝑛+1 + f𝑛+1
coh
+ f𝑛+1
correct

−1
11 Compute acceleration: a𝑛+1 = M𝑛+1 f𝑛+1
𝜎0
Second partial update for nodal velocities:
12
k𝑛+1 = k𝑛+(1/2) + (𝑡𝑛+1 − 𝑡𝑛+(1/2) )a𝑛+1
13 Update counter n
14 Output: if simulation is not completed, go to Step 4
a
2h
momentum equation for the newly added degrees of freedom
becomes

L f kin = f int − f ext + f coh − f correct , (19)

where ‖f correct ‖ tends to zero when the crack tip reaches

Figure 8: Geometry and loading of the semi-infinite plate example. the new edge, and thus the element becomes completely cut
by the discontinuity (see Figure 6(d)). Note that f coh is not
displayed in Figure 6. The initial value of the correction force
If the crack branching angle between crack 1 and crack 2 (when the crack tip is on the previous edge) is such that the
is acute, that is, both cracks cut the element edge of nodes 2 sum of the four forces in the equation above is zero as it
and 5 in Figure 5, the phantom node method cannot resolve is shown in Figure 6(b). At this point, the correction force
this case and in this case, minimal remeshing is required to is the same as the internal force, and thus no acceleration
properly model crack branching. occurs yet on the new additional degrees of freedom, denoted
by a in Figure 6. The flowchart for numerical computation
procedures is described in Table 1.
4. Correction Force The evolution of the correction force is shown in Figure 7
In this section, we aim at developing a method to deal with which describes the magnitude of the correction force as a
the release of the crack tip element when the crack propagates function of the crack tip position in the tip element. Indeed
through. It is to avoid sudden element release near the crack the correction force goes from the initial internal force to
tip during propagation and thus avoid unphysical stress wave zero when the crack tip propagates from one edge to another.
propagation due to the crack propagation. Between these two crack tip positions, the correction force
Figure 6(a) shows the crack tip element with the associ- is taken to be linear in our simulation. However this is not a
ated phantom nodes. When the crack tip reaches the next restriction to use a linear law. In other words, the correction
element, the new crack tip element is suddenly released force is only applied to the newly added degrees of freedom,
(see the sudden passage from Figures 6(a) to 6(d)) because for example, the additional degrees of freedom of the crack
the corresponding internal force takes a significant value tip element (see Figure 7). The law is
when the phantom nodes are injected (i.e, faint is nonzero 𝑎 (𝑡) int
in Figure 6(d)). Our proposed method makes a progressive f𝑒correct (𝑡) = (1 − ) f𝑒 (𝑡inj ) , (20)
𝑙𝑒
release of the crack tip element. This even happens with
the cohesive force. Figures 6(b) and 6(c) show the crack tip where 𝑡inj is the time the corresponding degrees of freedom
element on a dotted line and the additional correction force are injected, 𝑎(𝑡) is the crack propagation speed, and 𝑙𝑒 is the
acting in the momentum equation, which aims at releasing characteristic length of finite elements.
smoothly the element when the crack tip travels through The continuity of the internal force related to the new
from one edge to the next one. Thus the modified discrete additional degrees of freedom gives the same property to the
Mathematical Problems in Engineering 7

1.0926e + 09 1.0926e + 09
9.3749e + 08 9.3749e + 08
7.8241e + 08 7.8241e + 08

Syy-stress

Syy-stress
6.2734e + 08 6.2734e + 08
4.7226e + 08 4.7226e + 08
3.1719e + 08 3.1719e + 08
1.6211e + 08 1.6211e + 08
7.0381e + 06 7.0381e + 06
−1.4804e + 08 −1.4804e + 08
−3.0311e + 08 −3.0311e + 08

(a) (b)

Figure 9: Stress field in the fine mesh: (a) without correction and (b) with correction force.

1.2 1.2

1 1

0.8 0.8

K1 /𝜎0 √h
K1 /𝜎0 √h

0.6 0.6

0.4 0.4

0.2 0.2

0 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
t/tc t/tc

Analytical Analytical
With correction With correction
Without correction Without correction
(a) (b)
45 40
40 35
35 30
Relative error (%)

Relative error (%)

30
25
25
20
20
15
15
10 10

5 5

0 0
1.6 1.8 2 2.2 2.4 2.6 2.8 3 1.6 1.8 2 2.2 2.4 2.6 2.8 3
t/tc t/tc

With correction With correction

Without correction Without correction
(c) (d)

Figure 10: Normalized stress intensity factor as a function of time computed on the (a) coarse mesh and (b) fine mesh; relative error on stress
intensity factor as a function of time on the (c) coarse mesh and (d) fine mesh.

acceleration through the momentum equation. Therefore, the 5. Numerical Examples

velocity and displacement remains quite continuous in time
when additional degrees of freedom are injected, and thus 5.1. Moving Semi-Infinite Mode I Crack. The example consid-
the property of continuity in time remains in the strain and ered in this section is an infinite plate with a semi-infinite
stress field too, and a continuous progressive release of the tip crack [22] loaded as shown in Figure 8. A theoretical solution
element occurs. of this problem for a given crack velocity is given in Freund
8 Mathematical Problems in Engineering

node as shown in Figure 9(a). In contrary spurious stress

0.25W 0.25W waves appear in Figure 9(a) due to the crack propagation and
the sudden release of the crack tip elements.
Figures 10(a) and 10(b), respectively, present the nor-
malized stress intensity factor as a function of time for the
coarse mesh and fine mesh, respectively, with and without the
0.5W 0.27W correction force. Both figures underline that the correction
1.2W

force improves the result during propagation by decreasing

the magnitude of the oscillations due to the released crack tip
element. Indeed the number of oscillations are directly related
to the number of newly cracked elements. To evaluate the
improvement, Figures 10(c) and 10(d) show the relative error
0.325W

between the computations using the correction force and not

for the coarse and fine meshes. The error is decreased from
20% to 5% by adding the correction force on the newly added
1.25W degrees of freedom during the crack propagation.

Figure 11: Compact tension specimen geometry (𝑊 = 0.05 m) and

the thickness is 𝑊/2. 5.2. Stiffened Compact Tension Specimen. The stiffened com-
pact tension specimen is used in various experiments [24, 25].
The particularity of the stiffened test is that an additional part
[23]. According to the geometry described in Figure 8, the of material opposite to the initial crack has the effect of a
analytical solution is valid until time 𝑡 ≤ 3𝑡𝑐 = 3ℎ/𝑐1 stiffener. With such a configuration, the crack will not be able
(where 𝑐1 is the dilatational wave speed). Beyond that, the to propagate straight toward the stiffener, but an instability
reflected stress wave reaches the crack tip and the analytical will make the crack propagate up or down as a curve. Figure 11
solution is no longer valid. The dimensions of the structure presents the geometry of the specimen without showing the
are the following: the length is 𝐿 = 10 m, the initial crack stiffened part; the stiffened part is glued on the right edge of
length 𝑎 = 5 m, and the vertical position of the crack is the specimen as shown in Figure 12(b). A 𝐽2 plasticity theory
ℎ = 2 m. Two regular meshes are used: 78 × 39 and 120 × 59 4- is used to model the behavior of the specimen. The material
node elements. The material properties are Young’s modulus properties are Young’s modulus 𝐸 = 69 GPa, Poisson’s ratio
𝐸 = 210 GPa, Poisson’s ratio ] = 0.3, and density 𝜌 = ] = 0.3, density 𝜌 = 2, 780 kg/m3 , yield stress 𝜎0 = 275 MPa,
8, 000 kg/m3 . The tensile stress applied on the top surface is and hardening slope ℎ = 640 MPa. A constant velocity of
𝜎0 = 500 MPa. The crack velocity is imposed to be zero until 1.8 m/s is applied at the center of two steel bars located in each
1.5 𝑡𝑐 and 1, 500 m/s after. The mode 1 stress intensity factor hole of the specimen.
Figure 12(a) shows the final fracture pattern of the spec-
is normalized by the factor 𝜎0 √ℎ.
imen at the end of the computation. The numerical result
We study the effect of the correction force on the accuracy
shown in Figure 12(a) agrees well with the experimental
of the stress intensity factor of a moving crack. The analytical
result obtained by Galanis [24] as shown in Figure 12(b). The
relation between the stress intensity factor 𝐾1 and the velocity
computed load-deflection curve is also in good agreement
𝑎̇of the crack is given by [23]
with the experiments as shown in Figure 13.
𝐾1 (𝑎,̇𝑡)
5.3. Dynamic Multiple Crack Branchings in a Square Plate.
{ 0 if 𝑡 < 𝑡𝑐 For dynamic fracture problems, crack branching due to
{
{
{
{ 2𝜎0 𝑐1 (𝑡 − 𝑡𝑐 ) (1 − 2]) a dynamic instability is a common phenomenon. Several
{
{ √ if 𝑡𝑐 ≤ 𝑡 < 1.5𝑡𝑐
{
{ experimental results on crack branching have been previously
{
{1 − ] 𝜋 (21)
= { 2𝜎 reported [26–30]. However, because of difficulties in the
{
{ 0 √ 𝑐1 (𝑡 − 𝑡𝑐 ) (1 − 2]) representation of branched crack paths, only a few numerical
{
{
{
{ 1 − ] 𝜋 results have been reported [8, 9, 31] and so forth. Note that
{
{ 1 − ( ̇
𝑎/𝑐 )
{
{ × 𝑟 Belytschko et al. [8] allowed the original crack to branch
if 1.5𝑡𝑐 ≤ 𝑡,
{ ̇ 𝑟)
1 − (𝑎/2𝑐 only once; Xu and Needleman [9] used an element edge
crack model which is less complex than intraelement crack
where the Rayleigh wave speed is 𝑐𝑟 = 2, 947 m/s and the models but has a certain mesh sensitivity; for the issues
dilatational wave speed is 𝑐1 = 5, 944 m/s. regarding mesh sensitivity, refer to Belytschko et al. [8]. Also,
Figure 9 shows the stress contour at the end of the Rabczuk and Belytschko [31] discretely modeled the crack
computation for the two cases, that is, with and without the with the meshfree cracked particle method. In the following,
correction force. One can notice that the correction force we examine the performance of the proposed method in a
makes the stress fields smoother in the structure when the crack branching problem.
crack propagation occurs and eliminates majority of the We consider a 0.1 m by 0.04 m prenotched specimen as
released stress waves due to the abrupt injection of phantom shown in Figure 14. Tensile traction, 𝜎 = 1 MPa, is applied on
Mathematical Problems in Engineering 9

CT3E

(a) (b)

Figure 12: (a) Deformed mesh and (b) experimental postmortem specimen [24].

12 damage law [32] and a linear cohesive law was imposed once
a discontinuity developed. To capture the crack branching
phenomena, we monitored the maximum principal stress
8 criterion at several additional points around the crack tip. If
Force (kN)

the Lemaitre damage criterion is satisfied and the maximum

principal directions show relatively different crack growth
4
angles, we initiate crack branches.
The pattern of multiple crack branchings with damage
evolution is shown in Figure 15. The numerical simulation
0
is executed until one of the crack tips reaches the boundary
0 5 10 15 20 25
Displacement (mm)
of the specimen: 𝑡 = 56.02 𝜇s. The maximum crack tip
speed is around 1250 m/s and the Rayleigh wave speed is
Computation
2100 m/s. The overall dynamic crack branching pattern agrees
Experiment (Fineberg et al. 2003)
with the results which were already reported by Rabczuk and
Figure 13: Comparison of the load deflection curve between the Belytschko [31].
experimental data [24] and computational results.
5.4. Thick Cylinder under Internal Pressure. We consider
a thick cylinder under high internal pressure. The inner
𝜎
and outer radii of the cylinder are 80 mm and 150 mm,
respectively. An internal pressure, 𝑝 = 𝑝0 𝑒−𝑡/𝑡0 , is applied with
𝑝0 = 10 GPa and 𝑡0 = 0.1 ms. The material properties are
𝜌 = 7800 kg/m3 , 𝐸 = 210 GPa, and ] = 0.30. Also, we induced
20 mm

a ±5% perturbation to the elastic modulus to introduce some

asymmetry for initiating the cracks. We modeled the thick
cylinder with 20,000 uniform quadrilateral elements.
20 mm

Because of the high internal pressure, the fragmentation

process occurs only in the first 5.0 𝜇s; then each fragments
moves outward in the radial direction with no further cracks
initiating. Figure 16 shows the magnified deformed mesh
50 mm 50 mm at different time steps. In this simulation, we obtained 6
relatively big fragments and 10 strip shape fragments; for a
clear illustration of the fragments, see Figure 17.
Figure 14: Plate with an initial notch under tensile stress. This overall pattern of cylinder fragmentation is similar
to that already reported by Rabczuk and Belytschko [31].
However, the finite element simulation shown here does not
both the top and bottom edges as a step function in time. We exhibit the small fragments seen in [31]. It is also found
discretized the domain with 100 × 51 uniform quadrilateral that the reduced 4-node quadrilateral finite element with the
elements and used explicit time integration with a Courant hourglass control scheme is quite sensitive to mesh distortion;
number of 0.1. Material softening is modeled with a Lemaitre it would be desirable to use the smoothed finite elements
10 Mathematical Problems in Engineering

1.00 1.00

Damage parameter

Damage parameter
0.88 0.88
0.77 0.77
0.66 0.66
0.55 0.55
0.44 0.44
0.33 0.33
0.22 0.22
0.11 0.11
0.00 0.00

(a) (b)
1.00 1.00

Damage parameter
Damage parameter
0.88 0.88
0.77 0.77
0.66 0.66
0.55 0.55
0.44 0.44
0.33 0.33
0.22 0.22
0.11 0.11
0.00 0.00

(c) (d)

Figure 15: The pattern of crack branching and damage evolution at different time steps: (a) 𝑡 = 30.14 𝜇s, (b) 𝑡 = 45.24 𝜇s, and (c) 𝑡 = 56.02 𝜇s.

(a) (b)

Fragment

Crack

(c) (d)

Figure 16: Deformed shape of the thick cylinder under internal pressure at different time steps: (a) 𝑡 = 1.78 𝜇s, (b) 𝑡 = 3.42 𝜇s, (c) 𝑡 = 6.40 𝜇s,
and (d) 𝑡 = 13.50 𝜇s; the long elements are cracked as indicated in (d).
Mathematical Problems in Engineering 11

(a) (b)

Figure 17: Shape of the fragments of the thick cylinder under internal pressure at different time steps: (a) 𝑡 = 3.72 𝜇s and (b) 𝑡 = 6.71 𝜇s.

[33] or the method proposed by Areias and Rabczuk [34] for simplicity within the context of the conventional explicit
this type of simulations, that is, problems with severe mesh finite element method.
distortion.
Conflict of Interests
6. Conclusion The authors declare that there is no conflict of interests
A numerical method for the simulation of the dynamic regarding the publication of this paper.
propagation of multiple cracks is presented. The method
employs the phantom node method with a one point integra- Acknowledgment
tion scheme. Though the phantom node method is another
form of the standard XFEM, it provides us with a simple The support of the Office of Naval Research under Grants
implementation within the framework of the standard FEM. N00014-13-1-0386 and N00014-11-1-0925 are gratefully
Also, by using one point integration with hourglass control, acknowledged.
we can decrease the computational cost and circumvent the
subdomain integration which is generally used for cracked References
elements. Moreover a correction force can handle the pro-
gressive opening of the crack tip element due to the crack [1] T. Belytschko and T. Black, “Elastic crack growth in finite
propagation, and this improves the results in terms of stress elements with minimal remeshing,” International Journal for
intensity factors. To evaluate the applicability of the proposed Numerical Methods in Engineering, vol. 45, no. 5, pp. 601–620,
method, several numerical examples which have a certain 1999.
complexity in the representation of crack geometry have [2] N. Moës, J. Dolbow, and T. Belytschko, “A finite element method
been analyzed. The smooth crack propagation obtained with for crack growth without remeshing,” International Journal for
a correction force is first checked from an energy point Numerical Methods in Engineering, vol. 46, no. 1, pp. 131–150,
1999.
of view. Then the accuracy of the method is shown by
computating the stress intensity factors for a dynamic mode [3] C. Daux, N. Moës, J. Dolbow, N. Sukumar, and T. Belytschko,
“Arbitrary branched and intersecting cracks with the extended
1 crack propagation. The simulation of a compact tension
finite element method,” International Journal for Numerical
specimen with a stiffener is in good agreement with the Methods in Engineering, vol. 48, no. 12, pp. 1741–1760, 2000.
experiment observations. We simulated a dynamic multiple
[4] P. M. A. Areias and T. Belytschko, “Analysis of three-
crack branching problem and found that the method is dimensional crack initiation and propagation using the
particularly successful in this type of simulations. In the extended finite element method,” International Journal for
simulation of a multiple crack propagation problem, the Numerical Methods in Engineering, vol. 63, no. 5, pp. 760–788,
numerical example shows that the proposed model can 2005.
simulate the growth, interconnection and, finally, failure of [5] Q. Duan, J. H. Song, T. Menouillard, and T. Belytschko,
a plate containing multiple cracks. Also, the simulation of a “Element-local level set method for three-dimensional dynamic
fragment processes is quite effectively analyzed. An attractive crack growth,” International Journal for Numerical Methods in
feature of this method is its low computational cost and Engineering, vol. 80, no. 12, pp. 1520–1543, 2009.
12 Mathematical Problems in Engineering

[6] G. Zi and T. Belytschko, “New crack-tip elements for XFEM [22] K. Ravi-Chandar, Dynamic Fracture, Elsevier Science, New
and applications to cohesive cracks,” International Journal for York, NY, USA, 2004.
Numerical Methods in Engineering, vol. 57, no. 15, pp. 2221–2240, [23] L. B. Freund, Dynamic Fracture Mechanics, Cambridge Univer-
2003. sity Press, Cambridge,UK, 1990.
[7] P. M. A. Areias, J. H. Song, and T. Belytschko, “Analysis [24] K. P. Galanis, Fracture of aluminum naval structures [Ph.D.
of fracture in thin shells by overlapping paired elements,” thesis], Massachusetts Institute of Technology, 2007.
Computer Methods in Applied Mechanics and Engineering, vol.
[25] K. P. Galanis and V. J. Papazoglou, “Effect of stiffening configu-
195, no. 41–43, pp. 5343–5360, 2006.
rations on fracture of aluminium structures,” Strain, vol. 45, no.
[8] T. Belytschko, H. Chen, J. Xu, and G. Zi, “Dynamic crack prop- 2, pp. 131–138, 2009.
agation based on loss of hyperbolicity and a new discontinuous
[26] M. Ramulu and A. S. Kobayashi, “Mechanics of crack curving
enrichment,” International Journal for Numerical Methods in
and branching—a dynamic fracture analysis,” International
Engineering, vol. 58, no. 12, pp. 1873–1905, 2003.
Journal of Fracture, vol. 27, no. 3-4, pp. 187–201, 1985.
[9] X.P. Xu and A. Needleman, “Numerical simulations of fast crack
[27] E. Sharon, S. P. Gross, and J. Fineberg, “Local crack branching
growth in brittle solids,” Journal of the Mechanics and Physics of
as a mechanism for instability in dynamic fracture,” Physical
Solids, vol. 42, no. 9, pp. 1397–1434, 1994.
Review Letters, vol. 74, no. 25, pp. 5096–5099, 1995.
[10] M. Ortiz and A. Pandolfi, “Finite-deformation irreversible
cohesive elements for three-dimensional crack-propagation [28] E. Sharon and J. Fineberg, “Microbranching instability and the
analysis,” International Journal for Numerical Methods in Engi- dynamic fracture of brittle materials,” Physical Review B, vol. 54,
neering, vol. 44, no. 9, pp. 1267–1282, 1999. no. 10, pp. 7128–7139, 1996.
[11] E. A. Repetto, R. Radovitzky, and M. Ortiz, “Finite element [29] K. Ravi-Chandar, “Dynamic fracture of nominally brittle mate-
simulation of dynamic fracture and fragmentation of glass rials,” International Journal of Fracture, vol. 90, no. 1-2, pp. 83–
rods,” Computer Methods in Applied Mechanics and Engineering, 102, 1998.
vol. 183, no. 1-2, pp. 3–14, 2000. [30] J. Fineberg, E. Sharon, and G. Cohen, “Crack front waves in
[12] F. Cirak, M. Ortiz, and A. Pandolfi, “A cohesive approach to dynamic fracture,” International Journal of Fracture, vol. 119, no.
thin-shell fracture and fragmentation,” Computer Methods in 3, pp. 247–261, 2003.
Applied Mechanics and Engineering, vol. 194, no. 21-24, pp. [31] T. Rabczuk and T. Belytschko, “Cracking particles: a simplified
2604–2618, 2005. meshfree method for arbitrary evolving cracks,” International
[13] J.-H. Song, H. Wang, and T. Belytschko, “A comparative study Journal for Numerical Methods in Engineering, vol. 61, no. 13, pp.
on finite element methods for dynamic fracture,” Computa- 2316–2343, 2004.
tional Mechanics, vol. 42, no. 2, pp. 239–250, 2008. [32] J. Lemaitre and J. L. Chaboche, Mechanics of Solid Materials,
[14] J. H. Song, P. M. A. Areias, and T. Belytschko, “A method Cambridge University Press, Cambridge, UK, 1990.
for dynamic crack and shear band propagation with phantom [33] N. Nguyen-Thanh, T. Rabczuk, H. Nguyen-Xuan, and S. P. A.
nodes,” International Journal for Numerical Methods in Engi- Bordas, “A smoothed finite element method for shell analysis,”
neering, vol. 67, no. 6, pp. 868–893, 2006. Computer Methods in Applied Mechanics and Engineering, vol.
[15] T. Menouillard and T. Belytschko, “Correction Force for releas- 198, no. 2, pp. 165–177, 2008.
ing crack tip element with XFEM and only discontinuous [34] P. Areias and T. Rabczuk, “Finite strain fracture of plates and
enrichment,” European Journal of Computational Mechanics, shells with configurational forces and edge rotation,” Interna-
vol. 18, no. 5-6, pp. 465–483, 2009. tional Journal For Numerical Methods in Engineering, vol. 94,
[16] T. Menouillard, J. Réthoré, A. Combescure, and H. Bung, no. 12, pp. 1099–1122, 2013.
“Efficient explicit time stepping for the extended finite element
method (X-FEM),” International Journal for Numerical Methods
in Engineering, vol. 68, no. 9, pp. 911–939, 2006.
[17] T. Menouillard, J. Réthoré, N. Moës, A. Combescure, and H.
Bung, “Mass lumping strategies for X-FEM explicit dynamics:
application to crack propagation,” International Journal for
Numerical Methods in Engineering, vol. 74, no. 3, pp. 447–474,
2008.
[18] A. Hansbo and P. Hansbo, “A finite element method for the sim-
ulation of strong and weak discontinuities in solid mechanics,”
Computer Methods in Applied Mechanics and Engineering, vol.
193, no. 35, pp. 3523–3540, 2004.
[19] J. H. Song and T. Belytschko, “Dynamic fracture of shells
subjected to impulsive loads,” Journal of Applied Mechanics,
Transactions of ASME, vol. 76, Article ID 051301, 2009.
[20] G. Ventura, J. X. Xu, and T. Belytschko, “A vector level
set method and new discontinuity approximations for crack
growth by EFG,” International Journal for Numerical Methods
in Engineering, vol. 54, no. 6, pp. 923–944, 2002.
[21] G. Ventura, E. Budyn, and T. Belytschko, “Vector level sets for
description of propagating cracks in finite elements,” Interna-
tional Journal for Numerical Methods in Engineering, vol. 58, no.
10, pp. 1571–1592, 2003.
Hindawi Publishing Corporation
Mathematical Problems in Engineering
Volume 2013, Article ID 957286, 11 pages
http://dx.doi.org/10.1155/2013/957286

Research Article
Explicit Dynamic Finite Element Method for Predicting
Implosion/Explosion Induced Failure of Shell Structures

Jeong-Hoon Song,1 Patrick Lea,2 and Jay Oswald3

1
Department of Civil and Environmental Engineering, University of South Carolina, Columbia, SC 29208, USA
2
Theoretical and Applied Mechanics, Northwestern University, Evanston, IL 60208, USA
3
School for Engineering of Matter, Transport and Energy, Arizona State University, Tempe, AZ 85287, USA

Correspondence should be addressed to Jeong-Hoon Song; [email protected]

Received 30 July 2013; Accepted 3 September 2013

Academic Editor: Timon Rabczuk

Copyright © 2013 Jeong-Hoon Song et al. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.

A simplified implementation of the conventional extended finite element method (XFEM) for dynamic fracture in thin shells is
presented. Though this implementation uses the same linear combination of the conventional XFEM, it allows for considerable
simplifications of the discontinuous displacement and velocity fields in shell finite elements. The proposed method is implemented
for the discrete Kirchhoff triangular (DKT) shell element, which is one of the most popular shell elements in engineering analysis.
Numerical examples for dynamic failure of shells under impulsive loads including implosion and explosion are presented to
demonstrate the effectiveness and robustness of the method.

1. Introduction Mehra and Chaturvedi [15] used the smooth particle

hydrodynamics (SPH) method for simulations of tearing of
In this work, we describe a method for modeling fractured thethick plates. Combescure et al. [16] and Maurel and
discrete Kirchhoff triangular (DKT) shell elements [1] based Combescure [17] recently developed SPH shell formulations
on the extended finite element method (XFEM) [2, 3]. One of for explicit dynamic method and successfully applied the
the underlying key concepts in the XFEM is the partition of method to the prediction of dynamic fractures in shell
unity approach [4, 5]. In the partition of unity approach, the structures.
approximation basis is spanned by the standard finite element The described implementation scheme is mainly based on
approximation space and extended by the products of the the XFEM, but its actual implementation follows the phan-
standard finite element shape functions with special local tom node method [7, 18] that has been developed by the
characteristic functions which are constructed from knowl- author of this paper. In this approach [7, 18], the element
edge about the solution. The XFEM has since been developed which contains the crack is replaced by two superposed
for two-dimensional dynamic fracture problems [6, 7], static elements with additional nodes. Though this discontinuity
shell fractures [8], and implicit dynamic shell fracture prob- representation scheme uses the same linear combination of
lems [9]. enrichment functions as the conventional XFEM, it allows for
Even though numerous references are available for con- considerable simplifications in fractured thin shell element
tinuum shell elements, the literature on dynamic crack propa- formalisms and furthermore is applicable to arbitrary large
gation in shells is quite limited. Cirak et al. [10, 11] have devel- deformations.
oped a method for dynamic crack propagation in Kirchhoff An elementwise progression of the crack is also employed;
type shells based on interelement cohesive crack methods that is, the crack tip is always on an element edge. The ele-
[12–14]; in the interelement cohesive crack methods, the crack mentwise crack propagation scheme may cause some noise
is limited to propagation along the element edges with local during the crack propagation with coarse meshes. However,
remeshing. in Song et al. [7], it is shown that suchnoise diminishes with
2 Mathematical Problems in Engineering

Bending
3
𝜃̂x3
𝜃̂x1
Corotational coordinates
𝜃̂y3
𝜃̂y1
1
ẑ 3 ̂z3
̂z1
̂
y

1
2 𝜃̂x2
̂
x
𝜃̂y2

2 ̂z2

(a) (b)
Drilling 3
Membrane 3
𝜃̂z3
̂x3 1
̂y3 𝜃̂z1
̂x1 1
̂y1

2
2
̂x2
̂y2 𝜃̂z2

(c) (d)

Figure 1: Kinematic data of the DKT triangular element: (a) the corotational coordinates, (b) the bending degrees of freedom, (c) the in-plane
membrane degrees of freedom, and (d) the drilling degrees of freedom.

mesh refinements, and the crack propagation speeds con- where 𝛽𝑥 = 𝜃𝑦 and 𝛽𝑦 = −𝜃𝑥 . This sign difference is due to the
verge to the progressive crack propagation results [6, 19, 20]. orientation of the rotation, which will generate in-plane dis-
placements. However, with a linear discretization, these con-
2. Discrete Kirchhoff Triangular Shell Element ditions cannot be verified within the entire domain of the
shell finite element. One alternative approach is that they can
The main advantage of the DKT shell element is that a mesh be only verified at some discrete parts of the shell elements,
can easily be generated from any kind of surfaces. The geome- such as a midpoint of each side of the DKT elements; this is
try of the element is described by three linear shape functions the implication of the discrete Kirchhoff assumption in the
in the reference coordinates. The kinematic of the DKT shell DKT shell elements.
elements is described by superimposing the membrane, the To verify the discrete Kirchhoff assumption, one has to
bending, and the rotational (drilling) behavior of shells with add additional shape functions which do not change the
different corresponding degrees of freedoms (DOFs) as nodal values of any field but only are allowed to modify the
shown in Figure 1. values on the midpoint. Thus, the rotational DOFs are dis-
However, for further explanation on the salient features of cretized by
the DKT element, henceforth, we will use 𝛽 instead of 𝜃 for
the rotation as shown in Figure 2; note that this is only for a 3 6
clear illustration purpose. In this notation, 𝛽𝑥 is the rotation 𝛽𝑥 = ∑ 𝑁𝑖 ⋅ 𝛽𝑥𝑖 + ∑ 𝑃𝑖 ⋅ 𝐶𝑖 ⋅ 𝛼𝑖 , (2)
whose displacement is along the axis 𝑥. 𝑖=1 𝑖=4
The discrete Kirchhoff assumption [1] is the following: the 3 6
normal rotations must be equal to the first derivative of the 𝛽𝑦 = ∑ 𝑁𝑖 ⋅ 𝛽𝑦𝑖 + ∑ 𝑃𝑖 ⋅ 𝑆𝑖 ⋅ 𝛼𝑖 . (3)
transverse displacement. These constraints are imposed at a 𝑖=1 𝑖=4
discrete number of points, which leads to the relation bet-
ween the normal rotations and the displacements at the ele- As shown in Figure 3, the discrete Kirchhoff constraints along
ment joints. Let us consider the constraints in terms of 𝑥 and each side 𝑖-𝑗 are introduced at the midpoint 𝑘:
𝑦 directions as
𝛽𝑥 (𝜉) + 𝑤𝑥 = 0, 𝛽𝑠 (𝜉) + 𝑤𝑠 = 0, (4)
(1)
𝛽𝑦 (𝜉) + 𝑤𝑦 = 0, where 𝑠 is the absciss along the side of the element.
Mathematical Problems in Engineering 3

Z Z

𝛽y = −𝜃x 𝛽x = 𝜃y

Y X
(a) (b)

Figure 2: Positive directions of 𝛽𝑥 and 𝛽𝑦 and correspondence between the rotations 𝜃𝑥 and 𝜃𝑦 and 𝛽𝑥 and 𝛽𝑦 .

n
3 j (̂
s
Ck = cos 𝜃k = xji /L k ̂j )
xj , y

Sk = sin 𝜃k = xji /L k
6 5
̂j − x
xji = x ̂i k
̂
y yji = y
̂j − y
̂i
̂
x s n s 𝜃k
1 2 + y2
L k = √xji
s ji i ̂
x
4 2
n (̂ ̂i )
xi , y

Figure 3: Geometry and local tangential-normal coordinate system of the DKT element.

The different functions 𝑃𝑖 appearing in (2) and (3) are So, the expression of 𝛼4 is given as
given by
𝛽𝑥1 𝛽𝑥2
𝛼4 = 𝑤1 − 𝑤2 − − . (9)
𝑃4 = 4 (1 − 𝜉 − 𝜂) 𝜉, (5) 2 2

𝑃5 = 4𝜂𝜉, (6) Similarly, we can determine the other unknowns 𝛼5 and 𝛼6 as

𝛽𝑦1 𝛽𝑦3
𝑃6 = 4 (1 − 𝜉 − 𝜂) 𝜂. (7) 𝛼6 = 𝑤1 − 𝑤3 − − ,
2 2
(10)
Note that, in (2) and (3), 𝐶 and 𝑆 define cosinus and sinus √2
values of the current geometry in the reference coordinates as 𝛼5 = −√2𝑤3 + √2𝑤2 + (𝛽𝑥2 + 𝛽𝑥3 − 𝛽𝑦2 − 𝛽𝑦3 ) .
shown in Figure 4. The particular values for the reference tri- 4
angular element are 𝐶4 = 1, 𝑆4 = 0, 𝐶5 = −√2/2, 𝑆5 = √2/2, However, the other three unknowns 𝛼4 , 𝛼5 , and 𝛼6 in the
𝐶6 = 0, and 𝑆6 = −1. These values are the cosinus and sinus formalism of the rotation should be determined by solving
of the vector representing the side of the reference triangular system equations given by
element. Thus, for example, at the node 4 in Figure 3, we can
show that 𝑥 is the same𝑠 in terms of substituting (2) and (4). √2
[1 − 2𝜉 − 𝜂 − 𝜂 0 ]
Furthermore, for this particular edge that has the midpoint at [ 2 ]
(0, 1/2), we could further develop the following relation: [ ]
[ √ 2 ]
4⋅[ 0 𝜉 𝜉 + 2𝜂 − 1]
[ ]
1 [ 2 ]
−𝑤𝑠 = 𝛽𝑠 ( , 0) [ 𝜉 √2 𝜂 ]
2 − (𝜂 − 𝜉) (11)
[ 2 4 2 ]
1 𝛽 + 𝛽𝑥2
= 𝛽𝑥 ( , 0) = 𝑥1 + 𝛼4 (8)
𝛼4 𝛽𝑥,𝑥
2 2 [ 𝛽 ]
[ ]
⋅ 𝛼5 = [ 𝑦,𝑦 ].
𝑤 − 𝑤1 1
= −𝑤𝑥 = − 2 . [𝛼6 ] [ 2 (𝛽𝑥,𝑦 + 𝛽𝑦,𝑥 )]
2
4 Mathematical Problems in Engineering

Original node
Phantom node

Figure 4: Representation of the crack by the XFEM and the phantom node method with DKT elements.

Throughout this procedure, that is, determining three brevity. The velocity equation for the DKT shells does not
unknowns 𝛼4 , 𝛼5 , and 𝛼6 , the Kirchhoff assumption is vary from that of the other shell elements but the rotations do.
enforced in the middle of each side of the shell element. Keeping with the notation from the last section the rotations
Note that the nodal value is still intact because of the shape in 𝑥 and 𝑦 can be expressed as
functions 𝑃𝑖 which are zero at the nodes.

3. Representation of Fractured Shell Element 3

𝛽𝑥 = ∑ 𝑁𝑖 (𝐻 (−𝑓 (x, 𝑡))) 𝛽𝑥𝑖
Based on the phantom node approach [7, 18], the DKT shell 𝑖=1
element which contains a crack is replaced by two superposed
6
DKT elements with additional phantom nodes as shown in
+ ∑ 𝑃𝑖 (𝐻 (−𝑓 (x, 𝑡))) 𝐶𝑖 ⋅ 𝛼𝑖
Figure 4. 𝑖=4
As with the standard approach to phantom nodes [7, 18],
cracks will be inserted elementwise at propagation and the 3
crack surface will be limited to normal to the shells midsur- + ∑ 𝑁𝑗 (𝐻 (−𝑓 (x, 𝑡))) 𝛽𝑥𝑗
face. While the equations here will be for the phantom node 𝑗=1
method, the equivalence between phantom node method and
6
XFEM is shown by Song et al. [7].
+ ∑ 𝑃𝑗 (𝐻 (−𝑓 (x, 𝑡))) 𝐶𝑗 ⋅ 𝛼𝑗 ,
The discontinuous velocity fields in the midsurface of the
𝑗=4
fractured shell elements can be described by
(14)
3
3
𝛽𝑦 = ∑ 𝑁𝑖 (𝐻 (−𝑓 (x, 𝑡))) 𝛽𝑦𝑖
kmid = ∑ 𝑁𝑖 (𝐻 (−𝑓 (x, 𝑡))) k𝑖 𝑖=1
𝑖=1
(12) 6
3
+ ∑ 𝑁𝑗 (𝐻 (𝑓 (x, 𝑡))) k𝑗 , + ∑ 𝑃𝑖 (𝐻 (−𝑓 (x, 𝑡))) 𝑆𝑖 ⋅ 𝛼𝑖
𝑖=4
𝑗=1
(15)
3
+ ∑ 𝑁𝑗 (𝐻 (−𝑓 (x, 𝑡))) 𝛽𝑦𝑗
where 𝑖 and 𝑗 are the first and second halves of the pair of over- 𝑗=1
lapping element, respectively, 𝑓(x) is an implicit function that
can describe the geometry of the crack surface in the mid- 6
surface of the shell with iso-zero line, that is; crack path in the + ∑ 𝑃𝑗 (𝐻 (−𝑓 (x, 𝑡))) 𝑆𝑗 ⋅ 𝛼𝑗 ,
midsurface of shell can be represented by 𝑓(𝜉) = 0, and 𝐻(𝑥) 𝑗=4
is the step function given by

1, 𝑥 ≥ 0, where again an overlapping pair (elements 𝑖 and 𝑗) is used.

𝐻 (𝑥) = { (13) One advantage of using the phantom node method with DKT
0, 𝑥 < 0.
shells is that the procedure for finding 𝛼’s does not need to be
varied from the standard method. This greatly simplifies the
The velocity fields can also be expressed in corotational implementation of the XFEM for DKT elements, leaving (5)–
coordinates as is common in shell, but we have omittedit for (11) unchanged.
Mathematical Problems in Engineering 5

4. Computation Procedures enriched DOFs also depends on the enrichment function 𝐻

as follows:
4.1. Time Integration: Newmark Scheme. In this work, New-
𝑚 1
mark scheme for the explicit time integration is used. The 𝑀𝑖 = ∫ 𝐻2 𝑑Ω𝑒 ,
𝑛node 󵄩󵄩󵄩󵄩Ω𝑒 󵄩󵄩󵄩󵄩 Ω
(21)
time integration procedure is written as

Δ𝑡2 ̈ where 𝑚 is the total actual mass of the element, 𝑛node is the
𝑈𝑡+Δ𝑡 = 𝑈𝑡 + Δ𝑡 ⋅ 𝑈̇
𝑡+ 𝑈, (16) number of nodes of the element, and ‖Ω𝑒 ‖ is the measure of
2 𝑡
the finite element domain Ω𝑒 . For the particular case of dis-
𝑈̇ ̇ 1 ̈ ̈
𝑡+Δ𝑡 = 𝑈𝑡 + Δ𝑡 (𝑈𝑡 + 𝑈𝑡+Δ𝑡 ) , (17)
continuous enriched functions such as Heaviside function,
2 the term of the mass matrix corresponding to an enriched
node is in fact just a fraction of the regular finite element term;
𝑀 ⋅ 𝑈̈𝑡+Δ𝑡 = 𝐹ext − 𝐹int , (18) in other words, the lumped mass matrix for the enriched
nodes is written as follows:
where 𝑈𝑡 (𝑈̇ ̈
𝑡 and 𝑈𝑡 , resp.) denotes the displacement (velocity
and acceleration, resp.) at time 𝑡. Δ𝑡 is the time integration 𝑚 𝐴𝑖
𝑀𝑖 = , (22)
step, 𝑀 is the mass matrix, and 𝐹ext (𝐹int , resp.) is the external 𝑛node 𝐴
(internal, resp.) forces at time 𝑡.
A diagonal mass matrix is frequently used in this explicit where 𝐴 is the volume or area of the regular element and 𝐴 𝑖
time integration scheme because it allows us to avoid a ma- represents the fraction ratio of the cut element. One impera-
trix inversion for solving (14); that is, no matrix inversion ap- tive advantage of this mass lumping scheme is that this met-
pears in this scheme. Consequently, the main advantage of us- hod does not significantly decrease the critical time step of the
ing explicit time integration scheme is to speed up the compu- continuum element [22].
tation and use less memory by storing only vectors instead of
matrices to the computer. 5. Material Model and Modeling of Fracture
However, this explicit integration scheme is conditionally
stable, and the stability condition is defined in terms of a max- 5.1. Damage Plasticity Model. A damage plasticity model that
imum time step Δ𝑡𝑐 ; we usually name it a critical time step can account for the effects of stress triaxiality and Lode angle
since it is the largest time step that can be used. The critical was proposed by Xue [23] and Xue and Wierzbicki [24]. In
time step is evaluated from the eigenvalue analysis with the this constitutive model, damage of a material point is accessed
mass 𝑀 and stiffness matrix 𝐾 since by measuring the accumulation of the following damage
increment:
2
Δ𝑡𝑐 = , (19) 𝜀𝑝 𝑚−1
d𝜀𝑝
𝜔max
d𝐷 = 𝑚( ) , (23)
𝜀𝑓 𝜀𝑓
where 𝜔max is the maximum frequency determined by solving
eigenvalue of problems in (16): where 𝐷 is damage parameter, 𝜀𝑝 is the plastic strain, 𝜀𝑓 is a
reference strain envelope, and 𝑚 is a material constant. The
det (𝐾 − 𝜔2 𝑀) = 0. (20) reference strain envelope is a function of the pressure and the
Lode angle:
The stability of explicit time integration for the XFEM is de-
fined by the same condition on the mass and the stiffness; the 𝜀𝑓 = 𝜀𝑓0 𝜇𝑝 (𝑝) 𝜇𝜃 (𝜃𝐿 ) , (24)
computation frequency must be greater than the greatest vi-
bration frequency of the structure. where 𝜀𝑓0 is the initial reference strain, 𝑝 is the pressure, and
𝜃𝐿 is the Lode angle. The functions 𝜇𝑝 and 𝜇𝜃 are defined as
4.2. Computation of Lumped Mass Matrix for Cracked Ele- 𝑝
ments. In the explicit dynamic analysis method, construc- 𝜇𝑝 (𝑝) = 1 − 𝑞 log (1 − ),
𝑝lim
tions of lumped mass are essential to ensure the computation (25)
󵄨 󵄨 𝑘
of nodal accelerations without implicit solution procedures. 6 󵄨󵄨𝜃 󵄨󵄨
However, the mass lumping scheme for cracked elements 𝜇𝜃 (𝜃𝐿 ) = 𝛾 + (1 − 𝛾) ( 󵄨 𝐿 󵄨 ) ,
which employ the XFEM approach is not obvious. To circum- 𝜋
vent such difficulties, several methods have been proposed: where 𝑞 and 𝑝lim are material constants and 𝛾 and 𝑘 are
implicit (in cracked elements)-explicit (in continuum ele- parameters determining the shape of the strain envelope. The
ments) time integration scheme [6] and modified mass lump- weakening effect caused by the damage was also considered
ing schemes [21, 22]. in this model:
In this study, the lumped mass for regular DOFs is diag-
onalized by the conventional row sum mass lumping tech- 𝜎 = 𝑤 (𝐷) 𝜎𝑀 = (1 − 𝐷𝛽 ) 𝜎𝑀, (26)
nique, but, for the cracked elements, we used the mass lump-
ing scheme that was proposed by Menouillard et al. [22]. where 𝜎𝑀 is the stress of the undamaged material, 𝑤(𝐷) is the
Thus, the diagonal term 𝑖 of the mass corresponding to the weakening function, and 𝛽 is another material constant.
6 Mathematical Problems in Engineering

principal
𝜎max >0
r𝛼

lc

Crack

principal
𝜎max <0

(a) (b)

Figure 5: Schematic of averaging domain for the evaluation of the fracture criterion: (a) the size of averaging domain and (b) possible principal
stress states through the shell depth.

Rigid end plug

t
tip element; see Figure 5. Note that, for the computation of the
averaged strain 𝜀avg , we only included material points which
D show tension dominant states as shown in Figure 5(b).

Le L Le
5.3. Dissipation of Fracture Energy. In this study, a cohesive
Figure 6: Setup for implosion induced failure of cylinder. crack model is prescribed along the newly injected strong dis-
continuity surfaces until the crack opening is fully developed,
that is, until cohesive traction has vanished. The roles of a pre-
p scribed cohesive model can be summarized as follows.
Pcr (1) It can be a remedy to spurious mesh-dependent path-
ological behaviors by providing a bounded solution at
the crack tip. For linear elastic fracture simulations, if
the crack tip is not smoothly closed with cohesive for-
ces, finite element solutions are unbounded at the
crack tip due to the crack tip stress singularity and a
crack path is determined by the surrounding mesh re-
solution. Also, for fracture in plastic bulk materials,
t0 t the crack tip stress singularity can be slightly allevi-
ated by plasticity. However, the finite element solu-
Figure 7: Loading curve used in the simulation of Texas experi-
tions still depend on the mesh resolution.
ments.
(2) If the crack opening displacement is not governed by
a cohesive model, the normal stress component to the
5.2. Fracture Criterion. A critical strain based fracture crite- crack surface suddenly drops to zero due to lack of
rion is used to determine the onset point of a poststrain lo- fracture energy dissipations; note that injecting a
calization behavior of a material, that is, fracture. When the strong discontinuity without prescribing cohesive
strain at a crack tip material point reaches a fracture thresh- force is the same as creating two free surfaces without
old, we inject a strong discontinuity ahead of the previous dissipating new surface initiation energies. In this
crack tip according to maximum principle tensile strain di- case, the total system suffers from an excessive accu-
rection of an averaged strain 𝜀avg . For the computation of the mulation of elastic energy and this excessively accu-
averaged strain, 𝜀avg , we used a pointwise weighted averaging mulated energy accelerates the crack propagation
scheme which is given by speed; more discussions on the relationship between
crack propagation speed and dissipated fracture ener-
∑𝛼 𝑤 (𝑟𝛼 ) 𝜀𝛼 gy can be found in Rabczuk et al. [27].
𝜀avg = , (27)
∑𝛼 𝑤 (𝑟𝛼 )
In this study, we only prescribe the normal traction of a linear
where 𝑤(𝑟) is the cubic spline weight function, 𝑟𝛼 is the dis- cohesive model since the early stage of crack initiation due
tance from the crack tip to the material points 𝛼, 𝑟𝑐 (≃ 3ℎ𝑒 ) is to implosion or explosion is mostly due to mode 1 fracture
the size of the averaging domain, and ℎ𝑒 is the size of the crack behavior.
Mathematical Problems in Engineering 7

(a) (b)

(c) (d)

Figure 8: Snapshots of numerical results of IMP26 experiment: (a) initial configuration and the mesh, (b) deformed configuration at time
𝑡 = 1.48 ms; the center of the cylinder begins to collapse, (c) deformed configuration at time 𝑡 = 1.78 ms; buckling region enlarges toward the
two ends, and (d) final deformed configuration; the cylinder buckles in mode 2.

(a) (b) (c)

(d) (e)

Figure 9: Snapshots of numerical results of IMP25 experiment: (a) deformed configuration at time 𝑡 = 0.92 ms; cylinder collapses at mode
3, (b) deformed configuration at time 𝑡 = 1.08 ms; cracks initiated at the two ends, (c) the final deformed configuration, (d) local zoom view
of the crack, and (e) view from the axial direction.

(a) (b)

(c) (d)

Figure 10: Comparison of final configurations of cylinders with different imperfection magnitudes: (a) 0.05% imperfection, (b) 0.1%
imperfection, (c) 0.5% imperfection, and (d) 1.0% imperfection.
8 Mathematical Problems in Engineering

6000 cylinder along its longitudinal axis. The summation of the

magnitude of these concentrated forces is equivalent to the
4000
total force exerted on the end plug due to the pressure. All the
2000 nodes in the two extension parts are only allowed to move
along the 𝑥-axis, and all the other DOFs of these nodes,
Velocity (in/s)

0 including the translational DOFs 𝑢𝑦 and 𝑢𝑧 and the rotational

DOFs 𝜃𝑥 , 𝜃𝑦 , and 𝜃𝑧 , are constrained. The employment of
−2000 the applied concentrated forces and the sliding boundary
−4000 conditions is to model the effect of the plug.
A bilinear load curve for the pressure was used in both
−6000 simulations. The pressure started at zero and was increased
to 𝑝𝑐𝑟 , that is, the experimental critical buckling pressure, in
−8000 time 𝑡0 and then was kept constant at 𝑝𝑐𝑟 until the end of
0 0.2 0.4 0.6 0.8 1
t (ms) simulation. The schematic of time history of the pressure is
shown in Figure 7; 𝑝𝑐𝑟 = 1.15 MPa for IMP26 experiment and
1.0% 0.1% 𝑝𝑐𝑟 = 2.83 MPa for IMP25 experiment. We used the para-
0.5% 0.05% meter 𝑡0 = 5.0 × 10−5 s in both simulations.
Figure 11: Comparisons of time history of nodal velocity at the Geometry imperfection is introduced into the radius to
center node with different amplitudes of imperfection. evoke circumferential buckling easily. The actual radius with
imperfection has the form 𝑟 = 𝑟0 (1 − 𝑎 cos 𝑛𝜃), where 𝑟0 is the
unperturbed radius, 𝑎 is the imperfection magnitude, and 𝜃
is the circumferential angle. 𝑛 is the number of wavelengths
6. Numerical Examples in the circumferential direction, 𝑛 = 2 for IMP26 simulation,
and 𝑛 = 3 for IMP25 simulation. It should be noted that the
6.1. Implosion Induced Dynamic Failure of Cylinder. The buckling mode in numerical results does not depend on the
cylinder is of length 𝐿 with two extension parts of length 𝐿 𝑒 at imperfection mode. We also tested the IMP25 simulation
the ends; see Figure 6. Rigid plugs are perfectly bonded to the with injection of mode 2 imperfection and obtained very sim-
extension parts. The thickness of the cylinder is 𝑡, and the ilar results.
diameter is 𝐷. In the experiments, the specimens are loaded Figures 8(a)–8(d) show four snapshots of the numerical
with hydrostatic pressure and the pressure was increased until results for IMP26. The center of the specimen yielded first and
the cylinders buckled. It was observed in the experiment that then buckled in mode 2. The buckling region evolved toward
specimens will buckle in some specific modes, which have a the two ends, and the specimen entirely collapsed. 𝑎 = 0.1%
dependence on the geometry. was used in the simulation of IMP26, which is close to 0.107%,
In this example, we focused on predicting final fracture the maximum ovalization measured in the experimental
pattern of two experiments. The first specimen is denoted by specimen.
IMP26 experiment, where 𝐿 = 366.8 mm, 𝐷 = 38.085 mm, In IMP25 simulation, we allowed an injection of the dis-
and 𝑡 = 0.701 mm. The length of the extension parts is continuity near the interface of the main part and the ex-
𝐿 𝑒 = 25.4 mm. The material of the cylinder is aluminum tension part of the specimen. We also observed large plastic
alloy (AL) 6061-T6. In the experiment, the result indicated strain, large damaged or unstable material points along the
a mode 2 implosion, but there was no fracture observed. The central buckling lines. However, this may be due to the repul-
second specimen, denoted by IMP25, is 𝐿 = 143 mm plus two sive forces generated during the contact, so no crack was al-
25.4 mm extension parts at the ends, 𝐷 = 38.087 mm, and lowed to initiate in these regions.
𝑡 = 0.701 mm. Because it has a much shorter length than that Several imperfection magnitudes 𝑎 from 0.05% to 1%
of IMP26, IMP25 specimen buckled in a higher mode, that were tested in IMP25 simulation. The maximum oval imper-
is, mode 3 buckling, and showed fracture on the interfaces of fection measured in the experiment is 0.043%. It should be
the end plugs and the main specimen. noted there should not be imperfection as large as 1% in the
For numerical analysis, we modeled IMP26 and IMP25 actual specimen, and this is only to examine the effect of the
specimens with 49200 and 24000 shell elements, respectively. imperfection.
The average element size is about 1 mm. The material behavior Figure 9 shows four snapshots of IMP25 results of 𝑎 =
of AL6061-T6 is modeled with the damage plasticity [23, 24] 0.05% at different times. The central part of the cylinder col-
with Young’s modulus 𝐸 = 69.5 GPa, the density 𝜌 = 2780 lapses first, and then the collapse region enlarges toward the
kg/m3 , Poisson ratio ] = 0.3, yield stress 𝜎𝑦0 = 276 MPa, and ends. Some cracks were initiated at the ends and propagated
the hardening modulus 𝐸𝑇 = 634 MPa. along the circumferential direction.
The following coordinate system was used: the 𝑥-axis is The final configurations of different imperfection mag-
along the axis of the cylinder, and the 𝑦- and 𝑧-axes are in nitudes are compared in Figure 10. No large effect can be
the radial direction. A uniform surface pressure loading was observed for imperfection magnitudes though the crack
applied on all the shell elements. Another set of concentrated opening of 𝑎 = 1.0% looks larger than that of 𝑎 = 0.05%. The
forces along the axis direction were applied to the nodes on time histories of velocity 𝑦 of finite element node 11670 (at
the rim of the two ends. These forces served to compress the the center of cylinder) are plotted and compared in Figure 11.
Mathematical Problems in Engineering 9

L
Detonation wave Le

D
Preflawed
Detonation tube specimen
Extension tube
Initiation source Flange

Figure 12: Preflawed cylinder with gaseous detonation loading.

(a) (b)

Figure 13: Experimental results of Shepherd experiment [25, 26]: (a) both the forward and backward tips curved and (b) backward tip curving
and forward tip bifurcation.

Let the collapse time be defined as the time to reach the peak pressure wave, 𝑡𝑐𝑗 = 𝑥/V𝑐𝑗 is the time for the wave to travel
velocity. It can be observed that smaller imperfection leads to from the initiation point to the evaluation point, V𝑐𝑗 is the
longer collapse time, but the peak velocity of all the imper- velocity of the wave, and the time parameter 𝑇 = 3.0𝑡𝑐𝑗 . In the
fections does not show an obvious dependence on the imper- current simulation, 𝑝𝑐𝑗 = 6.1 Mpa and V𝑐𝑗 = 2404 m/s. The
fection magnitude. pressure was applied to all the elements from the internal
side. The material properties are Young’s modulus 𝐸 = 2780
6.2. Explosion Induced Failure of Cylinder. Chao [25] and kg/m3 , Poisson ratio ] = 0.3, initial yield 𝜎𝑦0 = 275 MPa, and
Chao and Shepherd [26] performed a series of experiments hardening modulus 𝐸𝑇 = 640 MPa.
with gaseous detonation loading. The schematic setting of the The configurations at different times of numerical results
experiment is shown in Figure 12. The preflawed cylinder was are shown in Figures 14 and 15. The stress concentration can
linked to a detonation tube and an extension tube. The initial be seen in front of the crack tip. The crack tips went straight at
surface notch was located at the center of the cylinder. All the the early stage of propagation. Both the forward and the back-
tubes were filled with explosive gas. The detonation source ward tips curved after around 0.3 ms. The forward crack tip
point is located inside the detonation tube, 1.52 m away from shows sharper curving at about 90 degrees to the axis. The
the left end of the preflawed tube. backward crack tip shows a slanted path.
The pressure wave was initiated at the source point and The final fracture patterns are shown in Figure 15. The
then passed the specimen and the extension tube, causing the crack propagation paths are similar to the experimental re-
original surface notch to form a crack cutting through the cyl- sults shown in Figure 13. The difference is that our results
inder wall and propagate. show a little shorter crack length, which may be due to the dif-
Chao [25] and Chao and Shepherd [26] launched 9 shots ference of the loading and boundary conditions between nu-
with the above specimen with the initial notch length 𝐿 𝑐 = merical modeling and real experiment.
25.4 mm. The length of the specimen is 𝐿 = 0.61 m, the dia-
meter 𝐷 = 0.038 m, and the shell thickness 𝑡 = 0.89 mm.
They found two types of fracture behaviors. One was that 7. Conclusion
both the forward and the backward crack tips curved after
they were formed and went straight for a short distance, as We described a new finite element method for prediction of
shown in Figure 13(a). The other type of results also showed a dynamic fractures in thin shells. The method is incorporated
backward crack curving but the forward crack tip bifurcated within an explicit time integration scheme and able to repre-
and finally cut the specimen into two segments. The configu- sent the crack paths free from initial mesh topologies. For the
ration of the second fracture pattern is shown in Figure 13(b). representation of discontinuities due to cracks, the described
The specimen is modeled with 40680 shell elements. The method employs a simplified version of the conventional
left and the right ends of the numerical models were fully XFEM based on the phantom node method. In this approach,
clamped in the simulation. The following fitted exponential- the cracked shell element is treated by two superimposed
decay curve [28] was used to represent the detonation pres- elements with newly added phantom nodes on the cracked
sure: portions.
The method is implemented for the DKT shell element.
{0, 0 < 𝑡 < 𝑡𝑐𝑗 ,
This facilitates the implementation of the method into stan-
𝑝 (𝑥, 𝑡) = { 𝑡 − 𝑡𝑐𝑗 (28)
𝑝𝑐𝑗 exp (− ) , 𝑡 > 𝑡𝑐𝑗 , dard finite element programs. Another attractive feature of
{ 𝑇 the method is that it provides an easy mesh generation and a
where 𝑥 is the distance away from the initiation point along relatively low computational cost and this allows large scale
the axial direction, 𝑡 is time, 𝑝𝑐𝑗 is the peak value of the nonlinear dynamic fracture problems to be solved efficiently.
10 Mathematical Problems in Engineering

(a) (b)

(c) (d)

Figure 14: Numerical results of Shepherd experiment at different times: deformed configurations with effective stress contour plots at
(a) time 𝑡 = 0.255 ms and (b) time 𝑡 = 0.3 ms and with damage plot at (c) time 𝑡 = 0.42 ms and (d) time 𝑡 = 0.54 ms.

(a) (b)

Figure 15: Final configurations of Shepherd experiment. (a) Top view. (b) Side view.

Conflict of Interests [6] T. Belytschko, H. Chen, J. Xu, and G. Zi, “Dynamic crack prop-
agation based on loss of hyperbolicity and a new discontinuous
The authors declare that there is no conflict of interests enrichment,” International Journal for Numerical Methods in
regarding the publication of this paper. Engineering, vol. 58, no. 12, pp. 1873–1905, 2003.
[7] J.-H. Song, P. M. A. Areias, and T. Belytschko, “A method
Acknowledgment for dynamic crack and shear band propagation with phantom
nodes,” International Journal for Numerical Methods in Engi-
The support of the Office of Naval Research under Grants neering, vol. 67, no. 6, pp. 868–893, 2006.
nos. N00014-13-1-0386 and N00014-11-1-0925 is gratefully [8] P. M. A. Areias and T. Belytschko, “Non-linear analysis of
acknowledged. shells with arbitrary evolving cracks using XFEM,” International
Journal for Numerical Methods in Engineering, vol. 62, no. 3, pp.
384–415, 2005.
References [9] P. M. A. Areias, J. H. Song, and T. Belytschko, “Analysis
of fracture in thin shells by overlapping paired elements,”
[1] J. L. Batoz, K. J. Bathe, and L. W. Ho, “Study of three-node
Computer Methods in Applied Mechanics and Engineering, vol.
triangular plate bending elements,” International Journal for
195, no. 41-43, pp. 5343–5360, 2006.
Numerical Methods in Engineering, vol. 15, no. 12, pp. 1771–1812,
1980. [10] F. Cirak, M. Ortiz, and A. Pandolfi, “A cohesive approach to
[2] T. Belytschko and T. Black, “Elastic crack growth in finite thin-shell fracture and fragmentation,” Computer Methods in
elements with minimal remeshing,” International Journal for Applied Mechanics and Engineering, vol. 194, no. 21-24, pp.
Numerical Methods in Engineering, vol. 45, no. 5, pp. 601–620, 2604–2618, 2005.
1999. [11] F. Cirak, R. Deiterding, and S. P. Mauch, “Large-scale fluid-
[3] N. Moës, J. Dolbow, and T. Belytschko, “A finite element method structure interaction simulation of viscoplastic and fracturing
for crack growth without remeshing,” International Journal for thin-shells subjected to shocks and detonations,” Computers and
Numerical Methods in Engineering, vol. 46, no. 1, pp. 131–150, Structures, vol. 85, no. 11-14, pp. 1049–1065, 2007.
1999. [12] X.-P. Xu and A. Needleman, “Numerical simulations of fast
[4] J. M. Melenk and I. Babuška, “The partition of unity finite crack growth in brittle solids,” Journal of the Mechanics and
element method: basic theory and applications,” Computer Physics of Solids, vol. 42, no. 9, pp. 1397–1434, 1994.
Methods in Applied Mechanics and Engineering, vol. 139, no. 1- [13] G. T. Camacho and M. Ortiz, “Computational modelling of
4, pp. 289–314, 1996. impact damage in brittle materials,” International Journal of
[5] I. Babuška and J. M. Melenk, “The partition of unity method,” Solids and Structures, vol. 33, no. 20-22, pp. 2899–2938, 1996.
International Journal for Numerical Methods in Engineering, vol. [14] M. Ortiz and A. Pandolfi, “Finite-deformation irreversible
40, no. 4, pp. 727–758, 1997. cohesive elements for three-dimensional crack-propagation
Mathematical Problems in Engineering 11

analysis,” International Journal for Numerical Methods in Engi-

neering, vol. 44, no. 9, pp. 1267–1282, 1999.
[15] V. Mehra and S. Chaturvedi, “High velocity impact of metal
sphere on thin metallic plates: a comparative smooth particle
hydrodynamics study,” Journal of Computational Physics, vol.
212, no. 1, pp. 318–337, 2006.
[16] A. Combescure, B. Maurel, and S. Potapov, “Modelling dynamic
fracture of thin shells filled with fluid: a fully SPH model,”
Mecanique & Industries, vol. 9, no. 2, pp. 167–174, 2008.
[17] B. Maurel and A. Combescure, “An SPH shell formulation for
plasticity and fracture analysis in explicit dynamics,” Interna-
tional Journal for Numerical Methods in Engineering, vol. 76, no.
7, pp. 949–971, 2008.
[18] J.-H. Song and T. Belytschko, “Dynamic fracture of shells
subjected to impulsive loads,” Journal of Applied Mechanics, vol.
76, no. 5, pp. 1–9, 2009.
[19] J. Réthoré, A. Gravouil, and A. Combescure, “An energy-
conserving scheme for dynamic crack growth using the
eXtended finite element method,” International Journal for
Numerical Methods in Engineering, vol. 63, no. 5, pp. 631–659,
2005.
[20] T. Rabczuk, G. Zi, A. Gerstenberger, and W. A. Wall, “A new
crack tip element for the phantom-node method with arbitrary
cohesive cracks,” International Journal for Numerical Methods in
Engineering, vol. 75, no. 5, pp. 577–599, 2008.
[21] T. Menouillard, J. Réthoré, A. Combescure, and H. Bung,
“Efficient explicit time stepping for the eXtended finite element
method (X-FEM),” International Journal for Numerical Methods
in Engineering, vol. 68, no. 9, pp. 911–939, 2006.
[22] T. Menouillard, J. Réthoré, N. Moës, A. Combescure, and H.
Bung, “Mass lumping strategies for X-FEM explicit dynamics:
application to crack propagation,” International Journal for
Numerical Methods in Engineering, vol. 74, no. 3, pp. 447–474,
2008.
[23] L. Xue, “Damage accumulation and fracture initiation in
uncracked ductile solids subject to triaxial loading,” Interna-
tional Journal of Solids and Structures, vol. 44, no. 16, pp. 5163–
5181, 2007.
[24] L. Xue and T. Wierzbicki, “Ductile fracture initiation and prop-
agation modeling using damage plasticity theory,” Engineering
Fracture Mechanics, vol. 75, no. 11, pp. 3276–3293, 2008.
[25] T. W. Chao, Gaseous detonation-driven fracture of tubes [Ph.D.
dissertation], California Institute of Technology, Pasadena,
Calif, USA, 2004.
[26] T. W. Chao and J. E. Shepherd, “Fracture response of externally
flawed aluminum cylindrical shells under internal gaseous
detonation loading,” International Journal of Fracture, vol. 134,
no. 1, pp. 59–90, 2005.
[27] T. Rabczuk, J.-H. Song, and T. Belytschko, “Simulations of insta-
bility in dynamic fracture by the cracking particles method,”
Engineering Fracture Mechanics, vol. 76, no. 6, pp. 730–741, 2009.
[28] W. M. Beltman and J. E. Shepherd, “Linear elastic response
of tubes to internal detonation loading,” Journal of Sound and
Vibration, vol. 252, no. 4, pp. 617–655, 2002.