See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/318925656

Phase-field simulation of interactive mixed-

mode fracture tests on cement mortar with
full-field displacement boundary...

Article in Engineering Fracture Mechanics · August 2017

DOI: 10.1016/j.engfracmech.2017.06.014

CITATIONS READS

0 101

4 authors, including:

T. Wu Martin Poncelet
Technische Universität Braunschweig Ecole normale supérieure de Cachan
17 PUBLICATIONS 70 CITATIONS 52 PUBLICATIONS 295 CITATIONS

SEE PROFILE SEE PROFILE

Laura De Lorenzis
Technische Universität Braunschweig
148 PUBLICATIONS 3,599 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Reproducing Sneddon's solution using fracture PhF formulations accurately is not easy View project

FRP Composites View project

All content following this page was uploaded by Laura De Lorenzis on 28 December 2017.

The user has requested enhancement of the downloaded file.

 1 Phase-field simulation of interactive mixed-mode fracture
2 tests on cement mortar with full-field displacement
3 boundary conditions
1,∗ 2,3 4 1
4 T. Wu , A. Carpiuc-Prisacari , M. Poncelet , L. De Lorenzis

1
5 Institute of Applied Mechanics, Technische Universität Braunschweig, Germany
2
6 Institut des Sciences de la Mécanique et Applications Industrielles, UMR
7 EDF-CNRS-CEA-ENSTA 9219 Université Paris-Saclay, 828 Boulevard des Maréchaux, 91762
8 Palaiseau Cedex, France
3
9 EDF Lab Paris-Saclay, 7, boulevard Gaspard Monge, 91120 Palaiseau, France
4
10 LMT-Cachan/ENS-Cachan/CNRS/Université Paris Saclay, 61 avenue du Président Wilson,
11 94235 Cachan Cedex, France

12

13 Abstract

14 Phase-field modeling is an elegant approach to simulate complicated fracture processes, in-

15 cluding crack initiation, propagation, merging and branching in a unified framework without
16 the need for ad-hoc criteria and on a fixed mesh. These capabilities can only be fully validated
17 through the comparison with experiments featuring crack development histories and patterns
18 of sufficient complexity. As opposed to conventional mixed-mode fracture tests with predefined
19 loading, interactive tests with multiaxial loading which are controlled during the propagation
20 of the cracks can create more complex and stable crack propagation patterns. Moreover, the
21 development of measurement techniques such as digital image correlation (DIC) provides the
22 possibility to quantitatively characterize the full-field kinematics during the tests. In this
23 work, full-field displacements measured by DIC during interactive mixed-mode fracture tests
24 on cement mortar specimens are adopted as boundary conditions for phase-field numerical
25 simulations. Qualitative and quantitative comparisons are illustrated, demonstrating the ca-
26 pability of the phase-field approach to predict complex mixed-mode fracture phenomena in
27 cement mortar and suggesting possible further developments.

28 Keywords: Cement mortar, Digital Image Correlation, Mixed-mode fracture, Phase-field

29 modeling, Full-field displacements

30

∗ Corresponding author. Tel: +49 531.391-94360. E-mail: t.wu@tu-braunschweig.de

 Wu et al.

31 1 Introduction
32 The prediction of fracture phenomena in concrete, which is of interest for several technical appli-
33 cations, still represents a challenge for the computational modeler. Within multiscale approaches,
34 concrete can be regarded as consisting of cement mortar, including the cement matrix and the
35 smallest aggregates, and the largest aggregates. In the majority of cases, cracking occurs in the
36 cement mortar and is a complex process featuring local mixed-mode conditions.
37 Among the best known experimental setups for the investigation of mixed-mode fracture of
38 concrete are the three-point bending tests with asymmetric notch location [1, 2, 3], the tests on
39 hollow cylindrical specimens adopted by Keuser and Walraven [4], the tests on double-notched
40 specimens by Nooru-Muhamed [5], the torsion fracture test developed by Brokenshire [6], and
41 the tests on L-shape specimens proposed by Winkler [7]. Each of these tests has advantages and
42 disadvantages, see [8].
43 New experimental techniques have recently brought the possibility of quantitatively investi-
44 gating mixed-mode fracture in cementitious materials (more general in brittle and quasi-brittle
45 materials) to a new level. In this paper, we will be focusing on the tests in [8]. Two crucial devel-
46 opments were exploited in this experimental campaign: (i) the use of a multiaxial electromechanical
47 testing machine, i.e. of a hexapod [9],[10] (6 degrees of freedom) with an optical 3D displacement
48 control loop [11]. This testing machine can apply a complete twist (3 translations+3 rotations),
49 wrench (3 moments+3 forces) or a combination of both to the boundary of the specimen; (ii) the
50 digital image correlation (DIC) technique [12], through which the full displacement field on each
51 face of the sample becomes available. This gives access to the actual boundary conditions, which
52 are essential for a more realistic computational modeling, and provides an enhanced capability of
53 quantitative comparison between numerical and test results. Last but not least, DIC computations
54 can be performed during the tests and lead to the interactive adjustment of the applied boundary
55 conditions. Testing with this combination of techniques can generate more complex and stable
56 crack patterns.
57 In the past few years, phase-field modeling of fracture, stemming from Francfort and Marigo’s
58 variational approach [13], has emerged as a very elegant and powerful framework to predict cracking
59 phenomena. Fracture is described by using a continuous field variable (order parameter), which
60 provides a smooth transition between intact and fully broken material phases, such that it avoids
61 modeling cracks as discontinuities, see e.g. [14, 15, 16, 17], the review in [18] and the references
62 therein. The phase-field framework leads to the ability to simulate complicated fracture processes,
63 including crack initiation, propagation, merging and branching in a unified framework without the
64 need for ad-hoc criteria and on a fixed mesh. The approach shares several features with gradient
65 damage models, yet they were established from different backgrounds [19].
66 Phase-field modeling of fracture has been the subject of numerous studies, encompassing brittle
67 fracture [20, 21, 14, 15, 16, 17, 18, 22], ductile fracture [23, 24, 25, 26, 27], fracture in shells
68 [28, 29, 30] and multifield problems [31, 32, 33, 34, 35, 36, 37]. Here we focus on brittle fracture
69 and are especially concerned with cementitious materials, which are known to be brittle or quasi-

1
 Wu et al.

70 brittle. In brittle materials, the force resisted across a crack is zero, whereas quasi-brittle materials
71 feature a softening behavior with a cohesive response. For both materials, no apparent plastic
72 deformation occurs before fracture. A series of computations for validation of the phase-field
73 approach to brittle fracture was conducted in [22], where numerical results were compared to results
74 of a few classical experimental tests, including the already cited Nooru-Mohamed, Brokenshire and
75 Winkler tests [5, 6, 7]. The tests were performed with the setup and instrumentation available
76 at the time, and only a limited set of experimental data is reported, so that the comparison in
77 [22] is almost exclusively of qualitative nature, i.e. it only focuses on the crack path. For ductile
78 fracture, a quantitative comparison between phase-field fracture computations and relatively simple
79 experimental results was conducted in [38]. To the best of our knowledge, no quantitative validation
80 of phase-field brittle fracture computations is currently available.
81 The objective of this work is to carry out qualitative and quantitative comparisons between
82 numerical simulations with the phase-field approach to brittle fracture and the complex interactive
83 mixed-mode fracture tests on cement mortar specimens conducted in [8]. As DIC provides the
84 full displacement field on the surface of the specimens, the measured fields at the boundaries of
85 the analyzed domain are employed as displacement boundary conditions for the numerical simula-
86 tions, whereas the displacement fields measured within the domain are used for the experimental-
87 numerical comparison. Forces and crack patterns are also compared. This paper is arranged as
88 follows. Section 2 introduces the fundamentals of phase-field modeling of brittle fracture follow-
89 ing the formulation in [14]. The material properties and especially the numerical calibration of
90 the fracture energy are discussed in Section 3. In Section 4, three mixed-mode fracture tests are
91 illustrated, and detailed comparisons between numerical and experimental results are presented.
92 Section 5 closes the paper with final remarks.

93 2 Phase-field modeling of brittle fracture

94 The phase-field approach used in this paper originates from the variational formulation of brittle
95 fracture by Francfort and Marigo [13]. For brittle materials, the fracture problem is formulated as
96 the minimization problem of the energy functional
Z Z
E(ε, Γ) = ψel,0 (ε)dΩ + Gc dΓ , (2.1)
Ω Γ

97 where Ω ⊂ Rn , n = 2, 3, is an open bounded domain, representing an elastic n-dimensional body,

98 Γ ⊂ Ω is the crack set, ψel,0 (ε) is the elastic strain energy density, Gc is the fracture energy, and
ε is the infinitesimal strain tensor, ε = 12 ∇u + ∇uT with u as the displacement field. With the


99

100 regularization treatment by Bourdin et al. [20], Equation (2.1) can be recast as

1
Z Z h i
El (ε, s) = ψel (ε, s)dΩ + Gc (1 − s)2 + ℓ|∇s|2 dΩ , (2.2)
Ω Ω 4ℓ

2
 Wu et al.

101 where s is the phase-field parameter which describes the state of the material and varies smoothly
102 between 1 (intact) and 0 (completely broken), 0 ≤ s ≤ 1. Here ℓ is a length-scale parameter
103 characterizing the width of the diffusive approximation of a discrete crack, i.e. the width of the
104 transition zone between completely broken and intact phases, which would be zero for a discrete
105 crack. The elastic strain energy density affected by damage, ψel , is given by

+ −
ψel (ε, s) = g(s)ψel (ε) + ψel (ε) , (2.3)

106 where the split into positive and negative parts is used to differentiate the fracture behavior in
107 tension and compression, as well as to prevent the interpenetration of the crack faces under com-
108 pression. Therefore, normal contact between the crack faces is automatically dealt with, however
109 frictional forces as a result of tangential relative sliding of the crack faces are excluded. In this
110 work, we adopt the volumetric-deviatoric split by Amor et al. [14]

+ 1
ψel := Khtr(ε)i2+ + G(εdev : εdev )
2 (2.4)
1
−
ψel := Khtr(ε)i2− ,
2
2G
111 where K = λ + 3 , hai± := 12 (a ± |a|), εdev := ε − 31 tr(ε)I, λ and G are the Lamé constants, and
112 I is the second-order identity tensor.
113 The degradation function g(s) is here given by

g(s) := s2 + η , (2.5)

114 where η is a very small dimensionless parameter used to prevent numerical difficulties in fully
115 cracked conditions.
116 The governing system of equations in strong form is obtained from the stationarity conditions
117 of functional (2.2). It is given by the balance equation

divσ(ε, s) = 0 , (2.6)

118 where the stress tensor σ(ε, s) is given by

+ −
∂ψel (ε) ∂ψel (ε)
σ(ε, s) = g(s) + , (2.7)
∂ε ∂ε

119 and the phase-field evolution equation

1−s g ′ (s) +
2ℓ∆s + = ψ , (2.8)
2ℓ Gc el

120 The weak form of the coupled system of Equations (2.6) and (2.8) can be solved monolithically or
121 with a staggered scheme [39]. Interpreting s as a damage variable, additionally an irreversibility

3
 Wu et al.

122 condition on s is needed to guarantee that damaged points do not “heal”. Within a staggered
123 solution scheme, Miehe et al. [40] proposed to introduce for this purpose the history variable
+
124 H+ (x) := maxτ ∈[0,t] ψel (ε(x, τ )) (x being the spatial coordinate and t the time) ensuring the
125 irreversibility of the evolution of the crack phase-field [40]. Hence, they rewrite (2.8) as follows:

1−s g ′ (s) +
2ℓ∆s + = H , (2.9)
2ℓ Gc

126 The boundary conditions are given by


 u = ū on ∂Ωū


σ · n = t̄ on ∂Ωt̄ , (2.10)

∇s · n = 0 on ∂Ω


127 where ū and t̄ are prescribed displacement and traction on the Dirichlet and Neumann portions
128 of the boundary ∂Ωū and ∂Ωt̄ , respectively, with ∂Ω = ∂Ωū ∪ ∂Ωt̄ and ∂Ωū ∩ ∂Ωt̄ = ∅ and n is
129 the outward normal unit vector to the boundary.
130 The governing equations (2.6) and (2.9), with the corresponding boundary conditions, consti-
131 tute a system of coupled non-linear equations in the unknown fields u and s. Their weak form is
132 appropriately discretized with the finite element method using linear finite elements for both dis-
133 placement and phase-field. Algorithmically, we adopt a staggered approach, solving alternatively
134 for the displacement and the phase-field [21]. For a detailed discussion of staggered and monolithic
135 schemes, see [39].
136 In a one-dimensional setting, the governing equations can be solved for a homogeneous stress
137 state deriving the tensile strength of the material (critical stress) and the corresponding critical
138 strain as [17]
r r
9 EGc Gc
σc = and εc = , (2.11)
16 6ℓ 6ℓE
139 where E is the elastic modulus (related to λ and G by well known relations). Thus, it is clearly
140 seen that the tensile strength (for a given E) is a function of ℓ and Gc , and increases as ℓ decreases
141 or Gc increases. This suggests the possibility to determine ℓ directly from the knowledge of the
142 tensile strength and the fracture energy of a given material. When the length parameter ℓ goes
143 to zero, the critical stress goes to infinity. This behavior of the phase-field model thus correctly
144 replicates linear elastic fracture mechanics, which is also not capable of nucleating fractures in the
145 absence of singularities.

146 3 Material properties and test setup

147 The material used in the tests is a cement mortar prepared using the mix in Table 1. The speci-
148 mens were casted vertically, i.e. the upper surface corresponds to the free surface during casting.
149 A tomographic analysis revealed no heterogeneity of the sand density in the region of crack prop-

4
 Wu et al.

Effective water [kg/m3 ] Cement [kg/m3 ] Sand 0/4 [kg/m3 ] Plasticizer [kg/m3 ]

319 611 1235 5.25

Table 1: Cement mortar mix.

150 agation, see [8] for more details. Due to the small aggregate size (< 4 mm) with no appreciable
151 aggregate interlocking, the behavior is expected to be quite brittle.
152 As follows, the material properties as determined through direct testing (tensile strength and
153 elastic modulus) or through a combination of testing and numerical modeling (fracture energy).
154 Subsequently, the setup used for the mixed-mode fracture tests is described.

155 3.1 Material properties

156 In order to experimentally determine the tensile strength and the Young’s modulus of the cement
157 mortar, a series of five three-point bending tests on 40 × 40 × 160 mm3 specimens were performed
158 in [8], obtaining the average values in Table 2. The Poisson’s ratio was given the value 0.2 which
159 is considered reasonable for cement mortar.
160 The fracture energy Gc is usually measured through a three-point bending test. Based on results
161 of this test, Gc can be approximately computed analytically from the area under the force-deflection
162 curve, the area of the ligament and the weight of the specimen. However, this computation relies
163 on highly simplifying assumptions. Hence, here we calibrate Gc by simulating the three-point
164 bending tests numerically and comparing numerical and test results.
165 Six three-point bending tests were carried out, see [8] for details. The dimensions of the
166 specimen are 70 × 70 × 284 mm3 with a central notch of 5 × 35 × 70 mm3 (Figure 1). The test is
167 performed in displacement control, and both the force and the mid-span deflection of the beam on
168 the neutral axis are recorded until failure. Figure 2 shows the experimental force-deflection curves.
u

5
70
35

210
284

Figure 1: Geometry and setup of the three-point bending test (dimensions in mm).

169 Gc is calibrated by minimizing the difference between the computed and experimental force-

5
 Wu et al.

170 deflection curves. Note that during calibration, for each choice of Gc , the length parameter ℓ is
171 recalculated using Equation (2.11). Since ℓ needs to be resolved by the finite element discretization,
172 it must be ensured that the mesh size is suitable to resolve the smallest value of ℓ. The fracture
173 energy obtained from this calibration is Gc = 0.01 N/mm, a reasonable value for cementitious
174 materials (see, e.g. [41]), for which the numerical force-deflection curve is shown in Figure 2. The
175 corresponding length parameter obtained from Equation (2.11) is 0.6241 mm. Figure 3 illustrates
176 the crack propagation pattern, where the crack is initiated from the notch and propagates vertically.

Property Values

E 18 (GP a) [8]
ν 0.2 (-)
σc 3.9 (M P a) [8]
Gc 0.01 (N/mm)
ℓ 0.6241 (mm)

Table 2: Mechanical parameters used in the numerical simulations.

2000
Exp1
Exp2
Exp3
1500 Exp4
Exp5
Exp6
Num
Force (N )

1000

500

0
0 0.02 0.04 0.06 0.08 0.1 0.12
Deflection (mm)

Figure 2: Experimental results and numerical curve for Gc = 0.01 N/mm.

177 3.2 Test setup

178 This section illustrates the setup of the three mixed-mode fracture tests to be analyzed in the
179 following, denoted as test I, II and III in order of increasing loading complexity. Test I includes
180 crack initiation, propagation and reorientation. In addition to these, test II also involves crack
181 closure, friction and merging. The aim of test III is to additionally obtain crack branching. These
182 three tests are carried out in a multiaxial testing machine with 6 degrees of freedom, as briefly
183 described in the introduction. The tests are interactive in the sense that the loading is adjusted

6
 Wu et al.

Figure 3: Crack initiation and propagation for the three-point bending test. The crack is characterized
by the iso-surface of the phase-field parameter with s 6 0.01.

184 during the crack propagation, in order to increase the complexity of the crack patterns over those
185 obtainable with standard tests with predefined loadings. Also, the use of in-plane rotations is able
186 to prolong the stable crack propagation period by generating tension/compression stress gradients.
187 For each test, only one crack is controlled at each time, despite the presence of two notches in
188 two of the specimens. The interested reader can refer to [11, 42, 8, 43, 44] for all details on the
189 experimental setup.
190 Three mortar specimens with the same dimensions of 50 × 200 × 200 mm3 but with different
191 sizes and numbers of notches are employed for the three tests, see Figure 4. In order to reduce
192 the noise in the boundary conditions from the experimental measurements, two rigid layers are
193 embedded in the upper and lower parts of the specimen in all three tests, see Figure 4. The value
194 of 10 mm was chosen based on some preliminary numerical tests. The material parameters were
195 the subject of Section 3.1 and are summarized in Table 2. The sub-volume between the two dashed
196 lines is the domain modeled numerically, a choice which is discussed in the following.
197 The full displacement field can be measured on the specimen surfaces every 5 seconds through
198 a regularized three-noded triangles (RT3) DIC algorithm, regularizing the sought displacement
199 field by the elastic solution (a mechanical filter), which allows the mechanical admissibility to be
200 enforced in a weak sense [45]. The DIC measurements can be directly adopted as displacement
201 boundary conditions for the numerical simulations. Due to the availability of the full field, the
202 decision can be taken whether the simulations should be carried out on the whole specimen or
203 only on a sub-volume. Both choices have advantages and drawbacks. Performing the simulation
204 on the entire specimen, the boundary conditions are far from the crack propagation zone, which
205 is an advantage. However, the fact that contact between the sample and the loading plates is not

7
 Wu et al.

0 0 0
25
10 10
100
75
200 150 5 5 5 5 5
25 25 40 40 25
75 100
10 10
Z
25
X• 200 200 200
Y (a) (b) (c)

Figure 4: Specimen geometry of (a) test I (b) test II (c) test III (grey colour indicates rigid layers). The
sub-volume between the two dashed lines is modeled numerically (dimensions in mm).

206 explicitly modeled could lead to loss of accuracy in the simulations. If simulations are run on a
207 sub-volume of the sample, the boundary conditions are closer to the crack propagation zone, so
208 that a small variation of the boundary conditions may greatly affect the local displacement field
209 and crack pattern. In the present work, we seek a compromise and perform the simulation on a
210 relatively large sub-volume of the sample, see Figure 4, such that the effect of the loading plates
211 can be alleviated, and at the same time the boundary conditions are sufficiently far from the crack
212 propagation zone.
213 In the interactive tests, the combination of three elementary loading types, e.g. tension, shear
214 and rotation, leads to the initiation, propagation, reorientation and branching of the crack. In
215 particular, in-plane rotations can increase the period of stable crack propagation. The rotations
216 are prescribed with respect to point 0 in the specimen, which is labeled in Figure 4.
217 The DIC provides the measurement of the full displacement field u (with components uX , uY ,
218 uZ ), see Figure 6(a). In particular, for the edges of the specimen corresponding to the dashed
219 lines in Figure 4 (shown in Figure 6(b)), Figure 6(c) illustrates examples of the distribution of the
220 displacement components uY and uZ at the final loading stage in test II. The displacement distri-
221 butions at these edges are used as displacement boundary conditions for the numerical simulations.
222 uX is not adopted in the numerical simulations, but only one node on the edges (the dashed lines
223 in Figure 4) is fixed in the X-direction in order to avoid rigid body motions. Since displacement
224 boundary conditions are needed on the whole top and bottom surfaces of the numerical specimen
225 and not only on their edges, a linear interpolation between the displacements at the two edges is
226 employed. In addition, the experiments provide displacements of discrete points along the edges
227 among which a further linear interpolation is conducted, hence, the number of elements also plays
228 a role on the quality of the imposed boundary conditions in the simulations. For an in-depth study
229 of the influence of the boundary conditions choice on the simulations, see [44].

8
 Wu et al.

uZ
uY R

Figure 5: Three elementary loading types (a) tension (b) shear (c) rotation with respect to point 0.

Face 1 Face 2

(a) (b) Z

(c)1 (c)2

(c)3 (c)4

Figure 6: (a) Displacement field measured by DIC, (b) edges on two faces of the specimen, (c)1 uY at the
bottom edge of face 1, (c)2 uZ at the bottom edge of face 1, (c)3 uY at the top edge of face 2 and (c)4 uZ
at the top edge of face 2.

230 4 Comparison between experimental and numerical results

231 In the following, we compare the experimental results with the numerical simulation results in
232 terms of displacement fields on the two side surfaces of the specimen, crack patterns and resultant

9
 Wu et al.

Z
X
Y

Figure 7: Test I: discretization of the specimen with tetrahedral elements.

233 normal and shear forces on the top surface of the specimen. Phase-field modeling is employed to
234 simulate three-dimensional crack propagation. In order to resolve the length parameter ℓ, the mesh
235 is refined in the central horizontal region of the specimen, where the crack is expected to propagate,
236 see Figure 7. In addition, the refinement of the mesh on the top and the bottom surfaces of the
237 specimen increases the resolution of the imposed boundary conditions. In some cases, oscillations
238 are observed in the numerical results due to oscillations in the boundary conditions coming from
239 the displacement measurements elaborated through DIC.

240 4.1 Test I

241 The objective of test I is to create successive crack initiation, propagation and reorientation. The
242 numerical specimen is discretized with 393961 four-node tetrahedral elements. Figure 8 illustrates
243 the imposed displacements and rotations on the specimen. Here the displacements are the mean
244 values of the displacement fields imposed on the edges of the top and bottom surfaces and the
245 rotations are prescribed with respect to point O (Figure 4). The displacement along X (uX )
246 and the rotations RY and RZ are quite small. The loading history of test I can be divided into
247 four steps, which are labeled with numbers from 0 to 3, see Figure 8. The comparison between
248 experimental and numerical results for the forces is shown in Figure 9, where the force is the
249 resultant on the top surface. In general, the numerical curves are quite jagged. This can be mainly
250 ascribed to the noise in the boundary conditions, coming from the experimental measurements.
251 In Figures 10 to 13 we compare the calculated fields of uY and uZ on the two side surfaces of
252 the specimen with those measured by the DIC at the instants of time highlighted in Figure 9.
253 The experimental displacement fields also contain information on the crack patterns. Figure 14
254 illustrates the contours of the phase-field on the two side faces of the specimen at different loading

10
 Wu et al.

255 steps, whereas Figure 15 shows the experimental crack pattern at failure. Two unstable crack
256 propagation phases are encountered in this test, the first one at the end of the step 1 and the
257 second one at the end. As follows, we compare experimental and numerical results for each step.
258 Step 0 : The test starts with an elastic loading-unloading step. The forces FY and FZ change
259 linearly, see Figure 9. Apart from checking the proper functioning of the devices, the main objective
260 of this step is to achieve the ’self-positioning’ of the specimen. As presented in Figure 9, the
261 calculated normal force FZ and shear force FY are in good agreement with the experimental data.
262 Step 1 : As shown in Figure 8, a displacement along Y together with a rotation along X are
263 imposed on the specimen. The shear force FY and the normal force FZ are greatly increased
264 (in absolute value). The calculated FY is in very good agreement with the experimental data,
265 see Figure 9. The first unstable cracking phenomenon occurs at t ≈ 6375 s, such that a sudden
266 change of FY and FZ is observed experimentally. Since the quasi-static phase-field model is not
267 able to accurately predict unstable crack propagation, this sudden change, especially in FY , is not
268 accurately reproduced. We define here stable propagation as the crack propagation which occurs
269 quasi-statically, i.e. with a very small increment of the crack length with time. We consider crack
270 propagation unstable when the crack advancement is dynamic, almost instantaneous. Usually
271 when unstable crack propagation occurs, the crack passes through the entire sample resulting in
272 its complete failure. In the present test, the first propagation of the crack was stable as it can be
273 seen in Figures 10-13 (a) (crack state before the unstable propagation). Unstable crack propagation
274 occurs between Figures 10-13 (a) and (b).
275 The displacement fields on the two side surfaces of the specimen right before the first unstable
276 crack propagation are shown in Figures 10-13 (a). They indicate that a crack initiates from
277 the notch and propagates toward the upper part of the specimen. A good agreement between
278 experimental and numerical results is observed. The unstable propagation at t ≈ 6375 s can be
279 observed by analyzing the two consecutive images in Figures 10-13 (a) and (b). Since the left side
280 of the sample is subjected to a compressive stress, the unstable crack propagation stops once the
281 crack gets close to the compression zone, see Figures 10-13 (b). As visible in Figures 14(a) and (b),
282 the phase-field contour resembles closely the experimental crack pattern on both side surfaces of
283 the specimen.
284 Step 2 : As illustrated in Figure 8, uY is increased (in absolute value) and the rotation is
285 kept constant. The shear force FY gradually goes from positive to negative for preparing the
286 reorientation of the crack while the normal force FZ remains negative, see Figure 9. A good
287 agreement between calculated and experimental data is observed for FZ . Instead, starting from
288 step 2, there is an offset between calculated and experimental FY , the former being larger than the
289 latter. This offset originates from the unstable crack propagation at the end of step 1 and is not
290 recovered afterwards. At the end of step 2, the calculated FY is ≈ −3 kN , while the experimental
291 FY is ≈ −5.5kN .
292 Step 3 : A displacement along Z is prescribed to lead the specimen to failure while the dis-
293 placement along Y and the rotation along X remain constant, see Figure 8. The second unstable

11
 Wu et al.

294 cracking phenomenon occurs at the end of the test. As illustrated in Figure 9, once again the
295 calculated and experimental FZ are in reasonably good agreement, whereas the offset between
296 computed and experimental FY already noted in step 2 remains approximately constant until the
297 end of the test. The displacement contours in Figures 10-13 (c) and (d) indicate a reorientation of
298 the first crack. Moreover, it is also seen that a second crack initiates from the left notch towards
299 the end of the test and then stably propagates upward. The first and second cracks approximately
300 merge in an unstable manner at the end of the test, see Figures 10-13 (d). The initiation of the
301 second crack is significantly influenced by local effects which are unavoidable due to the intrinsic
302 heterogeneity of the material. Moreover, the initiation and propagation period of the second crack
303 is quite short. This leads to the difficulty in not only replicating it by numerical simulations but
304 also experimentally reproducing it on different specimens. The calculated uY and uZ are overall
305 consistent with the experimental data. However, an offset between first and second cracks at the
306 left notch is only noticed in the experimental contours. A more pronounced reorientation of the
307 crack close to the left notch (corresponding to the second crack) is shown by the phase-field contour
308 in Figure 14, where however the final crack pattern once again ends up at the level of the left notch,
309 and not higher as in the experiment, see Figure 15.
310 We conclude this section with two general remarks, valid for each test in this work: 1. The
311 longer the test, the more difficult is the exact simulation of the behavior, due to the dependency
312 of the current simulation results on the preceding cracking history. Therefore, small errors which
313 are cumulated during the crack propagation process could result in rather different final crack
314 patterns in the simulations as opposed to the real experiment. 2. The thickness of the diffusive
315 approximation of the crack shows a fictitious increase in the regions where the crack passes from a
316 tensile to a compressive state, see e.g. Figure 14 (c) and (d). This is due to an inherent drawback
317 of the tension-compression split in Equation (2.4), and is still an open issue.

500 7
0 1 2 3 0 1 2 3
6
400 uX
uY 5
300 uZ
R(× 10−4 rad)

4
u(µm)

RX
200 3 RY
RZ
2
100
1
0
0
-100 -1
0 0.5 1 1.5 2 2.5 2.75 0 0.5 1 1.5 2 2.5 2.75
Time (× 104 s) Time (× 104 s)

Figure 8: Test I: (left) displacements and (right) rotations with respect to point O.

12
 Wu et al.

18 8
0 1 2 3 0 1 2 3
14 6
Exp
10 4 Num
2
Force (kN )

Force (kN )
5
0
0
-2
-5 Exp
-4
Num
-10 (a) (c)(d) -6 (a) (b) (c)(d)
(b)
-14 -8
0 0.5 1 1.5 2 2.5 2.75 0 0.5 1 1.5 2 2.5 2.75
Time (× 104 s) Time (× 104 s)

Figure 9: Test I: comparison between experimental and numerical results for (left) normal force FZ and
(right) shear force FY .

318 4.2 Test II

319 Test II is able to display the phenomena of crack initiation, propagation, reorientation, closure and
320 merging. The numerical specimen is discretized with 430636 four-node tetrahedral elements.
321 The prescribed displacements and rotations are shown in Figure 16. Once again the loading
322 history is subdivided in four steps which are labeled with numbers from 0 to 3. Figure 17 shows the
323 comparison between experimental and numerical results for the forces FY and FZ . The comparison
324 for the fields uY and uZ on the two side surfaces of the specimen at the instants of time shown in
325 Figure 17 is given in Figures 18-21. The contour plots of the phase-field at different loading steps
326 are shown in Figure 22. Figure 23 is a picture of the experimental final crack pattern.

327 Step 0 : The test is started with elastic loading-unloading and without crack initiation. The
328 calculated FZ and FY are very close to the experimental data, see Figure 17.
329 Step 1 : As shown in Figure 16, a positive rotation along X combined with a negative displace-
330 ment along Y (and with small displacements in the other two directions) are imposed to initiate a
331 crack at the right notch. From Figure 17 it is seen that the shear force FY increases almost linearly
332 with the applied displacements. Note that FY is positive, whereas uY is negative. This indicates
333 that FY is induced primarily by the rotations, rather than by the applied displacements. As soon
334 as the crack starts to propagate upward (t ≈ 5500 s), FZ is decreased due to the reduction of the
335 area in the specimen which can transmit the tensile force. The calculated FZ and FY are in good
336 agreement with the experimental data.
337 Figures 18-21(a) present the experimental and calculated contours of uY and uZ on the two
338 side surfaces of the specimen at the beginning of step 1. uY has a larger negative value on face 2
339 than on face 1. In terms of uZ , the region above the right notch (above the crack) shows positive
340 values indicating vertical tension whereas the rest of the specimen has negative values connected
341 to vertical compression. The experimental contours of uY and uZ are successfully reproduced by
342 the numerical simulations.
343 Figure 22(a) illustrates the calculated phase-field contours on the two side surfaces of the

13
 Wu et al.

(a)exp (a)num

(b)exp (b)num

(c)exp (c)num

(d)exp (d)num

Figure 10: Test I: contours of uY on face 1 of the specimen. On the left are the experimental data and on
the right are the numerical results. (a) Last image before the first unstable crack propagation (t ≈ 6375s),
(b) first image after the first unstable crack propagation (5 s between the two), (c) image during step 3 and
(d) image at the end of step 3. The exact points (a)-(d) are shown in Figure 9. The specimen geometry in
the simulation results is enlarged by a scale factor of 1.5 for better visualization.

344 specimen at the end of step 1. It is seen that the crack initiates from the right notch and propagates
345 upward, which is consistent with the displacement contours as well as with the experimental crack

14
 Wu et al.

(a)exp (a)num

(b)exp (b)num

(c)exp (c)num

(d)exp (d)num

Figure 11: Test I: contours of uZ on face 1 of the specimen. On the left are the experimental data and on
the right are the numerical results. (a) Last image before the first unstable crack propagation (t ≈ 6375s),
(b) first image after the first unstable crack propagation (5 s between the two), (c) image during step 3 and
(d) image at the end of step 3. The exact points (a)-(d) are shown in Figure 9. The specimen geometry in
the simulation results is enlarged by a scale factor of 1.5 for better visualization.

15
 Wu et al.

(a)exp (a)num

(b)exp (b)num

(c)exp
(c)num

(d)exp (d)num

Figure 12: Test I: contours of uY on face 2 of the specimen (mirrored). On the left are the experimental
data and on the right are the numerical results. (a) Last image before the first unstable crack propagation
(t ≈ 6375s), (b) first image after the first unstable crack propagation (5 s between the two), (c) image
during step 3 and (d) image at the end of step 3. The exact points (a)-(d) are shown in Figure 9. The
specimen geometry in the simulation results is enlarged by a scale factor of 1.5 for better visualization.

16
 Wu et al.

(a)exp (a)num

(b)exp (b)num

(c)exp (c)num

(d)exp (d)num

Figure 13: Test I: contours of uZ on face 2 of the specimen (mirrored). On the left are the experimental
data and on the right are the numerical results. (a) Last image before the first unstable crack propagation
(t ≈ 6375s), (b) first image after the first unstable crack propagation (5 s between the two), (c) image
during step 3 and (d) image at the end of step 3. The exact points (a)-(d) are shown in Figure 9. The
specimen geometry in the simulation results is enlarged by a scale factor of 1.5 for better visualization.

17
 Wu et al.

(a)f 1 (a)f 2

(b)f 1 (b)f 2

(c)f 1 (c)f 2

(d)f 1 (d)f 2

Figure 14: Test I: contour plot of the phase-field on the two side surfaces of the specimen. On the left are
results on face 1 and on the right are results on face 2 (mirrored). (a) Last image before the first unstable
crack propagation (t ≈ 6375 ), (b) first image after the first unstable crack propagation (5 s between the
two), (c) image during step 3 and (d) image at the end of step 3.

346 pattern shown in Figure 23.

347 At the end of step 1 the specimen is unloaded. Upon unloading, the experimental value FY
348 ≈ −1 kN indicates that the two faces may not be exactly superimposed during the crack closure.

18
 Wu et al.

Figure 15: Test I: final crack pattern on the two side surfaces of the specimen in the experiment (a) face
1 and (b) face 2 (mirrored).

349 This is possibly due to crack roughness effects which are not accounted for by the numerical model.
350 The calculated FY is close to 0 kN .
351 Step 2 : After unloading at the end of step 1, the specimen is reloaded by applying a negative
352 rotation along X combined with a positive displacement along Y (and smaller values of the other
353 displacements). This leads to the closure of the first crack and the initiation of a second crack
354 starting from the other (left) notch. The objective of applying displacement uY is to alleviate the
355 negative shear force induced by the rotation and to generate a positive shear load. The forces FY
356 and FZ have similar evolutions as in step 1. At t ≈ 13500 s, a second crack initiates at the left
357 notch and propagates downward. After a certain amount of propagation, the specimen is unloaded.
358 As the crack starts to propagate, FZ decreases both in the experiment and in the computations.
359 During step 2, a small difference between experimental and calculated FY is observed, which is
360 induced by the initial discrepancy at the end of step 1. At the end of unloading, similarly to what
361 observed at the end of step 1 after unloading, the experimental FY ≈ −3 kN indicates that the
362 two faces may not be exactly superimposed during the crack closure. The calculated FY is close
363 to 0 kN .
364 The comparison between experimental and numerical results for uY and uZ at the end of step
365 2 after unloading is shown in Figures 18-21 (b). It is observed that uZ in the right region of the
366 sample is negative, indicating that this region is subjected to a vertical compressive stress, such
367 that the first crack is closed. The calculated contours of uY and uZ match well the experimental
368 data.
369 The calculated phase-field contours on the side surfaces of the specimen at the end of step 2
370 after unloading are shown in Figure 22(b). The first crack, although closed, is still visible due
371 to the enforcement of the irreversibility condition, and another crack has initiated from the other
372 notch and propagated downward. This matches the displacement contours in Figures 18-21 (b).

19
 Wu et al.

373 Step 3 : Here a positive uZ and a negative uY are applied simultaneously. This makes the two
374 pre-existing cracks reorient and eventually merge. Merging of the two cracks occurs in an unstable
375 manner. Both the experimental and calculated FZ increase, yet following a very different trend.
376 The experimental and calculated FY , which start from different initial values at the end of step 2,
377 decrease also with different slopes. This difference can be explained by the fact that, as discussed
378 next, the final predicted crack pattern does not coincide with the experimental one.
379 Figures 18-21 (c) and (d) illustrate comparisons between experimental and calculated contours
380 of uY and uZ during and at the end of step 3, respectively. During step 3, before the final unstable
381 cracking, the numerical results of uY and uZ match the experimental data well, although the
382 experimental and numerical forces are already quite different at this stage. At the end of the step,
383 after the final unstable cracking, the experimental displacement contours indicate that the second
384 crack has partially closed, becoming significantly shorter, and a third “diagonal” crack has formed
385 joining the previous two. The formation of this “diagonal” crack is not visible in the numerical
386 displacement contours, which look similar throughout step 3. As the images in Figures 18-21
387 (d) have not fully converged, we leave the comparisons for uY and uZ between experimental and
388 numerical results out of this work.
389 Figures 22 (c) and (d) show the calculated phase-field contour on the two side surfaces of the
390 specimen during and at the end of step 3, respectively. As in the displacement contours, the two
391 cracks are seen to reorient, however, the third “diagonal” crack leading to their merge is not visible.
392 Instead, the final crack pattern in the experiment, see Figure 23, is consistent with the experimental
393 displacement contours. However, this consistency does not always hold, see Test III. The reasons
394 for the discrepancy between experimental and numerical results may be multiple. Certainly the
395 role of heterogeneities of the material is significant, as local defects leading to stress concentrations
396 can influence the evolution of the cracks in particular during unstable branches. Then, unstable
397 crack propagation is inaccurately reproduced by the numerical model as this ignores the possibly
398 significant dynamic effects. Finally, the material heterogeneities lead to roughness of the crack
399 surfaces which can induce interlocking and friction effects. These in turn may be responsible, as
400 already noted, for different crack closure behavior upon unloading, which may also influence the
401 subsequent evolution.

402 4.3 Test III

403 The goal of test III is to observe successive crack initiation, propagation, reorientation and branch-
404 ing. The numerical specimen is discretized with 305794 four-node tetrahedral elements.
405 The crack of test III is designed to follow a zigzag path, see Figure 24. A crack initiates from
406 the notch and propagates upward (branch (1) in Figure 24). Then it reorients downward (a) and
407 propagates (2). An upward reorientation (b) and a subsequent propagation (3) follow. The final
408 stages are a reorientation downward (c) and a last crack propagation (4). The crack path is kept
409 close to the central horizontal line of the specimen, in order to avoid the influence by imperfections
410 of the boundary conditions. Two crack propagation stages occur in the upward direction ((1) and

20
 Wu et al.

100 6
0 1 2 3 0 1 2 3
80
uX
4
RX
60 uY RY
uZ 2 RZ

R(× 10−4 rad)

40
u(µm)

0
20

0 -2

-20
-4
-40
-6
0 0.5 1 1.5 2 2.25 0 0.5 1 1.5 2 2.25
Time (× 104 s) Time (× 104 s)

Figure 16: Test II: (left) displacements and (right) rotations with respect to point O.

14 8
0 1 2 3 0 1 2 3
12 6

10 4
Exp
Num 2
8
Force (kN )

Force (kN )

0
6
-2
4
-4
Exp
2 -6 Num
0 (a) (b) (c)(d) -8 (a) (b) (d)
(c)
-2 -10
0 0.5 1 1.5 2 2.25 0 0.5 1 1.5 2 2.25
Time (× 104 s) Time (× 104 s)

Figure 17: Test II: comparison between experimental and numerical results for (left) normal force FZ and
(right) shear force FY .

411 (3)) and two in the downward direction ((2) and (4)).
412 Knowing the expected crack path in Figure 24, one can better understand the complex loading
413 history of this test in Figure 25, which is divided into 8 steps, labeled with numbers from 0 to
414 7. Generally speaking, steps 1, 3, 5, and 7 consist of applying tension/rotation, thus resulting in
415 crack propagation. Shear loading is imposed during steps 2, 4 and 6, in order to prepare the crack
416 reorientations. The comparison between experimental data and numerical results for the forces is
417 shown in Figure 26, whereas the displacement contours are given in Figures 27-30. The calculated
418 phase-field contours are illustrated in Figure 31 and the final crack pattern is shown in Figure 32.
419 Step 0 : The test is started with elastic loading-unloading without any crack initiation. The
420 calculated FZ and FY , as shown in Figure 26, are in good agreement with the experimental data.
421 Step 1 : As shown in Figure 25, a combination of displacements along Y and Z and a rotation
422 along X is imposed on the specimen to induce a gradient of tensile stress along Z and a shear
423 stress along Y , thus yielding crack initiation and propagation. At the beginning of step 1, the
424 shear force FY is approximately proportional to the applied displacement, where FZ remains close
425 to zero. At t ≈ 6000 s, the crack initiates and stably propagates upward ((1) in Figure 24). FZ
426 becomes more and more negative as the crack propagates. The calculated FZ and FY are in good

21
 Wu et al.

(a)exp (a)num

(b)exp (b)num

(c)exp (c)num

(d)exp (d)num

Figure 18: Test II: contours of uY on face 1 of the specimen. On the left are the experimental data and
on the right are the numerical results. (a) Image at the end of step 1, (b) image at the end of step 2, (c)
image during step 3 and (d) image at the end of step 3. The exact points (a)-(d) are shown in Figure 17.
The specimen geometry in the simulation results is enlarged by a scale factor of 1.5 for better visualization.

427 agreement with the experimental data.

428 Figures 27-30(a) give the experimental and calculated contours of uY and uZ at the end of step
429 1. The sign of uY is negative on both side surfaces of the specimen. The sign of uZ is positive in
430 the region above the notch, where the crack initiates, and negative elsewhere. The experimental

22
 Wu et al.

(a)exp (a)num

(b)exp (b)num

(c)exp (c)num

(d)exp (d)num

Figure 19: Test II: contours of uZ on face 1 of the specimen. On the left are the experimental data and
on the right are the numerical results. (a) Image at the end of step 1, (b) image at the end of step 2, (c)
image during step 3 and (d) image at the end of step 3. The exact points (a)-(d) are shown in Figure 17.
The specimen geometry in the simulation results is enlarged by a scale factor of 1.5 for better visualization.

431 contours of uY and uZ are well reproduced by the numerical model.

432 Figure 31(a) shows the calculated phase-field contours at the end of step 1. It is seen that
433 the crack initiates from the notch and then propagates upward, which is consistent with the
434 displacement contours in Figures 27-30(a).

23
 Wu et al.

(a)exp (a)num

(b)exp (b)num

(c)exp (c)num

(d)exp (d)num

Figure 20: Test II: contours of uY on face 2 of the specimen (mirrored). On the left are the experimental
data and on the right are the numerical results. (a) Image at the end of step 1, (b) image at the end of
step 2, (c) image during step 3 and (d) image at the end of step 3. The exact points (a)-(d) are shown in
Figure 17. The specimen geometry in the simulation results is enlarged by a scale factor of 1.5 for better
visualization.

435 Step 2 : The imposed loading history (Figure 25) prepares the first reorientation ((a) in Figure
436 24) by keeping RX constant and increasing uY (in absolute value) at a higher rate, thus inverting
437 the sign of FY . No visible crack propagation is observed in this step. The slope of FY is equal to
438 the one during step 0, indicating that the transmission of the load in Y direction through the crack
439 faces is unaltered. Also, it is seen that the calculated FZ and FY are in good agreement with the
440 experimental data.

24
 Wu et al.

(a)exp (a)num

(b)exp (b)num

(c)exp (c)num

(d)exp (d)num

Figure 21: Test II: contours of uZ on face 2 of the specimen (mirrored). On the left are the experimental
data and on the right are the numerical results. (a) Image at the end of step 1, (b) image at the end of
step 2, (c) image during step 3 and (d) image at the end of step 3. The exact points (a)-(d) are shown in
Figure 17. The specimen geometry in the simulation results is enlarged by a scale factor of 1.5 for better
visualization.

441 Step 3 : Linearly increasing rotations and displacements are imposed to make the crack prop-
442 agate downward ((2) in Figure 24). The forces FY and FZ remain approximately constant and are
443 consistent with the experimental data.

25
 Wu et al.

(a)f 1 (a)f 2

(b)f 1 (b)f 2

(c)f 1 (c)f 2

(d)f 1 (d)f 2

Figure 22: Test II: contour plot of the phase-field on the two side surfaces of the specimen. On the left
are results on face 1 and on the right are results on face 2 (mirrored). (a) Image at the end of step 1, (b)
image at the end of step 2, (c) image during step 3 and (d) image at the end of step 3.

444 As shown in Figures 27-30 (b), the experimental contours of uY and uZ at the end of step 3
445 match well the numerical results. In comparison to the end of step 1, uY is increased and the
446 region with positive uZ is enlarged. The displacement contours indicate that the crack propagates
447 downward. In addition, an initial reorientation ((b) in Figure 24) and propagation upward ((3)
448 in Figure 24) are observed. All these observations are consistent with the phase-field contour, see

26
 Wu et al.

(a) (b)

Figure 23: Test II: final crack pattern on the two side surfaces of the specimen in the experiment (a) face
1 and (b) face 2 (mirrored).

c a
4 3 1
2
b
2’

Figure 24: Test III: Sketch of the expected crack path.

449 Figure 31(b).

450 Step 4 : The prescribed loading history is meant to induce the reorientation at (b), similarly
451 to step 2. However, here the sign of FY is inverted from negative to positive. The slope of FY is
452 the same as during steps 0 and 2. During step 4, the calculated FY and FZ slightly overestimate
453 the experimental data.
454 Step 5 : Following the reorientation started in step 3 and promoted in step 4, now the crack
455 follows path (3) upward. The forces FY and FZ decrease at the beginning of this step and then
456 remain approximately constant in both the experimental and numerical curves.
457 As shown in Figures 27-30 (c), the calculated contours of uY and uZ at the end of this step
458 are in good agreement with the experimental data. The phase-field contour in Figure 31(c) is also

27
 Wu et al.

459 consistent with these results.

460 Step 6 : A further crack reorientation is induced by the imposed loading history, as in steps 2
461 and 4. FY is changed from positive to negative, similarly to step 2. The slope of FY is lower than
462 during steps 0, 2, and 4. This indicates that the crack faces no longer fully transmit the loading in
463 Y direction. Also, the reorientation induces a small decrease of FZ . A good agreement is observed
464 between numerical and experimental forces.
465 Figures 27-30 (d) provide the comparison between the numerical results and the experimental
466 data for uY and uZ . A good agreement is evident. Also, these results are consistent with the
467 phase-field contour in Figure 31 (d), which shows the incipient branching of the crack along path
468 (2’).
469 Step 7 : As only a positive displacement along Z is applied, FZ becomes positive. A branching
470 along path (2’) appears (Figure 24). FY approaches zero due to the shortening of the ligament
471 and the opening of the crack. At the end of this step, the crack propagation becomes unstable and
472 failure follows. A discrepancy between experimental and numerical results for the forces (especially
473 for FZ ) is observed due to this instability. In particular, the calculated FZ decays to zero when the
474 specimen is completely broken, which is different from the experimental value (FZ ≈ 8kN ) This
475 may be due to a residual cohesive behavior across the crack faces, which is not accounted for in
476 the numerical model. Instead FY attains zero in both the experimental and numerical curves.
477 As illustrated in Figures 27-30 (e), the experimental contours of uY and uZ are consistent with
478 the numerical results at the end of step 7. From the displacement contours, it is clearly seen that
479 the crack branches and propagates downward. This phenomenon is well reproduced by phase-field
480 modeling, see Figure 31(e). Here, spurious cracked regions close to the boundary of face 1 are
481 induced by oscillations of the DIC measurements.
482 Figure 32 shows the final crack pattern on both side surfaces of the specimen in the experi-
483 ment, where the branching is not visible, which is not completely consistent with the experimental
484 displacement contours. However, cross-sectional images obtained from computed tomography at
485 different X coordinates, shown in Figure 33, reveal that branching is indeed present in the interior
486 of the specimen, and its angle fits well the experimental displacement contours and the numerical
487 results (Figure 34).

488 5 Conclusions
489 This work provides a qualitative and quantitative comparison between mixed-mode fracture test
490 results in cement mortar and numerical simulations carried out with the phase-field approach. The
491 tests adopted for the comparison are conducted with a multiaxial testing machine and digital image
492 correlation (DIC) provides the full displacement field on each face of the sample. In addition, DIC
493 computations are performed during the tests and lead to the interactive adjustment of the applied
494 boundary conditions thorough a multiple input-multiple output integrated DIC algorithm. In
495 comparison to the conventional tests with predefined loadings, these interactive tests can generate

28
 Wu et al.

150 20
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
100
15 RX
50 RY
RZ

R(× 10−4 rad)

0 10
u(µm)

-50
uX 5
-100 uY
uZ
-150
0
-200
-250 -5
0 0.5 1 1.5 2 2.5 3 3.5 0 0.5 1 1.5 2 2.5 3 3.5
Time (× 104 s) Time (× 104 s)

Figure 25: Test III: (left) displacements and (right) rotations with respect to point O.

10 8
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
6
5
4
0
2
Force (kN )

Force (kN )

-5 0
-2 Exp
-10
Num
-4
-15 Exp (a) (b) (c) (d)(e) (a) (b) (c) (d)(e)
-6
Num

-20 -8
0 0.5 1 1.5 2 2.5 3 3.5 0 0.5 1 1.5 2 2.5 3 3.5
Time (× 104 s) Time (× 104 s)

Figure 26: Test III: comparison between experimental and numerical results for (left) normal force FZ
and (right) shear force FY .

496 more stable and complex crack patterns. Also, the use of the DIC gives access to the actual
497 boundary conditions, and thus provides an enhanced capability of quantitative comparison between
498 numerical and test results. Phase-field modeling is adopted to simulate the fracture phenomena.
499 Three mixed-mode fracture tests are carried out with the description of crack initiation, prop-
500 agation, reorientation, merging and branching. The experimental displacement fields on the upper
501 and lower surfaces of the specimen are adopted as boundary conditions for phase-field numerical
502 simulations. We compare the experimental and numerical results in terms of displacement contours
503 on the two side surfaces of the specimen, crack patterns and resultant normal and shear forces on
504 the top surface of the specimen.
505 In general, a remarkably good agreement between experimental and numerical results is found.
506 This demonstrates the capability of the phase-field approach to predict complex mixed-mode frac-
507 ture phenomena in cement mortar. The discrepancies observed in some cases can be attributed to
508 three main reasons: 1. the occurrence of unstable crack propagation phenomena, which cannot be
509 accurately captured by a quasi-static phase-field modeling approach; 2. the local effects generated
510 by heterogeneities in the material, which seem to play a role especially during unstable crack prop-
511 agation phases, and which could only be incorporated in the model through stochastic approaches;

29
 Wu et al.

512 3. the presence of residual cohesive and/or frictional effects across the crack faces, which are not
513 accounted for in the model. Finally, it should be mentioned that the adopted tension-compression
514 split leads to an increase in the thickness of the diffusive approximation of the crack once the crack
515 passes from a tensile to a compressive state. The incorporation of the aforementioned dynamic
516 and stochastic effects, as well as of cohesive and frictional behavior at the crack faces, and the
517 improvement of the tension-compression split, would lead to further progress in the predictive
518 capability of damage and cracking phenomena in brittle materials.

519 Acknowledgements
520 T. Wu and L. De Lorenzis would like to acknowledge funding provided by the European Re-
521 search Council under the European Union’s Seventh Framework Programme (FP7/2007-2013)/ERC
522 Grant agreement N. 279439 and by the German Project DFG GRK-2075. The experimental work
523 on which is based this numerical work has been supported and carried out within the EnerCampus
524 framework.

525 References
526 [1] Y. S. Jenq and S. P. Shah. Mixed-mode fracture of concrete. International Journal of Fracture,
527 38:123–142, 1988.

528 [2] S. E. Swartz, L. W. Lu, and L. D. Tang. Mixed-mode fracture toughness testing of concrete beams
529 in three-point bending. Materials and Structures, 21:33–40, 1988.

530 [3] J. C. Galvez, M. Elices, G. V. Guinea, and J. Planas. Mixed mode fracture of concrete under
531 proportional and nonproportional loading. International Journal of Fracture, 94:267–284, 1998.

532 [4] W. Keuser and J. Walraven. Fracture of plain concrete under mixed mode conditions. Fracture of
533 Concrete and Rock-Recent Developments, pages 625–634, 1989.

534 [5] M. B. Nooru-Mohamed. Mixed-mode fracture of concrete: An experimental approach. PhD thesis,
535 Delft University of Technology, Delft, The Netherlands, 1992.

536 [6] D. R. Brokenshire. Torsional fracture tests. PhD thesis, Cardiff University, Cardiff, UK, 1996.

537 [7] B. Winkler. Traglastuntersuchungen von unbewehrten und bewehrten Betonstrukturen auf der Grund-
538 lage eines Objektiven Werkstoffgesetze. PhD thesis, University of Innsbruck, Innsbruck, Austria, 2001.

539 [8] A. Carpiuc-Prisacari. Innovative tests for characterizing mixed-mode fracture of concrete: from pre-
540 defined to interactive and hybrid tests. PhD thesis, ENS Cachan, Cachan, France, 2015.

541 [9] V. E. Gough and S. G. Whitehall. Universal tyre test machine. pages 117–137. Proceedings of the
542 FISITA Ninth International Technical Congress, 1962.

543 [10] D. Stewart. A platform with six degrees of freedom. 180(15), pages 371–385. Proceedings of the
544 IMechE, 1965.

545 [11] J. Le Flohic, V. Parpoil, S. Bouissou, M. Poncelet, and H. Leclerc. A 3D displacement control by
546 Digital Image Correlation for the multiaxial testing of materials with a strewart platform. Experimental
547 Mechanics, 54:817–828, 2014.

30
 Wu et al.

548 [12] M. A. Sutton. Computer vision-based, noncontacting deformation measurements in mechanics: A

549 generational transformation. Applied Mechanics Reviews, 65(5):050802–(1–23), 2013.

550 [13] G. A. Francfort and J. J. Marigo. Revisiting brittle fractures as an energy minimization problem.
551 Journal of the Mechanics and Physics of Solids, 46:1319–1342, 1998.

552 [14] H. Amor, J. J. Marigo, and C. Maurini. Regularized formulation of the variational brittle fracture
553 with unilateral contact: Numerical experiments. Journal of the Mechanics and Physics of Solids,
554 57:1209–1229, 2009.

555 [15] C. Miehe, M. Hofacker, and F. Welschinger. A phase field model for rate-independent crack propa-
556 gation: Robust algorithmic implementation based on operator splits. Computer Methods in Applied
557 Mechanics and Engineering, 199:2765–2778, 2010.

558 [16] C. Kuhn and R. Müller. A phase field model for fracture. Engineering Fracture Mechanics, 77:3625–
559 3643, 2010.

560 [17] M. J. Borden, C. V. Verhoosel, M. A. Scott, T. J. R. Hughes, and C. M. Landis. A phase-field

561 description of dynamic brittle fracture. Computer Methods in Applied Mechanics and Engineering,
562 217-220:77–95, 2012.

563 [18] M. Ambati, T. Gerasimov, and L. De Lorenzis. A review on phase-field models of brittle fracture and
564 a new fast hybrid formulation. Computational Mechanics, 55:383–405, 2014.

565 [19] R. de Borst and C. V. Verhoosel. Gradient damage vs phase-field approaches for fracture: similarities
566 and differences. Computer Methods in Applied Mechanics and Engineering, 312:7894, 2016.

567 [20] B. Bourdin, G. A. Francfort, and J. J. Marigo. Numerical experiments in revisited brittle fracture.
568 Journal of the Mechanics and Physics of Solids, 48:797–826, 2000.

569 [21] B. Bourdin, G.A. Francfort, and J.J. Marigo. The variational approach to fracture. Journal of
570 Elasticity, 91:5–148, 2008.

571 [22] A. Mesgarnejad, B. Bourdin, and M.M. Khonsari. Validation simulations for the variational approach
572 to fracture. Computer Methods in Applied Mechanics and Engineering, 290:420–437, 2015.

573 [23] M. Ambati, T. Gerasimov, and L. De Lorenzis. Phase-field modeling of ductile fracture. Computational
574 Mechanics, 55:1017–1040, 2015.

575 [24] C. Kuhn, T. Noll, and R. Müller. On phase field modelling of ductile fracture. GAMM-Mitteilungen,
576 39:35–54, 2016.

577 [25] C. Miehe, S. Teichtmeister, and F. Aldakheel. Phase-field modelling of ductile fracture: a variational
578 gradient-extended plasticity-damage theory and its micromorphic regularization. Mathematical, Phys-
579 ical and Engineering Sciences, 374:1–1, 2016.

580 [26] L. De Lorenzis, A. McBride, and B.D. Reddy. Phase-field modelling of fracture in single crystal
581 plasticity. GAMM Mitteilungen, 39:7–34, 2016.

582 [27] M.J. Borden, T.J.R. Hughes, C.M. Landis, A. Anvari, and I.J. Lee. A phase-field formulation for
583 fracture in ductile materials: Finite deformation balance law derivation, plastic degradation, and
584 stress triaxiality effects. Computer Methods in Applied Mechanics and Engineering, 312:130–166,
585 2016.

586 [28] F. Amiri, D. Millan, Y. Shen, T. Rabczuk, and M. Arroyo. Phase-field modeling of fracture in linear
587 thin shells. Theoretical and Applied Fracture Mechanics, 69:102–109, 2014.

588 [29] M. Ambati and L. De Lorenzis. Phase-field modeling of brittle and ductile fracture in shells with
589 isogeometric NURBS-based solid-shell elements. Computer Methods in Applied Mechanics and Engi-
590 neering, 312:351373, 2017.

31
 Wu et al.

591 [30] J. Kiendl, M. Ambati, L. De Lorenzis, H. Gomez, and A. Reali. Phase-field description of brittle
592 fracture in plates and shells. Computer Methods in Applied Mechanics and Engineering, 312:374394,
593 2017.

594 [31] C. Miehe, F. Welschinger, and M. Hofacker. A phase field model of electromechanical fracture. Journal
595 of the Mechanics and Physics of Solids, 58:1716–1740, 2010.

596 [32] A. Abdollahi and I. Arias. Phase-field modeling of crack propagation in piezoelectric and ferroelectric
597 materials with different electromechanical crack conditions. Journal of the Mechanics and Physics,
598 60:2100–2126, 2012.

599 [33] C. Maurini, B. Bourdin, G. Gauthier, and V. Lazarus. Crack patterns obtained by unidirectional
600 drying of a colloidal suspension in a capillary tube: experiments and numerical simulations using a
601 two-dimensional variational approach. International Journal of Fracture, 184:75–91, 2013.

602 [34] B. Bourdin, J.J. Marigo, C. Maurini, and P. Sicsic. Morphogenesis and propagation of complex cracks
603 induced by thermal shocks. Physical Review Letters, 112:014301, 2014.

604 [35] C. Miehe, L. M. Schänzel, and H. Ulmer. Phase field modeling of fracture in multi-physics problems.
605 Part I. Balance of crack surface and failure criteria for brittle crack propagation in thermo-elastic
606 solids. Computer Methods in Applied Mechanics and Engineering, 294:449–485, 2015.

607 [36] C. Miehe, M. Hofacker, L. M. Schänzel, and F. Aldakheel. Phase field modeling of fracture in multi-
608 physics problems. Part II. Coupled brittle-to-ductile failure criteria and crack propagation in thermo-
609 elastic-plastic solids. Computer Methods in Applied Mechanics and Engineering, 294:486–522, 2015.

610 [37] T. Wu and L. De Lorenzis. A phase-field approach to fracture coupled with diffusion. Computer
611 Methods in Applied Mechanics and Engineering, 312:196–223, 2016.

612 [38] M. Ambati, R. Kruse, and L. De Lorenzis. A phase-field model for ductile fracture at finite strains
613 and its experimental verification. Computational Mechanics, 57:149–167, 2016.

614 [39] T. Gerasimov and L. De Lorenzis. A line search assisted monolithic approach for phase-field computing
615 of brittle fracture. Computer Methods in Applied Mechanics and Engineering, 312:276–303, 2016.

616 [40] C. Miehe, F. Welschinger, and M. Hofacker. Thermodynamically consistent phase-field models of frac-
617 ture: Variational principles and multi-field FE implementations. International Journal for Numerical
618 Methods in Engineering, 83:1273–1311, 2010.

619 [41] M. R. Taha, E. Soliman, M. Sheyka, A. Reinhardt, and M. Al-Haik. Fracture toughness of hydrated
620 cement paste using nanoindentation. pages 105–111. Proceedings of the International Conferences on
621 Fracture Mechanics of Concrete and Concrete Structures, Seoul, 2010.

622 [42] A. Carpiuc, M. Poncelet, K. Kazymyrenko, H. Leclerc, and F. Hild. Innovative tests for characterizing
623 mixed-mode concrete fracture. 2nd International Conference on Technological Innovations in Nuclear
624 Civil Engineering, Paris, France, 2014.

625 [43] A. Carpiuc-Prisacari, M. Poncelet, K. Kazymyrenko, H. Leclerc, and F. Hild. A complex mixed-
626 mode crack propagation test performed with a 6-axis testing machine and full-field measurements.
627 Engineering Fracture Mechanics, 176:1–22, 2017.

628 [44] A. Carpiuc-Prisacari, M. Poncelet, K. Kazymyrenko, and H. Leclerc. Comparison between experi-
629 mental and numerical results of mixed-mode crack propagation in concrete: Influence of boundary
630 conditions choice. (submitted).

631 [45] Z. Tomičevć, F. Hild, and S. Roux. Mechanics-aided digital image correlation. Journal of Strain
632 Analysis for Engineering Design, 48:330–343, 2013.

32
 Wu et al.

(a)exp (a)num

(b)exp (b)num

(c)exp (c)num

(d)exp (d)num

(e)exp (e)num

Figure 27: Test III: contours of uY on face 1 of the specimen. On the left are the experimental data and
on the right are the numerical results. (a) − (e) Images at the end of steps 1,3,5,6,7 respectively. The exact
points (a)-(e) are shown in Figure 26. The specimen geometry in the simulation results is enlarged by a
scale factor of 1.5 for better visualization.

33
 Wu et al.

(a)exp (a)num

(b)exp (b)num

(c)exp (c)num

(d)exp (d)num

(e)exp (e)num

Figure 28: Test III: contours of uZ on face 1 of the specimen. On the left are experimental data and on
the right are numerical results. (a) − (e) images at the end of steps 1,3,5,6,7 respectively. The exact points
(a)-(e) are shown in Figure 26. Specimen geometry of simulations is enlarged by a scale factor of 1.5 for
better visualization.

34
 Wu et al.

(a)exp (a)num

(b)exp (b)num

(c)exp (c)num

(d)exp (d)num

(e)exp (e)num

Figure 29: Test III: contours of uY on face 2 of the specimen (mirrored). On the left are the experimental
data and on the right are the numerical results. (a) − (e) Images at the end of steps 1,3,5,6,7 respectively.
The exact points (a)-(e) are shown in Figure 26. The specimen geometry in the simulation results is
enlarged by a scale factor of 1.5 for better visualization.

35
 Wu et al.

(a)exp (a)num

(b)exp (b)num

(c)exp (c)num

(d)exp (d)num

(e)exp (e)num

Figure 30: Test III: contours of uZ on face 2 of the specimen (mirrored). On the left are the experimental
data and on the right are the numerical results. (a) − (e) Images at the end of steps 1,3,5,6,7 respectively.
The exact points (a)-(e) are shown in Figure 26. The specimen geometry in the simulation results is
enlarged by a scale factor of 1.5 for better visualization.

36
 Wu et al.

(a)f 1 (a)f 2

(b)f 1 (b)f 2

(c)f 1 (c)f 2

(d)f 1 (d)f 2

(e)f 1 (e)f 2

Figure 31: Test III: contour plot of the phase-field on the two side surfaces of the specimen. On the left
are results on face 1 and on the right are results on face 2 (mirrored) (a) − (e) Images at the end of steps
1,3,5,6,7 respectively.

37
 Wu et al.

(a) (b)

Figure 32: Test III: final crack pattern on the two side surfaces of the specimen in the experiment (a)
face 1 and (b) face 2 (mirrored).

(a) (b) (c)

Figure 33: Test III: Cross-sectional images of the region of the specimen close to the crack branching
point in the displacement contour, obtained from computed tomography at X = (a) 15 mm (b) 0 mm (c)
-15 mm. Zinc iodide is used to enhance the crack detectability.

42 ◦ C 36 ◦ C
40 ◦ C
36 ◦ C 38 ◦ C
36 ◦ C

Figure 34: Test III: Comparison of the crack branching angle between (a) the tomographic image at
X=-15 mm, (b) the simulated result on face 1 and (c) the simulated result on face 2 (mirrored).

38

View publication stats