Soil Dynamics and Earthquake Engineering 124 (2019) 172–183

Contents lists available at ScienceDirect

Soil Dynamics and Earthquake Engineering

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

Inspection of three sophisticated constitutive models based on monotonic T

and cyclic tests on fine sand: Hypoplasticity vs. Sanisand vs. ISA
T. Wichtmanna,∗, W. Fuentesb, T. Triantafyllidisc
a
Chair for Geotechnical Engineering, Bauhaus-Universität Weimar, Coudraystraße 11c, 99423, Weimar, Germany
b
Universidad del Norte, Barranquilla, Colombia
c
Institute of Soil Mechanics and Rock Mechanics, Karlsruhe Institute of Technology (KIT), Engler-Bunte-Ring 14, 76131, Karlsruhe, Germany

ARTICLE INFO ABSTRACT

Keywords: The prediction quality of three of the most sophisticated constitutive models for sands has been examined based
Constitutive models on a comparison of results from element test simulations and laboratory tests. Hypoplasticity with intergranular
Calibration strain, the Sanisand elastoplastic model in the version of 2004 and the recently proposed ISA (intergranular
Element test simulations strain anisotropy) model have been inspected. The laboratory tests performed on Karlsruhe fine sand used in the
Comparison with laboratory data
present study are freely available on the homepage of the first author. The model predictions have been in-
Monotonic and cyclic tests
spected for drained and undrained monotonic triaxial tests with various densities and pressures and for
monotonic tests with a few number of un- and reloading cycles. The main focus, however, lies on undrained
cyclic triaxial tests with either stress or strain control and with different densities, initial stresses and stress/
strain amplitudes. The strengths and weaknesses of the different constitutive models for various test conditions
are worked out.

1. Introduction examination of the constitutive equations based on element tests cre-

ates confidence for an application of the model to different real pro-
Numerical simulations of geotechnical structures under earthquake blems.
loading, which may eventually be affected by soil liquefaction, require The present paper documents a numerical study in which the data
the use of sophisticated constitutive models for the soil. For these from Refs. [24–26] have been used to inspect three of the most so-
analyses, the selected constitutive model is expected to reproduce the phisticated models mentioned above: Hypoplasticity with intergranular
mechanical behavior of the soil at different stress/strain amplitudes, strain, the Sanisand elastoplastic model in the version of 2004 and the
especially under undrained cyclic loading, whereby cyclic mobility ef- first version of ISA model (version 2014). Alternative or improved
fects and the accumulation of pore water pressure should be adequately versions of Sanisand [21], Hypoplasticity [18,19] and ISA [8–10] have
described. Several advanced constitutive models for non-cohesive soils been proposed in the meantime, but this paper restricts to the original
have been developed during the last two decades, e.g. the hypoplastic and most popular versions of the models, already incorporating the
model with intergranular strain [20,22], the Sanisand elastoplastic most important features of those types of models. The results of element
model [3,4], or most recently the ISA (intergranular strain anisotropy) test simulations with the three models are confronted with the experi-
model [7]. Likewise, a huge number of high-quality laboratory tests mental data. The element test simulations have been performed using
with various boundary conditions and control has been performed by the program Incremental Driver of Niemunis [16] in combination with
the authors at the Institute of Soil Mechanics and Rock Mechanics (IBF) Abaqus implementations of the constitutive models (the so-called
at Karlsruhe Institute of Technology (KIT) during the last decade. The UMAT's) written by A. Niemunis (Hypoplasticity with intergranular
experimental database has been published in Refs. [25,26] and made strain) or W. Fuentes (ISA, Sanisand). First, the calibration of the three
freely available on the homepage of the first author [24]. The data models is briefly explained in the next section. All experimental and
enable the calibration, inspection and further development of con- theoretical research reported herein has been done during the colla-
stitutive models for granular materials, considering as many different borate work of all authors at the IBF at KIT.
boundary conditions and types of control as possible. Such close

Corresponding author.
∗

E-mail addresses: [email protected] (T. Wichtmann), [email protected] (W. Fuentes),

[email protected] (T. Triantafyllidis).

https://doi.org/10.1016/j.soildyn.2019.05.001
Received 14 April 2018; Received in revised form 20 April 2019; Accepted 3 May 2019
Available online 03 June 2019
0267-7261/ © 2019 Elsevier Ltd. All rights reserved.
T. Wichtmann, et al. Soil Dynamics and Earthquake Engineering 124 (2019) 172–183

Table 1
Material parameters of Hypoplasticity with intergranular strain for Karlsruhe fine sand used for the simulations. The last row presents the tests used for the
calibration of the corresponding parameter.

c [°] ei0 [−] ec0 [−] ed0 [−] hs [MPa] n [−] α [−] β [−] R [−] mR [−] mT [−] R [−] [−]

33.1 1.212 1.054 0.677 4000 0.27 0.14 2.5 10 4 2.2 1.1 0.1 5.5
CPi STii ST ST OEDiii OED DMTiv OED UCTvi UCT UCT UCT UCT

i
CP Cone pluviation test.
ii
ST Standard test on minimum and maximum density.
iii
OED Oedometric test.
iv
DMT Drained monotonic triaxial test.
v
UMT Undrained monotonic triaxial test.
vi
UCT Undrained cyclic triaxial test.
vii
CSL Critical state line in e-p space.

2. Parameter calibration

The parameters of Hypoplasticity with intergranular strain for

Karlsruhe fine sand are summarized in Table 1. The critical friction
angle c has been determined from the inclination of a loosely pluviated
cone of sand. Following the parameter determination guide proposed in
Ref. [11], the characteristic void ratios at p = 0 are estimated from the
relations ei0 = 1.15emax , ec0 = emax and ed0 = emin , where emax and emin are
the maximum and minimum void ratios determined from standardized
laboratory tests [1]. Parameters hs and n were carefully calibrated to
reproduce the reduction of the void ratio e with increasing mean (ef-
fective) pressure p under oedometric compression conditions on loose
samples. The e-p behavior of dense samples in oedometric compression
tests was then reproduced by calibrating the parameter β. The para-
meter α has been adjusted to reproduce the peak stress measured in Fig. 1. Critical state in the e-p diagram from numerous undrained monotonic
drained monotonic triaxial tests with dense samples. Parameters for the triaxial tests on Karlsruhe fine sand with a variation of initial density and mean
effective stress, and its approximation by the equations incorporated in the
intergranular strain extension, namely R, mR , mT , R , χ, have been ca-
three inspected constitutive models.
librated to reproduce the mechanical behavior of the sand in an un-
drained cyclic triaxial test, named TCUI7 in the database [24], with
constant deviator stress amplitude q ampl = 60 kPa, initial mean effec- state lines in the p-q plane for triaxial compression or extension, re-
tive stress p0 = 200 kPa and relative density spectively. These parameters can be obtained from the inclination of
Dr = (emax e)/(emax emin ) 100 = 67% (medium dense sample). The the final phase of the effective stress paths in undrained monotonic
parameters were chosen in order to achieve the best possible re- triaxial tests. In the present case, Mc and Me have been determined by
production of the measured curves of strain amplitude 1ampl (N ) and calculation using the Mohr-Coulomb relations and
pore water pressure accumulation u acc (N ) . A respective comparison is Me = 6 sin c /(3 + sin c ) with the critical friction angle c = 33. 1 .
provided in Section 4. Users interested in more details for the calibra- The parameters nd and fb0 controlling the intensity of dilatancy and the
tion of intergranular strain parameters are referred to Refs. [15,23]. peak of the stress-strain curves, respectively, were determined based on
Table 2 contains the parameters of the ISA model. For this work, the the curves of volumetric strain v ( 1) and deviatoric stress q ( 1) mea-
same parameters reported by Ref. [7] for the same sand are used, except sured in a drained monotonic triaxial test with initially dense state.
for the intergranular strain parameters mR , β, h and cz which were It is worth to mention, that the recent version of ISA model [10]
recalibrated herein to reproduce the mechanical behavior of the un- uses the same parameters of the hypoplastic model for sands for
drained cyclic test TCUI7. A detailed guide for the parameter de- monotonic loading (ei0 , ec0 , ed0 , hs , n, α and β), which makes its cali-
termination of the ISA model can be found in Ref. [7]. A brief de- bration simpler for users of Hypoplasticity.
scription is given in the following lines: The parameters ei0 , i , npi and ne The parameters of the Sanisand elastoplastic model for Karlsruhe
control the e-p behavior under oedometric compression of either loose fine sand are summarized in Table 3. The parameters were calibrated
or dense sand samples, and therefore, their calibration follows from according to the following procedure: Parameters e0 , λ and ξ describe
oedometric curves. Poisson's ratio ν has been calibrated from the initial the critical state in the e-p plane derived again from the data in Fig. 1.
inclination of the q ( 1) curves in drained monotonic triaxial tests. The Mc and Me are the same parameters as used in the ISA model and thus
parameters ec0 , c , npc describing the critical state line in the e-p plane calibrated in the same way. The parameter m = 0.05 describing the
have been determined from a fitting to the experimental data from shape of the cone-type yield surface has been overtaken from Ref. [3] to
numerous undrained monotonic triaxial tests performed on the fine enable a proper small-strain response. The shear modulus factor G0 was
sand (Fig. 1). Mc and Me with their ratio c = Me /Mc describe the critical chosen such that a realistic prediction for the initial stiffness in drained

Table 2
Material parameters of ISA (2014) model for Karlsruhe fine sand used for the simulations. The last row presents the tests used for the calibration of the corresponding
parameter (see Table 1 for acronyms).
ei0 [−] i [−] npi [−] ne [−] ν [−] ec0 [−] c [−] npc [−] Mc [−] c [−] nd [−] fb0 [−] R [−] mR [−] β [−] h [−] cz [−] rF [−]

1.21 0.0045 0.8 3.2 0.35 1.067 0.00573 0.68 1.34 0.7 0.5 1.8 10 4 1.7 0.1 11 50000 1.6
OED OED OED OED UMTv CSLvii CSL CSL UMT UMT DMT DMT UCT UCT UCT UCT UCT DMT

173
T. Wichtmann, et al. Soil Dynamics and Earthquake Engineering 124 (2019) 172–183

Table 3
Material parameters of Sanisand (2004) model for Karlsruhe fine sand used for the simulations. The last row presents the tests used for the calibration of the
corresponding parameter (see Table 1 for acronyms).
e0 [−] λ [−] ξ [−] Mc [−] Me [−] m [−] G0 [−] ν [−] h 0 [−] ch [−] nb [−] A 0 [−] nd [−] zmax [−] cz [−]

1.103 0.122 0.205 1.34 0.938 0.05 150 0.05 10.5 0.75 1.2 0.9 2.0 20.0 10000
CSL CSL CSL UMT UMT – DMT, UCT OED DMT, UCT DMT DMT DMT DMT UCT UCT

monotonic triaxial tests was obtained. A very low (and thus rather overestimates the stiffness at the reloading phase, leading to an over-
unrealistic) value of Poisson's ratio ν = 0.05 is necessary in this model shooting effect. The shear stiffness parameter G0 of the Sanisand model
to achieve an oedometric stiffness that comes close to the experimental has been calibrated from cyclic triaxial tests. Having parameter G0
data. The parameters h 0 and ch of the plastic modulus have been varied fixed, the oedometric stiffness can be solely influenced by Poisson's
until the best possible agreement of the simulated and the measured ratio ν. The diagrams in Fig. 2e and f reveal, however, that even a very
stress-strain relationships q ( 1) in the drained monotonic triaxial tests low value of = 0.05 still leads to an overestimation of stiffness, in
was reached. The parameter nb was adopted to reproduce the peak particular in case of loose samples. Probably, a better agreement of
deviatoric stresses measured in these tests, while the constants nd and experimental and numerical oedometric curves can be achieved by
A0 controlling the dilatancy were chosen to reproduce the volumetric employing the version 2008 of the Sanisand model [21], which in-
response v ( 1) best possible. The parameters zmax and cz were selected corporates a wedge-type yield surface with a cap.
to reproduce the cyclic mobility effect of the test TCUI7. The para- Simulations of three drained monotonic triaxial tests with different
meters G0 and h 0 obtained from tests with monotonic loading were densities and an initial mean effective stress p0 = 100 kPa are shown in
slightly adjusted based on the cyclic test in order to achieve a better Fig. 3. For this and subsequent simulations with isotropic initial
reproduction of stiffness and pore water pressure accumulation. stresses, a fully isotropic mobilization of the initial state of the inter-
granular strain has been assumed, i.e. hii = R/ 3 . This is justified
with the preceding isotropic increase of effective stress towards p0 .
3. Inspection based on data from monotonic tests From these simulations, one may note that the peak strength is fairly
well reproduced by the hypoplastic and ISA models. This could be ex-
Fig. 2 presents simulations of oedometric compression tests on loose pected since parameters α (Hypoplasticity), and fb0 (ISA) have been
and dense samples with a single un- and reloading cycle. The simula- carefully calibrated to reproduce the peak strength of these tests. The
tions have been started at 1 = 1 kPa. In case of the hypoplastic and ISA Sanisand model shows a disagreement of the q ( 1) -curves before
models, the intergranular strain has been assumed to be initially fully reaching the peak strength, because the parameters G0 and h 0 have been
mobilized in the vertical direction. Since the parameters hs , n and β of optimized to reproduce a cyclic loading test and not monotonic loading.
the hypoplastic model and i , npi and ne of the ISA model have been The volumetric strain behavior v ( 1) in Fig. 3 (lower row) is well re-
calibrated to reproduce the first loading curves of these oedometric produced by Sanisand and ISA models considering that both models
tests, the good prediction of these models at this phase of the test is offer the possibility to control these curves independently of the peak
evident. The oedometric stiffness upon unloading is also well re- strength behavior: parameters nd and rF of the ISA model and nd and A0
produced by both models. However, both models show a weak per- of the Sanisand model can be calibrated to reproduce the v ( 1) -curves
formance during the final stage of reloading for the loose case (upper once the parameters for the critical state line at the e-p space are fixed.
row in Fig. 2). For the dense case (lower row in Fig. 2), Hypoplasticity

Fig. 2. Simulations of oedometric compression tests on loose (Dr0 = 4% and 7%, upper row) or dense samples (Dr0 = 83% and 88%, lower row) with a single un-
and reloading cycle: a,b) Hypoplasticity with intergranular strain, c,d) ISA, e,f) Sanisand.

174
T. Wichtmann, et al. Soil Dynamics and Earthquake Engineering 124 (2019) 172–183

Fig. 3. Simulations of drained monotonic triaxial tests with different initial densities (Dr0 = 21%, 63%, 85%) and an initial mean effective stress p0 = 100 kPa: a,b)
Hypoplasticity with intergranular strain, c,d) ISA, e,f) Sanisand.

The hypoplastic model does not show an agreement on the v ( 1) curves, pressure-dependence of shear strength is well described by the three
because its parameter α controlling these curves was previously cali- models. Once more, the v ( 1) behavior is well assessed by the Sanisand
brated to reproduce the peak stress of the test. It is obvious that Hy- and ISA models, while a disagreement is observed in case of the hy-
poplasticity has a clear disadvantage against the other two models poplastic model.
considering that it does not offer any independent parameter to control Fig. 5 presents simulations of undrained monotonic triaxial tests
the v ( 1) behavior under drained triaxial conditions. starting at a mean effective stress of p0 = 200 kPa, and having different
The results from simulations of four drained triaxial tests with dif- initial densities. Compressional and extensional triaxial shearing are
ferent initial pressures p0 = {50,100,200,400} kPa performed on samples included. The simulations show a general agreement with the measured
with medium density (57 %< Dr0 < 68%) are provided in Fig. 4. The q ( 1) -curves (see Fig. 5, lower row) while a lack of accuracy on the

Fig. 4. Simulations of drained monotonic triaxial tests on medium dense samples (57% Dr0 68%) with different initial pressures p0 = 50, 100, 200 or 400 kPa:
a,b) Hypoplasticity with intergranular strain, c,d) ISA, e,f) Sanisand.

175
T. Wichtmann, et al. Soil Dynamics and Earthquake Engineering 124 (2019) 172–183

Fig. 5. Simulations of undrained monotonic triaxial tests with different initial densities (Dr0 = 34, 68, 87% in triaxial compression, Dr0 = 29, 58, 98% in triaxial
extension): a,b) Hypoplasticity with intergranular strain, c,d) ISA, e,f) Sanisand.

effective stress paths in the q-p space is observed. Probably, the best under undrained conditions is shown in Fig. 7. A loose sample (Dr0
simulations are provided by the hypoplastic model because it in- = 23%) was tested with an initial mean effective stress of p0
corporates a parameter mT controlling the transverse stiffness, which = 200 kPa. An unloading to q = 0 was performed every 1 = 0.2%.
strongly influences the behavior during an undrained shearing in the q- The hypoplastic simulation shows a clear example of ”overshooting” of
p space after the isotropic consolidation phase. The ISA and Sanisand the effective stress path during the first reloading. The overshooting
models do not incorporate parameters controlling the transverse stiff- effect is actually a serious limitation in this and other models. As a
ness, and therefore their simulations are deficient at this stage (a recent matter of fact, one of the authors of the model admits to be aware about
development of the ISA model has now tackled this issue [10]). All this drawback, which is shown by this and other models, such as the
models show a poor reproduction of the development of pore water Hardening Soil (HS) Small Strain model included in the software Plaxis
pressure, considering that most parameters controlling the q-p behavior [17]. Though this disagreement, the accumulation of the pore water
were calibrated to simulate the cyclic (and not monotonic) behavior pressure upon the cycles is realistically simulated by the hypoplastic
under undrained conditions. Note also, that the Sanisand model pre- model. The ISA and Sanisand models show a higher accumulation rate
dicts higher stress ratios q/p than measured in the experiment. This last of the pore water pressure. Such larger accumulation rate is certainly
observation is disappointing, considering that the parameter nb was accompanied by an unrealistically rapid decrease of the maximum de-
carefully calibrated with the maximum stress ratios observed in the viatoric stresses reached during the individual cycles (see q- 1 diagrams
drained triaxial tests to reproduce the bounding surface of the soil. This in Fig. 7c,e).
drawback corroborates some ideas of other authors pointing to the fact
that the bounding surface of the soil cannot be fully described with the
mean effective stress p and void ratio e for all stress-strain paths [19], as 4. Inspection based on data from undrained cyclic tests
assumed by the Sanisand model.
Fig. 6 presents simulations of a drained monotonic test interrupted Selected undrained cyclic triaxial tests performed on KFS have been
by four unloadings (to q = 0) with subsequent reloading phases. The simulated with the three constitutive models. Simulations of the un-
initial effective mean stress was p0 = 100 kPa and the sample was drained cyclic test TCUI7 already mentioned in Section 2 are shown in
medium dense (Dr0 = 62%). The axial strain increment between two Fig. 8. One may note that the effective stress path is fairly well re-
adjacent unloadings was chosen as 1 = 6%. The results show that the produced by the Sanisand and ISA models, showing an accurate re-
employed constitutive models deliver a lower stiffness during the final production of the butterfly-like shape during the cyclic mobility phase
phase of reloading as observed in the experiments. This issue is ex- (Fig. 8c,e). Hypoplasticity delivers a too contractive response during
plained with the fact, that although the three models incorporate some the first cycle and an unrealistic lens-shaped effective stress path during
mechanisms to capture the small-strain effects, the latter are limited to the final stage of the test. Contrary to the experiment, this model does
some strain amplitudes which depend on the selected parameters. In not reach a zero effective mean stress (p = 0) at the end of the test.
contrast, experiments involving unloading-reloading cycles of medium Evidently, this drawback is explained with the fact, that while the
or large strain amplitudes have shown an enlargement of the strain formulations of the ISA and Sanisand models have been extended by an
amplitude at which the small strain effects should act. This enlargement additional tensorial state variable to reproduce the cyclic mobility
is for example evident at the reloading stages of the oedometric tests, paths, Hypoplasticity uses the same equation without any modification.
especially on loose samples, see Fig. 2 (first row). Some new Sanisand- Looking at the stress-strain relationships (Fig. 8b,d,f), it is evident that
type models already include ”initialization tensors” intending to im- the hypoplastic and Sanisand models show a strong bias in strain ac-
prove this behavior [2] under undrained conditions. cumulation during the cyclic mobility phase. The latter drawback can
An inspection of the model performance based on a similar test be attributed to the following reasons: first of all, these models in-
corporate different slopes of the critical state lines for compression and

176
T. Wichtmann, et al. Soil Dynamics and Earthquake Engineering 124 (2019) 172–183

Fig. 6. Simulations of a drained monotonic triaxial test interrupted by an unloading to q = 0 after each strain increment 1 = 6% ( p0 = 100 kPa, Dr0 = 62%): a,b)
Hypoplasticity with intergranular strain, c,d) ISA, e,f) Sanisand.

Fig. 7. Simulations of an undrained monotonic triaxial test interrupted by an unloading to q = 0 after each strain increment 1 = 0.2% ( p0 = 200 kPa, Dr0 = 23%):
a,b) Hypoplasticity with intergranular strain, c,d) ISA, e,f) Sanisand.

extension in the q-p space, denoted by Mc and Me respectively. Hence, similar to other formulations in the literature [5,6,12–14]. With this
for these undrained cyclic paths, a larger strain accumulation is always mechanism, the model is able to reproduce almost symmetrical ”wings”
obtained on the extension side (q < 0 ) than on the compression side during the cyclic mobility phase in the q- 1 space (Fig. 8d). Despite of
(q > 0 ). The ISA model does not show this bias in strain accumulation this achievement, one may note that the ISA model ceases to accumu-
because although the model considers the slopes Mc and Me as well, its late the axial strain when reaching a value of approximately | 1 | 3%, in
dilatancy, and therefore its stress hardening under undrained shearing, contrast to the experimental result, where the strain amplitude during
is strongly controlled by a fabric tensor which evolves with the devia- the cyclic mobility phase grows with increasing number of cycles. The
toric strain tensor. This fabric tensor forces the critical state line in the latter drawback is related to the fact, that the model fails to degrade the
e-p space to be reached after a certain amount of deviatoric strain, stiffness during the cyclic mobility phase.

177
T. Wichtmann, et al. Soil Dynamics and Earthquake Engineering 124 (2019) 172–183

Fig. 8. Simulations of undrained cyclic triaxial test TCUI7 on a medium dense sample (Dr0 = 67%) with isotropic consolidation ( p0 = 200 kPa, 0 = 0) and stress
cycles (q ampl = 60 kPa): a,b) Hypoplasticity with intergranular strain, c,d) ISA, e,f) Sanisand.

Fig. 9a and b compare the measured and simulated curves in test medium dense sand (Fig. 8). All three models overestimate the rates of
TCUI7 of the normalized accumulated pore water pressure u acc (N )/ p0 pore water accumulation u acc . Once more, Hypoplasticity and the Sa-
and the axial strain amplitude 1ampl (N ) , respectively. The results show nisand model generate a bias in the extensional strain with almost
that the pore pressure accumulation rate and axial strain amplitude are constant rate during the cyclic mobility phase, while the strains pre-
well reproduced by the models. Since u acc has been evaluated at q = 0 dicted by the ISA model are limited to | 1 | 1.5%.
after the completion of a full cycle, a ratio of u acc / p0 = 1 is neither The variation of the pore water pressure accumulation rate with
reached by Hypoplasticity nor by ISA, but almost by Sanisand (compare density and stress amplitude is further inspected in Fig. 12. Therein the
the effective stress paths in Fig. 8a,c,e). Fig. 9c presents the develop- amplitude-pressure ratio CSR = q ampl /(2p0 ) is plotted versus the
ment of the maximum axial strain | 1 | reached under triaxial compres- number of cycles N required to reach the onset of large axial strain
sion (q > 0 ) and extension (q < 0 ) during the last 10 cycles. Once more generation | 1 |> 1%. This criterion is used as a synonym for initial li-
this diagram confirms that Hypoplasticity and the Sanisand model show quefaction since not all models reach u acc / p0 = 1 (see Fig. 9a). Simu-
a bias in strain accumulation upon the cyclic mobility phase. lations of undrained cyclic tests with an initial mean effective stress of
Simulations of a similar test (nominated TCUI1 in the database p0 = 100 kPa, considering three different densities (Dr0 = 26, 63 and
[24]) on a loose sample (Dr0 = 27%) are shown in Fig. 10. In contrast 81%, corresponding to average values from laboratory test series) and
to the experimental evidence, the three models predict a butterfly- different stress amplitudes have been performed. The CSR (N ) -curves
shaped effective stress path during the final stage of the test. The rate of resulting from these simulations are confronted with experimental data
pore water pressure accumulation u acc simulated by the hypoplastic (CSR versus number of cycles to | 1 |> 1%). Sanisand and ISA deliver
model is closer to the test results than that of ISA and Sanisand (too quite similar CSR (N ) -curves, being, however, much steeper than the
large u acc values). The axial strain amplitude during the cyclic mobility experimental CSR (N ) relationships. Hypoplasticity seems to capture
phase simulated by the ISA model obviously increases with decreasing better the CSR dependence. Nevertheless, all models show an under-
density (here | 1 | 8 %, Fig. 10d). estimation of the influence of the density on the CSR (N ) -curves. The
The performance of the three models in case of dense sand (Dr0 experience of the authors with these models is, that some parameters
= 87%, test TCUI17, Fig. 11) does not differ much from that for controlling the accumulation rate upon cyclic loading seem to show a

Fig. 9. Comparison of the results of the simulations using the three different constitutive models with the experimental data of test TCUI7: a) normalized accu-
mulated pore water pressure u acc (N )/ p0 , b) axial strain amplitude 1ampl (N ) , c) maximum absolute values of axial strain | 1 | in compression and extension during the
last 10 cycles.

178
T. Wichtmann, et al. Soil Dynamics and Earthquake Engineering 124 (2019) 172–183

Fig. 10. Simulations of undrained cyclic triaxial test TCUI1 on a loose sample (Dr0 = 27%) with isotropic consolidation ( p0 = 200 kPa, 0 = 0) and stress cycles
(q ampl = 30 kPa): a,b) Hypoplasticity with intergranular strain, c,d) ISA, e,f) Sanisand.

Fig. 11. Simulations of undrained cyclic triaxial test TCUI17 on a dense sample (Dr0 = 87%) with isotropic consolidation ( p0 = 200 kPa, 0 = 0) and stress cycles
(q ampl = 60 kPa): a,b) Hypoplasticity with intergranular strain, c,d) ISA, e,f) Sanisand.

density-dependence, although these models are thought to simulate a (compressional) axial strain compared to the minimum (extensional)
wide range of void ratios. one. This observation is best reflected by the ISA model. However, the
Fig. 13 shows simulations of a test (TCUA16) on a medium dense restriction of axial strains achievable by this model is visible also in
sample (Dr0 = 56%) with stress cycles after anisotropic consolidation. Fig. 13d. Similar to the isotropic tests, during cyclic mobility Hypo-
In this test the deviatoric stress amplitude q ampl was larger than the plasticity and Sanisand predict an accumulation of axial strain of almost
initial deviatoric stress q0 , i.e. the effective stress path crosses the p axis. constant rate (here in the compressional direction).
The stress paths in the q p space simulated by the hypoplastic model Fig. 14 inspects the case of another test on medium dense sand (Dr0
ends up with unrealistic lens-shaped loops, while the unsymmetrical = 64%) with anisotropic consolidation, but now q ampl < q0 holds, such
butterfly-shaped path generated by ISA and Sanisand is close to the that the effective stress path lies completely in the compressional re-
experimental observations. Larger discrepancies between the test re- gime of the q-p plane (q > 0 ). In agreement with the experiment
sults and the simulations can be found in the stress-strain relationships. (TCUA2), all simulations end up in a lens-shaped effective stress path
The experiment shows a somewhat larger increase of the maximum repeatedly passed through with each further cycle. However, the

179
T. Wichtmann, et al. Soil Dynamics and Earthquake Engineering 124 (2019) 172–183

cycles is generally less accurate (Fig. 16). In the laboratory test on dense
sand (Dr0 = 94%, TCUE17) the effective stress path finally ends up at
the origin p = q = 0 . In contrast, the hypoplastic simulation repeatedly
passes through an eight-shaped stress path, located far away from
p = 0. Consequently, a liquefaction as observed in the experiment is not
reached in the hypoplastic simulation. Of course, this is related to the
fact that Hypoplasticity does not incorporate any extension accounting
for cyclic mobility effects as the other models. If a dense material is
simulated by this model, the dilatant effect during the cycles is stronger
than the contractant effect. The large butterfly-shaped effective stress
path predicted by the Sanisand model contrasts the experimental results
as well, but at least p = q = 0 is passed temporarily. The ISA model
performs better in this test, predicting a smaller butterfly.
Fig. 12. Amplitude-pressure ratio q ampl /(2p0 ) vs. the number of cycles N re- A similar test (TCUE15) on loose sand (Dr0 = 29%) is investigated
quired to reach the onset of large axial strain generation (| 1 |> 1%). in Fig. 17. In congruence with the test results, the effective stress path
from the simulation with the ISA model comes close to p = q = 0 .
predicted lens is too narrow in case of Hypoplasticity and ISA, while the While the eight-shaped hypoplastic path stagnates at p 5 kPa in this
area encompassed by the stress path is more realistic in the results of case, the Sanisand simulation once again ends up with an unrealistic
the Sanisand simulation. The inclination of the final lens is in turn best butterfly. It is, however, somewhat smaller than in the case of dense
reproduced by Hypoplasticity, while ISA and Sanisand predict an un- sand.
realistic adaption of the lens to the failure line from undrained mono-
tonic tests. Once more, the Sanisand model fails to capture the max- 5. Summary and conclusions
imum stress ratio controlled by the bounding surface of the model.
In agreement with the experiment (TCUE1), all three constitutive In this work, the performance of three different constitutive models
models reproduce the state of zero effective stress p q 0 reached for the simulation of monotonic and cyclic loading of sand has been
after a certain number of strain cycles with relatively small amplitudes analyzed, namely Hypoplasticity with intergranular strain, the Sanisand
( 1ampl = 6 10 4 , Dr0 = 64%, Fig. 15). The fact that the resulting stress 2004 model and the ISA 2014 model. Their formulations consider the
amplitudes are similar to those of the test indicates that the small-strain influence of void ratio, stress tensor, and additional internal variables to
stiffness of the three models has been well adjusted, or at least, for this improve their prediction capabilities on cyclic loading. Hence, they are
particular test. All three models overestimate the pore pressure accu- assumed to reproduce well the mechanical behavior on a wide range of
mulation rate uacc , i.e. the number of cycles necessary to reach densities and confining pressures. An analysis of the simulations on
p q 0 is predicted too low. This is probably due to the fact, that different tests showed that although some experimental observations
the parameters controlling the accumulation rate were calibrated for a are satisfactorily captured, other important features are poorly re-
different test with different density and stress amplitude, and as men- produced. From this analysis, the following observations related to the
tioned before, these parameters have shown to be density-dependent. limitations of these models and their use on Boundary Value Problems
The stress-strain relationships provided in Fig. 15b,d,f show a quite (BVP) are highlighted:
good agreement between the experiment and the simulations, espe-
cially in the case of ISA and Sanisand. • Simulations on oedometric tests demonstrate that models as
The constitutive description of the material response to large strain Sanisand, which incorporates a wedge-type open yield surface, are

Fig. 13. Simulations of undrained cyclic triaxial test TCUA16 on a medium dense sample (Dr0 = 56%) with anisotropic consolidation ( p0 = 200 kPa, q0 = 50 kPa,
0 = 0.25) and stress cycles with q
ampl > q (q ampl = 60 kPa): a,b) Hypoplasticity with intergranular strain, c,d) ISA, e,f) Sanisand.
0

180
T. Wichtmann, et al. Soil Dynamics and Earthquake Engineering 124 (2019) 172–183

Fig. 14. Simulations of undrained cyclic triaxial test TCUA2 on a medium dense sample (Dr0 = 64%) with anisotropic consolidation ( p0 = 300 kPa, q0 = 150 kPa,
0 = 0.5) and stress cycles with q
ampl < q (q ampl = 120 kPa): a,b) Hypoplasticity with intergranular strain, c,d) ISA, e,f) Sanisand.
0

not able to capture the oedometric behavior. This disables the model loading. One can show that these monotonic tests are better re-
for example, to simulate correctly Boundary Value Problems (BVPs) produced with other sets of parameters, at the cost of reducing the
where settlement is of interest. In some way, this limitation suggests prediction capabilities on cyclic loading.
that the yield surface shape for Sanisand-type models should be • Simulations on cyclic loading showed overshooting effects on some
carefully reformulated in order to account for correct assessment of reloading paths, especially with the hypoplastic model. Actually,
volumetric behavior under oedometric conditions. this undesired effect has been frequently observed in many ad-
• Even though these models are aimed to simulate a wide range of vanced models. To the authors' opinion, this drawback emerges
stress/strain amplitudes, their calibration showed that it is im- from the fact that these models lack of proper mechanisms to cap-
possible to obtain a single set of parameters delivering accurate si- ture well the strain amplitude at which small-strain effects should
mulations on both monotonic and cyclic loading. To give an ex- act. It evidently cannot be captured based on an elastic locus of
ample, simulations of undrained monotonic shearing in Fig. 5 constant stress ratio amplitude (as by the Sanisand yield surface) or
showed in general a poor performance, considering that most constant strain amplitude (as by the ISA and hypoplastic models).
parameters were calibrated to match the behavior under cyclic Herein, more investigation is recommended to enhance model

Fig. 15. Simulations of undrained cyclic triaxial test TCUE1 on a medium dense sample (Dr0 = 64%) with anisotropic consolidation ( p0 = 200 kPa, q0 = 150 kPa,
0 = 0.75) and strain cycles of relatively small amplitude ( 1 = 6 10 4 ): a,b) Hypoplasticity with intergranular strain, c,d) ISA, e,f) Sanisand.
ampl

181
T. Wichtmann, et al. Soil Dynamics and Earthquake Engineering 124 (2019) 172–183

Fig. 16. Simulations of undrained cyclic triaxial test TCUE17 on a dense sample (Dr0 = 94%) with isotropic consolidation ( p0 = 200 kPa, q0 = 0, 0 = 0) and strain
cycles of relatively large amplitude ( 1ampl = 1 10 2 ): a,b) Hypoplasticity with intergranular strain, c,d) ISA, e,f) Sanisand.

Fig. 17. Simulations of undrained cyclic triaxial test TCUE15 on a loose sample (Dr0 = 29%) with isotropic consolidation ( p0 = 200 kPa, 0 = 0) and strain cycles of
relatively large amplitude ( 1ampl = 1 10 2 ): a,b) Hypoplasticity with intergranular strain, c,d) ISA, e,f) Sanisand.

Table 4
Qualitative ranking of the performance of the models on different observations based on the simulations of the present work (+ ++ is the best score while + is the
lowest one).
Description Sanisand Hypoplasticity ISA

Oedometric first loading + +++ +++

Oedometric cycles ++ + ++
Maximum stress ratio under drained conditions +++ +++ +++
Maximum stress ratio under undrained conditions + ++ ++
Pore pressure reproduction under undrained monotonic conditions + ++ +
Ability to avoid overshooting issues on reloading paths ++ + ++
Cyclic mobility effects ++ + ++
Accumulation rate under undrained stress controlled cycles + ++ +
Accumulation rate under undrained strain controlled cycles + ++ +

182
T. Wichtmann, et al. Soil Dynamics and Earthquake Engineering 124 (2019) 172–183

formulations considering the mentioned observation. its relation to critical state theory. Int J Numer Anal Methods Geomech 2010;35(8).

• The CSR (N )-curves showed that all three models are still not ac- [6] Fu P, Dafalias Y. Study of anisotropic shear strength of granular materials using
DEM simulation. Int J Numer Anal Methods Geomech 2010;35(10).
curate enough on the reproduction of the number of cycles required [7] Fuentes W. Contributions in mechanical modelling of fill material. Dissertation,
to reach the onset of large strain development, herein defined as | 1 | Veröffentlichungen des Institutes für Bodenmechanik und Felsmechanik am
= 1%. Particularly, the simulations showed an underestimation of Karlsruher Institut für Technologie. Heft 2014;179.
[8] Fuentes W, Hadzibeti M, Triantafyllidis Th. Constitutive model for clays under the
the influence of the density. This suggests that for a certain BVP, the ISA framework. In: Triantafyllidis Th, editor. Holistic simulation of geotechnical
calibration of a set of parameters should be limited to reproduce a installation processes: benchmarks and simulations. Springer International
particular small range of densities. Publishing; 2016. p. 115–29.

• The Sanisand and ISA models were able to capture fairly well the
[9] Fuentes W, Tafili M, Triantafyllidis T. An ISA-plasticity-based model for viscous and
non-viscous clays. Acta Geotech 2018;13(2):367–86.
cyclic mobility effect. A bias in strain accumulation was observed on [10] Fuentes W, Wichtmann T, Lascarro C. Extended hypoplasticity for sands with ISA
Hypoplasticity and the Sanisand model. To the authors' opinion, this for liquefaction analysis (submitted). Acta Geotech 2019.
[11] Herle I, Gudehus G. Determination of parameters of a hypoplastic constitutive
drawback may not significantly affect simulations of BVPs if they
model from properties of grain assemblies. Mech Cohesive-Frict Mater
are limited to a maximum strain of approximately = 5%. 1999;4(5):461–86.
Otherwise, excessive distortion of finite elements is also expected, or [12] Li X, Dafalias Y. Anisotropic critical state theory: role of fabric. J Eng Mech
even problems related to the loss of the stress rate objectivity. 2012;138:263–75.

•
[13] Li X, Li XS. Micro-macro quantification of the internal structure of granular mate-
Simulations of undrained cyclic loading with constant deviatoric rials. J Eng Mech 2009;135(7):641–56.
stress amplitudes reached the point of zero effective stress [14] Masin D. Incorporation of meta-stable structure into hypoplasticity. In:
(q = p = 0 ) only in case of loose sand. On the other hand, most Triantafyllidis Th, editor. Numerical modelling of construction processes in geo-
technical engineering for urban environment. Taylor and Francis; 2006. p. 283–90.
models showed a ratio between the accumulated pore water pres- [15] Niemunis A. Extended hypoplastic models for soils. Habilitation,
sure u acc and the initial mean effective stress p0 greater than 0.9, i.e. Veröffentlichungen des Institutes für Grundbau und Bodenmechanik, Ruhr-
u acc / p0 > 0.9 after reaching the onset of large axial strain | 1 |> 1%. Universität Bochum. Heft Nr 2003;34 available from www.pg.gda.pl/aniem/an-
liter.html.
This suggests that the mechanisms responsible for the cyclic mobi- [16] Niemunis A. Incremental driver user's manual. 2008 available from www.pg.gda.
lity effects are still density-dependent, and the parameters control- pl/aniem/an-liter.html.
ling this behavior cannot be calibrated for all densities. This me- [17] Niemunis A, Cudny M. Discussion on Dynamic soil-structure interaction, A three-
dimensional numerical approach and its application to the Lotung case study. Poor
chanism should be carefully revised and improved to capture better performance of the HSS model. Comput Geotech 2018;98:243–5.
the observed behavior. [18] Niemunis A, Grandas Tavera CE. Computer aided calibration, benchmarking and
• The model parameters were adjusted to capture the maximum stress check-up of constitutive models for soils. Some conclusions for neohypoplasticity.
In: Triantafyllidis Th, editor. Lecture Notes in Applied and Computational
ratio observed in monotonic drained tests. However, the outcoming
Mechanics. Vol. 82. Holistic simulation of geotechnical installation processes -
maximum stress ratio under undrained monotonic shearing was not theoretical results and applications. Springer; 2017. p. 168–92.
well reproduced with the Sanisand model. This suggests that the [19] Niemunis A, Grandas Tavera CE, Wichtmann T. Peak stress obliquity in drained and
bounding surface of the soil cannot be fully described only by the undrained sands. Simulations with neohypoplasticity. In: Triantafyllidis Th, editor.
Lecture Notes in Applied and Computational Mechanics. Vol. 80. Holistic simula-
void ratio e and mean effective stress p as assumed by this model. tion of geotechnical installation processes - benchmarks and simulations. Springer;
Experimental corroboration of this statement can be found in Ref. 2015. p. 85–114.
[19]. [20] Niemunis A, Herle I. Hypoplastic model for cohesionless soils with elastic strain
range. Mech Cohesive-Frict Mater 1997;2(4):279–99.
[21] Taiebat M, Dafalias Y. SANISAND, simple anisotropic sand plasticity model. Int J
Finally, the aforementioned limitations have been qualitatively Numer Anal Methods Geomech 2008;32(8):915–48.
ranked in Table 4. [22] von Wolffersdorff P-A. A hypoplastic relation for granular materials with a pre-
defined limit state surface. Mech Cohesive-Frict Mater 1996;1(3):251–71.
[23] Wegener D, Herle I. Prediction of permanent soil deformations due to cyclic
References shearing with a hypoplastic constitutive model. Geotechnik 2014;37(2):113–22.
[24] T.Wichtmann, www.torsten-wichtmann.de, Homepage.
[25] Wichtmann T, Triantafyllidis Th. An experimental data base for the development,
[1] DIN 18126: Bestimmung der Dichte nichtbindiger Böden bei lockerster und dich-
calibration and verification of constitutive models for sand with focus to cyclic
tester Lagerung. 1996.
loading. Part I: tests with monotonic loading and stress cycles. Acta Geotech
[2] Boulanger RW, Ziotopoulou K. Formulation of a sand plasticity plane-strain model
2016;11(4):739–61.
for earthquake engineering applications. Soil Dynam Earthq Eng 2013;53:254–67.
[26] Wichtmann T, Triantafyllidis Th. An experimental data base for the development,
[3] Dafalias Y, Manzari M. Simple plasticity sand model accounting for fabric change
calibration and verification of constitutive models for sand with focus to cyclic
effects. J Eng Mech 2004;130(6):622–34.
loading. Part II: tests with strain cycles and combined cyclic and monotonic loading.
[4] Dafalias Y, Papadimitriou A, Li X. Sand plasticity model accounting for inherent
Acta Geotech 2016;11(4):763–74.
fabric anisotropy. J Eng Mech 2004;130(11):1319–33.
[5] Fu P, Dafalias Y. Fabric evolution within the shear bands of granular materials and

183