You are on page 1of 18

Acta Geotechnica (2022) 17:3421–3438

https://doi.org/10.1007/s11440-021-01441-0 (0123456789().,-volV)(0123456789().
,- volV)

RESEARCH PAPER

Automated calibration of advanced soil constitutive models. Part I:


hypoplastic sand
Tomáš Kadlı́ček1 • Tomáš Janda2 • Michal Šejnoha2 • David Mašı́n1 • Jan Najser1 • Štěpán Beneš2

Received: 15 January 2021 / Accepted: 4 December 2021 / Published online: 16 January 2022
Ó The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature 2022

Abstract
This paper discusses an automated deterministic approach to parameters calibration of the hypoplastic model for sand. The
calibration is performed on results from basic laboratory experiments such as the oedometric test, isotropic compression
test, and the drained and undrained triaxial shear tests. The calibration method is structured in a hierarchical order and
implemented into a free-to-use online application called ExCalibre. The method is based on the sensitivity study performed
prior to the development of the calibration method. The calibration procedure respects the physical meaning of the
calibrated parameters and their influence on the stiffness and asymptotic states, rather than performing a blind optimization
of an objective function.

Keywords Calibration  Hypoplasticity  Software

1 Introduction laboratory protocol as a part of their service, the calibration


of more advanced models remains neglected. This can be
The development of the advanced constitutive soil model attributed to several issues. Firstly, there is a lack of
theories such as hypoplasticity [16, 32, 53, 56], barodesy knowledge or the tools necessary for the calibration of
[24, 35] or bounding surface plasticity [6, 47] continues advanced constitutive models. Secondly, it is often an
steadily. However, these theories are not well known in the unavailability of the necessary equipment to conduct suc-
engineering community and the basic elastoplastic models, cessful calibration exercises, such as local transducers,
such as the linear elastic and perfectly plastic model with which are essential when calibrating the small strain stiff-
the Mohr-Coulomb failure criterion or their enhanced ness [42] of the enhanced hypoplastic models.
versions are adopted. While laboratories often provide
values of the Mohr-Coulomb strength parameters and the
stiffness parameters for various ranges of stresses in the

& Tomáš Kadlı́ček


tomas.kadlicek@natur.cuni.cz
1
Faculty of Science, Charles University, Albertov 6,
128 00 Praha 2, Czechia
2
Faculty of Civil Engineering, Czech Technical University in
Prague, Thákurova 7, 166 29 Praha 6, Czechia

123
3422 Acta Geotechnica (2022) 17:3421–3438

Notation
Stress and strain quantities General
r Stress tensor NCL Normal compression line
r^ Stress tensor ratio INCL NCL at the isotropic state
r^ Deviatoric part of r^ CSL Critical state line
h Lode’s angle h[x] Heaviside function of x
r1 ; r2 ; r3 Principal stresses H(x) Hessian matrix
I1 , I2 , I3 Stress invariants SS Scaled sensitivity
ra Axial stress CSS Composite scaled sensitivity
rr Radial stress SBS State boundary surface
p Mean stress HS Hypoplastic sand
pe Hvorslev’s stress USCS Unified Soil Classification System

pc Overconsolidation stress Optimization


q Deviatoric stress S System
s Diameter of the Mohr’s circle A Action imposed on the system
t Radius of the Mohr’s circle M Constitutive model
u Pore pressure m Parameters of a constitutive model
K0 Lateral Earth pressure coeff. R Reaction of the system
e Strain tensor i Experimental value at the point i
Uexp
epl Plastic strain tensor i Numerical value at the point i
Unum
eel Elastic strain tensor Pi The value of the i-th parameter
etot Total strain tensor Wi Weight factor at point i
ea Axial strain N Number of observations/counts
ev Volumetric strain E(x) Objective function of the variable x

eeng Engineering strain Experiments


etrue True strains e Void ratio

Hypoplastic sand parameters ac Slope of the CSL


hs Granular hardness kc Slope of the NCL
n Par. controlling curvature of NCL einit;1 Initial void ratios
ei0 ei at p = 0 kPa pinit;1 Initial mean stresses
ec0 ei at p = 0 kPa ec0;1 Variation of the parameter ec0
ed0 ed at p = 0 kPa e1 Distinct void ratios on NCL
uc Crit. state friction angle p1 Distinct mean stresses on
a Par. controlling the peak friction angle kc;1 Distinct slopes of curved NCL
b Par. controlling the model’s stiffness a0 Variation of the parameter a

Hypoplastic equation quantities b0 , b1 , b2 Variation of the parameter b0


L; N Stiffness tensors emax Maximum void ratio
fd Pyknotropy function
fs ; fb Barotropy function
F; a; w; # Strength defining scalars
ei Void ratio at the lowest density
ec Void ratio at the the critical state line
ed Void ratio at the highest density
hi Barotropy denominator
dij Kronecker delta
1 Identity matrix
I 4th order identity matrix

*All stresses referenced in this article represent the effective portion of the total stress

The need for reliable calibration tools, not only for a numerous software that identifies the parameters of dif-
preliminary design but also for improvements of the design ferent soil models. The calibration procedures are regarded
during construction, has led to the development of as an inverse analysis, since the parameters m of the

123
Acta Geotechnica (2022) 17:3421–3438 3423

during sampling and handling, the field experiments have


the advantage of being able to execute the test on the on its
intact state.
Consequently, the calibration procedures can be per-
formed using a single element method in the case of labo-
ratory experiments or Finite Element Methods in the case of
problems concerning a geotechnical structure. A combina-
tion of both is often applied by the laboratory experimental
Fig. 1 Schematic representation of the direct and inverse analysis
results supported or further specified with the field mea-
constitutive model M are not known in advance but the surements. Such examples can be found in [4, 25, 39, 61].
The optimization methods used in the calibration pro-
reaction R of the system S(M,m) to the action A is
[59, 61], see Fig. 1. cess can be divided in deterministic, stochastic and their
The goal of calibration is to minimize the objective error combination.
The deterministic methods are optimization methods
function E(U), which is defined as a function of the dif-
ference between the observation or an experiment Uexp and that do not involve any random search and the same
solution is generated for a given input value. Common
a simulation Unum . The most common method used is the
examples are the simplex method [8, 34, 45, 59, 63] or
least square method [7, 45]. Nevertheless, more sophisti-
gradient-based methods such as the Levenberg–Marquardt
cated dimensionless formulae can be used, such as
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi algorithm [25, 61], the Newton or quasi-Newton opti-
u N !2ffi
uX U i
 U i mization [4, 5, 7, 26, 28, 38, 48, 49, 51]. In order to obtain
EðUÞ ¼ t
exp num ð1Þ
Wi i
; reliable results of the optimization, the initial values have
i¼1
U exp
to be relatively close to the global minimum, otherwise the
where Uexpi i
and Unum are the values of the experimental and optimization process could be trapped in a local minimum.
simulated data at the point of interest i, respectively, and Another option can be initiating the optimization process
PN from multiple initial values [45], which decreases the
Wi represents its weight satisfying i¼1 Wi ¼ 1. The
possibility of an incorrect solution. This problem is even
variable N defines the number of observational points
more relevant in the case of optimization with multiple
[17, 28, 60]. The obvious shortcoming of Eq. (1) is a
i variables.
presence of Uexp in the denominator. Any method thus
The stochastic methods are optimization methods that
i
should account for a case of Uexp ¼ 0. It should also be generate random variables in pursuit of the minimization of
noted that a perfectly minimized error function E(U) does the objective function. Before the optimization itself, a
not guarantee the correctness of the solution. This is even sampling process is undertaken that can be classified as
more relevant in the case when more parameters are opti- random or uniform. A variety of stochastic methods have
mized simultaneously as various combinations of the been employed to determine constitutive model parame-
parameters can provide a well minimized objective error ters. These include the genetic algorithm-based calibration
function E(U) while neglecting the physical meaning of the method [3, 9, 15, 17–20, 28, 29, 36, 38, 44–46, 49, 50, 60],
parameters. Furthermore, the calibration procedure itself the differential evolution method [52, 64], and other
can, in general, be regarded as an ill posed optimization stochastic methods such as the particle swarm optimization
problem, since the calibration offers a non-unique solution [2, 22, 23, 40, 62, 63, 65, 66], the simulated annealing
to a given problem. method [57] or neural network [43]. The stochastic meth-
In geotechnics, the problem is represented by a numer- ods are capable of determining stationary points over the
ical simulation that ought to imitate the observation of defined domain and thus can provide a good picture of the
laboratory or field measurements obtained either a priori or objective function properties. It can be decided which of
during the construction work. The laboratory measure- the solutions best agree with one’s experiences.
ments usually consist of compression and shear experi- A comparison of the calibration methods can be found in
ments that will provide an idea of the soil behaviour with [45], where the deterministic simplex method and the
regard to its compression, stiffness and strength charac- stochastic genetic method were compared on the calibra-
teristics [59]. Typical experiments conducted in the labo- tion of the elastoplastic Mohr-Coulomb model and the
ratory are the oedometric test, the isotropic compression strain hardening model. It was pointed out that the genetic
test, the drained and undrained triaxial test or the direct algorithm was 12 times more time demanding than the
shear test. In addition, these scheduled laboratory tests simplex method. Similar conclusions were found in [28].
usually precede the onset of construction work. While the However, it appeared that it is necessary to perform a
specimens intended for a laboratory test can be damaged stochastic optimization if the initial value of a parameter

123
3424 Acta Geotechnica (2022) 17:3421–3438

cannot be closely specified or if the problem contains more application ExCalibre. Examples of both the sensitivity
parameters to optimize. In addition, it has been shown in analysis and manual calibration are presented in this paper
[28] that it is extremely difficult to use a deterministic together with the results of the calibrations performed by
method to optimize several parameters simultaneously if the ExCalibre software. This paper is followed with the
they are highly sensitive or coupled towards the same part II [21], which focuses on the calibration of the
experiments. Therefore, the experiments used for the cali- hypoplastic clay and Modified Cam-Clay model.
bration should be fundamentally diverse since coupling
effects can appear between several parameters in a given
experiment [61]. The advantages of the deterministic and 2 Hypoplastic sand model
stochastic methods were combined in [5, 62, 63], where a
stochastic method was used to determine local minimums The initial studies of the hypoplastic models were per-
and a deterministic method was employed for a local formed in pursuit of finding tensor-valued functions satis-
optimization. fying the observed behaviour of particular materials [56].
Calibrations based on either the deterministic or Although further studies and enhancements were con-
stochastic methods often use only a portion of model’s ducted [1, 10, 13, 41, 55], the hypoplastic model for sand
parameters for the calibration and the rest is calibrated by (HS) considered in this paper as is referred to as the von
means of analytical relations. This assumption is also often Wolffersdorff’s hypoplastic model [53]. The behaviour of
based on sensitivity analysis [4, 28, 45, 58, 61]. the hypoplastic sand model is controlled by the following
Even though the stochastic methods are powerful tools eight parameters:
in the case of a poorly defined problem or when there is a
hs —Granular hardness
lack of quality data, it is of high computational cost. It is
n—Parameter controlling the curvature of the compres-
also necessary to ensure that a correct sampling is per-
sion line
formed and a proper set of parameters from the calibration
ei0 —Void ratio at the lowest density at p ¼ 0 kPa
results are chosen. Furthermore, the calibration process is
ec0 —Void ratio at the critical state line at p ¼ 0 kPa
focused on the optimization of the objective error function,
ed0 —Void ratio at the highest density at p ¼ 0 kPa
while the physical meaning of the model parameters is
uc —Critical state friction angle
somewhat sidelined and assured only through the preset
a—Parameter controlling the peak friction angle
parameter limits. Given that the constitutive models
b—Parameter controlling the model’s stiffness
themself are not perfect tools, focusing solely on the
optimization of the objective error function to its global
minimum might be misleading. Since advanced soil models
based on the critical soil mechanics have well defined
parameters according to the thoroughly examined asymp-
totic behaviour of soils [30], i.e., well illustrated by the
Normal Compression Line (NCL) or the Critical State Line
(CSL), it is advantageous to use analytical equations to
obtain the initial values of the model parameters and only
then use deterministic methods, such as the Newton opti-
mization method to improve the calibration. Granted, the
minimum necessarily does not have to be the global one,
however, it holds on to the physical meaning of the cali-
brated parameters. This approach was verified during the
development of the calibration method presented in this
paper. The effect of the parameters was studied with the aid
of the local sensitivity analysis and, consequently, a hier-
archical order of the calibration was established. Further-
more, the automated calibration was compared to a manual
calibration performed by a specialist familiar with the
adopted constitutive model. In addition, the calibration
method was thoroughly tested and fine-tuned on a data
library gathered from sixteen different locations consisting
of up to eleven separate experiments. The developed
method is currently implemented in the online calibration Fig. 2 Hypoplastic sand SBS

123
Acta Geotechnica (2022) 17:3421–3438 3425

Table 1 Hypoplastic sand parameters


Specimens hs (kPa) n (–) ei0 (–) ec0 (–) ed0 (–) a (–) b (–) u c ( )

Dobrany 65339 0.207 1.42 1.183 0.592 0.05 4.0 41.1


Hrusovany 1903 0.162 1.418 1.182 0.591 0.07 6.3 42.6
Jablonec 1037 0.232 1.478 1.232 0.616 0.02 4.1 42.6
Kralupy 1756 0.149 1.942 1.1618 0.809 0.23 4.8 41.2
Stvanice 853 0.199 2.026 1.689 0.844 0.23 4.9 35.8

Fig. 3 Calibrated specimen Dobrany

The hypoplastic sand model defines a unique relationship 3 The sensitivity analysis
between the mean stress p and the void ratio at its lowest
density ei as A properly defined calibration procedure calls for a sensi-
  n  tivity analysis as it provides great insight into the model
3p
ei ¼ ei0 exp  : ð2Þ behaviour and the effect of the parameters. The sensitivity
hs
of the selected types of loading towards given parameters
It is postulated that the void ratios at the critical state ec and helped to create a hierarchical order of the calibration
the maximum density ed evolve in an analogous manner as processes and defined crucial parameters driving the model
  n  behaviour. The sensitivity can be evaluated with the scaled
3p
ec ¼ ec0 exp  ; ð3Þ sensitivity
hs
oEi
and SSij ¼ Pj ; ð6Þ
oPj
  n 
3p
ed ¼ ed0 exp  : ð4Þ and composite scaled sensitivity for the parameter Pj
hs vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u N
The evolution of the limiting void ratios is displayed in u 1 X 2
CSSj ¼ t SSij ; ð7Þ
Fig. 2 with a portion of the State Boundary Surface (SBS) N i¼1
in the stress-void ratio space q  p  e.
Notice that the state boundary surface is bounded not as proposed in [14]. The constant Pj is the value of the j-th
only from above by the isotropic state designated by ei , but model parameter, Ei represents the objective error function
also from the bottom by the states of the highest density associated with the j-th value of the parameter P, see
defined by ed and therefore the condition Eq. (1). The sensitivity oEi =oPj is calculated using the
ed  e  ei ð5Þ central finite difference

should be fulfilled under all circumstances.

123
3426 Acta Geotechnica (2022) 17:3421–3438

Table 2 Sensitivity analysis characteristics single point of the parameter/error value space [65]. To that
OED CID
end, the parameters of five well calibrated specimens were
used as the initial stage for the sensitivity analysis.
HS e  ra E(e) q  ea E(q) The value of CSSj was also calculated separately for
q  ea Eðea Þ each specimen (N ¼ 1) in order to observe and compare
possible discrepancies in each individual specimen and
parameter. Eventually, the overall CSSj was calculated for
oEi Ei ðPj þ DPj Þ  Ei ðPj  DPj Þ all five specimens selected for the sensitive analysis.
:¼ ; ð8Þ
oPj 2DPj According to the USCS (Unified Soil Classification Sys-
tem), the soils were classified as SW (well grained sand)
where the step size of the parameter is set to a relatively
and SM (silty sand). The results of these calibrations can be
larger value
found in Table 1.
DPj ¼ Pj =100; ð9Þ For the sake of clarity, an example of the calibrated
specimen Dobrany is displayed in Fig. 3.
since the error developed during the integration of the
The sensitivity analysis of the hypoplastic sand model
constitutive equation can have unpredictable consequences
parameters was performed on a combination of one oedo-
which can greatly surpass the machine precision. Note that
metric test (OED) and one drained triaxial test (CID).
this parameter step size was used solely for the purposes of
Individual laboratory tests with the corresponding stress/
the sensitivity analysis. Summations are performed on all
strain and stress/void ratio spaces and their objective error
experimental observations, that is, for all N specimens.
functions E are shown in Table 2.
This type of the sensitivity analysis is regarded as a local
A graphical illustration of the objective error function E
sensitivity analysis since the sensitivity is evaluated in the
is displayed in Fig. 4. While one analysis was necessary in

i i
Fig. 4 Definition of objective function evaluated between the experiment and simulation (Uexp —value of an experiment at the point i, Unum —
value of a numerical simulation at the point i)

123
Acta Geotechnica (2022) 17:3421–3438 3427

3.1 Results

The sensitivity analysis of the oedometric test reveals the


importance of the three compression related parameters hs ,
n and ei0 , which control the position of INCL. This trend
was observed for all individual oedometric tests CSSj
(N ¼ 1 in Eq. (7)) and further supported by the overall
CSSj , see Fig. 5. The stiffness controlling parameter b has
almost identical effect as the parameter hs . The parameters
ec0 and uc controlling CSL have a similar effect. The
remaining parameters ed0 and a seem to have little influ-
Fig. 5 Overall CSSj , Oedometric test ence on the oedometric test. Since the oedometric test
specimen is relatively loose, the parameter ed0 does not
show much impact. Nonetheless, the effect might be dif-
ferent in the case of a more dense specimen.
In the triaxial test, the critical state friction angle uc had
the greatest impact on the objective error function F(q), see
Fig. 6. This tendency was observed in each CSSj (N ¼ 1 in
Eq. (7)). The second most significant parameter is ec0 .
These two parameters control the critical state in the q  p
and p  e space, respectively. In addition, the lower the
current void ratio e is bellow the critical state void ratio ec ,
the more significant the dilatancy and the higher the peak
friction angle up are. The sensitivity analysis thus provides
Fig. 6 Overall CSSj , Drained triaxial test Eðeq Þ reasonable and justifiable results. Other parameters do not
have an overall unique position and their effects are very
similar, see Fig. 6. It is worth pointing out that the
parameter a which affects the peak friction angle does not
demonstrate a significant role.
In the case of the objective function Eðea Þ, the results
show that two parameters ei0 and b have a distinct effect,
see Fig. 7. The effect of the parameters ei0 can be
explained when recalling the scalar functions in Eq. (39)
representing the barotropy fb . Different ranges between the
parameters ei0 , ec0 and ed0 influence the barotropy and
consequently the model stiffness. The parameter b influ-
ences the model stiffness by increasing the effect of the
limiting void ratios in Eq. (37) and Eq. (39).
Fig. 7 Overall CSSj , Drained triaxial test Eðea Þ
The remaining six parameters are split into two groups
where hs , ec0 and uc form the first group with approxi-
the case of the oedometric test regarding the void ratio
mately half the impact of parameter b. The parameters n,
E(e), see Fig. 4a, two analyses were conducted in the case
ed0 and a have only a small effect.
of the drained triaxial tests as seen in Fig. 4b, c. The first
The sensitivity analysis can be concluded by pointing
objective function is evaluated on the basis of the devia-
out the significance of the parameters controlling the
toric stress E(q) to observe sensitivity towards the strength
asymptotic states. The proportional compression test is
of the model. The second is obtained on the basis of axial
driven by the parameters hs , n and ei0 and the critical state
deformation Eðea Þ to test the influence of various param-
by ec0 and uc . These parameters appear to be the most
eters on the stiffness degradation. Each figure also illus-
i influential ones. Other parameters were shown to have
trates an example of the experimental (Uexp ) and numerical
gradually diminishing roles. Therefore, the parameters
i
(Unum ) value at the point i, Eq. (1). controlling the asymptotic states should be calibrated first
with high precision in order to obtain the most reliable
results. Recall that the experiments used for the sensitivity

123
3428 Acta Geotechnica (2022) 17:3421–3438

by the analytical calculation of the initial parameter values.


Subsequently, the optimization process in the form of the
Newton method is applied to find the correct solution. The
iterative process is defined by
xiþ1 ¼ xi  cHðxi Þ1 rFðxi Þ; ð10Þ

where x, FðxÞ, H ðxÞ, c represent the solution vector, the


vector of the function to be minimized, the Hessian matrix
and the step size correction, respectively.
Since the Newton optimization uses the second order
approximation at each step and converges towards the
closest stationary point, the method can converge to either
the maximum or minimum, see Fig. 8. When minimization
Fig. 8 Newton’s 2nd order approximation of the objective error function is of interest, control of the
gradient should be employed to force the convergence
analysis were conducted on specimens in a loose state, towards the minimum value of the objective function
which is preferable for the calibration of the asymptoticity E(x) [37]. It is also advantageous to restrict the increment
controlling parameters since achieving the asymptotic of the parameters during the optimization to avoid exces-
states is possible at lower stress levels. sive changes in the parameter values and adaptively
decrease the step size when it is not possible to find a new
minimum of the objective error function E(x) during sev-
4 Calibration testing and validation eral succeeding steps. Furthermore, the parameter values
should be constrained by the upper and lower bounds to
As was pointed out earlier, the deterministic method avoid absurd results. These limits are intrinsic to each
requires the initial parameter values to be in relative parameter.
proximity to the global minimum in order to find the cor- Convergence and the shape of the objective error
rect solution of the optimization process. In the case of functions E(x) were examined to test the Newton’s opti-
critical state soil models, this issue can be easily overcome mization method. This is well represented in Fig. 9, which

Fig. 9 Iteration procedure—parameters hs and n

Table 3 Hypoplastic sand parameters: Dobrany


Specimens hs (kPa) n (–) ei0 (–) ec0 (–) ed0 (–) a (–) b (–) u c ( )

Manual 405519.5 0.237 1.1953 0.9961 0.498 0.122 3.79 35.5


ExCalibre 52638.0 0.178 1.509 1.257 0.629 0.100 4.20 36.5

123
Acta Geotechnica (2022) 17:3421–3438 3429

Fig. 10 A comparison of the manual and automated calibration

Table 4 Hypoplastic sand parameters: Hrusovany specimen


hs (kPa) n (–) ei0 (–) ec0 (–) ed0 (–) a (–) b (–) u c ( )

Parameters 8350 0.204 1.131 0.943 0.471 0.14 5.5 36.8

depicts a converging iteration procedure for the parameters parameter ec0 to be the initial void ratio of the oedometric
hs and n on the oedometric test. Both parameters were test einit , see Fig. 10a, the automated calibration considers
optimized simultaneously. Notice that the rate of reduction the initial void ratio at the beginning of the oedometric test
of the objective error function E(e) is lowered in steps, as a to be positioned on the critical state line defined by the
progressive reduction of the step size is employed. limiting void ratio ec from Eq. (3). The parameter ec0 is
During the testing phases, the Newton iteration proce- then back calculated by Eq. (3) while optimizing the
dure proved to be sufficient. The calibration procedures parameters hs and n. The assumption ec0 ¼ einit adopted by
were applied in a hierarchical order that resulted from the the manual calibration was possible since the oedometric
performed Sensitivity analysis. specimen was specifically prepared so that the condition
The next step involved a validation of the developed einit ¼ emax is satisfied. Nonetheless, this assumption can-
procedures by comparing the results of the automated not, in general, be guaranteed.
calibration with a manual calibration performed by a spe- The optimization of the parameter b is designed to fit a
cialist familiar with the calibration of constitutive models. triaxial shear degradation in q  ea , see Fig. 10b. The
The results of the calibrations are compared in Table 3. parameter a is optimized to capture the peak state in q  ea .
The difference between the calibrations is mainly driven The evolution of the volumetric strain in ev  ea is not
by the approach to the parameter ec0 , which denotes the optimized.
initial void ratio of the critical state at the mean stress The triaxial test simulations in Fig. 10 clearly do not
p ¼ 0 kPa. Whilst the manual calibration considers the reach the critical state during the predefined deformation

123
3430 Acta Geotechnica (2022) 17:3421–3438

Fig. 11 HS asymptotic approach to the critical state

limits which correspond to ea  30%. To illustrate that the 5 Calibration procedure


simulations asymptotically reach the critical state defined
by constant stress ratio, simulations up to very large Analytical calibration techniques for the hypoplastic sand
(though unrealistic) axial strains are included in Fig. 11. can be found in [12, 31]. These analytical relations serve
Notice that the experimental Data in Fig. 11a are artifi- predominantly to establish the initial values of parameters
cially extended for a better comparison between the final for later optimization.
states of both experiments and simulations. Ultimate states A successful calibration of all parameters described in
of simulations are designated with grey circles as they this paper requires the combination of a compression and
coincide with the critical states indicated by CSL, see triaxial shear laboratory experiment. In the case of com-
Fig. 11b. pression, either an oedometric or isotropic compression test
is necessary. The triaxial shear test can be performed in
either a drained or undrained state.
Since the drained triaxial test is usually carried out for
coarse grained soils and the undrained triaxial test is per-
formed on fine grained soils, the calibration procedures
described in [12, 31] adopt these assumptions. Neverthe-
less, both triaxial experiments should be possible to use in
this calibration method.
In addition, the calibration should consider different
specimen states for different parameters. The parameter
controlling the asymptotic states should preferably be cal-
ibrated on the basis of specimens with a high relative
density, while the stiffness driving parameters should be
calibrated on a specimens with the relative density close to
one, found in close proximity to the examined area, in
order to capture the structure of the specimen. The

Fig. 12 Hypoplastic sand model calibration flowchart Fig. 13 Undrained triaxial tests in t  s

123
Acta Geotechnica (2022) 17:3421–3438 3431

calibration techniques described in this section are imple-


mented in the online calibration software ExCalibre
available at https://soilmodels.com/excalibre/.
Considering the Sensitivity analysis, the calibration is
structured in the hierarchical order as shown in Fig. 12.
Individual steps are discussed hereinafter.

5.1 Calibrate uc

This parameter is the only one that is not optimized and its
value remains the same throughout the calibration process.
The critical state friction angle uc can be determined in two
ways. The first option is to determine the critical state
friction angle uc as the angle of repose urep . This value can Fig. 15 The exclusion of the unstable section of the compression test
be specified in the input data uploaded to ExCalibre. The the critical state is considered to be reached at the highest
second option is to directly calculate uc from the results of deviatoric state q for undrained triaxial tests or at the end of
a triaxial shear test plotted in the invariant stress space loading for drained triaxial tests.
t  s, where s and t are the effective stresses defined in
terms of the effective principal stresses as 5.2 Modify Data
s ¼ ðr1 þ r3 Þ=2; ð11Þ
The obtained laboratory data must often be modified or
and other quantities must be calculated. Such a modification
t ¼ ðr1  r3 Þ=2: ð12Þ includes a transformation of the engineering strain eeng to
the true strain etrue due to the large deformation exhibited
To obtain reliable results, multiple triaxial shear tests are during the laboratory tests ( eeng a [ 20%). Additional
usually performed to evaluate uc . For sandy soils it is quantities, such as the void ratio e, the pore pressure u and
preferred to perform the drained triaxial test as it is the radial stress rr , are supplemented as a necessary input
promptly executed due to a highly porous structure of for simulations.
sands. The parameter uc is determined from the slope of Furthermore, notice that Eqs. (2)–(4) produce a double
the critical state line ac in Fig. 13 as curved INCL in the e  ln p space which flattens with the
uc ¼ arcsin ðac Þ; ð13Þ parameter n ! 0, see Fig. 14.
However, some experiments performed on very loose
where the slope ac is given by samples can provide a convex shape for NCL at the
PN beginning of the test, which implies a collapsible state.
si t i
ac ¼ Pi¼1
N 2
: ð14Þ This can results in extreme values of the parameter ec0 as
i¼1 si displayed in Fig. 15, because the void ratio at the begin-
The stresses si and ti represent the stress state at the critical ning of the compression test einit is considered for the
state for the ith triaxial test. In the automated calibration, calibration. It is therefore recommended that this prob-
lematic part of the experiment is omitted, as suggested in
Fig. 15.
Finally, because the basic version of the hypoplastic
sand model is not well suited for a cyclic loading, it is
better to leave out any cyclic section of tests to prevent the
unexpected effects that such sections can have on the
calibration.

5.3 Calibrate hs , n and ei0 , ec0 , ed0

The granular hardness hs and the exponent n appear in


Eqs. (2)–(4), which control the dependency of the bulk
modulus on the mean stress p. In order to accurately
determine these parameters, at least one of the specimens
Fig. 14 The evolution of INCL in e  ln p

123
3432 Acta Geotechnica (2022) 17:3421–3438

prepared for the compression test has to be prepared in a De


very loose, but not collapsible, state. Furthermore, the kc ¼  ð19Þ
D ln p
parameters can be determined on the basis of either
oedometric or isotropic compression test. The former test is or the rate form
simulated with a constant lateral earth pressure coefficient e_
kc ¼ p : ð20Þ
K0 . p_
The oedometric test is often preferred since it allows
higher stress levels to be reached and it is simple to per- To obtain the granular hardness hs we first differentiate
form. Since these five parameters are defined in the e  ln p Eq. (2) with respect to time to get
 
space, a modification of the oedometric data is necessary. ne 3p n
Given the fact that the sand specimen is prepared in a e_ ¼  _
p: ð21Þ
p hs
loose state and is thus close to the critical state, the Jaky
formula Combining Eqs. (20) and (21) then gives
 1n
K0 ¼ 1  sinðuc Þ; ð15Þ ne3
hs ¼ 3p3 : ð22Þ
kc;3
can be used and it follows that
ra þ 2K0 ra The most suitable way to obtain the parameter hs is from
p¼ : ð16Þ the central part of the compression line to capture an
3
overall slope, see Fig. 16, where the recommended state is
Equation (16) can be written in the logarithmic expression designated with subscript 3 [1, 33].
as The exponent n can be obtained when considering two
    different states on the proportional loading curve with
3p 3
ln ra ¼ ln ¼ ln þ ln p: ð17Þ unique values of the mean stress p and void ratio e. It is
1 þ 2K0 1 þ 2K0
preferable to choose states on the far extremes of the
Due to a possible confusion with the parameters of the compression curve labeled with subscripts 1 and 2 in
hypoplastic clay and Cam-Clay models, the slope of NCL Fig. 16 [1, 33]. This gives
in the e  ln ra space is designated here as  n  n
hs kc;1 hs kc;2
De n¼ ¼ ; ð23Þ
kc ¼ : ð18Þ 3p1 e1 3p2 e2
D ln ra
and finally
According to Eq. (17), the vertical distance (in terms of  
void ratio) between NCL of the oedometric test and INCL e1 kc;2
ln
of the isotropic compression test is a constant value; e k
n¼ 2 c;1
 : ð24Þ
therefore, the slope kc can also be expressed in the incre- p2
ln
mental form p1
It is worth pointing out that even though Eq. (2) fits the
experimental data for a wide range of stresses, it is less
reliable for very high (approx. p [ 10 MPa) mean stresses
p due to a grain crushing. The crushing limit depends
heavily on grains mineralogy [27], angularity and size
distribution [11].
Figure 17a, b show the effect of parameters hs and n.
The granular hardness hs controls the overall slope of the
proportional compression curves, whereas the exponent n
affects its curvature. The parameters hs and n derived by
the aforementioned procedure are accurate only for non-
crushing values of stress as hs and n change with changes
in the granulometric properties and non-zero initial stress
values [1, 12, 33].
An evolution of the critical void ratio ec with the mean
stress p is defined by Eq. (3), where the parameter ec0
refers to the critical void ratio at p = 0 kPa. Given that the
Fig. 16 Compression index

123
Acta Geotechnica (2022) 17:3421–3438 3433

Fig. 17 Parameters hs and n

parameters hs and n are already known, ec0 can be obtained experimental studies, the parameter can be estimated
from the oedometric or isotropic compression test. It has [12, 33] as
been experimentally confirmed that the void ratio at the
ed0 ¼ 0:5ec0 : ð26Þ
critical state ec closely corresponds to the maximum void
ratio emax and it was suggested in [12, 33] that equality emax
 ec0 is admissible. In addition, since the void ratio of 5.4 Optimize hs , n and ei0 , ec0 , ed0
loosely prepared specimen should be close to the critical
state, it is considered to be on CSL and the parameter ec0 is Once all the five parameters are evaluated analytically, the
back calculated from Eq. (3). calibration process advances to the optimization of the
An evolution of the isotropic void ratio ei with the mean parameters hs and n on all uploaded compression tests
stress p is defined by Eq. (2). The parameter ei0 is a more while the parameter ec0 and subsequently ei0 and ed0 are
fictitious value as it is represented by an array of identical recalculated from Eqs. (2)–(4). The objective error func-
spherical particles in the gravity free space. Although such tion E(e) is determined from the points distributed along
a state is experimentally unfeasible, it can be estimated the loading part of all the compression tests according to
[12, 33] as Eq. (1), i.e., the difference between the reconstituted or
ei0 ¼ 1:2emax : ð25Þ undisturbed specimen is not taken into account. This

The parameter ed0 represents the void ratio of maximal


density; its evolution is given by Eq. (4). Based on

Fig. 18 Calibration optimization Fig. 19 The effect of parameter a

123
3434 Acta Geotechnica (2022) 17:3421–3438

Fig. 20 The effect of parameter a under undrained triaxial conditions

optimization process is illustrated in Fig. 18 for the initial The parameter a is optimized on the basis of the
guess and the final optimized example. objective error function E(q) evaluated at a single point in
The simulation of the compression tests also requires the location of the peak state. The target is to achieve the
initial values of the parameters a and b, although they have same deviatoric stress at the peak state for both the simu-
not yet been determined. Therefore, the simulations are lation and experiment.
performed with initial guesses of a ¼ 0:2 and b ¼ 2. It was The effect of parameter a under undrained triaxial
observed that these values were appropriate. conditions is demonstrated in Fig. 20. Similarly to that
seen in the drained conditions, increasing the parameter a
5.5 Optimize a, b, hs , n, ei0 , ec0 , ed0 promotes more dilatative behaviour, see Fig. 20a. This
behaviour can also be expressed by the evolution of the
Although explicit procedures to obtain both exponents can mobilized friction angle umob or its counterpart g ¼ q=p
be found in [12], it seems to be advantageous to obtain from the meridian plane, see g plotted with the axial strain
these parameters by means of a parametric study provided ea in Fig. 20b. Therefore, under the undrained conditions,
that all other parameters were evaluated. The exponent a the parameter a is optimized on the basis of the objective
from Eq. (36) controls the value of the peak friction angle function E(q/p).
up as Fig. 19 suggests. Note that the hypoplastic sand model [53] experiences
an excessive accumulation of pore pressure under
undrained conditions [41]. This issue was improved with
the enhancement designated as the small strain stiffness
[41] or the implementation of Poisson’s ratio m [54] as the
additional parameter. Nonetheless, these modifications are
not accounted for in the presented calibration. The value of
the parameter a thus can reach as high values as a ¼ 0:8
when optimizing.
The parameter b in Eq. (37) influences the overall
stiffness for the current density and consequently the peak
state position of the triaxial stress/strain curve, see Fig. 21.
Therefore, the objective error function Eðea Þ is evaluated
from the q  ea chart while the points of interest i are
distributed within the boundaries qi 2 ð0; qmax Þ.
It is preferable to calibrate the parameters a and b from
the undisturbed specimens.
During a modification of the parameter b, changes of the
compression curve occur and it is therefore necessary to
Fig. 21 The effect of parameter b account for these changes and adjust the parameters driving

123
Acta Geotechnica (2022) 17:3421–3438 3435

triaxial tests. Two examples of the calibration performed


on the specimens selected from the ExCalibre library are
presented.
The hypoplastic sand model was calibrated on the
specimen designated as Hrusovany; this input file is also
available in the website library. The data file consists of
one oedometric test and three drained triaxial tests at initial
pressures of 50 kPa, 100 kPa, and 200 kPa. According to
the USCS, the specimen is classified as SW (well grained
sand). The results of the calibration are shown in Table 4.
The calibrated results for the tests are displayed in the
e  ln p space with the designated limiting void ratios ei , ec
and ed , see Fig. 22. It should be noted that the simulation of
the oedometric test does not start from the initial state
showed by the laboratory test. The initial stage that indi-
cates a collapsible behaviour was left out for illustrative
purposes as it produces a relatively low value of the
Fig. 22 All tests presented in e  ln p—Hrusovany specimen
parameter hs , if included. The critical state line designated
NCL, as proposed in Section 5.4. This loop is performed as by ec then passes through the initial state of the simulation
long as the iteration of the parameter b improves the indicated by the bold circle. The drained triaxial tests
solution. Nonetheless, the number of these cycles in converge towards the critical state line ec . However, it is
ExCalibre is limited to 15. not reached during the simulation. An evolution of the
volumetric deformation implies that the dilatative beha-
viour would continue and the critical state would be
6 ExCalibre reached later if the simulation continued.
While the peak friction angle up is well captured during
The calibration method described in this article is imple- the optimization of the parameter a, see Fig. 23a, the
mented in the online application ExCalibre available at: evolution of the volumetric strain ev in Fig. 23b does not
http://www.soilmodel.com/excalibre/. The application coincide well with the experimental results. This is due to
requires an Excel spreadsheet where the experimental the calibration setting since matching the stress evolution is
results can be input. The corresponding template is avail- more preferred to matching the evolution of strain.
able at the website of the application. The calibration
results are displayed via a user interface, which also allows
the user to further manually tune the parameters and 7 Conclusions
compare the updated simulations with the automatically
calibrated results. In addition, the website provides a The presented article introduces a technique for the cali-
library of experimental results from sixteen locations bration of the hypoplastic sand model. The key idea of this
consisting of various combinations of compression and method is to respect the asymptotic states defined by the

Fig. 23 Drained triaxial test—Hrusovany specimen

123
3436 Acta Geotechnica (2022) 17:3421–3438

model parameters rather than the whole optimization of the the Kronecker’s delta. Scalar functions a and F are
objective error function E as the asymptotes serve since expressed by
attractors for various stress states and loading directions. pffiffiffi
3ð3  sin uc Þ
To achieve this effect, records of experiments performed a¼ pffiffiffi ; ð31Þ
on reconstituted and undisturbed specimens are considered. 2 2 sin uc
While reconstituted specimens are preferred for the deter- and
mination of the asymptotic states, undisturbed specimens sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
should be used when assessing the soil stiffness. 1 2 2  tan2 w 1
F¼ tan w þ pffiffiffi  pffiffiffi tan w;
The developments of the method was preceded by local 8 2 þ 2 tan w cos 3# 2 2
sensitive analyses evaluated on five calibrated specimens.
ð32Þ
The calibration method was thoroughly tested on the
library of sixteen soils and various combinations of with invariants
experiments were used to assess the calibration capabili- pffiffiffi
ties. Furthermore, a comparison of the manual and auto- tan w ¼ 3jjr^ jj; ð33Þ
mated calibration was performed to remedy and improve and
the automated calibration. This process was illustrated on
pffiffiffi trðr^  r^  r^ Þ
an example of the Dobrany soil. cos 3# ¼  6 : ð34Þ
3
The calibration method is currently implemented in the ðr^ : r^ Þ2
online calibration application ExCalibre which is free to
Stress tensors r^ and r^ represent the stress ratio and the
use and available at https://soilmodels.com/excalibre/. The
deviatoric part of the stress ratio respectively
calibration method is fast and the current version enables
calibration of three constitutive models. Implementation of r
r^ ¼ ; ð35Þ
other constitutive models based on the critical state theory trr
is possible. The calibration algorithm presented in this Scalar functions fe and fd implement the effect of the
article represents the most reliable combination of experi- pyknotropy, i.e., dependency on a density. These functions
ments and numerical methods tested during the develop- are defined in the form
ment of the application ExCalibre.  
e  ed a
fd ¼ ; ð36Þ
ec  ed
Appendix: Hypoplastic sand model and
ec b
Based on the Bauers and Gudehus model, Wolffersdorff fe ¼ : ð37Þ
improved the hypoplastic equation for the predefined fail- e
ure surface; the hypoplastic model for sand can thus be The barotropy function fb incorporates dependency on the
expressed in the basic form as mean stress
r_ ¼ fb fe ðL : e_ þ fd Njjejj
_ Þ: ð27Þ 1
p ¼  I1 ; ð38Þ
3
The forth order tensor L and the second order tenor N are
expressed in the form respecting the Drucker-Prager and as
Matsuoka-Nakai failure surfaces as    
hs ei0 b 1 þ ei 3p 1n
fb ¼ ; ð39Þ
1  2  nhi ec0 ei hs
L¼ F I þ a2 r^ r^ ð28Þ
r^ : r^
with the scalar function hi defined by
and    
2
pffiffiffi ei0  ed0 a
aF hi ¼ 3 þ a  3a : ð40Þ
N¼ ðr^ þ r^ Þ; ð29Þ ec0  ed0
r^ : r^
where Equation (5) can be violated during some loading paths,
  such as unloading during oedometric or isotropic com-
1 pression tests and the situation ed [ e can occur. To rem-
I Iijkl ¼ ðdik djl þ dil djk Þ ; ð30Þ
2 edy this possibility, Niemunis [41] provided a solution that
enables the condition ed ¼ e to be fulfilled under all
represents the forth order symmetric tensor with dij being

123
Acta Geotechnica (2022) 17:3421–3438 3437

circumstances. However, a simpler and more straight for- 14. Hill, M.C.: Methods and guidelines for effective model calibra-
ward solution is to define a Heaviside function for fd as tion. In: Building partnerships, pp 1–10 (2000)
15. Javadi A, Farmani R, Toropov V, Snee C (1999) Identification of
fd ; ed  e; parameters for air permeability of shotcrete tunnel lining using a
h½fd ¼ ð41Þ genetic algorithm. Comput Geotech 25(1):1–24
0; ed [ e: 16. Jerman, J., Mašı́n, D.: Hypoplastic and viscohypoplastic models
for soft clays with strength anisotropy. Int J Numer Anal Methods
The hypoplastic differential equation is thus reduced to the
Geomech (in print)
hypoelastic form anytime Eq. (5) violates the lower limit. 17. Jin YF, Wu ZX, Yin ZY, Shen JS (2017) Estimation of critical
This alternative is considered during the calibration in this state-related formula in advanced constitutive modeling of
article. granular material. Acta Geotech 12(6):1329–1351
18. Jin YF, Yin ZY, Shen SL, Hicher PY (2016) Investigation into
Acknowledgements The financial support from research grant moga for identifying parameters of a critical-state-based sand
TA04031603 of the Technology Agency of the Czech Republic is model and parameters correlation by factor analysis. Acta Geo-
gratefully appreciated. The first, fourth and fifth authors appreciate tech 11(5):1131–1145
the financial support given by the INTER-EXCELLENCE project 19. Jin YF, Yin ZY, Shen SL, Hicher PY (2016) Selection of sand
LTACH19028 by the Czech Ministry of Education, Youth and Sports. models and identification of parameters using an enhanced
The second author acknowledges the financial support by the Euro- genetic algorithm. Int J Numer Anal Meth Geomech
pean Regional Development Fund within the Center of Advanced 40(8):1219–1240
Applied Sciences at the Czech Technical University in Prague 20. Jin YF, Yin ZY, Shen SL, Zhang DM (2017) A new hybrid real-
(CZ.02.1.01/0.0/0.0/16_019/0000778). The fourth author acknowl- coded genetic algorithm and its application to parameters iden-
edges the institutional support by the Center for Geosphere Dynamics tification of soils. Inverse Probl Sci Eng 25(9):1343–1366
(UNCE/SCI/006). 21. Janda, Kadlı́ček, Mašı́n, Šejnoha, Beneš, Najser (2021) Auto-
mated calibration of advanced soil constitutive models Part II:
Hypoplastic clay and modified Cam-Clay. Acta Geotechnica 1–2.
https://doi.org/10.1007/s11440-021-01435-y
References 22. Knabe T, Datcheva M, Lahmer T, Cotecchia F, Schanz T (2013)
Identification of constitutive parameters of soil using an opti-
1. Bauer E (1996) Calibration of a comprehensive hypoplastic mization strategy and statistical analysis. Comput Geotech
model for granular materials. Soils Found 36(1):13–26 49:143–157
2. Bharat TV, Sivapullaiah PV, Allam MM (2009) Swarm intelli- 23. Knabe T, Schweiger HF, Schanz T (2012) Calibration of con-
gence-based solver for parameter estimation of laboratory stitutive parameters by inverse analysis for a geotechnical
through-diffusion transport of contaminants. Comput Geotech boundary problem. Can Geotech J 49(2):170–183
36(6):984–992 24. Kolymbas D (2012) Barodesy: a new hypoplastic approach. Int J
3. Calvello M, Cuomo S, Ghasemi P (2017) The role of observa- Numer Anal Meth Geomech 36(9):1220–1240
tions in the inverse analysis of landslide propagation. Comput 25. Lecampion B, Constantinescu A, Nguyen Minh D (2002)
Geotech 92:11–21 Parameter identification for lined tunnels in a viscoplastic med-
4. Calvello M, Finno RJ (2004) Selecting parameters to optimize in ium. Int J Numer Anal Meth Geomech 26(12):1191–1211
model calibration by inverse analysis. Comput Geotech 26. Ledesma A, Gens A, Alonso E (1996) Estimation of parameters
31(5):410–424 in geotechnical backanalysis-I. Maximum likelihood approach.
5. Cekerevac C, Girardin S, Klubertanz G, Laloui L (2006) Cali- Comput Geotech 18(1):1–27
bration of an elasto-plastic constitutive model by a constrained 27. Leleu S, Valdes J (2007) Experimental study of the influence of
optimisation procedure. Comput Geotech 33(8):432–443 mineral composition on sand crushing. Géotechnique
6. Dafalias YF, Papadimitriou AG, Li XS (2004) Sand plasticity 57(3):313–317
model accounting for inherent fabric anisotropy. J Eng Mech 28. Levasseur S, Malécot Y, Boulon M, Flavigny E (2008) Soil
130(11):1319–1333 parameter identification using a genetic algorithm. Int J Numer
7. Finno RJ, Calvello M (2005) Supported excavations: observa- Anal Methods Geomech 32(2):189–213
tional method and inverse modeling. J Geotech Geoenviron Eng 29. Levasseur S, Malecot Y, Boulon M, Flavigny E (2009) Statistical
131(7):826–836 inverse analysis based on genetic algorithm and principal com-
8. Gioda G, Locatelli L (1999) Back analysis of the measurements ponent analysis: method and developments using synthetic data.
performed during the excavation of a shallow tunnel in sand. Int J Int J Numer Anal Methods Geomech 33(12):1485–1511
Numer Anal Meth Geomech 23(13):1407–1425 30. Masin D (2012) Asymptotic behaviour of granular materials.
9. Gras JP, Sivasithamparam N, Karstunen M, Dijkstra J (2017) Granular Matter 14(6):759–774
Strategy for consistent model parameter calibration for soft soils 31. Mašı́n D (2013) Clay hypoplasticity with explicitly defined
using multi-objective optimisation. Comput Geotech 90:164–175 asymptotic states. Acta Geotech 8(5):481–496
10. Gudehus G (1996) A comprehensive constitutive equation for 32. Mašı́n D (2014) Clay hypoplasticity model including stiffness
granular materials. Soils Found 36(1):1–12 anisotropy. Géotechnique 64(3):232–238
11. Hagerty M, Hite D, Ullrich C, Hagerty D (1993) One-dimen- 33. Masin D (2015) Hypoplasticity for practical applications, course
sional high-pressure compression of granular media. J Geotech handouts. https://web.natur.cuni.cz/uhigug/masin/hypocourse
Eng 119(1):1–18 34. Mattsson H, Klisinski M, Axelsson K (2001) Optimization rou-
12. Herle I, Gudehus G (1999) Determination of parameters of a tine for identification of model parameters in soil plasticity. Int J
hypoplastic constitutive model from properties of grain assem- Numer Anal Methods Geomech 25(5):435–472
blies. Mech Cohesive-Friction Mater Int J Exp Model Comput 35. Medicus G, Fellin W (2017) An improved version of barodesy for
Mater Struct 4(5):461–486 clay. Acta Geotech 12(2):365–376
13. Herle I, Kolymbas D (2004) Hypoplasticity for soils with low
friction angles. Comput Geotech 31(5):365–373

123
3438 Acta Geotechnica (2022) 17:3421–3438

36. Mendez FJ, Pasculli A, Mendez MA, Sciarra (2020) Calibration 53. Von Wolffersdorff PA (1996) A hypoplastic relation for granular
of the von wolffersdorff model using genetic algorithms. arXiv materials with a predefined limit state surface. Mech Cohesive-
preprint arXiv:2006.08433 Friction Mater Int J Exp Model Comput Mater Struct
37. Moin P (2010) Fundamentals of engineering numerical analysis. 1(3):251–271
Cambridge University Press, Cambridge 54. Wichtmann T (2005) Explicit accumulation model for non-co-
38. Moreira N, Miranda T, Pinheiro M, Fernandes P, Dias D, Costa hesive soils under cyclic loading. dissertation, schriftenreihe des
L, Sena-Cruz J (2013) Back analysis of geomechanical parame- institutes für grundbau und bodenmechanik der ruhr-universität
ters in underground works using an evolution strategy algorithm. bochum, heft 38
Tunn Undergr Space Technol 33:143–158 55. Wu W, Baurer E, Kolymbas D (1996) Hypoplastic constitutive
39. Nguyen LT, Nestorović T (2015) Nonlinear kalman filters for model with critical state for granular materials. Mech Mater
model calibration of soil parameters for geomechanical modeling 23:46–69
in mechanized tunneling. J Comput Civ Eng 30(2):04015025 56. Wu W, Kolymbas D (1990) Numerical testing of the stability
40. Nguyen-Tuan L, Lahmer T, Datcheva M, Stoimenova E, Schanz criterion for hypoplastic constitutive equations. Mech Mater
T (2016) A novel parameter identification approach for buffer 9(3):245–253
elements involving complex coupled thermo-hydro-mechanical 57. Yepes V, Alcala J, Perea C, González-Vidosa F (2008) A para-
analyses. Comput Geotech 76:23–32 metric study of optimum earth-retaining walls by simulated
41. Niemunis A (2003) Extended hypoplastic models for soils, vol annealing. Eng Struct 30(3):821–830
34. Inst. für Grundbau und Bodenmechanik, Vienna 58. Yin ZY, Hicher PY (2008) Identifying parameters controlling soil
42. Niemunis A, Herle I (1997) Hypoplastic model for cohesionless delayed behaviour from laboratory and in situ pressuremeter
soils with elastic strain range. Mech Cohesive-Friction Mater Int testing. Int J Numer Anal Methods Geomech 32(12):1515–1535
J Exp Modell Comput Mater Struct 2(4):279–299 59. Yin ZY, Jin YF, Shen JS, Hicher PY (2018) Optimization tech-
43. Obrzud RF, Vulliet L, Truty A (2009) Optimization framework niques for identifying soil parameters in geotechnical engineer-
for calibration of constitutive models enhanced by neural net- ing: comparative study and enhancement. Int J Numer Anal
works. Int J Numer Anal Methods Geomech 33(1):71–94 Methods Geomech 42(1):70–94
44. Pal S, Wathugala GW, Kundu S (1996) Calibration of a consti- 60. Yin ZY, Jin YF, Shen SL, Huang HW (2017) An efficient opti-
tutive model using genetic algorithms. Comput Geotech mization method for identifying parameters of soft structured
19(4):325–348 clay by an enhanced genetic algorithm and elastic-viscoplastic
45. Papon A, Riou Y, Dano C, Hicher PY (2012) Single-and multi- model. Acta Geotech 12(4):849–867
objective genetic algorithm optimization for identifying soil 61. Zentar R, Hicher PY, Moulin G (2001) Identification of soil
parameters. Int J Numer Anal Methods Geomech 36(5):597–618 parameters by inverse analysis. Comput Geotech 28(2):129–144
46. Pedroso DM, Williams DJ (2011) Automatic calibration of soil- 62. Zhang Y, Gallipoli D, Augarde C (2009) Simulation-based cali-
water characteristic curves using genetic algorithms. Comput bration of geotechnical parameters using parallel hybrid moving
Geotech 38(3):330–340 boundary particle swarm optimization. Comput Geotech
47. Petalas AL, Dafalias YF, Papadimitriou AG (2020) Sanisand-f: 36(4):604–615
Sand constitutive model with evolving fabric anisotropy. Int J 63. Zhang Y, Gallipoli D, Augarde C (2013) Parameter identification
Solids Struct 188:12–31 for elasto-plastic modelling of unsaturated soils from pres-
48. Rangeard D, Hicher P-Y, Zentar R (2003) Determining soil suremeter tests by parallel modified particle swarm optimization.
permeability from pressuremeter tests. Int J Numer Anal Methods Comput Geotech 48:293–303
Geomech 27(1):1–24 64. Zhao B, Zhang L, Jeng D, Wang J, Chen J (2015) Inverse analysis
49. Rechea C, Levasseur S, Finno R (2008) Inverse analysis tech- of deep excavation using differential evolution algorithm. Int J
niques for parameter identification in simulation of excavation Numer Anal Methods Geomech 39(2):115–134
support systems. Comput Geotech 35(3):331–345 65. Zhao C, Lavasan AA, Barciaga T, Zarev V, Datcheva M, Schanz
50. Rokonuzzaman M, Sakai T (2010) Calibration of the parameters T (2015) Model validation and calibration via back analysis for
for a hardening-softening constitutive model using genetic algo- mechanized tunnel simulations-the western scheldt tunnel case.
rithms. Comput Geotech 37(4):573–579 Comput Geotech 69:601–614
51. Tang YG, Kung GTC (2009) Application of nonlinear opti- 66. Zhao Hb, Yin S (2009) Geomechanical parameters identification
mization technique to back analyses of deep excavation. Comput by particle swarm optimization and support vector machine. Appl
Geotech 36(1–2):276–290 Math Model 33(10):3997–4012
52. Vardakos S, Gutierrez M, Xia C (2012) Parameter identification
in numerical modeling of tunneling using the differential evolu- Publisher’s Note Springer Nature remains neutral with regard to
tion genetic algorithm (DEGA). Tunn Undergr Space Technol jurisdictional claims in published maps and institutional affiliations.
28:109–123

123

You might also like