Вы находитесь на странице: 1из 12

Geo-Risk 2017 GSP 285 69

Probabilistic Calibration of a Modified Van der Poel Model Representing the


Viscoelastic Behavior of Sandstone

Yichuan Zhu, S.M.ASCE1; Lu Cong2; Xinli Hu, Ph.D.3; and Zenon Medina-Cetina, Ph.D.,
M.ASCE4
Downloaded from ascelibrary.org by Texas A&M University on 10/30/17. Copyright ASCE. For personal use only; all rights reserved.

1
Stochastic Geomechanics Laboratory, Zachry Dept. of Civil Engineering, Texas A&M Univ.,
College Station, TX 77843-3135. E-mail: yzhu@tamu.edu
2
Dept. of Civil Engineering, China Univ. of Geosciences, Wuhan, Hubei 430074. E-mail:
conglu0912@hotmail.com
3
Dept. of Civil Engineering, China Univ. of Geosciences, Wuhan, Hubei 430074. E-mail:
huxli2000@163.com
4
Stochastic Geomechanics Laboratory, Zachry Dept. of Civil Engineering, Texas A&M Univ.,
College Station, TX 77843-3135. E-mail: zenon@tamu.edu
`
Abstract

The study of rheological behavior of rocks is fundamental to the interpretation of long-term time-
dependent deformations and of fracturing processes. This work introduces the probabilistic
calibration method to account for the uncertainty resulting from the sources of evidence used to
conduct a model calibration (i.e. experimental observations, model prediction and expert’s
judgment). Conditioned on a series of triaxial compression creep experiments conducted on
sandstone, this research introduces a systematic methodology through the probabilistic
assessment of a modified Van der Poel model to simulate the viscoelastic behavior of a
sandstone specimen under a single loading condition. Results show the structural correlation
(linear and non-linear) between the model parameters, and the effects they pose on the
rheological behavior of the specimen.

INTRODUCTION

Rock rheology helps to describe the effects of a given force and temperature on
specimens showing pure elastic to fluids flow responses. Rheological properties of rocks are
greatly associated with the development of folds, faults and fractured zones, which are identified
as causes of weakening of rocks, and consequently of landslide developments (Chigira 1992).
Such fragmentation of rock masses can take place with or without a well-defined slip surface,
softening the bottom support of rocks and resulting in a long-term gravitational viscous
deformation (Deng et al. 2000).

© ASCE

Geo-Risk 2017
Geo-Risk 2017 GSP 285 70

Due to these unfavorable specimen deteriorating mechanisms, a full identification and


characterization of rock properties is recommended prior to implement risk assessment and
management engineering strategies (Chigira 1992; Comegna et al. 2013). Key elements to
develop these strategies is the integration of evidence stemmed from experimental standardized
rock mechanics testing (Jun 1993; Le et al. 2012; Li and Xia 2000; Sun and Wang 2000), and
from analytical and numerical models considering rock creep behavior (Boidy et al. 2002; Grgic
et al. 2003; Zhou et al. 2008, 2011). Based on a series of standardized triaxial tests, the main
Downloaded from ascelibrary.org by Texas A&M University on 10/30/17. Copyright ASCE. For personal use only; all rights reserved.

interest of this paper lies in a systematic quantification of uncertainty of rheological rock


mechanics parameters.
Bayesian probabilistic calibration framework has been used in many geotechnical
applications, including site investigation (Wang et al. 2016), and model uncertainty
quantification (Wang et al. 2012; Yang et al. 2010; Zhang et al. 2009). This framework allows a
systematic exploration of parametric space through a functional formulation combining the state
of knowledge of governing parameters with respect to the rheological behavior in this paper,
with the selection of model parameters that best resemble the corresponding experimental
observations. The sampling process to populate the resulting multivariate joint probability
density function called posterior distribution, is discussed as well (Esmailzadeh et al. 2015;
Rechenmacher et al. 2004). With this set up, a unique feature of the Bayesian analysis can be
observed on the assessment of the parameters correlation structure, which depicted with marginal
probability distributions of two parameters at the time (Gauer et al. 2009; Ranalli et al. 2010).
Once the posterior distribution is populated, it is then possible to generate model
realizations from the sampling combinations of the model parameters (Medina-Cetina and Arson
2014; Yang et al. 2010). Model responses are utilized to assess the accuracy of the model
predictions along the time domain and to assess the corresponding confidence levels of model
predictions. The benefits of this study will be translated in an improved understanding of
viscoelastic behavior of rock creep, and ultimately, in an improved decision-making process for
risk assessment and management.

Triaxial Compression Test of Sandstone

This work presents results of three triaxial compression tests on sandstone specimens
conducted at China University of Geosciences. These represent three confining stress conditions
at 3MPa, 5MPa and 7MPa, respectively, to capture the sandstone specimen’s short term strength
. A critical strain rate of 2 × 10 /ℎ is used to judge whether deformation has been stabilized
and ready to enter into another loading level. Applying Blotzmann Superposition, it is able to
depict the creep curves of strain against time under various confining pressures.
The Jurassic ( ) sandstone samples that used in this research are collected 40
underneath the Three Gorges Reservoir, the world's largest power station in terms of installed
capacity (Deng et al. 2000). Figure 1 depicts the stress-strain curves of sandstone samples under
the confining pressures 3 ,5 and 7 , respectively. In general, S-shaped stress-strain

© ASCE

Geo-Risk 2017
Geo-Risk 2017 GSP 285 71

curves are observed for all three tests. These show how the sandstone enters a first compaction
stage where the slope of these curves increased with the loading stress. The strain occurred in
this phase is reversible. With the increase of stress, the strain-stress curves become gradually
convex in shape. During this next compression stage, the micro-cracking inside of rock sample
grow, link, and finally lead to sample failure. At the ultimate stress limit, the peak value of
deviator stress obtained is what is defined as the triaxial compression strength ( ) of each
sample.
Downloaded from ascelibrary.org by Texas A&M University on 10/30/17. Copyright ASCE. For personal use only; all rights reserved.

Figure 1. Compression test results

Table 1 summarizes the result of triaxial compression tests of the sandstone specimens
subjected to different confining pressures. Where, ϵ represents the axial strain corresponding to
the sample strength σ ; and E is the slope of the line between the experimental starting point
and sample’s half strength 0.5σ .
Table 1. Results of compression tests
σ3(MPa) ( ) ( ) (%) ( )

3 85.6 63 1.212 6.54


5 110 74 1.208 9.27
7 160 94 1.237 13.31

Triaxial Rheological Test of Sandstone

The triaxial rheological tests of sandstone were carried out following the standard ASTM
D7070-08. Notice that although a number of studies are interested in rock rheological behavior
to associate the effect of temperature (Arabani and Kamboozia 2014; Sun and Wang 2000), this
is not addressed in this work. During the test, room temperature was maintained at 20 ± 1℃ in
order to eliminate the possible impact on sample’s rheological behaviour.
As a preliminary approach to discuss the benefits of the probabilistic calibration approach
as applied to the proposed forward model, only the tests with confining pressure at 3 is
considered in this work. Table 2 presents the loading stress levels of the triaxial rheological test
under a confining pressure of 3 .

© ASCE

Geo-Risk 2017
Geo-Risk 2017 GSP 285 72

Table 2. Stress levels of rheological test


Confining
0.50 0.55 0.60 0.65 0.70 0.75
Pressure
3Mpa 42.8 47.1 51.4 55.6 59.9 64.2

Figure 2 displays the 3 strain path with the corresponding domains of deviator stress
and time . The ultimate stress level is 0.75 which corresponds to the sample failure in the
long-term creep test. This demonstrates sandstone’s compression strength has decreased due to
Downloaded from ascelibrary.org by Texas A&M University on 10/30/17. Copyright ASCE. For personal use only; all rights reserved.

the rheological deformation.

Figure 2. Rheological test results in 3D space

Modified Van Der Poel model

Mechanical rheological models are generally formulated by a family of standard


components such as the Hooke spring or the Newtonian dashpot in its elastic or viscous response
of creep deformations respectively (Arabani and Kamboozia 2014). One of the objectives of this
work is to explore the relationship between the transient elastic modulus and the loading stress
under dynamic loading conditions. Kondner (1963) indicated that for a triaxial test, the
corresponding stress-strain relationship can be expressed by a hyperbolic equation:
− = (1)
Where , are material nonlinear constants. Eq. (1) can also be written as:
( )
= (2)
( )
Hence, the initial elasticity modulus considering the hyperbolic relationship between
stress and strain can established as:
= = = − (3)
It is shown in Eq.(3) that there is a linear relationship between initial elasticity modulus
and the stress level applied to the rock. Thus, the classic Van Der Poel model (Figure 3) can be
modified as:

© ASCE

Geo-Risk 2017
Geo-Risk 2017 GSP 285 73

= + (1 − ) (4)
Where and are material constants.

E2

σ σ
E1
Downloaded from ascelibrary.org by Texas A&M University on 10/30/17. Copyright ASCE. For personal use only; all rights reserved.

η1
Figure 3. Van Der Poel model

UQ framework

The uncertainty quantification of an inverse problem aims to identify, characterize and


simulate the various sources of uncertainty inherently participating in the physical process of
interest (Medina-Cetina 2006)). The expected output of the true process is represented by a set of
random vectors which based on the definition of the physical process at prescribed control
points. In practice, experimental observations can be retrieved from lab or field
measurments, and compared to predicting outcomes of the same process process (Medina-
cetina 2006). These are quantifiable vectors formed as a-priori information used to approximate
the true process . Accordingly, the uncertainty quantification framework in terms of ,
and can be summarized by:
= +∆
= +∆
− =∆ −∆ (5)

Eq.(5) illustrates the tradeoff between the scientific evidence and through the
gradients ∆ . The involved uncertainty incorporate random vectors both in experimental
observations and theoretical predictions. Herein, is comprised of mathematical predictions
stemming from a forward model () , which is governed or characterized by a set of parameters
. Noticed that can represent geometric and material properties, such as shape parameters or
rheological parameters in this case. Similarly, the uncertainty involved in the calibration of
vectors can be defined as:
= +∆ (6)
Where denotes mean of parameters and ∆ represent the uncertainty component. Other
than the deterministic methodology, the probabilistic calibration allows for an exhaustive
exploration of all potential combinations of the model parameters that best resemble the
experimental observations. As a result, the correlation structures of model parameters is
populated, which ultimately be translated in a better understanding of uncertainty inherent to
model nature (∆ ) with respect to the experimental observations. Noticed that in this paper,

© ASCE

Geo-Risk 2017
Geo-Risk 2017 GSP 285 74

the model prediction assumed to be unbiased with respect to process of interest, which implies
∆ −∆ = 0.

Bayesian probabilistic calibration

The proposed UQ framework requires assessment of plausible solutions of model


parameters conditioned on the data , which generates the need of a mathematical mapping
Downloaded from ascelibrary.org by Texas A&M University on 10/30/17. Copyright ASCE. For personal use only; all rights reserved.

regarding this inverse problem (Medina-Cetina 2006). From a deterministic standpoint, this can
be accomplished by selecting an ‘optimal’ set of model parameters, which maximize the
likelihood of fitting . However, the proposed calibration is an ill-posed inverse problem,
since many combinations of the model parameters can lead to the same experimental response.
To tackle this problem, the proposed probabilistic calibration method follows a Bayesian
approach, which accounts for the full probabilistic description of the model parameters through
probability maps. This starts from an expert’s belief setting up the prior ( ( )) about model
parameters before the experimental evidence is presented to the mechanical model (forward
model). This prior knowledge ideally facilitates calibration of model parameters by limiting and
defining plausible values in the form of probability distribution, which later updated
systematically via quantifying likelihood ( ( | )) between available observations and
model parameters . From the basic definition of Bayes theorem:
( | ( ), ) ( )
( | )= ∝ ( | ( ), ) ( ) (7)
( | ( ), ) ( )

The posterior ( | ) is the probability proportional to the prior ( ) and the


likelihood ( | ). This is because of the integral of denominator is a normalizing constant
over the parametric space Θ so that the integral of the posterior ( | ) can be 1.
The sampling of the posterior is based on a Markov Chain Monte Carlo approach coupled
with Metropolis Hastings criteria (MCMC- MH), which makes it possible to draw samples from
a proposing distribution to infer the target posterior distributions.

Probabilistic Calibration of Modified Van der Poel Model

The probabilistic calibration method introduced in the preceding sections is now applied
to the proposed rheological model. The confining stress tests at 3 MPa includes six loading
increments. This work will concentrate only on the first loading condition at 42.8 to
account solely for the specimen viscoelastic behavior. As illustrated in Eq. (4), the Young’s
modulus is expressed by a linear relationship between the material constants and .
Following a phenomenological logic, the calibration process starts with parameters and .
Once certain level of stress level applied to the sample, an immediate elastic strain can be
observed, which allows to compute the corresponding elastic modulus. Figure 4 presents the

© ASCE

Geo-Risk 2017
Geo-Risk 2017 GSP 285 75

algorithm of probabilistic calibration and Table 3 shows the experimental design for the
numerical calibration.
Prior π (θ 1 ) Likelihood f (d obs | g (θ1 ),θ1 )
Non-informative θ1 = {a, b} Model g (θ ) : E1 = aσ + b

Probabilistic Calinration

Posterior π ( θ 1 | d obs )
Downloaded from ascelibrary.org by Texas A&M University on 10/30/17. Copyright ASCE. For personal use only; all rights reserved.

Serving as informative prior for


next step calibration

Prior π (θ ) Likelihood f ( d obs | g (θ ), θ )


Informative θ1 = {a, b} E2
σ σ − t
Non-informative θ2 ={E2,η1} g (θ ) : ε = + (1 − e η ) 1

aσ + b E2
Parameters θ = θ1 + θ 2

Probabilistic Calinration

Posterior π (θ | d obs )

Statistical Analysis

Figure 4. Probabilistic calibration scheme

Table 3. Numerical experimental design


, 3
− , 42.8
{ , }
{ , }

Figure 5 displays the immediate elastic modulus for all loading conditions, along with
a best estimate of = + . The deterministic calibration results here are used as the starting
point for the calibration of the subsequent parameters of the model. The solution of the numerical
integration of the posterior was obtained after 1 + 6 MH-MCMC simulations where
convergence was guaranteed. The initial 20% samples are discarded as burn-in iterations, only
after which the posterior are used to make the statistical inference. Figure 6 presents the joint
relative frequency distribution of and constructed from randomly derived samples from the
target distribution. It reveals a clear negative relationship between and as suggested
theoretically in the preceding sections. The marginal distributions of and as well as their
correlation structure will be used as a-priori information in the calibration of the remaining
model parameters.

© ASCE

Geo-Risk 2017
Geo-Risk 2017 GSP 285 76
Downloaded from ascelibrary.org by Texas A&M University on 10/30/17. Copyright ASCE. For personal use only; all rights reserved.

Figure 5. Young’s modulus and linear fit

Figure 6. Joint relative frequency histogram (JRFH) of and

Figure 7 presents the probabilistic calibration result containing the parameter set
{ , , , }. A major benefit of the probabilistic calibration lies in the ability to retrieve the
correlation structure defining the degree of association between the parameter of interest (linear
and non-linear), which is technically unable to obtain by typical deterministic calibration (e.g.
optimization-based). The off-diagonal joint probability maps are comprised of all combinations
of the parameters taken two at a time. The marginal posterior distributions are presented on the
diagonal plots. Notice that in order to accommodate the variability of all parameters on this set,
the mean and standard deviation of { , } have shifted. This finding is in accordance with the
philosophy of Bayesian updating. Similarly, significant less variability is observed for
parameters and due to the effect of the informative prior introduced in the calibration. Apart
from this distinct negative correlation, it is observed significant correlation between and

© ASCE

Geo-Risk 2017
Geo-Risk 2017 GSP 285 77

parameters { , , }, indicating that under the first loading stress = 42.8 , the dynamic
viscosity coefficient in the Kelvin model is in fact related with the other components in the
viscoelastic model.
Downloaded from ascelibrary.org by Texas A&M University on 10/30/17. Copyright ASCE. For personal use only; all rights reserved.

Figure 7. Joint relative frequency histogram of model parameters

Table 4. Statistics of the probabilistic calibration


Parameters ( ) ( ∙ )
Mean ( ) 104.37 3725.07 9.78e + 04 1.20e + 07
SD( ) 0.32 14.10 213.69 7.51e + 04
Coefficient of 0.003 0.004 0.002 0.006
variation

Table 5. Correlation matrix of model parameters by probabilistic calibration


Parameters
1
−0.996 1
0.041 −0.057 1
0.227 −0.300 −0.218 1

Once the posterior probability map is generated, likely realizations of the rheological
model predictions can be computed by sampling random parameters combinations from the
MCMC. These model responses provide numerical evidence to estimate first and second order
statistics. Figure 8 presents the first order statistics of the model performance for the single load
case discussed here. This figure shows that despite the overall good fit of model and data, a
higher deviation between model predictions and data observation is found around the initial and
the final loading stages. This shows that the transitions between initial and final loading stages
with the loading steady-state stage, is a nonlinear process that is not fully captured by the
forward model. The w-shaped curve of standard deviation of model predictions illustrate that

© ASCE

Geo-Risk 2017
Geo-Risk 2017 GSP 285 78

model is more uncertain about its predictions at the initial and end area of the time domain, yet
its overall order of magnitude is rather small.
Downloaded from ascelibrary.org by Texas A&M University on 10/30/17. Copyright ASCE. For personal use only; all rights reserved.

(a)

(b) (c)
Figure 8. a) Model realizations; b) mean of model predictions; c) standard deviation of
model predictions

CONCLUSION

A Bayesian probabilistic calibration was conducted to investigate the effects of the


uncertainty of a modified Van Der Poel model, when presented to experimental observations of a
single loading cycle of a sandstone specimen ( = 42.8 at 3 MPa confining stress), taken
at 40 underneath the Three Gorges Reservoir. The main objective of this work was to identify

© ASCE

Geo-Risk 2017
Geo-Risk 2017 GSP 285 79

the uncertainty and correlation structure of the viscoelastic parameters capturing the sandstone
specimen behavior.
From the statistics of the calibration of the model parameters, a significant correlation
was observed between and all other model parameters{ , , }. Furthermore, by assessing
model responses and its statistics, the model has been validated with its ability to capture the
trend of rheological behavior, with higher prediction uncertainty observed at the start and at the
end of the loading cycle.
Downloaded from ascelibrary.org by Texas A&M University on 10/30/17. Copyright ASCE. For personal use only; all rights reserved.

This research provides a full probabilistic calibration of proposed models under a specific
confining pressure, further investigation can be developed based on the influence of various
confining pressures, temperatures, and degrees of saturation or sample locations (i.e. spatial
features of the sandstone). It is also of interest to consider assessing the confidence of competing
forward models when presented the same dataset.

REFERENCES

Arabani, M., and Kamboozia, N. (2014). “New achievements in visco-elastoplastic constitutive


model and temperature sensitivity of glasphalt.” International Journal of Pavement
Engineering, 15(9), 810–830.
Boidy, E., Bouvard, A., and Pellet, F. (2002). “Back analysis of time-dependent behaviour of a
test gallery in claystone.” Tunnelling and Underground Space Technology, 17(4), 415–424.
Chigira, M. (1992). “Long-term gravitational deformation of rocks by mass rock creep.”
Engineering Geology, 32(3), 157–184.
Comegna, L., Picarelli, L., Bucchignani, E., and Mercogliano, P. (2013). “Potential effects of
incoming climate changes on the behaviour of slow active landslides in clay.” Landslides,
10(4), 373–391.
Deng, Q. L., Zhu, Z. Y., Cui, Z. Q., and Wang, X. P. (2000). “Mass rock creep and landsliding
on the Huangtupo slope in the reservoir area of the Three Gorges Project, Yangtze River,
China.” Engineering Geology, 58(1), 67–83.
Esmailzadeh, S., Medina-Cetina, Z., Kang, J. W., and Kallivokas, L. F. (2015). “Varying
dimensional Bayesian acoustic waveform inversion for 1D semi-infinite heterogeneous
media.” Probabilistic Engineering Mechanics, Elsevier, 39, 56–68.
Gauer, P., Medina-Cetina, Z., Lied, K., and Kristensen, K. (2009). “Optimization and
probabilistic calibration of avalanche block models.” Cold Regions Science and
Technology, Elsevier B.V., 59(2-3), 251–258.
Grgic, D., Homand, F., and Hoxha, D. (2003). “A short- and long-term rheological model to
understand the collapses of iron mines in Lorraine, France.” Computers and Geotechnics,
30(7), 557–570.
Jun, L. J. S. (1993). “On mesocrack damage of brittle rocks and its time-dependent
characteristics.” Chinese Journal of Rock Mechanics and Engineering, (4), 001.
Kondner, R. L. (1963). “Hyperbolic stress-strainresponse: Cohesive soils.” Journalof Soil
Mechanics and Foundations Division, ASCE, 89, 115–143.
Le, T. M., Fatahi, B., and Khabbaz, H. (2012). “Viscous Behaviour of Soft Clay and Inducing
Factors.” Geotechnical and Geological Engineering, 30(5), 1069–1083.

© ASCE

Geo-Risk 2017
Geo-Risk 2017 GSP 285 80

Li, Y., and Xia, C. (2000). “Time-dependent tests on intact rocks in uniaxial compression.”
International Journal of Rock Mechanics and Mining Sciences, 37(3), 467–475.
Medina-Cetina, Z. (2006). “Probabilistic calibration of soil constitutive models.” Ph.D
dissertation, Johns Hopkins Univ., Baltimore
Medina-Cetina, Z., and Arson, C. (2014). “Probabilistic calibration of a damage rock mechanics
model.” Géotechnique Letters, 4(1), 17–21.
Ranalli, M., Gottardi, G., Medina-Cetina, Z., and Nadim, F. (2010). “Uncertainty quantification
in the calibration of a dynamic viscoplastic model of slow slope movements.” Landslides,
Downloaded from ascelibrary.org by Texas A&M University on 10/30/17. Copyright ASCE. For personal use only; all rights reserved.

7(1), 31–41.
Ranalli, M., Medina-Cetina, Z., Gottardi, G., and Nadim, F. (2014). “Probabilistic Calibration of
a Dynamic Model for Predicting Rainfall-Controlled Landslides.” Journal of Geotechnical
and Geoenvironmental Engineering, 140(4), 04013039.
Rechenmacher, A. L., Medina-Cetina, Z., and Ghanem, R. G. (2004). “Calibration of soil
constitutive models with heterogeneous parameters.” 9th ASCE Specialty Conference on
Probabilistic Mechanics and Structural Reliability, 133(December), 1567–1576.
Sun, J., and Wang, S. (2000). “Rock mechanics and rock engineering in China: developments
and current state-of-the-art.” International Journal of Rock Mechanics and Mining Sciences,
37(3), 447–465.
Wang, L., Ravichandran, N., and Juang, C. H. (2012). “Computers and Geotechnics Bayesian
updating of KJHH model for prediction of maximum ground settlement in braced
excavations using centrifuge data.” Computers and Geotechnics, Elsevier Ltd, 44, 1–8.
Wang, Y., Cao, Z., and Li, D. (2016). “Bayesian perspective on geotechnical variability and site
characterization.” Engineering Geology, Elsevier B.V., 203, 117–125.
Yang, S.-L., Medina-Cetina, Z., and Nadim, F. (2010). “Uncertainty analysis on remoulded
undrained shear strength of marine clay.” Georisk: Assessment and Management of Risk for
Engineered Systems and Geohazards, 4(1), 43–50.
Zhang, J., Zhang, L. M., and Tang, W. H. (2009). “Bayesian Framework for Characterizing
Geotechnical Model Uncertainty.” Journal of Geotechnical and Geoenvironmental
Engineering, 135(7), 932–940.
Zhou, H., Jia, Y., and Shao, J. F. (2008). “A unified elastic-plastic and viscoplastic damage
model for quasi-brittle rocks.” International Journal of Rock Mechanics and Mining
Sciences, 45(8), 1237–1251.
Zhou, H. W., Wang, C. P., Han, B. B., and Duan, Z. Q. (2011). “A creep constitutive model for
salt rock based on fractional derivatives.” International Journal of Rock Mechanics and
Mining Sciences, Elsevier, 48(1), 116–121.

© ASCE

Geo-Risk 2017

Вам также может понравиться