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

net/publication/284047437

Convergence-confinement method for simulating NATM tunnels evaluated by

comparison with full 3D simulations

Article · January 2010

CITATIONS READS

3 346

2 authors, including:

David Mašín
Charles University in Prague
116 PUBLICATIONS   1,554 CITATIONS   

SEE PROFILE

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

Effects of consolidation under a penetrating footing in carbonate silty clay View project

soilmodels.com View project

All content following this page was uploaded by David Mašín on 23 November 2015.

The user has requested enhancement of the downloaded file.

 Convergence-confinement method for simulating NATM
tunnels evaluated by comparison with full 3D simulations

T. Svoboda & D. Mašín

Charles University in Prague, Faculty of Science, Prague, Czech Republic

The convergence-confinement method (CCM) for simulating NATM tunnels using plane strain finite element method was
in the paper evaluated by comparison with fully 3D simulations. Three real shallow tunnels in urban environment in
different stiff clays were simulated. The soil behaviour was described by an advanced non-linear soil constitutive
model based on the hypoplasticity theory. It was shown that for an optimum value of the parameter λd the displacement
field predicted by the CCM method agrees well with the 3D simulations. The controlling parameter was, however, found
to be dependent on the problem simulated (for the same material) and also on the material properties (for the same
tunneling problem).

1. INTRODUCTION application to tunnel lining design. This method is

3D numerical analysis becomes an increasingly affordable nowadays commonly used to predict also displacement
tool for predicting deformations and stress redistribution field due to NATM tunnelling. Panet and Guénot (1982)
induced by tunnelling. As tunnel excavation is clearly a demonstrated that the 3D ground response to tunneling
three dimensional problem, considering the third could be analysed with a plane strain approach, provided
dimension should intuitively lead to more accurate a fictitious pressure σrf was introduced inside the tunnel
predictions. It might therefore be surprising that simplified area in the 2D model. This pressure could be derived from
procedures that allow us to consider 3D effects within a the initial stress in the ground σrO from
simplified 2D plane strain analysis are still popular in σ r
f
= (1 − λ )σ 0
r (1)
geotechnical design. This is because the 3D simulations
require accurate description of such aspects as the where λ is the stress release coefficient. Its value
excavation sequence, lining installation procedure, or corresponding to the moment of lining instalation is
time-dependent behavior of shotcrete, which may be denoted as λd . The method can be properly evaluated by
difficult to incorporate in a numerical model accurately means of comparison of 2D with fully 3D simulations.
and they might not be feasible at preliminary design stages Such comparison, however, are scarce in the technical
or for less demanding tunneling problems. 2D methods literature. The aim of this paper is to investigace:
typically require specification of just a few or only one • whether λd depends on the problem geometry, i.e.
parameter (denoted in the following as λd), which whether the same λd can be used for predictions of
integrates the influence of all of the aforementioned different tunnels in the same material.
factors that need to be considered in 3D analyses. • to what extent λd depends on material properties, i.e.
Calibrating λd based on monitoring results is popular as it whether the same λd can be used for predictions of a
is often possible to tune the model using a single empirical tunnel advancing through different geological materials.
parameter such that some monitored aspect is correctly
reproduced. Disadvantage of this approach is that it 3. 3D SIMULATIONS OF CASE STUDIES
hinders eventual inaccuracies in description of the ANALYSED
mechanical behaviour of the rock massif. This approach 3D finite element models of three case histories will be
thus does not provide any information on the suitability of used as a basis for the evaluation of the CCM method. All
the 2D method itself to reproduce the 3D effects. The only of the cases represent shallow NATM tunnels excavated
way to proper evaluation of the 2D method thus lies in in stiff clays in urban environment. Results of the 3D
comparison of its predictions with predictions by analyses were thoroughly checked with respect to
equivalent fully 3D analyses. monitoring data to demonstrate realistic representation of
the three dimensional effects. Soil properties were de-
2. STRESS-RELEASE (CONVERGENCE- scribed by means of an advanced nonlinear constitutive
CONFINEMENT) METHOD model for fine-grained soils, a hypoplastic constitutive
At present, a number of 2D methods that can be used to model for clays (Mašín 2005) enhanced by the
account for 3D effects within finite element analysis intergranular strain concept (Niemunis and Herle 1997).
framework is available. Probably the most popular This model has been shown to represent accurately the soil
method is the convergence-confinement method (CCM), behaviour from the very small strain range to large strain
introduced by Panet and Guénot (1982) with primary range, including high quasi-elastic stiffness in the very
1
small strain range and its non-linear decrease with
increasing strain level. Finite element simulations were
performed using software Tochnog Professional
(Rodemann 2008). In all cases, undrained analyses were
performed with reduced value of water bulk modulus Kw
to account for consolidation effects. Realistic excavation
sequence as applied in the field was simulated. Lining
behaviour was described by a linear elastic model with
time-dependent Young modulus following an exponential
expression by Pottler (1990). Continuum elements, rather
than shell elements, were used to simulate tunnel lining.
For a detailed description of the analysis procedures see
Mašín (2009).
3.1.Heathrow express trial tunnel
Heathrow express trial tunnel (Deane and Basset 1995), a (b)
NATM tunnel built in London Clay to test effectiveness of Figure 2: Stiffness degradation curves simulated by the
the shotcrete lining method, has since become a classical hypoplastic model with different parameters (from Mašín,
example for evaluation of different numerical tools. In this 2009). Experimental data on natural samples of London Clay
work, 3D analyses described in detail by Mašín (2009) from Gasparre (2005).
will be used as a basis for evaluation of the CCM method.
Hypoplastic constitutive model was calibrated using high Mašín (2009) performed a parametric study aimed at
quality experimental data on London Clay by Gasparre clarification of the influence of material parameters on the
(2005). All the simulations were performed with a predicted results. Figure 2 shows shear stiffness
constitute model calibrated solely on the basis of degradation curves for undrained shear triaxial tests on
laboratory experiments, without tuning material natural samples of London Clay simulated using the
parameters to obtain monitored deformations. High K0 hypoplastic model with different values of material
values varying with depth, as measured in situ by Hight et parameters. As is clear from Figure 2, the parameter mR
al. (2007), were considered. Finite element mesh and the controls the initial stiffness in the very small strain range,
modelled geometry is shown in Figure 1. while the parameter r controls the large strain stiffness
(constant initial stiffness was imposed in the analyses
while varying parameter r). Values mR = 9 and r = 0.5
represent the best the experimental data. Surface
settlement troughs predicted by the 3D model are shown in
Figure 3. With parameters calibrated solely on the basis of
laboratory experiments, the model is capable of
reproducing the settlement magnitude, while it slightly
overestimates the settlement trough width. Although the
influence of the large strain stiffness (parameter r)
appears to be insignificant in Fig. 2b, it has a substantial
Figure 1: Finite element mesh used in the analyses of the effect on the predicted settlement magnitude (Fig. 3b).
Heathrow express trial tunnel (from Mašín, 2009).

(a)
(a)

2
 assumes that overconsolidation is caused by creep
phenomena with K0 = 0.66 obtained from Jáky (1944)
formula. Surface settlement troughs predicted by the
hypoplastic model for the two K0 values are shown in Fig.
5a.

(b)
Figure 3: The influence of the small-strain stiffness
characteristics on the predicted settlement trough for the
Heathrow express trial tunnel (from Mašín, 2009; monitoring
data from Deane and Basset 1995).

3.2. Dobrovskeho exploratory adit (a)

The second case study analysed is an exploratory adit of
the Dobrovskeho tunnel, which is being excavated in
Brno, Czech Republic. These tunnels form the northern
part of the large city ring road. The tunnels consist of two
oval tunnel tubes with lengths 1.2 km with height of about
12 m, a section width of about 14 m. Both the tunnels are
led parallel at a distance of 70 m and are being excavated
by the NATM with vertical face sequence subdivided into
6 segments. For exploration purposes, three adits were
excavated. The exploratory adits had approximately
triangular cross sections with side length 5 m and were
situated in the tunnel top headings. The subsoil in which
the tunnels are excavated consists of Miocene limy, silty
stiff clay (Brno Clay). Full 3D numerical model of the
exploratory adit has been developed by Svoboda and (b)
Mašín (2009). The parameters of the hypoplastic model Figure 5: Surface settlement trough due to exploratory adit of
for clays were calibrated on the basis of quality laboratory Dobrovskeho tunnel predicted by the 3D analysis for two
experiments that included measurements of small strain different K0 values (a) and for different model parameters with
stiffness characteristics using local LVDT strain K0 = 1.25 (b).
transducers and bender elements. FE mesh and the model
geometry are shown in Figure 4. Analysis with K0 = 0.66 represents the monitored
behaviour better, but in general the K0 value does not have
a substantial effect on the settlement trough. Again,
realistic predictions were obtained with the constitutive
model calibrated solely on the basis of laboratory
experimental data. Additional analyses were performed
with variable parameters r (large strain stiffness) and mR
(small strain stiffness) for K0 = 1.25. The influence of
these characteristics on the small strain stiffness is similar
to the one from Figure 2. Figure 5b shows that both small
and large strain stiffness influence the settlement trough
Figure 4: Finite element mesh used in the analyses of the ex- predictions.
ploratory adit of Dobrovskeho tunnel (from Svoboda and
Mašín, 2009) 3.3. Dobrovskeho tunnel
In addition to the simulations of the exploratory adit of
As no measurements of the coefficient of earth pressure at Dobrovskeho tunnel, full 3D model of the whole tunnel
rest K0 have been performed on the site, simulations were has been created. The model considered rather complex
performed with two different extreme values of K0. One excavation sequence followed at the site, with the tunnel
considers the apparent overconsolidation of the soil face subdivided into 6 segments. Two cases were
deposit caused by mechanical unloading. K0 is then considered. One with original model parameters calibrated
calculated using an approach proposed by Mayne and on the basis of laboratory experiments, the second with
Kulhawy (1982) leading to K0 = 1.25. The second parameters optimised based on monitoring results from
3
exploratory adit (see Svoboda and Mašín 2008 and (Poeter and Hill 1998). For other applications of this
Svoboda and Mašín 2009 for details). Only simulations software in geotechnical engineering optimisation
with the original parameter set are considered in this problems see Finno and Calvello (2005). The following
paper, so that direct comparison with simulations of the procedure was applied. The vertical surface displacements
exploratory adit is possible. Simulations represent „class computed by the CCM method in different distances from
A“ predictions of deformations due to the tunnel, as the the tunnel axis (approx. 20 locations) were used to
tunnel has not been built by the time the authors were assemble a „simulation vector“ y´, whereas displacements
performing the simulations. Geometry of the 3D model obtained by the 3D analyses formed an „observation
during simulation of the complex excavation sequence is vector“ y. The difference was quantified by means of a
in Fig. 6a, predicted surface settlement troughs compared weighted least-squares objective function S(b) expressed
with the monitoring data for the two K0 states in Fig. 6b. as
[
S ( b ) = [ y − y´( b ) ] w y − y ´ ( b )
T
] (2)

where b is a vector containing values of parameters to be

estimated (in our case a single parameter λd) and w is a
weight matrix, considered as a unity matrix for simplicity.
Minimisation of the objective function S(b) was
accomplished by UCODE with the modified Gauss-
Newton method.

5. SUMMARY OF THE CCM ANALYSES AND

DISCUSSION OF RESULTS
Altogether 12 simulations have been performed and
evaluated. They correspond to the 3D simulations
described in Sec. 3. Results are summarised in Tab. 1,
which gives values of the CCM parameter λd
(a) corresponding to the minimum value of the objective
function S(b). In addition, relation between very small
strain and large strain shear moduli used in simulations
(G0(simul.) and Gls (simul.) respectively) and their original
values calibrated using laboratory experiments (G0 and Gls
respectively) is indicated. The table also indicates the
initial K0 conditions.
5.1. Dependency of λd on different studied factors
The following observations may be summarised based on
results presented in Tab. 1. λd depends on the assumed
material parameters, i.e. on the soil type. The very small
strain shear modulus G0 influences λd remarkably.
Interestingly, λd does not appear to be influenced
significantly by the soil behaviour in the large strain rage.
Varying the very small strain shear modulus G0 imposed
(b) changes of λd of the order of 0.1 in comparison with the
Figure 6: 3D model of the Dobrovskeho tunnel (a) and surface original values, while varying Gls had only slight effect on
settlement troughs predicted for two different K0 values (b)
compared with monitoring data (tunnel chainage).
λd of the order of 0.03 at maximum. This result might
appear surprising, as both G0 and Gls were shown to have
substantial effect on the predicted displacements, both for
4. 2D ANALYSES BY THE CONVERGENCE- the Heathrow express trial tunnel (Fig. 3) and
CONFINEMENT METHOD Dobrovskeho exploratory adit (Fig. 5).
To study the applicability of the CCM method, 2D
equivalents of all the 3D models presented have been Table 1: Summary of the CCM simulations
prepared. Basic version of the CCM was adopted, i.e.
time dependency of the lining stiffness, which has been
considered in the 3D analyses, was not modelled. In 2D,
truss-beam elements were used to represent the lining,
whereas in 3D, lining was modelled using continuum
elements. The CCM controlling parameter λd was
calibrated to ensure that the 2D and 3D analyses predicted
as closely as possible the surface settlement troughs (the
overall displacement fields were studied subsequently). In
order to prevent subjectivity of the λd determination, it was
calibrated using a software tool specifically devised for
optimisation analyses and inverse modelling UCODE

4
One of the consequences of this observation is that a axis. This is caused by high K0 conditions adopted in the
change of geological conditions during excavation of a Dobrovskeho case study analyses presented in this section
single tunnel might require appropriate modification of λd (K0 = 1.25). In the 3D analyses, this effect is not that
values used in the simulations. K0 does not appear to have significant and the method predicts more reasonable shape
substantial effect on λd. For the same soil type, the tunnel of the vertical displacement field in a close vicinity of the
size and geometry influences significantly appropriate adit.
values of λd . In the case of the Dobrovskeho study, λd ≈
0.5 was found for the exploratory adit, whereas λd ≈ 0.3
for the whole tunnel. Thus, if λd found on the basis of
results of an exploratory adit simulations was used for
predictions of the full tunnel response, it would lead to an
overestimation of the tunnel deformations. This is
demonstrated in Fig. 7, in which the Dobrovskeho tunnel
simulations are repeated with λd calibrated based on
simulations of exploratory adit. These simulations lead to
approximately 35% larger surface settlements.

(a)

Figure 7: The influence of λd on the CCM predictions of

Dobrovskeho tunnel.

5.2. Accuracy of the CCM predictions

Accuracy of the CCM predictions was studied on the
basis of analyses of the three case histories with the (b)
original parameter values (and K0 = 1.25 for Dobrovskeho
case studies). Figure 8 shows predictions by the CCM
method and full 3D method for the three case studies
analysed. Figure 8a gives surface settlement troughs, Fig.
8b shows horizontal displacements from an inclinometer
located approximately 1D from the tunnel boundary, and
Fig. 8c gives vertical displacements from an extensometer
located above the tunnel axis. The surface settlement
troughs by the 2D and 3D methods match very well. An
overall agreement could have been expected, as λd was
calibrated with the intention to match the surface
settlement trough as accurately as possible, but it is
interesting to observe that also the settlement trough shape
is predicted accurately by the 2D method. Plots of
variation of vertical and horizontal displacements with
depth show good agreement for the Heathrow express (c)
case. For both Dobrovskeho adit and Dobrovskeho tunnel, Figure 8: Comparison of surface settlement troughs (a),
the 2D method underestimates the displacements within a inclinometer results (b) and extensometr results (c) predicted
distance of approximately 1 tunnel diameter from a tunnel. by the 3D and 2D methods for the three case studies analysed.
The predictions match well outside this region. The
overall field of vertical displacements is shown in Fig. 9.
An overall agreement of 2D and 3D method is good. The
most notable discrepancy is in the predictions of the
exploratory adit. The 2D method predicts higher vertical
displacements above the sides of the adit then above its

5
 london clay from the Terminal 5 site at Heathrow airport.
Géotechnique, 57(1): 3–18.
Jáky, J. 1944. The coefficient of earth pressure at rest. Journal
for Society of Hung. Architects and Engineers,: 355–357.
Mašín, D. 2005. A hypoplastic constitutive model for clays.
International Journal for Numerice and Analytical
Methods in Geomechanics, 29(4): 311–336.
Mašín, D. 2009. 3D modelling of a NATM tunnel in highK0
clay using two different constitutive models. Journal of
Geotechnical and Geoenvironmental Engineering ASCE,
135(9): 1326–1335.
Mayne, P. W. and Kulhawy, F. H. 1982. K0–OCR
relationships in soil. In Proc. ASCE J. Geotech. Eng. Div.,
Figure 9: Qualitative comparison of vertical displacement field Volume 108, pp. 851–872.
predicted by the 3D and 2D methods for the three case studies Niemunis, A. and Herle, I. 1997. Hypoplastic model for
analysed (the same color scale for corresponding 2D and 3D cohesionless soils with elastic strain range. Mechanics of
analyses). Cohesive-Frictional Materials, 2: 279–299.
Panet, M. and Guénot, A. 1982. Analysis of convergence
behind face of a tunnel. In Proc. Tunnelling’82, pp. 197–
6. CONCLUDING REMARKS 203. IMM, London.
The 2D convergence-confinement method for simulating Poeter, E. P. and Hill, M. C. 1998. Documentation of
NATM tunnels using plane strain finite element method UCODE, a computer code for universal inverse modelling.
Technical report, US Geological Survey Water Resources
was in the paper evaluated by comparison with fully 3D
Investigations Report 98-4080.
simulations of three different case histories. It was shown Pottler, R. 1990. Time-dependent rock-shotcrete interaction. A
that for an optimum value of the CCM parameter λd the numerical shortcut. Computers and Geotechnics, 9: 149–
displacement field predicted by the CCM method agrees 169.
well with the 3D simulations. In some cases only, a Rodemann, D. 2008. Tochnog Professional user’s manual.
discrepancy was observed in a close vicinity of the tunnel. http://www.feat.nl.
The parameter λd was found to be dependent on the Svoboda, T. and Mašín D. 2008. Impact of a constitutive
problem simulated (for the same material) and also on the model on inverse analysis of a NATM tunnel in stiff clays.
material properties (for the same tunneling problem). In V. K. Kajnlia et al. Proc. ITA-AITES World Tunnel
Considering material properties, the very small strain Congress, Agra, India, Volume 2, pp. 627–636. Springer,
shear modulus was found to be more influential on the λd Berlin.
Svoboda, T. and Mašín, D. 2009. Optimisation of parameters
value than the large strain shear modulus. The initial K0
for simulating a NATM tunnel in stiff clays based on a 3D
stress state was not found to influence λd substantially. model of exploratory adit. In G. Meschke, G. Beer, J.
Eberhardsteiner D. Hartmann, and M. Thewes (Eds.),
ACKNOWLEDGEMENT Proc. 2nd International Conference on Computational
The authors acknowledge the financial support by the Methods in Tunnelling EURO:TUN 2009, Volume 1, pp.
research grants GACR 205/08/0732, GAUK 134907 and 249–256.
MSM0021620855.

REFERENCES

Deane, A. P. and Basset, R. H. 1995. The Heathrow express

trial tunnel. Proc. Instn. Civil Engineers, 113: 144–156.
Finno, R. J. and Calvello, M. 2005. Supported excavations:
Observational method and inverse modelling. Journal of
Geotechnical and Geoenvironmental Engineering ASCE,
131(7): 826–836.
Gasparre, A. 2005. Advanced laboratory characterisation of
London Clay. Ph. D. thesis, University of London,
Imperial College of Science, Technology and Medicine.
Hight, D. W., Gasparre, A., Nishimura, S., Minh, N. A.,
Jardine, R. J., and Coop, M. R. 2007. Characteristics of the

6
View publication stats