You are on page 1of 6

NATM-tunnelling in softening stiff clays and weak rocks

S.C. Möller & P. A. Vermeer


Institute of Geotechnical Engineering, University of Stuttgart, Germany
T. Marcher
ILF Consulting Engineers ZT GmbH, Innsbruck, Austria

ABSTRACT: This numerical study concerns the mechanical behaviour of tunnels in non-structured stiff clays
and weak rocks. A constitutive model will be presented, which accounts for isotropic hardening as well as for
softening. Attention will be focused on cohesion softening rather than friction softening. The hardening-
softening model is formulated by adopting a non-local continuum, as softening behaviour yields ill-posed
problems when formulated within classical continuum mechanics. For computing ground response curves,
numerical analyses are carried out for a very shallow and a medium deep NATM tunnel. It is shown that such
situations differ from very deep tunnels in the sense that material softening can produce progressive failure,
as demonstrated by a trough-like ground response curve.

1 INTRODUCTION the framework of a monitoring programme, whereas


more brittle cohesion softening may lead to a more
Stiff clays and weak rocks tend to show a peak sudden loss of stability. It is thus logic to concen-
strength and a much lower residual strength. The trate first of all on cohesion softening, at least when
transition from the peak to the residual strength is considering effects in tunnelling.
referred to as softening and it usual occurs in com-
bination with the localization of deformations in The constitutive model will be formulated within the
shear bands, i.e. thin zones of intensively shearing framework of rate-independent plasticity with iso-
material. Softening shear bands imply a reduction of tropic hardening and softening. A particular idea in
shear stresses both inside and outside the band; oth- which friction hardening is combined with softening
erwise there would be no equilibrium. In adjacent in the cohesion is elaborate. This idea seems promis-
regions outside the band the shear stress reduction ing, as it yields brittle behaviour at low stress levels
causes a quasi-elastic unloading so that one observes and increasing ductility with higher stress levels.
more or less rigid block movements. Moreover in this paper attention will be focussed on
isotropic ground without a macro structure due to
The softening related effect of progressive failure is stratification, schistarity or jointing.
well-known for clay slopes, e.g. by the papers of
Skempton (1964), Chen et al. (1992) and Potts et al. The hardening-softening model is formulated by
(1997). In tunnelling, softening may result in a con- adopting a non-local (De Borst et al., 1993) contin-
cave ground-response curve as considered e.g. by uum, as softening behaviour yields ill-posed prob-
Pacher (1964) and more recently by Vavrovsky lems when formulated within classical continuum
(1994). mechanics. A relatively simple method is obtained
by using a single non-local parameter, namely the
In stiff clays softening occurs both for the friction softening parameter. It will be shown that this leads
angle and the cohesion, but friction softening is not to mesh-independent finite-element analyses.
as dangerous as cohesion softening. This relates to
the fact that bonds between particles conferring ef- Results of numerical analyses will be shown for
fective cohesion are destroyed after small deforma- NATM-tunnelling. Here the so-called ground re-
tions, finally resulting in zero cohesion. In contrast, sponse curve, also referred to as Fenner-Pacher
friction angles tend to drop much more slowly down curve, is computed for two different tunnels. The
to a residual value often well above zero. A gradual hypothesis by Pacher (1964) concerning a trough-
loss of stability due to ductile friction softening is shaped ground response curve in softening ground is
thus less severe and more readily observed within to be investigated.
2 CONSTITUTIVE MODEL WITH COHESION For triaxial conditions with σ1=σ2, Equation 4 re-
SOFTENING duces to Equation 3b, but surfaces of constant mobi-
lized φm-values are smooth as indicated in Figure 1.
According to Equation 2, φm increases as a function
2.1 Plastic shear strain in primary loading
of γp and approaches the failure value φa. The as-
One of the ideas for the formulation of the model is ymptotic surface for φm= φa is a failure surface as
the hyperbolic relationship between the shear strain proposed by Matsuoka & Nakai (1982).
γ and the deviatoric stress q=σ1′-σ3′ in a standard
drained triaxial test. For primary loading, we adopt
the hyperbolic equation
2.2 Yield function for friction hardening
A proper elastoplastic model is obtained by convert-
γ p ≡ ε1p − ε 3p = γ 50p ⋅ q /(qa − q ) (1) ing Equation 4 into the yield function
where γ50p is the plastic shear strain for q=qa/2 and I1a ⋅ I 2a 9 − sin 2 ϕ m
qa is the asymptotical failure stress. The above equa- f = − (7)
I 3a 1 − sin 2 ϕ m
tion may also be written as
For plastic yielding with f=0, this equation reduces
γ p = α ⋅ sin ϕ m /(sin ϕ a − sin ϕ m ) (2) again to Equation 4. Stress-hardening is included as
where φa is the asymptotic friction angle and φm relates to γp. Indeed, Equation 2 may be inverted
to obtain
α = γ 50p ⋅ (1 − sin ϕ a ) (3a)
sin ϕ m = sin ϕ a ⋅ γ p /(α + γ p ) (8a)
sin ϕ m = q /(σ 1′ + σ 3′ + 2a) , a = c′ ⋅ cot ϕ a . (3b) with γp=∫ γ& p dt and
On using the mobilized friction angle φm instead of 1
γ& p = ⋅ (ε&1p − ε&2p ) 2 + (ε&2p − ε&3p ) 2 + (ε&3p − ε&1p ) 2 (8b)
the ratio q/qa as done in Equation 1, one obtains a 2
shear strain formula that can be extended to general
three dimensional states of stress and strain. Indeed, Equations 7 and 8 define an isotropic yield function
the formula implies Mohr-Coulomb type shear strain with the plastic shear strain as a hardening parame-
contours in principal stress space. As test data sug- ter. Important input parameters are used in the ex-
gest smooth cones rather than irregular hexagonal pression
cones, it is appropriate to define 1− m
 p′ + a 
9 ⋅ I 3a − I1a ⋅ I 2a α = (1 − sin ϕ a ) ⋅ γ p
50 = (1 − sin ϕ a ) ⋅ γ po
50 ⋅   (9)
sin 2ϕ m ≡ (4)  p0 + a 
I 3a − I1a ⋅ I 2a
where p0 is a fixed reference stress of 100 kPa, p′ is
where the effective mean stress and γ50p as well as m and φa
are material constants. For m=1 successive yield loci
I1a = σ 1a + σ 2a + σ 3a , I 3a = σ 1a ⋅ σ 2a ⋅ σ 3a ,
in p-q-plane are straight lines, but a curvature is ob-
I 2a = − σ 1a ⋅ σ 2a − σ 2a ⋅ σ 3a − σ 3a ⋅ σ 1a , (5) tained for m<1. In addition there is a curvature due
to the definition of the cohesion c´.
and
σ ia = σ i′ + a , a = c′ ⋅ cot ϕ ′ (6) 2.3 Extension to cohesion softening
The hardening function can easily be extended to in-
clude both friction softening (Marcher & Vermeer,
2001) and cohesion softening. To this end both φa
and c′ have to be related to a distortion measure γ*.
In the present paper, however, φa is taken constant
and only c′ is linked to γ* by the relation
c′ = c peak − hc ⋅ γ * (10)
where hc is a softening modulus. From a view point
of consistency it would be logical to take the (local)
plastic shear strain γp, but this (local) softening does
not allow for stable numerical procedures and mesh-
independent results. Hence, the model has to be
Figure 1. Generalized definition of mobilized friction using regularised by introducing either a polar continuum,
stress invariants. second-order gradients of strain or non-local strains.
In coarse-grained materials, particle rotations may
occur and one might adopt a polar continuum. In this
paper, however, the focus is on fine-grained materi-
als and we adopt the non-local plastic shear strain
1
γ * = ∫ γ& * dt , γ& * = ⋅ ∫ w(r ) ⋅ γ& p dV (11)
Vw
with
2
1 r Figure 2. Typical curve of a drained triaxial compression test.
w(r ) =
l⋅ π
⋅ exp−   ,
l
Vw = ∫ w dV (12)
denoted as νur. The cohesion dependent stress a and
where r is the distance to the material point consid- the input parameter m have already been used to de-
ered and l is an internal length. Marcher (2003) fine the yield function f. According to Equation 9 the
found a strong coupling between the internal length l yield function also involves a plastic shear strain as
and the thickness ts of shear bands in finite element an input parameter. As engineers have more affinity
analysis, namely such that ts≈2l. Hence the thickness with stiffness moduli than with magnitudes of strain,
of numerical shear bands is fixed by the choice of we consider stiffness moduli as more suitable input
the internal length. parameters. For this reason, we consider the secant
modulus E50, as indicated in Figure 2. This modulus
relates to the elastic and the plastic strain in a stan-
2.4 Completion of the model dard drained triaxial test. From the yield function for
In this section the model will be completed by add- the plastic strains and the formulation of elastic
ing a plastic potential function g. Adopting a non- strains, it can be deduced that
associated model the plastic potential differs from m
the yield function, but it is formulated in analogy to  p´ + a 
E 50 = E 0
50
 (16)
the yield function of Equation 7 by defining p
 0 + a 
I1b ⋅ I 2b 9 − sin 2 ψ m where E500 is obtained for p′=p0=100kPa. The se-
g = − (13)
I 3b 1 − sin 2 ψ m cant modulus E50 is closely related to the shear strain
γ50 at least for incompressible soils. Indeed, consid-
where ψm is a mobilized angle of dilatancy and the ering a triaxial test with ε2=ε3, it yields
invariants Ib are obtained by using σb=σ′+b. The
added stress b is computed such that we have g=0. γ = ε1 − ε 3 = ∆ε1 − ∆ε 3 (17)
Hence b depends both on the state of stress and the
mobilized dilatancy angle. The latter is computed where ∆ε1 and ∆ε3 are measured in the shearing
from Rowe’s so-called stress-dilatancy equation phase of loading after isotropic compression. On in-
troducing the equation ∆εvol=∆ε1+2∆ε3, it follows
sin ϕ m − sin ϕ cv that
sin ψ m = (14)
1 − sin ϕ m ⋅ sin ϕ cv
1 q50
The dilatancy angle is thus positive as soon as ψm
γ 50 = ⋅ 3∆ε 1 − ∆ε vol ≈ 1.5 ⋅ ∆ε 1 = 1.5 ⋅ (18)
2 50 50 E50
exceeds a constant-volume angle φcv. Considering
dense materials contraction is excluded by taking Especially in case of hard soils and soft rocks volu-
ψm=0 for φm<φcv. Together with the yield function metric strains tend to be small and can be disre-
the plastic potential determines the rates of plastic garded in the above expression. The subscript 50 im-
strain on the basis of the corresponding flow rule plies that the deviatoric stress is chosen such that
and the consistency condition. q50=qa/2. The plastic shear strain at q=qa/2 is ob-
Elastic strain rates are formulated using Hooke’s tained by subtracting elastic strains, i.e.
law in hypoelastic form. The stress dependent
Young’s modulus is chosen to be  1.5 1 + ν ur 
γ 50p = γ 50 − γ 50e =  − q50 (19)
m  E50 Eur 
0  p´ + a 
E ur = Eur   (15) The plastic shear strain constant γ50p0, as used in the
 p + a 
0
yield function f, is obtained for a stress state with
p′=p0.
where Eur0 is the tangent modulus for a reference For the analyses presented in this paper, tensile
mean stress of p′=p0=100kPa. The indices ur refer stresses are excluded by using a tension cut-off.
to unloading-reloading. Similarly Poisson’s ratio is
3 ON THE GROUND RESPONSE CURVE

In NATM-tunnelling the ground response curve , as


indicated in Figure 3, is used to illustrate the ground
pressure on the lining as a function of deformations.
A steep ground response curve with a low minimum
indicates a stiff and strong ground which needs little
support of a lining. Such a ground is able to carry
the overburden load by arching around the tunnel. Figure 3. Geometry and typical pressure-displacement curve
Vice versa, a relatively flat ground response curve (ground response curve).
with a high minimum corresponds to a relatively soft
ground that needs significant support from a lining. elements inside the tunnel. This does not disturb the
Attention will be focussed on relatively shallow equilibrium as equivalent pressures are applied on
tunnels and the constitutive model will be used in fi- the inside of the entire tunnel. The minimum amount
nite element analyses to compute ground response of pressure needed to support the tunnel is then de-
curves. Most recently this was done for deep tunnels termined by a stepwise reduction of the supporting
by Bliem & Fellin (2001) to find non-concave pressure.
curves. In contrast, we will consider shallow tunnels
to find trough-like ground response curves.
All subsequent analyses of the ground response 4 VERY SHALLOW UNLINED TUNNEL
curve have been obtained by using an earlier version
of the hardening-softening model. In this early ver- In this section a tunnel with a cover of H = 8 m is
sion we used the Mohr-Coulomb yield function considered. Firstly a response curve was computed
f=(σ1′-σ3′)/(σ1′+σ3′+2a)-sinφm. For f=0 this yields a for a non-cohesive ground to obtain the dashed up-
hexagonal yield surface in principal stress space, per curve in Figure 4 with a failure pressure of
rather than the smooth one after Matsuoka & Nakai pf = 0.4γD. Secondly a non-softening cohesive
(1982). ground with c′ = 40 kPa was considered to obtain
the lower curve in Figure 4 with a slightly negative
We consider an unlined circular tunnel with a di- failure pressure indicating a stable situation. Analy-
ameter of 8 m and concentrate on a plane strain ses involving cohesion softening should obviously
situation. As symmetrical tunnels are considered, render ground response curves in between the upper
calculations are based on only half a circular tunnel. curve for c′ = 0 kPa and the lower curve for
The ground is represented by 6-noded triangular c′ = 40 kPa.
elements. The boundary conditions of the finite ele- In order to model softening in narrow shear bands
ment mesh are as follows: The ground surface is free sufficiently accurate, we used a very fine mesh
to displace, the side surfaces have roller boundaries around the tunnel, as indicated in Figure 5. In fact,
and the base is fixed. Initial stresses are assumed to such a fine mesh is needed when applying a non-
be geostatic according to σ´h = K0 σ´v, where σ´h is local model in combination with a small internal
the horizontal effective stress and σ´v is the vertical length (see Section 2). First of all a softening analy-
one. The coefficient of lateral earth pressure at rest sis was carried out for a stiff clay with a softening
is taken as K0 = 1. modulus hc = 600 kPa. This yields the ground
The considered ground has a unit weight of
γ = 20 kN/m³, a friction angle of ϕ′a = 30° and an
initial effective cohesion of cpeak = 40 kPa and dila-
tancy angel ψ=0. The stiffness moduli are taken as
Eur0=90 MPa and E500=30 MPa. The Poisson’s ratio
is νur=0.2 and m=1.
The softening modulus hc can be obtained on the
basis of high-quality triaxial tests with a relatively
homogeneous post-peak sample deformations. At
Stuttgart University such tests were carried out on a
particular stiff clay, named Beaucaire Marl
(Marcher, 2002), to find hc = 600 kPa. As different
clays will give different values of hc, we used
600 kPa as a reference value in the context of a sen-
sitivity analysis, by varying hc=300; 600; 1200 kPa. Figure 4. Computed ground response curves for a shallow tun-
The first stage of the calculations is to remove the nel with H/D = 1.
response curve in the middle of Figure 4 with a 5 DEEPER UNLINED TUNNEL IN SOFTENING
marked peak in point A. Well before peak this curve GROUND
deviates already from the non-softening lower bound
and it meets the non-cohesive upper bound finally in In this section a deeper tunnel with a cover of
point B. The peak point A yields a peak pressure of H = 32 m is considered and all other parameters
pf = 0.23γD, being about half way in between the conform to the shallow tunnel of the previous sec-
failure pressures for non-softening materials with tion. The initial supporting pressure is given by
c′ = 0 kPa and c′ = 40 kPa respectively. p0 = γ (H + 0.5D) and this pressure is stepwise re-
As different clays will have different softening duced to failure. As in the previous section, upper
moduli, we have varied the softening modulus and lower bounds to the ground response curve are
around the above value of hc = 600 kPa. Resulting obtained for non-softening material with c′ = 0 and
ground response curves for hc = 300 kPa and c′ = 40 kPa respectively. Instead of showing the full
hc = 1200 kPa are indicated by the curves next to the curves starting at p = 4.5γD, Figure 6 focuses on the
middle one in Figure 4. A very slight decrease of the lower part from p = 2γD down to failure. As in the
minimum pressure is observed for hc = 300 kPa and previous section the upper curve reaches a failure
a noticeable increase for hc = 1200 kPa. pressure of pf = 0.4γD and the lower curve reaches a
Figure 5 shows a close up around the tunnel with slightly negative failure pressure.
softening zones at and beyond peak, i.e. for point A,
B and C of the middle curve in Figure 4. The lightest The computed ground response curve for the stiff
zones indicate regions where cohesion has softened clay with a softening modulus of hc = 600 kPa is
down to about 10 kPa. States B and C in Figure 5 found to be well in between the bound solutions.
show post-peak softening zones with a shear band However, there is a distinct difference to the re-
starting at the tunnel side and gradually growing to- sponse curve of a shallow tunnel. Instead of follow-
wards the surface. In the middle of these shear bands ing the lower bound for c′ = 40 kPa, the deep-tunnel
the soil has fully softened. Similar results have been response curve tends to remain closer to the upper
obtained by Schuller & Schweiger (2002) using a bound for c′ = 0 kPa. Accordingly the computed
multilaminate model that includes softening behav- peak of pf = 0.33γD at point A is only slightly below
iour. the upper bound of pf = 0.4γD. In fact, there is a dif-
ference of only ∆ pf = 0.07γD with non-cohesive
material. At the same time the deeper tunnel is sub-
ject to much larger deformations than the shallow
tunnel, as can be observed by comparing Figures 4
and 6. No doubt, the relatively large deformations in
deep tunnelling induce a relatively large amount of
cohesion softening.

Figure 7 shows the development of the softening


zone around the tunnel. In the following we concen-
Figure 5. Development of softening zones for materials with trate on the fully softened zone. At failure (state A)
hc = 600 kPa. Red indicates fully softened material with c´ = 0. one observes already a thin fully softened zone and
First for failure state, then for intermediate state and finally for post peak this zone increases rapidly. For state C,
residual state.
one observes the initiation of a shear band towards
the surface. For state D this shear band has extended
to the surface, but the fully softened part of the band
has not yet reached the surface.

Figure 6. Computed ground response curves for a deeper tun- Figure 7. Development of softening zones for a deep tunnel.
nel with H/D = 4.
6 CONCLUSIONS this finding applies to all regularisation methods, i.e.
non-local, second-order gradient and the polar con-
On tunnelling: Attention has been focussed on tun- tinuum. For clay problems, it would seam appropri-
nels in softening ground. To study consequences of ate to allow for strong (displacement) discontinuities
cohesion degradation, ground response curves have rather than for finite element meshes with weak dis-
been computed both for a very shallow tunnel and a continuities as considered in this paper.
deeper one. The computed ground response curves
appear to depend significantly on tunnel depth. For
the very shallow tunnel, a trough-like Fenner-Pacher 7 ACKNOWLEGMENT
curve is computed with a marked minimum as fail-
ure pressure. The deeper the tunnel, however, the The authors are indebted to Dr. Paul Bonnier of the
smaller the softening behaviour on the structural Plaxis software company for his work on the imple-
level of the tunnel. The present study suggests that mentation of the constitutive model.
ground response curves for very deep tunnels will
show no softening at all. This is conform to recent
numerical studies by Bliem & Fellin (2001). More- 8 REFERENCES
over it confirms practical experiences by Vavrovsky
(1994). Bliem, C. & Fellin, W. (2001). The ground response curve (in
In practical tunnelling one has to consider the sta- German). Bautechnik 78 (4): 296-305.
De Borst, R., Sluys, L.J., Mühlhaus, H.-B. & Pamin, J. (1993).
bility of the tunnel heading. In closed-face shield Fundamental issues in finite element analysis of localiza-
tunnelling the heading is supported by a support tion of deformation. Eng. Comp. (10): 99-121.
pressure, but in (open-face) NATM-tunnelling there Chen, Z., Morgenstern, N.R. & Chan, D.H. (1992). Progressive
is no support pressure and the stability of the head- failure of the Carsington Dam: a numerical study. Can.
ing depends heavily on the cohesion as shown by Geotech. 29 (6): 971-988.
Lupini, J. F., Sinner, A. E. & Vqughan, P. R. (1981). The drai-
Vermeer et al. (2003). Considering present computa- ned residual strength of cohesive soils. Géotechnique 31(2):
tional results for the ground-response curve, it would 181-213.
seem interesting to consider the three-dimensional Marcher, T., Vermeer, P. A. (2001). Macromodelling of Sof-
tunnel-heading stability also on the basis of cohesion tening in Noncohesive Soils. Continuous and Discontinu-
softening. ous Modelling of Cohesive-Frictional Materials, (Ed.)
For shallow tunnels, it may also be of importance Vermeer, P.A. (u.a.): 89 – 110. Berlin: Springer.
Marcher, T. (2002). Results of an experimental study on Beau-
to use the present model in 3D settlement analyses caire-marl (in German). Mitteilung des Instituts für Geo-
as described by Vermeer et al. (2003). technik der Universität Stuttgart 49.
Marcher, T. (2003). Non-local modelling of softening of dense
On modelling: Stiff clays and clay stones do not sands and stiff clays (in German). Mitteilung des Instituts
only show cohesion softening but also friction sof- für Geotechnik der Universität Stuttgart (doctoral thesis)
49.
tening. The latter is especially important when the Matsuoka, H. & Nakai, T. (1982). A new failure criterion for
clay fraction dominates the silt fraction (Lupini et soils in three-dimensional stresses. IUTAM Conference on
al., 1981) and it would seem that both phenomena Deformation and Failure of Granular Materials, Delft, 31
can be well-described by means of the non-local sof- August-3 September 1982. Rotterdam: Balkema.
tening parameter γ*. The related softening moduli hc Pacher, F. (1964). Deformation measurement in a test gallery
may for instance be measured in a direct-shear test for the exploration of ground response and tunnel design
(in German). Felsmechanik u. Ingenieurgeologie, Suppl. I.
or a ring-shear test. Unfortunately these softening Potts, D., Kovacevic, N. & Vaughan, P.R. (1997). Delayed col-
moduli relate directly to the internal length l, which lapse of cut slopes in stiff clays. Géotechnique 47, No. 5,
governs the thickness ts≈2l of shear bands. 953-982.
Schuller & Schweiger (2002). Application of a multilaminate
In reality clays have vanishing thin shear bands and model to the shear and formation in NATM tunneling.
Computers and Geotechnics 29, No. 7, 501-524.
a real thickness simulation requires vanishing small Skempton, A.W. (1964). Fourth Rankine Lecture: Long-term
finite elements. In order to limit the computational stability of clay slopes. Géotechnique 14, No. 2, 77-102.
effort, one has to simulate relatively thick shear Vavrovsky, G.-M. (1994). Development of groundpressure,
bands by means of an artificially large internal deformation and tunnel design (in German). Felsbau 12 (5):
length and an artificially high softening modulus. On 312-329.
Vermeer, P.A., Ruse, N., Marcher, T. (2002). Tunnel Heading
increasing this modulus one approaches an abrupt Stability in Drained Ground. Felsbau, Jg. 20(6): 8-18.
type of softening which renders numerical proce- Vermeer, P.A., Möller, S.C., Ruse, N. (2003). On the Applica-
dures unstable again. As a consequence there is an tion of Numerical Analysis in Tunnelling. Post proceedings
upper bound for the shear band thickness and thus 12th Asian Regional Conference on Soil Mechanics and
for the size of the elements. Geotechnical Engineering (12 ARC), Singapore, 4-8 Au-
gust 2003, 2: 1539-1549.
Therefore realistic simulations of yielding in stiff
clays will require extremely fine meshes. No doubt

You might also like