Sie sind auf Seite 1von 9

Underground Singapore 2005

Checking and reviewing the output from numerical


analysis
J. Nick Shirlaw
Golder Associates (Singapore) Pte Ltd, Singapore

D. Wen
Land Transport Authority, Singapore

ABSTRACT: The use of numerical methods as part of the design of underground works is now com-
mon. However, the increasing complexity of the programs also results in greater potential for error. To
minimize the risk of failure, designs should be reviewed by the original design team and checked ex-
ternally. However, in Singapore internal reviews are often perfunctory at best, and external checks
may not cover the full design process. To carry out a proper review, the reviewer must be fully aware
of what the designer has done. Checking should start with assessing the adequacy of the site investiga-
tion data, and an independent assessment of the geological model and geotechnical parameters. It
would help if common errors in the use of numerical methods are publicized. A few common prob-
lems in the use of such methods in Singapore are discussed.

1 INTRODUCTION

The use of numerical methods (finite difference, finite element, boundary element or discrete element
methods) is becoming an integral part of the design process for many geotechnical problems. While
such methods may appear highly sophisticated, the use of them does not guarantee good design. The
Committee of Inquiry into the Nicoll Highway collapse found that one of the errors in the design of the
retaining structure at Nicoll Highway was the use of the wrong soil model in the finite element analy-
sis of that structure. The soil model was part of the input into the program, and resulted in the moments
and movements of the wall being underestimated by a factor of about 2. This shows how the results
obtained from such programs can be highly dependent on the program user. Carter et al (2000) de-
scribe exercises carried out by the German Society for Geomechanics in which the same excavation
problem was analysed by fifteen users. The predictions for the movement of the top of the wall ranged
from -229mm to +33mm. Even where the same program and soil constitutive model were used, there
was a significant difference in the predictions. Potts (2003) refers to similar benchmarking exercises,
and states that ‘A significant amount of human error is involved’. These examples would imply that
there is no guarantee that just because a sophisticated numerical analysis is carried out, the result will
be free from error.

There is a mental barrier to overcome in terms of accepting the potential for error in the use of numeri-
cal analysis. The vast majority, in fact almost all, papers, on numerical analysis report close agreement
with measured data. For example, Negro and de Quiroz (1996) carried out a survey of 65 papers on the
analysis of settlement over bored tunnels, and assessed that over 70% of the ‘predictions’ were +/-
10% of the actual settlement. As pointed out by Shirlaw (1996), the natural variations in the measured
settlement over tunnels due to normal variations in workmanship and ground, in apparently the same
ground conditions, is much higher than +/- 10%. Taken as a whole, the reported results were clearly
skewed towards those that nearly matched the field results. This may be because few people publish
cases where the predicted and actual settlements are significantly different, or because the so called
‘predictions’ were in fact made after the event and adjusted to match the measured performance. The
examples assessed by Negro and de Quiroz (1996) were for tunnels, but a survey of papers on deep
excavations would probably reveal a similar degree of claimed accuracy for the movement of the walls
of excavations. This can be compared with the high degree of variability recorded by Carter et al
(2000) for the predicted movement of the walls of a relatively simple excavation. The same conclu-
sions can thus be made for excavations as for tunnels. The literature generally gives an impression of
readily attainable accuracy that it is impossible to achieve in practice. There are very few documented
cases of exceptionally poor prediction , such as those leading to failure.

Numerical methods for geotechnical engineering have developed rapidly over the past 25 years, from
the first ‘beam on elastic spring’ programs to the 3-D finite element programs that are beginning to be
used in commercial practice today. Advanced soil models have been developed and incorporated, to
give a better representation of real soil behaviour. The greater complexity of the programs should al-
low greater accuracy in prediction. However, the greater complexity also comes at a price. The greater
complexity means there is much greater potential for major errors by the user of the program. The
process has also become more opaque, making it more difficult to identify those errors. Vaughan
(2003) states that the ‘new methods are at least as easy to abuse as the old ones’, referring to numerical
analysis.

In order to reduce the risk of inadequate design it is therefore necessary that the output from numerical
analysis is checked, just as it is common to carry out checks on other design methods. However, tradi-
tional design methods, based on limit equilibrium methods and/or design charts, are much more simple
and transparent than numerical methods. This lack of transparency raises the issue of how to check de-
signs based on numerical analysis.

This paper will explore the issue of the checking of the design of deep excavations, although many of
the principles can be applied to other design situations.

2 CHECKING OR REVIEW?

There is a major difference between reviewing a design and checking it. A review is carried out to con-
firm that the design complies with the requirements of the owner, including any requirements in rele-
vant design codes. The purpose of design review is to identify errors in the original designers work. In
contrast, a design checker carries essentially the same responsibility as the original designer: the
checker warrants that the design is safe and fit for purpose, irrespective of how the original designer
carried out his work.

While a reviewer would normally be required to confirm that the design complies with the relevant
Codes of Practice or Standards, current codes and standards are typically written around limit equilib-
rium design methods, and are not explicit about how the output from numerical analysis should be
used in design. This is not usually an issue for serviceability limit states (SLS). However, for ultimate
limit states (ULS), where and how to apply factors of safety or partial factors within a numerical
analysis is still a matter of contention.

In the use of numerical methods, simply following the program manual or tutorials is no guarantee of
adequate design. The interpretation of the program manual was a major issue in the Nicoll Highway
design and at the Committee of Inquiry, and the reason given for the selection of an incorrect soil
model.

One of the limitations of reviewing designs based on numerical methods is that the primary designer,
rightly, chooses the program or suite of programs to be used. The choice of the program will be gov-
erned both by the nature of the problem and the familiarity of the designer with that program. The re-
viewer may be completely unfamiliar with that program. The reviewer is then in the position of re-
viewing the application of a program where the primary designer has a high level of experience and
confidence in the use of the program, but the reviewer has little or none of either.

A reviewer can, and should, go through the input to and output from numerical analysis and ensure
that these are sensible. However, the limitations discussed above mean that this is not sufficient to
warrant the design; hence the important distinction that is made between review and checking.
In contrast to the review process, which involves going through what the primary designer has done, a
design check involves an independent assessment of the safety of the design. The checker should have
all of the information available to the designer, and the design drawings and specifications. The
checker should independently assess the safety of the design as developed on the drawings and in the
specifications.

Under typical quality control procedures during design, the design team should have its own review or
checking process. It is only after the design has been through this internal process that it is then sub-
jected to external review or checking. Most commonly, an internal process consists primarily of re-
view, i.e. trying to find any errors in the design.

An external review/checking process may be governed by regulation, or by the owner, or by regulation


supplemented by additional requirements by the owner.

Current design processes commonly include both review and checking, with review carried out inter-
nally, within the design team, while external checking is now a common requirement for the design of
excavations.

Some of the issues in review or checking can be considered further assessing the various steps of a
typical design process.

3.0 DESIGN PROCESS

Typically there are a number of major steps in developing a design based on numerical methods. A
typical sequence is outlined in Table 1.

Table 1 – Design process for numerical analysis


Step
1 Select program or suite of programs
2 Idealise problem to fit program
3 Develop input to program
4 Run program
5 Use output to develop design

This process is essentially similar to a traditional design, with minor variations, as shown in Table 2:

Table 2 – Design process for traditional design


Step
1 Select method of analysis
2 Idealise problem to fit analytical method
3 Develop input to model
4 Carry out analysis
5 Use results of analysis to develop design

The steps in Table 1 are discussed further below.

3.1 Idealize problem


All analysis involves some degree of idealization. Part of this process involves constructing a geologi-
cal model of the site from the available site investigation and other geological data. Inevitably, the
geological model will not match precisely the actual geology of the site, as it involves extrapolation
between boreholes, and some degree of simplification.

Unless extremely complex 3-D modeling is being carried out, the next stage of idealization is to sim-
plify the problem to suit the program. Most programs are still 2-D programs, and part of idealization
involves how to model a three dimensional problem in two dimensions.
3.2 Develop input to program
Input to the program could include:

• The geological strata, and their boundaries


• The constitutive models to be used for the various strata
• The properties of the various strata
• Initial groundwater levels
• Boundary conditions
• Surcharges
• The mesh – this can be treated as input even if it is automatically generated, as the decision to
use automatic mesh generation is an input
• Structural elements, and their properties
• The sequence or stages of construction, together with the anticipated time for construction

Many programs also either allow or require the input of groundwater levels for the various stages of
construction, in which case this becomes part of the input.

3.3 Run program


This part of the exercise appears simple. However, no design should be based on a single run of the
program. The program should be run a number of times to explore the sensitivity of the results to the
possible ranges of the various input parameters, assumptions and simplifications.

3.4 Use output to develop design


For SLS analysis the output from the program is used directly in the design. However, for ULS design,
the relevant factors of safety have to be applied. This can be done either by factoring the input parame-
ters, or by factoring the output.

3.5 Discussion on design stages


Errors can occur in any step of this process, or in any part of each step. For example, the variety of in-
put required in step 3 provides plenty of scope for errors. A well known case of a failure related to
numerical analysis (although this was not the sole cause) was the Nicoll Highway collapse of 2004.
The report of the committee of inquiry (COI), Magnus et al. (2005), can be used to show the reported
errors in that case, in the context of the overall design process described above. In the COI report,
Chapter 5 assesses the causes of the collapse, including the numerical modelling used as the basis for
the design of the wall. Some of the problems in the original analysis identified in the report were:

1) Use of drained parameters with the Mohr-Coulomb criteria and ‘undrained’ material model to
represent the soft marine clay at the site
2) The assumption of undrained response to excavation in the Old Alluvium
3) The assumption (as input) that the pore pressures profile in the passive zone would be hydro-
static with respect to the base of the excavation
4) Inadequate consideration of ongoing settlement of the area, and the implications of this for the
degree of consolidation of the clay
5) The assumption of symmetry across the excavation was not representative of the actual ground
conditions
6) The partial factors required by BS 8002 were not applied to the soil parameters

In the context of the design process discussed above, item 5 relates to step 2, items 1, 2, 3 and 4 relate
to step 3 and item 6 to step 5.

One of these assumptions, that for the pore pressure in the passive zone, was not apparent in the input
to, or output from the program used, which was PLAXIS. As identified in the COI report, one part of
the PLAXIS manual was used in justification of the decision to represent the soft clay as outlined in 1)
above, although another part of the manual gave different advice.
The example of the Nicoll Highway analysis shows how a design based on numerical methods can
significantly in error, and that it may not even be apparent exactly what the user of the numerical
analysis has done, if only the printed input and output are reviewed. One way round the latter problem
is for the reviewer to be provided with the digital input files and the program. By rerunning the analy-
sis, the reviewer should be able to assess what the designer has done, even where this is not apparent
from the printed input and output files. However, this requires the reviewer to have some familiarity
with the program.

For a checker the essential requirement is to confirm that the structure, as shown on the drawings, is
safe. Any errors by the original designer are not an issue provided that the final structure is safe, meets
code requirements and does not have an unacceptable effect on adjacent structures and utilites. How-
ever, this then raises the issue of what happens when the checker finds that this is not the case. If both
the original design and the check are based on numerical methods, the question arises as to whether the
error(s) are in the design or the check. If the original design is based on numerical methods and the
check based on design charts or limit equilibrium methods, then it can be questioned whether the dif-
ference is due to an error in the numerical analysis, or simply due to a difference in the accuracy of the
analysis.

4.0 DISCUSSION

As numerical methods become more complex the potential for greater accuracy co-exists with the in-
creased risk of major errors in the analysis. The increasing complexity also reduces the benefits of a
simple review of the design. A reviewer attempts to identify errors in the original design, in contrast to
a checker, who determines whether the design is safe, irrespective of whether it contains errors. As
discussed above, the reviewer faces a number of problems in identifying errors in a numerical analysis.
These problems are specific to numerical analysis, and are such as to suggest that review is of limited
value in respect of designs based on numerical analysis. Logically, this would suggest that checking,
rather than review, should be a standard requirement for designs based on numerical analysis. How-
ever, the reviewer within the design team still needs to review the design for errors, so the review part
of the process cannot be totally avoided.

The following sections discuss what can be done to improve both review and checking in relation to
numerical methods.

4.1 Improving design review


A reviewer should go through every step in the design process given in Table 1, and every input, as
listed in 3.2, to verify that the assumptions and inputs to the model are reasonable. The reviewer
should also carry out some simple verification exercises to confirm that the output is reasonable. A
pre-requisite for such a review is that the reviewer is fully aware of what the original designer has
done. The input to and output from the model should be fully documented.

For example, for excavations the input documentation should include (unless not relevant to the prob-
lem being modelled):

• The stratigraphy and general structural arrangement


• All parameters for the soil and/or rock and / or treated ground
• All parameters for the stiffness and strength of the structural elements
• Whether, and how, the stiffness of the structural elements varies with deflection or curvature
• The boundary conditions for the model, including hydraulic boundary conditions (where ap-
plicable)
• Surcharges
• The sequence of construction
• How the interface between ground and structural elements is modelled
• Any simplifications or assumptions made to develop the model
• The failure criteria used for the various soil and rock strata
• The initial water pressure, and how the change in the water pressure with excavation is mod-
elled or input to the program
• The size and nature of the elements used
• A drawing showing the mesh used shall be included, with a note to say whether this was
automatically generated, input, or adjusted (from an initial mesh). Where ground treatment is
represented in the model, and is treated as a soil, an enlargement of the mesh specifically for
each area of ground treatment shall also be included

The documentation of the output will depend on the nature of the problem being modelled. Where
relevant to the particular problem the output should include, for each stage of an excavation:

• The magnitude and distribution of active earth pressures on the wall


• The magnitude and distribution of passive earth pressures on the wall
• The magnitude and distribution of ground water pressures on the active and passive sides of
the wall
• The magnitude and distribution of bending moments in the retaining wall(s)
• The magnitude and distribution of shear forces in the retaining wall(s)
• The magnitude of strut or anchor loads, and of the forces in any parts of the permanent struc-
ture used to resist ground and/or groundwater loads during construction
• The magnitude and distribution of wall movements
• The magnitudes and vectors of ground movements
• The magnitude and distribution of lateral and vertical movements on elements such a piles
• The location of any yield points
• The change in groundwater pressures in the zone from the excavation to the point at which
there is no significant change in ground water pressures, compared with the initial condition
• For zones of ground treatment, the magnitude of the minor principle stresses, to demonstrate
that there are no tension zones

Where the results are presented graphically, scales should be provided such that both the magnitude
and position of any information provided can be established, with a reasonable accuracy, by scaling
off the material submitted.

Where yield points are identified in the analysis, stress paths should be plotted at a representative
number of the yield points, and the stress at failure checked against the values assessed from the site
investigation data.

Unless the reviewer uses the program to identify for himself what has been done, the list of input and
output data given above is the minimum that a reviewer needs to understand what has been done in the
numerical analysis of an excavation.

Simple verification exercises that the reviewer can carry out could include, as appropriate:

• Hand calculation of basal stability


• Hand calculation of embedment depth
• Comparison of lateral pressures with limiting pressures based on charts, such as those given in
BS 8002
• Comparison of the stress at failure with the limiting values based on simple earth pressure the-
ory
• Calculation of strut loads, using apparent earth pressure diagrams
• Comparison with published case histories of similar excavations in similar strata

Based on current industry practice in Singapore, most internal review of numerical analyses within
consulting and other design organizations is perfunctory, and far from the level given above. However,
designers should be aware of the potential for error in the use of numerical methods, and the potential
liabilities associated with such errors.
4.2 Improving checking
A check for a design should be an independent assessment of all of the stages of design outlined in
Table 1. Some checkers in Singapore start from the geological model and geotechnical parameters
used by the original designer. Such a check is not complete, as it does not check whether the geologi-
cal model and geotechnical parameters are correct; if there are errors in these, which form the basis for
the analysis, the analysis itself will be in error.

The checker should start with the basic information from the boreholes and other site investigation
data. The first issue the checker has to assess is whether this information is sufficient, both in quantity
and quality, to form a reasonable basis for the analysis that will follow.

Once the checker has decided that there is sufficient geotechnical information, he should then proceed
to develop the design, using the steps in Table 1. One of the issues here is the program or method of
analysis adopted by the checker. Simply using the same parameters and the same program as the origi-
nal designer does not provide a check on the original design. The checker therefore has to decide
whether he will use a different program, or an entirely different method of analysis (such as design
charts or limit equilibrium analysis).

If the checker decides that the original design is unsafe, then there is the issue as to who is wrong, the
original designer or the checker? One approach is to follow the more onerous of the two designs, even
though this may mean that the work is much more expensive than originally estimated. Another ap-
proach is for the original designer and checker to compare the designs and determine the cause of the
difference(s), and determine what is the approach that both can accept.

4.3 Other improvements


Almost all literature about numerical methods is about ‘how to’ and about the high level of accuracy
that can be achieved with well conducted numerical analyses. However, there are very few cautionary
tales about what can go wrong, or ‘how not to’. More awareness of the potential pitfalls in the use of
numerical methods would help to give guidance to users. As a contribution to the latter, a number of
common problems in the use, or abuse, of numerical analysis in relation to excavations in Singapore
are covered in the sections below.

4.3.1 Symmetrical models (half mesh) in asymmetrical conditions


It is quite common for excavations to be modelled assuming an axis of symmetry down the centre of
the excavation. In this case only half of the excavation is actually modelled. However, sloping bounda-
ries between strata, unbalanced surcharges or other excavations on one side of the modelled excavation
can lead to conditions that are asymmetrical. In relation to symmetrical conditions, this will result in
one wall deflecting more than anticipated, and one less. The curvature, and thus the bending moments,
will also vary between the two walls. If symmetry is assumed in asymmetrical conditions, the diver-
gence from the predicted values can be very large. Programs for 1-D analysis inherently assume sym-
metry, and are not suitable for use in conditions where there is a significant degree of asymmetry.

4.3.2 Use of automatically generated meshes in the modelling of jet grout slabs or other forms of
ground treatment
A layer or zone of ground treatment, such as a jet grout slab, used in relation to a deep excavation, can
be subject to both compression and bending. There have been cases of automatic mesh generators not
providing a sufficiently dense mesh within the ground treatment zone to capture the effects of the
bending. When using an automatic mesh generator, the mesh should be carefully inspected for this
type of problem. When in doubt, test runs should be carried out to assess the sensitivity of the output
to the density of the mesh.

4.3.3 Use of effective stress parameters for undrained analyses in clay with an OCR<2
The shearing of soft clays generates positive excess pore pressures. Advanced soil models, such as the
Cam Clay model, can be used to model this. However, simpler models, such as the Mohr-Coulomb,
will not. If effective stress parameters are used for undrained analyses, together with one of the simple
soil models, then the strength of the clay can be significantly overestimated.
4.3.4 Calculation of strut stiffness for symmetrical analyses
Although the potential problems in the use of symmetrical analyses are covered in general terms
above, a particular issue is the calculation of strut stiffness. In symmetrical (half mesh) analyses (and
1-D analyses), this is commonly done by calculating the strut stiffness, or equivalent spring reaction,
based on half of the length of the strut. This inherently assumes that the strut compresses equally about
a point of symmetry in the centre of the strut. However, particularly for corner bracing and raking
struts, this may not be the case in practice. Where one end of the strut is subject to full active pressure,
but the other end needs to develop passive pressure as a resistance, then the assumption of symmetry is
not correct. Corner struts for long excavations

4.3.5 Setting the boundaries of the model too close to the excavation
If the model boundaries are set too close to the excavation, this will distort the results in a way that is
unconservative both for the wall design and the assessment of ground settlement. This can be assessed
by varying distance of the model boundaries from the excavation, and assessing the point at which fur-
ther movement of the boundaries has a negligible effect.

4.3.6 Setting of recharge (or constant head) boundaries for groundwater modelling
The selection of recharge (or constant head) boundaries can have a major effect on the results of
groundwater modelling. Many of the key aquifers in Singapore are relatively thin, confined aquifers.
Examples include fluvial sands below the marine clay and weathering Grades V and VI of the Bukit
Timah Granite. In these conditions, pore pressure changes due to seepage, resulting from excavation,
can be very widespread. Settlements due to dewatering effects transmitted through confined aquifers
have been monitored, in some cases, over 300m (in plan) from deep excavations. This would indicate
that the effective constant head boundary was even further from the excavation. However, if the aqui-
fer is not confined, or is receiving recharge from, say, recharge wells, then the constant head boundary
will be significantly closer to the excavation.

The selection of the location for the constant head boundaries can have a significant effect on some of
the ultimate and serviceability limit state calculations for an excavation. If the constant head boundary
is set too far from the excavation, the passive pressures can be overestimated, as the pore pressures in
the passive zone may be underestimated. Conversely, if the constant head boundary is set too close to
the excavation, the zone over which consolidation settlements will occur can be underestimated.

4.3.7 Running analyses for seepage without considering the effect of stress changes on pore pres-
sures
During excavation, the pore pressures in the active and passive zones will be altered first by changes in
stress, and then by seepage. If the effect of the changes in stress are not considered in the seepage
analysis, then the results will be misleading. For example, many excavations in Singapore are carried
out in areas of deep marine clay. In a simple seepage analysis the thick plug of low permeability ma-
rine clay below the base of the excavation acts as an aquitard, minimising seepage and so minimising
the reduction in pore pressures outside the excavation. However, the stress changes associated with
excavation result in reduced pore pressures in the marine clay, except in those zones where the genera-
tion of positive pore pressures due to shearing exceeds the effects of stress relief. The effect of the re-
duction in pore pressures due to stress relief typically depresses the pore pressures below the steady
state seepage pressure. Under these conditions, the marine clay in the passive zone no longer acts as an
aquitard.

4.3.8 Modelling of piles in the base of excavations


Piles in the base of the excavation can significantly increase the ultimate passive pressure (Broms,
Heng and Chong, 1987). Combinations of piles and one or more jet grouted slabs have been used suc-
cessfully for a number of excavations in Singapore, to control both basal stability and lateral move-
ment (Shirlaw, Tan and Wong, 2005). In such cases, the piles anchor down the slabs, providing a reac-
tion to base heave pressures. This reaction has the same effect as providing a surcharge on the passive
zone. Apart from such cases, where the piles are part of the retaining system, it is common for design-
ers to ignore the presence of the piles in their design of the retaining system. As the piles are benefi-
cial, this is conservative. However, piles should be included in the model in the following cases:

• If the piles form part of the overall retaining system


• In a back-analysis, as the back-analysis would not be representative if the piles were not in-
cluded

The piles are typically spaced intermittently along the excavation, and in a 2-D model, the modelling
of such intermittent piles requires care. A common method for including contiguous or secant piles
(used as retaining walls) into models is to ‘smear’ the stiffness of the piles so that the line of piles is
treated as a wall of uniform stiffness. However, if the same process of ‘smearing’ is applied to inter-
mittent piles in the base of the excavation, this will overstate the contribution of the piles, which is po-
tentially unsafe. Use of a ‘smeared’ stiffness to represent an intermittent line of piles treats those piles
as a continuous wall, potentially providing larger skin friction reaction and greater confinement to the
ground in the passive zone than can occur in practice.

5.0 CONCLUSIONS

Although there is very little literature on the subject, there is evidence that analyses based on numeri-
cal analysis can be seriously in error, usually because of errors by the original designer (program user).
Once this has been accepted, then it follows that there is a need for safeguards to avoid unsafe design
based on erroneous numerical analysis. Such safeguards typically consist of internal verification and
quality control procedures (described here as ‘design review’) within the original design organization,
and external checking. The purpose of these exercises is typically different, in that internal verification
and quality control is intended to identify and correct errors in the design, while checking is an inde-
pendent analysis to confirm the safety of the structure as designed.

Proper review can only be carried out if the reviewer is fully aware of what the original designer has
done. Based on typical documentation for such designs in Singapore, it is clear that internal review is
often perfunctory, and that the reviewer is often not properly aware of what the original designer has
done.

The checker should consider not just the structure, but also the adequacy of the ground information
used to develop the geological model and the geotechnical parameters.

Documentation of some of the common pitfalls in the use of numerical analysis would help users to
avoid these pitfalls, and improve the overall quality of the resulting designs.

REFERENCES

Broms, B.B., Heng, Y.S. and Chong, M.K. 1987. Influence of piles on the stability of excavations in soft clay.
Fifth International Geotechnical Seminar, Cas Histories in Soft Clay, NTI, Singapore, pp 121 - 126
Carter, J.P., Desai, C.S., Potts, D.M., Schweiger, H.F., and Sloan, S.W. 2000. Computing and Computer Model-
ling in Geotechnical Engineering, GeoEng2000, 19th-24th November, Melbourne, publ. Technomic, 1157-
1253
Potts, D.M. 2003. 42nd Rankine Lecture: Numerical analysis: a virtual dream or practical reality? Geotechnique,
53, no. 6, 535-573
Shirlaw, J.N, Tan, T.S. and Wong, K.S. 2005. Deep excavations in Singapore marine clay. Special lecture, 5th
International Symposium, Geotechnical Aspects of underground construction in soft ground. Amsterdam,
June.
Shirlaw, J.N. 2000. Discussion: Can settlements over tunnels be accurately predicted using advanced numerical
methods? Proc. Geotechncial Aspects of Underground Construction in Soft Ground, Tokyo. Kusakabe, Fu-
jita, Miyazaki, eds, publ Balkema 471-472
Vaughan, P.R. 2003. Vote of thanks to the 42nd Rankine Lecture. Geotechnique, 53, no. 6, 572-573

Das könnte Ihnen auch gefallen