Você está na página 1de 17

345

Analytical methods to reduce uncertainty in tunnel


construction projects
J.Y. Ruwanpura, S.M. AbouRizk, and M. Allouche

Abstract: This paper presents a method to quantify uncertainty using simulation techniques and approximate
geotechnical methods. Unknown soil conditions are major contributors to uncertainty in any underground construction
project. Soil conditions are unknown because generally soil samples taken from vertical boreholes show only the soils
present in the discrete borehole locations. The soil profiles between the boreholes therefore contribute to project uncertainty, and construction practitioners must make assumptions about these soil profiles for construction planning and
scheduling purposes. Analytical and simulation methods are presented to accurately predict soil profiles between boreholes and reduce uncertainty in a rough and ready fashion. These methods use existing borehole data to create an analytical model for soil prediction, which is then incorporated with a process interaction simulation model of the
construction project using special purpose simulation concepts and advanced geotechnical characterization techniques.
The application of these methods to an Edmonton tunnel construction project is also detailed. Construction engineers or
managers can use these simulation methods to strengthen the geological data obtained for the construction project.
Key words: borehole data, construction, risk, soil families, soil profiles, soil transitions, special purpose simulation, tunnelling, uncertainty.
Rsum : Cet article prsente une mthode pour quantifier lincertitude lors de lutilisation de techniques de simulation
et de mthodes gotechniques approximatives. Des conditions de sol inconnues sont le principal facteur dincertitude de
tout projet de construction souterrain, Les conditions de sols sont inconnues parce que les chantillons de sol, pris de
forages verticaux, ne montrent gnralement que les sols prsents aux endroits spcifiques du forage. Les profils de
sols entre les forages contribuent donc lincertitude du projet et les constructeurs doivent poser des hypothses sur les
profils de sols lors de la planification et de ltablissement de lchancier de construction. On prsente des mthodes
analytiques et de simulation afin de prdire prcisment les profils de sols entre les forages et ainsi rduire
lincertitude un niveau gnral et acceptable . Ces mthodes utilisent les donnes de forage existantes pour crer
un modle analytique de prdiction des sols, lequel est ensuite incorpor dans un modle de simulation de processus
dinteraction du projet de construction utilisant des concepts de simulation spciale et des techniques de caractrisation
gotechniques avances. Lutilisation de ces mthodes lors dun projet de percement dun tunnel Edmonton est expliqu en dtail. Les ingnieurs ou les gestionnaires de construction peuvent utiliser ces mthodes de simulation pour soutenir les donnes gotechniques obtenues pour le projet de construction.
Mots cls : donnes de forage, construction, risque, familles de sols, profils de sols, transitions de sols, simulation spciale, percement de tunnels, incertitude.
[Traduit par la Rdaction]

AbouRizk et al.

360

Introduction
Tunnel construction projects are considered to be highrisk projects. Identifying and quantifying uncertainty factors
in tunnel construction helps the project planners, engineers,
and constructors to mitigate risks during construction. Soil

conditions are major contributors to uncertainty in any underground construction projects. For typical utility tunnel
construction, soil samples from vertical boreholes spaced
about 300500 m apart show only the soil types that are
present in the borehole locations. The soil profiles between
the boreholes are therefore uncertain. Practitioners must

Received 3 July 2003. Revision accepted 20 November 2003. Published on the NRC Research Press Web site at http://cjce@nrc.ca
on 6 April 2004.
J.Y. Ruwanpura. Department of Civil Engineering, ENF 232, The University of Calgary, 2500 University Drive NW, Calgary, AB
T2N 1N4, Canada.
S.M. AbouRizk.1 Department of Civil and Environmental Engineering, 220 Civil Engineering Building, University of Alberta,
Edmonton, AB T6G 2G7, Canada.
M. Allouche. Research Associate, Hole School of Construction, 220 Civil Engineering Building, University of Alberta, Edmonton,
AB T6G 2G7, Canada.
Written discussion of this article is welcomed and will be received by the Editor until 31 August 2004.
1

Corresponding author (e-mail: abourizk@ualberta.ca).

Can. J. Civ. Eng. 31: 345360 (2004)

doi: 10.1139/L03-105

2004 NRC Canada

346

make assumptions about the soil profiles between the boreholes for construction purposes. There are various factors
considered in predicting soil conditions in an area. The
availability of a particular soil type, the start and end elevations of the soil layers, the thicknesses of the soil layers, and
the distribution of the soil layers between the boreholes and
water table are some of the major factors that can affect tunnel construction productivity.
This paper describes analytical methods to quantify and
reduce uncertainty in tunnel construction projects. This analysis helped the Asset Management and Public Works Department of the City of Edmonton to reanalyze the soil
profiles for a major tunnelling project in Edmonton, Alberta.
The soil profiles predicted using the analytical methods explained in the following sections were found to be accurate
during soil exploration and tunnel construction. Although
these methods are not a replacement for comprehensive
geotechnical investigation, construction engineers or managers can use them to confirm, validate, or complement such
investigations.

Risks and uncertainty in tunnel


construction
Risk can be defined as the possibility of suffering loss or
harm (Concise Oxford Dictionary 1991). AbouRizk (2000)
clarifies that the event and (or) its outcome must be associated with a certain degree of uncertainty (the possibility) for
risk to be an issue. Therefore, risk and uncertainty go hand
in hand and must be assessed concurrently. Palisade Corporation (1999) suggests that Risk derives from our inability
to see into the future, and indicates a degree of uncertainty
that is significant enough to make us notice it. Lifson
(1972) goes further in tying the two concepts together by indicating that Risk is the measure of uncertainty concerning
outcomes; it is the explicit, quantitative representation of the
uncertainty associated with the estimates of outcomes.
The most effective way to deal with uncertainty is to collect more information and knowledge. When that is expensive or infeasible, various researchers have tried to quantify
uncertainty using analytical or simulation techniques. The
remaining paper details how applying analytical and simulation techniques can reduce uncertainty in tunnel construction
projects.
To address uncertainty in tunnel construction, researchers
require an approach that provides better characterization of
soil conditions underground. Such approaches are often
time-consuming and laborious. The method described in this
paper provides such soil characterization in a rough-andready manner that maximizes the certainty while minimizing involvement of the practitioner by making use of existing knowledge and simple probability concepts. The soil
profiles are then predicted based on these probabilities using
simulation techniques. To carry out this characterization
method, the following must already be in place:
1. a database of data from borehole logs within the area of
the tunnel
2. an analytical model to predict soil conditions within the
tunnel alignment
3. a process interaction simulation model for the tunnel to
be constructed using special purpose simulation concepts

Can. J. Civ. Eng. Vol. 31, 2004

equipped with advanced geotechnical characterization


techniques.
This paper discusses an approach that uses the abovementioned requirements in quantifying uncertainty in tunnel
construction.

State-of-the-art in special purpose


tunnelling simulation
A special purpose simulation (SPS) template for utility
tunnel construction operations was designed in collaboration
with the Asset Management and Public Works Department
of the City of Edmonton. AbouRizk and Hajjar (1998) define SPS as a computer-based environment built to enable a
practitioner who is knowledgeable in a given domain, but
not necessarily in simulation, to model a project within that
domain in a manner where symbolic representations, navigation schemes within the environment, creation of model
specifications, and reporting are completed in a format native to the domain itself. These SPS tunnel construction
templates developed with simphony (Hajjar and AbouRizk
2000), enabled the City of Edmonton to evaluate various
tunnelling options expeditiously, primarily allowing the engineers to test the validity of their construction planning
strategies. A complete design, development, and implementation of this tool can be found in Ruwanpura et al. (2001a)
and Er et al. (2000).
Figure 1 depicts the modeling layout of the tunnel template including some of its input parameters, statistics, and
cost planning outputs. Particular attention should be directed
to the input parameters of the soil segment modeling element at the right side of the figure. The user could add many
tunnel segments to the model depending on the soil properties. In this example of the simulation template, there are
two soil segments in the figure for illustration purposes: the
first is excavated in bedrock (100%) and the second section
is excavated in a combination of bedrock (80%) and glacial
clay till (20%). The composition of the soils is subjectively
set by the modeller and is therefore approximate in this template.
The following are sample applications of the tunnel simulation template that were implemented with the City of Edmonton:
1. testing the template for tunnel construction projects by
changing different resources and setting up options for
undercut and shaft (see AbouRizk et al. 1999)
2. evaluating various alternatives at the project planning
stage of an Edmonton tunnel construction project (Er et
al. 1999, 2000)
3. evaluating two bidding scenarios and determining the
better of the two based on the outputs of the simulation
model (Ruwanpura et al. 2000)
However, the simulation results documented by
Ruwanpura (1999) demonstrated that uncertainty factors
could not be predicted using the current state-of-the-art described in Ruwanpura et al. (2001a). In this particular project, the adverse soil conditions along with several other
factors had caused very low productivity in terms of tunnel
progress. This identified other factors that affect tunnelling
productivity, such as soil conditions, human productivity,
tunnel boring machine (TBM) efficiency, and method of
2004 NRC Canada

Fig. 1. Special purpose simulation (SPS) tunnel template model layout with some input parameters, statistics and cost planning outputs.

AbouRizk et al.
347

2004 NRC Canada

348

construction operation, and the need for a methodology to


predict soil types and their transitions throughout the tunnel
trajectory for the purpose of accurately predicting production rates.

Establishing a database of borehole data in


the tunnel area
This section describes a method to reduce the uncertainty
between boreholes by dividing the distance separating the
boreholes into segments to identify the soil transitions. For
example, if there are two boreholes spaced at 250 m, the distance may be divided into a few sections based on the soil
types observed in the two boreholes and the surrounding
area of the tunnel. The construction manager could decide
whether he (she) needs to drill additional boreholes to further justify the existence of soil types other than those found
in the boreholes or to take extra measures to handle that portion of the tunnel. The overall process of this method involved five steps: gathering and characterizing the data,
analyzing and dividing the data into unique groups, clustering the soil families, developing algorithms to predict the
soil families, and applying the concepts to a real tunnel construction project.
Gathering and characterizing the data
Boreholes driven by the department of Asset Management
and Public Works of the City of Edmonton in the past
30 years and the boreholes selected by the Alberta Research
Council (ARC) to draw the graphical existence of the Edmonton geology (McPherson and Kathal 1972) were collected. The geology of the Edmonton area is categorized into
11 major soils: bedrock (soil 1), ice-shoved bedrock (soil 2),
disturbed Saskatchewan gravels and sands (soil 3), Saskatchewan gravels and sands (soil 4), glacial till (soil 5), glacial
sand and gravel (soil 6), lacustro-till (soil 7), glaciolacustrine deposits 1 and 2 (soils 8 and 9), aeolian deposits (soil
10), and alluvium (soil 11). The detailed description of these
soil types and general observations can be found in
Ruwanpura et al. (2001b).
Analyzing and dividing the data into unique groups
There are different soil types that exist vertically at discrete elevations between the surficial soil and the bottommost layer at a given point within an area. During the analysis of the Edmonton soil data, the following observations
were made:
only unique surficial soil types exist in the area
certain soils exist as the bedding layer to some other soils
soils exist as pockets within another soil layer
some soils co-exist with other soils
These observations suggest that there could be many vertical soil profile variations from the surface to the bedrock
layer from one point to another in the area. That means there
are several unique types of soils that interact with other soil
types in the area. To identify these interactions, the concept
of a soil family (Fig. 2) was introduced. A soil family represents the order of the different soil types that exist vertically at discrete elevations commencing from the surficial
soil to the bottom-most layer at a given point within an area.

Can. J. Civ. Eng. Vol. 31, 2004

According to ARC data, there are 59 soil families in the Edmonton study area. The most common families are given in
Table 1. The most common family in the Edmonton area is
851, which consists of soil type 8 (Lake Edmonton) the
surficial soil type, followed by soil type 5 (glacial clay till)
and soil type 1 (bedrock 1). This step provided a sound database of soil data in the Edmonton area.
Clustering the soil families
Any area can be divided into various soil clusters (groups)
depending on different dependant variables (Isaaks and
Srivastava 1989). In this study, the area was divided into various clusters based on the surficial soil type, irrespective of
the soil types under the surficial soil layer. However, there
could be several families of soils under one surficial soil; Table 2, for example, lists all the soil families where soil 8 is
the surficial soil. In Edmonton, there are 7 different soil
clusters: cluster 5 (CL5), cluster 6 (CL6), cluster 7 (CL7),
cluster 8 (CL8), cluster 9 (CL9), cluster 10 (CL10), and cluster 11 (CL11). There are 9 different soil families under CL5,
4 soil families under CL6, 10 soil families under CL7, 30 soil
families under CL8, and 1 soil family each under CL9, CL10,
and CL11 for the Edmonton study area. This clustering allows researchers to identify the coexistence of some soil
families with others.

Developing algorithms to predict the soil


families in the tunnel path
Using the steps explained above, it could be inferred that
a particular soil exists in the neighbourhood of a particular
borehole, although the borehole does not show any sign of
this soil. Algorithms were developed that are documented in
Ruwanpura et al. (2001b) to calculate the probability of the
coexistence of a particular soil family in the neighbourhood,
as shown in the boreholes, with respect to the other soil families within the tunnel area. The notation used to describe a
soil family is a list of numbers describing the soil layers in a
borehole from surface to bottom. For example, a borehole
with soil family 85651 has soil type 8 as its surficial soil,
with a layer of soil type 5 underneath, then a layer of soil
type 6, another layer of soil type 5, and a bottom layer of
soil type 1. The following equations explain the specific algorithms used in this calculation.
An example will serve to better illustrate the calculations.
Figure 3 shows several soil families in the boreholes in proximity to the tunnel that is 3440 m long. The families are 851,
751, 51, 75651, 7561, 85651, and 651 in the respective boreholes. The coexistence probability (PCEBase) of a particular
soil family (FamilyBase) in the neighbourhood of a borehole
in the tunnel area is dependent on the other families (Family1 to Familyn) in the boreholes (BH1 to BHn) and can be
calculated using eq. [1]. If FamilyBase is the same as the soil
family in the borehole, the PCEBase value is equal to 1 irrespective of eq. [1].
[1]

PCE Base, BHi =

F(Base) soil family A


F(Base) soil family + F(Base) soil family
2004 NRC Canada

AbouRizk et al.

349

Fig. 2. Soil families.

Table 1. Most common soil combinations (families) stratigraphy of Edmonton.


Total No. of
observations

Layer 1

Layer 2

Layer 3

Layer 4

Layer 5

Layer 6

79
76
18
14
8
6
6
5
5

8
8
5
7
8
8
8
8
7

5
5
1
5
5
7
5
5
5

1
3

1
6
5
2
6
6

5
1
5
5
5

1
3
1

where F(Base)soil family A is the number of occurrences of the


base soil family A in the database within a given area, and
F(BHi)soil family is the number of soil family occurrences in
borehole i that exist within a given area.
The following example illustrates the PCE for soil family
51 with respect to a borehole in a typical tunnel path, BH2,
which contains soil family 751:
F(Base)51 = 18 (see Table 1, line 3)
F(BH2)751 = 14 (see Table 1, line 4)
PCE

51,BH2

points along the tunnel path provide a distribution of a particular soil family from the start of the tunnel to the end of
the tunnel.
n

[2]

WFT, soil family =

1/dBH PCE BH, soil family A

BH =1

1/dBH

BH =1

= 18/32
= 0.5625

The next step is to calculate a weighted factor (WF) for a


particular soil family at a target point within the tunnel trajectory using the distance of the boreholes to the target point
and the calculated PCE values for a particular soil family in
the respective boreholes. The weighted factor at target point
T (WFT) for soil family A in the tunnel direction could be
calculated as per eq. [2] where dBH is the distance from
borehole (BH) to the target point T and PCEBH, soil family A is
the PCE value for soil family A at borehole BH. The use of
distances gives a higher weighted value if the target point is
closer to the borehole. The weighted factors at the target

The final step is to calculate the weighted coexistence


value (WCEV) for all soil families at each target point in the
tunnel trajectory using eq. [3]. The minimum (WFMinimum)
and maximum (WFMaximum) values for all WF values calculated using eq. [2] for each soil family are taken into consideration in calculating the WCEV for each soil family at each
target point. The WCEV for soil family A at target point T
can be calculated as per eq. [3]. The maximum value of
WCEV for soil families becomes the predicted soil family at
each target point. Equation [3] is a transformation formula to
normalize weighted factors calculated using eq. [2] and to
identify the optimum existence of the soil families along the
tunnel path.
2004 NRC Canada

34.80
33.48
3.96
3.08
2.64
2.64
2.20
1.76
1.76
1.76
1.76
1.32
1.32
0.44
0.44
0.44
0.44
0.44
0.44
0.44
0.44
0.44
0.44
0.44
0.44
0.44
0.44
0.44
0.44
0.44
100.00

Percentage

Total No.
79
76
9
7
6
6
5
4
4
4
4
3
3
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
227
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8

Layer 1
5
5
5
7
5
5
5
7
5
9
6
9
1
5
5
5
5
5
5
6
9
6
7
7
7
7
7
9
9
5

Layer 2

Table 2. Borehole details starting from soil type 8 (CL8).

1
3
6
5
2
6
6
5
2
1
5
5

4
6
3
2
3
6
1
8
5
5
5
5
6
8
5
5
6

Layer 3

1
5
1
5
5
1
3
3

3
3

5
5
5
5
5
5

9
1
2
6
6
1
5
1
6
5

Layer 4

1
3

1
1

1
1

1
2
1
3
3
2

5
1
5

5
6

Layer 5

1
1
5

3
5

Layer 6

1
2

Layer 7

Layer 8

Layer 9

350
Can. J. Civ. Eng. Vol. 31, 2004

2004 NRC Canada

AbouRizk et al.

351

Fig. 3. Sample tunnel and borehole locations (ARC data).

[3]

WCEVT,

=
WFT, soil family A WFMinimum, soil family A

soil family A

WFMaximum, soil family A WFMinimum, soil family A


Returning to our example, there are several soil families
in the boreholes in proximity to the 3440 m tunnel, as illustrated in Fig. 3. Figure 4 shows the predicted soil families
along the length of the tunnel based on WCEV values. The
results from Fig. 4 indicate the following soil family sequence for the tunnel: 751 (0 to 220 m), 51 (221 to 479 m),
7561 (480 to 719 m), 51 (720 to 823 m), 75651 (824 to
998 m), 85651 (999 to 1794 m), 651 (1795 to 2420 m),
85651 (2421 to 2761 m), and 851 (2762 to 3443 m). At this
stage, results are reevaluated using knowledge based on the
area by looking at previous, present, and following families
(e.g., 751, 51, 7561) in the area to determine the probability
of the sequence. For example, it is improbable that soil family 85651 would exist between soil family 851 and soil family 651 according to the knowledge base. These observations
amend the predicted results and are used to determine the
next best alternatives. The amended and final sequence is
751 (0 to 220 m), 51 (221 to 823 m), 75651 (824 to 998 m),
651 (999 to 2420 m) and 85651 (2421 to 3440 m) in the tunnel trajectory.
Applying the concept to a tunnel construction project
This section details the application of the modeling concepts to the North Edmonton sanitary trunk (NEST) tunnel
project in Edmonton. The NEST tunnel project initially had
few boreholes driven along the 1650 m-long tunnel path.
Figure 5a shows the location of the initial boreholes driven
for the project and the additional borehole data obtained
from the database created using ARC data close to the tunnel path. The boreholes driven by the City of Edmonton are
denoted with TH and the ARC boreholes are denoted with
BH.

The methodology presented in this paper was applied to


determine the existence of noncontinuous soil types in the
last 708 m of the tunnel. The main soil types available in the
tunnel path are glacial clay till (soil type 5), reworked clay
shale (soil type 2), and sand pockets (soil type 6). Boreholes
TH99-1 and TH99-2 contain soil family 8565251, boreholes
TH99-3 and BH287 contain soil family 85251, and boreholes TH99-4, TH6-2, and BH2115 contain soil family 851.
The last 708 m, as shown in Fig. 5a, contain several noncontinuous soil types including soil 6 and 2 in boreholes TH99-1
to TH99-3. The elevation of both soil types 2 and 6 are
within the dimensions of the tunnel or just below the bottom
elevation of the tunnel. Based on the borehole data, it could
be assumed that soil type 6 exists continuously from some
point left of borehole TH99-2 to the right of borehole TH99-1
(Fig. 5b). The borehole data also suggest that soil type 2
does not exist in the tunnel path. Figure 5b shows one of the
most likely profiles of soil types 2 and 6 between the boreholes using linear approximations or interpolations a typical industry practice.
Only soil family 851 was present in both the City of Edmonton and the ARC boreholes for the first portion of the
tunnel. However, the ARC borehole BH287 containing soil
family 85251 is about 370 m away from the tunnel. Since
BH287 has no evidence of the existence of soil type 6, and
the distance between TH99-1 and TH99-2 is approximately
400 m, the prediction of soil families was performed to discover the following:
1. the extent of soil type 6 between TH99-1 and TH99-2
2. the extent of soil type 2 between TH99-1 and TH99-3
An analysis was performed for the last portion of the tunnel using the methodology described above and submitted to
the design and construction department of the City of Edmonton. The conclusions are represented visually in Fig. 5c.
There was no need to analyze the rest of the tunnel, as there
was no evidence to show that there could be noncontinuous
soil types except in the last 708 m.
1. soil family 8565251 exists only about 54 m from borehole TH99-1 to borehole TH99-2 and for only 67 m in
the area of borehole TH99-2. This confirms that soil
type 6 is not a continuous soil layer between TH99-1
and TH99-2
2. soil family 85251 is the most likely soil family for
264 m from borehole TH99-3 to the end of the tunnel
and for 321 m between boreholes TH99-2 and TH99-1
The analysis also recommended that the City of Edmonton consider exploring the soil conditions further, especially
between boreholes TH99-1 and TH99-2. This is motivated
by the demand to deal with the fundamentals of uncertainty,
i.e., to collect more information but based on predictive
modeling outputs.
The City of Edmonton further explored the soil conditions
along the tunnel path by driving four additional boreholes
(TH00-1 to TH00-4), as shown in Fig. 6a. Borehole TH00-3
contains soil family 851, whereas all other boreholes have
soil family 85251. The borehole TH00-2 confirmed the prediction that soil type 6 does not exist continuously from
TH99-2 to TH99-1, although both TH99-1 and TH99-2 contain soil type 6. Further, the existence of soil family 85251
in boreholes TH00-4 and TH00-2 confirmed that soil type 2 is
continuous from left of TH99-3 to TH99-1. This prediction
2004 NRC Canada

352

Can. J. Civ. Eng. Vol. 31, 2004

Fig. 4. Weighted coexistence value (WCEV) of the soil families for sample tunnel.

and analysis is a clear indication of the validity of the methodology proposed for tunnel construction projects. Engineers
can use this method to explore the soil types along the tunnel path in addition to the usual geological explorations performed.
Ruwanpura and AbouRizk (2001) briefly discuss the soil
transitions along the tunnel trajectory, which is explained in
the next section. Another method to handle uncertainty explains how the prediction of soil families could assist in deducing the soil profiles between the boreholes logically to
determine the soil transitions along the tunnel path. Based
on the prediction and the two new boreholes, the tunnel
length between boreholes TH99-3 and TH99-1 can be divided
into seven sections. These seven sections are shown in
Fig. 6b and are listed below: (i) section 1 165 m in soil
family 85651, (ii) section 2 99 m in soil family 85251,
(iii) section 3 30 m in soil family 8565251, (iv) section
4 37 m in soil family 8565251, (v) section 5 271 m of
soil family 85251, (vi) section 6 52 m of soil family
85251, and (vii) section 7 54 m of soil family 8565251.
Using the prediction analysis of the soil families, it is possible to reduce the uncertainty of the soil conditions between
the boreholes by identifying the extent of each soil family.
The analysis proved that soil type 6 does not exist continuously. Because of the analysis and the availability of new
boreholes, it is easy to determine the soil profiles between
the boreholes more accurately and logically. Figure 6b
shows the most likely profiles of soil 2 and 6 within soil
type 5 along the tunnel path based on the soil family prediction methodology. Figure 6b can be divided into several segments to determine the transition from one soil to another
and the modeling algorithms for these soil transitions. These
algorithms can then be implemented using special purpose
simulation, as explained in the next section. The results
shown in this section further prove that this technique is very

useful in predicting the noncontinuous soil layers (soil pockets) along the tunnel path.
Predicting soil transitions along the tunnel path
Predicting soil transitions along a tunnel path is a challenging task. The boreholes driven for a tunnel construction
project only provide a handful of deterministic data points at
discrete locations in either the tunnel alignment itself or adjacent to the tunnel trajectory. The borehole data determine
soil types at discrete locations and produces deterministic
estimates of the type of material and the elevations of each
of the soil layers in the boreholes. Predicting soil compositions between the boreholes are generally achieved using approximate methods, as demonstrated by Ruwanpura et al.
(2001a).
A major deficiency of approximate methods is the determination of transition points from one soil type to another
when soil composition is mixed (e.g., clay material and
sand). This section describes an approach for modeling the
transition of soils between the boreholes for simulation purposes, and the hightlights of this approach are as follows:
1. develop an approach for calculating the transitional
probabilities to determine the transition from one soil
type to another in the tunnel trajectory
2. develop modeling algorithms based on the soil transition
patterns included in the database of soil transition scenarios
3. design of a tunnel construction simulation tool to incorporate the modeling algorithms
4. apply the tunnel simulation tool to an actual project to
validate its accuracy
Modeling transitional probabilities
Figure 7 shows a typical scenario of soil type transitions
among three borehole locations along the path of a hypothet 2004 NRC Canada

AbouRizk et al.

353

Fig. 5. (a) Preliminary boreholes; (b) approximate estimate of the soil types between the boreholes; and (c) analytical estimate of the
soil types between the boreholes.

ical tunnel. A simple two-state (soil type A, soil type B)


Markov chain can be used to determine the occurrence of
soil type A or B. Figure 7 shows that the transition from soil
type B to A at both the top elevation (elevationTop) and bottom elevation (elevationBottom) of the tunnel between boreholes BH1 and BH2. This two-state Markov chain is defined
by transitional probabilities of moving from one soil state to
another soil state. Transitional probabilities are dependant
on the location of the tunnel, diameter of the tunnel, start
and end elevations of the soil types, and the soil types in the
vicinity. Equation [5] defines the transitional probability of
transiting from state b (soil type B) at one known point
(BH1) in the tunnel to state a (soil type A) at a known point
(BH2) at the top elevation of the tunnel (elevationTop).
The probability of observing soil type A (state a) in the
tunnel path at elevationTop can be calculated using eq. [4]

[4]

PTop( A) =

N Top( A)
TTop

where NTop(A) is the number of observed transitions in soil


type A at elevationTop and TTop is the total number of all soil
types observed along the tunnel path at elevationTop.
There are two overall probabilities that define the model
in detail for one elevation, which are listed below:
PTop (A): overall probability of having soil type A at elevationTop
PTop (B): overall probability of having soil type B at elevationTop
[5]

PTop(B / A) =

N BA
nB

where NBA is the number of observed transitions generated


from soil type B to A at elevationTop along the tunnel path.
The number of observed transitions generated from soil
type B to all states including soil type A and B at elevationTop is denoted by nB. When two types of soil interact
within the tunnel alignment, four transitional probabilities
must be defined:
2004 NRC Canada

354
Fig. 6. (a) Location of all boreholes for NEST tunnel and (b) assumed soil profiles between the boreholes.

Can. J. Civ. Eng. Vol. 31, 2004


Fig. 7. Transition of soil type A and soil type B between boreholes.

Table 3. Transitional probabilities for the tunnel in


Fig. 8.

T
T
T
C
B+
B
B

PTop, BH2 (A/A): Probability of soil type A at location BH2


in the tunnel path at elevationTop given that soil type A can
be observed at location BH1 at elevationTop
PTop, BH2 (A/B): Probability of soil type B at location BH2
in the tunnel path at elevationTop given that soil type A can
be observed at location BH1 at elevationTop
PTop, BH2 (B/A): Probability of soil type A at location BH2
in the tunnel path at elevationTop given that soil type B can
be observed at location BH1 at elevationTop
PTop, BH2 (B/B): Probability of soil type B at location BH2
in the tunnel path at elevationTop given that soil type B can
be observed at location BH1 at elevationTop
Rules in transitional probability matrices for soil
prediction (two continuous soils)
Figure 8 is used to explain the modified transitional probabilities rules using a hypothetical example of a tunnel with
eight boreholes. There are only two soil types in the vicinity
of the tunnel. The transitions are enumerated separately in
seven elevation levels. The elevation level considers any gradients (if any) in the tunnel from its start to the end. The following are the elevation levels, which reflect the soil
combinations in the tunnel vicinity: (i) top of the tunnel elevation (T), (ii) bottom of the tunnel elevation (B), (iii) center
of the tunnel elevation (C), (iv) mid-point between top and
center of the tunnel elevation (T), (v) mid-point between
center and bottom of the tunnel elevation (B+), (vi) one point
(user-determined) above the top of the tunnel elevation (T+),
and (vii) one point (user determined) below the bottom of
the tunnel elevation (B).
Requirements (vi) and (vii) are user inputs and could be
determined based on the distribution of the soil profiles between the start and end of the tunnel. In this example, it is
limited to 1 m just below and above the tunnel to verify that

P(A/A)

P(A/B)

P(B/A)

P(B/B)

0.83
0.75
0.75
0.67
0.50
0.50
0.00

0.17
0.25
0.25
0.33
0.50
0.50
1.00

1.00
0.67
0.67
0.50
0.40
0.40
0.17

0.00
0.33
0.33
0.50
0.60
0.60
0.83

the transitions may consider the possible occurrence of soil


types closer to the top and bottom elevations of the tunnel.
Table 3 represents the transitional probabilities of the following combinations for the tunnel depicted in Fig. 8, based
on eq. [4]: T+ = AAAABAAA, T = BAAABBAA, T =
BBAABBAA, C = BAABBBAA, B+ = BABBBBAA, B =
BABBBBAA, and B = BBBBBBAB.
The transitional probabilities in the matrix are stationary
(or homogeneous) for the tunnel as the prediction analysis
uses these probabilities throughout the entire tunnel. It represents the probability of moving from one soil type to another
or remaining in the same soil type.
For example, elevation T generates all four transitional
probabilities. There are three transitions from A to A and
one transition from A to B with a total of four transitions
generated from A. This prompts the calculation of the transitional probability of A to A as 0.75 (3 of 4 transitions) and
transitional probability of A to B as 0.25 (1 of 4 transitions).
The transition probability of B to A is then 0.67 (2 of 3 transitions) and B to B is 0.33 (1 of 3 transitions).
Three transition probability matrices are created from the
data in Table 4: top matrix to calculate the transition point at
the top elevation; center matrix to calculate the transition of
soils in the middle of the tunnel, if any; and bottom matrix
to calculate the transition point at the bottom elevation. The
top matrix is made up of the transitions at elevations T+, T,
and T; the center matrix is made up of the transitions at elevations T, C, and B+; and the bottom matrix is made up of
the transitions at elevations B+, B, and B.
Modeling algorithms based on the transition scenarios
of the soils
There are several soil combinations that make up the stratigraphy of any area. The method explained above intro 2004 NRC Canada

AbouRizk et al.

355

Fig. 8. Transitions of soils in the tunnel.

duced the various families of soils in stratigraphy of


Edmonton. The modeling algorithms for identifying soil
transitions vary according to the following factors:(i) number of soils in the area, (ii) start and end elevations of the
soils, (iii) direction of the soil profiles, (iv) status of the soil
types (continuous or pockets), and (v) start and end elevation
of the tunnel between boreholes.
A sample database of the various combinations of two
continuous layer soils, which is used for simulation modeling, is shown in Fig. 9. Five soil-transition-modeling combinations can be derived from Fig. 9 based on the soil
transitions for the sample tunnel, as shown in Fig. 8:
1. Transition of one soil to another (soil type B to soil type
A) at both the top and bottom of the tunnel (for example, BH1 to BH2 and BH6 to BH7). This involves predicting the transition points at both the top and bottom
elevations of the tunnel.
2. Transition of soil type A to soil type B only occurs at
the top elevation (BH4 and BH5), requiring the prediction of the top transition point.
3. Transition of soil type A to soil type B only occurs at
the bottom (BH2 and BH3), requiring the prediction of
the bottom transition point.
4. There are no soil type transitions either at the top or
bottom of the tunnel. Transitions only occur between
the top and bottom elevations of the tunnel (BH3 and
BH4), requiring no transition point predictions at the top
and bottom of the tunnel.
5. There are no soil type transitions at the top, bottom, or
inside the tunnel elevations. Transitions only occur
above or below the tunnel (BH5 to BH6 and BH7 and
BH8) requiring no transition point predictions at the top
and bottom of the tunnel.
Although some of the scenarios in Fig. 9 are very similar
in terms of the transitions at top and bottom, the location of
the transition points could make a difference in determining
the productive boring rate. A complete list of all scenarios
including the soils that are not continuous (e.g., pockets of
sand within clay) can be obtained from the database of scenarios included in Ruwanpura (2001). Algorithms to calculate the transition points for two continuous soil layers vary
according to many factors in addition to transitional probability values, including the direction of the soil transition
profile, number of top transitions, and number of bottom

transitions. Figure 10 shows the module for scenario 1 of


Fig. 9.

Special purpose simulation tool based on


soil transition algorithms for tunnelling
The modeling algorithms documented above are implemented within a tunnel simulation tool described in
Ruwanpura et al. (2001a). The simphony simulation engine
documented in Hajjar and AbouRizk (2000) provides a flexible and easy-to-use modeling environment to implement the
soil transition concepts. The existing tunnel simulation template in simphony was embellished with additional modeling
elements and algorithms without losing its originality. The
major highlights of the embellishments are given below:
1. The ability to input the soil types based on the present
observation point and the next immediate observation
point along the tunnel path, and the ability to add data
about the appropriate transitional probability values between the two soil types to determine the transition
point at the top or bottom elevation of the tunnel.
2. The ability to input a user-defined method of calculating
the boring rate for soils. Based on the data survey and
interviews with tunnelling experts, the final tunnel construction boring rate may be calculated using one of the
three methods mentioned below, if boring is performed
in two soils:
(i) Percentage weighting based on the composition of
the soils. If soil type A represents 30% of the soil and
soil type B represents 70%, the boring rate is adjusted to
be 30% of the boring rate of A plus 70% of the boring
rate of B.
(ii) The lowest boring rate of the two soils.
(iii) Worse than the minimum boring rate of the two
soils. Tunnel personnel identified this as the most common situation, although the City of Edmonton cannot
justify it with supporting data.
3. The ability to model different transition combinations
involving two continuous soils, three continuous soils,
and continuous soils with soil pockets.
4. The ability to add the data, such as soil type and elevation of the soils, from the actual boreholes driven for a
particular project.
2004 NRC Canada

356

Can. J. Civ. Eng. Vol. 31, 2004


Table 4. Transitional probability matrices top, center, and bottom.
Top
Soil type A
Transitional probabilities
Soil type A
0.786
Soil type B
0.714
Transitions
Soil type A
11
Soil type B
5

Center

Bottom

Soil type B

Soil type A

Soil type B

Soil type A

0.214
0.286

0.667
0.500

0.333
0.500

0.400
0.313

3
2

6
6

3
6

2
5

Soil type B
0.600
0.688
3
11

Fig. 9. Some combinations of two continuous soil layers in the database.

2004 NRC Canada

AbouRizk et al.

357

Fig. 10. Algorithm module for scenario 1 in Fig. 9.

Application of the tunnel simulation


template to an actual project
This section presents a case study to validate the described modeling algorithms. The case study uses actual productivity data from a tunnelling project that was completed
in 19941995. The data were obtained from the daily report
logs and through consultation with the site supervisor and
the site engineer. The tunnel length used for validation purposes is 1651 m and the elevation varies from 675.79 m at
the tunnel start to 671.31 m at the tunnel end. This particular
tunnel was excavated in bedrock comprising two soil types:
shale and sandstone. The tunnel employed two separate construction methods. The first portion used for validation was a
2.9-m finished diameter tunnel excavated using an M-126
Lovat TBM lined with precast concrete segments. The second portion, which was 700 m long, was a 3.48-m finished
diameter tunnel lined with shortcrete. The first portion has
been selected for analysis as the tunnel simulation template
has been designed to simulate tunnels lined with precast
liner segments. The tunnel is about 2025 m below ground
level and has a gradient of 0.077% from the entry shaft to
the removal shaft. There are roughly 203 m of curved section starting at the 687th m of the tunnel. Although there
were 24 boreholes in the tunnel trajectory, only 17 boreholes
were driven in the first portion of the tunnel. Two of these
boreholes were very shallow and did not represent the soil
types in the tunnel elevation. Figure 11 illustrates the length
between the boreholes and the estimated soil combination
scenarios.
Several alternative models were created using the modeling approach presented in this paper with different calculation methods to determine the boring rate. Six models were
created, of which three used the approximate soil composi-

tions as per the method outlined in Ruwanpura et al. (2001a,


2001b). The best approximation is the linear interpolation of
the soil conditions between the boreholes, which uses fifteen
separate soil modeling elements with approximate values.
The other three models were created using the new modeling
approach with different calculation methods to determine the
boring rate, as explained in the previous section of this paper. The results of all six models were tested against the actual tunnel productivity. Since the final productivity of the
sixth model, which is based on the assumption that the boring rate is worse than the minimum boring rate of the two
soils (sandstone in this case), is very close to the actual productivity, Fig. 12 compares the results of the actual tunnel
advance rates to those predicted based on the model presented. The overall project productivity is very close to the
predicted productivity rates of this model. Up to about the
first 300 m, the actual tunnel advance rate is far below the
predicted-tunnel advance rate of the simulation model because of the learning curve typical of any project. For the remainder of the tunnel, the actual-tunnel advance rate
remains close to the simulated-tunnel advance rates. This
comparison indicates that the proper selection of boring rate
inputs could provide a more accurate prediction of tunnel
construction productivity.

Conclusions
The methods to reduce uncertainty explained in this paper
provide some logical input for the industry practitioners to
plan future tunnel construction projects. These methods not
only provided a database of borehole information collected
from two sources but also a logical method to assess uncertainty about soil types between the boreholes. Users can implement these methods using simulation to predict the
2004 NRC Canada

358

Can. J. Civ. Eng. Vol. 31, 2004

Fig. 11. Profiles of the actual tunnel and its transition scenarios.

Fig. 12. Tunnel advance rate actual versus new modeling approach.

various possible transitions between the boreholes, which reduce the uncertainty in assuming the soil content between
the boreholes.
The prediction of soil types has been identified during the
implementation of the special purpose simulation template
as a major factor in reducing uncertainty and improving tunnel construction operations productivity. The soil prediction
methodology uses an analytical approach in predicting soil
types that are beneficial to the project planners and engineers. The analysis is presented with the goal of demonstrating how this analysis could be useful for construction
purposes. This analysis also compliments the geotechnical
explorations conducted for tunnel construction projects as
tested in the NEST project. After the preliminary soil characterization, the concept of soil families was introduced
along with soil clusters commencing from the surficial soil
type in any given area. The probability analysis predicted the
combination of soil families and their distribution, and
thereby identified the existence of soil families along the
tunnel path. This prediction method provides insight into
two areas. First, since the prediction of the soil families is

established using an analytical method, it will provide further research to developing an analytical method to predict
the elevation of the soils using the concept of soil families.
An accurate prediction of soil elevations could further allow
the end users to determine the distribution of the soil profiles between the boreholes. Second, this method enables the
project engineers to further analyze the geological explorations for construction purposes.
The NEST tunnel analysis also proved how the prediction
of soil families could aid in accurately assessing the soil
profiles to determine the soil transitions along the tunnel
path for simulation purposes. The use of this analysis for actual tunnel construction will reduce the uncertainty of the
project by logically investigating the soil types in the tunnel
path before beginning actual construction. It also provides
an opportunity for the construction engineers and managers
to get acquainted with possible soil type occurrences along
the tunnel path that can be useful for applications in project
scheduling and estimating for tunnel construction operations. Although the application of this method was limited to
Edmonton geology, the method can be used for any other
2004 NRC Canada

AbouRizk et al.

city or area provided that the city or area has adequate published data or borehole data from past tunnel construction
projects.
This paper also presented the development of algorithms
to predict the soil transition points along a tunnel path using
transitional probabilities. This new approach predicts the
transition points along the tunnel using an analytical method
rather than using approximations or assuming arbitrary transition points. The soil composition and boring rates are calculated, based on the transition points, to arrive at the tunnel
construction productivity. Several soil transition combinations, which are implemented within a special purpose tunnelling simulation template for many scenarios, have been
presented. With this new approach, the end users can specify
the borehole data rather than approximating soil data for a
specific tunnel section. Based on the soil data input and the
user inputs, the template determines the best modeling scenario between the two boreholes and predicts soil transition
points and productivity values. This method also enables the
end users to specify the boring rate calculation method for
production purposes. The tunnel construction productivity is
determined through an analytical method based on the soil
transition points along the tunnel. The validated case study
proved that these modeling algorithms not only provide a
logical approach to predicting productivity based on the
transition of soils but also provide an accurate prediction
given the fact that the end user inputs the actual data. The
successful development and application of the soil transition
modeling algorithms, thus, reduce the risk and uncertainty in
predicting the tunnel advance rate and productivity. The application of these algorithms within special purpose simulation to future tunnel construction projects will provide better
project planning and decision-making for engineers before
actual construction begins.
The prediction of soil families and soil transitions could
also be extended to more advanced levels. The analysis
shown in this paper limits the prediction to major soil types.
However, there are various minor soil types within a major
soil category. For example, shale, clayshale, sandstone, bentonite, and siltstone are within the bedrock major category.
The prediction of these individual soil types along a tunnel
path could provide better inputs for the simulation model.
Further, a soil has various properties such as plasticity, moisture content, compressive strength, and granularity, and the
boring rate would differ based on the soil properties. The
following are recommended for further research:
1. Extend this soil prediction study to minor soil types to
predict the probability of their existence and their elevations.
2. Study the properties of the soils and develop an analytical model that could be added to the SPS tunnel template to derive boring rates.
3. All the soil transition scenarios depicted in paper show
somewhat smooth transition curves from one borehole
to another. It is possible to assume that there are many
transitions between the two boreholes rather than a
smooth transition from one soil to another. As there is
an uncertainty in predicting the shape of the profile between the boreholes, further analysis is required to determine the exact nature of the soil profile between the
boreholes, as a future embellishment.

359

Hence, it is recommended that further analysis in this area


provides an opportunity to improve the assessment of the
soil families and transition patterns between two boreholes
that can be modeled using special purpose simulation. The
successful development and implementation of these tools
will reduce the uncertainty in predicting soil conditions
along the tunnel path and will thereby provide a more comprehensive tunnel simulation template that could be very
beneficial to academia and the industry.

Acknowledgements
This research was conducted under the Natural Sciences
and Engineering Research Council of Canada (NSERC)/Alberta Construction Industry Research Chair in Construction
Engineering and Management. This work was funded by a
number of Alberta construction companies and by the
NSERC under project CRDPJ 249188-01, Productionbased framework for construction planning and execution.

References
AbouRizk, S.M. 2000. A structured process for risk analysis and
management. Proceedings of the 7th Canadian Construction Research Forum, Edmonton, Alta., 18 September 2000. NSERC/
Alberta Construction Industry Research Chair and Construction
Research Institute of Canada, Edmonton, Alta. pp. 108116.
AbouRizk, S.M., and Hajjar, D. 1998. A framework for applying
simulation in construction. Canadian Journal of Civil Engineering, 25: 604617.
AbouRizk, S.M., Ruwanpura, J.Y., Er, K.C., and Fernando, S.
1999. Special purpose simulation templates for utility tunnel
construction. Proceedings of the Winter Simulation Conference,
Phoenix, Ariz., 58 December 1999. Institute for Operations Research and Management Sciences, Linthicum, Md. pp. 948955.
Concise Oxford dictionary, The. 1991. 8th ed. Edited by R.E. Allen. Oxford University Press, New York, N.Y.
Er, K.C., Fernando, S., and Ruwanpura, J.Y. 1999. Application of
special purpose simulation modeling for tunnel construction.
Proceedings of the 6th Canadian Construction Research Forum,
Edmonton, Alta., 25 October 2000. NSERC/Alberta Construction Industry Research Chair and Construction Research Institute of Canada, Edmonton, Alta. pp. 19.
Er, K.C., Fernando, S., AbouRizk, S.M., and Ruwanpura, J.Y.
2000. Selecting the best construction option for the 1st stage of
the SESS tunnel using simulation. Proceedings of the 7th Canadian Construction Research Forum, Edmonton, Alta., 18 September 2000. NSERC/Alberta Construction Industry Research
Chair and Construction Research Institute of Canada, Edmonton, Alta. pp. 7686.
Hajjar, D., and AbouRizk, S.M. 2000. Application framework for
the development of simulation tools. Journal of Computing in
Civil Engineering, 14: 160167.
Isaaks, E.H., and Srivastava, R.M. 1989. Applied geostatistics. Oxford University Press, New York, N.Y.
Lifson, M. 1972. Decision and risk analysis for practicing engineers. Cahners Books, Boston, Mass.
McPherson, R.A., and Kathal, C.P. 1972. Stratigraphic sections and
drill hole logs. Alberta Research Council, Edmonton, Alta.
Palisade Corporation. 1999. @Risk: risk analysis and simulation
add-in for Microsoft Excel. Newfield, New York, N.Y.
Ruwanpura, J.Y. 1999. Combined Markovian and fuzzy based simulation for construction operations. Proceedings of the 6th Cana 2004 NRC Canada

360
dian Construction Research Forum, Edmonton, Alta., 25 October
2000. NSERC/Alberta Construction Industry Research Chair and
Construction Research Institute of Canada, Edmonton, Alta.
pp. 2132.
Ruwanpura, J.Y. 2001. Special purpose simulation for tunnel construction operations. Ph.D. thesis, University of Alberta, Edmonton, Alta.
Ruwanpura, J.Y., and AbouRizk, S.M. 2001. Design development
and application of soil transitions algorithms tunnelling using
special purpose simulation. Proceedings of the Winter Simulation Conference, Arlington, Va., 912 December 2001. Association for Computing Machinery, New York, N.Y. pp. 15121520.
Ruwanpura, J.Y., AbouRizk, S.M., and Allouche, M. 2000. Predicting
geological conditions for special purpose tunnel construction simulation. Proceedings of the Seventh Canadian Construction Research

Can. J. Civ. Eng. Vol. 31, 2004


Forum, Edmonton, Alta., 18 September 2000. NSERC/Alberta
Construction Industry Research Chair and Construction Research
Institute of Canada, Edmonton, Alta. pp. 202213.
Ruwanpura, J.Y., AbouRizk, S.M., Er, K.C., and Fernando, S.
2001a. Special purpose simulation templates for tunnel construction operations. Canadian Journal of Civil Engineering, 28:
222237.
Ruwanpura, J.Y., Allouche, M., and AbouRizk, S.M. 2001b. Predicting of soil transitions for tunnel construction operations using special purpose simulation. Proceedings of the 4th CSCE
Construction Specialty Conference, Victoria, B.C., 30 May 2
June 2001. CD-ROM. Canadian Society for Civil Engineering,
Montral, Que.

2004 NRC Canada

Copyright of Canadian Journal of Civil Engineering is the property of Canadian Science Publishing and its
content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's
express written permission. However, users may print, download, or email articles for individual use.