Você está na página 1de 24

Computer-Aided Civil and Infrastructure Engineering 32 (2017) 748–771

Probabilistic Risk Analysis of Diversion Tunnel


Construction Simulation

Jia Yu, Denghua Zhong*, Bingyu Ren & Dawei Tong


State Key Laboratory of Hydraulic Engineering Simulation and Safety, Tianjin University, Tianjin, China

&

Kun Hong
State Key Laboratory of Hydraulic Engineering Simulation and Safety, Tianjin University, Tianjin, China and
SINOHYDRO Bureau 14 Co., Ltd., Kunming, Yunnan, China

Abstract: Comprehensive and effective risk analysis is factors existing during construction. These factors not
significant for studying construction simulation of diver- only cause common variability of construction perfor-
sion tunnel. Existing tunnel risk simulation approaches mance, but also may lead to major losses and delays in
mainly analyze ordinary risk factors, and cannot quanti- tunnel projects. Therefore, finding a more realistic as-
tatively study risk events considering their causes. Addi- sessment approach of associated risks is of vital signifi-
tionally, in other tunnel probabilistic risk analysis meth- cance for schedule estimations.
ods, although some studies have made full probabilistic According to Isaksson and Stille (2005), the potential
estimates of tunnel schedule, risk factors are unable risk factors during tunnel excavation can be divided into
to be studied considering cyclic construction character- two categories. Some factors, named ordinary risk fac-
istics and occurrence probability of risk events cannot tors, will cause deviation in a normal range, which can
be determined quantitatively. To address the issues, a be represented with continuous probability distribution.
probabilistic risk analysis approach of diversion tunnel The other, defined as risk events in this study, may lead
construction simulation is proposed. Based on hierar- to major delays and exert a serious impact on construc-
chical simulation model, risk factors can be analyzed tion processes, such as tunnel collapse. It is evident that
at operation level of tunnel construction. Moreover, both ordinary risk factors and risk events should be an-
Bayesian network is embedded into simulation program alyzed, comprehensively and effectively, for more real-
to quantitatively calculate probability of risk events in istic risk assessment.
each simulation cycle, considering geology, design, con- Simulation technology is thought to be an effective
struction, and management conditions and their mutual tool for analyzing the uncertainty of tunnel construc-
dependence. tion. In existing tunnel risk simulation approaches,
the impacts of ordinary risk factors are able to be
1 INTRODUCTION considered. Some researchers pay attention to factors
affecting tunnel excavation, especially geological con-
Construction of diversion tunnel plays a significant role ditions, in simulation. However, they mainly analyze
in increasing social effect and economic benefits of hy- the factors’ influence on simulation parameters or
draulic engineering. Schedule is key factor in construc- construction modes. The impacts of potential risk
tion planning of diversion tunnel. However, tunnel ex- events, such as tunnel collapse, under diverse geolog-
cavation is an uncertain process due to the potential risk ical conditions and other conditions including design,
∗ To construction, management, etc., are not studied. It will
whom correspondence should be addressed. E-mail:
dzhong@tju.edu.cn. affect the comprehensiveness of risk analysis. In other


C 2017 Computer-Aided Civil and Infrastructure Engineering.

DOI: 10.1111/mice.12276
Probabilistic risk analysis of diversion tunnel construction simulation 749

tunnel probabilistic risk analysis methods, although from past case histories or from the early stages of a
some studies have made full probabilistic estimates of project. Vargas et al. (2014, 2015) proposed a simulation
tunnel schedule, the two kinds of risk factors cannot algorithm based stochastic probabilistic method, and
be analyzed considering cyclic features of tunnel con- adopted Monte Carlo method or Markov Chain Monte
struction. Especially, the occurrence probability of risk Carlo (MCMC) approach for decision making in the
events is unable to be determined quantitatively taking tunnel planning process. Rahm et al. (2016) evaluated
their causes into account. In light of this, a probabilistic the impact of three types of disturbances in mechanized
risk analysis method of diversion tunnel construction tunneling, including production disturbances, supply
simulation is proposed for realizing the comprehensive chain problems, and cascading disturbances. Construc-
and effective risk analysis of tunnel construction. With tion uncertainties are considered in these tunnel simu-
this method, two kinds of risk factors are allowed to lation approaches, which are mainly reflected with de-
be analyzed at operation level, including measurement, fined probability distributions of simulation parameters.
drilling, blasting, and other operations composing Some simulation approaches focus on study of the
tunnel excavation cycle. In addition, Bayesian network factors affecting tunnel excavation, especially geological
is embedded into simulation program to quantitatively conditions. Decision Aids for Tunneling (DAT) consid-
calculate probability of risk events in each simulation ered the uncertainties in geology, which were modeled
cycle, considering geology, design, construction, and by means of a Markov process (Chan, 1981), and used
management conditions and their mutual dependence. Monte Carlo simulation for probabilistic prediction of
Simulation easily lends itself to tunnel project plan- construction time, costs, and consumption of resources.
ning on a theoretical level because tunneling is both Likhitruangsilp and Ioannou (2003) considered the fact
repetitive and linear (Hajjar and AbouRizk, 1997). that tunnel excavation and support methods varied with
The start of modern construction simulation languages the change of geological conditions, and presented a
was led by Halpin (1977) with the introduction of stochastic methodology for evaluating tunneling per-
cycle operation network (CYCLONE). Then, diverse formance using discrete-event simulation performed in
simulation tools were developed, embracing RESQUE ProbSched. Bi et al. (2014) embedded Markov predic-
(Chang, 1986), UM-CYCLONE (Ioannou, 1989), tion model into CYCLONE model to consider litholog-
Micro-CYCLONE (Halpin, 1990), COOPS (Liu, 1991), ical uncertainty and automatically estimate construction
DISCO (Huang, 1994), STROBOSCOPE (Martinez parameters for each excavation cycle of long-distance
and Ioannou, 1994), Simphony (AbouRizk and Mo- diversion tunnels. These studies mainly consider the im-
hamed, 2000), activity-based construction (ABC) (Shi, pact that uncertainty of geological/geotechnical param-
1999), Vitascope (Kamat and Martinez, 2003), S3 (Lu eters (e.g., the type of rocks) exert on tunnel simula-
et al., 2008), COSYE (AbouRizk and Hague, 2009), tion parameters or excavation modes. The impact of
hybrid SD-DES (Alvanchi et al., 2011), etc. potential risk events, such as tunnel collapse, under di-
As an effective tool for planning repetitive construc- verse geological conditions and other conditions includ-
tion projects that involve uncertainties (Ruwanpura ing design, construction, management, etc., has not been
and Ariaratnam, 2007), simulation modeling has been taken into account.
widely used in forecasting schedules and analyzing risk Therefore, existing tunnel risk simulation methods
of tunnel projects. Ruwanpura et al. (2001) analyzed can realize the analysis of ordinary risk factors, but ne-
tunneling process of shielded tunnel-boring machines glect the impact of potential risk events on excavation
using a special purpose simulation tool developed with processes, which will affect the comprehensiveness of
Simphony. Liu et al. (2010) adopted CYCLONE model tunnel schedule risk analysis.
to predict the excavation schedule of Tunnel Boring In addition to simulation tools, other efforts have
Machine (TBM). Based on Simphony.NET, an inte- been made to probabilistically assess tunnel construc-
grated scenario-based solution for tunnel planning and tion risk. There are some studies that have made full
decision support was created using modular develop- probabilistic estimates of tunnel construction time,
ment and a high level architecture (HLA)-inspired com- taking both ordinary risk factors and risk events into
munications framework (Al-Bataineh et al., 2013). account. Isaksson and Stille (2005) and Grasso et al.
For accurate estimation of tunnel schedule, Chung (2006) took tunnel excavation section with certain
et al. (2006) and Zhang et al. (2014) updated the dis- length as object to study the influence of ordinary
tributions of simulation input parameters based on data risk factors, and proposed the idea of analyzing risk
collected from actual project. Rostami et al. (2014) stud- events from occurrence probability and consequence.
ied an advance rate simulation approach for hard rock However, specific approach for risk events analysis was
TBMs, which could predict the durations of all activities not introduced in their studies. Špačková et al. (2013)
in excavation based on their recorded time distributions utilized a combined probability distribution to consider
750 Yu et al.

the influence of common construction variability, and in this study for simulating tunnel cyclic excavation pro-
used database available in literature to analyze failure cess. Moreover, to reduce the number of nodes, hierar-
rate as well as consequence of geological risk events. In chical model coupling CPM and CYCLONE is adopted.
addition, Špačková and Straub (2013) defined unit ex- Because the variances in time caused by ordinary risk
cavation time with a conditional cumulative probability factors can be described with a continuous distribution,
distribution function for each construction method and appropriate probability distributions for measurement,
human factor, and described extraordinary events with blasting, ventilation, and other unit operations in CY-
failure mode and number of failures. Nevertheless, CLONE are determined with the BestFit technology.
there still exist some deficiencies when analyzing the In addition, occurrence probability analysis technology,
two types of risk factors. For ordinary risk factors, which takes the causes of risk events, is studied and
researchers mainly regard tunnel excavation section embedded into CYCLONE to facilitate quantitative
with certain length as object, which cannot analyze the analysis of risk events in each tunnel excavation cycle.
influence of these factors considering cyclic features Estimation of risk occurrence probability is critical in
of tunnel construction. It is because tunnel excavation tunnel risk analysis. Subjective estimation or statistical
section with certain length is composed of several exca- analysis are the most common methods (Shahriar
vation cycles, and each cycle can be divided into mea- et al., 2008; Sousa, 2010; Sousa and Einstein, 2012;
surement, drilling, blasting, ventilation, and other unit Špačková et al., 2013; Wu et al., 2015). Afterwards,
operations (Gonzalez Jimenez et al., 2015). It is be- other approaches enabling quantitative calculation of
lieved that studying the effect of ordinary risk factors at tunnel risk probability are developed, such as fault
operation level is more conducive to fine and effective tree, event tree analysis, risk matrices, etc. (Hong et al.,
analysis of the factors when compared with taking tun- 2009; Aliahmadi et al., 2011; Nývlt et al., 2011; Jurado
nel excavation section with certain length as research et al., 2012). As a common approach to quantitatively
object. In terms of risk events, occurrence probability analyze risk occurrence probability, Bayesian network
is usually obtained based on subjective estimation has been widely used in construction projects to study
or statistical analysis, and cannot be quantitatively slope collapse (Cheng and Hoang, 2014), fall risk of
calculated considering their causes. In addition, the cantilever bridge projects (Chen and Leu, 2014), deep
influence analysis of risk events on each excavation water drilling operations (Bhandari et al., 2015), etc.,
cycle has not been realized in existing studies. It will and has also been applied to other diverse research
affect the effectiveness of tunnel schedule risk analysis. fields (Sun and Bette, 2015; Wang et al., 2015; Yuen and
Therefore, an approach enabling comprehensive and Mu, 2015; Mu and Yuen, 2016). It not only combines
effective probabilistic risk study of diversion tunnel, the merits of fault tree and other risk probability
considering both ordinary risk factors and risk events, quantitative analysis methods, but also has distinct
is in urgent need to be proposed. advantages in describing dependence between vari-
As the first of modern construction simulation lan- ables quantitatively (Holický et al., 2013), and dealing
guages, CYCLONE is suitable for projects that consist with uncertainty information (Lee et al., 2008). These
of hundreds of repetitive cycles of tasks (Touran, 1987; merits make Bayesian network a more suitable tool
Halpin and Riggs, 1992). It is regarded as the widely in analyzing occurrence probability. Bayesian network
used and best-known discrete simulation method used has been used in risk analysis of tunnel construction
in construction engineering research (Martinez and (Zhang et al., 2014; Feng and Jimenez, 2015; Wu et al.,
Ioannou, 1999; Zhong et al., 2004). The satisfactory 2015). Nevertheless, these studies are unable to obtain
performance of CYCLONE in describing tunnel cyclic risk occurrence probability in each excavation cycle
construction is beneficial for the fine and effective by fully considering geology, design, construction, and
analysis of risk factors in tunnel excavation. In addition, management conditions and their mutual relationship,
CYCLONE is featured with simplicity and flexibility which are regarded as the main causes leading to tunnel
(Lau et al., 2014; Liu et al., 2014), and can nest within accidents (Sousa, 2010; Zhang et al., 2015). Therefore,
other tools (Bi et al., 2014; Zhang et al., 2014), which in this study, Bayesian network is embedded into
facilitates the combination of CYCLONE and other simulation program to calculate occurrence probability
methods, such as Bayesian network, for further analysis of risk events in each simulation cycle considering these
of ordinary risk factors and risk events. Furthermore, causes of the events.
the CYCLONE model has been effectively used in In summary, a probabilistic risk analysis approach of
forecasting tunnel schedules and analyzing risk (e.g., diversion tunnel construction simulation is proposed.
geological uncertainty) (Bi et al., 2014; Liu et al., 2014; It facilitates comprehensive and effective risk analysis
Zhong et al., 2016), which lays a solid foundation for the of tunnel construction, taking both ordinary risk fac-
study in this article. Therefore, CYCLONE is adopted tors and risk events into account. For ordinary risk
Probabilistic risk analysis of diversion tunnel construction simulation 751

factors, measurement, drilling, blasting, and other unit methods and parameters. The mathematical model is
operations in CYCLONE are regarded as study objects built to present the overall model architecture and logi-
to finely analyze the effect of these factors. For risk cal relationship between different methods. The mathe-
events, Bayesian network, fully considering geology, matical model of probabilistic risk analysis of diversion
design, construction, and management conditions and tunnel construction simulation is shown in Figure 2.
their mutual dependence, is embedded into the CY- The model is composed of four parts. Part  1 defines
CLONE model for quantitative analysis of the events in the theoretical method set M of the model framework,
each simulation cycle. The data and the application ex- including simulation method (MSIM ) and risk analy-
ample presented in this article are limited to the conven- sis method (MRISK ). MSIM can be subdivided into CY-
tional tunneling method, such as drilling and blasting CLONE modeling method (MCO ) and CPM network
method. The approach will be extended to the schedule planning modeling method (MCM ), which reflect the hi-
risk analysis of mechanized tunneling with use of TBM. erarchical characteristics of simulation structure. MRISK
The article begins with an overall research framework can be subdivided into analysis method of ordinary risk
presented in Section 2. Then, methodology of prob- factors (MOR ) and that of risk events (MER ). Part  2
abilistic risk analysis of diversion tunnel construction defines the parameter sets of simulation method, which
simulation is introduced in Section 3, including model- are the basic parameters of the model. Specifically, PCO
ing of the probabilistic risk analysis model, fine anal- is the parameter set of MCO . The input parameters refer
ysis of ordinary risk factors, and quantitative calcula- to parameters required to simulate tunnel construction
tion of risk event occurrence probability using Bayesian cycle considering ordinary risk factors and risk events,
network. Next, in Section 4, solving process based on including cyclic simulation parameters (ICO ), parame-
discrete event simulation (DES) and Monte Carlo simu- ters about the impact of ordinary risk factors (IOR ), and
lation is described in detail. Finally, the proposed proba- parameters about the impact of risk events (IER ). The
bilistic risk analysis approach is applied to an illustrative output parameters (OCO ) represent output of each sim-
case study, and the feasibility of the approach is verified. ulation cycle. For diversion tunnel, ICO embraces tunnel
structural parameters, blasting parameters, equipment
2 RESEARCH FRAMEWORK parameters, etc. IOR means the probability distributions
of durations of tunnel excavation operations. IER is on
The research framework of probabilistic risk analysis behalf of the occurrence probability and consequence
of diversion tunnel construction simulation is shown in of risk events in each cycle. OCO includes duration of
Figure 1. It consists of four layers, namely, (1) param- each footage driving cycle, equipment productivity, etc.
eter layer, (2) model layer, (3) method layer, and (4) PCM is the parameter set of MCM . The input parame-
index layer. ters include parameters required for schedule analysis,
Parameter layer embraces all the parameters needed such as simulation activities (ISIM ), nonsimulation activ-
for analysis, including simulation parameters, and pa- ity (INON ), and logical relationship between the activi-
rameters of ordinary risk factors and risk events. The ties (IR ). The output parameters (OCM ) consist of total
detailed explanation of input parameters can be found duration, completion probability of planned duration,
in Section 3. Model layer refers to the established activity critical index, etc. Because simulation activity
probabilistic risk analysis model of diversion tunnel is composed of several construction cycles, f1 reflects
construction simulation, which fully considers ordinary the relationship between MCM and MCO , and shows
risk factors and risk events during tunnel construction. the hierarchical characteristics of simulation structure.
Method layer consists of DES method and Monte Carlo Part  3 defines the parameter sets of risk analysis ap-

simulation, which are utilized to solve the model. The proach. These parameters determine the rationality and
detailed solving process is described in Section 4. Even- effectiveness of risk analysis method, and are the most
tually, construction duration, completion probability, important parameters in the mathematical model. POR
and activity criticality index (ACI) can be obtained for is the parameter set of MOR . The input parameters
tunnel schedule risk analysis. include statistical data of each operation time (IOR ),
which are collected from similar project or completed
part of the project, and output parameters (OOR ) re-
3 METHODOLOGY OF PROBABILISTIC RISK fer to distributions of operation durations. PER is the
ANALYSIS OF DIVERSION TUNNEL parameter set of MER . The input parameters represent
CONSTRUCTION SIMULATION parameters required for calculating occurrence proba-
bility of risk events with Bayesian network, incorporat-
The methodology of probabilistic risk analysis of diver- ing nodes of Bayesian network (INO ), joint probabil-
sion tunnel construction simulation consists of various ity distribution (JPD) of risk event (IJPD ), conditional
752 Yu et al.

Fig. 1. Research framework of probabilistic risk analysis of diversion tunnel construction simulation.

Fig. 2. Mathematical model of probabilistic risk analysis of diversion tunnel construction simulation.

probability table (CPT) between each node (ICPT ), and events (IOR ) can be analyzed with occurrence probabil-
parameters used for analyzing consequences of risk ity (OPRO ) and consequence (OCON ) of the events, f3
events (ICON ). When analyzing the risk events of di- reflects the relation between MCO and MER .
version tunnel, Bayesian network nodes embrace ge- The core of the mathematical model can be summa-
ology condition, design condition, construction condi- rized as defining simulation approach coupling CPM
tion, management condition, and the occurrence of risk and CYCLONE, defining risk analysis method studying
events. Output parameters embrace occurrence proba- both ordinary risk factors and risk events, and clarifying
bility (OPRO ) and consequence (OCON ) of risk events. relationship between the two approaches, so as to
Part  4 defines the relationship between simulation realize comprehensive and effective tunnel schedule
method (MSIM ) and risk analysis method (MRISK ). In risk analysis. Apart from diversion tunnel constructed
the CYCLONE modeling method, impact of ordinary with conventional tunneling methods such as drilling
risk factors (IOR ) can be expressed with fitted proba- and blasting approach, the mathematical model can be
bility distributions (OOR ). Therefore, f2 reflects the re- generalized in TBM tunneling (Borrmann et al., 2015)
lation between MCO and MOR . Because impact of risk or other construction engineering, whose construction
Probabilistic risk analysis of diversion tunnel construction simulation 753

processes are repetitive and affected by both ordinary described in detail in Section 3.2. In regard to risk events
risk factors and risk events. When applying to other such as tunnel collapse, a Bayesian network is embed-
projects, the input simulation parameters (ICO ), node ded into CYCLONE to calculate the occurrence proba-
parameters of Bayesian network (INO ), which represent bility P of risk event, with comprehensive consideration
the factors considered in risk events analysis, JPD of geological condition G, design condition D, construc-
of risk event occurrence (IJPD ), CPT between each tion condition C, and management condition M, and
node (ICPT ) need to be determined considering project their mutual dependence. The detailed calculation pro-
characteristics. For the TBM method, the biggest cess of occurrence probability is introduced in Section
difference from conventional tunneling approach lies 3.3.1, and the consequence analysis of risk events is de-
in the construction technology instead of potential risk scribed in Section 3.3.2.
events, as the working environment of the two methods The established probabilistic risk analysis model is
are similar. Therefore, when studying the schedule not a deterministic model, whose uncertainties are re-
risk of TBM tunneling, the simulation parameters flected in two aspects:
(ICO ) are mainly required to adjust according to the
cyclic feature of TBM construction, including TBM
relocation, checking, boring, transportation of muck, 1. The variance of model parameters. The durations
TBM advancing, shifting, liner installation, etc. The of operations in CYCLONE and nonsimulation
parameters about risk analysis can refer to that used in activities in CPM network are defined with certain
drilling and blasting method. probability distributions considering the effect of
ordinary risk factors. In addition, footage driving
cycle, blasting parameters, and other input param-
3.1 Overall structure of probabilistic risk analysis eters will change under diverse geological condi-
model of diversion tunnel construction simulation tions.
The probabilistic risk analysis model of diversion tunnel 2. The variance of model logical relationship. Dur-
construction simulation, which considers both ordinary ing tunnel construction, when risk events occur in
risk factors and risk events, is set up in this study, and an excavation cycle, recovery activity for event dis-
the overall structure of this model is shown in Figure 3. posing of is adopted, and the active vector line
The model is established based on a hierarchical simu- will connect to the upper available vector line, as
lation framework coupling the CPM network and CY- is shown in Figure 3. It will affect operation se-
CLONE. The CPM network embraces simulation activ- quences, and logical relationship between activi-
ities and nonsimulation activities. Simulation activities, ties will change.
such as excavation activities, are studied using the CY-
CLONE model. Nonsimulation activities include lining,
grouting, and other activities. The durations of the non-
3.2 Ordinary risk factors analysis of diversion tunnel
simulation activities are hard to measure directly and
are obtained with engineering analogy method (Bi et al., According to Isaksson and Stille (2005) and Hulett
2014). The CYCLONE model reflects the cyclic process (2009), the variation caused by ordinary risk factors can
of tunnel excavation. This process is repeated until all be described by a continuous distribution. Therefore,
the work of simulation activity is finished. For symbol the working efficiency of drills, dump trucks, and load-
definitions at the CYCLONE level, refer to Lin et al. ers, and operations in CYCLONE, except for drilling
(2011). and rubble removal, are endowed with certain probabil-
Based on the hierarchical simulation framework, the ity distributions. These distributions are determined us-
impacts that ordinary risk factors and risk events ex- ing the BestFit technology through analyzing data col-
ert on tunnel excavation are taken into account. For lected from similar project or completed part of the
ordinary risk factors, their impacts on measurement, project.
drilling, blasting, and other unit operations in CY- For lining, grouting, and other nonsimulation activi-
CLONE and nonsimulation activities in the CPM net- ties in CPM, these processes can be divided into quan-
work are studied. It needs to be noted that as durations tities of small operations, whose durations follow ran-
of drilling and rubble removal are calculated according dom distribution. According to central limit theorem
to the selected mechanical quantity and working effi- (Castillo et al., 2014), the duration distributions of these
ciency in simulation, the impact of ordinary risk factors activities can be approximately regarded as normal. The
on these two operations can be reflected by studying mean value and standard deviation can be determined
the factors’ effect on working efficiency of drills, dump with comprehensive consideration of project character-
trucks, and loaders. Analysis of ordinary risk factors is istic and expert experience.
754 Yu et al.

Fig. 3. Overall structure of probabilistic risk analysis model of diversion tunnel construction simulation.

3.3 Risk events analysis of diversion tunnel short enough to justify this assumption. In addition,
according to Špačková and Straub (2011), when conven-
Risk events, such as collapse, tunnel flooding, rockburst,
tional construction method, such as drilling and blasting
etc., will exert a serious impact on the construction
approach, is adopted, the length of section should
schedule. The occurrence of these events not only has a
not be selected shorter than the typical excavation
close relationship with geological conditions, which are
cycle length. Therefore, simultaneously considering the
regarded as external factors affecting tunnel construc-
two requirements, the segment length is selected as
tion, but also depends on other internal factors, such
excavation cycle length.
as design, construction, and management (Zhang et al.,
2015). Therefore, this study pays attention to geology, 3.3.1 Calculation of occurrence probability with
design, construction, and management conditions of Bayesian network. Bayesian network is a concise
diversion tunnel, takes advantage of Bayesian network graphical structure for representing the probabilistic
to model the dependencies between these causes, and relationships among plenty of stochastic variables and
finally calculates the occurrence probability of risk performing probabilistic interface with those variables
events in each excavation cycle. (Neapolitan, 2004; Rostami, 2012; Abimbola et al.,
It is assumed that only one risk event may occur 2015).
within each segment (Špačková and Straub, 2013). The Bayesian network is a type of graph model based
Therefore, the selected length of a segment must be on Directed Acyclic Graph (DAG) (Abimbola et al.,
Probabilistic risk analysis of diversion tunnel construction simulation 755

(UCS) of intact rock material, rock quality designation


(RQD), joint or discontinuity spacing, joint condition,
groundwater condition, and an adjustment for joint ori-
entation to classify rock mass (Hamidi et al., 2010).
Each of the six parameters is assigned a value derived
from field surveys and laboratory tests, and RMR value
is determined as an algebraic sum of ratings for the six
parameters, which lies between 0 and 100. On the basis
of RMR value, the rock mass is sorted into five classes,
namely, very good (class I, RMR 100–81), good (class
II, RMR 80–61), fair (class III, RMR 60–41), poor (class
IV, RMR 40–21), and very poor (class V, RMR < 20).
Fig. 4. Bayesian network model for calculating occurrence In this study, rock classes I and II are classified as good
probability of risk events. geology (G1 ), rock class III is identified as fair geology
(G2 ), and rock classes IV and V are categorized as poor
2015). It is composed of qualitative and quantitative geology (G3 ).
components (Cai et al., 2013). The qualitative element In addition, the nodes of design condition (D), con-
is represented by a network structure consisting of struction condition (C), and management conditon (M)
nodes and a set of directed edges (dependencies). The all have two states, corresponding to reasonable de-
quantitative component is expressed by the assignment sign (D1 ) and unreasonable design (D2 ), proper con-
of conditional probability distributions to the nodes. struction (C1 ) and improper construction (C2 ), and ef-
Considering the conditional dependencies of variables, fective management (M1 ) and poor management (M2 ).
Bayesian network can represent the JPD P(V) of vari- Specifically, unreasonable design includes wrong choice
ables V = {X1 , X2 , . . . ,Xn } as follows: of construction method, inappropriate support parame-
ters, inadequate planning for emergency measures, etc.

n
Improper construction embraces lag in primary support,
P(X 1 , X 2 , ..., X n ) = P(X i |Pai ) (1) disqualification in the lining thickness, etc. Poor man-
i=1
agement incorporates inexperienced site management,
where Pai is the parent node set of nodeX i , and n repre- inadequate emergency response measures, etc. It is be-
sents the number of variables. lieved that adverse geological condition will greatly en-
The node variables X i can be divided into discrete hance the difficulty in reasonable design of construc-
variables, continuous variables, and hybrid discrete- tion parameters, proper implementation of construction
continuous variables, which respectively correspond to technology, and efficient management of construction
discrete Bayesian network, continuous Bayesian net- process. Therefore, the design, construction, and man-
work, and hybrid Bayesian network. A Bayesian net- agement are all thought to be related to geological con-
work fully considering geology, design, construction, dition. Considering geology, design, construction, and
and management conditions and their mutual depen- management conditions and their mutual dependence,
dence is established in this article, as is shown in a Bayesian network, shown in Figure 4, is utilized to
Figure 4. In the network, the node of geology condition calculate occurrence probabilities of risk events in each
owns three states, namely, good, fair, and poor. In ad- excavation cycle under good geology (G1 ), fair geology
dition, the nodes of design, construction, and manage- (G2 ), and poor geology (G3 ). The occurrence probabil-
ment are all binary variables. Therefore, as all nodes are ities of risk events are respectively denoted as P(A|G 1 ),
discrete variables, the established Bayesian network in P(A|G 2 ), and P(A|G 3 ).
this article is discrete Bayesian network.
Geological condition is one of the significant factors 3.3.2 Consequence analysis of risk events. The occur-
affecting the occurrence of risk events such as tunnel rence of tunnel collapse and other risk events will lead
collapse (Wang et al., 2012; Zhang et al., 2015). The ge- to construction suspension and seriously affect tunnel
ological condition (G) in established Bayesian network schedule. Project delays resulting from risk events were
can be divided into three categories, including good ge- analyzed by Sousa (2010) and Špačková et al. (2013).
ology (G1 ), fair geology (G2 ), and poor geology (G3 ). Sousa studied 64 tunnel construction risk events. Nev-
Rock Mass Rating (RMR) system, a rock classifica- ertheless, among the 64 risk events, only one case of
tion system developed by Bieniawsk, is regarded as the a delay was shorter than 60 days. It may be likely that
judgment criteria. The RMR system employs six input events leading to short delays were not included in the
parameters, namely, the uniaxial compressive strength database. Based on the study of Sousa, Špačková took
756 Yu et al.

risk events, which led to delays between 15 and 60 days, within a specified time (Lee, 2005), and ACI, showing
into consideration and determined the cumulative prob- the probability that an activity falls on the critical path
ability distribution function of consequence, as is shown (Ghomi and Teimouri, 2002), are finally calculated af-
in Equation (2): ter repeating simulation for multiple times. The detailed
solving process based on DES and Monte Carlo simula-
F(d) = 1 − exp(−0.0062(d − 15)) (2)
tion is presented in Figure 5.
where F(d) is the cumulative probability density func- The number of simulation times N is determined
tion representing the probability that the delay is no according to the error range between duration sam-
longer than d days. ple average and expected value (Johnson, 2002). The
This article adopts the approach proposed by results of N simulations are respectively denoted  N as
Špačková to analyze the consequence of risk events. It [T1 , T2 , ..., TN ]. The mean value is T = 1/N i=1 Ti ,
is assumed that the delay resulting from risk events is N 2
and variance is S = 1/(N − 1) i=1 (Ti − T ) . μ is the
2
no less than 15 days and the probability density func- expected value of construction durations and σ 2 is the
tion Dgeo (d), which means the probability of delay with population variance. T1 , T2 , ..., TN are independent and
d days, can be obtained as the first derivative of F(d): identically distributed, and S 2 is the consistent estimate
Dgeo (d) = 0.0062exp( − 0.0062(d − 15)) (3) of σ 2 . Therefore, according to the central limit theorem,
when N is large enough, the following equations can be
obtained:

4 SOLVING PROCESS BASED ON DES AND Z = (T − μ)/ S 2 /N ∼ N (0, 1) (4)
MONTE CARLO SIMULATION (x) means probability distribution of random variable
x, which obeys standard normal distribution X N(0,1),
DES method and Monte Carlo simulation (Argyroudis and β is the confidence level:
et al., 2015; Tondini et al., 2015) are utilized to solve the  
probabilistic risk analysis model. Considering the estab-  T −μ 
 
lished hierarchical simulation structure, global simula- P(   ≤ V ) = (V ) − (−V ) = β (5)
 S 2 /N 
tion clock and local simulation clock (Zhong and Liu,

2003) are respectively set to advance the simulation pro- Denote ω = V. S 2 /N , when absolute error requires
cess of the CPM model and the CYCLONE model. It to be less than ω and confidence level is β, the error
allows for simulating the excavation process of diver- function can be expressed as
sion tunnel at both activity level and operation level,  
and guarantees the efficiency and refinement of simu- P(T − μ ≤ ω) = β (6)
lation process. The two simulation clocks adopt next- Considering the requirements of schedule risk analy-
event time advance (Law et al., 2013) to record running sis of diversion tunnel, β = 0.95, ω = 1.5 days. Accord-
trajectory of simulation time. In the running process of ing to normal distribution, (V ) = (β + 1)/2 = 0.975,
global simulation clock, the duration of nonsimulation and V = 1.96 after checking normal distribution ta-
activity is obtained with Monte Carlo sampling. When ble. For the pressure pipes analyzed in the case study,
global simulation clock reaches a simulation activity, lo- 3,000 simulations, which take into account both ordi-
cal simulation clock is invoked to simulate the cyclic ex- nary risk factors and risk events, are first conducted to
cavation process of tunnel. obtain sample variance S2 = 5294.934. To meet accuracy
In each simulation cycle, the operation time is gen- requirement, simulation number needs to be at least
erated with Monte Carlo sampling according to the fit- N = V 2 S 2 /ω2 = 9,040. For statistical convenience, the
ted probability distribution, which can reflect the impact simulation time is N = 10,000.
of ordinary risk factors. In addition, the impact of risk
events is considered in each cycle, whose occurrence
probability is calculated using a Bayesian network, with
5 CASE STUDY
comprehensive consideration of geology, design, con-
struction, and management conditions and their mu-
5.1 Project description
tual dependence. When local simulation clock comes
to the end, its state is fed back to global simulation Two pressure pipes are located behind a diversion tun-
clock and global simulation clock continues running un- nel to convey the water from diversion tunnel to the
til the end. Construction duration and critical path can turbine for power generation in a hydropower station.
be obtained in each simulation. Completion probabil- The two pipes comprise upper section, vertical shaft sec-
ity, which means the probability of completing a project tion, and lower section. Drilling and blasting method
Probabilistic risk analysis of diversion tunnel construction simulation 757

Fig. 5. Solving process based on DES and Monte Carlo simulation.

is adopted for the excavation of the pressure pipes. is extended to be a pilot shaft in the opposite direc-
The construction of upper section can start after the tion. Finally, the vertical shaft is excavated downwards
completion of M4 auxiliary tunnel, and the lower sec- with drilling and blasting method. The rock type and
tion is able to be excavated following the finish of M1 excavation process of pressure pipes are presented in
auxiliary tunnel. The vertical shaft can be constructed Figure 6.
when the upper section is finished. Specifically, a pilot The two pressure pipes are planned to be completed
hole is first drilled from top to bottom. Then, the hole within 1,176 days. The proposed probabilistic risk
758 Yu et al.

Fig. 6. Rock type and excavation process of two pressure pipes.

analysis method of diversion tunnel construction simu- inspection, clearing, shotcrete-bolt support, working
lation is applied to analyze excavation process and study efficiency of drill and loader, and velocity of transport-
schedule risk of the pressure pipes. ing rubble and return, are endowed with appropriate
probability distributions.
Distributions of these simulation parameters are
5.2 Establishment of CPM network of pressure pipes fitted based on data collected from completed part
The CPM network of pressure pipes is established based of the pressure pipes using the BestFit software.
on the layout and excavation direction, as shown in BestFit is a distribution-fitting software that performs
Figure 7. Squares in the network represent simulation well in finding the statistical distribution function
activities and the circles mean nonsimulation activities. that best fits a data set (Jankauskas and Mclafferty,
Definitions of diverse excavation sections of the pres- 1996). More than 60 distributions, including Normal,
sure pipes are shown in Figure 6. Gamma, Lognormal, Weill, etc., are provided in the
software. When selecting the best fit for input data,
for each distribution type, BestFit first estimates pa-
5.3 Probabilistic risk analysis of pressure pipes rameters values using maximum-likelihood estimators
(MLEs) (Teimouri and Nadarajah, 2013; Asgharzadeh
The impacts that ordinary risk factors and risk events
et al., 2013), then optimizes the parameters with the
exert on the construction of pressure pipes are ana-
Levenberg–Marquardt method (Ueda and Yamashita,
lyzed based on hierarchical simulation structure cou-
2010), an algorithm that maximizes the goodness-of-fit
pling CPM network and CYCLONE model.
between a data set and a distribution function. BestFit
offers three goodness-of-fit tests: chi-square (Adekped-
5.3.1 Analysis of ordinary risk factors. The effect that jou et al., 2015), Kolmogorov–Smirnov (Wang et al.,
ordinary risk factors make on construction can be 2011; Lekomtcev et al., 2014), and Anderson–Darling
reflected by continuous probability distribution. For (Coronel-Brizio and Hernandez-Montoya, 2010;
simulation activity, operations in CYCLONE model, Ashkar et al., 2012), and the function with the lowest
including measurement, blasting, ventilation, security goodness-of-fit values is considered as the best fit. In
Probabilistic risk analysis of diversion tunnel construction simulation 759

Fig. 7. CPM network of pressure pipes.

this article, chi-square is adopted for goodness-of-fit operations, the delay of material supply and worker
measurement. arrival will significantly increase the durations of op-
The data used for optimal distribution fitting are erations. Furthermore, in the skewed distributions, the
collected from the onsite construction record of the kurtosis values are all greater than three. It indicates
completed sections of the pressure pipes. Due to space that these distributions are heavy-tailed, which reflects
limitations, Figure 8 presents fitting results of security the large dispersion of activity durations due to impact
inspection and shotcrete-bolt support with the BestFit of risk factors. With the largest skewness and kurtosis of
technology. Detailed fitting results of each parameter duration distribution, shotcrete-bolt support is thought
are shown in Table 1. to have higher uncertainty than other operations and
It can be learned from Table 1 that except for ven- needs more attention during tunnel excavation.
tilation, whose duration is thought to be constant, the Chi-square test (Satorra and Bentler, 2001) is used to
durations of other operations, including measurement, justify the selected distributions, and the test results are
blasting, security inspection, clearing, and shotcrete- presented in Table 2. Because the ventilation time is de-
bolt support, all follow the skewed distributions. The fined to be constant, it is not required to be justified.
skewness values of these distributions are all positive, It can be learned from the results that, for each sim-
which indicates that the tail on the right side of the ulation parameter, the statistical value (Dn ) is smaller
distributions is longer than the left side, as is pre- than threshold (Dn,0.05 ) under the significance level of
sented in Figure 8. It is because during construction of 0.05, which justifies the rationality of the selected distri-
measurement, blasting, security inspection, and other butions of simulation parameters.
760 Yu et al.

Fig. 8. Fitting results of security inspection and shotcrete-bolt support.

Regarding lining, grouting, and other nonsimulation and management situations are all affected by geologi-
activities, their distributions are thought to obey normal cal condition. Therefore, geological condition (G) is the
distribution according to central limit theorem, as is ex- root node of this Bayesian network, and plays a critical
plained in Section 3.2. The mean value and standard de- role in analyzing occurrence of risk events.
viation are determined combining construction feature The determination of JPD and CPT is critical for the
of this project and expert experience. The duration dis- establishment of Bayesian network. There are generally
tributions and coefficients of variation of nonsimulation three approaches to constructing a Bayesian network:
activities are presented in Table 3. (1) learning from quantities of training data; (2) depend-
With comparison of the coefficients of variation of ing on experiences of domain experts; and (3) hybrid
diverse nonsimulation activities, it can be learnt that method (Chen and Leu, 2014). The second approach is
drilling of pilot hole and expanding of pilot shaft have chosen in this study as the training data are limited, and
higher uncertainty than that of other nonsimulation Delphi technology (Billig and Thrän, 2016) is adopted
activities, due to their larger coefficients of variation to build a consensus on the determination of JPD and
presented with bold font in Table 3. Therefore, more CPT.
attention should be paid to the drilling of pilot hole and
expanding of pilot shaft during construction. 
2 
2 
2

P(A |G i ) = P(A Dj Ck Mm Gi ).P(Dj |Gi ).
m=1 k=1 j=1
5.3.2 Analysis of risk events. Risk events such as col-
lapse are likely to occur in each excavation cycle. P(Ck |G i ).P(Mm |G i ) i = 1, 2, 3 (7)
Figure 4 shows that the occurrence of risk events (A)
has a relationship with geological condition (G), design Thirty experts, including senior engineers, project
condition (D), construction condition (C), and manage- managers, and consultants, make up the panel mem-
ment condition (M). Moreover, design, construction, bers. The probability scales shown in Tables 4
Probabilistic risk analysis of diversion tunnel construction simulation 761

Table 1
Detailed fitting results of simulation parameters

Parameters

Standard Probability
Name (unit) Mean Median Mode deviation Skewness Kurtosis distribution
Measurement (min) 60.64 59.06 56.19 12.96 0.686 3.706 Gamma
Blasting (min) 65.13 62.41 57.20 23.39 0.742 3.995 Lognormal
Ventilation (min) 70.00 70.00 70.00 0.00 – – Constant
Security inspection (min) 53.58 50.45 44.08 23.07 0.824 4.012 Gamma
Clearing (min) 61.26 58.51 52.15 24.96 0.601 3.191 Weibull
Shotcrete-bolt support (min) 160.03 148.28 127.44 63.54 1.325 6.275 Lognormal
Production rate of drilling 0.15 0.15 0.15 0.0196 0.00 3.0 Normal
machine (m/min)
Production rate of loading 60 60 60 1.98 0.00 3.0 Normal
machine (m3 /h)
Velocity of transporting (km/h) 20 20 20 2.32 0.00 3.0 Normal
Velocity of returning (km/h) 25 25 25 2.06 0.00 3.0 Normal

Table 2
Results of chi-square test

Simulation parameters Degrees of freedom n Significance level α Threshold Dn,α Statistical value Dn
Measurement 5 0.05 11.071 7.8526
Blasting 6 0.05 12.592 8.1743
Security inspection 6 0.05 12.592 4.3977
Clearing 6 0.05 12.592 5.9627
Shotcrete-bolt support 5 0.05 11.071 3.1519
Production rate of drilling machine 6 0.05 12.592 5.6892
Production rate of loading machine 7 0.05 14.067 7.1857
Velocity of transporting 5 0.05 11.071 6.5742
Velocity of returning 6 0.05 12.592 5.4983

(Edwards, 1995) and 5 (Wang, 2004) with both verbal of geological conditions. However, it is believed that
and numerical expressions are used for determining the there is still difference between occurrence probabili-
probability. Table 5 is particularly suitable for study- ties under diverse geological conditions. Therefore, oc-
ing event or condition with small occurrence probabil- currence probabilities of risk events under satisfactory
ity. Iterative questionnaires are generally required for design, construction and management conditions, and
reaching a consensus, and the results of the previous diverse geological conditions are determined according
questionnaires are returned to the respondents, who are to Table 5. In addition, when geology, design, construc-
then allowed to modify their responses. In this study, tion, and management are all under adverse conditions,
the degree of consensus is measured in terms of abso- the occurrence probability of risk events is 0.99 in accor-
lute deviation (e.g., standard deviation). When Table 4 dance with Table 4. However, experts point out that risk
is adopted for probability analysis, the target value is an event is certain to happen under such condition, then
average absolute deviation of less than 0.01 on proba- the occurrence probability is adjusted to be 1. When it
bility. When Table 5 is chosen, the target value turns to comes to conditional probability, because the probabili-
be 10−6 . ties of unreasonable design, improper construction, and
The JPD is determined using Table 4. Under the cir- poor management under diverse geological conditions
cumstance of reasonable design, proper construction, in each excavation cycle are thought to be quite small,
and effective management, the verbal expressions of the Table 5 is chosen for analysis. The determined JPD and
occurrence of risk events are all “impossible” regardless CPT are presented in Tables 6 and 7.
762 Yu et al.

Table 3
Duration distributions and coefficients of variation of nonsimulation activities

Activity number Activity name Duration distribution N (a, b) Coefficient of variation (×10−2 )
6 Lining of 1# shaft section N (240, 6.75) 2.81
7 Grouting of 1# shaft section N (135, 5.21) 3.86
8 Lining of section 2 of 1# upper section N (30, 2.12) 7.07
9 Lining of 1# lower section N (200, 6.51) 3.26
14 Steel liner installation of 1# lower section N (343, 7.37) 2.15
15 Plugging between M1 and 1# lower section N (60, 2.54) 4.23
13 Steel liner installation of 2# lower section N (425, 7.85) 1.85
17 Lining of 2# lower section N (217, 6.32) 2.91
18 Plugging of M6 auxiliary tunnel N (70, 2.62) 3.74
19 Plugging between M1 and 2# lower section N (50, 2.47) 4.94
20 Lining of 2# shaft section N (236, 6.64) 2.81
21 Lining of section 2 of 2# upper section N (32, 2.05) 6.41
22 Grouting of 2# shaft section N (120, 5.04) 4.20
23 Plugging between M4 and 2# upper section N (55, 2.68) 4.87
27 Drilling of 2# pilot hole N (44, 4.54) 10.32
28 Expanding of 2# pilot shaft N (53, 4.75) 8.96
30 Lining of section 1 of 2# upper section N (322, 7.17) 2.23
31 Lining of section 1 of 1# upper section N (388, 7.72) 1.99
32 Plugging of M5 auxiliary tunnel N (40, 1.95) 4.88
33 Plugging between M4 and 1# upper section N (30, 1.89) 6.30
36 Drilling of 1# pilot hole N (44, 4.42) 10.05
37 Expanding of 1# pilot shaft N (52, 4.68) 9.00
Note: In normal distribution N(a,b), a represents the mean value of activity duration, whose unit is day, and b means the standard deviation of
activity duration.

Table 4
Probability elicitation scale 1

Impossible Improbable Uncertain 50-50 Somewhat expected Probable Certain


1% 10% 20% 30% 40% 50% 60% 70% 80% 90% 99%

The occurrence probability of risk events under good Table 5


geology (G1 ), fair geology (G2 ), and poor geology (G3 ) Probability elicitation scale 2
can be calculated with Equation (7).
Risk level Verbal expressions of risk level Probability scale
The definitions of A, Dj (j = 1, 2), Ck (k = 1,
2), and Mm (m = 1, 2) are explained in Section 1 Very large [0.1, 1]
3.3.1. P(A|D j Ck Mm G i ) refers to the JPD presented in 2 Large [0.01, 0.1]
3 Medium [0.001, 0.01]
Table 6. P(D j |G i ), P(Ck |G i ), and P(Mm |G i ) can be ob-
4 Small [0.000001, 0.001]
tained according to conditional probability in Table 7.
5 Very small [0, 0.000001]
According to Table 6, Table 7, and Equation (7), oc-
currence probabilities of risk events under diverse geo-
logical conditions are calculated: P (occurrence| good
geology) = 7.2* 10−6 ; P (occurrence| fair geology) =
3.6* 10−5 ; P (occurrence| bad geology) = 1.02* 10−4 . agement conditions on the occurrence of risk events
The consequence of risk events can be expressed as is thought to be same, and is larger than that of ge-
Equation (3) in Section 3.3.2. ological condition. It reflects that although geological
More information can be found from Tables 6 and condition will influence the occurrence of risk events,
7. Table 6 shows the JPD of risk events. It can be experts believe that the performance of engineer in de-
learnt that the impact of design, construction, and man- sign, construction, and management is more important
Probabilistic risk analysis of diversion tunnel construction simulation 763

Table 6 stages of tunnel engineering, especially at construction


JPD of occurrence of risk events (A) stage.
Dj (j = Ck (k Mm (m
Gi (i = 1, 2, 3) 1,2) = 1,2) = 1,2) P(A|D j Ck Mm G i )
D1 C1 M1 0 5.4 Simulation result analysis
D2 C1 M1 0.15
Construction simulations considering and without con-
D1 C2 M1 0.15
G1 D1 C1 M2 0.15
sidering risk events are both conducted 10,000 times and
D2 C2 M1 0.5 simulation results are compared. The determination of
D1 C2 M2 0.5 simulation times refers to Section 4. With the proces-
D2 C1 M2 0.5 sor of Intel R
CoreTM i7-4790 CPU @ 3.60GHz and the
D2 C2 M2 0.7 RAM capacity of 12.0 GB, the CPU times required by
the probabilistic risk analysis model, which fully consid-
D1 C1 M1 0.00002
D2 C1 M1 0.2 ers ordinary risk factors and risk events, are 446.675 sec-
D1 C2 M1 0.2 onds under 10,000 times simulation.
G2 D1 C1 M2 0.2 As shown in Figure 9, mean values of simulation
D2 C2 M1 0.65 durations considering and neglecting risk events are
D1 C2 M2 0.65 1,180 and 1,165 days, standard deviations are 70.91 and
D2 C1 M2 0.65 11.34, respectively, and the completion probabilities of
D2 C2 M2 0.85 planned duration (1,176 days) are 0.841 and 0.771. Thus,
D1 C1 M1 0.00005 consideration of risk events in tunnel simulation will
D2 C1 M1 0.4 lead to larger mean value and standard deviation, and
D1 C2 M1 0.4 lower completion probability of planned duration. Par-
G3 D1 C1 M2 0.4 ticularly, the standard deviation of simulation duration
D2 C2 M1 0.8 increases considerably when considering risk events,
D1 C2 M2 0.8 which is also evident from the tail behavior presented
D2 C1 M2 0.8 in Figure 10. This is because once a risk event occurs,
D2 C2 M2 1
it will exert great influence on the tunnel construction
and increase schedule risk, despite the low occurrence
probability.
Table 7 Activity duration will change due to the influence of
CPT between each node ordinary risk factors and risk events, leading to variance
of the critical path. Figure 11 shows the criticality anal-
D2 C2 M2
ysis result of each activity after 10,000 times simulation
G1 0.000015 0.000018 0.000015 considering risk events.
G2 0.000025 0.00003 0.000025 It can be observed that the criticalities of activities
G3 0.00004 0.00005 0.00004 1, 10, 11, 13, 14, 15, and 19 are significantly larger than
those of other activities. Therefore, these activities are
critical activities affecting tunnel construction and need
to be attached great importance to in construction man-
in guaranteeing tunnel schedule. Table 7 reflects the agement.
influence of geological condition on design, construc- Not only can the results of 10,000 times simulation be
tion, and management. Experts believe that construc- analyzed, but detailed study of each simulation can be
tion is more sensitive to geology than design and man- realized. Taking one simulation as an example, the bar
agement, which have the same sensitivity to geologi- chart in Figure 12a shows that the total duration of this
cal condition. It is mainly embodied in two aspects. To simulation is 1,309 days and the critical path consists of
begin with, under same geological condition, the oc- activities 1, 10, 11, 13, 14, 15, and 19. The detailed ex-
currence probabilities of unreasonable design and in- cavation process of 2# lower section, embracing section
effective management are the same and both smaller 1 (activity 10) and section 2 (activity 11), is shown in
than that of improper construction. In addition, as ge- Figure 12b. As is shown, when the cumulative excava-
ological condition going adverse, the increase degree tion length reaches 562 m, a risk event will occur in the
of occurrence probability of improper construction is next excavation cycle and the corresponding risk treat-
larger than that of the other two. Therefore, geological ment time is 131 days, which leads to major delay for
condition should be taken into account at different tunnel excavation.
764 Yu et al.

Fig. 9. Completion probability curves of excavation duration.

Fig. 10. Exceedance probability curves of duration.

5.5 Model verification ule does. Especially in the 700–725 m excavation sec-
tion, as water inflow occurred, whose processing time
The excavation process of 2# upper section is taken
was 25 days, the actual excavation efficiency decreased
as the research object for verification of the estab-
obviously, which led to greater deviation between actual
lished probabilistic risk analysis model of diversion tun-
and planned schedule. The pictures of the water inflow
nel construction simulation. The schedule obtained us-
and the treatment are presented in Figures 14 and 15.
ing probabilistic risk analysis model, actual schedule,
Because the proposed probabilistic risk analysis model
and planned schedule are compared in Figure 13. Be-
can comprehensively consider the impact that potential
cause construction duration varies in each simulation,
ordinary risk factors and risk events exert on pressure
the 1st, 50th, and 99th percentile curves are respec-
pipe construction, the actual schedule of 700–725 m and
tively drawn after analyzing 10,000 simulation results.
subsequent excavation section still lies between the 1st
The figure shows that the 50th value performs better in
and 99th value. Therefore, when compared with tradi-
matching with the actual schedule than planned sched-
tional construction schedule plan, the probabilistic risk
Probabilistic risk analysis of diversion tunnel construction simulation 765

Fig. 11. Analysis of activity criticality after 10,000 times simulation considering risk events.

analysis model established in this article can fully take Furthermore, the coefficient of variation of simulation
potential risk into account, and realize effctive analysis durations without risk events is 0.97%, and the value
of construction process. increases to be 6% when considering risk events. The
difference demonstrates that the proposed probabilistic
risk analysis method facilitates comprehensive study of
6 DISCUSSION tunnel construction risk and is conducive for accurate
estimation of construction schedule.
Schedules are major concerns in construction planning Besides ensuring comprehensiveness, the proposed
(Adeli and Karim, 1997; Adeli and Wu, 1998; Karim method performs distinctive advantages in realizing ef-
and Adeli, 1999a,b; Adeli and Karim, 2001; Senouci and fective analysis of tunnel construction risk. For ordinary
Adeli 2001). The potential risk during construction may risk factors, this approach pays attention to unit oper-
cause delay and the planned schedule cannot be success- ations composing an excavation cycle, and studies the
fully implemented. factors’ impact on measurement, blasting, ventilation,
The proposed probabilistic risk analysis method of di- working efficiencies of drill, dump truck, and loader, etc.
version tunnel construction simulation guarantees com- When compared with existing approaches that take tun-
prehensiveness and effectiveness of tunnel construction nel excavation section with certain length as research
risk analysis, which allows engineers to accurately es- object for analyzing ordinary risk factors (Špačková
timate tunnel schedule and develop effective schedule and Straub, 2011, 2013; Špačková et al., 2013), this ap-
plan. Potential risk events (such as collapse) during proach carries on research combining the cyclic features
tunnel construction are taken into consideration in the of tunnel construction, and allows for analyzing the im-
proposed method. It can be learnt from Figure 9 that pact of ordinary risk factors at operation level. For
risk events influence mean value of simulation dura- risk events, the probabilistic risk analysis method takes
tion to some extent, and increase standard deviation, excavation cycle, the fundamental unit of tunnel con-
namely, fluctuation of total duration, considerably, struction, for research, and quantitatively analyzes the
which can be observed more obviously in Figure 10. impact of risk events on each excavation cycle. Espe-
The minimum and maximum simulation durations cially in terms of occurrence probability of risk events,
neglecting risk events are respectively 1,125 and 1,206 compared with other methods used in tunnel schedule
days, and the corresponding exceedance probability risk analysis, such as statistical analysis and expert ex-
curve, shown in Figure 10, is smooth without any perience (Špačková et al., 2013; Liu et al., 2014), the
inflection point. When risk events are taken into proposed method studies the occurrence of risk events
consideration, the minimum duration remains the from their root causes, namely, geological condition,
same, but the maximum value turns to be 2,320 days. design condition, construction condition, management
When it comes to the exceedance probability curve, condition, and the mutual dependence between these
obvious turning is presented at point T = 1,200 days, causes, and utilizes Bayesian network to quantitatively
and the curve tail appears to be gentle and long. This calculate risk occurrence probability in each excavation
phenomenon is closely related to the low occurrence cycle, which facilitates effective analysis of potential risk
probability and serious consequence of risk events. events during tunnel construction.
766 Yu et al.

Fig. 12. Result analysis of one simulation: (a) Gantt chart of pressure pipe construction schedule; (b) detailed excavation process
of 2# lower section.

The proposed probabilistic risk analysis method can archical simulation structure coupling CYCLONE and
be generalized in other tunnels of civil engineering con- CPM. Furthermore, based on the hierarchical model,
structions, such as highway tunnel, railway tunnel, etc., the impact of ordinary risk factors is allowed to be
regardless of whether conventional tunneling method or finely analyzed at the operation level, and the influ-
mechanical tunneling method is adopted. Tunnel con- ence that risk events exert on each excavation cycle can
struction processes with diverse methods are all repet- be studied, taking into account geology, design, con-
itive and involve numerous activities. In addition, the struction, and management conditions and their mu-
excavation process is affected by both ordinary risk fac- tual relationship. Therefore, the proposed probabilistic
tors and risk events. The occurrence of risk events (e.g., risk analysis method guarantees the comprehensive and
tunnel collapse) is related to geology, design, construc- effective assessment of tunnel schedule risk, which facil-
tion, and management conditions, which are not inde- itates the accurate estimation of construction schedule
pendent of each other. For these tunnel projects, the and effective development of schedule plan.
presented probabilistic risk analysis approach could ef- There are still some limitations requiring further re-
fectively describe cyclic construction process with hier- search in future study. In the proposed probabilistic
Probabilistic risk analysis of diversion tunnel construction simulation 767

Fig. 13. Comparison of schedule calculated with probabilistic risk analysis model, actual and planned schedule of 2# upper
section.

icance for accurately predicting tunnel construction.


Moreover, more efforts for collecting data about lining,
grouting, and diverse risk events such as water inflow,
rockburst, etc. should be made for further analysis of
tunnel schedule risk. In addition, as modern construc-
tion development tends toward digitalization and sys-
tematism, further research will aim to study a probabil-
ity risk analysis system of diversion tunnel. This system
can be applied to actual projects in practice and is more
convenient for designers and construction management
personnel.
Fig. 14. Water inflow happened in 2# upper section.

7 CONCLUSIONS

A probabilistic risk analysis method of diversion tun-


nel construction simulation is proposed. It enables com-
prehensive and effective risk analysis of tunnel con-
struction by considering both ordinary risk factors and
risk events, which facilitates accurate estimation of con-
struction schedule and effective development of sched-
ule plan.
Considering the cyclic construction features of diver-
sion tunnel, the proposed method defines optimal prob-
ability distribution for measurement, blasting, ventila-
Fig. 15. Treatment for water inflow. tion, and other operations in CYCLONE with BestFit
technology, so as to realize fine and effective analysis
of ordinary risk factors at the operation level of tunnel
risk analysis approach of diversion tunnel construc- construction.
tion simulation, simulation parameters, such as prob- Bayesian network is embedded into simulation pro-
ability distributions of operation durations, are deter- gram to quantitatively analyze the occurrence proba-
mined based on data collected from completed section. bility of potential risk events in each simulation cycle,
With the advancement of tunnel excavation, finding taking geology, design, construction, and management
an efficient approach for fully utilizing abundant con- conditions and their mutual dependence into account.
struction data to update parameters is of vital signif- It enables quantitative analysis of risk events from their
768 Yu et al.

underlying causes and facilitates effective analysis of im- Adeli, H. & Wu, M. (1998), Regularization neural network for
pact that risk events exert on each tunnel excavation construction cost estimation, Journal of Construction Engi-
cycle. neering and Management, 124(1), 18–24.
Al-Bataineh, M., AbouRizk, S. & Parkis, H. (2013), Using
The proposed probabilistic risk analysis method of simulation to plan tunnel construction, Journal of Construc-
diversion tunnel construction simulation is applied to tion Engineering and Management, 139(5), 564–71.
analyze excavation process of pressure pipes, which Aliahmadi, A., Sadjadi, S. J. & Jafari-Eskandari, M. (2011),
are located behind a diversion tunnel to convey the Design a new intelligence expert decision making using
water from diversion tunnel to the turbine for power game theory and fuzzy AHP to risk management in design,
construction, and operation of tunnel projects (case stud-
generation. Simulations considering and neglecting risk ies: Resalat Tunnel), The International Journal of Advanced
events are respectively conducted 10,000 times. The Manufacturing Technology, 53(5-8), 789–98.
mean values and standard deviations of construction Alvanchi, A., Lee, S. H. & AbouRizk, S. (2011), Modeling
durations and completion probabilities of planned du- framework and architecture of hybrid system dynamics and
discrete event simulation for construction, Computer-Aided
ration under the two situations can be obtained. The Civil and Infrastructure Engineering, 26(2), 77–91.
results show that risk events will lead to the increase Argyroudis, S., Selva, J., Gehl, P. & Pitilakis, K. (2015), Sys-
of mean simulation duration, especially significant in- temic seismic risk assessment of road networks considering
crement of duration fluctuation, and decrease of com- interactions with the built environment, Computer-Aided
pletion probability. In addition, critical activities can be Civil and Infrastructure Engineering, 30(7), 524–40.
Asgharzadeh, A., Esmaily, L. & Nadarajah, S. (2013), Ap-
found, which need to be attached great importance to proximate MLEs for the location and scale parameters of
in practical tunnel schedule management. Finally, the the skew logistic distribution, Statistical Papers, 54(2), 391–
feasibility of the proposed probabilistic risk analysis ap- 411.
proach in comprehensive and effective risk analysis is Ashkar, F., Aucoin, F., Choulakian, V. & Vautour, C. (2012),
validated. Cramér-von Mises and Anderson-Darling goodness-of-fit
tests for the two-parameter kappa distribution, Journal of
Hydrologic Engineering, 18(12), 1749–57.
Bhandari, J., Abbassi, R., Garaniya, V. & Khan, F. (2015),
ACKNOWLEDGMENTS Risk analysis of deepwater drilling operations using
Bayesian network, Journal of Loss Prevention in the Pro-
This work was financially supported by Natural Sci- cess Industries, 38, 11–23.
ence Foundation of China (51679165), the National Bi, L., Ren, B., Zhong, D. & Hu, L. (2014), Real-time con-
Key Research and Development Program of China struction schedule analysis of long-distance diversion tun-
nels based on lithological predictions using a Markov pro-
2016YFC0401806, and Innovative Research Groups of cess, Journal of Construction Engineering and Management,
the National Natural Science Foundation of China 141(2), 1–16.
(51621092). Billig, E. & Thrän, D. (2016), Evaluation of biomethane tech-
nologies in Europe – technical concepts under the scope of
a Delphi-survey embedded in a multi-criteria analysis, En-
ergy, 114, 1176–86.
REFERENCES Borrmann, A., Kolbe, T. H., Donaubauer, A., Steuer, H.,
Jubierre, J. R. & Flurl, M. (2015), Multi-scale geometric-
Abimbola, M., Khan, F., Khakzad, N. & Butt, S. (2015), Safety semantic modeling of shield tunnels for GIS and BIM ap-
and risk analysis of managed pressure drilling operation us- plications, Computer-Aided Civil and Infrastructure Engi-
ing Bayesian network, Safety Science, 76, 133–44. neering, 30(4), 263–81.
AbouRizk, S.M. & Hague, S. (2009), An overview of the Cai, B., Liu, Y., Liu, Z., Tian, X., Zhang, Y. & Ji, R. (2013),
COSYE environment for construction simulation, in Pro- Application of Bayesian networks in quantitative risk as-
ceedings of Winter Simulation Conference, IEEE, Texas, pp. sessment of subsea blowout preventer operations, Risk
2624–34. Analysis, 33(7), 1293–311.
AbouRizk, S.M. & Mohamed, Y. (2000), Simphony—an in- Castillo, E., Calviño, A., Nogal, M. & Lo, H. K. (2014), On
tegrated environment for construction simulation, in Pro- the probabilistic and physical consistency of traffic random
ceedings of Winter Simulation Conference, IEEE, FL San variables and models, Computer-Aided Civil and Infrastruc-
Diego, pp. 1907–14. ture Engineering, 29(7), 496–517.
Adekpedjou, A., De Mel, W. A. & Zamba, G. K. (2015), Data Chan, M. H. C. (1981), A geological prediction and updating
dependent cells chi-square test with recurrent events, Scan- model in tunneling, Ph.D. thesis, Massachusetts Institute of
dinavian Journal of Statistics, 42(4), 1045–64. Technology, Cambridge, USA.
Adeli, H. & Karim, A. (1997), Schedulingcost optimiza- Chang, D. Y. (1986), RESQUE: a resource based simulation
tion and neural dynamics model for construction, Journal system for construction process planning, Ph.D. thesis, Uni-
of Construction Engineering and Management, 123(4), 3– versity of Michigan, Ann Arbor, MI.
20(18). Chen, T. T. & Leu, S. S. (2014), Fall risk assessment of can-
Adeli, H. & Karim, A. (2001), Construction Scheduling, tilever bridge projects using Bayesian network, Safety Sci-
Cost Optimization, and Management-A New Model Based ence, 70, 161–71.
on Neurocomputing and Object Technologies, Spon Press, Cheng, M. & Hoang, N. (2014), Slope collapse prediction us-
London. ing Bayesian framework with k-nearest neighbor density
Probabilistic risk analysis of diversion tunnel construction simulation 769

estimation: case study in Taiwan, Journal of Computing in Jankauskas, L. & Mclafferty, S. (1996), BestFit, distribution
Civil Engineering, 30 (1), 1–8. fitting software by Palisade Corporation, in Proceedings of
Chung, T. H., Mohamed, Y. & AbouRizk, S. (2006), Bayesian the 28th Conference on Winter Simulation, IEEE, California
updating application into simulation in the north Edmon- pp. 551–55.
ton sanitary trunk tunnel project, Journal of Construction Johnson, D. J. (2002), A spreadsheet method for calculating
Engineering and Management, 132(8), 882–94. work completion time probability distributions of paced or
Coronel-Brizio, H. F. & Hernandez-Montoya, A. R. (2010), linked assembly lines, International Journal of Production
The Anderson–Darling test of fit for the power-law dis- Research, 40(5), 1131–53.
tribution from left-censored samples, Physica A: Statistical Jurado, A., De Gaspari, F., Vilarrasa, V., Bolster, D., Sánchez-
Mechanics and its Applications, 389(17), 3508–15. Vila, X., Fernàndez-Garcia, D. & Tartakovsky, D. M.
Edwards, L. (1995), Practical Risk Management in the Con- (2012), Probabilistic analysis of groundwater-related risks
struction Industry, Thomas Telford, London, 83 p. at subsurface excavation sites, Engineering Geology, 125,
Feng, X. & Jimenez, R. (2015), Predicting tunnel squeezing 35–44.
with incomplete data using Bayesian networks, Engineering Kamat, V. R. & Martinez, J. C. (2003), Validating complex
Geology, 195, 214–24. construction simulation models using 3D visualization, Sys-
Ghomi, S. F. & Teimouri, E. (2002), Path critical in- tems Analysis Modeling Simulation, 43(4), 455–56.
dex and activity critical index in PERT networks, Eu- Karim, A. & Adeli, H. (1999a), Conscom: an OO construc-
ropean Journal of Operational Research, 141(1), 147– tion scheduling and change management system, Journal of
52. Construction Engineering and Management, 125(5), 368–76.
Gonzalez Jimenez, J., Arevalo, V., Galindo, C. & Ruiz- Karim, A. & Adeli, H. (1999b), OO information model for
Sarmiento, J. R. (2015), An automated surveying and mark- construction project management, Journal of Construction
ing system for continuous setting-out of tunnels, Computer- Engineering and Management, 125(5), 361–67.
Aided Civil and Infrastructure Engineering, 31(3), 219–28. Lau, S. C., Lu, M. & Poon, C. S. (2014), Formalized ap-
Grasso, P., Xu, S., Pescara, M., Russo, G. & Repetto, L. proach to discretize a continuous plant in construction sim-
(2006), A methodology for the geotechnical design of long ulations, Journal of Construction Engineering and Manage-
high-speed rail tunnels under the conditions of uncer- ment, 140(8), 1–9.
tainty, in Proceedings of the ITA-Sponsored China Interna- Law, A. M., Kelton, W. D. & Kelton, W. D. (2013), Simulation
tional Symposium on High-Speed Railway Tunnels, Beijing, Modeling and Analysis, McGraw-Hill, New York.
China. Lee, D. E. (2005), Probability of project completion using
Hajjar, D. & AbouRizk, S. (1997), Development of an object stochastic project scheduling simulation, Journal of Con-
oriented framework for the simulation of earth moving op- struction Engineering and Management, 131(3), 310–18.
erations, in Proceedings of the Intelligent Information Sys- Lee, S. J., Kim, M. C. & Seong, P. H. (2008), An analytical
tems Conference, IEEE, New York, pp. 326–30. approach to quantitative effect estimation of operation ad-
Halpin, D. W. (1977), CYCLONE-method for modeling job visory system based on human cognitive process using the
site processes, Journal of the Construction Division, 103(3), Bayesian belief network, Reliability Engineering & System
489–99. Safety, 93(4), 567–77.
Halpin, D. W. (1990), Micro-CYCLONE User’s Manual, Di- Lekomtcev, D. & Marsalek, R. (2014), Evaluation of
vision of Construction Engineering and Management, Pur- Kolmogorov-Smirnov test for cooperative spectrum sens-
due University, West Lafayette, IN. ing in real channel conditions, in Telecommunications Fo-
Halpin, D. W. & Riggs, L. S. (1992), Planning and Analysis of rum Telfor (TELFOR), 2014 22nd, IEEE, pp. 557–60.
Construction Operations, Wiley, New York. Likhitruangsilp, V. & Ioannou, P. G. (2003), Stochastic eval-
Hamidi, J. K., Shahriar, K., Rezai, B. & Rostami, J. (2010), uation of tunneling performance using discrete-event simu-
Performance prediction of hard rock TBM using rock mass lation, in Construction Research Congress: Wind of Change:
rating (RMR) system, Tunnelling and Underground Space Integration and Innovation, HI, pp. 1–8.
Technology, 25(4), 333–45. Lin, C. T., Hsie, M., Hsiao, W. T., Wu, H. T. & Cheng, T.
Holický, M., Marková, J. & Sýkora, M. (2013), Forensic as- M. (2011), Optimizing the schedule of dispatching earth-
sessment of a bridge downfall using Bayesian networks, En- moving trucks through genetic algorithms and simulation,
gineering Failure Analysis, 30, 1–9. Journal of Performance of Constructed Facilities, 26(2),
Hong, E. S., Lee, I. M., Shin, H. S., Nam, S. W. & Kong, 203–11.
J. S. (2009), Quantitative risk evaluation based on event Liu, D., Xuan, P., Li, S. & Huang, P. (2014), Schedule risk
tree analysis technique: application to the design of shield analysis for TBM tunneling based on adaptive cyclone sim-
TBM, Tunnelling and Underground Space Technology, ulation in a geologic uncertainty–aware context, Journal of
24(3), 269–77. Computing in Civil Engineering, 29(6), 1–12.
Huang, R. Y. & Halpin, D. W. (1994), Visual construction op-
Liu, D. H., Zhou, Y. Q. & Jiao, K. (2010), TBM construction
eration simulation: the DISCO approach, Computer-Aided
process simulation & performance optimization, Transac-
Civil and Infrastructure Engineering, 9(3), 175–84.
tions of Tianjin University, 16(3), 194–202.
Hulett, D. (2009), Practical Schedule Risk Analysis, Gower
Liu, L. Y. (1991), COOPS: construction object-oriented sim-
Publishing, Ltd., United Kingdom.
ulation system, Ph.D. thesis, University of Michigan, Ann
Ioannou, P. G. (1989), UM-CYCLONE User’s Guide, Depart-
Arbor, MI.
ment of Civil Engineering, University of Michigan, Ann Ar-
bor, MI. Lu, M., Lam, H. C. & Dai, F. (2008), Resource-constrained
Isaksson, T. & Stille, H. (2005), Model for estimation of time critical path analysis based on discrete event simulation and
and cost for tunnel project based on risk, Rock Mechanics particle swarm optimization, Automation in Construction,
and Rock Engineering, 38(5), 373–98. 17(6), 670–81.
770 Yu et al.

Martinez, J. C. & Ioannou, P. G. (1994), General purpose sim- Špačková, O. & Straub, D. (2013), Dynamic Bayesian net-
ulation with stroboscope, in Proceedings of the 26th Con- work for probabilistic modeling of tunnel excavation pro-
ference on Winter Simulation, Society for Computer Simu- cesses, Computer-Aided Civil and Infrastructure Engineer-
lation International, FL, pp. 1159–66. ing, 28(1), 1–21.
Martinez, J. C. & Ioannou, P. G. (1999), General-purpose sys- Sun, H. & Betti, R. (2015), A hybrid optimization algo-
tems for effective construction simulation, Journal of Con- rithm with Bayesian inference for probabilistic model up-
struction Engineering and Management, 125(4), 265–76. dating, Computer-Aided Civil and Infrastructure Engineer-
Mu, H. Q. & Yuen, K. V. (2016), Ground motion prediction ing, 30(8), 602–19.
equation development by heterogeneous Bayesian learn- Teimouri, M. & Nadarajah, S. (2013), Bias corrected MLEs
ing, Computer-Aided Civil and Infrastructure Engineering, for the Weibull distribution based on records, Statistical
31(10), 761–76. Methodology, 13, 12–24.
Neapolitan, R. E. (2004), Learning Bayesian Networks, Pear- Tondini, N., Bursi, O. S., Bonelli, A. & Fassin, M. (2015), Ca-
son Prentice Hall, Upper Saddle River, NJ. pabilities of a fiber Bragg grating sensor system to moni-
Nývlt, O., Prı́vara, S. & Ferkl, L. (2011), Probabilistic risk as- tor the inelastic response of concrete sections in new tunnel
sessment of highway tunnels, Tunnelling and Underground linings subjected to earthquake loading, Computer-Aided
Space Technology, 26(1), 71–82. Civil and Infrastructure Engineering, 30(8), 636–53.
Rahm, T., Scheffer, M., Thewes, M., König, M. & Duhme, R. Touran, A. (1987), Simulation of tunnel operation, Journal of
(2016), Evaluation of disturbances in mechanized tunneling Construction Engineering and Management, 113(4), 554–68.
using process simulation, Computer-Aided Civil and Infras- Ueda, K. & Yamashita, N. (2010), On a global complex-
tructure Engineering, 31(3), 176–92. ity bound of the Levenberg-Marquardt method, Jour-
Rostami, J., Farrokh, E., Laughton, C. & Eslambolchi, S. nal of Optimization Theory and Applications, 147(3),
S. (2014), Advance rate simulation for hard rock TBMs, 443–53.
KSCE Journal of Civil Engineering, 18(3), 837–52. Vargas, J. P., Koppe, J. C. & Pérez, S. (2014), Monte Carlo
Rostami, N.K. (2012), Dynamic safety analysis using advanced simulation as a tool for tunneling planning, Tunnelling Un-
approaches, Ph.D. thesis, Memorial University of New- derground Space Technology, 40, 203–09.
foundland., St. John’s, Canada. Vargas, J. P., Koppe, J. C., Pérez, S. & Hurtado, J. P. (2015),
Ruwanpura, J. Y., AbouRizk, S. M., Er, K. C. & Fernando, Planning tunnel construction using Markov chain Monte
S. (2001), Special purpose simulation templates for tunnel Carlo (MCMC), Mathematical Problems in Engineering,
construction operations, Canadian Journal of Civil Engi- 2015, 1–8.
neering, 28(2), 222–37. Wang, C., Zeng, B. & Shao, J. (2011), Application of bootstrap
Ruwanpura, J. Y. & Ariaratnam, S. T. (2007), Simulation method in Kolmogorov-Smirnov test, in 2011 International
modelling techniques for underground infrastructure con- Conference on Quality, Reliability, Risk, Maintenance,
struction processes, Tunnelling and Underground Space and Safety Engineering (ICQR2MSE), IEEE, Xi’an, pp.
Technology, 22(5), 553–67. 287–91.
Satorra, A. & Bentler, P. M. (2001), A scaled difference chi- Wang, H., Jia, Y. & Cao, L. (2012), Cause analysis and preven-
square test statistic for moment structure analysis, Psy- tion of road tunnel collapse in complex soft strata, Energy
chometrika, 66(4), 507–14. Procedia, 16, 259–64.
Senouci, A. B. & Adeli, H. (2001), Resource scheduling us- Wang, H., Yajima, A., Liang, R. Y. & Castaneda, H. (2015),
ing neural dynamics model of Adeli and Park, Journal of Bayesian modeling of external corrosion in underground
Construction Engineering and Management, 127(1), 28–34. pipelines based on the integration of Markov chain Monte
Shahriar, K., Sharifzadeh, M. & Hamidi, J. K. (2008), Carlo techniques and clustered inspection data, Computer-
Geotechnical risk assessment based approach for rock Aided Civil and Infrastructure Engineering, 30(4),
TBM selection in difficult ground conditions, Tunnelling 300–16.
and Underground Space Technology, 23(3), 318–25. Wang, M. S. (2004), Tunnel Boring Machine Construc-
Shi, J. (1999), Activity-based construction (ABC) modeling tion Technology and Practices, China Railway Publishing
and simulation method, Journal of Construction Engineer- House, Beijing (in Chinese).
ing and Management, 125(5), 354–60. Wu, X., Liu, H., Zhang, L., Skibniewski, M. J., Deng,
Sousa, R. L. (2010), Risk analysis for tunneling projects, Ph.D. Q. & Teng, J. (2015), A dynamic Bayesian network
thesis, Massachusetts Institute of Technology, Cambridge, based approach to safety decision support in tunnel con-
MA. struction, Reliability Engineering & System Safety, 134,
Sousa, R. L. & Einstein, H. H. (2012), Risk analysis during 157–68.
tunnel construction using Bayesian networks: Porto Metro Yuen, K. V. & Mu, H. Q. (2015), Real-time system identi-
Case Study, Tunnelling and Underground Space Technol- fication: an algorithm for simultaneous model class selec-
ogy, 27(1), 86–100. tion and parametric identification, Computer-Aided Civil
Špačková, O., Sejnoha, J. & Straub, D. (2013), Probabilistic and Infrastructure Engineering, 30(10), 785–801.
assessment of tunnel construction performance based on Zhang, G. H., Jiao, Y. Y., Chen, L. B., Wang, H. & Li, S.
data, Tunnelling and Underground Space Technology, 37, C. (2015), Analytical model for assessing collapse risk dur-
62–78. ing mountain tunnel construction, Canadian Geotechnical
Špačková, O. & Straub, D. (2011), Probabilistic risk assess- Journal, 53(2), 326–42.
ment of excavation performance in tunnel projects using Zhang, L., Wu, X. & Skibniewski, M. J. (2015), Simulation-
Bayesian networks: a case study, in Proceedings of the 3rd based analysis of tunnel boring machine performance in
International Symposium on Geotechnical Safety and Risk, tunneling excavation, Journal of Computing in Civil Engi-
Bundesanstalt für Wasserbau, München, pp. 651–60. neering, 30(4), 1–14.
Probabilistic risk analysis of diversion tunnel construction simulation 771

Zhang, L., Wu, X., Skibniewski, M. J., Zhong, J. & Lu, Y. based on Markov chain Monte Carlo method, Science
(2014), Bayesian-network-based safety risk analysis in con- China: Technological Sciences, 59(2), 252–64.
struction projects, Reliability Engineering & System Safety, Zhong, D. & Liu, D. (2003), Construction System Simulation
131, 29–39. Theory and Application Study of Large Scale Underground
Zhang, S., Du, C., Sa, W., Wang, C. & Wang, G. (2014), Cavern Group, China Water & Power Press, Beijing (in
Bayesian based hybrid simulation approach to project com- Chinese).
pletion forecasting for underground construction, Jour- Zhong, D. H., Li, J. R., Zhu, H. R. & Song, L. G. (2004),
nal of Construction Engineering and Management, 140(1), Geographic information system-based visual simulation
1–9. methodology and its application in concrete dam construc-
Zhong, D., Bi, L., Yu, J. & Zhao, M. (2016), Robustness anal- tion processes, Journal of Construction Engineering and
ysis of underground powerhouse construction simulation Management, 130(5), 742–50.

Você também pode gostar