risk management in the construction industry using

Accep

ted M

anus

cript

Not Cop

yedit

ed

1

Risk Management in the Construction Industry Using

Combined Fuzzy FMEA and Fuzzy AHP

Mohamed Abdelgawad1 , Aminah Robinson Fayek2, A.M.ASCE

_____________________________________________________________________

Abstract: Failure Mode and Effect Analysis (FMEA) is recognized as one of the most beneficial

techniques in reliability programs. FMEA is a structured technique that can help in identifying

all failure modes within a system, assessing their impact, and planning for corrective actions.

Although this technique has been widely utilized in many industries, it has some limitations. The

purpose of this paper is to extend the application of FMEA to risk management in the

construction industry. Fuzzy logic and fuzzy AHP are utilized to address the limitations of

traditional FMEA. In essence, this method explores the concept of fuzzy expert systems to map

the relationship between Impact (I), Probability of Occurrence (P), and Detection/Control (D)

and the level of criticality of risk events. A case study is presented to validate the concept. The

results obtained confirm the capability of fuzzy FMEA and fuzzy AHP to address several

drawbacks of the traditional FMEA application. Use of this approach can support the project

management team to establish corrective actions in a timely manner.

CE Database subject headings: Risk management; Fuzzy sets; Failure modes

_____________________________________________________________________ 1 Ph.D. Candidate, Dept. of Civil and Environmental Engineering, Hole School of Construction Engineering, 3-013 Markin/CNRL, NREF, Univ.of Alberta, Edmonton AB, Canada T6G 2W2. Email: [email protected] 2 Professor, NSERC Associate Industrial Research Chair, Ledcor Professor in Construction Engineering, Dept. of Civil and Environmental Engineering, University of Alberta, Edmonton, AB, Canada T6G 2W2. Email: [email protected]

Journal of Construction Engineering and Management. Submitted April 27, 2009; accepted February 24, 2010; posted ahead of print February 26, 2010. doi:10.1061/(ASCE)CO.1943-7862.0000210

Copyright 2010 by the American Society of Civil Engineers

Accep

ted M

anus

cript

Not Cop

yedit

ed

2

Introduction

The construction industry is characterized by high levels of risks and uncertainties. Over the past

decade, many projects have experienced large variations in cost and/or schedule, turning these

projects into unsuccessful endeavours. The current trend in the construction industry is towards

increasing project size and complexity, both of which result in greater levels of risk and

uncertainty. Thompson and Perry (1992) attributed the failure of projects to the lack of effective

management of risk events, which often leads to overlooking of milestones and targets. Clearly,

the management of project risks has become fundamental to successful project management.

Failure Mode and Effect Analysis (FMEA) is one of the risk analysis techniques

recommended by international standards, such as (MIL-STD-1629A 1980). FMEA is a technique

that is based on identifying potential failures, analyzing root causes, and examining failure

impacts so that these impacts can be reduced. Within the context of the traditional FMEA, the

degree of criticality of a failure mode is determined by calculating the Risk Priority Number

(RPN). The RPN ranges from 1 to 1000 and is an index score calculated as the product of the

Severity (S), Occurrence (O), and Detection (D) of a failure mode. System components that are

assessed to have a high RPN are assumed to be more critical than those with lower values. The

Severity (S) rating is used to represent the potential effects associated with the occurrence of a

failure mode. Thus, it reflects the seriousness of the effects of the failure. The Occurrence rating

(O) is the frequency of the occurrence of the failure. Ayyub (2003) defined the Detection rating

(D) as “a measure of the capability of the current controls”. Within traditional FMEA, a



Accep

ted M

anus

cript

Not Cop

yedit

ed

3

numerical scale ranging from 1 and 10 is used to represent the universe of discourse for Severity

(S), Occurrence (O), and Detection (D). Based on the values assigned to these terms, the value of

the RPN is calculated.

FMEA and Risk Management Issues

Several studies suggest using FMEA in the context of risk management. For instance, Carbone

and Tippett (2004) introduced an application of FMEA to the context of project risk management

by evaluating the risk score and the RPN value to find the most critical risk events that require

immediate risk response.

From a management perspective, the traditional RPN calculation is easy to understand

and straightforward. However, from a technical perspective, several authors have noted concerns

related to using the traditional FMEA approach to calculate the RPN. For instance, Bowles and

Peláez (1995) and Puente et al. (2002) noted several shortcomings in both the ways in which

calculations are made and the ways in which the results are interpreted. For example, a failure

mode with the following assessment of Severity (S), Occurrence (O) and Detection (D) (say 9, 5,

and 5, respectively) may have a lower RPN (225) than one with high Severity (S), high

Occurrence (O), and a moderate level of Detection (D) (say 7, 7, and 5 yielding a RPN of 245).

Yet from the management perspective, the first failure induces higher priority for corrective

action. Bowles and Peláez (1995) also noted that the RPN may be underestimated when a failure

mode has multiple dimensions of effect, since only the most severe effect is used in the

calculation. Puente et al. (2002) and Pillay and Wang (2003) noted that the RPN does not take



Accep

ted M

anus

cript

Not Cop

yedit

ed

4

into account possible weighting for the importance of Severity (S), Occurrence (O), and

Detection (D). Xu et al. (2002) noted the difficulty of using numerical values to evaluate the

occurrence of the failure event and argue the advantage of using other techniques that can

support linguistic assessment of failure modes.

Ayyub (2003) noted that any failure mode with a RPN greater than 125 should be

considered seriously. However, such a rule of thumb for determining the RPN threshold is very

subjective. In fact, without linking the value of the RPN to linguistic terms describing the

priority to take corrective action, the project team will not be able to recognize the difference, for

example, between a risk with RPN equal to 140 versus one with RPN equal to 160.

Because of the limitations of the traditional calculation of the RPN, this paper focuses on

adapting the traditional FMEA to make it more suitable as a decision support tool to aid in risk

management in the construction industry. The proposed approach aims to address many of the

above mentioned limitations by incorporating fuzzy logic with traditional FMEA and AHP

techniques. The project manager can rely on the decisions obtained from this technique to aid in

successful completion of a project by identifying critical risk events that require immediate

corrective actions. Such a decision will provide the project team with sufficient time to respond

to those critical risk events. In addition, the results obtained from using this technique can aid in

identifying risk events that require comprehensive root cause analysis and detailed lessons

learned. Thus, more successful projects can be achieved in the future using this technique.

Fuzzy Logic for FMEA



Accep

ted M

anus

cript

Not Cop

yedit

ed

5

Fuzzy logic was first suggested by Zadeh (1965) to support reasoning in situations that require

approximate solutions rather than exact ones, and for complex processes for which there is no

simple mathematical model. Fuzzy logic is characterized by the ability to assign membership

values (x) expressing the degree to which a certain value of a variable fits a linguistic concept.

Accordingly, gradual transition between different states of a variable can be achieved through the

use of membership functions.

Fuzzy logic, as opposed to probabilistic techniques, is suitable for handling situations in

which data are not available or are difficult to obtain, or in which assessments are made in

linguistic and subjective terms. Using fuzzy logic, an assessment of the problem can be elicited

from experts in the form of linguistic terms such as ‘very low’, ‘medium’, ‘high’, etc. Since

fuzzy logic is based on a natural way of human communication, the subjective assessment of the

problem can be utilized to derive an acceptable approximation. Fuzzy logic is combined with

FMEA to overcome the deficiencies associated with the traditional approach of computing the

RPN number. Instead of depending on the multiplication of S, O, and D, to calculate the RPN,

the proposed approach uses a fuzzy expert system, based on information elicited from experts, to

analyse and prioritise different risk events.

In essence, any fuzzy expert system is composed of three processes referred to as

fuzzification, fuzzy inference and defuzzification. In fuzzy FMEA, the fuzzification process is

the process in which Severity (S), Occurrence (O), and Detection (D) are converted into their

fuzzy representations. During this process, the user provides assessments of Severity (S),

Occurrence (O), and Detection (D), as will be shown in the next sections, and the system

identifies the corresponding membership function for each associated linguistic term represented

by a fuzzy set. The fuzzy inference is composed of a knowledge base of if-then rules that define



Accep

ted M

anus

cript

Not Cop

yedit

ed

6

the relationships between the premise, represented by input variables, and the conclusion,

represented by the RPN in our case. During the fuzzy inference process, each rule is fired

according to the degree of matching between the input values and the premise of each rule;

implication and aggregation are used during this process to calculate the contribution of each rule

to the overall value of the RPN. The final step is the defuzzification process in which a crisp

output value of the RPN is calculated from the aggregated fuzzy set.

Proposed Framework for Fuzzy FMEA

Because FMEA is about failure modes and their effects, it is necessary to define the term

“failure”. From the production perspective, failure can be defined as “the inability of a design or

a process to perform its intended function” (Justiniano and Gopalaswamy 2003). Failure is not

limited to design or process weakness but can also be due to errors made during product or

process use. On the other hand, from a project risk management perspective, we will use the

“risk” definition as defined in the Guide to the Project Management Body of Knowledge

(PMBOK) to refer to “failure mode”, which is “an uncertain event or condition that, if it occurs,

has a positive or a negative effect on at least one project objective, such as time, cost, scope or

quality” (PMI 2004).

The following terminologies are used in this paper:

• Occurrence (O) is referred to as Probability of Occurrence (P) and is defined over the

range of 1 to 10.



Accep

ted M

anus

cript

Not Cop

yedit

ed

7

• Severity (S) is referred to as Impact (I) and has three dimensions: Cost Impact (CI), Time

Impact (TI), and Scope/Quality Impact (SI). They are all defined over the range of 1 to

10.

• Detection (D) is referred to as Detection/Control (D) and is defined over the range of 1 to

10.

• RPN is referred to as Risk Criticality Number (RCN) and is defined over the range of 1 to

1000.

According to our proposed approach presented in Figure 1, there are two main phases. Phase

1 is concerned with developing the fuzzy FMEA expert system. During this phase, linguistic

definitions are established to define Probability of Occurrence (P), Impact (I), the level of

Detection/Control (D), and the risk criticality number (RCN). According to these definitions,

membership functions for Impact (I), Probability of Occurrence (P), Detection/Control (D), and

RCN are defined. Linguistic definitions of corrective actions are established, and if-then rules are

elicited and used to build the knowledge base. Fuzzy AHP is conducted to aggregate CI, TI, and

SI into a single variable entitled Aggregated Impact (AI).

Phase 2 involves use of this model to analyse risks and recommend corrective actions.

During phase 2, each project is broken down to its main components using the work breakdown

structure (WBS), and each work package is analyzed to identify different risk events.

Determining which work package is expected to be affected by each risk event will aid risk

analysts in assessing the level of Impact (I) of each risk event on each work package. Root cause

analysis is conducted to identify root causes of different risk events. Understanding the root

causes can help the risk analysts to estimate the Probability of Occurrence (P) of each risk event

based on its root causes, and to suggest appropriate risk response strategies to minimize these



Accep

ted M

anus

cript

Not Cop

yedit

ed

8

root causes. An evaluation of the level of Detection/Control (D) for each response strategy is

conducted, and the RCN is calculated using the fuzzy expert system. The calculated RCN is used

to aid in selecting the required corrective action for each risk event. The following steps describe

the approach taken to develop the fuzzy FMEA expert system.

Step 1: Linguistic Definition of Probability of Occurrence and Impact

The first step in integrating fuzzy logic and FMEA is to define Probability of Occurrence (P),

Impact (I), and Detection/Control (D) using linguistic terms. In this regard, each variable is

defined using membership functions (MFs) that cover the universe of discourse of each variable.

To define the linguistic terms for each variable, several interviews were arranged with a senior

risk coordinator working at one of the largest pipeline companies in North America.

The objective of the first meeting was to introduce our proposed framework and to

explore options to implement this technique. Fortunately, the design of the risk matrix, as

implemented by the company, is based on linguistic definitions for both Probability of

Occurrence (P) and Impact (I). Table 1 and Table 2 present five linguistic terms and their

definition for both Probability of Occurrence (P) and Impact (I), as defined in the company’s risk

management framework. In addition, according to their existing risk management framework,

each project is broken down into work packages, and risk identification sessions are normally

arranged to identify different risk events and root causes, assess Probability of Occurrence (P)

and Impact (I) of different risk events, and populate the risk register. Table 1 and Table 2 are



Accep

ted M

anus

cript

Not Cop

yedit

ed

9

used to define Probability of Occurrence (P) and Impact (I) in the proposed approach; however,

the meaning of each linguistic term can be calibrated to suit a different organization or context.

Step 2: Linguistic Definition of Detection/Control

To define the Detection/Control linguistic terms, some deviations were considered during this

study from the traditional FMEA definition. The meaning of control of root causes and

controlling the effect of the risk event have been incorporated in the definition of

Detection/Control. The need for this deviation can be justified as follows. As the project evolves

to the execution phase, detection of any risk event, even with sufficient time, can not be

considered in isolation of the ability to control the risk to judge the level of criticality of the risk.

Some of the risk events might be easily detected before they are realized, yet team members have

little influence in controlling these risk events.

The idea of altering the meaning of Detection is not new to our proposed approach. In

fact, some of the previous work in this area noted the requirement of considering some

modification to the meaning of Detection. For instance, Rhee and Ishii (2003) noted confusion

and difficulty in understanding the Detection definition and asked the following questions: “Are

we trying to measure how easy it is to detect where a failure has occurred or when it has

occurred? On the other hand, are we trying to measure how easy or difficult it is to prevent

failures?”. Bowles (2003) stated that “the category name is misleading since it is a measure of

whether subsequent testing will show that the failure mode actually exists rather than whether the

failure will be detected when it occurs”. Carbone and Tippett (2004) have adopted a different

meaning from the traditional definition of Detection. They define Detection as: “the ability of



Accep

ted M

anus

cript

Not Cop

yedit

ed

10

detection technique or method(s) to detect the risk event with enough time to plan for a

contingency and act upon the risk.”

Several discussions were conducted with three risk experts in the participating

organization. According to these discussions, Detection/Control (D) is defined as “the ability of

the risk response strategy to detect and control the root causes before they lead to the occurrence

of the risk event, and to control the effect given the occurrence of the risk event“. Table 3

provides five linguistic definitions of the Detection/Control rating as developed with these

experts. Again, the meaning of each of these linguistic terms can be calibrated to suit a different

organization or context.

Step 3: Membership Function Definition for Input Factors

The third step entailed the development of the membership functions for each of the input

factors. During this process, the senior risk coordinator was asked to define five membership

functions for each of Probability of Occurrence (P), Impact (I), and Detection/Control (D)

according to the definitions shown in Table 1, Table 2, and Table 3 respectively. Triangular and

trapezoidal membership function shapes were chosen, since they are intuitive to experts. The

direct method with one expert (Klir and Yuan 1995) was used to elicit the ranges of each

membership function. To elicit the membership functions from the expert, questions similar to

this were asked: "What is the degree of membership of 1 in “very low”?". Figure 2 shows the

findings from the elicitation process. The membership functions for (P) and (I), represented by



Accep

ted M

anus

cript

Not Cop

yedit

ed

11

(CI), (TI), (SI), are similar as they both cover the same universe of discourse. The membership

functions for (D) have the reverse order of linguistic terms as compared to (P) and (I).

Step 4: Membership Function Definition for RCN

The senior risk coordinator was also involved in defining the membership functions for the RCN.

It was agreed that nine linguistic terms (very low, very low-low, low, low-medium, medium,

medium-high, high, high-very high, and very high) would be sufficient to cover the universe of

discourse for the RCN. As per the expert’s request, eight options were proposed for further

examination. Each of the nine linguistic terms was bounded by the range of values of the RCN.

Figure 3 shows the RCN membership functions and a tabular definition of the corrective actions

associated with each linguistic term. If the RCN falls within the range defined by categories 6 to

9 shown in Figure 3, then the “Yes” path (i.e., recommend corrective actions) in Figure 1 is

chosen. The threshold category for recommending corrective actions (e.g., 6 in this case) can be

modified depending on the risk tolerance of the user of the system.

Step 5: Fuzzy Expert System for FMEA

The next step in the construction of the fuzzy FMEA expert system is to build the rules relating

the input variables (P, I, and D) to the output variable (RCN). As noted previously, any fuzzy if-

then rule is composed of two parts: a premise and a conclusion. The premise considers different



Accep

ted M

anus

cript

Not Cop

yedit

ed

12

combinations of input variables, while the conclusion represents the output given the selected

interaction between the input variables. Since we have three inputs (P, I, and D) and each one is

represented with five linguistic terms, 5 3 =125 rules can be generated to cover all of the possible

input combinations. An interview was arranged with the senior risk coordinator to elicit the rules.

The following is a sample if-then rule:

If Impact (I) is “very low” and Probability of Occurrence (P) is “very low” and Detection

/ Control (D) is “high” then RCN is “very low”.

The membership functions for each variable together with the fuzzy if-then rules were

utilized to build the fuzzy expert system. The minimum operator was selected for input

aggregation; the product operator for selected for implication; the maximum operator was

selected for rule aggregation; and, the center of area was selected for defuzzification. The fuzzy

FMEA decision support system was implemented using Fuzzy Tech 5.72 © (Inform GmbH 2007)

Step 6: Fuzzy AHP for Cost, Time, and Scope/Quality Impact

In order to address the fact that the RPN may be underestimated when a failure mode has

multiple dimensions of effect (Bowles and Peláez 1995), the fuzzy AHP concept has been

adopted in this study. The Analytical Hierarchy Process (AHP) was first established by Saaty

(1982) to aid in decision making for problems that involves multiple criteria. Applying this

concept entails hierarchical formatting of the problem, establishing the pairwise comparison

matrix in accordance with the pairwise comparison scale, as shown in Table 4, and determining

the aggregated weight (AW) for each alternative.



Accep

ted M

anus

cript

Not Cop

yedit

ed

13

AHP has been utilized extensively by researchers in many applications including risk

management (e.g., Dey 2003). However, one shortcoming of using the AHP method is attributed

to the level of uncertainty and subjectivity of selecting a single number from the pairwise

comparison scale. Zeng et al. (2007) noted that experts sometimes found difficulties in selecting

a single number for comparison and argue the advantage of allowing for a range values for

comparison, such as 3 to 6, rather than a definitive number. To address this deficiency, the fuzzy

AHP approach, as proposed by Zeng et al. (2007), is adopted in this study. The pairwise

comparison scale, shown in Table 4, was presented to the senior risk coordinator who was asked

to do a pairwise comparison of the relative importance to the project objectives between Time

(T) vs. Cost (C), named (atc); Cost (C) vs. Scope/Quality (S), named (acs); and, Time (T) vs.

Scope/Quality(S), named (ats). The preference of the expert was captured using a Standard

Trapezoidal Fuzzy Number (STFN), i.e., each preference was captured using four parameters (a,

b, c, d) as shown in Figure 4. For instance, when comparing Cost (C) vs. Scope/Quality (S) to

calculate (acs), the expert believed that the minimum is that Cost (C) is between equal to

moderate more importance than Scope/Quality (S) (i.e., a=2), the most likely is that Cost (C) is

moderately more important than Scope/Quality (Q) (i.e., b=c=3), and the maximum is that Cost

(C) is between moderate to strong more importance than Scope/Quality (S) (i.e., d=4). The

defuzzified value for each preference was calculated according to equation [1]:

aij = 6

)(*2 dcba +++ [1]

Where aij is the relative importance of factor i over j. For example, the defuzzified value

of Cost (C) vs. Scope/Quality (S) is calculated as follows:

acs = 6

4)33(*22 +++ = 3 [2]



Accep

ted M

anus

cript

Not Cop

yedit

ed

14

The corresponding reciprocal value 1/3 is used for the reverse comparison, i.e.,

Scope/Quality (S) vs. Cost (C). The same concept was applied to calculate the defuzzified value

of Time (T) vs. Cost (C), and Time (T) vs. Scope/ Quality (S) and their reciprocal terms.

Thereafter, standard AHP calculations and consistency analysis were conducted as shown

in Table 5 (Saaty 1982). Equation [3] shows the calculation for the consistency index.

Consistency Index =1

max

−−

n

nλ=

13

3006.3

−−

=0.003 [3]

Where is the maximum eigenvalue, and n is the dimension of the pairwise matrix.

The result of the analysis is consistent since the consistency index is less than 0.1 (Pedrycz and

Gomide 2007).

According to the AHP results in Table 5, the aggregated impact (AI) is calculated as

shown in equation [4], and is extended over the universe of discourse of 1 to 10 with

membership functions similar to CI, TI, and SI in Figure 2.

AI= 0.40 * Cost Impact + 0.46 * Time Impact + 0.14 * Scope/Quality Impact [4]

To show how this system works, let us assume that we have the following assessments

for one of the risk events:

Probability of Occurrence (P) = 7, Cost Impact = 5, Time Impact= 5, Scope/Quality Impact= 3,

and Detection/Control = 5

Firstly, the AI is calculated according to equation [4], as shown in equation [5]

AI= 0.40 * 5 + 0.46 * 5 + 0.14 * 3= 4.72 [5]

Secondly, the assigned value for P and D together with the calculated AI are used as input to the

fuzzy FMEA expert system. Input values are fuzzified according to Figure 2, and the

membership values (x) of each linguistic term are calculated. The fuzzy inference process fires

the appropriate rules, aggregates the inputs, performs rule implication and aggregation, and



Accep

ted M

anus

cript

Not Cop

yedit

ed

15

defuzzifies the result to obtain a value of RCN of 286. Thus, according to Figure 3, based on the

value of RCN, there is a moderate priority to take corrective action(s), and the organization

should consider mitigation or transfer of the risk.

Step 7: System Implementation (Risk Criticality Analyzer)

The proposed approach has been used to design software entitled: “Risk Criticality Analyzer “

(RCA) implemented using Visual Studio 2008 ©. RCA was built to support decision making for

matters related to risk criticality by providing the top management with valuable risk reporting

features. Figure 5 shows the main screen of the RCA software.

RCA is composed of two modules. The first module is based on manual manipulation of

data, in which the user provides an assessment of (P), (CI), (TI), (SI), and (D) using the

numerical scale 1 to 10, which is associated with linguistic terms. The system automatically

calculates the (AI) according to equation [4]. The software also supports the user in conducting

the fuzzy AHP analysis to update the weightings in equation [4]. By clicking on the “Update”

button, the user is directed to the fuzzy AHP Excel sheet, which supports conducting fuzzy AHP

calculations. The last step is to calculate the (RCN) by clicking on “Calculate” button, which

automatically connects with the fuzzy expert system, export values of the input variables and

conducts the fuzzy inference process to calculate a value of RCN. This value of RCN is

presented to the user together with the recommended corrective actions.

The second module is the “Automated Risk Criticality Analysis”. The first step in using

this module is to load data from the risk register template by clicking on the “Load Data” button.



Accep

ted M

anus

cript

Not Cop

yedit

ed

16

The risk register contains several fields for each risk event including: Risk ID, Risk Description,

Root Causes, Risk Impact, Affected Work Package, Risk Response Strategies, and an assessment

of (P), (CI), (TI), (SI) and (D). Data are automatically imported to the software and presented to

the user in the data grid as shown in Figure 5. By clicking on the “Calculate” button, RCA will

read (P), (CI), (TI), (SI), and (D) for each risk event, export these values to the fuzzy expert

system to calculate the RCN, and present the resultant RCN and the recommended corrective

actions to the user in an easy to use Excel report.

Model Verification

To verify the validity of the fuzzy FMEA expert system and fuzzy AHP approach for risk

management, a meeting was conducted with a group of six experts at the participating

organization consisting of three risk analysts, the manager of the project management office, one

practices and standards coordinator, and the senior risk coordinator. The purpose of the meeting

was to present the traditional approach of applying FMEA, its drawbacks, the proposed approach

to address these limitations, and an introduction to RCA. Thereafter, the expert group was

encouraged to raise questions and provide feedback. The feedback received on the proposed

approach from all experts was positive. The next step was to validate RCA with a case study

project, utilizing the risk register from an actual project. The approximate capacity of the chosen

pipeline project is 70,000 barrels per day (bpd) and is composed of new and refurbished pump

stations, metering facilities, and substations. The anticipated in-service date of the project is mid-

year 2009. The risk register of this project is populated with data related to 41 active risk events,



Accep

ted M

anus

cript

Not Cop

yedit

ed

17

their Probability of Occurrence (P), and their Cost Impact, Time Impact, and Scope/Quality

Impact. In order to assess the level of Detection/Control (D), another meeting was conducted

with the risk analyst in charge of the project. RCA was then utilized to calculate the RCN and the

required corrective actions for each risk event.

The results obtained from the case study project using RCA were validated with both the

senior risk coordinator and the risk analyst in charge of the project, both of whom confirmed the

validity of the results and the usefulness of RCA in practice. The experts consulted in the

development of RCA noted several different practical uses of RCA, as follows:

To support comparison at the project level as well as at portfolio level by comparing total

RCN for each of the nine corrective action categories shown in Figure 3.

To aid in identifying which risk events should undergo a detailed root cause analysis. For

example, all risk events that fall under categories 6 to 9 in Figure 3 should be considered

for detailed root cause analysis.

To aid in determining which risk events should be used for lessons learned. For example,

all risk events that fall under categories 6 to 9 in Figure 3 should be tracked monthly, and

detailed lessons learned should be documented.

To determine the training needs for different risk individuals by identifying areas of

weakness within each project.

Conclusions and Future Work

The model presented in this paper offers the contribution of combining fuzzy logic with both

FMEA and AHP in a comprehensive framework that provides a practical and thorough approach



Accep

ted M

anus

cript

Not Cop

yedit

ed

18

for assessing the level of criticality of risk events in the construction domain. The combination of

fuzzy FMEA with fuzzy AHP and their integrated application to risk criticality assessment in

construction are unique. The model is currently in use by the company as a risk monitoring and

control tool to support its top management to identify critical risk events in a timely manner so

that corrective actions can be established more effectively. Use of the model over time will help

in validating its applicability and usefulness in practice.

Fuzzy logic was utilized to address the limitations that some combinations with lower RCN

should be given more attention than others with higher RCN, and to take into account the relative

importance of input factors in calculating the RCN. Moreover, instead of relying on rules of

thumb to identify the thresholds for the RCN, Figure 3 was created to support the project team

members in communicating the required corrective actions in a more objective way that suits the

construction domain.

The fuzzy AHP approach has been adopted in this study to solve the multi-criteria decision

making problem, in which Cost Impact, Time Impact, and Scope/Quality Impact are required to

be aggregated into one term. Risk Criticality Analyzer (RCA) was developed to implement the

proposed framework. The results obtained confirmed the capability and the usefulness of the

method to produce valid FMEA results.

Despite these improvements, future research is needed. The direct method with one expert

was used to elicit the different membership functions. Any modification to the linguistic terms,

for instance using seven linguistic terms to define Detection/Control (D) instead of five, will

require re-elicitation of the relevant membership functions. Further research is required on how

to calibrate the membership functions without requiring full re-elicitation. In addition, the

recommendations of RCA are limited to establishing the priority for corrective actions and



Accep

ted M

anus

cript

Not Cop

yedit

ed

19

recommending a category for the response strategy (i.e., accept, mitigate, transfer, or avoid).

RCA does not provide recommendations for specific corrective actions. Future work is required

to address this limitation by developing a database of recommended corrective actions that are

suitable for each specific risk, partially based on historical data and lessons learned.

The application of FMEA to the construction domain entails several challenges. For

instance, in order to implement an effective FMEA at each stage of the project life cycle, the up-

front allocation of resources is required, which is not always feasible in the construction industry.

Identification of the potential risk events at each stage in the project life cycle is another

challenge, since many root causes may interact to cause the risk event to occur. To overcome

such challenges, organizations need to create a risk-based culture in each functional area within

the organization. Life-cycle cost analysis can be used to demonstrate the potential cost savings

that can be gained by applying FMEA to the risk assessment of an organization’s projects. Buy-

in to the benefits of applying FMEA by top management is crucial to the successful

implementation of this technique.

Acknowledgements

During this study, a great deal of support was received from many people within Enbridge

Pipelines. The authors would like to express their appreciation to Mr. Guy Krepps and Mr.

Fernando Martinez for all the time, effort, and feedback received during this research. We would

also like to thank the entire stakeholder group for their input and valuable feedback. This work

was financially supported by the NSERC Associate Industrial Research Chair in Construction

Engineering and Management at the University of Alberta under a Natural Sciences and

Engineering Industrial Research Chair grant NSERC IRCPJ 349527-05.



Accep

ted M

anus

cript

Not Cop

yedit

ed

20

References

Ayyub, B. M. (2003). Risk analysis in engineering and economics, Chapman & Hall/CRC, New

York, Chapter 2, 33-113.

Bowles, J. B., and Peláez, C. E. (1995). “Fuzzy logic prioritization of failures in a system failure

mode, effects and criticality analysis.” Reliability Engineering & System Safety, 50(2),

203-213.

Bowles, J. B. (2003). “An assessment of RPN prioritization in a failure modes effects and

criticality analysis.” Proc., Ann. Reliability and Maintainability Symp., IEEE, Tampa,

Florida, 380-386.

Carbone, T. A. and Tippett, D. D. (2004). “Project risk management using the project risk

FMEA.” Engineering Management Journal, 16(4), 28-35.

Dey, P. K. (2003). “Analytic hierarchy process analyzes risk of operating cross-country

petroleum pipelines in India.” Natural Hazards Review, ASCE, 4(4), 213-221.

Justiniano, J. M. and Gopalaswamy, V. (2003). Practical design control implementation,

Interpharm/CRC, NY, Chapter 8, 133-154.

Inform GmbH (2007). FuzzyTECH ® user’s manual, Inform Software Corporation.

Klir and Yuan (1995). Fuzzy sets and fuzzy logic: theory and applications, Prentice Hall PTR,

Chapter 10, 280-301.

MIL-STD-1929A (1980), Military Standard- Procedures for performing a failure mode effects

and criticality analysis, 2, US Department of Defense, Washington DC, USA.

Pedrycz, W., and Gomide, F. (2007). Fuzzy systems engineering: toward human-centric

computing. John Wiley and Sons, Hoboken, N.J, Chapter 4, 67-99.



Accep

ted M

anus

cript

Not Cop

yedit

ed

21

Pillay, A. and Wang, J. (2003). “Modified failure mode and effects analysis using approximate

reasoning.” Reliability Engineering & System Safety, 79(1), 69–85.

Puente, J., Pino, R., Priore, P., and Fuente, D. (2002). “A decision support system for applying

failure mode and effects analysis.” Int. J. of Quality & Reliability Management, 19(2),

137-150.

Project Management Institute (PMI) (2004). “A guide to the project management body of

knowledge.” 3rd Ed., Wexford, Pa, Chapter 11, 237-268.

Rhee, S. J. and Ishii, K. (2003). “Using cost based FMEA to enhance reliability and

serviceability.” Adv. Eng Inf., 17(3-4), 179–188.

Saaty, T. L. (1982). Decision making for leaders: the analytical hierarchy process for decision in

a complex world, Lifetime learning, Belmont, California, Chapter 5, 75-92.

Thompson, P., and Perry, J. G. (1992). Engineering construction risks: a guide to project risk

analysis and risk management. Thomas Telford, London.

Xu, K., Tang, L. C., Xie, M., Ho, S. L., and Zhu, M. L. (2002). “Fuzzy assessment of FMEA for

engine systems.” Reliability Engineering & System Safety, 75(1), 17-29.

Zadeh, L.A. (1965). “Fuzzy sets.” Inf. Control., 8(3), 338-353.

Zeng, J., An, M., and Smith, N. J. (2007). ”Application of a fuzzy based decision making

methodology to construction project risk assessment.” Int. J. Proj. Manage., 25(6), 589-

600.



22

Table 1. Linguistic definition of Probability of Occurrence (P)

Description Term Probability of Occurrence

Very High (VH) > 67% (2/3) chance. Event will certainly occur High (H) Between 33% - 67% (2/3) chance. Event is expected to

occur. Medium (M) Between 10% - 33% (1/3) chance. Event may occur. Low (L) Between 1% - 10% chance. Event is unlikely to occur. Very Low (VL) Less than 1% chance. Event is highly unlikely to occur.

Accepted Manuscript Not Copyedited



23

Table 2. Linguistic definition of Impact (I)

Terms Impact Categories

Cost Time Scope/Quality Very High

(VH) ≥ 10% of project cost. In service date delayed ≥ 10 % of project

duration. Project scope or quality does not meet business expectations.

High (H)

Cost increase is ≤ 7% and < 10% of project cost.

In service date delayed ≤7% and < 10% of project duration.

Scope changes or quality are unacceptable to project sponsor.

Medium (M)


In service date delayed ≤ 4% and < 7% of project duration.

Major areas of scope or quality are affected.

Low (L)


In service date delayed ≤ 1% and < 4% of project duration.

Few areas of scope or quality are affected.

Very Low (VL)

<1% of project cost. Insignificant schedule slippage. Scope change is not noticeable/ quality degradation is not noticeable.




24

Table 3. Linguistic definition of Detection/Control (D)

Term Detection /Control Very Low

(VL) The project team was unable to identify a risk response strategy capable of detecting the risk event, controlling root causes, and controlling the consequence of the risk event.

Low (L) The project team has identified a risk response strategy with a low chance of detecting the risk event, controlling the root causes, and controlling the consequence of the risk event.

Moderate (M) The project team has identified a risk response strategy with a moderate chance of detecting the risk event, controlling the root causes, and controlling the consequence of the risk event.

High (H) The project team has identified a risk response strategy with a high chance of detecting the risk event, controlling the root causes, and controlling the consequence of the risk event.

Very High (VH)

The project team has identified a risk response strategy that has been proven in the past to have high effectiveness in detecting the risk event, controlling the root causes, and controlling the consequence of the risk event.




25

Table 4. The pairwise comparison scale (Adapted from Saaty 1982)

Scale Definition 1 Equal importance of both elements 3 Moderate importance of one element over another 5 Strong importance of one element over another 7 Very strong importance of one element over another 9 Absolute importance of one element over another

2, 4, 6, 8 Intermediate values




26

Table 5. Pairwise comparison matrix and consistency check

C T S Normalized C T S ∑ AW

C 1.000 0.800 3.000 C 0.387 0.375 0.429 1.191 0.397 T 1.250 1.000 3.000 T 0.484 0.469 0.429 1.381 0.460 S 0.333 0.333 1.000 S 0.129 0.156 0.143 0.428 0.143 ∑ 2.583 2.133 7.000 ∑ 1.000 1.000 1.000

C*0.397 T*0.460 S*0.143 ∑ λλλλmax Overall λλλλmax C 0.397 0.368 0.428 1.193347 3.006748911

3.006 T 0.496 0.460 0.428 1.384649 3.007507038

S 0.132 0.153 0.143 0.428475 3.002354524




Accep

ted M

anus

cript

Not Cop

yedit

ed

27

List of Figures

Figure 1 Proposed framework for applying FMEA

Figure 2 Membership functions for (P), (CI), (TI), (SI), (AI) and (D)

Figure 3 Membership functions of the RCN and corrective action categories

Figure 4 Fuzzy analysis (cost, time, scope/quality)

Figure 5 Risk Criticality Analyzer V (1.04)



Figure 1




Figure 2




Figure 3




Figure 4




Figure 5




risk management in the construction industry using

Documents