Page 108 - Reliability and Maintainability of In service Pipelines

Methods for Structural Reliability Analysis 97

Compared with other models, ANN modeling necessitates the use of all vari-
ables that influence the failure of a pipe. Higher degree of nonlinearity can be
taken into consideration by ANN models. Because these models basically depend
on actual data parameters, a limitation may be the lack of data utilities possess.
An increased level of skill and training is also required in order to develop the
complex networks in ANN models (Clair and Sinha, 2012).
Fuzzy logic models use engineering judgment and professional experience in
order to predict pipes’ deterioration processes. This type of model is often used
when data is scarce and observations and model criteria are expressed in vague or
“fuzzy” terms. This technique implements expert opinions. A high level of skill is
required in constructing the rule set and deciding the defuzzification process for a
reliable output.
Kleiner et al. (2005) proposed a fuzzy Markov deterioration process to predict
the future condition of CI pipes. The fuzzy set techniques help to incorporate the
imprecision and subjectivity of the data. Rajani and Tesfamariam (2007) applied
a fuzzy set theory for consideration of uncertainties to estimate the structural
capacity of aging CI water mains. Fares and Zayed (2010) developed a hierarchi-
cal fuzzy expert system to determine the risk-of-failure of water mains. The
system consists of 16 risk-of-failure factors within four main categories: environ-
mental, physical, operational, and post failure.
Compared with other deterioration models, fuzzy logic models necessitate the
use of fuzzy logic-based techniques that possess the ability to incorporate engi-
neering judgment to predict pipe deterioration. Fuzzy logic models are used for
systems that are subject to uncertainties, ambiguities, and contradictions. The pri-
mary limitation for fuzzy logic models is the challenges that exist in constructing
a fuzzy rule set, selecting a membership function, and determining a defuzzifica-
tion process (Clair and Sinha, 2012).
Heuristic models are rare and limited in nature, but can illustrate how method-
ologies incorporate engineering knowledge in the determination of deterioration
rates. This technique is a structured way of capturing expert opinions. A limita-
tion of using engineering knowledge is the inconsistency in the expert judgments
from individual to individual and/or lack of personal experience in making the
judgments. However, model capabilities can be improved by considering addi-
tional expert knowledge and opinions (Clair and Sinha, 2012).
Watson (2004) proposed a Bayesian methodology that combines engineering
knowledge with recorded failure data to establish failure rates. This hierarchical
model works to combine information from various sources of data. Al-Barqawi
and Zayed (2008) presented an integrated model utilizing an analytic hierarchy
process (AHP) and ANN. The AHP is first utilized to assign weights and assess
the current condition of a water main based on physical, environmental, and oper-
ational factors. Then an ANN, which is trained with the available data set,