Page 107 - Reliability and Maintainability of In service Pipelines

96 Reliability and Maintainability of In-Service Pipelines

A reliability-based methodology for assessment of corroded steel gas pipelines
was presented by Mahmoodian and Li (2017). They developed a stochastic model
for strength loss which relates to key factors that affect the residual strength of a
corroded pipe. The failure of pipeline was defined as when pipe residual strength
falls below its operating pressure. An analytical time-variant method was
employed to quantify the probability of failure due to corrosion so that the time
for the pipeline to fail and hence require repairs, could be determined with confi-
dence. To deal with the assessment of pipelines with more than one corrosion pit,
they employed a system reliability analysis method. Monte Carlo simulation tech-
nique was applied to verify the results of the analytical method.

3.6.3 OTHER MODELS

Statistical models initially forecast the number of pipe failures with the use of
maintenance records and failure data. Recently, some researchers have also used
statistical models to predict pipe failure. Kleiner and Rajani (2008) examined the
use of a nonhomogenous Poisson model and evaluated factors that affect water
pipes. Berardi et al. (2008) proposed the application of performance indicators to
model pipe deterioration rate using the method of evolutionary polynomial regres-
sion (EPR). Wang et al. (2009) developed deterioration models in their statistical
analysis to predict the annual break rates of water mains considering the pipe
material type diameter, age, and length.
Compared with deterministic models, statistical models take into account his-
torical data and their variables in the prediction of pipe failures. Statistical models
can be applied to pipes that have an adequate and reliable historical database over
time. Therefore, applicability of statistical models is limited when considering
cases with insufficient monitoring data.
Artificial Neural Networks (ANN) predict pipe deterioration rates by utilizing
all variables that influence the service life of a pipe. Each analyzed parameter can
increase the system performance and reliability. Parameters can be prioritized by
applying weights and learning algorithms. A high level of skill is involved in
developing these complex networks. Data preprocessing, training, and testing
methods for selecting appropriate network are also required.
Christodoulou et al. (2004) examined the deterioration of water mains using
ANN techniques from parametric and nonparametric analyses. Al-Barqawi and
Zayed (2008) developed a condition rating model to assess the rehabilitation pri-
ority for water pipes using an ANN. The output variable consists of a condition
rating scale from 0 to 10 with 0 representing a critical condition and 10 represent-
ing an excellent condition.