A Nonparametric Bayesian Network Model for Predicting Corrosion Depth on Buried Pipelines

Abstract

The present study develops a nonparametric Bayesian network (NPBN) model to predict the corrosion depth on buried pipelines using the pipeline age and local soil properties. The dependence structure and parameters of the NPBN model are extracted from Velázquez’s dataset, which consists of 250 samples of corrosion depths, pipeline age, and such local soil properties as the water content, redox potential, and pH value. The NPBN models the joint distribution of the corrosion depth, pipeline age, and local soil parameters by a Gaussian copula. The five-fold cross-validation is used to examine the predictive capability of the developed NPBN model. The results indicate that the predicted mean values of corrosion depths in general agree well with the corresponding field measurements, and more than 95% of the field-measured depths are within the 5 to 95 percentile range of the predicted distribution for the corrosion depth. The NPBN and the associated model mining method provide an effective data-driven approach to develop predictive models of corrosion depths using soil parameters as predictors. The developed NPBN will benefit the corrosion management of pipelines for which direct inspections are infeasible.