| dc.description.abstract | As the safest means of transporting gas and hazardous materials, pipelines transport invaluable petroleum material. However, a considerable number of accidents have happened involving these facilities, leading to economic losses and environmental impacts. Several inspection techniques are used to provide safety for pipelines. Despite their accuracy, these techniques are time-consuming and costly. Some failure prediction and condition assessment models were recently developed to tackle these inefficiencies. However, most of these models only predict one failure source or they rely on subjective expert surveys. This research developed three objective models based on artificial neural network (ANN) and multinominal logit (MNL) regression to predict failure sources in oil pipelines. An ANN model was developed for prediction among mechanical, corrosion, and third-party failures with an average validity percentage (AVP) of 73.7%. Another ANN model was developed for prediction between corrosion or third-party failures with an AVP of 72.8%. In addition, an MNL model was developed for prediction among mechanical, corrosion, and third-party failures with an AVP of 73.7%. Pipeline operators and decision makers can use these models to identify pipeline failure sources. They can also be applied to prioritize in-line inspection to carry out appropriate maintenance. | |
