Search

• Expert system to aid microbiologically influenced corrosion failure assessments. • The model is based on machine learning Artificial Neural Networks (ANNs). • Fifteen experts evaluated 65 case studies related to internal pipeline failures. • Two ANN models were developed: one including confidence level and the other not. • This expert system is a potential tool for non-experts to screen pipeline failures. This study highlights the modeling of an expert system for the classification of internal microbiologically influenced corrosion (MIC) failures related to pipelines in the upstream oil and gas industry. The model is based on artificial neural networks (ANNs) and involves the participation of 15 MIC experts. Each expert evaluated a number of model case studies ranging from MIC- to non-MIC-driven upstream pipeline failures. The model accounts for variations in microbiological testing methods, microbiological sample types, and degradation morphology, among other variables. It also accounts for missing datasets, which is commonly the case in actual failure assessments. The outcome is an expert system model whose outputs are classes (classification ANN) which comprises MIC potential and data confidence. The performances of two approaches are contrasted in this study. One classifies the output in 5 classes, a 5-output classification (5OC) model; the other in 3 classes, a 3-output classification (3OC) model. The 5OC model had an accuracy of 62.0% while the simpler 3OC model had a better accuracy of 74.8%. This modelling exercise has demonstrated that knowledge from experts can be captured in a reasonably effective model to screen for possible MIC failures. It is hoped that this study not only may contribute to a better understanding of the prevalence of MIC in the oil and gas sector, but also may highlight the key areas necessary to improve the diagnosis of MIC failures in the future. [ABSTRACT FROM AUTHOR]