Insights on an artificial neural network (ANN) based expert system for screening failures due to microbiologically influenced corrosion (MIC) in oil and gas upstream pipelines

The analysis of pipeline failures due to Microbiologically Influenced Corrosion (MIC) is challenging due to the complex interaction of many influencing parameters including pipeline operation conditions, fluid chemistry and microbiology, as well as the analysis of corrosion features and products. To help address this challenge, an expert system was developed to assist specialists and non-specialists in screening internal pipeline corrosion failures due to the threat of MIC. To that end, 15 MIC subject matter experts (with a total of 355 man-years of accumulated MIC based experience) were recruited to evaluate a total of 65 MIC failure cases based on real-life scenarios. These case study parameters and the expert elicited results were input into an Artificial Neural Network (ANN) model to create a model system which can screen whether a given failure scenario is one of three outcomes: a) failure is likely due to MIC, b) failure is likely not due to MIC, or c) the conclusion is inconclusive (analysis needs more data/information). The model system had an overall accuracy of 74.8% and it showcases that knowledge from subject matter experts can be captured in a reasonably effective way to screen for possible MIC failures. Based on that, this presentation will provide details of the model development process and key results to date. Important considerations regarding the level of confidence of the diagnoses and variation between expert opinion will also be discussed alongside with ideas on how to improve the model for field applicability.