Development of Machine Learning based Model for Anomaly Detection and Fault Cause Diagnosis of Assets in Petrochemical Industries

Abstract

Petrochemical companies put much effort into maximizing productivity and optimizing TCO(Total Cost of Operation) by reducing the unplanned downtime for stable operation of assets since unplanned downtime of assets leads to colossal production loss and environmental safety accidents. The PdM (Predictive Maintenance) solution is required to predict prognostic abnormal behavior of assets before the time when asset fault occurs, give warning alarm to engineers, and help them take proactive measures by diagnosing the fault cause and guiding suitable measures.In this research, the PdM model has been developed using Variational AutoEncoder and Isolation Forest algorithms to detect the prognostic abnormal behavior of assets before the unplanned shutdown. Moreover, PdM model for diagnosing the possible causes of abnormal behavior of the centrifugal compressor has also been developed to help domain field engineers take the suitable measures before the unplanned shutdown of the asset. By applying the PdM model to actual data of centrifugal compressor in petrochemical process, the PdM model has been successfully validated and shown feasible results.