Expert System Development to Predict Canned Motor Pump Status

Abstract

This research presents the development of an Expert System to predict Canned Motor Pump (CMP) Status by applying a machine learning (ML) algorithm with domain expert knowledge in the case study plant. A Case study plant is a petrochemical plant that uses CMP to transfer process medium within inside plant battery limit (ISBL). At present, The CMP maintenance strategy is improving from condition-based maintenance to predictive maintenance. To archive desired level of predictive maintenance need CMP domain expert knowledge to find potential failure signs. This expert system is contributing to reducing expertise human load by substitution with the system. The research contains identifying system framework, experiment steps, including dataset preparation and model testing. The experiment result shows Random Forest (RF) algorithm is suitable for this system due to model performance evaluation comparing four algorithms with confusion matrix and similar data resampling and hyperparameter tuning method. Further on, this contribution is a role model, and enrolling in other equipment in the case study plant is a benefit of this work. Recommendation and key success factors found during this research are also mentioned in the conclusion for further work as a continuous improvement process cycle.