Explainable Artificial Intelligence for Predictive Maintenance in Industry: A Survey, Methodology, and CNC Machine Case Study

Abstract

Predictive maintenance (PdM) is the use of both AI and sensor data to improve the ability to accurately determine when equipment will fail and therefore provide users with an optimal maintenance strategy. The combination of machine learning (ML) and deep learning (DL) technology provides even more accurate predictions on when the equipment will fail. The black-box characteristics of ML and DL models can prevent transparency and trust for industrial implementation. As a result, Explainable Artificial Intelligence (XAI) has been put forth as a means to provide understandable explanations regarding model performance and build trust between the end-user (technician) and the model. To close this gap, we propose a comprehensive review of existing PdM techniques, review existing XAI models and methodologies, identify key research gaps, and ultimately present an explanation-enabled methodology for implementing PdM via XAI. Our proposed methodology has been validated on an actual machine tool located at a CNC manufacturer. This case study demonstrates the benefits of providing an explanation for how the model predicted machine failure while simultaneously increasing trust in the model, enhancing fault diagnosis, and improving decision-making abilities without negatively affecting prediction accuracy.