Automating Motor Predictive Maintenance (PdM) by real-time Explainable AI (XAI) and Deep Learning

Unplanned machine failures may interrupt regular operations resulting in production losses. These incur financial losses as well as additional time required to service, procure, and maintain components. Industry 4.0 revolution embodies the use of modern tools and technologies to establish predictive maintenance (PdM), where integrating sensors, IoT devices, and Machine Learning (ML) capabilities and enable fault prediction, diagnosis, and maintenance optimization can be achieved. However, only a fraction of industries currently deploy predictive measures due to its complexity, accuracy, and explainability limitations. This research presents a two-step approach for PdM, first, industrial sensors record motor shaft vibrations, misalignment, and equipment temperature to model faults. Several ML models were trained on this data to accurately classify the fault. Next, the Explainable AI (XAI) model SHapley Additive exPlanations (SHAP) was used to explain the fault nature, and the selection of most important features used by the model. This was further used to tune the Long-Section Term Memory Auto-Encoder (LSTM-AE) Network model accordingly, resulting in 98. 4% precision. Finally, the models were evaluated on several datasets at various motor speeds and shaft alignments, resulting in a consistent accuracy score.