An explainable predictive maintenance strategy for multi-fault diagnosis of rotating machines using multi-sensor data fusion

Abstract

Industry 4.0 denotes smart manufacturing, where rotating machines predominantly serve as the fundamental components in production sectors. The primary duty of maintenance engineers is to upkeep these vital machines, aiming to reduce unexpected halts and extend their operational lifespan. The most recent development in smart maintenance is Predictive Maintenance (PdM). Due to the diversity of machinery and the diverse behaviour of each machine in different fault conditions, the most challenging task in predictive maintenance is to detect the fault, diagnose the type of fault, and explain why a particular fault is predicted. This study proposes an effective Explainable Predictive Maintenance strategy considering (1) test setup building, (2) low-cost Fast Fourier Transform (FFT) raw data using multiple sensors, (3) multi-sensor data fusion, (4) comparing various multi-class classification algorithms, (5) analysis of cases concerning multi-sensor versus single sensor and multi-location versus single location, and (6) explainable predictive maintenance. Quantitative results from this study reveal a remarkable multi-fault detection accuracy and multiple fault type classification, with the highest accuracy of 100 percent. Furthermore, multi-sensor data fusion significantly outperforms single-sensor approaches, demonstrating an enhancement in fault prediction accuracy of all models. Using Explainable Artificial Intelligence methods contributes to the interpretability of fault diagnoses, making it a critical advancement in Intelligent Manufacturing and Predictive Maintenance in Industry 4.0. The study's novelty is using Explainable Artificial Intelligence (Local Interpretable Model Agnostic Explanation (LIME) and Random Forest) for multi-fault classification of rotating machines using multi-sensor data fusion.