Data-Driven Diagnostics Based on Non-invasive Monitoring Using Electrical Signals: Application to Rotating Machines

Nowadays, industrial machinery companies provide a wide propagation of manufacturing, in particular induction motors, due to their robustness and low costs. Indeed, with the advancement of power electronic converters, their integration offers promising perspectives for high reliability, maintainability, availability and safety systems. However, because of switch commutations in the converters, they affect the quality of data processing analyses for fault detection and diagnostics and therefore, more challenging for the system health assessment. In this regard, it is necessary to develop a practical methodology, based on the monitoring of converters measurements, to assess the system health state. This paper aims to propose a data processing technique based on the time-domain analysis. This technique allows features extraction to build an efficient health indicator that separates the different health states of the system. The health indicator is constructed using the Concordia transform applied to the converter of electrical signals such as three-phase current and voltage signals. The obtained results are then injected into machine learning classifier for fault detection and diagnostics. The performance and robustness of the proposed method are highlighted through an experimental test bench taking into account different fault types and various operating conditions.