An Investigation of Fault Signatures in Online Condition Monitoring Methods for Shaft Misalignment and Mass Unbalance Fault Diagnosis of PMSM Driven Systems

Abstract

Mechanical faults in permanent magnet synchronous motor (PMSM)-driven systems can introduce disturbances and system interruptions leading to reduced performance and reliability. Effective fault diagnosis is essential for early fault detection and identification, which enables timely maintenance, reduced downtime, and efficient operation of the system. This article presents an investigation of fault signatures in online condition monitoring methods, focusing on mechanical fault diagnosis of PMSM driven systems. Angular shaft misalignment and mass unbalance faults are investigated under two different severity levels. Three axis vibration, acoustic emission, stator currents, axial and radial stray flux, and shaft torque are evaluated for fault detection, severity assessment and fault discrimination. In the first place, a theoretical framework describing the influence of the specific mechanical faults on the mechanical and electromagnetic behavior of the PMSM is established, enabling the identification of characteristic fault-related signatures. An experimental test bench is developed, integrating a PMSM with vibration, current, stray flux, acoustic emission, and torque sensors. Fault signatures are analyzed in both nominal operating condition and variable speed and load levels. Moreover, load and speed transients are investigated. Finally, to extract the most informative sensors, a data driven approach with mutual information, random forest, and Shapley additive explanations is employed. Results from this experimental investigation highlight the tradeoffs between diagnostic performance, practical implementation, and the effectiveness of combining domain knowledge with data-driven approaches for accurate, early, and cost-effective PMSM fault diagnosis.