Online health monitoring of rotating machine elements using statistical spectral distances

Abstract

The industry is increasingly recognizing the necessity for predictive maintenance solutions that are applicable for identifying various machinery faults and can be seamlessly integrated into online vibration-based condition monitoring systems. This article presents a methodology suitable for online monitoring based on statistical spectral image similarity analysis to diagnose incipient machine fault conditions induced under both stationary and nonstationary operating regimes. Discrete multivariate statistical distances are employed as similarity metrics for comparing the frequency/order spectra images computed for healthy and faulty conditions. An upper control limit to identify the anomalies and subsequently a root cause analysis to determine the underlying cause of the fault are proposed in the current work. The effectiveness of this method is illustrated by analyzing vibration data collected from extensive testing of gears, bearings, and shafts on various in-house test rigs, under both their normal and defective conditions. In addition, this method is employed for NASA IMS run-to-failure bearing dataset and found to be successful in identifying the initiation of the fault. The analyses demonstrate the proficiency of the proposed methodology in early fault detection and diagnosis, showcasing its effectiveness even with minimal raw vibration data.