Vibration Signal-Based Fault Detection for Rotating Machines

Abstract

Fault detection in rotating machinery has applications in fields such as wind turbines and helicopter transmissions. Detecting and diagnosing faults is important to maintenance planning, preventing equipment damage, and preventing failure. During this presentation, two novel signal-based methods will be presented; one based on adaptive control theory, one based on deconvolution. The adaptive control theory approach is an adaptive sum-of-sinusoid model used for one-step ahead prediction. The presented deconvolution approach is a periodic expansion of the well established Minimum Entropy Deconvolution method. Results are presented on simulated signals, acceleration data from a gearbox with seeded gear tooth faults, and bearing proximity sensor data from two 50MW back pressure steam turbine generators with suspected rotor-to-stator rubbing regions.