Model-based adaptive frequency estimator for gear crack fault detection

Detection of gear cracks from vibration data is a difficult task. This paper investigates an alternative to the linear predictor residual fault detection based on the nonlinear adaptive control system concept of frequency estimators. The frequency estimator model takes advantage of the sinusoidal nature of vibration and adapts the system model during operation. The low-computational requirements, no-priori knowledge, sinusoidal-based prediction, and on-line model adaption makes this model ideal for on-line gear crack fault detection. Performance is evaluated through both synthetic and experimental data while comparing to the autoregressive linear predictor model.