Abstract: An autoregressive model-based technique to detect the occurrence and advancement of gear shaft crack is proposed. The order of autoregressive (AR) model is selected by using Akaike information criterion on the time synchronous averaging signal of the gear shaft in its healthy-state, the coefficients of the AR model is determined by the solution of the Yule-Waker equations with the recursion algorithm. The established AR model is then used as a linear filter to process the future-state signal under the same condition. The Kolmogorov-Smirnov test is then performed over the prediction error signals. The statistical distance of the test is used as a fault parameter which is capable of diagnosing the gear shaft crack effectively with a higher level of confidence. The other statistical measures such as kurtosis, variance are also analyzed and computed. The performance of the technique is demonstrated by using a full lifetime gear shaft vibration data history.

Key words: Autoregressive model, Fault diagnosis, Gear shaft fault, Gearbox vibration, Kolmogorov-Smirnov test