A comparison of autoregressive modeling techniques for fault diagnosis of rolling element bearings

The paper introduces the concept of fault diagnosis using an observer bank of autoregressive time series models. The concept was applied experimentally to diagnose a number of induced faults in a rolling element bearing using the measured time series vibration signal. Three distinct techniques of autoregressive modeling were compared for their performance and reliability under conditions of various signal lengths. The results indicate that backpropagation neural networks generally outperformed the radial basis functions and the traditional linear autoregressive models. This modeling technique for fault diagnosis was found to require much shorter lengths of vibration data than traditional pattern classification techniques used in the field of machine condition monitoring. (C) 1996 Academic Press Limited