Autoregressive Model-Based Vibration Fault Diagnosis of Rolling Bearing

Abstract

In order to overcome low resolution and poor variance performance in the classic power spectrum, autoregressive model in modern spectrum estimation was developed to diagnose the vibration fault of rolling bearing with noise so that it can get high resolution, smooth power spectrum, and it can be able to identify the fault of rolling bearing, to determine the fault type and improve fault diagnosis rate. The resolution of autocorrelation method, Burg method, and improved covariance method with the simulated vibration signal of rolling bearing with noise was studied. The method to determine optimal order and the resolution of different sampling point for autoregressive model were presented. The results showed that Akaike information criterion, that is AIC- p, is a good method to obtain power spectrum on the order of smooth curve rather than the order of minimum AIC. If the sampling point number is larger, three methods can obtain the same resolution, but if the sampling point number is less, Burg method and improved covariance method can get better resolution than autocorrelation method.