Fault diagnosis of rolling bearings based on an improved time-varying autoregressive model

An improved time-varying autoregressive model was established for rolling bearing fault diagnosis based on combination of forward and backward estimation. By adopting time-varying forgetting factor, mean squared error based on forward and backward estimation was defined and partial derivatives were derived for weighted coefficients of basis functions to obtain their calculation formulas. Then, the recursion formulas of the weighted coefficients were derived using recursive least squares (RLS). Time-frequency analysis for simulation and experimental signals of a faulty bearing inner ring was conducted using the improved and unimproved models. The results showed that the improved model can overcome the unavailability of frequency estimation at the initial time, it has higher accuracy in temporal and frequency estimation, and better anti-noise performance; so the improved model can extract fault feature frequency of rolling bearing more effectively.