Fault identification and classification of rolling element bearing based on time-varying autoregressive spectrum

Rolling element bearing faults are among the main causes of breakdown of rotating machines and its condition monitoring based on vibration signal has been used extensively. For obtaining more accurate time–frequency spectrum estimation, time-varying autoregressive method based on Kalman smoothing algorithm is utilized to realize parametric modeling of non-stationary signal so as to obtain high resolution time–frequency spectrum. Singular value decomposition (SVD) method is adopted to obtain the first left and right singular vectors of time–frequency spectrum. And by down sampling and preprocessing, these singular vectors are taken as feature vectors of time–frequency spectrum. Moreover, radial basis function (RBF) neural network is adopted to realize the automated classification. By classification of rolling element bearing in four kinds of different status, the results show that algorithm mentioned above can realize the automated and accurate diagnosis of bearing fault.