Sparse representation-based classification for rolling bearing fault diagnosis

Rolling element bearing is a vital but also an easily damageable part for rotating machinery. In this paper, sparse representation-based classification is introduced in the method of fault diagnosis for rolling bearing. In the method, Wavelet Packet Transform and Fisher discrimination criterion are used to construct the sample feature dictionary. It takes advantage of the Orthogonal Matching Pursuit algorithm to get sparse coefficients and diagnose fault according to the minimal ratio of reconstruction error to L1-norm of the sparse coefficients. Its performance is demonstrated in test of faulty bearing vibration signals, and its advantage is further validated through comparison with other classification-based diagnosis methods such as support vector machine.