Detection of Broken Rotor Bars Fault in Induction Motors by Using an Improved MUSIC and Least-Squares Amplitude Estimation

Abstract

The frequencies and amplitudes of the broken rotor bar (BRB) fault features are the basis for the accurate diagnosis of the BRB fault. However, how to accurately detect their frequency and amplitudes has always been a difficult problem for induction motor fault detection. For this problem, a new fault detection method based on an improved multiple signal classification (MUSIC) and least-squares magnitude estimation is proposed. First, since the fixed-step traversal search reduces the computational efficiency of MUSIC, a niche bare-bones particle swarm optimization (NBPSO) for multimodal peaks search is proposed to improve MUSIC, which is used to compute the frequency values of fault-related and fundamental components in stator current signal. Second, using these frequency values, a fault current signal model is established to convert the magnitude estimation problem into a linear least-squares problem. On this basis, the amplitudes and phases of fault-related and fundamental components could be estimated accurately with the singular value decomposition (SVD). A simulation signal is used to test the new method and the results show that the proposed method not only has higher frequency resolution, but also improves estimation accuracy of parameters greatly even with short data window. Finally, experiments for a real induction motor are performed, and the effectiveness and superiority of the proposed method are proved again.