关键词: 滚动轴承, 故障诊断, 稀疏分解, 相关正交匹配追踪, 相关系数

Abstract: Aiming at the problems that effective impulse feature was difficult to detect because the fault signals collected from rolling bearings in wind turbines was non-stationary and submerged by strong noises，a sparse fault diagnosis method was proposed based on COMP algorithm，the correlation coefficients between atoms and residuals were calculated according to the inner product size after each iteration，the atom with the largest correlation coefficient was merged with other qualifying atoms，and the merged atom was treated as a new atom.Then，a new dictionary with strong correlation with the signals was reconstructed by these new atoms，and the signals were sparse represented by the dictionary.Finally，the accurate diagnosis of rolling bearing failures was achieved by analyzing the envelope spectrum of the results of sparse representation.Because of the strong correlations between the new atoms reconstructed by this method and the residuals，a high sparse representation precision may be obtained with only a small number of iterations.Simulations and engineering applications were performed to verify the validity and practicability of the proposed method.

Key words: rolling bearing, fault diagnosis, sparse decomposition, correlation orthogonal matching pursuit（COMP）, correlation coefficient