Multiangle improved SVD method and its application in failure feature extraction of bearing

Abstract

Singular value decomposition (SVD) is widely applied in fault identification, but current studies focus on single objects. Meanwhile, the performance of SVD is dependent to the matrix to be constructed and option of valid order of singular value. Unexpectedly, current research usually does not take into account the feature of the rotating machinery itself and the fault signals, which lead to less-than-ideal results in denoising and fault identification of SVD. To solve these problems, according to the close relation of fault information and feature period of rotating machinery, a multiangle improved SVD (MI-SVD) method is proposed to extract information of compound failure of rolling bearing and intermediate bearing of aeroengine. The MI-SVD method constructs Hankel matrix based on periodicity of signals, and its embedding dimension is determined by the period of the minimum feature frequency, which solves the problem of randomness in determining them based on experience. Simultaneously, the proposed method performs a noncontinuous adaptive selection of singular values related to richness of fault information with multiple signal evaluation metrics. This solves the problem that important fault information is missed when selecting singular values continuously by dividing bounds. To assess the applicability of MI-SVD method for multiple objects, vibration signals of intermediate bearing of real aeroengines with different sensors mounting directions and rolling bearings with different composite failure types are verified and compared with a variety of different comparison methods. Results show that MI-SVD method can better control noise and accurately identify composite failures of intermediate bearings and rolling bearings, which has greater engineering application value.