Fault feature extraction based on multifractal and singular value decomposition for reciprocating compressors

Here, a fault feature extraction method based on multifractal and singular value decomposition of multi-sensor was presented, aiming at interference and coupling of fault information and complex non-linear, and non-stationary characteristics of vibration signals in a reciprocating compressor. The generalized fractal dimension number could characterize local scale behavior of a signal more appropriately, so an initial feature matrix was built by calculating the generalized fractal dimension number of multi-sensor signals. The matrix was compressed with the singular value decomposition method, and its eigenvalues were taken as feature vectors. Taking a reciprocating compressor transmission mechanism as a study object, feature vectors of bearing clearance faults of different positions were extracted from vibration signals. A support vector machine was established as a pattern classifier to identify faults. Compared with results of the single sensor multifractal method and the multi-sensor single fractal method, the validity of this proposed method was verified.