Feature extraction & application of engineering non-stationary signals based on EMD-AR model and SVD

According to the non-stationary characteristics of the vibration signals from reciprocating machine and the situation that it´s hard to obtain enough fault samples, a method based on empirical mode decomposition (EMD), auto regression(AR) mode, singular value decomposition (SVD) and support sector machine (SVM) is proposed in this paper. Firstly, with the help of EMD, the vibration signals are decomposed into a finite number of intrinsic mode functions (IMF), then AR model of each IMF components are established. The AR model parameters and variance of remnant are regared as initial feature vectormatrixes. Thirdly,by applying SVD technique to the vectormatrixes, the singular values are obtained and serve as the fault characteristic vectors to be input to SVM classifier. So the mechanical working condition and faults are classified. The results of engineering application show that this method have high accuracy and good generalization abilities even in the case of small number of samples and can also be applied to the fault diagnosis of other equipment.