Incipient Feature extraction based on singular value decomposition and undecimated lifting scheme packet

Abstract

Vibration signal measured from machinery is often heavily interfered with by various noises. This paper puts forward a joint method to reduce noises, acquire the enhanced signals from the decomposed subbands and extract the incipient fault features. First, the signals are denoised by the method of singular value decomposition (SVD). Then, the denoised signal is decomposed into four layers by undecimated lifting scheme packet (ULSP). Finally, all 16 subbands of the fourth layer are plotted and the rich-fault-information subbands are used to extract incipient features. The effectiveness of the proposed method is validated with simulated data. Furthermore, in the processing of engineering signal, the weak feature caused by the fault of a valve in reciprocating compressor is bulged and the early failure of spring is detected.