Sparse feature extraction method for fault signal of rotating machine

Abstract

The invention provides a sparse feature extraction method for a fault signal of a rotating machine. The method includes the steps of constructing a fault overcomplete atom library according to a faultgeneration mechanism of an impact fault of the rotating machine; calculating a multi-resolution time-frequency spectrum of a to-be-analyzed signal through multi-resolution generalized S-transform resolution and normalization processing according to a mathematical model of a to-be-observed signal and the fault overcomplete atom library; calculating a maximum value of the normalized multi-resolution time-frequency spectrum, and calculating an atom set capable of best performing sparse representation on the to-be-observed signal by combination with an orthogonal matching tracking algorithm; andaccording to the fault sparse representation signal and fault feature frequency of a device and rotating speed information, determining a fault type, to achieve a rapid fault diagnosis of the mechanical device. The sparse feature extraction method has the advantages that the complexity of the atomic search and the search time are greatly reduced, the efficiency of sparse decomposition is improved,and the efficiency of fault diagnosis of impact faults such as crack, pitting or spalling is improved.

Description

technical field [0001] The invention relates to the field of fault diagnosis of rotating machinery, in particular to a method for extracting sparse features of fault signals of rotating machinery. Background technique [0002] During the operation of mechanical equipment, rotating parts are the key equipment to ensure the safe and stable operation of the equipment. Once a failure occurs, the light one will affect the production accuracy, and the serious one will cause major equipment failure and huge economic losses. There are two main types of vibration signals of rotating machinery, one is the smooth fault signal caused by wear, misalignment, unbalance, etc., and the other is the non-stationary impact fault signal caused by cracks, pitting, peeling, etc. Shock faults are more likely to cause equipment breakdown in a short period of time. Therefore, the feature extraction and fault diagnosis of shock faults are of great significance. [0003] The traditional Fourier analys...

AI Technical Summary

Problems solved by technology

[0005] However, there are two major defects in the above-mentioned patents: ① The window width of the short time window is fixed and not adaptive, especially for oscillation attenuation signals, the selection of the short time window directly affects the time-frequency resolution of the time spectrum ② The harmonic estimation algorithm involved in the recognition of non-periodic and sparse fault signals under intermittent working conditions is not good, and the feature extraction is difficult

Images