Wavelet packet frequency domain signal manifold studying failure diagnosis method

Abstract

The invention relates to a wavelet packet frequency domain signal manifold studying failure diagnosis method which comprises the steps of acquiring vibration data on rotating mechanical equipment, performing N-layer wavelet packet decomposition on the acquired vibration data for obtaining 2N component time domain signals; performing Hilbert envelope demodulation on each component time domain signal generated through decomposition, and extracting a modulation signal; performing FIR filtering on a demodulation signal; performing resampling on each component time domain signal in a low sampling frequency for reducing data length; performing autocorrelation calculation on the layered resampled data and performing normalization for forming an autocorrelation coefficient; calculating the power spectrum of the autocorrelation coefficient after normalization of each component, performing threshold processing on power spectrum data by means of a preset threshold, forming high-dimension data vector from the power spectrum data after threshold processing; performing dimension reduction on the high-dimension data vector, combining the component power spectrum data for forming an L*2N-dimensional matrix, finally forming a two-dimensional or three-dimensional manifold, and determining the fault state of the rotating mechanical equipment through a manifold result.

Description

technical field [0001] The invention relates to a fault diagnosis method for mechanical equipment, in particular to a fault diagnosis method for wavelet packet frequency domain signal manifold learning of rotating mechanical equipment. Background technique [0002] Collecting vibration signals from key parts of rotating machinery, extracting and analyzing sensitive features related to faults from vibration signals is the main fault diagnosis method for rotating machinery at present. After denoising and purifying the vibration signal, it is often difficult to effectively judge the fault state of the equipment only by extracting a single feature information. Using a variety of feature extraction methods to obtain multiple features and comprehensively use them can judge the equipment status more accurately, but the increase of feature information leads to a large increase in the dimension of information, which brings difficulties to engineering applications. The structure of h...

AI Technical Summary

Problems solved by technology

The manifold learning algorithms that have been studied more now mainly include principal component analysis (PCA) algorithm, local linear embedding (LLE) algorithm, isometric mapping (Isomap) algorithm, local tangent space arrangement algorithm (LTSA), etc. These manifold learning algorithms There have been many applications in the field of mechanical fault diagnosis, but most of these algorithms in the application of mechanical fault diagnosis have problems such as nonlinear data congestion, low-dimensional manifold expression is not clear enough, etc.

Images