Rolling bearing fault diagnosis method based on improved variational mode decomposition and extreme learning machine

Abstract

The invention discloses a rolling bearing fault diagnosis method based on improved variational mode decomposition and extreme learning machine, which is characterized in that: the vibration signals of rolling bearings under different types of faults are collected, and the maximum correlation kurtosis deconvolution is used to filter the vibration signals, Using the particle swarm algorithm to optimize the parameters of the maximum correlation kurtosis deconvolution method, the envelope energy entropy after signal deconvolution is proposed as the fitness function; the energy threshold is proposed to improve the number of modes in the variational mode decomposition , realize the improved variational mode decomposition of the filtered vibration signal, and obtain the modal matrix of the corresponding vibration signal; perform singular value decomposition on the modal matrix, obtain a singular value vector and construct a rolling bearing fault feature set; use extreme learning The computer trains the fault feature set to establish a rolling bearing fault diagnosis model. The invention realizes the stable feature extraction of the complex vibration signal of the rolling bearing, thereby improving the diagnostic accuracy.

Description

Technical field [0001] The invention belongs to the field of mechanical fault diagnosis, and specifically relates to a rolling bearing fault diagnosis method based on an improved variational modal decomposition and an extreme learning machine. Background technique [0002] Rolling bearings are one of the indispensable general mechanical parts in rotating machinery, and are widely used in metallurgy, aerospace, chemical and other fields. Due to the long-term repeated action of the working surface of the rolling bearing and the contact stress and the long-term exposure to high speed, high load, high temperature and other environments, the rolling bearing is prone to fatigue, cracks and other failures, resulting in abnormal vibration of the rolling bearing and affecting its associated equipment Even the performance (accuracy, reliability, etc.) of the entire machine, in actual production, the possibility of equipment failure caused by rolling bearing failure is the greatest. Theref...

AI Technical Summary

Problems solved by technology

[0004] Wavelet transform has unique advantages in dealing with nonlinear signals. It can perform local refinement analysis on the time domain and frequency domain of the signal. However, the selected wavelet bases are different, and the decomposition results are different. At present, there is still no good theory for the selection of wavelet bases.

EMD can adaptively decompose the signal into multiple intrinsic mode functions, but due to recursive decomposition, it is easy to make the decomposition error pass down layer by layer, causing problems such as mode aliasing, over-envelope, and endpoint effects. Based on the development of EMD EEMD still doesn't solve the problem very well

HHT is through the EMD decomposition of the signal, which gets rid of the shackles of linearity and stability, but lacks rigorous physical meaning, requires complex recursion, and takes a long time to calculate, making HHT more effective only under the condition of relatively simple combination

The variational mode decomposition algorithm overcomes the modal aliasing phenomenon of EMD, but there is no theoretical basis for the setting of the mode number parameter, which has a great influence on the signal decomposition

BP neural network is a more commonly used state recognition method at present. It has the advantages of strong generalization ability and fault tolerance, nonlinear mapping, etc., but there are also problems that the network structure is difficult to determine and the convergence speed is slow.

Compared with BP neural network, support vector machine has faster convergence speed, but it has some problems such as difficult to determine parameters and long training time.

[0005] In the existing technology, the fault diagnosis of rolling bearings has problems such as inaccurate feature extraction, long time for establishing a diagnostic model, and low diagnostic accuracy.

Images