Motor bearing fault diagnosis method

Abstract

The invention relates to a motor bearing fault diagnosis method, which comprises the following steps of S1, building a generative adversarial network of a small sample data category through a GAN training method based on a discrimination model and a generation model, and generating a data set conforming to small category features; S2, adding the generated data set into an original small class sample training set to form a balance data set; S3, constructing deep convolutional neural networks of the discrimination model and the generation model, wherein the deep convolutional neural network of the discrimination model comprises three convolutional layers and three corresponding pooling layers, two full connection layers being arranged behind the third pooling layer, and taking the optimizedbalance data set as a training set of the deep convolutional neural networks; and S4, for the training set, learning fault features from training data in a self-adaptive layer-by-layer mode, and carrying out diagnosis and recognition of different fault types through a classifier. Compared with the prior art, the method has the advantages of learning the fault features in a self-adaptive mode, andimproving the diagnosis and recognition rate of faults with small data volume and the like.

Description

technical field [0001] The invention relates to the field of fault diagnosis of mechanical equipment, in particular to a fault diagnosis method for motor bearings. Background technique [0002] Traditional fault diagnosis mainly processes data. Through a series of time-domain and frequency-domain analyzes of motor current, vibration, electromagnetic, temperature and other data, the characteristics containing fault information are obtained, and then evaluated, such as empirical mode Decomposition, morphological filtering and wavelet packet transform, etc. With the development of intelligent algorithms such as machine learning and pattern recognition, using these intelligent algorithms to extract features from data, and then through various optimization algorithms, and finally using specific classifiers, the diagnosis and identification of motor faults can be completed. The classic Classification algorithms include support vector machines, neural networks, random forests, ker...

AI Technical Summary

Problems solved by technology

[0003] In the early stage of the motor fault, the influence on the current and vibration signals is small, and the acquisition accuracy of the acquisition equipment and the accuracy of the fault diagnosis method are required, so there is a certain cost problem and algorithm difficulty

Traditional fault diagnosis methods cannot realize the identification of weak signals

Machine learning algorithms rely on signal processing technology and diagnostic experience to extract fault features and can complete fault diagnosis of fixed equipment. However, complex industrial environments have a great impact on equipment operating data, and traditional methods are difficult to accurately complete fault diagnosis.

[0004] Chinese patent CN201811590223.2 discloses a motor fault diagnosis method and system, which uses kernel density to process training data, and uses a supervised learning model to compare fault data with normal data to judge fault categories, but there are insufficient data feature extraction capabilities Defects, the supervised learning model is difficult to apply to actual production and life, and does not take into account the small number of fault categories

[0005] Chinese patent CN201811082332.3 discloses a GAN-based interpolation method for the missing data of wind turbine hub wind measurement, which applies the generated confrontation network in the field of fault diagnosis, and inputs the training samples into the generator model and the countermeasure model for iterative training, and completes the Data interpolation work, but it only uses the generative confrontation network to generate virtual data and interpolate it into the data set. It only considers strengthening the correlation of the original data set, and does not take into account the fault categories with small data volumes. At the same time, there is no diagnostic model designed to match

Images