Five-phase asynchronous motor rolling bearing fault diagnosis method based on single-channel graph data enhancement and migration training residual network

Abstract

The invention discloses a five-phase asynchronous motor rolling bearing fault diagnosis method based on a single-channel graph data enhancement and migration training residual network, is suitable for unbalanced small sample conditions, and belongs to the technical field of diagnosis. The method comprises the following steps: dividing samples into a training set and a test set, and coding signals of the training set by using a GASF to obtain a single-channel graph; taking the single-channel graph as a sample to train a WGAN-GP network to obtain a generator network corresponding to the sample; afterwards, according to a migration training principle, constructing a residual classification network by using an ImageNet pre-trained residual network convolution layer as a feature extraction layer; performing training: training a full connection layer by using a sample generated by the generator network; training the full connection layer and the feature extraction layer in sequence by using actual samples as a training set; and carrying out fine tuning on the whole residual network. Diagnosis can be completed by storing and calling a trained model, and the accuracy rate is not lower than 99.4%.

Description

technical field [0001] The invention relates to a method capable of diagnosing rolling bearing faults of five-phase asynchronous motors under the condition of unbalanced small samples, which belongs to the technical field of diagnosis. Background technique [0002] Because of its high reliability, fault-tolerant operation and many other advantages, five-phase asynchronous motors have been used in special fields such as ships and submarines, and rolling bearings are used in five-phase asynchronous motors with overwhelming advantages. Rolling bearings are composed of outer raceway, inner raceway, cage and rolling elements rotating between them. Under normal working conditions, fatigue failure starts from tiny cracks, which gradually expand, and then cause material fragments to fall off, leading to failure. Therefore, the fault diagnosis of rolling bearings in five-phase asynchronous motors is of great significance. [0003] For now, the frequency spectrum analysis of vibrati...

AI Technical Summary

Problems solved by technology

Therefore, for the fault diagnosis of rolling bearings of five-phase asynchronous motors, the above-mentioned commonly used machine learning algorithms are not applicable, or cannot achieve ideal results

[0006] More importantly, since high-quality simulation data cannot be generated for bearing faults, and bearing faults in engineering practice are relatively low-probability events compared to normal states, the opportunity to collect actual bearing fault signals is not sufficient, so the obtained bearing normal and Failure samples are usually unbalanced, and the total number of failure samples is too small

However, the commonly used machine learning algorithms cannot handle the situation of unbalanced small samples well, so naturally they cannot be ideally used for fault diagnosis of rolling bearings of five-phase asynchronous motors.

Images