Adversarial Diagnosis Method for Bearings Across Working Conditions

Abstract

The invention discloses a confrontation diagnosis method for bearings across working conditions. In the method, the vibration data of the bearing in the running state is collected, the vibration data is divided to generate signal samples, and the samples are divided into test samples for testing. Set and the training set of training; Construction training module, it comprises the feature extractor that extracts signal feature, the classifier to the classifier of bearing fault classification and the working condition of distinguishing feature and the discriminator of fault; Said training module is based on training set training, wherein , using the BP method to update the feature extraction parameters and classifier parameters according to the loss function; fixing the feature extractor parameters, using the loss function to update the discriminator parameters; fixing the discriminator parameters, using the adversarial loss function to update the feature extractor parameters ; Build a test module based on the updated classifier and feature extractor, and input the test set and / or target domain working condition samples into the test module for fault diagnosis.

Description

technical field [0001] The invention belongs to the technical field of intelligent diagnosis of bearing faults, in particular to a confrontation diagnosis method for bearings across working conditions. Background technique [0002] Rolling bearings are important key parts in rotating machinery, and the safety of the bearings inside is directly related to the normal operation of the machinery. Traditional intelligent diagnosis relies on experts to extract features and complex signal processing. Due to the high reliance on expert knowledge, traditional methods cannot be widely used. [0003] Traditional artificial intelligence fault diagnosis methods based on deep learning rely on samples from new working conditions to participate in training in the field of variable working conditions (the speed and load of the training set are different from the speed or load of the test set), and retraining is required when new working conditions occur in actual operation , wasting a lot o...

AI Technical Summary

Problems solved by technology

[0008] Aiming at the problems existing in the prior art, the present invention proposes a confrontation diagnosis method for bearings across working conditions, which does not require manual feature extraction and reduces dependence on expert knowledge, while overcoming the shortcomings of traditional artificial intelligence semi-supervised variable working conditions training, That is, unlabeled target field data is required for training, and retraining is required once the target field changes

Images