Based On Deep Learning Approach For Rolling Bearing Fault Diagnosis

The main function of the rolling bearing is to support the rotating parts and reduce the friction between the shaft and the shaft seat.Rolling bearings are one of the irreplaceable parts in mechanical equipment.The health status of rolling bearings is related to the normal operation of the production equipment.But in the actual working condition,various failures and damages of the bearings are inevitable,which will affect the production of products.Therefore,quickly and accurately diagnosing the fault category has a great practical effect.However,due to the precision and coupling of the equipment,it is difficult to acquire the signal of the failed bearing.At the same time,affected by environmental changes,the bearing signals of the same fault type may be quite different,so that the subsequent diagnosis effect is not obvious.In view of the above problems,this work using the model based on deep learning to study the bearing fault diagnosis.The proposed method can solve the above problems in a targeted manner,and has a good diagnosis effect.The specific work is as follows:(1)A learning model based on few-shot learning is proposed for fault diagnosis in the case of insufficient samples.This method uses the down-sampled original vibration signal as input,and judges the distance between two samples by comparing the distance between the two samples in the feature space,that is,whether the input sample pair belongs to the same category.The proposed Siamese network sub-network adopts the combination of convolutional neural network and long-term memory network,which can fully extract the local and global features of vibration signals for fault diagnosis.Experiments show that this method still has high accuracy and noise resistance even when samples are not sufficient.When a new category appears,the diagnostic accuracy increases from 87.09% to 91.35% with only 90 training samples.(2)In order to reduce the influence of the variation of the working conditions of the motor on the inconsistent distribution of the collected data,this paper proposes an unsupervised domain adaptive network based on adversarial for bearing fault diagnosis.This method is based on the adversarial training idea of Generative Adversarial Networks to learn the same distribution feature representation between different domains by adding a gradient flip layer in front of the classifier to minimize the Wasserstein distance between the source and target domain distributions,while using the joint maximum distribution difference For aligning empirical joint distributions across different domains.Diagnosis and analysis of unlabeled target domain data sets are realized through labeled source domain data sets,so as to complete bearing fault diagnosis under variable working conditions.Experiments show that the proposed method achieves high results in diagnosing faults under variable working conditions,with an average accuracy of 97.22%.To sum up,the bearing fault diagnosis method based on deep learning proposed in this paper can better solve the problem of insufficient samples and variable load conditions.By directly modeling and analyzing the original vibration signal,the end-to-end bearing fault diagnosis is effectively realized,and the disadvantage of requiring manual intervention in the early stage is solved.Compared with the existing methods,the proposed method improves the high precision and adaptability of bearing fault diagnosis to a certain extent.