Research On Data-driven Diagnosis Method Based On Deep Learning For Bearing

As an "industrial joint",bearings are widely used in various rotating machinery and equipment.The normal operation of the bearing is the guarantee for the safe operation of the equipment.Research on practical and reliable bearing fault diagnosis methods is the key to ensure the normal operation of bearings.Traditional bearing fault diagnosis methods rely on signal processing methods and priori expert knowledge,which cannot meet the needs of remote intelligent diagnosis of some bearings in the complex environment of the "big data" era.In view of this,this article mainly carried out the following work:(1)Aiming at the problem of poor anti-noise ability of the traditional bearing diagnosis model,a method of bearing fault diagnosis under the noisy environment that combines denoising convolutional autoencoder(DCAE)and convolutional neural network(CNN)is proposed.The noisy signal is input into DCAE to complete denoising,and then CNN is used to identify and classify bearing faults.Both DCAE and CNN can directly process one-dimensional time-domain signals without any form of signal processing.The experimental results show that DCAE-CNN has achieved better performance than the traditional methods in seven different noisy environments.(2)Aiming at the problem that the traditional bearing diagnosis model has low recognition accuracy under variable operating conditions,a bearing fault diagnosis method under variable operating conditions based on domain adaptation is proposed.By calculating the multi-layer maximum mean difference(MMD)of data samples under two different operating conditions,the domain adaptation is introduced and added to the optimization objective function of model training to achieve adaptive diagnosis of the model under variable operating conditions.The experimental results show that in six different diagnosis tasks,the domain adaptation method has improved the model recognition accuracy.(3)Aiming at the problem of poor generalization ability of traditional bearing diagnosis model in cross-domain diagnosis(from artificial damage to real damage),a cross-domain fault diagnosis method for bearings based on transfer learning is proposed.By pre-training a large number of source domain samples and using the method of "freezing" the parameters of each convolution layer,it is transfered to the target domain model with only a few samples to achieve cross-domain fault diagnosis.The experimental results show that the transfer learning method can not only improve the recognition accuracy of the model in the target domain,but also save training samples and greatly shorten the training time.(4)Aiming at the problem of poor real-time performance of local offline diagnosis and low diagnosis efficiency,a cloud-based remote intelligent diagnosis system for bearings is designed and developed.The system is based on the B/S(browser/server)structure,and the operating status of the bearing can be remotely monitored,historically queried,data downloaded and intelligently diagnosed through the browser.The test results show that the system is stable and reliable,which realizes online collection,remote transmission,cloud storage and intelligent diagnosis of bearing operating data,which makes up for the shortcomings of traditional offline diagnosis and greatly improves the diagnosis efficiency.