Research On Bearing Fault Diagnosis Based On Edge Computing And Deep Learnin

With the rapid development of the Internet of Things and artificial intelligence technology,the massive data generated by mechanical equipment has pushed the fault diagnosis technology into the era of "big data".The analysis,diagnosis and prediction of these data have become the key to ensure the smooth and safe operation of machinery and equipment.Rolling bearings are the core components of modern mechanical equipment.Carrying out bearing fault diagnosis research not only helps to improve the level of fault diagnosis and ensure the healthy operation of mechanical equipment,but also has good research value.With the successful application of deep learning technology in the field of bearing fault diagnosis,this "end-to-end" adaptive fault diagnosis method overcomes the drawbacks of manual feature extraction.However,the direct application of deep learning models often cannot achieve good diagnostic results.In addition,deep learning model training and sample labeling require a lot of time,which is not suitable for scenarios with high real-time requirements for bearing faults.How to modify the deep learning model structure to improve the performance of fault bearing diagnosis,and on this basis,further improve the real-time diagnosis of faulty bearings are valuable research topics.Therefore,this thesis combines deep learning technology and edge computing technology to carry out the following research work:Firstly,aiming at the problem of feature loss when one-dimensional convolutional neural network(1DCNN)processes time series signals,a two-dimensional convolutional neural network(2DCNN)combined with an extra trees regressor(ETR)for bearing fault diagnosis method is proposed.First,reconstruct the collected time-series vibration signals into grayscale images,and then use 2DCNN to adaptively extract bearing fault features and input them into ETR for training and testing.On this basis,random search(RS)is used to find the optimal parameter combinations in ETR,so as to realize the intelligent diagnosis of adaptive bearing faults.The proposed method is applied to the intelligent diagnosis of fault data on the CUT-2 experimental platform.The experimental results show that the proposed signal-image preprocessing method can well preserve the characteristics of the original signal and effectively improve the ability of the model to identify bearing faults.Compared with traditional intelligent fault diagnosis methods,it is verified that the proposed method has excellent diagnostic accuracy,and has good robustness and generalization ability.Secondly,in view of the problem of poor real-time performance in the current fault diagnosis research based on deep learning technology,this thesis proposes a rapid fault diagnosis scheme for rolling bearings based on cloud-edge collaboration.On the basis of a cloud-edge collaborative diagnosis framework adapted to bearing fault diagnosis,an improved depthwise separable convolutional neural network and global average pooling(DSCNN-GAP)lightweight algorithm is used to carry out diagnosis research in this framework.The universal model trained on the cloud is sent to each edge node,and a small number of fault samples collected by each edge node are used to perform personalized finetuning of model parameters,and then carry out diagnostic experiments.In addition,this thesis innovatively cites multi-sensor technology and uses dempster-shafer(DS)evidence theory to perform decision fusion on the diagnosis results at each edge node.It has been verified by experiments that compared with the traditional diagnosis method in the case of a small number of samples,the method of cloud-edge collaboration can greatly improve the diagnosis accuracy and save a lot of training time.The DS evidence theory is used to make decision on the diagnosis results at each edge node,and the fusion result is better.The use of transfer learning to realize cloud-edge collaborative work enhances the adaptability of fault diagnosis algorithms to personalized applications and the real-time performance of fault diagnosis,while ensuring the privacy and security of terminal equipment data.