Research On Intelligent Diagnosis Of Rolling Bearing Fault Based On Deep Transfer Learning

In recent years,with the development of mechanical equipment more and more towards the direction of intelligence and information technology,in order to enable mechanical equipment to operate efficiently in different complex environments,it is necessary to monitor its operating status in real time to reduce the economic and safety losses caused by the occurrence of failures.Rolling bearings are the core components of machinery and equipment,and their health status directly determines whether the machinery and equipment can operate normally.Therefore,it is necessary to monitor their health status.For the problem that the distribution of fault data across working conditions and equipment is different,which leads to the unsatisfactory fault diagnosis effect,this paper adopts the method of deep migration learning to study the rolling bearing fault diagnosis method in two cases of cross working conditions and cross equipment.The main research contents of this paper are as follows:(1)A multi-layer composite domain adaptation-based sparse diagnosis method for rolling bearings across operating conditions is proposed to address the problem of sparse representation classification performance degradation due to differences in rolling bearing vibration data distribution under variable operating conditions.The method converts the one-dimensional vibration signal into a two-dimensional time-frequency image as the input of the residual convolutional network,and uses a multilayer composite domain adaptation strategy to reduce the difference between the source and target domain data distributions;then the learned features are used as the input of the label consistent KSVD algorithm,and a discriminant dictionary and a linear classifier are learned through training to achieve the diagnosis and identification of rolling bearing faults,thus improving the performance of sparse representation classification in cross-service bearing fault diagnosis.Two rolling bearing datasets are used in the paper to verify the effectiveness of the proposed method.(2)In response to the problem of greater variability in cross-equipment rolling bearing data distribution leading to degraded model diagnosis performance,an attention-based deep dense capsule network method for cross-equipment fault diagnosis of rolling bearings is proposed.First,the one-dimensional vibration signals are converted into two-dimensional grayscale maps through grayscale transformation and input to the constructed attentionbased deep dense capsule network model to extract fault features;then,the global domain discriminator and local domain discriminator are used to extract domain-invariant features of source and target domain data,and fine-tuning is used to further enhance the learning ability of the model for target domain data features to achieve rolling bearing fault diagnosis.The effectiveness of the proposed method is verified in the paper using three rolling bearing datasets from different experimental equipment.