Research On Cross-device Fault Diagnosis Method Based On Deep Transfer Learning

The data-driven equipment fault diagnosis method diagnoses or predicts the operation status of equipment through analyzing a large amount of historical data.With the development of Internet of Things,data-driven equipment fault diagnosis methods are widely used.However,data-driven equipment fault diagnosis relies on a large amount of labeled data,and it requires large labor cost,time cost and equipment wear and tear to collect a large amount of data and build a diagnosis model in industrial scenarios.In addition,the fault diagnosis model trained by utilizing data under limited conditions has poor generalization.To mitigate the above problem,this thesis proposes corresponding improvements from two aspects of example-based deep transfer learning and model-based deep transfer learning via combining the advantages of deep learning and transfer learning,respectively,and the main contents of this thesis are as follows:1.A cross-equipment gearbox fault diagnosis algorithm based on adversarial deep transfer learning is proposed to address the problems of scarcity of effective samples and weak feature extraction ability affected by external environment in actual production equipment fault diagnosis.First,the source and target domains are domain adapted based on the deep adversarial algorithm,and the multi-granularity domain adaptation of the full domain and fault category is considered simultaneously to improve the fault diagnosis performance in the target domain.Secondly,the fault classifier of gearbox is trained using multiple source domains to obtain a diagnostic model,subsequently,fine-tuned using the target domain data based on the model transfer method.Finally,based on the cross-equipment fault diagnosis algorithm proposed in this thesis,it is validated in a case of deep transfer fault diagnosis from a publicly available dataset to a coal mining machine gearbox dataset,and the results show that the method has great accuracy and generalization.2.A transfer convolutional neural network fault diagnosis algorithm combining subspace migration techniques is proposed for the cross-domain transfer problem in real production equipment fault diagnosis.The algorithm is divided into two main steps.In the first step,the model is trained on the basis of a large amount of source domain data,and the training efficiency of the diagnostic model is improved by adding Standout layer to reduce the use of training parameters and improve the classification effect of the diagnostic model.In the second step,the diagnostic model is transferred to the target domain task by the model migration scheme,and the optimal number of fixed layers and the tuning of non-fixed layer parameters by the tuning layer are experimentally analyzed to improve the efficiency of transferring the diagnostic model of large source domain data to the target domain task,facilitating the fast construction of the target diagnostic model,and improving the training efficiency as well as the diagnostic performance of the target domain diagnostic model.Finally,it is experimentally verified that the method in this thesis has remarkably accuracy and adaptability.Figure [22] Table [7] Reference [111]...