Industrial Equipment Based On Fractional Order Fourier Transform Research On Micro-fault Diagnosis Method

As industrial electrical equipment operates continuously under long-term high load,the possibility of equipment failure increases significantly.In order to ensure the safe and stable operation of industrial systems,fault diagnosis technology has gradually become an important research direction.However,detecting minor faults remains a challenge,as they not only occur in the early stages but are often mixed with other types of faults.Currently,fault diagnosis methods based on Fourier transform have achieved good results.However,they lack robustness in diagnosing fractional frequency faults and cannot achieve fine diagnosis of multiple types of equipment faults.Based on the strong feature extraction ability of the fractional Fourier transform in detecting small faults,this thesis focuses on industrial equipment fault diagnosis,and combines the fractional Fourier transform(FRFT),neural network algorithms,and federated learning to improve fault diagnosis performance and solve the problem of identifying and diagnosing fractional-order small faults in industrial equipment.The main innovations and research contents are as follows:(1)Proposed a refined fault diagnosis method based on FRFT feature extraction.Firstly,based on the FRFT signal processing method,the feature information of the fractional frequency fault contained in the signal is extracted.Then,the extracted fault features and corresponding prior information of fault categories are input into an SVM classification model to train a more accurate and efficient fault classification model.Finally,the effectiveness of the method is verified on a motor rotor dataset.(2)Proposed a composite fault diagnosis method based on FRFT feature extraction and transfer learning.Firstly,based on a small number of samples annotated with manual knowledge,an expert label system is constructed to label a large number of samples,achieving knowledge transfer from small to large samples.Then,the neural network model is trained using the large samples labeled by the expert system and the small samples annotated with manual knowledge.Finally,a composite fault diagnosis model is established,and the effectiveness of the new method is compared and verified with the method in(1).(3)Proposed a multi-client federated learning fault diagnosis method based on FRFT feature extraction.To address the issues of small and imbalanced client samples,the method first trains local network models based on the constructed samples.Then,it uses a feature function filter-based method to adaptively update the diagnostic model parameters of the client locally and upload them to the federated cloud center.Next,a sequential model parameter fusion method suitable for wireless network transmission with data latency and packet loss is designed at the cloud center,and the fused model parameters are transmitted to each client for sharing.Finally,the effectiveness of the method is validated on a rolling bearing dataset.