Research On Compound Fault Diagnosis Method Based On Multi-label Theory

In recent years,the concept of multi-label classification has been widely recognized and applied in various fields such as image recognition and language text classification due to its prevalence and importance in data processing.Although multi-label classification methods have been partially used in compound fault diagnosis research,they are limited by their inability to adapt to the interdependencies between different faults occurring in compound faults.As a result,traditional multi-label classification methods do not exhibit significant advantages compared to conventional compound fault diagnosis methods.To address this issue,this study focuses on the core idea of multi-label classification and conducts in-depth research on several key issues in the field of compound fault diagnosis,including multi-label learning,deep neural network models,fault data processing,and convolutional neural networks.The main contributions of this research are as follows:1)To tackle the challenges of fault recognition difficulties and inadequate fault identification methods in compound fault diagnosis,multi-label learning methods are applied to the field of compound fault diagnosis.Firstly,by utilizing the transformationbased multi-label processing method and adapting traditional single-classification algorithms,a multi-label learning model is constructed.In this study,compound fault data is processed using multi-label techniques,and four different multi-label learning methods are employed for compound fault diagnosis,followed by a comparative analysis.The results demonstrate the effectiveness of multi-label learning in the domain of compound fault diagnosis.2)To address the issue of poor classification performance caused by the interdependencies between different faults in compound fault diagnosis,a compound fault diagnosis algorithm that combines multi-label classification and deep learning theory is proposed.This algorithm utilizes deep learning for compound fault feature extraction,thereby overcoming the decoupling problem among different fault features in multi-label classification.The algorithm is validated using publicly available datasets,and the results show a significant improvement in compound fault diagnosis compared to traditional multi-label classification approaches.3)To tackle the problems of insufficient fault data and poor data quality in compound fault diagnosis,a transfer learning-based multi-label compound fault diagnosis method using two-dimensional image processing is proposed.This method leverages the characteristics of transfer learning by learning from a source domain to a target domain.The source domain model is a pre-trained deep convolutional neural network model,which effectively transfers the image recognition and classification capabilities to the domain of compound fault diagnosis,thereby addressing the issue of insufficient fault data of the same type in the source domain.The proposed method is validated using a compound fault data set from bearings,and the results demonstrate its ability to address the issue of limited data in compound fault diagnosis.