Research On Intelligent Fault Diagnosis Method Of Equipment Based On GAN And Transfer Learning

Equipment is the cornerstone of modern industry,and it is important to monitor the operation status and diagnose fault of equipment to ensure production safety and reduce costs and increase efficiency for enterprises.In recent years,with the development of computer information technology,equipment fault diagnosis has entered a new stage,and some researchers call it the "big-data-era" of equipment,in which many data-driven intelligent fault diagnosis methods have been born.However,in the real application scenario,big data for equipment has the problem of being big but not complete:condition monitoring data are mostly normal state data,and fault data are few and incomplete.This problem makes the existing intelligent diagnosis methods weak in generalization and unable to establish a complete diagnosis model,i.e.,there are unknown faults.To solve this problem,a framework of intelligent fault diagnosis method for equipment based on GAN and transfer learning is proposed in this paper,which combines the advantages of domain knowledge and artificial intelligence technology to build a personalized diagnosis model for each equipment,which fundamentally avoids the problem of weak generalization ability.At the same time,the diagnosis model has self-learning capability when the unknown faults appear.To realize the method framework,this paper consists of VSPR method and PCCNN based intelligent fault diagnosis method.The details of research contents are described as follows.(1)Research on virtual sample generation and its pattern recognition method.The D3IF principle of virtual sample generation is proposed,the virtual sample generation model is established by domain knowledge under the guidance of D3IF principle,and the general application process of VSPR method is proposed.(2)Application of VSPR method in equipment fault diagnosis.The virtual fault sample generation method based on transfer learning and the data enhancement method based on GAN are proposed for the problem of more normal data and less fault data in real scenarios.The virtual fault sample generation method is based on the fault mechanism and the normal data of the equipment itself,which solves the problem of no-fault data at the early stage of fault diagnosis model establishment on the one hand,and on the other hand,the generated virtual samples are all for specific equipment,so they have good generalization performance.(3)Application of VSPR method in pointer meter image recognition.Aiming at the problem of difficult sample acquisition in CNN-based pointer meter image recognition method,the virtual sample generation method of pointer meter image based on prior knowledge is proposed,which effectively solves the problem of small samples in this application scenario.(4)Research on the intelligent diagnosis method of equipment fault based on PCCNN model.For the problem of incomplete fault diagnosis model,the unknown fault identification method and model self-learning method based on PCCNN are proposed.In the unknown fault identification,the Early-Stopping strategy is adopted to alleviate the High-Confidence problem in the training process of PCCNN method;in the model self-learning process,two guidelines are proposed that the model self-learning should comply with,and the Fine-tune and Class-Weight strategies are adopted to improve the performance of the diagnostic model self-learning process.The study makes up for the shortcomings of the VSPR method in real applications and achieves diagnostic model self-learning.