Research On Fusion Type Deep Learning Identification Algorithm Of Subhealth

With the modern industrial productive process gradually developing into the production equipment of large-scale,intelligence,complexity,automation,if one of the parts has something wrong,the whole productive process will be influenced to some extent.But if we can identify the current state of the system accurately and replace the equipment of relevant failures in a timely manner,so that we can effectively prevent the occurrence of failure.Those emergent system failures are generally due to the unpredictable external force caused by the system,where there are generally no signs and which are uncontrollable.However,the failure of the industrial process control system is mostly due to the delay fault caused by the equipment wear or the component aging,which is not obvious when the equipment is working on.Therefore,the research of delay fault and equipment reliability has become an important part in the field of fault diagnosis about industrial equipment,which has aroused great concern of many experts and scholars.In this paper,we mainly study the deep automatic encoder and its improvement,and use the ‘Subhealth' state recognition of rolling bearing as the application scene.After learning a lot of methods about deep learning and troubleshooting,I found that a well-designed learning rate strategy can significantly improve the convergence rate of the deep learning model.Therefore,this paper proposes an adaptive learning rate of deep learning model to improve the convergence speed of the deep network model for the deep learning model.Meanwhile,this paper improves the cost function of the original depth model and increase the generalization ability and robustness of the network model adopting the advantages of sparse automatic encoders and edge noise reduction encoders.In this paper,the cascade encoder is used as the network structure of deep learning.Through this structure,the noise in the mechanical vibration signal can be filtered to facilitate the extraction of favorable features.The experimental results show that the convergence rate of the model is accelerated in the case of the basic guarantee accuracy.This paper uses the relevance vector machine to replace the depth of the Soft Max layer used in traditional deep learning.The kernel function selection of the relevance vector machine and the selection of the kernel parameters are especially important for the final classification.In this paper,kernel parameter selection method for the optimal mapping of the relevant vector machine is proposed according to the fisher criterion and the maximum entropy criterion.The experimental results show that the selected kernel parameters can improve the recognition accuracy of the model.In order to further improve the accuracy of recognition,this paper will be normalized by the relevance vector machine as the first evidence space of D-S evidence theory.The results obtained by the original Soft Max classification are normalized as the second evidence space,then the two evidence spaces are merged according to the fusion rule of D-S evidence theory to get the final recognition result.The experimental results show that the method is effective to improve the accuracy of the ‘Subhealth' state of rolling bearing.