The Application Of Deep Belief Network And Chaos Immune Algorithm In Fault Diagnosis

With increasingly complicated,larged and automated of the electromechanical equipment,the complexity of daily condition monitoring and fault diagnosis of electromechanical equipment is getting higher and higher,and the mapping relationship between fault status and fault feature is more and more complicated.Traditional fault diagnosis method based on the signal processing and artificial intelligence can not meet the needs of practical use.Deep learning can learn the mapping relationship between original fault signals and fault types through the nonlinear transformation of multiple hidden layers,which provides a new idea for fault diagnosis of complex electromechanical equipment.Deep Belief Network(DBN)is taken as an example to study the fault diagnosis method based on deep learning in this paper.Based on the principle of DBN,this paper introduces the network structure and parameter training methods of DBN,and constructs the DBN fault diagnosis model.The fault diagnosis performance of the DBN fault diagnosis model is verified by the bearing fault data set published by the Western Reserve University.The influence of the hyperparameters and weight initialization methods of the DBNfault diagnosis model on its fault diagnosis performance is discussed,which provides theoretical support for model parameter setting.Firstly,the traditional DBN fault diagnosis model is trapped in local search during the learning and training process,which affects the fault diagnosis performance of the DBN fault diagnosis model.The improved chaos immune algorithm is used to optimize the connection power and the bias of the networks,and an optimized deep belief network fault diagnosis model(ICIM-DBN)is proposed in this paper,which has higher training efficiency and fault diagnosis accuracy.The improved chaos immune algorithm adopts chaos initialization and adaptive mutation to improve the diversity of antibody population,and the variable selection operator is introduced to accelerate the optimization speed of the algorithm.Secondly,a optimized DBN multi measure point fault diagnosis method is proposed based on the optimized DBN fault diagnosis model.By using fault signals of different measure point to train the ICIM-DBN model,the preliminary diagnosis results obtained from each measure point are fused with DS evidence theory.This method makes full use of the fault information of each measure point,avoids the fault misjudgment caused by fault information transmission or sensor performance difference,and improves the fault diagnosis accuracy of complex electromechanical equipment.Finally,selecting the asynchronous motor as experimental object to test the performance of the proposed methods.The experimental results showthat the accuracy of the ICIM-DBN model proposed by this paper is 7.19%higher than that of the traditional DBN fault diagnosis model and 3.7% higher than the DBN model optimized by genetic algorithm.The accuracy of the optimized DBN multi measure point fault diagnosis method for asynchronous motors fault diagnosis is as high as 99.45%,which is 1.59% higher than the average fault diagnosis accuracy of the single measuring point.