Rolling Bearing Fault Diagnosis Based On Phase Space Reconstruction Phase Diagram And Deep Belief Network

Among various types of mechanical equipment,rolling bearings are the most widely used with a very important position.Considering the research background of rolling bearing fault diagnosis and the current research status at home and abroad,based on the chaotic characteristics of the rolling bearing signal,this paper proposes a rolling bearing fault diagnosis method based on the phase space reconstruction phase diagram deep belief network.The fault features are extracted by reconstructing the phase diagram in the phase space,and combined with the deep belief network(DBN)classifier to diagnose the rolling bearing fault.First,the chaotic characteristics and phase space reconstruction theory are analyzed.According to the theory of chaotic attractor and phase space reconstruction,the selection method of relevant parameters during signal phase space reconstruction is discussed.The complex autocorrelation method,mutual information method,false nearest neighbor method,CAO method,saturation correlation.After analyzing these methods,the simpler and more efficient algorithm C-CMethod is used to select the parameters delay time and the embedding dimension 8)during phase space reconstruction.Next,on the basis of several different chaos statistics methods and chaos characteristics analysis,the method of calculating the maximum Lyapunov exponent proves that the rolling bearing signal has chaotic characteristics.Furthermore,the feature extraction method based on reconstructed phase diagrams in phase space was proposed.The reconstruction of the chaotic dynamic system Lorenz,Duffing and Rossler equations in the XY plane confirms the effectiveness of the feature extraction method.By comparing different damage locations,different damage levels,and the same damage level,three different bearing fault signal phase diagrams are used to detect the similarity of phase diagrams using perceptual hash algorithm.Experimental results proved the feasibility of this feature extraction method.Then,the fault diagnosis method based on deep belief network(DBN)was researched.Restricted Boltzmann machine(RBM)quickly trains RBM by contrasting divergence algorithm to fit the distribution of training samples.DBN trains the network through semi-supervised learning,trains RBM from bottom to top in an unsupervised pre-training mode,obtains high-order data features from low-order data features,and then supervises and finetunes each layer of RBM parameters from top to bottom.It avoided the shortcomings of the network falling into local optimality.And the impact of the DBN model parameter settings was discussed.Through experiments on bearing small sample failure data,the network layer,the influence of the error rate and training time by the parameters such as hidden layer neurons,learning rate and iteration number.The appropriate parameters were chosen to improve the diagnostic effect of the model.Finally,taking the rolling bearing fault data of Case Western Reserve University in the United States as the research object,combining the bearing fault features extracted by phase space reconstruction phase diagram and DBN,the DBN-based rolling bearing model diagnosis method is constructed.Diagnosis and analysis are carried out from different degrees of damage of rolling bearings,different fault parts and ten kinds of faults.The experimental results showed that the feature extraction method based on phase space reconstruction is effective,and combined with DBN for fault diagnosis.DBN classification model in a multi-task achieved better diagnosis using the comprehensive evaluation model by evaluating results of precision,recall and F values,etc.Applying the method in this paper to the signal collected by Shanghai Baosteel has achieved good results.