An Adaptive Gaussian-Bernoulli Deep Belief Extreme Learning Machine Network Rolling Bearing Fault Diagnosis Method

Rolling bearing is one of the important parts of rotating machinery,and its fault diagnosis is of great significance for the safe operation of mechanical equipment.Due to the complicated working environment of the rolling bearing,the collected vibration signal contains a lot of noise.How to extract the effective fault information in the noise environment and realize the more accurate fault identification is the focus and difficulty of the current research.Aiming at the non-stationary and nonlinear characteristics of the rolling bearing vibration signal,this paper proposes a rolling bearing fault diagnosis method based on adaptive Gaussian-Bernoulli deep belief extreme learning machine network.Firstly,for the problem that the original signal data dimension leads to the complexity of the layers of the deep belief network and affects the efficiency of fault diagnosis,a set of empirical mode decomposition-singular value decomposition(EEMD-SVD)rolling bearing feature extraction method is proposed.The method reduces the interference of the noise signal on the effective information,extracts the signal singular value vector as the sample data,and reduces the input data dimension.Secondly,for the problem that the deep belief network(DBN)visible layer input is limited by binary value,an adaptive Gaussian-Bernoulli restricted Boltzmann machine(AGBRBM)method is proposed to pre-train the sample and solve the pre-training process.The learning rate parameter value is difficult to achieve optimal setting and the DBN input data is limited.In the pre-training process,the learning rate can be adjusted according to the reconstruction error,the network convergence speed is accelerated,and the network training accuracy rate is improved.Then,in the process of optimizing the network parameters for DBN fine tuning,the back propagation(BP)algorithm has a long iteration time,which affects the diagnostic efficiency and is easy to fall into the local optimal value in the iterative process.The adaptive Gaussian-Bernoulli restricted Boltzmann machine-extreme learning machine(AGBRBM-ELM)optimization network parameter method is proposed.The weight and offset of AGBRBM training are used as the initialization input parameters of ELM,and the output parameter matrix is calculated by ELM algorithm to realize network parameter optimization,which improves the speed and accuracy of model training,and solves the problem of network instability caused by ELM randomly generating input parameters.Finally,the singular value vector extracted by EEMD-SVD feature is taken as a network construction sample.Combined with AGBRBM pre-training and AGBRBM-ELM network parameter optimization method,an adaptive Gaussian-Bernoulli deep belief extreme learning machine network(AGBDBEN)is constructed.Complete fault diagnosis method for rolling bearings.AGBDBEN and DBN,ELM,BPNN,SVM and other models are compared.The experimental results demonstrate the effectiveness of the proposed method.