Rolling Bearing Fault Diagnosis Method based on EEMD and GBDBN

Abstract

Aiming at the complexity, nonlinearity, and non-stationarity of the rolling bearing vibration signal, a fault diagnosis method based on Ensemble Empirical Mode Decomposition (EEMD) and Gauss Bernoulli Deep Belief Network (GBDBN) model is proposed. The method first carries out EEMD on the vibration signal; second, the eigenvalues of each intrinsic mode function (IMF) are statistically analyzed; then, the feature vectors are constructed by selecting less change features; finally, the normalized feature vectors are input into the GBDBN to identify the fault types. The experimental results show that the proposed method achieves more than 90% recognition rate of fault types and has better fault diagnosis ability, which can provide convenience for maintenance.

Research on EMD improvement methods

“…At the same time, the white noise is basically canceled after multiple averaging, but there is residual, and the reconstructed noise cannot be ignored. [5] Since EEMD [6] adds Gaussian white noise when improving EMD drawbacks, this method is to get rid of the Gaussian white noise first, and then perform EEMD processing to get the individual IMF components, and then screen the appropriate IMF components for signal reconstruction (a technical means to recover the whole signal with a part of the known signal). The impact of processing Gaussian white noise on signal processing is achieved through three main steps [7] .…”

Icing Prediction of Fan Blade based on a Hybrid Model

“…Self-cluster prediction based on multi-source eigenvalues predicted by the Elman neural network can be understood as a nonlinear system, and the BP neural network is a kind of nonlinear optimization. It adjusts the weights and thresholds of the network according to the error back propagation so that the sum of the system errors is the smallest [12][13]. Kolmogorov's theorem states that the three layers of the BP neural network are sufficient to approximate complex nonlinear systems with arbitrary precision [14][15][16].…”

“…Refer to the empirical formula = √ + 0 + , where is the number of neurons in the input layer, 0 is the number of neurons in the output layer, n is the number of neurons in the hidden layer, and the range is from [1,10]. The number of nodes in the hidden layer calculated by the above formula is [1,13]. The mean square errors (MSE) of nine different hidden layer neural nodes are compared and analyzed.…”

Comparison of the effects of multiple empirical mode decomposition on vibration signals

“…To solve the problem of mode aliasing, the proposed method is to add random white noise before signal decomposition, which is a derivative method of EMD called Ensemble Empirical Mode Decomposition (EEMD) [4][5]. But individual differences of added white noise, in the next step to carry on the empirical mode decomposition, can create new aliasing, at the same time, there may be cannot eliminate accumulated in the original signal, signal influence signal analysis and feature extraction, finally, because want to set the average of the component composition, caused the increase of operation time.…”