Mechanical Fault Prediction Based On Full Vector Singular Value Decomposition

The mechanical fault prediction technology can detect the faults that may occur in the mechanical equipment in advance,which is of great significance for ensuring the smooth operation of the equipment.The spectrum of the mechanical equipment vibration signal can reflect its operating state.By predicting the spectrum,it is not only able to judge the severity of the fault,but also the type of fault.The prediction of the spectrum is different from the amplitude prediction at a single characteristic frequency and belongs to multivariate time series prediction.Multivariable time series has the characteristics of large amount of information,noise and variable correlation.The single variable prediction model is not suitable for multivariable prediction.Extreme Learning Machine(ELM)is a single hidden layer feedforward neural network with high learning accuracy and fast operation speed.It is suitable for processing multivariable data.Combining it with the idea of dimensionality reduction of singular value decomposition,a model of ELM based on singular value decomposition is proposed to realize spectrum prediction.The main research contents of this paper are as follows:1.The signal processing method based on the full-vector singular value decomposition is studied.Firstly,based on the principle of singular value decomposition,the singular value decomposition signal processing model is established.The denoising process is analyzed by simulation and experiment.Aiming at the problem that the information contained in the single-channel signal is not comprehensive,the full-vector technique and the singular value decomposition method are combined to establish a full-vector singular value decomposition model for spectrum analysis of the two-channel signal of the rolling bearing.Comparing the obtained full-vector spectrum with the single-channel spectrum,the results show that the full-vector singular value decomposition method can obtain a more complete spectrum.2.The multivariate extreme learning machine prediction model is studied.In order to solve the prediction problem of spectrum structure,based on the extreme learning machine,the prediction method of multivariate extreme learning machine is proposed.Firstly,the sliding window method is used to process the multivariate time series of the spectrum to construct the training and test sample matrix.When the sample matrix is input to the limit learning machine,the sample matrix is vectorized.This model is used to predict the spectrum structure of the inner ring vibration data of rolling bearings.The research shows that the multivariable extreme learning machine can predict the fault characteristics,but the prediction accuracy of the whole spectrum structure is not high and needs to be improved.3.The SVDELM prediction model was studied.Extreme learning converts the multivariate time series in matrix form into vector form during sample input,destroying the structure of the original data.In order to solve this problem,based on the extreme learning machine,the singular value decomposition layer is added between the input layer and the hidden layer to reduce the dimension of the sample,and the SVDELM prediction model is constructed.The principle and prediction process of the method are analyzed,and the spectrum obtained by the full-vector singular value decomposition method is predicted.Studies have shown that the SVDELM method can achieve accurate prediction of the spectral structure.