Research On Fault Prediction Of Mine Ventilator Based On DBN And QPSD

The ventilator is the necessary equipment of the mine ventilation system which plays a dual role of transporting fresh air to the underground and discharging the harmful gas out of the mine.A great threat will be brought to the safety of the person,so as to the economic benefit once a failure occurs.The bearing is the most vulnerable part of the ventilator according to the research,therefore,the fault of ventilator bearing is the main object that is going to be predicted and diagnosed in the thesis.In the part of feature extraction,the Restricted Botlzmann Machine(RBM)and deep confidence network(DBN)are introduced.The DBN is used to extract the features of the original signal.in order to reduce the effect of manual participation on the quality of extracted features.Firstly,the influence of the number of iteration times and the number of hidden layer nodes on the feature extraction ability of single layer stacked RBM features was verified by using the image signal as the research data.Then,the feature extraction capability of DBN is verified by DBN network reconstruction,comparison between DBN network reconstruction and other feature extraction methods,and extracting feature visualization.Finally,bearing signal data is used as the research object to verify its capability from DBN network reconstruction.Experiments show that DBN has a strong feature extraction capability for both image signal and bearing signal.In the part of fault prediction,the RBF neural network algorithm and quantum particle swarm optimization(QPSO)are introduced.Aiming at the defect that the RBF neural network is easy to fall into the local optimum,a QPSO-optimized RBF neural network model is proposed and used to perform the fault prediction which step length is 2 based on the time series.In this paper,three basic functions were used to verify the strong global optimization ability and convergence ability of QPSO,and comparing the predictive ability of qpso-rbf,pso-rbf and RBF,the results show that qpso-rbf model is the most accurate.In the part of fault diagnosis,DBN is used to perform fault diagnosis.the data output from the prediction model is processed and then input into the pre-trained DBN model for fault classification.Compared with other traditional classification models,the results show that the accuracy of the DBN model is high and the classification ability is very strong.