Research on Fault Diagnosis Method Based on RSAPSO-DBN

Abstract

In view of the fact that the existing traditional methods of mechanical equipment have a large dependence on the data signal processing method, this paper uses the Deep Belief Network (DBN) based fault diagnosis method. The DBN is made up of a number of restricted Restricted Boltzmann Machines (RBM). The last layer uses a back propagation network (BP network) to fine tune the network. DBN directly uses the original data as input to reduce the influence of human factors in feature extraction, but the excessive interference factors in the original data make the diagnosis result difficult to reach the ideal result. Therefore, in order to further improve the diagnostic accuracy of DBN, this paper proposes a random self-adapting particle swarm optimization algorithm (RSAPSO) to optimize the BP classifier of the last layer of DBN. Through simulation experiments, it is found that the use of particle swarm optimization DBN effectively improves the accuracy of fault diagnosis than the standard DBN.