Research on intelligent fault diagnosis of mechanical equipment based on sparse deep neural networks

Abstract

In the big data background, the accuracy of fault diagnosis and recognition has been difficult to be improved. The deep neural network was used to recognize the diagnosis rate of the bearing with four kinds of conditions and compared with traditional BP neural network, genetic neural network and particle swarm neural network. Results showed that the diagnosis accuracy and convergence rate of the deep neural network were obviously higher than those of other models. Fault diagnosis rates with different sample sizes and training sample proportions were then studied to compare with the latest reported methods. Results showed that fault diagnosis had a good stability using deep neural networks. Vibration accelerations of the bearing with different fault diameters and excitation loads were extracted. The deep neural network was used to recognize these faults. Diagnosis accuracy was very high. In particular, the fault diagnosis rate was 98 % when signal features of vibration accelerations were very obvious, which indicated that using deep neural network was effective in diagnosing and recognizing different types of faults. Finally, the deep neural network was used to conduct fault diagnosis for the gearbox of wind turbines and compared with the other models to present that it would work well in the industrial environment.