An Improved Fault Diagnosis Framework Based on Deep Belief Networks

Abstract

Real-time and accurate fault diagnosis can provide early warning of system failure and support decision-making of maintenance and replacement processes, enhancing reliability of the dynamic system and reducing costs for maintenance. Deep belief networks, as one of the deep learning methods, can extract features from monitoring data and establish nonlinear relationship between extracted features and comprehensive system conditions. It has potentials for fault diagnosis. In this paper, a complete fault diagnosis framework starting from FFT(Fast Fourier Transform) to health condition prediction is proposed. Bearing vibration data is employed to verify the proposed approach. The results show that the proposed model has high and stable prediction accuracy. These results demonstrate the effectiveness, stability, and robustness of the fault diagnosis framework based on deep belief networks.