Neuromorphic computing-enabled generalized machine fault diagnosis with dynamic vision

Abstract

Rotating machinery plays a critical role in modern industry, and accurate fault diagnosis is essential for ensuring reliable operations. Event-based cameras have recently emerged as a promising tool for non-contact vibration measurement and fault diagnosis of rotating machines. However, the camera factors such as monitoring angles, lighting conditions, etc., have a noticeable influence on the diagnostic performance. A simply trained model cannot be well deployed in new testing scenarios with factor variations. To address this issue, this paper proposes a neuromorphic computing-enabled method for generalizing non-contact fault diagnosis. Dynamic vision data is captured using event-based cameras under varying operational conditions. A dynamic vision data representation method is developed to transform event streams into features that are well-suited for processing by neuromorphic spiking neural networks. Furthermore, a specially designed neuromorphic domain generalization approach is proposed to improve generalization ability across different working conditions. Extensive experiments are conducted to validate the domain generalization performance of the proposed method, along with comparisons with popular domain generalization techniques. The results demonstrate that the proposed approach achieves robust diagnostic performance under different conditions, validating its effectiveness for potential industrial applications.