New way of detecting vibration of mechanical systems by explainable deep learning

초록

Nowadays, AI technology is rapidly developing and can be applied to various tasks. Even in the mechanical engineering field, many tasks are automated, become very convenient, and its performance is often improved thanks to AI. Therefore, many industries also try to apply this kind of advanced AI technology for machine health monitoring tasks to reduce costs. In this research, we suggest a novel way to detect abnormal vibration and failure by deep learning algorithms. In the past, most industries diagnose abnormal vibration of equipment based on an expert's experience or vibration-related sensor readings. Unlike what industry typically does with a vibration sensor in the past, we propose a new way of detecting vibration and its regime by image-based pattern recognition deep learning algorithm and interpretable AI technology. We can establish a cost-efficient system, so reduce the cost and resources by using the camera only. We demonstrate the proposed algorithm with vibrating cantilever and pumps. © Proceedings of 2020 International Congress on Noise Control Engineering, INTER-NOISE 2020. All rights reserved.