Dynamic Vision-Enabled Machine Condition Monitoring: A Point Cloud-Based Diagnostic Methodology

Event-based cameras are an emerging type of Dynamic Vision Sensor. Unlike traditional standard cameras, they record the occurrence of brightness changes, which enables them to achieve low latency, high dynamic range, and high temporal resolution. To explore the application of contactless event data in machine fault diagnosis, this article draws inspiration from the similarity in data format between event data and 3D point cloud data. While point cloud data is widely employed in computer vision tasks, particularly in intelligent systems like autopilots and robots, its usage in industrial settings is scarce. This article presents one of the pioneering attempts in the current literature on using a 3D point cloud data algorithm to process event data for machine fault diagnosis. A geometric data structure is proposed to represent event information and a deep learning method on point cloud data is employed for classification to perform diagnosis tasks. The article conducts experiments on diagnosing rotating machine faults using event-based cameras. It has been extensively validated that utilizing point sets derived from dynamic vision data captured by event-based cameras can achieve high accuracy in fault diagnosis. Given their economic viability, installation flexibility, and data efficiency, event-based cameras emerge as a promising solution for non-contact monitoring of machine health conditions and fault diagnosis.