Abnormal Vibration Detection by $2\mathrm{D}$ Digital Image Correlation Analysis

In industrial IoT, sensors are embedded in machinery to monitor critical factors such as bearing wear, motor degradation, and environmental conditions. Traditional sensor-based IoT setups are costly and invasive. With the advancements in computer vision, image-based defect detection has made significant progress across various domains. This study proposes to integrate deep learning, via a high-speed camera, with DIC (digital image correlation) techniques for non-intrusive, full-field anomaly detection of a vibrating object, such as industrial motors. Through image processing, the surface of the vibrating object is partitioned into sub-regions, each undergone FFT analysis. By combining the results from all sub-regions, the normal/abnormal state of the vibrating object is determined. Experimental results demonstrate the feasibility of this approach. We also contribute a dataset with detailed annotations collected from our experimental environment for future work.