Vision-based Non-contact Vibration Measurement and Fault Diagnosis of Rolling Bearings

Abstract Accurate vibration measurement is crucial for reliable fault diagnosis of rotating machinery. Conventional contact-based sensors are limited by installation constraints and environmental noise, reducing their applicability in complex industrial environments. High-speed video, as a non-contact vibration sensing tool, offers great potential but is challenged by variations in illumination and noise, making it difficult to extract vibration signals effectively. To address these challenges, we propose a novel non-contact visual measurement method. Video frames are processed into multi-scale representations using a Complex Controllable Pyramid (CCP) structure, and inter-frame phase differences are utilized to reconstruct vibration signals. Quadratic Convolutional Networks (QCNN) are applied to extract fault-related features from the visual vibration signals, enhancing fault diagnosis. Experimental validation on rolling bearings and rotors demonstrates the method's robustness under varying illumination and noise conditions. The results show that the proposed approach significantly improves fault diagnosis accuracy and reliability compared to traditional methods. This method offers a promising solution for intelligent, non-contact machinery health monitoring, with practical feasibility for integration into existing diagnostic frameworks.