Noncontact Vibration Measurement and Bearing Fault Feature Extraction Method Based on Machine Vision

With the development of intelligent and interconnected industrial equipment, the demand for equipment diagnosis and prediction promotes the optimization of smart factory maintenance strategy. As the core transmission component of rotating machinery, the bearing directly affects the production efficiency and operation safety, so that the efficient performance evaluation of the bearing is very important for the predictive maintenance and intelligent management of the equipment. The traditional signal acquisition method is affected by the challenging industrial field condition. In recent years, non-contact methods have been gradually applied to industrial measurement. However, causes such as difficulties in calibration and image recognition, weak vibration, etc., visual condition monitoring of complex systems such as bearing faults has not been widely and effectively explored in existing research. Therefore, this paper incorporates visual displacement tracking into the field of bearing fault detection, and uses Lucas-Kanade optical flow to extract sub-pixel vibration from equipment operation video, the optimal filter size is searched by Fibonacci, and the pulse of each fault period is enhanced by Multipoint Maximum Kurtosis-Harmonic to noise ratio Deconvolution, which directly expresses the accurate fault characteristic frequency through the spectrum. The signal extraction and analysis of bearings at different fault positions under various rotational speed are carried out respectively, and the contact acceleration sensor is used to collect the vibration signal synchronously for comparison and verification. Furthermore, the experiment is carried out by using the main light source with decreasing power, the robustness of the vibration extraction method under different illumination environments is verified.