Visual vibration measurement of rotating bodies with effective time–frequency characterization at constant and variable rotational speeds

Abstract

Non-contact visual vibration measurement methods are gradually applied to vibration signal analysis of rotating bodies. However, the displacement fitting accuracy of existing visual methods needs to be improved in constant speed vibration measurement, and they are rarely used in variable speed vibration measurement. Aiming at the needs of rotating machinery condition monitoring, the paper proposes a non-contact vibration measurement method integrating deep learning technology. The method uses a high-speed industrial camera to capture rotor vibration images, and obtains vibration displacement under constant speed and variable speed conditions through instance segmentation network processing. By constructing a new instance segmentation network architecture, the target segmentation accuracy and vibration measurement accuracy are improved, and Feature Enhancement Module(FEM) and improved Protonet are introduced to further improve the measurement accuracy. Combining vibration displacement data with spectrum analysis enriches the vibration monitoring methods of rotating machinery. Experiments show that the method performs better than target detection and segmentation algorithms in constant speed and variable speed rotor vibration measurement, and has application potential in rotating machinery condition monitoring.