This study aims to address the challenge of precise vibration signal monitoring in thermal power unit rotor shafts by proposing a non-contact measurement method based on high-speed stereo vision technology.
A laser point-projection vision system integrated with a 10,000-fps high-speed camera was deployed for vibration signal extraction. An image enhancement model incorporating hybrid attention mechanisms into the UNet architecture was developed to improve feature recovery in complex environments. In addition, a sub-pixel keypoint fitting algorithm was proposed to enhance measurement accuracy. Field experiments were conducted to validate the system.
Experimental results demonstrate that the system stably acquires multi-point non-contact vibration signals from rotating shafts. Compared with existing methods, this approach exhibits superior performance in detection precision.
The originality and value of this research lie in the pioneering application of high-speed stereo vision technology to rotor shaft vibration monitoring, with the proposed model and algorithms substantially enhancing robustness and achieving sub-micron-level measurement precision. This approach provides a high-precision solution for rotating machinery condition monitoring, demonstrating significant engineering application value.
