基于深度学习与立体视觉的四大管道位移监测系统

Displacement Monitoring System of Four Pipelines Based on Deep Learning and Stereo Vision

摘要: 火力发电厂的“四大管道”长期处于高温高压的运行状态下，在机组的常年运行过程中，由于调峰等电力任务的存在，“四大管道”的管道会发生蠕变从而产生位移，因此需要对其进行监测进而保障机组运行安全。为了提高测量精度和降低测量用时，从而达到高精度三维管道位移实时监测的目的，提出了一种基于立体视觉与深度学习相结合的非接触式四大管道位移在线测量系统。该系统使用两台相机对固定于四大管道上的棋盘格标靶进行拍摄，捕获图像利用深度学习神经网络进行处理，无需单独对相机进行畸变矫正与双目矫正，实验室标定之后即可用于现场实际环境的测量。在实现了四大管道三维位移的高精度监测的同时，使得测量操作更加简洁、系统智能化程度进一步提高。实验结果表明：该系统测量的标准差小于0.24 mm，位移测量误差小于0.3%，单点测量时间小于0.1 s，是一套适用于四大管道高精度三维位移实时监测的测量系统。

Abstract: The “four pipelines” of thermal power plant have been operating under high temperature and high pressure for a long time. During the perennial operation of the unit, due to the existence of peak shaving and other power tasks, the pipelines of the “four pipelines” will creep and produce displacement. Hence, the monitor of pipelines is needed to ensure the safe operation of the unit. A non-contact online measurement system is advanced in this paper to improve the measurement accuracy and reduce the measurement time, so as to achieve the purpose of high-precision three-dimensional pipeline displacement real-time monitoring. The system used two cameras to capture the chessboard grid target fixed on the four pipelines, and the captured image is processed by deep learning neural network, so that the system could be used in the measurement of actual environment without the process of distortion and binocular correcting of cameras after calibration in the laboratory. While realizing the high-precision monitoring of the three-dimensional displacement of the four pipelines, the measurement operation was more concise and the degree of system intelligence is further improved. The experimental results showed that the standard deviation of the system is less than 0.24 mm, the displacement measurement error is less than 0.3%, and the single point measurement time is less than 0.1 s. It was a set of measurement system suitable for high-precision three-dimensional displacement real-time monitoring of four pipelines.