智能视觉辅助的燃气管道泄漏激光遥测研究

【作者】 杨雪；

【导师】 李金义； 刘金星；

【作者基本信息】 天津工业大学 ， 电子信息（专业学位）， 2023， 硕士

【摘要】 随着燃气管道规模的日益增长,高效智能的泄漏检测对保障石油和天然气行业安全生产愈发重要。本文提出一种视觉机械臂辅助的燃气管道泄漏遥测技术,将激光甲烷遥测传感器与经过手眼标定的深度相机和机械臂相结合,模拟人工手持式泄漏巡检过程,旨在实现燃气管道的智能化泄漏遥测。主要研究内容和成果如下:(1)小型化轻量化的单端式激光甲烷遥测传感器研制。基于可调谐半导体激光吸收光谱(TDLAS)技术,选用1654nm可调谐激光器、Grin透镜光纤准直器、菲涅尔透镜和光电探测器,基于小型化的光机设计和3D打印技术设计了可搭载至机械臂末端的甲烷遥测传感器。实验确定激光器最佳调谐范围为30-70m A,最佳正弦调制电流峰峰值为22.4m A。基于波长调制一次谐波归一化的二次谐波(WMS-2f/1f)检测方法设计了免标定的甲烷浓度反演算法。分析了不同硬地形目标和不同距离下传感器的信噪比、检测精度和灵敏度,结果表明在积分时间为0.1s时传感器的检测限为1.60ppm·m,单程探测距离可达10m。(2)视觉引导机械臂的定位实现。针对泄漏智能检测中管道识别和三维信息获取的问题,选用Intel Real Sense D415深度相机用于实验和数据集采集并完成对相机的参数标定和图像配准。通过添加自行设计的m-att模块、修改用于学习残差特征的C3模块以及引入可变形卷积模块改进深度学习算法Yolov5s,实现管道的实时高精度检测,识别准确率达99.5%。基于入射光垂直于非合作目标表面的切面的原则,选择Dobot机械臂,完成机械臂的MD-H建模、正逆解的验证。同时利用MATLAB的Robotics Toolbox工具包对模型仿真验证。最后,用眼在手外的方式进行手眼标定,完成激光遥测传感器、深度相机与机械臂的协同工作,实现视觉引导机械臂携带激光遥测传感器指向管道目标。(3)智能激光遥测系统的集成及实验验证。对遥测系统的重复定位精度进行评价,有无负载情况下机械臂末端定位偏差均值分别为1.74mm和1.48mm,而经过视觉引导定位后遥测系统的平均定位误差为14.47mm。将100ppm·m的甲烷充入透明气袋模拟燃气管道泄漏的气团进行实验,遥测系统针对3m处泄漏位置的定位误差约为30.00mm。针对泄漏位置进行连续监测,实验结果表明系统可以跟随动态变化的泄漏气体,在石油和天然气工业中具有巨大的应用潜力。更多还原

【Abstract】 Effective and intelligent leakage detection is becoming more crucial as gas pipelines scale expand in order to guarantee safe production in the oil and gas industry.In this dissertation,a visual robotic arm-assisted gas pipeline leakage remote detection technique is proposed,which combines a laser methane standoff sensor with a handeye calibrated depth camera and robotic arm to simulate a manual hand-held leakage inspection process,is aimed to realize the intelligent leakage remote sensing of gas pipelines.The main research elements and results are summarized as follows.Firstly,a miniaturized and lightweight single-ended laser methane standoff sensor is developed.Based on tunable diode laser absorption spectroscopy(TDLAS)technology,a methane sensor is integrated by means of a miniaturized optical machine design and 3D printing technology with a 1654 nm tunable laser,a Grin lens fiber collimator,a Fresnel lens and a detector,which could be mounted on the end of the robotic arm.The tuning range for the laser is experimentally determined to be 30-70 m A,while the ideal peak to peak value of sinusoidal modulation current is determined to be22.4 m A.A calibration-free concentration inversion algorithm for methane is developed on the basis of the wavelength modulation spectroscopy with 1f normalized 2f(WMS-2f/1f)detection method.The signal-to-noise ratio,detection accuracy,and sensitivity of sensor are assessed at various hard terrain targets and distances.The results show that the detection limit of the sensor is 1.60 ppm·m at an integration time of 0.1s and the one-way detection distance of sensor can reach 10 m.Secondly,a vision-guided robotic arm positioning strategy is implemented.In response to the problems of pipeline identification and three-dimensional information acquisition in leakage intelligent detection,the Intel Real Sense D415 depth camera is chosen as the tool for the acquisition of experiments and datasets,and simultaneously parameter calibration and image registration of the camera is completed.Yolov5 s,a deep learning method,is enhanced with the addition of a self-designed m-att module,a modified C3 module for learning residual characteristics and an introduced deformable convolution module.As a consequence,the pipeline is accurately and quickly detected in real time,with a 99.5% identification accuracy.Then,based on the principle that the incident light is perpendicular to the tangent surface of the non-cooperative target,a Modified Denavit-Hartenberg(MD-H)modeling of the robotic arm is finished,along with the verification of the forward and inverse solutions for the chosen robotic arm.The model is also validated and simulated utilizing MATLAB Robotics Toolbox toolkit.Finally,the hand-eye calibration is carried out using the eye-to-hand approach to complete the collaborative work of the standoff methane sensor,the depth camera and together with the robotic arm.As a result,the visually guided robotic arm carrying the standoff sensor is achieved to automatically identify and point at the pipeline target.Thirdly,the intelligent laser remote sensing system is integrated and experimentally verified.The average deviation of the robotic arm end positioning with and without load was 1.74 mm and 1.48 mm,respectively,while the average positioning error of the standoff system is 14.47 mm after the vision-guided positioning.Subsequently,100ppm·m methane standard gas is filled into a transparent gas pocket to simulate the gas mass leaking from the pipeline,the positioning error of system is about 30.00 mm at 3m.And then continuous monitoring of the location of gas leakage demonstrated that the system could track dynamic changes in the leaking gas,denoting great potential for its application in oil and gas industries.更多还原