摘要：During the pipeline detection of the in-pipe robot, obstacles such as welds inside the pipeline will interfere with the movement of the robot. It can result in changes of the flow field environment around the robot. Its velocity and attitude can also change, making the errors of pipeline defect positioning. And it cannot meet the detection requirements. Therefore, a perception model of the in-pipe robot's drive system was established. It can adjust the driving torque in real time by sensing the velocity of the external flow field, the velocity and the attitude of the robot. And then the motion state of the robot can be adjusted to better adapt to the changes of the external environment. First, a composite-driven in-pipe robot was designed. And the dynamic analysis of the in-pipe robot was carried out to calculate the driving torque under different combinations of fluid velocities, velocities of robot motion and attitude angles. An offline database was constructed. Second, support vector regression (SVR) method was used to establish the prediction model of drive system. And the hyperparameters of the model were optimized through double deep Q-network (Double DQN) to improve the prediction performance. Finally, the optimal model parameters were selected through the comparison between the different parameters of the model. The results showed that the prediction error of the established perception model was controlled within 10%, which improved the accuracy of the pipeline detection. ? 2022 IEEE.

ISSN号：9781665406482

卷、期、页：页322-327

发表日期：2022-01-01

期刊分区（SCI为中科院分区）：无

收录情况：EI（工程索引）

发表期刊名称：Proceedings - 11th Electrical Power, Electronics, Communications, Control, and Informatics Seminar, EECCIS 2022

通讯作者：苗兴园,马英涵

第一作者：赵弘

论文类型：会议论文

论文概要：苗兴园,赵弘,马英涵，Perception Modeling of In-Pipe Robot based on Machine Learning，Proceedings - 11th Electrical Power, Electronics, Communications, Control, and Informatics Seminar, EECCIS 2022，2022，页322-327

论文题目：Perception Modeling of In-Pipe Robot based on Machine Learning