摘要：Compared with traditional plugging methods, the in-pipe intelligent isolation plugging tool (IPT) is advantageous in safety and work efficiency. However, during the plugging process, the flow field around the IPT changes drastically, resulting in vortex-induced vibration and potential failure of the plugging operation. In this study, three foldable spoilers were designed at the tail of the IPT to optimize the flow field. The vibration of the IPT can be alleviated by adjusting the angles of the spoilers. A vibration reduction control system of the IPT was designed based on deep reinforcement learning. First, we conducted an experiment for vibration reduction system. Second, a nonlinear model of the pressure difference based on experimental data was established. Then, a multi-agent self-learning system based on the deep Q-network (DQN) was designed, and the optimal actions were selected in each agent to adjust the spoiler angles during the plugging process. Finally, a controller based on fuzzy reinforcement learning was proposed to flip the spoilers to the optimized angles. The results show that the vibration reduction control system of the IPT reduced the pressure difference by an average of 28.32%, which indicates the stability of the plugging process and a successful reduction of the IPT vibration.

关键字：Intelligent isolation plugging tool; Vibration reduction; Spoiler; DQN; Multi-agent; Fuzzy reinforcement learning

ISSN号：2288-6206

卷、期、页：卷: 9期: 6页: 1477-1491

发表日期：2022-11-01

影响因子：5.671300

期刊分区（SCI为中科院分区）：二区

收录情况：SCI（科学引文索引印刷版）,EI（工程索引）,SCIE（科学引文索引网络版）

发表期刊名称：INTERNATIONAL JOURNAL OF PRECISION ENGINEERING AND MANUFACTURING-GREEN TECHNOLOGY

通讯作者：苗兴园,高博轩,吴婷婷,侯岩光

第一作者：赵弘

论文类型：期刊论文

论文概要：苗兴园,赵弘,高博轩,吴婷婷,侯岩光，Vibration Reduction Control of In-Pipe Intelligent Isolation Plugging Tool Based on Deep Reinforcement Learning，INTERNATIONAL JOURNAL OF PRECISION ENGINEERING AND MANUFACTURING-GREEN TECHNOLOGY，2022，卷: 9期: 6页: 1477-1491

论文题目：Vibration Reduction Control of In-Pipe Intelligent Isolation Plugging Tool Based on Deep Reinforcement Learning