Scholarlink

Purpose <p>A T-shape movable three-coupled flexible beams (MTCFB) system has characteristics of small damping and close modes. A hybrid control approach that integrates motion trajectory optimization with piezoelectric active control is proposed to mitigate vibrations.</p> Methods <p>Finite element modeling (FEM) and experimental system identification are employed to develop the piezoelectric driving model and motor acceleration driving model for the MTCFB system. The mathematical model identification is accomplished by wavelet analysis and black widow optimization algorithm (BWOA). The identified model is helpful to strengthen the actual environment of learning controller training. The wave search algorithm (WSA) is employed to obtain the optimal vibration reduction trajectory. Basing on multi-agent variance exploration network (MAVEN), a reinforcement learning algorithm is constructed to obtain the MAVEN controller, which is applied to vibration control experiment.</p> Results and Conclusions <p> Compared with the proportional and derivative (PD) controller, the MAVEN controller has better suppressing vibration ability. Combined with the optimized trajectory of MAVEN controller, the vibration after translation can be effectively suppressed and the residual vibration can be reduced.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Trajectory Planning and Multi-Agent Reinforcement Learning Vibration Control of T-Shape Movable Three-Coupled Flexible Beam

  • 期刊论文

  • 发表日期：

    2025年3月12日

查看全文
  • Hao-Ran Liu,
  • Zhi-Cheng Qiu

摘要

Purpose

A T-shape movable three-coupled flexible beams (MTCFB) system has characteristics of small damping and close modes. A hybrid control approach that integrates motion trajectory optimization with piezoelectric active control is proposed to mitigate vibrations.

Methods

Finite element modeling (FEM) and experimental system identification are employed to develop the piezoelectric driving model and motor acceleration driving model for the MTCFB system. The mathematical model identification is accomplished by wavelet analysis and black widow optimization algorithm (BWOA). The identified model is helpful to strengthen the actual environment of learning controller training. The wave search algorithm (WSA) is employed to obtain the optimal vibration reduction trajectory. Basing on multi-agent variance exploration network (MAVEN), a reinforcement learning algorithm is constructed to obtain the MAVEN controller, which is applied to vibration control experiment.

Results and Conclusions

Compared with the proportional and derivative (PD) controller, the MAVEN controller has better suppressing vibration ability. Combined with the optimized trajectory of MAVEN controller, the vibration after translation can be effectively suppressed and the residual vibration can be reduced.