Heterogeneous reinforcement learning vibration control of coupling system with four flexible beams connected by springs

• Published in: Mechatronics (Oxford) Vol. 95; p. 103063
• Main Authors: Qiu, Zhi-cheng, Yang, Yang, Zhang, Xian-min
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.11.2023
• Subjects: Experimental identification; Four-flexible beam coupling system; HATRPO; RL algorithm; Vibration control; Vibration control; RL algorithm; Experimental identification; HATRPO; Four-flexible beam coupling system
• ISSN: 0957-4158; 1873-4006
• DOI: 10.1016/j.mechatronics.2023.103063

•A coupling system of four flexible beams connected by springs is designed.•Wavelet transform and MVO algorithm are used to identify the model parameters.•Vibration controller is trained by the HATRPO algorithm.•The experiments verify the vibration control effectiveness of the HATRPO algorithm. Aiming at studying the vibration characteristics and active control of a coupling system with four flexible beams connected by springs, an experimental platform is built. The dynamic equation of the system is solved by finite element method (FEM), and the parameter model based on state space equation is deduced. In order to ensure the accuracy of the parameter model, an experimental identification method based on wavelet transform and optimization algorithm is adopted. The state matrix, observation matrix and control force coefficient matrix in the parameterized model are solved in turn. A multi-agent based Heterogeneous-Agent Trust Region Policy Optimization (HATRPO) reinforcement learning (RL) algorithm is designed. The HATRPO RL algorithm interacts with the identified parameter model. After several rounds of training, the HATRPO RL vibration controller is finally obtained. The simulation and experimental results show that the HATRPO RL controller can well compensate for the nonlinearity and uncertainty in the multi-flexible beam coupling system. In addition, the nonlinear characteristics of the HATRPO RL algorithm effectively solve the problem of insufficient control power of traditional linear controller in small vibration amplitude, and realize faster vibration suppression. [Display omitted]