Vibration control of thin structures using a reinforcement learning approach

Date

2023-9

Author

Wanyonyi, Sandra Nafuna

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

134
views

63
downloads

Cite This

Vibration control in thin structures is an important research topic with numerous practical applications in engineering and robotics. With the continued use of thin structures for space applications like solar sails, space reflectors, and satellite antennas, there is an ever-growing interest in effective and robust vibration suppression methods. This thesis presents a comprehensive study into the application of Reinforcement Learning (RL) as a control approach to suppress the vibration in thin beams with pinned-pinned boundary conditions. The primary focus of this research is to contribute a novel insight into the field of vibration control for thin structures. With this goal in mind, a novel approach for vibration control in thin beams with pinned supports is introduced. A robust RL controller is developed to effectively handle parameter uncertainties and varying external disturbances. In this case, they are modeled as varying natural frequency and initial displacements of the the beam-actuator environment, respectively. By incorporating these uncertainties, the research offers one of the earliest illustrations of the impact of parameter uncertainty on the performance of an RL controller for vibration control in thin beams. The learning and control performance of off-policy and on-policy RL algorithms for vibration control in thin pinned-pinned beams is evaluated, reviewed, and discussed with an aim to identify the most suitable RL approach for effectively suppressing vibrations in such structures. A detailed comparative study is provided on the reward-shaping process of the problem by two different reward function schemes being implemented, and the results obtained from both are compared and discussed. The controllers developed from the different reward schemes illustrate a significant difference in their performance and reinforce the importance of the process in the RL controller development process. Despite the simple uncertainty model, the results indicate that the developed RL controller can cope with changing system parameters and varying initial excitations. The controlled response of the beam system vibration for agents that were trained in environments with different dynamics also indicates the success of the RL controller application.

Subject Keywords

Reinforcement learning, Vibration control, Active control, Thin structures

URI

https://hdl.handle.net/11511/106046

Collections

Graduate School of Natural and Applied Sciences, Thesis