Bistable DVR meta-structure: A reinforcement learning approach to vibration control

Abstract

This research introduces a data-driven approach to vibration control, utilizing programmable bandgap metastructures tailored for applications such as cargo protection. Central to this design are bistable dynamic vibration absorbers (DVRs), which leverage a dual-state mechanism to enhance energy dissipation and improve vibration isolation under varying dynamic loads. The study employs a finite element approach to model the dynamic behavior of these metastructures, simulating their response to different forcing conditions. We then utilize artificial neural networks (ANNs) for predictive modeling of the metastructures’ characteristics to various DVR configurations, thereby enhancing their real-time adaptability. Additionally, we integrate reinforcement learning (RL) algorithms to dynamically adjust the bandgap behavior of the metastructures, allowing for on-the-fly tuning of their vibrational characteristics. This approach proves highly effective in scenarios with diverse vibration profiles. Experimental results demonstrate the efficacy of this comprehensive strategy, showcasing significant resonance frequency shifts and improved vibration absorption characteristics. This approach offers a robust solution for dynamic adaptation in various vibrational environments.