Structural optimization of an arch-structured epoxy/rubber composite vibration isolator using deep Q-value neural network reinforcement learning

Abstract

Vibration isolators are used to reduce unexpected vibrations that can negatively affect the system performance. For high load-bearing applications, vibration isolators must have high vibration reduction capabilities and strong load-carrying capacities. However, most studies on vibration isolators have focused only on vibration reduction characteristics. In this study, we performed structural optimization of an arch-structured epoxy/rubber composite vibration isolator using deep Q-value neural network (DQN) reinforcement learning. Our optimization framework, based on DQN reinforcement learning, iterated through modeling and structural/vibration analysis. Consequently, an optimal model with a high load-carrying capacity and excellent vibration reduction performance was obtained. This optimal model had a high safety factor even under a high pre-load and an elastic wave transmissibility value of less than −20 dB in the target frequency range. To validate the optimization results, an arch-structured epoxy/rubber composite vibration isolator was fabricated and vibration experiments were conducted. The fabricated vibration isolator showed good agreement with the vibration analysis results and excellent vibration reduction characteristics in the target frequency range, as confirmed by the transient response tests. This optimization method has potential applications in various fields because it can derive an optimal model considering multiple objectives with an efficient computational cost.