We design a reinforced learning (RL) algorithm which learns to solve the multiobjective problem of the design of acoustic materials made of Helmholtz resonators minimizing both the volume and the resonance frequency. RL is based on a reward system that evaluates the quality of the decisions made. The idea is to mimic human learning, which, through interaction with the surrounding environment, learns to choose the best action to take in various contexts. The two entities in the RL model are the environment and the agent that makes decisions influencing the environment. Each interaction is called step, while the situation presented by the environment at each step is called state. The agent, observing the current state of the environment, takes a decision called action. The criterion for choosing actions is refined through experience and guided by the reception of a reward that either rewards or penalizes the agent’s work at each step. Starting from an initial state, the agent performs a sequence of steps until reaching a final state. This sequence is called an episode. Through numerous episodes, each consisting of several steps, the algorithm learns to navigate optimally in any situation. The optimal geometries are 3-D printed and experimentally evaluated.
Novelli et al. (Wed,) studied this question.