AI-Driven Acoustic Metamaterial Optimization for Broadband Noise Control

The acoustic metamaterials have some transformational advantages in terms of noise reduction, but it is still a challenge to design them to be broadband. Conventional design methodologies are based on parametric sweeps and trial-error system, not effective to the complicated metamaterial geometries. The paper presents an artificial intelligence (AI)-based design methodology of broadband acoustic metamaterials inverse optimization. In a hybrid approach, deep reinforcement learning (DRL) is combined with finite element simulations to discover, with ease, design spaces and estimate by trial and error structures with small area of search that obtain desired acoustic performances over broader frequency bands. The framework uses surrogate model through neural networks which minimizes the computational expenses allowing evaluation of candidate geometries quickly. Examples of case studies are the design of labyrinthine panels and locally resonant metal panels used in aircraft engines noise and industrial machines. Findings show the highest sound transmission loss to the conventional designs with the optimized geometries that are tunable as well as structurally robust. The uniqueness of this study is the integration of AI with computational acoustics to limit the drawback of manual design, which can offer a scaling channel of broadband noise control solutions. The framework not only makes the process of metamaterial discovery faster but it also creates a pipeline that allows replication of AI integration in the engineering of computational acoustics.