Interpretable SHAP-bounded Bayesian optimization for underwater acoustic metamaterial coating design

The paper presents a framework to design underwater acoustic coatings made from polyurethane elastomers that include metamaterial-like geometric features. The goal is to improve acoustic performance, specifically sound absorption, by selecting appropriate geometrical design variables. A data-driven model is used to learn how the design variables affect sound absorption. To make these relationships understandable, the authors apply SHAP (SHapley Additive exPlanations) to identify which parameters most influence the objective, both across the whole design space and for local regions. The SHAP-derived insights are then used to automatically tighten (refine) the bounds of the design space, steering the Bayesian optimization toward more promising regions. The approach is tested on two polyurethane materials with different hardness levels and is reported to yield improved optimal designs compared to standard Bayesian optimization without increasing the number of simulations.