Achieving broadband electromagnetic absorption in laminated composites through progressive Bayesian optimization

As radar detection technology advances, achieving broadband electromagnetic (EM) absorption in laminated composites remains a critical challenge due to the intricate nonlinear coupling between structures/compositions and frequency-dependent EM responses, which demands specialized expertise and extensive trial-and-error efforts. Here, this paper presents a data-driven machine learning (ML) framework based on progressive Bayesian optimization (PBO), which addresses traditional Bayesian optimization's (BO) limitations to local optima. Within this framework, a predictive model is trained iteratively for EM performance of variable ply configurations, which exhibits desirable robustness and generalization ability. Notably, this framework is parameter-free, making it applicable to the design of EM properties in similar laminated structural systems. Through this PBO strategy, a gradient laminated structure with a thickness of 5.1 mm has been proposed, which achieves an effective absorption bandwidth (EAB) of 9.6 GHz benefiting from enhanced impedance matching. Compared to traditional BO and “native” random search, the EAB of this optimal design represents an increase by ∼20 % and ∼36 %, respectively. During the iterative process, the PBO-ML framework progressively elucidates the EM response mechanisms associated with variations in composition and structures, which endeavors to reconcile the trade-off between composition and structures from high to low frequencies.