Machine Learning Assisted Worst-Case Radar Cross Section Optimisation Using Geometry-Aware Surrogate Modelling

Radar Cross Section (RCS) reduction is a key consideration in the early-stage design of low observable aerospace platforms. Direct optimisation of RCS using full-wave electromagnetic solvers is often computationally prohibitive when iterative design exploration is required. To address this challenge, this paper presents a software-oriented framework for worst-case RCS optimisation based on a physics-inspired synthetic benchmark model coupled with machine learning and genetic algorithm optimisation. A compact set of geometry and material descriptors, including platform dimensions, surface complexity, coating thickness, and dielectric loss tangent, is employed to parameterize the design space. A synthetic RCS benchmark model is used to generate training data capturing dominant qualitative scattering trends, which are subsequently learned by a Random Forest regression model. The trained surrogate enables rapid evaluation of candidate designs within a genetic algorithm framework that minimizes the worst-case RCS over the full angular domain. The proposed framework is implemented as a standalone graphical software tool supporting geometry input, surrogate model training, optimisation, visualization, and automated report generation. Results obtained within the synthetic benchmark en- vironment demonstrate stable optimisation behavior and physically consistent design trends, highlighting the utility of the approach for rapid design space exploration and methodological evaluation rather than final electromagnetic validation.