A hybrid AI-driven framework for wide-angle low-RCS stealth satellite design and adjustable nose cone implementation

This paper presents an efficient optimization framework to address the critical challenge of designing satellite with robust wide-angle stealth capabilities. The framework first parameterizes the satellite forebody using B-splines and constructs a high-fidelity dataset via electromagnetic simulations. A deep neural network is then trained as a surrogate model, achieving a Mean Absolute Error (MAE) of only 2.42 dB in predicting the Radar Cross Section (RCS). By integrating this high-precision surrogate model with an Ant Colony Optimization (ACO) algorithm, the framework rapidly identifies optimal low-RCS shapes within the vast design space. The framework’s efficacy was experimentally validated: a fabricated prototype of an optimized design achieved a minimum RCS of − 50 dBsm in anechoic chamber tests, while maintaining the wide-angle RCS below − 30 dBsm within the ±45° elevation range, demonstrating excellent agreement with simulation predictions. Furthermore, the framework’s capability was extended to design and implement an adjustable nose cone with actively adjustable stealth characteristics. This intelligent structure exhibits a maximum dynamic RCS adjustment range of up to 25 dB between its extended and retracted states; specifically, the adjustable design achieves RCS reductions of 20 dB at 2 and 5.5 GHz, and 10 dB at 18 GHz, providing a novel and validated paradigm for designing adjustable intelligent stealth systems.