A physics-informed neural network surrogate for multi-band reflection-loss prediction of carbon-fiber-reinforced polymer / Ti-6Al-4V ELI radar absorbing materials for UAV airframe integration

Abstract Radar absorbing materials (RAM) are critical for stealth performance of unmanned aerial vehicles (UAVs), but their design typically requires expensive parametric sweeps over frequency, incidence angle, and material thickness. This work presents a physics-informed neural network (PINN) surrogate model for rapid prediction of reflection loss (RL) of a metal-backed carbon-fiber-reinforced polymer (CFRP) absorber over the 2–18 GHz band and 0–80° transverse-electric (TE) incidence range. The model uses random Fourier feature embedding to mitigate spectral bias, combined with physics-based monotonicity and angular boundary regularization. Trained on a transfer matrix method (TMM) reference dataset of 6,400 samples in 42 s on a single CPU, the surrogate achieves a root-mean-square error of 0.22 dB and mean absolute error of 0.12 dB against the analytical baseline. Multi-band thickness optimization identifies the quarter-wave optima at S-band (3.9 mm, $$RL = -4.64 dB$$ R L = - 4.64 d B ), X-band (1.9 mm, $$RL = -8.93 dB$$ R L = - 8.93 d B ), and Ku-band (1.5 mm, $$RL = -11.31 dB$$ R L = - 11.31 d B ). Practical structural integration of the CFRP / Ti-6Al-4 V ELI bilayer onto medium-altitude long-endurance UAV airframes is assessed quantitatively, demonstrating mass overheads of 1.3–8.5% for the Ku-band design. The proposed framework provides a computationally efficient pathway for coupled material and structural UAV design. Highlights Physics-informed neural network predicts CFRP absorber reflection loss approximately fifty times faster than the analytical baseline. Random Fourier feature embedding and physics regularization yield sub-decibel accuracy across the full angular range. Multi-band thickness optimization yields a CFRP coating compatible with UAV airframes at modest mass overhead.