A partition-based point-cloud regression framework is developed for fast prediction of multi-angle radar cross section (RCS) responses of three-dimensional radome geometries. A family of 125 radomes is generated by independently scaling a reference geometry along the three coordinate axes. After surface meshing, METIS is used to divide each radome into 2000 connected surface regions. The center coordinates and unit normal vector of each region are extracted to form a six-dimensional geometric feature. The partition-center coordinates are normalized using a common spherical normalization scheme, while the normal vectors are retained in their unit-normalized form. The operating frequency is standardized, and the incident angles are encoded using trigonometric functions. The geometric features are processed by shared point-feature transformations and mean--max aggregation, while the electromagnetic conditions are encoded separately. The resulting geometry and condition features are fused and mapped to a complete 360-sample angular RCS response. Numerical results show that the proposed framework can effectively reproduce the main variations of the simulated RCS responses under different excitation conditions, while significantly reducing the computational cost of repeated electromagnetic simulations.

radar cross section, radome, point cloud, electromagnetic scattering, multi-angle RCS prediction