RCS Minimization using Bayesian Optimization

This work proposes to use Bayesian optimization and shows that RCS minimization can be achieved with a very small number of initial observations and function evaluations. We form a surrogate model of mean RCS of a radome shaped distribution of scattering centers with respect to its geometric parameters using Gaussian process regression. Based on the mean and the variance of the response surface, the next RCS computation point is chosen where the expected improvement in mean RCS is maximized. In this way, the surrogate model is updated with a new RCS computation and the process is repeated until convergence. Since each iteration predicts the next best design point for RCS minimization, the surrogate model may not be fit for RCS prediction in the entire design space, but it reduces the total number of function evaluations that are required to minimize RCS. This work is particularly useful in multidisciplinary optimization where RCS is a parameter such as antenna, ship, and aircraft structure optimization.