Machine Learning-Based Hybrid Random-Fuzzy Modeling Framework for Antenna Design

A machine learning-based framework is proposed to evaluate the effect of design parameters, affected by both aleatory and epistemic uncertainty, on the performance of antennas. In particular, possibility theory is leveraged to define aleatory and epistemic uncertainty in a common framework. Then, a method combining Bayesian optimization and Polynomial Chaos expansion is applied to accurately and efficiently propagate both uncertainties throughout the system under study. A suitable application example validates the proposed method.