Multi-fidelity surrogate optimization of an electrothermal ice protection system for in-flight anti-icing

Abstract

This study proposes a novel multi-fidelity surrogate optimization (MFSO) framework for evaluating and optimizing an electrothermal ice protection system for in-flight anti-icing. This approach effectively integrates high-fidelity and low-fidelity simulations to significantly reduce the number of training samples required to train an accurate surrogate model, thereby lowering computational costs without compromising accuracy. The surrogate model was constructed via proper orthogonal decomposition (POD) and multi-fidelity Gaussian process regression (MFGPR). To perform high-fidelity simulations of the electrothermal anti-icing process, a unified finite volume framework was employed. This framework efficiently combines several in-house solvers, including a compressible Navier–Stokes–Fourier (NSF) airflow solver, an Eulerian droplet impingement solver, a partial differential equation (PDE)-based ice accretion solver, and a multilayer heat conduction solver. The optimization framework minimizes total electric power consumption while satisfying regime-specific constraints in both running-wet and evaporative modes. The search for optimal heater power distributions is guided by the mesh adaptive direct search (MADS) algorithm. Our findings show that the multi-fidelity surrogate model reduces training cost by 75% by lowering the number of required high-fidelity simulations from 400 in the single-fidelity approach to 100. The optimization results achieved up to 38.5% reduction in total electric power consumption for the running-wet regime and 40% for the evaporative regime under representative operational constraints for the NACA0012 airfoil.