Optimization of hot-air anti-icing structures based on neural networks and genetic algorithms

Hot-air anti-icing systems are critical for ensuring the safety and reliability of aircraft operations in icing conditions. Traditional design methods for these systems depend heavily on numerical simulations to predict the surface temperature distribution and optimize the structural parameters, which are inefficient. This study proposed an innovative optimization method that integrated neural networks with genetic algorithms to address these challenges. A multi-objective optimization method based on the Non-dominated Sorting Genetic Algorithm II (NSGA-II) was applied to minimize the required bleed air quantity and maximize the average temperature within the anti-icing region. The method efficiently identifies a Pareto-optimal solution set, offering diverse design options that meet the constraints. Experimental results show that the optimized designs significantly reduce the required air bleed while enhancing the anti-icing performance. A data-driven prediction model, named Multi-CNNs with GRU (MCG), for the surface temperature distribution, was developed using optimized Latin Hypercube Sampling (OLHS) and high-dimensional numerical simulation data. The proposed neural network model demonstrates high accuracy, achieving a mean absolute error of less than 1.5 K and a mean prediction accuracy exceeding 96%. The optimization framework used the MCG model as an individual evaluation algorithm to achieve fast prediction of high-dimensional temperature distributions with a computation time of about 5.5 ms per sample, which is thousands of times more efficient compared to traditional numerical simulations. Comparative analysis with a reduced-order model, POD-AlexNet, highlights the superior accuracy and stability of the MCG model, particularly in predicting complex three-dimensional surface temperature distributions. Additionally, the optimization results demonstrate that the proposed method provides a reliable trade-off between improving anti-icing performance and minimizing energy consumption, thereby enhancing the practical applicability. This study provides an efficient tool for optimizing hot-air anti-icing structures, enabling rapid and accurate decision-making in engineering design. The proposed framework holds significant potential for advancing the development of anti-icing systems in aviation and related fields. This study presents a novel data-driven approach with potential for engineering application, facilitating rapid and efficient optimization design of hot-air-anti-icing systems.