自然语言驱动的翼型设计与优化方法研究OA

Airfoil design and optimization,as a critical link in the aerodynamic design of aircraft,directly determines the lift-drag characteristics and overall performance of flight vehicles.However,existing artificial intelligence methods based on neural networks and reinforcement learning often suffer from limited generalization ability,large data requirements,and difficulties in achieving cross-domain collaboration.To address these limitations,this study introduces generative artificial intelligence(Generative AI),particularly technologies related to large language model(LLM),into airfoil design and optimization tasks.By integrating the class function/shape function transformation(CST)method with natural language interactive modeling,we propose an engineering method for natural language-driven airfoil design and optimization,namely the"LLM+CST"framework.Taking small general aviation aircraft and long-endurance unmanned aerial vehicle(UAV)as case studies respectively,we verify the effectiveness of LLMs in the conceptual design phase using an evaluation index system tailored for"LLM-based airfoil design".In three sets of distinct LLM-based airfoil optimization experiments,we assess the applicability of various optimization workflows and identify the superior-performing methods by employing an evaluation index system and scoring rules specifically designed for"LLM-based airfoil optimization".Using the NACA 2418 as the baseline airfoil,we compare the optimization results from DeepSeek-R1 with those obtained via genetic algorithm optimization.The results demonstrate that the proposed method not only achieves a significant improvement in aerodynamic performance but also substantially reduces the optimization time.This research confirms that the"LLM+CST"method exhibits excellent generalization and interactivity,providing novel insights and practical evidence for the application of LLMs in the field of aerodynamic design in aviation.