A Target Testing Conditional Generative Adversarial Network (TT-CGAN) is developed and applied to airfoil inverse design. This network extends the Conditional Generative Adversarial Network (CGAN) by integrating a Multi-Layer Perceptron (MLP) tester, so as to enhance the capability of CGAN in evaluating the impact of additional conditions on target testing. Utilizing the UIUC airfoil database, 797 real airfoils were selected, and their corresponding aerodynamic parameters were calculated by solving the Reynolds-Averaged Navier-Stokes (RANS) equations to construct a comprehensive airfoil database. The airfoil shapes were parameterized using the Class Shape Transformation (CST) method, transforming the geometric parameters from 100 to 14 CST parameters. Multi-modal aerodynamic parameters, including lift coefficient, drag coefficient, and surface pressure distribution, were fused using the feature-level fusion approach. These parameters were compared with the aerodynamic parameters based solely on lift and drag coefficients, which served as auxiliary conditions for the network during the airfoil inverse design process. The results indicate that the TT-CGAN based inverse design method generates more accurate airfoils, with an average root mean square error of 1.779×10^-3 and an average mean absolute error of 1.351×10^-3 in airfoil geometry. The generated airfoils were further validated through numerical simulations by solving the RANS equations, demonstrating an average relative error of 3.599 8% for the lift coefficient and 3.723 9% for the drag coefficient, confirming that the generated airfoils can meet the specified design criteria. Analysis of the lift-to-drag ratio distributions reveals that 40% of the test airfoils achieved lift-to-drag ratios within the [20, 30) range, compared to only 16% in the training set. This finding highlights the method's capability to make accurate predictions even within data-sparse regions, showcasing its generalizability.

It is of great significance to find out how to deform a three-dimensional deformed wing independently to meet the requirements of aerodynamic performance and the basic function of mission adaptive deforming under variable operating conditions. In this study, an RLID (reinforcement learning inverse design) framework is proposed and applied to the reverse design of three-dimensional morphing wings for adaptive morphing flight missions under variable operating conditions. The CST(class-shape function transformation) parameterization method is chosen to define three-dimensional morphing wings, and the Latin hypercube sampling method is used to sample in the design space and generate sample points. Computational fluid dynamics simulations are performed to obtain corresponding aerodynamic parameters, and the deep belief network surrogate model is constructed to map the input-output relationship between morphing design parameters and aerodynamic parameters. To address the variable operating conditions, a DQN (deep Q-network) reinforcement learning agent, leveraging unsupervised learning, is used to provide real-time morphing strategies, and the results meet about 70% of the expected aerodynamic performance requirements, and the average aerodynamic performance reaches more than 98% of the requirements. Furthermore, the design results via the DQN agent are compared with those via the G-CGAN (greedy-based conditional generative adversarial network) agents. The results indicate that the proposed RLID framework efficiently obtains a satisfactory strategy of morphing wings under variable operating conditions and that the DQN agent focuses more on overall task rewards than the G-CGAN agent.

For the purpose of reducing the number of the design variables of the wing, the PCA (principal component analysis) technique is employed to improve the CST parameterization method. A database of the wings is established, including geometrical parameters and aerodynamic parameters. Based on the database, the DBN (deep belief network) is established and trained as the surrogate model in the aerodynamic optimization. In order to improve the convergence rate and global searching ability of the standard MOPSO (multi-objective particle swarm optimization) algorithm, an improved MOPSO algorithm is developed based on the α-stable distribution functions. By embedding the DBN surrogate model into the improved MOPSO algorithm, the multi-objective aerodynamic optimizations are performed for the wing of a general aviation airplane. The optimization results indicate that, the computation amount is decreased dramatically by introducing the PCA technique. Numerical validation is conducted for the designed wing configuration, and the results indicate that, comparing with the baseline configuration, the lift coefficients of the designed configuration are increased obviously under multiple states with the drag coefficients not increased.