We propose a steady-state aerodynamic data-driven method to predict the incompressible flow around airfoils of NACA (National Advisory Committee for Aeronautics) 0012-series. Using the Signed Distance Function (SDF) to parameterize the geometric and flow condition setups, the prediction core of the method is constructed essentially by a consecutive framework of a convolutional neural network (CNN) and a deconvolutional neural network (DCNN). Impact of training parameters on the behavior of the proposed CNN-DCNN model is studied, so that appropriate learning rate, mini-batch size, and random deactivation rate are specified. Tested by “unseen” airfoil geometries and far-field velocities, it is found that the prediction process is three orders of magnitudes faster than a corresponding Computational Fluid Dynamics (CFD) simulation, while relative errors are maintained lower than 1% on most of the sample points. The proposed model manages to capture the essential dynamics of the flow field, as its predictions correspond reasonably with the reconstructed field by proper orthogonal decomposition (POD). The performance and accuracy of the proposed model indicate that the deep learning-based approach has great potential as a robust predictive tool for aerodynamic design and optimization.

To alleviate the computational burden associated with the computational fluid dynamics (CFD) simulation stage and improve aerodynamic optimization efficiency, this work develops an innovative procedure for airfoil shape optimization, which is implemented through coupling the genetic algorithm (GA) optimizer with the aerodynamic coefficients prediction network (ACPN) model. The ACPN is established using a fully connected neural network with the airfoil geometry as the input and aerodynamic coefficients as the output. The results show that the ACPN's mean prediction accuracy for the lift and drag coefficient is high up to about 99.02%. Moreover, the prediction time of each aerodynamic coefficient is within 5 ms, four orders of magnitude faster compared to the CFD solver (3 min). Taking advantage of the fast and accurate prediction, the proposed ACPN model replaces the expensive CFD simulations and couples with GA to force the airfoil shape change to maximize the lift–drag ratio under multiple constraints. In terms of time efficiency, optimized airfoils can be fast obtained within 25 s. Even considering an extra 50 h spent on data preparing and 20 s for model training, the overall calculation cost is reduced by a remarkable 62.1% compared to the GA-CFD optimization method (5.5 days). Furthermore, the GA-ACPN model improves the lift–drag ratio with and without constraint by 51.4% and 55.4% for NACA0012 airfoil, respectively, while 50.3% and 60.0% improvement achieved by the GA-CFD optimization method. These results indicate that the GA-ACPN optimization approach significantly enhances the optimization efficiency and has great potential to address varying constraint optimization problems.

Computational efficiency and precision pose a classic contradiction in aerodynamic shape optimization. To address this challenge, this study introduces an effective optimization framework based on multi-fidelity fully connected neural network (MFFCN). The framework utilizes transfer learning (TL) to train a multi-fidelity surrogate model that establishes direct mappings between geometric configuration parameters and aerodynamic performance by adaptively capturing linear or nonlinear relationships concealed between high-fidelity (HF) and low-fidelity (LF) information. The HF and LF data are derived from fine and coarse grids, respectively, evaluated using the same computational fluid dynamics (CFD) model. The MFFCN-TL framework is applied to optimize the National Advisory Committee for Aeronautics 0012 (NACA0012) airfoil (12 design variables) and the Office National d'Études et de Recherches Aérospatiales M6 (ONERA M6) wing (50 design variables). Simulation results demonstrate that the NACA0012 airfoil achieves a 69.47% enhancement in lift–drag ratio in 1.069 s, compared to a 12.76% gain over 24.8 h in single-fidelity CFD-based optimization. The ONERA M6 wing achieves a 24.66% reduction in drag coefficient in 694 ms compared to 18.37% over 237.3 h in the CFD model. Statistical results show that the MFFCN-TL framework can reduce optimization cost by more than 90% compared to the single-fidelity CFD-based model. These findings suggest that the MFFCN-TL framework significantly enhances optimization efficiency and provides superior feasible solutions over single-fidelity methods.

This study applies deep-reinforcement-learning algorithms to integrated guidance and control for three-dimensional, high-maneuverability missile-target interception. Dynamic environment, reward functions concerning multi-factors, agents based on the deep-deterministic-policy-gradient algorithm, and action signals with pitch and yaw fins as control commands were constructed in the research, which control the missile in order to intercept targets. Firstly, the missile-interception system includes dynamics such as the inertia of the missile, the aerodynamic parameters, and fin delays. Secondly, to improve the convergence speed and guidance accuracy, a convergence factor for the angular velocity of the target line of sight and deep dual-filter methods were introduced into the design of the reward function. The method proposed in this paper was then compared with traditional proportional navigation. Next, many simulations were carried out on high-maneuverability targets with different initial conditions by randomization. The numerical-simulation results showed that the proposed guidance strategy has higher guidance accuracy and stronger robustness and generalization capability against the aerodynamic parameters.

This paper proposes a novel guidance law for intercepting a high-speed maneuvering target based on deep reinforcement learning, which mainly includes the interceptor–target relative motion model and value function approximation model based on deep Q-Network (DQN) with prioritized experience replay. First, a method called prioritized experience replay is applied to extract more efficient samples and reduce the training time. Second, to cope with the discrete action space of DQN, a normal acceleration is introduced to the state space, and the normal acceleration rate is chosen as the action. Then, the continuous normal acceleration command is obtained using numerical integral method. Third, to make the line-of-sight (LOS) rate converge rapidly, the reward function whose absolute value tends to zero has been constructed. Finally, compared with proportional navigation guidance (PNG) and the Q-Learning-based guidance law (QLG), the simulation experiments are implemented to intercept high-speed maneuvering targets at different acceleration policies. Simulation results demonstrate that the proposed DQN-based guidance law (DQNG) can obtain continuous acceleration command, make the LOS rate converge to zero rapidly, and hit the maneuvering targets using only the LOS rate. It also confirms that DQNG can realize the parallel-like approach and improve the interception performance of the interceptor to high-speed maneuvering targets. The proposed DQNG also has the advantages of avoiding the complicated formula derivation of traditional guidance law and eliminates the acceleration buffeting.