MLP-residual surrogate model for aerodynamic prediction in projectile external flows

Area of Science:

• Aerospace Engineering
• Computational Fluid Dynamics
• Machine Learning

Background:

• Accurate aerodynamic prediction is crucial for real-time projectile trajectory simulation and digital twin systems.
• Conventional computational fluid dynamics (CFD) methods are too computationally expensive for real-time applications.

Purpose of the Study:

• To develop a rapid and accurate aerodynamic prediction method for exterior ballistics.
• To improve the performance of surrogate models for aerodynamic parameter prediction.

Main Methods:

• A primary-residual dual-stage surrogate modeling approach using a multilayer perceptron (MLP).
• A primary model maps global nonlinear aerodynamics using CFD data (Mach number, angle of attack).
• A residual network compensates for fitting deviations in highly nonlinear regions.

Main Results:

• The proposed dual-stage MLP model significantly improves prediction accuracy compared to polynomial fitting, RBF, and single-stage MLP.
• Reduced Root Mean Square Error (RMSE) for lift prediction from 2.91 to 0.87.
• Reduced RMSE for pitching moment prediction from 0.634 to 0.086.
• Achieved an average RMSE reduction of approximately 50% across all aerodynamic parameters.

Conclusions:

• The primary-residual dual-stage surrogate model offers superior accuracy, robustness, and generalization capability.
• The method is highly applicable for digital twin-driven exterior ballistic prediction systems.
• Demonstrates practical engineering applicability for real-time aerodynamic prediction.