Neural Network Application for Aerodynamic Force Estimation in Heavy Artillery Projectile Trajectories

Resumo

This work proposes the use of a neural network to estimate components of the aerodynamic forces acting on the trajectory of a 155 mm projectile. The approach aims to replace analytical expressions with a machine learning-based model, in order to compare the computational efficiency and flexibility of ballistic modeling techniques. The network was designed to estimate, based on the projectile's aerodynamic coefficients, the lift, drag, Magnus effect, rotational damping forces, and the projectile's angular adjustment. The dataset used for training was generated through the analysis of aerodynamic coefficients using the ballistic software PRODAS, considering different elevations, muzzle velocities, and atmospheric conditions. For validation, real firing data under various conditions were employed. The results show a mean squared error on the order of 3.10⁻³ and a correlation coefficient (R²) above 0.9, indicating that the neural network is capable of adequately reproducing aerodynamic effects and presents itself as a viable alternative to traditional models.