Open Access Open Access  Restricted Access Subscription Access

DETERMINATION OF BALLISTIC LIMIT CURVES USING PHYSICS-INFORMED OVERSAMPLING MACHINE LEARNING ALGORITHMS

Teresa Fras, Tomasz Hachaj, Paul Baillard, Gregory Vincent

Abstract


Machine learning methods allow us to model phenomena that lack analytical or phenomenological formulas and for which finite element method (FEM) modelling is computationally very expensive. An example of such a phenomenon is the investigation of the relationship between input velocity and residual velocity during projectile penetration as a function of a given material's thickness. In this paper, we present a numerical solution to this problem. We will use the well-known ML algorithms Random Forest and XGBoost, trained on real data from ballistic experiments, supplemented with numerical data from the FEM. We will then augment the dataset with physics-informed oversampling, resulting in a numerical model of the phenomenon capable of approximating residual velocities for material thicknesses for which no experiments or FEM simulations were performed. Furthermore, we will examine the accuracy of the approximations of ballistic limit curves and the model’s ability to predict ballistic limit velocities.


DOI
10.12783/ballistics26/37596

Full Text:

PDF

Refbacks

  • There are currently no refbacks.