Predicting Warhead Fragment Dispersion Using Stepwise Machine Learning Regression

Abstract The dispersion characteristics of warhead fragments—direction, velocity, and mass—are crucial for assessing weapon effectiveness and informing the design of efficient weapon systems and operational strategies. Conventional methods for obtaining fragment data of a warhead, such as arena fragmentation tests and the finite element Method (FEM), are often limited by high costs and extended processing times. A faster approach is needed to generate fragment dispersion data with acceptable accuracy for supporting real-time decision-making and iterative design cycles. This study introduces a machine learning-based regression framework to develop a reduced-order model (ROM) for rapid prediction of fragment dispersion characteristics. A stepwise regression approach is utilized to enhance predictive accuracy. In the first stage, warhead parameters are used to estimate fragment velocity and direction at discrete warhead positions. In the second stage, these outputs serve as additional inputs to predict spatial dispersion, defined as fragment distributions across polar zones. The proposed framework achieves accuracy comparable to FEM while significantly reducing computational time. This efficient, cost-effective machine learning approach is well-suited for rapid weapon effectiveness evaluations in wartime scenarios and for optimizing next-generation weapon system designs.