On the Prediction of Fragment Distribution of Naturally Fragmenting Warheads: a Machine Learning Approach

Abstract

The prediction of fragment data for naturally fragmenting warheads, that are representative of the actual warheads, is of vital importance to lethality and vulnerability modelling, and also to warhead design per se. Fragment speeds can be predicted by means of Gurney (or modified) equations and the fragment trajectory direction by means of an adapted Taylor approximation. Considerable research has been conducted on the prediction of the fragment mass distribution, which includes the models of Mott, Weibull and Held. Specifically non-cylindrical warheads, with varying casing thickness, require additional segmentation and the empirical parameters for the fragment mass distribution are then adjusted manually such that the overall calculated mass distribution corresponds as close as possible to test data. In this study, to acquire a more accurate estimate of the fragment mass distribution for noncylindrical warheads, a machine learning regression algorithm is proposed. As the amount of test data increases, a function model is optimized to best represent the data. The function model is then used to provide a distribution of the test data for simulating new parameters for further analyses.