Surrogate drag model of non-spherical fragments based on artificial neural networks

The hazard evaluation of the fragments produced from an explosion requires an accurate prediction of the motion of fragments subjected to gravity and aerodynamic forces. The drag coefficient (Cd) is among the most crucial components of various aerodynamic models. Limited experiments cannot reproduce the Cd (Mach) curves of all shapes of fragments. In this study, we develop a surrogate model of the average drag coefficient, C¯d, for the randomly tumbling non-spherical fragment using artificial neutral networks. To train and validate the surrogate model, a comprehensive dataset was developed through mesoscale simulations of Cd for a wide variety of fragment shapes combined with icosahedron average method. The surrogate model shows that the dependence of C¯d on different Mach numbers varies with fragment shape. The fully validated surrogate model of C¯d allows us to derive the statistics of C¯d for a host of fragments from a specific explosion, subsequently gaining insight into the terminal ballistics of the fragments. [Display omitted] •Icosahedral average method can effectively estimate the Cd of irregular particles.•Indicating the mapping between fragment shape, Mach number and Cd.•Improving the estimation accuracy of fragments' trajectory via the surrogate model.