Upscaling Reaction Rates From Mesoscale Simulations of HMX-TNT Explosive Mixtures Using Machine Learning Methods

Reaction rates are one the most complicated and uncertain components of continuum reactive flow models for heterogeneous explosives. They are influenced not only by chemistry but also by thermomechanical properties, microstructure, and loading conditions. In this study, we perform mesoscale simulations with the LLNL hydrocode, ALE3D, to investigate shock-induced reactions in HMX-TNT mixtures. The thermochemical code, Cheetah, provides temperature-dependent heat capacity, kinetics, and equation-of-state (EOS) properties for the energetic constituents. We employ an isotropic HMX material model has been updated from previous studies to incorporate pressure-dependent shear modulus as well as pressure-dependent melt curve and pressure/temperature-dependent shear viscosity based on molecular dynamics studies. Reaction rates are computed from mesoscale simulations for a range of HMX-TNT mixture ratios, applied pressures, porosity, and pore sizes. Rates are used to inform an existing reactive flow model and to train a surrogate rate model using machine learning methods. The performance of the reactive flow and surrogate models are evaluated and discussed. These studies are important for developing formulation-sensitive reactive flow models for explosive mixtures.