Unsupervised learning-based multiscale model of thermochemistry in 1,3,5-trinitro-1,3,5-triazine (RDX)

The response of high-energy-density materials to thermal or mechanical insults involves complex physical and chemical processes coupling over disparate temporal and spatial scales. Reactive molecular dynamics (MD) simulations can contribute to a predictive understanding of their behavior under dynamical loading; however, computational intensity limits MD to the nanoscale, making it difficult to simulate eventual shock to detonation transition. Therefore, we developed a multiscale model for RDX, where a continuum description is informed by reactive and non-reactive MD simulations to characterize chemical reactions and thermal transport. Coarse graining is done using dimensionality reduction via unsupervised learning to establish a two-step reduced order chemical decomposition model. From homogeneous isothermal and adiabatic simulations, we extract chemical kinetics and heat of reaction parameters. The multiscale approach is validated from hot spot calculations where we find excellent agreement in the hotspot temperature fields. Finally, the continuum model is used to quantify the effects of uncertainties for various materials’ parameters and assess the criticality of hotspots involving scales beyond the reach of atomistic simulations. Approved for unlimited release LA-UR-20-29003.