High-fidelity neural network model of shock to detonation transition

Thermo-mechano-chemical interactions between shock fronts and defects at the meso-scale play a crucial role in the study of sensitivity during the shock to detonation (STD) transition of energetic materials. In general, the characteristic size of defects, such as cracks and voids, is around 3 – 4 orders of magnitude below the required distances to develop a steady Chapman-Jouet (CJ) state during the run to detonation for a particular input. Thus, explicitly modeling shock-defect interactions and the subsequent thermo-chemical kinetics at the meso-scale is highly inefficient due to the natural discrepancy in the kinetics and length scales of these two phenomena. We couple a finite element simulation with meso-scale resolution with nano-scale molecular dynamics model using a neural network called MISTnet to predict the sensitivity of a system with an explicitly described microstructure. The model describes the STD transition in which the chemical evolution of each grid point is predicted by the subgrid MISTnet model at the shock front, and the trailing kinetics are modeled explicitly using a chemical reaction model.