Modeling meso-scale energy localization in shocked HMX, Part I: machine- learned surrogate model for effect of loading and void size

TL;DR

This paper develops a machine learning surrogate model to predict hotspot ignition and growth in pressed HMX, using meso-scale simulation data to improve multi-scale modeling of energetic materials.

Contribution

It introduces a Bayesian Kriging-based surrogate model trained on meso-scale simulation data to predict hotspot criticality and ignition behavior in HMX.

Findings

Surrogate model accurately predicts ignition and growth rates.

Criticality envelope depends on shock pressure and void size.

Classifies hotspot criticality into plastic collapse and jetting regimes.

Abstract

This work presents the procedure for constructing a machine learned surrogate model for hotspot ignition and growth rates in pressed HMX materials. A Bayesian Kriging algorithm is used to assimilate input data obtained from high-resolution meso-scale simulations. The surrogates are built by generating a sparse set of training data using reactive meso-scale simulations of void collapse by varying loading conditions and void sizes. Insights into the physics of void collapse and ignition and growth of hotspots are obtained. The criticality envelope for hotspots is obtained as the function {\Sigma}_cr=f(P_s,D_void ) where P_s is the imposed shock pressure and D_void is the void size. Criticality of hotspots is classified into the plastic collapse and hydrodynamic jetting regimes. The information obtained from the surrogate models for hotspot ignition and growth rates and the criticality envelope can be utilized in meso-informed Ignition and Growth (MES-IG) models to perform multi-scale simulations of pressed HMX materials.