Computer vision and statistical ML to analyze PBX microstructure, initiation threshold, and self-similarity in explosive hotspots

Detonation properties of heterogeneous solid explosives are known to depend strongly on the material microstructure. Variations in synthesis/manufacturing details, as well as exposure to aging conditions under different environmental stressors can lead to changes in both microstructure and detonation performance. To quantify such changes and build useful correlation models one requires developing appropriate summary features from microstructure characterization data. To this end we will discuss two different applications of computer vision to analyze microstructure of a plastic-bonded HE (PBX) system: (1) scalar-field topology to summarize 3D X-ray CT data, and (2) bag-of-visual-words (BOVW) to summarize surface-profilometry data. Correlation between summarized structural features and initiation threshold is explored via Generalized Linear 'probit' Model within a Bayesian formalism. We will also discuss a recently developed Principal-Component based 'image distance' metric that is used to quantify stochasticity in molecular-dynamics-generated property fields, and how we utilize it to explore scale-invariance in explosive hotspot structures characterized by pore-collapse features and shear bands. Such analysis can provide an insightful link between microstructural defects within an HE system and its grain-scale shock response.