Heterogeneous energetic damage simulator (HEDS): Deep Learning for Synthetic PBX Microstructures with Controlled Damage and Porosity

Damage within polymer-bonded explosive (PBX) microstructures profoundly affects safety and performance. We develop HEDS (Heterogeneous Energetic Damage Simulator) framework which incorporates deep learning techniques using conditional diffusion for generating and manipulating PBX microstructures, which can generate stochastic microstructures with highly controlled damage morphologies and porosity. By systematically regulating binder fraction, crack shape, and crack fraction, we achieve microstructures that span a wide porosity range and capture critical morphological features of damaged PBXs. Coupling these synthetic microstructures with Direct Numerical Simulations (DNS) enables rapid in silico assessments of shock sensitivity and energy release in response to different damage configurations. Through iterative sampling and fine-tuning, we replicate various microstructural scenarios, ultimately providing a robust toolset for studying the influence of porosity and damage on PBX performance.