Dept. of Mechanical Engineering Graduate Seminar

PRESENTATION: The performance and sensitivity of polymer-bonded explosives (PBXs) are strongly influenced by their complex heterogeneous microstructures. Microstructural characteristics such as binder fraction, porosity, and particle morphology govern mesoscale processes including hotspot formation and shock propagation. However, systematic computational investigation of these effects is constrained by the limited availability of experimentally resolved microstructures and the difficulty of generating datasets that span the relevant microstructural design space.

This work presents a framework for generating realistic synthetic energetic material microstructures using generative machine learning and algorithmic microstructure construction methods, enabling high-fidelity simulation studies of energetic material behavior. Multiple deep learning approaches are employed, including CycleGAN, conditional diffusion, and fine-tuning foundational models. In addition, new particle packing techniques that relax traditional geometric constraints enable the construction of stochastic microstructure configurations and allow the generation of representative microstructures exceeding domain sizes typically available in public data.

Together, these approaches provide a generalizable methodology for synthesizing microstructures with tunable characteristics and for systematically exploring how microstructural variability influences material response, creating new opportunities for simulation-driven investigation and design of heterogeneous energetic materials.

PRESENTER: Irene Fang is a Ph.D. candidate in Mechanical Engineering at the University of Iowa. Her research focuses on using machine learning to generate synthetic microstructures of energetic materials and investigate how microstructural features influence damage and shock sensitivity. She previously conducted research at Lawrence Livermore National Laboratory and Los Alamos National Laboratory.