A machine-guided approach to connect porosity characteristics to corner turning performance in PBX 9502

Machine Learning (ML) methods have been used to connect microstructural details in energetic materials (EMs) to macroscale performance parameters. One such EM, PBX 9502, has historically shown a wide range of corner turning behavior, likely due to microstructural manufacturing variations. The 10 nm – 10 µm void size distribution (VSD) influences EM behavior; here we investigate 'corner turning'. In this work, our ML algorithm learns the correlation between VSD details and the corresponding corner-turning performance. Training data is generated by simulating experiments using a physically-informed reactive flow model running on HPC resources. The study observes the robustness of the approach and, as there is limited experimental data, we heuristically evaluate the predictions. We conclude that our algorithm did robustly learn the correlations, finding that porosity characteristics in the 100-200 nm range have the strongest effect. This agrees with our qualitative knowledge and we discuss the validity of that prediction. The ML algorithm provides a powerful tool to assess and predict the effects of manufacturing variability on the corner turning behavior of PBX 9502, through which it can guide manufacturing and formulation. The method can be extended to other EMs.