An Ensemble-Learning approach to connect porosity characteristics to corner turning performance in PBX 9502

Machine Learning (ML) methods have become popular in the field of High Explosives (HEs) due to the capability of producing reliable datasets using numerical simulations. Previously, πSURF was used to generate Run to Detonation (RtD) data from theoretical microstructures of PBX 9502. This microstructure-sensitivity linkage was then robustly learned using the Sen-Rai Ensemble Learning (EL) method. However, the RtD metric covers the low-pressure regime; for the high-pressure regime, PBX 9502 historically shows a wide range of sensitivity, as measured by the Enhanced COrner Turning (ECOT) test. This variation is likely due to subtle microstructural variations that arise during manufacturing. In this work, we hypothesize that the Sen-Rai EL method can robustly learn the relationship between theoretical microstructure and corner turning behavior. Further, we hypothesize that the algorithm can distill the most important microstructural parameters in determining corner turning behavior. Our results show that the Sen-Rai method did learn the relationship and was able to predict corner turning behavior; however, some nuanced details were missed. The algorithm found that three microstructural parameters had nearly equal importance in determining corner turning behavior. Comparison with the previous RtD study reveals that significantly more data was required to form a robustly learned relationship for the ECOT study. This is likely because a single microstructural parameter dominated RtD behavior.