Transfer-Learning for Rapid Predictions of CHNO Energetic Material Sensitivity

The sensitivity of an energetic material (EM) is strongly influenced by its molecular and microstructure. For a particular CHNO EM, the shock sensitivity is challenging to predict as each reacting differently under various loading conditions and microstructural characteristics. In this study, we introduce a transfer-learning approach for swiftly and accurately predicting the sensitivity of a general CHNO energetic material. Our approach uses knowledge of sensitivity knowledge of a small number of CHNO species to transfer learn the behavior of the side class of such materials and make sensitivity predictions with sparse data. To validate our approach, we conducted reactive void collapse simulations on one specific CHNO materials, namely HMX, under varying shock loading conditions. We then applied our transfer-learning model to rapidly adapt the knowledge gained from HMX to predict the sensitivity of other species, RDX, TATB, PETN, and TNT. Our findings showcase a high level of agreement between results obtained from the transfer learning algorithm and direct numerical simulations, despite training the transfer learning algorithm on a relatively small dataset. In order to measure macro-scale sensitivity, we also created Pop-plots for the different materials and showed the validity of the approach by comparing with experimental data. The present transfer-learning model can be used for rapid prototyping to improve the efficacy and safety of CHNO energetic materials.