In this work, we report the first attempt of the development of a universal method for impact sensitivity prediction extrapolated to trending energetic materials – nitrogen-rich compounds. A simple two-step methodology including both manual and machine calculation usually takes a few minutes to provide this paramount-importance property with an acceptable accuracy.

Abstract

In this paper, we report the first example of impact sensitivity prediction based on the genetic function approximation (GFA) as a regression method. The prediction is applicable for a wide variety of chemical families, which include nitro compounds, peroxides, nitrogen-rich salts, heterocycles, etc. Within this work, we have obtained 7 empirical models (with 27–32 basis functions), which all provide 0.80≤R ²≤0.83 and 7.2 J≤RMSE≤7.8 J (for 450 training set compounds) and 0.64≤R ²≤0.70 and 11.2 J≤RMSE≤12.4 J (for 170 test set compounds). The models were developed using Friedman Lack-of-Fit as a scoring function, which allows avoiding an overfitting. All the models have simple descriptors as basis functions and include linear splines. Furthermore, the applied descriptors do not require expensive calculation procedures, namely, non-empirical quantum-chemical calculations, complex iterative procedures, real space electron density analysis, etc. Most descriptors are based on structural and topological analysis and a part of them require very cheap semi-empirical PM6 calculations. The prediction takes a few minutes as an average, and most of the time is for the structure preparation and manual calculation of the descriptor “Increment”, which is based on our recent incremental theory.