Investigation of the Various Structure Parameters for Predicting Impact Sensitivity of Energetic Molecules via Artificial Neural Network



A generalized scheme is introduced for predicting impact sensitivity of any explosives by using artificial neural networks. Experimental values for the impact sensitivity for 291 compounds containing C, H, N and O have been used for training and testing sets. The input descriptors include aromatic character, heteroaromatic character, the number of N[BOND]NO2 bonds and the number of α-hydrogen atoms as well as the number of carbon, hydrogen, nitrogen, and oxygen divided by molecular weight. The reliability of the proposed model was assessed by comparing the results against measured values as well as five models of complicated quantum mechanical computed values of 14 CHNO explosives from a variety of chemical structures. The model gives root mean squares errors of 41 cm and 56 cm for training and test sets, respectively, of the H50 quantity.