Reverse design of high-detonation-velocity organic energetic compounds based on an accurate BPNN with wide applicability

Abstract

In this work, a new effective development strategy for designing new energetic compounds with higher and higher detonation velocity (D) was proposed based on the machine learning (ML) technique. First of all, a Back Propagation Neural Network (BPNN) model used for predicting D of the five most common kinds of energetic compounds (C–NO₂, N–NO₂, O–NO₂, long nitrogen-chain, and cage) was constructed and trained successfully, by using an optimal combination of nine descriptors (ρ, nO/V_m, nN/V_m, nC/V_m, nH/V_m, V_m, OB, F, and M) which can be easily obtained. The prediction error for various energetic compounds of this BPNN model is as low as 2%, showing its high accuracy and wide application range. Then, a new concept for designing new energetic compounds with high D was extracted, that is, improving the ρ, nO/V_m and nN/V_m by structural modification. Finally, new organic energetic compounds with comparable D (8.1–9.6 km s⁻¹) to three famous energetic compounds CL-20, HMX and RDX can be easily and rapidly designed from several reported compounds with mediocre D, based on the new design concept and BPNN model investigated in this work. In this study not only is a new accurate ML model developed for predicting D of various energetic compounds, but also a new insight is provided for developing new energetic compounds with higher and higher energy.