Exploring New Energetic Materials Based on Materials Genome Engineering Approach

Siwei Song, Yi Wang^*, Fang Chen, Mi Yan, Qinghua Zhang*

Institute of Chemical Materials, China Academy of Engineering Physics, Mianyang

ABSTRACT: As a class of metastable energy storage materials, energetic materials can instantaneously provide huge power, which are widely used in military and civilian fields such as space exploration, mining, construction, demolition, drilling oil wells, etc. However, finding new high-performance energetic materials is always a significant challenge due to the intrinsic energy-safety contradiction and relatively low efficiency of traditional research paradigm of trial and error. In recent years, machine learning technology has demonstrated the great promise in the development of new materials. So far, to the best of our knowledge, applying machine learning for the discovery of new energetic materials has not yet been reported.

   In this study, we first extracted and quantified the “genome’’ feature of insensitive energetic material from the views of elemental constitution, molecular structure and crystal packing. Then, we trained four models covering density, detonation velocity (D_v), detonation pressure (P) and decomposition temperature (T_d) through kernel ridge regression (KRR) algorithm to guarantee the basic properties’ prediction of energetic materials on the molecular level. At the crystal level, a convolutional neural network (CNN) using one-hot encoding as input was used to train a classification model for identifying specific graphite-like layered crystal structure. After integrating these machine learning models with a high-throughput molecule generation module, a whole high-throughput virtual screening (HTVS) system was finally established for high-efficiency design, prediction and screening of promising energetic materials. With the aid of materials genome engineering approach and this self-established HTVS system, we finally targeted two new insensitive energetic materials with specific graphite-like layered crystal structures from a huge chemical space. Our work demonstrates the potential of materials genome engineering approach in accelerated discovery of advanced energetic materials and opens an avenue to develop new promising energetic materials such as melt-cast carrier explosive and primary explosive in future.

Keywords: Energetic materials; materials genome engineering; machine learning; high-throughput virtual screening (HTVS)

* Corresponding author: qinghuazhang@caep.cn