Energetic materials and machine learning. Requirements and application

Machine Learning is a rapidly developing field, where data availability and data structure are key features that determine the quality and performance of models, which can be used to predict and optimize a variety of variables. In the case of energetic materials, these predictable variables can be material properties, for example. In the last few years, different methods and applications have been published about the use of machine learning to predict different properties of energetic materials, such as enthalpy of formation, sensitivity, or specific impulse. In this work, the use case of machine learning to predict the enthalpy of formation based on the ICT Thermodynamics database is shown. Important aspects such as data fusion from various sources, dealing with substantial amounts of data and anomaly detection are discussed from the scientist’s (i.e., the chemist’s) view. Special attention is paid to the interface between the world of computer sciences and chemistry and how these two can and must interact with each other to obtain good and reliable results. Typical problems and basic knowledge to assess literature in this field will be presented.