Machine Learning-Driven Discovery of High-Performance Solid Propellants

Solid propellants are the primary sources of propulsion energy for rockets. Their energy characteristics determine the payload capacity and range of rockets. To design higher-performance solid propellants, it is often necessary to perform a high-precision quantification of the enthalpy of formation (EOF) for the component materials before conducting thermodynamic calculations. However, this process is inefficient and time-consuming. Herein, a machine learning (ML) framework integrating ML with genetic algorithms (GAs) was introduced to accelerate the design of solid propellants, allowing for accurate and rapid prediction of energy characteristics of propellants, only with the mass ratio and chemical formulation of each component as input. Leveraging the proposed framework, three propellant formulations with the ratios very close to the best-reported ratios were identified by using GAs, thereby validating the reliability of this framework for designing solid propellants. By applying high-throughput screening within this framework, seven promising energetic compounds (ECs) were identified from over 1000 candidates, with the potential to increase the specific impulse ( I sp ) to 278 s and to enhance the rocket range by up to 45%. This study highlights the practical application of ML in predicting energy characteristics of solid propellants and establishes methodologies for advancing their intelligent design.