Machine learning innovations for reliable gurney energy estimation in energetic materials

Accurately predicting the Gurney energy (E_G) of energetic materials is crucial for industrial applications, yet it traditionally requires costly, time-intensive empirical testing and complex thermochemical modeling. This study aims to overcome these limitations by developing a robust, high-throughput machine learning framework to predict E_Gdirectly from intrinsic molecular and thermodynamic descriptors. Following rigorous data standardization, multiple machine learning algorithms were optimized and evaluated to capture complex structure-property relationships. Quantitative analysis revealed that the artificial neural network (ANN) significantly outperformed baseline models, yielding exceptional predictive accuracy (R² = 0.995 and MRD <1.5%). Feature importance analysis demonstrated that produced nitrogen gas (N₂), oxygen balance, and solid carbon yield (C) are the most dominant predictors, strongly aligning with fundamental detonation physics. Ultimately, this data-driven approach provides a highly accurate and efficient alternative to traditional measurements, offering profound practical significance for the rapid in silico screening and targeted formulation design of energetic materials.