Exploring an accurate machine learning model to quickly estimate stability of diverse energetic materials

Area of Science:

• Materials Science
• Computational Chemistry
• Chemical Engineering

Background:

• High energy and low sensitivity are key goals for new energetic materials (EMs).
• Accurate and rapid methods for evaluating the stability of diverse EMs are currently lacking.
• Bond dissociation energy (BDE) is a critical parameter for assessing EM stability.

Purpose of the Study:

• To develop a highly accurate machine learning (ML) model for predicting the BDE of EMs.
• To establish a reliable and representative dataset for training and validating the ML model.
• To improve the characterization and prediction of EM stability.

Main Methods:

• Construction of a comprehensive BDE dataset using 778 experimental energetic compounds and quantum mechanics calculations.
• Development of a hybrid feature representation coupling local bond information with global structural characteristics.
• Application of pairwise difference regression as a data augmentation technique to enhance dataset diversity and reduce errors.
• Utilizing the XGBoost algorithm for the ML prediction model.

Main Results:

• The XGBoost model achieved a high prediction accuracy for BDE, with an R² of 0.98 and a Mean Absolute Error (MAE) of 8.8 kJ mol⁻¹.
• The hybrid feature representation effectively captured essential characteristics for BDE prediction.
• Pairwise difference regression improved model robustness and data utility.
• The developed model significantly outperformed other competitive ML models.

Conclusions:

• The developed ML model provides a rapid and accurate method for evaluating EM stability through BDE prediction.
• This approach facilitates the design and discovery of novel energetic materials with improved safety and performance characteristics.
• The methodology offers a valuable tool for computational screening of energetic materials.