Pioneering detonation pressure in energetic materials

Area of Science:

• Computational Chemistry
• Materials Science
• Machine Learning Applications

Background:

• Predicting detonation performance of energetic materials is challenging due to complex physicochemical and molecular interactions.
• Accurate prediction is crucial for efficient design and screening of novel energetic compounds.

Purpose of the Study:

• To develop a comprehensive data-driven machine learning (ML) framework for predicting detonation pressure.
• To systematically evaluate and compare various ML algorithms for this prediction task.
• To identify key descriptors governing detonation pressure and provide interpretable insights.

Main Methods:

• Curated a dataset of 222 energetic compounds and divided it into training, validation, and testing sets.
• Employed 18 input parameters including oxygen balance, density, HOMO level, elemental composition, and detonation products.
• Evaluated 14 ML models (e.g., Gaussian Process, Linear Regression, CNN, KNN) and incorporated Monte Carlo outlier detection and hyperparameter optimization.

Main Results:

• Gaussian Process, Linear Regression, and Convolutional Neural Network models achieved high prediction accuracy (R² > 0.960) with low deviation (< 2.7%).
• Pearson correlation analysis identified strong relationships between detonation pressure and oxygen balance, N₂ production, density, and solid carbon.
• SHAP-based explainable ML identified key governing factors and provided physically interpretable insights into structure-property relationships.

Conclusions:

• The developed ML framework provides a reliable and efficient tool for predicting detonation pressure.
• The study enhances understanding of structure-property relationships in energetic materials.
• This approach accelerates the rational design and discovery of high-performance energetic materials.