New Insights Into Drop Height of Energetic Materials Via Data Driven Models

ABSTRACT Accurately predicting drop height (H50) in energetic materials is a challenging task due to the intricate relationships between their chemical and physical properties. In this work, a comprehensive data‐driven methodology is proposed, utilizing advanced machine learning (ML) models to estimate drop height. The study leverages a dataset consisting of 155 training samples, 33 validation samples, and 34 test samples, incorporating 18 distinct input features such as oxygen balance, heat of explosion, product gases (CO 2 , N 2 ), aromaticity, and density, among other critical structural attributes. Pearson correlation analysis reveals significant relationships, with drop height showing negative correlations with heat of explosion ( r = −0.59) and CO 2 product gas ( r = −0.51), while it is positively correlated with product solid C ( r = 0.47) and HOMO level ( r = 0.35). We evaluated a range of machine learning approaches, from standard regression algorithms to advanced ensemble techniques and deep neural networks. Gradient Boosting and Random Forest delivered exceptional results, posting test R 2 scores of 0.9998 each, followed closely by CatBoost at 0.9396, all with minimal mean squared error (MSE) and mean relative deviation (MRD%). Meanwhile, algorithms such as Lasso Regression, K‐Nearest Neighbors, and Ridge Regression fell short in performance. SHAP analysis revealed that heat of explosion, CO 2 gas product, oxygen balance, and product solid C were the most influential parameters impacting the prediction of drop height. The innovation of this research lies in its combination of extensive chemical and physical descriptors with powerful ML algorithms, resulting in high prediction accuracy and enhanced interpretability. This approach provides valuable insights into the structure–property connections in energetic materials, offering a reliable tool for the rational design and evaluation of high‐performance materials in various applications.

Ҳали таржима қилинмаган