Scholarlink

<p>Terminal ballistics and penetration mechanics are inherently intricate domains, governed by complex, nonlinear interactions between projectile characteristics, target properties, and impact conditions. Although analytical and numerical methods provide valuable insights, they are often constrained by prohibitive computational expense. In contrast, machine learning models present a computationally efficient alternative to these traditional methodologies. Nevertheless, the practical application of many existing models is hampered by their limited generalizability across a wide range of target materials and thicknesses. To overcome these limitations, this study develops a mechanics-guided machine learning framework that augments classical models and enables the efficient evaluation of ballistic performance under varied target conditions. This framework adopts a bifurcated strategy. First, an XGBoost-based classification model delineates perforation outcomes with 97.95% accuracy. Second, regression models estimate Kinetic Energy Reduction (KER) using two distinct datasets: the complete dataset (<b>ALL-dataset</b>), which includes both perforation and non-perforation cases, and a subset containing only perforation cases (<b>PER-dataset</b>). Models trained on <b>ALL-dataset</b> (<b>ALL-models</b>) demonstrate superior predictive fidelity for lower KER values, while models trained exclusively on the <b>PER-dataset</b> (<b>PER-models</b>) yield greater accuracy for higher KER values. A feature importance analysis identifies target thickness and projectile material as the dominant mechanical factors, an observation that aligns with foundational principles of penetration mechanics. These results demonstrate that the proposed framework achieves high predictive fidelity across the diverse range of material and thickness combinations studied, yielding robust models for the rapid evaluation of new impact scenarios within this domain. This study thus presents a robust and interpretable framework that forges a critical link between data-driven machine learning and the classical mechanical understanding of terminal ballistics.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Mechanics-Guided Machine Learning for Predicting Penetration and Energy Dissipation in Terminal Ballistics

  • 期刊论文

  • 发表日期：

    2026年5月7日

查看全文
  • Yi Zhang,
  • James Ross,
  • Brianna Thompson,
  • Jesse Sherburn,
  • Ashley Abraham,
  • Lisa Kubiak

摘要

Terminal ballistics and penetration mechanics are inherently intricate domains, governed by complex, nonlinear interactions between projectile characteristics, target properties, and impact conditions. Although analytical and numerical methods provide valuable insights, they are often constrained by prohibitive computational expense. In contrast, machine learning models present a computationally efficient alternative to these traditional methodologies. Nevertheless, the practical application of many existing models is hampered by their limited generalizability across a wide range of target materials and thicknesses. To overcome these limitations, this study develops a mechanics-guided machine learning framework that augments classical models and enables the efficient evaluation of ballistic performance under varied target conditions. This framework adopts a bifurcated strategy. First, an XGBoost-based classification model delineates perforation outcomes with 97.95% accuracy. Second, regression models estimate Kinetic Energy Reduction (KER) using two distinct datasets: the complete dataset (ALL-dataset), which includes both perforation and non-perforation cases, and a subset containing only perforation cases (PER-dataset). Models trained on ALL-dataset (ALL-models) demonstrate superior predictive fidelity for lower KER values, while models trained exclusively on the PER-dataset (PER-models) yield greater accuracy for higher KER values. A feature importance analysis identifies target thickness and projectile material as the dominant mechanical factors, an observation that aligns with foundational principles of penetration mechanics. These results demonstrate that the proposed framework achieves high predictive fidelity across the diverse range of material and thickness combinations studied, yielding robust models for the rapid evaluation of new impact scenarios within this domain. This study thus presents a robust and interpretable framework that forges a critical link between data-driven machine learning and the classical mechanical understanding of terminal ballistics.