Results in Engineering (Mar 2025)
Predicting penetration depth in ultra-high-performance concrete targets under ballistic impact: An interpretable machine learning approach augmented by deep generative adversarial network
Abstract
In recent decades, numerous explosions and ballistic attacks have caused significant global loss of life and property. Ultra-high-performance concrete (UHPC) minimizes blast and impact damage to structures and can be applied to protective walls and bunkers. Many researchers have proposed methods to estimate projectile penetration depth in concrete, but accurate estimation remains unresolved due to the phenomena's complexity. This paper explores interpretable machine learning (ML) methods to estimate penetration depth in UHPC targets under ballistic impact using 103 data points from the literature. To address the limited experimental data for assessing UHPC impact resistance, a deep generative adversarial network (DGAN) is used for data augmentation. The comparison of real and DGAN-generated data showed the DGAN's effectiveness in replicating the real data's probability distribution. The ML models exhibited excellent accuracy in estimating penetration depth in UHPC targets, with correlation (R) exceeding 0.855 for DGAN data and 0.954 for real data. The extreme gradient boosting (XGBoost) model achieved high accuracy with R and mean absolute error (MAE) values of 0.990 and 4.933, respectively. A comparison with existing empirical models indicated the superiority of the ML-based models. Interpretability techniques revealed that projectile features like impact energy, velocity, diameter, and mass, along with the target's compressive strength and fiber addition, significantly influence penetration depth. The study underscores the potential of explainable ML augmented by DGAN as a valuable tool for predicting penetration depth in UHPC targets under ballistic impact, particularly with limited data.
