A methodological investigation of physics-informed neural networks for multi-mechanism coupled damage effectiveness prediction
Yalong Wang, Xianming Shi, Mifen Yang, Penghua Liu, C. H. Hu, Chao Hsi Chang · Advances in Theoretical and Applied Mechanics · 2025
Damage effectiveness prediction is a pivotal issue in weapon system design and operational application. However, the coupled multi-physics effects in complex environments and the scarcity of experimental data pose significant challenges to traditional predictive methodologies. To address these limitations, this paper proposes a damage effectiveness prediction method based on Physics-Informed Neural Networks (PINN). This approach integrates key physical mechanisms, including kinetic energy penetration, blast shockwave propagation, fragment lethality, and ricochet effects, and quantifies them as residual loss functions under mathematical constraints to construct a deep learning framework that incorporates prior physical knowledge. Experimental results demonstrate that the proposed model achieves a 63.3% reduction in Root Mean Square Error (RMSE) and a 42.3% decrease in Mean Absolute Error (MAE) on the test set. Furthermore, under conditions of 30%-70% training data, the model exhibits superior generalization capabilities compared to traditional Multilayer Perceptron (MLP) models. Physics-consistency validation confirms that the prediction results strictly adhere to the law of conservation of energy. Nevertheless, the model’s capability to capture abrupt changes in physical parameters requires further enhancement, indicating a direction for future research. This study provides a novel, high-precision, and highly interpretable technological pathway for damage effectiveness prediction, possessing both theoretical innovation and engineering application value.