.png){kind=icon}
.png){kind=icon}
Return
Inverse prediction and data augmentation of concrete failure behavior under high impact loads based on physics-consistent machine learning
.png){kind=icon}
PRE
AI
X
C
N
Y
DOI：10.1016/j.ijsolstr.2026.113951.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The inverse prediction of concrete failure behavior under high impact loads is a critical challenge in impact dynamics, where traditional empirical, theoretical, and numerical methods face significant limitations. Machine learning provides a promising alternative due to its strong nonlinear fitting capability. However, the scarcity of concrete failure data hinders the robustness and generalization of these models. To address this challenge, a data augmentation method named Uncertainty-Quantified Physics-Informed Generative Adversarial Network (UQ-PIGAN) and a data-driven inverse prediction model are developed. Traditional theoretical models are integrated into the data augmentation model as constraints. The inherent uncertainty of these physical models is quantified to prevent theoretical errors from misleading the training of neural network. Experimental results demonstrate that UQ-PIGAN efficiently generates synthetic data with high physical consistency. The augmented datasets significantly improve the performance and generalization of machine learning models. Furthermore, the inverse prediction model trained on these datasets exhibits excellent accuracy and physical reliability, providing an effective way of disaster reconstruction.
Keywords:
Concrete
Impact loads
Neural networks
Data augmentation
Inverse prediction
Journal
IF:
3.8
Papers:
1.1W
Citations:
3.1W
Organization
Cited Papers