.png){kind=icon}
.png){kind=icon}
Return
Predicting probabilistic flexural strength of corroded reinforced concrete columns based on physics-informed GPR model
.png){kind=icon}
PRE
AI
P
Q
W
B
DOI：10.1016/j.compstruc.2025.108074.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
A physics-informed Gaussian process regression (PI-GPR) model for predicting the probabilistic flexural strength of corroded reinforced concrete (RC) columns was developed based on the multi-level embedding strategy. According to the moment equilibrium conditions, a physical model representing the flexural mechanism of corroded RC columns was established first. Subsequently, the multi-level embedding strategy was adopted to develop the PI-GPR model by constraining the mean, kernel, and loss functions hierarchically. Meanwhile, an adaptive trade-off between physical priors and data features was achieved by adding the dynamic coefficients that are continuously updated as training data changes. Then the PI-GPR model was applied to predict the probabilistic flexural strength of corroded RC columns. Finally, the effectiveness of the PI-GPR model was validated by comparing it with traditional prediction methods. Analysis results show that the PI-GPR model, which integrates the physical constraints with data-driven learning, demonstrates excellent performance and robust uncertainty quantification. Compared with traditional GPR, the PI-GPR ensures high physical consistency of predictions, which not only achieves an average improvement of 20% in confidence interval coverage when the training set proportion was only 30%, but also reduces the mean absolute error and root mean square error for edge samples by 23% and 21%, respectively.
Journal
C
IF:
4.8
Papers:
6.2K
Citations:
1.7W
Organization
Cited Papers