.png){kind=icon}
.png){kind=icon}
Return
A sample efficient deep learning framework for bond performance assessment of FRP bar reinforced seawater sea sand concrete
.png){kind=icon}
OA
AI
L
W
R
S
G
F
G
Z
J
DOI：10.1016/j.cscm.2026.e06359.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
To address the small sample challenge in predicting the bond performance between fiber reinforced polymer (FRP) bars and seawater sea sand concrete (SSC), this study develops a deep learning framework integrating transfer learning and data augmentation. A source domain dataset for FRP bar normal concrete bond performance and a target domain dataset for FRP-SSC bond performance were established with consistent input-output definitions. A CNN-Transformer model was constructed to predict both bond strength and failure mode, and three transfer strategies were compared. Results showed that partial freezing achieved the best transfer performance, with a test RMSE of 2.06, MAE of 1.45, R2 of 0.89, classification accuracy of 91.67%, and weighted F1 score of 92.86%. Incorporating TabSyn augmentation further improved performance, reducing RMSE and MAE by 38.19% and 27.59%, respectively, and increasing R2 to 0.9512. The TL2-TabSyn model also achieved an accuracy and weighted F1 score of 94.29%. Data efficiency evaluation further confirmed the robustness of TL2-TabSyn under reduced training data, with an RMSE of 1.658, an MAE of 1.265, and an R2 of 0.938 at the 90% data scale, and a classification accuracy of 88.9% at the 60% data scale. SHAP analysis further identified physically meaningful controlling factors for bond strength and failure mode prediction. These results provide a practical modelling approach for FRP bar reinforced SSC structures with limited experimental data.
Keywords:
Fiber reinforced polymer
Seawater sea sand concrete
Bond strength
Machine learning
Journal
IF:
6.6
Papers:
6.1K
Citations:
2.0W
Organization
Cited Papers