همایش بین المللی هوش مصنوعی

صفحه اصلی / دومین همایش بین المللی هوش مصنوعی

Cross-domain prediction of ultimate strength in corroded reinforced concrete columns using domain adaptation techniques

نویسندگان :

Melina Sadeghi¹ Pooria Poorahad² Mahmoud R. Shiravand³

1- دانشگاه شهید بهشتی 2- دانشگاه شهید بهشتی 3- دانشگاه شهید بهشتی

کلمات کلیدی :

artificial intelligence،machine learning،domain adaptation،corrosion،RC columns

چکیده :

Reinforced concrete (RC) columns in corrosive environments undergo complex nonlinear deterioration, making accurate seismic performance assessment challenging. While machine learning (ML) offers predictive capabilities, conventional models often fail to generalize across datasets from different laboratories due to domain shifts in material properties and loading protocols. This study addresses this limitation by employing two domain adaptation techniques: correlation alignment (CORAL) and domain-adversarial neural networks (DANN). Using a comprehensive database of experimental tests on corroded RC piers, the methods are evaluated across three distinct domain split scenarios: failure mode, axial load ratio, and corrosion ratio. The study focuses on predicting the ultimate strength of the structures. Results indicate that both techniques significantly outperform baseline ML models in cross-domain scenarios, with DANN showing superior robustness. The findings demonstrate that domain adaptation effectively mitigates distribution discrepancies, offering a reliable framework for assessing aging infrastructure even when training and target data diverge.