Prediction of moment improvement in UHPC strengthened damaged RC beams based on data augmented machine learning

Strengthening of damaged reinforced concrete (RC) structures with ultra high performance concrete (UHPC) can increase their load carrying capacity and durability. However, there are limited studies that forecast the moment improvement ( M u ) in strengthening damaged RC beams. The aim of this study is to develop a reliable model that can precisely predict M u . Initially, the researchers gathered 173 datasets from experimental studies. Due to the limited amount of data available, kernel density estimation (KDE) was employed to expand the data. Subsequently, six machine learning algorithms were developed to predict the M u . In addition, a new prediction model was constructed by considering the failure modes of the strengthened beams. Finally, Shapley Additive Explanations were employed to conduct an evaluation of model explainability. The results show that KDE can improve the robustness and accuracy of the model. Extreme gradient boosting performed best in predicting M u and considering the failure mode could improve the accuracy of the model. The height of the RC beam, the reinforcement ratio of the UHPC, and the width of the RC beam have a large and proportional effect on M u . This study can provide guidance for the engineering design of UHPC strengthened damaged RC beams.