Machine learning-based shear strength prediction of exterior RC beam-column joints

Abstract

Column-beam joints are of critical importance in cast-in-situ reinforced concrete structures. Under the effect of earthquake, no damage should occur in these regions that would disrupt the integrity of the load-carrying system of a structure. In particular, shear damage in this area is very critical. Exterior joints that are not fully confined by the beams are forced under the effect of more shear force during an earthquake. Therefore, it is very important that the shear capacity of these areas is sufficient. Although there are empirical approaches for the estimating of the shear capacity of these areas in the seismic codes, the development of software that can predict with high accuracy using more innovative methods is also very important in terms of structural engineering. With this motivation, this study has developed a new method that uses machine learning to predict the shear strength in exterior column-beam joints of RC structures. Firstly, 270 shear damaged exterior joint experimental data sets were collected from the existing literature. Then, the shear strength capacity was determined according to 14 different parameters related to cross-section and material properties using different machine learning (ML) methods such as Multiple Linear Regression (MLR), Decision Tree (DT), Random Forest (RF), Support Vector Regression (SVR), Bootstrap Aggregating (Bagging), Gradient Boosting Regression (GBR), Adaptive Boosting (AdaBoost) and Extreme Gradient Boosting (XGBoost). Among the algorithms used, the GBR model provided the most successful predictions, according to achieved R-squared (R2) value and MAE and RMSE error metrics calculating %95.45, 42.10 and 62.98, respectively. Furthermore, ML models have more predictive power than the conventional building and seismic code approaches given for the shear capacity of column-beam joints. © 2024, The Author(s), under exclusive licence to Springer Nature Switzerland AG.