Estimating the joint shear strength of exterior beam–column joints using artificial neural networks via experimental results

Beam–column joints play an important role in resisting lateral loads induced by earthquakes. Previous post-earthquake reports have indicated that the failure of beam–column joints was a primary cause of the collapse of numerous existing reinforced concrete structures, even when designed in accordance with seismic building code requirements. This study aims to design artificial neural networks (ANNs) model to evaluate the joint shear strength capacity of exterior beam–column joint and, in addition, assess the adequacy of existing code provisions utilized in estimating the shear strength of exterior beam–column joints. A database comprising 127 tested specimens from the existing literature encompassing a wide range of variables was constructed and utilized for developing, training, and validating the ANN-based model and assess the building code provisions. The performance of the designed ANN models and building code provisions was evaluated using metrics such as Pearson correlation coefficient (R), mean value (μ) of the ratio of predicted joint shear strength to experimental, standard deviation (SD), coefficient of variation (COV), and the mean absolute errors (MAE). The result demonstrates that ANN designed model shows high performance with accurate and reliable means of predicting joint shear strength of beam–column, offering potential improvements over existing national building code provisions. In addition, the present study drives a weight factor for input variables making the use of designed model easily accessible to anyone interested in simulating the procedure using a spreadsheet.