Optimal Prediction of Shear Properties in Beam-Column Joints Using Machine Learning Approach

The failure of shear-type beam-column joints in reinforced concrete (RC) frames during severe earthquake attacks is a critical concern. Traditional methods for determining joint shear capacity often lack accuracy due to improper consideration of governing parameters, impacting the behaviour of these joints. This study assesses the capabilities of machine learning techniques in predicting joint shear capacity and failure modes for exterior beam-column joints, considering their complex structural behaviour. An artificial neural network (ANN) model is proposed for predicting the shear strength of reinforced exterior beam-column joints. ANN, a component of artificial intelligence that learns from past experiences, is gaining popularity in civil engineering. The ANN model is developed using a dataset comprising material properties, specimen dimensions, and seismic loading conditions from previous experimental investigations. The model considers twelve input parameters to predict shear strength in exterior beam-column joints. Training and testing of the ANN model are conducted using established design codes, empirical formulas, and a specific algorithm. The results demonstrate the superiority of the proposed Shallow Feed Forward Artificial Neural Network (SFF-ANN) compared to previous approaches. The effectiveness of an Artificial Neural Network (ANN) model was quantitatively assessed in this study, with a focus on its performance in comparison to various design codes commonly used in structural engineering. The model was assessed using the coefficient of determination (R2) and achieved R-squared values of 99%, 94%, and 98% during the training, testing, and validation stages, respectively. The study highlights the significance of beam reinforcement as a key element in estimating shear capacity for exterior RC beam-column connections. Although the proposed models exhibit a high degree of precision, future research should focus on developing improved models using expanded datasets and advanced algorithms for enhanced pattern recognition and performance.