Mechanics guided data-driven model for seismic shear strength of exterior beam-column joints

Abstract

This study presents an enhanced predictive model for the seismic shear strength of exterior beam-column joints (BCJs). Initially, the principles of strut-and-tie mechanism and variable selection procedures were first utilized to identify the most influential parameters. Subsequently, an evolutionary algorithm, specifically multigene genetic programming (MGGP), was utilized to search for the near-optimal predictive model. The dataset used to develop, train, and test the proposed model was compiled from previously published tests, focusing specifically on cyclically loaded exterior BCJs that encountered shear and flexure -shear failures. The prediction performance of the developed model was assessed through various statistical measures, and then compared with that of other existing models. Additionally, sensitivity analyses were also performed to identify the influence and importance of each design parameter. The results demonstrated that the methodology employed in this study yielded an elegant model that adheres to the underlying mechanics and provides higher prediction accuracy compared to existing models. Furthermore, the sensitivity analyses showed that BCJ shear strength positively correlates with concrete compressive strength, beam reinforcement, joint transverse reinforcement, column intermediate vertical reinforcement, and axial load ratio, while it negatively correlates with the joint aspect ratio. Among these design parameters, beam reinforcement has the greatest influence on the model response, followed by concrete compressive strength. Conversely, column intermediate vertical reinforcement and axial load ratio have the least impact on the model response. The notable prediction capabilities and robustness demonstrated by the developed model render it an efficient design tool with promising potentials for adoption by practicing engineers and for consideration in design guidelines.