Plastic hinge length of RC shear walls: Practical approximation via machine learning and probabilistic assessment

Abstract

This study presents practical solutions for estimating the plastic hinge length of RC shear walls using machine learning (ML)-based approaches and incorporates probabilistic assessment to investigate the robustness of the developed solutions. A comprehensive dataset, comprising 234 experimental tests and 487 numerical simulation cases, was used to develop and refine the models, accounting for diverse wall configurations and material properties. Feature selection identified key influencing factors such as wall dimensions, reinforcement ratios, and material strengths. Two solutions were developed: a simpler, more straightforward model and a more complex model with enhanced performance. Both models outperform existing empirical methods in terms of accuracy and reliability. The SHapley Additive exPlanations (SHAP) analysis reveals that wall length and shear span are the most influential features. Using the Monte Carlo Simulation (MSC) method, the reliability of the existing and proposed solutions was analyzed to evaluate the robustness of their predictions. The results demonstrated that the predictions of the developed closed-form solutions are more reliable and robust than those of existing solutions in estimating plastic hinge length. While the second solution offers higher accuracy, the first one provides a simpler and more practical solution without significant compromise in performance.

Keywords: Machine learning models, Monte Carlo simulations, Plastic hinge length, RC shear walls, Reliability analysis