Machine learning-integrated nonlinear modeling of ductile reinforced concrete shear walls

Reinforced concrete (RC) structural walls are critical structural components in lateral load resisting systems as they provide significant lateral stiffness and ductility under seismic loads. However, predicting the cyclic responses of structural walls using finite element (FE) models introduces a number of challenges, especially related to predicting local responses, such as rotations and strains. To address these issues, this study introduces an integrated framework that combines validated OpenSees material models with machine learning-based parameter estimation to enhance the physical transparency and generalizability of RC wall simulations. The computational FE platform, Open System for Earthquake Engineering Simulation (OpenSees), is used to simulate the cyclic behavior of RC structural walls and accounts for strain penetration, shear deformation effects, bar rupture, buckling and low-cycle fatigue, as well as numerical challenges such as strain localization issues in walls with softening behavior. To improve the prediction accuracy, ML algorithms are employed to predict calibration parameters (peak strain of confined concrete) based on geometrical and material properties. Using datasets from 140 structural wall test specimens, model and test results are compared using various ML techniques and Simple Weighted Ensemble Models to demonstrate the validity and efficiency of the proposed approach. Findings highlight the potential of ML-driven FE modeling to enhance seismic performance assessment and structural design optimization for RC structural walls.