Backbone Curve Development for Shear-Dominated Nonductile Circular Bridge Piers: An Explainable Machine Learning Approach

Nonlinear analysis procedures are commonly employed to evaluate the force–deformation behavior of reinforced concrete (RC) bridge piers, with backbone curves frequently used to approximate the nonlinear response and identify performance limit states. Developing simplified backbone curves representing the critical damage states of a shear-dominated, nonductile RC bridge pier is challenging due to their complex behavior involving the interaction of flexure, shear, and bar slip displacements. This study presents a novel machine learning (ML)–based approach, specifically an artificial neural network (ANN), to predict seismic performance and generate backbone curves for nonductile circular RC bridge piers. A set of prototype nonductile bridge piers is modeled using OpenSees and validated against experimental results. A data set comprising of 1,000 circular RC piers, developed using Monte Carlo simulation, is employed to train, test, and validate different ML models through (1) hyperparameter optimization using grid search algorithm, (2) direct comparison with test results, (3) a fivefold cross-validation process, and (4) benchmarking against recommended modeling parameters prescribed in different guidelines. The developed ANN model turns out to be the best ML model with the highest coefficient of determination (R2) value of 0.998 on the test set among all algorithms investigated. ANN results are interpreted using Shapley additive explanation to identify the most influential parameters defining nonlinear behavior compared with existing mechanics-based knowledge. A user-friendly graphical user interface is also developed to facilitate rapid and accurate seismic performance assessments of nonductile RC bridge piers.