Residual capacity prediction of post-impact precast concrete bridge piers with grouted splice sleeve connectors under sequential vehicle impact and seismic loading using an explainable boosting model

Precast concrete (PC) bridge piers with grouted sleeve splice (GSS) connectors are increasingly popular because they accelerate construction while providing effective seismic resistance. However, in seismic regions with heavy traffic, a pier may be struck by a vehicle and subsequently subjected to an earthquake before repairs, creating a cumulative vulnerability that cannot be captured by evaluating each hazard in isolation. This issue is particularly critical for lightweight vehicle or low-velocity impacts, where external damage may appear negligible, yet internal damage, such as bond deterioration within the GSS or local cracking in the grout, can still develop and silently undermine future load-carrying performance. To date, the sequential impact-seismic behavior of PC piers with GSS connections has not been systematically investigated, and no engineer-oriented tool exists for rapid post-collision residual capacity estimation. To address these gaps, this study develops LS-DYNA finite element models of three pier configurations: cast-in-place (CIP), GSS with couplers in the column (GSS-C), and GSS with couplers in the footing (GSS-F), subjected to a sequential impact-seismic protocol with three vehicle types (0.9-, 2-, and 8-ton). A 318-case dataset is generated via hybrid one-factor-at-a-time (OFAT) sampling and Latin Hypercube Sampling (LHS) and used to train XGBoost models achieving R 2 > 0.99. SHAP-based interpretation is employed, and a Tkinter-based GUI translates the model into a rapid post-impact assessment tool. In contrast to seismic-only conditions, under sequential loading GSS-C exhibits lower residual capacity than GSS-F, and the axial load ratio has a nonlinear influence that plateaus near 0.3. This research delivers a validated machine learning (ML) -based predictor interface supporting rapid post-event decisions when detailed simulations are infeasible or internal GSS damage is not visually observable.