Answers are grounded in this paper and cite their sources.
Nonlinear seismic time-history analysis of concrete dams requires repeated high-fidelity finite element simulations, limiting its use in large-scale probabilistic assessment. This study proposes an Energy-Adaptive Kernel Gaussian Process (EAKGP) surrogate for rapid and uncertainty-aware prediction of full-field and full-time displacement responses over a representative dam section. Equivalent nodal seismic forces and displacement fields are spatially reduced by proper orthogonal decomposition, while reduced-force trajectories and ground-motion intensity measures characterize the seismic excitation. An acceleration-derived energy-adaptive kernel adjusts the temporal correlation scale according to the evolving strong-motion state, and predictive uncertainty is reconstructed in the physical field. Parent ground motions are separated between model development and evaluation. On the independent test record, EAKGP achieves L 2 accuracies of 94.59% and 98.67%, with RMSEs of 6.153 mm and 5.047 mm for the horizontal and vertical displacement fields, respectively. For a fully held-out ground motion, the RMSEs are 9.957 mm and 4.869 mm. Approximately 1700-node displacement histories are reconstructed in about 15 s, providing a 480–720-fold speedup over comparable two-dimensional nonlinear FEM analysis. The predicted fields also support subsequent strain-exceedance-based probabilistic assessment.
No takes yet. Share an insight, caveat, or question.
Hao et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: