Quantifying modeling uncertainties in seismic analysis of dams: Insights from an international benchmark study

Abstract

Advances in nonlinear dynamic analysis of dams have not completely resolved concerns over modeling confidence and analysis accuracy. Verification and validation offer accuracy assessment, but uncertainties persist during performance evaluation due to both epistemic (modeling) and aleatory (parametric) sources. Epistemic uncertainties arise from simplifications and modeling techniques. This paper addresses epistemic uncertainties in a gravity dam seismic analysis using data from the International Comnission on Large Dams (ICOLD) benchmark study. While the benchmark formulation included the finite element model of the dam, mechanical material properties, and dynamic loads, participants retained the flexibility to opt for best-practice modeling assumptions, simplifications, and other specifics. Notable response variability emerged, particularly in crack profiles and damage predictions. This study examines sources of variability, quantifying modeling uncertainty for the benchmark problem. More specifically, the modeling variability is quantified using the logarithmic standard deviation, also known as dispersion. This metric enables its incorporation into other seismic risk assessment and fragility studies. Under relatively low-intensity motion (peak ground acceleration [PGA] of 0.18 g in this case), modeling dispersion of 0.45, 0.30, 0.32, and 0.30 were calculated for the maximum dynamic crest displacement, maximum hydrodynamic pressure at the heel, heel and crest maximum acceleration, respectively. Additionally, the dispersion of the failure PGA was determined to be 0.7. Findings underscore the need for systematic seismic response modeling in dam engineering to enhance prediction accuracy. A better understanding of the sources and magnitudes of modeling uncertainties can help improve the reliability of dam seismic analysis and contribute to the development of more effective risk assessment and mitigation strategies.