MCS-based response surface metamodels and optimal design of experiments for gravity dams

Deterministic seismic analysis of dam-foundation-water does not usually reflect the true behavior of the system due to presence of different epistemic uncertainties. On the other hand, the crude Monte Carlo simulation (CMCS) of such a system is computationally expensive and often practically impossible. This paper investigates the potential application of so-called design of experiments (DOE) in order to develop appropriate metamodels and present an explicit expression of dam response. More than ten DOE techniques are compared and contrasted. Accuracy of metamodels is evaluated using about 120,000 finite element simulations from CMCS. Furthermore, the application of several sampling reduction techniques (i.e. Latin Hypercube, Sobol, and Halton) is evaluated. Next, the impact of polynomial degree is studied for an actual gravity dam with five random variables. Results are generalised for 12 different dam classes where the displacements, stresses and failure modes are quantified. Last but not least, a metamodel is proposed for seismic uncertainty of gravity dam in which the dam response is estimated using different features of the input ground motions.