Use of metamodels in uncertainty analysis of concrete arch dams

Uncertainty analysis of large coupled systems, such as those formed by concrete arch dams, their reservoirs and surrounding foundations, is typically addressed with Monte Carlo simulations in literature. Although it is a robust method, the high computational effort required makes it impractical for everyday engineering. This research investigated the applicability of the so called metamodels, specifically of the Response Surface Method, as a more efficient alternative. The goal was to develop polynomial surrogate models that could estimate diverse structural responses with acceptable precision, allowing for uncertainty quantification at a much cheaper computational cost. For that end, two case studies located in Italian territory were taken into consideration, namely the Lumiei dam (cupola) and the Pertusillo dam (arch-gravity). Starting from original documentation, linear and non linear numerical models of both of them were developed. Cantilever joints were introduced in the non linear models, located in their real positions, and modeled with a hard-contact formulation. Material and rock foundation were considered as linear elastic materials. Dynamic interaction between the main structure and the reservoir was accounted for with the aid of acoustic finite elements. Their use and precision were evaluated in detail, comparing the method with classic approaches such as those of Westergaard and Zangar. Five Design of experiment techniques were adopted, and the resulting metamodels precision/ computational cost ratio analyzed. Finally, it was concluded that metamodels are a valid an efficient alternative for the task proposed, and furthermore can be used in other applied sciences as well.