Transforming results from model to prototype of concrete gravity dams using neural networks

摘要

A new method using neural networks for the transformation of results from dam models to prototypes has been proposed and validated through application to Koyna and Pine-Flat Dams, which have also been investigated by other researchers. The neural network has been called the neurotransformer. The common method for building a suitable experimental model for a dam to be tested on a shaking table is linear dimensional analysis or simply linear scaling (LS). However, because LS is theoretically applicable to linear systems, it generally provides imprecise results of transformation for extreme loading when the model or the prototype experiences noticeable nonlinearity. In this paper, it is shown through numerical simulation of the dynamic behaviour of Koyna Dam and its 1/50 model under strong earthquakes, which cause nonlinear behavior in both the dam and its model, that transformation by neural networks is considerably more precise than LS. To show the method can also be applied to other dams, the same procedure was successfully applied to Pine-Flat Dam; again, the neurotransformer outperformed the LS.