USE OF ARTIFICIAL NEURAL NETWORKS IN SEISMIC ANALYSIS OF CONCRETE GRAVITY DAM

Abstract

In the study Artificial Neural Network is used for seismic analysis of concrete gravity dam. An Artificial Neural Networks (ANN) model "neuro-modeler" is built and then verified for quick estimation of the parameters such as maximum crest displacement, maximum heel stresses and maximum toe stresses developed on a concrete gravity dam section due to seismic excitation. Typical geometries of dam which is generally used in the construction of dam are considered. Incremental Dynamic Analysis (IDA) is performed in Abaqus software on each geometries to collect the adequate amount of data for training of the neuro-modeler so that the precision of the method could be assessed. Neuro-modeler is designed in two stages. In first stage ANN-I is designed without taking the geometrical configurations in the input and in ANN-Il model is designed taking the consideration of geometrical configurations in the input. Training of the neuro-modeler is done using generalized statistical tool "MATLAB". The neuro-modeler is trained for four geometries of dam considered on the information gained from the IDA analysis of geometries of dam under the different earthquake. After successful training, the precision and generalization capabilities of the neuro-modeler is tested and it is expected to be able to provide reliable, convenient and precise results about the response of one of the typical geometry considered under any given earthquake. Neuro-modelers have been successful in providing precise results in this numerical simulation of the nonlinear behavior of dams.