Comparative evaluation of MFP and RBF neural networks’ ability for instant estimation of r/c buildings’ seismic damage level

Abstract

The problem of the seismic damage prediction of reinforced concrete (r/c) buildings utilizing two types of Artificial Neural Networks (ANN) is investigated in the present paper. More specifically, the problem is formulated and solved in terms of the Function Approximation problem as well as of the Pattern Recognition problem using Multilayer Feedforward Perceptron Networks (MFP) and Radial-Basis Function (RBF) networks. The required training data-sets are created by means of Nonlinear Time History Analyses of 90 r/c buildings which are subjected to 65 earthquakes. The selected buildings differ in total height, in structural system, in structural eccentricity as well as the existence or not of masonry infills. The seismic damage index which is used to describe the seismic damage state is the Maximum Interstorey Drift Ratio. The influence of the parameters which are used for the configuration and the training of MFP and RBF networks on the reliability of their predictions is also investigated. The generalization ability of the best configured ANNs is examined by means of two categories of seismic scenarios. The most significant conclusion that turned out is that the trained ANNs can reliably and rapidly classify the r/c buildings into pre-defined damage classes provided they are appropriately configured.