COMPUTER AIDED DESIGN OF COLUMNS USING ARTIFICIAL NEURAL NETWORKS

Abstract

Artificial neural network tries to mimic the information processing principles of biological neural networks. In recent past, it has attracted wide attention and found a growing number of applications in various areas of engineering. The neural network learns to map an input into a desired output by self adaptation and so the ANN is treated as model free estimators. In the present work this quality of ANN is used. Column is an important element of all the structural engineering systems. Design of column is a trial and error procedure. Design of beam columns is more complicated process. The calculations involved are tedious. To tackle this problem, SP-16 "Design Aids" provide charts to simplify design procedure. In the present work these charts are used to develop training set for the development of an artificial neural network. This type of network is> used directly in a fully computer based design procedure for the design of R.C.0 columns subjected to combined axial compression and bending in the present work. Parametric studies to find out the appropriate ANN architecture and learning rate for the RCC columns of rectangular section having reinforcement (Fe 415) distributed equally on four sides is taken as the objective of the present work. The ANN parameters selected in this study are, training sample size, number of hidden layers, number of hidden nodes and learning rate. In each experiment, first, training was carried out with a given training sample and then the performance of ANN has been evaluated in terms of prediction accuracy on test data. Various neural network architectures have been evaluated to assess their performance on the prediction and generalization capability. From these, an optimal iii neural network model has been identified. Finally, this network is coupled to a fully computer based design procedure for the design of R.C.0 columns subjected to combined axial compression and bending The findings of the present work show that the developed ANN models have immense potential to predict reasonably accurate values of percentage steel for columns subjected to combined axial compression and bending.