Modeling the Capacity of Pin-Ended Slender Reinforced Concrete Columns Using Neural Networks

Abstract

This study demonstrates the feasibility of using multilayer feedforward neural networks to model the complicated nonlinear relationship between the various input parameters associated with reinforced concrete columns and the actual ultimate capacity of the column. The neural network models were constructed directly from a fairly comprehensive set of experimental results and were found to be tolerant of certain levels of errors in the original testing results. Comparison with the original testing data and theoretical model showed that the ultimate capacity of reinforced concrete columns predicted by the neural network models is reasonably accurate. Parametric analysis indicates that the neural network model has reasonably captured the behavior of reinforced concrete columns. Numerical studies are conducted to investigate modeling issues such as different data scaling schemes and dimensionless representation schemes. Nonlinear transformation of the output values resulted in an overall improvement in the generalization capabilities of the neural network model. Preliminary studies using a limited data set of 54 test results on high strength concrete columns also showed promising results. The neural network model can be useful in checking routine designs because it provides instantaneous results once it is properly trained and tested.