Failure Load Prediction of Castellated Beams Using Artificial Neural Networks

This paper explores the use of artificial neural networks in predicting the failure load of castellated beams. 47 experimental data collected from the literature cover the simply supported beams with various modes of failure, under the action of either central single load, uniformly distributed load or two-point loads acting symmetrically with respect to the center line of the span. The data are arranged in a format such that 8 input parameters cover the geometrical and loading properties of castellated beams and the corresponding output is the ultimate failure load. A back-propagation artificial neural network is developed using Neuro-shell predictor software, and used to predict the ultimate load capacity of castellated beams. The main benefit in using neural network approach is that the network is built directly from the experimental or theoretical data using the self-organizing capabilities of the neural network. Results are compared with available methods in the literature such the Blodgett’s Method and the BS Code. It is found that the average ratio of actual to predicted failure loads of castellated was 0.99 for neural network, 2.2 for Blodgett’s Method, and 1.33 for BS Code. It is clear that neural network provides an efficient alternative method in predicting the failure load of castellated beams.