A COMPARATIVE STUDY ON DIFFERENT PROPAGATIONS FOR DEVELOPMENT OF ARTIFICIAL NEURAL NETWORK MODEL TO STUDY THE BEHAVIOUR OF CST (COMPOSITE STEEL TUBES) UNDER COMPRESSION

Mohammed Jawahar Soufain, Syed Jawad Sujavandi, Umar Ahad Shariff, N.S Kumar

Abstract: In this research work, the present study is emphasized on the behavior of Composite Steel Tubes filled with concrete under monotonic loading. The prime factors considered to get ultimate axial load and corresponding axial shortening under axial compression are cross-sectional area, wall thickness of the steel tube, strength of infilled concrete. Also this study focuses on the development of Artificial Neural Network architecture for the prediction of ultimate load carrying capacity of CST infilled with different grades of concrete. Artificial Neural Network for different propagations- Feed forward backdrop, Cascade backdrop, Elman backdrop, Time delay, Layer recurrent are developed. The developed ANN models were verified with the experimental results conducted on Composite Steel Tubes. The compressive strength of composite steel tubes was modeled as a function of Eight variables: Diameter, Thickness, Length, Grade on concrete, Yield strength of steel, Epoxy, L/D ratio, D/T ratio. The effects of each parameter on networks were studied for different propagations. The cascade forward back propagation performed better than other propagations. The error in the Cascade backdrop propagation almost subsidized to 0.932594% and is approximately providing results coinciding with Experimental values.

Keywords: Artificial Neural Network, Composite Steel Tubes, Feed Forward Back Propagation, Cascade Back Propagation, Elman Propagation, Time Delay Propagation, Layer Recurrent Propagation.

DOI: https://doi.org/10.15623/ijret.2017.0602008