An automated computationally efficient two-stage procedure for service load analysis of RC flexural members considering concrete cracking

Abstract

An automated computationally efficient two-stage procedure has been proposed for service load analysis of reinforced concrete (RC) flexural members considering concrete cracking and tension stiffening. The proposed procedure yields cracked lengths, redistributed bending moments and inelastic deflections. The computation of final state, including cracked lengths and interpolation coefficients for tension stiffening, is automated by the initialization of second stage from the results of first stage. The procedure combines the analytical–numerical procedure developed by authors and the neural networks methodology. The cracked lengths and corresponding interpolation coefficients are rapidly estimated in the first stage using the closed-form expressions which are obtained from the trained neural networks. Eight separate neural networks are trained for the estimation of final cracked lengths and interpolation coefficients at in-span locations and supports. The use of estimated cracked lengths and interpolation coefficients of the first stage, in the beginning of second stage, significantly reduces number of iterations in the second stage. The deflections and bending moments obtained from the two-stage procedure are compared with those from the analytical–numerical procedure for a number of beams. The analytical–numerical procedure requires around six analyses to yield results with sufficient accuracy for design purpose (within 2–3%); whereas, in the two-stage procedure, only two analyses, one in each stage, are required to yield results with similar accuracy. The developed two-stage procedure requires a significantly small computational effort as compared to similarly accurate methods available in literature.