NEURAL NETWORKS FOR DEFLECTIONS IN CONTINUOUS COMPOSITE BEAMS CONSIDERING CONCRETE CRACKING

Abstract

Maximum deflection in a beam is a design criteria and occurs generally at or close tothe mid-span. A methodology has been developed for continuous composite beams to predict theinelastic mid-span deflections, d i (considering the cracking of concrete) from the elastic mid-spandeflections, d e (neglecting the cracking of concrete). Nine significant structural parameters havebeen identified that govern the change in mid-span deflections. Six neural networks have beenpresented to cover the entire practical range of the beams. The proposed neural networks havebeen validated for a number of beams with different number of spans and the errors are small forpractical purposes. The methodology enables rapid estimation of inelastic deflections incontinuous composite beams and requires a computational effort that is a fraction of that requiredfor the conventional iterative or incremental analysis. The methodology can easily be extended forlarge composite building frames where a huge savings in computational effort would result.