Deflection estimation of edge supported reinforced concrete slabs using artificial neural network

Abstract

Excessive deflections of reinforced concrete slabs can cause severe serviceability problems. In recent years, realistic estimation of slab deflection under service loads has become more important due to the increasing use of high strength materials and due to the ultimate limit state design, which generally result in thinner members. Deflection calculations of slabs using nonlinear Finite Element (FE) analysis are complicated and time consuming due to the fact that it is affected by cracking, creep and shrinkage etc. The main objective of this work is to develop an easy method of deflection estimation using Artificial Neural Network (ANN), which will be useful in estimat;'1g deflection of edge supported slabs. Hossain (1999) developed a nonlinear FE module, which was incorporated in the finite element software FE-77 (Hitchings, 1994) to model the effect of cracking using ACVBranson's equation. This has been used in the current work after proper validation against experimental results. A large number of FE analysis has been carried out on slabs with varying support condition, span, aspect ratio, loads, material properties etc. and a database has been created for training the ANN prediction tool. The purpose-built ANN program has been trained using the database until the amount of error become very small. Once the network has been trained, the prediction tool was validated against experimental and numerical results from previous FE analysis. Use of the trained ANN software to estimate short-and long term deflections has been demonstrated with example. Now, a designer will be able to estimate deflection of an edge-supported slab for any span, aspect ratio, support condition etc. easily by using the developed ANN prediction tool.