Neural Network for Creep and Shrinkage Deflections in Reinforced Concrete Frames

A recently developed accurate procedure termed as the consistent procedure (CP) for evaluation of creep and shrinkage behavior in reinforced concrete (RC) frames is elaborate and requires large computational effort. An approximate procedure (AP) that has been available and widely used is simple and requires much less computational effort but can be erroneous. The feasibility of using the neural network model to simulate the inelastic deflections of CP from the results of AP for a class of RC frames is investigated. This model would enable rapid estimation of inelastic deflections of CP and would be useful at the planning stage. For this purpose, a ratio η of inelastic deflections of CP, to corresponding deflections of AP, designated as inelastic deflection ratio is defined as the output parameter. The sensitivity of η with the probable structural parameters in the practical range of values is studied and governing input parameters identified. The training is carried out for a practical range of the governing structural parameters. Trained network is validated for a number of example buildings.