Prediction of restraint in second cast sections of concrete culverts using artificial neural networks

Abstract

Estimation of restraint is very important for accurately predicting the risk of early thermal and shrinkage cracking in concrete structures. The stress in young concrete is affected by changes in its dimensions during hydration and the restraint imposed by adjoining structures. In concrete culverts, the restraints from existing structures acting upon the first and second casting sections to be cast are different, causing them to exhibit different early cracking behaviour. This work presents a new method for predicting restraint in complex concrete structures using artificial neural networks (ANNs). Finite element calculations were performed to predict restraint in 108 slabs, 324 walls and 972 roofs from second sections of concrete culverts, and the results obtained were used to train and validate ANN models. The ANN models were then used to study the effects of varying selected parameters (the thickness and width of the roof and slab, the thickness and height of the walls, and the length of the culvert section) on the predicted restraint. Mathematical expressions for predicting restraint values in slabs, walls and roofs were derived based on the ANN models’ output and implemented in an Excel spreadsheet that provides a simple way of predicting restraint in practical applications. Restraint values predicted in this way agree well with the results of finite-element calculations.