Neural networks for predicting properties of concretes with admixtures

Abstract

Backpropagation neural networks were used to predict the strength and slump of ready mixed concrete and high strength concrete, in which chemical admixtures and/or mineral additives were used. Although various data transforms were tried, it was found that models based on raw data gave the best results. When non-dimensional ratios were used, arranging the ratios such that their changes resulted in corresponding changes in the output (e.g. increases in ratios to cause increases in output values) improved network performance. The neural network models also performed better than the multiple regression ones, especially in reducing the scatter of predictions. Problems associated with models trained on non-dimensional ratios were uncovered when sensitivity analyses were carried out. A rational approach was used for carrying out sensitivity analyses on these mix design problems by constraining the sum of input values. These analyses, using the raw data based model, showed that the modelling had picked up not only the fundamental domain rules governing concrete strength, but also some well-known second order effects.