Models for Predictions of Mechanical Properties of Low-Density Self-compacting Concrete Prepared from Mineral Admixtures and Pumice Stone

This study applies the principle of artificial neural networks for modelling the mechanical characteristics of a lightweight self-compacting concrete containing pumice and mineral admixtures. Models for predicting compressive strength, split tensile strength and flexural strengths were developed based on several measures of the materials as obtained from the experimental stage. The input parameters for the model were contents of cement, ground granulated blast furnace slag (GGBS), rice husk ash (RHA), fine aggregates, coarse aggregates, pumice stone, water, super-plasticizers and micro-silica. Three output parameters, including compressive strength, tensile strength and flexural strength were considered. The data were trained, tested and validated using the feedforward backpropagation algorithm. The study established the best model for the tested concrete, based on the minimal error criteria, as 9 (input), 12 (hidden layer) and 3 (output layer). This model is expected to serve as a useful tool for concrete designers and constructors.