Compressive strength prediction for glass aggregates incorporated concrete, using neural network and reviews

Abstract

Production of concrete by use of conventional materials is unsustainable due to high demand. Henceforth, there is need to upscale the use of alternative materials, including those from waste streams, in concrete. This research aims at developing a suitable predictive model of concrete having partial or 100% glass aggregates. 50 datasets reviewed from 9 sources were adopted and artificial neural network (ANN) models were developed in GNU Octave. The trial models had 7 input variables and 1 output variable (compressive strength) and 1 hidden layer. The selected model, having 24 nodes in the hidden layer and 90.000 iterations, indicated overall root mean square error (RMSE), mean absolute errors (MAE), mean absolute percentage errors (MAPE) and absolute factor of variance (R2) of 2.679 MPa, 1.422 MPa, 6.951% and 0.996 respectively. The glass fine aggregates between >40% and 50% indicated just over 11% average strengths from the controls. Generally, RMSE, MAE, MAPE and R2 values showed that the selected model had a good accuracy level and good generalization, particularly considering that the datasets were not from the same experimental program. The study recommends research and utilization of glass fine aggregates up to 50% by weight, with consideration to other influencing factors and also research in cost-effective and environmentally friendly additive and assessment on waste glass aggregates incorporated concrete.