PARAMETRIC MODELING OF RECYCLED BRICK AGGREGATE CONCRETE USING NEURAL NETWORK AND REGRESSION

Abstract

Concrete made from recycled materials is usually produced through a challenging design process. This implies that there is difficulty in producing low-cost concrete with suitable mechanical properties. Hence, this study was conducted to utilize parametric modeling for the optimum mix design of recycled brick aggregate concrete (RBAC). The artificial neural network (ANN) and multiple linear regression (MLR) were utilized to generate the model. The result of the back-propagation algorithm in the ANN model showed that Bayesian regularization with nine (9) neurons is the best mathematical model, with regression (R) and mean squared error (MSE) values of 0.86499 and 0.007996756, respectively. The correlation coefficient for the Multiple Linear Regression (MLR) model, on the other hand, was 0.6508. The results clearly showed that the prediction using a neural network model is more accurate than using a multiple linear regression model. A parametric study was done in Bayesian regularization with nine (9) neurons in the model to assess the effect of each independent variable on the compressive strength of concrete by varying the amount of one independent variable and setting the other independent variables to a constant value. Based on the result of the parametric study, the recommended amounts for each material are as follows: cement = 500 kg; water-cement ratio = 0.4; recycled aggregates = 20–40%; and natural aggregates = 60–80%.