Design and Modelling the Compressive Strength of High-Performance Concrete with Silica Fume: A Soft Computing Approach

Abstract

Soft computing methods were used in this research to design and model the compressive strength of high-performance concrete (HPC) with silica fume. Box-Behnken design-based response surface methodology (RSM) was used to develop 29 HPC mixes with a target compressive strength of 80 ± 10 MPa. Cement (450–500 kg/m3), aggregates (1500–1700 kg/m3), silica fume (SF) (20–45% weight of cement) and water-binder (w/b) ratio of (0.24–0.32) were provided as input factors while the compressive strength at 7 and 28 days were analysed as responses. Datasets for the artificial neural network (ANN) prediction were generated from 87 experimental observations from the compressive strength test. Performance indicators such as p-value, coefficient of determination (R2), and mean square error (MSE) were used to assess the models. Results demonstrated that RSM worked relatively well in projecting compressive strength with model p-values < 0.05 and R2 values of 0.913 and 0.892 for compressive strength at 7 and 28 days, respectively. In addition, RSM performed better in detecting the synergistic effects of the variables on the responses. On the other hand, ANN best generalised the relationship between independent and dependent variables considering the low MSE of 12.32 and 14.60, and high R2 values of 0.912 and 0.946 for compressive strength at 7 and 28 days, respectively. Model equations were developed to predict the compressive strength of silica-based HPC after 7 and 28 days. It is considered that adopting components from both approaches could help the design process for developing consistent mixes of HPC with supplementary cementitious materials (SCMs).