Geopolymer concrete (GPC) is an environmentally friendly and sustainable concrete produced through the geopolymerization process, involving the combination of alumino silicate materials with alkaline solutions. It has unique advantage of being carbon-negative which helps reduce carbon dioxide levels in the atmosphere. One challenge with making GPC is the significant variation in compressive strength that arises from differences in constituent proportions, specimen ages, andcuring conditions. There is a need to optimize the GPC constituents to achieve the desired compressive strength. To address this issue, a comprehensive dataset comprising 1242 data points from existing literature on fly ash-based GPC (FA-based GPC) was compiled for the purpose of optimizing the constituent proportions of GPC using statistical analysis. Nine independent variables were selected to develop the predictive models for compressive strength using linear (LRA), multilinear (MRA), and non-linear (NRA) regression analyses. The models were assessed using scatter plots between experimental and predicted responses and adopting the performance criteria such as the statistical significance of the model unknowns, R2 values, root mean square error (RMSE), mean absolute error (MAE), and scatter index (SI). LRA revealed that the compressive strength of GPC is not dependent on any single independent variable. Therefore regression analysis with multiple independent variables was conducted. From MRA, ±15 % error for training dataset and (15–20) % error for validating dataset was estimated. The R2, RMSE, MAE, and SI values were found to be 0.8086, 6.11, 5.35, and 0.21. NRA predicted the compressive strength more efficiently because it provides the error ±10 % for both training and validating datasets with R2, RMSE, MAE, SI values of 0.8321, 5.46, 4.77, and 0.19. Further, curing temperature, specimen age, Na2SiO3/NaOH ratio, I/b ratio, and aggregate content were found as the most significant independent variables. The time and cost-effective design mix for FA based GPC can be prepared using the regression models presented in the current work.
Read full abstract