Prediction of engineering properties of fly ash-based geopolymer using artificial neural networks

Fly ash-based geopolymer has been studied extensively in recent years due to its comparable properties to Portland cement and its environmental benefits. However, the uncertainty and complexity of design parameters, such as the SiO2/Na2O mole ratio in alkaline solution, the alkaline solution concentration in liquid phase, and the liquid-to-fly ash mass ratio (L/F), have made it very difficult to create a systematic approach for geopolymer mix design. These mix design parameters, along with fly ash properties and curing conditions (temperature and time), significantly influence key properties of the material, such as setting time and compressive strength. In this study, an artificial neural network (ANN) was used to develop models for predicting the key properties of high-calcium fly ash-based geopolymer according to its mix design parameters. The correlations between experimental measurements and ANN model predictions of setting time, compressive strength, and heat of geopolymerization were established based on the results of tests on 36, 273, and 72 geopolymer mixes, respectively. The results show that the correlations between the experimental measurements and ANN model predictions of the properties studied are all strong. ANN modeling was found to be a suitable computing method to analyze the effects of design parameters on geopolymer properties and showed thatL/Fexhibited the greatest effect on setting time, alkaline solution concentration had the greatest influence on compressive strength, and a mole ratio larger than 1.5 significantly impacted heat at the geopolymerization peak. The developed ANN models can be used as guidance for mix design of high-calcium fly ash geopolymer in engineering applications.

Automated summary

Fly ash-based geopolymer has been extensively studied due to its comparable properties to Portland cement and environmental benefits. However, the uncertainty and complexity of design parameters make it difficult to create a systematic approach. Artificial neural network models can predict key properties of high-calcium fly ash-based geopolymer, providing guidance for engineering applications.