Predicting the geopolymerization process of fly ash-based geopolymer using deep long short-term memory and machine learning

Abstract

In this study, the geopolymerization process of fly ash-based geopolymer was estimated using the deep long short-term memory (LSTM) and machine learning models. The k-Nearest Neighbors (kNN) and support vector regression (SVR) approaches were used in the machine learning models. The percentages of major chemical components in fly ash, the alkaline solution concentration, the mole ratio, the liquid-to-fly ash mass ratio, and the curing temperature were selected as input variables in the models. The geopolymerization peak time, geopolymerization peak heat, dissolution peak time, and dissolution peak heat obtained from the calorimetric curve of fly ash-based geopolymer were used as the output variables. The deep LSTM, SVR, and kNN models estimated the geopolymerization peak time with 99.55%, 98.83%, and 91.62% accuracy, respectively. The geopolymerization peak heat was estimated by the deep LSTM, SVR, and kNN models with 99.69%, 98.91%, and 88.36% accuracy, respectively. The deep LSTM, SVR, and kNN models predicted the dissolution peak time with 99.49%, 99.43%, and 92.86% accuracy, respectively. The dissolution peak heat was estimated using the deep LSTM, SVR, and kNN models with 99.67%, 99.38%, and 90.60% accuracy, respectively. This study found that the deep LSTM model can be estimated the geopolymerization process of fly ash-based geopolymer with higher accuracy than the SVR and kNN models.