Determination of compressive strength of perlite-containing slag-based geopolymers and its prediction using artificial neural network and regression-based methods

Abstract

Listen

Translate article

• The parameters affecting the CS of perlite-containing slag-based geopolymers were investigated. • The regression-based CRA, MARS, and TreeNet methods as well as ANN methods were used. • RMSE, MAE, SI and NS performance statistics were used to evaluate the CS estimation capabilities of the methods. • The highest performances in regression-based models were obtained from the MARS method. • The ANN method showed higher estimation performance than the regression-based methods. This study has two main objectives: (i) to investigate the parameters affecting the compressive strength (CS) of perlite-containing slag-based geopolymers and (ii) to predict the CS values obtained from experimental studies. In this regard, 540 cubic geopolymer samples incorporating different raw perlite powder (RPP) replacement ratios, different sodium hydroxide (NaOH) molarity, different curing time, and different curing temperatures for a total of 180 mixture groups were produced and their CS results were experimentally determined. Then conventional regression analysis (CRA), multivariate adaptive regression splines (MARS), and TreeNet methods, as well as artificial neural network (ANN) methods, were used to predict the CS results of geopolymers using this experimentally obtained data set. Root mean square error (RMSE), mean absolute error (MAE), scatter index (SI) and Nash-Sutcliffe (NS) performance statistics were used to evaluate the CS prediction capabilities of the methods. As a result, it was determined that the optimum molarity, curing time, and curing temperature were 14 M, 24 h, and 110 ℃, respectively and 48 h of heat curing did not have a significant effect on increasing the CS of the geopolymers. The highest performances in regression-based models were obtained from the MARS method. However, the ANN method showed higher prediction performance than the regression-based methods. Considering the RMSE values, it was seen that the ANN method made improvements by 24.7, 2.1, and 13.7 %, respectively, compared to the MARS method for training, validation, and test sets.