APPLICATION OF ARTIFICIAL NEURAL NETWORK FOR PREDICTING COMPRESSIVE STRENGTH OF GEOPOLYMER CONCRETE

摘要

In the present study, four numbers of two-layered ANN models were used to predict the compressive strength of geopolymer concrete. The differences in the models were the number of neurons in the hidden layer and the method of termination of training. Eight input parameters were employed which included curing time, Na_2SiO_3 / NaOH ratio, alkaline liquid/fly ash ratio, plasticizer, rest period, water content, NaOH concentration and curing temperature. There was one output parameter that was a compressive strength. A total number of 51 datasets were utilized among which 35 were used for training, eight were applied for validation, and the remaining eight were employed for testing. It was detected that as the number of neurons in the hidden layer increased, there was an improvement in the result and errors decreased. Also, the termination state affects the results. ANN Ⅲ and ANN Ⅳ were found to exhibit the best results.