PREDICTION OF COMPRESSIVE STRENGTH OF CONCRETE CONTAINING POZZOLANIC MATERIALS BY APPLYING NEURAL NETWORKS

The most interesting aim of the study is to assess and compare the dependability of using the multiple linear regressions (MLR) model and the artificial neural networks (ANN) model to predict the concrete compressive strength using metakaolin (MK) and silica fume (SF) admixtures materials. A proposed prediction model of artificial neural network (ANN) for concrete compressive strength. That proposed model is trained, validated and tested using the available test data of 132 concretes with various mixture proportions that were collected from different technical literature. Next the prediction of concrete compressive strength is conducted on those models. The collected data organized in a form of eight input variables (parameters) which includes concrete specimen age, water, fine aggregate, metakaolin, cement, coarse aggregate, silica fume, and superplasticizer. Relating to these input parameters in the ANN model, the concrete compressive strength containing MK and SF, are predicted. The results from the training, validation, and testing stages from making use of the ANN model showed that neural networks (NN) have strong potential use for the prediction of concrete compressive strength that contain materials such as MK and SF. The correlation coefficient for the ANN model in the training, validation, and test stages that achieved are equal to 0.99661, 0.99093, and 0.98577, respectively. Whereas the correlation coefficient for the the MLR model was 0.794. The results suggest that the prediction using ANN model is more accurate than when using the MLR model.