부산대학교 도서관

The purpose of this study is to propose a 40MPa concrete blending design model that applies the neural network theory to minimize the effort wasted in trial and error. A mixed design model was applied to each of the 180 data using fly ash, blast furnace slag, and rice husk powder. And in the neural network model, the optimized connection weight was obtained by repeatedly applying it to the back-propagation algorithm. The completed mixed design model was demonstrated by analyzing and comparing the predicted values in the mixing design model with those measured in the actual compressive strength test. The factors in the neural network theory affect the compressive strength as the input values and collect the mixed design data through references related to the compressive strength characteristics. To validate the model, a compressive strength test was conducted based on the prediction of the mix design factor, comparing the target compressive strength and the test value in the mix design. According to the results of the mix design verification experiment, the lowest error rate in 40MPa class strength cementless concrete composed of fly ash and rice husk powder was 4.7%, and in 40MPa class strength cementless concrete composed of fly ash, blast furnace slag, and rice husk powder, the lowest error rate was 4.3%. In addition, if the error rate decreases according to the test conditions and environment, a more accurate value could be obtained through the mixed design model.