Designing the Composition of High-Strength Self-Compacting Concrete Using Neural Network

Purpose: To extend the possibility of interpreting the results of a limited number of physical experiments while selecting the components for the creation of concrete mixes and concrete on its basis with the required output properties. Methods: Using a neural network in the analytical environment Loginom, a model with six input variables representing introduced special additives and seven output variables representing the quality parameters of the concrete mix and concrete has been built. Results: In the process of training a three-layer network model has been formed, with the help of which the predicted output characteristics of concrete mix and concrete have been obtained. Experimental data and neural model prediction for each output variable have been compared in a clear graphical form. Practical significance: Prediction of the required characteristics of the synthesized material by means of the neural model response to various combinations of input data allows to carry out an optimal number of physical experiments. It is possible to apply the proposed modeling method for various multicomponent materials.