High performance concrete strength prediction using artificial neural network

Abstract

An artificial neural network of the fuzzy-ARTMAP type was applied for predicting strength properties of high-performance concrete (HPC) mixes. Composition of HPC was assumed simplified, as a mixture of six components (cement, silica, superplasticizer, water, fine aggregate, and coarse aggregate). The 28-day compressive strength value was considered as the only aim of the prediction. Data on about 340 mixes were taken from various recent publications. The system was trained based on 200 training pairs chosen randomly from the data set, and then tested using remaining 140 examples. A significant enough correlation between the actual strength values and the values predicted by the neural network was observed. Obtained results suggest that the problem of concrete properties prediction can be effectively modeled in a neural system, in spite of data complexity, incompleteness, and incoherence. It is demonstrated that the approach can be used in multicriterial search for optimal concrete mixes.