Using an Artificial Neural Network to Predict Mix Compositions of Steel Fiber-Reinforced Concrete

Abstract

An artificial neural network (ANN) has a wide application field for mathematical problems. Specifically, an ANN is successfully applied to problems that are difficult to solve or do not have any information on their operating techniques. In this article, an ANN was applied to predict the concrete mix composition for steel fiber-reinforced concrete (SFRC). Thus, an ANN model was developed and trained with data collected from literature. These data have SFRC mix compositions, workability measurements of fresh SFRC, compressive strength of SFRCs, and additional information that affects concrete quality. Additionally, the ANN included steel fiber volume fraction in the SFRC and steel fiber properties. With the goal of determining the concrete mix composition, which is cement dosage, amount of water, coarse aggregate content, fine aggregate content, and chemical admixture, an ANN model was developed. The inputs for the ANN were consistency class of SFRC, compressive strength of SFRC, maximum size of aggregate, steel fiber volume fraction, steel fiber length, and diameter. At the end of the study, a feed forward ANN model with six inputs and five outputs was successfully trained and used to produce the correct responses to testing data. Designing SFRC requires more trial mixtures to obtain the desired quality than does conventional concrete. In conclusion, artificial neural networks have a strong potential for predicting concrete mix composition for SFRC such that without trial mixes and loss of time, an SFRC design is possible with the desired workability and mechanical properties.