Prediction of compressive strength ultra-high steel fiber reinforced concrete (UHSFRC) using artificial neural networks (ANNs)

Abstract

For several years, experts have explored machine learning (ML) models in civil engineering to identify many concrete industry-related issues, such as getting the optimized mixture. This paper investigates a wide range of mixtures to perform ultra-high steel fiber reinforced concrete (UHSFRC) compressive strength (CS) by ML. The most important component of UHSFRC uses various ingredients incorporated with the mixture, i.e., ordinary Portland cement (OPC), silica fume (SF), fly ash (FA), aggregate (AG), water/ binder (W/B), aspect ratio (AR), water (W), superplasticizer (SP), steel fiber (Sf), fiber length (Fl), fiber diameter (Fd), and curing age (CA). In this study, artificial neural networks (ANNs) are applied to forecast the CS of UHSFRC. Moreover, the fiber properties and mixture properties are considered in this study at different CA. ANN models are created using the MATLAB R2017b program. 762 experimental data points with 12 input variables are created from the literature to generate the ANN models. The database is separated into three parts. 70% is used for training, 15% for validation, and 15% for testing. Different hidden layers and neurons also consider finding the model’s best accuracy. MSE, RMSE, R, and R2 are figured out for the UHSFRC mixture to verify the ANN model in different conditions. Sensitivity analysis is also carried out to see how each parameter affects the ANNs model. Inspired by the ANN model, the MATLAB GUI toolbox can use the UHSFRC mixture best to get the optimized mixture in the future.