Prediction of Tensile Properties of Ultra-High-Performance Concrete Using Artificial Neural Network

A multilayer perceptron artificial neural network (MLP-ANN) was developed to calculate the cracking stress, tensile strength, and strain at tensile strength of ultra -high-performance concrete (UHPC), using the mixture design parameters and strain rate during testing as inputs. This tool is envisioned to provide reference values for direct tension test results performed on UHPC specimens, or to be employed as a framework to determine the tension response characteristics of UHPC in the absence of experimental testing, with minimal computational effort to determine the tensile characteristics. A database of 470 data points was compiled from 19 different experimental programs with the direct tensile strength, cracking stress, and strain at tensile strength corresponding to different UHPC mixtures. The model was trained, and its accuracy was tested using this database. A reasonably good performance was achieved with the coefficients of determination, R 2 , of 0.91, 0.81, and 0.92 for the tensile strength, cracking stress, and strain at tensile strength, respectively. The results showed an increase in the cracking tensile stress and tensile strength for higher strain rates, whereas the strain at tensile strength was unaffected by the strain rate.