Abstract

Predicting the static yield stress (τ0S) is relevant to design innovative concrete technologies. This paper evaluated three machine learning methods to choose the most accurate model to predict the growth of τ0S of blended cement pastes. Eight input parameters were used, including supplementary cementitious material (SCM) properties, cement reactivity, mixture design parameters, and resting time. Among the evaluated methods, the multilayer perceptron was the most accurate. Results showed that τ0S during the initial linear growth stage is controlled by the SCM properties. However, the τ0S during the later exponential growth stage is controlled by the amount of pseudo-contact zones.