Predictive analytics for fresh concrete rheological characteristics using artificial intelligence approaches

Acquiring precise knowledge of the rheological characteristics of fresh concrete is particularly important for ensuring its pumpability and flowability. The present study aims to design reliable predictive tools for the yield stress (YS) and plastic viscosity of fresh concrete, modeled as a Bingham fluid. Artificial intelligence (AI) approaches, including Gaussian process regression (GPR), multilayer perceptron neural network (MLP-NN), and radial basis function neural network (RBF-NN), are utilized to achieve this goal. The proposed models are validated using 142 experimental data gathered from the literature. The analyzed data enveloped YS and PV of fresh concrete in extensive ranges of time after mixing and the contents of various constituents, including water, cement, sand, aggregates of various sizes and additives. According to statistical investigations, all AI-based models present satisfactory outcomes for both rheological factors analyzed. However, the most trustworthy results are obtained by the RBF-NN based model, with average absolute relative error (AARE) of 8.84 % and 7.65 % for YS and PV, respectively, in the testing (validation) data. Furthermore, these models accurately estimate the rheological characteristics of fresh concrete within most operational ranges, with relative errors less than 10 %. An analysis based on the William's plot confirms the high credibility of both the experimental data and the models for YS and PV. Ultimately, the order of significance of operating parameters in controlling the rheological properties of fresh concrete is identified through a sensitivity analysis.