The analysis of rheological properties of suspensions requires the use of models such as Einstein’s formulation for viscosity in dilute conditions, but its effectiveness diminishes in the context of concentrated suspensions. This study investigates the rheology of suspensions containing solid particles in aqueous media thickened with starch nanoparticles (SNP). The goal is to model the viscosity of these mixtures across a range of shear rates and varying amounts of SNP and SG hollow spheres (SGHP). Artificial neural networks (ANN) combined with swarm intelligence algorithms were used for viscosity modeling, utilizing 1104 data points. Key features include SNP proportion, SGHP content, log-transformed shear rate (LogSR), and log-transformed viscosity (LogViscosity) as an output. Three swarm algorithms—AntLion Optimizer (ALO), Particle Swarm Optimizer (PSO), and Dragonfly Algorithm (DA)—were evaluated for optimizing ANN hyperparameters. The ALO algorithm proved most effective, demonstrating strong convergence, exploration, and exploitation. Comparative analysis of ANN models revealed the superior performance of ANN-ALO, with an R2 of 0.9861, mean absolute error (MAE) of 0.1013, root mean absolute error (RMSE) of 0.1356, and mean absolute percentage error (MAPE) of 3.198%. While all models showed high predictive accuracy, the ANN-PSO model had more limitations. These findings enhance understanding of starch suspension rheology, offering potential applications in materials science.
