Shear thickening fluid: A multifaceted rheological modeling integrating phenomenology and machine learning approach

Abstract

Amorphous silica and polyethylene glycol (PEG)-based shear thickening fluid (STF) having 30 % (w/w) was synthesized. The complex viscosity pattern for PEG-silica STF examined at varying temperatures (25–50 °C) and shear rates (1–1000 1/s). It was predicted using Galindo-Rosales technique-based phenomenological model that utilizes piecewise functions to predict viscosity in different shear rate zones. Moreover, machine learning (ML) based models namely, support vector regression (SVR) and artificial neural networks (ANNs), were developed to forecast the nonlinear nature of the viscosity of STF as a function of temperature and shear rates. The phenomenological model performs well during training (R² > 0.99) but has low prediction the for unknown test data (R² > 0.90), i.e. it overfits the data. Additionally, for phenomenological modeling, too many parameters need to be evaluated using complex equations based on iterative calculations, whereas ML based models are more accurate, quick, and generalized for both training and testing regimes for all zones. The ML models projected outstanding match between the predicted and experimental viscosities. For every zone, the present study concludes that the ML approach, with its better generalization outcomes, is a robust technique to estimate the rheology of STF. As a result, viscosity behaviour prediction by ML could assist in designing custom fluid formulations with desired viscosity properties for specific applications in varied environmental conditions.