Abstract Summary (Max 250 words)
Inline rotor–stator mixers are widely used in consumer goods manufacturing to accelerate mixing and control microstructure, both critical for product performance and consumer satisfaction. Power input in these mixers is typically described using a two term correlation that separates tank type and flow driven contributions, from which the coefficients Poz and k1 are obtained by regressing experimental or CFD data. Although these coefficients vary significantly with geometry, industrial scale up practice often applies fixed values based on limited data for specific mixer designs, leading to inaccurate power predictions and inconsistent product quality. Given the wide variation in geometries across manufacturing sites, direct measurements or CFD simulations for every configuration is impractical. This work presents an integrated approach combining validated CFD simulations, geometry parameterisation, and data driven modelling to predict Poz and k1 across diverse inline high shear mixer designs. Building on the geometric combinations described in literature, this study replaces design specific investigations with a generalised parameter space that systematically represents key rotor-stator features, including rotor diameter, rotor and stator open area, gap width and stator thickness across multiple scales. Reduced order models are developed using several machine learning methods, with Gaussian process modelling selected for its accuracy and uncertainty quantification. Bayesian optimisation is used to guid additional simulations and enhance predictive capability. The resulting models reproduce literature-reported values and capture the nonlinear influence of geometric parameters on Poz and k1, providing a scalable and reliable workflow for predicting mixer power behaviour and improving scale up fidelity.
