Application of Recurrent Neural Network in Actual Shear Rate Prediction Under Wall Slip Phenomenon

Abstract

Wall slip refers to the phenomenon where particles in a suspension move away from the boundary wall, creating a thin liquid-rich layer nearby. This occurrence can significantly affect rheological measurements, notably viscosity, shear stress, and shear rate. Suspensions find widespread use in various industries such as food processing, personal care products, pharmaceuticals, paints, medicine, and agrochemicals. Predicting the actual shear rate traditionally proves challenging, time-consuming, and cost-intensive. Hence, there's a pressing need for a computational model to perform this task with acceptable accuracy. Leveraging the precise input-output mapping capability of recurrent neural network (RNN), it was employed to develop a model for the actual shear rate prediction. Evaluation of these models through statistical analyses reveals that RNN model III outperforms others, boasting the highest coefficient of determination (0.9998), lowest mean squared error (0.000721), root mean squared error (0.001361), most negative Akaike information criterion (-18646.3), Bayesian information criterion (-18635.9), and the smallest percentage error (15%). This developed model provides an alternative means to predict suspension shear rate under experimental constraints.