Optimizing a Physics-Informed Machine Learning Model for Pulsatile Shear-Thinning Channel Flow

Abstract

This paper explores the application of physics-informed neural networks for solving pulsatile shear-thinning flows in a two-dimensional channel. To identify an optimal model, models of varying implementations of boundary conditions, network sizes, number of training points, activation functions, and loss weights are investigated through case by case studies complemented by Gaussian-processes based Bayesian optimization. The final model demonstrates a high level of agreement with a reference numerical solution, with an error of less than 2%. This result indicates that appropriately trained PINNs can be utilized as a method for simulating transient shear-thinning flows.