Pipe flow of thixotropic fluids: Experiments, simulations and neural networks

Thixotropy refers to the time-dependent evolution of a material’s microstructure, where structural agglomerates are progressively broken down under shear and rebuild when flow ceases. This study examines how such microstructural dynamics influence the transient pressure response of thixotropic yield stress fluids in laminar pipe flow. Experiments on Laponite suspensions performed with a custom built pipe rig reveal a pronounced decay in pressure drop over time at constant flow rate, reflecting shear induced structural breakdown. Complementary computational fluid dynamics simulations based on the Houska thixotropic constitutive model accurately reproduce the experimentally measured transient pressure drop evolution, with relative errors below 5% for most flow conditions, and elucidate the interplay between structural breakdown and recovery that governs the evolution of flow. The simulations capture the reduction of plug flow velocity profile due to thixotropic effects as flow rate increases, accompanied by a reduction of unyielded regions. To reduce experimental effort and computational cost, a feed-forward neural network surrogate trained on a limited set of validated experimental and simulated data accurately predicts the transient pressure drop evolution across a wide range of flow conditions with greater than 95% accuracy. The integrated experimental, computational, and data-driven framework provides an efficient approach for predicting transient pressure losses in thixotropic pipe flow, substantially reducing the need for extensive experiments and computationally intensive CFD simulations.