Simulation of multi-phase flow of non-Newtonian fluid via phase field-LBM an‎d FSI in python; accelerated with parallel computing

10/21/2026 12:00:00 AM

This study presents a unified computational framework for simulating non-Newtonian multiphase flows using the Lattice Boltzmann Method (LBM) an‎d the Phase-Field (Cahn-Hilliard) approach, coupled with the Immersed Boundary Method (IBM) for Fluid-Structure Interaction (FSI). A Power-law model is implemented to capture shear-thinning rheology by self-consistently updating the relaxation time within a Multiple-Relaxation-Time (MRT) LBM scheme. To address computational costs, the code is developed in Python with a modular architecture an‎d GPU acceleration. Validation was performed via three key benchmarks: fundamental multiphase flow tests, analytical comparison of two-layer non-Newtonian Poiseuille flow (maximum error ≈9.68%), an‎d FSI scenarios involving elastic boundary deformation, which showed good agreement with reference data. All simulations employed the MRT operator to ensure higher stability an‎d accuracy compared to SRT. Performance analysis demonstrated that the PyTorch-based GPU implementation provided speedups of 7.32× (RTX 3070) an‎d 16.30× (RTX 4070) compared to CPU baselines. Moreover, integrating custom CUDA kernels with manual memory management offered additional acceleration of 2.81× an‎d 3.53× over the PyTorch version, facilitating the stable simulation of complex geometries. Results from non-Newtonian FSI simulations indicated that local viscosity distribution an‎d boundary deformation aligned with flow patterns, where increased velocity in the shear-thinning fluid led to greater longitudinal deformation. Ultimately, this work provides a reproducible, stable, an‎d high-performance platform for modeling complex multiphase FSI problems in biofluids an‎d microfluidics.

روش بولتزمن شبكه‌اي , ميدان فازي , سيالات غيرنيوتني , برهم‌كنش سيال-ساز , پردازش موازي روي GPU

Lattice Boltzmann method , Phase field , Non-Newtonian fluids , Fluid-structure interaction , Parallel processing on GPU

Mohammadreza Ahdi

Pr.Hoseinalipour