Computational Biofluid and Multiphysics Modeling
Abstract
This project advances high-fidelity simulation of complex fluid systems in two distinct but computationally intensive areas: Computational Biofluid Dynamics and Multiphysics Modeling. Both branches rely on large-scale numerical simulations to investigate unsteady, three-dimensional flows that are beyond the reach of conventional experimental or analytical methods. The project continues and expands work supported by previous NAISS Medium Compute and Large Storage allocations, with new methodological developments, higher resolution, and data-driven modeling components.
The Computational Biofluid Dynamics branch focuses on patient-specific simulations of cardiovascular flow derived from medical imaging. Using both ANSYS Fluent/CFX and IBAMR immersed-boundary framework with large-eddy simulation (LES) methods, the project models pulsatile blood flow in anatomically realistic, moving geometries reconstructed from photon-counting CT (PCCT) and 4D flow MRI. Applications include (i) atrial flow dynamics in atrial fibrillation (AF), (ii) hemodynamics of left-ventricular assist devices (LVADs) and aortic valve insufficiency, (iii) turbulent flow in a total artificial heart prototype, and (iv) preoperative prediction of valve-implantation outcomes in collaboration with the Mayo Clinic. These studies address key clinical questions on thrombus formation, device–heart interaction, and surgical optimization. By combining high-resolution imaging with CFD, the work pushes the frontier of patient-specific modeling and contributes to the development of physics-based digital twins for cardiovascular research.
The Multiphysics Modeling branch investigates complex engineering flows where fluid mechanics interacts with heat transfer, cavitation, or multiphase transport. Large-scale CFD simulations using ANSYS Fluent/CFX, OpenFOAM, and in-house Python/PyTorch frameworks are performed to (i) develop machine-learning–assisted surrogate models for turbulent flow and active flow control, (ii) study coupled thermal–fluid behavior in gyroid-type lattice structures for additive manufacturing, (iii) model cavitation-induced erosion in hydraulic systems, and (iv) quantify multiphase mixing dynamics in biogas reactors. These efforts aim to create benchmark-quality datasets for training and validating data-driven models, improving predictive capability, and accelerating design optimization in multiphysics engineering applications.
Both branches rely critically on high-performance computing (HPC) to achieve the required spatial and temporal resolution. Typical models consist of 10–100 million computational cells, requiring 128–512 CPU cores per simulation and producing 0.5–2 TB of data each. The combined computational demand for 2026 amounts to approximately 3 million core-hours. The increased allocation relative to 2025 is justified by the transition to higher-fidelity imaging-based geometries, ensemble simulations for uncertainty quantification, and machine-learning workflows that require extensive CFD-generated training data.
Through the integration of advanced numerical methods, medical imaging, and machine learning, this project aims to generate new scientific knowledge on flow physics in both biological and engineered systems, strengthen Sweden’s computational research capacity, and contribute to the development of predictive, data-driven simulation methodologies across disciplines.
