Blood-device design drops mesh-by-mesh CFD: a mesh-free PINN enforces mass conservation and reproduces FDA nozzle shear layers
Synopsis
Hydrodynamic design of blood-contacting medical devices has long relied on meshed CFD, where mesh structure and compute cost slow optimization. A mesh-free physics-informed neural network writes the axisymmetric mass-conservation equation into its loss function and is trained on experimental PIV/LDV data over the FDA reference nozzle, spanning laminar, transitional and fully turbulent regimes (Re = 500–6500), reconstructing flow fields and turbulent kinetic energy at high resolution and yielding an empirical correlation for peak TKE as a rapid hydrodynamic indicator.
Interpretation
Hydrodynamic optimization of blood-contacting devices can move away from generating a mesh for each CFD run, with flow fields and turbulent kinetic energy distributions reconstructed directly by a mesh-free model. The prior route was meshed CFD, whose results depend heavily on mesh structure and carry high computational cost, creating a bottleneck in optimization. On FDA reference nozzle geometries the model predicted sharp velocity gradients in the shear layers with high accuracy, trained and validated on experimental PIV/LDV data covering laminar, transitional and fully turbulent regimes.
The model does not merely fit experimental data; it places a physical constraint directly in the training objective so reconstructions satisfy axisymmetric mass conservation. Common artificial-intelligence algorithms only fit data and do not impose conservation equations as mathematical constraints. The axisymmetric mass-conservation equation is integrated into the loss function, and the model is trained across Re = 500–6500, with publicly available FDA benchmark experimental data used for validation.
Within the investigated conditions, peak turbulent kinetic energy can be estimated quickly from an empirical correlation that serves as a hydrodynamic indicator. Obtaining peak turbulent kinetic energy previously required a full flow solution or post-processing. The correlation was derived within the investigated conditions, limited to this nozzle geometry and the Re = 500–6500 range.
Perspective
The result is validated on FDA reference nozzle geometries, with training and validation data covering laminar, transitional and fully turbulent regimes at Re = 500–6500, using publicly available FDA benchmark PIV/LDV measurements. The peak-TKE empirical correlation is limited to the investigated conditions. It is aimed at readers working on hydrodynamic design of blood-contacting devices, post-processing of experimental flow fields, and physics-constrained learning.
The peak-TKE correlation was derived only within the investigated conditions, and extrapolation beyond that range is not given in the material. Model weights, source code and post-processed data are available from the corresponding author upon reasonable request, so public availability is not guaranteed. The method addresses axisymmetric sudden-expansion flow, and behavior on other geometries is not stated in the material.