Learning rheological parameters of non-Newtonian fluids from velocimetry data

We solve a Bayesian inverse Navier–Stokes (N–S) problem that assimilates velocimetry data by jointly reconstructing a flow field and learning its unknown N–S parameters. We devise an algorithm that learns the most likely parameters of a Carreau shear-thinning viscosity model, and estimates their uncertainties, from velocimetry data of a shear-thinning fluid. We conduct a magnetic resonance velocimetry experiment to obtain velocimetry data of an axisymmetric laminar jet in an idealised medical device (US Food and Drug Administration’s benchmark nozzle) for a blood analogue fluid. The algorithm successfully reconstructs the flow field and learns the most likely Carreau parameters. Predictions from the learned model agree well with rheometry measurements. The algorithm accepts any differentiable algebraic viscosity model, and can be extended to more complicated non-Newtonian fluids (e.g. Oldroyd-B fluid if a viscoelastic model is incorporated).