Assessment of turbulence models for pulsatile flow inside a heart pump

Computational fluid dynamics (CFD) is applied to study the unsteady flow inside a pulsatile pump left ventricular assist device, in order to assess the sensitivity to a range of commonly used turbulence models. Levels of strain and wall shear stress are directly relevant to the evaluation of risk from haemolysis and thrombosis, and thus understanding the sensitivity to these turbulence models is important in the assessment of uncertainty in CFD predictions. The study focuses on a positive displacement or pulsatile pump, and the CFD model includes valves and moving pusher plate. An unstructured dynamic layering method was employed to capture this cyclic motion, and valves were simulated in their fully open position to mimic the natural scenario, with in/outflow triggered at control planes away from the valves. Six turbulence models have been used, comprising three relevant to the low Reynolds number nature of this flow and three more intended to investigate different transport effects. In the first group, we consider the shear stress transport (SST) \[k - \omega \] k−ω model in both its standard and transition-sensitive forms, and the ‘laminar’ model in which no turbulence model is used. In the second group, we compare the one equation Spalart–Almaras model, the standard two equation \[k - \varepsilon \] k−ϵ and the full Reynolds stress model (RSM). Following evaluation of spatial and temporal resolution requirements, results are compared with available experimental data. The model was operated at a systolic duration of 40% of the pumping cycle and a pumping rate of 86 BPM (beats per minute). Contrary to reasonable preconception, the ‘transition’ model, calibrated to incorporate additional physical modelling specifically for these flow conditions, was not noticeably superior to the standard form of the model. Indeed, observations of turbulent viscosity ratio reveal that the transition model initiates a premature increase of turbulence in this flow, when compared with both experimental and higher order numerical results previously reported in the literature. Furthermore, the RSM is indicated to provide the most accurate prediction over much of the flow, due to its ability to more correctly account for three-dimensional effects. Finally, the clinical relevance of the results is reported along with a discussion on the impact of such modelling uncertainties.