Machine Learning based blade design optimisation of the NeoVAD

The NeoVAD is a proposed paediatric axial-flow Left Ventricular Assist Device (LVAD), small enough to be implanted in infants. The design of the impeller and diffuser blades is important for hydrodynamic performance and hemocompatibility of the pump. This study aims to implement blade optimisation utilising Computational Fluid Dynamics (CFD) modelling, machine learning and global optimisation techniques.

CFD simulations were first validated by comparison of the pressure-flow curves with those measured for prototype pumps. Blade geometry was parameterised with five variable parameters: impeller chord length, inlet and outlet angle; diffuser chord length and inlet angle. CFD models of 32 base geometries were created. For each of the pump designs, a range of mass flow rate conditions were simulated. As fluid simulations are computationally expensive, a surrogate model was required to allow the optimisation routine to conduct an efficient search; a Neural Network predicted the optimisation objective at design points not explicitly simulated. A Genetic Algorithm was used – operating on the surrogate model – to search for an optimal design of the five variable parameters.

Blade geometries were successfully optimised, and the optimal designs were each better than the original 32 base geometries. Two objective functions were considered as analogues to hydrodynamic performance and hemocompatibility: respectively, maximising pressure head (75 mmHg above base average, 47 mmHg above best base design) and maximising efficiency (9.5 % above base average, 0.98 % above best base design).

An optimisation method for the blade design of LVADs has been shown to work for a single objective function. The simulations that form the basis of the surrogate model, however, must be revisited to ensure better agreement with experimental data in future optimisations. A multi-objective optimisation routine is the next step in this research.