The goal of the project was to optimise the Impeller and Diffuser for a novel Left Ventricular Assist Device (the NeoVAD) for paediatric patients.
To aid this, an automated framework was generated to create a new geometry, update the mesh, and run a simulation for multiple experiments.
An initial model was created using values obtained from mean-line theory, and a Design of Experiments (DoE) was generated using theD-optimal method.
Steady-state simulations were run using the Frozen Rotor method, and the pressure head results from each experiment were fed into a Generalized Regression Neural Network (GRNN). This was used within the NSGA-II Genetic Algorithm to determine the optimum blade parameters based on the required pressure increase.
The initial simulation run using blade angles determined from theory had an error of 0.5% compared with the expected value. A total of eighty simulations were run as part of the DoE through the automated framework, 5% of which did not mesh successfully. The GRNN was generated with a spread of 0.5, and an error of 5.5% when validated using the training data.
However, the simulated pressure increase using the parameters determined by the genetic algorithm had an error of 25% compared with the expected value. The likely cause of this was a poorly refined DoE with input ranges that were too wide, and the limited dataset. The automated workflow was created successfully and provided a significant boost in efficiency, doubling the number of simulations run per day. The framework for the optimisation process was generated and verified. A brief investigation into a numerical model to calculate blood Haemolysis was conducted.
Master’s Dissertation | Summary | Aims and Objectives | Background | Literature Review | Methods | Results and Discussion | Conclusions | References