Approaches commonly used in the design and optimization of centrifugal blood pumps, based on manual selection of design parameters or systematic modification of design configurations, aim to identify the best candidate among a restricted set of design points rather than the global optimum. In contrast, automated optimization techniques aim to reach the near-global optimum by exploring the design space from a broader perspective while identifying the best parameter combinations with minimal human intervention. This study presents an automated procedure developed for the blade optimization of a centrifugal blood pump by integrating an adaptive neuro-fuzzy inference system modeling with a genetic algorithm. A baseline pump model was designed using conventional turbomachinery methods to meet the requirements of a left ventricular assist device. Geometric variations were generated using a Latin Hypercube Sampling approach, and computational fluid dynamics simulations were carried out to produce a dataset covering pressure head, hydraulic efficiency, and hemolysis index. These performance parameters were incorporated into the optimization framework, enabling simultaneous evaluation. The optimized impeller blade geometry demonstrated notable improvements over the baseline pump, including a 7.2% increase in pressure head, a 2.4% enhancement in hydraulic efficiency, and a 4.78% reduction in hemolysis index. Flow field analyses indicated reduced recirculation and improved suction performance, as well as decreased hemolysis accumulation near the impeller blade outlet. The results indicate that the simultaneous use of machine learning and evolutionary optimization has the potential to enable the design of centrifugal blood pumps with improved hydraulic and hemolytic performance.
Önder et al. (Fri,) studied this question.