Multi-Objective Genetic Algorithm Assisted by an Artificial Neural Network Metamodel for Shape Optimization of a Centrifugal Blood Pump

A centrifugal blood pump is a common type of the pump used as a left ventricular assist device (LVAD) in the medical industries. The reduction of the LVADs hemolysis level to reduce the blood damage is one of the major concerns in designing of such devices. Also, the enhancement of the LVADs efficiency to decrease the battery size is another design requirement. The blood damage critically depends on the state of the blood being pumped. Besides the blood state, the blood damage also depends on the pump impeller and volute geometries. In this research, a multi-objective optimization of a centrifugal blood pump is performed. A complete 3D-optimization platform is established for both impeller and volute of a centrifugal blood pump consisting of parametric modeling, automatic mesh generation, computational fluid dynamics (CFD) simulation, and optimization strategy. A vast number of cases with various impeller and volute shapes are numerically simulated. Three different metamodels are created using artificial neural networks (ANNs) in order to approximate the pump hydraulic efficiency, hemolysis index (HI), and pressure head. The inverse of the relative pressure head is defined as the first objective and the summation of relative hemolysis index and the inverse of the relative efficiency is assumed as the second objective. Non-dominated Sorting Genetic Algorithm-II (NSGA-II) is used to find the Pareto Front. A set of optimal points is selected. Finally, for the physiological flow conditions, the optimum design that provides 11.9% HI reduction and 7.2% efficiency enhancement is selected.