Genetic algorithm on the design optimization of a micro gas turbine centrifugal compressor impeller

Authors

Konstantinos Braziotis  - Aristotle University of Thessaloniki, Greece
Konstantinos Eleftheriou - Aristotle University of Thessaloniki, Greece
Anestis Kalfas - Aristotle University of Thessaloniki, Greece

Abstract

A numerical analysis based on a 1-D model and Computational Fluid Dynamics (CFD) simulations is caried out in order to investigate the process of optimizing the rotor design of a centrifugal compressor, for Micro Gas Turbine (MGT) application. Scope of this work is the increase of the adiabatic efficiency of the impeller, during the preliminary design. The geometry of the blade at the meridional flow path is parameterized by using Bezier-Bernstein polynomial curves. In that manner the 2-D shape of the flow path is differentiated with the use of control points. The proposed approach of the efficiency evaluation is based on an in-house one-dimensional model, which is tailored by using some empirical loss correlations models. This work focuses on the implementation of a genetic algorithm, for the purpose of investigating new geometries in an automated/ stochastic way, with the aim of minimizing the losses and thus increasing the isentropic efficiency of the rotor. The verification of the model is done by simulating the performance in Ansys CFX. The purpose of this work is to create a model that performs quasi 3-D performance evaluation of preliminary/initial design geometries with sufficient accuracy, but also with significant time savings compared to similar traditional three-dimensional fluid-mechanical simulations. The results of the one-dimensional analysis model of the present work showed that from the original to the final/optimal geometry we have an increase in the efficiency of 1.2%. This was confirmed by simulating the 3-D flow in the CFD of the same geometries, initial and final, as derived from the genetic algorithm, with the same distribution of the beta angles for both cases. The results showed a 2.16% increase in efficiency and a slight increase in the pressure ratio from 1.63 to 1.65. It is worth noting that the pressure ratio resulting from the one-dimensional analysis was 1.61, the same order of magnitude as the CFD. In the end, it was confirmed that the genetic algorithm produces a better geometry in terms of the isentropic efficiency imp. The CFD analysis even predicts an increase in efficiency of 0.936% more than GA's prediction.