A high-dimensional design optimisation method for centrifugal impellers

Abstract

The metamodel-based design optimisation (MBDO) method has been gaining increasing popularity in the field of turbomachinery design. Nevertheless, the ‘curse of dimensionality’ remains a formidable obstacle for the application of MBDO in high-dimensional design problems, which, in the context of turbomachinery design, are problems with at least 10 design variables, if expensive simulations such as computational fluid dynamics are required. To address this issue, the present study proposes to utilise a novel high-dimensional model representation (HDMR) method to support the process of MBDO and applies it to the design optimisation of a centrifugal impeller. A total of 15 geometrical design variables and three performance objectives are considered in the optimisation. In particular, the HDMR method is assisted by support vector regression (SVR) metamodels to explicitly represent the relationships between design variables and performance objectives. Apart from high metamodelling accuracy and efficiency, the SVR-HDMR metamodelling technique enables designers to refine design space as well as to identify correlations between design variables during the metamodelling process. Analysis of the final optimisation results show that a reduced hub-line curvature, a negative rake angle, a positive splitter blade lean angle and an increased splitter blade inlet angle are beneficial for enhancing efficiency of this impeller. This work provides valuable references for further high-dimensional design optimisation of turbomachine components.