Optimization of Axial Thrust Balancing Swirl Breakers in a Centrifugal Pump Using Stochastic Methods

Abstract

Hydraulic axial thrust balancing in single-stage radial pumps is a frequently applied procedure to reduce the bearing load to a reasonable level. It leads to considerable efficiency loss, which gives reason to investigate the optimization potential of the common balancing methods. This paper’s focus is on so-called casing ribs, which are used in the default design of an examined industrial pump. Radial vanes on the casing wall of the front impeller side chamber work as swirl breakers by decreasing rotating flow whereby the static pressure at the shroud increases and counteracts the resulting axial thrust. The objective is to retain the reduction of axial thrust and to improve the internal efficiency simultaneously. Therefore a CFD model of the industrial radial pump is created with Ansys CFX. Sufficient numerical quality is ensured whereby consistency is verified by a mesh study. The model is validated by integral values of the characteristic curve and axial thrust measurements as well as by experimental transient static pressure measurement at different locations of the pump flow. Probes are placed in the suction port, the volute and the impeller side chambers, where most balancing methods are implemented. Since the side chamber contains a complex flow, the effect of geometry changes is hard to predict. For this reason a stochastically based sensitivity analysis using a comprehensively parameterized geometry of the front side chamber domain with the included casing ribs is carried out. For this purpose 110 design points are calculated and evaluated with support of the software Optislang. Correlations of parameters are suggested and important parameters regarding the objective are identified. Some reasonable model simplifications are conducted to reduce the computational time. According to the acquired findings a local optimization is executed using the best sample of the sensitivity analysis as start design. An evolutionary algorithm method determines a best design with an efficiency improvement of 0.26 percentage points. It is discussed in detail conclusively.