A Novel Approach to Parametric Design of Centrifugal Pumps for a Wide Range of Specific Speeds

This work presents a novel approach to parametric design of centrifugal pumps. The outcome is represented by an industrial design tool for a whole family of pumps with single shaft centrifugal impeller, volute, horizontal suction duct and vertical discharge diffuser. In order to reduce costs and time to market for a given design, the main industrial goal consisted in developing a quick and flexible design tool, capable of describing in a continuous manner the whole range of specific speeds of interest. This task should be achieved without reusing or re-adapting existing geometries, as this usually leads to poor performance. To guarantee the continuity over the whole design space, when varying the specific speed, the entire family of pumps had to be geometrically described by means of the same parameterization.

By varying all the geometrical parameters, a performance database was created and used to train a meta-model (artificial neural network or support vector machine), which provides a global response surface describing the entire design space for any specific speeds. The response surface is representative of the impeller performance, that were evaluated using a three-dimensional viscous flow solver (TRAF). The accuracy of the meta-model in generating the response surface was evaluated by comparing the operating curves of several geometries with CFD results, demonstrating the ability of the meta-model to well reproduce performance.

The flange-to-flange pump performance are then calculated by coupling the response surface for the impeller with correlations for the static components. In this way, the design tool is able to predict the performance of any combination of impeller and static components. The proposed design tool allows the designer to calculate a reliable performance map as a guide to the expected operating flow range and the sensitivity to the choice of the static components, thus assessing the most suitable solution.