MULTI-PARAMETERS PUMP IMPELLER OPTIMIZATION

Rotodynamic pumps designed by traditional methods have persisted at their current levels of maturity, vis-a-vis performance and efficiency, for decades now. These historical statistics of performance have been formulated as industry standard charts of efficiency based on specific speed and flow-rate (Fig. 1). Through the last decade, research activity aimed at challenging and surpassing these historical performance levels, via the application of various optimization methods and tools, has been slowly but surely building up. The work presented in this paper is one such effort to enhance the performance of a pump, with the impeller blade of a medium specific speed centrifugal pump chosen as its subject, the blade being the heart of any turbo-machine. This study is somewhat unique in the respect that intuitive, high-level features were used to optimize the blade design rather than inanimate variables. A radical and proprietary blade development scheme allowed a 75% reduction in the number of parameters needed to fully describe the blade, compared to conventional blade layout and development schemes. A software suite comprised of an intimate blend of proprietary and commercial tools, for design, performance simulation and optimization, constituted the engine for this study. A template was generated with all software linked in a seamless bi-directional communication loop for hands-off execution of the optimization study (Fig. 3). A Design of Experiments (DoE) scheme is used to generate several hundred design points corresponding to different geometry and flow configurations. Computational Fluid Dynamics (CFD) software was used to simulate the performance and yield integrated output parameters, viz., head, power, and efficiency, for each design point at multiple flow-rates. Data from the simulated DoE table was then used to generate an n-dimensional Response Surface (RS), which established pre-validated relationships of the output to the input parameters. Finally, a Multi Objective Genetic Algorithm (MOGA) technique was used to extract the most efficient designs satisfying other weighted goals or objectives. The weighting process forces some objectives to be sacrificed to a certain extent in favor of others, which is a good reflection of reality, since a trade-off between different objectives is an essential element of any design process. The predictions from the optimizer were then subjected to a final round of CFD simulations for validation. Two discrete application examples were used as test cases for this study, using the optimization template described in the foregoing. This multi objective optimization approach resulted in a performance improvement of 2 efficiency points over a mature baseline blade design, developed using an advanced proprietary design code. Traditional CFD-assisted design enhancements were found to provide less than 1 point of efficiency improvement over this baseline design.