These slides were presented at the NAFEMS World Congress 2025, held in Salzburg, Austria from May 19–22, 2025.

Abstract

At Grundfos, mechanical vibrations and acoustic noise has an increasing focus. As new products have increasing power density and variable speed, dynamics and noise are a main concern in development projects. Since dynamics in such complex systems are difficult to model and understand, dynamic issues are unfortunately often found too late. Centrifugal pumps experience a very complex dynamic load pattern which depends both on rotational speed, duty point, pump media, temperature and assembly tolerances. The load spectrum includes distinct frequency loads, such as unbalance, blade-pass load and EM-motor loads, but also broadband hydraulic loads from turbulence and recirculation. To predict noise and vibrations from products at distinct operational frequencies using simulation, a very high level of accuracy and fidelity is required. Many modelling aspects such as interface characterization, accurate material models, added mass effects, FSI, system damping, and dynamic forces must be mastered to expect reasonable accurate results. Mastering all these aspects at once is very difficult on a product level, hence we seek a way to model and interpret result which can still provide usable result although uncertaintiesfrom the before mentioned modelling aspect is practically unavoidable. To achieve this, an approach for hydraulic force modelling is presented in this paper. The paper presents an approach to model blade-pass loads and pressure pulsations. The load calculation is based on a geometrical approach and the load magnitude and shape can be tuned to match CFD, empirical or experimental results. The mapping of the load cases is done in such a way that loads represented as unit-loads can be combined and scaled to match many different load scenarios, which cover both different operational points and cases, but also variance due to production and assembly tolerances. Results are processed as waterfall plots which makes it possible to interpret result in a less error-prone way and to comprehend the response of a whole response range at once. Using this approach, we are able to model dynamic response of large assemblies and complex products and process the result in a way which reveals the dominating dynamics leading to vibration and/or noise issues. A case is presented where experimental and simulation results are compared.