Impeller Dynamics in a Diesel Engine Converter

To be able to avoid fatigue problems in impeller pumps for torque converters, the engineer must have a thorough understanding of the nature of the fluid-structure interaction characteristics of the impeller pump. In this article, we show how this was obtained by combining the outcome of fluid-structure interaction simulations with results obtained through experiments and CFD-calculations.

Fig. 1: Torque Converter Work Schematic

Introduction
With increasingly more powerful tools for the computation of physical quantities (e.g. software for structural analysis and for computational fluid dynamics), it has become possible to design lighter and more energy-efficient mechanical devices like torque converters used for cars, excavators, and a variety of other drive-trains. However, limit design with respect to certain features often causes new and unknown problems to occur. Among many "nasty" phenomena that may be difficult to get a grip on, are flow-induced vibrations. Flow-induced vibrations may result in fatigue problems and ultimately failure of the subject. In this paper, we show how the results from prototype testing of a new torque converter could be explained by means of combining the outcome of structural, CFD, and Fluid-Structure Interaction simulations, thereby establishing the foundation for an advanced and reliable design. The work was performed in cooperation between a major Swedish producer of movable equipment and its German supplier of torque converters with the assistance of ANKER - ZEMER Engineering AB.

The working Principle of a torque Converter
A torque converter has three main parts: Impeller (or Pump), the Stator, and the Turbine (see Fig. 1).

Impeller Problems
During prototype testing, it became apparent that very small changes in the geometry of the impeller can lead to serious fatigue problems and ultimately total failure of the converter (a typical torque converter is shown in Figure 2).
However, the influence of the various geometrical parameters was very difficult to apprehend, as the physics of the problem was not very well understood.
Therefore, a project with the purpose to investigate the problem was initiated.

Fig. 2: Torque Converter

The Project
At the start of the project, it was clear that the problem could not be efficiently studied by only using testing and/or conventional numerical simulations alone, as the project was most likely facing a Fluid-Structure Interaction (“FSI”) problem involving relatively high frequencies. To simulate a flow field having high frequency oscillations due to phenomena’s such as rotor-stator interaction, vortex shedding etc, very fine grids and small time steps in the unsteady simulations yielding excessive computer running times are required. Furthermore, if fluid–structure resonance points are to be found, an excessively high number of CFD simulations interacting with structural simulations have to be performed. Due to the presumed difficulties of the task, it was decided to perform the project based on concerted testing and numerical simulations to find the cause of the vibrations resulting in limited fatigue life of the impeller. Testing would yield factual data to be used in their own right and also data needed for the calibration of the numerical simulations, and the numerical simulations would reveal the influence of the various parameters and (hopefully) give a better understanding of the physics of the problem. The project setup included the following tasks and tools to be utilized:

1. Investigate several impeller configurations by testing.
   
2. Perform steady and unsteady CFD fluid dynamic analysis of the initial design and suggested design changes.
   
3. Perform Fluid-Elastic analysis of the initial impeller design in the frequency domain. This was performed by ANKER-ZEMER Engineering AB (Sweden).

The numerical simulations performed under tasks 2 and 3 above comprised structural dynamics (utilizing the ANSYS Finite Element program), fluid dynamics (utilizing the FLUENT CFD software), and fluid-elastic analysis (applying the LINFLOW Fluid-Structure Interaction analyzer utilizing modes and eigenfrequencies computed in ANSYS).

Fig. 3: Pressures in Impeller as computed in Fluent

Tasks and Findings
The experimental work (Task 1) showed that small changes in impeller geometry would result in significant variations in impeller fatigue life. However, testing did not give any conclusion as to why.

The CFD simulations of steady and unsteady fluid flow (Task 2) did not reveal any significant changes in impeller blade load due to geometrical modifications. This may sound like a surprise considering the large variations in fatigue life, but was not totally unexpected. A picture of the pressures on the impeller from the CFD calculations is shown in figure 3.

The evaluation of the fluid-elastic characteristics of the impeller was performed as Task 3. Since the concept utilized for determining the fluid-elastic characteristics is not widely known, it will be briefly described here:

• The dynamics of the system is studied in modal coordinates, hence
  
• The dynamic properties of the structure is established based on modal information (i.e. eigenfrequencies, mode shapes).
  
• The dynamic properties of the fluid is established based linearized fluid dynamics due the characteristics of the participating modes.
  
• Structure and fluid must be in equilibrium at any point in time, this can be expressed as an eigenvalue problem.
  
• The characteristics of the fluid–structure interaction problem is given by the solution(s) to the eigenvalue problem

Following the concept outlined above, a structural finite element model of the impeller was built and modal analysis of the model was performed in order to establish the structure dynamic characteristics of the impeller. With this information included as the structure dynamics model, a series of fluid-elastic eigenvalue analyses were performed. A picture of the structural dynamics model is shown at figure 4.

Fig. 4: Impeller as Modelled for ANSYS	Fig. 5: Impeller Modelled for LINFLOW (Note Wake Elements)

The LINFLOW unsteady fluid flow model of the system is shown in figure 5. The dark coloured elements in the model are the wake elements, which are attached for lift generating surfaces in LINFLOW. The picture is a graphical representation of the model, the actual analysis model include 3 of the blades only.
The reason for modelling 3 blades of the impeller only and not including turbine and stator is that experience shows that this is sufficient as long as long as only the fluid-elastic characteristics of the impeller is considered. Figure 6 shows steady flow vectors for the flow field at the operating point at which stability of the system has been investigated.
When studying the fluid-elastic characteristics of the impeller at the high pressure operational conditions it was found that there are fluid-elastic modes that pick-up energy for the fluid dynamics if excited. On the other hand the fluid-elastic modes involved most likely have a much larger frequency than the frequencies of pressure oscillations appearing in the flow field (the effect has so far not been studied). An example of a impeller fluid-elastic mode is displayed in figure 7.

Fig. 7: Impeller Modelled for LINFLOW	Fig. 8: Impeller Modelled for LINFLOW

Fig. 8: Damping Requirements for Critical Modes

Conclusions
A conclusion that was drawn by studying the fluid-elastic mode animations was that if there is a strong pressure pulse propagating through the impeller channels (even if the frequencies are lower then the frequencies of the modes), this pressure pulse will make the impeller structure deform radially in a way that will generate large strain levels in the region where cracks had been found to develop. As the crack grows in length the trailing end of the blade will become increasingly unstable and a faster failure will appear. By reviewing the fluid-dynamic behaviour in the converter seen in the performed CFD calculations, it could be concluded that there was indeed a large difference in the pressure pulse propagation between the designs that experimentally gave short fatigue life for the impeller and the impeller geometry that showed fatigue life above the requirements set in the specification.
A final remark is that, through the combined use of tests, structural FEA analysis, fluid dynamic CFD analysis, and LINFLOW fluid-elastic analysis it was possible to get an understanding of why one design work well and the others did not. It is also clear that without performing all 4 tasks (testing, structural FEA analysis, CFD, and FSI), the insight needed to arrive at a final design and have confidence in that the system is not fatigue sensitive would have been difficult.

For more information, please contact:
Ing. Giovanni Falcitelli - info@enginsoft.it
www.anker-zemer.com