Education & Training

www.nafems.org

Confessions of Two Non-R egister ed Analysts

Jan Christian Anker and Jari Hyvrinen of ANKER - ZEMER Engineering, Norway, revisit an analysis described in a previous issue of BENCHmark to explore the effect of aero-elasticity.
n the October 2003 issue of Benchmark, there is an article entitled Confessions of a Registered Analyst. The article deals with the estimation of the 20 first eigenmodes and eigenfrequencies of a 4x4 millimetre square membrane of thickness 0.003 millimetres. Material data is a Modulus of Elasticity of 121000 N/mm2, a Poissons Ratio of .21, and a mass density of 2.7e-9 Mg/mm3 (or - in units we like - 2.7e9 N x sec2/mm4). The boundary conditions are clamped edges. Among other things, the (anonymous) author compares the computed eigenfrequencies for the first (2122 Hz) and fifth (7761 Hz) modes with the analytical solutions for the eigenfrequencies for various discretisations. It is shown that excellent agreement with the analytical solution can be obtained, given that the modelling is performed with insight and care. For the simulation referred to, it is implied that the membrane is situated in absolute vacuum. Looking at its material properties and the dimensions, we felt that different eigenfrequencies and mode shapes will be observed if the membrane is supposed to be located in air and not in vacuum. Also, we thought it would be interesting to know what the eigenfrequencies would be if the air at one side of the membrane would be moving, say at a speed of 40m/sec. So, we decided to investigate the supposed frequency shifts. This article deals with our approach to the problem and our findings. Below, we briefly describe the tools we used for the study and the model(s). Then we discuss the results for air at zero velocity and for air at 40m/sec with a description of the physics of the problem. After that, we discuss related problems and concepts in LINFLOWTM. Finally, we give some examples of engineering problems where aeroelastic effects should be considered.

In the case at hand, we are dealing with an assumed fluidstructure interaction (FSI) problem. In LINFLOW, this class of problems is solved as an equilibrium problem formulated as an eigenvalue problem in general coordinates. The equilibrium problem can (somewhat inaccurately) be described as equilibrium between the dynamics of the fluid and structural eigenmodes. LINFLOW can utilize the eigenmodes from any structural analysis program; for the study reported here, we utilized ANSYS. The reason for choosing ANSYS (apart from its excellent reputation) was the tight integration that exists between ANSYS and LINFLOW. The structural model is based on a grid of 26 by 26 higher order solid elements. The reason for choosing solid elements and not shells was the assumption of fluid (i.e. air) acting on both sides of the membrane. LINFLOW will not handle this unless solid elements are utilized because of coincident elements.

R esults and Physics of the Problem

The eigenfrequencies obtained in ANSYS are shown in Table 1 together with the values in the reference and the findings for air (at rest) and air at 40m/sec. The results indicate that our feeling that the influence of air would be pronounced was correct. It should be mentioned that we did not fine-tune the structural model to compute the theoretical eigenfrequencies with great accuracy (with a finer mesh ANSYS came very close to the theoretical values) in order to reduce running time in LINFLOW to a few minutes. Figures 1 10 show the mode shapes as contour plots of the absolute displacements for modes 1 to 4 in vacuum, for modes 1 to 4 in air, for modes 1 4 in air at a velocity of 40 m/sec at one side of the membrane, and zdisplacement plots of modes 2 and 4 for the latter case. From the plots it can be seen that two mirror-symmetric modes at 4380 Hz in vacuum are missing (modes 2 and 3 are symmetric by rotation). The eigenfrequency of the first mode is 1856 Hz when accounting for air, a reduction of approx. 13% compared to being in a vacuum. Modes 2 and 3 show a reduced eigenfrequency of approx. 9%, mode 4 shows a reduction in frequency of approx. 7% in this case.

Tools
The main tool utilized for this study is LINFLOW. LINFLOW is based on the Boundary Element Method (BEM). The LINFLOW Boundary Elements have been designed to solve the velocity potential equation for steady and unsteady fluid dynamics. The unsteady flow is solved in the frequency domain assuming harmonic motion. The formulation is such that compressible as well as incompressible fluids may be accounted for. Page 2

April 2004

www.nafems.org

Education & Training

The changes in frequency are more pronounced at an air speed of 40m/sec. The first eigenfrequency is now 949 Hz, a reduction of 55% compared to vacuum. The fifth eigenfrequency has been computed to be 6815 Hz, a reduction of approx. 14%.

Mode 1 2 3 4 5

Theory (Vacuum) 2122 7761

Reference 1 (Vacuum) 2130 7948

Model in Vacuum 2139 4380 4380 6495

Model in Air 1856 3999 3999 6034 7382

Model in Air at 40m/sec 949 3226 3867 5768 6815

Table 1: Eigenfrequencies in [Hz]

The changes in frequency are more pronounced at an air speed of 40 m/sec. The first eigenfrequency is now 949 Hz, a reduction of 55% compared to vacuum. The fifth eigenfrequency has been computed to be 6815 Hz, a reduction of approx. 14%. Before discussing the results as depicted in Figures 1 14, we want to briefly touch on the physics we are simulating. We suppose that the physical significance of the eigenmodes of the structure in vacuum is clear to the reader. The physical significance of the aeroelastic modes of structure and air at rest might not be so obvious at a first glance, but if one considers them to be a spring-mass system (air or aeroelastic medium) connected to another spring- mass system (elastic structure), the picture should become clearer. However, an aeroelastic mode will usually be formed by a combination of multiple structural modes and the fluids corresponding reaction, a reaction that is dependant on the velocity of the medium. The combination of modes is in most cases not intuitive. The structural modes in vacuum as shown at Figures 1 4 are as expected (see confessions at the end). The combined modes shown at Figures 5 8 are similar to the structural modes, and shall not be discussed further. Interesting are the combined modes for air at 40 m/sec (from left to right). Here, the influence of the moving air is clearly visible. From Figure 13, it can be seen that the membrane (plate) takes the shape of a wave travelling in the direction of the air. The only conclusion we will draw from Figures 12 and 14, is that the shape of the combined modes can look very simple, albeit they are built up of rather complicated combinations of aeroelastic and structural modes.

Fig. 1: Mode 1; Vacuum

Fig. 2: Mode 2; Vacuum

Fig. 3: Mode 3; Vacuum

Fig. 4: Mode 4; Vacuum

Fig. 5: Mode 1; Air at v = 0 m/sec

Fig. 6: Mode 2; Air at v = 0 m/sec

Fig. 7: Mode 3; Air at v = 0 m/sec

Fig. 8: Mode 4; Air at v = 0 m/sec

April 2004

Page 3

Education & Training

www.nafems.org

The second example is a ventilation fan. The fan was to be optimized with respect to vibrations (and consequently with respect to noise). The computed aeroelastic response agreed extremely well with measurements, and the design of the fan blade could be successfully optimized. Figure 16 shows one of the aeroelastic modes.

Confessions
The authors have some confessions to make:
Fig. 9: Mode 1; Air at v = 40 m/sec Fig. 10: Mode 2; Air at v = 40 m/sec

We have not checked the analytical solutions for the eigenfrequencies of our model in vacuum; this should always be done (however, we have used various models to verify the results). If you look at the modal plots shown in Figures 2 and 3, you will notice that the modes are cyclic symmetric. Two other mode shapes can be generated by mirroring of modes 2 and 3, corresponding to 4 modes at 4 identical eigenvalues, however, only two eigenvalues were found. As stated above, an element mesh of 26x26 elements was utilized. Now Figures 17 and 18 show the modes for a 25x25 mesh. Here, we find modes that are symmetric

Fig. 11: Mode 3; Air at v = 40 m/sec

Fig. 12: Mode 4; Air at v = 40 m/sec

R elated Problems and Concepts in LINFLOW

A related problem that is easy for the reader to study is the following (the authors will assume no liability whatsoever with respect to the suggested study): Roll down the window in your car and mount a relatively thin piece of plastic in the opening (car at rest!). Now start driving at an increasingly higher speed and you will see that with increasing speed, it becomes easier to make the plastic vibrate. Other related problems are vibrations of pipes and ducts due to internal flow, flutter of panels, vibrations of submerged structures, vibrations of propellers, etc.

Fig. 13: Mode 2; v = 40 m/sec (z-disp.)

Examples
The tools and methods mentioned here can be used for many classes of problems, ranging from fluid flow via fluid-structure interaction to acoustics in flowing media. We will introduce only two examples. The first example is a ventilation duct. In Figure 15, one of the aeroelastic modes is shown. This mode is a combination of structural modes and the corresponding response of the fluid. Determining the aeroelastic modes is important for the prediction of noise due to instabilities. In this particular case LINFLOW predicted that a velocity of approx. 15 m/sec. would lead to vibrations, confirming what was observed. Page 4 April 2004

Fig. 14: Mode 4; v = 40 m/sec (z-disp.)

www.nafems.org

Education & Training

about the diagonal of the plate; a mode-shape can be created from the other by a rotation and will remain the same if it is mirrored about the diagonal. The difference in eigenfrequencies is not significant, a 26x26 mesh gives an eigenfrequency of 4380 Hz, the 25x25 mesh an eigenfrequency of 4369 Hz.

Fig. 15 Aeroelastic modes of Duct

Our computations in LINFLOW are hence based on eigenmodes which are not fully resolved. However, our experiences with LINFLOW indicate that the results should be fairly accurate, albeit with inaccurate eigenmodes. We have not undertaken any attempt to verify the results provided by LINFLOW. The methods used in LINFLOW have of course been verified against experimental and analytical results during the development off the tools. However, we think that we have demonstrated that you might get surprising results when you account for fluidstructure interaction when calculating eigenfrequencies.

R eference
1. Confessions of an RA, BENCHmark October 2003, NAFEMS
Fig. 16: Fan (Courtesy of ABB Ventilation)

Contact
Jan Christian Anker ANKER - ZEMER Engineering E [email protected]

The essential concepts in LINFLOW (related to this article) can be summarized as follows:

Fig. 17 Structural Mode #2, 25x25 Mesh

Boundary Element Method Inviscid and irrotational flow Velocity potential solution Compressible flow Incompressible flow Steady flow Harmonic flow Fluid-Structure Interaction Aeroelastic stability & response

Fig. 18:Structural Mode #3, 25x25 Mesh

April 2004

Page 5