Beruflich Dokumente
Kultur Dokumente
Jari Hyvrinen Anker-Zemer Engineering AB, Sweden Jan Sderkvist Colibri Pro Development AB, Sweden
Abstract The interaction between fluids and structures is often larger for smaller structures. This is due to the use of narrow gaps and to scaling making the fluid properties more pronounced. MEMS devices are characterized by small dimensions and the dynamic behavior of their moving parts can therefore noticeably be affected by the surrounding media (e.g. air). The fluid-structure interaction (FSI) involves several mechanisms for how the dynamic properties, such as resonant frequencies and damping, can be affected. Some of these can be explored by iterating between the fluid and structural domains (the sequential method) while others require addressing the two domains simultaneously (the direct method). The complication is that no CPU-efficient and affordable stand-alone software packages can address the direct interaction for FSI. A CPU-efficient procedure for handling both the sequential and direct interaction effects is outlined in this paper. The procedure uses the finite element software ANSYS complemented by the aeroelastic boundary element 3rd party software LINFLOW that can be accessed from ANSYS menu system. ANSYS is used primarily for the structural part of the analysis and for some of the FSI-effects. LINFLOW is used to handle the direct FSI. The paper includes several examples to illustrate the procedure and how the fluid can affect the mechanical system. The examples explore the general dynamic behavior, the influence of holes in plates, and the influence of an external wind. The paper also includes a description of how the FSI can affect the dynamic properties of the system.
based on these approximations [1, 2]. Very few MEMS-related publications questions the accuracy of the results obtained with these approximations.
Background simulations
Modern simulation tools are becoming more frequently used also for MEMS components. However, it is still a complicated task to study the fluid-structure interaction (FSI) if most of its aspects are to be accounted for. In addition, assumptions made when developing the fluidic parts of classical aeroelastic simulation tools used by the aeronautical industry, such as the Doublet Lattice Method (DLM) [3], do not apply to MEMS type applications due to that the underlying approximation in the method and to the small dimensions used in MEMS change the relative importance of FSI-effects. The FSI is an area into which ANSYS puts much effort [4]. A convenient procedure for linking the structural and fluid domains has been implemented in ANSYS that is based on iterating between the two domains. In addition, ANSYS has added elements that allow the calculation of squeeze film and slide film damping. Many of the FSI-effects can be handled with ANSYS today and fairly accurate predictions can be made. Nevertheless, there are limitations. Some of these will be more pronounced for small dimensions and when the interaction is stronger. The limitations include: o o The CFD-calculations are very computer-intensive which results in very long simulation times for the implemented FSI iteration procedure also for simple models. The iteration procedure treats the two domains separately. This means that effects that depend on a very tight interaction between the two domains will not be caught by ANSYS.
A complication is that no CPU-efficient and affordable stand-alone software packages can address these limitations. However, new software packages emerge that enables procedures that expand the possibilities. It is now possible to study both domains simultaneously and more FSI effects can be explored. One can get the best of both worlds if combining tools, for example by continuing to use ANSYS and adding help from new software packages. The procedure proposed by us is to use the 3rd party programs LINFLOW as a help to ANSYS. An advantage is that LINFLOW can be run from within ANSYS just as if it was part of ANSYS. LINFLOW is based on the boundary element method (BEM) which results in very fast simulation times compared with conventional FSI approaches [5, 6]. The procedure uses ANSYS to compute the structural fluid-free eigenmodes. The aeroelastic frequencies and the fluids contribution to the damping are thereafter calculated in LINFLOW using the ANSYS model in modal coordinates as the description of the structure dynamics. This step accounts for the simultaneous interaction between the fluid and structural domains. The data from LINFLOW can thereafter be transferred back to ANSYS if further simulations are needed, such as the calculation of the squeeze film and slide film effects. The implementation used in LINFLOW is based on a small-disturbance linearization of the fluid dynamics. The solution is obtained in the frequency domain. Few limiting approximations have been made in the implementation apart from these. A more detailed description of the implementation is found in references [5-8]. Several general types of geometries that are used in MEMS components have been used in this presentation to illustrate the effects of the fluid-structure interaction. For example, a membrane illustrates the general dynamic aeroelastic behavior, an inertial sensor illustrates the effect of holes in the structure, and a micromirror illustrates the effect of air and liquid flow and is used as yet another application example of the usefulness of ANSYS+LINFLOW.
Mechanism: Energy dissipated via the acoustic sound wave. The energy is transferred away from the moving via collisions between the molecules participating in the sound wave. Consequences: The damping is affected. It often increases, but the opposite can also occur. Resonance frequencies are also affected, but to a lesser degree. The effect is often non-linear, and a phase difference between the motion of the mechanical part and the medias counter-pressure can be expected. Procedure for predictions: This effect is most accurately and efficiently predicted using a direct interaction procedure in which the fluid and structural domains are evaluated simultaneously, preferably in the frequency domain. This is due to the tight energy transfer in both directions between the two domains. This type of loss may also be analytically approximated for simplified systems as described in reference [9]. Importance: Important not only for acoustic and sonar applications. It can also introduce visible undesired behavior. Moving media that cause energy dissipating and transfer Mechanism: Media that flow away from the structural region will dissipate energy from the structure in the form of kinetic energy. The effect can change the nature if the media is moving, e.g. due wind. Consequences: Both the resonance frequencies and the damping are affected. It often increases, but the opposite can also occur. Resonance frequencies are also affected, but to a lesser degree. This effect is non-linear. This effect can transfer energy between different parts of the structure which can result in large negative consequences, such as, instabilities in which the vibration amplitude increases in an uncontrolled avalanche way. Procedure for predictions: This effect is most accurately and efficiently predicted using a direct interaction procedure in which the fluid and structural domains are evaluated simultaneously, preferably in the frequency domain. This is due to the tight energy transfer in both directions between the two domains. Importance: This effect can cause very noticeable undesired behavior.
same time as the fluid dynamics is solved. The result from LINFLOW are the aeroelastic resonance frequencies and mode shapes (will be a mode superposition of the structural modes), the fluids contribution to the damping. The pressure and velocity field in the fluid can thereafter be determined via a subsequent flow-harmonic analysis, if desired. The data from LINFLOW can thereafter be transferred back to ANSYS if further simulations are needed, such as the calculation of the squeeze film and slide film effects. Thus, LINFLOW will solve the structural and fluidic domains simultaneously in contrast to conventional approaches that handle the domains sequentially. The combined fluid-structure equation system solved by LINFLOW is in aeroelasticity called the aeroelastic equation of motion. In summary, the procedure is as follows: 1. 2. 3. 4. Perform a modal analysis in ANSYS with no air present eigenmodes and resonance frequencies Transfer the eigenmodes to LINFLOW and use them to describe the structural part via modal coordinates Perform a modal analysis in LINFLOW with air (no viscosity) mode coordinates (complex), aeroelastic eigenmodes, and aeroelastic resonance frequencies Perform a harmonic analysis in ANSYS around each aeroelastic resonant frequency using the mass-normalized mode shape as a mechanical load and ANSYS squeeze film and/or slide film elements activated (coupled modes may require superimposing the results for simulations on the two phase components of the mode coordinates) viscous damping and squeeze stiffness at the correct frequency and mode shape Perform a full harmonic analysis in LINFLOW using the calculated viscous damping as modal damping aeroelastic response that includes both the viscous and aeroelastic damping (only a small modification of mode shapes and resonance frequencies are expected) Repeat steps 4 and 5 if the added viscous damping has a noticeable influence on the aeroelastic system
5.
6.
It is important to have short simulation times in the early stage of the design phase of a project. One need rapid feedback from tests, including from simulations, so that one can chose the optimal design route early on before spending large amounts of resources on fruitless designs. The analysis time required in the aeroelastic analysis described in this paper will depend on the number of structural modes that need to be included to describe the structure dynamics and also on the number of aeroelastic modes that need to be converged. A typical example for the simulation times involved was that it takes 87 seconds (on an ordinary x MHz PC) when using the ANSYS/LINFLOW combination for predicting the dynamic characteristics for the case 3 model presented below. The case 3 model included about 2000 boundary elements and 884 structural solid elements, and 5 structural modes was included when converging the first aeroelastic mode. The case 1 and case 2 models require even less resources. The procedure has been validated in many industrial projects over several years. This includes both macrosized geometries (aircraft parts, missiles, race cars, fans, etc) [13, 14] and micro-sized geometries [6]. The results are therefore considered to be in agreements with reality.
gyroscopes. In such applications, the media (often gas) is allowed to flow out of the gap-region, in contrast to pressure sensors for which the gap-region forms a closed cavity. The basic geometry chosen for illustrating the presented simulation procedure consists of a membrane/plate. Three different clamping conditions are used in order to address different types of motions used for different application areas: o o Case 1: Membrane with fixed clamping on all sides. Typical membrane deflection. Case 2: Membrane hinged on two of its sides via springs. Tilting motion around the line of the springs. Similar geometries are used for electrostatically excited micro-mirrors. This example illustrates that the presented procedure equally well can be applied to structures with other vibration modes (torsion in this case). Case 3: Membrane hinged on all four sides via springs. The deflection of the membrane approaches a rigid-body motion. Similar geometries are used for capacitive inertial sensors.
Each of the three geometries is used to study a different type of behavior, from the simple response characteristics to more design-related topics. The first case explores the dynamic response in general. The second case focuses on how the mode shapes are changed when the surrounding media moves, for example due to the wind from a cooling fan. The third case focuses on the influence of holes in the moving plate, holes that can be needed for successful under-etch or to reach the desired response. The dynamic response has primarily been characterized via changes in the resonance frequencies and in the mediums contribution to the damping. The media is assumed to be gas (air: po=1 atm, vsound=343 m/s and =1.4). The membrane is assumed to have elastic modulus 121 GPa, poissons ratio 0.21 and =2700 kg/m3 [15]. The simulations could equally well have been carried out for structures submerged into liquids. Although the illustrative geometries are chosen to be very simple there is no limitation in the presented procedure as to the complexity of the geometry.
Figure 1a. The simulation model (left) and the fundamental mode shape in vacuum (right). The 3 m thick plate has a side-length of 4 mm. The fixed surface under the moving plate is larger than the plate so to avoid edge effects of the fixed surface.
Figure 1b. The dependence of the fundamental resonant frequency on the gap to the fixed surface under the moving plate. The large gap-dependency in Figure 1b shows that the squeeze film damping plays an important role for the membranes behavior. It has therefore been motivated to carry out a harmonic analysis using ANSYSs element FLUID136 to determine the squeeze film effect. This analysis was done after that LINFLOW determined the aeroelastic resonance frequencies so that the harmonic sweep that is used for FLUID136 to evaluate the squeeze film effect could be done at the correct frequency range.
Figure 1c. The harmonic response near the fundamental resonant frequency without (left) and with (right) squeeze film damping activated. The gap under the membrane is 150 m. The surrounding can influence the behavior in more ways. For example, one can get a strong change in behavior if the gas is moving, for example due to a wind. This effect is illustrated in Figure 1d. Here, a gas flow has been added on top of the membrane. The effect illustrates that the wind initiates a mass-transport of energy from one part of the structure to the other, and that the resulting counter-pressure can be said to make the membrane weaker. An even stronger wind than used in Figure 1d would have resulted in the system being statically unstable.
Figure 1d. The dependence of the fundamental aeroelastic resonant frequency on the speed of the airflow over the membrane (left) and the velocity profile (right).
micro-mirror array if the arrays package is not covered with a protecting glass lid. This flow can influence the dynamic response, as explored below. The general behavior of a micro-mirror can be illustrated by clamping the membrane in case 1 via springs at two opposite points (see Figure 2a). The resulting plate can rotate along the axis of the springs, and is fairly stiff compared with the springs. The plate is surrounded by gas, and a fixed surface is placed under the plate to create a thin air-gap region. A velocity has been added to the surrounding gas. This is equivalent to saying that the structure is moving relative the gas. The structural resonance frequencies are calculated using ANSYS. The first five modes are shown in the left column is Figure 2. The modes are well-behaved and symmetric. The aeroelastic frequency was thereafter calculated in LINFLOW using the ANSYS model in modal coordinates as the description of the structure dynamics. The mechanical structure was moved with a velocity of 10 m/s in the LINFLOWsimulations. This created a wind that affects the acoustic field generated by the vibrating mechanical structure. The wind will transfer acoustic energy between different parts of the structure, and it can be expected that this will affect the resonant frequencies, the mode shapes, and the damping. The right column in Figure 2b shows the first four stable modes when the aeroelastic effects and the wind is taken into account. Noteworthy is how much the resonant frequencies change when the effect of the moving air was accounted for. The main part of this change is due to the movement of the air, and only to a smaller part to the damping in the air gap. The air gap was, on purpose, made large to illustrate that this effect exists also when the gap does not exist. The aeroelastically calculated modes are a linear combination of the structural modes. The interaction between the gas and the structure will determine the mixing. The wind contributes strongly to the mixing, and the aeroelastically calculated modes differ substantially from the structural as can be seen when comparing the two columns in Figure 2b. The wind will also cause the aeroelastically calculated modes to be non-symmetric in the direction of the wind. This is due to the wind carrying acoustic energy from one side of the structure (left) to the other (right). The energy will be fed back to the structure as an acoustic pressure. The location and phase of this pressure force can be such that it increases the amplitude of the mechanical vibration, and an avalanche process is initiated that will cause this vibration mode to be aeroelastically instable. In fact, this is the case for one of the modes calculated in this example, which explains why one frequency is missing in the right column in Figure 2b. Animations of the aeroelastic mode shapes clearly show that the modes consist of a mode-superposition of the structural modes. Different parts of the structure will move with different phase angles.
Figure 2a. Geometry used for studying the micro mirror. The 3 m thick plate has a side-length of 4 mm. The width of the 0.8 mm long springs is one 7th of the plates width. The gap to the fixed surface under the mirror is 150 m. The fixed surface under the moving plate is larger than the plate so to avoid edge effects of the fixed surface.
Figure 2b. Left column: The first five structural resonance frequencies for the structure in Figure 2a. Right column: The first four aeroelastic resonance frequencies for the structure in Figure 2a when the structure is affected by a 10 m/s wind. The first mode is aeroelastically unstable and is not included
(no hole)
(one hole)
(four holes)
(nine holes)
Figure 3. The fundamental mode shape for a simplified inertial sensor moving vertically. The 3 m thick plate has a side-length of 200 m. The width of the 100 m long springs is one 7th of the plates width. The gap to the fixed surface under the mirror is 1 m. The diameter of the holes is 20 m, and the cc-distance between the holes equals the distance between the outermost hole and the plates edge. The fixed surface under the moving plate is larger than the plate so to avoid edge effects of the fixed surface. The surrounding air has no background velocity (no fan). Several observations can be made from the tables below: o The simulations focus on the frequency shift. Viscous effects have a small influence on this parameter, and it has therefore not been necessary to include the last iteration step in the proposed procedure, i.e. to iterate back to ANSYS to determine the squeeze film effects via the FLUID136 element. The damping contributions listed in Tables 3a-3c do therefore not include the viscous effects. Including the last step would not have caused any complications. The presence of the gas reduces the fundamental resonant frequency, and especially for large plates (Table 3c). Introducing holes reduces the influence of the gas (frequency shift and aeroelastic damping). The influence of the gas is reduced more if the area of a hole is split on many small holes instead of a single large hole (the largest hole in Table 3b has an area that falls between the 4 and 9 holecases in Table 3a).
o o o
Opening up a small hole will immediately reduce the aeroelastic damping, but the reduction is not increased that much if the hole is increased until the hole starts to approaches the plates sidelength (see Table 3b). This is due to that the first hole allows the air to flow in towards the center and a boundary zone at a radial distance out from the hole is creates outside which the gas continues to flow outwards when the plate is pushed down while the gas inside this boundary reverse direction and flows towards the hole at the center. The location of this boundary zone does not change much if the hole is made larger, until the hole is large. The aeroelastic frequency shift behaves in the same way. The same effect applies also to the viscous damping. Larger plates have a smaller damping contribution from the air (Table 3c). This is partly an illusion since the damping listed in the tables has a frequency dependency. The behavior is expected to be slightly different had the viscous damping been included. The procedure used in ANSYS does not account for these aeroelastic frequency shifts, and the structural mode coupling as pointed out before. It is therefore natural that the damping calculated by ANSYS would differ from that given in Table 3c, had only ANSYS been used.
Table 3a. The dependence of the dynamic properties on the number of holes. Dimensions are as described in the legend of Figure 3. Number of holes Structural frequency [kHz] 117.2 117.3 117.8 118.5 Aeroelastic frequency [kHz] 113.9 115.1 116.5 117.8 Aero. minus struct. frequencies [kHz] -3.3 (-2.8%) -2.2 (-1.9%) -1.3 (-1.1%) -0.8 (-0.7%) Aeroelastic damping contribution -0.00087 -0.00061 -0.00036 -0.00020
0 1 4 9
Table 3b. The dependence of the dynamic properties on the size of the hole (the one hole-case in Table 3b). Dimensions are as described in the legend of Figure 3 except for the diameter of the hole. Hole diameter [m] no hole 10 (one hole) 16 (one hole) 20 (one hole) 50 (one hole) Structural frequency [kHz] 117.2 117.2 117.3 117.3 118.2 Aeroelastic frequency [kHz] 113.9 114.7 115.0 115.1 116.7 Aero. minus struct. frequencies [kHz] -3.3 (-2.8%) -2.5 (-2.1%) -2.3 (-2.0%) -2.2 (-1.9%) -1.5 (-1.3%) Aeroelastic damping contribution -0.00087 -0.00063 -0.00062 -0.00061 -0.00042
Table 2c. The dependence of the fundamental resonant frequency on the size of the hole (the no hole-case). Dimensions are as described in the legend of Figure 2 except for the plates side-length (the width of the springs is increased accordingly). Size of plate (no holes) [m2] Structural frequency [kHz] Aeroelastic frequency [kHz] Aero. minus struct. frequencies [kHz] Aeroelastic damping contribution
200 200 500 500 1000 1000 2000 2000 4000 4000
-3.3 (-2.8%) -1.38 (-7.32) -0.735 (-15.5%) -0.334 (-28.2%) -0.125 (-42.1%)
Conclusion
The fluid structure interaction plays a key role in many industrial MEMS applications, for example due to the use of narrow gaps. Some FSI effects cannot be addressed with conventional FEA software, and additional tools and procedures are needed. The conclusion is that best results are reached if a combination of tools is used. For example, ANSYS is well suited for the structural domain and for the viscous squeezed and slide film damping, while the fluid dynamic and its intimate coupling to the structural part (aeroelasticity) are well handled with a BEM program such as LINFLOW. The proposed procedure combines these tools. The examples illustrate that the procedure is useful for a large range of MEMS geometries. The simulation times are quick, and the results agree with expected values. The examples also illustrate some results that are hard to predict without accounting for the full aeroelastic effect. For example: The approximations made in ANSYS and other similar software are valid in many of the cases, but that there are cases for which the errors in the results are noticeable Frequency shifts for resonating plates are observed that are not predicted using ANSYS alone The flow velocity of the surrounding air influences the performance, especially for structures with holes
That LINFLOW and its results can be accessed from within ANSYS creates an interesting possibility to perform Multiple Object function optimization. ANSYS own optimization module and the probabilistic design can be used. External programs, such as modeFRONTIER can also be used to perform parameters studies.
References
1. 2. 3. 4. 5. 6. 7. 8. J. Mehner, S. Kurth, D. Billep, C. Kaufmann, K. Kehr and W. Dtzel, Simulation of gas damping in microstructures with nontrivial geometries, Proc. MEMS 98, 98CH36176 (1998) 172-177. J.B. Starr, Squeeze-film damping in solid-state accelerometers, Proc. IEEE Solid-State Sensor and Actuator Workshop, Hilton Head Island, SC, U.S.A., IEEE: 90CH2783-9 (1990) 44-47. M. Blair, A Compilation of the mathematics leading to the doublet lattice method, US Air Force Book Series, WL-TR-92-3028, 1992. www.ansys.com www.linflow.com J. Hyvrinen and J. Sderkvist, Aeroelasticity The interplay between fluids and solids, J. Micromech. Microeng, 11 (2001) 416-422. J. Hyvrinen, Generalized methods for aeroelastic analysis (thesis), The Royal Institute of Technology, Stockholm, Sweden, ISSN 1651-7660. LINFLOW Theory manual, Anker-Zemer Engineering, Sweden.
9.
R.B. Blevins, Flow induced vibration, Krieger Publishing Company, 2nd ed. reprint 1994, ISBN 0-89464-823-3.
10. E.H. Dowel, et al, A modern course in aeroelasticity, 2nd Ed., Kluwer Academic Press, 1989. 11. J.J. Blech, On isothermal squeeze films, J. Lubrication Technology, 105 (1983) 615-619. 12. ANSYS Theory Reference, ANSYS, Inc., USA. 13. S. Mancino and A. Sollo, Aeroelastic response comparison of two different A/C tail configurations by using LINFLOW code, Proc. Virtual prototyping Today: Industrial Impact and Future Trends (EnginSoft 2002 Conference and Users Meeting), Bergamo, Italy (3-4 Oct. 2002). 14. J. Hyvrinen, Aeroelastic analysis of fan system dynamics, Proc. CADFEM Users Meeting 2005, Bonn, Germany (Nov. 2005). 15. Anonymous author, Confessions of a Registered Analyst, Benchmark (NAFEMS) (Oct. 2003). 16. Market analysis for microsystems 1996-2002, Nexus Office c/o FhG-ISiT, 1998. 17. L.J. Hornbeck, Digital light processing: A new MEMS-based display technology, Techn. Digest IEEJ 14th Sensor Symp., Kawasaki, Japan (June 4-5 1996) 297-304. 18. A.A. Seshia, M. Palaniapan, T.A. Roessig, R.T. Howe, R.W. Gooch, T.R. Schimert, and S. Montague, A vacuum packaged surface micromachined resonant accelerometer, J. Microelectromechanical Systems, 11(6) (2002) 784-793.