Desktop version

Home arrow Engineering arrow Small Unmanned Fixed-Wing Aircraft Design. A Practical Approach

Source

Analyzing Decode-1 with Fluent

If a more detailed analysis of the fundamental aerodynamics is required, Fluent is able to model the entire airframe including fuselages, undercarriage elements, wheels, and deployed control surfaces, and even allows for propeller models to be included (e.g., as actuator disks). However, this increased capability comes at a significant extra cost in terms of the effort required to set

University of Southampton flight simulator

Figure 13.28 University of Southampton flight simulator.

up and carry out an analysis. It would be wasteful to commence such studies until a full XFLR5 analysis has been completed and any changes made to the design that might be indicated as being necessary from that analysis. At the very least, the main wing and elevator setting angles should be chosen using XFLR5 before commencing detailed work with Fluent. Note that we generally do not use Fluent to calculate stability derivatives; even at the most basic level, doing this requires six additional runs of the RANS code to carry out finite differencing with respect to the six degrees of freedom of flight and this cost is seldom warranted. Moreover, those working in this field now tend to use fully nonsteady RANS approaches, adjoint sensitivity calculations, or frequency-domain techniques which are even more involved to set up, see, for example, Mader and Martins [26]. If more precise estimates of stability derivatives than can be supplied by XFLR5 are required, it is generally more effective to use experimental-based approximations as available either through the USAF Stability and Control Digital DatCom datasets[1] or the UK ESDU datasets[2].

Because we only use Fluent for dealing with more complex geometries, we build meshes for these using the Harpoon mesher.[3] To get the best results, we try and use the k - m SST approach and so require a y+ value below 1 for the meshes. However, since our flight speeds are typically 30 m/s or less, the flight Reynold’s numbers are lower than for the previously presented validation trials, meaning that a half airframe model including wing, elevator, fin, control surfaces, fuselage, and undercarriage elements can typically be built with between 8 and 9 million cells, although careful y+ adaptation around control surfaces can take this up to 2 or even 3 times this. By comparison, a simple wing, elevator, and fin model typically requires less than 4 million cells. Figures 13.29 and 13.30 illustrate a typical mesh that includes fuselage and landing gear, but not control surfaces, together with the equivalent histogram of

Decode-1 mesh shown inside Harpoon along with wake surfaces and refinement zones

Figure 13.29 Decode-1 mesh shown inside Harpoon along with wake surfaces and refinement zones.

the y+ values. Figure 13.31 shows the Fluent convergence history for this model; note the steps in convergence caused by shifts from first to second order and following mesh refinements. Figure 13.32 shows the resulting Fluent lift/drag polars for the aircraft at the cruising speed of 30m/s (as both a full model and one with just the lifting surfaces used in XFLR5), together with results already obtained from the XLFR5 analysis and shown in Figure 13.24. Also shown here are experimental results recorded for the final aircraft in our wind tunnel, but corrected for blockage effects by using CFD results for a model of the aircraft in a model of the tunnel at 6° AoA (discussed further in the section on wind tunnel testing in Chapter 16). The AirCONICS models used to prepare these Fluent polars are shown in Figures 13.33 and 13.34.

Again, a NACA 23012 section has been chosen for the main wing and NACA 0212 for the elevator and fin elements. The setting angle of 2.53° calculated as being necessary for trimmed flight at 30 m/s by XLFR5 has also been used in these calculations. Notice that the slopes of the lift curves do not quite agree between the two codes and there are significant differences in the drag values as would be expected from the previous validation studies already described. Also, XFLR5 predicts lift all the way up to 13.2° without any fall-off in the lift slope, beyond which some sections start to stall, and the XFLR5 wing analysis code no longer returns data. For the full aircraft model, Fluent shows a roll-off in lift slope beyond 9° and a drop in lift beyond 12° with a peak Cl of around 0.97, while the measured results gave lift coefficients as high as 1.14 at 11 °. As before, it is clear that Fluent is being pessimistic and XFLR5 optimistic in regard to the actual lift performance, which lies somewhere between the two. Interestingly, the Fluent analysis for the wing only gives peak lift values that are very similar to those seen for the whole aircraft in the tunnel.

The measured drag is 50% higher than the Fluent predictions for fully attached flows, even though Fluent is generally pessimistic in drag predictions. This is because the actual aircraft includes a range of additional small drag elements and unknown surface roughness. In the previous spreadsheet analysis, a parasitic drag coefficient of 0.0375 was assumed. If this is

Fluent mesh on the center plane for the Decode-1 airframe k — a SST analysis at 30 m/s, together with resulting y+ histogram

Figure 13.30 Fluent mesh on the center plane for the Decode-1 airframe k — a SST analysis at 30 m/s, together with resulting y+ histogram.

Fluent convergence plot for Decode-1 whole aircraft model at 30 m/s

Figure 13.31 Fluent convergence plot for Decode-1 whole aircraft model at 30 m/s.

Polar plot for Decode-1 airframe at 30 m/s showing both Fluent and XLFR5 results for lift and drag

Figure 13.32 Polar plot for Decode-1 airframe at 30 m/s showing both Fluent and XLFR5 results for lift and drag. Those for Fluent include results for just the lifting surfaces and with the complete airframe fuselage, control surfaces, and undercarriage gear; those for XFLR5 show also the impact of adding a fixed parasitic drag coefficient of 0.0375.

AirCONICS model of Decode-1 lifting surfaces

Figure 13.33 AirCONICS model of Decode-1 lifting surfaces.

AirCONICS model of complete Decode-1 airframe with control surfaces, undercarriage, and propeller disk

Figure 13.34 AirCONICS model of complete Decode-1 airframe with control surfaces, undercarriage, and propeller disk.

added to the XLFR5 drag coefficient curve, reasonable agreement with experiment is reached, as shown in the figure. A value of only 0.025 is sufficient to correct the Fluent drag coefficient (calculated for the lifting surfaces only) to reach similar levels of agreement. Again, Fluent is somewhat pessimistic in its drag predictions. Fluent does, however, allow investigations of

Streamlines colored by velocity magnitude around the complete Decode-1 airframe with deflected ailerons

Figure 13.35 Streamlines colored by velocity magnitude around the complete Decode-1 airframe with deflected ailerons.

aspects such as lift generated by the fuselage; a well-designed fuselage can augment lift by 5% or more, and Fluent studies can reveal these benefits.

The next set of calculations that may be undertaken with Fluent concerns the behavior of the various control surfaces that will be required on the UAV. Typically, moving surfaces are added to give pitch, roll, and yaw control and to augment lift at take-off and landing. On a conventional aircraft, these take the form of elevators, ailerons, rudders, and flaps, respectively. It is clearly useful if the size of these elements can be checked prior to final airframe construction, as initial estimates will have been based on one of the many standard aircraft design texts. As already mentioned, we generally opt for large surfaces to maintain control authority, particularly during the low-speed take-off and landing phases. As already seen, AirCONICS can be used to add the various moving control surfaces to the model, and these can be at arbitrary angles of deflection, and the resulting geometry can then be analyzed in Fluent, though care has to be taken to mesh the zones between the moving elements and the main lifting surfaces in sufficient detail and to ensure that the resulting flow field remains attached to the surfaces; it is unwise to rely on steady-state RANS results for separated flow studies. When carrying out such work, it is often sensible to study just the lifting surfaces in the first instance so as to reduce the mesh sizes being used: for roll moment calculations, a half model is no longer appropriate and so mesh sizes immediately double. Figure 13.35 shows the velocity streamlines past a full Decode-1 model with deflected control surfaces.

It should be noted, however, that the most desirable result from such analysis would be to find the peak lift that can be obtained in the landing configuration with flaps fully deployed. Unfortunately, even a finely meshed model in Fluent is unlikely to be able to do this very accurately. For commercial airliners, it is still common practice to validate the high-lift design using wind tunnel tests. However, whether the effort expended in building complex and highly refined CFD models for small UAVs is really worthwhile remains debatable. We have access to large-scale wind tunnels and rely on these to confirm the high-lift behavior prior to flight trials. Nonetheless, once one becomes familiar with the behaviors of XLFR5 and Fluent (or other equivalent codes) and how best to use them, it is relatively straightforward, albeit sometimes expensive, to validate a good number of the earlier design decisions taken in the spreadsheet analysis prior to detail design and airframe construction.

  • [1] http://www.pdas.co./datcom.html.
  • [2] http://www.esdu.com. - the former Technical Department of the Royal Aeronautical Society
  • [3] http://www.sharc.co.uk/.
 
Source
Found a mistake? Please highlight the word and press Shift + Enter  
< Prev   CONTENTS   Next >

Related topics