External stores mounted on aircraft generate loads which need to be estimated before first takeoff. These loads can be measured in a wind tunnel but since the possible store configurations are basically endless, testing them all is neither economically feasible nor time efficient. Thus, scaling based on geometrical similarity is used. This can, however, be a crude method. Stores with similar geometrical properties can still behave in different ways due to aerodynamic interference caused by adjacent surfaces.
To improve the scaling performance, this work focuses on investigating two CFD codes, ADAPDT and Edge. The CFD simulations are used to derive the difference in aerodynamic coefficients, or the Δ-effect, between a reference store and the new untested store. The Δ-effect is then applied to an existing wind tunnel measurement of the reference store, yielding an estimation of the aerodynamic properties for the new store.
The results show that ADAPDT, using a coarse geometry representation, has large difficulties predicting the new store properties, even for a very simple store configuration on the aircraft. Therefore it is not suited to use as a scaling tool in its present condition. Edge on the other hand uses a more precise geometry representation and proves to deliver good estimations of the new store load behavior. Results are well balanced and mainly conservative. Some further work is needed to verify the performance but Edge is the recommended tool for scaling.
Aerodynamic scaling is a difficult topic and most of the research done in the field is focused on scaling wind tunnel data from small scale models to full-size applications. Scaling data from one full size application to another with similar appearance is a sparsely investigated field. Applying a single scale factor for an entire range of α or β is not sufficient. It is easy to illustrate the difficulties associated with this form of scaling. Figure 4 shows a comparison of wind tunnel measured CMZ coefficients for Store1 and Store2.
Panel methods can be used with both two and three dimensional body representations and ADAPDT uses the simpler, two dimensional representation. Successful attempts to integrate three dimensional representation in ADAPDT have been made but they are not yet implemented in the available code. In ADAPDT, bodies are approximated with thin sheets in one plane or in a cross configuration, see Figure 5. Each sheet is then divided into a number of smaller panels.
Two major designs were tested for the stores; body-shaped sheet design (BSD) and rectangular sheet design (RSD), shown in Figure 6. The RSD was based on using the smallest possible rectangle to fit the planform. The RSD is currently used at Saab Aerosystems for various simulations with good results. The BSD was chosen to evaluate if using the contours of the body would improve upon the results of the RSD design.
Since no previous dwfSumo models existed and the software lacks the ability to import CAD geometry, every body had to be modelled, including the Gripen fighter, see Figure 8. Fine details were excluded to keep the mesh size to a minimum. Also, the vertical stabiliser was removed, since its effect on the store was assumed to be negligible. The meshes were generated focusing on refinement in the region of interest; close to the store.
The mesh size was approximately 700,000 nodes, varying with store complexity. Mesh independence was tested using the Edge ’hadaption’ feature which refines the mesh based on gradients in the flow parameters. For more information regarding this feature, please view the Edge manual. Figure 9 illustrates the mesh representation in the vicinity of the store.
Scaling was less accurate for yaw and roll moments, Figure 19. The slopes of the curves were overestimated in both cases but Edge picked up the opposite direction of slope for the yaw moment. The pitch moment, Figure 20, was well predicted in shape but slightly underestimated, mainly for positive β. All β -sweep scaling charts are available in appendix I.
The results from the scaling tests were very different for the two CFD methods. Edge performed relatively well for most aerodynamic coefficients, delivering loads resembling the wind tunnel measured results. ADAPDT performance on the other hand left much to be desired in terms of both shape and magnitude for all but one coefficient. Looking at the time consumed for the total scaling process, the two CFD methods were comparable. The slower simulation times of Edge were compensated by a long and complicated process for adjusting the ADAPDT models to fit free flight aerodynamic data.
Based on the results presented in this report, there is little reason to choose the ADAPDT software as the scaling tool for the Loads Department at Saab Aerosystems. ADAPDT predicted only one out of five coefficients well. The results for the other coefficients were unpredictable and exaggerated. However, it would be valuable to repeat the scaling performance test if three dimensional geometry was implemented. Currently, ADAPDT shows no advantages over Edge. Scaling results are less accurate and the total amount of time for the scaling is basically the same due to the long process of model adaption.
Edge results were consistent and proved to deliver fairly accurate results also in raw CFD values. Scaling performance was good throughout the tests. Cases with less accuracy were mainly conservative and never overestimated by more than a factor 2.5 at maximum value.
Edge will be a good choice for a scaling tool if some time is spent verifying the results for more store configurations. Overall, this scaling method will most likely improve the accuracy of the scaling performed at Saab Aerosystems and, more importantly, make the scaling process more efficient.
SUMMARISING THE METHODOLOGY
Four main steps are needed to perform the scaling using Edge. Modelling, where the geometries are created. Preprocessing, which generates the mesh and prepares it for the simulation. The actual simulation and then postprocessing, where the data is managed. See Figure 21 for an illustration of the work flow, including an iteration for investigating the mesh independence.
Source: Linköping University
Author: Christian Spjutare