Document
Wind Tunnel Data Fusion and Immersive Visualization: A Case Study
Kurt Severance Paul Brewster Barry Lazos Daniel Keefe NASA Langley Research Center NASA Langley Research Center NASA Langley Research Center Brown University k.severance@larc.nasa.gov p.f.brewster@larc.nasa.gov b.s.lazos@larc.nasa.gov dfk@cs.brown.edu
ABSTRACT 2. WIND TUNNEL EXPERIMENTS
This case study describes the process of fusing the data from The experiments were conducted in the Basic Aerodynamics several wind tunnel experiments into a single coherent Research Tunnel (BART) [9] at the NASA Langley Research visualization. Each experiment was conducted independently and Center. The model was a generic 4-wheel landing gear was designed to explore different flow features around airplane configuration with components scaled to 31% of the main landing landing gear. In the past, it would have been very difficult to gear of a Boeing 757 (Figure 1). The model was simplified by correlate results from the different experiments. However, with a eliminating detailed components such as tubing, wheel covers, single 3-D visualization representing the fusion of the three braking mechanisms, bolt heads, etc. Three experiments were experiments, significant insight into the composite flowfield was performed on this model: observed that would have been extremely difficult to obtain by studying its component parts. The results are even more compelling when viewed in an immersive environment.
CR Categories : I.3.7 [Computer Graphics]: Three-Dimensional Graphics and Realism – color, shading, and texture; I.3.8 [Computer Graphics]: Applications.
Keywords: Data fusion, Photogrammetry, Line Integral Convolution, Reconstruction, Oil flow, Particle Image Velocimetry, Wind tunnel, Landing gear, Texture mapping, Image-based rendering, VRML.
1. INTRODUCTION
Experimental wind tunnel data come in various forms and with varying quality. Unlike computational fluid dynamics (CFD), where the values at each grid point are calculated, wind tunnel experiments use various sensors to try to directly and indirectly measure values in certain areas. Measuring the complete flow around an object, similar to a CFD solution, is often impractical, Figure 1. Inverted landing gear model in the wind tunnel.
if not impossible. The data could be quantitative information like Flow direction is from left to right.
pressure on the surface, or qualitative information, like pictures or images of flow patterns. It can also be difficult to correlate data 1. A mixture of oil (kerosene) and titanium dioxide powder was taken from different measurements of the same object, even at the applied to a single wheel. The tunnel was rapidly brought to same flow conditions. This case study gives an example of fusing test speed, and shear lines formed in the mixture as air three different kinds of image and point measurements together flowed over the wheel. Photographs of the established flow into a single 3-D flow visualization that can be displayed in an patterns were then obtained from multiple angles using a immersive environment.
digital camera. A sample photograph of the front wheel is shown in Figure 2. This same wheel was cleaned and Aircraft noise has been one of the key difficulties in building mounted on the rear axle, the mixture was reapplied, and the an efficient civil air network. Because of the noise, airports are wind tunnel was rapidly brought to the same flow conditions.
often built farther from urban populations than is practical. Next After the mixture was dry, similar images were taken of the to the engine, the landing gear during final approach can be one of established shear stress patterns on the rear wheel.
the largest generators of noise. Noise is created by pressure fluctuations in the fluid flow around the gear. Three experiments 2. Surface pressures were measured using a line of pressure were conducted to analyze the mean flow of air around the taps along the wheel periphery from the inside center (near landing gear of a Boeing 757. This information may be used to the axle) to the outside center. Pressure readings over most of validate computational results and highlight areas where more the surface of the wheel were then obtained by rotating the detailed time dependent measurements are important. The wheel about its axle in 2-degree increments in a stepwise experiments are explained in detail in [8], but a brief explanation fashion. At each step, 30,000 data samples were taken per of pertinent parts of the experiment follow.
sensor. These samples were averaged together to compute the mean pressure at each point.
3. Digital Particle Image Velocimetry (DPIV) [7] was used to determine the mean velocities in a vertical plane bisecting the inline wheels. This procedure involved illuminating flow particles with a laser light sheet and recording their instantaneous positions using two photographs taken Given eight oil flow images of one wheel, our goal was to microseconds apart. One hundred image pairs were used to combine these photographs into a single 3-D visualization. We determine the mean velocities at 160 locations that formed followed a procedure very similar to the photogrammetric the data plane (Figure 4). reconstruction described in [10, 11] which created 3-D oil flow and light sheet visualizations from 2-D video images. In fact, the camera calibration and resection procedure is the same, and is shown schematically in Figure 5. This procedure is based upon the collinearity condition in which the object point, image point, and camera lens center all lie on a straight line. Well-placed and accurately measured fiducial marks are a key input into the process. Further details on this procedure can be found in [2, 5].
The result of the camera calibration and resection process is the X,Y,Z position and Ω,Φ,Κ orientation angles of the camera.
This procedure was performed on each of the eight camera shots.
Given the camera parameters and a known geometry, the procedure in [11] can be used to perform a forward projection from a single camera onto a 3-D surface. However, when working with multiple cameras, we found it more effective to implement a backward projection scheme. This approach is similar to the image-based rendering techniques in [4] and the Figure 2. Photograph of the oil flow on the outside of the wheel.
multiple camera mapping technique in [6]. The algorithm examines each quadrilateral of the known surface geometry
3. VISUALIZATION METHODS
(described as a PLOT3D structured grid) and determines which The geometry of the landing gear model was well known before camera had the best view of that facet, based upon the surface the experiments. A PLOT3D structured grid of 50x181, or 9050, normal (Figure 6). Next, the 3-D vertices of that quadrilateral are points was constructed based upon the as-built measurements of back-projected toward the selected camera to determine the the wheel. The points on the grid coincided with the pressure tap corresponding 2-D image texture coordinates (Figure 7). The measurement locations so that surface pressures could be easily pixels from this region of the camera image are then applied to the visualized. Fiducial marks were also placed at regularly spaced quadrilateral using standard texture mapping. This procedure is locations around the surface of the wheel. Their coordinates were carried out for every facet of the geometry, until the entire object measured to later facilitate the calibration of the digital camera.
is texture mapped. (Currently, this algorithm uses only the surface normal to determine the visibility of a facet, thus only convex
3.1. Oil Flow surfaces are supported at this time.) The output is expressed as a
Once the oil flow pattern had dried on a wheel after a tunnel test, single platform-independent VRML 1.0 file. The result is a single a digital camera was used to capture eight high-resolution images 3-D visualization with very high-resolution textures that convey (3060x2036 pixels). The first image was obtained while the the complex oil flow patterns around the entire model. It can be model was still in the tunnel (Figure 2). Then the wheel was viewed interactively on any platform with adequate texture removed and oriented horizontally by inserting the axle into a mapping hardware.
temporary mount. Six more images were then captured around the periphery of the wheel, at locations aligned with fiducial
3.2. Surface Pressures
points on the model. A final image was captured from the side of Since the vertices of the PLOT3D grid describing the geometry of the wheel containing the axle. A schematic showing the relative the wheel were identical to the pressure tap locations, adding placement of seven digital images is shown in Figure 3.
pressure information to the oil flow visualization was straightforward. VRML 1.0 allows both texture and color information for each vertex, so we simply added the appropriate color from a rainbow color map to represent the pressure at each vertex. The result of combining scalar pressures represented by color with the grey-scale oil flow texture is shown in Color Plate 1. Note the correspondence of the oil flow features with both low (blue) and high (white) pressure regions. The combination of these two surface data sets into a single 3-D visualization made identification of the surface topology features far easier and more accurate than the manual comparison of numerous experimental 2-D images and pressure measurements.
3.3. Digital Particle Image Velocimetry
The DPIV process resulted in a steady 2-D Cartesian vector data set. Because the plane was measured in piecewise fashion, the data was divided into 160 cells of 60x60 data points each, with 4 points overlapping neighboring cells (Figure 4). A Gaussian filter was applied to smooth the edges between cells and combine the data into one Cartesian 1236x676 data set. Vectors with a value of Figure 3. Relative placement of camera images around the wheel.
zero were placed in the parts of the data set corresponding to the correlate the data from multiple experiments all at once. This interior of the wheels. capability provides valuable insight into the flowfield that would be nearly impossible to obtain by studying its component parts.
The Line Integral Convolution (LIC) [1] technique was The alternative, a manual comparison of more than a dozen 2-D applied to visualize the data in a manner visually similar to the oilflow photographs with numerous pressure and velocity flow of oil on the wheels. Color was used to visualize the measurements, would be very impractical. The 3-D visualization magnitude of the velocity at each point. This process resulted in a also serves as a useful record of the experiment that can be shared single color image showing the flow of air across the plane (Color with others on the Web via VRML. It also became apparent that Plate 2). A vortex is apparent between the wheels, and a saddle the best quality visualizations were obtained when the point and slower moving air appear behind the wheels. experiments are designed from the beginning with visualization in mind. Specifically, well placed and accurately measured fiducial marks, a high quality digital camera, sufficient lighting, and proper camera settings are necessary to produce high-quality results.
5. ACKNOWLEDGEMENTS
The visualization work was performed in the Data Visualization and Analysis Lab at NASA Langley Research Center. Special thanks to Patricia Kerr for geometry preparation, David Wolverton for camera calibration, and Chris Sandridge for providing the ImmersaDesk and CAVE. Thanks also to Jay Lambiotte, Kathy Stacy, and Ronnie Gillian for their advice and support.
6. REFERENCES
Figure 4. Diagram of the DPIV data divided into cells.
[1] B. Cabral, and L. Leedom. Imaging Vector Field Using Line Flow direction is from left to right.
Integral Convolution. In Computer Graphics (SIGGRAPH 93 Proceedings) , pg 263-270.
3.4. Immersive Visualization [2] CRAMPA-Close Range Multistation Photogrammetric
Adjustment Reference Guide. Version 1.1, Geometric The on-surface oil flow and pressure visualizations for the front and rear wheels were combined into a single VRML 1.0 file. Software (Coburg, Austral.) , May 1993.
[3] C. Cruz-Neira, D. Sandin, and T. DeFanti. Surround-Screen Flow symmetry was assumed to produce visualizations of all four Projection-Based Virtual Reality: The Design and wheels. The 2-D LIC image (from Color Plate 2) representing the off-surface velocity field data was then texture-mapped onto the Implementation of the CAVE. Computer Graphics (SIGGRAPH 93 Proceedings) .
appropriate 3-D plane between the wheels. Color was removed and transparency was added to the LIC image to facilitate [4] P. Debevec, S Gortler, L. McMillan, R. Szeliski, and C.
comparison with on-surface flow features. A few well-placed Bregler. Image-Based Modeling and Rendering. SIGGRAPH 98 Course Notes for Course 15 , July 1998.
streamlines were also included to highlight certain flow features in the data set. The final scene (Color Plate 3) was presented on [5] H. Karara. Non-Topographic Photogrammetry. Second ed.
American Soc. For Photogrammetry and Remote Sensing , an ImmersaDesk and in a CAVE [3], where the researcher could 1989.
navigate around a compelling visualization of the experiments and obtain a fully 3-D view from any point. This immersive [6] L. Keely and S. Uselton. Development of a Multi-Source Visualization Prototype. IEEE Visualization ’98. IEEE, environment was particularly effective for examining flow October 1998.
behavior in very tight locations, such as in-between wheels. It was very straightforward in this environment for the researcher to [7] C. Landreth and R. Adrian. Measurement and Refinement of Velocity Data using High Image Density Analysis in Particle correlate the off-surface flow features (e.g., vortices and Image Velocimetry. Applications of Laser Anemometry to stagnation points) with on-surface flow signatures (e.g., th convergence and separation zones). The point of view within the Fluid Mechanics, 4 Internation Symposium. Springer- Verlag. July 1988.
interior of a (hollow) wheel was also effective within the CAVE.
[8] B. Lazos. Mean Flow Features Around The Inline Wheels Of This vantage point provided the user with an overview of the flow patterns around that wheel, any part of which could be observed a 4-Wheel Landing Gear. To be published in AIAA Journal , 2001.
simply by turning their head. In the future, more user-interaction [9] W. Sellers III, and S. Kjelgaard. The Basic Aerodynamics will be added to this environment, to select from options such as animated LIC, multiple flow conditions, and scalar color control. Research Tunnel-A facility Dedicated to Code Validation.
AIAA-88-1997 , May 1988.
[10] K. Stacy, K. Severance, and B. Childers. Computer-Aided
4. CONCLUSIONS
Light Sheet Flow Visualization. AIAA-93-2915 , July 1993.
This case study described the process by which data from several [11] K. Stacy, K. Severance, and B. Childers. Computer-Aided very different wind tunnel experiments can be combined into a Light Sheet Flow Visualization Using Photogrammetry.
single 3-D scene by applying appropriate visualization techniques.
NASA TP-3416 , July 1994.
The result is an accurate high-resolution reconstruction of the composite flowfield that allows the researcher to visualize and Figure 5. Camera Resectioning.
Figure 6. Camera Selection.
Figure 7. Texture Mapping Color Plate 1. Texture mapped oil flow images colored by pressure on the front wheel. Flow direction is from lower left Color Plate 2. LIC technique applied to to upper right.
velocimetry data, colored by velocity magnitude. Flow is from top to bottom.
Color Plate 3. Final data fusion of oil flow images, surface pressures, and velocity measurements for the landing gear configuration. Flow direction is from left to right.