Extraction and Visualization of Swirl and Tumble Motion from Engine Simulation Data

Christoph Garth1 , Robert S. Laramee2 , Xavier Tricoche3 , Jrgen u Schneider4 , and Hans Hagen5
1 2 3 4 5

Visualization Group, Univ. of Kaiserslautern ([email protected]) VRVis Research Center, Vienna ([email protected]) SCI Institute, Univ. of Utah ([email protected]) AVL, Graz ([email protected]) Visualization Group, Univ. of Kaiserslautern ([email protected])

Summary. An optimal combustion process within an engine block is central to the performance of many motorized vehicles. Associated with this process are two important patterns of ow: swirl and tumble motion, which optimize the mixing of uid within each of an engines cylinders. Good visualizations are necessary to analyze these in-cylinder ows. The simulation data associated with in-cylinder tumble motion within a gas engine, given on an unstructured, time-varying and adaptive resolution CFD grid, demands robust visualization methods that apply to unsteady ow. We present a range of methods including integral, feature-based, and image-based schemes with the goal of extracting and visualizing these two important patterns of motion. We place a strong emphasis on automatic and semi-automatic methods, including topological analysis, that require little or no user input. We make eective use of animation to visualize the time-dependent simulation data and describe the challenges and some of the implementation measures necessary in an application of the presented methods to unstructured, time-varying and volumetric grids.

1 Introduction
Among the many design goals of combustion engines, the mixing process of fuel and oxygen occupies an important place. If a good mixture can be achieved, the resulting combustion is both clean and ecient, with all the fuel burned and minimal exhaust remaining. In turn, the mixing process strongly depends on the inow of the fuel and air components into the combustion chamber or cylinder. If the inlet ow generates sucient kinetic energy during this valve cycle, the resulting turbulence distributes fuel and air optimally in the combustion chamber. For common types of engines, near-optimal ow patterns are actually known and include, among others, so-called swirl and

Garth, Laramee, Tricoche, Schneider, Hagen

tumble motions. With the general progress of state-of-the-art CFD simulations, the discipline of engine design is made accessible to both numerical simulation and visualization of the resulting datasets, allowing for rapid testing of engine designs. Laramee et al. [9] took preliminary steps towards the visualization and analysis of in-cylinder ow. Using a combination of texture-based and geometric techniques, they were able to indirectly visualize the key swirl and tumble patterns in two engine simulation datasets. The approaches they used were essentially manual and they did not consider time-dependent ow. It is the aim of this paper to expand on this previous study by applying additional feature-centric visualizations. Here, we focus on topological methods and volumetric approaches, namely cuttingplane and boundary topology and direct volume rendering, examine hybrid visualizations that combine dierent techniques, emphasize schemes that can extract swirl and tumble characteristics semiautomatically, and handle full time-dependent ow on a time-dependent geometry.

With application by engineers in mind, we present a survey of methods that are useful in this context and demonstrate how they can be eectively applied in engine simulation analysis. Of particular interest are the time-varying nature of the simulation and the interconnection between visualization methods that treat data of dierent dimensionality (typically boundary vs. volume data). We study to what extent an analysis of the boundary ow permits reliable insight into the volume of the combustion chamber on the presented examples. As an example, we examine the eect that vortices have on the topological structure on the boundary. We describe the techniques employed and present a critical discussion of the resulting visualizations from an application standpoint. Although the application domain covered in this work is specic, the conclusions reached can be leveraged in many areas of engineering. The paper is structured as follows. In Section 2, we describe the application that we based our analysis on, namely two important patterns of in-cylinder ow. Section 3 is concerned with the criteria for our choice of methods. We briey describe the methods and how they contribute to a satisfactory extraction and visualization of swirl and tumble motions in Section4. Some of the technical aspects involved in time-varying unstructured grids are detailed there as well. Hybrid combinations of methods are examined in Section 5, before we conclude on the presented work in Section 6. Remark: In the course of this work, we have found that the possibility of interactive and animated viewing of visualization results greatly enhances the comprehension of occurring structures. For an impression, and for additional images, the reader is referred to the accompanying video [2].

Extraction and Visualization of Swirl and Tumble Motion

2 Engine Simulation Data

From a simplied point-of-view, there are two types of ideal ow patterns in an engine cylinder: swirl motion and tumble motion. Both are rotational motions, however, the axis of rotation is dierent in each case. Depending on the type of engine, one of these patterns is considered optimal because it maximizes mixing of injected fuel and air, resulting in homogeneous combustion. In this paper, we treat two datasets showing each of these two types of ow patterns (henceforth termed swirl motion and tumble motion). The basic geometries of the datasets and the respective desired motion patterns are shown in Figure 1. Although they were generated in the same problem context, the simulation datasets dier in a number of ways. Swirl Motion in a Diesel Engine This simulation is the result of a the simulation of steady charge ow in a diesel engine, based on a stationary geometry, resulting in a simple and stable ow. The main axis of motion is aligned with the cylinder axis and is constant in time. The spatial resolution of the single timestep is high with a total of 776,000 unstructured cells on an adaptive resolution grid. Tumble Motion in a Gas Engine This dataset results from an unsteady simulation of the charge phase of a gas engine. As the piston moves down, the cylinder volume increases by an order of magnitude and the fuel-air mixture entering the cylinder is drawn into a gradually developing tumble pattern. The overall motion is highly transient

stationary rotation axis

Fig. 1. (Left) Stable, circulating ow pattern in a diesel engine designated as swirl motion, with the cylinder axis as the axis of rotation. The ow enters tangentially through the intake ports. (Right) Transient tumble motion in a gas engine. The axis of motion moves as the cylinder expands and stays halfway between the top cylinder wall and the piston head at the bottom (not shown).

moving rotation axis

Garth, Laramee, Tricoche, Schneider, Hagen

and unstable. Both spatial and temporal resolution are relatively low, with the data given on 32 timesteps and the grid consisting of roughly 61,000 unstructured elements at the maximum crank angle. It is interesting to note that the actual mesh topology remains constant throughout all timesteps. This is accomplished by the use of virtual zero-volume cells at the piston head that expand as the piston moves down; only the mesh vertices are changing in time. Both simulations were computed at the Department of Advanced Simulation Technologies (AST) at AVL (www.avl.com) for the design and analysis of specic ow in-cylinder types. The commercially available AVL Fire solver was used for the solution of the compressible Navier-Stokes equation with a Finite Volume Method. In addition to the ow vector eld, the datasets encompass a number of additional attributes such as temperature and pressure. In this work, we focus on the analysis of the swirl and tumble aspects of the ow vector eld. Although the highest priority is given to the visualization of the pattern structures themselves, causes for their absence are also sought.

3 Choice of Methods
Here, we describe the criteria for our selection of visualization methods. Visualization Goals The main interest in the visualization of the in-cylinder ow is the extraction and visual analysis of the swirl and tumble motion patterns. Therefore, the ow vector eld and its derived quantities are of primary interest. For the use in design analysis, the constructed visualizations need to be objective and reproducible, meaning that the quality of the visualization result must not depend on vital parameters to be supplied by the user. This results in comparable visualizations for dierent simulation results of the same prototype or possibly even among dierent design prototypes. Therefore, in the selection of methods, we have put an emphasis on automatic schemes that require little or no user input. Data Dimensionality The simulation results are given in the form of attributes dened in the interior of the respective cylinder geometries. As is quite common in CFD simulations, the ow is required to vanish on the domain boundary (no-slip condition) in order to correctly model uid-boundary friction. Nevertheless, values on the boundary of the domain are easily inferred by e.g. extrapolation of volume values next to the boundary. It is also notable that in classical engineering analysis, visualization is widely performed on two-dimensional slices. Overall, the level of information that can be provided by a visualization technique increases with the dimension of the data it treats. At the same time, the visualization result need not necessarily improve due to perceptual

Extraction and Visualization of Swirl and Tumble Motion

issues such as cluttering. Finally, there is usually a price to pay in algorithmic complexity and computational cost as one progresses to higher dimensions. Therefore, for the case of our examples, we examine in some detail how the analysis of boundary and slice data allows to draw reliable conclusions on the pattern of the volume ow. We achieve this by a pairing of methods that combine boundary and volume techniques.

4 Extraction and Visualization of Swirl and Tumble Motion

In this section we present our choice of methods along with corresponding visualization results and discuss their relevance with respect to the realization of the visualization goals. Due to limitation of available space for images, many of the gures used for illustration of individual methods actually show a combination of dierent visualization approaches. We discuss the benets of such combinations in detail in Section 5. Moreover, the companion video [2] provides additional images and animations. 4.1 Global Flow Behavior using Integration-based Methods Integration-based methods are well suited to the analysis of time-dependent ows. Their common application to stationary ows is only a special case. We study the applicability of this class of methods on two examples. Particles and Pathlets Despite their simplistic nature, particle visualization can provide valuable insight into the overall structure of a ow dataset (cf. [1, 12] among others). This is especially true for time-dependent data. While the basic principle is similar to that of streamlines or pathlines, an animation of moving massless particles manages to convey the dynamic nature of the ow much better that static imagery alone. In the general case, integral methods suer from seeding issues, although strategies have been proposed to circumvent this (e.g. [19]). However, none of these approaches is concerned with time-varying data. Fortunately, engine geometries oer the inlet pipe as a natural choice of a seeding region. Integration of pathlines in time-dependent 3D ows is straightforward through the application of standard numerical integration algorithms that only require the integrand at a sparse set of points. While interpolation in time-varying grids is usually problematic, we were able to exploit the topologically invariant structure of the grid to simplify point location. Figure 2 depicts a frame from an animation of massless particles moving with the ow during the early stage of the valve cycle, seeded at positions in the intake pipe. The particles are of uniform size and color-coded according to ow velocity magnitude. The image allows an easy identication of zones

Garth, Laramee, Tricoche, Schneider, Hagen

where the velocity is lower than average, hinting at a non-optimal inow pattern at the side of the valve. Our general experience with this technique is that in spite of being visually imprecise, it greatly furthers the comprehension of the dynamic of the time-dependent ow by providing good overview. Stream surfaces Stream surfaces as an extension of streamlines are of great value in some applications because they manage to convey a spatially coherent picture of ow structures (cf. [3, 5, 14]). We found them to be of limited use in our case, for two reasons. First, both swirl and tumble are large-scale motions that are locally overlaid by other small-scale ow patterns and therefore, the resulting stream surfaces are complicated to interpret since the small details are emphasized by the surface nature of this primitive. Secondly, it is unclear how they can be applied satisfactorily in a time-dependent context. Although visualizations of some value can be generated, determination of good (in terms of visualization result) starting curves is dicult. Hence, we do not consider stream surfaces an objective technique by the criterion stated above. 4.2 Topology-Based Visualization of Flow Structures Topological methods provide ecient means for the visualization of essential structures in steady ows. As opposed to the integral methods described previously, they oer a fully automatic way to gain insight from vector data sets. The topological technique is typically applied in the visualization of planar ows [13] for which it yields synthetic graph representations. It consists of critical points (vector eld zeros) and connecting separatrices. The threedimensional case, however, remains challenging. Besides frequent occlusion problems that must be addressed specically [16], prominent features of interest like vortices cannot generally be identied as elements of 3D topology. In this paper, we address these deciencies by combining a topological analysis of the boundary ow with a hybrid approach that leverages 2D topology to explore the 3D structure of a vector eld [17].

Fig. 2. A frame from a time-varying tumble motion visualization using a combination of particles and vortex cores (lines, extracted by cutting-plane topology). Particle velocity magnitude is color-coded. Lines are colored according to path type (saddles - red, sources - green, sinks - blue). In this frame, some of the particles have been captured in the vicinity of vortex cores resulting in lost energy (lower velocity) for the creation of the tumble pattern.

Extraction and Visualization of Swirl and Tumble Motion

Fig. 3. Visualization of swirl motion using boundary topology. Critical points are colored by type (cf. Fig. 2), and separatrix color varies with separation/attachment behavior from dark blue (weak) to cyan (strong). Separatrices indicate the separation between neighboring vortices on the boundary. (Left) Combination with volume rendering with a transfer function of 2 only. On the bottom left of the cylinder, the recirculation zone causes a non-ideal o-center rotation, as visualized by topology. (Right) In combination with LIC.

Boundary Topology To our knowledge, a general algorithm for vector eld topology on 2D unstructured grids embedded in three-space has not yet been described in the visualization literature (although algorithms exist for parameterized grids, e.g. [4]). Thus, we propose the following approach. In each triangle, we use the well-dened local tangent plane to perform a cell-wise search for critical points and determine their type. The construction of separatrices from saddle points is performed using a streamline integration approach based on geodesics as introduced by Polthier and Schmies [11]. A specic characteristic of triangulated surfaces for topological analysis is the existence of what we term singular edges 6 . Since the tangent plane is discontinuous across surface edges, the ow on both sides can be contradictory. Singular edges are typically found along the sharp contours of the geometry where they must be integrated in the topological analysis to account for the possibly contradictory ow behavior between neighboring cells. For viscous ows, the information conveyed by the boundary topology can be enhanced naturally by showing the strength of ow separation and attachment along separatrices. Flow separation occurs when the ow surrounding an embedded body interrupts its tangential motion along the objects boundary and abruptly moves away from it. The opposite phenomenon is called ow attachment. As pointed out by Kenwright [7], separatrices of the boundary vector eld constitute so called closed separation or attachment lines. To quantify ow separation and attachment along a separatrix, we compute the divergent (resp. convergent) behavior of neighboring streamlines by evaluating the local divergence of the vector eld [18]. This is illustrated in Fig. 3
6

Integral curves cannot continue over these edges since the interpolants on both sides of the edge are incompatible.

Garth, Laramee, Tricoche, Schneider, Hagen

by the color coding of the intensity of ow separation and attachment along separatrices. Direct visualization of the boundary topology produces images such as Figure 3 (swirl dataset) and can also be applied in a time-dependent context. The combination of linear interpolation in time and in space often produces artifacts (such as articial pairs of critical points that appear and quickly vanish quickly). Nevertheless, animations that show the temporal evolution of these instantaneous graphs provide valuable means to track the dynamics of important ow patterns (cf. Fig. 8). In our experience, boundary topology is most eective if used in conjunction with methods that visualize additional properties of the ow, both in the volume and on the boundary. Such combinations permit to determine the mutual inuence between boundary and volume. We will discuss these issues when we examine further combinations of methods presented in Section 5. Cutting-Plane Topology It was shown previously [17] that a moving cutting plane (traversing the volume of the dataset) on which the vector eld is resampled and projected at regular intervals can be a powerful tool in the analysis of 3D datasets. The projection of the vector eld on the plane eectively manages to discard structures orthogonal to the plane, but preserves plane-parallel ow patterns. If assumptions on the orientation of features are given, this property can be exploited. Cutting-planes are hence well suited for the qualitative analysis of swirl or tumble motion, since its axis of rotation is known. Furthermore, the (discretized) continuum of cutting-planes allows for the application of criticalpoint tracking over the plane parameter range. In the case of vortical motion that is intersected orthogonal to the rotational center, a reproduction of the vortex core as the path of a critical point over the parameter range should, in theory, work well. In practice, it is hard to intersect (a-priori not known) structures exactly. This results in approximations of the vortex cores. Still, a qualitative analysis is viable. In Figure 4, the results of this approach applied to the swirl motion dataset are displayed. As the cutting planes are applied orthogonal to the cylinder axis,

Fig. 4. Cutting-plane topology applied to the diesel engine. Plane separatrices are colored gray. Despite the visual inexactness, swirl structures emerge clearly. Critical points paths are colored according to nature. It appears the overall swirl motion is fueled by several parallel vortices at the top of the cylinder. The main swirl motion core is disrupted near the middle.

Extraction and Visualization of Swirl and Tumble Motion

coherent swirl-type structures emerge at the top of the cylinder. Rotation cores orthogonal to the planes are visualized by critical point paths over the plane continuum. Interestingly, the main swirl core is supposed to extend through the whole cylinder, but is actually interrupted near the middle. Figure 5 shows frames from an animation of the tumble dataset. The moving cutting planes have been applied orthogonal to the tumble axis and are color coded by their distance to the back wall of the combustion chamber for increased visual clarity. Although the visualization is not exact, the prevalent tumble structure is captured well in spite of its overall weakness and instability. Again, the center of the respective motions is given by the critical points paths. The tumble motion is found to consist of several smaller vortices, of which some have a diagonal orientation that looks like a simultaneous combination of swirl and tumble. In the full animation (cf. [2]), the interaction of the dierent smaller tumble patterns can be observed as they split and merge. Using only the critical point paths for visualization, it is possible to observe the main tumble vortex, as shown in Figure 9. It does not completely match the desired axis and is highly o-center on the curved wall of the combustion chamber. Here too, the diagonal nature of the main rotation is conrmed. Volume Visualization Among the many region-based vortex denitions, the 2 -criterion [6] has a well established tradition in engineering use. The criterion is given as a scalar quantity derived from the ow eld Jacobian matrix and related to a minimum in pressure in the Navier-Stokes equation [10]. According to its denition, a vortex is present at a point if 2 < 0. Traditionally, isosurfaces are used to

Fig. 5. Two frames from an animation of the tumble motion simulation. Cutting plane topology is applied to visualize ow eld structures in the plane orthogonal to the tumble axis. Color of separatrices varies from blue to red on successive cutting planes. Tumble-like ow structures emerge clearly from the otherwise incoherent lines. The paths of critical points over the cutting plane continuum are displayed in green. In the last frame (right), the diagonal main tumble axis can be observed together with a large recirculation zone (closed path on the left).

10

Garth, Laramee, Tricoche, Schneider, Hagen

visualize vortices. However, this approach does not fare well in complicated datasets with many vortices. The resulting isosurfaces do not separate individual vortices and are prone to visual complexity. This is only avoided by a careful manual selection of the isovalue. In addition, the strength of the rotation (as given by the modulus of 2 ) is not visualized. Interactive volume rendering has also been applied in this context [15, 17] and is able to overcome most of the diculties related to isosurfaces. As described in [17], we have not attempted to apply volume rendering directly to the unstructured grid but have employed a resampling scheme that results in a rectilinear grid covering the region of interest (in our cases, the cylinder). This grid is then used in a direct volume rendering approach. Although artifacts are incurred in naive sampling, post-sampling scale-space ltering is very eective in removing these artifacts. Using this approach, Figure 3 (left) illustrates the vortices in the context of the swirl motion using a simple one-dimensional transfer function that indicates vortex strength by color. As already visible in the results of the Sujudi-Haimes method, the actual swirl motion consists of several independent vortices, most prominently a strong vortex that spans almost the entire length of the cylinder. It is clearly o-center. A close-up of the upper cylinder region including the region just below the intake port is detailed in Figure 6. It is apparent that the fuel-air mixture entering the cylinder is drawn into a strong rotational motion. Whether this is good for the overall mixing process or an obstacle that results in a pressure loss with negative consequences remains unclear. Technically, we have limited the transfer function range to 2 -values between 106 and 102 to lter vortical motions at the small scales. Recently, the use of multi-dimensional transfer functions [8] in ow visualization was investigated [17]. We make use of this technique to visualize the rotational orientation of individual vortices, allowing more insight into the mutual interaction in the observed vortex systems. Adding normalized helicity as a second variable, we manage to both strengthen the vortex region criterion as well determine the orientation of the rotational motion. Normalized helicity is given as the angle between the velocity vector and the vorticity vector and has a range of [1, 1], with positive values in the case of counterclockwise rotation in ow direction and near-zero values indicating very weak vortical behavior. The obtained visualization (cf. Fig 6) indicates several mutually counter-rotating vortices in close proximity near the intake ports. This

Fig. 6. Visualization of the rotational directions in the vortex system at the top of the diesel engine cylinder. The transfer function is identical to that of Fig. 7. The counterrotating vortices appear in blue and in red depending on rotation direction. On the boundary, the topological analysis extracts and visualizes separation lines between individual vortices (color coding as in Figs. 2 and 3).

Extraction and Visualization of Swirl and Tumble Motion

11

Fig. 7. Unsteady visualization of vortices from in-cylinder tumble motion in a gas engine and its relationship to the boundary. During the valve cycle (top left to bottom right), the piston head that shapes the bottom of the geometry moves down (not shown). The volume rendering shows vortices using a two-dimensional transfer function of 2 and normalized helicity (legend). The main tumble vortex is extracted and visible as o-center and with an undesired diagonal orientation. The ow structure on the boundary is visualized using boundary topology, with critical points colored according to their nature (cf. Fig. 2). A direct correspondence between the volume and boundary visualizations can be observed. In the third image, the intersection of the main vortex with the boundary results in critical points on the front and back walls.

is highly undesired, since part of the energy contained in the inow is used up by this vortex system. The use of multidimensional transfer functions (used commonly in medical image generation) can require some degree of interactivity in the determination of the transfer functions in order to specify variable ranges manually for most satisfactory results. We wish to emphasize that in our case, transfer functions based on physical criteria (such as 2 < 0) are automatic in the sense the interesting value ranges are dictated by the laws of uid dynamics. Using simple linear interpolation in time, an animation (cf. [2]) can depict the temporal evolution of vortices associated with swirl and tumble motion (see Fig. 7).

5 Hybrid approaches
While performing experiments with the dierent approaches detailed above, it became apparent that a combination of visualizations can provide an even more thorough understanding of the simulation results. In this section, we describe examples of particularly eective combinations and how they contribute to the swirl and tumble analysis.

12

Garth, Laramee, Tricoche, Schneider, Hagen

Fig. 8. Two frames from a time-varying visualization of tumble motion using a combination of cutting-plane topology (green lines) and boundary topology (blue lines). Where the critical point paths computed over the cutting-plane parameter range intersect the boundary, singularities appear there, too. This is a prime example of a hybrid approach being used to investigate the relationship between boundary and volume methods.

5.1 Boundary and cutting plane topology In uid ows, complex ow structures such as vortices are often caused by the interaction of the ow with boundaries. Depending on this interaction, dierent patterns appear on a boundary that in turn allow one to infer properties of the volume ow. However, in complicated geometries such as in-cylinder ow, this approach alone introduces visualization complexity due to the high number of vortices involved and generally complex ow structure. The combination of boundary topology and cutting plane topology is an eective approach. Figure 8 illustrates this in time-slices from an animation. For the tumble ow, this type of visualization provides valuable insight into the development of the diagonal tumble motion. In the early stages of the valve cycle (left two images), the ow pattern is very incoherent and unstable. Roughly at the middle of the cycle (second image from the right), a swirl pattern occurs in the front half of the cylinder, mainly constituted by two large vortices. In the very last timesteps, the rotational axis tilts towards the desired tumble axis, but fails to reach it completely. Here, a large recirculation zone can be observed that may hinder the development of the tumble motion. 5.2 Sparse and dense methods On the boundary, the topological graph as a visually sparse method is eectively combined with dense methods, such as texture-based methods or volume visualization. While texture-based methods are built on the capability of the human visual system to identify patterns in the ow, the topological graph serves as a terse structural picture that relies on cognitive interpolation on behalf of the viewer. It is therefore a very natural combination. Figure 3 (swirl, right image) provides an example, showing a very strong vortex near the inlet that is drawing away energy from the creation of the ideal swirl pattern. We refer the reader to previous work [9] for other applications of texture-based techniques in this problem context that we believe will benet strongly from a pairing with a feature-based visualization.

Extraction and Visualization of Swirl and Tumble Motion

13

A similar conclusion is reached by combining boundary topology and the volume visualization of vortices. Again looking for the swirl pattern, the imperfection of the actual ow motion is visible in Figure 3 (left). While the volume rendering shows the correctly oriented but o-center main vortex, the topology graph on the lower cylinder boundary complements this visualization by showing corner regions of the ow that are not taking part in the swirl pattern. As expected, the topological graph also serves to show separation and attachment boundaries that delimit the regions of inuence of the dierent vortices. As conrmed by the rotational direction analysis (cf. Fig. 6), these vortices are rotating in dierent directions, which is considered destructive ow behavior. There are of course other possible combinations: for example, the SujudiHaimes vortex core line visualization can play a similar role as cutting-plane topology. In our experiments we found however that it often detects only the strong vortices at the intake portions of both datasets. Combining particles and vortex core lines oers insights into how exactly the vortices are created and where kinetic energy is lost in small scale structures (cf. Fig. 2).

6 Conclusion and Future Work

By using a number of visualization techniques that we selected as automatic and objective, we were able to extract and create visualizations of the swirl and tumble datasets that allow an in-depth visual analysis of the actually occurring patterns. The visualizations are comparable between similar datasets and can thus be employed in design prototype analysis. By using hybrid combinations of dierent techniques, we were able to determine the extent to which the desired pattern is established and also detect inuences that hinder its formation. In summary: The swirl motion in the diesel engine is visible in the form a prominent main vortex spanning the entire cylinder (Figure 3, left). It is nonoptimally o-center. A recirculation zone is present in the lower corner of

Fig. 9. Close-up of Figure 8. Tumble motion center lines are extracted using cutting-plane topology and are colored according to the type (cf. Fig. 2) of the critical point path. Boundary singularities appear at intersections of the rotation centers with the boundary.

14

Garth, Laramee, Tricoche, Schneider, Hagen

the cylinder. This may be the cause for the eccentricity of the swirl motion. A vortex system at the intake valve (Figures 3 and 6) prohibits the full conversion of energy contained in the incoming ow into the swirl motion. The observed tumble motion in the gas engine diers from an optimal pattern in several aspects: it is unstable and sporadic over time rather weak and o-center (Figs. 7 and 8) and resembles a hybrid of both swirl and tumble patterns (Figures 8 and 9). Here too, a prominent recirculation zone induces the swirl component into the ow. Again, the ow distribution at the intake valve is not optimal.

The unstructured, adaptive and time-varying nature of the tumble datasets poses a technical diculty that we were able to circumnavigate through the choice of schemes and appropriate extensions where needed. The resulting visualizations are of high quality and provide valuable insight into the application. There are many possible avenues that future work might take. High up on the priority list are improvements in the eld of topological visualization. Full three-dimensional topology has not been completely realized, and while delivering viable visualization results, replacements such as cutting-plane topology are not completely satisfactory. The inclusion of features such as vortex cores into the topological skeleton is desirable, but may not be possible. Furthermore, the interplay between the topologies of the boundary and volume ows need to be investigated on a more systematic basis. In general, it seems desirable to examine the hybridization of dierent visualization approaches in the same context. Future work could also feature improvements to many of the schemes presented here to allow for the treatments of larger datasets, such as entire engine blocks. It remains to be seen in how far this is feasible from a technical point of view. Acknowledgements The authors would like to thank all those who have contributed to this research including AVL (www.avl.com), the Austrian research program Kplus (www.kplus.at). The CFD simulation datasets are courtesy of AVL.

References
1. D. Bauer and R. Peikert. Vortex Tracking in Scale-Space. In Data Visualization 2002. Proc. VisSym 02, 2002. 2. C. Garth, R. S. Laramee, X. Tricoche, J. Schneider, and H. Hagen. Extraction and visualization of swirl and tumble motion from engine simulation data (companion video). available at http://www.vrvis.at/scivis/laramee/MotionExtracted/.

Extraction and Visualization of Swirl and Tumble Motion

15

3. C. Garth, X. Tricoche, T. Salzbrunn, and G. Scheuermann. Surface Techniques for Vortex Visualization. In Proceedings Eurographics - IEEE TCVG Symposium on Visualization, 2004. 4. A. Globus, C. Levit, and T. Lasinski. A tool for visualizing the topology if three-dimensional vector elds. In IEEE Visualization Proceedings, pages 33 40, October 1991. 5. J. P. M. Hultquist. Constructing Stream Surfaces in Steady 3D Vector Fields. In A. E. Kaufman and G. M. Nielson, editors, IEEE Visualization 92, pages 171 178, Boston, MA, 1992. 6. Jinhee Jeong and Fazle Hussain. On the Identication of a Vortex. Journal of Fluid Mechanics, pages 6994, 285 1995. 7. D. Kenwright, C. Henze, and C. Levit. Feature extraction of separation and attachment lines. IEEE Transactions on Visualization and Computer Graphics, 5(2):135144, 1994. 8. J. Kniss, G. Kindlmann, and C. Hansen. Multidimensional transfer functions for interactive volume rendering. IEEE Transactions on Visualization and Computer Graphics, 8(3):270285, July-September 2002. 9. R. S. Laramee, D. Weiskopf, J. Schneider, and H. Hauser. Investigating Swirl and Tumble Flow with a Comparison of Visualization Techniques. In Proceedings IEEE Visualization 04, pages 5158, 2004. 10. R. Peikert. Tutorial on Feature Oriented Methods in Flow Visualization, IEEE Visualization 2004. 11. K. Polthier and M. Schmies. Straightest Geodesics on Polyhedral Surfaces. In Hans-Christian Hege and Konrad Polthier, editors, Mathematical Visualization, pages 135150. Springer Verlag, Heidelberg, 1998. 12. F. H. Post, B. Vrolijk, H. Hauser, R. S. Laramee, and H. Doleisch. Feature Extraction and Visualization of Flow Fields. In Eurographics 2002 State-of-theArt Reports, pages 69100, 26 September 2002. 13. G. Scheuermann and X. Tricoche. Topological methods for ow visualization. In C.D. Hansen and C.R. Johnson, editors, The Visualization Handbook, pages 341356. Elsevier, 2005. 14. D. Stalling. Fast Texture-based Algorithms for Vector Field Visualization. PhD thesis, Freie Universitt Berlin, 1998. a 15. S. Stegmaier and T. Ertl. A Graphics Hardware-based Vortex Detection and Visualization System. In Proceedings of IEEE Visualization 04, pages 195202, 2004. 16. H. Theisel, T. Weinkauf, H.-C. Hege, and H.-P. Seidel. Saddle Connectors - An Approach to Visualizing the Topological Skeleton of Complex 3D Vector Fields. In IEEE Visualization 03, 2003. 17. Xavier Tricoche, Christoph Garth, Gordon Kindlmann, Eduard Deines, Gerik Scheuermann, Markus Rtten, and Charles Hansen. Vizualization of Intricate u Flow Structures for Vortex Breakdown Analysis. In Proceedings of IEEE Visualization, pages 187194, October 2004. 18. Xavier Tricoche, Christoph Garth, and Gerik Scheuermann. Fast and robust extraction of separation line features. In Proceedings of the Dagstuhl Scientic Visualization Seminar, 2003. to appear. 19. Verma V., D. Kao, and A. Pang. A Flow-guided Streamline Seeding Strategy. In Proceedings of IEEE Visualization 00, 2000.