One more example of using scalar attributes to identify areas of interest. Attributes depicted in scatter plots are particle velocity mean and maximum. We can clearly see how brushing different clusters in the scatter plot views (bottom views) results in pathlines of different behavior being selected (top views). Also similarities within the groups are very obvious. 

Context in source publication

Context 1

... data set consisting of 36524 cells for the first step and about 2 million cells for the second step of the analysis. This exhaust manifold is used in a six cylinder car, but we have only analyzed one side (exits from three cylinders) since the sides are symmetrical. From the simu- lation data set pathlines and their selected attributes are computed. The data set proved to be very challenging because of its pulsing nature, where only one pipe (on one side) is active at specific time points and for only a short period of time. Also, as velocity values of the flow field are very high we had to pay special attention to how to compute pathlines, as they would otherwise leave the, in our case, relatively small volume very fast. To cope with that we decided to use cellular integration instead the one with constant time step. Cellular integration iterates cell by cell and builds pathline, thus making sure that all the vectors of the flow field along the pathline way are taken into account. This frees the user of sometimes difficult decision of which integration time step to use, especially in cases of very high velocities. Resulting data sets consisted of up to 10000 pathlines and their attributes (5 scalar and 11 time series attributes). Every internal combustion engine has an exhaust manifold. It is generally a simple unit that collects engine exhaust gases from multiple cylinders and deliv- ers them to the exhaust pipe. Each cylinder has its own exhaust head-pipe, and they usually converge into one tube called a collector. When an engine starts its exhaust stroke, the pis- ton moves up the cylinder bore, decreasing the total chamber volume, which increases the pressure in the cylinder, and when the cylinder’s valve opens, the high pressure exhaust gas exits into the exhaust manifold, creating an exhaust pulse comprising three main parts. They are: high-pressure head, medium-pressure body, and the low-pressure tail component. The momentum of the high and medium pressure components reduces the pressure in the combustion chamber to a lower than atmospheric level. This relatively low pressure helps to extract all the combustion products from the cylinder and induct the intake charge during the overlap period when both intake and exhaust valves are partially open. The effect is known as scavenging. Length, cross-sectional area, and shape of the exhaust ports and pipeworks influence the degree of scavenging effect, and the engine speed range over which scavenging oc- curs. Selecting the length and diameter of the primary tubes must be done very carefully depending on what we want to accomplish (more power, lower fuel consumption, ...). To get a better overview over the time series attributes some first order statistics (mean, max, ...) was computed. This newly computed scalars were depicted using scatter plot, parallel coordinates and histogram views. Also we had the 3D view active showing pathlines with color mapping [12] (mapping of the colors to the values) enabled, so we could see if the features found in other views as interesting are the result of the interesting behavior of the pathlines. For color mapping we use time, particle velocity or any other time series attribute of the ones listed in Table 1. The flow field features stored as time series along the pathline proved to be very useful, since they give us an insight into what made pathlines behave as they do. One such example can be seen in Figure 2. In the scatter plot (lower-left views), depicting relative start- end distance and length attributes, clusters are clearly recognizable, so we wanted to investigate what do we get when we brush some of them. The most interesting proved to be clusters with relatively high length values. We have selected pathlines with high length, and medium (Figure 2a) and high (Figure 2b) distance between start and end. There are other pathlines that trav- eled the same distance (gray points in scatter plot be- low the brush), but they have shorter length. The higher length of the brushed pathlines is the result of vortices along their way, and this is confirmed when looking at the parallel coordinates view with relatively high maximums of torsion, curvature, vorticity magnitude, and swirling strength values. As, in the case of an ideal exhaust manifold, gases would travel as fast as possible to the exhaust, this is certainly an unwanted behavior. We continue with exploration now, trying to find other attribute combinations which will support ana- lysts in getting insight into complex flows. Having multiple ways to identify possible problems gives us a better chance to identify all of them. Since different attribute combinations (both scalar and time series) are used in these approaches, it also gives us a better un- derstanding of the attributes. We can find correlations between them, and thus get better insight into the data set. Another such example, using scalar attributes, but in a different combination, can be seen in Figure 4. We use average and maximum particle velocity attributes. As can be seen, using different combination results in somewhat different clustering. In this case clusters are even more easy to notice and brushing them reveals groups of pathlines, with behavior of pathlines being very similar within group, which was not always the case in the previous approach were these three groups were mixed into two. So it seems we have found a better way (that gives us better control) for finding pathlines of interest. Using time series attributes we get a similar result, as can be seen in Figure 5a. Two curve views are used to depict time series attributes, which are in this case curvature and torsion. Again almost the same pathlines are selected, but in this case we do not have as good control over what is selected as in previous cases. To gain better control we can use the projection views with their direct pathline brushing ability, as can be seen in Figure 5b. What can be seen in all cases is that none of the brushed pathlines goes out through the expected pipe, but instead they finish in the left or middle pipe, which is undesirable. Also, almost all of these pathlines are seeded or pass near the edges of the pipes. The according histogram views show us that this happens only in few time steps, and for a small amount of cases, but still this could be an indication of a problem in the exhaust manifold. To gain better understanding why is this happening we use finer (higher resolution) data set for those time steps and seeding points and then do a more detailed analysis to see what can be done, if any- thing, to reduce them even more. Now higher resolution data set is used to compute pathlines seeded only at the spatio-temporal areas identified in the previous step. This data set is then analyzed to try to get better understanding of the unwanted behavior. What is immediately obvious when looking at Figure 7, showing a 3D view depicting pathlines computed in higher resolution data set, but only for the seed cells of pathlines brushed in Figure 2b, is that, unlike in lower resolution case, now we have pathlines ending in the middle pipe also. This shows how complex the flow is and that very small changes can drastically change the result. Next we focus on one interesting pathline from the pathlines shown in Figure 7. Using color mapping with time series attributes and also different types of views with attributes we try to discover as much as we can about behavior of that pathline, i.e., behavior of the particle in the flow. Figure 6, which shows four 3D views depicting the same pathline, but with different attributes color mapped, shows an example of such an analysis. Our focus here was on particle velocity (color mapped in the top-right view in the figure). We can see that the particle velocity dropped very rapidly once it started changing direction, and the reason that it did change direction is because it entered areas of high swirling strength and vorticity magnitude which can be seen in the two bottom views. With this we have shown an example of how our tool can be used in the analysis of the exhaust manifold data set. The domain expert found it very interesting, and thinks it has great potential, especially when used in combination with other tools, and we will continue our collaboration. In this paper we have shown how pathlines in combination with a carefully designed tool can give a user new abilities in analyzing the fluid flow field dynamics simulation results. Pathlines and their attributes are used to help identify possible problems in the exhaust manifold which can cause loss of engine power and increase of fuel consumption. We show how computing power limitations can be avoided by first using lower resolution to identify spatio-temporal areas of interest and then using higher resolution for detailed analysis of those areas. Furthermore, direct pathline brushing by using projections is introduced. Also, usage of pre-configured view arrangements, to help with screen space organization, is proposed. The domain expert gave us positive feed- back and we plan to continue our work on this subject in the future, specially focusing on the selection of the attributes available and combining our tool’s results with other tools. We think that our tool is a step forward and can lead to further developments in this area. The project SemSeg acknowledges the financial support of the Future and Emerging Technologies (FET) programme within the Seventh Framework Programme for Research of the European Commission, under FET- Open grant number ...