Development of a Pre-processor for Grid-Free Solvers K. Anandhanarayanan, R. Krishnamurthy and Debasis Chakraborty Scientists, Defence Research and Development Laboratory, Hyderabad Abstract A versatile pre-processor has been developed to generate data-structure for grid-free solvers. The grid-free solver requires a distribution of points in the computational domain and a set of neighbours around each point. The pre-processor uses overlapped unstructured grids to obtain the distribution of points. The points that lie inside the body are removed using hole-cutting methods and the neighbours around each point are obtained using the grid information. The neighbours of the point in the overlapping region are obtained using donor search method. Approximate inverse map and alternating digital tree methods are used to accelerate the donor search. The implicit hole cutting method is used to reduce the overlapping region, by which the computational time can be reduced and as well the solution accuracy can be improved. The pre-processor has been applied to generate connectivity for multi-body problems such as wing-store, aircraft-store and launch vehicle with fairings and grid-free solver is applied to solve subsonic flows around the above configurations. 1. Introduction The Grid-free Euler and Navier Stokes Solver (GEANS)1 has been developed to operate on an arbitrary point distribution. The solver requires a distribution of points and a set of neighbours or connectivity for each point. The governing partial differential equations (PDE) for fluid flow are solved at each point using the least squares method to approximate the spatial derivatives of fluxes in the PDE using the flow field variables available at the neighbouring points. The solver has been validated for geometrically simple configurations using cloud of points that are obtained using structured or unstructured grids and the neighbours are obtained using the grid connectivity information. With the objective of extending the application of the solver to multi-body problem, a pre-processor has been developed to generate a distribution of points and a set of neighbours around each point. The purpose of the pre-processor is to utilize a set of overlapping grids, known as overset or chimera grids2,3, as shown in Figs. 1 and 2, to get the distribution of points. The complex geometry can be considered as blending of various simple components. The component grids are generated around portions of the geometry without regard for other portions of the geometry, which greatly simplifies the grid generation process. It has been widely used to simplify the grid generation requirements for complex geometries. The use of an overset grid system enables for the simulation of bodies in relative motion, such as a store separation and rotorcraft4, 5. Due to overlapping of multiple grids, some points may lie inside the body of other component, such points are referred to as hole points and such points should be excluded from the computations. There could be some points in the overlapped region that may be outside the body, but to be excluded in favour other component grid points. The points adjacent to the holes become inter-grid boundary points, termed as “fringe" points. The marking of holes and the surrounding fringe boundaries forms the first phase of an overset grid assembly process. In the grid based solvers, the boundary values required by the flow field solution at the fringe points are obtained by interpolating the solution from appropriate 2 donor elements using information from other grids that overlap the region. Whereas, in the grid-free solver, the fringe points are also considered as active points but they include points from other overlapping grids6. Therefore, there is no interpolation involved in the grid-free solver. The fringe points, essentially active points in the grid-free method, couple the solutions between component grids using points from all the overlapping grids as neighbours7. The donor cell of a point is a cell in the overlapping grid that contains the point. The pre-processor identifies the hole points and fringe points, known as hole cutting. Then, the connectivity of fringe points is enhanced by identifying the donor cell using donor search method from overlapping grids. The following sections give details about the various algorithms to realize the above tasks and in addition, implementation of those algorithms. 2. Overset Hole Cutting The hole-cutting procedure is a critical element of the overset grid assembly process. It is desirable for the hole-cutting process to be fast, efficient, and require a minimal amount of user input and control. The surface meshes of component grids are used to represent the geometry for hole-cutting. The unstructured surface meshes are composed of triangles and a triangular surface face is a planar geometric simplex with well known geometric properties, such as the face normal. The intersection tests and other operations can be easily performed for a triangular face, since analytical solutions are available for planar elements. The following methodologies have been implemented in the pre-processor for hole-cutting. 2.1. Implicit hole cutting The implicit hole cutting8 is the simplest hole-cutting approach that uses only a standard donor search procedure (which will be discussed in the later section) to find the potential candidate donor elements. It is basically selection of the quality cells in the overlapped grid for computation while other cells are blanked. In the present work, wall distance, distance to the closest boundary, is used for selection of cells. The wall distances are computed at all the points in a grid from the wall boundaries within the grid. In the next step, the donor cell is obtained for each point and the wall distance of a point is compared with the wall distances of the vertices of the donor cell. The point is marked as inactive if it is closer to the other boundary than to its own boundary. The effect is to locate the fringe boundary equidistant between the bodies and minimize the overlap. In the grid-based method, since the fringe points do not take part in computation, the fringe points isolate solid boundary and therefore, the name is implicit hole-cutting which does not require separate hole cutting procedures. In the grid-free method, there is no interpolation involved and therefore, the points inside the body should be blanked. In the present work, implicit hole cutting is used only for optimising the overlap region and one of the following hole cutting procedure is used for blanking solid cells. This method works well when the grid has homogeneous spacing in the overlap region but requires a donor search for all points in the mesh. If there is insufficient overlap in the grid system, the approach may fail with much of the grid system erroneously marked as holes. 2.2. Analytic Shapes The simplest, and typically fastest, hole cutting method is the use of analytic shapes. Simple analytical shapes such as cubes, spheres, cones, cylinder, parallelepiped as holecutting geometry9. Testing of a point, that is inside the body or not, can be easily done against analytical shapes. However, this approach can be time consuming as the set of analytic shapes must be chosen and positioned to approximate the actual geometry. Sometimes, it may 3 not be possible to define such shapes for closely positioned multi-bodies. This method is implemented for cases when the bodies are positioned at farther distances. 2.3 Point-Normal Comparison The computational surface grids are used as hole-cutting surfaces. The dot product between the surface normal vector and a vector drawn from the surface to a point in question is used to decide if the point lies inside the surface or not10. An example of showing the point-surface normal test is shown in Fig. 3. The cost of the approach is proportional to the number of component grid points being tested and the number of component used to define the cutting surface. The cost can be reduced by considering only the points that are within the bounding box of cutting surface. The hole cutting surface should be convex for successful application of this method. To overcome this difficulty, a closest point on the cutting surface is used to apply this test. 2.4 Cartesian Grid Approximations to the Geometry: The Cartesian hole-map offers an efficient means of hole point determination. A closed surface S can be enclosed within a uniform Cartesian grid. Points in the grid can be marked as being inside or outside of S very easily. As a pre-processing step, the bounding box of surface faces are compared with the cells of Cartesian grid and the intersecting cells are marked as cut-cells. Then, a cell far away from the boundary is marked as outside and traversing through neighbours recursively, all the unmarked neighbour cells are set as outside cells. After marking all the outside cells, the rest of cells are marked as inside cells. A uniform Cartesian grid so marked becomes an approximate representation of S and is referred to as a hole-map11 (Fig. 4). The proximity of any point P with respect to surface S can be determined by checking the hole-map of S. Given the coordinates of point P, the corresponding bounding hole-map element can be computed directly as ( x p xmin ) ( y p y min ) ( z p z min ) (1) i ,j ,k dx dy dz where xmin, ymin, zmin are the coordinates of the hole-map origin, and dx, dy, dz are the holemap spacings in three co-ordinate directions. If the eight vertices of hole-map are all marked as a hole, then P is inside the hole-cutting surface. If the eight vertices are all unmarked, P is outside the surface. However, if the eight vertices are of mixed type (marked and unmarked), P is near a hole-cutting plane and a vector intersection test can be used to determine the actual status of P. The test of a point status relative to the hole-map can be made completely robust and very fast. But, the memory required to store hole-maps of cutting surfaces can be a serious drawback to this approach for some applications. Hole-maps require 3-D arrays that can be very large when precise hole-cuts are needed. 2.5 Object X-ray Object x-ray12 method is similar to Cartesian hole-map, but approximates the geometry in two dimensions and rigorously evaluates the geometry in the third dimension. In a pre-process step, a 2-D uniform Cartesian mesh is defined for a plane located at an elevation below an object to be used for hole-cutting purposes. From each vertex in the 2-D plane, a ray is defined normal to the plane and is of sufficient magnitude to reach above the object. Intersections (i.e., pierce-points) between the rays and the object are then computed and saved as shown in Fig. 5. The blank status of a point P can then be determined as a function of the pre-computed ray casts in the vicinity of the projection of P onto the 2-D 4 plane. In the present implementation of this method, the 2-D plane, or "image plane," is always aligned with the Cartesian X-Y plane, and the rays are aligned with the Cartesian Z direction. Accordingly, the complete definition of an object can be efficiently represented with a single plane of data, analogous to how common x-ray images for medical purposes can be stored on the photograph paper. Given this construct, the location of P on the image plane is simply defined by xp, yp. The corresponding bounding element i,j in the image plane can be found from Eqn. 1 , where xmin, ymin are the coordinates of the image plane origin and dx, dy are the resolution of the object x-ray. Hence, the robustness of the ray casting method is coupled with the efficiency derivable from the structure of uniform Cartesian grids. Further, the memory limitations of the hole-map method discussed earlier do not exist, since only planar data (as opposed to volumetric data) is required here. Use of object x-rays to cut chimera holes is simple. Object x-rays can be used any number of times, applied to any combination of grid components, and used to cut any size of hole about the respective objects. The hole cutting operations can be carried out in a fully automatic way. A crosssection of the geometry and object x-rays that intersecting the geometry are shown in Fig. 6. Consider a point P that P is associated with one of the grid components to be cut by the object. Further, any point within a small distance of the surface defined by the object x-ray is to be part of the hole. Their indices of the x-ray element that bounds P can be computed using Eqn. (1). Determination of the blank status requires comparison of the Z coordinate of P with that of pierce-points on the four rays that surround P in the image plane. The hole generated by the object ray is called the "minimum hole" because it is a cut from the actual object x-ray. Given the minimum hole, however, it is easy to expand the hole any distance away from the object. The x-ray method reduces the memory required while providing a better approximation of the geometry. 2.6 Direct cut approach A direct cut approach13 uses the surface grid of a geometry to remove the cells inside the component and is applied in two distinct processes. The first process will identify the grid elements that intersect the cutting faces. The second process is required to identify the remaining elements or points that are inside the geometry and should also be cut. In the first process, the direct cut method identifies the elements cut by the geometry. This is typically performed by finding the intersection of the cutting surfaces with the target grid and breaking the links in the grid at these intersections. It is the most accurate method, since it does not use an auxiliary approximation to the geometry. Once the cutting phase is completed, a fill operation is required to mark the points that lie interior to the bodies. The fill operation is initiated from some point that is known to be inside the geometry and iteratively sets neighbours to be inside the geometry. Neighbours across the links broken by the cutting process are not set as inside the geometry and iteration halts in that direction. If the cut is not complete and water tight, this fill operation can leak out of the actual geometry and erroneously mark the entire grid as inside the geometry. The “hole” in the geometry where the leak initiates can be difficult to locate. A robust fill procedure is critical to the success of the direct cut method. An efficient search process is required to limit rigorous tests to only reasonable cells. The present approach uses Alternating Digital Tree (ADT)14 for this purpose as it is a very efficient spatial decomposition data structure. The ADT optimizes the search operation to O(Log2 N) as opposed to O(N). A search tree is built for the cells of each grid with the root covers the complete domain. The tree is used to identify the possible overlapping cells of a grid with each cutting face of other grids. A list of candidate cells that overlap with the bounding box of a specific cutting surface face are found using ADT tree. Once the list of candidate cells has been obtained, each cell must be tested to determine if the 5 candidate cell overlaps the specific cutter surface face. Any edge of the element that is intersected by the face is marked as cut and the point of edge on the positive side of the face normal is marked as exterior to the body while the opposite point of the edge is marked as interior to the body. Intersection of an edge with a surface triangle and the points are marked as interior (I) and exterior (E) based on surface normal (n) are shown in Fig. 7. There are certain cases where a point can be marked as interior to a face and the point will be marked as exterior to some other face. This can occur for the thin-cut case where an element is cut by two faces that point in opposite directions such as an airfoil trailing edge. Therefore, both points of the edge are set as exterior, but the edge is set as cut so that they do not form neighbours during the connectivity generation. Determining the intersection between element edges and the cutting face is a critical component of the current hole cut procedure. Since, computing the intersection can be expensive, simpler tests are used to eliminate candidate intersections. An intersection can only occur if the end points of the edge lie on both sides of the face, i.e. one is above and other is below the face. Therefore, checking for an intersection of the element edge with the cutting face is only performed if one of the element edge nodes is above while the other is below the cutting face. The current process begins by testing the edges of the element for intersection with the cutting face. All element edges must be checked for intersections so that all appropriate element edges will be marked as cut. The pyramid, prism or hexahedral elements will have quadrilateral faces. The bilinear representation of the quadrilateral face is used to maintain consistency of representation of the face between two cells. The intersection tests use a parametric representation of the line segment, triangle, or quadrilateral with an intersection occurring when the parameters are between 0 and 1. The initial phase of the direct cut approach will only mark those elements intersected by the cutting faces as shown in Fig. 8. A second phase is required to mark the remaining nodes or elements that are interior to the body. This is usually implemented as a flood/fill operation that starts at a location know to be interior to the body. Neighbouring points or elements that were not cut are iteratively also marked as interior to the body. The procedure is terminated when no additional elements or points are marked. Once, the points are identified as interior, they are set as hole points, the exterior points of cut edges are set as fringe points and rest of the points are assigned as active points. The flow field information across various overlapping grids are exchanged through fringe points. The grid-based solver interpolates the flow field at the fringe point using the value in the cell, called as donor cell, which contains the fringe point. In the grid-free solver, the fringe points are also active points, but its connectivity include neighbours from all the overlapping grids. It is obvious that the vertices of the donor cell are neighbours of the fringe point. Therefore, it is critical to identify the donor cell for the fringe points from the overlapping grids. The process of identifying the donor cell is known as donor search. 3. Donor Search The grid-free method does not require donor cell for interpolation as in grid-based method. But, identification of donor cell helps in identifying the closest neighbours from the overlapping grids. The donor cell is identified using stencil jump algorithm15. The central element of this algorithm is testing whether a point is inside a cell or not. The fringe point is tested with a possible donor cell using the shape functions of the donor cell. The preprocessor has been developed to operate on hybrid grids whose elements are tetrahedron, pyramid, prism and hexahedron. Therefore, the shape functions are defined for the above element types and the local co-ordinates of the fringe point are iteratively solved. The local co-ordinates indicate whether the fringe point lies inside the cell or not. If the point does not lie inside the donor cell, then the adjacent cell in the direction of local co-ordinates of the 6 fringe point is checked. The cells are traversed till the actual donor cell is identified. The efficiency of the search depends on selection of the good initial condition for the donor search. An approximate inverse map (AIM)16 is used to identify the possible donor cell. In this method, as an initial step, virtual uniform Cartesian grid is defined over the domain. The grid cell whose centre closest to the Cartesian cell centre is stored in the Cartesian cell. During the donor search, using the co-ordinates of the fringe point, the Cartesian cell is identified and in turn possible donor cell is identified. This strategy is very efficient but typically results in a few off-body orphans, fringe points without donor cell. These orphans are generated because the donor-search fails near partition boundaries. There is no way of jumping across grid boundaries in this approach and therefore, the donor search fails. The method is improved by storing all the intersecting boundary cells, whose one or more faces are boundary faces, in the virtual Cartesian cell. During the donor search, if a boundary face is reached, the virtual Cartesian cell that contains the centre of the boundary face is identified. Starting from other boundary cells, that are stored in the corresponding virtual cell, the search is continued till the donor cell is identified. The search, with the present modification, fails only for the hole points and those points are set as hole points. This approach helps in hole cutting also and no separate hole cutting procedure is required. The holes are generated as a byproduct of the donor search. An Alternating Digital Tree (ADT) is also implemented for the donor search. The ADT method provides the most accurate and robust donor search, i.e., no orphans are generated. However, it can be more than an order of magnitude slower than the AIM method, which is unacceptable especially for dynamic mesh problems wherein domain connectivity needs to be carried out frequently during the solution process. 4. Development of pre-processor The connectivity generation requires various search algorithms and therefore, good data structure makes this process efficient and cost effective. The data structure consists of lists of cells, faces, edges, nodes and link them mutually with each other. Generalisation of this process implies minimal user input. Further, problems like store separation dynamics of missile from aircraft, require repeated connectivity generation by moving some component grids relative to other component grids. Some of the data structure related to a particular component grid does not change during relative displacements. Considering all these facts, the connectivity generation process is split into two codes. The first code generates all linked lists, surface and volume details which are fixed for a particular component grid. In this code, the component grids can be rotated and translated so that the component grids can be positioned relative to other grid or the component grids can be aligned to a particular axis for convenience. The memory intensive and complex operations are restricted to this stage. The wall distance which is used in the selection process of implicit hole cutting is also obtained by the code. The wall distances can also be used in the viscous simulation in the wall damping terms of turbulence model. The second code reads the above details and generates the connectivity information. The component grids can be again translated, if required at this stage. In this stage, the pre-processor carries out hole-cutting and donor search and generates the data-structure needed for a grid-free solver. During dynamic simulation, the second code is applied repeatedly to generate the data structure. The minimal inputs such as component name, component grid file name, three translation distances and three rotation angles are only needed to be given for both the codes. The pre-processor has been completely automated and various options are given for hole cutting and donor search. The pre-processor first marks the points as hole, fringe or active respectively using hole cutting information. Donor cell is identified for all the fringe points. The connectivity is generated for active and fringe points using grid information and edge cut status. Further, the vertices of the donor cell are added to 7 the connectivity of the fringe point and flag it as an active point. The boundary flag is set to the corresponding boundary points. The surface tangents and normal (pointing inward) at the boundary points are estimated using the area averaged of those values of the surface faces that share the points. The wall surface and volume information is written for force integration and flow visualisation. 5. Applications The pre-processor has been applied to three cases to verify the capability of the code. The unstructured grids are generated around each component and the pre-processor is applied to generate the data structure. The geometries are very complex and therefore, X-ray hole cutting is used to identify the hole points. Then, the donor search is applied along with AIM to obtain the donor cells for all the points except hole points. The ADT method is applied only for the orphan points which could not find donor cell in the previous step. It was observed that there was no orphan point present in all the cases. Then, the implicit hole cutting method is applied to reduce the active region. After identification of active points and fringe points, the connectivity is generated using respective grid information for those points and neighbours of fringe points are updated using donor cells of overlapping grids. Then, the surface outward normal for the boundary points are estimated. The results for three cases are presented in the following sub-sections. (i)Wing-Store The combined wing-store geometry is shown in Fig. 9. In this case, unstructured grids with size of 125 and 79 thousand cells are generated around wing and store respectively. A typical cross section of overlapped grids is shown in Fig. 10. The red and green colour grids are around wing and store respectively. After applying the pre-processor, the points are set as active, fringe and hole points. The cells connecting hole points are shown in Fig. 11. It can be observed that apart from the cells inside wing in the store grid, the cells above the wing are also marked as hole cells. It can be recalled that the implicit hole cutting optimise the overlap region by the cell selection process. Therefore, minimum numbers of cells are marked as active and hence, the flow information exchange takes place across the component grids at minimum number of points. This can clearly be seen in the Figs. 12 and 13 which show the fringe cells and active cells respectively. The grid-free flow solver GEANS is applied on this data-structure to simulate the inviscid flow past wing-store configuration. The free-stream condition of Mach 0.8 and AOA 0o has been simulated. The pressure contours and Mach contours are shown in Figs. 14 and 15. The contours are smooth and all the flow features are captured. (ii) Fighter aircraft-store The pre-processor is applied to a fighter aircraft with an axi-symmetric store. The configuration is shown in Fig. 16. The unstructured grid sizes of 2.5 and 0.6 million cells are generated around aircraft and store respectively. A typical cross-section of the overlapped grids is shown in Fig. 17. The blue and red colour grids are around aircraft and store respectively. The wall distances, that is used in the cell selection process, at a cross-section is shown in Fig. 18. The hole cells, fringe cells and active cells are shown in Fig. 19, 20 and 21 respectively. As mentioned earlier, the minimum overlap of cells are marked as fringe cells. This helps in stability by marking as many as bad cells as hole cells and hence, improves the solution accuracy. The grid-free flow solver is applied on this cloud of points at a free stream Mach number 0.8. The surface pressure contours are shown in Fig. 22 and it can be observed all the flow features such as compression near nose and wing, tail leading edges and flow expansion on the aircraft fuselage are captured accurately. 8 (iii) Separation of a Fairing of a launch vehicle Next case considered is a launch vehicle and its two separated fairings. The unstructured grids are generated around vehicle and two fairings with size of 1.4, 0.17 and 0.19 million cells respectively. The geometry is shown in Fig. 23 and the overlapped grids are shown in Fig. 24. The red, blue and green colour grids are around launch vehicle and two fairings respectively. The pre-processor is applied to generate connectivity. The hole cells, fringe cells and active cells are shown in Figs. 25, 65 and 27 respectively. The effect of wall distance in the cell selection process can be clearly seen from the figures such that fringe cells are far from the solid boundaries. The flow solver is applied at free stream Mach number 0.7. The surface pressure contours are shown in Fig. 28. The stagnation pressure is observed near the nose and vehicle attachment plates. The pressure gradients along the body is less except near stagnation points. 6. Conclusions and further work A pre-processor has been developed to operate on overlapped grids to generate connectivity for grid-free solvers. Efficient algorithms have been implemented for hole cutting and donor search. The pre-processor has been applied to three cases, viz., wing-store, aircraft-store and launch vehicle with fairings to generate connectivity. The grid-free solver is applied and proven the functioning of solver on the connectivity. As an immediate next goal would be application of the pre-processor on hybrid grids to generate connectivity for viscous solver and to be integrated with rigid body dynamics code for store separation studies. Acknowledgements The authors express their sincere gratitude to the Director, DRDL for his support and encouragement throughout the work. They wish to thank Mr. Ankit Raj, Scientist, CFD Division, DRDL for providing unstructured grids around various geometries. References 1. Anandhanarayanan, K., “Development of a Grid-free Viscous Solver", DRDL report no. DRDL/5261/DOCD/CFD/TECH/08-04, 2008. 2. Benek, J. A., Steger, J., and Dougherty, F., “A Flexible Grid Embedding Technique with Applications to the Euler Equations,” Paper 83-1944, AIAA, 1983. 3. Belk, D. and Maple, R., “Automated Assembly of Structured Grids for Moving Body Problems”, Proceedings of 12th AIAA Computational Fluid Dynamics Conference, AIAA Paper 95-1680-CP, San Diego, CA, 1995. 4. R. L. Meakin, “Computations of the Unsteady Flow about a Generic Wing/Pylon/Finned-Store Configuration”, Paper AIAA 92-4568, AIAA, AIAA Atmospheric Flight Mechanics Conference, 1992. 5. R. L. Meakin, “Unsteady Simulation of the Viscous Flow about a V-22 Rotor and Wing in Hover”, AIAA Atmospheric Flight Mechanics Conf., 95-3463-CP, 1995, pp. 332-344. 6. Anandhanarayanan, K., “Development of Three Dimensional Grid-Free solver and its Applications to Multi-body Aerospace Vehicles", Defence Science Journal, Vol. 60, No. 6, Nov. 2010, pp. 653-662. 9 7. G. Harish, M. Pavanakumar, and K. Anandhanarayanan, “Store separation dynamics using grid-free Euler solver”, AIAA Paper No. 2006-3650, (2006). 8. Lee Y, Baeder J. Implicit hole cutting—a new approach to overset grid connectivity. Sixteenth AIAA Computational Fluid Dynamics Conference, AIAA2003-4128, Orlando, FL, 2003. 9. R. L. Meakin, “A New Method for Establishing Intergrid Communication Among Systems of Overset Grids”, Tech. rep., Honolulu, HI, 1991. 10. K. Anandhanarayanan, M. Nagarathinam, and S.M. Deshpande, “Development and applications of A Grid-free Kinetic Upwind Solver to Multi-body Configurations”, AIAA Paper 2005-4628, June 2005. 11. Suhs, N., Rogers, S., and Dietz, W., “PEGASUS 5: An Automated Pre-processor for Overset-Grid CFD,” Paper 2002-3186, AIAA, 2002 12. R. L. Meakin, “Object X-Rays for Cutting Holes in Composite Overset Structured Grids”, Paper 2001-2537, AIAA, 2001. 13. R. W. Noack, “A Direct Cut Approach for Overset Hole Cutting”, Paper 20073835, AIAA, 18th AIAA Computational Fluid Dynamic Conference, Miami, FL, 2007. 14. J. Bonet and J. Peraire, “An Alternating Digital Tree (ADT) Algorithm for 3D Geometric Searching and Intersection Problems”, International Journal of Numerical Methods in Engineering, Vol. 31, 1991, pp. 1–17. 15. Lohner, R., “Robust, Vectorized Search Algorithms for Interpolation on Unstructured Grids,” Journal of Computational Physics, Vol. 118, 1995, pp. 380– 387 16. Roget, B. and Sitaraman, J., “Robustness and Accuracy of Donor Search Algorithms on Partitioned Unstructured Grids”, 10th Symposium on Overset Composite Grids and Solution Technology, NASA Ames Research Center, Moffett Field, CA, September, 2010 hole points Fig. 1 Overlapped structured grids around the flight vehicle 10 Fig. 2 Overlapped unstructured grids around launch vehicle and fairings P nQ Q S rP rQ Z Y X Fig. 3 Surface normal test Fig. 4 Hole map x-rays . P Image plane Image plane xmin,ymin Fig. 5 Object x-ray and image plane E V1 Fig. 6 Hole point marking using object x-ray n grid1 V3 Cut-cells I V2 Fig. 7 Intersection of an edge with a triangle Fig. 8 Cut cells of grid1 by surface triangles 11 Fig. 9 Wing-store geometry Fig. 10 Overlapped meshes around wing-store Fig. 11 Hole-cells of wing-store grids Fig. 12 Fringe cells in wing-store grid Fig. 13 Active cells in wing-store grid Fig. 14 Surface Mach contours on wing-store 12 Fig. 15 Surface Pressure contours on wing-store Fig. 16 Aircraft store geometry 6000 z 4000 2000 0 -2000 0 2000 4000 6000 8000 y Fig. 17 Overlapped grids around aircraft-store Fig. 19 Hole-cells of aircraft-store grids Fig. 18 Wall distances around aircraft-store Fig. 20 Fringe cells of aircraft-store grids 13 Fig. 21 Active-cells of aircraft-store grids Fig. 22 Surface Mach contours on aircraft-store Fig. 23 Geometry of launch vehicle with fairings Fig. 25 Hole-cells of launch vehicle grids Fig. 24 Overlapped grids around launch vehicle Fig. 26 Fringe cells of launch vehicle grids 14 Fig. 27 Active-cells of launch vehicle grids Fig. 28 Surface pressure contours on launch vehicle 15