Fully Automated, Parallel and Topology Agnostic Domain Connectivity For Overlapping Surface Meshes Fields

Transcription

1 Overset and Composite Grid Symposium, University of Maryland Oct 1-3, 2018 Fully Automated, Parallel and Topology Agnostic Domain Connectivity For Overlapping Surface Meshes Fields Approved for public release; distribution unlimited. Review completed by the AMRDEC Public Affairs Office (PR3924, 18 June 2018) Jay Sitaraman Mark Potsdam Beatrice Roget PGA LLC Army ADD Science and Technology Corporation Aviation and Missile Research, Development and Engineering Center Moffett Field, CA

2 Who is AMRDEC? U.S. Army Aviation and Missile Research, Development, and Engineering Center Core Competencies Life Cycle Engineering Research, Technology Development and Demonstration Design and Modification Software Engineering Systems Integration Test and Evaluation Qualification Aerodynamics/ Aeromechanics Structures Propulsion Guidance/Navigation Autonomy and Teaming Radio Frequency (RF) Technology Fire Control Radar Technology Image Processing Models and Simulation Cyber Security 2

3 AMRDEC Mission Deliver collaborative and innovative aviation and missile capabilities for responsive and cost-effective research, development and life cycle engineering solutions. 3

4 AMRDEC Priorities #1: Readiness Provide aviation and missile systems solutions to ensure victory on the battlefield today. #2: Future Force Develop and mature Science and Technology to provide technical capability to our Army s (and nation s) aviation and missile systems. #3: Soldiers and People Develop the engineering talent to support both Science and Technology and the aviation and missile materiel enterprise 4

5 Introduction and Motivation Overlapping surfaces is ubiquitous in structured grid based overset methods Unstructured grid based overset methods most often do not require surface overlap solution since body can be easily closed with a single mesh But.. extending the solution of surface overlap problem to general (unstructured and structured) systems can support advanced simulations with sliding contacts. 5

6 Motivation Sikorsky Raider V-280 Valor Gaps are often left in models since sliding surfaces are hard problems 6

7 Motivation Upper hub Middle body For example Raider Coaxial Hub has no gaps But we have modeled them to date with gaps because of inability to model sliding contacts. Lower hub 7

8 Motivation Shaft grids! Upper hub Very easy to construct as structured grids Middle body Lower hub 8

9 Design Goals Run-time overset grid assembly, i.e. not a pre-processor Fully parallel - will use solver generated mesh partitioning at run time. Topology agnostic - individual meshes can be solved by structured grid based or unstructured grid based solvers in the same framework. Fully automated minimal additional information for bodies with overset surfaces, i.e. no X-rays or other hole-maps that need to be generated ahead of time. Will be integrated into CREATE A/V TM Helios and Kestrel for production use. 9

10 Software PUNDIT Parallel Unsteady Domain Information Technology Developed under the HPCMP CREATE A/V TM and manages all overset grid assembly functions in Helios and Kestrel production codes. Modular and fully automated -- only minimal essential data on a per process basis, i.e. following items per mesh partition Coordinates of grid nodes Volume connectivity of grid nodes (i.e. tets/prisms/pyra/hex ) Index of nodes on a wall boundary Index of nodes on an overset boundary Implicit Hole-cutting approach Active load balancing available Domain connectivity successfully demonstrated up to 20k cores. 10 journal papers (

11 Multi-code Operation in Helios overflow/mstrand fun3d,/kcfd/nsu3d/mstrand 11

12 Implicit-hole-cutting (1) Domain connectivity procedure attempts to find donor cells for all mesh points. Donors are selected if they have better resolution capacity Resolution capacity : Heuristic parameter that quantifies solution quality 12

13 Implicit-hole-cutting (2) Hole points: Mesh points that are inside a solid wall 13

14 Implicit-hole-cutting (3) Receptor Points: Mesh points that could find donor cells of better resolution capacity flow solution will be interpolated to these points Some points are mandatory receptors: Neighbors of hole points Neighbors of outer boundary 14

15 Implicit-hole-cutting (4) Automated procedure to identify point type (field, hole or receptor) minimal mesh overlap 15

16 Identifying Minimum Holes (1) Find bounding box of wall faces (gather info from all processors) Partitioned mesh Outer boundary Hole Bounding Box 16 Local wall bounding boxes

17 Identifying Minimum Holes (2) Create structured auxiliary mesh (SAM), refine until no sub-block (SB) contains both an outer face and wall face Exact Face-to-Cube intersection Checks Hole (wall) sub-blocks 17 Sparse matrix type storage for the hole-map using linked lists Hole sub-block containing outer cell faces (at coarser level)

18 Identifying Minimum Holes (3) Perform flood-fill to identify all hole sub-blocks Hole (inner) sub-blocks Outer layer of sub-blocks tagged as OUT (seeds of flood-fill algorithm) 18 Query point in hole SB AND no donor in hole mesh HOLE POINT Query point in hole SB, but donor cell exists in hole mesh Not a Hole Point

19 19 Exact Minimum Holes

20 Challenges Two challenges 1. Extension of the minimum hole-cutting algorithm for topologically open meshes 2. Projection problem to deal with non-matching discrete surfaces can occur between multiple topologically different meshes. Extra inputs needed: 1. Sub-body list for each composite body, i.e. compbody[1]=[1,2,3] etc 2. Projection threshold (optional) 3. Dominance of each body, i.e. cutter vs cuttee (optional) Maintain the original design philosophy, i.e. fully parallel run-time connectivity and no user input or intervention other than the strictly required ones 20

21 Composite Bodies Bodies composed of several meshes with matching surface overlap Meshes can add body protrusions: or remove parts of body: Mesh 2 Mesh 2 Mesh 1 Composite Body Composite Body Meshes can have relative motion: Mesh 1 Mesh 2 21

22 Finding Hole Points Due to Composite Bodies Each mesh computes its own wall map based on global SAM Outer boundary Wall Outer boundary Wall Refine until no SB contains both Wall and Outer faces Exclude fringe to make this feasible 22

23 Finding Hole Points due to Composite Bodies Each mesh computes its own wall map based on global SAM Outer boundary Wall Outer boundary Wall Refine until no SB contains both Wall and Outer faces Exclude fringe to make this feasible 23

24 Finding Hole Points Due to Composite Bodies Each mesh computes its own wall map based on global SAM Outer boundary Wall Outer boundary Wall Refine until no SB contains both Wall and Outer faces Exclude fringe to make this feasible 24

25 Finding Hole Points Due to Composite Bodies Combine into complete wall map for entire composite body: True wall SB after flood fill 25 Flood fill from outer SB until encountering Potential Wall True Wall Propagate True Wall tag to face neighbors that are Potential Wall ONLY IF tagger meshes are identical

26 Finding Hole Points Due to Composite Bodies Combine into complete wall map for entire composite body: Final wall map 26 Flood fill from outer SB until encountering Potential Wall True Wall Propagate True Wall tag to face neighbors that are Potential Wall ONLY IF tagger meshes are identical

27 Finding Hole Points Due to Composite Bodies Decision logic for hole points: Wall SB tagged by mesh 2 Wall SB tagged by both mesh 1 and mesh 2 no Point in INSIDE SB? no Point in (true) WALL SB? yes Donor cells in ALL meshes tagging SB as wall? yes yes no Wall SB tagged by mesh 1 Not a Hole Point Hole Point 27

28 Finding Hole Points Due to Composite Bodies Decision logic for hole points: HOLE POINTS Wall SB tagged by mesh 2 Wall SB tagged by both mesh 1 and mesh 2 no Point in INSIDE SB? no Point in (true) WALL SB? yes Donor cells in ALL meshes tagging SB as wall? yes yes no Wall SB tagged by mesh 1 Not a Hole Point Hole Point 28

29 Cutter Meshes Parts of the body can be removed by cutter meshes: Outer Mesh 2 boundary Wall Mesh 1 Outer boundary Wall Wall Composite Body 29

30 Cutter Meshes Extra step: 1/ tag SAM sub-blocks containing cells from cutter mesh, except wall SB in the exclude region Mesh SB These SB can not be WALL For cuttee mesh Cutter mesh Cuttee mesh SB in exclude region will not be tagged as WALL Let cuttee mesh tag as wall 2/ Prevent cuttee mesh from tagging those as WALL. 30

31 Cutter Meshes Create wall maps Outer boundary These SB can not be WALL Wall Wall 31

32 Cutter Meshes Combine wall maps Outer boundary 32

33 Cutter Meshes Flood fill + propagate Outer boundary Flood fill Neighbor propagation 33

34 Cutter Meshes Decision logic for hole points: Wall SB tagged by mesh 2 no Point in INSIDE SB? no Point in (true) WALL SB? yes yes Donor cells in ALL meshes tagging SB as wall? yes no Wall SB tagged by mesh 1 Not a Hole Point Hole Point 34

35 Cutter Meshes Decision logic for hole points: Wall SB tagged by mesh 2 no Point in INSIDE SB? no Point in (true) WALL SB? yes yes HOLE POINTS Donor cells in ALL meshes tagging SB as wall? yes no Wall SB tagged by mesh 1 Not a Hole Point Hole Point 35

36 Projection problem Discrete geometry causes mismatch between grids that are made on the same CAD surface Problem: 36 Find the best possible interpolant - i.e match the local wall distance of the donor with the receptor

37 Projection Algorithm Isolate nodes that need projection Have to be part of a mesh in the composite body list Have to be in the group of points isolated to be send for searching to mesh partition belonging to a sister mesh. Have to be at a distance within the projection threshold from its wall. For each node that need projection (1) Find closest point on the wall of its own mesh (A) (2) Find closest point to A on wall of mesh B (3) Move P by vector (B-A), with a decay P d P = P + (B-A)*decay(d) A B 37

38 Projection Smooth decay of displacement such that at a large distance the original and projected points match 38

39 Closest point on Quadrilateral fix Wall distance/closest point evaluation between a bilinear quadrilateral and a point in space requires iterative solution that performs a constrained minimization. Wall distance evaluation between a triangle and a point does not need an iterative solution -- for wall distance purposes quads are split into triangles However, closest point found on the sub-triangle may not lie on the actual bilinear quadrilateral. Q = T1 U T2 Find (u,v ) in Q that correspond to (u,v) in T1 or T2 T2 T1 (u,v ) can be solved using a quadratic formed by elimination of terms in the mapping equations. Once (u,v ) is found exact point on the quad face can be obtained 39 Lagae and Dutre, Efficient Ray Quadrilateral Intersection Test, Journal of Graphics Tools, 2005

40 40 RESULTS

41 Case1: Rotating Cylindrical Bearing FUN3D SAMCART OVERFLOW Before Domain Connectivity After Domain Connectivity 41

42 42 Flow Field

43 Case 2: PSUHUB wing interaction Hub and stand used FUN3D unstructured sovler Wing used OVERFLOW structured solver Off-body cartesian system uses SAMCART PUNDIT performs Domain Connectivity between all grids 43

44 Mesh Trimming Tunnel walls use prism layers to better handle viscous and inviscid wall spacings Allow for sufficient off-body overlap Cut distance hub: 1.4 inch stand: 4.0 inches 44

45 Unstructured Mesh Hole-cutting hub mesh stand mesh blanked out near-body near-body hole cuts 45

46 Wing With Unstructured/Structured Surface Overlap Unstructured stand / tunnel wall mesh (no wing) (FUN3D) 2.3M nodes Structured wing grid (OVERFLOW) 6.6M nodes (311 x 301 x 71) Improved spacing compatibility structured/ unstructured overlapping surfaces unstructured structured structured Cartesian Cartesian 46

47 OVERFLOW Wing 47 Good overset compatibility and contour continuity in overlapped surface area

48 Trailing edge view 48 Good overset compatibility and contour continuity in overlapped surface area

49 Movie Baseline Hub Overset grid assembly time/step : 7.5 seconds (about 20% of overall time step) Composite body processing and Projection overhead : 0.5 seconds 49

50 50 Movie Low Drag Hub

51 Hub Drag Coefficient A = in 2 Helios mean and phase-averages exclude the first 3 revolutions C D run repeatability < 0.6% C D rev-to-rev variability < 0.8% C D PSU Exp Phase III 0.88 Helios baseline ± Helios baseline (OV) ± Helios low drag ± Helios low drag (OV) ± baseline low drag

52 Instantaneous Flow Fields Vorticity Magnitude FUN3D wing OVERFLOW wing 52

53 Case3: Coaxial hub problem Counter rotating hubs on top of each other Upper hub Middle body Lower hub Hubs are complex, usually hard to grid with purely structured grids and quite easy to grid with unstructured/strand grid Modeling the non-physical gaps between the hubs sometimes cause the flow solvers to converge poorly within the gaps degrading the overall convergence. How can we prevent these gaps? 53

54 Coaxial hub problem Shaft grids! Upper hub Very easy to construct as structured grids Middle body Lower hub 54

55 Shaft middle body Middle body 55

56 56 Lower Hub

57 57 All meshes (Shaft added)

58 58 After Domain Connectivity (Hub and Middle body get cut)

59 Conclusions and Future Work Extended parallel implicit hole-cutting algorithm for composite bodies with surface overlap Projection and minimum hole-cutting overhead is minimal ~10% Demonstrated successful application to model problems and several real problems (only two presented here). Reduction of forces accurately is on going work Will follow USURP type approach by conservatively closing the cut surface grids. Force computation in Helios is done outside of flow solvers, i.e the cloud of points with forces on it that each solver sends, need to be weighted by a fraction. Determining this fraction to satisfy conservation constraints is the bulk of this work. 59

60 AMRDEC Web Site Facebook YouTube Public Affairs 60