Flame Front for a Spark Ignited Engine Using Adaptive Mesh Refinement (AMR) to Resolve Turbulent Flame Thickness New Trends in CFD II Fully Coupled Automated Meshing with Adaptive Mesh Refinement Technology: CONVERGE CFD
OVERVIEW 1. Introduction - CONVERGE CFD 2. Background 3. Automated Meshing in CONVERGE 4. Grid Control 5. Efficient mesh utilization and Grid Convergence 6. Applications Slide 2 of 29
CONVERGE CFD CONVERGE is a multi purpose 3D CFD Software. Special features, e.g. : Automatic grid generation with adaptive refinement Genetic algorithm Detailed chemistry Fluid Structure Interaction Conjugate Heat Transfer Very efficient where complex geometries and chemistry are to be modeled Specialization on engine simulation Auto-ignition locations (red) and liquid spray droplets for dual fuel (Isooctane+Diesel) test case. Temperature contours for stratified charge SI engine using the SAGE detailed chemistry solver coupled with adaptive mesh refinement (AMR)
GEOMETRY DEFORMATION IN CONVERGE Geometry deformation (translation, rotation, arbitrary) is handled via input files or automatically (Piston) Mixer Geometry for throttling analysis (throttle shown in red) Moving Tailgate Gas turbine Nozzle flow simulation Two stroke engine with special crank kinematics
TRADITIONAL CFD METHOD FOR MOVING BOUNDARIES Requires a grid to be made before the calculation Significant amount of user time and effort is needed to create the grid Since mesh is made a-priori, the user must guess where mesh resolution is required Experienced meshing engineer needed with extensive training Once the mesh is generated, it is fed into a solver and then mesh motion is imposed Poor quality cells resulting from mesh motion Additional numerical errors associated with convection terms Crashes (solver problems) often accompany such mesh motion Special treatment for seating valves is often required Once a mesh is made, and the solver runs without crashing with mesh motion, the models are typically tuned to achieve desired results Grid independence studies are almost never performed due to the time and effort required to make a moving mesh User to user variations in results due to manual meshing approach make coordinating a team of CFD engineers difficult Meshing issues are often the limiting bottleneck for both accuracy and productivity in engine CFD analysis User 1 User 2 User 3 Representative plot showing user to user variation in results for same geometry using same CFD code
CONVERGE METHOD Mesh is made at runtime automatically by CONVERGE Meshing algorithm is extremely fast and runs in parallel The base mesh size is specified as a simulation input This easily allows for a suite of runs, each with a different base mesh size, to be performed which can help judge the grid dependence of the result The mesh elements are orthogonal and stationary Ideal numerically and eliminates convection errors All boundary motion and valve closure handled automatically Adaptive mesh refinement (AMR) is used to automatically enhance mesh resolution where it is needed based on gradients of field variables Adequate mesh resolution is essential for accurate predictions (without tuning) CONVERGE always keeps the initial geometry surface definition regardless of the mesh resolution Mesh elements are added when and where they are needed to minimize grid dependency of results at the lowest possible computational cost Different analysts using the standardized mesh settings will get essentially the same answer User Velocity 1 contours for gas exchange simulation User 2 User 3 Representative plot showing consistent user to user results using CONVERGE
AUTOMATED MESH GENERATION IN CONVERGE STL geometry is immersed within a Cartesian Block Bounding box The surface is mapped onto a Cartesian grid located within the bounding box Cells intersected by the boundary surface are trimmed or cut to generate body fitted mesh Bounding box Cut cell Cartesian method Interior cells are orthogonal hex elements and remain stationary, with a base grid size specified by the user Cells intersected by the geometry surface are cut by the surface, thus forming arbitrary sided polyhedra The resulting mesh used by the solver is body fitted and is an exact representation of the surface geometry Representative Cut Cell Mesh - Generated Automatically at Interior cells are orthogonal and stationary Runtime by CONVERGE Polyhedra cut cell elements generated at surface intersection
CONVERGE WORKFLOW Import surface mesh (STL format) and repair surface (if needed) Flag Boundaries Export Surface File Specify Inputs Solve Post-Process Results At no point does the user spend any time meshing Most engine cases can be running within an hour even with moving valves (assuming a relatively clean CAD file is used)
GRID CONTROL With CONVERGE, the user has full control over the mesh sizing through four methods: 1. The base mesh size 2. Adaptive Mesh Refinement (AMR) mesh automatically enhanced based on gradients 3. Grid Scaling the entire mesh (i.e. base mesh size) is scaled at prescribed times 4. Grid embedding - local refinement based upon user inputs. The user can specify as many grid embedding methods as desired Add mesh elements only when and where The gridding methods can be activated and deactivated at various times throughout the they are needed such simulation that results can be trusted (grid independent) The original surface definition is always maintained
ADAPTIVE MESH REFINEMENT AMR enhances the mesh resolution automatically based upon gradients in field variables (temperature and velocity, for example) With AMR, the user specifies a maximum number of cells, which in conjunction with the base grid size, determines the overall mesh resolution and run times The maximum cell limit forces AMR to make difficult tradeoffs adding mesh resolution when and where it is needed This results in a very efficient usage of mesh elements y+ AMR is available to resolve the wall y+ down to a targeted level Add mesh elements only when and where they are needed such that results can be trusted (grid independent) Boundary embedding used (extra mesh added to valve seat immediately before valve motion begins to resolve small flow area) AMR used to resolve flow where gradients are large Interior nodes are stationary and orthogonal Representative Mesh in Valve Region Both AMR and Grid Embedding Used
Boundary embedding is commonly used to add mesh resolution near a boundary (typically a wall) For a moving surface, like a valve, the embedding will move with the surface automatically The user can specify between which crank angles the embedding is active EMBEDDING Sphere embedding allows for the mesh resolution within a given sphere to be altered at a specified time. This is commonly used to resolve the spark plug region immediately before the spark event. Intake Valve Seat Flagged for Boundary Embedding Cylinder embedding is commonly used to control the ports, valves and cylinder mesh densities Nozzle and Injector embedding will enhance the resolution of the spray region The position and targeting of the embedding is linked to the injector / nozzle. Boundary Embedding Entire Valve Surface
GRID CONTROL EXAMPLE - 1 Boundary embedding to resolve valve seat area Cylinder embedding (two total) used for initial mesh AMR used to capture flow areas of large gradients Add mesh elements only when and where they are needed such that results can be trusted (grid independent) Representative mesh for a clip plane through throttled port Representative Meshing Strategies Used for GDI engine
GRID CONTROL EXAMPLE - 2 Nozzle Embedding Boundary Embedding (1) Coarse base grid run during compression for minimal run time (3) Before injection, both nozzle and boundary embedding are used to further resolve spray and wall effects (2) Near TDC, base cell size is reduced by a user specified factor using grid scaling (4) Final mesh using embedding and AMR Representative Meshing Strategies Used for Diesel Sector Case Convergent Science Inc. Proprietary
EFFICIENT MESH UTILIZATION Embedding and AMR allow the mesh to be used very efficiently providing mesh resolution only when and where it is needed to minimize the cell count To show this, Case 2 was run using a constant mesh size equivalent to the minimum mesh size from Case 3 (with embedding and AMR) As seen below, the results for Case 1 and Case 2 are essentially identical However, the mesh size at +100 degrees is 28 times higher for Case 2 Therefore, AMR and embedding allow the same answers to be obtained as a fine, uniform mesh at a fraction of the computational cost Case 1 With spray embedding With adaptive mesh refinement (AMR) Minimum cell size = 0.625 mm Cell count at +100degrees = 46,000 Case 2 Constant cell size = 0.625 mm Cell count at +100degrees = 1,290,000 Temperature contours for Diesel simulation Pressure versus crank angle for Case 1 (red) and Case 2 (green).
GRID CONVERGENCE 14 Grid-Convergent Engine Simulations 14 14 2 million cells utilized 12-2SOC Measured CONVERGE - Full Geometry 12 2SOC Measured CONVERGE - Full Geometry 12 6SOC Measured CONVERGE - Full Geometry 10 10 10 Pressure (MPa) 8 6 Pressure (MPa) 8 6 Pressure (MPa) 8 6 4 4 4 2 2 2 0-40 -20 0 20 40 60 Crank Angle (deg. ATDC) 0-40 -20 0 20 40 60 Crank Angle (deg. ATDC) 0-40 -20 0 20 40 60 Crank Angle (deg. ATDC) NOx (ppm) 400 350 300 250 200 150 100 50 Measured CONVERGE - Full Geometry 0-2 0 2 4 6 SOC (deg. ATDC) OH* Chemiluminescence vs. OH iso-surface, 7.5 degrees after SOC The grid-convergent methodology results in excellent agreement with global combustion behavior as well as flame lift-off length and flame location Collaboration with Caterpillar Inc. and Sandia National Laboratories *Senecal et al. ASME 2013
ENGINE AND NON-ENGINE APPLICATIONS Slide 16 of 29
Any motion can be pre-described FLUID STRUCTURE INTERACTION Forces on the surfaces can be used to calculate movement or deformation Resulting motion or deformation can be calculated via UDF (User defined boundary deformation) or in conjunction with FE tools (e.g. stresses) or coupled to GT-SUITE or Matlab 1 3 2 4 17
GEOMETRY OPTIMIZATION Coupling to 3 rd Party optimization tools (e.g. DOE software) Included Genetic Algorithm: CONGO automates the process of running a Genetic Algorithm (GA) in CONVERGE A GA takes a survival of the fittest approach to optimize a design Geometry optimization possible with CONGO by parameterization of features Coupling to external surface deformation tools possible: Piston Bowl Optimization Control Points Arbitrary geometry optimization (not restricted by finite set of parameters) 18
SUMMARY Automated meshing in CONVERGE eliminates all user meshing time As the base grid resolution is specified in an input file, grid sensitivity studies can be easily performed Mesh is extremely high quality leading to accurate results Internal cells are orthogonal with their neighbors which is ideal for the solver algorithm Internal cells remain stationary thus removing numerical diffusion associated with a moving mesh Meshing algorithm can easily handle any geometry of interest including irregular piston bowls, spark plugs, glow plugs, closely spaced valves, pre-chambers, etc. CONVERGE always maintains the correct geometry surface definition, regardless of the mesh density Areas and volumes are always exact regardless of mesh resolution This allows for extremely course meshes to be used while setting up a case The solver and meshing algorithm are tightly coupled (mesh files are never created), which makes parallel processing simple and easy to the user User-to-user meshing variations are mostly removed allowing for a team of CFD analysts to work together with standardized workflow Geometry changes can be considered quickly and easily (i.e. changing a piston shape) Training new users is very quick (approximately two days) as there is no meshing to learn
THANK YOU FOR YOUR ATTENTION Flame propagation in a gasoline direct injected engine