sans contributions to case cr1 - mit · sans contributions to case cr1 marshall c. galbraith...
TRANSCRIPT
SANS Contributions to Case CR1
Marshall C. Galbraith
Aerospace Computational Design LaboratoryDepartment of Aeronautics and Astronautics
Massachusetts Institute of Technology
Galbraith (MIT) SANS Contributions to Case CR1 6 January 2018 1 / 12
SANS
Solution Adaptive Numerical Simulator (SANS):PDE agnostic (advection-diffusion-reaction)
RANS: negative-SAAutomatic Differentiation
Stabilized-CG, DG, HDG, EDG discretizationsDG-BR2 used here
Linear SolversGMRESUMFPACK, MKL-PARDISO, and PETSc interfaces(PETSc GMRES with ILU used here)
Non-linear SolverNewton-Raphson with line searchPTC and P-Sequencing used here
Output-based mesh adaption using MOESSminimize estimated error in output functionalsubject to DOF count ≤ set cost
Galbraith (MIT) SANS Contributions to Case CR1 6 January 2018 2 / 12
Joukowski Airfoil Verification
SUPG O(h2P), asymptotically dual consistent?
Galbraith (MIT) SANS Contributions to Case CR1 6 January 2018 3 / 12
CRM High-order Workshop Meshes: CD
Mach=0.3. Decreased variation with increased P and Q.Galbraith (MIT) SANS Contributions to Case CR1 6 January 2018 4 / 12
CRM High-order Workshop Meshes: CL
Mach=0.3. Relatively small variation.Galbraith (MIT) SANS Contributions to Case CR1 6 January 2018 5 / 12
Mesh Adaptation
Initial Mesh
Boeing metric based mesh generator EPICCurving not robust enough for adaptationPoor DG P1 Q1 discretization for adaptation
Target: 2.0M DOF (625 k elements) with drag adjoint
Galbraith (MIT) SANS Contributions to Case CR1 6 January 2018 6 / 12
Mesh Adaptation
Initial Mesh
Boeing metric based mesh generator EPICCurving not robust enough for adaptationPoor DG P1 Q1 discretization for adaptation
Target: 2.0M DOF (625 k elements) with drag adjointNot enough DOF for poor P1 Q1 DG discretizationI’m gonna need a bigger box
Galbraith (MIT) SANS Contributions to Case CR1 6 January 2018 6 / 12
Adapted Mesh: Isometric View
Initial0.65M DOFs
Adapted2.3M DOFs
Galbraith (MIT) SANS Contributions to Case CR1 6 January 2018 7 / 12
Adapted Mesh: Side View
Initial0.65M DOFs
Adapted2.3M DOFs
Galbraith (MIT) SANS Contributions to Case CR1 6 January 2018 8 / 12
Adapted Mesh: Side View Crinkle Cut
Initial0.65M DOFs
Adapted2.3M DOFs
Galbraith (MIT) SANS Contributions to Case CR1 6 January 2018 9 / 12
Adapted Mesh: Fairing Side View
Initial, 0.65M DOFs Adapted, 2.3M DOFs
Galbraith (MIT) SANS Contributions to Case CR1 6 January 2018 10 / 12
Adapted Mesh: Wing Root
Initial, 0.65M DOFs Adapted, 2.3M DOFs
Galbraith (MIT) SANS Contributions to Case CR1 6 January 2018 11 / 12
Summary
Small variation in CDFor P2 less than 1 count across all meshesHow coarse can we go?
Relatively small variation in lift
Mesh AdaptationNeed robust metric conforming mesh curvingPoor P1 Q1 needs more DOF.Lack of resolution in boundary layers.
Galbraith (MIT) SANS Contributions to Case CR1 6 January 2018 12 / 12