Public Release: Native Overset Mesh in FOAM-Extend

Transcription

1 Public Release: Native Overset Mesh in FOAM-Extend Vuko Vukčević and Hrvoje Jasak Faculty of Mechanical Engineering and Naval Architecture, Uni Zagreb, Croatia Wikki Ltd. United Kingdom 5th OpenFOAM UK and Eire User Group Meeting, Dublin, January 2017 Public Release: Native Overset Mesh in FOAM-Extend p. 1

2 Outline Objective Announce public release of native Overset Mesh capability Present validation and verification and parallel scaling results for the Overset mesh solver Topics Introduction Native implementation of the Overset Mesh support infoam-extend Overset Mesh: Validation and parallel scaling results Pre-announcement of NUMAP Summer 2017 and OFW 12 Summary Public Release: Native Overset Mesh in FOAM-Extend p. 2

3 Introduction Open Source Native Overset Mesh Capability infoam-extend We have been involved in several previous Overset projects... but none reached the maturity or provided flexibility required for a general-purpose CFD platform Project Objective: Complete rewrite without limitations on release Native implementation of Overset Mesh technology in FOAM-Extend without reference to external libraries Operator-based support for Finite Volume (FV) discretisation: implicit and explicit operators (fvc andfvm) Setup viaoverset patch type, with minimal impact on existing code components: full re-use of existing physical models Consistent support for strict mass conservation and second order accurate discretisation in space and time Further requirements Overset mesh functionality must operate on unstructured polyhedral mesh hierarchies Basic layout for massive parallelisation in domain decomposition mode Support for dynamic overset mesh simulations Public Release: Native Overset Mesh in FOAM-Extend p. 3

4 Overset Mesh Overset Mesh Capability infoam-extend: An Update Revisited native implementation of Overset Mesh: public release Rewrite of basic Overset functionality to improve performance and parallel scaling Overview of New Overset Capability Explicit and Implicit fringe treatments: chosen per-field, depending on application Rewrite of parallelised Overset assembly New research item: Implicit conservative Overset Additional fringe reduction algorithms Rewrite of parallel communications to improve scaling on HPC machines Public release of the Overset Solver compatible with foam-extend-4.0 Public Release: Native Overset Mesh in FOAM-Extend p. 4

5 Overset Mesh Explicit Fringe Treatment Interpolation is performed before the solution of each equation, setting values in acceptor cells (careful with the pressure equation!) It is also performed in certain places in the code where the interpolation is necessary (eg. 1/A update, etc.) Benefits Better parallel scaling than implicit fringe (linear solver) Region-wise mass conservative without special treatment It seems to be much more robust than implicit (especially for the pressure equation), which is probably linked to its conservative properties Drawbacks At least one non-orthogonal corrector is needed for the pressure equation in order to have regions behaving in a coupled way (otherwise, you can see that regions are not coupled and the mean value of pressure is different) Public Release: Native Overset Mesh in FOAM-Extend p. 5

6 Overset Mesh: New Development Implicit Fringe Treatment Interpolation is performed within the matrix-vector multiplication in the linear system solver Benefits No need for non-orthogonal correctors in order to update the Overset after the pressure equation (regions will always be strongly coupled) Drawbacks Some reduction in parallel scaling compared to implicit fringe Without special treatment, implicit Overset interpolation is non-conservative! Public Release: Native Overset Mesh in FOAM-Extend p. 6

7 Overset Mesh: New Development New Parallelised Overset Assembly Algorithm Using new fringe detection and minimisation variants Improved performance of fringe detection and prioritisation of domains Choice of Fringe Algorithms 1. manual fringe - user specifiedcellset for acceptors 2. facecells fringe - acceptors are face cells of a given patch 3. overlap fringe - automatically calculates overlap between regions, which can be minimised using different Donor Suitability Functions Choice of Donor Suitability Functions 1. nosuitability: full overlap (expensive interpolation!) 2. patchdistance: pushes the overlap at a given distance between specified patches, 3. cellvolumes: cells with similar volumes constitute donor/acceptor pairs 4. facearea: cells with similar minimum face areas constitute donor/acceptor pairs 5. cellboundingboxdiagonal - cells which have similar sizes of their bounding box diagonals constitute donor/acceptor pairs Public Release: Native Overset Mesh in FOAM-Extend p. 7

8 Overset Mesh: New Development New Overset Interpolation Algorithms Transfer of data from donors to acceptors is carried out using run-time selectable oversetinterpolation schemes Choice of Overset Interpolation Schemes 1. injection: simply takes the donor value and injects it into the acceptor (first order accurate, bounded) 2. averagevalue: average value of all donors (first order accurate, bounded) 3. inversedistance: weights for donor cells calculated using inverse distance from a given donor to acceptor (first/second order accurate and bounded) 4. leastsquaresgradient: value in acceptor calculated using least squares fit through its donors (second order accurate/unbounded) Public Release: Native Overset Mesh in FOAM-Extend p. 8

9 Overset Mesh: Implicit Conservation Implicit Conservative Overset If performed implicitly, Overset interpolation is not conservative because of arbitrary interpolation From the continuity equation, the net flux through fringe layer must be zero: this represents an additional constraint on the global pressure equation Handling Conservation at Overset Interfaces The pressure equation consists of three parts 1. Pressure gradient flux (implicit, on LHS) 2. Pressure gradient in non-orthogonal correction (explicit, on RHS) 3. Divergence of the intermediate flux (explicit, on RHS) Last two parts should be adjusted across the fringe (since they are explicit) before the insertion into the pressure equation How to handle the first, implicit part? Public Release: Native Overset Mesh in FOAM-Extend p. 9

10 Overset Mesh: Implicit Conservation Handling Conservation at Overset Interfaces Notes We need to add an additional constraint in the pressure equation, stating that the sum of the pressure gradient flux in fringe cells needs to be zero This constraint is enforced using the additive correction approach In each linear solver iteration, we calculate an additive correction to the non-conservative Overset interpolation based on fringe mass imbalance and add scatter this value across all acceptors This approach ensures mass conservation up to specified pressure equation tolerance If explicit Overset is used (no interpolation during linear system solver), the pressure gradient will take care of ensuring global mass conservation Public Release: Native Overset Mesh in FOAM-Extend p. 10

11 Overset Mesh: Example Handling Conservation at Overset Interfaces: Example Moving 2-D square with constant velocity through the stagnating fluid Public Release: Native Overset Mesh in FOAM-Extend p. 11

12 Overset Mesh Handling Conservation at Overset Interfaces: Example Pressure equation solved to the tolerance of 1e-7 in all cases Continuity error, fvc::div(phi) for implicit Overset: Order of magnitude 0.1 without conservative correction approach Order of magnitude 1e-5 with conservative correction approach Public Release: Native Overset Mesh in FOAM-Extend p. 12

13 Overset Mesh: Example Overset Mesh Validation: DTC Propeller Case Transient propeller simulation for a model scale propeller for DTC hull (Duisburg Test Case, a container ship) Case details Model scale propeller for DTC hull, Diameter, D = 0.15 m n = 890 RPM ( RPS) Advance coefficient (J = v/(n*d) = 0.75) Simulation set-up deltat = 2 degrees per time step endtime = 10 periods (10 full rotations) 1.85M cells (1.16M for propeller M for background) Max Courant number during simulation is approximately 15 Euler on ddt, convection/diffusion all standard second order k-omega SST turbulence model, Explicit Overset Public Release: Native Overset Mesh in FOAM-Extend p. 13

14 Overset Mesh: Example Overset Mesh Validation: DTC Propeller Case Simulation time: approximately 4 hours for a single rotation on a Desktop machine Errors compared to experiments: 4.3% under-prediction in thrust 6.6% over-prediction in torque Errors probably due to large time step and poor mesh 0.4 Thrust & torque coefficients; K T & 10K Q K T, CFD (overset,transient) K T, EXP K Q, CFD (overset, transient) K Q, EXP Time; t, s Public Release: Native Overset Mesh in FOAM-Extend p. 14

15 Overset Mesh: Parallel Scaling Parallel Efficiency of the Overset Mesh Implementation of Overset interpolation performed similar to GGI Interpolation performed in out-of-core multiplication with parallel comms Parallelised usingmapdistribute tool Parallel scaling test case Scaling test performed on 20M cells submarine mesh Approximately 40K donor/acceptor cells (0.4% of total cell count) Performed 20 iterations with explicit and implicit Overset fringe Parallel speed-up on 64 cores: 41 (implicit) and 46 (explicit) Parallel efficiency on 64 cores: 64% (implicit) and 74% (explicit) Explicit fringe Implicit fringe Linear scaling Parallel speed-up, S Parallel efficiency, E, % Explicit fringe Implicit fringe Number of cores, n Number of cores, n Public Release: Native Overset Mesh in FOAM-Extend p. 15

16 Overset Mesh: Parallel Scaling Overset Mesh: Parallel Scaling Test Public Release: Native Overset Mesh in FOAM-Extend p. 16

17 Overset Mesh: Future Work Future Work Improved overlap fringe algorithm: the current algorithm is under-performing both in serial and in parallel because we first find the full overlap and then minimise it. I think this is not a small task - funding is required Need to improve the parallelisation of the donor search Validation and verification! Public Release: Native Overset Mesh in FOAM-Extend p. 17

18 NUMAP Summer 2017 NUMAP Summer 2017: 10-day Summer School at University of Zagreb NUMAP Spring not planned for 2017 At Uni Zagreb, we may do a new Programming course in OpenFOAM based on new training materials Pre-announcement for NUMAP Summer 2017: 21/Aug 1/Sep/2016. Applications open from 1/Mar/2017 ESI/OpenCFD sponsoring the 2017 NUMAP? (Always ask in public!) Public Release: Native Overset Mesh in FOAM-Extend p. 18

19 Summary UK and Eire OpenFOAM User Day Who s next for the OpenFOAM UK and Eire User Group Meeting? 12th OpenFOAM Workshop July 2017 Exeter, UK German OpenFOAM Stammtisch United 2017, 20-21/Feb/2017, Kassel NUMAP-FOAM Summer School 2017 Zagreb Croatia, 21/Aug 1/Sep/2016 Public Release: Native Overset Mesh in FOAM-Extend p. 19