2008 Int ANSYS Conf Implementation Hybrid Navier Stokes

Embed Size (px)

Citation preview

  • 8/6/2019 2008 Int ANSYS Conf Implementation Hybrid Navier Stokes

    1/27

    2008 GE. All rights reserved. 1

    2008 International

    ANSYS Conference

    Implementation of a Hybrid Navier-Stokes / Vortex Panel Method for WindTurbine Aerodynamic Analyses in CFX

    Mark E. Braaten1, Kevin Standish2, Slawomir Kolasa3,Emad Gharaibah4

    1 GE Global Research, Schenectady, NY

    2 GE Energy, Greenville, SC

    3 GE Polska, Warsaw, Poland4 GE Global Research, Munich, Germany

  • 8/6/2019 2008 Int ANSYS Conf Implementation Hybrid Navier Stokes

    2/27

    2008 GE All rights reserved. 2

    Outline of Talk

    Overview of hybrid CFD method

    Implementation in CFX Validation

    Sensitivity Studies

    Concluding Remarks

  • 8/6/2019 2008 Int ANSYS Conf Implementation Hybrid Navier Stokes

    3/27

    2008 GE All rights reserved. 3

    Hybrid CFD for Wind Turbine Aero

    Conventional CFD methods based on Navier-Stokes haveserious shortcomings for wind turbine analyses Very large domain required to enforce far-field boundary conditions

    sufficiently far from blade Inlet and outlet conditions imposed many blade radii upstream

    and downstream

    Periodicity requires large sector (120for 3 blades) to be

    modeled Results in very large meshes (> 10M nodes), long run times, difficult

    post-processing

    Prevents use of fine enough mesh near blade needed to capture

    turbulence transition effects Unsteady CFD simply not practical on such large meshes

    Wake behind blade dissipates too quickly due to numericaldissipation

    Prior studies have shown need to resolve wake as far as 10-20blade radii downstream for accurate power predictions

  • 8/6/2019 2008 Int ANSYS Conf Implementation Hybrid Navier Stokes

    4/27

    2008 GE All rights reserved. 4

    Full Domain CFD

    Inlet

    Outlet

    50 m

  • 8/6/2019 2008 Int ANSYS Conf Implementation Hybrid Navier Stokes

    5/27

    2008 GE All rights reserved. 5

    Hybrid Navier-Stokes/Vortex Panel method proposed bySchmitz, Chattot (UC Davis) looks very promising

    Small Navier-Stokes region (~1 - 5 M nodes) around the bladecomputes near field

    Vortex Panel method computes far field using Biot-Savart law

    Effect of blade on far field represented by lifting line,

    helicoidal paths from prescribed wake

    Hybrid CFD (contd)

    Small Navier-Stokes domain

    Prescribedwake shape

  • 8/6/2019 2008 Int ANSYS Conf Implementation Hybrid Navier Stokes

    6/27 2008 GE All rights reserved. 6

    Hybrid CFD (contd)

    Coupling between near and far fields

    Navier-Stokes code computes circulation on polylines aboutspanwise blade sections needed by Vortex panel solver

    Vortex panel method computes induced velocities on boundariesof NS region computed from Biot-Savart law

    Effect of blade on far field represented by lifting line, helicoidalpaths from prescribed wake

    These provide coupling between Navier Stokes and Vortex Panelsolvers

  • 8/6/2019 2008 Int ANSYS Conf Implementation Hybrid Navier Stokes

    7/27 2008 GE All rights reserved. 7

    Goals for Hybrid CFD for WindTurbine Aero Design

    Goal is to develop design system using hybridmethodology that can be routinely used by GE engineersto design wind turbine blades for optimum aeroperformance

    Scripting of meshing, pre- and post-processing is essential

    Hybrid CFD analysis in CFX must be no more difficult to set upand run than conventional analysis

    Best practices established and embedded in process

  • 8/6/2019 2008 Int ANSYS Conf Implementation Hybrid Navier Stokes

    8/27 2008 GE All rights reserved. 8

    The hybrid method requires:

    The computation of the circulation on closed loops that

    include all of the sources of vorticity The calculation of the induced velocities and their imposition

    on the boundaries

    These calculations are now done as part of a singleCFX solver run, instead of requiring a separateprogram to be run between successive runs of CFX

    The vortex panel solver is implemented in CFX User

    Fortran (CFX Version 11.0 and up)

    Implementation of Hybrid CFD inCFX

  • 8/6/2019 2008 Int ANSYS Conf Implementation Hybrid Navier Stokes

    9/27 2008 GE All rights reserved. 9

    User Fortran Routines

    CFX provides two types of User Fortran routines:

    Junction Box routines, which are User Fortran routines that arecalled and executed at particular times during a CFX run.

    CEL routines These evaluate a function in a similar fashion to CFX expression

    language

    We basically need routines to:

    Read the user defined loops upon which to compute the circulation, andsave these in the CFX MMS (memory management system) for lateruse by the solver

    Compute the helicoidal wake paths

    Compute the circulation on the polylines at the start of each CFXiteration

    Impose the induced velocities on the outer boundary as a prescribedvelocity opening boundary condition

  • 8/6/2019 2008 Int ANSYS Conf Implementation Hybrid Navier Stokes

    10/27 2008 GE All rights reserved. 10

    Basic Hybrid CFD Flow Chart

    Iteration Loop

    Note

    User Fortran routines

    shown in yellow

    Initial Setup

    Read user input

    Compute circulation

    Compute induced velocities

    Solve equations

    Output results file

    data files Generate polylines

    Compute helix

  • 8/6/2019 2008 Int ANSYS Conf Implementation Hybrid Navier Stokes

    11/27 2008 GE All rights reserved. 11

    Computing Circulation in CFX

    The circulation is computed on closed loops (polylines) thatinclude all of the sources of vorticity The circulation calculation is made for a number of spanwise

    positions along the blade This calculation is updated every iteration using the currentvalues of the velocity field

    Polyline points are mapped to the closest mesh points

    Velocity vector and gradient returned at mapped points User Fortran provides capability to recover gradient operator

    Use Taylor expansion to interpolate solution from mapped point (mp)to polyline point (pp)

    Vpp = Vmp + V * dr Circulation calculation requires parallel implementation

    Simplest parallel implementation is sufficient, as the line integrals are1D calculations:

    Routines use CFX Flow Parallel message passing calls for parallelcommunication (as does rest of CFX solver)

  • 8/6/2019 2008 Int ANSYS Conf Implementation Hybrid Navier Stokes

    12/27 2008 GE All rights reserved. 12

    Helicoidal paths are computed based onsimple actuator disk theory

    Paths depend on estimated powercoefficient Paths are periodically updated during

    CFX calculation based on latestcomputed power coefficient

    Helicoidal paths computed redundantly oneach processor using a Junction Boxroutine

    This allows each processor to

    compute induced velocities in perfectlyparallel fashion

    Computation of Helicoidal Paths

    Winddirection

    Navier-Stokesdomain

  • 8/6/2019 2008 Int ANSYS Conf Implementation Hybrid Navier Stokes

    13/27 2008 GE All rights reserved. 13

    Induced Velocities

    The induced velocities are computed on the outer boundary ofthe domain, using the Biot-Savart law

    This involves the line integration along a helicoidal path, starting out

    from each spanwise location on the trailing edge Outer boundaries of Navier-Stokes domain are treated as

    openings

    Induced velocity on boundary is defined as an expression, which is

    provided by a user CEL function Influence coefficients computed at beginning of run, and stored in

    MMS

    Subsequent updates of induced velocities use stored coefficients

    Influence coefficients updated periodically to reflect updatedhelicoidal paths

  • 8/6/2019 2008 Int ANSYS Conf Implementation Hybrid Navier Stokes

    14/27 2008 GE All rights reserved. 14

    Induced Velocities (Biot-Savart Law)

    Influence Coefficients

    All equations from

    Schmitz thesis

    Equations for Induced Velocities

  • 8/6/2019 2008 Int ANSYS Conf Implementation Hybrid Navier Stokes

    15/27 2008 GE All rights reserved. 15

    Computation of Induced Velocities

    Total storage requirement (6 influence coefficients) X (nbf

    boundary faces) X (jx polylines)

    For 100,000 boundary faces, 40polylines 100 MB storage

    Typically represents about 20%additional storage for solver

    Each processor needs only tocompute influence coefficients,induced velocities just for itsboundary faces

    Naturally parallel !

    Storage is distributed acrossprocessors

  • 8/6/2019 2008 Int ANSYS Conf Implementation Hybrid Navier Stokes

    16/27 2008 GE All rights reserved. 16

    Specification of BCs in CFX

  • 8/6/2019 2008 Int ANSYS Conf Implementation Hybrid Navier Stokes

    17/27 2008 GE All rights reserved. 17

    Mesh Generation

    The blade geometry is created inUnigraphics, and imported into ICEM-CFD for meshing

    Restrictions on Navier-Stokes domain: Needs to contain polylines for computation of

    the circulation

    Exit plane should be orthogonal to wake path

    A grid template with a C-H topology hasbeen developed to simplify the meshgeneration process

    The blocking file can be imported to anyICEM project with the same parts andadapted to a new blade geometry

  • 8/6/2019 2008 Int ANSYS Conf Implementation Hybrid Navier Stokes

    18/27 2008 GE All rights reserved. 18

    Creating the polylines

    The polylines over which the circulation is computed aregenerated using two utility programs:

    BladeContours

    A CFX-POST Power Syntax script file that extracts thecontours of the blade cross sections at the spanwise locations

    where the polylines are desired These blade contours are then input to

    PolyGen

    A program that created closed polylines that are offset fromthe blade contours

    The polylines, and the location of the leading and trailing edgelocations of the spanwise blade sections are written into files

    that are input files for the CFX run

  • 8/6/2019 2008 Int ANSYS Conf Implementation Hybrid Navier Stokes

    19/27 2008 GE All rights reserved. 19

    Screen shot from BladeContours script,showing spanwise blade contours

    Polylines for NREL wind turbineblade

    Generation of the polylines

  • 8/6/2019 2008 Int ANSYS Conf Implementation Hybrid Navier Stokes

    20/27 2008 GE All rights reserved. 20

    Post-processing

    Post-processing of the results is also performed using aCFX-Post Power Syntax script

    Normal and tangential directions input for each spanwise bladesection

    At each section, script calculates:

    Normal and tangential forces Pressure coefficients

    Torque force

    Same post-processing script used for full-domain andhybrid CFD computations to facilitate comparisons

  • 8/6/2019 2008 Int ANSYS Conf Implementation Hybrid Navier Stokes

    21/27 2008 GE All rights reserved. 21

    Pressure distribution at 63% span 7 m/s wind speed

    Hybrid CFD

    NREL Validation Case

    National Renewable Energy Laboratory (NREL) performed detailedwind tunnel experiments on 10m diameter wind turbine in NASA Ameswind tunnel

    This was the test case used by Sven Schmitz (UC Davis) to help

    develop the hybrid methodology Calculated torque values, pressure distributions match experiment very

    well for low wind speeds 7 & 9 [m/s] cases are attached flow

  • 8/6/2019 2008 Int ANSYS Conf Implementation Hybrid Navier Stokes

    22/27 2008 GE All rights reserved. 22

    Navier-Stokes domain

    Computed Mach Numbers (TSR=8)

    PowerCoefficient

    Tip Speed Ratio

    Run Time Comparison(equal # iterations)

    CPUtim

    e(min)

    0

    50

    100150

    200

    250

    300

    350

    400

    Full CFD 366Hybrid CFD 46.9

    Solver time [min]

    Pressure Coefficient (PC) Comparison

    GE46 Validation Case

    This was the first GE blade runusing the hybrid CFD method

    Comparisons made to full-domain

    CFD, other analytical methods Power predictions match well for

    attached flow cases

  • 8/6/2019 2008 Int ANSYS Conf Implementation Hybrid Navier Stokes

    23/27 2008 GE All rights reserved. 23

    Sensitivity Study

    Sensitivity studies are being performed to investigate theeffects of a number of parameters, and to establish bestpractices

    Parameters under study:

    Size of Navier-Stokes domain

    Grid size

    Number of polylines

    Location, orientation of polylines

    Results appear to be relatively insensitive

  • 8/6/2019 2008 Int ANSYS Conf Implementation Hybrid Navier Stokes

    24/27

    2008 GE All rights reserved. 24

    (a) Full domain CFD(b) Hybrid (C=1)(c) Hybrid (C=)

    (d) Hybrid (C=)

    (a) (b)

    (c) (d)

    Vary size of Navier-Stokes domain, from

    one chord (C=1)down to chord Vorticity field

    starts to lookunphysical ifdomain is madetoo small

    Vorticity at 20% Span

    Example: Effect of Domain Size

  • 8/6/2019 2008 Int ANSYS Conf Implementation Hybrid Navier Stokes

    25/27

    2008 GE All rights reserved. 25

    Future Work

    Hybrid methodology allows fine enoughgrid in Navier-Stokes domain to allow

    transition to be modeled Initial calculations with Langtry-Menter

    transition model look very promising

    Reduction in run time makes transientsimulations possible

    Unsteady hybrid analysis capabilitycurrently under development (w/ UC Davis)

  • 8/6/2019 2008 Int ANSYS Conf Implementation Hybrid Navier Stokes

    26/27

    2008 GE All rights reserved. 26

    Concluding Remarks

    Hybrid CFD approach looks very promising for windturbine aerodynamic analyses

    Grid size and run times much less than full domain CFD

    Results similar to full-domain CFD Results reasonably insensitive to size of Navier-Stokes domain

    and location of polylines

    User Fortran in CFX solver and Power Scripting in CFX-POST allows the development of an integrated hybridCFD design system

  • 8/6/2019 2008 Int ANSYS Conf Implementation Hybrid Navier Stokes

    27/27

    Hybrid CFD Represents a unique differentiatingtechnology that will be a prime enabler for nextgeneration quiet, efficient wind turbine blades

    Hi-fidelity 3D aerodynamic design

    Low noise design with CFD-based or direct CAAnoise prediction

    Aero-elastic fully-coupled design methods

    Impact