Simulations of Rotary Wing Systems Using Line-Based ...2018.oversetgridsymposium.org/assets/presentations/Wednesday/S… · NREL Phase VI wind turbine ... Available Data Campaigns,”

Simulations of Rotary Wing Systems Using Line-Based Unstructured/Structured Grids

Yong Su JungGraduate Research Assistant

October 2018 14th Symposium on Overset Composite Grids And Solution Technology 1

Presented at 14th Symposium on Overset Composite Grids and Solution TechnologyOctober 1-4 2018, College Park, MD

James BaederProfessor

Department of Aerospace Engineering, University of Maryland

Dylan JudeGraduate Research Assistant

Outline

• Introduction

– Motivation and Selected Previous Works

– Objectives

• Methodology

– Mercury framework

– Hamiltonian-Strand solver

• Results

– NREL Phase VI wind turbine

– Slowed rotor at high advance ratio

• Concluding Remarks

October 2018 214th Symposium on Overset Composite Grids And Solution Technology

Motivation

October 2018 3

• Line-based multiple mesh paradigms using overset grid system for realistic complex geometries

• Multiple computing architectures (CPU and GPU) in Python Integration Framework

• Improved airload prediction considering transition effect using laminar-turbulent transition model

• Coupling with CSD solver allows elastic deformation for slender and flexible blades

Helios Multi-Solver Paradigm Applied to a rotor and fuselage configuration.Wissink et al., AIAA 2018-0026

Background Cartesian Gridon GPU

Line-based Unstructured Grid on CPU

Horizontal-axis wind turbine (NREL Phase VI) simulation

14th Symposium on Overset Composite Grids And Solution Technology

strand

unstructured

Complex rotor hub geometryLee et al. (2018)

Selected Previous Works

October 2018 4

• Graphics processing unit (GPU) accelerated solvers

– Sebastian and Baeder (2013), Jude and Baeder (2016): GPU RANS

– Jude et al. (2018), Lee et al. (2018): Coupled CPU/GPU Framework

• Finding structures in unstructured grids

– Meakin (2007), Wissink (2009), Katz (2011), Lakshminarayan (2016-2018): Strand grids

– Sitaraman and Roget (2014): Hamiltonian paths in 2D

– Govindarajan et al. (2015), Jung et al. (2016-2018): Hamiltonian paths/ strand grids

• Wind turbine flow simulation using CFD (NREL Phase VI)

– Sorensen et al. (2002,2009,2014): RANS, RANS-transition, DDES-transition

– Gomez et al. (2009), Zahle et al. (2009): Blade-tower interaction

Simulation of complete rotorcraftJude et al. (2018)


• Slowed Mach-Scale Rotor simulation using CFD-CSD method

– Xing et al. (2018) : Pressure/Airload comparison with Experimental data

Objectives

October 2018 5

• Couple unstructured grid based CPU solver and structured GPU accelerated solver

- Exploit advantages of both multiple mesh paradigms and line-based methods

- Apply to full wind turbine simulation for investigating the effect of flow unsteadiness

- Apply laminar-turbulent transition model on wind turbine simulation to study the effect

on turbine performance

• Couple CFD framework with CSD solver for trimmed rotor with aeroelastic effect

− Adopt deforming mesh technique for elastic deformation of blade

− Compare pressure/airload prediction with experiment on slowed-rotor at high advance

ratio


Outline

• Introduction


– Objectives

• Methodology



• Results





CFD Framework : Mercury

October 2018 7

• CPU/GPU Structured/Unstructured grid solvers- Multiple computing architecture- All solvers use line-based implicit time marching

• Each code wrapped in python - light-weight and flexible for program languages- no expensive operations performed during

communication between solvers

• TIOGA* is used for overset connectivity

• Separate mesh deformation routine using algebraic method for Structured and Strand volume mesh

Can be interfaced with a structured solver

Framework Initialization(Solvers, Mesh Motions, TIOGA)

Main Iteration (t=t+dt)

Mesh Motion Update Mesh Deformation

TICOGA Overset Connectivity

Newton Sub-iteration (i=1,2,..,m)

Flow SolverOVERTURNS

TIOGA Data Exchange

YESNo

Finalization

Flow SolverGARFIELD

Flow SolverHAMSTR

*Brazell, et al., “An overset mesh approach for 3D mixed element high-order discretizations,” Journal of Computational Physics, Vol. 322, 2016, pp. 33 – 51.


HAMSTR flow solver (1)

< Subdivision of triangle into quadrilateral >

Strand grids

• Extend the formulation to three-dimensions

• Extrude surface mesh in wall-normal direction

• Each cell has three coordinate directions

< Strand template >

Unit directional vectors

Root

Tip

Node distribution along strand

October 2018 8

Hamiltonian paths

• Begin with original unstructured surface mesh

• Subdivided triangles and quadrilaterals

• The colored loops are Hamiltonian paths

< Strands and associated cell coordinates >

Multiple layers

Strand

Cell coordinates

Two-dimensional surface loop

Reference: Sitaraman and Roget (2014), Govindarajan et al. (2015), Jung et al. (2016-2018)


HAMSTR flow solver (2)

October 2018 9

Spatial reconstruction along “loops” –equivalent to traditional lines

face j

face j-1qL

qR

qR

qL

j,k

j-1

j+1

• Inviscid reconstruction can use standard line-methods (e.g. MUSCL, WENO, CRWENO)

• Viscous flux computed using finite difference / linear least-square

• Interface fluxes computed using Roe’s scheme with all-Mach correction

Reference: Sitaraman and Roget (2014), Govindarajan et al. (2015), Jung et al. (2016-2018)

• Large sparse matrix for typical unstructured grid replaced with many of smaller block tri-diagonal matrices

• Matrix inversion performed along each Hamiltonian path using Diagonally Dominant Line Gauss Seidel (DDLGS)

∆qj

I+x xx I+x x

x I+x xx I+x x

x I+x x x I+x x

x I+x

= ∆qj

Banded block tri-diagonal system along each loop(top: closed loop, bottom: open loop)

I+x x xx I+x x

x I+x xx I+x x

x I+x x x I+x x

x x I+x

∆qj = ∆qj

Ignored for first-order linearization

• Hamiltonian path is correspond to the mesh coordinate direction in fully structured mesh

• 3D unsteady compressible Navier–Stokes formulation (Finite volume)


• Preconditioned GMRES works using DDLGS

Laminar-Turbulent Transition Model

October 2018 10

𝜸𝜸 − 𝑹𝑹𝑹𝑹𝜽𝜽𝜽𝜽 − 𝑺𝑺𝑺𝑺 Transition Model for Spalart-Allmaras Turbulence Model (Medida, 2014)

• Two transport equations

- New correlation for 𝑹𝑹𝑹𝑹𝜽𝜽𝜽𝜽

- Constant freestream turbulence intensity

- Modified production and destruction terms in intermittency equation

- Omission of the separation-induced transition modification

• Only production term need to be scaled in Spalart-Allmaras Turbulent Model

𝐷𝐷 𝑅𝑅𝑅𝑅𝜃𝜃𝜃𝜃𝐷𝐷𝐷𝐷 = 𝑃𝑃𝜃𝜃𝜃𝜃 − 𝐷𝐷𝜃𝜃𝜃𝜃 + 𝑀𝑀𝜃𝜃𝜃𝜃

𝐷𝐷𝛾𝛾𝐷𝐷𝐷𝐷 = 𝑃𝑃𝛾𝛾 − 𝐷𝐷𝛾𝛾 + 𝑀𝑀𝛾𝛾

Experimental Correlations Onset criterion

𝑷𝑷𝜸𝜸 = 𝝆𝝆 � 𝑭𝑭𝒐𝒐𝒐𝒐𝒐𝒐𝑹𝑹𝜽𝜽� 𝑮𝑮𝒐𝒐𝒐𝒐𝒐𝒐𝑹𝑹𝜽𝜽 � 𝒎𝒎𝒎𝒎𝒎𝒎𝛀𝛀

𝟏𝟏𝟏𝟏.𝟏𝟏 ,𝟏𝟏.𝟒𝟒

New non-local in the wall normal direction

𝑫𝑫𝜸𝜸 = 𝜸𝜸𝝆𝝆𝛀𝛀(𝟏𝟏.𝟏𝟏 − 𝑮𝑮𝒐𝒐𝒐𝒐𝒐𝒐𝑹𝑹𝜽𝜽)

𝐷𝐷�𝜈𝜈𝐷𝐷𝐷𝐷 = 𝛾𝛾𝑃𝑃𝜈𝜈 − 𝐷𝐷𝜈𝜈 +

1𝜎𝜎 𝛻𝛻 � 𝜈𝜈 + �𝜈𝜈 𝛻𝛻 �𝜈𝜈 + 𝑐𝑐𝑏𝑏𝑏 𝛻𝛻 �𝜈𝜈 𝑏

Reference: Medida, S., Correlation-based Transition Modeling for External Aerodynamic Flows (2014)


w/o 𝑮𝑮𝒐𝒐𝒐𝒐𝒐𝒐𝑹𝑹𝜽𝜽

Improved intermittency recovery in turbulent boundary layer

w/ 𝑮𝑮𝒐𝒐𝒐𝒐𝒐𝒐𝑹𝑹𝜽𝜽

Outline

• Introduction

• Methodology

• Results


• Steady axial flows for rotor-alone configuration

• Steady axial flows for full configuration (upwind)

• Steady axial flows for full configuration (downwind)

• Effect of wind shear flows (upwind)




NREL Phase 6 Reference Wind Turbine

October 2018 12

Properties

configuration, No. of blade Upwind/downwind, 2 blades

Rotor speed, Tip pitch 72 RPM, 3°

rotor diameter and root cut 10.058 m, 0.508m

Tower diameter, clearance 0.4 m, 1.401 m

hub height 12.192 m

Test wind speed 7 m/s, 10 m/s, 20 m/s

Shaft tilt, precon angle 0°, 0° (upwind) 3°, 3.4° (downwind)

Ref. Hand et al., ”Unsteady Aerodynamics Experiment Phase VI: Wind Tunnel Test Configurations and Available Data Campaigns,” NREL/TP-500-29955, 2001.

NREL Phase VI turbine blade


Overset grid system for NREL Phase 6

October 2018 13

Near body (CPU) Cylindrical nest (GPU) Background (GPU)

5.1 millions 5.8 millions 4.1 millionsOverset mesh for blade-tower interaction

Mixed element blade surface mesh

Unstructured quads.

Structured quads.

Initial wall spacing

Number of strands

9,000 240x90 1e-5 chord 52

Inviscid wall B.C

Far-field B.C

Viscous wall B.C

Off-body domain mesh and boundary condition

*Dimensions are based on NASA Ames wind tunnel section

Nested mesh


Outline

• Introduction

• Methodology

• Results


• Steady axial flows for rotor-alone configuration

• Steady axial flows for full configuration (upwind)

• Steady axial flows for full configuration (downwind)

• Effect of wind shear flows (upwind)




Rotor alone analysis in Steady axial flow

October 2018 15

V∞(m/s)

Torque[N-m]

Experiment[N-m]

7 766 (696) 805

10 1,196 (1,172) 1,340

20 1,134 (1,084) 1,110

𝑽𝑽∞=7 m/s

Chord-wise Pressure distribution

𝑽𝑽∞= 10 m/s

𝑽𝑽∞= 20 m/s• The use of transition model better estimates the torque• At high wind speeds (10, 20 m/s), separated flow present on the suction side• Transition onset along with laminar separation was predicted at mid-chord along the span at 7m/s

Comparison of torque prediction for NREL Phase VI Turbine


Contour of skin friction with streamline on suction side of blade at 7 m/s

Transition Fully turbulence

Effect of tower influence on rotor inflow

October 2018 16

Upwind configuration(Upstream view)

Downwind configuration(Downstream view)

𝝍𝝍 =

𝑤𝑤/𝑉𝑉∞

𝝍𝝍 =

𝑤𝑤/𝑉𝑉∞

Inflow Inflow

Nondimensional inflow distribution on the upstream plane Comparison of sectional pressure distribution at 𝝍𝝍 = 𝟏𝟏𝟏𝟏𝟏𝟏𝟏

• Inflow velocity distribution on sectional plane of 3% rotor radius upstream from rotor disk plane• Upwind configuration reduced inflow at the tower due to tower blockage effect• Downwind configuration reduced inflow at the tower due to tower shedding• Tower interference was much severe at downwind configuration, results in loss of suction peak

Side view

Tower shedding

Tower blockage

upwind downwind

𝟗𝟗𝟏𝟏𝟏

𝟏𝟏𝟏𝟏𝟏𝟏𝟏

𝟐𝟐𝟐𝟐𝟏𝟏𝟏

𝟏𝟏𝟏

𝟗𝟗𝟏𝟏𝟏

𝟏𝟏𝟏𝟏𝟏𝟏𝟏

𝟐𝟐𝟐𝟐𝟏𝟏𝟏

𝟏𝟏𝟏


Upwind full configuration results

October 2018 17

Azimuthal variation single blade torque and thrust

Azimuthal variation sectional normal and tangential force

0.63R

0.47R0.3R

0.8R

0.63R

0.47R

0.3R

0.8R

Inflow distribution along blade path• Slower axial velocity due to tower blockage reduce blade aerodynamic force• During half revolution, performance was affected by tower• Inner section experience more interference due to lower tangential velocity and shorter blade path

10 %

Blade path

affected region

decelerated


6 %

60 %

35 % 25 %15 %

Downwind full configuration results

October 2018 18

Snapshots of axial velocity contour at sectional plane located 0.63R

• Abrupt reduction of sectional blade airload due to tower shedding at 180° azimuth angle over entire span• Good agreement with experimental data in sectional normal force history during a rotor revolution

Unsteady blade aerodynamic loads

𝑤𝑤 [m/s]

98 %72 % 55 %

28 %

Azimuthal variation single blade thrust and torque

33 % 54 %• Over 30 % reduction in single blade thrust at 𝜓𝜓 = 180°

• Over 50 % reduction in single blade torque at 𝜓𝜓 =180°

0.63RTower

shedding Downwash effect


Effect of wind shear on upwind configuration

October 2018 19

[Ref.] International Standard IEC 61400-1. Wind turbines, part 1:DesignRequirements.

𝑉𝑉 𝑧𝑧 = 𝑉𝑉ℎ𝑢𝑢𝑏𝑏(𝑧𝑧

𝑧𝑧ℎ𝑢𝑢𝑏𝑏)𝛼𝛼

𝑉𝑉ℎ𝑢𝑢𝑏𝑏 = 7 𝑚𝑚/𝑠𝑠𝑧𝑧ℎ𝑢𝑢𝑏𝑏 = 12 𝑚𝑚, 𝛼𝛼 = 0.2

Normal wind profile (NWP)

• Asymmetric force on rotor disk due to presence of wind shear, towerblockage effect, and swirl effect

• Opposite effects between two blades canceled each other

Dimensional sectional force distribution along radial direction

𝑭𝑭𝜽𝜽𝒐𝒐𝒕𝒕𝒕𝒕𝒕𝒕𝑹𝑹(𝑵𝑵/𝒎𝒎)

Wind shear Uniform flow

Tower blockage

Tangential force distribution on rotor disk plane

• Large airload variations of sectional forces near blade tip• Tower effect was minor compared to wind shear effect

Azimuthal variation of torque (Left: single blade, right: rotor)

40 % 6 %


𝝍𝝍 = 𝟏𝟏𝟏 𝝍𝝍 = 𝟏𝟏𝟏

Outline

• Introduction

• Methodology

• Results



• Coupled CFD-CSD method from trimmed rotor

• Pressure/Airload correlation with experimental data



Slowed Rotor at High Advance Ratios

October 2018 21

Test Rotor Properties

Rotor type 4-blade articulated

Radius (in) 33.5 (AR=10.63)

Chord (in) 3.15

Solidity 0.120

Hinge offset 6.4%

Root cutout 16.4%

Lock number 4.96

Nominal RPM 2300

Airfoil NACA0012

The blades are untwisted and untampered*

Test hover-stand* (Glenn L. Martin Wind Tunnel)

• Slowing down the rotor RPM is a viable method to alleviate the compressibility effect in forward flight• A series of wind tunnel tests were conducted in Glenn L. Martin Wind Tunnel (UMD)*• Pressure sensors were installed at 30% radial section of blade• Correlation study using CFD-CSD coupled analysis were performed for selected cases


* Xing et al., CFD Prediction/Airload Correlation with Experimental Data on Slowed Mach-Scaled Rotor at High Advance Ratios, 74th AHS forum, 2018

RPM Collective sweep conducted at

700 𝜇𝜇 = 0.3~0.8 at 0 deg shaft tilt,𝜇𝜇 = 0.8 at ±2 deg shaft tilt

900 𝜇𝜇 = 0.3~0.7 at 0 deg shaft tilt,𝜇𝜇 = 0.7 at ±2 deg shaft tilt

Wind Tunnel test matrix*

Overset grid system for Slowed Rotor

October 2018 22

Rotor blade with hub/shaft model after hole cutting(overset: strand volume mesh and background)

Sharp blade tip shape (200x100 structured quads., 47 strand layers)

• Strand volume domains were embedded in Cartesian background grid of 10 % chord length• Initial wall normal spacing of 5e-5 chord (𝑦𝑦+=0.7)• Each blade tip was modeled without tip cap which is similar with experimental blade shape• Far-field boundary condition was imposed for background

Total rotor blades Hub/Shaft Background

4.0 millions 0.51 millions 4.4 millions


Coupled CFD-CSD method

October 2018 23

Comprehensive Analysis : PrasadUM

• Structural model• Euler-Bernoulli beam theory (flap, lag and torsion)• Finite element method with modal reduction (8 modes)

• Aerodynamic model• Blade element theory with 2D airfoil tables• Leishman-Beddoes dynamic stall model• Bagai-Leishman freewake model• Weissinger-L near wake representation

• Trim method• Prescribe collective (experiment)• Adjust cyclics to match zero hub moment

CFD flow solver: HAMSTR• RANS simulation with SA turbulence model• Implicit time marching using BDF2• Time step size of 1 degree using 15 sub-iterations• WENO5 reconstruction for inviscid flux

• In-house comprehensive code PrasadUM was used for CSD part.

• Coupled CFD and CSD using delta-coupling process for airload correction

• Simulation focused on a representative high 𝜇𝜇− 700 RPM, 𝜇𝜇=0.8, 𝜃𝜃0 = 3𝟏and 11𝟏

Flow chart for delta coupling method*


* Xing et al., CFD Prediction/Airload Correlation with Experimental Data on Slowed Mach-Scaled Rotor at High Advance Ratios, 74th AHS forum, 2018

CFD-CSD Study Convergence

October 2018 24

Case Control Test PrasadUM CFD/CSD

𝜽𝜽𝟏𝟏 = 𝟑𝟑𝟏𝜽𝜽𝟏𝟏𝟏𝟏 2.79° -0.20° 0.36°

𝜽𝜽𝟏𝟏𝒐𝒐 -5.27° -3.54° -3.58°

𝜽𝜽𝟏𝟏 = 𝟏𝟏𝟏𝟏𝟏𝜽𝜽𝟏𝟏𝟏𝟏 5.51° 1.17° 2.09°

𝜽𝜽𝟏𝟏𝒐𝒐 -13.94° -12.80° -12.69°

Convergence history of control cyclics (𝝁𝝁=0.8, 𝜽𝜽𝟏𝟏=11°)

• Control cyclics and airloads are used for checking coupled study convergence• CFD-CSD coupled study converges within 10 steps• Both comprehensive and CFD-CSD results show discrepancy with experimental data• Minor changes in longitudinal cyclic between comprehensive and CFD-CSD results


Convergence history of control cyclics (𝝁𝝁=0.8, 𝜽𝜽𝟏𝟏=3°)

Summary of control angle results

Sectional Airload at 30% R, (𝝁𝝁=0.8, 𝜽𝜽𝟏𝟏=11°)

October 2018 25

Normal force (30% R) Pitching moment (30% R)

Interference with hub wake


Less accurate in reverse flow region

Lack of details(vortex interaction)

Less accurate in reverse flow (nose-up peak)

• Comprehensive analysis results show good agreement in general trend, but lack of details in vortex interaction and reverse flow region

• Prescribed CFD results (rigid blade) using experimental control setting is no better than comprehensive analysis results

• CFD-CSD with hub model improves agreement with test data significantly• Effect of hub wakes are well captured in both normal force and pitching moment


Improved in reverse flow region

Pressure data at 30% R, (𝝁𝝁=0.8, 𝜽𝜽𝟏𝟏=11°)

October 2018 26

• 7 pairs of pressure sensors are located on top and bottom surface of blade (8% - 80% chordwise) • CFD-CSD pressure results exhibit good correlation with test data• Pressure fluctuation on advancing side and pressure loss on retreating side due to dynamic stall• Dynamic stall pressure loss is overpredicted near the trailing edge

Upper surface (7 sensors) Lower surface (7 sensors)

CFD/CSD with hub Test data


Dynamic stall peak was captured

Pressure fluctuation on advancing side

Sectional Airload and Pressure data at 30% R, (𝝁𝝁=0.8, 𝜽𝜽𝟏𝟏=3°)

October 2018 27

Normal force (30% R) Pitching moment (30% R)

• CFD-CSD result captures general trends of normal force• Effect of hub wake reduces airloads significantly and well

captured • Compared with test data, CFD-CSD results show stronger

dynamic stall phenomena• Pressure predictions of CFD-CSD shows overall good

agreement with test data• Some discrepancies at upper surface leading edge on

advancing sideUpper surface Lower surface

CFD/CSD with hubTest data




Stronger dynamic stall

Outline

• Introduction


– Objectives

• Methodology



• Results





Concluding Remarks

• Explored CPU/GPU heterogeneous CFD framework for wind turbine simulation– Observed improvements in turbine performance prediction using 𝛾𝛾 − 𝑅𝑅𝑅𝑅𝜃𝜃 − 𝑆𝑆𝑆𝑆 transition

model, especially in attached flow condition– Investigated blade-tower interference for both upwind and downwind configurations at wind

speed of 7 m/s– Studied effect of free-stream wind shear on the rotor performance

• Applied coupled CFD/CSD method for Slowed Rotor at High Advance Ratio– Deformed Hamiltonian-Strand mesh using algebraic method– Observed improved correlation with experimental data using coupled method in terms of

pressure and sectional airloads

• Future Works– Near-wall strand generation needs to be more robust for complex geometries– For real size wind turbine model (e.g. 5MW), coupled CFD-CSD study can be conducted to

consider deforming structures in rotor performance prediction.


Acknowledgements

October 2018 30

• Air Vehicle element of the HPCMP CREATE program

• Rajneesh Singh (ARL) and Roger Strawn (AMES) for continued support

• Xing Wang and Dr. Chopra (University of Maryland) for providing experimental

data for rotor simulation

• UMDs Deepthought II high-performance computing facility

• Johns Hopkins Maryland Advanced Research Computing Center


Simulations of Rotary Wing Systems Using Line-Based Unstructured/Structured Grids

October 2018 14th Symposium on Overset Composite Grids And Solution Technology 31

Presented at 14th Symposium on Overset Composite Grids and Solution TechnologyOctober 1-4 2018, College Park, MD

Department of Aerospace Engineering, University of Maryland

Yong Su JungGraduate Research Assistant

James BaederProfessor

Dylan JudeGraduate Research Assistant

Documents

Simulations of Rotary Wing Systems Using Line-Based ...2018.oversetgridsymposium.org/assets/presentations/Wednesday/S… · NREL Phase VI wind turbine ... Available Data Campaigns,”