1 1 Best Practice Guidelines for Computational Turbulent Dispersed Multiphase Flows René V.A. Oliemans ERCOFTAC Seminar, Innventia, Stockholm, June 7-8, 2011 Department of Multi-Scale Physics Acknowledgements • Prof. Jos Derksen Un. of Alberta • Dr. Muhamed Hadziabdic Un. Serajevo • Prof. Hans Kuipers/dr. Niels Deen Un. Eindhoven • Prof. Rob Mudde Delft Un. of Technology • Prof. Dirk Roekaerts Delft Un. of Technology • Prof. Alfredo Soldati/dr. Marchioli Un. Udine • Prof. Martin Sommerfeld Martin-Luther Un. • Prof. Berend van Wachem Imperial College
42
Embed
Best Practice Guidelines for Computational Turbulent ... · Gas-solid Gas-Droplet Pneumatic conveying, cyclones, fluidized beds Spray-drying, -cooling, -painting,- scrubbers Liquid-solid
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
1
Vermelding onderdeel organisatie
1
Best Practice Guidelines for Computational Turbulent Dispersed Multiphase Flows
René V.A. Oliemans
ERCOFTAC Seminar, Innventia, Stockholm, June 7-8, 2011
Department of Multi-Scale Physics
Vermelding onderdeel organisatie
Acknowledgements
• Prof. Jos Derksen Un. of Alberta • Dr. Muhamed Hadziabdic Un. Serajevo • Prof. Hans Kuipers/dr. Niels Deen Un. Eindhoven • Prof. Rob Mudde Delft Un. of Technology • Prof. Dirk Roekaerts Delft Un. of Technology • Prof. Alfredo Soldati/dr. Marchioli Un. Udine • Prof. Martin Sommerfeld Martin-Luther Un. • Prof. Berend van Wachem Imperial College
2
3
Contents
• Industrial flow challenges • Requirements for CFD of dispersed multiphase flow • Fundamentals • Some Examples (bubbles, drops and particles flows) • Sources of Errors • Checklist of Best Practice Advice
Liquid-solid Hydraulic conveying, particle dispersion in stirred vessels, liquid-solid separation
Liquid-droplet Stirred tank reactor, liquid-liquid extraction
Liquid-bubble Bubble columns, aeration of sewage water, flotation, gas lift
3
CFD simulation of bubble column
Lehr et al. (2002)
5
Grid configuration for bubble column
36288 cells, no wall refinement 247050 cells+wall refinement
1 7
CFD simulations by Laborde-Boutet et al. (2009)
6
4
RANS (standard k-ε) simulations
Laborde-Boutet et al. (2009)
7
Turbulence modelling and bubble population balance
Av. Eddy Size: 22 mm 19 mm 10 mm
Measured average bubble size: 4.5 mm
Computed liquid eddy length scales:
8
5
3D simulation with bubble population balance and ten-fold bubble breakage
Chen et al. 2005b
9
Liquid turbulent shear stress
Chen et al. 2005b
RANS +
Exp. by Ong (2003)
10
6
11
Fluidized Beds
12
Fluidized Riser Flow
Particle segregation dependent on size and turbulence
Schuurmans (1980)
7
13
Riser reactor performance
Conversion variation along the riser height
Schuurmans (1980)
14
Contents
• Industrial flow challenges • Requirements for CFD of dispersed multiphase flow • Fundamentals • Some Examples (bubbles, drops and particles flows) • Sources of Errors • Checklist of Best Practice Advice
8
15
Requirements for CFD of Industrial Turbulent Dispersed Multiphase Flows
Aim:Mixing to optimize heat/mass transfer
Fluid Flow aspects: • Particles with various sizes in wall bounded flow fields • Spatial and temporal distribution of the phases • Particle coalescence and break-up • Fluid/particle interactions • Model validation • Upscaling from laboratory
16
CFD Requirements for industrial dispersed multiphase flows
• Tool to examine complex, large scale systems • Produce accurate wall-bounded turbulent flow fields • Spatial and time-dependent distribution of phases • Closure relations for averaged equations • Fast, reliable and accurate computations • Guide to handle enormous amount of simulation data
9
17
Particle volume fractions and sizes
Bubble column αp = 0.10-0.40 d= 1-30 mm
Stirred tank reactor αp = 0.02-0.40 d= 5 – 500 µm
Fluidized Beds αp = 0.40 d = 50 – 1000 µm
Cyclone αp = 0.001 d = 1 - 50 µm
18
Particle/Fluid interactions
10
19
Turbulence modulation
(Gore and Crowe, 1989)
20
Contents
• Industrial flow challenges • Requirements for CFD of dispersed multiphase flow • Fundamentals • Some Examples (bubbles, drops and particles flows) • Sources of Errors • Checklist of Best Practice Advice
11
21
Fundamentals
• Turbulence model? • Particle/Fluid and Particle/Particle interactions? • Computational Multiphase Flow Model? • Numerical Solver and Grid Generator?
22
Turbulence Model for carrier fluid
Numerical model Details of turbulence
RANS (Reynolds Average Navier Stokes)
No dynamics: just an integral length scale
LES (Large Eddy Simulation) Dynamics of the most energetic eddies
DNS (Direct Numerical Simulation)
Dynamics of all eddies
12
23
Dispersed Phase fraction
Ni = number of particles in size fraction i
VPi = π Dpi3 /6 = particle volume
Volume fraction Mass fraction
UP , UF are particle/fluid velocities
24
Inter-particle spacing
For αp = 0.01: spacing ~ 4 Dp
For αp = 0.10: spacing ~ 2 Dp
13
25
Dispersed two-phase flow regimes
Dilute Dispersed Two-Phase Flow
One-Way Coupling
Two-Way Coupling
Dense Dispersed Two-Phase Flow
Four-Way
Coupling
26
Particle Tracking coupling mechanisms
14
27
Particle Response time
Particle response time Stokes number
Fluid flow time scale:
28
Models for Computational Dispersed Multiphase Flows
DNS resolving the particles Extremely demanding in computing time and memory
Discrete Particle Model (DPM) Used as learning tool in academia to study closure relations
Euler-Lagrange (Particle Tracking) For dilute flows with various turbulence models: LES, RANS
Euler-Euler (Two-Fluid) For intermediate to dense loadings with RANS
15
29
Eulerian-Lagrangian RANS: k-ε model with particles
30
Euler-Euler (Two-Fluid) simulations
k=1: continuous phase; k=2:dispersed phase
16
31
Contents
• Industrial flow challenges • Requirements for CFD of dispersed multiphase flow • Fundamentals • Some Examples (bubbles, drops and particles flows) • Sources of Errors • Checklist of Best Practice Advice
32
Bubbles in a Bubble column
1. Turbulence generated by bubbles injected at bottom
2. Radial distribution? 3. How to maximize bubble
liquid interface? 4. Residence time of
bubbles? 5. Flow regime affected by
upscaling?
17
33
Simulation approach
• Euler-Euler needed due to high loading • Turbulence with RANS • Drag and gravity forces • Wide size distribution: broad range of bubble velocities • 2D steady state simulation on 20,000 nodes takes
several hrs. On a single processor (1GHz) • 3D transient simulations needed for oscillations • 3D transient requires weeks per simulation! • Compared to single-phase Two-Fluid model has to use
coarse grids to keep computing times realistic
34
Influence of sparger
Flow field Gas fraction (Double and single ring spargers, left and right, resp.)
18
35
Axial time-averaged liquid velocity
Sanyal et al. (1999)
36
Gas fraction profile
Sanyal et al. 1999
19
37
3D simulation Bubble column
• RANS with drag and gravity forces • Lift force, needed for near-wall behaviour, unreliable • Two-Fluid model with all bubbles moving at ensemble
averaged mean velocity of dispersed phase • Mixture model with all bubbles moving at ensemble
averaged mean velocity of dispersed phase • Mixture model with N+1 phases and different bubble
size moving at different velocities
38
Radial gas fraction
Chen et al. (2005)
20
Axial liquid velocity profiles
39
Deen et al 2001
Turbulent kinetic energy profiles
40
Deen et al 2001
21
41
Status Bubble column simulation
• Mixture model is a cheap way for qualitative trends • For dense bubbly flows coalescence and break-up can
seriously change the flow field • Multi-Fluid model for quantitative results • Computational costs of adding more phases is large • Bubble-break-up model needs attention • Break-up and coalescence very sensitive to surfactants • Euler-Euler with LES turbulence model promising
42
Stirred tank with glass beads in water LES snapshot
Derksen (2003)
22
43
Stirred Tank Reactor
• Complex turbulent flow field • Drop size • Shape of drop size distribution • Flow field variation with stirrer speed
44
Drops in a stirred tank reactor
Experimental findings: • Drop size decrease with increasing vessel size • Shape of drop size distribution independent of vessel
size and stirrer speed • After change in stirrer speed a new steady state drop
size distribution requires more stirrer revolutions for larger vessels
(Colenbrander, 2000)
23
45
RANS results for bubbles in liquid
Sliding mesh with matching frames and 48,000 cells
(Issa, 1998)
Drag and gravity forces for particles in k-ε turbulent field
46
Summary RANS findings
• Fine to establish impeller location • Multiple frame approach useful • Convergence problems due to equation coupling • Liquid velocity close to impeller erroneous • Gas hold-up values only qualitatively correct • Turbulence intensities not accurate enough for break
up and coalescence sub-models
(Montante et al. (2001) and Issa (1998))
24
47
LES for ‘drops’ in stirred vessel
• Flow driven by Rushton turbine • Lattice Boltzmann scheme • Smagorinski constant 0.12 • Uniform, cubic grid with 6 million nodes • Parallel shared memory: CPU= 34 hrs./simulation • Adaptive force fields for impeller and walls
(Derksen et al. (1998,1999)
48
Angle resolved results simulation
experiment
0.0
0.045 0.030
0.075 0.060
0.090
0.015
0.135 0.120 0.105
k/vtip2 simulation
experiment
Kinetic energy at 19o behind the blade
25
49
Average turbulent fields Turbulent kinetic energy
k/vtip2
0.1
0.025
0.006
0.003
0.0015
0.0004
Dissipation rate ε/εav
10.0 3.0 1.0 0.3 0.1 0.03 0.01 0.003 0.001
50
Evaluation of LES and RANS Guha et al. AIChE Journal 54 (2008) 766-778
• Experiments with Computer Automated Radioactive Particle Tracking (CARPT)
• Euler-Lagrange simulation with LES • Euler-Euler simulation with RANS • Overall solids hold-up of 1% by volume in STR with
impeller speed of 17 revolutions/s and Re=74000 • T=0.2 m, H=T; turbine blade D=T/3; water with solids
of mean diameter of 0.3 mm
26
51
Simulated Flow pattern
CARPT RANS LES
52
Radial profiles of solids radial velocity
27
53
Radial profiles of solids turbulent kinetic energy
54
Slip Reynolds number at impeller cross section
28
55
Axial variation of mean sojourn time
56
Conclusions LES/RANS evaluation Guha et al. (2008)
• Observed bottom recirculation loop stronger than top not found in either simulation
• Azimuthally averaged solid velocities with LES slightly better than with Euler-Euler RANS
• Solids turbulent kinetic energies, similar for both models, over-predict at impeller plane and under-predict at all other axial locations
• Slip velocities from RANS order of magnitude lower than those from LES
• Mean sojourn times from LES agree with experiments
29
Euler-Lagrange with bubble breakup and coalescence (Sungkorn et al 2011)
57
Euler-Lagrange with bubble breakup and coalescence (Sungkorn et al 2011)
58
30
59
Status stirred tank simulations
• Currently only data for bubbles and particles • RANS fails to correctly predict both k and ε • LES required for proper turbulence intensity • LES applied in Euler-Lagrange with solids and gas • With Bubble break-up and coalescence implemented,
mean-diameter and av. liquid velocities agree with measurements in laboratory-scale reactors
• LES in lattice Boltzmann technique ready for higher loading of industrial interest
Recent literature on Stirred Tank Reactor simulations
• J.J. Derksen, “Numerical simulation of solids suspension in a Stirred Tank”, AIChE Journal 49 (2003) 2700-2714
• D.Gunha, P.A. Ramachandran, M.P. Dudukovic and J.J. Derksen, “Evaluation of Large Eddy Simulation and Euler-Euler CFD models for solids flow dynamics in a stirred Tank reactor”, AIChE Journal 54 (2008) 766-778
• R. Sungkorn, J.J. Derksen and J.G. Khinast, “Euler-Lagrange modeling of gas-liquid stirred reactor with consideration of bubble breakage and coalescence”, accepted paper AIChE Journal (2011)
31
61
Particle flow in a riser
LES/DNS for continuous phase • One-way coupling with gravity, drag, lift forces
in a riser or channel • Two-way coupling effects • Collisions
62
Particle Tracking for riser flow
Uijttewaal and Oliemans, 1996
One-way
coupling
32
63
Particle dispersion
64
Particle deposition
Dotted: exp.
Diamonds: DNS
33
65
Particle wall transfer in upward turbulent pipe flow
(Marchioli et al. (2003))
66
Elongated particle clusters in a channel
From top to bottom: t+= 0, 706, 1412, 2118
(Marchioli and Soldati (2002))
One-way DNS in vertical upward channel flow with Stokes drag, gravity and lift
34
67
Particle near-wall behaviour
68
Near-wall particle distribution and turbulent coherent structures
t+= 1412 t+=1450
Green:counterclockwise Red: clockwise rotating vortices Blue dots:descending, black dots: ascending particles
35
69
Particle-Tracking results for wall-bounded flows
LES/DNS for continuous phase • One-way coupling with gravity, drag, lift forces
in a riser • Two-way coupling effects in channel flow • Collisions
70
Two-way point particle results
(Portela, 2000)
Normalwise Turbulence Intensity
for increasing particle density
Channel flow with point particles
36
71
Low speed streaks
unladen laden
72
Particle-Tracking with collisions
(Li et al., 2001)
37
73
Particle affected Cµ
0.09 for Single phase flow
74
Status particle pipe flow simulation
• LES well established for Euler-Lagrange scheme • Dispersion depends on particle size and turbulence • Only point particles considered • Drag, gravity and lift forces determine particle
distribution in inhomogeneous turbulence • Physics close to the wall important • Particle wall accumulation affected by two-way
coupling and collisions • LES still too expensive for industrial use
38
75
Contents
• Industrial flow challenges • Requirements for CFD of dispersed multiphase flow • Fundamentals • Some Examples (bubbles, drops and particles flows) • Sources of Errors • Checklist of Best Practice Advice
76
Sources of Errors
• Physical Mechanisms • Closure models • Time and length scales • Governing Equations • Numerical errors • User errors
39
77
Contents
• Industrial flow challenges • Requirements for CFD of dispersed multiphase flow • Fundamentals • Some Examples (bubbles, drops and particles flows) • Sources of Errors • Checklist of Best Practice Advice
78
Best Practice Advice -1
1. Determine size distribution and volume fraction of particles injected
2. Assess particle forces: drag, gravity, lift,..? 3. Select appropriate models for the particle forces 4. For dilute flows (αd<10-3) use Euler-Lagrange 5. For intermediate flow (10-3< αd<4x10-1): Euler-Euler 6. Assess whether dynamics of carrier phase is needed 7. Select turbulence model
40
79
Best Practice Advice -2
8. Make sure wall roughness is accounted for 9. For bubbles or droplets ensure break-up and
coalescence models are available 10.Start with single-phase mesh-independent simulation 11.Continue multiphase simulations until for converged
results sufficient data are available for particle statistics
80
Possibilities
• Multi-Fluid model for dense dispersed flow with turbulence modulation and collisions
• Euler-Lagrange models with LES/DNS for understanding basic physics
• Particle segregation in riser flow quantified • Dependence of turbulence model constants on
particle loading and size can be established
41
81
Limitations
• Closure of Multi-Fluid model • Euler-Lagrange for point particles with volume
fractions up to 0.01 • Computing time/memory • Validation • Dealing with large amount of simulation data • Transfer from Academia to Industry
82
Best Practice Guidelines
• Best Practice Guidelines for Industrial Computational Fluid Dynamics, ERCOFTAC (2000)
• Best Practice Guideliens for Computational Dispersed Multiphase Flows, ERCOFTAC-SIAMUF (2008)
• Available via ERCOFTAC website (www.ercoftac.org)