cfd-based modeling of inflight mercury capture

CFD-based Modeling of Inflight Mercury Capture

Hg

HgCl2

Jens Madsen Thomas O’BrienAnsys / Fluent Inc. NETL / U.S. Dept. of Energy

Morgantown, West Virginia Morgantown, West Virginia

10th Electric Utilities Environmental Conference Tucson, January 22-24. 2007

cwright

Text Box

DOE/NETL-IR-2007-069

EUEC-10, Tucson, January 2007

Background and Motivation

• 1,100+ coal-fired units in the US− 40% of man-made mercury emissions− Annual sorbent cost for typical 300 MW

plant estimated to $1-2 million − DOE’s programmatic goal is to reduce

cost by 25-50% over baseline estimates• Project Goal

− “Use CFD-based tools to simulate and improve the understanding of sorbent-based mercury control processes”

− Flow modeling support for DOE/NETL field test sites over the past three years

• CFD benefits− Enables parametric study and

optimization of capture processes − This may substantially reduce the cost of

CAMR compliance

Monroe (Detroit-Edison) Brayton Point (PG&E Natl.Energy)

Meramec (Ameren-UE) Yates (Southern Co.)

Presque Isle (Wepco)



• Practical questions answered−Optimize injection grids−Predict necessary sorbent feed rates− Inexpensive what-if studies

• Detailed information provided −Flue gas conditions

• Velocity• Temperature• Mercury (elemental/oxidized)

concentrations [µg/m3] • ….

−Sorbent conditions• Dispersion and residence time• Where the capture takes place

Data for Model Calibration & Validation

Support of DOE/NETL field test activities



• Few existing models of mercury capture− Typical simplifications include:

• Plug gas flow (1D models) • Uniform sorbent dispersion• No velocity slip between particles and flue gas

• CFD-based model without such simplifications− Based on first principles (conservation laws)− Considers adsorption of Hg(o) and HgCl2− Known partitioning between these species (oxidation fraction) used as model input

Hg

HgCl2


Presentation Outline

• Consider distinct mass transfer processes− Occur on multiple scales − Any single process may limit the overall mercury capture

1. Injection and dispersion of sorbent• Discrete Particle Modeling (DPM)• Injection Lance design (single- or multi-nozzle lances?)• Injection Grid design (impact on sorbent dispersion and Hg

capture) 2. Duct-scale transport of gaseous mercury species

• Transport by convection and turbulent diffusion• Mercury sink term determined from DPM simulations

3. Film mass transport• Transfer from bulk gas phase to sorbent exterior

4. Pore diffusion through sorbent’s interior• Molecular (laminar) or Knudsen diffusion transport

5. Surface adsorption on internal sites• Adsorption rates calculated using Langmuir theory

DUCT SCALE

SORBENT SCALE


Discrete Phase Model (DPM)

• Trajectories of particles are computed in a Lagrangian frame− Each trajectory represents a group of particles with same initial properties− Particle-Particle interaction is neglected (dilute solid-gas flows)− Particle force balance determines trajectory (Newton’s 2nd Law)− Effect of turbulence modeled by stochastic tracking

•Particles behave differently based on size

•Particle size distribution (PSD) represented by discrete bins

•Typically ten size bins (dp= 1 … 100µm)•Trajectory flow rates weighted by PSD

PSD for DARCO-Hg

Considered size range


Injection Lance Design

0.30

0.39

0.19

0.080.03 0.01

0.61

0.84

0.94 0.96 0.97 0.98

0.0

0.2

0.4

0.6

0.8

1.0

1 20 40 60 80 100Diameter [micron]

Particle Size Fraction Lower Nozzle Fraction

• Determine sorbent split for multi-nozzle injection lances • Multi-nozzle lances offer a false sense of security

− Sorbent split can be very uneven (here 81% exits lower set of nozzles)− Sorbent coverage very similar to that of a much simpler single-nozzle lance− Staggered lance arrangements is a preferable approach to achieve good coverage

from top-to-bottom of duct− Note: Smaller size fraction does most the capture

Four-nozzle lanceUsed at DE-Monroe


Sorbent Dispersion – DTE Energy’s Monroe Plant

• Monroe plant has a very wide rectangular duct (51.5ft)

• Major stratification problems (temperature/sorbent/capture)

• Five multi-nozzle injection lances provide only partial coverage

• Stratification causes packages of gas to pass untreated by ACI

• Overall CFD predictions agree with outlet mercury sampling and analysis of hopper ash mercury content

InletAC Injection

Outlets (~50% flow each)

Ladder vanes

Splitter plate

Perforated plate


Navigation Slide

1. Injection and dispersion of sorbent• Discrete Particle Modeling (DPM)• Injection Lance design (single- or multi-nozzle lances?)• Injection Grid design (impact on sorbent dispersion and

Hg capture)

2. Duct-scale transport of gaseous mercury species • Transport by convection and turbulent diffusion• Mercury sink term determined from DPM simulations




Compute sorbent trajectories using DPM

Solve scalar equation(s) for Hg transport

Compute Hg adsorption rates

Convergence?No

Solve gas phase momentum equations

START

Yes

STOP


Duct-Scale Mercury Transport

• Mercury transport equations• Determines distribution of gas-phase

mercury in duct

• Turbulent diffusion dominates

• µt >> µmol

• Similar diffusivity for Hg(o) and HgCl2

Hgi

Hg

t

tmolgi

i

SxSc

µµcρux

=

∂

∂

+−

∂∂ c

sink term

Inlet Hg level

Carrier Gas(no Hg)

Hg level tapers off


Navigation Slide

1. Injection and dispersion of sorbent• Discrete Particle Modeling (DPM)• Injection Lance design (single- or multi-nozzle lances?)• Injection Grid design (impact on sorbent dispersion and

Hg capture) 2. Duct-scale transport of gaseous mercury species

• Transport by convection and turbulent diffusion• Mercury sink term determined from DPM simulations




Compute sorbent trajectories using DPM

Solve scalar equation(s) for Hg transport

Compute Hg adsorption rates

Convergence?No

Solve gas phase momentum equations

START

Yes

STOP


Capture Modeling – Film Mass Transfer

c∞

Dp

position

concentration

co

δ

r

• Film layer effect: drop in Hg concentration from bulk phase to sorbent surface − Concentration drop across film determined from flux conservation

• Mass Transfer Coefficient kf from empirical relation for Sherwood number

( )off cckJ −⋅= ∞Film mass flux:


Capture Modeling – Porous Diffusion

Molecular Diffusion – intermolecular collisions • Binary system: air+mercury (Hgo or HgCl2)• Chapman-Enskog (molecular) theory

Ω

+⋅= 2

12 p2/11/13/2T3-10 1.86 molD

σ

MWMW

Knudsen Diffusion – molecule/walls collisions

• Does not depend on gas composition nor pressure

MWT

2pored

9700 KnD =

• Two modes of porous diffusion• Less diffusive mode limiting (Molecular or Knudsen Diffusion) Effective diffusivity Deff

• Added resistance from tortuosity of the porous media • In both diffusive modes: DHgCl2< DHg(o) (by about 35…40%)

1-

KnD1

molD1

p

p effD

+=

τ

ε


Capture Modeling – Porous Interior

• Closed-form solution for concentration profile within perfect sphere

• Introduction of sorbent effectiveness η

• Allows use of surface values co throughout volume when calculating adsorption rates

Concentration Profile in spherical particle First order reactions

0

0.2

0.4

0.6

0.8

1

1.00 0.75 0.50 0.25 0.00 r/rp

cp/co

Thiele - Low Thiele - Medium Thiele - High

η=0.939

η=0.48

η=0.115

Reaction limited

Diffusion limitedc∞

Dp

position

concentration

co

δ

drr

Thiele Modulus:


Capture Modeling – Surface Adsorption

• Mercury adsorption rates computed using Langmuir isotherms• Separate isotherm parameters for each mercury species• Capture by UBC may be accounted for by separate particle stream with own rates

• Langmuir: net adsorption rate = forward rate (k1) minus desorption rate (k2)

• Here θ is the sorbent utilization (ω / ωmax ), ie. fraction of occupied sites• ωmax is the maximum number of available sites (sorbent capacity)

• Isotherm parameters (ωmax, k1, and b = k1/k2) are temperature-dependent• Getting proper isotherm data for a sorbent is challenging• When determined from packed bed breakthrough curves, adsorption process is essentially

lumped with film transfer and pore diffusion

[ ] θθ max2omax1 ωω kc1k −−=ℜ


Capture Modeling – Summary

Sorbent trajectories

HgCl2 sink terms

HgCl2 concentration (no ACI)

HgCl2 concentration (with ACI)


Capture Modeling Results – Brayton Point

• Comparison: model predictions vs. field test measurements− Quantitatively, capture is under-predicted− Qualitatively, trends are well represented (as illustrated in normalized plot)

Normalized Capture vs. Injection Rate

0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

0 5 10 15 20 25 30 35

lb/Macf

Nor

mal

ized

cap

ture

Field Test CFD

Capture Efficiency

0

1020

3040

50

6070

8090

100

0 5 10 15 20 25 30 35lb/Macf

Field Test CFD


Parametric Study – Size Effects

• Capture efficiency for different uniform particle sizes− Maintain constant injection rate− No appreciable capture for size

fractions dp>10µm

• Conclusion: Size does matter! − Focus on smallest size fraction− Darco-Hg: ~1/3 smaller than 10µm − Modeling-wise, this means sensitivity

towards discrete representation of size distribution (number of bins)

Normalized Mercury Capture

0

0.5

1

1.5

2

2.5

3

0 10 20 30 40

Particle Size [micron]

Normalized Capture


Concluding Remarks

• CFD-based mercury capture model − Enables cost-effective optimization of injection systems

− Directly addresses major cost component of ACI control technology

− Detailed modeling framework exist

− More work pending on determining proper set of adsorption rates


Extra slides


Validation of Discrete Particle Model

• Can these predictions of sorbent dispersion be trusted?− Dispersion data not available for real power plants− Circumstantial evidence exist in the form of dispersion results that match capture

stratification patterns at Monroe field test site− A more thorough model validation required

• Model validation based on well-documented experiments *− Dispersion of particle jet in isotropic turbulence− Turbulence is generated in experiment using a screen− CFD model closely match experimental conditions (such as inlet turbulence parameters

y

* W.H. Snyder and J.L. Lumley : “Some measurements of particle velocity autocorrelation functionsin a turbulent flow”, Journal of Fluid Mechanics, 1971, vol. 48 (No.1), pp 41-71.


-1

0

1

2

3

4

5

6

7

0 50 100 150 200 250 300 350 400 450 500

Time (msec)

Dis

pers

ion

(cm

2)Stochastic ModelExperimentalCloud Model

Validation of Discrete Particle Model (2)

NavgYY

Dispersion2)( −

=

• Stochastic Tracking under-predicts the particle dispersion by 5 … 30%

cfd-based modeling of inflight mercury capture

Documents