Está en la página 1de 275

A NUMERICAL INVESTIGATION OF GAS CYCLONE SEPARATION

EFFICIENCY WITH COMPARISON TO EXPERIMENTAL DATA AND

PRESENTATION OF A COMPUTER-BASED CYCLONE DESIGN

METHODOLOGY

A Thesis

Presente d to

The Graduate Faculty of the University of Akron

In Partial Fulfillment

of the Requirements for the Degree

Master of Science

Steve Kegg
A NUMERICAL INVESTIGATION OF GAS CYCLONE SEPARATION

August, 2008
A NUMERICAL INVESTIGATION OF GAS CYCLONE SEPARATION

EFFICIENCY WITH COMPARISON TO EXPERIMENTAL DATA AND

PRESENTATION OF A COMPUTER-BASED CYCLONE DESIGN

METHODOLOGY

Steve Kegg

Thesis

Approve Accepte
d: d:

Advisor Dean of the College


Dr. Minel J. Braun

Faculty Dean of the Graduate


Reader Dr. School Dr. George R.
Scott Sawyer Newkome

Department Chair Date


Dr. Celal Batur

ii
ABSTRACT

Cyclone separators have existe d since the 1800s an d are still widely use d in

many industries to separate particles from gases. Although cyclones are geometrically

simple, the physics describing the flow and separation processes which occur in them is

complex. Over the decades many researchers have studied these devices and have

developed a number of theories and empirical models for design purposes. In practice,

most cyclones are designed using some type of empirical information. Physical prototypes

are then built, tested and tuned until an acceptable level of performance is obtained.

Recent advancements in numerical methods and in the performance capabilities of

moderately price d computers have opene d the possibility of developing computer-base d

methods, which are not primarily based on empirical models that can be effectively use d

for cyclone design. Cyclone flows are characterized by high swirl and stream wise

curvature. This paper presents a description of the numerical models that can be used to

calculate the performance of cyclones including the gas flow and the particle tracking

processes.

A commercially available computational fluid dynamics (CFD) computer program

using these numerical models was used to calculate the performance of a cyclone at

several operating points. The calculated performance was then compared to experimental

data. One of the characteristics of the separation process which was

3
observed was the short circuiting of particles. This short circuiting allowed some particles

to leave the system shortly after they were injecte d. The phenomenon of particle re-

entrainment from the dust bin, which re duces the effectiveness of the cyclone, was also

observed in the calculated results.

As a result of the work done in this study, a computer-base d cyclone separator

design methodology is presented. The methodology is believed to be unique in that it

takes advantage of the cyclone design knowledge which has been gained over the years of

research by others as well as current state-of-the-art numerical methods. The existing

knowledge is use d to provide quick starting geometry at the beginning of a new design

process when no other information is available. This knowledge is presented in the form

of cyclone performance and sizing correlations. Also, to aid in the design process,

guidelines have been assembled from the literature, which help the designer decide which

types of geometry changes should be considered to affect the performance characteristics

with which he is most concerned. The heart of the methodology is a CFD-based approach

that provides detailed cyclone performance calculations. This methodology provides the

potential to produce cyclone designs with the required performance characteristics more

quickly and more economically than older methods which use empirical and experimental

design approaches exclusively.

4
TABLE OF CONTENTS

Page

LIST OF TABLES ............................................................................................................. xi

LIST OF FIGURES .........................................................................................................xiii

CHAPTER

I. INTRODUCTION ......................................................................................................... 1

1.1 Cyclone Separator Overview................................................................................... 1

1.2 Literature Review.................................................................................................... 4

1.3 ScopeofWork ........................................................................................................ 13

II. ANALYTICAL MODEL ........................................................................................... 15

2.1 Intro duction .......................................................................................................... 15

2.2 Flow Governing Equations ................................................................................... 20

2.3 Turbulence ............................................................................................................ 21

2.3.1 Intro duction to Turbulence Mo dels .............................................................. 21

2.3.2 Type I: Eddy Viscosity Models Using the Boussinesq Assumption ............. 25

2.3.3 Type II: Models Based on the Reynolds-Stress Equation ............................. 29

2.3.4 Type III: Models Not Based Completely on the Reynolds-Stress Equation...

31

2.3.5 The Turbulence Mo del Use d for this Work ................................................. 32

5
2.4 Particle Dynamics Motion Equations ................................................................... 32

6
2.5 Chapter Summary 34

III. THE NUMERICAL MODEL AS IMPLEMENTED IN FLUENT ......................... 36

3.1 Intro ducti on ......................................................................................................... 36

3.2 .......................................................................................................................

Implementation of the Flow Governing Equations ...................................................... 36

3.3 Choice of Turbulence Model and its Implementation .......................................... 37

3.4 .......................................................................................................................

Two Phase Flow: Air an d Solid Particle Mixture........................................................ 41

3.4.1 Basic Particle Trajectory Calculations ........................................................... 41

3.4.2 The Turbulent Dispersion of Particles ........................................................... 42

3.5 The Fluent Solver an d Solution Metho dology ................................................... 45

3.5.1 Overview of Control Volume Metho d .......................................................... 45

3.5.2 Choice of Equation Discretization Schemes .................................................. 48

3.5.3 Solving the Equations .................................................................................... 50

3.5.4 Convergence Criteria ..................................................................................... 52

3.6 Chapter Summary ................................................................................................. 53

IV. EXPERIMENTAL DATA........................................................................................ 55

4.1 Intro ducti on ......................................................................................................... 55

4.2 150 mm Cyclone Geometry.................................................................................. 55

4.3 Test Dust Characteristics ...................................................................................... 55

4.4 Experimental Collection Efficiency Results......................................................... 58

4.4.1 Grade Efficiency Curves ................................................................................ 58

vii
5.10 Chapter Summary 127

4.4.2 Absolute Collection Efficiency...................................................................... 60

4.4.3 Measurement Uncertainty Estimates ............................................................. 65

4.4.3.1 Uncertainty Estimates for Efficiency Measurements.............................. 68

viii
70
4.4.3.2 Uncertainty Estimates for Velocity Measurements ...................

4.4.3.3 Overall Uncertainty of the Measured Test Result for Efficiency ........... 71

4.4.3.4 Overall Uncertainty of the Measured Test Result for Particle size ........ 72

4.4.3.5 Complete d Uncertainty Results for all Three Inlet Velocities ............... 74

4.5 Chapter Summary ................................................................................................. 77

V. NUMERICAL STUDY.............................................................................................. 79

5.1 .......................................................................................................................

Intro ducti on ................................................................................................................ 79

5.2 Preparation of the Geometry for Grid Generation ................................................ 79

5.3 Computational Grid .............................................................................................. 81

5.3.1 Grid Construction Sequence .......................................................................... 83

5.4 .......................................................................................................................

Material Properties an d Boun dary Con ditions .......................................................... 89

5.5 Time Step Size ...................................................................................................... 92

5.6 Fully Unsteady Continuous Phase and Particle Tracking Calculations ............... 94

5.7 Quasi-Unsteady Particle Tracking Calculations in a "Frozen" Flow Field ....... 108

5.8 .......................................................................................................................

Sources of Error an d Grid Refinement Stu dy ........................................................... 112

5.8.1 Sources of Simulation Error ......................................................................... 112

5.8.2 Estimating Discretization Error using a Gri d Refinement Study ............... 113

5.8.3 Results for the Coarse and Fine Grid Models at 15.1 m/s ........................... 120

5.8.4 Static Pressure Drop Discretization Error Estimate..................................... 122

5.8.5 Particle Cut Size Discretization Error Estimate ........................................... 122


9
5.10 Chapter Summary 127
5.8.6 Cyclone Absolute Efficiency Discretization Error Estimate ....................... 123

5.8.7 Summary of Grid Refinement Discretization Error Calculations ................ 124

5.9 Results for 150 mm Cyclone at Three Flow Rates with Discretization Error ... 125

10
70

VI. RESULTS AND DISCUSSION ............................................................................. 129

6.1 Introduction ........................................................................................................ 129

6.2 A Comparative Study: Numerical Results and Experimental Validation .... 129

6.2.1 The 10.3 m/s Inlet Velocity Case ................................................................. 129

6.2.2 The 15.1 m/s Inlet Velocity Case................................................................. 131

6.2.3 The 19.7 m/s Inlet Velocity Case................................................................. 133

6.3 Observations ....................................................................................................... 135

6.3.1 Cut Size ........................................................................................................ 136

6.3.2 Absolute Efficiency ..................................................................................... 137

6.4 Discussion ........................................................................................................... 138

6.4.1 .................................................................................................................

Review of the Stairmand Grade Efficiency Diagram from Chapter 1 .................... 139

6.4.2 .................................................................................................................

Separation Efficiency for Particles Larger than the Cut Size ................................. 140

6.4.2.1 Turbulent Eddies ................................................................................... 141

6.4.2.2 Particle Bouncing.................................................................................. 142

6.4.3 Separation Efficiency for Particles Smaller than the Cut Size .................... 142

6.4.3.1 Particle Collisions and Agglomeration ................................................. 142

6.5 Suggested Future Areas of Investigation ............................................................ 143

6.5.1 Turbulence Modeling ................................................................................... 144

6.5.2 Particle Collisions and Agglomeration ........................................................ 144

11
5.10 Chapter Summary 127

6.5.3 Fully Unsteady Continuous Phase and Particle Tracking Calculations ...... 144

6.6 Chapter Summary ............................................................................................... 145

VII CYCLONE DESIGN METHODOLOGY ............................................................... 146

12
7.1 Intro duction 146

7.2 Design Methodology Flow Chart ....................................................................... 147

7.3 Description of Methodology Process Steps for a Typical Design Cycle ............ 149

7.4 .......................................................................................................................

Detaile d Cyclone Design Example Problem ............................................................. 152

7.4.1 Step 1: Define the Design Requirements ..................................................... 153

7.4.2 Step 2: Does Geometry Already Exist? ....................................................... 153

7.4.3 Step 3: Calculate Initial Geometry ............................................................... 153

7.4.4 Step 4: Create Model of the Geometry and the Computational Grid .......... 154

7.4.5 Step 5: Solve Flow Problem and Obtain Performance Information ............ 156

7.4.6 Step 6: Compare Performance of Current Design with Requirements ........ 159

7.4.7 Step 7: Review Design Guidelines an d Specify Mo difications ................. 160

7.4.8 Re-enter Step 4: Revise the Geometry and the Computational Grid ........... 161

7.4.9 Re-enter Step 5: Calculate Performance of Revise d Cyclone .................... 163

7.4.10 ...............................................................................................................

Re-enter Step 6: Compare Revise d Design with Requirements ............................ 166

7.5 Chapter Summary ............................................................................................... 166

VIII. CONCLUSIONS .................................................................................................... 167

8.1 Accuracy an d Efficiency of the CFD Particle Separation Calculations ............. 167

8.2 Limitations of the Model .................................................................................... 168

8.3 Computer-Base d Cyclone Design Methodology ............................................... 169

REFERENCES ............................................................................................................... 170

xiii
APPENDICES ................................................................................................................ 175

A. SVAROVSKY EULER NUMBER - STOKES NUMBER CORRELATION ...... 176

B. DERIVATION OF THE SVAROVSKY CYCLONE SIZING EQUATION ....... 179

xiv
LIST OF TABLES

C. GUIDELINES FOR CYCLONE PROPORTIONS ................................................ 186


D. ............................................................................................................................
CYCLONE OPTIMIZATION STUDY FROM LEITH ET AL .................................... 189

15
Table Page

3.1 Turbulence mo del coefficient values an d source references ................................... 38

4.1 Grade efficiency data for 150 mm Ogawa cyclone ................................................... 59

4.2 Efficiency vs. particle size for absolute efficiency calculation example .................. 61

4.3 Particle size ranges for absolute efficiency calculation example.............................. 63

4.4 Mass fraction in each particle size range for absolute efficiency example............... 64

4.5 Mass collecte d in each size range for absolute efficiency calculation..................... 65

4.6 Absolute fractional collection efficiency data forOgawa150 mm cyclone ............... 66

4.7 Absolute efficiency data for 150 mm cyclone in the19.7 to21.4 m/s range .............. 68

4.8 Uncertainty estimates for absolute collection efficiency measurements .................. 69

4.9 ...........................................................................................................................

Uncertainty estimates for velocity measurements ............................................................. 70

4.10 Final uncertainty estimates for efficiency & particle size ...................................... 75

5.1 Material properties .................................................................................................... 89

5.2 Flow boun dary con ditions ....................................................................................... 90

5.3 Quantities relate d to time step size selection ........................................................... 93

5.4 Fine and coarse grid performance results for 15.1 m/s inlet velocity ..................... 121

16
LIST OF TABLES

5.5 Discretization error calculations summary for 15.1 m/s inlet velocity case ........... 124

5.6 Summary of 150 mm cyclone calculation results ................................................... 126

17
6.1 Experimental vs. calculated efficiency at 10.3m/s inlet velocity ............................ 130

6.2 Experimental vs. calculated efficiency at 15.1m/s inlet velocity ........................... 132

6.3 Experimental vs. calculated efficiency at 19.7m/s inlet velocity ........................... 134

149

7.1 Design methodology process steps: descriptions & instructions

7.2 Material properties .................................................................................................. 156

7.3 Flow boundary conditions ...................................................................................... 157

7.4 Comparison of performance of initial design with requirements ........................... 159

7.5 Comparison of design ratios for cyclone inlet and outlet ducts.............................. 161

7.6 Flow boundary conditions ...................................................................................... 163

7.7 Performance comparison of initial and revised design with requirements ............. 166

18
LIST OF FIGURES

Figure Page

1.1 Cyclone components ................................................................................................... 1

1.2 Flow patterns in a cyclone separator ........................................................................... 2

1.3 Basic cyclone dimensions ........................................................................................... 3

1.4 Tangential velocity characteristics from Stairmand (1951) ........................................ 4

1.5 Diagram of forces acting on a single particle ............................................................. 5

1.6 Diagram of grade efficiency curve characteristics from Stairmand (1951) ................ 8

2.1 Cyclone coordinate system an d velocity component description ............................ 15

2.2 Turbulent boundary layer developing on a flat plate................................................ 22

2.3 Relationships of instantaneous, mean and fluctuating velocitycomponents ............ 23

3.1 Control volumes use d to illustrate the transport of quantity (/) ............................... 47

3.2 General segregate d solver solution process steps .................................................... 51

4.1 Ogawa 150 mm cyclone geometry ........................................................................... 56

4.2 Kanto-loam test dust size distribution ...................................................................... 57

4.4 Experimental absolute collection efficiency data ..................................................... 66

4.5 Cyclone at 10.3 m/s grade efficiency points with experimental error ..................... 76

4.6 Cyclone at 15.1 m/s grade efficiency points with experimental error ..................... 76

xix
4.7 Cyclone at 19.7 m/s grade efficiency points with experimental error ..................... 77

xx
5.1 Cyclone geometry features relate d to the grid generation process .......................... 80

5.2 Inlet pipe to cyclone bo dy transition region ............................................................ 81

5.3 Cyclone grid construction overview ......................................................................... 83

5.4 Plane where grid construction starts ......................................................................... 84

5.5 Starting grid size and distribution ............................................................................. 84

5.6 Hex grid in cyclone body and exit pipe .................................................................... 85

5.7 Hexahedral cells are connected to tet cells with pyramid cells ................................ 86

5.8 Tetrahe dral cells complete mesh in the transition region ........................................ 87

5.9 Prism cells used in the inlet pipe extension .............................................................. 87

5.10 Grid in cone of cyclone ........................................................................................... 88

5.11 Complete d Grid ...................................................................................................... 89

5.12 Overview of separation process .............................................................................. 95

5.14 Particle injection at 0.04 seconds ............................................................................ 97

5.15 Short circuiting of particles at 0.14 seconds ........................................................... 98

5.16 Short circuiting an d filling of dust bin at 0.24secon ds ......................................... 99

5.17 Continue d dust bin filling an d short circuiting at 0.34 secon ds ......................... 100

5.18 Process at 0.44 secon ds ........................................................................................ 101

5.19 Process at 1.0 secon d ........................................................................................... 102

5.20 Process at 6.0 secon ds .......................................................................................... 103

5.21 Process at 12.5 seconds ......................................................................................... 104

5.22 Cyclone fractional efficiency vs. particle size vs. time ........................................ 105

5.23 Cyclone fractional efficiency vs. particle sizevs. time with curve fitting ............. 107

21
5.24 Cyclone grade efficiency curve at 15.1 m/s .......................................................... 107

22
5.25 Collection efficiency vs. particle size vs. time using alternate approach ............. 109

5.26 Gra de efficiency curves comparing both tracking approaches ............................ 110

5.27 Starting grid surface for fine grid discretization error study ................................ 118

5.28 Complete d fine grid cyclone mo del mesh .......................................................... 119

5.29 Close up of completed fine grid cyclone model mesh .......................................... 120

5.30 Coarse vs. fine grid gra de efficiency curves at15.1m/s ....................................... 121

5.31 Grade efficiency curves at three inlet velocities from calculations ...................... 126

6.1 Experimental vs.calculated grade efficiency forthe 10.3 m/s case ..................... 131

6.2 Experimental vs.calculated grade efficiency forthe 15.1 m/s case ..................... 133

6.3 Experimental vs.calculated grade efficiency forthe 19.7 m/s case ..................... 135

6.4 Experimental vs.calculated cut size comparison .................................................. 136

6.5 Experimental vs. calculate d absolute efficiency comparison ................................ 138

6.6 Diagram of grade efficiency curve characteristics from Stairmand (1951) ........... 139

7.1 Cyclone design methodology flow chart ................................................................ 148

7.2 Sprea dsheet to calculate initial cyclone dimensions .............................................. 154

7.3 Cyclone solid model geometry ............................................................................... 155

7.4 Preprocessor with completed grid for the 297mm cyclone model ......................... 155

7.5 Convergence history of static pressure during solution process ............................. 157

7.6 The 297mm cyclone particle tracking history with curve fits ................................ 158

7.7 Grade efficiency curve for initial 297mm cyclone design ...................................... 159

7.8 Solid mo del of revise d cyclone geometry ............................................................ 162

7.9 Revised computational grid .................................................................................... 162

23
7.10 .........................................................................................................................

Convergence history of static pressure for revise d cyclone mo del ............................... 164

24
7.11 Collection efficiency vs. time an d size for revise d cyclone design .................... 164
7.12 Grade efficiency curve for revised 297mm cyclone ............................................. 165

25
CHAPTER I.

INTRODUCTION

1.1 Cyclone Separator Overview

A cyclone separator is a device that separates particles from a gas stream using

Axial Exit
Pipe

Cylindrical
Body

Dust Discharge at
bottom of conical
base (apex)

Dust Bin (dust


bunker or
hopper)

centrifugal force. Figure 1.1 shows the basic components of a typical cyclone separator.

Figure 1.1 Cyclone components

1
Gas containing particles of various sizes is brought tangentially into a cylindrically shape d

bo dy. This results in the development of a vortex, which subjects the particles to strong

centrifugal forces. The larger particles move toward the outer wall of the system as a result

of the centrifugal effects. This type of cyclone is also referred to as a reverse flow cyclone

because the flow enters tangentially near the top of the unit, spirals downward along the

outer walls, carrying the larger particles which are deposited in a collection bin under the

Gas that contains small


particles not separated by
Gas and particles of the system leaves through
various sizes enter the exit pipe.
the cyclone
tangentially.
A rotating vortex
develops that separates
dust as a result of the
Smaller, less dense centrifugal forces
particles that are acting on the particles
less effected by the
centrifulgal forces Larger, more dense particles
are able to find migrate to the outer walls
their way toward and spiral downward where
the center of the they fall into the dust bin.
cyclone and be (Solid lines)
carried out of the
system in the
central vortex core
that has a net Separated particles
upward velocity accumulate in bin.
component.
(Dashed Lines)

Figure 1.2 Flow patterns in a cyclone separator

unit.

The flow, reverses as it eventually migrates inward to the axis of the unit, changes

2
direction, flowing upward in a central rotating core and leaves the system at the top of the
cyclone through an exit pipe. The heavier particles are collected in the bin under the unit

3
and smaller particles will escape the system with the exiting gas. Figure 1.2 shows the

basic flow pattern in a cyclone. Figure 1.3 shows the basic dimensions of a cyclone.

4
Cyclone separators are devices that have been aroun d since the late 1800s. They

are used widely in many industrial applications where it is necessary to remove dust or

particles from gases. These devices are simple with no moving parts and are easy to
exit pipe diameter,
De

inlet pipe diameter, distance outlet pipe extends


Di into main body of cyclone, S

main body diameter, main body height, h


D

cone height, (H-h)


where H is the
separated dust exit total height of the
diameter, B main body and
cone

dust bin diameter,


D dust bin height, HB

Figure 1.3 Basic cyclone dimensions

5
maintain. Although the construction of these devices is simple, the physics governing the

flow processes in them is complex. Cyclones have been the subject of much stu dy over the

years yet in practice, most designs are base d on empirical information. There are a

6
number of standard geometries whose performance is well documented and it is common

to use scaling laws to adjust the standard designs to a specific set of operating conditions.

1.2 Literature Review

An important, early study of cyclone separation performance was presented by C. 6

Stairmand in (1951). This paper as well as one published three years earlier (Stairman d,

1949) which discusses pressure drop in cyclone separators has been reference d by

researchers over the years an d up to to day. In his 1951 paper, Stairman d describe d the

characteristics of the vortex that develops in a cyclone, ... which will centrifuge the dust

particles to the walls, whence they can be transported to the dust collecting hopper out of

the influence of the spinning gases.

Tangential Velocity, ut

Figure 1.4 Tangential velocity characteristics from Stairmand (1951)


4
Stairmand indicated that the vortex has characteristics of a free vortex (i.e. The angular

velocity is constant.) in the region outside of a diameter of about one fourth that of the exit

pipe. Inside of about one fourth of the exit pipe diameter the vortex has characteristics of a

forced vortex (i.e. The tangential velocity is proportional to the radius as sketche d in

Figure 1.4.). In addition to the effect of the tangential velocity profiles, there is a general

movement of the gas from the periphery of the system toward its axis. Stairmand referre d

to this as inward drift, which in combination with the spin of the gas creates an inward

spiraling motion of the gas.

The basis for many of the empirical separation efficiency models is the condition in

which a particle of a particular size and density would orbit indefinitely around the axis of

the cyclone. This condition would be satisfied if the drag force on the particle due to the

inward drift was in equilibrium with the outwardly directe d centrifugal force resulting

from the rotation of the particle (Leith, 1984).

5
This force balance acting on a particle is shown in Figure 1.5. Equations 1.1 through 1.3

show the mathematical description of this force balance, which is written in terms of a

polar coordinate system and acts along a radial line passing through the cyclone axis, which

is considere d to be the axis of rotation. The equations are written in a reference frame

rotating with the particle.

dr
F F Fc - Fdr = (1.1)
mp-
c dr p
dt2

where Fc is the centrifugal force, Fdr is the drag force on the particle acting in the radial

direction, mp is the particle mass and d2r/dt2 is the acceleration of the particle in the radial

F=
nd VX (12)
Dr

direction.

where d is the particle diameter, pp is the particle density, utp is the component of

particle velocity in the tangential direction and r is the radial distance of the particle from

the axis of the cyclone.

F
dr = nF
3 d(U
r - U
rp )
(1.3)

where jl is the viscosity of the gas. The quantity urp is the radial velocity component of

the particle itself, which would be zero if the particle was rotating indefinitely at a

constant ra dius, r. The quantity ur is the radial component of the velocity of the gas an d

represents the inward drift talked about by Stairmand (1951).

6
Stairmand (1951) described the collection efficiency curve which indicates the efficiency

of the cyclone for separating particles of a given density over a range of sizes.

7
See Figure 1.6. This curve is also known as the grade efficiency curve. The x-axis plots

particle size, usually in units of microns. The scale of the x-axis is drawn to encompass the

particle size range where the collection efficiency changes from zero to 100%. The y- axis

plots the collection or separation efficiency either as a fraction or as a percent. One point on

the grade efficiency curve which is typically identifie d is the cut size. The cut size is

defined as the particle size at which the separation efficiency is 50%. The cut size is

identified in Figure 1.6. Theoretically, the grade efficiency curve consists of a vertical line

segment at a particle size which is referred to as the theoretical cut size. For particles

smaller than the theoretical cut size, the collection efficiency is zero. For particles larger

than the theoretical cut size the collection efficiency is 100%. The theoretical grade

efficiency curve is shown as a solid line in Figure 1.6. Actual grade efficiency curves

typically have an S-shape as shown in the dashed line of Figure 1.6. At particle sizes

larger than the (actual) cut size, the collection efficiency is lower than the theoretical value

of 100% because some of the larger particles escape the cyclone due to the effects of

turbulent e d dies an d the effects of bouncing, according to Stariman d. At particle sizes

smaller than the (actual) cut size, the collection efficiency is higher than the theoretical

value of zero because some of the smaller particles escape the cyclone as a result of

particle-to-particle collisions. Small particles can also agglomerate to form larger, heavier

groups, which together have enough mass to be retained in the cyclone.

8
Figure 1.6 Diagram of grade efficiency curve characteristics from Stairmand (1951)

Researchers over the years have produce d a number of predictive models that use

empirical information related to the geometry and operating conditions of specific cyclone

designs and are intended to estimate cyclone collection efficiency. Leith (1984) summarize

d a number of these models. His list include d models by Stairmand (1951), Barth (1956),

Lapple and Shepherd (1940), Davies (1952), Lapple (1951) and Leith and Licht (1972).

Ogawa (1984) also reviewed a number of predictive models including Barth (1956), Lapple

and Shepherd (1940) and Stairmand (1951). He also presented his own theoretical

predictions.

Leith and Iozia (1989a; 1989b; 1990) presented what they considered to be the

most accurate empirical model for cyclone separation efficiency and compared its

performance to several other empirical models. Leith, Ramachandran, Dirgo and Feldman

(1991) combine d the empirical mo del for separation efficiency of Leith and

9
Iozia (1989b) with a new empirical mo del for cyclone pressure drop base d on a stu dy of

98 cyclone designs. It is useful to remember, however, what Leith stated in reference

(1984) that, ...while these theoretical calculations of fractional efficiency are useful, they,

... may predict performance substantially in error. Leith (1984) later stated The best way

to determine cyclone fractional efficiency characteristics is to test the cyclone in the

laboratory. The phrase fractional efficiency used by Leith refers to the separation or grade

efficiency curve of the cyclone as discussed in the previous paragraphs and sketched in

Figure 1.6.

Svarovsky (1992) studied a number of commercial cyclones and showed a

relationship between cyclone diameter, flow rate, pressure drop and cut size and the

dimensionless Euler number and Stokes number. As explained earlier, the term cut size

refers to the particle size that is recovered at 50% efficiency. Also see Figure 1.6 above.

The Euler number is considered to be a resistance coefficient that represents the ratio of

static pressure drop between the inlet and outlet of the cyclone to the dynamic pressure of

the flow within the cyclone and is


Ap
defined below as

E=

(14)

Ap
where is the static pressure drop measure d between the inlet an d the gas outlet of a

cyclone, p is gas density and v is the body velocity based on the flow rate and the cross-

section of the cylindrical body of the cyclone as


4Q
_

v=
(15)

10
nD
2

v=
(15)

11
where Q is the gas flow rate and D is the cyclone body inside diameter. The Stokes

number is commonly use d to characterize particle laden flows. It can be thought of as the

relation between the particle response time, Td and the system response time, t, and in

general can be written as


Stk T (1.6)
t
,
T_Pddd2
where T
d^ (1.7)
d
18^,

where p is the particle density, dd is the particle diameter and fic is the viscosity of the gas

and

L
(1.8)
vs

where L, is the characteristic length of the system an d Vs is the characteristic velocity of

the system. The Stokes number written for a particle of size x50 will be define d as

x
>, v
5
Stk 18^D (1.9)
50

where p is the gas viscosity, p, is the solids density, the velocity v is define d by Eqn.1.5

and x50 is the cut size. Svarovsky (1992) found that for the cyclones he studied Eu was

relate d to Stk50 by Eqn. 1.10 below.

12
E
Stk5 (110)
0
Appen dix C shows the cyclone data Svarovsky (1992) use d an d his curve fit.
More recently, numerical approaches to cyclone efficiency estimation have been

propose d. Boysan, Ayers an d Swithenbank (1982) an d Boysan, Ayers, Swithenbank an


d
10
Ewan (1985) presented a somewhat simplified three dimensional method in which they

assumed that the flow field was axisymetric. They recognized that the standard k-e

turbulence models, which assumed isotropy of the Reynolds-stress terms, were not

applicable in the case of highly swirling flows. (Please note that turbulence models are

discussed in more detail in Chapters II and III.) For turbulence closure they used an

algebraic Reynolds-stress model (RSM) to account for the non-isotropy of the flow. Once

a solution for the gas flow field was obtained, they tracked particles through the domain to

estimate the collection efficiency curve. They showed fairly good agreement with the

grade efficiency curves for the Stairmand High Efficiency cyclone and High Flow Rate

cyclone data (Stairmand, 1951) as well as with collection efficiency data for a cyclone that

had three different exit pipe diameters.

Griffiths an d Boysan (1996) use d a Computational Fluid Dynamics (CFD)

computer code to model the collection performance and pressure drop of two small

cyclone aerosol samplers and a Stairmand type cyclone. They compare d the CFD results

with experiments and three empirical theories. Full three-dimensional models of the

geometry were used. A steady-state solution to the continuity and momentum equations

was obtained. The turbulence model used was the RNG-based k-e model originally derive

d by Yakhot an d Orszag (1986) with improvements by Yakhot an d Smith (1992) and

Yakhot, Orszag, Tangham, Gatski and Speziale (1992). RNG stands for renormalization

group theory. It is not as computationally expensive as the Reynolds- stress model but is

an improvement over the standard k-e model for highly swirling flows. The RNG k-e

model includes an additional term in the dissipation rate transport equation which gives

11
improved pre dictions for flows that have high amounts of swirl and stream

12
wise curvature. The empirical model of Barth (1956) predicte d the collection efficiency

curves for the two small cyclones fairly well but did not predict the performance of the

Stairmand cyclone well. The Iozia and Leith (1989a; 1990) theory did a good job of

predicting the performance of the Stairmand design, but not for the two small samplers.

The CFD collection efficiency and pressure drop results compared well with the

experimental data for all three cyclones.

Hoekstra, Derksen and Van Den Akker (1999) compared computational results

using three different turbulence models with laser-doppler velocimetry (LDV)

measurements in a cyclone that had three different exit pipe diameters. The aim of their

study was to evaluate the performance of the standard k-e model, the RNG-k-e model and

an RSM in predicting the gas flow field of a cyclone separator. Their CFD model used the

assumption of incompressible air, steady, axisymetric flow. They compared the tangential,

axial velocity profiles predicted using the three different turbulence models with the LDV

measurements for the experimental cyclone, which was varied by using three exit pipe

diameters. They concluded that both the k-e model and the RNG-k-e model predicte d

unrealistic distributions of axial and tangential velocities, and therefore were unsuitable

for mo deling cyclonic flow. The results using the RSM model were in reasonable

agreement with the experimental data.

Slack, Prasad, Bakker and Boysan (2000) modele d a Stairmand type cyclone in

three dimensions without assuming axisymetric flow. They compared the performance of

a steady-state solution using an RSM turbulence model with the time-averaged results of a

large e d dy simulation (LES) mo del. In the LES mo del, the large turbulent e d dies are

13
computed explicitly while the small scale turbulent features are modeled. The LES

14
approach requires a finer grid than the RSM does. In this case the LES model used

approximately 640,000 computational cells while the RSM model used approximately

40,000 cells. Predictions of the tangential and axial velocity profiles from these two

approaches were compared with LDV measurements. The authors indicated that the

steady-state RSM results on a relatively coarse mesh provided a computationally

inexpensive method to simulate the approximate time-averaged flow details in a cyclone

of the Stairmand type. The computationally more expensive LES mo del was useful in

revealing time dependent vortex oscillations.

1.3 Scope of Work

The literature review in the previous section indicates that the use of empirical

models to predict cyclone performance has limitations. Numerical methods have also been

proposed to model flow fields in these devices. Cyclone flows are threedimensional,

unsteady and have non-isotropic turbulence characteristics because of the high swirl and

stream-wise curvature, whose effects need to be included to provide accurate tangential

and axial velocity profiles. Absent in these studies have been consideration of the effects

due to the influence of the dust bin that is commonly an integral part of the cyclone

system. Particle re-entrainment is an important consideration in the separation

performance of cyclones. This is because particles initially deposite d in the dust bin may

migrate under the influence of secondary flow features to locations

15
where they can eventually be carried out of the system. Re-entrainment can be aggravated

by high inlet velocities as well as the geometry of the dust bin.

The scope the work presente d in this paper is limite d to the stu dy of a 150 mm

cyclone with geometry an d test data presente d by Ogawa (1984). The commercial

computational fluid dynamics (CFD) program, Fluent is used for this study. The analysis

evaluates the effectiveness of using a three- dimensional, unsteady numerical model of the

flow process with a Lagrangian particle tracking method to calculate the collection

efficiency of the cyclone run at different operating conditions. The cyclone model

includes the dust bin geometry so that particle re-entrainment and its effects on collection

efficiency can be reflecte d in the results. Based on the knowledge gained in this study, a

computer-based cyclone separator design methodology will be presented. While pressure

drop is calculate d an d reporte d, it is not within the scope of this project to perform

detailed studies of pressure drop in cyclone separators.

16
CHAPTER
II.
ANALYTICAL MODEL

2.1 Intro duction

The equations needed to model the flow are the continuity equation and the

momentum equations. Figure 2.1 below shows a Cartesian coordinate system in an

inertial reference frame where the equations are applied.

z (x3 or
xk)

y (x/ or
x,) u< (u or
uk)

Ux ->% (u2 or
(u9 or u,)
uO

Figure 2.1 Cyclone coordinate system and velocity component description

17
The flow is treate d as turbulent. Therefore a turbulence model is needed. The

assumption of turbulent flow can be checked by considering the Reynolds number of the

system. The Reynolds number is the ratio of inertia forces to viscous forces in a flow and

pDv
Re = (2.1)

is shown below.

where p is the density of the fluid, D is a characteristic length scale of the system, v is a

characteristic velocity of the system and n is the viscosity of the fluid. The fluid is air. The

density use d for the air in this study is 1.225 kg/m3 and the viscosity is 0.00001789 kg/m-

s. The characteristic length scale for a cyclone is the diameter of the cylindrical bo dy, D.

The diameter, D for the cyclone in this stu dy is 0.150 meters. The characteristic velocity

is often considered to be the velocity in the inlet duct as it enters the cyclone, which is a

measure of the tangential velocity. The three inlet velocities evaluated are 10.3, 15.1 and

19.7 m/s. Svarovsky (1994) used the body velocity which was define d above in eqn. 1.5.

The body velocity is a measure of the axial velocity in the system. The body velocities

corresponding the these three flow rates, using eqn. 1.5 are 1.1, 1.7 and 2.2 m/s

respectively. Using this information the Reynolds numbers were calculated for these

different conditions using eqn. 2.1. The Reynolds numbers based on body velocity ranged

from 12,000 to 22,000. The Reynolds numbers based on inlet velocity varied from

106,000 to 202,000. For flow in a pipe with a sharp-e dged entrance, the Reynolds number

where turbulent flow begins, also call the critical Reynolds number is approximately 2300

(Schlichting, 1979). Based on these results, the assumption of

18
turbulent flow is reasonable.

19
The solid particles which are separated from the air flow in a cyclone represent a

secon d phase that must be mo dele d. For particle-laden flows, the amount of particulate

loading and volume fraction of the particulate phase can be used to determine how best to

model the process (Fluent 6.3 CFD Users Guide [hereafter indicate d as F6.3UG], 2006).

The Stokes number, discussed above, will also provide understanding as to how the

particles will behave. The volume fraction of the particulate phase is defined below as

mp
ap
= ( 2 .
2 ) PP

where mp is the particle concentration in the flow and pp is the particle density. Particle

concentrations typically foun d in cyclone applications can vary from 1.5 grams/m3

(Ogawa, 1984) to 2,000 grams/m3 (Leith, 1984). Using the higher concentration value for

this calculation and the density of the particles use d in this study, which is 2970 kg/m3 the

2000 (g / m3) S 1 (kg)


a
P
0.00 (2.3)
2970 (kg / m3) 1000(g) 1

volume fraction is

The discrete particle tracking scheme used in this study assumes that the particulate

volume fraction is less than about 10 % to 12 %. For this case the volume fraction is about

0.1 %, so the discrete tracking model is applicable. The volume fraction for the gas phase

is

ag = 1 - ap = 1 - 0.001 = 0.999 (2.4)

The particulate loading is defined as the mass density ratio of the dispersed phase (the

20
particles) to that of the carrier phase (the gas) and is written as

21
a
p Pp
fi= (2.5)
a
g P
Using the densities and volume fractions listed above the particulate loading is calculated

as

0.001*2970
= 2.4 (2.6)
0.999*1.225

The material density ratio is defined as

Y= (2.7)
Pg

which is typically greater than 1000 for gas-solid flows, about 1 for liquid-solid flows and

less than 0.001 for gas-liquid flows (F6.3UG). Calculating the material density ratio for

the flow being studied here is

2970
Y= ------- = 2400 (2.8)
1.225
Using the parameters and y it is possible to estimate the average distance between

individual particles in the particulate phase. An estimate of this distance, L compared to a

particle of diameter dp given by Crowe et al. (1998) is

0/
L_ 3
~d (2.9)
6K
~

where K = fiy. For this exercise K = ap = 0.001. Using Eqn. 2.9 the distance to diameter

ratio is

L = (n 1 + 0.001 V/3 = 8 1
dp = 16 0.001 ) = . (2.10)

22
An interparticle space (another name for the L/dp ratio) of 8 in dicates that the particles

can normally be treated as being isolated from each other. This indicates the that the

particulate loading is low. For low loading, the coupling between the phases is

predominantly one-way. The gas (which is the carrier fluid) influences the particles via

drag and turbulence, but the particles have little influence on the gas. A Lagrangian

discrete particle tracking scheme is well suited to model this type of flow.

The Stokes number for the flow being studied will now be considered. Using a cut

diameter of 1.5 microns, which is the value Ogawa measure d for the me dium flow rate

case, the corresponding inlet velocity of 15.1 m/s and the cyclone diameter of 0.150

meters with the already listed material properties, the following value for Stokes number

is obtaine d as

4 PPV (1.5e-6)2 (m2)*2970(kg/m3)*15.1(m/s)


Stk 18 yD 0.002 (2.11)
50 18*0.00001789(kg- / m - s)*0.150(m)

For Stokes numbers, Stk <<1.0, the particle will follow the flow closely. For Stk >1.0 the

particles will move independently of the flow. For this study, the Stokes numbers are

small enough that the particles of interest are expected to closely follow the flow. The

Lagrangian discrete particle tracking scheme can handle flows with large and very small

Stokes numbers. Having gone through this exercise, a better understanding of how the

particles will behave is gained.

From the above considerations, the flow can be modeled as turbulent as the

Reynolds numbers characterizing the flow are well over the critical pipe Reynolds number

of 2300. The Lagrangian discrete particle tracking scheme to be used will be appropriate

to model the solid particle separation process because the particulate loading

23
is low an d the particle volume fraction is much less than 10 %. The particles are expecte

d to follow the flow closely. The next sections look at the details of the flow and particle

tracking calculations.

2.2 Flow Governing Equations

The continuity equation, assuming incompressible flow and with no mass source

terms, is shown in a Cartesian coordinate system form below as equation 2.12 (Tannehill,

Anderson andPletcher, 1997; Wilcox, 1994).

d ui
=0
d x. (2.12)

In equation 2.12, u, is the velocity component in the /-coordinate direction and x, is the

distance in the /-direction where / = 1, 2, 3 in the case of three dimensional problems. The

momentum equation for incompressible flow in a non-accelerating reference frame may be

written as equation 2.13 below from Wilcox (1994).

du. du. dp + d ti
P J
+ pu. dt y
dx. dx, dx. (2.13)

In equation 2.13, p is pressure, p is density, w is velocity in the i-th direction, x, and x. are

spatial coordinates in the i-th an d j-th directions respectively, t is time an d t,. is the

viscous stress tensor define d by

^ = 2Psy (2.14)

where p is the molecular viscosity and Sy is the strain-rate tensor define d by

24
1 ( dut dUj'
S
v/ V dxs + dX y
(2.15)
2.3 Turbulence

Turbulence is discusse d in the following section. A brief overview is presented,

which describes three approaches for handling turbulence. The turbulence model chosen

for this work as well as the reasons for its selection will also be discussed.

2.3.1 Intro ducti on to Turbul ence Mo del s

6 O. Hinze (1975) describe d turbulence as follows: Turbulent fluid motion is an

irregular condition of flow in which various quantities show a random variation with time

and space coordinates so that statistically distinct average values can be discerned.

Wilcox (1994) explains that turbulence consists of a continuous spectrum of scales that

vary from smallest to largest over several orders of magnitude. The idea of a series of

turbulent e ddies is often use d. Wilcox (1994) states that, A turbulent e d dy can be

thought of as a local swirling motion whose characteristic dimension is the local

turbulence scale. He explains that these eddies overlap in space and that the large ones

carry the smaller ones. The conversion of energy in a turbulent flow follows a cascading

process where the kinetic energy is transferred from the larger eddies to the smaller ones.

The energy in the smallest eddies is finally dissipated in heat. For illustration purposes

Figure 2.2 shows a flow visualization of a turbulent boundary layer from a group of

25
photos assembled by Van Dyke (1982). The photo shows streak lines pro duce d in a

turbulent boun dary layer on a flat plate by a smoke wire. The photograph is by Thomas

Corke, Y. Guezennec and Hassan Nagib (Van Dyke. 1982). The presence of turbulent

eddies as well as the intermittent nature of the outer part of the boundary layer can be

seen. If, for example, a hot wire anemometer were used to measure the instantaneous

velocity of the flow across the boundary layer, one would expect to see a velocity profile

similar to the one that is sketche d in re d in Figure 2.2.

Figure 2.2 Turbulent bound ary layer developing on a flat plate

Reynolds (1895) introduce d a proce dure by which, all instantaneous quantities

may be expressed as the sum of mean and fluctuating components. The instantaneous

velocity profile sketche d on the photo in Figure 2.2, could be represente d as a

combination of a mean velocity component and a fluctuating velocity component as

shown below in Figure 2.3. In general, an instantaneous quantity, w, can be written as


22
u
, = Ut + u' (2.16)

where U, is the mean component of u, and u' is the fluctuating component of u,. Then

using this expression for the instantaneous velocity, the time-averaging of the continuity

equation (2.12) and the momentum equation (2.13) may be carrie d out.

To make the time-averaging process simpler, Wilcox (1994) shows that the

convective term of the momentum equation, which is the second term on the left hand side

of equation 2.13, can be rewritten in conservation form as follows.

du. d d
u.1 dx du; (u u )
, 1
.j
dx 1 1l
j
1
(u ,u,) - u dx, dx, ' (2.17)

23
The term (u.du, /dx;. ) in equation 2.17 above is zero because of the continuity equation

2.12. Using the relationship in equation 2.17 and combining equations 2.13 and 2.14

results in the Navier-Stokes equation in conservative form shown below (Wilcox, 1994).
du1 d dp d
P dt
dx dx, dx. (2.18)

Wilcox (1994) as well as Tannehill et al. (1997) show the steps in the Reynolds averaging

process carrie d out on equations 2.18 an d 2.13 that result in the Reynolds time average d

equations of motion shown below.

dul
=
dxi 0
(2.19)

pPi_+pL (u u + u ) = -+(2ps .)
dt dx, )
(; ;
dx dx, ( ^ p)
(2.20)

where the upper case symbols U , U,, P, and S,. represent mean values of the velocity

components in the i-th an d j-th directions, pressure an d strain-rate tensor respectively an

d
the symbols u'; an d u' represent the fluctuating values of velocity in the i-th an d j-th

directions respectively. The appearance of the term u'u' represents the difference in the

time average d and the instantaneous momentum equations other than the replacement of

the instantaneous velocity, pressure and strain-rate values with mean values. The term

u'u' is a statistical correlation that resulted from the time-averaging method which, in

general, is not equal to zero and which represents the mean value of the product of

velocity fluctuations which, are related to turbulent effects in the flow. The purpose of a
turbulence model is to provide a prescription for computing u'u' . When a method for

24
computing ujuj is available then it is possible to solve the continuity equation 2.19 an d

the momentum equation 2.20 for the mean quantities of the turbulent flow under

consideration. Equation 2.20 can be rearranged, again using the relationships shown in

equation 2.18, but in reverse, to write what is calle d the Reynolds-average d Navier-

Stokes equation in its most recognizable form as


dU, TT dU, dP d
P~ + pU}1- ----- 1--- =2 US , - pu .Uj )
dt 3x
, dx dx, = ^p 11 > =2.21
)
where the quantity - pu.jujj is known as the Reynolds-stress tensor =Wilcox, 1994) also

denoted as

_//
T
v = -pu, ui =2.22)

By inspection t1}. = Tjt which means that the Reynol ds-stress tensor is a symmetric

tensor. A symmetric tensor has six independent components.

There have been a number of approaches to modeling the Reynolds-stress tensor

shown in equation 2.22 above. Tannehill et al. =1997) divide the approaches into three

types of models, which vary in their level of complexity and also the level of

computational power needed to solve the resulting turbulence equations. These three

general types of models are described briefly in the following sections.

2.3.2 Type I: E d dy Viscosity Mo dels Using the Boussinesq Assumption

Boussinesq =1877) suggested that the apparent stresses in turbulent flows might be relate

d to the strain rate through a scalar turbulent eddy viscosity. Following the

25
Boussinesq assumption, the Reynolds-stress tensor can be written for an incompressible

fluid as
pu]u' = 2^tSt ij (2.23)

where lt is the turbulent viscosity and the mean strain-rate tensor, S. define d as

S, = 9 1 . dx. dxi
1
2V J
(2.24)

Using an analogy with kinetic theory of gases and the way in which the molecular

viscosity for gases can be evaluated, one might expect that the turbulent eddy viscosity

could be proportional to a characteristic eddy velocity and a characteristic eddy size or

length scale (Tannehill et al., 1997; Wilcox, 1994) . Turbulent viscosity could therefore be

modele d as

l = Pvth (2.25)

where vt is the characteristic velocity scale for the turbulence and lt is the characteristic

length scale for the turbulence. The function of a turbulence model that uses the

Boussinesq assumption is to prescribe appropriate values for vt and lt.

One of the simplest turbulence models that uses the Boussinesq assumption was

suggeste d by Pran dtl in the 1920s when he presente d the idea of a turbulent mixing

length, where lmix could be thought of as the transverse distance over which the fluid

particles could retain their original momentum. The characteristic velocity, vmix could then

be relate d to lmix as:

du
= l mi d
x y
(2.26)
26
Substituting equation 2.26 into 2.25 results in the mixing length mo del:

du
Mt = d
P$ y (2.27)

The mixing length model assumes that the turbulent viscosity is a scalar and has been use

d successfully for a number of applications. However, the evaluation of lmix varies with the

type of flow being considered. One example of good results for this model is for flow very

near a solid wall where:

1
mix = *y (2.28)

where y is the distance from the wall and & is known as the von Karman constant, which

has been foun d from experimental results to be approximately equal to 0.41. The

importance of the von Karman constant is that it is universal in nature. This constant is

found in the work of both von Karman and Prandtl in the derivation of a universal velocity

law for turbulent channel flow. This universal velocity distribution law can be use d to

closely calculate the observed velocity profiles for turbulent flow in ducts, except for

regions very close to the wall and very close to the duct centerline. See Schlichting (1979).

While there are a number of one equation as well as two equation models that use

the Boussinesq assumption, perhaps the best known turbulence model of this type is the

two equation k ~ model. Prandtl, in 1945, presented the notion of computing the

turbulence characteristic velocity scale so that one would not need to assume that
vmix = lmix |3u/3y| (equation 2.26 above). He use d the kinetic energy per unit mass of the

turbulence, k as the basis for his velocity scale where

27
1
i / / 1 /- /2 . >2.
>2\ (2.29)
k = uu.= (u + v + w ) 2 2
There are still two unknown quantities (the turbulent characteristic length and velocity

scales) that must be resolved and in two equations models, as the name implies, two

transport equations are used. A turbulence kinetic energy transport equation is used for the

velocity scale, which is k from equation 2.29. This transport equation usually assumes the

following form (Wilcox, 1994):

dk dk dUt d ^ p \ dk
t
P + pu,------- u + ------
dt dx, Ti; - - p +
dx dx O dx (2.30)
. , x
k Jo j

where is the dissipation rate of turbulent kinetic energy per unit mass and is commonly

modeled as

de de d EL de
P --- + PU, ------ = C1 T, (
2 P^~ + P O
dt 1 dx, 1
k y dx. k 3x, + dx (2.31)
JXj

where the Prandtl number for k, ok = 1.0 . The Prandtl number for is O = 1.3 . The

values for the two closure coefficients, C1 and Ce2 are C1 = 1.44, andCe2 = 1.92

respectively. The values for these coefficients and also for the two Prandtl numbers are

from Launder and Spalding (1972). The turbulent viscosity, pt is related to k and by

PCMk2
(2.32)

c 0 09
where the value of the coefficient, E = , which is also from Launder and Spalding

(1972). The turbulent length scale, I is related to k and e by

C^k 2
I (/.33)
=

28
2.3.3 Type II: Models Base d on the Reynolds-Stress Equation

The Reynolds-averaging process described above resulted in additional unknowns

due to the appearance of the Reynolds-stress tensor. In order to close the system, ad

ditional equations are nee de d. Wilcox (1994) shows how a differential equation for the

Reynolds-stress tensor can be derived by taking moments of the Navier-Stokes equation.

To do this the Navier-Stokes equations are multiplied by a fluctuating property and then

the product is time averaged. Wilcox (1994) shows this process and on a term-by-term

basis and shows the resulting differential equation, known as the Reynolds-stress

equation, which is written below as

dr. dr. dU dU duj du'


- + Uk = ~tlk --------- L -Tk -------- L + 2u ---- ...
dt dxk dxk dxk dxk dxk

, dp' , dp' d dT.


. + u,dx+ u , + dx.
- V dx, + Pu,u uk
rr

dx. - j
(2.34)
,
Equation 2.34 describes the transport of terms in the Reynolds-stress tensor and as such it

generates six new equations that can be used to solve for the six independent component

terms of the tensor. However, in the derivation process, twenty-two new unknowns have

also been generated. To use the Reynolds-stress equation, in practice or in other words to

close the system in practice, requires that modeling be done to reduce the number of

unknowns so that the number of unknowns is equal to the number of available equations.

The terms in equation 2.34 can be rearranged into the following.

29
dT dU, dU,
dT = -T--T, du, dU: /duf t du ^
-- '- + Uk- L
k ------------- + 2u! ----------- - - p
i+J
dt dx, dx, dx, dx, dx
Vdx; dx
iJ

T dr /// //c / / c l
+ d V dx. + ^[P U,U iUk + P U S
jk + P U
dx - dx, J
i Sik (2.35)
,, ,
where the kinematic viscosity, V is define d as V = p/p. The term Sjk is the Kronecker

delta which is defined as

f 1 (if i = j)
Sr (2 36)
' |o(i/i * ,) '
To simplify the terms in equation 2.35 the following expressions are defined:

c=U
C
= Uk dx, (2-3E)

where Q represents the transport of Reynolds stresses due to convection.

p = dUJ dU, 11Tk dxk Tk


(2.38)
dxk

where P y represents the transport of Reynolds stresses due to stress


production.
_ dU , d n

ii = 2p dx 1 dx 1 (2.39)
k k

where % represents the transport of Reynolds stresses due to dissipation.

A / du' + du'j ^ fa =
p L + -
B
I dx; dxi J (2.40)

where represents the transport of Reynolds stresses due to pressure strain.

d T
D L,i dxk
dx (2.41)
j
k

30
where DL!j represents the transport of Reynolds stresses due to molecular diffusion.
D
T, y = \PUV 'jU * + P'U'< 5jk + PU'j 5r k ] (2.42)

where DT!- represents the transport of Reynolds stresses due to turbulent diffusion.

Substituting the above expressions from equations 2.37 to 2.42 into equation 2.35 the

Reynolds-stress equation can be written in the following form:


dr,.,.
dt (2.43
)
y
+ C- = P- + + D, .. + D* ..
T
This model and the way y V . 'V L.ij ,.j in which it is implemented

for practical use is discussed in Chapter 3, section 3.3.

2.3.4 Type III: Models Not Based Completely on the Reynolds-Stress Equation

A model that falls into this category is Direct Numerical Simulation (DNS), which

is a complete time-dependent solution of the Navier-Stokes and continuity equations. No

modeling assumptions are required but the computational requirements make this model

unpractical for engineering calculations at the present time. Another example of a type III

turbulence model is the Large Eddy Simulation (LES). In the LES model, the evolution of

the larger eddies is calculated directly. The effects of the small eddies are modeled.

Simulations involving type III models are more computationally expensive than the type I

and II models and were not considered for this study. The interested reader can consult

Tannehill et al. (1997) as well as Wilcox (1994) for more


information on these models.

31
2.3.5 The Turbulence Mo del Use d for this Work

Previous researchers (Boysan et al. 1982; Ayers et al. 1985; Hoekstra et al., 1999;

& Slack et al., 2000) have shown that a Reynolds-stress model (RSM) is able to calculate

the turbulence characteristics of cyclone flows to a reasonable degree. These researchers

have also shown that models using the eddy viscosity assumption are not adequate to

model the flow characteristics of cyclones. While LES simulations were shown to be an

improvement over the RSM models (Slack, Prasad et al., 2000) the computational

expense was considered too high for the purposes of this study. Therefore the RSM mo

del has been chosen for this work. Section 3.3 describes the way RSM mo del is

implemented numerically.

2.4 Particle Dynamics Motion Equations

The equation of motion for a single particle written in the x-coordinate direction

in Cartesian coordinates and on a per unit of particle mass basis is

(p
^ = FD (ug - up ) + ~Pg > (2.44)
dt Pp

where the subscriptp refers to the particle and the subscript g refers to the gas. The term
FD (ug - up ) is the drag force per unit particle mass. The second term on the right-hand

32
side of eqn.2.44 is the gravitational force on the particle (per unit particle mass), where gx

is the acceleration of gravity. The coefficient, FD is define d below as equation 2.45.


F _ 18 p CD Re

p dp 24
p (2.45)

In the two equations above, u is the fluid velocity, up is the particle velocity, p is the

molecular viscosity of the fluid, p is the fluid density, pp is the density of the particle and

dp is the particle diameter. A particle Reynolds number, Re for a spherical particle is

defined as
pdp u - up
Re _ ----- ! ------- 1
F (2.46)

where + is the absolute viscosity of the gas.

At small Reynolds numbers of Re = 0.1, the flow is known as Stokes flow and the

24
CD
Re (2.47)

drag coefficient, CD is

At Reynolds numbers above Re = 103the value of CD is approximately constant at about

0.4. In Reynolds number ranges between these extremes the value for CD varies in a

complex manner. Morsi an d Alexander (September 26, 1972) develope d functions which

describe the variation of CD for the regions where CD varies in a complex manner. These

functions are use d in the numerical particle tracking model used in this study. This is

discussed more in Chapter III.

33
2.5 Chapter Summary

The basic equations that are required to solve the flow field in a cyclone separator

were presented. The flow is turbulent and creates a problem of closure with the

momentum equation. Osborne Reynolds (1895) presented an averaging method that when

applied to the continuity and momentum equations allows turbulence effects to be treated

in a statistical fashion. The averaging process results in the appearance of the Reynolds-

stress tensor. In general, some type of modeling for turbulence is require d to close the

governing equations. There are three classifications of turbulence models. The first type

assumes that turbulence influences the flow by effectively increasing its viscosity. The

turbulent e ddy viscosity is a function of a characteristic velocity related term, the

turbulent kinetic energy, k, and a characteristic length scale that can be a function of k and

e. The second type of model is based on the use of transport equations for each of the six

component terms of the Reynolds-stress tensor. This model accounts for more of the real

effects observed in turbulent flows but also contains terms with additional unknowns.

Modeling needs to be used to reduce the number of new unknowns so that closure is

obtained. The third type of model is one in which all or part of the turbulent effects are

solved for directly. These approaches require significantly more computational effort and

at this time and are not considered practical for many engineering calculations. The

turbulence model used for this work is from the second type, in which, a modeled form of

the Reynolds-stress equation is solved. To calculate the particle separation efficiency of a

34
cyclone requires the ability to track the paths of

35
2.5 Chapter Summary

particles that are introduced into the flow. To provide for this, the equation governing the

motion of a discrete particle was introduced.

36
CHAPTER III.
THE NUMERICAL MODEL AS IMPLEMENTED IN FLUENT1

3.1 Intro duction

The commercial CFD computer program, Fluent is used to solve the governing

equations discussed in Chapter II for the cyclone flow field in this study. The

implementation of the governing equations and the way in which Fluent solves the

governing equations are outlined in this section.

3.2 Implementation of the Flow Governing Equations

The continuity equation 2.19 is shown again below. It is unchange d from the
previous section.

auL
=0
3x,
(2.19)

The momentum equation in the form of the Reynolds-averaged Navier-Stokes equation is

also unchange d from its form presente d in the previous section and is repeated below.

1
Fluent is a commercially available computational fluid dynamics (CFD)
computer program.
37
(2.21)
dU, dU, dP d

p- + pU,- = ------------ +---- (2uS,, pu ,u,)
yJ
dt dx, dx, dx,. ^ '
J J
3.3 Choice of Turbulence Mo del an d its Implementation

As indicated in the previous section, a Reynolds-stress model (RSM) (Launder,

1989a; Launder, 1989a; Gibson and Launder, 1978; Launder, Reece and Rodi, 1975) was

used in the present study and the way it is implemente d in Fluent is described in this

section. For the RSM model the Reynolds-stress equation is used as the starting point and

is shown below as it was presented in the previous section.

at
+ C. = P.. +.. 6- + DT .. + D*..
T, (2.43)
V . . 'V V *

The Reynolds-stress equation does provide six additional equations, one for each of the

six independent components of the Reynolds-stress tensor. In the process, however, the

derivation of the Reynolds-stress equation introduced an additional twenty-two new

unknown quantities. These new unknowns in equation 2.43 appear in the terms ., 6 and

D*! The expressions for these terms were presented in the previous section and are

repeated below.

du' du';
, = 2u !
dxk dxk (2.39)

6' du' + du) ^


= P L + -
B
I dx, ax, J (2.40)
D
*,ij =~^~ [Pu'u.',u'k + P 'u' 38
S
-k + P u'- S,k
] ox.
(2.42)
In the implementation of the Reynolds-stress model in Fluent, the three terms listed above

need to be modele d. The following sections describe the models used in Fluent for these

terms. A number of empirical constants are used in the following sections. Table

3.1 is presente d below an d lists the coefficients, their values, the equations in which they

first appear and the reference describing the determination of the value used.

Table 3.1 Turbulence model coefficient values and source references


Equation of First
Symbol Value Reference
Appearance
0.82 (3.1) (Lien & Leschziner,
0k 1994)
(Launder &
0.09 (3.2)
Spalding, 1972)
(Fluent 6.3 CFD
1.0 (3.4) Users Guide,
2006)
( (Launder &
\ 1.44 (3.4)
Spalding, 1972)
(
l 1.92 (3.4) (Launder &
Spalding, 1972)
(Fluent 6.3 CFD
K 0.4187 (3.7) Users Guide,
2006)

Beginning with equation 2.42 DT!j is modele d in Fluent from Lien and Leschziner

D*. =
T,IJ
3xt 3/ /
J
<7,,d x,, (3.1)

(1994) as equation 3.1 shown below (Fluent 6.3 CFD Users Guide[F6.3UG], 2006):

Lien and Leschziner (1994) derived a value of ok = 0.82 and the turbulent viscosity,

is computed using equation 3.2 shown below where the constant, CM= 0.09.

39
r k2 H =PCU~

(3.8)
where kis the turbulence kinetic energy an d is obtaine d from equation 3.3 below.

k = U:U:

2 ll (3.3)

In equation 3.2 the scalar dissipation rate, , is compute d with a transport equation

shown below (F6.3UG, 2006):


d TT d 1 d d
p
- + u , = ClPh ----- C2 p + - U
;
dx, 1 11
2k 2
k dx, + dx
J XJ (3.4)

In equation (3.4), <J = 1.0, C1 = 1.44, and C2 = 1.92 . Pu is from equation 2.38.

The pressure-strain term, fa- shown in equation 2.40 is modeled based on

proposals by Gibson and Launder (1978), Fu, Launder and Leschziner (1987) and Launder

(1989a; 1989b) in a decomposed form as follows (F6.3UG, 2006).

fa- =fa ,1 +01J ,2 + (3.5)


fav
where


fa,1 = -18
PT U'U fak
1
J 3 lJ (3.6)

(
fa- ,2 = -060 Py - C)-f fay I 1Pk - 2Ckk
(3.7)

where the terms Plj and Pkk are defined as in equation 2.38 and the terms Q and Ckk are

defined as in equation 2.37. The term, fa,w is modeled using equation 3.8 below (F6.3UG,

2006).
40

. r , c (-^-y- x 3-r-r l~r-r ^ Clk/l
K,* = k I UkUnknAj- /UUnjnk - -u 1uk##k
05
(3.2)
. xV
/2
e 3 3 ^ C,k
+ 0 3 I K 2 ^^i, --$k,2 n;#k --K;k,2 #nk

where nk is the xk component of the unit normal to the wall, d is the normal distance to the

wall and C$ = C^ /fc, where CH=0.09 and & is the von Karman constant, which in this

case is given a value of 0.4187. The dissipation tensor, , shown as equation 2.36 is

2
n = -4 (P + )
(3.9)

modeled as shown below (F6.3UG, 2006).

In equation 3.9 above, = 2p, where a is the speed of sound. The scalar

dissipation rate, e, is computed with a model transport equation. See equation 3.4 above.

To summarize, the Reynolds-stress turbulence model implemented in Fluent uses

six transport equations to solve for the six unique terms in the Reynolds-stress tensor. The

other terms in the Reynolds-stress equation that contain additional unknowns are modeled

as functions of the six Reynolds-stress terms with the addition of a transport equation for

the scalar dissipation rate, e. By this approach the original Reynolds-stress equation with a

total of twenty-eight unknowns is reduced to a system that involves seven unique unknown

terms and is solved using a total of seven transport equations.

41
3.4 Two Phase Flow: Air an d Solid Particle Mixture
The modeling of a cyclone separator involves two distinct phases. The gas flow

must be modele d. The paths of the solid particles which are transported by the gas must

also be calculate d.

3.4.1 Basic Particle Trajectory Calculations

As was discusse d in Chapter II, the trajectory of a discrete particle is pre dicte d

by integrating the force balance on the particle and is written in a Lagrangian reference

frame. The force balance equates the particle inertia with the forces acting on the particle

and is shown below for the x-direction in Cartesian coordinates as shown below in

equation 3.10 (F6.3UG, 2006).

dup P -P>
(
FD (U - uv)
dt + p
p (3.10)

The first term on the right han d side of equation 3.10 is the drag force per unit particle

mass where the coefficient, FD is define d below as equation 3.11.

18^ CD Re
FD
, PX 24
(3.11)

In the two equations above, u is the fluid velocity, UP is the particle velocity, p is the

molecular viscosity of the fluid, p is the flui d density, pp is the density of the particle and

42
dp is the particle diameter. Re is the relative Reynolds number, which is shown below as

equation 3.12.

Pdp\up -u
Re = ----- ! ------ 1

P (3.12)

The drag coefficient, CD, in equation 3.11 is for smooth spherical particles an d is shown

below as equation 3.13.

CD = " + +3
Re Re2 (3.13)

In equation 3.13 the constants aj, a2 and a3 are for smooth spherical particles over several

ranges of Re ( Morsi & Alexander, 1972; F6.3UG, 2006).

3.4.2 The Turbulent Dispersion of Particles

Fluent provides a means to account for the dispersion of particles due to turbulence

using a stochastic tracking model. The stochastic model, also referred to as the random

walk model, includes the effect of instantaneous turbulent velocity fluctuations on the

particle trajectories. The basic trajectory calculations described above in section 3.4.1 will

use the mean fluid phase velocity, u , by default. Optionally the instantaneous value of the

fluctuating gas flow velocity, u(( can be used where

u(t) = U + u '(t) (314)

to predict the dispersion of the particles due to turbulence. The term, u'(t), in equation 3.14

43
is the fluctuating component of velocity. In the stochastic tracking approach, the

44
turbulent dispersion of particles is pre dicte d by integrating the trajectory equations for

individual particles using the instantaneous fluid velocity, u(t), along the particle path.

By computing the trajectory in this manner for a sufficient number of representative

particles (termed the number of tries), the random effects of turbulence on the particle

dispersion is accounted for. The model as implemente d in Fluent is calle d the discrete

random walk (DRW) model where the fluctuating velocity components are discrete

piecewise constant functions of time. Their random value is kept constant over an interval

of time given by the characteristic lifetime of the turbulent eddies. This prediction of

particle dispersion uses the concept of the integral time scale, *, which describes the time

spent in turbulent motion along the particle path, ds where

(3.15)

The integral time is proportional to the particle dispersion rate, where larger values

indicate more turbulent motion in the flow. It can be shown that the particle diffusivity is
given by u'u'T (F6.3UG, 2006). For small tracer particles that move with the fluid,
*3

the integral time becomes the fluid Lagrangian integral time, * L. This time scale can be

approximated as

*L = C -
(3.16)

where CL is to be determined as it is not well known. By matching the diffusivity of the

tracer particles, u'u'*L, to the scalar diffusion rate predicted by the turbulence model, Htj(p

o), one can obtain

45
k
Tr 0.3 -

(3.17)

where the coefficient 0.3 is appropriate to use with the RSM turbulence model (F6.3UG,

2006; Daly & Harlow, 1970). In the discrete ran dom walk (DRW) mo del the interaction

of a particle with a succession of discrete stylized fluid phase turbulent eddies is simulated.

Each e ddy is characterized by

o a Gaussian distributed random velocity fluctuation, u', v', and w'


o a time scale, ?e

The values of u', v', and w' that prevail during the lifetime of the turbulent eddy are

sampled by assuming that they obey a Gaussian probability distribution, so that


/ ^ /2

u
= gu (3.18)

where g is a normally distributed random number and the remainder of the right-hand side

of equation 3.18 is the local root-mean-square (RMS) value of the velocity fluctuations.

Because the RSM is use d, the nonisotropy of the stresses is included in the derivation of

the velocity fluctuations so that equation 3.18 is valid calculating u'. The other fluctuating

components are define d by


/ ^ /2
v
=g v (3.19)

and

w
= g' (3.20)

See references (F6.3UG, 2006; Zhou & Leschziner, 1991).

The characteristic lifetime of the eddy is defined either as a constant by

46
T=2* (3.21)

where TL is given by equation 3.17 or as a ran dom variation about TL by the expression

T = -TL log(^> (3.22)

where r is a ran dom number between 0 an d 1 an d TL is given by equation 3.17. The

option of the random calculation of T yields a more realistic description of the

correlation function. The random calculation of T is the option used in the present study.

When the eddy life time has been reached, a new value of the instantaneous velocity is

obtaine d by applying a new value of g in equations 3.18 through 3.20, (F6.3UG, 2006).

3.5 The Fluent Solver and Solution Methodology

An overview of the approach used by Fluent to solve the governing equations is

presente d in the following section.

3.5.1 Overview of Control Volume Method

Fluent uses a control-volume based method to integrate the governing equations

about each computational volume or cell, creating discrete algebraic equations that

conserve each quantity on a control-volume basis. An integral conservation equation for

47
the transport of a scalar quantity, $ and for an arbitrary control-volume, V can be written

f
J
V dt dV + fs p$v dA = fs r,V$-dt + jvSfdV
(3.15)

as

where p is the density, V is the velocity vector (= ui + v j in 2D), A is the surface area

vector, r$ is the diffusion coefficient for $, V$ is the gradient of $


( V$ = (d$/3x)i + (d$/dy)j in 2D) and S$ is the source of $ per unit volume.

Figure 3.1 shows several control volumes created by the division of the domain

using a computational grid. Applying equation 3.15 to each control volume or cell, one can

p N N
v
+1 p$fvr a=I rv$f a+S$v
,
dt
(3.16)

obtain a discrete form of the transport equation for $:

where N is the number of faces enclosing a given cell, $f is the value of $ convected

through face f pvf Af is the mass flux through the face, Af is the area of the face,
V
$f is the gradient of $ at face f and V is the cell volume. Discrete values of the

quantity $ are stored at the cell centers. Values of $ at the cell faces, $f are needed to

compute the convective term in equation 3.16.

48
face,/

cell
center
displacement
vectors
Figure 3.1 Control volumes use d to illustrate the transport of quantity $

This is accomplished using an appropriate discretization scheme. The discretized scalar

transport equation (3.16) contains the unknown scalar variable $ at the cell center as well

as the unknown values in surrounding neighbor cells. This equation will, in general, be

non-linear with respect to these variables. A linearized form of equation (3.16) can be

written as

"p $^ anb$nb + 6
nb (3.17)
where the subscript nb refers to neighbor cells,

and aP and anb are the linearized coefficients for $ an d $nb an d b is a source term

(Patankar, 1980). The number of neighbors for each cell depends on the grid topology, but

will typically equal the number of faces enclosing the cell (boundary cells being the

exception). Similar equations can be written for each cell in the grid. This results in a set

of algebraic equations with a sparse

49
coefficient matrix. For scalar equations, Fluent solves this linear system using a point

implicit (Gauss-Seidel) linear equation solver in conjunction with an algebraic multigrid

(AMG) method.

3.5.2 Choice of Equation Discretization Schemes

For cyclone numerical modeling it is recommended that higher order

discretization schemes be use d (Thematic Network for Quality and Trust in the Industrial

Application of CFD [QNET-CFD], 2003; Slack & Harlow, 2000). For this reason, the

highest order schemes available in Fluent were chosen for this study. Diffusion terms in

the governing equations are central difference d an d are secon d order accurate. For the

convection terms, a third order scheme based on the Monotone Upstream-Centered

Schemes for Conservation Laws (MUSCL) (Van Leer, 1979) was chosen. The MUSCL

scheme achieves third order accuracy by a blending of a central differencing scheme and a

second order upwind scheme and is available for all cell types (F6.3UG, 2006). The third-

order scheme is shown below as


<pf = d<f)f ,CD + (1 - ff)(pf sou (3.18)

where </>f,CD is a central difference scheme define d by equation (3.19) an d (/>f,sou is a

secon d-order upwin d scheme define d by equation (3.20) below an d 0 is a coefficient

use d to determine the degree of blending of the central difference and second-order

upwind schemes and usually takes a value of 1/8. The central difference scheme is
shown below as

50
fi
f ,CD /(& + fi)
+ 2 (Vfi 0 ' G + Vfi
1 r)
(3.19)

where the indices 0 and 1 refer to the cells that share face f V fi0 and Vfi are the

gradients at cells 0 and 1 respectively, and r is the displacement vector directed from the

cell centroid to the face centroid. Also see Figure 3.1 above for a sketch of the cell

relationships. The second-order upwind scheme is shown below as


0f,sou =0 + V 0 - r (320)

where f i and V f i are the cell-centered value and its gradient in the upstream cell and r is

the displacement vector from the upstream cell centroid to the face centroid. This

formulation requires the calculation of the gradient V f i in each cell. The determination of

the gradient of the scalar f i at the cell center c0 is computed using the Green-Gauss

theorem and is written in discrete form as

(3.21)
(Vfi) C0 =V^rA,
v
f

where fif is the value of fi at the cell face centroid and the summation is over all faces

enclosing the cell and v is the number of faces surrounding the cell. A node-based
gradient evaluation is used to compute , which is the arithmetic average of the nodal

values of (/> on the face shown below as

1 Ni
# (3.22
)

where Nf is the number of nodes on the face. The nodal values, in equation (3.22) are

51
constructed from the weighted average of the cell values surrounding the nodes. The

52
scheme reconstructs the exact values of the linear function at a node from surrounding

cell-centered values on arbitrary unstructured meshes by solving a constrained

minimization problem and preserves a second-order spatial accuracy (F6.3UG, 2006).

Second-order discretization in time derivative term is chosen for this study.

Temporal discretization involves the integration of every term in the differential equations

over a time step At. A generic expression for the time evolution of a variable 0 is given by

d0
=F
~d (0) (3.23)
t

30+1 - 40 + 0-1
= F (0)
2At (3.24)

where the function F incorporates any spatial discretization. The second-order accurate

discretization scheme used for the first order time derivative is used in this study which

given by

where 0 is a scalar quantity, n+1 is the value of 0 an d the next time level, t + At an d n is

the value of 0 and the current time level, t and n-1 is the value of 0 and the previous time

level, t -At (F6.3UG, 2006).

3.5.3 Solving the Equations

The discrete governing equations are linearized to form a system of equations for

the depen dent variables in each computational cell. The resulting system of equations are
solve d using a segregated method. The segregated method solves for a single variable at
50
a time such as u-velocity considering all cells in the problem. It then solves for the next

variable in turn, say y-velocity component, etc., until all variables are up date d.

Figure 3.2 General segregate d solver solution process steps

An implicit formulation of the equations is use d meaning that each unknown will appear

in more than one equation in the system of equations and that the equations for a given

unknown must be solved simultaneously. Because the equations are non-linear and

coupled, a number of iterations must be performed before a converged solution is reached

for each time step. Each time iteration consists of the steps shown in Figure 3.2.

51
3.5.4 Convergence Criteria

At the end of each solver iteration, the residual sum for each of the conserved

variables is computed and stored. The calculation of the residuals is described here. After

discretization the linearized equation for the general variable $ at cell P was presente d

above as equation (3.17) an d is repeate d below.

"p $ = Y "nb$nb + 6
nb (3.17)

If equation (3.17) is summe d over all the

computational cells P in the domain, there will be an imbalance in the left-hand and right-

hand sides of the summed result. The imbalance in the expression for the sum with respect

to the general variable $ over all of the P cells is termed the residual, R$. This residual, also

R$ = Y Y "nb $nb + 6"p $P


unscaled n (3.25)
Cells P b

called the unscaled residual, RLcaied and is define d below as

where the various terms are the same as define d for equation (3.17) earlier an d where $P

is the value of the general variable $ for cell P. In general, it is difficult to judge

convergence by examining the residuals define d by equation (3.25) since no scaling is use

52
(3.26)

d. Fluent scales the residual using a scaling factor representative of the flow rate of $

through the domain. This scale d residual is define d as

53
3.5.4 Convergence Criteria

S S
a
nb*nb + 6 "p
R* = Cel *P
nb
cells
ls P P S\"p *

For the momentum equations, the term *P is replaced by vP .which is the magnitude of the

particular velocity component at cell P. For the continuity equation, the unscale d residual

for the solver used in this study is defined as

Runscaled = S I r"te f mass creation in cell P |


cellsp (3.27)

The scaled residual for the continuity equation is defined as

RC

RC
v
unscaled-
iteration N
RC (3.28)
v
unscaled-
where the denominator is the largest absolute
iterationvalue
5 of the unscale d continuity residual in

the first five iterations. The criteria used to determine convergence for this study require d

that the scale d residuals for all variables as define d by equations (3.26) an d (3.28)

decrease to 10' , that is decrease by three orders of magnitude for each time step.

3.6 Chapter Summary

Chapter III described briefly how the governing flow equations are implemented

in the program Fluent. The modeling used to obtain a closed form of the Reynolds-stress

equation was discussed. This modeling results in what is called the Reynolds-stress model

or RSM. This model is more appropriate to54use for cyclone simulations because of the

strong swirl present in the flow. The particle trajectory calculations are explained,
which also include an option that accounts for some effects of turbulent dispersion on the

particles called the discrete random walk or DRW model. A brief discussion of the control

volume method was presented as well as the way the governing equations are discretized.

The general process used by Fluent to solve the resulting system of equations was

outlined. Finally, the convergence criteria that is used to determine when the iterative

solution process is stopped was discussed. The interested reader can consult the references

liste d in the chapter for more detail on the subjects discusse d in this section.

55
CHAPTER IV.

EXPERIMENTAL DATA

4.1 Intro duction

This chapter presents the geometry, test dust properties experimental results and

an estimate of the 95% confidence limits for the test data. The numerical results will be

compare d with the data in this chapter.

4.2 150 mm Cyclone Geometry

Figure 4.1 below shows the 150 mm cyclone geometry given in Ogawa (1984). This

geometry was chosen because Ogawa provided a detailed description of the geometry,

including the dust bin as well as test data at several flow rates.

4.3 Test Dust Characteristics

The test dust used in the experiments was Kanto-loam. Kanto-loam is a naturally

occurring volcanic dust found primarily in Japan. It comes in several grades and is a

56
standard dust use d in Japan to evaluate filters. The density of the particles is 2970

kg/m3. The size distribution of the dust is shown in the following figure. The

equation for the curve fit shown in the graph was supplied by Ogawa (1984) and is

the following:

R(xp )(%) = 100 * exp(-0.0848x^93) (41)

where xp is the particle size in microns and R(xp) is called the residue as a percent.

The residue, R(xp) is interpreted as the percentage of total test dust based on mass

that is greater than the particle size, xp.

De=
50mm

S=
80mm

h=
150mm

H-h = 300mm
(where H is the total
height and H = 450mm)

HB =
225mm

size of 15 microns is 35%. This means that 35% of the dust (by mass) is composed
Figure 4.1 Ogawa 150 mm cyclone
geometry
of
For example, referring the graph below and using the curve fit, the residue at a
particle
56
particles greater than or equal to 15 microns. The curve is also use d to determine

the percentage of the dust that is within a particular particle size range. For

example, to calculate the percentage of Kanto-loam test dust that was between the

sizes of 2.5 an d 3.5 microns the following procedure would be used:


2.5^m < (%dust) < 3.5^m = ^(2.5) -^(3.5)

^(2.5) - ^(3.5) = 100 exp(-0.0848 * 2.5093) - 100exp(-0.0848 * 3.5093) (43)

^(2.5) - ^(3.5) = 81.96899 - 76.19448 = 5.8% (44)

2.5^m < (%dust) < 3.5^m = 5.8% (45)

The above type of calculations will be used in following sections to calculate the

absolute collection efficiency of the cyclone.

Kanl
o-Loam Test Dust Distribution dust density: 2970
kg/m3
Test Data Curve Fit
100.0 -
QH n sN
\
on n
- 7n n
A
R0 0
t/> W
00.0

^Ann
30.0

-inn
0.0 - 0
05 0 35 .0 40.0
.0 20
).0 15 0 25 xp .0 30
0 10 icle S
Part (micri ons)
ize,

Figure 4.2 Kanto-loam test dust size distribution


57
4.4 Experimental Collection Efficiency Results

Ogawa provided collection efficiency results both in terms of grade efficiency

curves and as well as the absolute collection efficiency operating with Kanto loam

test dust.

4.4.1 Grade Effi ciency Curve s

Ogawa provides fractional collection or grade efficiency data as a function of

particle diameter for the 150 mm cyclone at several different inlet velocity con ditions.

The data points are listed in Table 4.1 below and are plotte d in Figure 4.3. These curves

of collection efficiency verses particle size at a specified particle density are called grade

efficiency curves. The grade efficiency curves are a function of particle size particle

density, gas properties, cyclone geometry and flow rate. The collection efficiency is also

affected by the dust concentration in the air being processed. At dust concentrations

higher than approximately 5 to 10 g/m3 the collection efficiency increases a small amount.

The data presented here had dust concentration levels of 1.5 to 2.5 g/m3.

58
Oc
jawa 150mm Cyclone Grade Efficiency Experimental Data

19.7 m/s data 015.1 m/sdata A10.3m/sdata

1 nn J_
n V n B V
>> n on 11 X A A A A A
n.9n 11
I
X
c < A
n an
n8n
n 7n
C ii I I

o n n
*->
o ^1
o n
o
O n ac\
RJ
? nm
y n on
LL.
nm
n nn
nn 1 n 2 n3 n4 n9 n ic .n ii .n 1 2 .n
n6 n7
n5 n8
Partit
:le e
rons)
Siz (mic
Figure 4.3 150 mm cyclone grade efficiency curves

Table 4.1 Grade efficiency data for 150 mm Ogawa cyclone


Particle Fractional collection efficiency at various inlet
diameter velocities
(microns) 19.7 (m/s) 15.1 (m/s) 10.3 (m/s)
1.0 0.686 n/a n/a
1.5 0.775 0.559 n/a
2.0 0.891 0.654 0.556
3.0 0.945 0.818 0.662
4.0 0.978 0.916 0.828
5.0 0.989 0.956 0.888
6.0 0.992 0.966 0.911
7.0 0.993 0.974 0.914
8.0 0.986 0.977 0.910
9.0 0.990 0.976 0.910
10.0 0.995 0.983 0.922
11.0 1.00 0.991 0.910

59
4.4.2 Absolute Collection Efficiency

Absolute collection efficiency, noveraii, means the fraction of total dust, by mass,

which is collected compare d to the total dust fed into the unit. The absolute collection

efficiency depends on the grade efficiency characteristics of the cyclone being used and it

also depends on the size distribution and density of the particles that make up the test

dust. To calculate Voveraii one must multiply the efficiency, V(XP ) , which is a function

of particles size, xp by the mass fraction of particles, G(xp) in the dust, which is also a

function of particle size, xp . An expression for overall efficiency can be written as

x=max xp I
overall = \n X
( ) dG

x=o (4.6)

where max xp is the largest particle size present in the test dust. Generally, overall

efficiency is calculate d in a piece-wise manner by dividing the separation efficiency or

grade efficiency curve of the cyclone into segments. The dust particle size distribution is

divided into the same segments used for the grade efficiency curves. The overall

efficiency is then the sum of the products of separation efficiency and fraction of dust in

each of the segments, which together cover the entire range of particle sizes in the test

dust. An expression for this piece-wise approach to calculating overall efficiency can be

written as

noverall = VlGl + VG2 + VaGA ^ + VG (4.E)

60
where the subscripts 1, 2 and 1 represent the discrete particles size ranges into

which the grade efficiency and dust size distribution have been divide d. The

subscript, n is the total number of segments use d for the calculation. If the grade

efficiency data at a known particle density is available and the particle size

distribution for the test dust also at a known density is available, then the absolute

collection efficiency of the cyclone for the given test dust can be calculate d. The

following example shows how this is done.

Example: Calculation of Absolute Collection Efficiency Fin d:

Absolute Collection efficiency of the 150mm cyclone operating at an

inlet velocity of 19.7 m/s filtering Kanto-loam test dust.

Given:

o 150mm cyclone collection efficiency verses particle (for particle density of

2970 kg/m3) size for an inlet velocity of 19.7m/s (in Table 4.1) o Residue

verses particle size for Kanto-loam dust at density 2970 kg/m3 from equation

(4.1)

Solution:

Step 1. Construct a table of the collection efficiency verses particle size. (Use
Table 4.1)
Table 4.2 vs. particle size for absolute
Efficiency efficiency
calc ation example
Particle u 19.7 m/s data
size fractional collection
(microns) efficiency
1.00 0.686
1.50 0.775
2.00 0.891
3.00 0.945

61
Table 4.2 Efficiency vs. pai tide size for absolute efficiency
calcu ation example (continue d)
4.00 0.978
5.00 0.989
6.00 0.992
7.00 0.993
8.00 0.986
9.00 0.990
10.00 0.995
11.00 1.000

Step 2. The above table gives separation efficiency for finite particle sizes. The

test dust contains particles of continuously varying sizes from almost zero microns to

some particles greater than 40 microns. What is done next is to divide up the range of

particle sizes into finite divisions about the known collection efficiency data points that

we have for the cyclone. This is done in the last two columns in the following table. There

exists a total of twelve particle sizes with known collection efficiencies. The range of dust

particle sizes is divided so that each of the known particle size efficiencies can be use d to

estimate the portion of dust in a size range about the known collection efficiency point.

The first row of Table 4.3 represents the collection efficiency at 1.0 micron. The one

micron efficiency is used as the efficiency at which particles from the low end size of zero

microns to the high end size of 1.25 microns is separated. In a similar manner, the

collection efficiency for the 1.5 micron particle is the efficiency at which particles

between 1.25 and 1.75 microns is collected. This process is continued until a particle size

of 11 microns is reached. The 11 micron efficiency is used as the collection efficiency at

which particles from a size of 10.5 microns up to some large size that would include all

the particles in the test dust. In this case a value of 1000 microns is use d for the upper
size limit of the dust size distribution.

62
Table 4.3 Particle size ranges for absolute efficiency calculation
example
19.7 m/s data
Particle fractional High end of
size collection Low end of size size
(microns) efficiency range range
1.00 0.686 0.00 1.25
1.50 0.775 1.25 1.75
2.00 0.891 1.75 2.50
3.00 0.945 2.50 3.50
4.00 0.978 3.50 4.50
5.00 0.989 4.50 5.50
6.00 0.992 5.50 6.50
7.00 0.993 6.50 7.50
8.00 0.986 7.50 8.50
9.00 0.990 8.50 9.50
10.00 0.995 9.50 10.50
11.00 1.000 10.50 1000.00

A dust size range has now been assigne d to each of the 12 cyclone collection efficiency

points. The dust size range that gets separated at the 1.0 micron efficiency of 0.686 is for

dust particles from zero to 1.25 microns. The dust size range that gets separate d at the

1.5 micron collection efficiency of 0.775 is for dust particles from 1.25 to 1.75 microns,

etc.

Step 3. Equation (4.1) describes the particle size distribution. The process shown

above in equations (4.2) through (4.5) is use d to estimate the portion of the test dust that

is contained in each particle size range. The results of this process are shown in the fifth

column in the table below. The last column title d Mass Fraction shows the fraction of

test dust that is contained in each of the 12 size ranges we have set up. This last column

sums to the value 1.000 indicating that all of the mass in test dust sample has been

accounted for.

63
Step 4. In Table 4.4 the fifth column contains the mass fraction of dust contained

in each of the designated particle size ranges. The second column contains the collection

efficiency of the cyclone in the corresponding particles size ranges. The product of the

values in columns 2 and 5 is the amount of dust that is collected in each size range.

Table 4.4 Mass fraction in each particle size range for absolute efficiency
example
19.7 m/s data
Particle High end of
fractional Low end of
size size
collection size range Mass Fraction
(microns) range
efficiency in this range
1.00 0.686 0.00 1.25 0.099
1.50 0.775 1.25 1.75 0.034
2.00 0.891 1.75 2.50 0.047
3.00 0.945 2.50 3.50 0.058
4.00 0.978 3.50 4.50 0.053
5.00 0.989 4.50 5.50 0.048
6.00 0.992 5.50 6.50 0.044
7.00 0.993 6.50 7.50 0.041
8.00 0.986 7.50 8.50 0.038
9.00 0.990 8.50 9.50 0.035
10.00 0.995 9.50 10.50 0.033
11.00 1.000 10.50 1000.00 0.470
Sums: 1.000

Step 5. A new Table 4.5 is created in which an additional column is added for the

fraction of dust collected for each particle size range. This additional column is the pro

duct of the values from columns 2 an d 5. The sum of the entries in this column is the

absolute efficiency.

Answer: The absolute efficiency of the unit operating at 19.7 m/s inlet velocity

when separating Kanto-loam test dust is 95.0%.

64
Table 4.5 Mass collected in each size range for absolute efficiency calculation

19.7 m/s data


Particle fractional Low en d High end Mass
size collection of size of size Fraction in Mass collected in
(microns) efficiency range range this range this range
1.00 0.686 0.00 1.25 0.099 0.067980204
1.50 0.775 1.25 1.75 0.034 0.026263649
2.00 0.891 1.75 2.50 0.047 0.042166566
3.00 0.945 2.50 3.50 0.058 0.054569146
4.00 0.978 3.50 4.50 0.053 0.051479037
5.00 0.989 4.50 5.50 0.048 0.04773355
6.00 0.992 5.50 6.50 0.044 0.044072464
7.00 0.993 6.50 7.50 0.041 0.040724306
8.00 0.986 7.50 8.50 0.038 0.037407833
9.00 0.990 8.50 9.50 0.035 0.034805191
10.00 0.995 9.50 10.50 0.033 0.032461151
11.00 1.000 10.50 1000.00 0.470 0.469884002
Sums: 1.000 0.950

4.4.3 Measurement Uncertainty Estimates

Ogawa (1984) also provided absolute collection efficiency data for the 150mm

cyclone at several different conditions. The data points are listed in the table below and

are plotted in the following figure. The data presente d in Table 4.6 and plotted in Figure

4.4 is clustered in three groups in the general inlet velocity ranges of interest in this study.

There is variation in both the fractional efficiency and the velocity test data. Ogawa did

not present any uncertainty estimates with his published results.

65
Table 4.6 Absolute fractional collection efficiency data for Ogawa 150 mm cyclone
Absolute fractional collection efficiency Inlet velocity (m/s)
0.880 10.3
0.848 11.0
0.866 11.0
0.880 11.0
0.914 15.1
0.912 15.2
0.904 15.4
0.877 15.5
0.956 19.7
0.879 20.9
0.924 21.3
0.933 21.4

The data from Table 4.6 was used to generate an estimate of test uncertainty because

measurement errors occur in all experiments. The test apparatus and measurement system

is not described. Therefore, where necessary, assumptions will be made in order to

complete the analysis.


Absolute Collection Efficiency for 150mm Cyclone

10.3 to 110 m/s range 151 to 155 m/s range A 19.7 to 21.4 m/s range

1.00 -
A
> 1
O L
O <>
<> 4
LU
re
c
o
000
re
LL

0.70 - 5 .0 250
0 10 .0 1 Inlet Velc 0 20 city (m/s)

Figure 4.4 Experimental absolute collection efficiency data

66
Kline and McClintock (January 1953) and Moffat (1988) describe types of

experimental error, which are usually classified as either bias (fixed) or precision

(random) error. Precision errors are presume d to behave randomly with a zero

mean. Bias errors are fixed and result in an offset or shift in the data so that the

mean of the set of measurements is shifted away from the true mean of the variable

being measured. The overall uncertainty is a combination of both the bias and

prec

ision

error sources and is commonly expresse d as

(4.8)

where U'r*ss is the root sum-of-squares value of the uncertainty of the test result, ; is

the bias error and Pr is the precision error. For cases where the sample sizes are

small, that is less than 30, the precision error, in equation (4.8) is replaced by the

product of the precision index, Sr and the t-statistic so that the resulting expression

for the overall uncertainty becomes

U
,SS =[;/ + (r ) F^
(4.9
)

wher
e the
preci
sion index is

(4.10)

where N is the number of observations in the sample set and

67
X=
(4.11)

For the test data shown in Table 4.6 the three groups of data are clustered about the

three inlet velocity ranges of interest. The variation in efficiency, for the purposes

of this

68
study, is assumed to be cause d by precision errors in the measurement of the

efficiency values. A bias error component of 1% will be used to account for fixed

errors in the particle sizing and size distribution measurement process. The

variation in inlet velocities within each of the three groups of data points will be

treated as precision error in the flow measurement system. A velocity bias error

component of 1%will be used to account for error that may be present due to

flow measurement system calibration.

4.4.3.1 Uncertainty Estimates for Efficiency Measurements

Table 4.7 Absolute efficiency data for 150


mm cyclone in the 19.7 to 21.4 m/s
range
Absolute fractional Inlet velocity (m/s)
collection efficiency
0.956 19.7
0.879 20.9
0.924 21.3
0.933 21.4

Referring to Table 4.7, the following computations can be made for efficiency in

the 19.7 to 21.4 m/s range. Using equation (4.11) with N = 4, the mean value is

=1 (0.956 + 0.879 + 0.924 + 0.933) = 0.923 4 (4.12)

Using equation (4.10) the precision in dex is

1
S,e = J1L [(0.956 - 0.923)2 2 /
[ (4 - 1) 2 2
+ (0.879 - 0.923) + (0.924 - 0.923) + (0.933 - 0.923)
] /
Se = 0.032280025 = 0.032 2

(4.13)

The degrees of freedom (DOF) is defined as

69
v = ( 5 -1) (4.14
)
(4 - 9)
Therefore, for this example, v
= = 3.

Looking in a t-distribution table for the t- statistic for a DOF ofv = 3, one finds that

t = 3.182 (Hines & Montgomery, 1980).

The bias error for efficiency was assume d to be 1% of the mean value of

efficiency, so one would have


(4.15
Be = 0.01*0.923 = 0.009 The overall )

uncertainty is then calculated using equation

(4.9) as follows:
ue = [B/ + (tse)2 j12 = [0.0092 + (3.182 * 0.032)2 j12 =0.102 ( 11.1%) (4 16)

In a similar manner, the errors for the other two groups of data can be

calculated. The following table shows the final values for errors for all three

flow range groups. This completes the uncertainty estimates for the measured

test result of efficiency.

70
Table 4.8 Uncertainty estimates for absolute collection
efficiency
10.3measurements
to 11.0 15.1 to 15.5 19.7 to 21.4
m/s range m/s range m/s range
obs V1 0.880 0.914 0.956
obs V2 0.848 0.912 0.879
obs V3 0.866 0.904 0.924
obs V4 0.880 0.877 0.933
Mean Eff 0.869 0.902 0.923
Se 0.015 0.017 0.032
Be 0.009 0.009 0.009
sample size, N 4 4 4
DOF = N-1: 3 3 3
t value 95% conf for
3.182 3.182 3.182
DOF:
Ue (fractional
efficiency) 0.049 0.055 0.103
Ue (%) 5.6% 6.1% 11.2%

71
4.4.3.2 Uncertainty Estimates for Velocity Measurements

Referring to Table 4.7, the following computations can be made for inlet

velocity for the test data in the 19.7 to 21.3 m/s range. Using equation (4.11) with

N = 4, the mean value is


<v = I (19.7 + 20.9 + 21.3 + 21.4) = 20.8
4 (4.16)

Using equation (4.10) the precision in dex is

1 [(19.7 - 20.8)2 + (20.9 - 20.8)2 + (21.3 - 20.8)2 + (21.4 - 2


S 2
,, .(4 -1) 20.8) H
Sv = 0.78102497 = (4.17
0.781 )
As was the case above, v = 3 an d t = 3.182.

The bias error for efficiency was assumed to be 1%of the nominal velocity, so

one would have

;v = 0.01*20.8 = 0.208 The overall (4.18


)
uncertainty is then calculated using equation (4.9) as follows:

Uv =[;v2 + (tSv)2Y2 = [0.2082 + (3.182* 0.781)2 Y2 = 2.493 (12.0%) (4 19)

In a similar manner, the errors for the other two groups of data can be calculated.

The following table shows the final values for uncertainties for all three flow range

groups.
Table 4.9 Uncertainty estimates for velocity measurements
10.3 to 11.0 m/s 15.1 to 15.5 19.7 to 21.4
range m/s range m/s range
obs V1 10.300 15.100 19.700
obs V2 11.000 15.200 20.900

72
Table 4.9 Uncertainty estimates for velocity measurements (continued)
10.3 to 11.0 m/s 15.1 to 15.5 19.7 to 21.4 m/s
range m/s range range
obs V3 11.000 15.400 21.300
obs V4 11.000 15.500 21.400
Mean Velocity 10.825 15.300 20.825
Sv 0.350 0.183 0.780
Bv 0.108 0.153 0.208
sample size, N 4 4 4
DOF = N-1: 3 3 3
t value 95% conf for 3.182 3.182 3.182
DOF:
Uv (m/s) 1.119 0.601 2.492
Uv(%) 10.3% 3.9% 12.0%

4.4.3.3 Overall Uncertainty of the Measured Test Result for Efficiency

The measure d results of the experiments are collection efficiency and particle size. The

general expression used to calculate the uncertainty in a measured result, Umr is

Vf 2 2
dr dr dr
+ U2 + + dXi Ui
2
va<2 , (4.20)

were Xj, X2 and Xi are the measured variables that affect the test result, r. The quantity r is

the measured test result which is a function of the measured quantities X;, X2 and Xi-. The

variables, U;, U2 and Ui are the uncertainty values for the measured quantities X; X2 and Xi.

The overall uncertainty for efficiency has only one component and the r- expression for

efficiency is trivial in this case:

r= (4.21)

dr = i
d (4522
)

73
From equation (4.16) we obtained:
U
, =05102 (4.23)

Substituting equations (4.21) through (4.23) into (4.20) results in the an expression for

overall uncertainty for the measure d collection efficiency, U, as follows.


2-i
'*U. 1/ [(1* U, )2 f = U
U ,3e
(4.24)

The uncertainty for the measured result, efficiency has already been calculated for all

three flow ranges because of the relation ship shown in equation (4.24). These uncertainty

values are shown in the last two rows of Table 4.9 above.

4.4.3.4 Overall Uncertainty of the Measured Test Result for Particle size

The cyclones collection efficiency characteristics change with flow (inlet velocity)

changes. Stairmand (1951) provided similarity laws or transformations that allow one to

adjust the performance characteristics of cyclones for changes in gas density, cyclone

diameter and inlet velocity. These relationships have been used by many researchers such

as Dirgo an d Leith (1985). The basic transformation expression is shown below as

d Pp 1Vi1 D

V P P 2 V i 2 D1
d2
(4.25)

where di is the particle diameter collected at a given efficiency and under operating con

ditions for particle density, pp 1 an d inlet velocity, v using a cyclone of size Di. d2

74
is the particle diameter which would be collected at the same efficiency as that of di but
under operating conditions reflected by a new particle density, pp2 and a new inlet

velocity, vz2 and a new cyclone size, D2. Equation (4.25) can be simplified because we are

only intereste d in changes in particle size as a result of changes in inlet velocity, which

results in the following:

d2 dj il

i (4.26)
2

Vil Vi2
If the variables di and are reference or nominal values, then d
2 and would be the

dependent and independent variables respectively for the purposes of this exercise. Then

equation (4.26) can be written, with modified subscripts as

d = d ref
(4.27)

where d is the measured result for particle size and is a function of the inlet velocity vz

which has uncertainty, Uv associated with it. The derivative of d with respect to vz can be

written as

dd 2
dv
z
(4.28)

Up
Now an expression for the uncertainty in particle size, can be written based on

equation (4.20) as

(4.29)

75
For the cyclone calculations the reference size, dre) will be taken as the cut diameter, dc

available from the test data and the reference inlet velocity, Vre/ will be taken as the mean

velocity of the test data, velocity, vmea#. Substituting the above variables into equation

(4.29) and simplifying results in the following:

2 d
c J
v
d v
d
Ur 3/ U U U
3/ 3/
2 v/2 2 v/2 2 v/2 (4.30)

An example of the particle size uncertainty estimate for the particles taken at an inlet

velocity v of 19.7 m/s, a reference mean velocity, vmea# of 20.825 m/s and based on a

cut size, dc of 0.9 microns and a velocity uncertainty , Uv of 2,492


m/s is shown below

using equation (4.30).

U= ^U = 09^20 8j25* 2.492 = 0.059 microns (6.5% of d )


2
vf2 2 19 2
* .^ (4.31)

4.4.3.5 Complete d Uncertainty Results for all Three Inlet Velocities

Based on the procedures described in the above two sections, calculations of

efficiency uncertainty and particle size uncertainty can be calculated for the three test flow

rates and are shown in Table 4.10 below.

76
Table 4.10 Final uncertainty estimates for efficiency & particle size

Test Inlet Velocity (m/s) 10.3 15.1 19.7


nominal fractional
0.869 0.902 0.923
Efficiency efficiency
Uncertainty Ue (fraction) 0.049 0.055 0.103
Ue (% of nominal eff.) 5.6% 6.1% 11.2%
nominal size (microns) 2.0 1.5 0.9
Particle
Size Up (microns) 0.111 0.030 0.059
Uncertainty Up (% of nominal size) 5.6% 2.0% 6.5%

If it is assumed that the variation in efficiency results applies equally to each particle size

category, then it is possible to show that the confidence interval for the collection

efficiency of each particle size is the same confidence interval as calculated above for the

absolute collection efficiency. Figures 4.5, 4.6 and 4.7 show the experimental grade

efficiency curves for 10.3, 15.1 and 19.7 m/s inlet velocities as reported by Ogawa (1984)

with error bars indicating the estimated efficiency and particle size measurement

uncertainty from Table 4.10.

77
Ogawa 150mm Cyclone Grade Efficiency with Estimated 95%
Confidence limits: 10.3 ms Inlet Velocity

1iiiif i
rj

3
j

0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 10.0 11.0
Particle Size (microns)

Figure 4.5 Cyclone at 10.3 m/s grade efficiency points with experimental error

Ogawa 150mm Cyclone Grade Efficiency with Estimated 95%


Confidence limits: 15.1 ms Inlet Velocity

iiiiii

<>

< >

0.0 1.0 2.0 30 4.0 50 60 70 8.0 9.0 10.0 11.0


Particle Size (microns)

Figure 4.6 Cyclone at 15.1 m/s grade efficiency points with experimental error

78
Ogawa 150mm Cyclone Grade Efficiency with Estimated 95%
Confidence limits: 19.7 ms Inlet Velocity

T T T I I I I I ;r

T >
S
T
c> 1

0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 10.0 11.0
Particle Size (microns)

Figure 4.7 Cyclone at 19.7 m/s grade efficiency points with experimental error

4.5 Chapter Summary

This chapter presented a description of the cyclone geometry, test dust

characteristics and collection efficiency data presented by Ogawa (1984) and use d in this

study for comparison with the numerical model. The grade efficiency curve of the cyclone

was explained. The absolute collection efficiency was explaine d and an example of how

to calculate the absolute collection efficiency was given. Ogawa (1984) gave no

experimental error estimates for the grade efficiency curves he provided. He did, however

provide several data points for absolute collection efficiency that provided

79
information on the variation in experimental data from his studies. This available data was

use d to estimate measurement uncertainty and to calculate 95% confidence limits for the

grade efficiency point test data. The uncertainty estimation was briefly explained an d

calculation examples were given.

80
CHAPTER V.

NUMERICAL STUDY

5.1 Intro duction

This chapter discusses the details of the numerical mo deling done for this project.

A description of the construction of the computational grid and the boundary conditions

use d will be given. A study of the separation process is conducted using two approaches

for the particle tracking calculations. A study of discretization error is conducted. Finally,

simulation of the separation performance of the 150 mm cyclone at three flow rates will

be conducted.

5.2 Preparation of the Geometry for Grid Generation

The basic cyclone dimensions are given by Ogawa (1984) along with the locations

of the upstream and downstream pressure taps. For this study the system inlet was located

1.5 inlet pipe diameters before the upstream pressure tap. The exit pipe of the model was

extended 6 outlet pipe diameters beyond the downstream pressure tap. The following

figure shows a side view of the geometry with these features identified.

81
Figure 5.1 Cyclone geometry features relate d to the grid generation process

A region was created in the geometry for the transition area where the inlet pipe enters the

cyclone body that consisted of part of the inlet pipe and a segment create d by rotating a

rectangular section 25 mm by 80 mm through a 90 degree angle. The following figure

shows the transition region for the computational model. This transition region provided a

flexible means to connect any shape of inlet duct to the body of the cyclone. Once the

regions described in this section were defined, the computational grid could then be

constructed.

82
I

Figure 5.2 Inlet pipe to cyclone bo dy transition region 5.3 Computational

Grid

Slack and Harwood (2000), QNET-CFD (2003) along with studies by Hoekstra et

al. (1999) an d Slack, Prasad et al. (2000) provide guidance for setting up the

computational grid. In general, it is possible to model the flow features with relatively

coarse grids totaling from approximately 30,000 cells (Slack & Harwood, 2000) to 40,000

cells (Slack, Prasad et al., 2000; QNET-CFD, 2003). It is important, however, to resolve

the swirling fluid core of the cyclone in order to calculate the radial, tangential and axial

velocities as well as pressure drop adequately. It is not necessary to have an extremely fine

mesh at the solid cyclone walls because the important turbulence generation affecting

performance occurs within the body of the cyclone due to the strong

83
centrifugal effects that result in large shear between fluid layers at varying radial

distances from the axis of fluid rotation (Slack & Harwood, 2000; QNET-CFD, 2003). In

general, the following rules of thumb may be use d.

o Hexahe dral cells are recommende d to the extent possible.

o Avoid highly stretche d cells. Cell aspect ratio should not excee d 1:5 as high aspect

ratios in swirling flows may cause divergence problems. o Use a fine mesh along the

axis of the cyclone to resolve the central core. o Boundary layer resolution near walls

is not extremely critical as turbulence is generated in the main flow.

Figure 5.3 below shows the grid use d for this study. Hexahe dral cells are used in

the dust bin, exit pipe, cone and most of the cylindrical body of the cyclone. A tetrahedral

grid is used in the transition region describe d in the previous section, where the inlet pipe

connects to the cyclone body. The only case in which other modelers were able to use hex

cells for the whole grid was when the inlet duct had a rectangular crosssection. In the

present analysis, the inlet duct consists of a cylindrical pipe. The most practical approach

to grid the transition was to use the tetrahedral cells. From the transition back to the inlet

boundary prism type cells were used. These cell regions are note d in Figure 5.3. This

basic approach to grid construction is able to accommo date all types of inlet duct cross-

sections including rectangular and elliptical shapes.

84
Figure 5.3 Cyclone grid construction overview

5.3.1 Grid Construction Sequence

The grid was created using the following steps.

Step 1. Grid begins in a horizontal plane that passes through the cyclone bo dy at

the bottom surface of the exit duct. Cells within the exit pipe diameter are generated at a

target size of 3.5% of the cyclone main diameter, D, which for this model is 3.5% of 150

mm or a target size of 5.25 mm. See Figures 5.4 and 5.5.

85
Figure 5.4 Plane where grid construction starts

Hex cells
across exit
pipe opening
are sized at
3.5% of the
cyclone
diameter, D to
resolve
Hex cells
cyclone core
transition to
flow features.
< size 7% of
cyclone
diameter, D.

Figure 5.5 Starting grid size an d distribution

86
Step 2. The cells within the exit pipe are extru de d upward in the positive Z-

direction using an axial cell length that was chosen so that the cell aspect ration was less

than 5:1. The cells in the bo dy of the cyclone, exclu ding the transition region, were

extruded upward in the positive Z-direction using an axial cell length of 7% of the

diameter D. The grid in the body of the cyclone is extrude d downward in the negative Z-

direction to grid the cylindrical body of the cyclone below the transition region. Note that

some grading of the grid in the axial direction is used to reduce the cell size near the outlet

pipe end wall while still keeping cell aspect ratios less than 5:1. Figure 5.6, below, shows

the grid create d in this process.

Figure 5.6 Hex grid in cyclone body and exit pipe

87
Step 3. The transition region is grid de d using tetrahe dral cells. In this process, the

mesh generation software uses pyramid-shaped cells which connect the hex cells in the

body to the tet cells in the transition region. Then the inlet pipe extension upstream of the

first pressure tap is extruded using prism cells to complete the inlet pipe grid. The next

three figures show the pyramid cells, the fully meshed tet transition region and finally the

inlet extension region filled with prism cells.

Figure 5.7 Hexahe dral cells are connecte d to tet cells with pyramid cells

88
Figure 5.9 Prism cells use d in the inlet pipe extension

89
Step 4. The grid in the cyclone cone is extruded next. Grading must be used to

make the grid finer toward the cone apex, where the dust bin is located, in order to keep

aspect ratios within the recommen de d 5:1 range. The grid in this region is shown below

in the following figure.

Step 5. Finally the dust bin is meshe d. Cell size is increase d smoothly in the radial

direction from the size that exists at the cone apex to a maximum of 7% of the cyclone

main diameter on the top of the dust bin. Then the grid at the top of the dust bin is

extruded downward in the negative Z- direction, again, adjusting the axial cell size so that

aspect ratios do not excee d 5:1. This completes the cyclone mesh, which is shown in the

following figure. The final grid consisted of 60,327 cells.


88
Figure 5.11 Complete d Grid

5.4 Material Properties an d Boun dary Con ditions

The fluid is air, assumed to be incompressible, with the properties shown in the

table below.
Table 5.1 Material properties
Density (kg/m3) 1.225
Air
Absolute Viscosity (kg/m-s) 0.000017894
Kanto-loam test dust Particle Density (kg/m3) 2970

A mass flow boundary condition was imposed at the inlet. On the outlet, a zero gradient

boundary condition (BC) was applied. The walls were modele d as no-slip boundaries.

89
Three flow rates were simulate d. The turbulence BC at the inlet was specific d using an

intensity of 5% and a hydraulic diameter of 0.050 m, the diameter of the inlet pipe. These

values are listed in the table below.


Table 5.2 Flow boun dary con ditions
BC type Case Use d Value
Low Flow Case 0.0247744
Mass Flow (kg/s) 0.0363198
Medium Flow Case
Inlet High Flow Case 0.0473841
Turbulence Intensity (%) All Cases 5%
Turbulence Duct Hy draulic Diameter
All Cases 0.050
_________ (m) _________
No input
Outlet Zero Gra d ient All Cases value
required
No input
Wall No-Slip All Cases value
required

The mass flows for the three flow rates were based on the three inlet velocities given by

Ogawa (1984) which are 10.3, 15.1 and 19.7 m/s. Based on the fluid density, the cross-

sectional area of the inlet duct, the inlet mass flow rates in kg/sec are calculated using the

following equation:
m
= V1Aip (5.1)

where V1 is the inlet velocity in m/s, pis the air density and A1 is the inlet duct area in

meters squared given by the following:

A1 = - D
1
4 1 (5.2)

where D1 is the inlet duct diameter in meters. Turbulence intensity, I is defined here as the

root-mean-square of the fluctuating component of velocity, u' divide d by the mean

90
flow velocity, uavg. Turbulent kinetic energy, k is related to turbulence intensity, I by the

following equation:
3
k
= ^avg1 ) (5.3)

From equation (5.3) the required values of the

Reynolds stresses at the inlet are derived assuming that the flow at the inlet is isotropic.

Isotropic turbulence means that the following expressions hold true:


--- 9
u'2 = - k (i= 1,2,3)
3 v
(5.4)

u'u= 0.0 (5.5)


J

From the hydraulic diameter, D? specified for the inlet of the cyclone duct, the turbulence

length scale, is determined. The turbulence length scale, , is a physical quantity related

to the size of the large eddies that contain most of the turbulent energy in turbulent flows.

In fully-developed duct flows, is restricted by the size of the duct, since the turbulent

eddies cannot be larger than the duct. An approximate relationship between and the

physical size of the duct is

= 007D? (5.6)

where D? is the hydraulic diameter of the duct. In the case of circular ducts, the hydraulic

diameter is the duct diameter. A transport equation for e, the dissipation rate of turbulent

kinetic energy is also used in the RSM. From the turbulent kinetic energy, k and the

turbulence length scale, , the boundary condition for e is determined using the following

relation.

91
k^_
_ C 3/
4 (5.7)
1
4

where C4 is an empirical constant specified in the turbulence model which has a value of

0.09.

5.5 Time Step Size

Several gui delines are available for determining an appropriate time step size. A

characteristic residence time (QNET-CFD, 2003) which is defined as


_ V_
tc V
_ (5.8)

where V is the volume of the model and V is the volume flow rate for the problem.

The recommendation given was that the time step, At be less than 0.01*t c. The iterative

solver used for the transient calculations, does not have a restriction on At from a stability

stand point (F6.3UG, 2006). That is, it is not necessary for the Courant number be

restricted to values less than 1.0. The Fluent users guide does recommend that, for

efficient calculations, the time step size be set so that 5 to 10 iterations are nee de d to

reach a converged solution for each time step. Another means of determining At can be

set by looking at the values of the cell Courant number in the flow domain. The cell

Courant number is defined as

Courant ,, _At |V OUtgoig JluXeS )


I vlume ) cell (59)

92
For a stable efficient calculation, the cell Courant number should not excee d a value of

20 to 40 in the most sensitive transient regions of the domain. Again, please note that for

stability considerations the Courant number, which is usually required to have a value less

than 1.0, is not restricted when using Fluent's transient iterative solver. In this case,

monitoring the values of the cell Courant number is used as a guide for determining an

appropriate time step size. The Table 5.3 below shows the various quantities that can be

part of the choice for time step size. In practice, running these simulations, the limiting

factor in determining the time step size was that the number of iterations to converge each

time step be approximately 5 to 10.

Table 5.3 Quantities related to time step size selection

Inlet actual time At Number of


Velocity step size, At t
max Courantcell
c solver iterations
(m/s) (sec)
per time step
10.3 0.0002 Actual Value 0.0004 3 5
15.1 0.0002 Actual Value 0.0006 5 7
19.7 0.0002 Actual Value 0.0008 6 8
Recommended
< 0.01 < 20 5 to 10
limit for value

Based on this information, time step sizes were initially set for 0.0004 tc < At < 0.0008 tc

then as the simulation ran, the step size was adjusted as neede d so that the number of

iterations per time step was in the range of 5 to 10.

93
5.6 Fully Unstea dy Continuous Phase an d Particle Tracking Calculations

A number of exploratory simulations were performe d to provide insight into the

overall separation process. In general the model was set up for a specific flow rate.

Unsteady calculations were then performed until the solution exhibited a steady-state or

periodically steady characteristic. Once the unsteady solution had stabilized, the dust

particles were introduced into the system and the unsteady calculations were continued

while tracking the particles through the domain. Figures 5.12 and 5.13 below show plots

of the changes in particle mass in the cyclone over time during separation. This was a

simulation in which particles were injecte d an d then tracke d in time during the unsteady

flow simulation. The air flow rate was set at a 15.1 m/s inlet condition. The re d line

represents the amount of particle mass present in the whole cyclone system. The blue line

represents the amount of particle mass that is contained only in the dust bin. The process

can be broken up into different stages as described in the following paragraphs.

The total particle mass in the system is considered first, which is tracke d by the red

curves in Figures 5.12 and 5.13. It can be seen that at time zero the particle injection

process begins. As particles continue to be injecte d, the particle mass in the system rises

sharply to a value of 1.0 in the first one tenth of a second. (Note that the total mass is

normalized so that a value of 1.0 indicates that the mass in the cyclone is 100% of the total

mass injected.) After the amount of particle mass peaks at 1.0, mass begins to leave the

system as the particles find their way from their location in the upper portion of the

cyclone directly to the outlet pipe. These particles are able to leave the system before they

begin to be influence d significantly by the centrifugal forces cause d by the flow

94
rotation. These particles leave the system by what might be calle d a short circuiting of

the separation process. The majority of the short circuiting of particles occurs, in this

example, from a time of about 0.1 second to about 0.4 or 0.5 seconds. This is a fairly rapid

decrease in mass and is most apparent in Figure 5.12. After the initial exit of particles by

short circuiting, the rate of mass leaving the system begins to decrease and approaches an

asymptotic, steady-state value of about 0.9 as seen in Figure 5.12 at a time of about 12

seconds.

Cyclone Separation Process


15.1 m/s inlet velocity

1
0.9
C
o 0.8
+J
o
0.7
(0
1
0.6
0.5
to
</) 0.4
E 0.3
_a
> 0.2
o
'e 0.1
ra
Q_ 0

0 4 6 8 10 12 14
time (sec)

Figure 5.12 Overview of separation process

95
The blue curves in Figures 5.12 and 5.13 show how particle mass changes within

the dust bin. At a time equal to zero particles begin to enter the system but do not reach

the dust bin imme diately. This can be seen most clearly in Figure 5.13 where the dust bin

mass is zero until about 0.14 seconds. The dust bin mass then increases and peaks at a

time of approximately 1.0 second. After 1.0 second, the curve for dust bin mass begins to

follow the curve for total system mass and continues in this manner for the remainder of

the simulation. Also, note that in Figures 5.12 and 5.13 there are notations referring to

Figures 5.14 through 5.21 in which plots of particle locations in the cyclone at different

times is shown.

Early Part of Separation Process


15.1 m/s inlet velocity

o
CO
M

U)
co
E
_C
D
O
CO
Q.
0 0.2 0.4 0.6 0.8
time (sec)

Figure 5.13 Early part of separation process

96
Figure 5.14 Particle injection at 0.04 seconds

97
Figure 5.15 Short circuiting of particles at 0.14 seconds

98
I .

15. 0 0e-06

4.52e-
4.0 5e~
0 6
06
3.57e-

0 6 3. 1 0

e- 0 6

2.62e-

06
7
2.1 5e~
V

0 6 1. 6 7
Particles colored by
diameter (meters)
e - 0 6

1Figure
. 2 05.16e Short
- circuiting an d filling of dust bin at 0.24 secon ds

0 6
99
7.25e-

07
100
I
5.0 0 e-
0 6
4.52e~0
6
4.0 5e-0
6
3.57e~0
6
3.10 e- .*

0 6
>
I
2.62e- Vv **
06 Vir
- vf
2.1 \ ' s.: * :
1? %
5e~06
-- '
v -C r .r
Is m
y&fr

7
1.2 0 e-
0 6 %
Y Vi,

7.25e-
07 Particles colored by
diameter (meters)

Figure 5.18 Process at 0.44 seconds

101
102
5.0 0e-
06
4.52e-

06

4.0 5e-

0 6

3.57e-

0 6 3. 1 0

e- 0 6

2.62e-

06

2.1 5e-

0 6 1. 6 7 Figure 5.20 Process at 6.0 seconds

e- 0 6
103

1.2 0 e-
5.0
0e~06
4.55e-
0e
4.10 e -
0 6
3.65e-
06
3.20e-
0E

1.8 5 e
- 01
7
1.4 0 e
- 01
Y

Particles colored by
5.0 0 e diameter (meters)
- 0 7
Figure 5.21 Process at 12.5 seconds

104
The plots in Figures 5.14 through 5.21 are colored base d on particle size. For this

simulation particles ranging from 0.25 microns to 5 microns were used. The dust particles

had the properties of Ogawas Kanto-loam. As describe d in Chapter IV, particle sizes

were chosen to represent a particular size range from the Kanto-loam distribution

provided by Ogawa. One can also track particles of the different sizes over time. Figure

5.22 below shows how the number of particles in the system of different sizes changes

with time. One thousan d particles of each size were injecte d. Figure 5.22 plots the

fraction of particles that remain in the system over time, which is the fractional efficiency

of the cyclone for each particle size.


Figure 5.22 Cyclone fractional efficiency vs. particle size vs. time
105

Cyclone S eparation Efficiency vs Time & Size

^0.25 micron ^0.5 micron ^1 micron ^1.5 micron


^2.0 micron ^2.5 micron ^3 , 4 & 5 micron

0.0 2.0 40 6.0 8.0 10.0 12.0 140


time (sec)
As can be seen from Figure 5.22, the 3, 4 and 5 micron particles are all retained by the

system. A small amount of the 2.5 an d 2 micron particles escape early in the process

through short circuiting. After the initial short circuiting losses the remainder of these

particles are retained because they are too heavy to be re-entrained once they enter the dust

bin. A number of the 1.5 micron particles exit through short circuiting, after which there is

a slow decline in the number of particles in the system. It is somewhat unclear whether all

of the particles would eventually escape or whether a certain number will be retained

indefinitely. The 1 micron particles show a more rapid decline and by about 12 secon ds

only 9 % of the original number remain. The 0.25 an d 0.5 micron particles all leave the

system.

The fate of the 1.5 and 1.0 micron particles is not clear and several ways of dealing

with this situation were considered. One approach was to omit particles that did not

approach a single-value d result. A straight line would then be drawn between the known

points. This did not prove to be a very satisfactory method. For the purposes of this study,

it was decided to fit second order polynomials to the efficiency vs. time curves of particles

whose fates could not be determined with confidence during a reasonable time period,

which is determined by the investigator. The zero slope location of the polynomial would

be calculated and the collection efficiency at the zero slope location would be use d as the

estimate for the ultimate efficiency of the particle size in question. All points would then

be used to construct the grade efficiency curve. This method was used for the data

generated above.

106
Separation Efficiency vs Time& Size with Curve Fits to 1.0 and
1.5 micron Points

- 3 , 4 & 5 micron ------- Poly. (1 ------- Poly. (1.5 micron)


micron)

0.0 2.0 40 6.0 8.0 10.0 12.0 140


time (sec)
Figure 5.23 Cyclone fractional efficiency vs. particle size vs. time with
curve fitting
at 15.1 m/s
Grade Efficiency

4n

0Q-

0./
o

05
05
'
c
0044 -
RJ
03-

00-
0 50
2 0 3 tide size
0 0 1.0 0 4 ns)
Par (micrc

Figure 5.24 Cyclone grade efficiency curve at 15.1 m/s

107
Figure 5.23 shows the data which was plotte d in Figure 5.22 with the curve fits to the 1.0

an d 1.5 micron data. Base d on the results of the particle tracking process a grade

efficiency curve for this cyclone was plotte d an d is shown in Figure 5.24.

An obvious way to improve the calculation of the grade efficiency curve would be

to run the simulation for the above example for a longer period of time. The simulation

described above, which computed about 12.8 seconds of the process took a significant

amount of actual time. Using a computer running Windows XP using a single Intel Xeon

3.20 MHz processor with 2 GB of RAM took 20 secon ds of wall clock time to calculate a

simulation time step size of 0.0002 seconds. Using this information the total wall clock

time for running the 12.8 simulation can be shown to be

^ ,. , , , * 20 ( clock
sec) * (time step)
Total time = 12.8= simulated sec)* --------------------- * ------------------------------
{time (5.8)
step ) 0.0002 (simulated sec)
= 1,280,000 (clock sec) ~ 14.8days

This amount of time is not practical for use as a design tool. The next section describes a

more efficient approach for this type of calculation.

5.7 Quasi-Unstea dy Particle Tracking Calculations in a "Frozen" Flow Field

In the above process, the greatest amount of time was spent during the particle

tracking process as both unsteady particle and unsteady fluid calculations were carried out

108
until most of the particles had exited the domain. An alternative approach was

Figure 5.24 Cyclone grade efficiency curve at 15.1 m/s

109
adopte d, which significantly reduced the time required for the particle tracking process

but gave comparable collection efficiency results to the approach describe d above in

section 5.5. In the same manner as was done in section 5.5, a flow rate was specified and

unsteady calculations were performe d until the solution reache d a stable, periodically

steady characteristic. At this point, the unsteady fluid simulation was stopped. Using what

might be calle d a frozen continuous phase flow field, particles were injecte d into the

frozen fluid domain and tracke d until their fate could be determined. The results of the

approach are shown below in Figures 5.25 and 5.26.


Collection Efficiency vs time & size
Tracking in 'frozen' flow field with alternate tracking approach

0.25micron (1 min) 0.5micron (1 min) lOmicron (9 min)


1.5micron (42 min) 2.0micron (20 min) 2.5micron (17 min)
- 3, 4 & 5micron (4 min ) -------- Expon. (lOmicron (9 min)) --------- Poly. (1.5micron
(42 min))

Figure 5.25 Collection efficiency vs. particle size vs. time using alternate approach

110
The graphs in Figures 5.23 an d 5.25 were pro duce d using the different tracking

approaches and show the rate at which the particles of different sizes leave the system. The

general shapes of the curves for corresponding sizes are similar, however, the particles

tracked using the frozen flow field approach leave the system at a slower rate than the

particles tracked in the fully unsteady process. It is believed that the particles take longer to

leave the in the frozen flow field approach at least partly because the variation with time of

the mean flow component of velocity has been eliminated from the tracking computation.

The time dependent fluctuations in the mean flow would aid in dispersing the dust particles

so that the particles would have a more opportunity to escape the system, compared to the

particles tracked using the frozen flow field approach.

Grade Efficiency Curves for Both Tracking Approaches


--fully unsteady eff - quaz-unsteady eff
unsteady cut size = 1.274 quaz-unsteady cut size = 1.396

m _ r 1 11 11

r
0.0 1.0 2.0 3.0 4.0 5.0
Particle size (microns)

Figure 5.26 Gra de efficiency curves comparing both tracking approaches

111
The advantage of the frozen flow field tracking is that the grade efficiency curve can be

constructed in much less time. For this example the time needed to track the particles using

fully unsteady calculations was on the order of days, where, for the frozen field approach,

the time nee de d was less than two hours. The graph in Figure 5.27 shows the grade

efficiency curves for both approaches. The resulting grade efficiency curves are not

identical but, in general show the same trends. The cut size was estimate d for both curves

and these points are plotted also. As a result of the work summarized in the previous

sections, it was decided that, for the purposes of this study, the alternate or quasi-unsteady

approach would be used because of the unreasonably long calculation times required by the

fully unsteady approach. The following procedure would be used for calculating the grade

efficiency curve of cyclones for the rest of this study:

1. After setting up the geometry an d grid for the cyclone to be mo dele d, a flow rate

is specified and the simulation is run until a stable solution characteristic has been

reache d. The unsteady flow computations are then stopped.

2. Particles of appropriate sizes are created and tracke d through the frozen flow

field for a long enough period that the collection efficiency values for all particles

within reason reach a single-valued result.

3. Secon d order polynomials will be fit to the efficiency vs. time curves of particles

whose fates could not be determined with confidence during a reasonable time

period. This time period is determine d by the investigator.

4. The zero slope location of the polynomial (or minimum value of the function)

would be calculated and this minimum collection efficiency value would be used

112
as the estimate for the ultimate separation efficiency of the particle size in question.

5. All points would then be use d to construct the gra de efficiency curve.

5.8 Sources of Error and Grid Refinement Stu dy

The following section discusses several sources of error that arise in numerical

models. A grid refinement study is then use d to estimate the discretization error.

5.8.1 Sources of Simulation Error

There are several sources of error that affect the results of computations done using

CFD mo dels. Oberkampf an d Blottner (1998) discuss two main categories of errors. One

category is called physical modeling errors. These errors are caused by the inaccuracies in

the mathematical models used to describe the physics being simulated. Included in this

type of error are errors in the partial differential equations (PDEs) describing the flow,

auxiliary (or closure) physical models, for example, the turbulence model use d. Also the

boundary conditions for the PDEs can be sources of error. The choice of numerical models

described earlier in this study was done in a careful manner so as to re duce physical

modeling errors to the extent possible while still being able to complete the calculations

within a reasonable time with the computational tools available. The other category of

simulation error that is related the numerical aspects of the problem

113
include discretization, incomplete convergence and numerical round-off errors. The largest

contributor to numerical solution error, according to Oberkampf and Blottner (1998), and

the one that has caused the most inaccuracy in CFD solutions is caused by inadequate grid

resolution. In order to quantify this grid resolution or discretization error a grid refinement

study has been conducted and is discussed in the following section.

5.8.2 Estimating Discretization Error using a Grid Refinement Study

The purpose of a grid refinement study is to estimate the grid convergence

accuracy of the CFD models used in this work. Richardson extrapolation (Richardson,

1908; Richardson, 1927) has been use d by a number of researchers for this purpose

(Hutton & Casey, 2001; Oberkampf & Blottner, 1998; Roache, 1994 an d Celik &

Karatekin, 1997). Roy (2003) shows that when a differential equation is solve d

numerically, the discretization error (DE) on a mesh with grid density level k can be

written as

DE
k fk fexact (5.9)

where fk is a discrete solution value on mesh level k and fexact is the exact solution to the

continuum partial differential equation. If the discretization error is represented as a series

expansion, which is substituted into equation (5.9) and the terms are rearranged, we can

write the following expression:

fk = /exact + Sfk + S 2 @1 + S 3 @k + 0(K ) (5.10)

114
where g is the i-th order error term coefficient and hk is a measure of the grid spacing on

mesh k. In the case of a formally second order scheme and the use of central differencing

the g9 coefficient will be zero (Roy, 2003). The general procedure for the grid refinement

stu dy is to write equation (5.10) for a number of different mesh levels an d then to solve

the system of equations for an approximation to fexact. The value of fexact is then an estimate

of the solution extrapolated to a mesh, in which the grid spacing, hk is zero. Having an

estimate of fexact allows one to estimate the error in the solution by comparing the difference

between fk and fexact.

Roache (1994) presente d a generalize d expression to estimate fexact without

assuming the absence of odd powers in the expansion. His generalized expression applied

top-th order methods and an r-value for the grid refinement ratio as follows.

fexact f 1 + a - A)
(rp -1 (5.11)

where )i is the fine grid solution an d f is the coarse grid solution. In this expression, fexact
2

is estimated as a correction the fine grid solution. If centered differences were used then

the extrapolation is (p+2) order accurate. But, generally, and notably if upstream- weighted

methods are used, the extrapolation is (p+1) order accurate. Roache also defines the actual

fractional error, 4 of the fine grid solution as


1

(Tifexact 1
fexact (5.12)

Roache gives an expression for the estimated fractional error, E for the fine grid solution, )
1 1

as

115

P
(r -1) (5.13)

where

= (L - /i)

/i (5.14)

He describes E1 as being an ordere d error estimator since it takes into account the order, p

of the numerical method as well as the grid refinement ratio, r and that it can be a good

approximation of the fine grid error when the solution is of reasonable accuracy, i.e.,

when Ei << 1. He states that the non-ordered error, is generally the quantity which is

commonly reported in grid refinement studies, but that it is not a good error estimator by

itself since it does not take into account the order of the method or the grid refinement

ratio. Roache then proposes a Grid Convergence Index (GCI). The idea of his GCI is to

approximately relate the obtained by whatever grid refinement study is performed (what

ever the p and r values used) to the same that would be expected from a grid refinement

study of the same problem with the same fine grid except having used p = 2 an d r = 2, i.e.,

a grid doubling with a secon d order metho d. He defines the GCI for the fine grid solution

as
3
[ ]
GCI finegfid
(r -9)
(5.15)

When a refinement study is done using grid doubling (r = 2) with a second-order method

(p = 2), one obtains GCI [ fine grid] = . Roache also recognizes that it is not always

practical to perform all computations for an engineering numerical study on the fine grid.

He derives and expression for fexact as a correction to the coarse-grid solution as

116
f
J exact
(r -1) (5.16)

From this expression, he defines coarse-grid GCI, which can be presented in

a forms as shown below. several

3 \d rp
GCI [coarse grid] = --
(rp -1) (5.17)

GCI [coarse grid] = rp * GCI [ fine grid] (5.18)

GCI [coarse grid ] = GCI [ fine grid ] + 3 |f| (5.19)

In the present study both second-order and third-order methods are used. As

a result the numerical models are consi dere d to be mixed order. For mixed order

numerical models Roy (2003) presents approaches for estimating grid convergence

errors, which involve performing more than two refinement grids. This provides

additional equations of the type shown above as equation (5.10). By performing

these ad ditional stu dies one can calculate an estimate of the observed order of the

mixed order scheme. For example if the numerical model used both second and

third order methods, the observed order of the model should be somewhere

between second and third-order. Roache however, suggests that for mixed order

models, the error estimates should be reported conservatively as being the lowest

order metho d in the mo del. For example, in the case of methods which use higher

order schemes for advection than for diffusion, the error will be dominated

asymptotically by the lower order term. The present model uses secon d-order

schemes for time discretization an d for diffusion an d third order for


advection. Therefore, for this exercise a conservative value of p = 2 will be chosen
for
116
the purpose of the error analysis. Also, Fluent does provide validation studies

showing that its second-order numerical schemes are indeed second-order accurate.

Because the cyclone problem is three dimensional and uses an unstructured

type mesh. It is not straight forward to create a fine grid model that has a refinement

ratio, r that is exactly two, for example. In this case, Roache suggests that the

effective refinement ratio be reporte d in terms of the total number of cells used in

the coarse (N2) and fine (N9) grids. along


( 5 Vd v N
2
with the dimensionality, (D) of the
j
problem as

r = r effective

(5.20)

The cyclone grid was create d basically by using an unstructure d two dimensional

hex grid on a horizontal plane at the bottom of the exit pipe. (See section 5.3.1 an d

Figures

5.4 an d 5.5.) Then, this basic 2D grid was extru de d in the axial direction to create

most of the 3D grid. Of course, the exception to this was for the transition from the

inlet pipe to the cyclone body, which was created with tetrahedral cells. The

approach used to create the fine grid model was to mesh the starting 2D horizontal

surface with unstructured cells half the size of the cells used in the coarse grid. This

was done so that to the extent practical there were four times the number of 2D cells

for the fine as compare d to the coarse grid. See Figure 5.27 below for the starting

117
2D grid surface for the fine grid.

118
Figure 5.27 Starting grid surface for fine grid discretization error stu dy

The grid spacing in the extruded (axial) direction could be controlled precisely and

thus set up having exactly twice the number of grid points compare d to the coarse

grid. For the tetrahedral transition section, the target cell size for the fine grid model

was set to be one-half the target size used for the fine grid model. The next two

Figures show the complete d fine grid mo del.

119
Figure 5.28 Complete d fine grid cyclone mo del mesh (See Figure 5.29 for a
close up view of the mesh.)
As a result, the fine grid rnesh contained 447,665 cells. The coarse grid model contained
60,3/7 cells. From these results, the
effective refinement ratio, re could be
calculate d as
447,665
60,327
= 1.95

(5.21)

The order, p then used for the error estimates was chosen, as indicated above,

conservatively as p = 2. From this information discretization error estimates could be

calculated for cyclone static pressure drop, overall separation efficiency and particle cut

size as shown below:

120
Figure 5.29 Close up of complete d fine grid
cyclone mo del mesh

5.8.3 Results for the Coarse and Fine Grid Models at 15.1 m/s

Having run both the fine and coarse grid models at 15.1 m/s inlet velocity,

the pressure drops, grade efficiency curves and absolute efficiencies were calculate

d using the procedures described in the previous sections of this document. Table

5.3 below provides a summary of the results. Figure 5.30 which follows is a plot

that also shows the resulting grade efficiency curves. It is important to note that the

refinement was done in the three space coordinates and also in the time dimension.

The grid was refine d by a factor of two (1.95 was the effective refinement ratio).

The time step size was also refine d by a factor of two compare d to that use d in the

121
coarse grid mo del.

122
Table 5.4 Fine and coarse grid performance results for 15.1 m/s inlet
velocity
Quantity Cut Size Absolute Efficiency Pressure Drop
(microns) (fraction) (Pa)
fine grid result (447
1.440 0.884 1544
,665 cells)
coarse grid result 1.396 0.887 1578
(60,327 cells)

Fine vs. Coarse Grid Grade Efficiency Curves at 15.1 m/s fine grid abseff:
0.884 coarse grid abseff: 0.887

Fine Grid Coarse Grid o fne grid cut size: 1.440 o coarse grid cut size:
1.396

1 n 1 ------------- ------- -------


n Q

n ft
>
O0
0 7
7 --
0) 0 6
<D 0 6
0.5
c
0 044
(15
fc n ft J

n 1

0 0.5 1.0 1 5 2 02 5 3 03 5 4 04 5 5 05
0.0 - 0
particl size icron
e (m s 5
Figure 5.30 Coarse vs. fine grid grade efficiency curves at 15.1 m/s

From these computed results, the discretization errors for pressure drop, particle cut

size and absolute collection efficiency can be calculated. These calculations are

shown in the following sections.

123
5.8.4 Static Pressure Drop Discretization Error
Estimate

Using a fine grid static pressure drop, /i = 1544 (Pa), a coarse grid static pressure

drop, /2= 1578 (Pa), an order, p = 2 an d a refinement ratio of r = 1.95, the

following can be compute d. Using equation 5.11 an d estimate of fexact can be

calculate d as (1544 -1578)


1 C / I / I
f
J exact = 1544 + ^ ---- ----- - _
1532 (5.22
(1.952 -1) )
Using equation 5.14 a non-ordere d fine grid error, can be
calculate d as
_ (1578 -
0.022
1544) _1544 0 (5.23
)
Using equation 5.15 a fine grid GCI can be
calculate d as
3 0.022 !
L_ 0.0236 (1.952 -
1) (5.24
)
Using equation 5.19 a coarse grid GCI can be calculate d
as

GCI [coarse grid] _ 0.024 + 310.022 _ 0.0896 (5.25


)

5.8.5 Particle Cut Size Discretization Error Estimate

Using a fine grid cut size, f1 = 1.440 (microns), a coarse grid cut size, f2=

1.396 (microns), an order, p = 2 and a refinement ratio of r = 1.95, the following

can be compute d. Using equation 5.11 an d estimate offexact can be calculate d as

124
fexact 1.440 (1.440 - 1.45
+ 1.396) 6 (5.26
(1.952 -1) )
Using equation 5.14 a non-ordere d fine grid error, can be
calculate d as
_ (1.396 -
-
1.440) _1.440 0.0305 (5.27
)
Using equation 5.15 a fine grid GCI can be
calculate d as
31-
GCI [ fine grid ] 0.032
_ 0.0305| 7 (5.28
(1.952 -1) )
Using equation 5.19 a coarse grid GCI can be calculate d as

GCI [coarse grid] _ 0.0327 + 31- 0.03051 _ 0.124


(5.29
)

5.8.6 Cyclone Absolute Efficiency Discretization Error


Estimate

Using a fine grid absolute efficiency,/1 = 0.884, a coarse grid absolute

efficiency, f2= 0.887, an order, p = 2 and a refinement ratio of r = 1.95, the

following can be compute d. Using equation 5.11 an d estimate of fexact can be

calculate d as (0.884 -
f 0.884 0.88
exact
+ 0.887) 3 (5.30
(1.952 -1) )
Using equation 5.14 a non-ordere d fine grid error, can be
calculate d as
_ (0.887 -
0.003
0.883) _0.883 3 (5.31
)
Using equation 5.15 a fine grid GCI can be
calculate d as
3|0.0033|
GCI [ fine grid ] 0.003
_ (1.952 -1) 5
(5.32)
125
Using equation 5.19 a coarse grid GCI can be calculate d as

GCI [coarse grid] = 0.0035 + 3|0.0033| = 0.0132 (5.33)

5.8.7 Summary of Grid Refinement Discretization Error Calculations

Table 5.5 summarizes the above calculations.


Table 5.5 Discretization error calculations summary for 15.1 m/s inlet velocity
case
Absolute
Cut Size
Quantity Efficiency Pressure Drop (Pa)
(microns)
(fraction)
fine grid result
1.440 0.883 1544
(447,665 cells)
coarse grid result 1.396 0.887 1578
(60,327 cells)
estimate of exact
value,
1.456 0.883 1532
f exact
relative error, e -0.0305 0.0033 0.0220
GCIQfine grid] 0.0327 0.0035 0.0236
GCIQfine grid] % 3.3% 0.3% 2.4%
GCIQcoarse grid] 0.124 0.0132 0.0896
GCIQcoarse grid] % 12.4% 1.3% 9.0%

As was discussed previously, the quantity fexact is an estimate of the asymptotic

value that the numerical solution would approach as the grid was refined until the

computational cells approached a very small size (close to zero). By comparing the

solution obtained from the existing grid to the asymptotic value, a judgment can be made

as to how much error the model has due to discretization error. Because numerical

methods can use discretization schemes with that have different levels of truncation error

or order of accuracy an d because grid refinement ratios can vary, Roache (1994)

proposed his Grid Convergence Index (GCI). The GCI takes into account the actual grid

126
refinement ratio used and the actual order of the discretization schemes used by a

researcher and presents the results in a uniform manner as if it had been performed

using grid doubling and second order accurate numerical schemes. In this way the

grid refinement studies done by different researchers can be compared fairly with

each other. There was no specific range of CGI values given, by the authors of the

references reviewed by this writer, which was considered either good or poor

for a given numerical model. It would seem that a user would have to make a

judgment for himself as to whether the discretization error reflected in the GCI

values of his model was acceptable for the purposes of his investigation. For the

purposes of the present study the GCI values obtaine d were judged to be

acceptable. For example, the coarse grid GCI for cut size was the largest calculated

and had a value of 12.4%. For a cut size of 1.5 microns the error based on a GCI of

12.4% would be less than 0.2 microns. This amount of error was considere d to be

acceptable for this work.

5.9 Results for 150 mm Cyclone at Three Flow Rates with Discretization Error

Simulations for the three flow rates of interest in this study were conducted.

Calculations of the performance quantities of interest, cut size, absolute efficiency

and pressure drop were made. The GCIQcoarse grid] values from the previous

section were use d as estimates for the numerical error for these computations.

Table 5.6 shows these values along with the grade efficiency data points from the

127
simulations.

128
Table 5.6 Summary of 150 mm cyclone calculation results
Particle size Fractional efficiency Fractional efficiency Fractional efficiency
(microns) at 10.3 m/s inlet at 15.1 m/s inlet at 19.7m/s inlet
velocity velocity velocity
0.5 0.000 0.000 0.000
1.0 0.000 0.000 0.000
1.5 0.110 0.631 0.962
2.0 0.907 0.968 0.989
2.5 0.972 0.996 1.000
3.0 0.991 0.998 1.000
4.0 1.000 1.000 1.000
5.0 1.000 1.000 1.000
Cut size (microns) 1.75 1.40 1.26
Error estimate using
GCIQcoarse] for cut 0.22 0.17 0.16
is 12.4%
Absolute efficiency
0.867 0.887 0.899
(fraction)
Error estimate using
GCIQcoarse] for 0.011 0.012 0.012
Abs. eff. is 1.3%
Static pressure drop
719 1578 2723
(Pal
Error estimate using
GCIQcoarse] for 65 142 245
press. drop is 9.0%
Calculated Grade Efficiency Curves

10.3 m/s 15.1 m/s 19.7 m/s

1 1 --------


1 0 4

0 3
3 -

0 . 0
2 0 3 article 0 4 jter 0 5
0
. 0 1 . 0 0 6 . 0
P Diam (micron s)

Figure 5.31 Grade efficiency curves at three inlet velocities from calculations
126
5.10 Chapter Summary

A description of the grid generation process for the cyclone was given.

Guidelines for cell size and aspect ratio limits were discussed. An acceptable grid

for cyclone modeling can be made using hexahe dral cells to the extent possible

depending, primarily on the geometry of the system where the inlet pipe enters the

body of the cyclone. To resolve the cyclone core adequately a cell size of 3.5% of

the cyclone diameter is recommende d in the center of the cyclone in a horizontal

plane cut through the system. The cell size can then be smoothly increased to 7.0%

of the cyclone diameter in the outer regions of the model. The axial height of the

cells is limited by the requirement that the cell aspect ratio be less than or equal to

5.

Initial simulations using fully unsteady continuous phase computations and

fully unsteady particle tracking were performed. While these simulations helped

give insight into the separation process, it was found that the amount of time

require d to run a single model took on the order of two weeks. This was

unacceptable for design purposes in a typical industrial engineering environment.

An alternate approach was adopted which shortene d the time needed significantly,

to on the order of hours, while giving similar results as the fully unsteady method.

Discretization error was discussed, a fine grid mo del of the cyclone was

built and a grid refinement study was conducted. Discretization error estimates for

127
cut size, absolute efficiency and pressure drop were calculated.

128
Simulations of the cyclone at three inlet velocity conditions were run.

Grade efficiency curves, cut sizes, absolute efficiency and pressure drop for the

three models were calculated. Discretization error results were used to estimate the

error in the computed quantities.

129
CHAPTER VI.

RESULTS AND DISCUSSION

6.1 Intro duction

In this chapter the numerical results will be compared to the experimental

results. Differences in the results will be discussed and reasons for any

discrepancies will be given. Suggestions for possible future work will be given.

Note that Ogawa did not provide experimental pressure drop data for these cases.

Therefore pressure drop comparisons cannot be made.

6.2 A Comparative Study: Numerical Results and Experimental Validation

The following section provides a comparison of the experimental and

numerical results from this stu dy.

6.2.1 The 10.3 m/s Inlet Velocity Case

Table 6.1 compares the experimental and calculated results for the 10.3 m/s

inlet velocity operating con dition. Below the table is Figure 6.1, which shows a

130
Ogawa 150mm Cyclone Grade Efficiency Experimental vs. Calculated
Results: 10.3 ms Inlet Velocity

O experiment calculated - - - Ogawa curve fit


graph of the O exp. cut size: 2.00 A calc, cut size: 1.75

131
resulting grade efficiency curves. The curve fit through the test points provided by

Ogawa as well as the experimental and calculated cut diameters are also plotted.

Table 6.1 Experimental vs. calculated efficiency at 10.3m/s inlet


velocity

Particle size (microns) Calculate d Experiment

0.50 0.000 n/a


1.00 0.000 n/a
1.50 0.110 n/a
2.00 0.907 0.556
2.50 0.972 n/a
3.00 0.991 0.662
4.00 1.000 0.828
5.00 1.000 0.888
6.00 (1.000) 0.911
7.00 (1.000) 0.914
8.00 (1.000) 0.910
9.00 (1.000) 0.910
10.00 (1.000) 0.922
11.00 (1.000) 0.910
Cut size (microns) 1.75 2.00

Cut size error estimate


0.22 0.112
(microns)

Absolute Efficiency (fraction) 0.867 0.888

Absolute Efficiency error 0.011 0.050


estimate (fraction)

132
Ogawa 150mm Cyclone Grade Efficiency Experimental vs. Calculated
Results: 10.3 ms Inlet Velocity

O experiment calculated - - - Ogawa curve fit


O exp. cut size: 2.00 A calc, cut size: 1.75

1 1 1 1

r, .1
i

1
1
1
5J

I
\}

t
i
i -

r
/. 2 i
C
\ r h hed line is
X Ogawa ' s which
/ c eletermine
das
he
the used to
experimental
c ; ur
r ;ve fiut size. It is included
# ' here or reference.

0.0 1.0 2.0 3.0 4.0 50 50 7.0 50 9.0 10.0 11.0


Particle S ize (microns)

Figure 6.1 Experimental vs. calculate d grade efficiency for the 10.3 m/s case

6.2.2 The 15.1 m/s Inlet Velocity Case

Table 6.2 compares the experimental and calculated results for the

15.1m/s inlet velocity operating condition. Below the table is Figure 6.2, which

shows a graph of the resulting grade efficiency curves. The curve fit through the

test points provided by Ogawa as well as the experimental and calculated cut

133
diameters are also plotted.

134
Table 6.2 Experimental vs. calculated efficiency at
15.1m/s
inlet velocity
Ogawa 150mm Cyclone Grade Efficiency Experimental vs. Calculated
Results: 10.3 ms Inlet Velocity
Particle size (microns)
O experiment
Calculated
calculated
Experiment
- - - Ogawa curve fit
O exp. cut size: 2.00 A calc, cut size: 1.75

0.50 0.000 n/a


1.00 0.000 n/a
1.50 0.631 0.559
2.00 0.968 0.654
2.50 0.996 n/a
3.00 0.998 0.818
4.00 1.000 0.916
5.00 1.000 0.956
6.00 (1.000) 0.966
7.00 (1.000) 0.974
8.00 (1.000) 0.977
9.00 (1.000) 0.976
10.00 (1.000) 0.983
11.00 (1.000) 0.991
Cut size (microns) 1.40 1.50

Cut size error estimate


0.17 0.03
(microns)

Absolute Efficiency
0.887 0.914
(fraction)

Absolute Efficiency error


0.012 0.056
estimate (fraction)

135
Ogawa 150mm Cyclone Grade Efficiency Experimental vs. Calculated
Results: 15.1 ms Inlet Velocity

O experiment calculated - - - Ogawa curve fit


O exp. cutsize:1.50 A calc, cut size: 1.40

Particle S ize (microns)


Figure 6.2 Experimental vs. calculate d grade efficiency for the 15.1 m/s
case

6.2.3 The 19.7 m/s Inlet Velocity Case

Table 6.3 compares the experimental and calculated results for the

19.7m/s inlet velocity operating condition. Below the table is Figure 6.3, which

shows a graph of the resulting grade efficiency curves. The curve fit through the

test points provided by Ogawa as well as the experimental and calculated cut

diameters are also plotted.

136
Ogawa 150mm Cyclone Grade Efficiency Experimental vs. Calculated
TableResults: 19.7 ms Inlet Velocity
6.3 Experimental vs. calculated efficiency at 19.7m/s inlet velocity

O experiment calculated - - - Ogawa curve fit


O exp. cut size: 0.90 A calc, cut size: 1.26
Particle size (microns) Calculated Experiment

0.50 0.000 n/a


1.00 0.000 0.686
1.50 0.962 0.775
2.00 0.989 0.891
2.50 1.000 n/a
3.00 1.000 0.945
4.00 1.000 0.978
5.00 1.000 0.989
6.00 (1.000) 0.992
7.00 (1.000) 0.993
8.00 (1.000) 0.986
9.00 (1.000) 0.990
10.00 (1.000) 0.995
11.00 (1.000) 1.000
Cut size (microns) 1.26 0.90

Cut size error estimate (microns) 0.16 0.059

Absolute Efficiency (fraction) 0.899 0.956

Absolute Efficiency error


0.012 0.107
estimate (fraction)

137
Figure 6.3 Experimental vs. calculate d grade efficiency for the 19.7 m/s
case

6.3 Observations

The following section lists observations made as a result of comparing the

experimental results with the calculated results. This is done with respect to cut

size and absolute efficiency.

138
6.3.1 Cut Size

Figure 6.4 shows the comparison of experimental and calculated cut size

Experimental vs. Calculated Cut S ize Comparison


experiment calculated
^2.5on- i 20
i

2
O
d) 15
4
rf
N
w
Ii
3
o 1 n
results with error bars using the error estimates done earlier in the study.

0.5
9 10 11 12 13 14 15 16 17 18 19 20 21
Inlet Velocity (rms)

Figure 6.4 Experimental vs. calculate d cut size comparison

It should be noted that Ogawa determined the experimental cut sizes by fitting a

curve through the test data and extrapolating to the 0.5 efficiency point. The curve

fits given by Ogawa are shown in the above Figures 6.1, 6.2 and 6.3. As seen in

Figure 6.4 there is more error due to discretization in the calculate d cut size results

compare d to the error that has been estimated for the experimental cut size results.

The error bands for the 10.3 and 15.1 m/s cases overlap indicating that the

experimental and calculated results are in agreement for these inlet velocity

conditions. The difference in the 19.7 m/s results is larger than the estimated errors

139
indicating that the experimental and calculated results are

140
not in agreement with each other. At 10.3 m/s the calculated cut size is smaller

than the experimental cut size. At 15.1 m/s the calculate d an d experimental cut

sizes have nearly the same values. At 19.7 m/s the calculate d cut size is larger than

the experimental cut size.

6.3.2 Ab solute Effi ci ency

Figure 6.5 shows the comparison of experimental and calculated absolute

efficiency results with error bars using the error estimates done earlier in the study.

In contrast to the cut size results, there is more uncertainty in the experimental

absolute efficiency results compared to the discretization error in the calculated

results. The error bands overlap in all three cases indicating that, within

experimental and discretization error estimates, the calculated and experimental

results are in agreement. At each inlet velocity, however, the calculate d efficiency

is less than the corresponding experimental efficiency. The under-prediction of

absolute efficiency is expected based on the calculated grade efficiency curves.

The calculated grade efficiency curves at all three inlet velocities pre dict zero

collection efficiency at particles less than 1.0 micron. See Figures 6.1, 6.2 and 6.3.

However, approximately 8.1% of the dust is in the 1 micron and smaller size range.

This means that the highest absolute efficiency the models (at these operating

conditions) could predict would be 0.919 or 91.9%.

141
Experimental vs. Calculated Absolute
Efficiency
Comparison
calculated experiment
1.00

0.95 <
O
0 90 < i
0
c
< >* t
)
o
H 085
S
3
0 80
.a
<

075
9 10 11 12 13 14 15 16 17 18 19 20 21
Inlet Velocity (m/s)

Figure 6.5 Experimental vs. calculate d absolute efficiency


comparison

6.4 Discussion

The following section discusses experimental and calculated results and

compares this information to the grade efficiency curve diagram of Stairmand

(1951). Differences between the experimental and calculated separation

performance is related to differences between the theoretical and actual grade

efficiency curves as explained by Stairmand in order to evaluate the performance

of the numerical mo del.

142
6.4.1 Review of the Stairmand Grade Efficiency Diagram from
Chapter I

To aid in the discussion of these results it seeme d appropriate that the

diagram given by Stairmand (1951), which was first presente d in Chapter I as

Figure 1.5 be redrawn below as Figure 6.6. Some insight into the causes of the

differences observe d in the calculated and experimental grade efficiency results

presented earlier in this chapter may be gained by referring to Figure 6.6.

Cut size: is the particle theoretical grade efficiency curve (solid


size at which eff is 50%
line)

100%

% zone of re duce d efficiency


removal at due to eddying, bouncing,
stated 50% etc.
particle size actual grade efficiency curve (dashed
line)

zone of increased efficiency due to


0 collisions, agglomeration, etc.
size of particles
increasing theoretical cut size

Figure 6.6 Diagram of grade efficiency curve characteristics from Stairmand


(1951)

The experimental data along with the Ogawa curve fits show similarities to the

dashed line representing the actual grade efficiency curve in Figure 6.6. The

curves are not strongly S-shaped. The cut sizes of the 150mm cyclone analyze d
experimental
in this
139
study are at small enough diameters that a definite inflection point at the cut size

along with a significant change in curvature between the cut point and the origin is

not evident. The form of Ogawas function is such that for cyclones with operating

con ditions resulting in larger cut sizes a definite S-shape will be apparent. The

knee of the experimental grade efficiency curves at particle sizes larger than the cut

size is very similar to the shape shown in Stairman ds diagram. The calculate d gra

de efficiency curves appear closer to the vertical theoretical grade efficiency

curve in Stairman ds diagram. The 19.7 and 15.1 m/s calculated curves do not

exhibit an S-shape but do have rounded knees in the curves at particle sizes

somewhat larger than the cut size. The calculated 10.3 m/s curve does have an S-

shape. The slope of the calculated curves are steep, that is, they have high

sharpness indices but are not vertical like the theoretical curve.

The purpose of going through this comparison, of the calculated and

experimental grade efficiency curves, an d to consider how they generally relate to

Stairman ds diagram, is to try to get an understanding of the possible ways in

which the computational model may be deficient. Then it will be possible to

identify options that could be investigate d which might improve the computational

mo del.

6.4.2 Separation Efficiency for Particles Larger than the Cut Size

Stairmand, from Figure 6.6, indicates that the reduction in theoretical

140
separation efficiency for particles larger than the cut size is due to effects that

include e ddying an d

141
bouncing. The computational model did show separation efficiencies less than 1.0

in this region, which this writer describe d as short circuiting and was discussed

in Chapter V. However, the computational model did not show efficiencies in this

region as low as those observed in the experimental data. Based on the numerical

model, this effect was found to occur early in the separation process as particles

first enter the cyclone body.

6.4.2.1 Turbulent Eddies

The effect of turbulent eddies on the dispersion of particles in the numerical

mo del is accounte d for to a certain extent by the use of the Discrete Ran dom

Walk (DRW) mo del describe d in Chapter III, section 3.4.2 along with the

Reynolds-stress turbulence model (RSM). Disadvantages of the present method

including the particle tracking approach include the following:

o Although the RSM is an improvement over e d dy viscosity type mo dels

using the Boussinesq assumption, the details of the turbulent eddies are not

resolved. The effect of the eddies on particle dispersion is a modeled

process and not calculated directly.

o The particle tracking approach, which was adopted because of time and

computational resource limitations, uses the quasi-unsteady tracking

calculations in a frozen flow field approach describe d in section 5.7. While

142
making the calculations more practical the frozen flow field removed the

time varying

143
aspects of the flow which, if present, may have resulted in more particles

escaping the system.

6.4.2.2 Particle Bouncing

Bouncing of particles at the walls are accounted for in the computational

model. The default boundary condition, which was used in this study, for discrete

particle collisions with walls is that the particles reflect off the walls perfectly,

retaining 100% of their pre-collision momentum. Bouncing or scattering of the

particles as a result of collisions of particles with other particles is not accounted

for.

6.4.3 Separation Efficiency for Particles Smaller than the Cut Size

The following section considers separation efficiency characteristics of the

experimental and calculated results for particle sizes which are smaller than the cut

size.

It is interesting to note that Ogawa had no experimental data points below the cut

size values he reporte d.

6.4.3.1 Particle Collisions and Agglomeration

Stairmand, from Figure 6.6, indicates that the increase in theoretical

144
separation efficiency for particles smaller than the cut size is due to effects that

include collisions an d agglomeration. For all three cases, the numerical mo del pre

dicte d a separation

145
efficiency of zero for all particles 1.0 micron and smaller. The particle tracking

method did not model any effects of particle-particle collisions or the

agglomeration of particles, so it is not surprising that the computed grade efficiency

curves would not exhibit features relate d to these effects. Fluent does have a

particle collision and agglomeration modeling capability which was investigated to

a great extent during this study. Unfortunately, a problem in the code was

discovered, which caused the program to lockup when using the collision model.

The problem was never resolve d and so it was not possible to evaluate the

collision model and investigate the extent to which it might be able to improve the

calculated results for this study. In Chapter II, section 2.1 the particulate loading

was calculated to be low. Therefore it was assumed that the particles could be

treated as being isolated from each other. If this assumption held true for the

cyclone flow process, then perhaps the use of a properly working collision model

would have not shown results much different from those of the present stu dy.

6.5 Suggested Future Areas of Investigation

There are several aspects of the numerical mo del, in this writers opinion,

that should be investigated and which may result in improvements to the separation

efficiency calculation results. These are discussed below.

146
6.5.1 Turbulence Modeling

Improvements in this aspect should result in better accuracy in modeling the

grade efficiency curve at particle sizes larger than the cut size diameter. It is

suggested that turbulence mo dels identifie d as type III an d discusse d briefly in

section 2.3.4 be investigate d. These are models not based completely on the

Reynolds-stress equation. The Large Eddy Simulation (LES) model, which

resolves the larger energy-containing eddies would be a good candidate if

appropriate computational resources were available.

6.5.2 Particle Collisions and Agglomeration

Improvements in this aspect of the model might result in better accuracy in

modeling the grade efficiency curve at particle sizes smaller than the cut size

diameter. This assumes that the chaotic flow in the cyclone provide d enough

opportunities for the particles to interact with each other in spite of the low

particulate loading calculated in section 2.1. Because Fluent already has a collision

modeling capability, additional efforts should be made to get the bug in this

feature corrected. Efforts were made to do this during this stu dy but were not

successful.

6.5.3 Fully Unsteady Continuous Phase and Particle Tracking Calculations

147
This was use d earlier in the stu dy an d was discusse d in section 5.6 but

was abandoned because of the long time periods needed to complete a simulation.

The

148
computer available for use during this study was about four years old. A more

capable computer would make the fully unsteady calculations run faster and so

might make the use of this approach more practical.

6.6 Chapter Summary

The numerical and the experimental results of the study were discussed in

this chapter. A diagram by Stairmand (1951) which explained how the shape of an

actual cyclone grade efficiency curve differs from and a theoretical grade

efficiency curve was use d as a reference. Processes that affect the shape of the

grade efficiency curve, which in turn affect the cut size and the absolute collection

efficiency include the collision and agglomeration of particles. Also affecting the

shape of the grade efficiency curve is the transport of particles due to the effects of

turbulent e ddies. Improvements in the ability of the numerical model to account

for particle collisions and particle agglomeration should help in calculating the

separation efficiency for particles smaller than the cut size. Improvements in the

modeling of turbulent e ddies should help to improve the separation efficiency

calculations for particles larger than the cut size.

149
CHAPTER VII.

CYCLONE DESIGN METHODOLOGY

7.1 Intro duction

This chapter describes a methodology for cyclone design based on the work

done in this study. It takes into consideration the significant amount of work that

has gone into the study of cyclone separators over a number of decades as well as

the powerful numerical methods available currently. A number of researchers have

propose d various empirical methods, which allow quick estimates of cyclone

performance to be made, yet can provide results with significant error if used for

cyclone geometries that are of a nonstandard nature. That is, the empirical methods

can provide misleading results if they are applied to geometries or operating

conditions outside the bounds of the original test data which was used to develop

those empirical models. Advanced numerical methods such as the one use d as the

basis for the work done in this study have a number of advantages over the

empirical models. For example, the numerical methods are able to provide much

more information about the details of the flows such as velocity and pressure

distributions and particle tracking capabilities within the cyclone model as well as

150
to calculate the overall performance characteristics of the device such as the grade

151
efficiency curve, cut size and absolute collection efficiency. The results of the

numerical methods are considere d to be more accurate over a broader range of

geometries and operating conditions, but more time is required to complete a

performance calculation than does an empirical mo del. The motivation for the

design metho dology describe d in this section is to take advantage of the best of

what each approach has to offer. The methodology uses an empirical correlation

from the work of Svarovsky (2000) to provide basic sizing and performance

information early in the design process. The output of the empirical model is a

description of initial geometry that provides approximately the desire d separation

performance and pressure drop. The detaile d numerical model is then use d to

refine the original geometry so that the final design specifications are met. Also use

d in the refinement process are guidelines on cyclone design taken from the

literature and base d on the experience of others in the field of cyclone design. This

makes the refinement process more efficient.

7.2 Design Methodology Flow Chart

The diagram in Figure 7.1 shows the main process steps of the design

methodology in the form of a flow chart. This diagram provides an overview of the

design process.

152
Figure 7.1 Cyclone design methodology flow chart

153
7.3 Description of Metho dology Process Steps for a Typical Design Cycle

Each activity step shown in the flow chart in Figure 7.1 has an identifying number

associate d with it. The process is described in more detail in Table 7.1. This is done in

the form of descriptions and instructions to be followed for a typical cyclone design

sequence.

Table 7.1 Design methodology process steps: descriptions & instructions

Process
Step Description Instructions
Step No.
Performance requirements for the design
are specified.
The needed information is gas properties Specify: ^ ps,^s,p, x5o, as well
1. and flow rate, particle density and
as constraints on system size
required cut size. If there are other
and allowable AP
constraints such as space limitations or
pressure drop limits, they should be
determined
This is a Decision Step: Do you have If you are starting from
existing geometry to analyze? scratch an d dont have an
idea of what geometry you
You have the option of getting help with a want to start with, the
new design so that with the basic answer is No an d you go to
performance requirements from Step 1, a Step 3 to use the Svarovsky
starting design can be calculated which sizing estimator.
shoul d give you approximately the
performance you need. If you have an existing
2. design that you wish to
If you have existing geometry then you modify or trouble shoot, etc.
will skip the estimation process and go on then the answer is Yes and
to creating the mo del. you go on to Step 3a. It is
assumed that you have the
information that describes
the device.

154
Table 7.1 Design methodology process steps: descriptions & instructions
(continue d>
Process
Step Description Instructions
Step No.
From the performance requirements,
this step uses the correlations and Appendix A gives the data
recommendations from L. Svarovsky and correlations presented
to define a starting design that will by Svarovsky. Review this
come close to satisfying the information. Appendix B
3. specifications. gives the equations and a
worked example of sizing
a cyclone for a given set of
requirements. These
equations can be put in a
spreadsheet very easily
Having an existing design
Step 3 a assumes you know or have in min d, gather the
available in some form the geometry to dimensions, CAD files, etc
3a.
be analyzed and you are ready to go on that give the detaile d
to step 4. geometric description of
the mo del you want to
analyze.
Once the geometry is known, the
cyclone needs to be defined in the
form of a 3D solid model. This can be
done with a CAD program such as
ProEngineer from Parametric 1. Create a 3D solid mo
Technologies. The solid model can del of the cyclone.
also be create d within Fluents
4. preprocessor/mesh generator program 2. Import the geometry
called Gambit. into the preprocessor and
After the solid model has been created create the grid using the
the grid is created in Gambit. NOTE: It instructions from Chapter
is not the intent of this study to teach V, Sections 5.2 and 5.3.
the basics of CAD or the details of the
CFD preprocessor. Information is rea
dily available on these subjects
through outside sources.

155
Table 7.1 Design methodology process steps: descriptions & instructions
(continue d>
Process
Step Description Instructions
Step No.
1. Set up boun dary
conditions per Chapter V,
sections 5.4 and 5.5 and
run solver until a stable
solution is obtained.
The mesh create d in Step 4 is importe 2. Create particles over
d into the CFD solver, boundary a range of sizes in 0.5
conditions are applied and a converged micron increments so that
solution is obtained. the grade efficiency curve
5. Particles are created and passed can be defined.
through the system an d the grade 3. Track particles through
efficiency curve, cut size, pressure the domain using Chapter
drop and other quantities of interest are V, section 5.7 as a guide.
calculated and plotted as neede d. 4. Create the grade
efficiency curve, calculate
the cut size, pressure drop
and other quantities that
you wish to use as part of
the design evaluation.
1. Compare the results of
Step 5 with the
requirements of Step 1.
Determine if the current model meets 2. If the requirements
the design requirements. If the answer have been met, you are
is Yes, the design process is finished. finished. STOP.
6. 3. If the requirements
If the answer is No, proceed to Step 7
to determine what revisions need to be have not been met, the
made. design needs to be
modified and another
analysis iteration nee ds to
be done. From here procee
d to Step 7.

156
Table 7.1 Design methodology process steps: descriptions & instructions
(continued)
Process
Step Description Instructions
Step No.
1. From the results of
Step 6 you have
identified
characteristics of the
system that are not
acceptable.
2. See Appendices C & D.
Consult the design guideline These provide suggestions
information in Appendix C & Appen for the types of changes
dix D. Other sources an d/or design that can be made to adjust
optimization methods can be used at the performance in the
7. this step as appropriate. The output of desired manner.
this step is a set of revised cyclone 3. Other sources of
dimensions that are expected to bring information can be used,
the design into agreement with the but guidelines in these two
requirements. Appendices should be
sufficient in most
instances.
4. At this point revisions
to the design for the next
iteration are specified.
5. Go to Step 4 to begin
preparing the revise d
model for analysis.

7.4 Detailed Cyclone Design Example Problem

The following section presents a worke d example problem for the design

of a cyclone using the proposed design methodology.

157
7.4.1 Step 1: Define the Design Requirements

A tangential entry single cyclone is to be designed for use for experimental

purposes. It will be used to measure collection efficiencies for droplets of a mineral

oil called Acroprime 200. The following specifications need to be met.

o Gas: air with density, Ps = 1 200islm , and viscosity, = 115e 5 k g f m s o The air flow
rate, Q = .14 m I s
1D0 m
o Particle material, Acroprime 200, a mineral oil with density, p ^ > ^sl
o Performance criteria is that cut size , *50 30 03 m/crons o There are
no requirements for pressure drop or overall cyclone size.

7.4.2 Step 2: Does Geometry Already Exist?

The answer to this question is No. A starting geometry will need to be

calculated. Therefore, procee d to Step3.

7.4.3 Step 3: Calculate Initial Geometry

Using the information found in Appendices A and B, the equations for the

calculation are written into a spreadsheet. Graph the resulting pressure drop vs.

diameter curve to help visualize the results. See Figure 7.2 for the spreadsheet

appearance as implemented for the Svarovsky (1992) design correlations.

158
7.4.4 Step 4: Create Mo del of the Geometry an d the Computational Grid

The output from Step 3 is the list of eleven cyclone dimensions, which may

be seen in the blue and yellow boxes in Figure 7.2. Use these dimensions to create

the geometry
CYCLONE_T/ ihare\DATA2\Working Datatskegglcyc lone\geo\CYCLONE. TJ tT.11 - Pro/ENGINEER

Figure 7.2 Spreadsheet to calculate initial cyclone dimensions


154
CYCLONE_T/ ihare\DATA2\Working Datatskegglcyc lone\geo\CYCLONE. TJ tT.11 - Pro/ENGINEER
Analysis Info Windo

H a QApplications
a w
||a[Fg a. n &- a: .sail s a a fall
|J X

Paramete
rs

CYCLONE_TAN_REC_297M CYCLONE_TAN_RE
M_EX1 BRIGHT C_297MM_EX1

O DTM1
JNLET_PIANE
Z7DTM2 Mar 17.2009
tangential, rectangular inlet
cyclone
Htot
al
q71 Extrude Hbo
2 cp Extrude dy
3 q71 Extiude Ho
4 tp Extrude one
5 ejd
Revolve 3
+ Insert Here
140
445.5
(1.5"Dbody)
297
(1.0Dbody)
D cyclone =
297 Htotal =
1337 Hbody
= 535

Hbin = 1.5*D
eye lone Dbin =
1.0eD eye lone

Diameter = 307.000
^Select a surface for Diameter analysis.
^Selectitems such as Dimensions, Sections, Trajectories, Tolerances, Surface Finish or
ether features for editing
Datum planes will not be displayed.

Figure 7.3 Cyclone solid model


geometry
Operation

|g m\ *i|
Bill

Global
Transcript Description Control
Active B | ffl | B | ffl | All |
. gnu. org/copyleft/lesse r.html) GRAPHICS
WIHD0W-
Command> window modify visible mesh QUADRAHT
Command) default set "GRAPHICS. GENERAL.
WIND0S_BACKGR0UBD_

Comman
d:

Figure 7.4 Preprocessor with complete d grid for the 297mm cyclone
model
155
in a 3D Solid Modeling CAD package. Figure 7.3 is a screen print of the 297 mm

diameter cyclone base d on the dimensions calculate d in Step 3. The solid mo del is

imported into the preprocessor where the grid is created. The grid is create d base d on the

guidelines given in Chapter V, sections 5.2 an d 5.3. Figure 7.4 is a plot of the

preprocessor user interface with the completed mesh for this cyclone.

7.4.5 Step 5: Solve Flow Problem an d Obtain Performance Information

Table 7.2 Material properties


Density (kg/m3) 1.200
Air
Absolute viscosity (kg/m-s) 0.0000185
Acroprime 200 Particle density (kg/m3) 860

The required flow rate was given as Q = 0.140 m3/s. A mass flow inlet will be use d

so the mass flow rate is calculated as


m = Qp = 0.140 m f s *1.200kgjm3 = 0.168 kg/s
3
(7
1)

The turbulence inlet boundary conditions are the turbulence intensity of 5.0% and the

hydraulic diameter of the inlet. The hydraulic diameter of a non-circular duct is calculated

as

4 * cross - sec tional area 4ab 4* 187mm *77mm


DH 109 mm
wetted perimeter 2 (a + b) 2 (187mm + 77mm) (7
2)
The time step size was 0.0008 seconds so that the number of solver iterations for each

time step was in the range of 5 to 10 in accordance to the findings in Section 5.5. The

156
solver was run such that a stable behavior was obtained. The convergence behavior

of the inlet and outlet static pressure was monitored. When the fluctuations in static

pressure stopped changing or changed in a stable periodic manner, the solver was

stoppe d. Figure 7.5 is a plot of the static pressure history for this problem.
Table 7.3 Flow boundary conditions
BC type Value
Mass Flow (kg/s) 0.168
Turbulence Intensity 5%
Inlet (%)
Turbulence
Duct Hy draulic 109
Diameter (mm)
No input value
Outlet Zero Gradient
required
Wall No-Slip No input value
required

pstat
pstat
80 00
ANSY
0
160 00
0
S
40 00
0
120 00
0
000 00

Average 80 00
of 0
Facet 60 00
0
Values 40 00
(pascal; 0
200 00

00

200 00
0000 500 0000 500 0000 500 0000 500
0 0 0 0
Flow Time

74600
cells
Convergence history of pstat-1 pstat-2 Apr 03,
S 297-7-3.
(Time=3.20Q8e+0u) 2008
FLUEN1 6.4 (3d. pbns, Rblvl,
^V9.7 unsteady;
Figure 7.5 Convergence history of static pressure during solution
process

157
At this time particles of the mineral oil material were create d an d injecte d

into the system using the quasi-unsteady tracking in a frozen flow field approach

as described in Section 5.7. Figure 7.6 shows the particle history leaving the

cyclone over time as describe d in Section 5.7.

S297 Vin=9.7: Collection Efficiency vs time & size Tracking in 'frozen' flow
field
^1.5 micron 1008(2 m i n ) 2micron 1008(3 m i n )
^2.5 micron 1008(4 m i n ) 3.5 micron 1008(19 m i n )
^ 4 micron 1008(27 m i n ) ^ 5 micron 1008(41 m i n )
^5.5 micron 1008(46 m i n ) ^4.5 micron 1008(64 m i n )
Poly. (4 micron 1008(27 min)) ----- Poly. (3.5 micron 1008(19 min))
Poly. (5 micron 1008(41 min)) ----- Poly. (4.5 micron 1008(64 min))

Figure 7.6 The 297mm cyclone particle tracking history with curve fits

From the information represented in Figure 7.6, the grade efficiency curve is

create d and is shown in Figure 7.7. The appearance of the grade efficiency curve

is not smooth. This is primarily caused by the curve fitting done with the 3.5, 4.0,

4.5 and 5.0 micron particles in order to estimate a final efficiency. In particular, the

4.5 an d 5.0 efficiency

158
verses time curves in Figure 7.6 are almost straight lines. This makes it difficult to fin d

the zero slope point for a second order polynomial fit to points that almost lie on a straight

line. One could choose not to use these points, but for this exercise all points are included.

Grade Efficiency Curve for Initial 297mm Cyclone Desiqn Row rate:
0.140 m3/sec; Static Pressure drop: 560 Pa Cut size: 4.5 microns;
Particle materia: mineral oil at 860 kq/m3

10
1
0.9
s. n fi
o

07
06
05
(C
n A

.4 0 3
0.3
2

0.2 -0 1 -
nn
304 07 09

00 1 0 20 0 5 time 0 6 (s) 08 0 10.0

Figure 7.7 Grade efficiency curve for initial 297mm cyclone design

7.4.6 Step 6: Compare Performance of Current Design with Requirements

Table 7.4 Comparison of performance of initial design with requirements


Required
Variable Name Initial Design Performance
Performance
Cut size (microns) 4.5 3.0 0.3
Pressure drop (Pa) 560
no constraint specified
Flow rate (m3/s) 0.140 0.140

159
Table 7.4 contains the results of the performance calculated in section 7.4.5, Step

5 along with the requirements set up in section 7.4.1, Step 1. The cut size is not within the

required range, therefore the design nee ds to be modified.

7.4.7 Step 7: Review Design Guidelines and Specify Modifications

The cut size needs to be decreased. No other performance variables need to be

change d. After reviewing Appendices C and D, one sees that in Appendix D, Table D.1,

the second row says that cut size is very sensitive to changes to the inlet duct height, a.

Also, in Figure D.1 the optimal curves relating a, 6 and De can be consulte d. Table 7.5

summarizes the decisions made here an d shows Leiths optimal ratios, the ratios use d in

the initial model and the new ratios to be tried in the revised design along with the

reasoning for choosing the values. The last column lists the changes to be incorporate d in

the revised design along with the reasoning for selecting the values used. Since the inlet

duct height was changed, it was decided to reduce the length that the exit pipe extended

into the cyclone, which is dimension S. S was shortene d by the same amount that a was

shortene d. This resulte d in a new value for S of 188mm. From the guidelines in

Appendices C and D, this change in S will not have a significant effect on the cut size.

Having deci de d on the changes to be made to the initial design and following the Flow

chart in Figure 7.1, we re-enter Process Step 4 to revise the solid model geometry and re-

mesh the model.

160
7.4.8 Re-enter Step 4: Revise the Geometry and the Computational Grid

The solid model is modified with the changes specified in Table 7.5. Figure 7.8 is a

screen print of the new geometry create d in the CAD program. Figure 7.9 is a screen

print of the new computational grid.

161
7.4.8 Re-enter Step 4: Revise the Geometry and the Computational Grid

Table 7.5 Comparison of design ratios for cyclone inlet and outlet ducts

Optimal ratios Ratios


Ratio from Leith, from Choice of Ratio to be use d in the Revise d
Name Appendix D, Initial Design
Figure D.1, at Design
x50 = 3.0 /am
The ratio a/D has a strong influence on x
50

therefore it was decided to reduce a/D by 25%


(and not make an extremely large change) for
the revise d design so set: a/D = 0.47, a =
140mm Because a is change d, it was decide d
0.35 to also shorten the exit pipe dimension S by the
(D=254mm) to same amount. The dimension a was shortened
a/D 0.63
~0.2 by a total of 47mm from 187 mm to 140mm
(D=318mm) (25% reduction). Therefore S was shortene d by
the same 47mm from the initial value of
235mm to the new value. So set: S = 235-47
=188mm

To move this ratio closer to the optimal, it was


b/D 0.2 0.26 decided to also reduce b/D by 25% for the
revise d design so set: b/D = 0.20, b =60mm

This ratio could be increased, which should


lower the pressure drop. However since there
D/D 0.56 0.47 was no constraint in the requirements, and for
the purposes of this exercise De/D was not
changed.
Keep: De/D = 0.47 , De = 140mm

162
Analys Windo
_shareVDATA2\Worl<lng&ata\skegg\cyclone\geolCYCLONE_T< naaifl --E
is w
QUA-a;
HHK
_R E C_297M
CYCLOMEJ M_EX1 _R E V*
a RIGHT
atop
O FRONT
^-
7 _C:
PRT
DTM1JNLET_PLAN
E 7 DTM2
Relations

qv1 Extrude
2 a? Extrude CYCLO N
3 cp Extrude E_TAN_REC_297MM_EX1_RE VI
4 ip Extrude
5 ojD
Revolve 3
4 Insert Here M ar 17.2008; Mar 18.2008
Revision #1 tangential
rectangular inlet cyclone

235; now shorten by 47 to


188 187; now use: 140
(reduction of 25%] 77; now
use: 80 (reduction cf 25%]

=A is total
height

Successfully changed to Z:\nco _share\D AT A21 Working


Datalskegg\cydone\geo\ directory. BDCould not read from Model Tree
config file "x:\skegg\tree.cfg".
Datum planes will not be displayed.

Figure 7.8 Solid mo del of revise d cyclone


geometry
Operation

\ m g]

Windows H ffl B ffl AH

r *j
P jj i
J Vertices
J B. Layers
*i
I^*J
J Visible

J Silhouette On ^ Off

J Render
J Lower topology

Global
Transcript Description Control
Active B | ffl | B | ffl | All |
. gnu. org/copyleft/lesse r.html)

Command) window modify shade


Command) default set "GRAPHICS. GENERAL.
WIND0WS_BACKGR01M]_

Comman
d:

Figure 7.9 Revised computational


grid
162
7.4.9 Re-enter Step 5: Calculate Performance of Revise d Cyclone

The same procedure is used as was shown in section 7.4.5 above except that

because the inlet duct dimensions have changed, the new hy draulic diameter for

the revise d model needs to be calculated as

^ 4 * cross - sec tional area 4ab 4 *140mm *


60mm
DH = -----------------------------------------= = ------------------------- = 14mm (7.3
wetted perimeter 2 (a + b) 2(140mm + 60mm) )

Table 7.6 Flow boundary conditions


BC type Value

Mass Flow (kg/s) 0.168


Inlet
Turbulence Intensity (%) 5%
Turbulenc
e Duct Hy draulic Diameter 84
(mm)
Outlet Zero Gra d ient No input value
required
Wall No-Slip No input value
required

The time step size was 0.0004 seconds so that the number of solver iterations for

each time step was in the range of 5 to 10 in accordance to the fin dings in Section

5.5. The solver was run such that a stable behavior was obtained. The convergence

behavior of the inlet and outlet static pressure was monitored. When the

163
7.4.9 Re-enter Step 5: Calculate Performance of Revise d Cyclone

fluctuations in static pressure change d in a stable periodic manner, the solver was

stopped.

164
pstat-2
__ pstat-1 _____
1400.00
ANSY
1200.00
S
1000.00

800.0
Average
of 0
Facet
Values
(pascal) 600.0

400.0
0.00
0. 0000 0.2500 0.5000 0.7500 1.0000 1.25001.5000 1.7500 2.0000
2.2500
0
Flow Time
200.00
75354 cells __
Convergence history of pstat-1 pstat-2 Apr 04. 2008
(Time=2.1410e+00) S 297-7-3.5R1 FLUENT 6.4 (3d, pbns, RSM,
unsteady):
Figure 7.10 Convergence history of static pressure for revise d cyclone
mo del
S297 Revision #1 Vin=16.7: Collection Efficiency vs time &size
Tracking in 'frozen' flow field
----- 1 micron 1008 (34min) 2 micron 1008 (49min)
^3 micron 1008 (76min) ^4 micron 1008 (l44min)
^5 micron 1008 (28min) ^6 micron 1008 *14min)
3.5 micron 1008 (69min) ----- Poly. (3 micron 1008 (76min))
----- Poly. (4 micron 1008 (144min)) ---- Poly. (3.5 micron 1008 (69min))

Figure 7.11 Collection efficiency vs. time and size for


revised cyclone design
164
Grade Efficiency Curve for Revise Row rate: 0.140 5d 297 mm C
m3/sec; Static Pres Cut size: 3.25 microns; >sure rop: /clone H0 Pa
Particles: min d ieral 10 l at kg/m3
oi 86(
in k __________________

nQ
n ft
n7

<v
E n fi -
n 5
rc
5n a
o
&n ft _
LL
n9
n1

nn
n.n 9 n A n8 n inn
n6
time (s)

Figure 7.12 Grade efficiency curve for revised 297mm cyclone

Figure 7.10 is a plot of the static pressure history for this problem. Figures 7.11

and 7.12 show the particle tracking history results and the grade efficiency curve

respectively for the revised design. The procedure used is the same as has been

described in earlier sections.

165
7.4.10 Re-enter Step 6: Compare Revised Design with Requirements

As indicate d in Table 7.7, the cut size of the revised design is within the required

performance limits. Because the required performance has been met the design process

can be stopped, thus completing this sample problem.

Table 7.7 Performance comparison of initial and revise d design with requirements
Initial Design Revision V1 Required
Variable Name
Performance Performance Performance
Cut size (microns) 4.5 3.25 3.0 0.3
no constraint
Pressure drop (Pa) 560 1010
specified
Flow rate (m3/s) 0.140 0.140 0.140

7.5 Chapter Summary

Chapter VII presented the details of a cyclone design methodology. The

methodology uses the numerical methods evaluated in this study. The methodology also

incorporates performance correlations, which can quickly provide initial geometry for a

new design. Guidelines for the design process were also provide d. These guidelines are

able to show the designer how various dimensional changes will affect the cyclone

performance. This is intende d to make the design process easier and more efficient for

less experience d designers, in the spirit of an expert system. An overview was presented

in the form of methodology flow chart. The steps of the process were then described along

with specific instructions for each step. Finally, a worked design example was presente d,

166
which included the details of the various steps used in the process.

CHAPTER VIII.

167
7.4.10 Re-enter Step 6: Compare Revised Design with Requirements

CONCLUSIONS

This chapter presents a set of conclusions that were made after reviewing

the results of Chapter VI and Chapter VII.

8.1 Accuracy an d Efficiency of the CFD Particle Separation Calculations

CFD can be used successfully to predict particle separation the performance

of cyclone separators. In a test case using three operating points the absolute

collection efficiency prediction agreed with the test data within the experimental

test error estimates and numerical discretization errors of the results. The prediction

of cut size agreed with the test data within test and discretization error limits for

two of the three cases run. A procedure for making the particle tracking

calculations was adopted that reduced the require d simulation time from days to

hours.

168
8.2 Limitations of the Mo del

The numerical model under-pre dicte d separation efficiency for particles

smaller than the cut size. Stairmand (1951), who is a recognized pioneer in the

study of cyclones, has explained that differences in observed and theoretical

cyclone performance for particle sizes smaller than the cut size are caused by

particle collisions and particle agglomeration. In considering the implementation of

the numerical particle tracking scheme, it is believed that a lack of a properly

working particle collision model in the CFD code is a possible reason for this

discrepancy in the calculated performance curves.

The numerical model over-predicted separation efficiency for particles

larger than the cut size. Stairmand has explained that differences in observed and

theoretical cyclone performance for particle sizes larger than the cut size are

caused by particle bouncing and the effects of turbulent eddies. The numerical

model did predict a reduction in separation efficiency for particles larger than the

cut size but not to the extent observed in the test data. The effects of turbulent

eddies on particle dispersion are accounted for, to some degree, by the RSM

turbulence mo del (section 3.3) along with the DRW (section 3.4) tracking mo del,

which estimates turbulent particle dispersion using a statistical approach. It is

believed that the full effects of turbulent eddies are not reflected in the computed

results. The reason for this conclusion is that the eddies are modele d but they are

169
not resolve d explicitly with the RSM turbulence model. The bouncing of particles

at walls is accounted for with particles retaining all of their momentum after any

particle-wall collisions.

170
It is believed that the use of the quasi-unsteady tracking in a frozen flow

field approach describe d in section 5.7 affecte d the accuracy of the results. This

simplification to the particle tracking process re duced the time required to perform

the simulations, as indicated earlier from days to hours, but it also removed the

time-varying effects of the flow field during the tracking calculations. Although

done for good reason, this simplification most probably had a negative effect on

the accuracy of the results.

8.3 Computer-Based Cyclone Design Methodology

A design methodology was presente d that makes use of empirical

information for initial sizing of the system as well as CFD calculations for a

detailed analysis. Also included are design guidelines from the literature to guide

the design process in the spirit of an expert system. This methodology forms a

complete computer-based cyclone separator design system. Compared to older

design approaches, the use of this methodology will save time, effort and expense

by reducing the amount of physical prototype design, prototype construction and

prototype testing required to produce a cyclone with the specified performance

characteristics.

171
REFERENCES

Ayers, W. H., Boysan, F., Swithenbank, J., Ewan, B. C. R. (1985). Theoretical


Modeling of Cyclone Performance, Filtration and Separation, 22(1), 39-43.
Barth, W. (1956). Brennst.-Waerme-Kraft, 8,1.

Batchelor, G. K. (1967). An Introduction to Fluid Dynamics. Cambridge, England:


Cambridge Univ. Press.

Boussinesq, J. (1877). Essai Sur La Theorie Des Eaux Courantes, Mem. Presentes
Acad. Sci. (Paris), 23, 46.

Boysan, F., Ayers, W. H., Swithenbank, J. (1982). A Fundamental Mathematical


Mo deling Approach To Cyclone Design, Transactions of the Institution of
Chemical Engineers, 60, 222-230.

Cebeci, T. an d Smith, A. M. O. (1974). Analysis of Turbulent Boundary Layers, San


Diego, California: Academic Press, Inc.

Celik, I. an d Karatekin, O. (September 1997). Numerical Experiments on


Application of Richardson Extrapolation with Nonuniform Grids,
Transactions of the ASME, 119(3), 584-590.

Corrsin, S. and Kistler, A. L. (1954). The Free-Stream Boundaries of Turbulent


Flows. NACA TN 3133.
Crowe, C., Sommerfield, M. and Tsuji, Y. (1998). Multiphase Flows with Droplets
and Particles: CRC Press.

Daly, B. J. and Harlow, F. H. (1970). Transport Equations in Turbulence. Phys.


Fluids, 13, 2634-2649.

Davies, C. N. (1952). Proc. Inst. Mech. Engs. (London), 10, 185.

Dirgo, J. and Leith, D. (1985). Cyclone Collection Efficiency: Comparison of


Experimental Results with Theoretical Predictions. Aerosol Science and
Technology, 4, 401-415.

172
Drain, D. L. (1982). The Laser Doppler Technique. Wiley International.

Dyke, M. Van (Ed.). (1982). An Album of Fluid Motion. Stanford, California: The Parabolic Press.

Fluent 6.3 CFD Users Guide (2006). Fluent Inc. Lebanon, NH.

Fu, S., Launder, B. E. and Leschziner, M. A. (1987). Modeling Strongly Swirling Recirculating Jet Flow with Reynolds-Stress Transport Closures. In Sixth
Symposium on Turbulent Shear Flows, Toulouse, France.

Gibson, M. M. and Launder, B. E. (1978). Ground Effects on Pressure Fluctuations in the Atmospheric Boundary Layer. J. FluidMech., 86:491-511.
[fluent #119]

Goldstein, R. J (Ed.). (1983). Fluid Mechanics Measurements, New York: Hemisphere Publishing Corporation.

Griffiths, W. D. and Boysan, F. (1996). Computational Fluid Dynamics (CFD) and Empirical Modeling of the Performance of a Number of Cyclone Samplers, J.
AerosolSci., 27(2), 281-304.

Hines, W. W. and Montgomery, D. C. (1980). Probability and Statistics in Engineering andManagement


Science (2nd e d.). New York: Wiley & Sons.

Hinze, J. O. (1975). Turbulence (2nd e d.). New York: McGraw-Hill.

Hoekstra, A. J., Derksen, J. J., Van Den Akker, H. E. A. (1999). An Experimental and Numerical Study of Turbulent Swirling Flow in Gas Cyclones,

Chemical Engineering Science, 54, 2055-2065.

Hutton, A. G. and Casey, M. V. (January 2001). Quality and Trust in Industrial CFD: A European Initiative, AIAA Paper 2001-0656.

Iozia, D. L. and Leith, D. (1990). The Logistic Function and Cyclone Fractional Efficiency, Aerosol Sci. Technol., 12, 598.

Jones, W. P. and Launder, B. E. (1972). The Prediction of Laminarization with a Two- Equation Model of Turbulence, Int. J Heat Mass Transfer,
15, 301.

Kline, S. J. and McClintock, F. A. (January 1953). Describing Uncertainties in SingleSample Experiments, Mech. Eng., 3-8.

Lapple, C. E. and Shepherd C. B. (1940). Ind. Eng. Chem., 32, 605.

173
Lapple, C. E. (1951). Chem. Eng. (New York), 58, 144.

Launder, B. E., Reece, G. 6 and Rodi, W. (1975). Progress in the Development of a Reynolds-Stress Turbulence Closure. J. FluidMech., 68(3), 537-566.

Launder, B. E. and Spalding, D. B. (1972). Lectures in Mathematical Models of Turbulence. London: Academic Press.

Launder,B. E. (1989a). Second-Moment Closure and Its Use in Modeling Turbulent

In dustrial Flows. International Journal for Numerical Methods in Fluids, 9, 963985.

Launder, B. E. (1989b). Second-Moment Closure: Present... and Future?


Inter. J Heat Fluid Flow, 10(4), 282-300.

Leer, B. Van. (1979).Toward the Ultimate Conservative Difference Scheme. IV. A Secon d Order Sequel to Go dunov's Metho d. Journal of
Computational Physics, 32, 101-136.

Leith, D. (1984). Cyclones. In M. E. Fayed, L. Otten (Eds.), Handbook of Powder Science and Technology. (pp. 730-745). City:
Van Nostrand Reinhold Company.

Leith, D. and Iozia, D. L. (1989a). Effect of Cyclone Dimensions on Gas Flow Pattern and Collection Efficiency, Aerosol Sci. Technol., 10,491-500.

Leith, D. and Iozia, D. L. (1989b). Cyclone Optimization. Filtration and Separation, 26, 272-274.

Leith, D. and Licht, W. (1972). A.I.Ch.E. Symposium Scr., 68, 196.

Leith, D., Ramachandran, G., Dirgo, 6, Feldman, H. (1991). Cyclone Optimization Based on a New Empirical Model for Pressure Drop, Aerosol Science
and Technology, 15, 135-148.

Lien, F. S. and Leschziner, M. A. (1994). Assessment of Turbulent Transport Models Including Non-Linear RNG Eddy-Viscosity Formulation and Second-Moment

Closure, Computers and Fluids, 23(8), 983-1004.

Moffat, R. J. (1988). Describing the Uncertainties in Experimental Results, Experimental Thermal and Fluid Science, 1, 3-17.

Morsi, S. A. and Alexander, A. J. (September 26, 1972). An Investigation of Particle Trajectories in Two-Phase Flow Systems, J FluidMech., 55(2), 193-
208.

174
Oberkampf, W. L. and Blottner, F. G. (May 1998). Issues in Computational Fluid Dynamics Code Verification and Validation, AIAA Journal, 36(5), 687-
695.

Ogawa, A. (1984). Separation of Particles from Air and Gases Vol. II. CRC Press.

Patankar, S. V. (1980). Numerical Heat Transfer and Fluid Flow. New York: Hemisphere Publishing Corporation.

Reynolds, O. (1895). On the Dynamical Theory of Incompressible Viscous Fluids an d the Determination of the Criterion, Philosophical
Transactions of the Royal Society of London, Series A, 186, 123.

Richardson, L. F. (1908) The Approximate Arithmetical Solution by Finite Differences of Physical Problems Involving Differential Equations with an Application to

the Stresses in a Masonry Dam, Transactions of the Royal Society of London, Series A, 210, 307-357.

Richardson, L. F. (1927). The Deferre d Approach to the Limit, Transactions of the Royal Society of London, Series A,
226, 229-361.

Roache, P. 6 (1994). Perspective: A Method for Uniform Reporting of Grid Refinement Stu dies, Journal of Fluids Engineering, 116(3),
405-413.

Roy, C. 6 (April 2003). Grid Convergence Error Analysis for Mixed-Order Numerical Schemes, .AIAA Journal, 41(4), 595-604.

Schlichting, H. (1979). Boundary-Layer Theory (7th ed.). New York: McGraw-Hill Book Company.

Slack, M. and Harwood, R.(November 2000). Cyclone Modeling at Blue Circle Industries Plc. Sheffield, UK: Fluent Europe
Ltd.

Slack, M. D., Prasad, R. O., Bakker, A., Boysan, F. (2000). Advances in Cyclone Modeling Using Unstructured Grids, Chemical Engineering
Research and Design, 78(8), 1098-1104.

Stairmand, C. 6 (1949). Pressure Drop in Cyclone Separators. Engineering, 168, 409412.

Stairmand, C. 6 (1951). The Design and Performance of Cyclone Separators. Transactions Institution of Chemical Engineers,
29, 356-373.

Svarovsky, L. (1981). Handbook of Powder Technology. Vol. 3, City: Elsevier Scientific Publishing Company.

175
Svarovsky, L. (June 1992). Gas Cyclones in Fine Particle Separation, Fluid/P article
Separation Journal. 5(2), 72-74.

Svarovsky, L.(2000). Solid-Liquid Separation (4th ed.). Boston: Butterworth-


Heinemann.

Tannehill, J C., Anderson, D. A. and Pletcher, R. H. (1997). Computational Fluid


Mechanics and Heat Transfer (2nd ed.). Philadelphia: Taylor & Francis.

Thematic Network for Quality and Trust in the Industrial Application of CFD
(QNET- CFD) (2003). Best Practice Advice for AC3-03 Cyclonic Separator.
Sheffield, UK: Fluent Europe Ltd.
Wilcox, D. C. (1994). Turbulence Modeling for CFD. La Canada, California: DCW
Industries, Inc.

Yakhot, V. and Orszag, S. A. (1986). Journal Scientific Comput., 1, 1-51.

Yakhot, V. and Orszag, S. A., Tangham, S., Gatski, T. B. and Speziale, C. G.


(1992). Phy. Fluid A., 4, 1510-1520.
Yakhot, V. and Smith, L. M. (1992). Journal Scientific Comput., 7, 35-61.

Zhou, Q. and Leschziner, M. A. (1991). Technical report: 8th Turbulent Shear Flows
Symp., Munich.

176
APPENDICES

177
APPENDIX A.

SVAROVSKY EULER NUMBER - STOKES NUMBER CORRELATION

The information presented here is from the work of Svarovsky (1992). Table

A.l below shows the dimensions and performance data compiled by Svarovsky.

The average proportions for the cyclones used is shown in Table A.2. The data is

plotted in Figure A.l and shown with his correlation.

Table A.l Commercial cyclone information used by Svarovsky (1992)


Cyclone D Q p gp
Ps X50 AP Eu
Name (mm) (m3/s) (Ns/m2) (kg/m3) (kg/m3) (;m) (Pa) Stk5o
C
1.85E-05

CO
0.18519

CO -C _
1.205

CO o O CN
CN
CN 0
o o
E ,s> o O CD
CD
-
LU
O)
S- T TO
CN
CN CD N 0
ro ii or w CD

CO
1.83E-05

1
ID O O
1.200

h- O O
OO OO
LU
h- ID
ul o " CD csi CD
CN
CO CN CO O)
X
Tongeren

1.83E-05

N
AC 435

135.24

O
2.731

1.200

O
Van

O O OO
O
O O h--
O) LU
CN OO
CD

N
150.00

O O
CO
O
CO CO CO CO CO CO LU
H c c c c c c CN
O
CD
16

178
Table A.1 Commercial cyclone information use d by Svarovsky (1992)
p
(continued)
Cyclone D Q p g Ps *50 AP Eu Stfeo
Name (mm) (m3/s) (Ns/m2) (kg/m3) (kg/m3) (;m) (Pa)
N

1.96E-05

256.82
O

0.214
CN o o o CO LU
H CO oo oo CD
CO
CO O) CN
CO
cc .c CO
o oo ID G>
O CO
o h-

1.85E-05

1.19E-04
.s> g 5
CD

0.06173
CN
CO

1.205
E CN
CN
W LU CO

ID

353.00
N - o 03 03 03 03 03 03 O
CO LU
H CD C C C C C C CO
CD
CO
Tongeren

ID
1.83E-05
AC 850

678.48
oo
1.200
O o
Van

O h- o o LU
O CD
N-
CO o
CN
CD
CO
CO
1.96E-05

ID
h-
0.024

o O G> co O
H G>
ID
oo O
O
CN CO
ID
CO
o LU
CO N -

CO

ID
1.83E-05

1238.71
120-20
F-KXQ

O O
1.200

ID
ID
ID
LO O O CO LU
CO csi CD
CN O) CN
CN
N-
CD
CO

Table A.2 Average cyclone dimensions use d in


the Svarovsky (1992) correlation
Cyclone Dimension Average Proportion
D 1
H 4.5D
h 1.8D
De 0.47D
a 0.63D
b 0.26D
S 0.79D
B 0.34D
H; 1.50D
D; 1.00D

179
Euler vs. Stokes Number for Cyclone Separators from Svarovsky
+ Commercial Cyclones -----------Eu=V(12/Stk50)

10000 - 1000 -

3
LU

100 -

10 - 0.0C

001 0.0 )001 0. Stk50 001 0.01

Figure A.l Euler vs. Stokes number correlation from Svarovsky (1992)

180
APPENDIX B.

DERIVATION OF THE SVAROVSKY CYCLONE SIZING EQUATION

The information presented here is from the work of Svarovsky (1992). The

expression for the empirical sizing and performance estimator used in the cyclone design

methodology is derived here. This expression uses equations (1.1) through (1.4) which

were presented in Chapter I. They are written below as equations (B.1) through (B.4). The

Euler number is considered to be a resistance coefficient that represents the ratio of static

pressure drop between the inlet and outlet of the cyclone to the dynamic pressure of the

flow within the cyclone and is


Ap
defined below as

E=

(B.1)

Ap
where is the static pressure drop measure d between the inlet an d the gas outlet of a

cyclone, p is gas density and v is the body velocity based on the flow rate and the cross-

section of the cylindrical body of the cyclone as


4Q
v=
nD2 (B.2)

where Q is the gas flow rate and D is the cyclone body inside diameter. The Stokes number

181
is commonly used to characterize particle laden flows. It can be thought of as the

182
relation between the particle response time and the system response time. The

5>,
X V

Stk 18^D
5 0 (B3)
Stokes number written for a particle of size x50 and is defined as

12
Eu
Stk
(B.4
)
where p is the gas viscosity, p, is the solids density, the velocity v is defined by

equation (B.2) and x50 is the cut size (equiprobable size). Svarovsky found that for

the cyclones he stu die d Eu was relate d to Stk50 by the equation (B.4) below.

Appendix A shows the cyclone data Svarovsky used and his curve fit.

The performance of the cyclone is related to the particle cut size, and the

flow rate of particle laden gas that must be processed. The costs associated with

processing this particle laden gas has two components. The first component is

operating cost which is a function of the pressure drop through the cyclone. The

pressure drop is considered an operating cost because power has to be input to the

system proportional to the product of Flow rate, Q an d pressure drop . A cyclone

system that has a large pressure drop costs more to operate. The secon d

component of cost is the capital expen diture nee de d to set up the system. A larger

cyclone costs more capital than a small cyclone to buy and install. To design the

most cost effective system requires consideration of the capital and operating costs

183
of the system in comparison to the performance obtaine d from the system.

Depending on the costs of the system as well as other constraints such as available

space, it may be advantageous to design a system that uses several cyclones in

parallel. This type of arrangement may have advantages because the flow is split

between the several

184
cyclones and the pressure drop (operating cost) is less than if all the flow were forced

through a single cyclone. Because of these trade offs the option of cyclone sizing for

multiple cyclones operated in parallel needs to be available. More will be said about

cyclones in parallel later. In terms of the operating cost, pressure drop, Svarovsky relates

this to the head loss, H in meters of the gas used in the cyclone. For design purposes he

recommen ds that the head loss for a cyclone be in the range of 40 to 100 meters of gas.

The expression relating head loss, H (m) to pressure drop, is shown below as equation
(B.5)

AP
H=
Pg (B5)

where AP is the pressure drop through the cyclone in Pascals, p is the density of the

working gas in kg/m3 and g is the acceleration of gravity in m/s2.

Starting with equations (B.1) through (B.5) an expression estimating the required cyclone

diameter, D needed to separate particles of cut size x 50, in a gas flowing at a rate Q with a
12 k
Eu
Stk, Stk 50 (B4.1)

where k is a coefficient with the value 3.46 and n is an exponent with a value of

0.5. Step 2. Substitute equation (B.5) into (B. 1).

head loss, H can be derived as shown in the following section.


Step 1. Recast equation (B.4) as

185
Ap Pg? _ 2 gH
E. PV PV 2
. I27 I2J

(B.1.1)

Step 3. Substitute equation (B.2) into equation (B.1.1).

2gH 2gH 7T gHD4


2

E =
2 2 8 Q2
'4Q_
nD2
(B.1.2)

Step 4. Substitute equation (B.2) into equation (B.3).

o.r = 50PSV = X 50 Ps 4 = 2 ps X50Q


X Q

(B.3.1)
18pD 18pD nD 9n p D3

Step 5. Rearrange equation (B.4.1)


Eu Stk 50 = (B4.2)
k
Step 6. Substitute equation (B.1.2) and (B.3.1) into equation (B.4.2)

n
V gHD4" s
2P X
50 Q
=k
1

1
______

---------
00

<N

9n p D3
(B.4.3)

Step 7. Rearrange equation (B.4.3).

n
n2g
2P X
s 50Q 1 n
HD 9np =k
8Q2 4 D3 (B.4.4)

Step 8. Define a new coefficient, Ct.

2
2 n
C= ng Ps X50 Q
1
______
00
<N

9np (B6)
1

Step 9. To allow the final expression to be used for the modeling of several cyclones in

parallel, replace the flow rate, Q in equation B6 with the term (Q/Nc) where Nc is the

number of cyclones operating in parallel in the system and name the new coefficient Ctp.
182
2
2p X
C n * 50Q

1 ____
Wp 9npNc

OO
(B.7)

to
Step 10. Substitute equation (B.7) into (B.4.4) and
rearrange.
D4 (4 / )
H 3# (
tp = HD ) (
tp =k
D (B.4.5)

Step 11. Solve equation (B.4.5) for D as a function of


H.
-i(3/)
k{
D = HC p _ t

(B.8)

Equation B8 is an expression that provides an estimate of the required cyclone diameter,

D for a quantity of Nc cyclones operating in parallel such that they are able to separate

particles of cut size, *50 and particle density, Ps from a volume flow rate, Q of gas with

the gas properties p and p . The cyclone diameter, D is a function of the permissible

cyclone pressure drop expressed as a head loss, H which is in turn expressed in meters of

the gas being processe d.

As in dicate d earlier in Appen dix B, Svarovsky (1992) recommen ds that for

design purposes, the permissible cyclone head loss, H be in the range of 40 to 100 meters

of gas. Equation (B.8) forms the basis of the size and performance estimation used in the

cyclone design methodology presente d in the thesis. An example of the results is shown

in the following section.

Example Sizing Problem:

Given the following system requirements and material

properties: Gas flow rate: Q = 0.03 m /5 3

183
Gas density: p = 1.2kg/m3 Gas viscosity: p= 1.85E - 5 kg/m - s Particle density: ps =
2
970kg/m3 Particle cut size: *50 = 3 0E - 6 m Acceleration due to gravity: g = 9.81m/s2

Determine the diameter, D of a single cyclone that would process the above particle laden

gas with a head loss, H through the cyclone of 80 meters of gas. Use the model parameters

developed by Svarovsky of k=3.46 an d #=0.5.

Solution:

Step 1. Substitute the required values for the system in the expression for the coefficient

Ctp
p X
2
s 50
2 # " 7T *9.81
2
"2*2970*(3.0E-6) *0.03" 0
2
ng * Q " * which
.5
1
______

<N

9quNc 8 * (0.03) 9n*1.85E - 5*1


00

2
1

is

equation (B.7).

23.5487

Step 2. Substitute the values of H, k, # and Ctp above into equation (B.8) to solve for the

estimated diameter, D.
(3 (* 3 0 = 0.094 m = 94.0 mm
k #4) ' 3.46 "
D= U)
1

80*23.5487 _ If desired, the cyclone


_

pressure drop in Pascals can be

calculated based on the relationships shown in equation (B.5).

AP = pgH = 1.2*9.81*80 = 942 Pa The cyclone sizing and

performance calculations use the above equations. The methodology calculated a range of

cyclone diameters that correspond to the head loss

184
range recommended by Svarovsky (1992), which is: 40 < H < 100. The designer can then

choose a cyclone diameter for his starting geometry base d on the pressure drop that can

be accepted for the system. In general, for a new design head losses in the middle of the

40 to 100 range are recommen de d. Usually in the design process, changes nee d to be

made to the inlet geometry to obtain the desired cut size, which results in an increase in

pressure drop for the final design.

185
APPENDIX C.

GUIDELINES FOR CYCLONE PROPORTIONS

This appendix provides a list of ranges for the various cyclone dimensions from

the literature (Ogawa, 1984; Svarovsky, 1992; Leith & Iozia, 1989b; Leith et al., 1991).

Table C.1 is a list of various dimensions and groups of dimensions with ranges of values

that are commonly found in the literature. This information is intended to be use d as a

guideline for the designer when considering the changes to his design in order to adjust

it's performance. Comments are also include d in the table to help the designer understand

how the particular dimensions or groups of dimensions affect a particular aspect of the

cyclones performance. A diagram showing the dimensions is shown at the en d of the

appen dix in Figure C.1.

186
Table C.1 Cyclone proportioning guidelines and typical values
Cyclone Typical Commonly Occurring range of
Dimension Comments
Value values
or Group
Increasing H re
H duces AP without 4.5D 3D < H < 6D
changing xso.
Small changes to h
h have small effects 1.8D 1D < h < 2D
on AP & x5 o
S In 0.8D 0.4D < S < 1D
general: a < S < h
Inlet height for
rectangular inlet 0.6D 0.3D < a < 0.8D
ducts, Ai=ab
Inlet width for < f D " De ]
6

b rectangular inlet 0.3D 0.15D < b < 0.4D \ J


2

ducts, Ai=ab

Di Inlet diameter for 0.33D Dt < De


circular inlet pipes,
4 =- 4
Outlet pipe
De diameter, 0.5D 0.3D < De < 0.8D
A =^
A
4
AP = vf 4
ab D 2 X X
D e
B 0AD < B < De 0.3D 0.3D < B < 0.5D
This dimension is
DB not commonly 1D X
reported.
This dimension is
HB not commonly 1.5D X
reported.

187
Figure C.l Diagram of cyclone dimensions

188
APPENDIX D.

CYCLONE OPTIMIZATION STUDY FROM LEITH ET AL.

This appendix presents information from Leith, Ramachandran, Dirgo and

Feldman (1991) to be used as part of the cyclone design methodology presented at part of

this study. Leith et al. (1991) report on the results of an optimization study done using

empirical models for cyclone separation performance and pressure drop. The objective of

the study was to develop design guidelines for cyclones such that the pressure drop was a

minimum for the given cut size at the design flow rate. Although empirical models have

limitations, it is believed that the resulting relationships especially with respect to the

relative sizing of the inlet and outlet ducts compared to the cyclone diameter are valuable.

The more useful of the findings of that study have be placed in this appendix as a design

aid.

Figure D.1 shows what these researchers believed to be the optimum ratios for

inlet duct dimensions a and 6 and outlet diameter, De with respect to the cyclone diameter,

D. Note also, that the performance models were based on tangential entry cyclones

considered to be of the high efficiency type. They do not apply to high flow type cyclones,

which normally have a wrap around or scroll type inlet. This type of

189
cyclone does not have the restriction on the inlet duct width, b. See Stairmand (1951) for

more information on high flow type designs.

190
Other curves similar to the ones shown in Figure D.1 are given in the paper, but provide

results that are consistent with the findings of others with respect to most other cyclone

Optimal Curves for D e vs. a vs. b from Leith et. al. (1991)
[Q=0.094 m3/s, H/D=5, B/D=0.375, h/D=1.5, S=a]
Note: the study was done on tangential entry cyclones using empirical performance models.

0.0 1.0 2.0 3.0 4.0 5.0 6.0


x5 (microns)
y= -0.0089x2+ 0.075x + 0.0577

Figure D.1 Optimal curves for De, a Kb from Leith et. al. (1991)

191
dimensions. Their findings 123 indicated that the best ratios for De and 6 did not

change as flow rate, Q, diameter, D and total cyclone height, H changed. Changes

to the inlet height, a had a strong influence on x 5 0 as can be seen in Figure D.l.

Table D.l below summarizes many of the findings. A diagram of the cyclone

dimensions is shown in Figure D.2 at the end of the appendix.


Table D.l Cyclone proportioning guidelines from Leith et al. (1991)
optimization study
Cyclone
Comments
Dimension
De was the primary dimension modified in the
optimization study. Changes to De change pressure drop.
For the study De was changed and then changes were
made to a, 6 and S to bring pressure drop back to its
original value. The
combination of a, 6 and S that resulte d in the lowest Xb
De was use d as the starting point for the next iteration. An
optimal design was considered to be one that produced the
lowest
pressure drop for a given cut size, XB .
For all of the optimize d designs: Ae > A
For the optimize d designs: D - De - D
0 4 0 6

Ae =nD2
a Cut size, X5 was verye sensitive
4
e to changes in a.
Aj=a6
D-6- D
For the optimize d designs: 0 1 5
.22
6
Aj=a6

H Increasing H re duces AP without changing xb . For the


stu dy: 4D - H - 6 D
h For the study: h = D 15

S For the stu dy: S = a


B For the study: B = 0.375D
dB Dust bin geometry was absent.
HB Dust bin geometry was absent.

192
12312321 3123

193

También podría gustarte