MIXING DURING THE FILLING PROCESS

by

GRACE NEAL

A thesis submitted to

The University of Birmingham

for the Degree of

MASTERS OF RESEARCH
 University of Birmingham Research Archive
e-theses repository

This unpublished thesis/dissertation is copyright of the author and/or third

parties. The intellectual property rights of the author or third parties in respect
of this work are as defined by The Copyright Designs and Patents Act 1988 or
as modified by any successor legislation.

Any use made of information contained in this thesis/dissertation must be in

accordance with that legislation and must be properly acknowledged. Further
distribution or reproduction in any format is prohibited without the permission
of the copyright holder.
 Abstract

By adapting the techniques Planar Laser Induced fluorescence (PLIF) and Particle

Image Velocimetry (PIV) for a transient system with a rising free surface the mixing

that occurs as a vessel is filled process has been investigated.

The PLIF technique has been demonstrated to give good repeatability for the same

flow conditions and qualitative examination of the images has revealed a change in

the mixing mechanism from laminar to turbulent above a critical value of the

Reynolds number (Re~1000), with cases above this reaching a log variance of -2

which equates to 92% mixed. PIV allowed the distribution of momentum and shear

in the tank to be investigated, highlighting dead regions in the bottom corners of the

tank and showing the top half of the tank was relatively quiescent. To improve the

mixing performance, changes to the flow rate and the nozzle design were made to

increase the movement in these regions.

Sinusoidal variations in the inlet velocity increased the level of mixing from an

unmixed state so that the vessel reached 84% mixed. With a swirl insert in the end of

the nozzle the level of mixing in the tank reached 95% (log variance of -2.6) at one

point in the fill. Combining these improvements may produce a more sustained and

reliable improvement and should feature in any future work.

 Acknowledgements

Firstly I would like to thank my supervisor Dr Mark Simmons for his support and

guidance throughout the last 6 years. I'd also like to acknowledge Dr Andreas

Tsoligkas for teaching me to use the PIV and PLIF equipment. Thank you also to

Alan Hough at Unilever for his support.

Thank you to all the technical staff who contributed to the building of my

experimental rig: everyone in the Mechanical Engineering Workshop, also Derek

Green and Andrew Tanner from the Biosciences Engineering Workshop. Thanks go

to Gordon Redwood at Unilever for making my nozzle designs a reality.

Thanks go to my family and friends, for picking me up when I'd lost hope. To

everyone at Unilever who made me feel so welcome and kept me sane with endless

cups of green tea. Thanks also to my friends at the University for the hours of

entertaining lunchtime banter.

Most of all thank you to the EPSRC and to Unilever for funding this research.
Contents

Chapter 1 Introduction and motivation ............................................................. 1

1.1 Industrial Challenge ................................ ............................................. 1

1.2 Late Variant Addition ............................................................................ 2

1.3 Project objectives ................................................................................... 4

1.4 Thesis Layout .......................................................................................... 5

Chapter 2 Literature review .................................................................................. 6

2.1 Introduction ............................................................................................ 6

2.2 Mixing ....................................................................................................... 6

2.2.1 Quantifying Mixing.................................................................. 7

2.2.2 Mixing in pipe ......................................................................... 11

2.3 Jet structure and mixing ................................................................... 13

2.3.1 Regions of an impinging jet ............................................... 13

2.3.2 Jets in large storage tanks ................................................. 16

2.3.3 Impinging jets ........................................................................ 18

2.4 Mixing and flow diagnostics ........................................................... 19

2.4.1 Planar Laser Induced Fluorescence ................................ 20

2.4.2 Particle image Velocimetry …............................................ 21

2.5 Summary .............................................................................................. 24

Chapter 3 Materials and Methods .................................................................... 25

3.1 Introduction ........................................................................................ 25

3.2 Mixing tank .......................................................................................... 25

3.2.1 Dimensions and equipment set up for static pipe

system ......................................................................................25
 3.2.2 Moving nozzle set up and control ................................ 26

3.2.3 Set up of case studies ....................................................... 28

3.3 Fluid properties ............................................................................. 30

3.3.1 Newtonian Fluids ............................................................. 30

3.4 PLIF .................................................................................................... 31

3.4.1 Normalisation ................................................................... 33

3.4.2 Log variance and image analysis ................................ 37

3.4.3 Calibration and error .................................................... 39

3.5 PIV .................................................................................................... 40

3.5.1 Assumptions and analysis techniques ..................... 41

Chapter 4 PLIF – Concentration fields ...................................................... 43

4.1 Introduction .................................................................................. 43

4.2 Static pipe ...................................................................................... 43

4.2.1 Log Variance analysis ................................................... 43

4.2.2 Flow regime in the vessel ............................................ 52

4.3 Moving pipe ................................................................................... 54

4.3.1 Mixing performance and mixing mechanism ........ 56

4.3.2 Obstructed jet. ................................................................. 58

4.4 Conclusions .................................................................................... 60

Chapter 5 Velocity Fields ............................................................................... 61

5.1 Introduction .................................................................................. 61

5.2 Velocity profiles ........................................................................... 61

5.3 Shear rates ..................................................................................... 67

5.4 Conclusions .................................................................................... 68

Chapter 6 Flow improvements .................................................................... 69

 6.1 Introduction ................................................................................ 69

6.2 Variable flow rate ...................................................................... 69

6.3 Nozzle improvements .............................................................. 71

6.4 Conclusions ................................................................................. 74

Chapter 7 Conclusions ................................................................................. 75

7.1 Mixing in a filling vessel .......................................................... 75

7.2 Future work ................................................................................ 76

References ............................................................................................................. 78

Appendix: Matlab analysis code ..................................................................... 84

List of illustrations

Figure 1.1 Illustration of tank lengthscales…………….……………………….…. 5

Figure 2.1 Illustration of the three different scenarios described by Pan and Meng

(2001) ……………………………………………………………….…. 11

Figure 2.2 Illustration of the regions of an impinging jet, Reungoat et al. (2001) ... 13

Figure 2.3 (Unger and Muzzio 1999) Geometry of the impinging jet mixers ….…. 19

Figure 2.4 PIV analysis ……………………………………………………………. 22

Figure 3.1 Schematic of the experimental rig and PLIF set up ……………………. 26

Figure 3.2 Schematic of the nozzle traverse ……………………………………..... 28

Figure 3.3 Schematic of a straight pipe and the three nozzles improvements …….. 30

Figure 3.4 Plot of greyscale as a function of tracer concentration at two different laser

intensities ………………………………………………………............ 33

Figure 3.5 Calculation of elements of the matrices M and K from the linear relation

between greyscale and tracer concentration for each pixel, (only two

pixels are plotted for demonstration) ………………………….………. 35

Figure 3.6 Definition of the location of the fluid surface via the change in the average

row greyscale values as a function of the vessel height. The interrogation

zones are shown and also the maximum fill height …............................. 36

Figure 4.1 Evolving concentration fields at different fill heights for the 6 mPa s fluid

at (a) h = 2.2 cm; (b) h = 4.7 cm; (c) h = 7.3 cm; (d) h = 10.3 cm at

Re = 661 (Q = 1.67 × 10-5 m3 s-1) ……………………………...………. 44

Figure 4.2 Evolving concentration fields at different fill heights for the 3 mPa s fluid

at (a) h = 2.2 cm; (b) h = 4.7 cm; (c) h = 7.3 cm; (d) h = 10.3 cm at

Re = 1245 (Q = 1.67 × 10-5 m3 s-1) .……………………………………. 44

Figure 4.3 Mixing performance for three different fluid viscosities at a fill rate of

Q = 1.67× 10-5 m3 s-1 (a) log variance plotted against dimensionless fluid

height (H/T); Image of the vessel filled with fluid at (b) Re=343

(12 mPa s), (c) Re=661 (6 mPa s), (d) Re=1245 (3 mPa s) ……………. 46

Figure 4.4 Mixing performance for three different fluid viscosities at a fill rate of

Q = 3.33× 10-5 m3 s-1 (a) log variance plotted against dimensionless fluid

height (H/T); Image of the vessel filled with fluid at (b) Re=687

(12 mPa s), (c) Re=1322 (6 mPa s), (d) Re=2490 (3 mPa s) ………..... 48

Figure 4.5 Mixing performance for three different fluid viscosities at a fill rate of

Q = 5.00× 10-5 m3 s-1 (a) log variance plotted against dimensionless fluid

height (H/T); Image of the vessel filled with fluid at (b) Re=1030 (12

mPa s), (c) Re=1983 (6 mPa s), (d) Re=3735 (3 mPa s) ……………..... 50

Figure 4.6 The dimensionless height (h/T) of the centre of the vortex from the base of

the tank, measured for three Reynolds numbers at flow rate of Q = 10-5

m3 s-1…………….………………………………………………..…….. 51

Figure 4.7 The dimensionless distance of the centre (h/T) of the vortex from the edge

of the inlet pipe, measured for three Reynolds numbers at flow rate of Q =

1.67× 10-5 m3 s-1 ……………………………………………….……… 52

Figure 4.8 Mixing performance for three different fluid viscosities at a fill rate of

Q = 5.00×10-5 m3 s-1 (a) log variance plotted against fluid height; Image

of the vessel filled with the (b) 16 mPa s fluid, (c) 8 mPa s fluid, (d) 4

mPa s fluid ……………………………………………….…………… 55

Figure 4.9 Plot of the minimum log variance reached in each experiment versus the

jet Reynolds number ……………………………….…….…………... 56

Figure 4.10 Evolving concentration fields at different fill heights for an 8 mPa s

fluid, Re = 1494 at (a) h = 2.2 cm; (b) h = 4.7 cm; (c) h = 7.3 cm; (d) h =

10.3 cm at a flow rate Q = 3.33 × 10-5 m3 s-1 through a 6 mm dip pipe

…………………………………………………………………..……… 58

Figure 4.11 Evolving concentration fields at different fill heights for an 8 mPa s

fluid, Re = 1494 where a bubble has formed at (a) h = 2.2 cm; (b) h = 4.7

cm; (c) h = 7.3 cm; (d) h = 10.3 cm at a flow rate Q = 3.33 × 10-5 m3 s-1

through a 6 mm dip pipe ………………………………………..……… 59

Figure 4.12 The log variance plot of two fills of the 8 mPa.s fluid at a flow rate

Q = 3.33 × 10-5 m3 s-1 where a bubble has formed under the pipe in one

(blue) and the jet is unobstructed in the other (red) ……….…..……….. 60

Figure 5.1 An example of the velocity profile in a vessel showing the features

present in all cases …………………………………………….….…… 62

Figure 5.2 Jet impingement zone, shaded to indicate velocity magnitude ….….… 62

Figure 5.3. The change in normalised velocity along the jet centreline for the four

different points in the fill ………………………………………………. 63

Figure 5.4. Vertical velocity profile across horizontal slices taken at heights h above

the base of the tank ……………………………………………….…… 64

Figure 5.5. The centreline velocity profile for low Reynolds number fills when the

injection height has reached 9 cm ………………………………...…….65

Figure 5.6. The centreline velocity profile for high Reynolds number fills when

the injection height has reached 9 cm ……………………………….… 66

Figure 5.7. Velocity magnitude contour plots for flow at Re = 517 and Re = 1911

…………………………………………………………………………. 67
Figure 5.8. The shear rate calculated from the velocity field measured at Re = 518

……………………………………………………………………...……68

Figure 6.1 Mixing performance for three nozzles at a viscosity of 18 m Pa s (a) log

variance plotted againsts fluid height (cm); Image of the vessel filled with

(b) constant flow rate of Q = 3.33×10-5 m3 s-1 (c) sinusoidal flow rate

varying between 1.67 x 10-5 m3 s-1 and 5 x 10-5 m3 s-1 (d) flow rate

increasing from 0 to 6.67 x 10-5 m3 s-1 ………………………………… 70

Figure 6.2 Mixing performance for three nozzles at a fill rate of Q = 3.33×10-5 m3 s-1

and a viscosity of 18 m Pa s (a) log variance plotted against fluid height

(cm); Image of the vessel filled with (b) straight pipe (c) cone 1 nozzle

(d) cone 2 nozzle ……………………………………………….……… 72

Figure 6.3 Mixing performance for two nozzles at a fill rate of Q = 3.33×10-5 m3 s-1

and a viscosity of 18 m Pa s (a) log variance plotted against fluid height

(cm); Image of the vessel filled with (b) straight pipe (c) swirl nozzle

……………………………………………………………………..…… 73
List of Tables

Table 2.1 Measurement techniques ........................................................................... 19

Table 3.1 Experimental parameters, where Reynolds number is defined as

ρd j u j
Re = .............................................................................................. 31
µ
Nomenclature

PLIF Planar Laser Induced Fluorescence

PIV Particle Image Velocimetry

CoV Coefficient of Variance

TKE Turbulent Kinetic Energy

Ci Concentration measurement point

Cmean Mean concentration

ξAij Reduced mean concentration

ξABij Contacting Parameter

k Sensitivity exponent, strength of interation

θ Mixing time

G0 Greyscale when concentration is zero

Gm Greyscale when vessel is completely mixed

Gn Normalised greyscale

Gi Measured greyscale values

DM Molecular Diffusivity

dj Diameter of the jet

H, h Height of fluid in the tank

K Intercept

M Gradient
Pixel size (Houcine et. al. 1996)

N Number of sampling points

Q, q, Flow rate (principal flow and side stream)

Re Reynolds number

T Tank width

D Tank Diameter
tD Diffusion time

uj, U Jet velocity

ε Specific energy

λ Kolmogoroff length scale

μ: Dynamic viscosity

ν Kinematic viscosity

ρ Density

σ2 Variance

Li Impingement height

We Weber number

γ Shear

x Horizontal positon

y Vertical position

vx, vy Velocity components

σ
Surface tension
 1

Chapter 1 Introduction and motivation

1.1 Industrial Challenge

Manufacturers will often produce a range of very similar products. This allows the

product and the marketing to be targeted towards different consumer groups,

increasing sales and expanding the market. There is usually a choice of colour,

fragrance or flavour within each group of products so the consumer has a full range of

options to cover their needs. This increased diversity has been driven by the

competition and consumer demand in the market and it results in a number of

challenges for manufacturing.

The shelf space available in retail outlets for each type of product has stayed constant

despite increased diversity. Where previously a whole shelf would be dedicated to

one product, retailers now only display a small quantity of each variety. Storing large

amounts of stock is expensive and so retailers want to order small amounts of a wide

range of products regularly. This in turn requires the manufacturer to either store

large amounts of each variety or to make smaller batch sizes, in reality both of these

have occurred.

Traditional manufacturing methods for fast moving consumer good involve

production in large batches followed by filling into packaging. As the batch size

required has decreased the efficiency of the production decreases due to the increase

in the down time and cleaning which required between batches. The down time

increases costs due to the reduction in output, the increase in cleaning produces both

an environmental and financial cost due to the disposing of the waste.

 2

A further challenge to the industry comes in the form of regulation; the European 7th

Amendment Legislation means that there can be no cross contamination between

fragrances above a specific low level. This is so the potential sensitizers which are

contained within the fragrance are controlled in each product. Since this legislation

was implemented in 2004, the level of cleaning between each batch has increased, this

in turn has caused an increase in the quantity of waste generated and hence the cost

associated with disposing this waste.

Changes to the manufacturing methods that would allow the flexibility to be increased

would not only alleviate the increased waste and loss of efficiency, but would also

provide the flexibility to react quickly to changes in the market with new products.

1.2 Late Variant Addition

To maximise the efficiency and flexibility of production, all products need to share as

much of the process as possible. Products are often made using the same base fluid

and a variant such as a fragrance or emotive which differentiates between the

products. The aim is to find a method for mixing these variants into the base product

at as late a stage as possible.

A technology that would allow the products to be mixed on the production line would

reduce the waste and improve efficiency. The increased flexibility would also give

the manufacturer greater ability to quickly respond to customer demand. Large

amounts of stock could be released generating cash flow and also reducing storage

costs.
 3

The two major technical challenges for achieving this late variant addition are the

accurate dosing of the variant and the effective mixing. In the case of personal care

products the variants particularly the fragrances are usually oil based liquids which

are being mixed into oil in water emulsions. The microstructure of the resulting

product is greatly affected by both the quantity of the variant and the length scale to

which it has been mixed. It is important that the microstructure of the product is

correct as it has a large effect on the efficacy, stability and appearance and therefore

customer satisfaction.

Late variant addition is already practiced by Unilever in the manufacture of their

aerosol deodorants. Accurate weighing machines allow the dosage of the variants to

be accurately measured and controlled; it is then mixed during the gasification of the

aerosol. A similar dosing machine could be used for roll on deodorants however the

mixing challenge is much greater.

There are a number of mechanisms that could provide the level of mixing required for

late variant addition in roll on deodorants. These include a form of mechanical

agitation such as a stirrer, mixing in the filling pipe using a static mixer, shaking or

through the use of jet mixing. Of this list jet mixing would provide the most suitable

solution as a stirrer or static mixer would require cleaning and shaking the product

would require a large change to the filling lines which would involve a large capital

expenditure.
 4

1.3 Project objectives

Extensive studies over many years have provided a very strong understanding of

mixing in large vessels; the role of impellers/agitators, recirculation through

submerged jets and geometry (e.g. the effect of liquid height). However what has

received less attention is the mixing that occurs as ingredients are added to vessels of

small length scale, where there is no mechanical agitation. The study of such

processes provides a number of experimental and modelling challenges in that start-

up effects account for a significant proportion of the mixing time. There can be large

disparities between the properties and ratio of the materials being added, the level of

the free surface changes and the entrainment of air makes visualization difficult.

The aim of this project is to understand the mixing processes that occur during jet

filling, this will characterise the range of products that can be produced in this

manner. This overall goal can be broken down into smaller objectives:

• To develop the technique Planar Laser Induced Fluorescence (PLIF) to allow

the level of mixing in the vessel to be measured throughout the filling process.

• Determine the parameters that set the limits for which mixing can be achieved

to the required lengthscales.

• Using the technique Particle Image Velocimetry (PIV) gain an understanding

of the flow structure in a filling tank

• Improve the mixing performance in the vessel

The lengthscales of the bottle to be filled are demonstrated in fig. 1.1, the flow rates

range from 1.67 × 10-5 m3 s-1 to 5.00 × 10-5 m3 s-1.

 5

d = 0.6 cm

hmax = 10 cm

hinit= 1 cm

T = 6 cm

Figure 1.1. Illustration of the tank lengthscales

1.4 Thesis Layout

The first stage in this work was to investigate the available literature, this is included

in Chapter 2. A full discussion of the experimental work carried out is described in

Chapter 3. Chapter 4 is a description of the mixing performance in the tank with a

standard straight inlet pipe while Chapter 5 goes on to examine the velocity field in

the tank. Chapter 6 investigates whether improvements can be made to the mixing by

using alternative pipe designs. Conclusions have been drawn in Chapter 7 along with

recommendations for future work.

Work from this thesis has also been published in the archival Journal Experiments in

Fluids (Neal et al.,. 2008), and presented at both the International Symposium for

Mixing in Industrial Processes (ISMIP 2008) (Neal et al.,. 2008), and the World

Congress in Chemical Engineering (Neal et al.,. 2009).

 6

Chapter 2 Literature review

2.1 Introduction

Although there is little published research into the mixing that occurs as a vessel is

filled there is a large body of work into mixing by a range of other mechanisms.

Principles from other mixing techniques and impinging jets can be applied to the

current problem and for this reason a brief outline of these are included here. Also of

importance to this project is the measurement techniques which have been used to

characterise the flow and mixing within the vessel: Particle Image Velocimetry (PIV)

and Planar Laser Induced Fluorescence (PLIF).

2.2 Mixing

Mixing is the reduction of an inhomogeneity such as concentration, phase or

temperature in order to achieve the desired process result (Paul et al. 2004). The

quality of mixing achieved in a process can affect product quality from appearance

through to the creation of microstructure, as a result an understanding of how

materials mix is essential for any successful process.

The mixing mechanism depends on the flow of the fluids being mixed, turbulent

flows where the velocity is constantly fluctuating reduces the scale of segregation

quickly down to the size of the smallest eddies while diffusion reduces the intensity of

the segregation (Kresta and Brodkey, 2004). Laminar flows do not contain the eddies

seen in turbulent flow and so there is little reorientation of the fluid particles; to create

mixing in a laminar flow an element of periodicity is necessary. Creating a chaotic

 7

flow provides the stretching and reorientation required to reduce the scale of

segregation (Szalai et al. 2004).

A large body of work has been done on a range of different mixing problems

especially the flow in stirred tanks; jets have been used for mixing in a range of

different situations from large storage tanks to two impinging jets on a smaller scale.

Mixing in transient flows such as filling processes is less well researched.

2.2.1 Quantifying Mixing

There are a number of ways that the degree of mixing in a vessel can be quantitatively

measured. The most common of these is a coefficient of variance approach defined

by Kukukova et al. (2009) as:

2
1 C −C 
CoV =
N
∑  i C mean  (2.1)
 mean 

Where Ci is the concentration measurement at a point, Cmean is the average

concentration and N is the number of sample points. An alternative is to take the log

of the variance as defined by Brown et al.. (2004). The principle of this

measurement is to take the standard deviation of a measurement that indicates the

concentration and determine when it has decreased to an acceptable level. The time

taken to achieve this level of mixedness is called the mixing time. This has been

determined experimentally within a vessel via a range of methods such as measuring

changes in conductivity, pH or temperature or by looking at the mass fractions of the

products from competing reactions (Brown et al.. 2004). Alternatively a method such

as PLIF can be applied which measures changes in the optical properties of the

system.
 8

To allow the measurement of these properties a tracer is needed, this is added to the

flow and causes a change that can be monitored. In the case of conductivity a salt can

be used so areas of higher concentration have higher conductivity, for pH an acid will

fulfil the same role. These techniques are usually measured at single points; for

whole field measurements such as the optical technique PLIF a fluorescent tracer such

as Rhodamine is used. When Rhodamine 6G is excited by a laser it emits light at an

intensity that is directly proportional to the concentration so measurements can be

made from a greyscale image of the tank. This allows a large sample of concentration

values across a whole plane of the vessel, these concentrations have a distribution and

this can be quantitatively analysed.

Houcine et al. (1996) and Fall et al. (2001) used a number of different statistics to

analyse the concentration distributions obtained for continuously fed vessels, where a

fluorescent tracer was continuously added into one feed. The first was the field of

reduced mean concentration ξAij, where the value at each pixel represents the

probability of the molecules of tracer being in the volume represented by that pixel.

This is calculated by dividing the mean concentration at the pixel Cij , obtained over a

large number of samples, by the concentration in the feedstream CA0:

Cij
ξA = (2.2)
ij
C A0

The second statistic used was the contacting parameter, similar to the reduced

concentration but where the contrast is increased, by multiplying through by one

minus itself: this allows the areas of different mixing performance to be more easily

identified but does not contain any new information:

 9

ξ A B = ξ A ⋅ ξ B = ξ A (1 − ξ A )
ij ij ij ij ij
(2.3)

The third statistic was the temporal variance where the concentration at each pixel

was compared with the temporal average. Houcine et al. (1996) proposed that the

contacting parameter and the field of variance are the most important parameters to

characterize, since the former gives the average state of the mixing at each point and

the latter characterizes the mixing dynamics of the concentration fields. However

these fields are defined by the geometry of the system so a local mixing parameter

was defined as the ratio of the contacting parameter at each point to its maximum.

This could be used to compare mixing performance in different mixing situations by

including the pixel size M and a sensitivity exponent k. An overall parameter more

sensitive to the non-homogeneous pixels was also defined and used to compare the

mixing performance for a selection of different stirrers:

1
 1 ξ AB  k

ω = 1− 
M ∑ (1 − ξ i
) k 
 (2.4)
 i A BMAX 

Fall et al. (2001) used a different mixing parameter to compare between systems: this

was based on the difference between the concentrations of two passive tracers at each

point.

C A − CB
α =1− (2.5)
C A + CB

Whilst these statistics have been designed to give a mixing criterion for continuous

processes; applications of PLIF to characterise mixing in batch vessels include studies

by Hall et al. (2004) and Chung et al. (2006). They determined mixing times within
 10

small high throughput experimentation reactors by performing a log-variance

analysis. The variance in this case is calculated by comparing each measured value,

Cn, at a specific time with the concentration when the vessel is completely mixed, Cm,

these values are normalised using the concentration before the tracer is added, C0:

Cn' (t ) =
(Cn (t ) − C0 ) (2.6)
(Cm − C0 )

Feng et al. (2005) examined mixing in planar-jet reactors, also by measuring the

variance of the intensity across the image. This statistic is also used as a mixing

parameter in work by Mortensen et al. (2004) who examined a coaxial jet in a pipe

and by Pan and Meng (2001) whose work focused on the behaviour of a tee mixer.

While CoV statistics give a measure of the intensity of the mixing it fails to describe

the degree of segregation. To achieve this a measure of the structure within the

mixture is required. One way to achieve this is to measure the striation thickness, as

this decreases the scale of segregation has decreased, alternative methods such as that

introduced by Guillard et al. (2000) use correlation functions to extract structural

information and length scales.

Kukukova et al. (2009) introduced a further dimension to evaluating the mixing in a

vessel which is analogous to the rate of mass transfer, it is the rate of change of

segregation which they labelled the exposure, which they define as:

Nt Nb
1
E = ∑∑ Kaij (Ci − C j ) (2.7)
1 1 2

Where Nt is the total number of measured points, Nb is the number of neighbouring

points, K is the strength of the interaction and a is the contact area per side, Ci-Cj is
 11

the concentration difference. Kukukova et al. (2009) state this measurement is

particularly important when the problem is dominated by the mixing timescale, such

as problems where there are mixing sensitive reactions.

2.2.2 Mixing in pipe

Jet-impaction regime

Jet-mixing regime

Wall-source regime

Figure 2.1 Illustration of the three different scenarios described by Pan and Meng (2001).

Pan and Meng (2001) identify three different scenarios when a jet enters a turbulent

crossflow; the jet-impaction regime, the jet-mixing regime and the wall-source

regime, these are demonstrated in Figure 2.1. In the jet-impaction regime the jet hits

the opposite wall and the large vortical structures are broken down improving the
 12

mixing. The disadvantage of this is the stress it exerts on the opposite wall of the tee

mixer.

The jet-mixing regime is when the jet penetrates the primary flow and bends until it is

aligned. Turbulent entrainment creates macromixing in the pipe and the most

efficient mixing in this regime occurs when the jet stream is placed so it aligns in the

centre of the primary flow (Cozewith and Busko 1989). This flow has very similar

flow characteristics to a jet in an unconfined crossflow (Pan and Meng 2001).

The wall-source regime is not efficient for mixing as the jet does not penetrate the

pipe flow, Forney and Lee (1982) found in these cases the mixing length in the pipe

(the distance downstream at which homogeneity is reached) was 50 to 100 diameters.

This mixing length is dependent on the relative diameters of the two streams and the

ratio of their flowrates. They came up with the optimum relation for the volumetric

flow in the pipe (Q) and side stream (q) and their respective pipe diameters (D,d)

1.5
q d
=  (2.8)
Q D

Pan and Meng (2001) used a pipe inlet where 6d=D, they tested two velocity ratios

v/V of 3.05 and 5.04, which give a flow rate ratio q/Q of 0.085 and 0.14 respectively.

According to equation 2.8, a centred jet would be produced with a flow rate ratio of

0.068. The lower inlet flow was closer to this ideal ratio and produced an

approximately centred jet, the higher flow rate was well above the ideal ratio which

caused the jet to impinge on the far side of the main pipe.
 13

While these rules apply to turbulent flows, for laminar flow, mixing will rarely occur

in a tee mixer or co axial pipe inlet due to the lack of radial mixing, these flow

conditions require a static mixer to enforce this radial flow (Etchells and Meyer,

2004).

2.3 Jet structure and mixing

2.3.1 Regions of an impinging jet

The structure of a turbulent jets impinging on a flat surface has been widely

investigated in the literature, largely due to their use in cooling systems (Chen et al.,

2005). A large amount of work has been carried out looking at planar or slot jets

(Law and Wang, 2000, Fondse et al., 1983, Kim et al. 2007) but Reungoat et al.

(2007) used PIV to measure the turbulent mixing in a round impinging jet at different

impinging distances. There are three main regions used when describing an

impinging jet; the free jet region, the impingement zone and the wall jet region, these

are illustrated in Figure 2.2.

Figure 2.2 Illustration of the regions of an impinging jet, Reungoat et al. (2007).
 14

The free jet region describes the jet between the injection point and where the

impingement begins to have an effect, in this zone the jet behaves as a free jet. Near

the point of injection, the momentum is maintained in the centre of the jet, this is

referred to as the potential core. It was observed by Reungoat et al. (2007) who

looked at the centreline velocity of the jet, the diameter of this core gradually

decreases as the mixing layer around it expands until about 6 nozzle diameters from

the injection where the jet moves into the developing zone. This agrees with Revill

(1985) who stated the same relation for a free jet in an unbound bulk liquid.

In the region of 6-50 nozzle diameters, Reungoat et al. (2007) found the centreline

velocity decreases linearly with distance, this is the developing zone and is in

agreement with previous work on slot jets such as Law and Wang (2000). Fondse et

al. (1983) describe this region as extending to 70 nozzle diameters away from the

injection point where the zone of established flow or self similar zone begins. Once

self similarity has occurred the cross sectional flow profile of the jet can be shown to

be Gaussian (Reungoat et al. 2007).

In the impinging region the jet is redirected from an axial into a radial direction, this

region is composed of a stagnation zone and then three wall zones (Chen et al. 2005).

The concentration of the tracer fluid in the Reungoat el al. (2007) investigation

appears to be high in the stagnation zone where the velocity of the fluid is low. They

also observe vortical structures that roll along the wall, these are four times larger

then the observed boundary layer.

 15

The concentration distribution of a passive tracer in a jet was examined by Guillard et

al. (1998) who used the difference between the 25% and 10% loci to determine the

mixing layer. This mixing layer increased linearly away from the inlet until it reached

a point 1.5 nozzle diameters from the impingement wall where the increase became

exponential; this curvature was also found by Reungout et al. The diffusion of the jet

is accelerated due to the deceleration of the jet caused by impingement on the wall.

Ding et al. (2003) found that the concentration core in the jet decays much quicker

than the velocity core of the jet, they attribute this to the faster transport of a scalar

quantity such as concentration than momentum, due to the pressure gradient in the

momentum transfer equation suppressing the turbulent transfer.

The studies above are all carried out on turbulent jets; Revill (1985) states that a jet is

fully turbulent at Re above or about 1000-2000 and laminar for Re below 100. Kim et

al. (2007) investigated water jets between Rej = 404 and Rej = 1026 and found that

the jet underwent a transition to unsteady within this range. Pawlak et al. (2007) look

at the dynamics of a starting laminar jet, the defining feature of these jets is the

leading vortex ring, which is followed by a secondary vortex, behind this leading

vortex pair further vortex pairs develop and grow.

The effects of the boundary layer on the mixing performance of a jet has been

investigated, Ding et al. (2003) used PLIF with a local injection point at the edge of

the jet entry to look at the mixing layer, the concentration fluctuation at each point

was measured and it’s maximum coincided with the area of maximum shear stress.

Fondse et al. (1983) looked at the influence of the exit conditions on the entrainment

rate and found the most important factor was whether the boundary layer was laminar
 16

or turbulent, with laminar jets entraining 15% more than turbulent jets. Placing a grid

across the outlet to promote turbulence in both the jet core and boundary layer

reduced entrainment by 40%.

In all the current jet impingement studies the distance between the nozzle and the

impingement wall remained constant throughout the experiment. The fluid height in

the vessel was maintained using a side weir (Reungoat et al. 2007) or similar

recirculation and the vessel was large enough that the effects of the side walls on the

jet are not considered.

2.3.2 Jets in large storage tanks

Jet mixing is also found in large storage tanks where fluid from an outlet is circulated

and jetted back in. Initial work on a simple system was carried out by Fossett and

Prosser (1949) who were investigating underground fuel storage tanks during World

War II. They proposed a relationship between the mixing time in the vessel, the

vessel diameter (T), the nozzle diameter (D) and the jet velocity (U):

T2
θ∝ (2.9)
UD

Although other relations have been suggested this was found to be the most accurate

in a number of studies (Grenville and Nienow 2004).

The effect of the jet angle, the vessel geometry, symmetry and aspect ratio, the nozzle

diameter and the jet Reynolds number are all considered in the recent literature. One

of the underpinning papers in this field is Lane and Rice (1982) who compared three

different designs of jet mixing. The first was an inclined side entry jet near the base
 17

of a flat bottomed cylinder, this design is refined to maximise the length of the jet, the

second design was an axial vertical jet in a flat based cylinder, and the third was an

axial vertical jet in a hemispherical based cylindrical tank. They found that the

hemispherical base is the best for mixing. This is attributed to reduced stagnant areas

in the bottom corners of the tank and that the liquid is constantly being redirected

from the edge back into the jet.

Patwardhan and Gaikwad (2003) use a side entry cylindrical tank to investigate the

nozzle angle and diameter. They use the same power input at different angles and

find that the optimum mixing occurs when the jet is angled at 45 degrees to the base,

they believe this is due to maximising the length of the jet so it entrains more of the

surrounding liquid. The mixing is aided in this case by the outlet, which is situated in

the poorly mixed bottom corner opposite the jet. The nozzle diameter is also varied;

they found that increasing the diameter improves the mixing at the same power

consumption.

The effect of jet angle can be modelled using computational fluid dynamics (CFD).

Zughbi and Rakib (2004) used FLUENT with a tetrahedral mesh for their simulations,

they validated their results using the experimental results of Lane and Rice (1982).

They used the temperature rather than the concentration to measure the mixing time, it

was assumed that density and viscosity were constant over the temperature range used

so that the flow properties are unaffected. Their results showed that the optimum

angle is around 30 degrees to the base of the tank for entry at the base. This appears

to disagree with the theory presented originally by Lane and Rice that the optimum

mixing occurs when the jet length is maximised.

 18

A theory to explain the results was suggested in a following paper (Zughbi and

Ahmad 2005). It states that the jet length does need to be maximised but this does not

necessarily occur at 45 degrees. The jet may have dissipated before reaching the

opposite corner, however if the jet is deflected off the opposite wall a greater jet

length is achieved. This theory agrees with the evidence from tee mixer experiments

where the mixing is also improved though deflection of the jet.

2.3.3 Impinging jets

Studies have also been carried out on the mixing that occurs between two impinging

jets. The work by Unger and Muzzio (1999) investigates the two different geometries

for mixing jets shown in figure 2.3. Low Reynolds number jets (Re < 80) were better

mixed in the asymmetric geometry this was due to the swirling motion created, little

mixing occurred when these jets were fired directly at each other. For unsteady

laminar flows (80 < Re < 300) the symmetrical geometry created better mixing, this

was thought to be due to the oscillations in the jet streams which are more effective at

mixing than the swirling nature of the flow in the asymmetric case. High Reynolds

number jets were effectively mixed in both geometries although the asymmetric case

was more efficient as it reduced the dead zones in the vessel.

 19

Figure 2.3 (Unger and Muzzio 1999) Geometry of the impinging jet mixers.

2.4 Mixing and flow diagnostics

Mixing and flow diagnostic techniques can be either single point or whole field

measurements. Single point measurements detect the instantaneous flow velocity or

concentration at a fixed point at a defined frequency in time, while the whole field

techniques measure over an entire area or volume at an instant in time, examples are

given in table 2.1.

Single Point Whole field

Velocity Field Hot wire anemometry (HWA), Particle Image

Positron Emission Particle Tracking Velocimetry (PIV),
(PEPT), Laser Doppler Velocimetry Magnetic Resonance
(LDV), Computer Automated Imaging
Radioactive (MRI).
Particle Tracking (CARPT)
Concentration Conductivity, pH or temperature Planar Laser Induced
probes Fluorescence (PLIF), mass
fractions of the products
from competing reactions
Table 2.1 Measurement techniques
 20

The techniques relevant to this work are the optical techniques of PLIF and PIV, this

review is focused on these two methods.

2.4.1 PLIF

Planar Laser Induced Fluorescence (PLIF) is a technique based on the measurement

of the instantaneous concentration distribution of a fluorescent material in a specific

plane. To relate the intensity of the fluorescence of the tracer to its concentration it is

important that there is a linear relationship between them; this is the case for dilute

tracers and is verified and calibrated before each study. PLIF has been used as a

technique in a number of different application areas including combustion of gases

(Hult et al., 2005; Degardin et al., 2006; Weigand et al., 2006), flows of liquids in

vessels (Hall et al., 2004; Dazin et al., 2006) and impinging liquid jets (Unger et al.,

1999).

Analysis of the data obtained from PLIF analysis requires the creation of a matrix of

greyscale values from the black and white digital images obtained, this matrix is then

used to compute various statistics such as those outlined in section 2.2.1. Due to the

large amount of data to be handled, analysis using computation is essential. The use

of the software package, Matlab® to analyze the results was investigated by Golnabi

(2006) who described how Matlab can be used to import and convert an image into a

two dimensional array with values based on the intensity of the image data at each

pixel.

There are a number of error sources in PLIF measurements, mainly associated with

the optics, camera sensitivity, the Gaussian distribution of the laser intensity and

temporal laser fluctuations. Golnabi (2006) used a continuous flow at a constant

concentration illuminated from the side by a Nd:YAG laser, the signal was then
 21

averaged along each row and column. The signal intensity was found to decrease

linearly across the cell, which was attributed to a decrease in camera sensitivity away

from the centre of the image. The errors that occur in PLIF images were also

analyzed by Law et al. (2000) who state the importance of the perpendicular

positioning of the camera relative to the light sheet to avoid deformation in the

images. The unsteadiness of the laser sheet was also measured in this study by

sampling 32 frames of uniform concentration; they found the maximum standard

deviation for individual pixels to be 2 grayscale units on an 8 bit (256) grayscale,

which was determined to be insignificant.

The above studies show that there is a body of work on the application and validation

of PLIF to obtain an understanding of mixing in steady flows with minimal free-

surface disturbance. However, the use of PLIF to gauge mixing performance in

unsteady (time-dependent) flows with moving free-surfaces has received very little

attention.

2.4.2 PIV

Particle Image Velocimetry (PIV) is a whole field visualisation technique which

yields an instantaneous velocity field obtained over a short time interval over a given

flow area of interest. This is achieved through measuring the displacement of

particles within a thin laser sheet. PIV is non intrusive, indirect velocity measurement

technique has developed rapidly recently with the improvements in cameras lasers,

optics and software (Raffel et al. 2007).

The PIV technique is demonstrated in Figure 2.4 This area of interest is illuminated

by a laser sheet and images of this sheet are captured using a camera. Each image is
 22

then split into a grid of interrogation windows and statistical techniques are used to

assign a velocity vector to each window.

frame 1

Crosscorrelation

Particle
displacement
frame 2

Vector
field

Figure 2.4 PIV analysis

Pawlak et al. (2007) look at the dynamics of a starting laminar jet using a dynamic

particle image velocimetry (DPIV) system and PLIF. The DPIV system uses cross

correlation to produce a coarse grid of velocity vectors, filters this to remove outliers

then a fine grid analysis is performed, the resulting velocity field is filtered again

using a threshold median filter to eliminate outliers. The individual velocity fields are

then presented alongside the PLIF images for this transient process. Six different

integrals are calculated in the region of the leading vortex, the first is proportional to

the axial momentum flux exiting a disc perpendicular to the axial direction, the

second is related to the flux of vertical momentum and the third is responsive to radial

velocity changes. These three integrals are used to determine the exact axial location
 23

of the vortex and a further three similar cases are used to determine the radial

position. The vorticity of the vectors was also used to analyse the local recirculation

in the vortex.

Turbulence is where there are temporal fluctuations in the fluid velocity at any point,

this can be measured through the PIV technique. It is assumed that the instantaneous

velocity can be decomposed into an average velocity and a fluctuating part. This

fluctuating part is usually characterised through an RMS value, the turbulent kinetic

energy (TKE) is the portion of the kinetic energy that is due to these fluctuations. As

PIV measures the velocity in a plane the third dimension is estimated from the two

dimensions in the frame by assuming the turbulence is isotropic. Saarenrinne et al.

(2001) states that the spatial resolution needed to capture 90% of the turbulent energy

is twice the Kolmogorov scale however Chung et al. (2006) obtain 95% accuracy with

a lengthscale of 20 times the Kolmogorov lengthscale.

Turbulent Kinetic Energy is a common statistic used in the analysis of turbulent jet

flow (Weisgraber and Liepmann, 1998, Bi et al., 2003,) however it may not be

possible to perform this on a transient flow, other analyses performed on a turbulent

jets include proper orthogonal decomposition. Bi et al. (2003) used DPIV and proper

orthogonal decomposition (POD) to determine the coherent structures in the flow.

The principles behind this involve maximising the correlation between the flow

description and the experimental results from the PIV, this is done by finding the

eigenmodes of the flow. They captured the velocity field at a frequency of 1000Hz,

using this data they plotted the spectral density of the frequency of the velocity

fluctuations in the centre of the shear layer on one side of the jet and found that the
 24

preferred frequency of the fluctuations was approximately 11Hz. The PIV images

were then grouped into 80 blocks each containing 256 files giving a bandwidth of

3.9Hz, these were then analysed using POD, they note that the first six spatial

eigenmodes capture over 50% of the energy, lower than in traditional turbulence as

more eigenmodes were needed to capture the higher number of degrees of freedom of

turbulence caused by carrying out the analysis in the streamwise direction and in the

temporal domain. Kim et al., (2007) state that the eigenvalue represents the energy of

each eigenmode, they note that the proportion of energy in the first six modes

increases as the Reynolds number increases, which implies the flow has lower

dimensional structures.

2.5 Summary

Although there is little published research into the mixing that occurs as a vessel is

filled there is a large body of work into mixing by a range of other mechanisms. This

mixing is measured in a variety of ways from full field techniques such as PILF to

single point measurements of concentration and is often quantified using a coefficient

of variance statistics. From studies on mixing involving jets it can be seen that the

mixing performance improves if the length of the jet, including any impinged flow, is

maximised.

The techniques PLIF and PIV have been used to measure a variety of flow scenarios,

however in most cases the system is a steady state. There has been no study of

transient systems with a moving free surface, this work will adapt these techniques to

measure the mixing in this case.

 25

Chapter 3 Materials and Methods

3.1 Introduction

A number of experiments were carried out throughout the project to determine the

flow properties and mixing performance for filling systems. Initially work was done

using a square section tank and a stationary pipe, this was then modified so that the

pipe could retract above the surface of the fluid. In both cases experiments were

carried out using PLIF and PIV to determine the level of mixing and the velocity field

respectively. Parameters such as nozzle design and flow rate were adapted based on

these results and the modifications were examined experimentally.

3.2 Mixing tank

3.2.1 Dimensions and equipment set up for static pipe system

A schematic of the experimental set-up is shown in Figure 3.1. The vessel used was a

T = 0.06 m square section tank; the square shape was chosen to eliminate any errors

due to refraction through curved surfaces. The fluid enters the tank through a 0.006 m

diameter vertical dip pipe aligned with the central axis of the tank. The tank was

filled to a height, H, of ~ 0.1 m (H/T = 1.67). A digital dispensing gear pump

(MicroPump 75211-35, Cole-Palmer USA) was used to produce a continuous pulse-

free flow. The volumetric flow rate of the pump was calibrated to the pump speed

(controlled digitally) for each viscosity of fluid used using the bucket and timer

method. Additional checks on the flow rate supplied by the pump were made by

monitoring the rise in liquid level in the tank as a function of time.

 26

Figure 3.1 Schematic of the experimental rig and PLIF set up.

3.2.2 Moving nozzle set up and control

Entrainment of air in the jet causes bubbles to be formed in the tank, when the laser

sheet hits these bubbles it is scattered causing shadows in the plane of interest. Bin

(1988) calculated the impingement height of the jet Lj for which air entrainment

begins in the region where the plunging jet would split into individual droplets region

as:

Lj −1.79
= 1.4 × 10− 5 d 0 We0j .937 (3.1)
d0

Where We is the Weber number which is calculated from the density (ρ) the velocity

(v), the lengthscale (l) and the surface tension (σ):

ρv 2 l
We = (3.2)
σ
 27

This gives an approximate value of Lj as 16 cm for the highest velocity fluid used in

these experiments, however in the continuous jet region entrainment will occur at a

shorter length. Whether the jet is in the continuous or droplet regions is dependent on

the inlet turbulence parameters. Initial test cases demonstrated air entrainment when

the pipe was fixed above the maximum fill level of 10cm, an impingement height of

3cm prevented most entrainment however a pipe fixed in this position would quickly

become submerged. To allow the jet to impinge on the surface of the fluid while

preventing the impingement height from exceeding this limit the nozzle was raised so

it stayed just above the level of the fluid in the tank.

To adapt the filling set up to allow the pipe to be moved, a purpose built traverse was

designed (Figure 3.2) which allowed the pipe to move in the vertical direction at a

controlled speed and with minimum vibration. This consisted of a motor which

rotated a vertical screw thread along which the pipe moved guided by two supports.

The speed was controlled using a specifically written control program and was

manually set before filling began. This speed was chosen to be the same as the rate at

which the free surface rose which was calculated from the flow rate. The rig included

cut off switches positioned to limit the pipe’s range of movement to keep it within the

range of the screw and also to prevent it from colliding with the vessel. The traverse

was started manually at the beginning of the filling process to coincide with the

initialization of the pump.

 28

Motor controlling
traverse
Traverse path

Clamp for
nozzle
Limit switches

Pyrex square
section tank

Control box with

connections to PC
Perspex platform with
machined hole to allow
laser sheet pass
through and vessel
drain to sit flat.

Laser source
supported by
bread board

Figure 3.2 Schematic of the nozzle traverse.

3.2.3 Set up of case studies

A number of improvements to the system were investigated experimentally. These

improvements fell into two groups; varying the flow rate and changing the pipe

geometry.
 29

A variable flow rate was created using a computer program to control the pump, this

was written by Andrew Tanner from the Biosciences Workshop at the University of

Birmingham. The program was written so that an excel file read the time and the

flow rate and any function that could be written in excel could be used to describe the

flow, the program then runs until it reaches an empty cell, at this point the pump is

stopped. The actual flow rate achieved is fed back to the computer and is displayed

graphically alongside the prescribed function this can then be exported to excel. Due

to limitations of the pump there was a lead time between the function and the actual

speed achieved and depending on the desired flow rate there was a certain length of

time to start up. This could be up to 2.5 seconds for the highest flows which meant

that there was a limit on the rate of change of the flow. For a sinusoidal function the

period achievable was 4 seconds, the flow rate was varied from 1.67 x 10-5 m3 s-1 to 5

x 10 -5 m3 s-1. For the ramped flow the flow rate started at zero and accelerated up to

the highest rate of 6.67 × 10-5 m3 s-1. In both cases the average flow rate was 3.33 x

10-5 m3 s-1.

Three nozzle designs were made to improve the mixing in the tank, two of these had

cone inserts in the base and one had a ‘swirl’ insert. For the cone insert the pipe was

flared to accommodate it, as shown in Figure 3.3, in practice spot welds were

required to attach the inserts to the outside of the pipe. The swirl nozzle was created

by inserting and welding a twisted piece of metal into the end of the pipe.
 30

Cone 1 Cone 2 Swirl

Straight

6mm 6mm 6mm 6mm

10mm

8mm 6mm
10mm 8mm

Figure 3.3 Schematic of a straight pipe and the three nozzles improvements.

3.3 Fluid properties

3.3.1 Newtonian Fluids

The experiments were carried out using Newtonian fluids. It was necessary for the

fluids to be transparent to allow the laser beam to pass through for the PLIF and PIV

diagnostic techniques. Aqueous glycerol solutions were chosen for initial studies,

with viscosities ranging from 0.003 Pa s to 0.016 Pa s measured using a rheometer

equipped with a cone and plate geometry (TA AR1000, TA instruments). The fluid

properties are described in Table 3.1

 31

Stationary pipe experiments Moving pipe experiments

Energy dissipated,

number, Re (-)

number, Re (-)
length scale, λ

length scale, λ
ε (W kg-1)
Fill time, t

Kolmogoroff

Kolmogoroff
Flow rate,

Viscosity, μ

Viscosity, μ
Density, ρ

Density, ρ
(m3s-1)

Reynolds

Reynolds
(kg m-3)

(kg m-3)
(Pa s)

(Pa s)
(s)

(μm)

(μm)
Q

3 × 10-3 1056 1245 231 4 × 10-3 1081 956 282

1.67 × 10-5 21.6 0.008 6 × 10-3 1121 661 389 8 × 10-3 1126 498 459

12 × 10-3 1164 343 654 16 × 10-3 1170 259 751

3 × 10-3 1056 2490 131 4 × 10-3 1081 1911 168

3.33 × 10-5 10.8 0.064 6 × 10-3 1121 1322 221 8 × 10-3 1126 996 273

12 × 10-3 1164 687 372 16 × 10-3 1170 517 446

3 × 10-3 1056 3735 94 4 × 10-3 1081 2867 124

5.00 × 10-5 7.2 0.217 6 × 10-3 1121 1983 158 8 × 10-3 1126 1494 202

12 × 10-3 1164 1030 266 16 × 10-3 1170 776 329

ρd j u j
Table 3.1 Experimental parameters, where Reynolds number is defined as Re = .
µ

3.4 Planar Laser Induced Fluorescence (PLIF)

PLIF images were obtained using a TSI Powerview system (TSI Inc. USA)

comprising of a single 1000×1016 pixel, 8 bit Charged Coupled Device (CCD)

camera (TSI PIVCAM 10-30, TSI Inc, USA), synchroniser and a dual head Nd-Yag

laser emitting at 532 nm (New Wave Inc., USA) and equipped with laser sheet optics.

The lenses used were a -15 mm cylindrical lens and a 200 mm spherical lens and the

laser was positioned 150 mm from the base of the vessel; this combination was

chosen so that thickness of the sheet was a minimum in the middle of the vessel. The
 32

laser sheet thickness throughout the imaging plane was less than 1 mm. The system

was controlled using a Dell Precision 620 workstation (Dell Corp. USA) running

INSIGHT 6.0 software. The images were captured at a rate of 15 Hz, saved as .tif

files and exported to the Matlab software for analysis.

The fluorescent tracer used was Rhodamine 6G which emits light at 566 nm when

excited using radiation at 532 nm, Rhodamine was chosen as the relationship between

the concentration and the intensity of the laser light emitted and hence the grayscale

of the images is linear over a certain range of concentration, this is known as the

Lambert-Beer law. This linear relationship was measured for different

concentrations, this was carried out at two different laser intensities; the graph

obtained is shown in Figure 3.4. The linear relationship of concentration is seen to

hold up to a concentration of 0.0001 kg m-3 (0.1 mg L-1) and of the two laser

intensities used the higher gave a better resolution over the required range and was

therefore used for the experiments. Before the fluid injection began 1 × 10-6 m3 of the

tracer was spread across the bottom of the vessel, as this tracer is then diluted by the

inflowing fluid the concentration of the tracer was chosen so that the concentrations in

the vessel would be within the linear range when the level of mixing was being

measured. Measurement took place when the fluid level reached approximately 2 cm

until the vessel was full, this meant the tracer would be diluted by a factor of 100

before measurement took place, so the concentration used was 0.008 kg m-3

(8 mg L-1), as this would ensure that the concentration was within the linear region

before measurement began. The camera was fitted with a sharp cut-off high-pass

wavelength filter of 545 nm (ALP545, TSI Inc.) to eliminate the light from the laser at
 33

532 nm and allow only the fluorescent light from the tracer fluid at 566 nm to pass

through.

Figure 3.4 Plot of greyscale as a function of tracer concentration at two different laser intensities.

3.4.1 Normalisation

As shown above, the relationship between the measured grayscale values and the

tracer concentration was found to be linear throughout the concentration range found

when the fluid heights were above 2 cm. To determine this linearity precisely a set of

images must be taken to calibrate the grayscale values measured for each pixel in the

image to the tracer concentration.

There are two main considerations to take into account when determining the pixel by

pixel values of concentration from the grayscale values; the laser sheet intensity has a

Gaussian profile across the plane so a uniformly mixed tank will not emit a uniform
 34

intensity of light, and the individual pixels of the CCD array in the camera do not

have a uniform sensitivity. To eliminate these sources of error each pixel is calibrated

independently. This is accomplished by taking 50 pictures at one known

concentration with the vessel fully mixed; this process is then repeated with double

the known concentration. The average grayscale value for each pixel over the 50

images is then calculated at each concentration and hence two arrays containing the

pixel by pixel values of grayscale are created. The linear relationships connecting the

corresponding pixel by pixel values of greyscale with concentration in the two

matrices are determined and a further two matrices are created containing the gradient

(M) and intercept (K) of these linear equations. This is demonstrated in Figure 3.5

using the first two pixels as examples, the nth element of the M or K matrix is

calculated from the measured greyscale at the nth pixel of the image when the

concentration is x and the greyscale when the concentration is 2x.

The zero concentration case is not used in the calibration as it does not include the

error associated with any non-uniformity in the laser sheet. This is because with no

concentration there is no fluorescence is emitted which the only detection method for

the local intensity. However they do allow the level of this variation to be quantified.

If the linear relationship obtained from two non-zero concentrations is extrapolated

back to the zero point there is found to be an error of 2.26 % between the calculated

‘K’ values and the experimental greyscale values for an image of an illuminated

vessel with no tracer.

 35

Figure 3.5 Calculation of elements of the matrices M and K from the linear relation between greyscale

and tracer concentration for each pixel, (only two pixels are plotted for demonstration).

Each image is then loaded one at a time into Matlab as an array of greyscale values

(see appendix for sample code). The image is then cropped so that only the area of

the image containing fluid is analyzed. The surface of the fluid is located by

calculating the average of each row and determining when this average is more than

10% higher than the background grayscale, the first point at which this is the case is

taken to be the free surface of the fluid. This was checked and found to be the best

cutoff by viewing cropped images to ensure that the fluid area had been correctly

identified, the calculated height of the fluid is also monitored to ensure it increases

linearly throughout the filling processes. If the height does not increase steadily with

image number then the images are not accurate and are discarded, this occurs when a

droplet forms on the pipe of edge of the tank which reflects the light higher in the
 36

vessel. For the purposes of this analysis the image is split into two interrogation

zones, one on each side of the inlet pipe. The change in the average greyscale values

at the surface and the area used for the analysis are shown in Figure 3.6.

Figure 3.6 Definition of the location of the fluid surface via the change in the average row greyscale

values as a function of the vessel height. The interrogation zones are shown and also the maximum fill

height.

The height of the liquid in the vessel is then calculated from the number of pixels that

cover it using a scaling factor calculated using the base of the tank as a standard

distance, and from this the average concentration of the fluid in the vessel at that

height. This is then converted into a grayscale value (Gn) at each pixel using the

calibration matrices. The average concentration in the vessel is the concentration

each pixel would be at if the vessel were fully mixed; it is dependent on the fluid

height in the vessel, the concentration of the tracer and the volume of tracer used.
 37

This concentration decreases as the vessel is filled which means that the degree of

mixing is being measured against a coarser scale when the vessel is nearly full.

3.4.2 Log variance and image analysis

To obtain the degree of mixedness in the vessel as a function of height a log-variance

analysis is used as described by Brown et al. (2004). The grayscale value at the nth

pixel is normalized at each time t by considering the grayscale values when the

concentration is zero i.e. background (G0) and the value if the vessel was completely

mixed at that fluid height (Gm);

(Gn (t ) − G0 )
Gn' (t ) = , (3.3)
(Gm − G0 )
where G’n (t) is the normalized grayscale value and Gn (t) is the grayscale value of the

nth pixel in an image taken at a time t. The log-variance (σ2) of the concentration in

the interrogation area is then calculated, this is defined as;

1
∑ (G )
N
2
log σ 2 = log '
n −1 , (3.4)
N n =1

where N is the total number of pixels used in the image. The log-variance gives a

direct measure of the level of mixing on the scale of a single pixel and as the mixing

improves this number decreases. For example, when the vessel is 95% mixed the log-

variance is log (1 - 0.95)2 = -2.6.

The number of pixels included in the variance increases with the height of the fluid

due to the larger area covered, this causes the error in the measured log variance to be

reduced though the process as the sample size increases. However the scale of the
 38

mixing being measured does not change throughout the fill, for this to be reduced a

higher resolution camera would need to be employed.

In order to ensure that the equipment has sufficient resolution to resolve mixing

events across all the length-scales, a characteristic length scale equivalent to the

Kolmogoroff scale for turbulent mixing was calculated. Although the mixing

mechanism was shown later to be predominantly laminar for the higher viscosity

fluids, hence the Kolmogoroff scale would not be expected to apply, calculation of

this scale provides an order of magnitude check on the likely accuracy of the

technique. The Kolmogoroff length scale is a function of the energy input per unit

mass to the fluid, ε, and the kinematic viscosity of the fluid, ν:

1/ 4
ν 3 
λ =   (3.5)
ε 

In a filling system, if the fill rate is constant, the mass of fluid in the vessel increases

linearly in time. The energy input per unit mass to the fluid from the liquid jet was

calculated on the basis of the dissipation of the kinetic energy into the bulk fluid in the

fully filled vessel; this provides the minimum value of specific energy, ε, input

during the filling process,

1
ρQu 2j
ε= 2 , (3.6)
ρT 2 H

where Q is the volume flow rate of liquid, ρ is the density, u is the velocity of the jet

and T and H are the height and width of the tank.. Values of λ and ε, calculated using

(3.5) and (3.6) respectively are given for each experiment in Table 1. The range of λ

is from 94-650 μm, this compares well with a pixel by pixel resolution in the images

of 130 μm.
 39

3.4.3 Calibration and error

Using the calibration method, the pixel by pixel variations in the laser sheet and in the

camera sensitivity are taken into account, however, this analysis assumes that the

intensities within the laser sheet are constant each time the laser head is triggered.

The fluctuation in the laser sheet with time was examined by observing multiple

calibration images. The concentration and liquid height were kept constant for 50

images so that temporal fluctuations in the signal could be attributed to the laser. The

laser sheet fluctuation had an average standard deviation of 2% for each pixel. This

error propagates through the calculations, there will be a 2% error in the two

intensities used to calculate the values of M and K, giving a 2.8% error in each pixel.

These are in turn used to calculate the fully mixed concentration and the normalized

concentration. Despite this propagation of error the large number of pixels used for

each log-variance value reduces the error. If more than 120 pixels are used it is

reduced to below 1% giving an error due to laser fluctuation in the log value of less

than 0.004, for a full vessel more than 15000 pixels can be used.

A further cause of anomalous concentration results occur when there are pixels which

have very low sensitivity or have been damaged. When the difference between the

grayscale value at zero concentration, Go and at the fully mixed condition, Gm, is too

small, their use in the denominator of Equation 3.2, leads to a very coarse

measurement which is not statistically significant. Any damaged pixels in the camera

appear as permanently saturated with light so the differences between the grayscale

values from these pixel is zero, this means that when the concentration values are

normalized an infinite answer is obtained. To eliminate both these problems all pixels

where the difference between the concentration at zero and the concentration at fully
 40

mixed is less than a certain cut off value, usually taken as 10 are discarded prior to

taking the log-variance.

3.5 Particle Image Velocimetry (PIV)

PIV images were obtained using a TSI high-speed PIV system. A high repetition rate

laser (10mJ at 1kHz) was used for illumination (New Wave) and a high frame rate

camera (Photron APX RS) to capture data at 500 Hz with a spatial resolution of 1025

× 1025 pixels2. The lens set up used was identical to that used in the PLIF

experiments described above.

The particles used to trace the fluid were silver coated glass particles of diameter 10

μm which reflected the light at the same wavelength as the initial laser plane (532

nm). These particles have a sufficiently small relaxation time that they can be

assumed to be following the fluid motion; they were well mixed into the fluid before

the injection of the fluid into the vessel during the experiment. The instantaneous

spatial displacement of the particles in the vessel was calculated by splitting the

images into small interrogation windows 32x32 pixels. Fast Fourier Transform and

Gaussian statistics were then used to cross correlate between the images from the two

time exposures. The spatial displacement that produces the maximum cross-

correlation statistically approximates the average displacement of the particles in the

interrogation window (Hall et al. 2005). The velocity of the particles can then be

calculated by combining this information with the known pulse separation. The

resulting matrix of velocity vectors defines the instantaneous flow field in the whole

image.
 41

The resolution of the velocity field measured using the PIV is dependent on the

magnification of the camera, the size of the interrogation window and the pulse

separation. The optimum pulse separation is calculated by defining the maximum

displacement between the pulses as one quarter of the interrogation window. The

amount of particles in each interrogation window also affects the accuracy of the

cross correlation, the optimum seeding for the fluid is 10-12 particles per

interrogation window.

3.5.1 Assumptions and analysis techniques

To obtain an accurate picture of the flow it is necessary to average over a number of

images, however in a transient flow this is not possible. An assumption that the fluid

height was stationary throughout the capture of 200 images was made so that

averaging could be used. The time needed to capture 200 images was 0.4 sec, for the

worst case scenario, a flow rate of 5.00 × 10-5 m3 s-1 the fluid height changed by

0.0056 m in this time; for the assumption to be valid the flow conditions must not

change significantly due to this height change. Sets of images were taken when the

fluid level reached 3 cm, 5 cm, 7 cm and 9 cm at two different pulse separations, this

was necessary due to the large difference in the velocity magnitude between the bulk

fluid and the jet. The data from both pulse settings was then combined during

analysis to give a complete picture of the velocity fields, an example MATLAB code

is included in the appendix.

As a large amount of the mixing that occurs is laminar the areas of high shear in the

vessel are important, as it is the elongation and shear as well as folding of fluid that

causes it to mix. The magnitude of the shear rate in the x and y direction is calculated

by:
 42

∂v x ∂v y
γ = + (3.7)
∂y ∂x

This is then plotted as a colour map with the velocity vector field, to show the regions

of high shear.
 43

Chapter 4 PLIF – Concentration fields 1

4.1 Introduction

The results presented in this Chapter provide basic knowledge of the mixing in the

simplified systems described in section 3.2. The first of these had the inlet pipe fixed

1 cm from the base of the tank; later experiments used a traverse to keep the pipe

above the liquid free-surface as the vessel filled. Jets with nine different Reynolds

numbers from 259 to 2867 were investigated by changing both the jet exit velocity

and the viscosity of the fluid, this range allowed the transition from relatively

unmixed laminar flows to better mixed turbulent flows to be observed and identified.

The technique employed in this Chapter to determine the mixing properties of the

flow was Planar Laser Induced Fluorescence (PLIF) as described in section 3.4. The

results obtained from this technique are discussed and where possible conclusions

have been drawn.

4.2 Static pipe

4.2.1 Log Variance analysis

The simplest system set up investigated using PLIF had a fixed pipe inlet (as

described in section 3.2) this system was used to develop the PLIF measurement

technique over an appropriate range of flow conditions chosen for investigation; these

conditions are described in Table 3.1.

1
A selection of the PLIF results presented in this Chapter for the fixed pipe inlet have been published

in the archival Journal Experiments in Fluids (Neal et al.,. 2008)

 44

Figure 4.1 Evolving concentration fields at different fill heights for the 6 mPa s fluid at (a) h = 2.2 cm;

(b) h = 4.7 cm; (c) h = 7.3 cm; (d) h = 10.3 cm at Re=661 (Q = 1.67 × 10-5 m3 s-1).

Figure 4.2 Evolving concentration fields at different fill heights for the 3 mPa s fluid at (a) h = 2.2 cm;

(b) h = 4.7 cm; (c) h = 7.3 cm; (d) h = 10.3 cm at Re = 1245 (Q = 1.67 × 10-5 m3 s-1).
 45

Figure 4.1 and Figure 4.2 show greyscale images taken by the CCD camera for the 6

and 3mPas fluids at values of Re of 661 and 1245 respectively. Both fluids generate a

similar flow pattern; a toroidal vortex is formed in the bottom quarter of the vessel

(Figure 4.1a, Figure 4.2a) whilst the flow in the upper half of the vessel is comparably

quiescent and does not have a well defined structure (Figure 4.1c-d, Figure 4.2c-d).

The striations created by the fluorescent tracer remain more defined in the low

Reynolds number fluid than for the high Reynolds number fluid when the vessel is

filled; although clearly in neither case is the vessel fully mixed.

Using the naked eye, the degree of mixedness appears to be higher for the higher

Reynolds number fluid as both the scale of the striations is smaller and the contrast

between them is lower, suggesting qualitatively a reduction in both the scale and

intensity of segregation respectively. The existence of the striations throughout the

filling process is indicative of a predominantly laminar mixing process (Ottino, 1989)

since eddy diffusion would be expected to eliminate them if the flow were turbulent.

The images shown here may form a basis for a mechanistic determination of the

transition to turbulence, since, if molecular diffusion is considered negligible, a

laminar mechanism would be characterised by clearly identified striations of black

and white, whilst eddy diffusion would be expected to cause regions of ‘grey’.

Comparing Figure 4.1d and Figure 4.2d, regions of the vessel appear more mixed in

Figure 4.2d, which is perhaps indicative of the beginning of this process.

 46

Figure 4.3 Mixing performance for three different fluid viscosities at a fill rate of Q = 1.67× 10-5 m3 s-
1
(a) log variance plotted against dimensionless fluid height (H/T); Image of the vessel filled with fluid

at (b) Re=343 (12 mPa s), (c) Re=661 (6 mPa s), (d) Re=1245 (3 mPa s).

Figure 4.3a shows the log-variance, as described in section 3.4.3, as a function of

dimensionless fluid height (h/T) in the vessel for three fluids of decreasing viscosity

at the same fill rate of Q = 1.67×10-5 m3s-1, giving Reynolds numbers of 343, 661 and

1245 respectively. Each experiment was repeated a minimum of two times. At a

value of Re of 1245 (red line), the log-variance reaches a minimum of ~ -1.8 (87%

mixed) once the vessel is filled beyond 3 cm in height and remains constant until the
 47

fill is complete. At values of Re of 661 and 343 (blue and green lines respectively),

the log-variance stays at a value close to zero throughout the fill, indicating that the

vessel remains unmixed.

There is a high level of noise in the log-variance plot, a small amount of which will be

due to the measurement technique however comparison between the low and higher

Reynolds number flows show that there is an increase in the noise for the higher

Reynolds number, this could be caused by areas of high or low concentration moving

through the measurement plane tangentially causing the average concentration in the

measurement plane to fluctuate.

These results are borne out by the images shown in b-d which display an image of the

filled vessel taken at the instant after completion of the filling process at values of Re

of 343, 661 and 1245 respectively. The tracer remains unmixed with the bulk of the

fluid in Figure 4.3b and c and is located either in the toroidal vortices at the bottom of

the vessel or is swept up close to the fill pipe or the free surface. At Re = 1245, as

shown in Figure 4.3d, the mixing performance is much improved and the tracer is

spread over the entirety of the vessel, although striations of black, where no tracer is

present, are still visible. Comparison of the visual images with the log-variance

analysis is most revealing; although the log-variance analysis can be used to

determine that the vessels are in an unmixed state, at Re = 1245, the degree of

information on the level of mixedness shown in the image is much higher than can be

represented by a single parameter such as log-variance. The results are also

remarkable in that a threefold increase in fluid viscosity causes the vessel to go from a

relatively well mixed state to almost unmixed at the same flow conditions.
 48

Figure 4.4 Mixing performance for three different fluid viscosities at a fill rate of Q = 3.33× 10-5 m3 s-
1
(a) log variance plotted against dimensionless fluid height (H/T); Image of the vessel filled with fluid

at (b) Re=687 (12 mPa s), (c) Re=1322 (6 mPa s), (d) Re=2490 (3 mPa s).

Figure 4.4 shows similar data for fluids with the same three decreasing viscosities at a

higher fill rate, Q of 3.33×10-5 m3s-1, giving values of Re of 687, 1322 and 2490

respectively. Again, at the lowest value of Re (687), the log-variance remains close to

zero (Figure 4.4a) and the tracer remains unmixed through the bulk of the vessel
 49

(Figure 4.4b). The mixing performance at the higher values of Re of 1322 and 2490

improves noticeably, with the log variance reaching ~ -1.5 (82% mixed) and -2.5

(94% mixed) respectively. Notably, for the experiment with the value of Re equal to

1322 (6 mPa s fluid, Figure 4.4c), the degree of mixedness is qualitatively similar to

the experiment obtained at Re = 1245 (3 mPa s fluid, Figure 4.3d); it would be

expected that Re would be an important part of any scaling criterion. At Re = 2490,

the tracer appears visually to be almost uniformly mixed through the vessel and very

few striations are present; this may be indicative of a change of mixing mechanism

from laminar to turbulent.

The change in mixing mechanism is even more apparent at the highest flow rate used,

Q = 5 ×10-5 m3s-1. The change in flow is still not sufficient to mix the highest

viscosity fluid at Re = 1030, (Figure 4.5b) but for Re = 1983 and Re = 3735 the

distribution of the tracer in the tank (Figure 4.5c,d respectively) is almost uniform.

The log-variance drops to -2.2 (92% mixed) and -3 (97% mixed) for the lower

viscosity fluids at Re = 1983 and Re =3735 respectively when the dimensionless fill

height (h/T) is of the order 0.5-1. However, upon further filling of the vessel, the

mixing performance actually worsens and the log-variance begins to rise again,

reaching – 1.8 at Re = 1983 (87% mixed) and -2.4 (94% mixed) at Re = 3735. The

worsening in mixing performance can be attributed to the quiescent nature of the flow

in the top half of the vessel and to the linear drop in specific energy input as the vessel

is filled.
 50

Figure 4.5 Mixing performance for three different fluid viscosities at a fill rate of Q = 5.00× 10-5 m3 s-
1
(a) log variance plotted against dimensionless fluid height (H/T); Image of the vessel filled with fluid

at (b) Re=1030 (12 mPa s), (c) Re=1983 (6 mPa s), (d) Re=3735 (3 mPa s).

For flows that appear to be dominated by laminar mixing, the toroidal vortices are

visible, Figure 4.6 and Figure 4.7 show the position of this vortex in relation to the

base of the tank and the inlet pipe at values of Re of 343, 661 and 1245 respectively

(Q = 1.67×10-5 m3s-1). It can be seen from Figure 4.6 that as the viscosity increases

the vortex centre moves closer to the base of the tank. For all three viscosities the
 51

height of the vortex at the beginning of the fill is less than 50% of the fluid height,

perhaps caused by interactions with the free surface of the fluid. As the fluid height

increases so does the vortex height until it reaches a maximum value which increases

with Re. At Re = 343, the vortex height that does not fluctuate significantly from its

maximum. However as the viscosity decreases there appears to be more instability in

the vortex position; this instability is also evident in Figure 4.7. At Re = 1245, the

fluctuations have a standard deviation of 8%, this may be the first signs of a more

turbulent flow regime.

Figure 4.6 The dimensionless height (h/T) of the centre of the vortex from the base of the tank,

measured for three Reynolds numbers at flow rate of Q = 1.67× 10-5 m3 s-1

Figure 4.7 shows that the average horizontal distance of the vortex from the pipe does

not vary significantly across the values of Re used in these experiments which

suggests that it is affected only by the geometry. This implies that the vortices are the

same size for the different viscosities but they have just moved further up the vessel.
 52

While this is true for the higher two viscosities which are still within the laminar flow

regime, the 3 mPa s fluid shows a large amount of distortion of the vortex which

could again be evidence of the beginning of turbulent fluctuations. It may prove

possible to exploit these vortices to further improve the mixing performance by

manipulating the geometry: clearly this will also alter the ratio of the jet to tank

Reynolds numbers and the distribution of local specific energy dissipation rate.

Figure 4.7 The dimensionless horizontal distance of the centre (h/T) of the vortex from the edge of the

inlet pipe, measured for three Reynolds numbers at flow rate of Q = 1.67× 10-5 m3 s-1

4.2.2 Flow regime in the vessel

The transition between laminar and turbulent flow is normally defined on the basis of

a critical value of Re determined experimentally for a given system. This value can

vary substantially, since the choice of length and velocity scales depends on the

geometry in question.. As shown in section 3.3.1, the values of Re based upon the dip
 53

pipe diameter and jet exit velocity are of O(103). An alternative would be to consider

a tank Reynolds number based upon the flow in the vessel, using the tank width and

the superficial velocity of the fluid (based on the cross-sectional area of the tank

bottom) as the length and velocity scales respectively. This would give Reynolds

numbers of O(10). Clearly, for the constant geometry used in these experiments, the

ratio of the jet and tank Reynolds numbers is also constant.

However, it is not sufficient to use Re as the sole parameter to characterize the mixing

performance. In the characterization of mixing in stirred vessels of constant volume,

a critical parameter is the energy input per unit mass, ε, which can be evaluated

locally and also averaged over the whole vessel. Indeed Kresta & Wood (1993) stated

that prediction of the local specific energy dissipation rate, ε T , is, in many cases, the

key to successful process modeling. In the filling case, the analysis is complicated

since the mass of liquid in the vessel increases linearly in time. At the start of the

filling process, when the mass in the tank is small, the energy input per unit mass is

high which causes intense mixing. As shown in Figure 4.1a and Figure 4.2a, the

initial charge of tracer is swept off the bottom of the vessel and the initial mixing

performance is commensurate with the high specific energy of the fluid. As the

vessel is filled further, the energy input per unit mass drops linearly with time and the

mixing performance noticeably worsens: Figure 4.1 and Figure 4.2 (b-d) demonstrate

that the tracer is swept upwards near the dip tube as it exits the toroidal vortex and

does not appreciably mix further with the fresh fluid entering the vessel. This effect is

more noticeable with the higher viscosity fluid, which is to be expected since the

value of Re is half that of the lower viscosity fluid. The extra mixing of the low

viscosity fluid may be either due to the creation of eddies via turbulence or due to
 54

molecular diffusion. Assuming the Stokes-Einstein relationship between viscosity

and molecular diffusivity holds (μ ~ DM-1); mixing due to molecular diffusion would

be twice as rapid in the low viscosity fluid. If one were to consider the remainder of

the mixing process being completed by this mechanism, assuming that

DM = O(10-8) m2s-1, then if the striations were 10% of the vessel diameter, the

diffusion time,

td =
λ2
=
(0.1T )2
DM DM

This gives tD ~ 3600 s for T = 0.06 m, or about one hour. This would be possible for

single phase bottled products during shipping. Over a fill time of ~ 20 s, assuming the

same value of DM, molecular diffusion would only occur over a length-scale of

~500 μm, an almost negligible amount.

4.3 Moving pipe

Figure 4.8 shows the log variance as a function of fluid height for three Reynolds

numbers using the moving pipe described in section 3.2.2; the mixing improves as the

Reynolds number increases as in the fixed pipe experiments. The lowest Re = 257

shows very little mixing throughout the fill with the log variance staying around zero,

this is comparable to the static pipe results at similar Reynolds numbers where the

mixing does not improve throughout the fill. For the highest Reynolds number

Re = 2867 the log variance decreases to ~ -2.6 indicating that it has reached 95%

mixed. Between these two is the Re = 996 case where some mixing occurs but not

sufficient to reach the 95% level, here the log variance reaches ~ -1.5. For both cases

where mixing occurs once the fill height has reached 5 - 6 cm, there is little further

improvement to the mixedness and in some cases the log variance increases indicating
 55

that the ‘fresh’ fluid entering the vessel is no longer mixing as effectively. For the

highest Reynolds number this may be due to the fact that there is always an area of

lower tracer concentration around the jet, so despite the homogeneity in the

surrounding liquid the variance cannot be decreased further.

Figure 4.8 Mixing performance for three different fluid viscosities at a fill rate of Q = 5.00×10-5 m3 s-
1
(a) log variance plotted against fluid height; Image of the vessel filled with the (b) 16 mPa s fluid, (c)

8 mPa s fluid, (d) 4 mPa s fluid

Figure 4.8b-d show a very clear change in the mixing mechanism, the Re = 257 fluid

(Figure 4.8b), the tracer is distributed around the vessel but there are two toroidal

vortices and obvious striations which indicate laminar flow conditions. The highest

Reynolds number fluid Re = 2867 (Figure 4.8d), is a more turbulent flow and as a

result it shows a more uniform distribution of dye, the striations have been removed

through eddy diffusion. For the Re = 996 (Figure 4.8c) case there are still striations
 56

visible but there is more uniformity than the Re = 257 case, this shows the transition

between the two different flow regimes. This agrees with the fixed pipe case

described in section 4.2.1 where the transition occurred around Re~1000.

A further indication of the onset of turbulence is the stability of the jet, the more

laminar jet (Figure 4.8b) shows the ‘fresh fluid’ travelling directly to the base of the

tank without undergoing spreading or jet break up and hence little mixing. The

transition jet shows similar travel to the base of the tank of the fresh fluid, however as

the jet becomes longer with the surface rising instabilities appear towards the base

(Figure 4.8c) indictating the jet is no longer purely laminar. The turbulent jet (Figure

4.8d) shows the jet mixing quickly with the surrounding fluid due to the jet spreading

and eddies forming.

4.3.1 Mixing performance and mixing mechanism.

Figure 4.9 Plot of the minimum log variance reached in each experiment versus the jet Reynolds

number.
 57

Figure 4.9 shows the minimum log variance reached by each experiment compared

with the Reynolds number of the jet in each case. The log variance in the tank

decreases rapidly as the Reynolds number increases for small values of Re. However

once the log variance decreases to below -2.5 this improvement in mixing becomes

more gradual, this occurs for Reynolds numbers above 1000. The mixing

mechanisms appear from the images to be a combination of laminar and turbulent,

with the turbulent mechanism becoming dominant above a value of Re of ~1000, the

evidence for this being the disappearance of the toroidal vortices and the appearance

of instabilities in the jet. This agrees with Kim et al. (2007) who observed the

stability of a water jet and found it was steady at Re = 404 but had become unsteady

when it reached Re = 1026.

In comparison with the stationary pipe results (Figure 4.9) using a moving pipe

appears to improve the mixing performance at the same liquid flow rate. From

observing the images the flow appears to be more turbulent with the moving pipe,

possibly because of the increasing length of the free jet which allowed instabilities to

develop. Also mixing will also have been improved by increasing entrainment into

the jet from the top half of the vessel.

It should be noted that the level of mixing in the tank has not reached the limit of the

resolution of the camera. For very highly mixed conditions the ability of this

technique to measure the fluctuations in concentration will be limited by the pixel size

in the captured image.

 58

4.3.2 Obstructed jet.

Figure 4.10 shows the flow pattern at different points in the fill for the Re = 1494

fluid, this clearly shows the striations that form in the tank and the improvement in the

mixing as the jet length increases. However a small number of nominally identical

experiments showed a different flow pattern, as shown in Figure 4.11a visual

comparison between Figure 4.10 and Figure 4.11 shows that the tank in Figure 4.11

appears to be better mixed, this is verified by comparing the log variance of the two

which is shown in Figure 4.12.

Figure 4.10 Evolving concentration fields at different fill heights for an 8 mPa s fluid, Re=1494 at (a)

h = 2.2 cm; (b) h = 4.7 cm; (c) h = 7.3 cm; (d) h = 10.3 cm at a flow rate Q = 3.33 × 10-5 m3 s-1

through a 6 mm dip pipe.

 59

Figure 4.11 Evolving concentration fields at different fill heights for an 8 mPa s fluid, Re=1494

where a bubble has formed at (a) h = 2.2 cm; (b) h = 4.7 cm; (c) h = 7.3 cm; (d) h = 10.3 cm at a flow

rate Q = 3.33 × 10-5 m3 s-1 through a 6 mm dip pipe.

Unlike Figure 4.10, in Figure 4.11a-b the dark jet is not clearly visible down the

centre of the tank although small areas of low concentration can be seen towards the

base, this suggests that the jet has been obstructed in some way and the fluid is now

entering the tank away from the centre plane. Figure 4.11c shows a bubble being

released into the flow, behind which the jet becomes clearly visible; in Figure 4.11d

the jet is now flowing through the centre of the tank. It appears that this bubble was

obstructing the jet, the dark eddies seen towards the base in Figure 4.11a-b suggest

that this obstruction has made the jet instable fortuitously creating a more turbulent

flow and better mixing. This mixing improvement could be created in the process by

putting a permanent obstruction in the end of the dip pipe, this is investigated in

Section 6.3. However, not surprisingly, the random presence of the bubble led to

unreliable reproduction of the result in Figure 4.11.

 60

Figure 4.12 The log variance plot of two fills of the 8 mPa s fluid at a flow rate Q = 3.33 × × 10-5 m3

s- where a bubble has formed under the pipe in one (blue) and the jet is unobstructed in the other (red)

4.4 Conclusions

The PLIF technique has been demonstrated to give good repeatability for the same

flow conditions and qualitative examination of the images has revealed a change in

the mixing mechanism from laminar to turbulent above a critical value of the

Reynolds number (Re~1000) which agrees with previous literature. For the cases

above Re~1000 the log variance reaches -2 which equates to 90% mixed. The use of

a retracting inlet pipe is shown to give a small increase in the mixing performance,

above Re~1000 the log variance reaches -2.2 in each case which equates to 92%

mixed. The retracting pipe also allows the stability of the jet to be examined giving

further evidence of the mixing regime in the tank. A bubble causing an obstruction in

the end of the pipe improved the mixing performance and could be replicated in the

nozzle design to provide this improvement in a more reliable way.

 61

Chapter 5 Velocity Fields

5.1 Introduction

To gain a better understanding of the mechanisms effecting the mixing in the tank the

velocity distribution was investigated using PIV. A comparison between the flow

patterns in the cases where the jet was able to mix the fluid and those where the

mixing was unsuccessful was made. In these cases the dead regions in the tank could

be identified and the areas where the shear rates are higher and more mixing occurred

highlighted. This chapter looks at the velocity profile and structure in the jet, a

comparison between jets at different Reynolds numbers is then made. The links

between the mixing in the tank and the flow patterns are made.

The PIV was set up as described in section 3.5, this allowed the flow field in the

vessel to be measured for the same values of Reynolds numbers as for the PLIF

results, these conditions are shown in Table 3.1

5.2 Velocity profiles

The general structure of the jet impinging on a pool of liquid is shown in Figure 5.1;

this jet structure in this region agrees with previous works as described by Reungoat

et al. (2007). The fluid enters the tank at the top of the vessel and travels vertically

downwards until it reaches the impingement zone, it is then forced along the base of

the tank towards the walls, at this point the fluid has lost the majority of its

momentum and the subsequent upwards movement is slow. This slow upward
 62

velocity is directed slightly towards the jet as it replaces the fluid that has been

entrained as well as fuelling the rising free surface.

Tank walls Jet inlet is centred just

above surface of fluid.
Fluid surface

Relatively
quiescent, but
with a slow
drift towards
the centre

Fluid flows along

Jet base of tank
impingement towards the walls
zone

Figure 5.1 An example of the PIV velocity profile in a vessel showing the features present in all cases.

Figure 5.2 Jet impingement zone, shaded to indicate velocity magnitude

Figure 5.2. shows the jet impingement zone at the base of the tank, the stagnation

region in the centre of the jet is clearly visible however the low accuracy of the results

next to the glass base means that the flow in the boundary layer cannot be analysed

further. Dead zones can be seen in both the bottom corners of the tank where the
 63

velocity is low caused by the abrupt change in direction of the jet along from the base

to moving up the wall.

Figure 5.3. The change in normalised velocity along the jet centreline for the four different points in

the fill

Further insight into the dynamics of the jet can be gained from Figs. 5.3 and 5.4.

Figs 5.3 shows the decrease in the centreline velocity of the jet as it moves from the

point of injection towards the point of impingement. The velocity has been

normalised by dividing though by peak velocity at the end of the pipe to allow

comparison. The deceleration of the jet is caused by the viscous drag of the

surrounding fluid, this viscous drag causes energy to be dissipated and also the

surrounding fluid to become entrained into the jet, the more the surrounding fluid is

entrained the more kinetic energy is transferred to the rest of the tank. The

deceleration of the jet is relatively high near the point of injection and decreases to a
 64

steady rate when the normalised velocity reaches ~ 0.8. The effect of the

impingement can be seen when the fluid reaches approximately 1 cm from the base

where the velocity suddenly begins to decrease more rapidly, this point is this same

regardless of how high the injection point is.

0.05

-0.05
Vertical velocity (m/s)

-0.1

-0.15

-0.2
h= 0
h= 0.46
-0.25
h= 1.02
h= 1.48
-0.3
h= 2.04
h= 2.50
-0.35

-0.4
3 2 1 0
Distance From Centre

Figure 5.4. Vertical velocity profile across horizontal slices taken at heights h above the base of the

tank

Figure 5.4. shows the velocity at points away from the centreline for the heights in

and near the impingement zone. At 2.5 cm the peak has well defined edges, moving

closer to the wall the peak spreads and there is a larger amount of upward velocity

towards the tank wall. For the data closest to the tank base this upward motion next to

the wall is not present, this is due to the dead zones in the corners of the tank
 65

Figure 5.5. The centreline velocity profile for low Reynolds number fills when the injection height has

reached 9 cm

Figure 5.5 shows the centreline velocity of the lower Reynolds number jets where the

flow is mostly laminar, Figure 5.6 shows the same plot for the higher Reynolds

numbers where flow appeared more turbulent, Re=959 appears in both graphs. For

the laminar flows there is a change in the gradient of the jet velocity is as the

Reynolds number is increased; the lower Reynolds numbers show the largest drop in

the velocity along the jet. As the fluid is more viscous the drag on the edges of the

fluid flow is larger and so the jet’s kinetic energy is lost. At high Reynolds numbers

the jet is turbulent, Figure 5.6 shows that once the transition to turbulence has

occurred the gradient of the velocity decrease is approximately constant for all values

of Reynolds number. The case with Re=996 does not fit into either catagory but is in

transition between the two, this agrees with previous observations from the PLIF data

of the onset of turbulence around this Reynolds number.

 66

Figure 5.6. The centreline velocity profile for high Reynolds number fills when the injection height has

reached 9 cm

The transition in the flow can be seen from the velocity magnitude plots (fig 5.7). For

low Re there is very little observable detail of the velocity change in the top half of

the vessel and the contours are fairly well defined; this flow showed the dye trapped

along the wall or up the side of the vessel in the PLIF data. For the transitional flow

there is some movement in the top of the vessel and the contours loose their

smoothness, this flow exhibited the definite striations in the PLIF data. For the higher

Re plots there is a larger amount of velocity fluctuation across the top of the tank, as

this becomes more pronounced the corresponding PLIF data shows the jet becoming

more unsteady and the mixing performance is improved. There is an error around the

jet injection points caused by the moving pipe.

 67

(a)

(b)

Figure 5.7. Velocity magnitude contour plots for flow at a) Re=517 and b) Re=1911

5.3 Shear rates

As a large amount of the mixing that occurs is laminar the areas of high shear in the

vessel are important, as it is the elongation and shear as well as folding of fluid that

causes it to mix. The magnitude of the shear rate in the following plots is calculated

using equation 3.6.

From Figure 5.8 it can be seen that the values of the shear are highest at the outsides

of the jet, where there is a sharp change in the velocity and along the base of the tank

where the jet moves radially away from the centre of the tank. These areas are where
 68

the striations seen in the laminar flow originate; this is then folded by the recirculation

in the tank.

Figure 5.8. The shear rate calculated from the velocity field measured at Re=517

5.4 Conclusions

To examine the flow regime in the tank further PIV was used; the resulting vector

fields allowed the distribution of momentum and shear in the tank to be investigated.

For all flows, dead regions were observed in the bottom corners of the tank and the

top half of the tank was relatively quiescent, however this was more prominent for

more laminar flows. For the more turbulent flows (Re > 1000) there was a higher

level of movement in the top half of the tank which corresponds to the increase in the

mixing performance shown in Chapter 4. To improve the mixing performance in the

laminar cases this movement must be created and accentuated by changing parameters

such as tank or nozzle geometry.

 69

Chapter 6 Flow improvements

6.1 Introduction

To improve the level of mixing in the tank for lower Reynolds number it is necessary

to alter the flow patterns so that there is more energy being discipated in the top half

of the tank. There are several ways in which this can be achieved, two different

methods have been investigated, varying the flow rate of the jet and changing the

design of the inlet nozzle. Section 6.2 examines the improvements due to ramping the

flow rate or of applying a sinusoidal variation to the flow rate. Section 6.3 describes

the increased level of mixing caused by changing the nozzle design.

6.2 Variable flow rate

As the tank is filled the energy from the jet dissipated per unit volume in the vessel

decreases. This means that the level of mixing in the tank is greater at the start of the

filling process than at the end, an effect that was demonstrated in Chapter 4. To

prevent this drop off in the mixing performance the flow rate of the jet could be

increased throughout the filling process. A case with this ramped flow was set up as

described in section 3.2.3.

An alternative way of improving the mixing is to increase the perturbations in the top

half of the jet, this has been achieved by varying the flow rate as a sinusoidal, this

case was designed to provide fluctuations in the flow field that would increase the

levels of turbulence.
 70

Figure 6.1 Mixing performance for three nozzles at a viscosity of 18 mPa s (a) log variance plotted

against fluid height (cm); Image of the vessel filled with (b) constant flow rate of Q = 3.33×10-5 m3 s-1

(c) sinusoidal flow rate varying between 1.67 x 10-5 m3 s-1 and 5 x 10-5 m3 s-1 (d) flow rate increasing

from 0 to 6.67 x 10-5 m3 s-1

Figure 6.1 shows the log variance of these two scenarios next to a case with constant

flow rate at the average flow rate for each case (3.33 x 10-5 m3 s-1), the case at

constant flow rate reaches a log variance of -0.6 which is 50% mixed . Both the sine

and the ramped flow result in a marked improvement in the level of mixing within the
 71

tank. For the ramped flow the low flow rate at the beginning of the fill means the

decrease in the log variance is more gradual throughout the process, the log variance

reaches approximately -1.4 which is 80% mixed. In contrast the sine function reaches

a log variance of -1.6 which 84% mixed. Figures 6.1c and 6.1d confirm this, the sine

case appears to be more homogeneous than the ramped case, however both are a large

improvement on figure 6.1b in which little mixing has taken place.

6.3 Nozzle improvements

The obstruction at the end of the pipe seen in section 4.3.2 resulted in a great

improvement in the level of mixing seen. To replicate this two nozzles were built

with a cone insert in the base as described in section 3.2.3. A third nozzle with a swirl

insert was also investigated; this nozzle was designed to give the jet angular

momentum so it would spread to greater regions of the tank.

Figure 6.2 shows the log variance achieved in the tank for the two cone nozzles in

comparison with the simple straight flow that reaches 50% mixed. The mixing

performance has dramatically improved. The second cone which is the smaller of the

two shows the best improvement in the mixing performance reaching a log variance

of -1.7 (86% mixed), however due to the level of noise in the log variance signal it is

difficult to differentiate between the two cases. Comparison of the images in Fig

6.2.c and Fig 6.2.d does indicate that the smaller nozzle has produced a more

homogeneous vessel.
 72

Figure 6.2 Mixing performance for three nozzles at a fill rate of Q = 3.33×10-5 m3 s-1 and a viscosity of

18 mPa s (a) log variance plotted against fluid height (cm); Image of the vessel filled with (b) straight

pipe (c) cone 1 nozzle (d) cone 2 nozzle

Figure 6.3 shows the improvement achieved using a swirl nozzle, this nozzle provides

a much greater improvement than the cone nozzles with the log variance reaching -2.6

(95% mixed) at points in the fill. Figure 6.3c shows that the fluid in the tank is much

more homogeneous than in Figures 6.2c and 6.2d. It appears that the angular

momentum applied to the fluid by the swirl insert in the end of the nozzle has
 73

increased the level of mixing particularly in the top half of the tank, by directing the

flow towards these stagnant regions.

Figure 6.3 Mixing performance for two nozzles at a fill rate of Q = 3.33×10-5 m3 s-1and a viscosity of

18 mPa s (a) log variance plotted against fluid height (cm); Image of the vessel filled with (b) straight

pipe (c) swirl nozzle

 74

6.4 Conclusions

Improvements to the mixing performance have been made by both changing the inlet

pipe geometry and varying the inlet flow rate. Two flow rate variations were

investigated, ramped flow and a sinusoidal flow, the sinusoidal function variations in

the inlet velocity increased the level of mixing from 50% mixed so that the vessel

reached 84% mixed. Changing the nozzle design had a greater effect on the level of

mixing, three nozzle designs were examined, the most successful of these had a swirl

insert in the bottom of the pipe. With the swirl insert the level of mixing in the tank

reached 95% (log variance of -2.6) at one point in the fill, however this was not

maintained throughout. To further improve the mixing it may be beneficial to

combine these improvements, such as by using a sinusoidal flow rate through a swirl

nozzle. This may produce a more sustained and reliable improvement and should

feature in any future work.

 75

Chapter 7 Conclusions

7.1 Mixing in a filling vessel

There is a large body of work into mixing by a range of mechanisms, however very

little has been carried out on the mixing that occurs as a vessel is filled. This mixing

process has been investigated using the techniques PLIF and PIV. These techniques

have been used to measure a variety of flow scenarios in the literature, however in

most cases the system is a steady state. This work has adapted these techniques to

measure the mixing of transient systems with a moving free surface.

The PLIF technique has been demonstrated to give good repeatability for the same

flow conditions and qualitative examination of the images has revealed a change in

the mixing mechanism from laminar to turbulent above a critical value of the

Reynolds number (Re~1000) which agrees with previous literature. For the cases

with a retracting pipe above Re~1000 the log variance reaches -2.2 in each case which

equates to 92% mixed.

PIV was used to measure the vector fields in the tank, allowing the distribution of

momentum and shear in the tank to be investigated. For all flows dead regions were

observed in the bottom corners of the tank and the top half of the tank was relatively

quiescent, however this was more prominent for more laminar flows. For the more

turbulent flows (Re > 1000) there was a higher level of movement in the top half of

the tank. To improve the mixing performance in the laminar cases this movement

must be created and accentuated, changes to the flow rate and the nozzle design were

made to achieve this.

 76

Sinusoidal function variations in the inlet velocity increased the level of mixing from

an unmixed highly striated case so that the vessel reached 84% mixed, however

changing the nozzle design had a greater effect on the level of mixing. Three nozzle

designs were examined, the most successful of these had a swirl insert in the bottom

of the pipe. With the swirl insert the level of mixing in the tank reached 95% (log

variance of -2.6) at one point in the fill, however this was not maintained throughout.

To further improve the mixing it may be beneficial to combine these improvements,

such as by using a sinusoidal flow rate through a swirl nozzle. This may produce a

more sustained and reliable improvement and should feature in any future work.

7.2 Future work

PLIF has proven to be a useful tool in measuring the mixing within a filling tank, the

development of the mixing analysis allows this technique to be used for other systems

with moving free surfaces.

There are a number of avenues that can be explored to improve the mixing in the tank

further; combining the variable flow rate with the nozzle design changes, changing

the tank geometry, adding internal baffles to the tank, adding the variant and the base

product simultaneously, or mixing in the nozzle. There are also a number of

challenges to overcome in translating the understanding of the Newtonian mixing and

flow patterns into rules that will work with real products.

One option for the screening of different improvements to the flow is the use of

Computational Fluid Dynamics, however its application requires a number of

 77

challenges to be overcome. The first is the difficulties in solving a flow with such

high viscous stresses and the accuracy of the results obtained, as the pressure

correction solver in the commercially available code often assumes that the viscous

stresses are much smaller than the inertial stresses. Furthermore while the mixing in

the tank can be modelled using the commercially available solvers the accuracy is

greatly dependent on the mesh. A very fine mesh is required as the lengthscale of the

mixing can only be determined down to the size of the mesh being used. Also the

mesh quality has to be high with predominantly hexahedral elements, and the time

step must be carefully chosen to reduce the numerical diffusion. These factors can

lead to long run times which reduces the effectiveness of CFD as a screening tool.

It would be a useful measure of the mixing improvement, gained through the nozzle

and flow rate changes made in this thesis, to test the techniques on real products.

Limited understanding of the flow structure would be gained due to the opaque nature

of the product but the level of mixing in the bottle can be measured using a dye and

measuring the colour of the resulting liquid. This would allow a correlation between

the Newtonian PLIF results and the real product to be made.

 78

References

Bi, W., Sugii, Y., Okamoto, K., Madarame, H. (2003) "Time-resolved proper

orthogonal decomposition of near-field flow of a round jet measured by

dynamic particle image velocimetry" Measurement Science and Technology

14 L1-L5

Bin, A. K. (1988). "Minimum air entrainment velocity of vertical plunging liquid

jets." Chemical Engineering Science, 43(2), 379-389.

Brown, D. A. R., Jones, P. N., and Middleton, J. C.(2004). "Experimental

methods." In: Handbook of industrial mixing: science and practice, Paul E.L.,

Atiemo-Obeng V.A., and Kresta S.M., eds., John Wiley & Sons Inc., New

Jersey, 145-256.

Chen, Y. C., Ma, C. F., Qin, M., and Li, Y. (2005). "Theoretical study on

impingement heat transfer with single-phase free-surface slot jets."

International Journal of Heat and Mass Transfer, 48(16), 3381-3386.

Chung, K. H. K., Barigou, M., and Simmons, M. J. H. (2006). "Reconstruction of

3-D flow field inside miniature stirred vessels using a 2-D PIV technique."

12th European Conference on mixing.

Cozewith, C., and Busko, M. (1989). "Design Correlations for Mixing Tees."

Industrial & Engineering Chemistry Research, 28(10), 1521-1530.

Dazin, A., Dupont, P., and Stanislas, M. (2006). "Experimental characterization of

the instability of the vortex ring. Part I: Linear phase." Experiments in Fluids,

40(3), 383-399.
 79

Degardin, O., Renou, B., and Boukhalfa, A. M. (2006). "Simultaneous measurement

of temperature and fuel mole fraction using acetone planar induced

fluorescence and Rayleigh scattering in stratified flames." Experiments in

Fluids, 40(3), 452-463.

Ding, R., Revstedt, J., Fuchs, L. (2003) "LIF Study of Mixing in Circular Impinging

Jets; Effects of Boundary Conditions" Proceedings of PSFVIP-4, Chamonix,

France F4015

Etchells, A.W., Meyer, C.F. (2004). " Mixing in pipelines." In: Handbook of

industrial mixing: science and practice, Paul E.L., Atiemo-Obeng V.A., and

Kresta S.M., eds., John Wiley & Sons Inc., New Jersey, Ch 7, 391 - 478.

Fall, A., Lecoq, O., and David, R. (2001). "Characterization of mixing in a stirred

tank by planar laser induced fluorescence (PLIF)." Chemical Engineering

Research & Design, 79(A8), 876-882.

Feng, H., Olsen, M. G., Liu, Y., Fox, R. O., and Hill, J. C. (2005). "Investigation

of turbulent mixing in a confined planar-jet reactor." Aiche Journal, 51(10),

2649-2664.

Fondse, H., Leijdens, H., and Ooms, G. (1983). "On the Influence of the Exit

Conditions on the Entrainment Rate in the Development Region of A Free,

Round, Turbulent Jet." Applied Scientific Research, 40(4), 355-375.

Fosset, H., Prosser, L. E., (1949). “The application of free jets to mixing of fluids in

bulk.” Proceedings of the Institute of Mechanical Engineers. 160, 224-232

 80

Forney, L. J., and Lee, H. C. (1982). "Optimum Dimensions for Pipeline Mixing at

A T-Junction." Aiche Journal, 28(6), 980-987.

Golnabi, H. (2006). "Precise CCD image analysis for planar laser-induced

fluorescence experiments." Optics and Laser Technology, 38(3), 152-161.

Grenville, R.K., Nienow, A.W. (2004). " Blending of miscible liquids." In:

Handbook of industrial mixing: science and practice, Paul E.L., Atiemo-

Obeng V.A., and Kresta S.M., eds., John Wiley & Sons Inc., New Jersey, Ch

9, 507 - 542.

Guillard, F., Fritzon, R., Revstedt, J., Tragardh, C., Alden, M., and Fuchs, L. (1998).

"Mixing in a confined turbulent impinging jet using planar laser-induced

fluorescence." Experiments in Fluids, 25(2), 143-150.

Guillard, F., Tragardh, C., and Fuchs, L. (2000). "A study on the instability of

coherent mixing structures in a continuously stirred tank." Chemical

Engineering Science, 55(23), 5657-5670.

Hall, J. F., Barigou, M., Simmons, M. J. H., and Stitt, E. H. (2004). "Mixing in

unbaffled high-throughput experimentation reactors." Industrial &

Engineering Chemistry Research, 43(15), 4149-4158.

Hall, J. F., Barigou, M., Simmons, M. J. H., and Stitt, E. H. (2005). "Comparative

study of different mixing strategies in small high throughput experimentation

reactors." Chemical Engineering Science, 60(8-9), 2355-2368.

 81

Houcine, I., Vivier, H., Plasari, E., David, R., and Villermaux, J. (1996). "Planar

laser induced fluorescence technique for measurements of concentration fields

in continuous stirred tank reactors." Experiments in Fluids, 22(2), 95-102.

Hult, J., Meier, U., Meier, W., Harvey, A., and Kaminski, C. F. (2005).

"Experimental analysis of local flame extinction in a turbulent jet diffusion

flame by high repetition 2-D laser techniques and multi-scalar measurements."

Proceedings of the Combustion Institute, 30, 701-709.

Kim, K. C., Min, Y. U., Oh, S. J., An, N. H., Seoudi, B., Chun, H. H., and Lee, I.

(2007). "Time-resolved PIV investigation on the unsteadiness of a low

Reynolds number confined impinging jet." Journal of Visualization, 10, 367-

379.

Kresta, S. M., Brodkey, R.S. (2004). "Turbulence in Mixing Applications." In:

Handbook of industrial mixing: science and practice, Paul E.L., Atiemo-

Obeng V.A., and Kresta S.M., eds., John Wiley & Sons Inc., New Jersey, Ch

2, 19 – 87.

Kresta, S. M., Wood, P.E. (1993). “The flow field produced by a pitched blade

turbine:characterisation of the turbulence and estimation of the dissipation

rate.” Chemical Engineering Science, 48, 1761-1774

Kukukova, A., Aubin, J., and Kresta, S. M. (2009). "A new definition of mixing and

segregation: Three dimensions of a key process variable." Chemical

Engineering Research & Design, 87(4A), 633-647.

 82

Lane, A. G. C., and Rice, P. (1982). "Comparative-Assessment of the Performance

of the 3 Designs for Liquid Jet Mixing." Industrial & Engineering Chemistry

Process Design and Development, 21(4), 650-653.

Law, A. W. K., and Wang, H. W. (2000). "Measurement of mixing processes with

combined digital particle image velocimetry and planar laser induced

fluorescence." Experimental Thermal and Fluid Science, 22(3-4), 213-229.

Mortensen, M., Orciuch, W., Bouaifi, M., and Andersson, B. (2004). "Mixing of a

jet in a pipe." Chemical Engineering Research & Design, 82(A3), 357-363.

Neal, G., Simmons, M. J. H., Hough, J. A., and Fryer, P. J. (2008). "Criteria for

measuring mixing performance in a vessel with a rising free surface."

Experiments in Fluids, 45: 3-12.

Neal G.., Simmons M.J.H., Hough J.A., Fryer P.J. (2008) “Mixing in a vessel with a

rising free surface” Proceedings of the International symposium on mixing in

industrial processes (ISMIP 2008),

Neal G., Simmons M.J.H., Hough J.A., Fryer P.J. (2009) “Mixing in a vessel with a

rising free surface” Proceedings of the World Congress in Chemical

Engineering 2009.

Ottino, J. M., (1989). The kinematics of mixing: stretching, chaos, and transport

Cambridge University Press, Cambridge.

Pan, G., and Meng, H. (2001). "Experimental study of turbulent mixing in a tee

mixer using PIV and PLIF." Aiche Journal, 47(12), 2653-2665.

 83

Patwardhan, A. W., and Gaikwad, S. G. (2003). "Mixing in Tanks Agitated By

Jets." Trans IChemE, 81(A), 211-220.

Paul, E.L., Atiemo-Obeng, V.A., and Kresta S.M. (2004). "Introduction" In:

Handbook of industrial mixing: science and practice, Paul E.L., Atiemo-

Obeng V.A., and Kresta S.M., eds., John Wiley & Sons Inc., New Jersey.

Pawlak, G., Cruz, C. M., Bazan, C. M., and Hrdy, P. G. (2007). "Experimental

characterization of starting jet dynamics." Fluid Dynamics Research, 39, 711-

730.

Raffel, M., Willert, C. E., Wereley, S. T., and Kompenhans, J. (2007). Particle

Image Velocimetry, A Practical Guide, Second Edition Ed., Springer.

Reungoat, D., Riviere, N., and Faure, J. P. (2007). "3C PIV and PLIF measurement

in turbulent mixing - Round jet impingement." Journal of Visualization, 10(1),

99-110.

Revill, B. K. (1985). "Jet Mixing." In: Mixing in the process industries, Harnby, N.,

Edwards, M.F., Nienow, A.W., eds., Butterworths, 145 -169.

Saarenrinne, P., Piirto, M., Eloranta, H., (2001) “Experiences of turbulence

measurement with PIV,” Measurement Science Technology 12, 1904

Szalai, E. S., Alvarez, M. M., Muzzio, F. J. (2004). " Laminar mixing: A

dynamical approach." In: Handbook of industrial mixing: science and

practice, Paul E.L., Atiemo-Obeng V.A., and Kresta S.M., eds., John Wiley &

Sons Inc., New Jersey, Ch 3, 89 - 144.

 84

Unger, D. R., and Muzzio, F. J. (1999). "Laser-induced fluorescence technique for

the quantification of mixing in impinging jets." Aiche Journal, 45(12), 2477-

2486.

Unger, D. R., Muzzio, F. J., and Brodkey, R. S. (1998). "Experimental and

numerical characterization of viscous flow and mixing in an impinging jet

contactor." Canadian Journal of Chemical Engineering, 76(3), 546-555.

Weigand, P., Meier, W., Duan, X. R., Stricker, W., and Aigner, M. (2006).

"Investigations of swirl flames in a gas turbine model combustor - I. Flow

field, structures, temperature, and species distributions." Combustion and

Flame, 144(1-2), 205-224.

Weisgraber, T. H., and Liepmann, D. (1998). "Turbulent structure during transition

to self-similarity in a round jet." Experiments in Fluids, 24(3), 210-224.

Zughbi, H. D., and Ahmad, Q. (2005). "Mixing in liquid-jet-agitated tanks: Effects

of jet asymmetry." Industrial & Engineering Chemistry Research, 44(4),

1052-1066.

Zughbi, H. D., and Rakib, M. A. (2004). "Mixing in a fluid jet agitated tank: effects

of jet angle and elevation and number of jets." Chemical Engineering Science,

59(4), 829-842.
 85

Appendix: Matlab analysis code

PLIF Analysis Script

T=6; %the width of the tank in cm

R=8; %the mg of Rhodamine in 1 ml put in bottom of vessel initially
Irange=(31:199); % image range
vbase=1003; %vessel base cut off
vleft=200; % vessel left cut off
vright=647; %vessel right cut off
inletleft=189;%the position of the inlet pipe in relation to the tank
inletright=279;
heightlogsigma=zeros(length(Irange),3);
size=zeros(length(Irange),2);

[M,K]=fcalibration(R,T);

for count=1:length(Irange)
cd 'E:\Experiments\0606PLIF8\4run\Image\Single' %the directory with the images in it
I=zeros(1016,1000);
I=imread (sprintf('4run0%04.4d.tif',Irange(count))); %reads the images
cd 'E:\mfiles' %changes to the directory with the mfiles in it
[A,Mi,Ki]=ffluidarea(I,vbase,vleft,vright,M,K);
clear I
[Cinf,H]=finfconc(A,R,Mi,Ki,T);
[C1,C2]=fnormconc(A,Ki,Cinf,inletleft,inletright);
clear A Cinf
[V,E]=flogsigmaRMS(C1,C2);
heightlogsigma(count,1)=H;
heightlogsigma(count,2)=V(1,1);
heightlogsigma(count,3)=V(1,2);
size(count,1)=E(1,1);
size(count,2)=E(1,2);
clear V Mi Ki C
count=count+1;
end

clear count M K T R Irange vbase vright vleft inletleft inletright

plot(heightlogsigma(:,1),heightlogsigma(:,2),'r',heightlogsigma(:,1),heightlogsigma(:,3),'b')
 86

Associated functions

function [M,K,I]=fcalibration(R,T)
%a function to give the gradient and intercept of graph relating the
%concentration to the greyscale. The output is two mattrices with these
%values for each pixel in the field.

cd 'E:\Experiments\0606PLIF8\cal\Image\Single'

irange1=(0:49);%input('enter cal image numbers for 1 dose:\n');

irange2=(50:99);%(input('enter cal image numbers for 2 dose:\n');
height=11.4;%input('enter height of liquid in calibration images:\n');

I=imread (sprintf('cal0%04.4d.tif',irange1(1)));
for count=2:length(irange1);
I1=imread (sprintf('cal0%04.4d.tif',irange1(count)));
I=cat(3,I,I1);
clear I1
count=count+1;
end

I1=mean(I,3);
clear I

I=imread (sprintf('cal0%04.4d.tif',irange2(1)));
for count=2:length(irange1);
I2=imread (sprintf('cal0%04.4d.tif',irange2(count)));
I=cat(3,I,I2);
clear I2
count=count+1;
end

I2=mean(I,3);
%clear I

K=2*I1-I2;

Conc=R/(T*T*height);
M=(I2-I1)/Conc;

function [A,Mi,Ki]=fluidarea(I,vbase,vleft,vright,M,K)
%program for focusing on the area that contains fluid in the
%vessel described in image I. This function takes the image determines the
%area with a greyscale 10% higher than the background greyscale and outputs
%a mattrix with greyscale values for the fluid area. It then scales the
%concentration gradient and intercept so that they are mattrices of the
%correct size.

B=mean(I(:,2));
cutoff=mean(B)+0.07*mean(B); %determines greyscale cut off value
%cutoff=77;

G=I(1:vbase,vleft:vright); % cuts off side and base of fluid

 87

F=mean(G');
E=find(F>cutoff);%creates vector containing the rows above the cut off greyscale

A=G(min(E)+30:vbase,:);
Mi=M(min(E)+30:vbase,vleft:vright);%cuts the gradient matrix to size
Ki=K(min(E)+30:vbase,vleft:vright);%cuts the intercept matrix to size
end

function [Cinf,H]=infconc(A,Cinit,M,K,T)
%function to deterime what the final greyscale would be if the vessel was
%well mixed at that fill height. A is the fluid area as defined by the
%fluid area function, Cinit is the concentration of the 1ml of tracer
%intially placed in the bottom of the tank in mg. M is the gradient and K the
%intercept of the linear relationship between concentration(x)and greyscale(y),
%T is the tank width in centimetres.

[B,D]=size(A); % tells us the no of rows (B) and the number of columns (D)

H=(B+30)*(T/D); % gives us the height of the liquid in centimetres

V=T*T*H; %the volume of fluid in the tank in ml

C=Cinit/V; %the concentration in the tank

Cinf=C*M+K; %calculates the infinite concentration greyscale value at that height

end

function [C1,C2]=normconc(A,Czero,Cinf,inletleft,inletright);
%outputs a normalised concentration - Cinf is the greyscale that the
%vessel would be at if completely mixed at the fill height of that image,
%Czero is the greyscale value when the concentration is zero, A is the area
%of fluid that has been previously determined by the fluid area function

B=double(A);

Dt=(B-Czero);
Du=(Cinf-Czero);

Ct1=Dt(:,1:inletleft);
Ct2=Dt(:,inletright:end);
Cu1=Du(:,1:inletleft);
Cu2=Du(:,inletright:end);

Ct1=Ct1(:);
Cu1=Cu1(:);

y=find(Cu1<2);
Ct1(y)=[];
Cu1(y)=[];
C1=Ct1./Cu1;

Ct2=Ct2(:);
Cu2=Cu2(:);
 88

y=find(Cu2<2);
Ct2(y)=[];
Cu2(y)=[];
C2=Ct2./Cu2;

end

function [V,E]=logsigmaRMS(C1,C2)
%function to determine the log of the RMS variance, C is the normalised
%concentration mattrix

A1=C1-1;
A2=C2-1;

B1=A1.^2;
B2=A2.^2;

D1=sum(B1);
D2=sum(B1);
E1=length(B1);
E2=length(B2);

D=[D1,D2];
E=[E1,E2];

V=log(D./E);

end
 89

PIV Analysis Script

cd ('F:\PIV\graceeee\3ml02car7J\Analysis');

range1=00:299;

A=csvread
(sprintf('3ml02car7J00%04.4d.T000.D000.P001.H001.L.vec',range1(1)),1,0);%front vector
file behind which the others are stacked
l=2;
for l=2:length(range1)
B=csvread (sprintf('3ml02car7J00%04.4d.T000.D000.P001.H001.L.vec',range1(l)),1,0);
A=cat(3,A,B);
l=l+1;
end
clear B

cd ('F:\mfiles');

%remove false values

[D,L]=find(A(:,5,:)<1);%finds all where 'false' is indicated by the software
for l=1:length(D);
A(D(l),3:5,L(l))=0;%sets the velocities and the 5th column to zero for the false vectors
l=l+1;
end
%remove false x velocity values
j=1;
[D,L]=find(abs(A(:,3,:))>20);%finds all where x velocity is greater that 2m/s-unrealistic in
this flow context
for j=1:length(D);
A(D(j),3:5,L(j))=0;%sets the velocities and the 5th column to zero for the false vectors
j=l+1;
end
%remove false y velocity values
j=1;
[D,L]=find(abs(A(:,4,:))>20);%finds all where y velocity is greater that 2m/s-unrealistic in
this flow context
for j=1:length(D);
A(D(j),3:5,L(j))=0;%sets the velocities and the 5th column to zero for the false vectors
j=l+1;
end

%time average velocity

ubar=zeros(size(A,1),5);
B=sum(A,3);%sums all the velocities
ubar(:,1:2)=A(:,1:2,1);%sets the position vectors so they are the same as the orginal
ubar(:,3:4)=B(:,3:4)./(B(:,5)*ones(1,2));%gives average velocity by dividing sum by number
of valid vectors
ubar(:,5)=B(:,5);
clear B

%root mean square

l=1;
B=A;
 90

for l=1:(size(A,3))
B(:,3:4,l)=(A(:,3:4,l)-ubar(:,3:4)).^2;
l=l+1;
end

for l=1:length(D);
B(D(l),3:5,L(l))=0;%resets the velocities and the 5th column to zero for the false vectors
l=l+1;
end
C=sum(B,3);
urms2=C(:,3:4)./(C(:,5)*ones(1,2));%where this is the rms value squared

clear B C l
clear D L

%TKE
k=ones(length(A),3);
k(:,1:2)=A(:,1:2,1);
k(:,3)=0.75*(urms2(:,1).*urms2(:,2));

%reorganise for plotting!

d=find(k(:,2)<k(1,2));
e=length(k)/(min(d)-1);
K=reshape(k(:,3),(min(d)-1),e);
K=K';

c=min(ubar(:,1));
b=max(ubar(:,1));
f=min(ubar(:,2));
g=max(ubar(:,2));
lim=[c-0.5 b+0.5 f-0.5 g+0.5];

clear k d e c b f g

%code to create average ubar, velocity magnetude field, and contour plot,
ubarg=ubar9J;
ubarj=ubar9g;
kj=K9g;
K=K9J;

limit=1.4;%speed in m/s of the cutoff point between the two versions of ubar and K;

U=(ubarg(:,3).^2+ubarg(:,4).^2).^0.5;%0.5; give column with average speed

B=find(U>limit);

ubar=ubarg;
for l=1:length(B);
ubar(B(l),:)=ubarj(B(l),:);
l=l+1;
end

ubar=ubar./10;
U=(ubar(:,3).^2+ubar(:,4).^2).^0.5;%0.5; give column with average speed with jet values for
larger numbers
 91

U=reshape(U,size(K'));

B=find(U>limit);
K=K';
for l=1:length(B);
K(B(l))=kj(B(l));
l=l+1;
end

K=K'./100;
U=U';
K=flipud(K);
U=flipud(U);

contour(U)