Project FOOS

Report 1

to the

MEMBER COMPANIES OF THE INSTITUTE OF PAPER SCIENCE AND TECHNOLOGY

January 2000
 INSTITUTE OF PAPER SCIENCE AND TECHNOLOGY -

Atlanta, Georgia

FORMING HYDRODYNAMICS

Project FO05

Report 1

A Progress Report

to the

MEMBER COMPANIES OF THE INSTITUTE OF PAPER SCIENCE AND TECHNOLOGY

C.K. Aidun, V.K.S. Bandhakavi, and J.S. Hooda

January 2000
An average person in the United States consumes approximately 675
pounds of paper per year as against a world average of about 100
pounds.
Contents

List of Tables iv

1 Paper - an Indispensable Commodity 1

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Need for Rapid Advances in Pulp and Paper Industry . . . . . . . . . . . . 1

2 Key Issues in Paper Forming Process 4

2.1 Headbox Requirements .............................. 4

2.2 Product Uniformity ............................... 4

2.3 Large Scale vs. Small Scale Nonuniformities .................. 5

3 Hydrodynamics of Streaks on the Forming Table 7

3.1 Moisture Streaks and their Causes . . . . . . . . . . . . . . . . . . . . . . . 7

3.2 Variations of Streaks on the Forming Wire . . . . . . . . . . . . . . . . . . 7

3.3 Amplification of Free Surface Waves at the Impingement Site . . . . . . . . 10

3.4 Origin and Nonuniformities of Streaks . . . . . . . . . . . . . . . . . . . . . 12

3.5 Dynamics of the Streamwise Streaks in the Forming Jet . . . .3” . . . . . . . 14

3.6 The Impact of the Streak Dynamics on Small Scale Physical Properties . . 16

4 Streak/Wave Analysis on the Forming Table 18

4.1 Schematic of a Drainage Model ......................... 18

4.2 Fourier Analysis of Wet Streaks ......................... 19

4.3 Results. ...................................... 21

 5 Effect of Secondary Flow on Fiber Orientation 24

5.1 Two Types of Secondary Flow. ......................... 24

5.2 Governing Fluid Flow Equations ........................ 26

5.3 Numerical Method ................................ 28

. 5.4 Headbox Secondary Flows Due to Sidewalls .................. 30

5.5 Uniform Inflow .................................. 32

5.6 The Influence of Nonuniform Inflow on the Cross-flow at the Slice ...... 36

5.7 Secondary Flows Due to Turbulence Anisotropy ................ 38

5.7.1 Origin of secondary flows of the second kind .............. 38

5.7.2 Numerical modeling with the non-linear K - c method ........ 41

5.8 Summary ..................................... 43

6 High-Speed Digital Imaging of Paper Forming 45

6.1 Introduction. ................................... 45

6.2 The Imaging Hardware Systems ......................... 45

6.3 Preparation of the Raw Images ......................... 46

6.4 Averaging and Time Series Image Processing . . . . . . . . . . . . . . . . . 48

6.5 Spectral Analysis ................................. 50

6.6 Evaluation of Time Series Images Using FFT Methods ............ 50

6.7 Evaluation of Averaged Images Using FFT Methods ............. 51

6.8 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

7 Quantitative Evaluation of the Forming Jet Delivered from Four Different

Headboxes 53

7.1 Operational Parameters of Four Hydraulic Headboxes . . . . . . . . . . . . 53

ii
 7.2 Results and Analysis ........................ . . . . . . . 54

8 Investigations of Experimental Headboxes 67

8.1 Effect of Contraction Ratio on Fiber Orientation ........ . . . . . . . 67

8.2 Effects of Guide Sheets on Headbox Hydrodynamics ...... . . . . . . . 67

8.3 Measurements at the Jet Exit . . . . . . . . ........... . . . . . . . 68

8.4 Measurements in the Headbox . . . . . . . . ........... . . . . . . . 73

8.5 Summary . . . . . . ........... . . . . . . . 75

9 Conclusions 77

APPENDIX A 79

Laser Doppler Velocimetry . . . . . . . 79

APPENDIX B 80

Finding Jet’s Velocity Profile using Cross-Correlation of Digital Images . . . . 80

REFERENCES 85

...
111
List of Tables

1 Headbox parameters and flow conditions . . . . . . . . . . . . . . . . 30

2 Conversion factor for nondimensional to dimensional units . . . . . 31

3 List of parameters for the four paper machines . . . . . . . . . . . . 53

4 Averaged Images - Range of Table Covered .............. 56

5 Time Series Images - Positions of Samples ............... 56

iV
 1 Paper - an Indispensable Commodity

1.1 Introduction

Very few products play as dominant a role as paper does in the lives of human beings. Var-
ious paper products that are used in our daily activities include newsprint, coated ground-
- wood (for advertising flyers), business and typing papers for recording, storing, and dissem-
inating information, cardboard boxes for packaging, wrapping paper, towels, and tissues,
just to name a few. It is an undeniable fact that the per capita consumption of paper of
any country can be used as an index of development like other indices such as consumption
of electricity and gas.

Since 105 A.D. when the Chinese invented paper, the paper industry has passed through
various phases of development. The 20th century has seen the rapid refinement and mod-
ification of the earlier crude technologies, along with the development of such multistage
techniques as cooking and bleaching of pulp, papermaking with synthetic fibers, paper dry-
ing, paper coating, and computerized process controls, etc. Today the pulp and paper
industry provides employment for millions of people all around the world, and especially in
the United States and Canada, where it plays a vital role in the overall economy. Also, the
rapid increase in automation of various processes in this industry require large investments,
which in some cases can reach 1 million dollars per worker and, hence, qualifies the pulp
and paper industry as a “capital-intensive industry” [l].

1.2 Need for Rapid Advances in Pulp and Paper Industry

The factors that were responsible for North American countries leading the world in the
pulp and paper industry are abundant timber, energy, water resources, modern technologies,
skilled labor supply, and accessibility to markets. Fortunately, this resulted in a situation
where there is no major competition from foreign sources in the domestic market. Contin-
uing globalization of the pulp and paper industry is resulting in the growth of this industry
elsewhere in the world. Areas such as Africa, Brazil, and Asia may become large producers
of pulp in the coming years. Also, as a result of economic growth in the developing coun-

1
 tries, the utilization of paper is steadily increasing. Some of the requirements placed on the
American paper mills by this increased demand and global competition are faster machine
speeds, greater product uniformity, and increased production rates. These factors continue
to drive the development of new technologies in various spheres of the paper industry such
as wet-end operations, pressing/drying technologies, multiple coating trends, to name a
- few [2].

One of the most significant areas of the current research in the papermaking process is
the improvement of wet-end processes, which include the headbox and forming zone. The
headbox is the first piece of equipment on a paper machine and a proper design of the
same is of vital significance to ensure the quality of the end product. The purpose of
headbox is to discharge a rectangular, free surface jet of water with uniformly dispersed
wood fibers and fillers. This jet is directed onto the forming table where a fiber mat is
formed. Providing a proper turbulence level in the jet, ensuring a uniform distribution
of the stock across the machine, with no net cross flow, enhancing fiber dispersion, and
suppressing the fiber flocculation at the jet exit are some of the fundamental functions of
the headbox. A thorough understanding of the complex hydrodynamics of the headbox,
wet-end dynamics such as turbulence, flocculation, stock jet velocity, and impingement
angle goes a long way in bridging the gap between the expected quality of the finished
paper and the limitations of the existing technologies.

During the last decade, the Fluid Dynamics Group at IPST has investigated these problems
from various perspectives. This article presents the efforts and the progress of the more
recent work in understanding the headbox hydrodynamics. A number of investigations (both
experimental and computational) carried out during the last few years are incorporated in
this review report.

In this report we start with the key issues in the paper forming process, with an emphasis on
headbox with large-scale and small-scale variations (§2), followed by a detailed analysis of
streaks on the forming wire ($3). In 54 we present a Fourier analysis of the surface waves on a
forming table and a correlation between the dominating Fourier modes and the length-scale
of wet streaks. Effects of two types of secondary flow are discussed in 55 using a numerical

2
 model of the headbox. In 56 we describe the technique of High-Speed Digital Imaging of
Paper Forming, which combines high- speed digital imaging, computational image analysis
and spectral analysis. This technique is later used for quantitative evaluation of the forming
jets from four commercial headboxes which is presented in $7. This is followed by the
velocity measurements and analysis of forming jets from laboratory scale headboxes with
- and without the guide sheets ($8). Conclusions and caveats comprise 59.
 2 Key Issues in Paper Forming Process

21. Headbox Requirements

Proper design and operation of the headbox is absolutely critical for an efficient performance
of the papermaking process. Readers are referred to a number of books and articles [1,3-
- 51 for a fundamental description of headboxes and various terminology involved in this
technology, such as forming table, forming wire, wet streaks, dry line, etc. Here we briefly
mention the functions of a modern headbox. The primary function of an ideal headbox is
to take the stock (a mixture of pulp and water), delivered by a pipeline, via a fan pump,
and transform it into a uniform, rectangular turbulent flow. The width of the jet is the
same as that of the paper forming table and the jet has uniform velocity profiles in the
machine direction (MD), cross direction (CD), and in the direction normal to the forming
table (ZD). While spreading, the jet should evenly disperse the stock across the length,
width, and breadth of the forming table. Prom a practical point of view, this could only
be achieved if the jet’s velocity profiles have no gradients in MD, CD, and ZD; but do have
a controlled level of turbulence to eliminate the fiber flocculation at a scale essential for
producing the different grades of paper.

The significance of the headbox in product uniformity is discussed in the next section.

2.2 Product Uniformity

Uniformity of physical properties of paper manufactured for printing, pAaging, and vaxi-

ous other applications is the most serious quality issue in today’s mills. Three fundamental
physical characteristics of paper are the density, the basis weight or the weight per unit in-
plane area of the sheet, and the fiber orientation. These characteristics are of immediate
importance since many other physical properties are directly influenced by them to varying
degrees. Thus, any variations in basis weight, density, and fiber orientation decide the qual-
ity of the finished paper. So, the greater product uniformity coupled with today’s faster
machine speeds and increased production rates place an enormous demand/responsibility
on the papermaking operation. But the quality of sheet formation does not begin at the

4
 forming table, instead, it starts at the headbox. The fundamental characteristics mentioned
above depend on various aspects of hydrodynamics of the flow inside the headbox, viz., the
secondary flow structure, the homogeneity and uniformity of turbulence at the slice, growth
of instabilities after the jet impingement, and the evolution of layer along the forming sec-
tion. In other words, if the flow from the headbox is not at its best, if the stock is not
. uniformly distributed across the width of the forming table, if the impingement angle is not
properly oriented, if there are too much turbulence, if there are too many velocity gradi-
ents (in CD, MD, and ZD), or if there is consistency variation in CD, then too much of
the forming table needs to be used to even out the fiber distribution rather than for water
draining. This will eventually result in a loss of production rate and/or product quality. In
view of the importance of the headbox in the paper formation process, a number of artic les
have appeared in recent times documenting the challenges facing the papermaking technolo-
gies [4,6-81. Also, in recent years, fine-scale velocity measurements using the High Speed
Digital Video Analyzer [g-11] and the Laser Doppler Velocimetry (LDV) [12] (see Appendix
A), in conjunction with controlled numerical simulations of headbox flows [10,13-161, have
considerably improved our understanding of the complex hydrodynamical processes inside
the headbox and at the wet end.

In the next section, we discuss the different scales of nonuniformities.

2.3 Large Scale vs. Small Scale Nonuniformities

There are several length scales associated with the physical characteristics of commercially
manufactured paper and board. The largest length scale in CD, obviously, is the width of
the paper machine. Most paper manufactured today have variations in physical properties
of the order of this length scale, such as the variation of average fiber angle in CD. For
most of the present-day machines this corresponds to a span greater than 6 in. in CD.
These variations are the so-called large scale variations and may result due to the following
reasons:

l Overall imbalance in the pressure manifold

l Edge effects in the headbox

5
 0 Cross flows due to slice profiling

0 Secondary turbulent structures [13]

0 Misalignment of a section of the liquid jet with the forming wire

l Misalignment of rolls
.

0 Nonuniform tension and edge effects in the press and dryer sections

various remedies to correct these large-scale variations are available for implementation.
TT l

However, some high resolution measurements show that today’s commercially manufactured
paper and board, have small-scale basis weight variations ranging from 5 to 20%, with peak-
to-peak CD spans of 0.5 - 4 in. For example, basis weight measurements and ultrasonic
stiffness measurements by Hall [17] show variations of the order of this length scale. It is
not clear at this time if the basis weight variation or the drying rate nonuniformity is the
primary cause of the nonuniform elastic stiffness properties. However, the length scale of
the fluctuations in physical characteristics is the same as the width of the wet streaks on the
forming table. In his experiments, Hall collected extensive measurements of fluid velocities
in CD, MD, and ZD on linerboard samples, using ultrasound techniques. The measurements
were repeated several times to confirm the existence of small-scale fluctuations. The results
from similar measurements overlap, showing real variations in the length scale mentioned
above. This reproducibility also shows that with sufficient care, the small-scale variations
can be reliably measured and characterized for linerboard and other grades. Since the
performance and strength of the product usually depend on the magnitude of the variation
in physical properties, the small-scale fluctuations could be more important than the overall
averaged nonuniformities. As a result, in this report, we focus our attention mostly on this
scale of nonuniformities.

In the subsequent sections, we present a discussion of various investigations (numerical,

experimental, and analytical) of the hydrodynamics of headbox and the forming table,
mainly carried out by our group at IPST. References to research carried out elsewhere
appear where necessary.

6
3 Hydrodynamics of Streaks on the Forming Table

3.1 Moisture Streaks and their Causes

Variations in basis weight, which at smaller scale correlate with the distribution of moisture
streaks, could have many causes. We categorize these into three classes: (1) nonuniformity
in fiber concentration or consistency, (2) nonuniformity in mass distribution on the forming
wire or forming table, and (3) nonuniform drainage in the forming section. Of these, we are
concerned with the second and third mechanisms, that is, the dynamics of the formation
process. The possible nonuniformities caused by the structural defects in the headbox and
the forming section are another issue which will not be discussed here.

The smaller scale variations in basis weight, which we consider here, are not well under-
stood and are, therefore, more difficult to eliminate. The magnitudes of the small-scale
fluctuations in basis weight are often considerably larger than the magnitude of the basis
weight variations recorded by the on-line sensors. Since the performance and strength of
the product usually depend on the magnitude of the variation in physical properties, the
small-scale fluctuations could be more important than the overall averaged nonuniformities.
The small scale fluctuations in elastic stiffness can be caused either by variations in the local
sheet density, small-scale variations in fiber orientation, or by the nonuniform drying rate
due to moisture streaks in the sheet.

In the next subsection we analyze the streaks on the forming wire.

3.2 Variations of Streaks on the Forming Wire

Turbulent flow, by definition, is a combination of high frequency indeterministic random

flow fluctuations superimposed on a deterministic mean flow. The time-frozen snap shot
of the velocity profile in a turbulent flow shows many fluctuations with an average trend.
If a large number of the instantaneous turbulent profiles are added and averaged together,
then a smooth profile, referred to as the mean turbulent profile, emerges. In contrast to the
random turbulent fluctuations, the mean turbulent profile is a deterministic or predictable
quantity.
In order to understand the flow in the forming section over a scale larger than l/2 inch, we
need to focus on the mean turbulent flow on the forming wire. The turbulent fluctuations
in the flow are also very important in terms of formation and effects on the evolution of
streaks.

Detailed high-speed digital imaging of the forming table is one of the methods that we
use in our program to study the physics of paper forming. An elaborate discussion of this
technique is presented in a later section ($6). The high-speed images can be taken up to 4000
frames per second with an exposure time as short as 1 microsecond. The small exposure
time of a single frame captures the forming layer in a time-frozen form showing the details
of the small-scale fluctuations due to turbulent flow. An example of this approach will be
presented here with analvsis of various regions of the forming section, as outlined in Fig. 1.

Free-surf ace
Forming Jet

r
Physical Property
Variations Due to
Streaks

Figure 1: Streaks on the forming table. Locations of the images presented in Figs. 2-8 are
outlined.

A single frame of the flow over a large section of the forming table is shown in Fig. 2.
The machine speed is at 3000 fpm making base sheet for a coated grade. In this view, the
slice and the impingement line are covered by the headbox, as indicated in the figure. This
image shows many small-scale fluctuations on the forming table. A time-frozen image, as
indicated below, is a superposition of randomly occurring flow fluctuations on the mean flow.
However, from this image, it is not clear whether there are consistent regularly patterned

8
streaks on the forming table. To examine this, we need to remove the random fluctuations
and look at the mean flow.

Figure 2: Time-frozen images of the flow over the forming table. (Machine speed =
3OOOft/min. Exposure time of the image = 50~s)

By averaging these images over several hundred frames, we can remove the small and
randomly occurring features from the image and examine the mean flow. Since the images
that we obtain are in digital form, we have developed methods to greatly enhance the image
and to take the average of a specified number of images [ll]. For this machine speed, an
exposure time of 50 microseconds is found to be sufficient. The average flow over 500 frames,
each with an exposure time of 50 microseconds taken at a rate of 1000 frames per second
covering 0.5 set in real time, is shown in Fig. 3. This image clearly shows that there is a
consistent streak pattern in the mean flow over the forming table. To determine the length
scale of the streaks, a grid line with tick marks is superimposed on the image. Each division
represents 12 inches.
 .

Lomtioa
H Slice
Locrtion
H

Figure 3: Average of 500 images of the forming table in Fig. 1, covering a period of 0.5 sec.
The lines show a 12-inch-long scale in MD and CD.

3.3 Amplification of Free Surface Waves at the Impingement Site

As the forming jet delivered from the slice impinges on the wire, the variations in the free
surface height, or in other words, the free surface waves or disturbances, amplify due to the
forcing of the jet flow along a curved surface. This results in local centrifugal acceleration of
the fluid particles normal to the curved surface. At this surface, the local acceleration forces
the jet to lose a part of its average kinetic energy to the random fluctuations (the physics
is similar to the Rayleigh-Taylor instability [l&21]). The amplification of the disturbances
increases with the magnitude of the acceleration or the jet angle: the larger the impingement
angle, the greater will be the magnitude of the acceleration. This is due to an acceleration-
induced instability which is well understood. The top view of a typical stock jump is shown
in Fig. 4 where the machine direction is from right to left of the figure. This is a single
frame taken directly from the top of the wire immediately after the impingement line.

The question is whether one can eliminate the streaks (see Fig. 3) on the forming table by
strong disturbance of the forming jet and creation of large stock activities at the impinge-
ment region. By averaging the top view images of the stock activity after the impingement
line, as shown in Fig. 5, we examine the character of the flow in this region.

10
Figure 4: Time-frozen top view image of the stock jump after the impingement line. The
MD is from right to left.(Machine speed=30OOft/min, exposure time of the image=50pS)

Figure 5: Average of 100 images of the top view image of the stock jump after the impinge-
ment line shown in Fig. 4, covering a period of 0.1 sec. The lines show a l-inch grid line.
The MD is from right to left.

It is clear from this figure that the mean free-surface flow of the forming layer at the post-
impingement line is nonuniform. A regular streak pattern develops in the mean flow even
with severe stock jump. This clearly demonstrates that at least in the situations examined
in this study, the regular stock activity on the wire cannot eliminate the streaks in the

11
forming jet.

In the next subsection we discuss the origin of these streaks.

3.4 Origin and Nonuniformities of Streaks

Let us take a closer look at the forming jet delivered from the slice. The enhanced time-
frozen image of the forming jet and the impingement line is shown in Fig. 6. This image
clearly shows streaks in the forming jet that amplify upon impingement on the forming wire
and result in some degree of stock jump. Some believe that the streaks in the forming jet
are due to the turbulent fluctuations inside the headbox and, therefore, occur randomly in
the system. To examine this point, we again take an average of the flow over 0.5 sec. The
averaged image is presented in Fig. 7. It is clear from the averaged result that the streaks
appear consistently, not randomly, in the forming jet. Turbulent fluctuations, however,
appear randomly. Therefore, the streaks in the forming jet are not the result of turbulent
fluctuations.

Figure 6: Time-frozen image of the forming jet, jet impingement region, and the stock
activity after the impingement line. The marks on the line at slice show a l-inch-long scale
(Machine speed=3OOOft/min, exposure time=50pS.)

12
Figure 7: Average of 500 images of the forming region shown in Fig. 4, covering a period
of 0.5 sec.

Using large-scale computational methods, we show (1996 IPST report to member compa-
nies) that different forms of small-scale secondary flows result from the interaction of the
jets from the tubes in the tube bank. The nature of the secondary flows depends on the
geometry and size of the tubes and their arrangement pattern in the tube bank.

Now, consider the jets that are delivered by these tubes. The streaks in the forming jet,
such as those in Fig. 7, are caused by the secondary flows induced by the jets from the
tubes. These secondary flows are spatially periodic in nature. That is, there is a given
CD span over which the structure of the flow pattern repeats. Let us refer to the CD
dimension of this periodic pattern as the width or the wavelength of the secondary flow.
The wavelength of the tube jet-induced secondary flow is not necessarily equal to the CD
spacing of the tubes in the tube bank. The secondary flows often appear in the form of
vortex tubes and recirculating vortices, which are superimposed on the main stream flow.
Secondary flows, not to be confused with turbulent fluctuations, are a deterministic part
of the flow generated by various mechanisms and can be computed using the mean form of
the momentum equation and appropriate turbulent models [13].

In the converging section of the headbox, the vortex tubes stretch along the direction of the
flow and cause pressure fluctuations, as shown in Fig. 8 [22]. The pressure fluctuations, in
turn, deform the free surface of the liquid into the streamwise streaks (SS) that we see in
Fig. 6. The upward and downward streams in the secondary flow form the crest and the

13
 trough of SS, respectively. The structure of SS that form at the slice depends directly on
these secondary flows in combination with the nature of the headbox converging zone (i.e.,
the outlet of the tubes in the tube bank) and the slice geometry. In some cases, there are
two SS generated for each column of tubes. Every pair of SS corresponds to a column of
tubes. Other types are when SS form from jets delivered by staggered tubes where a weak
a secondary SS exists next to a stronger primary SS.

Figure 8: Different types of streak creation, a) Vortex stretching, b) Centrifugal instability,

tv
c) boundary layer transition, and d) Wave instability.

Our computational analysis of the interaction between the tube jets inside the headbox
reveals the precise form of these secondary flows. Here we have discussed the origin of SS
in the forming section. In the next subsection, we discuss the dynamics of SS.

3.5 Dynamics of the Streamwise Streaks in the Forming Jet

SS on the forming table are caused by nonuniform or secondary flows in the headbox, the
jet interaction with the wire, and the fluid dynamics of the free surface flow on the forming
table. With most forming sections, one can immediately see the narrow SS that form near

14
the slice. These SS are l/2 to 4 inches wide and correspond to the small-scale secondary
flows emanating from the tube jets inside hydraulic headboxes or the wake of holes in the
slice rectifier roll.

In order to analyze the dynamics of SS in the forming jet, we use a time-line projection
imaging method. Events occurring in a fixed section of space are collected and assembled
sequentially along a time line for analysis. This is shown in Fig. 9 where we have recon-
structed the images of the forming jet over an area of 0.6 by 6.2 inches at 1 inch from the
slice over a period of 30 milliseconds in time.

Time
r
Figure 9: Time-line projection of the forming jet at 1 inch from the slice showing the
dynamics of streaks delivered from the headbox.

This figure shows 30 sections outlined by red lines. The distance between adjacent red lines
is 0.6 inch. The blue lines show a l-inch square grid representing the length scale of the
image. The images are taken at 1000 frames per second and each frame has an exposure
time of 50 pS. Each 0.6-inch section, outlined by the red lines, represents the image of
the same location of the forming jet (i.e., at 1 inch from the slice). Since the forming jet
is moving on average at 3000 fpm or 0.6 inch per millisecond, the 0.6-inch sections move
by exactly 0.6 inch in MD at each millisecond. The time line in Fig. 9 covers 30 images
of the 0.6-inch sections from time 0 to time 30 milliseconds stacked in sequence from top
to bottom. Time-line images up to several seconds can be constructed for analysis of the
forming jet dynamics.

15
 The physics and dynamics of the forming jet delivered by the headbox can be analyzed with
these time-line images, as shown in Fig. 9. Considering that in this headbox, the spacing
between the tubes in each row is about 1 inch and the tubes have a staggered pattern, and
considering the average SS width in Fig. ‘7 and 9, one can state with some confidence that
SS are due to the secondary flows from the jets delivered from individual tubes in the tube
- bank. In Fig. 9, we see the effect of jets interacting from adjacent tubes. Careful analysis
of these time-line images shows that the jets from adjacent tubes from two rows (i.e., w
0.5 inch apart in CD) interact and compete with each other to leave a SS in the forming
jet. Through this interaction, SS are deflected sideways. The interaction and subsequent
deflection of the tube jets in the headbox result in sideways flapping-like motion of SS in
the forming jet. The sideways motion of SS in the forming jet have significant impact on
the small-scale fiber orientation and formation in the sheet. In fact, we believe that the
small-scale variations in fiber orientation are caused by the local dynamics of SS in the
forming jet.

Although here we have only presented the images of the top of the forming section, we note
that SS should be viewed in terms of secondary flows throughout the forming jet. Similar
SS exist at the underside of the forming jet. Interaction of these SS with the wire determine
the structure of the sheet.

In the next subsection, we discuss the relation between the dynamics of SS in the forming
section with the small-scale physical properties of the sheet.

3.6 The Impact of the Streak Dynamics on Small Scale Physical Proper-
ties

We can approximate the variation in angle of the forming jet SS with MD from the time-
line images in Fig. 9. If the distance from the tube outlet to the slice is 29 inches and the
distance from the slice to the impingement line is 6 inches, then for every 0.1 inch that a SS
moves in CD of Fig. 9, the angle of the SS changes roughly by tan1(0.1/30) = 0.2 degrees.
Considering that fiber orientation is governed by the direction (or angle) of the velocity
field relative to MD in the forming layer, this variation in the local forming jet angle would

16
result in significant fiber misalignment and fiber orientation nonuniformity over a small CD
length scale. The magnitude of the variation would depend on the forming jet angle and
the rush/drag ratio (i.e., the shear rate in the forming layer). We predict that the length
scale of the in-plane average fiber orientation variation in the first layer from the wire side
would be in the order of the length scale of SS in the forming jet.

Jet/wire interaction and the evolution of SS along the forming section have significant
impact on the small scale basis-weight and moisture profile. The mechanism and interaction
between the layer on the forming section and the formation characteristics on the wire are
currently under study.

In order to eliminate the nonuniformities in physical properties caused by SS in the forming

section, we need to eliminate the small-scale secondary flows induced by the tube jets inside
the .headbox.

17
4 Streak/Wave Analysis on the Forming Table

4.1 Schematic of a Drainage Model

liquid surface

height Liquid layer with low fiber

concentration MD

\ Zone 2,
wire f Fiber mat with high
fiber concentration

Figure 10: A simple drainage model showing the two different liquid layer zones on the
forming wire.

Consider the drainage model on the forming table outlined in Fig. 10. Zone 1 has the
same fiber concentration as the liquid inside the headbox, whereas Zone 2 has much higher
concentration (-5 to 7 percent). The dry line forms at the point along MD where Zone 2
has completely drained. The shape of the dry line, visible on most forming tables, depends
on the CD variations in the thickness of Zone 1. If Zone 1 has nonuniform thickness in CD,
then the dry line becomes streaky, as demonstrated schematically in Fig. 1. Wet streaks
are regions at the dry line which contain more liquid with low concentration (headbox level,
approximately) of dispersed fiber than the rest of the layer (see Fig. 1). Assuming that
the fibers are uniformly dispersed, this results in variation in mass density of the fibers in
the sheet. Also, since the streaks contain more liquid, the moisture content and drying rate
vary according to the width of the streaks. In most cases, the streaks result in detectable
basis weight variations in the sheet. The length and the width of the streaks are important
factors in the overall effect on nonuniform physical properties. A very short or narrow streak
has no detectable effect on the physical characteristics. In contrast, long and relatively wide
streaks result in detectable and often troublesome nonuniformities.

18
As mentioned earlier, SS on the forming table are caused by nonuniform or secondary flows
in the headbox, the jet interaction with the wire, and the fluid dynamics of the free surface
flow on the forming tables. With most forming tables, one can immediately see the narrow
streaks that form near the slice. These streaks are 1 to 5 cm wide and correspond to the jets
emanating from the tubes inside hydraulic headboxes or the wake of the holes in the slice
rectifier roll. The streaks at the dry line are 0.2 to 0.5 m wide, depending on the headbox
and the forming section design and flow parameters.

In order to eliminate the nonuniformities in physical properties caused by wet streaks in

the forming section, we need to understand the origin of streak/wave formation and the
physical mechanism of their evolution on the forming table to the relatively wide structures
at the dry line. In the following subsection, we explain the physics of wave/streak formation
and evolution on the forming table.

4.2 Fourier Analysis of Wet Streaks

The wave analysis is based on the spectral decomposition of the liquid surface height near
the slice and computation of the wave evolution on the wire from the slice to the dry line.
The purpose of this analysis is to explain the physical mechanism of the evolution of the
apparently narrow streaks near the slice to the relatively wide streaks (0.2 to 0.5 m) at the
dry line.

To analyze the evolution of the waves on the forming table, we use a filtration
-.Y model of
paper forming on the table, as demonstrated in Fig. 10. We distinguish two zones on the
wire. One, where the fiber mat is forming on the wire, and the second zone where the
liquid has a low concentration of fiber equal to that of the headbox. The analysis takes the
amount of the drainage rate measurements to compute the average height of Zone 1. The
waves or streaks are variations in the height of Zone 1 in the cross-machine direction.

Since the wave amplitude is smaller than the wavelength of the range of streaks that we
are concerned about, one can show that in this case, the convective terms in the momen-
tum equation are much smaller than the other terms. Therefore, for this application, the

19
momentum equation can be reduced to

where u’ and p are the velocity vector and the pressure, respectively, p is the density of the
liquid, Dij is the rate of increase of strain, and the viscosity term j!Z is the sum of dynamic,
effective viscosity of fiber suspension and turbulent viscosities, i.e.,peff + pt. We use the
Prandtl’s mixing length theory, given by

where, I is the mixing length (i.e., Eq. 28 which appears in a later section) and

(3)

Because of conservation of mass and the incompressibility condition, the velocity vector is
a divergence-free function.

The height of the fluid surface defined as, C, is expanded in terms of Fourier modes, as

C(x,y) = 2 $[G] exp(ik,x + an;) (4)

n=l

where
G = A&xp(knJ) - exp(-knc)] + iC&xp(L~) - exp(-l,c)]

and
%z is the nth mode wave amplitude;
xn is the nth mode wave length;
s is the layer thickness on the wire (height functions);
kn & L are the nth mode wave numbers (kn = 2n/Xn and Zz = ki + an/D);
fi= ueff + Q is the sum of the kinematic viscosity of the fiber suspension plus
the eddy viscosity;
A2 & Ga are constants which depend on the nth mode initial disturbance
amplitude;
c is the average layer thickness (varies due to drainage along the
table);
=x:&Y are cross-machine( CD) and machine direction( MD), respectively.

20
 The free surface conditions are the balance of tangential and normal stresses along with
the dynamic condition at the free surface which constrains the free surface to a material
surface, as it physically is.

The expanded series is substituted in the governing equations, and using the free surface
conditions, the relation for the coefficients of the modes is derived. This relation is in the
.
form of a set of nonlinear algebraic equations which are solved using the nonlinear equation
solver in IMSL.

The relation between the exponential growth rate, a,, and the wavelength, A,, or the wave
number, kn, is given by a nonlinear algebraic equation:

where

veff = + + $;q+~ +27+24] l -•[23]

u is the viscosity of white water;
TI, ratio of the fiber length to diameter;
4 is the fiber volume fraction;
VT, fi, &rlL are functions of rp;
s~=,/~,G=~+w,
ZJO is the drainage rate on the wire;
9 is gravity; and
cr is the surface tension of the fluid.

4.3 Results

By calculating the growth and decay rates of the waves in the machine direction, y, along
the forming wire, we compute the surface height of the liquid along the full forming table.
Note that the results are functions of several parameters on the forming wire including the
drainage rate, the kinematic and turbulent viscosities, the average layer height, the machine
turbulent viscosities, the average layer height, the machine speed, and the location of the
drainage element. Therefore, the growth and decay rates for each mode vary along the
forming table. The streaks at the slice have a broad spectrum of wavelengths ranging from
a few centimeters to more than a meter. The characteristic length scale of the streaks at

21
the dry line are determined by the critical mode which becomes dominant as the streaks
are evolving on the forming table. The reason for this is that the modes with smaller
wavelengths grow and decay very rapidly before they reach the dry line. The streaks with
larger wavelengths take too long to grow and, therefore, they are of no consequence. Fig.
lla and llb show the growth and decay rates of the waves at a given MD location on the
wire. Only the critical mode grows to its maximum amplitude just in time to reach the dry
line and form the dry-line streaks.

Wave Length (m)

-8

‘*(b)
.8

1.8

Wave Length (m)

Figure 11: Rates of growth and decay of streaks on the forming table
(a) Decay of streaks (b) Growth of streaks

22
 6.00E-02

[Link]-02 E

% [Link]-02
w
I!

Jii [Link]-02
.
8
2.00E-02

[Link]-02

[Link]+OO ’
0 1 2 3 4 5 6 7 8

CD Location trn)

Figure 12: Computation of surface height across the machine at about 6’ from the slice

For the headbox and the forming table considered in this study, the wave amplitudes on
the wire near the slice and at about 2 m from the slice are plotted in Fig. 12. Note that
the streaks quickly evolve to the dominant mode which in this case is about 0.45 m wide.

The measurements of the layer thickness on the wire at the same thickness from the slice
show good qualitative agreement with the calculations.

The streaks that are computed at the dry line have the same 0.4 to 0.5 m width as observed
on the machine. Also, the results show the same qualitative length scale of fluctuations as
found in the ZD ultrasound measurements presented in [17].

23
5 Effect of Secondary Flow on Fiber Orientation

5.1 Two Types of Secondary Flow

Secondary flows in a headbox are weak, three-dimensional flow patterns that superimpose
on the basic or primary two-dimensional flow due to hydrodynamic effects alone. These
secondary patterns have small velocity components in the CD. Although the magnitude of
the CD velocity components is small compared with the MD velocity, its effect on fiber
orientation is significant. Cross-flows originating in the headbox result in nonuniform shear
on the wire in the CD. Considering that the shear stress on the wire is the main factor de-
termining fiber orientation, the secondary flows inside the headbox have profound influence
on the average fiber angle. This causes nonuniform fiber orientation in the CD.

Two kinds of secondary flows could occur in the headbox. The first kind results from
flow retardation due to boundary layer formation at the side wall of a converging channel.
Examples of secondary flow induced by boundary layers at side walls occur in various hy-
drodynamic systems [24]. The second kind of cross-flow is from the anisotropy in turbulent
Reynolds stresses. A discussion follows later.

Consider a straight rectangular channel. Here, the simple geometry of the channel provides
one type of fluid problem. It is possible to isolate the physics of the formation of secondary
flows of the second kind [25,26] from other cross-flow generating effects. In this geometric
configuration, the structure of turbulent fluid motion is the sole generator of the secondary
mean flows. The weak secondary flow is in a plane normal to the streamwise axis of the
channel. Its magnitude is approximately l-3% of the main flow. This flow configuration
also provides the opportunity to develop and test turbulent closure models, considering the
correct and accurate simulation of secondary currents [27,28].

Recent investigators [29,30] have presented direct numerical simulations (DNS) of low
Reynolds number turbulent flows in straight channels. These simulations provide a database
from which it is possible to calculate the turbulent velocity and pressure correlations. The
computed instantaneous velocity field near the channel will capture the bursting events
also observed in some experiments [31-331. Averaging the flow field in both the streamwise

24
direction and in time provides mean flow streamwise velocity and secondary velocity dis-
tributions. Ejection structures produced during bursting events create streamwise vortices
[29]. Near the corners, the reduced mean shear prohibits the occurrence of ejections. Mean-
while, there are strong ejections along the walls away from the corners. This distribu tion of
ejections is the major contributor to the anisotropy of the normal Reyn .olds stresses. It also
generates the mean secondary circulations carrying streamwise momentum from the core
of the channel towards the corners. The earlier investigators also observed a nonlinear in-
teraction between simultaneous ejections produced along perpendicular walls near a corner
[29]. This interaction results in the redistribution of turbulent intensities there.

Simulating flow problems of technological interest with DNS is beyond the scope of cur-
rently available computers. For example, to achieve a sufficient spatial resolution for the
turbulent channel flow, Gavrilakis [30] had to use 16.1 million grid nodes and more than
60 days on a supercomputer for one run. One must, therefore, rely on a turbulence model
that can correctly account for the anisotropies in a more realistic computational time.
Speziale [28] evaluated different turbulent closure models. He suggests the application of
the two-equation, nonlinear K - c method for engineering calculations. An advantage of this
method compared to the linear (standard) K - E method is the more accurate prediction of
the turbulent channel flows. The consequence is better turbulent secondary flow prediction.
The derivation used an asymptotic expansion subject to the constraints of dimensional and
tensorial invariance and realizability [34]. This model obtains the turbulent kinetic energy,
K, and dissipation rate, E, from a modeled version of transport equations [34,35]. These
are the same form as those used in the linear K - e method. The closure formulation for
the Reynolds stress tensor, rt, takes a simple non-linear represent ation introducing new
quadratic terms in the mean velocity gradients. This formulation is similar to the constitu-
tive equation of dilute polymers. It contains two more constants compared with the linear
model. Speziale [34] concludes that future research is necessary to obtain a more detailed
picture of the turbulent structure in rectangular channels predicted by the nonlinear model.

Colombini [36] applied the nonlinear K - E method in the linear stability analysis of the
formation of longitudinal ridges. In straight, infinitely wide channels, the free surface tur-

25
bulent flow over a deformable and erodible bed is unstable to transverse disturbances with
a wavelength equivalent to the flow depth. The nonlinear K - c model has been successful
in modeling the turbulence-driven, cellular secondary motions directed to amplify the bed
perturbations. The interactions of the flow and the erodible bed thus lead to the formation
of longitudinal ridges even without side walls.

Nezu et al. [37] have studied secondary currents in straight channel free surface flows of
different aspect ratios. They address the initiation and maintenance mechanism of cellular
secondary currents in a straight, wide channel with a flat solid bed. They concluded that
secondary currents appear only near the side walls. If the ratio of width to depth, B/h, is
greater than 4, there are no cellular secondary flows in the central region of the channel.
Numerical simulations of other investigators [38] agree with these findings. Our work will
systematically evaluate the effect of aspect ratio in confined and free surface flows.

Here, we consider the analysis of secondary flows and their effects on fiber orientation. The
following subsections outline the governing equations and the methods applied in the nu-
merical simulations. Applications to the headbox hydrodynamics focusing on the secondary
flows of the first and second kind are in the following subsections. This section concludes
with a brief summary of present results.

5.2 Governing Fluid Flow Equations

The following momentum conservation and continuity equations govern [Link] flow of an in-
compressible and homogeneous viscous fluid:

dU
P ;+ii-vii
>
= -vp+pg+v*7 (7)
v*ii=o (8)
where
u’ = ui = (u, v, w) = instantaneous fluid velocity vector,
P = pressure,
-
9 - gravitational acceleration or body force per unit of mass,
P = density, and
T = 2~Dij = fluid stress tensor, where Dij is the rate of increase of strain (Eq. 3).
For a Newtonian fluid with constant viscosity, the last term in the momentum conservation

26
equation is the following:
v l r = pv2ii
(9)
where

P = dynamic viscosity of the fluid.

In a turbulent flow, the velocity and pressure field decompose into ensemble mean and
fluctuating parts, respectively, as the following shows:

u’= ii+ u” (10)

p = jJ+p’. (11)

In this equation, the overbar denotes the ensemble mean and the prime denotes the fluctu-
ating components. Figure 13 shows the mean and fluctuating velocity components in time
for steadv and unsteady
r/ flows, respectively.

Figure 13: Fluctuating and mean velocity components in steady and unsteady flows.

After substituting Eqs. 9 and 10 into Eq. 7 and averaging Eq. 7 in time, the governing
equation of a turbulent flow reduces to the following:
%
dU
P ~fibVii =-vp+pg+v-? (12)
>
v*u’=() (13)

In Eq. 12, the stress tensor, ?, includes both the viscous and turbulent Reynolds stress
tensors, such that the following equation is true:

V-?=pV2~+v*Tt (14)

27
The following equation provides the components of the Reynolds stress tensor [39], rt:

rt9 = -pu:u~. (15)

The Reynolds stress tensor, & introduces additional unknowns for a turbulent flow problem
governed by Eq. 12. To describe the mean velocity and pressure fields, a closure formulation
is necessary to relate the components of the Reynolds stress tensor to the mean flow veloc-
ity or velocity gradients. Without a closure formulation, DNS can theoretically solve the
Navier-Stokes equations-Eq. 7 combined with Eq. g-to obtain the instantaneous velocity
and pressure fields. This task demands a very fine space discreatization, however. It is not
a practical approach for many industrial problems.

5.3 Numerical Method

To calculate secondary flows of the first kind, one can apply the standard, linear PC- c model
for closure of the averaged momentum equation, Eq. 12. Any numerical method to obtain
the second kind of secondary currents in confined and free-surface flows should consider the
anisotropy of the turbulent field. The computational method for these calculations uses
the volume-of-fluid (VOF) computational technique [40,41]. Original development of this
technique was for both confined and free-surface flows. Our work extends its application
from laminar to turbulent flows employing a nonlinear TV- E described below.

McKibben and Aidun [40,41] described the VOF solution of the governing equations, i.e.,
Eq. 7 with Eq. 9 or Eq. 12 with Eq. 14. There has been modification of their numerical
method with the introduction of implicit shear stress calculation and the advancement of
the advection terms using the Adam-Bashforth scheme. These modifications enhance the
efficiency of the numerical simulations [42]. Other researchers have used similar numerical
techniques [30,43].

This subsection summarizes the steps of a cycle in the VOF procedure. In describing this
time-splitting method, the overbars are eliminated, and the procedure is applicable for both
turbulent and laminar flows with the inclusion of an appropriate closure for the stress terms.

First step: The initial procedure isl to calculate an intermediate or provisional 2 velocity

28
field using the following equation:

In this equation, the superscripts n and rz - 1 denote values calculated in previous n and
n - 1 cycles. The superscripts m and m - 1 refer to the iterative nature of the solution of
Eq. 16. At is the time increment, and r(m) symbolizes a functional relation and closure for
the stress calculation.

Second step: One next applies the continuity constraint to obtain the Poisson equation
for pressure:
AtV l 08~ = V .? WI

where

-- the pressure change needed to satisfy mass conservation.

Third step: Using the following equations, one next updates the pressure and velocity
fields:

n+l = pn + dl,
P (18)
-+l=,;
U - AtVap (19)

Earlier work [40,41] details the space discreatization of the different terms in Eq. 16-19 on a
staggered grid. They also describe the boundary conditions for the Navier-Stokes equation
and the solution of the pressure Poisson equation, Eq. 17. There has been extensive testing
of the code for laminar, free-surface and confined-flow benchmark problems.

The current study proposes the inclusion of a nonlinear K - c model to this code to capture
the secondary flows of the second kind induced by turbulent flow. The finite element
method in a fluid dynamics analysis package [44] solves the standard (linear) K - e model.
This provides the secondary flows of the first kind presented in this paper for confined flow
in a headbox.

29
5.4 Headbox Secondary Flows Due to Sidewalls

This subsection focuses on the secondary flow of the first kind. This is the main cause of
nonuniform fiber orientation in CD. Table 1 provides the specifications for the full-scale
headbox considered in this work. Figure 14 is a diagram showing the headbox and an
individual tube, and Fig. 15 shows the geometry and the computational grid.

Since the solution for the governing equations is in nondimensional form, the results are
also nondimensional. The length scale is the size of the box inlet, and the velocity scale is
the inlet velocity of the fluid into the box. Table 2 offers a conversion for the dimensionless
members in the figures to dimensional quantities for the case of 45,000 gal/min total flow
rate through a headbox with a 1.5in. slice opening. Here the average forming jet speed will
be 2400 ft/min with a higher or lower machine speed, depending on the rush or drag mode
of operation.

Since the boundary conditions and the geometry considered in this study are symmetrical
in the Z or MD plane of the headbox, we present the results for half the box. The result
for the full box is a mirror image projection for the other half.

Table 1: Headbox parameters and flow conditions

Parameters Dimensional
Width 300 in.
Inlet height 15 in. 1 ..
Total flow rate 45,000 gal/min
Slice opening, B 1.5 in. 01.
Average jet velocity 2,400 ft/min 10
No vanes
No side ejection

30
 Tube bank

Figure 14: Schematic of a headbox and an individual tube

Figure 15: The geometry and computational grid system for the headbox

Table 2: Conversion factor for nondimensional to dimensional units

7
For the case of 45,000 gal/min
To convert To dimension and 1.5 in. slice multiply by
Length m @I 0.381 (1.25)
Velocity m/s (ft/min) 1.22 (240)
Pressure N/m2 (psi) 1488 (0.216)

31
5.5 Uniform Inflow

Consider first the case with uniform pressure at the inlet to the box. Figure 16 shows the
resulting pressure contour lines in the symmetry plane. As expected, the contour plots show
that pressure decreases as the fluid accelerates through the box. The pressure contour lines
become denser near the outflow. This indicates sharper decrease in pressure (larger pressure
gradient) due to the rapid velocity increase toward the channel exit. Figure 17 illustrates a
velocity vector plot. These plots only show a cross section of the headbox. To learn more
about the global features of the flow, it is necessary to examine the pressure variations in
the CD.

Figure 16: Pressure contour plot at the middle cross section of the symmetry plane

Figure 17: Velocity vector plot at the middle cross section of the symmetry plane

To investigate the pressure distribution in the headbox, Figure 18 provides the pressure
contour plots on the inner surface showing a three-dimensional view of pressure on the
boundaries of the headbox. Figure 19 shows the pressure contours or isobars in a horizontal
plane located in the middle of the slice. Although the boundary conditions are uniform in
the CD, the isobars show that the variation in the pressure field is three dimensional. The

32
pressure changes in both the CD and the MD.

Figure 18: Pressure contour plot on the boundaries of the headbox

Figure 19: Top view of the pressure contour plot at the horizontal plane intersecting the
middle of the slice

For quantitative information, one can compare the magnitude of the pressure variation at
various levels of the slice presented in Figure 20 to the plot of the MD pressure drop shown in
Figure 21. The pressure variation in CD is primarily due to the retardation of velocity near
the sidewall. Our objective was to study the generation and features of cross flows because
of their effects on fiber orientation. It was therefore important to examine the influence of
CD pressure variation on the flow. We wanted to see if the CD pressure gradient generated

33
 finite cross-flow.

-
2.06 3.00 4.00 S.00 6.00 7.00 8.00 9.00
Y

Figure 20: CD pressure variation in the headbox at the slice (rO is fluid density, U is
characteristic inflow velocity)

Figure 21: MD pressure drop along the axis of symmetry

To address this question, it was necessary to examine the CD(or y) component of velocity(v)
inside the headbox. The contour plots of v shown in Figures 22 and 23 clearly indicate a
local minimum near the sidewall. The values of this local minimum is small compared to
the positive value of the cross-flow velocity. Figure 22 illustrating the ‘distribution of the
y-component of velocity in the slice also indicates this. The relative magnitude of the cross-
flow velocity components is the critical factor in generation of nonuniform fiber orientation.
Figures 24 and 25 show the MD and CD components of velocity along a line in the middle
of the slice, respectively. Besides the local minimum near the sidewall, these figures clearly
indicate a positive cross-flow. This reaches a maximum at approximately 3.3 ft (1 m) from
the sidewall of the headbox. Comparing the x- and y-components of velocity shows that
the maximum cross-flow is approximately 1% of the x-component of velocity. Generation
of the cross-flow is due to the converging section of the headbox in the x-z plane bounded
by the vertical sidewalls. The boundary layer at the vertical sidewall retard the flow in the

34
 streamwise direction. Mass conservation therefore force the flow to move in the CD.

Y
- 9 .I?91

Figure 22: Contour plot of the y-component of fluid velocity in CD inside the headbox

Figure 23: Top view of the horizontal plane intersecting the middle of the slice

640 . \r

6.06 .

4.00 l

2.66 .

'0 c 1.00 200 3.00 4.00 500

. 600
. 700
. 8.00 900
. ,&ii

Figure 24: The x- or MD component of fluid velocity, u, at the middle of the slice

35
 Figure 25: The y- or CD component of fluid velocity, u, at the middle of the slice

Even when a perfect pressure and flow rate distribution exists in the tube bank, these
results emphasize that the flow inside the headbox will be three-dimensional because of the
secondary flows. In addition, the results show the location of the maximum cross-flow.

Turbulent flow characteristics partly control the formation and flocculation. Figure 26 is
a plot of the computed turbulent flow characteristics inside the headbox. The individual
components of this figure present three-dimensional views of the turbulent kinetic energy,
K, dissipation, E, and turbulent viscosity, pt. This figure shows a turbulent intensity in the
headbox that is almost uniform. The only regions with large variations in turbulent kinetic
energy, dissipation, and viscosity are near the solid walls of the converging channel. As one
would expect, there are large velocity gradients here resulting in a more intense turbulent
energy production and dissipation.

5.6 The Influence of Nonuniform Inflow on the Cross-flow at the Slice

Dividing the inlet into three sections as Figure 27 shows helps in examining the effect of
nonuniform influx at the inlet to the headbox on the cross-flow at the slices. Figure 28
presents the cross-flow at the slice when the flux in section B is 4~ 10% of the flux in region
A. The results show that the flux into the headbox has significant effect on the cross-flow
at the slice. This result is consistent with the physical explanation of the nonuniform CD
velocity by the retardation of the flow due to the sidewall. These results demonstrate that
by computationally optimizing the influx, we can substantially reduce CD flow and fiber
disorientation.

36
 .

Turbulent kinetic energy

Turbulent dissipation

Figure 26: Three dimensional view showing turbulent kinetic energy, turbulent dissipation
rate, and turbulent (eddy) viscosity

A -- B
b A

X
Y t,

Figure 27: Schematic of regions showing variation in influx profile

37
 ‘*% I 100
. 2.00 300
. 400
. 500
. 600
. 100
. 000
. 900
. 10.0

f Y (CD) LOCATION

Figure 28: Variation of the CD velocity profile across the headbox

5.7 Secondary Flows Due to Turbulence Anisotropy

5.7.1 Origin of secondary flows of the second kind

To understand the formation of secondary currents in straight rectangular channels, it is

necessary to transform the governing equations into a vorticity stream function formulation.
This demonstrates the condition for the existence of a vorticity vector in the axial (main)
flow direction. Figure 29 illustrates the geometry of a rectangular channel.

Normal wall
bisector

Figure 29: Geometry of a straight rectangular duct

Taking the x-direction to be the axial streamwise flow direction and assuming that the
steady flow does not change in this direction provide the following set of equations:

38
 2 t
( dW dW a( &) a2rt a29
rYY - ZY ZY
P a --+a$
dY >
= pv2w, +
dydz +
---
dX2 dy2
(21)
v21i! = wz (22)
d!P
fj=- (23)
ax
a=-- ar!
dy
(24)
where
wx = axial vorticity
XP -- stream function

Speziale [45] provides the derivations for Eqs. 20-24. On the right hand side of Eq. 21,
the vorticity equation, the terms formed from the Reynolds stress components represent
the source term for vorticity production. If the difference of the normal Reynolds stresses
equals zero, i.e., $y = &7 then no generated vorticity results. Then wz = XP = ‘is = 0 = 0.
In other words, no secondary flow can exist in the duct. This would contradict earlier
experimental observations [46]. To model these flow phenomena numerically, the turbulent
closure model must be able to predict the anisotropy of turbulent normal stresses.

Figures 30 and 31 present the instantaneous velocity field - (u, v, w) - in a straight duct
calculated by DNS [29]. Figure 30 shows the streamlines of the main flow, and Figure 31

illustrates the instantaneous secondary flow vectors. Averaging the instantaneous stream-
lines of the main flow, U, and the secondary velocity vector field, v and w, in time and along
the duct axis provides the mean flow quantities. Figures 32 and 33 display%v these. Figure 32
shows the mean streamwise flow, a. Due to the secondary flow circulations directed from
the center of the duct toward the corners, these streamlines bulge along the corner bisectors.
Figure 33 presents the mean secondary velocity field, v and a, causing this bulging.

39
 Qs

0.4

0.3
)r
0.2

0.1

0
0 0.1 0.2 0.3 0.4 0.S 0.6
2

Figure 30: Instantaneous velocity field in the lower left corner of a straight duct showing
main flow

Figure 31: Instantaneous velocity field in the lower left corner of a straight duct showing
secondary velocity field [29]

Figure 32: Ensemble mean velocity field in the lower left corner of a straight duct showing
mean main flow

40
 . ..-. . . . ,
0.4;l.;;,f;.H : - ..: .....:I....
I : ---..
. -. ... .
.
.
*
.
.
..

Figure 33: Ensemble mean velocity field in the lower left corner of a straight duct showing
mean secondary velocity field [29]

5.7.2 Numerical modeling with the non-linear K - c method

In closing Eqs. 12 and 14, the non-linear K -e method provides a formulation that one would
expect to simulate secondary flow currents. This formulation uses the secondary solution
of two more transport equations that are identical to those of the standard (linear) K- E
model.

The version of these equations in the present study comes from Speziale [34]. Eqs. 25 and
26 describe the transport of turbulent kinetic energy, K, and dissipation rate, E, respectively:

In these equations, Cc.) et al. are dimensionless constants.

In the lowest order nonlinear K - 6 model, it is possible to calculate the shear stress [34]
using the following equation:

The following equations provide terms for this equation:

41
In Eq. 27, the additional constants, Co and DE, are necessary in comparison to the linear
tc- E method. Determination of these parameters comes from calibration with the help of
experimental or DNS results.

Considering only the first two terms on the right-hand side of Eq. 27, one can obtain the
shear stress formulation of the linear K - E method. Remembering this, note that the linear
K - c model predicts the following for Eq. 21:

rt%% t t --2PK (31)

= ryy = rxx = 3

The linear formulation consequently does not generate vorticity. In contrast to the linear
model, the nonlinear method yields the following [34]:

copz2 da 2

3 4 ( ax >
g N 2PK

(32)
-- - m

+
%% =

- -- 2PK + -copz2 m
t rv aii 2
TYY
3 (33)
4 ( dy >

Equations 32 and 33 satisfy the necessary condition for the development of the secondary
flows, i.e., riy # &. This justifies the choice of the nonlinear K - c method for the present
study. %.

The simultaneous solution of Eqs. 25-30 with Eq. 12 provides the turbulent mean velocity
and pressure fields. Figure 34 displays the secondary flow streamlines calculated by a
simplified form of the nonlinear K - c method [34].

42
 Figure 34: Mean secondary flow streamlines [34]

5.8 Summary

In principle, secondary flow in a hydrodynamic system is a weak stream with a more complex
flow pattern superimposed on the primary flow pattern. In a headbox, the base flow is the
two-dimensional flow pattern with zero CD velocity component. The secondary flows are
weak three-dimensional flow patterns superimposed on the primary flow. The result is a
three-dimensional flow pattern. The secondary flows that are important in this study are
of two kinds:

l Type I: Secondary flow induced by the curvature in streamlines in the developing flow
inside the headbox and the transverse velocity that satisfy the continuity condition

l Type II: Secondary flow caused by the anisotropy of turbulent flows.

We have identified these two types of secondary flow inside the converging section of the
headbox. Such secondary flows result in nonuniform cross-flows in the headbox and at the
slice. The secondary flow of the first type is due to the flow retardation at the slice resulting
in misalignment of a section of the forming jet with the wire. The nonuniform shear from
this misalignment results in a nonuniform fiber orientation. The effects of the secondary
flows of the second type are not well established yet. We believe that the cross streams due
to the turbulent anisotropy result in more local nonuniformities in CD and MD. This gives
nonuniform fiber orientation in MD.

43
We can adiust the insert tubes at the tube bank to minimize the cross flow at the slice
and to eliminate the problems associated with nonuniform fiber orientation. We have used
this approach evaluate the hydrodynamics of a commercial headbox. The objective was
to optimize the influx for a more uniform fiber orientation. Computation provided the
optimized influx profile. The location of the insert tubes achieved the same influx profile
obtained computationally. During a shutdown period, adjustment of the insert tubes at the
tube bank accordingly gave notable improvement in the fiber orientation even at low rush
to drag ratios.

The only way to obtain the secondary flows of the second kind is by direct numerical
simulation of the full Navier-Stokes equations. This is very time consuming, and it is
not practical for engineering applications. We are extending this work to calibrate the
parameters in a nonlinear K - E model for computation of the secondary flows of the second
kind in irregular geometries.

In the next section we discuss a technique

I called High-Speed Digital Imaging which our
group at IPST uses for quantitative evaluation of paper forming hydrodynamics. A brief
mention of this technique has been made in $3, though no particular details have been
given. The next section presents this technique in greater detail. This same technique is
later used in 57 for a quantitative analysis of the forming jet delivered from four different
commercial headboxes.

44
 6 High-Speed Digital Imaging of Paper Forming

6.1 Introduction

It is often important to know the characteristics of the forming jet to evaluate the headbox
performance or to examine the wet end as a potential source of defects in the sheet proper-
- ties. In this section, we present a method for quantitative evaluation and comparison of the
forming jet hydrodynamics in commercial machines. This method is useful for investigat-
ing and eliminating wet end problems or for routine quality control of the forming section
hydrodynamics. Our interest in implementing this method is primarily to investigate the
source of the nonuniformities in formation, the methods to improve formation, and the re-
sult of any modifications on the formation characteristics. Visual inspection of the forming
section has been a useful tool for many years in various studies of paper and board form-
ing [10,47-501. The ability to see the often important details in the forming layer depends
on the visualization method. Simple inspection with the naked eye could reveal large and
dynamically slow features of the flow. Frozen images using strobe lights or short exposure
time photography (16 mm or 30 mm silver halide films) or videography (short exposure at
30 frames per second) are useful to some degree; however, many important details occur in
fraction of a second and cannot be captured in slow framing rate visualization. High-speed
digital imaging(lOOO frames per second or greater) can reveal new phenomena, opening the
door to significant improvements in the forming process. These images are recorded on an
optical disk in digital form allowing computational analysis of the images. In this section,
we describe a method for quantitative analysis of the forming section using high-speed dig-
ital imaging and image processing. This technique is used in the next section for evaluating
the forming jets delivered from four commercial headboxes.

6.2 The Imaging Hardware Systems

The imaging system used in this study consists of a transportable Kodak Ektapro Motion
Analyzer, Model 1000 HRC and a desktop Silicon Graphics workstation. The Kodak camera
system is capable of collecting 1000 to 4000 images per second. The images are then archived
to optical disk for later analysis. Archives of 500 images series (0.5 set) with exposure times

45
of 0.06 msec are typical for this study. Each image is 512 pixels wide by 384 pixels high. One
byte per pixel is used to digitize a gray level. This corresponds to 256 gray tones possible
per pixel. The sensor uses CCD technology to expose the image. In this particular sensor,
16 CCD’s are used to collect the image information. The image is divided into sixteen 64
by 192-pixel sections each employing a single CCD sensor. The pixel data are combined
with a [Link] footer containing framing information to produce an image file of 197,632
bytes. A typical 500-image acquisition will require 94 megabytes of disk space.

6.3 Preparation of the Raw Images

The quality of the raw images acquired at the mill is limited due to poor lighting and limited
access to the imaged area. We have developed methods to overcome these limitations and to
considerably enhance the raw images for more effective analysis. After images are acquired,
downloaded to optical disk, and transferred to a Silicon Graphics workstation they are
ready for processing. We have written special software for various computational analyses
of these image sets. All raw images are first patch equalized. Patch equalization is an image
processing procedure based on histogram equalization. Histogram equalization is used to
maximize the contrast (range) of each pixel in the raw image, however, it is not effective
in balancing any uneven illumination. Patch equalization is an extension of histogram
equalization that we have developed specifically to maximize contrast and to balance spot
lighting.

At a mill, powerful halogen lights are used to illuminate the scene to a point where 0.06-
msec exposures becomes possible. Even with a good on site lighting system, images will
have bright areas and as a result the camera system will not take advantage of the full
8-bit grayscale. The standard technique in commercial image analysis programs to stretch
.
an image over the full 8-bit range in histrogram equalization. However, histogram will not
balance uneven lighting conditions. To counter spot lighting we have developed methods
where the image is divided up into many patches. Each of these patches is histogram
equalized. This will balance uneven lighting, however, the seams between patches become a
visible artifact. To solve this problem, a moving patch is developed which is centered on the

46
 current pixel. A neighborhood of pixels around the current pixel is used to form the local
histogram for the equalization. Accumulation of histograms and equalization procedures
for each pixel of an image in a 500-image set can become a computationally time-consuming
effort. To accelerate the process, the requirement to accumulate histograms for each pixel is
relaxed and the same histogram is used over small groups of pixels (4 x 4 is typical). Since
. the patch over which the histogram is gathered is larger than the group it is applied to,
seams at the patch boundaries are effectively equalized and the algorithm time is reduced
significantly.

The patch and window sizes that display effective corrections are found to be 4 x 4 patch sizes
with a 25 x 25 pixel floating window. Comparisons of the different histogram equalization
methods are found in Fig. 35. A slight hallowing artifact may be introduced due to patch
equalization under certain conditions. This can sometimes be seen on image edges and
around sharp transitions to large constant features, such as headbox faces and large sharp
shadows. These static features, if present, are of scales much larger than the dynamic
features of interest and, as such, do not influence the final results.

(a) (b) (d)

Figure 35: Comparison of Equalization Techniques, (a) Original Raw Image, (b) Standard
Histogram Equalization, (c) Fixed Window Patch Equalization, (d)Moving Window Patch
Equalization

47
6.4 Averaging and Time Series Image Processing

Once the raw series of images have passed through the patch-equalization stage they may
be examined in detail. Two basic operations are performed to reduce the data. The first
type of operation is averaging. The second type of operation is time series compilation.

The flow in the headbox and on a significant part of the forming section is turbulent. Turbu-
lent flow is characterized by presence of flow fluctuations in time around a mean flow. The
mean flow can be unsteady, however, the time scale for the mean flow is much larger than
the time scale for the turbulent fluctuations. The mean flow is deterministic and predictable
whereas the turbulent fluctuations have random characteristics. Given “enough” compu-
tational resources, the Navier-Stokes equations can be solved using direct computational
methods for the turbulent velocity and pressure. However, except for very simple turbulent
flows, today’s computational resources are not adequate for direct turbulent analysis using
the Navier-Stokes equtions. The more practical approach is to solve for the mean velocity
and pressure, defined by
+T +T
ui = -1 1
uidt and jj = - Pdt i =1,2, and 3 (34)
T Jt T st

where the bar over the variable ui and p represents the mean velocity and pressure,
respectively, and the fluctuating component is defined by ut = ui - ?I&. This requires
modeling the Reynolds Stress term u&$, which arises in the mean N-S equations. In many
cases, visualization of the mean flow becomes very important for understanding the flow
characteristics. We have developed methods for construction of the mean flow pattern from
the high-speed digital images of the turbulent flow in the forming section. In this process
each pixel is accumulated for the entire image set and then divided by the total number of
images. This operation removes electrical noise and random turbulent components. What
remains is the mean flow field. Typically, 10, 50, 100, and 5004mage sets are used to
construct the mean flow. Once the average is produced it is patch-equalized to rescale the
range of an 8-bit pixel range.

Time series compilation is used to record activity that passed a particular point over the
acquisition time. The best way to illustrate this procedure is with an analogy. A room has

48
 were not accurately measured.

The dynamics of the streaks on the forming table have been captured using a digital video
system. Off-line image processing analysis and signal processing analysis and signal analysis
techniques were employed to quantify the scale and regularity of the streaks, as outlined
by Aidun and Ferrier [ll] and in Appendix B. To facilitate the comparison of one paper-
-
making machine to another, the data are plotted with similar scales and the image data
are printed with overlaid grids. FFTs for the averaged images and the time- series images
are presented and discussed. Furthermore, the FFT data may be directly compared across
machines by plotting the ratio of standard deviation to the average value.

7.2 Results and Analysis

In the imaging of the forming jet and the forming layer, we have focussed on the fluid
mechanics after the impingement line region. Despite our attempts, in some cases we were
unable to access and record the images at the same relative distance from the slice and
the impingement line. This is due to the differences in the headbox slice design and the
location of the catwalk on the forming table. For the sake of accurate documentation, we
have made careful records of the exact location on the forming table that has been imaged.
Figures 37 and 38 show the typical overhead view layout and a map of the locations where
each machine has been imaged. The distance from the imaging region to the slice has been
outlined in Figure 38.

54
 6.5 Spectral Analysis

The goal of the study is to examine and characterize nonuniformities on the table. At
this point, we have the high-speed image data in a form that can meet this goal. Fourier
methods are utilized to measure dominant wavelengths in the CD direction. The two types

e of reduced images formed display the streak features but in different ways. The averaged
view removes any turbulent fluctuations and electronic noise leaving mean features, while
the time series images retain the turbulent fluctuations and any electronic noise but show
feature dynamics over time. Any one scanline in the CD direction may be passed through
a Fast Fourier Transform (FFT) to recover any dominant wavelengths that make up these
streak features. The frequencies of features across image may be correlated to physical
parameters.

The FFT is a powerful statistical tool and much has been written on its use. An FFT
of any signal can contain useful informaion about that signal, however, with some signal
preconditioning misinterpretation of signal results can be reduced. In this study, signal
amplitudes are de-trended and then windowed using a Hann function. For a full discussion
of FFT methods the reader is referred to Bloomfield [51] or Press et al. [52]. Typical
ranges of frequencies resolvable are l/(2 pixels) to l/(512 pixels). The relative distance the
sensor positioned above the wire will determine the length scale within the image, however,
75 pixel/in. is typical. This scale results in frequencies of 38 cycles/in. to 0.2 cycle/in.
No aliasing w
of harmonics has been seen as most of the information is typically below 5
cycles/in., well below the highest frequencies resolvable .
511

6.6 Evaluation of Time Series Images Using FFT Methods

The raw data have been patch equalized and recomposed into two different forms: averaged
images and time series images. The time series images were analyzed using FFT methods
and will be discussed here. Averaged images will be discussed in the next subsection. To
begin, a typical CD scanline was chosen and an FFT was made of this scanline. Since the
time series data contain turbulent information, a single scanline may not show a frequency
of a large scale streak. To examine the range of turbulent frequencies and expose any non-

50
 turbulent effects an FFT was taken, accumulated, and averaged. This exposes dominant
frequencies and shows that turbulent frequencies cover a continuous range.

6.7 Evaluation of Averaged Images Using FFT Methods

The average images were also analyzed using FFT methods. Similarly, a typical CD scanline
.
was chosen and an FFT was made of this scanline. However, since the time series data con-
tain little turbulent information, a single scanline in this case distinctively shows frequencies
of any large-scale streaks. The averaged overhead views often display the evolution of mean
features down the viewed portion of the table; it is useful then to examine the change in
the FFT for each CD scanline down the table. The top 4 peaks are recorded for the FFT
of each scanline and plotted against scanline number (position down the table). Any shift
in dominant frequencies down the table can be illustrated. Similarly, if the magnitudes of
the major peaks are recorded then the relative significance of the peaks can be seen. In this
way the continuity of the streaks down the table can be illustrated.

To find the peaks of an FFT and to rank them by magnitude, the following algorithm was
developed. First, an array of zeros is created such that the length of the array is the same
as the number of points in the FFT. This is the ‘accountant’ array. The maximum and
minimum values of the FFT are found. Second, the value of the maximum is replaced with
the value of the minimum such that it will not be found again. At the position of the last
found maximum, a 1 is recorded in the accountant array. The position in the accountant
array is examined. If this position has no l’s as immediate neighbors, then its magnitude
and value are recorded. Third, the next maximum is found and the process continued until
the required number of ranked peaks are recorded or the next found maximum equals the
minimum. The recorded peaks will be ordered from highest to lowest magnitude.

6.8 Conclusions

Much can be learnt by digital imaging of the forming section and subsequent analysis and
manipulation of the images. The raw images can be enhanced to better capture the details
of the forming jet, the jet impingement, and the forming layer dynamics along the forming

51
table. Spectral analysis of the images and timeline imaging can be used to evaluate the
performance of the headbox and the quality of the forming jet. In fact, detailed imaging
can be used to obtain a qualitative level of surface turbulence (RMS) in the forming jet,
in addition to the mean flow structure. Even though here we have concentrated on light
reflection from the surface of the forming jet, the same approach can be used for light
transmission imaging through the forming to evaluate the uniformity of fiber dispersion
and the degree and rate of floe formation on the forming table.

In the next section, we use this high-speed digital imaging technique to quantitatively
evaluate the forming jets from four different commercial headboxes.

52
7 Quantitative Evaluation of the Forming Jet Delivered
from Four Different Headboxes

7.1 Operational Parameters of Four Hydraulic Headboxes

Considering that there are a variety of headbox and forming sections, we selected four
machines which cover the different design features that are most popular in commercial
headboxes. Also, we have chosen two of the headboxes to have similar designs for com-
parison of the imaging method. These machines all have Fourdrinier forming sections with
relatively easy access for side and top views. During the imaging studies, these commercial
machines were operating under slightly different parameters, but all producing base sheet
coated paper. The four Fourdrinier paper machines considered in this study operate under
somewhat similar conditions, although being commercial, the operating parameters are not
exactly the same. Table 3 summarizes these parameters. Note that all of the data have
been provided by the mill.

Table 3: List of parameters for the four paper machines

Machine Machine Headbox Wire Rush Consistency Slice
Width Flow Rate Speed I Drag I Opening
( in. > (gpm) (fpm) (in. >
A 332 18,922 2917 0.997 0.61 0.45(2)
B 234 (1) 2640 0.992 0.65 0.55(2)
C 265 17,808 3000 (3) 0.90 0.55(4)
D 302(5) (6) 0 1.005(S) (9) 0.49(10)
Notes:
(1) not reported
(2) These values may not be very accurate(MO%)
(3) not reported
(4) given as: uert. = 0.55”) hori. = -0.46”
(5) given as: wire = 302”, slice = 294”) reel = 272”
(6) range given as: 16,000 - 24,000
(7) range given as: 2,000 - 3,400O
(8) given as: J/W = 1.005
(9) range given as: 0.40 - 0.75
(10) range given as: 0.4” - 0.6”; typical values listed in the table.

Some of the parameters are given as a range within an estimate of the operating value. In
these cases, the note below the table specifies the range. Some parameters given by the mill

53
 were not accurately measured.

The dynamics of the streaks on the forming table have been captured using a digital video
system. Off-line image processing analysis and signal processing analysis and signal analysis
techniques were employed to quantify the scale and regularity of the streaks, as outlined
by Aidun and Ferrier [ll] and in Appendix B. To facilitate the comparison of one paper-
-
making machine to another, the data are plotted with similar scales and the image data
are printed with overlaid grids. FFTs for the averaged images and the time- series images
are presented and discussed. Furthermore, the FFT data may be directly compared across
machines by plotting the ratio of standard deviation to the average value.

7.2 Results and Analysis

In the imaging of the forming jet and the forming layer, we have focussed on the fluid
mechanics after the impingement line region. Despite our attempts, in some cases we were
unable to access and record the images at the same relative distance from the slice and
the impingement line. This is due to the differences in the headbox slice design and the
location of the catwalk on the forming table. For the sake of accurate documentation, we
have made careful records of the exact location on the forming table that has been imaged.
Figures 37 and 38 show the typical overhead view layout and a map of the locations where
each machine has been imaged. The distance from the imaging region to the slice has been
outlined in Figure 38.

54
 a
I
aNear Time Series Far Time Series Very Far Time Series
: Sample Point Sample Point Sample Point
I

I T I
8
I
I
I
I
a
I
I
I
I
8
I
I
I
8 Close Image View Far Image View
I
I
8
Slice

Figure 37: Typical overhead view layout showing image positions and time series sample
points

Machine Direction (inches)

-10 0 10 20 30 40 50 60 70
1”‘/“‘1”‘~“‘~“‘1”‘~“‘i”‘l

7-l
I
I
I
I
I

Machine A I
I
I
I
I
I
I
I sl

I
I
I
I
I
I
I
I

MachineB :
I
I
I
I
I
I

Machine C

Machine D
D
10
Figure 38: The locations imaged for each machine

55
The range of the forming table covered in the imaging study and the position of the samples
are outlined in Tables 4 and 5. For machines A and B we were unable to image the forming
jet before the impingement line. Therefore, in the comparison of the images, we focus on
the region after the impingement line for all cases.

Table 4: Averaged Images - Range of Table Covered

Start of End of Start of End of
Machine Close Image Close Image Far Image Far Image
(in. from slice) (in. from slice) (in. from slice) (in. from slice)
A 11.5 23.4 28.8 42.3
B 40. 21.5
C -11. 61. 13.0 20.1
D 00. 50. 90. 17.4
6

Table 5: Time Series Images - Positions of Samples

Impingement
Machine Near Far Very Far Line
(in. from slice) (in. from slice) (in. from slice) (in. from slice)
A 12.8 20.9 35.5 40.
B 70. 15.5 40.
C 05. 35. 16.5 48.
D 05. 45. 13.5 55.

Typical enhanced images of the forming jet are shown in Figs. 39a-d,“for machines A to
D, respectively. These images are taken at 0.06-millisecond exposure time at 1000 images
per second. The forming jet is illuminated by 1000 W Tungsten light from the side of the
forming table. The streaks in the forming jet reflect the incident light at varying intensites
to the camera depending on the nonuniform curvature of the liquid surface. In principle,
from these images, one can calculate the amplitude of the streaks or waves from the intensity
of the recorded light. However, this approach requires a more precise control on the light
intensity, a requirement not easily achieved in the mill during a commercial run without
interference with the operation and production of the machine.

56
 MD

Figure 39: (a) Single frame image of the forming jet after impingement on the forming wire
of machine A. The gridlines show a l-inch length scale

MD

Figure 39: (b) Single frame image of the forming jet after impingement on the forming wire
of machine B. The gridlines show a l-inch length scale

Visual inspection of the images in Figures 39(a-d) provides some view of the forming jet
characteristics. One can immediately see slight nonuniformities in CD as well as MD with
more or less in all of the forming layers. Jet roughness and small structures in the forming
jet result from turbulent flow inside the headbox as well as the mean secondary flows. The
scale of the surface roughness seen on the forming layer in Figures 39(a-d) is related to the

57
turbulent eddy size and strength. There are some similarities between images for machines
A and B. Also, machines C and D, having very similar features show similar forming layer
patterns after the impingement line. The impingement angles in these machines are differ-
ent, therefore, the degree of stock jump after impingement would be different. However, we
are more concerned about the surface pattern and regularity of flow on the forming table.

Figure 39: (c) Single frame image of the forming jet after impingement on the forming wire
of machine C. The gridlines show a l-inch length scale

58
Figure 39: (d) Single frame image of the forming jet after impingement on the forming wire
of machine D. The gridlines show a l-inch length scale

Comparison of the A and B forming layers, shown in Figures 39(a and b), with C and D
forming layers, shown in Figures 39(c and d) reveals a fundamental difference in the pattern.
The A and B layers show jagged streaky pattern where C and D layers show a relatively
straight streak in the machine direction. The reason for this difference, we believe, is a
fundamental difference between the A & B headboxes and those on machines C & D. The
headbox on machine A has one set of tube bank, the tube column is inclined, and there
are vanes or sheets in between the rows of tubes. Headbox B also has one tube bank with
a regular vertical column and vanes or sheets between the tube rows. Machines C and D
have almost identical headboxes. These headboxes have a set of manifold tubes, a ‘stilling
chamber’ and 3 rows of staggered tubes in the second bank of tubes prior to the converging
nozzle. There are no vanes or sheets inside the converging nozzle of the C and D headboxes.

As we have outlined elsewhere [lo], the interaction of the jets from the tubes inside the
headbox generates mean secondary flows which result in a regular streaky pattern on the
forming table. The jets from the tubes are squeezed onto neighboring jets, below and above,
generating mean secondary flows in the form of counter-rotating vortices. With headboxes
A and B, the sheets delay the interaction of the jets in this manner.

We take the average of the flow over 0.1 sec. (i.e., 100 frames) and present the results in
Figures 40(a-d) for machines A - D, respectively. We see deterministic streak patterns in

59
all the cases, although case B shows much noise (random variations) in the averaged image.
The grid lines superimposed on the images represent a l-in. length scale. To evaluate the
degree of regularity of the streaks, we cannot rely on the visual inspection of these images
or the averaged images. For a quantitative method of evaluation, we compute the average
and the standard deviation of the FFT of each CD scanline of the image averaged over 100
frames (i.e., the image presented in Figures 4O(a-c) for each machine, and averaged over
300 frames for Figure 40d).

-.r

MD

Figure 40: (b) Image averaged over 100 frames (0.1 set) of the post-impingement forming
layer on the forming wire of machine B. The gridlines show a l-inch length scale

60
 I
MD

Figure 40: (c) Image averaged over 100 frames (0.1 set) of the post-impingement forming
layer on the forming wire of machine C. The gridlines show a l-inch length scale

The more regular and persistent the streaks are, the smaller the standard deviation of the
FFT divided by the mean value of the FFT will be. For a rigorous evaluation, we need
to make the comparison at equal headbox and forming section parameters: that is, equal
slice opening, machine speed, average jet velocity, jet impingement angle, and the ratio of

61
the average jet velocity to the wire speed. However, we could not find two machines that
operated with exactly the same values of these parameters. the four machines considered
in this study were the closest we could get to the ideal case.

With the understanding that the difference in the operational parameters, as listed in
Table 3, between these machines could influence the quantitative results, we present the
comparison of the FFT standard deviation/averaged FFT for 100 and 500 frames (i.e., 0.1
and 0.5 set) in Figures 41 and 42, respectively. We have included the 500-frame averaged
FFT result to demonstrate that the number of frames considered in this evaluation, as long
as at least 100 frames are considered, does not change the conclusions. The frequency range
of interest is between 0.5 - 2 in.- l, that is 2 to 0.5 in. length scale. The dominant streaks
in machines C and D are in the range of 0.5 to 1 in. length scale, where for machines A and
B, the dominant length scale is 1 to 2-in. In the 0.5 to 2 in. range, the standard deviation
of machines C and D are lower than the standard deviation of images taken from machines
A and B. This shows more regularity in streaks for machines C and D.

3 J 0 J 0 7

muefq IMti=l
. .

Figure 41: The FFT standard deviation/average for all scanlines after the jet impingement
line - 100 frames

62
Figure 42: The FFT standard deviation/average for all scanlines after the jet impingement
line - 500 frames

In general, the results show that the images of the streaks for machines A and B are less
regular than the streaks generated in machines C and D about the l-in. length scale. About
the 2-in. length scale, machines C and D show the most regular streak pattern. This method
of comparison does not make an evaluation of the amplitude of the streaks; it is just the
persistence or regularity of the streaks that is measured. The amplitude of the streaks is
strongly influenced by the jet impingement angle and the jet velocity.

The question is why do headboxes A and B, in general, deliver jets with less regular streaks
than headboxes C and D in the 0.5 to l-in. range. Although we have not rigorously analyzed
the difference between these headboxes, some ideas can be considered. The main difference
between the converging nozzle of these headboxes is that both A and B have sheets extending
from the tube outlet to a few inches from the slice separating the jets from each row of
tubes, where headboxes C and D have no sheets. Therefore, the jets from adjacent rows
in headboxes C and D can interact immediately starting from the outlet of the tubes. One
may conjecture two mechanisms based on this difference in the converging nozzle geometry,
as outlined below:

One, the sheets inside the headbox provide ‘additional turbulence’, as reported. The higher

63
turbulence level (we assume higher turbulence level referred to in the industry is higher value
for turbulent kinetic energy defined as the trace of the Reynolds Stress Tensor) results in
damping or flattening the velocity profile from the tube jets and provide a more random
jet delivered from the slice. We doubt that this is the mechanism since our measurements
of the turbulent kinetic energy in a headbox very similar to headbox A shows an increase
in the turbulent kinetic energy (most likely due to an increase in the turbulent eddy scale)
by removing the sheets from the headbox [see 95.11.

The second mechanism is based on isolating the jets from each row of tubes and therefore,
preventing the interaction of the tube jets with their neighbors in the rows above and below.
Computational analysis of the flow in the converging zone of the nozzle shows that as the
jets from two rows interact, a mean secondary flow of counter-rotating vortices form for
each pair of jets [lO,ll]. These vortices generate streaks at the slice and on the forming
table. When a solid surface isolates the rows of jets, the early interaction of the jets with
the adjacent jets from the rows above and below are delayed until further downstream in
the converging nozzle. This changes the mechanism of interaction among adjacent jets and,
therefore, alters the pattern of the streaks on the forming table. The relatively straight
streaks are replaced with considerably jagged streaks. However, the jagged streaks are
still formed along the machine direction, as the averaged image in Figures 40(a and b)
show. At this time, the exact mechanism for the formation of streaks in headboxes with
vanes is not clear to the author. We are now engaged in detailed computational analysis
of a headbox with sheets in the converging zone. The results which will appear in future
publications would provide some insight into the fluid mechanics of the forming jet by this
class of headboxes.

To further investigate the characteristics of the forming jet from the four headboxes consid-
ered in this study, let us analyze the timeline images of the forming sections as presented in
Figures 43(a-d) for machines A to D, respectively. An explanation of the timeline imaging
has been given in 56 and Appendix B.

64
 OmS looms

Figure 43: (a) lOO-ms time-series image of the forming section of machine A at 15” from
the slice. The blue gridline shows a l-inch length scale and the sections bounded by the red
lines represent a 1OOms increment

OlllS 1OOms

Figure 43: (b) lOO-ms time series image of the forming section of machine B at 15” from
the slice. The blue gridline shows a l-inch length scale and the sections bounded by the red
lines represent a 1OOms increment -.w

OmS looms

Figure 43: (c) lOO-ms time-series image of the forming section of machine C at 15” from
the slice. The blue gridline shows a l-inch length scale and the sections bounded by the red
lines represent a 1OOms increment

65
Figure 43: (d) lOO-ms time-series image of the forming section of machine D at 15” from
the slice. The blue gridline shows a l-inch length scale and the sections bounded by the red
lines represent a 1OOms increment

These images show the dynamics of the flow pattern in the forming jet. The images are
taken at a section of the forming jet about 15 inches away from the slice. The impingement
line in all cases is about 4 to 6 inches from the slice. Machines A and B, in general, show a
different surface flow pattern than machines C and D. The streaks in machines C and D are
more elongated in the machine direction. We hesitate to make any conclusions from visual
inspection of these images.

In general, one assumes that the smaller scale eddies in the forming layer provide a more
effective mechanism for preventing fiber flocculation. Some previous studies have addressed
this issue [48-501.

In the next section we study an experimental headbox to understand the mechanism of

turbulence in CD, MD, and ZD.

66
 8 Investigations of Experimental Headboxes

8.1 Effect of Contraction Ratio on Fiber Orientation

Headbox guide sheets and headbox contraction ratio are among the structural parameters
that greatly influence the orientation of fibers in the finished paper. This has been demon-
- strated by Mats Ullmar and Bo Norman [15]. Contraction Ratio (CR), as defined by them,
is the ratio of headbox nozzle inlet area to nozzle outlet area. They constructed experi-
mental equipment to study the fiber orientation in a headbox nozzle by mounting a CCD
camera over the outlet together with a stroboscope. Image analysis has been used to eval-
uate fiber orientation from the recorded images. The fibers used in this investigation were
polymer fibers of mean length and diameter of 3 mm and 45 ,wm, respectively. Experiments
have been conducted for different contraction ratios by altering the slice opening at the
nozzle end. They also conducted experiments for various flow velocities for each of these
contraction ratios. Their results indicate that the effect of CR is more significant in fiber
orientation than in flow velocity. The fibers have been found to be more strongly oriented
in the machine direction for higher CR’s. This is also supported by an earlier paper by
the same group [22] which showed that an increase in the CR results in an increase of the
MD/CD tensile stiffness ratio of the paper due to more fibers oriented in the MD.

8.2 Effects of Guide Sheets on Headbox Hydrodynamics

To complement the above study, we present the results of two sets of experiments done by
our group in which we measured velocity profiles and turbulence intensities using LDV [53]
(also see Appendix A) at the jet exit and in the headbox, with and without the guide
sheets in place. Our results show that the guide sheets effectively reduce both the average
cross-flow velocities and normal Reynolds stresses in CD, MD, and ZD, and provide more
uniform velocity profiles than the headboxes without them. Both sets of experiments were
done with and without the guide sheets in place. All measurements are at a flow rate of

50.5L/s k3 % (800 gpm) with a slice opening of 12.7 mm (0.5 in.). The headbox width
is 406 mm (16 in.); so the average jet velocity at the exit is 9.8 m/s. The headloss in the
headbox is somewhat higher with the sheets than without.

67
8.3 Measurements at the Jet Exit

Vertical profiles of the variation of the horizontal streamwise, u, and cross-flow velocity, v,
at the jet exit were made at five horizontal locations as shown in Figure 44. The horizontal
locations were at y = 2, 4, 8, 12, and 14 in. The vertical coordinate, x, of the point
of measurement is the distance from the Lucite attached to the lower surface of the slice.
The laser probe is placed below the Lucite looking upwards as shown in Fig. 44. Three
measurements were recorded at each location. Data were obtained both with the separating
plastic sheets inside the headbox and with them removed.

Da
Ins

domain signal Fiber optic

nrorR,ssnr cable

Figure 44: Schematic depiction of LDV experimental configuration

The mean streamwise and cross-flow velocities are shown in Figures 45 and 46. Typical
streamwise velocities are greater than the average velocity at the jet exit due to the vena-
contracta effect. It can be seen that the streamwise velocity is not exactly uniform. The
variation of velocity over the cross section is about 3% of the mean value. Without the
sheets, the variation is almost 10%. The results also imply some asymmetry of the flow
about the jet exit centerline. The flow closer to the headbox inlet (y = 14 in.) is less
uniform than farther away (y = 2 in.).

The cross-flow velocity is very small, essentially zero, with the sheets installed, implying
essentially straight streamlines parallel to the headbox axis with negligible secondary cir-
culations. The magnitudes are substantially higher without the sheets. For this case, the

68
velocities are generally negative to the left of the headbox centerline, and positive to the
right. This implies diverging flow to the headbox walls. At the far wall (y = 14 in.) es-
pecially, the velocities are negative near the free surface and positive near the middle and
lower boundary. This implies a rather strong secondary circulation within the jet, which is
sunDressed bv the sheets.

“.CN , “.“”

free surf8ce free surface

m mm r 0.3- 000
om -8 1 : m m
m I

- 0.15 -
0m mm 00
q a I m
m Ip 3 : rm 0
m 4 0.05- mm
plexiglass r3 - plexiglass
. . . . . . . ..1....1......-.......1........- Ol.........,..............,..........-.-.
8 8.5 9 9.5 10 10.5 11 116 12 8 8.5 9 9.5 10 10.5 11 11.5 12
Slrormwlu voloclty In ml8 Streunwiw voloclty in m/s

$ :
t 0.1 m n 8
23 1 00 I
; 0.05 I s 0.05 -
J : plexiglass
plexiglass

Stroamwfaovolodtyfnmh Stroamwiro volodty In ml@

yZl4’
0.85,
a free surface
2E 0.3 0 m I
000 I
4 025 I 00 888

plexiglasa

9.5 10 10.5
Streamwise volodty In ml.

Figure 45: Mean streamwise velocity, u, at the jet exit with and without sheets

69
 free surface sufface
0
0 0
q m
m
0 0
om I m 0
m 0 I8 0 00
mo 00
mo 0
0 0 0
0 0 -4 ,
om _ N’
plexiglass
000 .
. .+o
-0.3 -0.2 -0.1 0 0.1 0.2 0.3
Crou4low velocity in mh Crosr4low velocity In m/8

1 free surface

0 0
0 0 0
0 00
0 00
[PO
00
m

-0.2 0 0.2 0.4 0.6 0.8 1

Cro884low vdoclty In mh Ctos*-flow velocity In m/a

yn14’
0.4
free surface
E
cI13 . E
Im -
ml
*m C[ OS3 i
m 000
m- mo 0.2 ii
I’ 00 :
m a g
cI- mu 0 m 0.1 8
* 000 3
u
om
plexiglass J
0 .oo
. . . 0 5Y
-0.4 -0.2 0 0.2 0.4 0.6 0.8
Croos-flow velocity In mh

Figure 46: Mean crossflow velocity, v, at the jet exit, with and without sheets

The streamwise and crossflow variations of turbulence intensity (proportional to the normal
Reynolds stresses) are shown in Figures 47 and 48. They are shown as the magnitude of
the root mean square (rms) value of the velocity fluctuations from the local mean value.

70
 y&2’
0.35 9 0.35
free surface free surface
fj8 0.3 -* mm 0 au m 0 00
I m
= - =
+zj) 0.25
0.2 -.-i mmmI mm
q om 0: mK]
g3 0.25
0.2 I : 000
m- Fl
2 - P 00 0
E 0.15 - Rrn
I8 0 MO Cl 0 5 0.15 -, I 00 0
e : I 00 0 f . m 00 0
g 0.1; mm 00 g 0.1. I mo
6 .
d
i!l -: m 00 0 z . mm 000

0.05 8I : [Link] : mm
J : plexiglass
plexiglass .
oi.... I . . . . I., * *. . . -.
1

RMS in m/8 RMS In m/8

0.35
free surface free surface
g 0.3 1 m n 00 0 f .
031 mm m 0
z : mw m 0 2 : mm 0m
s 0.25 - I mo = 0.25 - m 00 0

a : I om i : I om
9 0.2 - mm m 0 z 0.2 - I m

2i 0.151* I
mm am 0 E]
e , mm cm

6
E 0.1 mm
m
0 a
0 00
j O*’ :
n
II
mm 0
m
0
zu 2 .
0.05 I - 0.05 - I8
IJ plexiglass rJ : plexiglass
i
04 . . . . . . . . . . . . . . . -. -. o$.........,..-.,....
1 0 1 1.s 2
RMS In m/s RMS in m/s

I free surface 000

I 0 a
0.25 0 0
9s I’ m 0
$f 0.2 I m0
S; 0.15 I 8 03
B m q pi!tzij
= i

6 m m0
s
E 0.05 I
plexiglass
i
0.5 1
RMS in m/s

Figure 47: rms streamwise velocity, u’, at the jet exit, with and without sheets

71
 free surface

q m
cm
on 00 0
1130 OW
om w
0 00 mo
00
mo
00
00 0
03
PO
plexiglass m
o!..............,.........,....I
0 0.2 0.4 0.6 0.8 1 1.2
RMS In m/s RMS in ml8

q IZI
0
0m
000 0

00 m
q m m
[Ip mo
m 0 PO
0 00 0 m

0 0 0 mo

0.6
RMS In m/s RMS In m/e

I free surface
F]

m
m

$0.1 I 00
b om
5 00 0
hi- plexiglass mo
i
o!. -.-1,-*. I.‘.... . . . ..‘......‘I
0 0.2 0.4 0.6 0.6 1 1.2
RMS In m/s

Figure 48: rms crossflow velocity, v’, at the jet exit, with and without sheets

The rms of the streamwise velocity is smaller and more uniform with the sheets than
without. Typical streamwise values with the sheets are about 0.3 - 0.6 m/s, corresponding
to relative turbulence intensities of about 3 - 6%. The crosswise rms values are smaller,
indicating that the turbulence is not isotropic. Without the sheets the rms values are higher
and less uniform. Streamwise turbulence intensities are typically 10% greater and are higher
near the lower boundary and at the free surface.

72
 8.4 Measurements in the Headbox

Measurements of the streamwise velocity, U, and vertical velocity, w, were made inside the
headbox along the longitudinal axis. The measurements were made at a distance y = 3”
from the Lucite wall along the centerline of the headbox. The longitudinal variations of the
mean streamwise and vertical velocities and their rms values are shown in Figures 49 and
.
50 respectively.

Horizontal velocity along headbox centerline

tube bank slice

1.~..,~.~~,...,,.,,.,...,,,...~0
-30 -25 -20 -15 -10 -5 0
x, distance from slice in inches

Vertical velocity along headbox centerline

t 1.8
1.6
1.4 w ‘*
2
1.2 ,c
,l .z
B
r 0.8 g
: 0.6 $
I 0.4 g
tube bank slice ;
0.2
0
-0.2
-25 -20 -15 -10 -5 0
x, distance from slice in inches

Figure 49: Mean streamwise velocity, u, and vertical velocity, v, along headbox centerline

73
 RMS of horizontal velocity along headbox centerline

0.8

tube bank slice :

x, distance from slice in inches

RMS of vertical velocity along headbox centerline

Ll 1
WI
0

I0 u - 0.8
cl 1I with sheets 1
Iq without sheets I
0.6 g
E
slice
F 0.4 ii

tube bank

-30 -25 -20 -15 -10 -5 0

x, distance from slice in inches

Figure 50: rms streamwise velocity, u’, and vertical velocity, v’, along headbox centerline

The streamwise velocity increases along the headbox due to the tapering cross section.
Without the sheets, the streamwise velocity is larger near the tube and smaller near the
slice than with the sheets. With the sheets, the velocity increases more uniformly along
the headbox. The vertical velocities with the sheets are very small. Without the sheets,
the vertical velocity is quite strong, up to about 1.8 m/s near the inlet tube bank. This
velocity is positive, i.e, upwards. It is also positive near the jet exit but with a much smaller
magnitude, about 0.1 m/s.

The streamwise rms intensities are smaller with the sheets installed than without them as
shown in Figure 50. This is particularly true near to the inlet tube bank, where typical rms

74
values are about twice as large without the sheets. The difference decreases with distance
along the headbox until the jet exit, where the rms values are very close. The relative
turbulence decreases rapidly with distance along the headbox. With the sheets, it is from
about 40% near the tube bank to about 3% near the jet exit. The high values near the inlet
tubes are presumably due to the shear induced by individual jets issuing from the tubes.
This turbulence decays with distance until by the jet exit it has been replaced by turbulence
due to shear at the top and bottom boundaries of the headbox, with typical values around
3!70.

8.5 Summary

1. The streamwise velocity of the flow leaving the headbox at the exit is more uniform
with the sheets.

2. The cross-flow velocities are very small with the sheets. This indicates that secondary
circulations are virtually eliminated by the sheets.

3. Cross-flow velocities without the sheets can be quite substantial and spatially variable.
This indicates’ the presence of secondary circulations.

4. The magnitude of the turbulent fluctuations(expressed as their rms values) are smaller
and more spatially uniform over the jet exit with the sheets than without. Relative
turbulence intensities over the jet exit are typically about 3 - 6% with
%” the sheets and
larger without.

5. The streamwise velocity inside the headbox increases more uniformly with the sheets
than without them. Vertical velocities in the headbox are essentially eliminated by
the sheets but can be quite large without them.

6. With the sheets, the relative turbulence intensity decreases rapidly with distance
inside the headbox. It is about 40% near the inlet tubes and about 3% near the jet
exit. This is presumably due to the decay of the turbulence intensity associated with
the multiple inlet jets.

75
In the next section we examine five different types of tubes from various tube banks to
study the effect of spiral fins, placed inside the tubes, on the average fluid velocities and
turbulent intensities at their outlets.

76
9 Conclusions

Three of the major quality concerns in today’s paper mills are the uniformity of basis weight,
fiber network structure (formation) and fiber orientation. Variation of these fundamental
characteristics influence the physical properties of the paper manufactured for printing,
packaging, and other applications. Two of the three characteristics mentioned above, the
fiber network structure and the basis weight uniformity, are affected by (a) the secondary
flow structure of the jet before the impingement line, (b) amplification of secondary flow
instabilities at the impingement line, and (c) the evolution of the layer along the forming
section. For this reason, the headbox and the forming section are considered to be the most
critical components in the paper making machine. With an increasing trend towards faster
and more efficient paper machines and higher standards for the quality and the strength
of the paper, a fundamental understanding of the wet-end dynamics, such as turbulence,
flocculation, stock jet velocity, and impingement angle has become an important area of
research (see 82).

In recent years, fine-scale velocity measurements, using the high speed digital imaging and
LDV, in conjunction with numerical modeling, have considerably improved our understand-
ing of the complex hydrodynamics of the headbox. To summarize, the turbulent flow from
the headbox nozzle comprises complex stream-wise streaks in MD, a nonuniform profile in
CD, and a large number of swirling eddies in ZD (which more or less form the cross sections
of stream-wise streaks in MD). The distribution of stream-wise streaksin MD, usually, is
quasi-periodic and is a manifestation of large-scale structures, superimposed on fine-scale
turbulence. These complex patterns when combined with the free surface instabilities of the
forming table produce wet streaks at the dry line. Larger and wider wet streaks degrade
the homogeneity of the paper web resulting in a poor quality paper and cause variations in
physical property ranging from 0.5 to 4.0 in. in CD. In stratified headboxes, the swirling
motion in ZD may also cause mixing among the different layers and thus can jeopardize the
grade of the paper (see $3).

The headbox geometry and dimension, along with the flow rate and fluid properties set
the level of turbulence in the forming jet. The dimension, separation distance, and the

77
pattern or arrangement of the tubes in the tube bank along with the converging angle of
the headbox and the slice lip determine the magnitude, dimension, and the dynamics of the
small scale secondary flows and, consequently, the characteristics of stream-wise streaks in
the free surface forming jet delivered at the slice (see $6).

The stream-wise streaks form in the forming jet, amplify upon impingement, and evolve
along the forming wire with a characteristic growth and decay rate which determine the
rate of growth and decay of a streak on the forming wire. Analysis of the streak formation
and decay ($4) shows that these rate factors depend on the turbulence level of the flow,
the average thickness of the layer, and the width (or wavelength) of SS. In general, the
very narrow streaks grow and decay very rapidly. The growth rate of the wide streaks is
relatively small. There exist a range of streaks with wave lengths in between, that grow to
form wet streaks at the dry line.

The interaction and dynamics of the tube jets inside the headbox result in small scale CD
oscillation or flapping of stream-wise streaks in the forming jet. This small scale CD motion
results in small scale variations in fiber orientation and fiber network in the sheet. The most
effective mechanism to eliminate this fundamental defect in paper forming is to control the
small scale secondary flows induced by the tube jets. This can be achieved by modifications
to the tubes in the tube bank.

Numerous experiments have been conducted on commercial headboxes ($7) using the high-
speed digital imaging technique described in $6. Also experiments have,been conducted on
laboratory scale headboxes with and without guide sheets (58). Current work is focussed
on velocity profile measurements for individual tubes with the presence of fiber floes that
feed into the headbox.

78
APPENDIX A
Laser Doppler Velocimetry

An LDV system measures the velocity of suspended particles in the liquid, which pass
through the interaction point of two laser beams. The backscattered light is interfered
with a reference beam, and the Doppler signal from the LDV optical system is sent by
optical fiber to a photodetector and converted to an electronic analog signal, resulting in a
Doppler burst. The analog signal is connected to the digitizer input of a processor. Once
the signal is digitized, the Doppler Signal Processor (DSP) board performs Doppler Fast
Fourier Transform (DFFT) for each Doppler burst and calculates the Doppler frequency and
the fluid velocity. Without frequency shift, the fringe pattern in the measurement probe is
stationary. A particle crossing the beam generates a Doppler signal, I$. With dual beam
frequency shifting, one beam is shifted at a higher frequency, I?‘H, while the other beam is
shifted at a lower frequency, I?‘L. This causes the optical fringes in the measurement probe
to move in a direction from 8’~ shifted beam to the FL shifted beam at a frequency equal
to the difference of the two frequencies, F,h+ where F,hift = FH - 8’~. Thus, a particle
passing through this probe and moving against the fringe generates a signal with an output

frequency Fmeasured = Fshift + Fo. The particle’s moving direction is defined as positive in
the output of the processor. It means that positive direction in the flow field is defined as
the direction of lower frequency-shifted beam toward higher frequency-shifted beam. The
flow velocity Vreal is calculated by \v

Vreal = FD * -& = (-6neaswed - Fshif t) * DS

where D, is the fringe separation distance.

79
APPENDIX B
Finding Jet’s Velocity Profile Using Cross-Correlation of Digital Images

This method involves cross-correlation of subsequent high-speed digital images to determine

the surface velocity profile of the forming jet. Use of cross-correlation methods to obtain
velocity field is not a new practice. Cross-correlation flow meters [54-551 and Particle Image
Velocimeters (PIV) are well established methods. The PIV, in particular, is a powerful
method for velocity field measurements [56]. This method, however, is limited to transparent
fluids with well-defined seeded particles. The fiber suspension in the headbox and the
forming section are not transparent, and therefore, not accessible by PIV methods.

Here, we describe a simple method which has been used to measure the two-dimensional
surface velocity profile of the forming jet. It takes advantage of the nonuniform patterns on
the surface of the forming jet, recording the positions of the patterned structures using a
high speed digital camera, and using cross-correlation techniques to produce a jet velocity
profile.

In order to analyze the hydrodynamics of the forming table, a high-speed digital imaging
system is used for high resolution imaging of the forming jet as it leaves the headbox. The
individual grey scale images, or frames, taken 1 msec apart, are digital images of 384 x 512
pixels.

Camera orientation is very important to the success of this method. It iYis important that
the camera be positioned normal to the forming table. If the images are from a different
angle, velocity measurements would have to be adjusted and projected on the coordinate
axes to account for the view angle. Furthermore, the camera must be angled so that the
mean flow is in a north/south or east/west direction within the images produced. This
will facilitate the line-by-line cross-correlation method described below. Samples of digital
images are shown in previous publications [lo].

These sample images show the type of data obtained by the digital camera. The camera
picks up the detailed surface features of the forming jet. These features include turbulent
patterns as well as mean flow streaks, crests, and troughs in the surface of the fluid. The

80
movement of these patterns during the 1 msec time lag between frames cross-correlated to
estimate the forming jet surface velocity profile.

In order for this method to be most effective, the images must have a strong contrast with
well-defined features. The original images may or may not have sharp contrast, depending
on the lighting situation, the pattern of the forming jet, and other surrounding conditions.
To account for this, contrast is enhanced using image analysis procedure [ll]. Then two
subsequent frames are compared to see how far the features of the image move during [Link]-
set interval between exposures. This program outputs a scalar value for the streamwise
surface velocity component of the forming jet.

Figure B.l shows two subsequent digital photographs of a section of a forming jet. Notice
that the second image is very similar to the first, with the surface features shifted slightly
to the left. The cross-correlation method takes advantage of this similarity by comparing
the two images. It determines the distance that the features of the second image have been
shifted relative to the first. This distance, the “lag” is then divided by the time elapsed
between exposures to provide an estimate for the jet velocity profile.

Figure Bl: Digital images of the turbulent flow on the forming table, taken 1 msec apart.
The MD is from right to left, and the photograph on the left is taken first. Notice the
patterns due to turbulence superimposed on the mean streak patterns within the images.
Each image represents a 4.7” x 6.3” region on the forming table

81
The features are compared via analysis of the gray scales, or light intensities of the digital
images. The signals compared in this method are the gray scale values of bands of pixels
aligned in the machine direction. Since this is the direction of the main flow, almost all
of the motion occurring during the lag between the two frames occurs in this direction.
Figure B.2 shows an example of these bands.

....
l 9.s

Figure B2: The top two lines are line segments from two digital images similar to those used
in the cross-correlation program. At the bottom, the two images are lined up according to
their respective gray scale values. The displacement of the bottom line is equal to the lag,
or the distance, the line of fluid has moved within the 1-msec time interval between exposures

Gray scale intensities are converted to numerical values, and a FFT one-dimensional cross-
correlation is carried out. If bands running the entire width of the image are used in the
cross-correlation, an average jet velocity profile can be generated. This profile describes
the jet velocity as a function of the location in the cross-machine direction. An average jet
velocity profile generated for the case shown in Figures B.l and B.2 is given in Figure B.3.

82
Figure B3: Average forming jet velocity profile determined by line-by-line cross-correlation

The trendline shown in Figure B.3 is a five-point moving average of the raw data produced
by a cross-correlation routine. This graph shows the type of data that can be obtained
via line-by-line cross-correlation of the entire image. One-dimensional cross-correlation can
also be used to determine the local jet velocity at most points along the machine and cross-
machine directions. This is done by choosing a different interrogation area. Instead of the
entire bands used to determine the average velocity profile, subsets of these bands are used
to determine local jet speeds.

To map the local jet speed profile of the forming jet, interrogation bands 64 pixels long,
aligned along the mean flow direction, are defined. These interrogation bands are centered
on each pixel of the image. Corresponding interrogation bands (bands centered at the
same pixel location) from the two subsequent images are cross-correlated. The location of
the correlation peak specifies the distance that the fluid surrounding the point has moved
between exposures. As the time interval between the two images is known, the data can be
used to calculate the local velocity of the forming jet.

The one-dimensional cross-correlation method, outlined above, does not provide the value
of velocity for every point in the image. Since each interrogation band is 64 pixels long, and
the local velocity at each pixel is determined by centering such a band at the pixel, velocity

83
values for areas to the extreme right and left of the image (with 32 pixels of the edge) are
not available.

With the interrogation band length of only 64 pixels used in this method, small inconsisten-
cies between the two images can cause relatively large errors. Even in the small time frame
of 0 .OOl set, many of the turbulent patterns within the digital frames may be distorted.
For this reason, an estimate of the average jet velocity based on the total flow rate at the
of the slice is used in the program, along with a set of user-defined tolerance values.
The tolerance of the program is left to the discretion of the user. If the correlation peak is
not within the allowed parameters, it is discarded.

84
REFERENCES

1 . Kenneth W. Britt (ed.), Handbook of PuZp and Paper Technology, Van Nostrand
Reinhold Company, 1970.

2 . J. Cox, Headboxes : New Technologies, New Challenges, PIMA ‘s PaperMaker, March

1997.

3 . G.A. Smook, Handbook for Pull, & Paper Technologists, TAPPI PRESS, Atlanta.

4 . S.C. Pentaleo, Modern Headboses - Their Role and Capabilities in the Sheet Forming
Process, Lecture Notes.

5 . B. Norman, Overview of the Physics of Forming, lecture notes.

6 . K.L. Patrick, New Papermaking Technologies Pave Way to Higher Quality and Effi-
ciency, Pulp & Paper, April 1998.

7 . K.L. Patrick, New Papermaking Technologies Spur Competition in P & W Papermar-

kets, Pulp & Paper, May 1998.

8 . K.L. Patrick, Paper Machine Developments Open Door to Higher Quality, Production
Efficiency, PuZp & Paper, June 1998.

9 . C.K. Aidun, Quantitative Evaluation of the Forming Jet Delivered from Four Different
Hydraulic Headboxes Using High-Speed Digital Imaging, EngineeGng & Papermakers
Conference, 1997, 11754195.

10 . C.K. Aidun, Hydrodynamics of Streaks on the Forming Table, Tappi Journal, V.80(8),
1997.

11 . C.K. Aidun and A.C. Ferrier, High-Speed Digital Imaging of Paper Forming: A
Method for Quantitative Evaluation of Paper Forming Hydrodynamics, Engineering
& Papermakers Conference, 1997, 883-887.

12 . J. Zhu, Laser Doppler Velocimetry for Flow Measurements in Pulp and Paper Re-
search, TAPPI Engineering Conference, 1996.
 13 . C.K. Aidun and A.E. Kovacs, Hydrodynamics of the Forming Section: the Origin of
Nonuniform Fiber Orientation, Tappi Journal, V.78(11), 1997.

14 . C.K. Aidun, Formation Hydrodynamics and Effects on Paper Physics, International

Paper Physics Conference, 1995, 1-3.

- 15 . M. Ullmar and B. Norman, Observation of Fiber Orientation in a Headbox Nozzle at

Low Consistency, Engineering & Papermakers Conference, 1997, 865-873.

16 . A.C. Li, E.B. Neil1 and T.D. Rogers, Jet Trajectory from a Converging Nozzle, Engi-
neering & Papermakers Conference, 1997, 1197-1204.

17 . M.S. Hall and T.G. Jackson, Small-Scale Variations in CD Strips Observed Using
Ultrasonic Velocity Measurements, Proceedings of International CD Symposium, XIV
IMEKO World Congress, Tampere, Finland, 1997, 174-179.

18 P.G. Drazin and W.H. Reid, Hydrodynamics Instabilities, Cambridge University Press,
1987, Cambridge, England.

19 . C. Yih and A.C. Spengos, TAPPI, 42(5): 398(1959).

20 . C. Yih and S.P. Lin, TAPPI, 47(2): 88(1964).

21 . J. Shands, J. PuZp Paper Sci., 17(3): J92(1991).

22 . L.D. Soderberg and P.H. Alfredsson, Engineering & Papermaker Conference, 1997,
1205-1222.

23 . I.L. Claeys, Hydrodynamic Transport Properties of Suspensions of non- Brownian

Prolate Spheroids, Ph.D. Thesis, California Institute of Technology, 1991.

24 . I.M. Hutten, TAPPI 1994 Nonwovens Conference Proceedings, TAPPI Press, Atlanta,
p.241.

25 . C.K. Aidun, N.G. Triantafillopoulos and J.D. Benson, Phys. Fluid, A.3, p.2081, 1991.

26 . J.A. Liggett, Fluid Mechanics, 2nd edn., McGraw-Hill, New York, 1994, p.256.

86
27. A.O. Demuren and W. Rodi, J. Fluid Mech., 140:189, 1984.

28. C.G. Speziale, Ann. Rev. Fluid Mech., 23:107, 1991.

29. A. Huser and S. Biringer, J. Fluid Mech., 257:65, 1993.

30. S. Gavrilakis, [Link] Mech., 244:101, 1992.

31. B.F. Gessner and J.B. Jones, J. Fluid Mech., 23:689, 1965.

32. A. Melling and J.H. Whitelaw, J. Fluid Mech., 78:289, 1976.

33. T. Wei and W.W. Willmarth, J. Fluid Mech., 204~57,1989.

34. C.G. Speziale, J. Fluid Mech., 178:459, 1987.

35. K. Hanzalic and B.E. Launder, J. Fluid Mech., 52:609, 1972.

36. M. Colombini, J. Fluid Mech., 254:701, 1993.

37. I. Nezu, H. Nakagawa and A. Tominaga, Turbulent Shear Flows IV, Springer Verlag,
New York, 1985, pp.246-260.

38. D. Naot and W. Rodi, ASCE 108(HY8): 948, 1982.

39. J.O. Hinze, Turbulence, McGraw-Hill, New-York, 1975.

40. J.F. McKibben and C.K. Aidun, J. Comput. Phys., in press. %v

41. J.F. McKibben and C.K. Aidun, Tappi J., 77(5):143, 1994,in press.

42. J. Kim and P. Moin, J. Comput. Phys., 59:308, 1984.

43. A. Huser and S. Biringer, Int. J. Num. Methods in Fluids, 14:1087, 1992.

44. M.S. Engelman, FIDAP: Fluid Dynamics Analysis Puckage, Fluid Dynamics Interna-
tional, Evanston, 1992.

45. C.G. Speziale, Transactions of the ASME, 108:118, 1986.

46. H.J. Perkins, J. Fluid Mech., 44:721, 1970.

87
47. C.K. Aidun, A Fundamental Opportunity to Improve Paper Forming, Tappi J., 79(6),
55, 1996.

48. H. Hauptmann, R. Vyse and J. Mardon, The Wake Effects as Applied to Modern
Hydraulic Headboxes, Part I and II, Pulp and Paper Canada, 91(9 and lo), T357 and
T369, 1990.

49. P.E. Wrist, Dynamics of Sheet Formation in the Fourdrinier Machine, in Formation
of Structure of Paper - 2, Trans. Symp., held at Oxford, London, Sept. 1961 (F.
Bolam ed.), Tech. Ser. Brit. Paper and Board Makers Assoc., 839, 1962.

50. L. Lindquist, Structure of the Flow from Paper Machine Headboxes: A Model Study,
Licentiate Thesis, Lulea University of Technology, Sweden.

51. Bloomfield, Fourier Analysis of Time Series: An Introduction, 1976, John Wiley &
Sons, Inc.

52. W.H. Press, S.A. Teukolsky, W.T. Vetterling and B.P. Flannery, Numerical Recipes

in C, Second Edition(ch. 12 & 13), Cambridge University Press, 1992.

53. C.K. Aidun, P.J. McKay, and X.L. Ye, Fundamentals of Headbox and Forming Hy-
drodynamics, Status Report for Project FO05, IPST, 1998.

54. C.K. Aidun, X.L. Ye, and P.J. McKay, Headbox and Forming Hydrodynamics, Status
5v
Report for Project FO05, IPST, 1997.

55. M.S. Beck, Correlation in Instruments: Cross-Correlation Flowmeters, J. Phys. E.

Sci. Instrum., 14, 1981, pp. 7-19.

56. F. Frigerio and D.P. Hart, Velocity Field Measurements of a Confined Swirling Flow
Using Digital Particle Image Velocimetry Cinematography, 1997 ASME Fluids Engi-
neering Division Summer Meeting, June 1997.

88