may 2004 journal - tappi
TRANSCRIPT
MMAAYY 22000044
WWWWWW..TTAAPPPPII..OORRGG JOURNALThe effect of slot
shape on pressure
screen performance
3Near-infrared
multivariate image
regression
8Online Exclusive
Concentric mixing
of hardwood pulp
and water
14Advanced wet-end
system with CMC
15Coating layer
formation in
substrates
20H
2SO
4emissions
from combination
boilers
25
2 TTAAPPPPII JJOOUURRNNAALL MAY 2004
TABLE OF CONTENTS MAY 2004 VOL. 3, NO. 5
EDITORIAL BOARD
SENIOR TECHNICAL EDITORS
William S. Fuller,Weyerhaeuser Co.,[email protected], +1 253 279-0250
Jere W. Crouse, JWC Consulting,[email protected], +1 608 362-4485
TECHNICAL EDITORS
Raimo J.Alén, University of Jyvaskyla,[email protected], +358-14-2602562
Terry L. Bliss, Engineering Consultant,[email protected], +1 770 579-8820
Peter W. Hart,Westvaco Corp.,[email protected], +1 409 276-3465
Bernard Kessler, Ponderay Newsprint Co.,[email protected], +1 509-445-2189
Reid A. Miner, [email protected], +1 919 941-6407
Richard J. Spangenberg, Retired,[email protected], +1 206 527-5490
Nick G.Triantafillopoulos, OMNOVA Solutions,Inc.,[email protected],
+1 330 794-6249
CONSULTING TECHNICAL EDITORS
James W.Atkins,Atkins, Inc.,[email protected], +1 908 806-8689
David J. Bentley, Jr., RBS Technologies, Inc.,[email protected], +1 505 299-6871
Kasy King, Papermaking Process Consulting,LLC, [email protected], +1 920 991-9102
Ted McDermott, McDermott ConsultingServices, [email protected],
+1 847 934-6386
GGrroouupp DDiirreeccttoorr,, PPuubblliisshhiinngg MMaarryy LLyynnnn MMiilllleerr,, [email protected]
+1 770 209-7263PPuubblliisshhiinngg DDiirreeccttoorr,,
MMaarryy BBeetthh CCoorrnneellll, [email protected]+1 770 209-7210
EEddiittoorr,,JJaanniiccee BBoottttiigglliieerrii, [email protected]
+1 847 466-3891AArrtt DDiirreeccttoorr,,
AAllaann SSccootttt, [email protected]+1 770 209-7557
PPuubblliisshhiinngg SSppeecciiaalliisstt,,LLiissaa HHiigghhttoowweerr, [email protected]
+1 770 209-7313WWeebb DDooccuummeenntt SSppeecciiaalliisstt,,
CCaarrllaa MM.. JJoonneess, [email protected]+1 770 209-7205
CCoonnssuullttiinngg EEddiittoorr,,DDoonnaalldd GG.. MMeeaaddoowwss, [email protected]
TAPPI, 15 Technology Parkway S., Norcross, GA, 30092, publishes TAPPIJOURNAL monthly. This is Volume 3, Number 5.
AATTTTEENNTTIIOONN PPRROOSSPPEECCTTIIVVEE AAUUTTHHOORRSS:: All papers published are subjectto TAPPI JOURNAL’s peer-review process. Not all papers accepted forreview will be published. Before submitting, check complete author guidelines at http://www.tappi.org/index.asp?pid=20845.
Statements of fact and opinions expressed are those of individualauthors. TAPPI assumes no responsibility for such statements and opinions.TAPPI does not intend such statements and opinions or construe them as asolicitation of or suggestion for any agreed-upon course of conduct or con-certed action of any sort.
Copyright 2003 by TAPPI with all rights reserved. To obtain copyright per-mission to photocopy pages from this publication for internal or personal use, contact Copyright Clearance Center, Inc. (CCC) via theirwebsite at www.copyright.com. If you have questions about the copy-right permission request process, please contact CCC by phone at +1 978750-8400. To obtain copyright permission to use excerpts from this publi-cation in another published work, send you specific request in writing toEditor, TAPPI JOURNAL, 15 Technology Parkway S., Norcross, GA 30092,USA; or by fax to +1 770 209-7400.
Send address changes to TAPPI, P.O. Box 105113, Atlanta, GA 30348, USATelephone +1 770 446-1400, or FAX +1 770 446-6947. www.tappi.org
JOURNALEDITOR’S NOTES JANICE BOTTIGLIERI, Editor; [email protected]
BETTER, STRONGER, FASTER
3 The effect of slot shape on the performance of a pressurescreen SUQIN DONG, MARTHA SALCUDEAN, AND IAN GARTSHORE
8 Using near-infrared multivariate image regression to predictpulp properties MANISH H. BHARATI, JOHN F. MACGREGOR,
AND MARC CHAMPAGNE
14 Online ExclusiveConcentric mixing of hardwood pulp and waterAKLILU T.G. GIORGES, DAVID E. WHITE, AND THEODORE J. HEINDEL
15 Advanced wet-end system with carboxymethyl-celluloseMASASUKE WATANABE, TOMOHISA GONDO, AND OSAMU KITAO
20 Coating layer formation in dry surface treatment of papersubstrates JUHA MAIJALA, KAISA PUTKISTO, AND JOHAN GRÖN
25 Sulfuric acid emissions from combination boilersVIPIN K. VARMA AND ASHOK K. JAIN
Since our relaunch in April 2002, the “new” TAPPI JOURNAL hascontinued to grow and develop. Here is a brief look at theadvances we have made in the two years since reintroducing (byTAPPI membership demand) our publication:
• Now at more than 5,000 strong, our circulation (combined print and electronic)has almost tripled in the past year alone.As a subscription-only publication, ourreaders must directly request TAPPI JOURNAL. Most subscribers are also TAPPImembers, and view the substantial member discounts on subscription prices as animportant member benefit.Thank you for your support!
• We have decreased the backlog of papers awaiting publication and reduced thewaiting time for papers in our system. TAPPI JOURNAL is currently reviewingmore than 40 research papers, more than half of which have been submitted within the past six months.
• We have clarified our author requirements and our author communications process,making it easier for authors to know exactly what we need before considering theirwork for peer-review; this helps us move manuscripts through the process morequickly. Because authors must now submit their work to us directly, our editors havefound it easier to stay in contact with them, making sure that every paper acceptedfor final publication is timely, complete, and includes any necessary revisions. Detailedauthor requirements are available at <http://www.tappi.org/index.asp?pid=20845>.
TAPPI JOURNAL welcomes research on any topic of interest to the pulp and papercommunity, and encourages authors to submit their work. Our peer review process hasbeen designed to secure readers the most timely,significant and important research avail-able.As a standard-bearer for the TAPPI organization,TAPPI JOURNAL can accept no less.Contact me at any time with questions, ideas or suggestions.
THIS MONTH'S COVER features a forester working for UPM's Blandin Papers unit performing habitat typing, a unique forestry classification system. Photo courtesy of UPM
PEER-REVIEWED SCREENING
Paper mills use pressure screens toremove contaminants from pulp and
to separate fibers according to their char-acteristics. Fractionation describes theprocess whereby a pressure screen differ-entiates between fibers having differentfiber properties such as fiber length.
The most important parts of the screenare the screen plate and rotor.The screenplate has holes or slots through which the“accept” pulp passes.These holes or slotsmay have different shapes.The rotor intro-duces pressure pulses to help prevent thescreen plate from plugging, and it alsotangentially accelerates the pulp at thesurface of the screen plate.The fiber sus-pension feed enters tangentially at thetop of the screen and travels through thegap between the screen plate and therotor. “Accept” pulp passes through thescreen plate to the accept outlet, and the“reject” pulp goes into the reject outlet.
Computational Fluid Dynamics (CFD)is a fast and economical way to obtain apractical understanding of an existingequipment design and to predict thecharacteristics of new designs. Our objec-tive was to study the effects of differentslot shapes on fiber passage using a simulation tool that was developed basedon CFD.
THEORETICAL BACKGROUNDFirst defined as an important perform-ance parameter by Gooding [1] andGooding and Kerekes [2], the “passageratio” relates the contaminant removalefficiency and the reject rate.The passage
(2)
where m is experimentally determined tobe 3.22.
Olson explored the probability func-tion of fiber passage using a two-dimen-sional, rigid-fiber trajectory simulationwith random initial orientations [4]. Heapproximated the probability function as
(3)
Equation 3 shows that only if the fibercenters are initially within the exit layer,they will pass through the aperture.
By assuming that all the fibers initiallyare aligned with the flow, Lawryshyn [6]approximated p(z) by
(4)
where Hcri
is the height of one criticallayer that is lower than the exit layer.
Recently, experimental work has beendone to study the fiber length fractiona-tion by pulp screening with smooth-holed or slotted screen plates [7–9].
As noted, both Eq. 3 and Eq. 4 arebased on two-dimensional simulations,neglecting the effects of the entry flow ofdifferent slot shapes. As mentioned, ourobjective was to use the completelydeveloped, three-dimensional simulationtool to study the effects of different slotshapes on fiber passage, or the turning
ratio is defined as the ratio of the con-centration of pulp fibers or contaminantsin the flow through the aperture, C
a, to
the corresponding concentrationupstream of the aperture, C
u.
Gooding and Kerekes also filmed thetrajectories of individual nylon fibers anddeduced that screening takes place bytwo mechanisms—a wall effect and aturning effect.The wall effect results fromthe fact that the concentration of fibers inthe exit layer Hex, as shown in FFiigg.. 11, islower than that in the mean flow. Theturning effect describes the way in whichthe aperture entry flow affects the pas-sage of different types of fibers.
Kumar experimentally demonstratedsome important factors that affect thefiber passage [3]. He found that the pas-sage ratio varied significantly with theratio of fiber length L to slot width W.
To investigate in detail the importantfactors, Olson [4] and Olson and Kerekes[4, 5] used both the mean and local vari-ables to express the passage ratio as
(1)
where v(z) is fluid velocity at height zabove the wall, C(z) is the upstream con-centration of fibers at height z reflectingthe wall effect, and p(z) is the probabilityof the passage of fibers at height z reflect-ing the turning effect. Olson theoreticallyestimated the probability of passage func-tion and experimentally determined C(z)as a function of fiber length [4]:
The effect of slot shape on the performance of a pressure screen
SUQIN DONG, MARTHA SALCUDEAN, AND IAN GARTSHORE
ABSTRACT: A new, comprehensive simulation tool was used to examine the flow and fiber behavior ina pressure screen having a single slot of any reasonable slot shape. This simulation tool includes three coupled models: the flow model to predict the flow field, the fiber model to trace the fiber trajectory, and thewall model to deal with the case in which a fiber touches the wall of the equipment. This simulation tool wasused to study the passage ratio of fibers moving through slots of several types of geometries. The slot shapewas found to have a critical influence on the fiber passage ratio. Among all the slot shapes tested, a slot withsloping sides upstream and downstream provides the best passage for the same kind of fibers, the same flowconditions, and the same overall slot width. For these contour slots, the contour height is also critical. A contour depth of 0.5 mm was found to be the best for all slot contour shapes examined, for fiber lengthsof 1 mm to 3 mm.
AApppplliiccaattiioonn:: The computer program can be used to predict the characteristics of pressure screens havingdifferent slot shapes, so as to optimize the screen slot design.
VOL. 3: NO. 5 TTAAPPPPII JJOOUURRNNAALL 3
∫∞
=
0 )(
)()(1
L
dzzp
C
zC
V
z
PP
uuef
ν
≥<
mLz
mLzLzm
/1/ if 1
/1/ if /
≥<
=Hexz
Hexzzp
if 0
if 1)(
≥
<=
cri
cri
Hz
Hzzp
if 0
if 1)(
=u
C
zC )(
4 TTAAPPPPII JJOOUURRNNAALL MAY 2004
SCREENING
effect. The wall effect has been studiednumerically with a concentration curve[10] that matches reasonably well withOlson’s observation [4].
NUMERICAL SIMULATIONTOOL
The CFD simulation tool we used isdescribed in detail elsewhere [17]. Thissimulation tool combines three relatedmodels: the flow model, the fiber model,and the wall model.
Flow modelThe mean velocities are solutions of theReynolds-averaged Navier-Stokes andcontinuity equations, which are closedby the standard k – e model for turbu-lence. For solving the flow equations, weused a finite volume method in conjunc-tion with general curvilinear grids andthe domain segmentation method.A CFDprogram developed at UBC [11] wasused to solve the relevant equations.
Fiber modelThe fiber model is critical for predictingthe fiber motion in the screen. For thispurpose,we adapted a three-dimensionalflexible fiber model proposed by Rossand Klingenberg [12]. Each fiber is mod-eled as a chain of spheroids linkedtogether by joints. Both rigid and flexiblefibers can be modeled by changing thebending constant of the joints. Themotion of the fiber is then determinedby solving equations for the translationand rotation of each spheroid.
In the overall model,we have assumedthat the fluid inertia is negligible and thatthe hydrodynamic force has a linear rela-tion with the relative velocity betweenthe fluid and the fiber. We have alsoassumed that the fiber suspension isdilute, and we have ignored thefiber–fiber interaction and the effects ofthe fiber on the turbulent flow.
Wall modelA three-dimensional universal wallmodel has been developed for the casein which a fiber contacts a wall. Thismodel deals with fiber interaction in anywall geometry. This procedure gives thesimulation tool great flexibility so that itcan apply to any other piece of pulp andpaper equipment. This simulation toolhas been used to investigate the fiber ori-entation in a headbox [13, 14] and fiberseparation in a hydrocyclone [15, 16].
RESULTS AND DISCUSSIONGeometry and boundary
conditionsThree kinds of slot shapes were studied:the smooth slot (FFiigg.. 22), the “step–step”contour slot, and the “slope–slope” con-tour slot (FFiigg.. 33).The flow field consistsof three parts, as shown in Fig. 2 for asmooth slot.
Part 1 is the channel. For all the calcu-lations, the channel height H is 20 mm,which is typical of the distance betweenthe rotor and the screen plate in anindustrial screen. The channel consid-ered in this model had a length of 400mm.The upstream length l
u, which is the
distance between the inlet of the calcu-lation domain and the middle of the slot,was 300 mm and was set long enough toallow the entry flow to attain a fullydeveloped profile.
Part 2 is the slot with a width of 0.5mm for all the calculations. The slotheight is 1 mm.
Part 3 is a small chamber with a widthof 3 mm and a length of 60 mm. Thislength ensures that the flow is fullydeveloped at Outlet 1.
The boundary conditions are as fol-lows. There is uniform velocity at theinlet, and the walls have no-slip condi-tions. Outlets 1 and 2 have what we calla “gradient condition,” which means thenormal derivatives of all variables are setequal to 0 and a fixed mass flow rate isspecified.
The geometry of each contour isabout the same as that of the smooth slotexcept for the slot shape. For thestep–step contour, three heights wereinvestigated: 0.1 mm (C0), 0.5 mm (C1),and 1.2 mm (C2). For the slope–slopecontour, two heights were examined: 0.5mm (L1) and 1.3 mm (L2), which wereabout the same heights as C1 and C2,respectively.The boundary conditions forall contours were the same as those forthe smooth slot.
For all the calculations, the upstreamvelocity was 6.5 m/s. To examine theeffects of slot velocities on fiber passage,we used six slot velocities: 1.5 m/s, 2.4m/s, 3.3 m/s, 4.4 m/s, 5.4 m/s, and 6.5m/s, as used by Lawryshyn [6].
Flow fieldThe smooth slot is the simplest of all thescreen slots. FFiigguurree 44 shows the stream-lines for the flow near the smooth slotfor several different slot velocities.As thefigure makes clear, a vortex appears onthe upstream side of the slot no matterhow large the slot velocity is.A larger slotvelocity produces a vortex of a smallersize.
The flow fields in the contoured slotsare quite different.The streamlines in thestep–step contour slots with differentheights are shown in FFiigg.. 55. In Fig. 5,Frames A, B, and C show streamlines inC0, C1, and C2 for the low slot velocity(2.4 m/s). Frames D, E, and F showstreamlines in C0,C1,and C2 for the highslot velocity (5.4 m/s).The flow field dif-fers considerably for different heights forthe step–step contour.The flow featuresin the C0 slot are closer to those in thesmooth slot.There is a small vortex in theupstream corner and a relatively biggervortex in the slot. The vortex in theupstream corner of C1 is much smallerthan that in C2, and the flow approach-ing the slot is smoother in C1. A smallvortex appears in the downstream side
L
w
Hex
x
z
“dividing”streamline
uC
aC
ua CCp /=
uV
sV
V(z)
1. Definition of the passage ratio.
uV
inlet
outlet 1
outlet 2
H
sV
ul
Part 1
Part 3
Par
t 2
2. Geometry of a smooth slot.
0.3=cW2.1 ,5.0 ,1.0=cd
W
1.3 ,5.0=cd
W
Step-step contour slot geometry: C0, C1, C2
Slope-slope contour slot geometry: L1, L2
80˚ 10˚
5.3=cW
3. Geometry of step–step contour slot
and slope–slope contour slot (with all
dimensions in mm).
VOL. 3: NO. 5 TTAAPPPPII JJOOUURRNNAALL 5
of the slot in C2 for the low slot velocity and disappears for highslot velocities.
FFiigguurree 66 shows the streamlines in the slope–slope contourslots L1 and L2 with low slot velocity in Frames A and B andhigh slot velocity in Frames C and D.The streamlines, the size,and the shape of the vortex are also quite different for theslope–slope slot shape with different heights. These resultsshow that the contour height is a critical factor in the design ofthe contour slot.
Three-dimensional fiber motionFiber motion in the slots is important in understanding andjudging the performance of the screen with different plates.Thesimulation tool can provide the detailed, three-dimensional fiberbehavior in any reasonable slot contour shape.
The fiber position (for a straight fiber) in three-dimensionalspace is identified by three parameters, as illustrated in FFiigg.. 77:the center position, the angle θ (which is the angle between thefiber main axis and the z axis), and the angle φ (which is theangle between the projection of the fiber main axis in the xyplane and the y axis). For each different flow field, we studied adifferent number of fibers and found that 4000 fibers wereenough to provide statistically steady results.
The fibers are initially set with a fixed upstream distance tothe slot, a random height, and random orientations θ and φ.Afterrelease, each fiber passes through the slot, goes out throughOutlets 1 or 2, or is stapled to a position in the slot.These three possibilities are denoted in the figures that followby three symbols. If a fiber passes through the aperture, its fateis denoted by symbol “+”. If it continues down the channel, itsfate is denoted by the symbol “o”. If it “staples” at the contourcorner or the slot corner, its fate is denoted by the symbol “ ”.
“Stapling” means that some fibers become immobilized by abalance of forces, which may include hydrodynamic forces andthe wall forces. In the calculation, we assumed that the fiber is
stapled if it does not leave through either the slot or the chan-nel exit within a large number of time steps (such as 200,000time steps). Fate plots, or plots of the fiber’s fate, show the fibertrajectory as a function of initial height and orientations θ andφ [4].
Because the flow field is much different for slots with differ-ent shapes (Figs. 4–6), the fiber behavior is also expected to bedifferent in the slots with different shapes.For example, FFiiggss.. 88aaaanndd 88bb show the fate plots for 1-mm rigid fibers in the smoothslot with a slot velocity of 5.4 m/s as a function of initial heightand the orientations θ and φ. (Other fate plots, which were notreproducible in black and white, are available online at
4. Streamlines in the smooth slot for different slot veloci-
ties (x, z dimensions in mm).
5. Flow field for step–step contour with different contour
heights (x, z dimensions in mm).
6. Flow field for slope–slope contour with different contour
heights.
6 TTAAPPPPII JJOOUURRNNAALL MAY 2004
SCREENING
<http://www.tappi.org/index.asp?pid=29253>
Fibers with initial orientations θgreater than π/2 have greater tendenciesto go into the aperture than do fiberswith θ orientations less than π/2. If thefiber is initially above the exit layer, itgoes into the aperture only if φ is close toπ/2, which means that the fiber is initial-ly in or close to being in the xz plane.Through two-dimensional calculations ofa smooth slot, Olson observed that thefibers with initial negative orientation (φ
0
< 0) could be drawn from a regionbeyond the exit layer [4], where φ
0is the
angle between the fiber and x axis.Olson’s observation is a special case ofthe three-dimension observations. If thefiber is initially within the exit layer, thefiber tends to pass through the aperturewhen its initial angle φ is close to 0 orπ/2, which means that the fiber isapproximately perpendicular to theflow. This three-dimensional phenome-non could be observed only in three-dimensional calculations such as thosepresented here.
Similar plots can be made for the C1slot. (See FFiiggss.. 88cc aanndd 88dd online at<http://www.tappi.org/index.asp?pid=29253>) The difference of the fiber trajec-tory between the smooth slot and the C1slot cases is that for the C1 slot almost allfibers initially within the exit layer willgo into the aperture no matter what theirinitial orientation is. For fibers above theexit layer, the fibers with θ > π/2 alsohave more chance to pass through theaperture.The fibers are stapled near theexit layer, and the three-dimensional phe-nomenon is not obvious from the figure.
The fate plots for 3 mm rigid fibers inthe smooth slot with the same slot veloc-ity as for the 1 mm rigid fiber show that,again, more fibers go into the slot with
higher slot velocities (see FFiiggss.. 88ee--hhonline at <http://www.tappi.org/index.asp?pid=29253>) The tendencyfor fibers with an initial orientation θgreater than π/2 to go into the slots iseven greater than that for 1 mm rigidfibers in this smooth slot. Fibers with aninitial φ close to 0 or π are more likely togo into the slot, which means that fibersin or close to the vertical plane go intothe slot more easily. It appears that longrigid fibers have a stronger tendency tobehave according to the three-dimen-sional phenomenon than do short rigidfibers in the smooth slot.The fate plots offibers (both rigid and flexible) in otherslot shapes can be found elsewhere [17].
From these fate plots,we see that fiberbehavior differs much in the slots withdifferent shapes. Equations 3 and 4 maynot be appropriate for predicting theturning effects of the slots with differentshapes.
Fiber passage ratioFrom the fate plots for different slotshapes, the probability of fiber passagep(z) can be calculated. Along with Eq. 2for the wall effect, Eq. 1 can be used toobtain the fiber passage ratio.
FFiigguurree 99 shows the comparison of thecalculated passage ratio for 1 mm rigidfibers with other available results in thesmooth slot.The calculated value is a lit-tle under Olson’s prediction, whichshows that not all the fibers initiallywithin the exit layer go into the slot.Thefact that Lawryshyn’s simulation is muchlower than all the other results showsthat his calculation with only one initialorientation is not accurate. Our calcula-tion is close to Kumar’s experimentalresults for low slot velocities but is muchlower than Kumar’s results for high slotvelocities. This outcome may resultbecause our simulation does not includeturbulence effect and fiber–fiber interac-tion, which may be important in the realsituation.
FFiigguurree 1100 shows the comparison ofthe simulated passage ratio for 1 mmrigid fibers in five different slot shapes:smooth, C1, C2, L1, and L2.As shown, theslot shape has an important influence onthe passage ratio. Not all of the contourslots have higher passage ratios than thesmooth slot has. C2, for example, has alower passage ratio than the smooth slothas. For the step–step contour C2 and
slope–slope contour L2, which haveabout the same height, L2 has a higherpassage ratio than C2 has. Also, the pas-sage ratio for L1 is higher than that forC1.The slope–slope contour has a higherpassage ratio overall than the sameheight step–step contour has. For thesame kind of contour C1 and C2, theshallow step–step contour C1 has a high-er passage ratio than that of C2. For thesame kind of contour L1 and L2, the shal-low slope–slope contour L1 has a higherpassage ratio than that of L2. Passageratios for longer fibers and for flexiblefibers are discussed elsewhere [17].
SUMMARY AND CONCLUSION
A complete numerical tool to simulatefiber motion in pulp and paper equip-ment has been developed using threemodels: a fluid model to predict the flowfield, a fiber model to predict the fibermotion,and a wall model to deal with thecase in which a fiber touches a solidwall. This tool was used to examine theflow and fiber behavior near a pressurescreen slot.
To examine the flow field in slots withdifferent shapes, we used the mean flowfield obtained from the Reynolds-aver-aged Navier-Stokes method togetherwith the k – e turbulence model.We thenstatistically determined the behavior offibers with different length in those slots.Three-dimensional phenomena wereobserved from these simulations.
Slot shape has the most importantinfluence on fiber passage ratio. For thesame kind of fiber, the same flow condi-tion, and the same slot width, the slotwith sloped sides provides the best pas-sage for the fiber among all the slotshapes tested. For the same kind of con-tour slot, the contour height is also criti-cal, and 0.5 mm seems best for all slotcontours examined, with the qualifica-tion that only fibers in the range oflengths 1 mm to 3 mm were studied.
In conclusion, the simulation toolgives us a way to predict the passage offibers of any length and flexibilitythrough a screen slot with any reason-able shape.The tool can also be used tostudy the behavior of fibers in othercomplex turbulent flows and types ofequipment. TJ
X
Y
Z
ø
cP
O
O
7. Initial position of the fiber.
Figures 8 a-h can be found, in color, at the end of this paper.
VOL. 3: NO. 5 TTAAPPPPII JJOOUURRNNAALL 7
LITERATURE CITED1. Gooding, R. W., “The passage of
fibres through slots in pulpscreening,” M.Sc. thesis,University of British Columbia,Vancouver, BC, 1986.
2. Gooding, R. W., and Kerekes, R. J.,TAPPI J. 75(11): 109(1992).
3. Kumar, A., “Passage of fibersthrough screen apertures,” Ph.D.thesis, University of BritishColumbia, Vancouver, BC, 1991.
4. Olson, J. A., “The effect of fiberlength on passage through narrowapertures,” Ph.D. thesis, Universityof British Columbia, Vancouver,BC, 1996.
5. Olson, J. A., and Kerekes, R. J.,“Fiber passage through a singlescreen aperture,” paper presentedat 51st Appita Conf., APPITA,Melbourne, 1997.
6. Lawryshyn, Y. A., “Statics anddynamics of pulp fibers,” Ph.D.thesis, University of Toronto,Toronto, ON, 1997.
7. Olson, J. A., Roberts, N., Allison, B.J., et al., J. Pulp Paper Sci. 24(12):393(1998).
8. Olson, J. A., Allison, B. J., andRoberts, N., J. Pulp Paper Sci.26(1): 12(2000).
9. Olson, J. A., J. Pulp Paper Sci.27(8): 255(2001).
10. Dong, S., Feng, X., Salcudean M.,et al., Intl. J. Multiphase Flow29(1): 1(2003).
11. Nowak, P., “A multi-grid and multi-block method,” technical report,University of British Columbia,1992.
12. Ross, R. F., and Klingenberg, D. J.,J. Chem. Phys. 106(7): 2949(1997).
13. Zhang, X., “Fiber orientation inheadbox,” M.Sc. thesis, Universityof British Columbia, Vancouver, BC,2001.
14. Dong, S., Feng, X., Salcudean, M.,et al., “Turbulence and fiber orien-tation in the converging section ofa paper-machine headbox,” paperpresented at The 4thASME/JSME/KSME Symposium onComputational Techniques forFluid/Thermal/Chemical Systemswith Industrial Applications, ASMEPVP, V448(2), 221(2002).
15. Wang, Z., “Numerical simulation offiber separation in hydrocyclones,”M.Sc. thesis, University of BritishColumbia, Vancouver, BC, 2002.
16. Wang, Z., Dong, S., Salcudean, M.,et al., Proceedings of the 2002TAPPI Fall Technical Conferenceand Trade Fair, San Diego, TAPPIPRESS, p. 1357.
17. Dong, S., “Modeling of fibermotion in pulp and paper equip-ment,” Ph.D. thesis, University ofBritish Columbia, Vancouver, BC,2002.
Received: March 6, 2003Accepted: December 17, 2003
This paper is also published on TAPPI’sweb site <www.tappi.org> and
summarized in the May Solutions! forPeople, Processes and Paper magazine
(Vol. 87 No. 5).
PA
SS
AG
E R
AT
IO
1.4
1.2
1.0
0.8
0.6
0.4
0.2
00 0.2 0.4 0.6 0.8 1.0 1.2
SLOT VELOCITY/CHANNEL VELOCITY
Smooth slot: current simulationC1 slot: current simulationC2 slot: current simulationL2 slot: current simulationL1 slot: current simulation
Fiber: 1 mm, rigid
10. Passage ratio of 1 mm rigid fibers
in slots of different shapes.
PA
SS
AG
E R
AT
IO
1.2
1.0
0.8
0.6
0.4
0.2
00 0.2 0.4 0.6 0.8 1.0 1.2
SLOT VELOCITY/CHANNEL VELOCITY
Smooth slot: current simulationSmooth slot: Kumar's experimentSmooth slot: Olson's predictionSmooth slot: Lawryshyn's simulation
Fiber: 1 mm, rigid
9. Passage ratio of 1 mm rigid fibers
in a smooth slot.
8a and 8b. Smooth slot: The effects of initial height and orientations θθ and φφ on
the passage of a 1-mm rigid fiber for a slot velocity of 5.4 m/s.
INITIAL ANGLE θθ, radiansa
INIT
IAL
HE
IGH
T
INITIAL ANGLE φφ, radiansb
INIT
IAL
HE
IGH
T
INSIGHTS FROM THE AUTHORSScreening is an important operation in the mill, and itwould be helpful to optimize the screen slot design.However, research on fiber motion and the fluiddynamics of slots is complex, pertaining to both themotion of the fiber and Computational Fluid Dynamics(CFD) modeling. This type of research lies within ourgeneral area of mathematical modeling of pulp andpaper processes and is a further development of ourprevious work on slots in screens.
We found it interesting to watch the “fate of thefibers” as they moved through slots of differentgeometries. It was also interesting to visualize the flowand correlate it with the behavior of fibers. Our nextsteps will be to address fiber–fiber interaction and to
apply fiber motion tracking to such equipment as head-boxes.
Dong is now working at Process Simulations Ltd., Vancouver,206-2386 East Mall, BC V6T 1Z3, Canada. Salcudean andGartshore are at the Dept. of Mechanical Engineering, Universityof British Columbia, 2054-2324 Main Mall, Vancouver, BC, V6T1Z4, Canada. Email Salcudean at [email protected].
Dong Salcudean Gartshore
++
+
++
+
+
++
++
+
+
+
+
+
+
+
+++
+
+
+
++
+
+
+
+
+
+
+
+
+
++
+
++ +
++
++
++
++
++
+
+++
+
++
+
++
+
+
+
+ +
++
+
+
+ +
+
+
++ +
++
+
+
+
+
+ +
+
+
+
+
+
+
+
+
+
+
+ +
+
+
++
+
++++
+
+
+
+
+
++
++
+
+
+
+
+
++
+
++
++
+
++
+ +
+
+
+
+
+
+
+++
+
+
+
+
+
+
+
+++
+
+
+
+
+ ++
+
+
++
+
++
+
++
+
++ +
+
+ +
+
+
+
+
+ +
+
+ +
+
++
+
+
+
+
+
+
++
+
+
+
+
++
+
+
++
+
+
+
+
++
++
+
++
+
+
+
+
+
+
++
++ +
+
+
+
++
+
+
+
+
+
+
+
+++ +
+ ++
++
+ ++
++
+
+
+
++
+
+++ +
+
+++
+
+
+ ++
+
++
+
+
++
+
+
+
+
++
+
++
+
++
+
+
+
+
+
+
+
+
+
+
+
+++
+ +
+
+
+
+
++
+
++
+
+
++
+
+
+
+
+
+
+
+
+
+
++
+
+
+
+
+
+
++
+
+
++ +
+
+
+
+
+
+
+
+
+
+
+
+++
+
+
+
+
++
+
+
+
+
+
++
+
+
+
+
+
+
+++
++
+
++
++ +
+
+ ++ +
+
+++
+ ++
++
+++
++ ++
+
+
+
+
++
+
+
+
+ +
+
+
+
++
+
+
+
+
+
+
+
+
++
+
+
++ +
+
+
+
+
+ +
+
+
+ +
+
++
+
+
+
+ ++ +
+
+
+
+
+
+++
+
++
++
+
+ +
+
++
+
+
+
++
+
++
++
++
+
++
+
+
+
+ +
+
+
+
++ +
+
+
+
+
+
+
+
+
+
+
+
+
+ ++
+
++
+ +
+
+
+
+
+
+
++
+
+
+
+
+
+ +
+ ++
++
+
+
+
+
+ +
+
+
+
+
+
+
+
+
+
+
++
+
++
+
+
+
+
+++
+
+
+
+
+
+
+
+
++
+
++
++
++
+
+
+ ++
+
+
+
++
+
+
+
+
++
+
+
+
++
+
+
+
+
+++ +
+
+
+
+ +
+
++
+
+ +
+
+
+
+
+
+ +
++
+
+
++
+++
+
+++
+
+
+
+
+ +
+
+
+
+
++
+
+
+
+
++
+
+
+
+
+
+
+
+
+
++
+
+
+
++
+
+
++ +
+
+
+ ++
+
+
+
+
+
+
+
+
++
+
++
+
+
+
++
+
+
+
+
+ +++
+
++
+
++
+
+
+
+
++
+
+
+
++
++
+
+
+
++ +
+
++
+
++
+
+++
+
+
++
+ +
++
+
+
++ +
+
+ ++
++
+ ++ +
+
+
++
++ +
+
+++
+
+
+
+
+
+
+
+
+
++
+
+
++
+
+
+
+
+
+
+
+
+
+
++
++
+
+
++
+
+
+
++
+ +++
+
+ +
+ +
++
+
+
+
+
+
+
+
+
+
+
+
++
+
++
+
+
+
++
++
+
+
+
++
+
+
+
++
+
+
+++
+
+
+
++
++
+
++
+
+ +
+
+ ++
++
+
+
+
+
+
+
+
++
++
+
+
+
++
+
+
+
+
+
+
+++
+
+
+
++
+
+
++ +
+
++ + ++
++
+
++
++
++
+
+
+
+
+++
++ +
++
++
++ +
+
+
++ + +
+
+
+
+
++
++
+
++
+
++
++
+
+
+
+
+
+
+
+
++
+
++
+
+
+
++++
+
+++
+
+
++
++
++
+
++
+
+
++ +
+
+
++
+
+ +
+
+
+ +
+
+
++
++
++
+
+
+++
+ +
+
+
++
+
+
+
+ ++
++ ++
+
+ +
+
+
+
+
+
+
++
+
++
+
+
+
+++
+
+
+
+
+
+
+
++ +
+
++
++
++
++
+
++
+
+
++ ++
+
+
+
+
+
+
+
+
+
+
++
+ +
+
+
+
+
+
+++
+
+
+
++
+
+
++
++
+ +
+
+
+
+
++
++
+
+
++
+++
+
+
+
+
++
++
+
+
+++
+
+
+
+
+
+
+
+
++
+
+
+
+
+
++
+
+
+
+
+++
++ +
++ +
+
+
+
+
+
+
++ +
+
+
+++
++
++++
+
+
++
++
+
+
+
+
+
+
+
++
+
+
+
++
++
+
+
+
++
+
+
+
+
++ +
+
+
+
+
+
+
+
++
+
+
+
++
++
+
+ ++
+
++
+
+ + ++
+
++
+
+
+
+ ++
++
+
+
+
+++
+
+
++
+
++
+
+
+ ++ +
+
+
++
+
+
+
+
+
++
++
++
+
++
++
+
+ ++++++
+
+
+
++
+
++
++
+ +
+
+
+
+
+
+
+
+
+
+
+
+
++ +
+
++
+
+
++
+ +++
++
+
+
++
+
+
++
+
+
+
+
+ +
+
+
+
++
+
+
+
+
+
++
++
+
++
+
+
+
+
++
+
++
++
+
+
+
+
++ +
+
++
+ +
+
+
++
+
++
+++
+
+
++
+
+
+
+
++
+
+ ++ +
+
++
+
+
++
+
+
+
++
+
+
+
+
+
+
+
++
+
++
+
+
+
++
+
+
++
+
+
+
+
+
++ +
+
++
+
++
+
+
+
+
+
+
++
+
+
+
+ ++ +
+
+
+
+
+
+
+
+
+
+
+
+
+
++
+
+ ++
+
+
+
+
+
+
+
+
+
+
+
+
+++
++
+
+
++
++
++
+
+
+
++
+
+
+ +
+
++
+
++
++
++
+
+
+
+
+
+
+
+
++
++
++
+
+
++++
+
+
+
+
+++
+
+ ++
+
+
++
+
+
++
++
+
+
+
+++
++
+
++
+
+
+
++
+
+
+
++
+
++
++
+
+
+
+
+
+
+ ++
+
++
++
+
+
+ +
++
+
++
+
+
+
+
+
+
+
+
+
++
+
+
+
+++
+
+ +
+
+
+
++
+
+
+
++
+
++
+++
+
+
+
+
++
+
+
+
+
+
+
++
+
+
+
+
+
+
++
++
+ +
+++
++
+
+
+
+
++
+
+
+
+ +
++
+ ++
++
+
+
++
+
+
+
+
+
+
+
+
++
+
+
+ +
+
+
+
++
+
+
++
+
++
+
+
+
++
+
+
++
+
++
++
+
+
++
+
+
+
+++
+
+
++
+
++
+
+
+
+
+
++ +
+
++
+
++
++
+++
+
+
+
+
+
+
++
++
+ +
++ +
++
+
++
+
+
+ +
++
+
+
+
+
++
+
+
++
+
+
+
+
+
++
+
+
+
++
+ +
+
+
+
+
+ +
+ +++
+
++
+
+
++
++ +
+
+
+
+
+
+
+
+
+
+ +
+
+
+++
+
+
+
+
+ +
+
+ +
+
++
+
In itia l a n g le
Initialheight(mm)
0 .5 1 1 .5 2 2 .5 3
0 .1
0 .2
0 .3
0 .4
0 .5
0 .6
0 .7
0 .8
0 .9
1
θ ( ra d ia n s )
c fd 5 V s= 5 .4 m /s
( ra d ia n s )
++
+
+ +
+
+
++
++
+
+
+
+
+
+
+
+ ++
+
+
+
++
+
+
+
+
+
+
+
+
+
++
+
++ +
++
++
++
++
++
+
+++
+
++
+
++
+
+
+
+ +
++
+
+
++
+
+
+ + +
++
+
+
+
+
++
+
+
+
+
+
+
+
+
+
+
++
+
+
++
+
++
++
+
+
+
+
+
+ +
++
+
+
+
+
+
++
+
++
++
+
++
++
+
+
+
+
+
+
+++
+
+
+
+
+
+
+
+++
+
+
+
+
+++
+
+
++
+
++
+
++
+
++ +
+
+ +
+
+
+
+
+ +
+
+ +
+
++
+
+
+
+
+
+
++
+
+
+
+
++
+
+
++
+
+
+
+
++
++
+
++
+
+
+
+
+
+
++
+ + +
+
+
+
+ +
+
+
+
+
+
+
+
+++ +
+++
++
+ ++
++
+
+
+
++
+
+++ +
+
++
+
+
+
+++
+
++
+
+
++
+
+
+
+
++
+
+ +
+
+ +
+
+
+
+
+
+
+
+
+
+
+
+++
+ +
+
+
+
+
++
+
++
+
+
++
+
+
+
+
+
+
+
+
+
+
++
+
+
+
+
+
+
++
+
+
+++
+
+
+
+
+
+
+
+
+
+
+
++ +
+
+
+
+
++
+
+
+
+
+
+ +
+
+
+
+
+
+
++ +
++
+
++
+++
+
++++
+
++ +
+ ++
++
++ +
++ + +
+
+
+
+
++
+
+
+
++
+
+
+
++
+
+
+
+
+
+
+
+
++
+
+
+++
+
+
+
+
++
+
+
++
+
++
+
+
+
++++
+
+
+
+
+
+++
+
++
++
+
++
+
++
+
+
+
++
+
++
++
++
+
++
+
+
+
++
+
+
+
+ ++
+
+
+
+
+
+
+
+
+
+
+
+
+++
+
++
+ +
+
+
+
+
+
+
++
+
+
+
+
+
+ +
+++
+ +
+
+
+
+
++
+
+
+
+
+
+
+
+
+
+
++
+
++
+
+
+
+
+ ++
+
+
+
+
+
+
+
+
++
+
++
++
++
+
+
+++
+
+
+
++
+
+
+
+
++
+
+
+
++
+
+
+
+
++ + +
+
+
+
++
+
++
+
+ +
+
+
+
+
+
+ +
+ +
+
+
+ +
+++
+
+++
+
+
+
+
++
+
+
+
+
++
+
+
+
+
++
+
+
+
+
+
+
+
+
+
++
+
+
+
++
+
+
++ ++
+
++ +
+
+
+
+
+
+
+
+
++
+
++
+
+
+
++
+
+
+
+
+++ +
+
++
+
++
+
+
+
+
++
+
+
+
++
++
+
+
+
+++ +
++
+
++
+
+++
+
+
++
++
++
+
+
++ +
+
+++
+ +
+ ++ +
+
+
++
++ +
+
++
+
+
+
+
+
+
+
+
+
+
++
+
+
++
+
+
+
+
+
+
+
+
+
+
++
++
+
+
++
+
+
+
++
+ ++ +
+
++
++
++
+
+
+
+
+
+
+
+
+
+
+
+ +
+
++
+
+
+
++
++
+
+
+
++
+
+
+
+ +
+
+
+ ++
+
+
+
++
++
+
++
+
+ +
+
+++
++
+
+
+
+
+
+
+
++
++
+
+
+
++
+
+
+
+
+
+
+++
+
+
+
++
+
+
++ +
+
++ ++
+
++
+
++
++
++
+
+
+
+
+++
++ +
++
+ +
++ +
+
+
++ + +
+
+
+
+
++
+ +
+
++
+
++
++
+
+
+
+
+
+
+
+
++
+
+ +
+
+
+
+++
+
+
+ ++
+
+
++
++
++
+
++
+
+
+++
+
+
+ +
+
+ +
+
+
+ +
+
+
+ +
++
++
+
+
+ ++
+ +
+
+
+ +
+
+
+
+++
++++
+
++
+
+
+
+
+
+
++
+
++
+
+
+
++ +
+
+
+
+
+
+
+
+++
+
++
++
+ +
++
+
++
+
+
++++
+
+
+
+
+
+
+
+
+
+
++
+ +
+
+
+
+
+
+ ++
+
+
+
++
+
+
++
++
+ +
+
+
+
+
++
++
+
+
++
++
+
+
+
+
+
++
++
+
+
++ +
+
+
+
+
+
+
+
+
++
+
+
+
+
+
++
+
+
+
+
+ ++
++ +
+ ++
+
+
+
+
+
+
++ +
+
+
++
+
+ +
++
+ +
+
+
++
++
+
+
+
+
+
+
+
++
+
+
+
+ +
+ +
+
+
+
++
+
+
+
+
+++
+
+
+
+
+
+
+
++
+
+
+
+ +
++
+
+ ++
+
++
+
++ ++
+
++
+
+
+
+++
++
+
+
+
++
+
+
+
++
+
++
+
+
+ +++
+
+
++
+
+
+
+
+
++
++
+ +
+
++
++
+
+ +++ ++ +
+
+
+
++
+
++
++
+ +
+
+
+
+
+
+
+
+
+
+
+
+
++ +
+
++
+
+
+ +
+ +++
++
+
+
+ +
+
+
++
+
+
+
+
++
+
+
+
++
+
+
+
+
+
++
++
+
++
+
+
+
+
+++
++
++
+
+
+
+
++ +
+
++
+ +
+
+
+ +
+
++
++
+
+
+
+ +
+
+
+
+
++
+
+ + ++
+
++
+
+
+ +
+
+
+
++
+
+
+
+
+
+
+
++
+
++
+
+
+
+ +
+
+
++
+
+
+
+
+
+++
+
++
+
++
+
+
+
+
+
+
+ +
+
+
+
+ ++ +
+
+
+
+
+
+
+
+
+
+
+
+
+
++
+
++ +
+
+
+
+
+
+
+
+
+
+
+
+
+ ++
++
+
+
+ +
++
++
+
+
+
+ +
+
+
+ +
+
+ +
+
+ +
++
++
+
+
+
+
+
+
+
+
++
++
++
+
+
++
++
+
+
+
+
++ +
+
+++
+
+
++
+
+
+ +
++
+
+
+
+ ++
++
+
++
+
+
+
++
+
+
+
++
+
++
++
+
+
+
+
+
+
+ ++
+
+ +
++
+
+
+ +
++
+
++
+
+
+
+
+
+
+
+
+
++
+
+
+
+ ++
+
++
+
+
+
++
+
+
+
+ +
+
++
++
+
+
+
+
+
++
+
+
+
+
+
+
++
+
+
+
+
+
+
++
++
+ +
+++
+ +
+
+
+
+
++
+
+
+
++
++
+ ++
++
+
+
++
+
+
+
+
+
+
+
+
++
+
+
+ +
+
+
+
++
+
+
++
+
+ +
+
+
+
++
+
+
++
+
++
++
+
+
++
+
+
+
+ ++
+
+
++
+
++
+
+
+
+
+
+++
+
++
+
++
++
+++
+
+
+
+
+
+
+ +
++
++
+++
++
+
+ +
+
+
++
++
+
+
+
+
++
+
+
+ +
+
+
+
+
+
++
+
+
+
++
+ +
+
+
+
+
+ +
+ +++
+
++
+
+
++
+ + +
+
+
+
+
+
+
+
+
+
++
+
+
+++
+
+
+
+
++
+
+ +
+
++
+
In itia l a n g le
Initialheight(mm)
0 .5 1 1 .5 2 2 .5 3
0 .1
0 .2
0 .3
0 .4
0 .5
0 .6
0 .7
0 .8
0 .9
1
φ ( ra d ia n s )
c fd 5 V s= 5 .4 m /s
( ra d ia n s )
Figure 8a (top) and 8b (bottom) Smooth slot: The effects of initial height and orientation θ and φ
on passage of 1mm rigid fiber for slot velocity 5.4 m/s.
++
+
++
+ +
+
++
+
+
+
+
+
+
+
+
+ +
+
+
+++
+
+
++
++
+
++
+
+
+
+
+
++
++
+
+
+
+
+
+
+
+
+
+
++ +
+
+
+
+
++
+
++
++
++
+
+
++
+
+
+
++
+
+
+
+
++
+
+
+
+ +
++
+
+
+
+
+
+
+
+
+
+
+
+
++
++ +
++
+
+
+
+
+ +
+
+ +
+ +
+
+
+
++
+
++
+
++
++
++
+
+ +
+
+
+
++
+
+
+
+
++
+
+
++
+
+
++
+
+
++
+
++
+
+
+
+++
++
+
++
+
+
+
+
+
+
+
+ +
+++
++
+ +
++
+
++
+
++
+ +
+
++
+
+
+
+
+
+
++
+
++
+
+
+
+
+
+
++ +
+
+++
++
+
+
+
++
+
+
+ + +
+
++
+
+
+
+++
+
+
+
+
+
++
+
++
+
+
++
+
+
+
++
++
+
+
+
+ +
++
+
+
+
+
++
+ +
+
+ +
++
+
+
+
+
++
+
+
++
+
++
+
++
+
+
+
+
+
+
++
++
+
++
+
+
++
+ +
+
++
+
+
+
+
+
++
++
+
+
+
+
+
+
++
+
++
+
+
+
+
+
+
++
++
+
+
++ +
+
+
+
++
+
+
+
++
+
+
+
+
+
+
+
+
+
+
++ +
+
+ +
+
+
++
++
+
++
++
+
++
+
++
+
+
+
+
++
+
+ +
+
+
+
++
++
+
++
+ ++
+
+ +
+
+
+
+ +
++
+
+
+
+
+
+
+
+
+
+
++
+
++
+
+
+
+
+
+
+
+
+
+
+
+ +
+
+
+
+
+
++
+
+
+
+
+
+
+
+
+ +
++
+
+
+
+
+
+ ++
+
+
++
+
+
+
+
++
+
++
+
+
+
+
+
+
+
+
+
+
+
+
+
+
++
+
+
+
+
+
+
+
+
++
++
+
+
+
++
++
+
+
+
+
+
+
+
+
+
++
+
+
+
+
+
+
+
+
+
+++
+
+
++
++
+
+
+
+
++
+
+
++
+
+
+
+
+
+
+ +
+
++
+
+
+
+++
++
++
+
+
+
+
++
+
++ +
++
+
++
++ +
+ +
+
+
++
+
+
++
+
+
+
++
++
+
+++
++
+
+++
+
+
+
+
+
++
++
+
++
+
+
+
+
+
+
+
++
+
+
++
+
+
+
+
+
+
+
+
++
+
+
+
++
++
++
+
+
+
+
+
+
+ +
+
++
+
+
+
+
+
++
+
+
+
+
+++
+
+ ++
+
++
+
+
+
+
+
+
+
++
+
++
+
++
++
+
+ ++
+
++
+
+ ++
+
+
+
+
+
+
+
+++
+
++
+
+
+
+
+
+
+
+
++
+
+
+
+
+
+
+
+
+
+
+
+
+
+
++
+
+
+
+
++
+
+
+
+
+
+
+
+
+ ++
+
+
+
+
+ +
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+ ++
++
+
+
++
+
+
+
+
+
+
++
+
+
+
+
+
+
++
+
+
+
+
+
+
+
++
+
+
++
++
+
+
+
+
++
+
++
+
+
+
+
+
+
+
+
+
+ +
+
++
++
+
+
+
+
+
+
+
+
+
++
+
+
+
+ +
+
++
+
+
++
+
+
+
+
+
+
+
+
++
++
++
++
+
+
+
++
+
+ ++
+
+
+
+
+
++ ++
+
++
+
+ +
+
++
+
+
+++
+
+
+
+
+
++
+
++
+
+ +
++
+
+
+
+
+ +
+
+++
+
+++
+
+
+
+
+
+ +
+
+
++ +
+
+
+
++
+
+
+
+
++
+
+
+
+
+
+
++
++
+
+
+
+
++
+
+
+
+
+ +
+++
+
+
+
++ +
+++
+
+
++
+
+
++
++
+
+
+
+
+
+
++
+
+
+
++
+
++
+
+
+
+
++
+
+
+
+
+
+ +++
+
+
+
+
+
++
++
+
+
+
+
+
++ +
+
++
+
+
+
+
++
+
++
+
++
+
++
+
+
++
++
+
+
+
++
+
+
+
+
+
+
++
+
+
+
+ +
+
+
+
+
+
+
++
+
+
+
+
++ +
+
+
+
+
+
+
+
++
+
++ ++
+
++ +
+
+
+
++
+
++
++ +
+
+
+++
+
+
+
+
+
+
+
+
+
++
+
+
++
+
+
+
+
+
+
+
+
+
+
++
++
+
+
++
+
+
+
++
+ +++
+
+ +
+ +
++
+
+
+
+
+
+
+
+
+
+
+
++
+
++
+
+
+
++
++
+
+
+
++
+
+
+
++
+
+
+++
+
+
+
++
++
+
++
+
+ +
+
+ ++
++
+
+
+
+
+
+
+
++
++
+
+
+
++
+
+
+
+
+
+
+++
+
+
+
++
+
+
++ +
+
++ + ++
++
+
++
++
++
+
+
+
+
+++
++ +
++
++
++ +
+
+
++ + +
+
+
+
+
++
++
+
++
+
++
++
+
+
+
+
+
+
+
+
++
+
++
+
+
+
++++
+
+++
+
+
++
++
++
+
++
+
+
++ +
+
+
++
+
+ +
+
+
+ +
+
+
++
++
++
+
+
+++
+ +
+
+
++
+
+
+
+ ++
++ ++
+
+ +
+
+
+
+
+
+
++
+
++
+
+
+
+++
+
+
+
+
+
+
+
++ +
+
++
++
++
++
+
++
+
+
++ ++
+
+
+
+
+
+
+
+
+
+
++
+ +
+
+
+
+
+
+++
+
+
+
++
+
+
++
++
+ +
+
+
+
+
++
++
+
+
++
+++
+
+
+
+
++
++
+
+
+++
+
+
+
+
+
+
+
+
++
+
+
+
+
+
++
+
+
+
+
+++
++ +
++ +
+
+
+
+
+
+
++ +
+
+
+++
++
++++
+
+
++
++
+
+
+
+
+
+
+
++
+
+
+
++
++
+
+
+
++
+
+
+
+
++ +
+
+
+
+
+
+
+
++
+
+
+
++
++
+
+ ++
+
++
+
+ + ++
+
++
+
+
+
+ ++
++
+
+
+
+++
+
+
++
+
++
+
+
+ ++ +
+
+
++
+
+
+
+
+
++
++
++
+
++
++
+
+ ++++++
+
+
+
++
+
++
++
+ +
+
+
+
+
+
+
+
+
+
+
+
+
++ +
+
++
+
+
++
+ +++
++
+
+
++
+
+
++
+
+
+
+
+ +
+
+
+
++
+
+
+
+
+
++
++
+
++
+
+
+
+
++
+
++
++
+
+
+
+
++ +
+
++
+ +
+
+
++
+
++
+++
+
+
++
+
+
+
+
++
+
+ ++ +
+
++
+
+
++
+
+
+
++
+
+
+
+
+
+
+
++
+
++
+
+
+
++
+
+
++
+
+
+
+
+
++ +
+
++
+
++
+
+
+
+
+
+
++
+
+
+
+ ++ +
+
+
+
+
+
+
+
+
+
+
+
+
+
++
+
+ ++
+
+
+
+
+
+
+
+
+
+
+
+
+++
++
+
+
++
++
++
+
+
+
++
+
+
+ +
+
++
+
++
++
++
+
+
+
+
+
+
+
+
++
++
++
+
+
++++
+
+
+
+
+++
+
+ ++
+
+
++
+
+
++
++
+
+
+
+++
++
+
++
+
+
+
++
+
+
+
++
+
++
++
+
+
+
+
+
+
+ ++
+
++
++
+
+
+ +
++
+
++
+
+
+
+
+
+
+
+
+
++
+
+
+
+++
+
+ +
+
+
+
++
+
+
+
++
+
++
+++
+
+
+
+
++
+
+
+
+
+
+
++
+
+
+
+
+
+
++
++
+ +
+++
++
+
+
+
+
++
+
+
+
+ +
++
+ ++
++
+
+
++
+
+
+
+
+
+
+
+
++
+
+
+ +
+
+
+
++
+
+
++
+
++
+
+
+
++
+
+
++
+
++
++
+
+
++
+
+
+
+++
+
+
++
+
++
+
+
+
+
+
++ +
+
++
+
++
++
+++
+
+
+
+
+
+
++
++
+ +
++ +
++
+
++
+
+
+ +
++
+
+
+
+
++
+
+
++
+
+
+
+
+
++
+
+
+
++
+ +
+
+
+
+
+ +
+ +++
+
++
+
+
++
++ +
+
+
+
+
+
+
+
+
+
+ +
+
+
+++
+
+
+
+
+ +
+
+ +
+
++
+
In itia l a n g le
Initialheight(mm)
0 .5 1 1 .5 2 2 .5 3
0 .1
0 .2
0 .3
0 .4
0 .5
0 .6
0 .7
0 .8
0 .9
1
θ ( ra d ia n s )
c fd 5 V s= 5 .4 m /s
( ra d ia n s )
++
+
+ +
++
+
++
+
+
+
+
+
+
+
+
+ +
+
+
+ ++
+
+
++
++
+
++
+
+
+
+
+
++
++
+
+
+
+
+
+
+
+
+
+
++ +
+
+
+
+
++
+
++
++
++
+
+
++
+
+
+
++
+
+
+
+
++
+
+
+
+ +
++
+
+
+
+
+
+
+
+
+
+
+
+
++
+ + +
++
+
+
+
+
++
+
++
+ +
+
+
+
++
+
++
+
++
++
++
+
++
+
+
+
++
+
+
+
+
++
+
+
++
+
+
+ +
+
+
++
+
++
+
+
+
++ +
++
+
++
+
+
+
+
+
+
+
++
+++
++
++
++
+
++
+
++
++
+
++
+
+
+
+
+
+
++
+
++
+
+
+
+
+
+
+ + +
+
+++
++
+
+
+
++
+
+
+++
+
++
+
+
+
+++
+
+
+
+
+
++
+
++
+
+
++
+
+
+
++
++
+
+
+
+ +
++
+
+
+
+
++
+ +
+
+ +
++
+
+
+
+
++
+
+
++
+
++
+
++
+
+
+
+
+
+
++
++
+
+ +
+
+
++
+ +
+
++
+
+
+
+
+
++
++
+
+
+
+
+
+
++
+
++
+
+
+
+
+
+
++
++
+
+
+ + +
+
+
+
++
+
+
+
+ +
+
+
+
+
+
+
+
+
+
+
++ +
+
++
+
+
++
++
+
++
++
+
++
+
++
+
+
+
+
++
+
++
+
+
+
++
++
+
++
+++
+
+ +
+
+
+
++
++
+
+
+
+
+
+
+
+
+
+
++
+
+ +
+
+
+
+
+
+
+
+
+
+
+
+ +
+
+
+
+
+
++
+
+
+
+
+
+
+
+
+ +
++
+
+
+
+
+
+++
+
+
++
+
+
+
+
++
+
++
+
+
+
+
+
+
+
+
+
+
+
+
+
+
++
+
+
+
+
+
+
+
+
++
++
+
+
+
++
++
+
+
+
+
+
+
+
+
+
++
+
+
+
+
+
+
+
+
+
+ ++
+
+
++
++
+
+
+
+
++
+
+
+ +
+
+
+
+
+
+
+ +
+
++
+
+
+
++ +
++
++
+
+
+
+
++
+
+++
+ +
+
++
+++
++
+
+
++
+
+
++
+
+
+
++
++
+
++ +
++
+
+ + +
+
+
+
+
+
+++
+
+
+ +
+
+
+
+
+
+
+
++
+
+
++
+
+
+
+
+
+
+
+
++
+
+
+
++
++
++
+
+
+
+
+
+
++
+
++
+
+
+
+
+
++
+
+
+
+
+++
+
+++
+
++
+
+
+
+
+
+
+
++
+
++
+
++
++
+
++ +
+
++
+
+++
+
+
+
+
+
+
+
++
+
+
++
+
+
+
+
+
+
+
+
++
+
+
+
+
+
+
+
+
+
+
+
+
+
+
++
+
+
+
+
++
+
+
+
+
+
+
+
+
+++
+
+
+
+
+ +
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+++
+ +
+
+
++
+
+
+
+
+
+
++
+
+
+
+
+
+
++
+
+
+
+
+
+
+
++
+
+
++
++
+
+
+
+
+ +
+
++
+
+
+
+
+
+
+
+
+
+ +
+
++
++
+
+
+
+
+
+
+
+
+
++
+
+
+
++
+
++
+
+
++
+
+
+
+
+
+
+
+
++
++
++
++
+
+
+
++
+
++ +
+
+
+
+
+
+ + ++
+
++
+
++
+
++
+
+
++ +
+
+
+
+
+
++
+
++
+
+ +
+ +
+
+
+
+
++
+
+++
+
+++
+
+
+
+
+
++
+
+
+++
+
+
+
++
+
+
+
+
++
+
+
+
+
+
+
++
++
+
+
+
+
++
+
+
+
+
++
+ ++
+
+
+
++ +
++ +
+
+
+ +
+
+
+ +
+ +
+
+
+
+
+
+
++
+
+
+
++
+
++
+
+
+
+
++
+
+
+
+
+
+++ +
+
+
+
+
+
++
++
+
+
+
+
+
+++
+
++
+
+
+
+
++
+
++
+
++
+
++
+
+
++
++
+
+
+
++
+
+
+
+
+
+
++
+
+
+
++
+
+
+
+
+
+
++
+
+
+
+
++ +
+
+
+
+
+
+
+
++
+
++++
+
++ +
+
+
+
++
+
++
++ +
+
+
++
+
+
+
+
+
+
+
+
+
+
++
+
+
++
+
+
+
+
+
+
+
+
+
+
++
++
+
+
++
+
+
+
++
+ ++ +
+
++
++
++
+
+
+
+
+
+
+
+
+
+
+
+ +
+
++
+
+
+
++
++
+
+
+
++
+
+
+
+ +
+
+
+ ++
+
+
+
++
++
+
++
+
+ +
+
+++
++
+
+
+
+
+
+
+
++
++
+
+
+
++
+
+
+
+
+
+
+++
+
+
+
++
+
+
++ +
+
++ ++
+
++
+
++
++
++
+
+
+
+
+++
++ +
++
+ +
++ +
+
+
++ + +
+
+
+
+
++
+ +
+
++
+
++
++
+
+
+
+
+
+
+
+
++
+
+ +
+
+
+
+++
+
+
+ ++
+
+
++
++
++
+
++
+
+
+++
+
+
+ +
+
+ +
+
+
+ +
+
+
+ +
++
++
+
+
+ ++
+ +
+
+
+ +
+
+
+
+++
++++
+
++
+
+
+
+
+
+
++
+
++
+
+
+
++ +
+
+
+
+
+
+
+
+++
+
++
++
+ +
++
+
++
+
+
++++
+
+
+
+
+
+
+
+
+
+
++
+ +
+
+
+
+
+
+ ++
+
+
+
++
+
+
++
++
+ +
+
+
+
+
++
++
+
+
++
++
+
+
+
+
+
++
++
+
+
++ +
+
+
+
+
+
+
+
+
++
+
+
+
+
+
++
+
+
+
+
+ ++
++ +
+ ++
+
+
+
+
+
+
++ +
+
+
++
+
+ +
++
+ +
+
+
++
++
+
+
+
+
+
+
+
++
+
+
+
+ +
+ +
+
+
+
++
+
+
+
+
+++
+
+
+
+
+
+
+
++
+
+
+
+ +
++
+
+ ++
+
++
+
++ ++
+
++
+
+
+
+++
++
+
+
+
++
+
+
+
++
+
++
+
+
+ +++
+
+
++
+
+
+
+
+
++
++
+ +
+
++
++
+
+ +++ ++ +
+
+
+
++
+
++
++
+ +
+
+
+
+
+
+
+
+
+
+
+
+
++ +
+
++
+
+
+ +
+ +++
++
+
+
+ +
+
+
++
+
+
+
+
++
+
+
+
++
+
+
+
+
+
++
++
+
++
+
+
+
+
+++
++
++
+
+
+
+
++ +
+
++
+ +
+
+
+ +
+
++
++
+
+
+
+ +
+
+
+
+
++
+
+ + ++
+
++
+
+
+ +
+
+
+
++
+
+
+
+
+
+
+
++
+
++
+
+
+
+ +
+
+
++
+
+
+
+
+
+++
+
++
+
++
+
+
+
+
+
+
+ +
+
+
+
+ ++ +
+
+
+
+
+
+
+
+
+
+
+
+
+
++
+
++ +
+
+
+
+
+
+
+
+
+
+
+
+
+ ++
++
+
+
+ +
++
++
+
+
+
+ +
+
+
+ +
+
+ +
+
+ +
++
++
+
+
+
+
+
+
+
+
++
++
++
+
+
++
++
+
+
+
+
++ +
+
+++
+
+
++
+
+
+ +
++
+
+
+
+ ++
++
+
++
+
+
+
++
+
+
+
++
+
++
++
+
+
+
+
+
+
+ ++
+
+ +
++
+
+
+ +
++
+
++
+
+
+
+
+
+
+
+
+
++
+
+
+
+ ++
+
++
+
+
+
++
+
+
+
+ +
+
++
++
+
+
+
+
+
++
+
+
+
+
+
+
++
+
+
+
+
+
+
++
++
+ +
+++
+ +
+
+
+
+
++
+
+
+
++
++
+ ++
++
+
+
++
+
+
+
+
+
+
+
+
++
+
+
+ +
+
+
+
++
+
+
++
+
+ +
+
+
+
++
+
+
++
+
++
++
+
+
++
+
+
+
+ ++
+
+
++
+
++
+
+
+
+
+
+++
+
++
+
++
++
+++
+
+
+
+
+
+
+ +
++
++
+++
++
+
+ +
+
+
++
++
+
+
+
+
++
+
+
+ +
+
+
+
+
+
++
+
+
+
++
+ +
+
+
+
+
+ +
+ +++
+
++
+
+
++
+ + +
+
+
+
+
+
+
+
+
+
++
+
+
+++
+
+
+
+
++
+
+ +
+
++
+
In itia l a n g le
Initialheight(mm)
0 .5 1 1 .5 2 2 .5 3
0 .1
0 .2
0 .3
0 .4
0 .5
0 .6
0 .7
0 .8
0 .9
1
φ ( ra d ia n s )
c fd 5 V s= 5 .4 m /s
( ra d ia n s )
Figure 8c (top) and 8d (bottom) C1 slot: The effects of initial height and orientation θ and φ on
passage of 1mm rigid fiber for slot velocity 5.4m/s.
+
+
+++
+
+
+
+
+
+
+
+
+
+
+
++ +
+
+
+
+++
+
+
++
+
++
+
+
+
+
+
+
+
+
+ ++
+
+++
+
+
+
+ +
+
+
+
+
+
+
+ +
+
++
+++
++
+
+
+
+++
+
+ +
+
+
++
+
+
++
++ +
+
+
+
+
+
+
+
++++
++
+++
+
+
+
++ +
++
+
+
+
+
+
++
+
+++
+
+
+ +
+
+
+ +
++ +
+
+
+
++
+
+
+
+
+
+
+
+
+
++
+
++
+
+
+
+ ++
+
+
+
+
+
++
+
++
+
+
+
+
+
+
++
+
+
++
++
+
+
+
++
+
+
+
+ +
+
+
+
++
+
+
+
++
++
++
+
+
+
+
+
+
+
+
+
+
+ ++
+
+
+
+
+
+
+
+ ++
++
+
+
++
+
+
+
++
++
+
++
+++
+
++ ++
+
+
++
++
+
++
+
+
+
+
+
++
++
++
+
+
++
+
++
+
++
+
+
+
++
+
+
+
+
+
+ +
+
+
+
+
+
+
++
+
++
+
+
++
+
+
+
+
+
++
++
++
+
+
+
+
++
++
+
++
+
++
++
+++
++
+
++
+
+
+
+
+
++
+
+
++
+
+
+
+
+
+
++
+
+
+
++
+
+
+
+
++
++
+
+
+
+
+
+
+ +++
+ +
++
+
+
+
++
++++ +
+
++ +
+
+
++
+
++ +
++
+
+++
+
++
+
+
+
+
+
++
+ +
+
++
+
+
+
+
+
+ +
+
+
+
+
+
+
++
+
+
+++
+
+
+
+
+
+
+
+
+
+
+
Initial angle
Initialheight(mm)
1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
θ (radians)
Smoothslot:cfd2 Vs=2.4 m/s
+
+
+++
+
+
+
+
+
+
+
+
+
+
+
+++
+
+
+
++ +
+
+
++
+
++
+
+
+
+
+
+
+
+
+ ++
+
+ ++
+
+
+
+ +
+
+
+
+
+
+
+ +
+
++
+++
+ +
+
+
+
+++
+
++
+
+
++
+
+
++
++ +
+
+
+
+
+
+
+
++++
++
++ +
+
+
+
+++
++
+
+
+
+
+
++
+
+ ++
+
+
++
+
+
+ +
++ +
+
+
+
++
+
+
+
+
+
+
+
+
+
++
+
++
+
+
+
+++
+
+
+
+
+
++
+
++
+
+
+
+
+
+
++
+
+
++
+ +
+
+
+
++
+
+
+
++
+
+
+
++
+
+
+
++
++
++
+
+
+
+
+
+
+
+
+
+
+++
+
+
+
+
+
+
+
+++
++
+
+
++
+
+
+
++
++
+
++
++
+
+
++ ++
+
+
++
++
+
++
+
+
+
+
+
++
++
++
+
+
++
+
++
+
++
+
+
+
+ +
+
+
+
+
+
++
+
+
+
+
+
+
++
+
++
+
+
+ +
+
+
+
+
+
++
++
++
+
+
+
+
+ +
++
+
++
+
++
++
+++
++
+
+ +
+
+
+
+
+
++
+
+
++
+
+
+
+
+
+
++
+
+
+
++
+
+
+
+
++
+ +
+
+
+
+
+
+
++++
+ +
++
+
+
+
++
++ +
++
+
+++
+
+
++
+
++ +
++
+
+++
+
++
+
+
+
+
+
++
++
+
++
+
+
+
+
+
++
+
+
+
+
+
+
+ +
+
+
+++
+
+
+
+
+
+
+
+
+
+
+
Initial angle
Initialheight(mm)
0 0.5 1 1.5 2 2.5 3
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
φ (radians)
Smoothslot: cfd2 Vs=2.4 m/s
+
++
+
+
++
+
++
++++
++
+
+
++
+
+
+
+
+
+
+
++
+
+++
++
+
+
+++
+
+
+ +
++
++
+
++++
+
+
+++
+
+
++
+
+
+ +
+
+
+
+
+
+
+
+
+ +
+ ++
+
+
+
++
++
+
+
+
+
+
+
+
+
+
++
+
+
++
+
+
+
+
+
+
+ +
++
+
+
+
+
+ +
++
++
+
+
+
+
+
++
++
++
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+ +
+
+
++
+
+
+
+
+
+
+
+
+
++
+
+
+
++
+
++
+
+
+
+
+
+
++
+
+
++
+
+
+
+
+
+
+
+
+ +
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
++
+
+
+
+
++
+
+
+
+
+
+
+
+
+
+
+
++
++
+
+
+
+
+
+
+
+
+
+
++
+
+
+
+
+
+
+
+
+
+
+
+
+
+
++
+
+ ++
+
++
+
+
+
+
+
+
+
+
+
+
+
+
+ +
+
+
+
+
+
++
+
+
+
+
+
++
+
+
++
+
++
+
+
+
+
++
+
+
+
+
+
+
+
+
+
++ +
+
+
+
+
+ +
+
++
+
+
+
+
+
+
+
+
+
+
+
+
+
++
+
++
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
++
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
++
+
+
+
++
++ +
+
+
+
++
++
+
+
+
+
+
++
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
++
+
+
+
+
++
+
+
+
+
+
+
+
++
+
+
+
++
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+ +
+
++
+
+
+
+
++
+
+
++
+
+
+
+
++
+
+
+
++
+
+
++
+
+
+
++
+
+
+
+
+
+
+
+
+
+
+
+
+ ++
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
++
++
+
+
+
+
+
+++
+
+
+
+
+
+
+
+
+
+
+
+
++
+
++
+
+
+
+
++
+
+
+
+
+
+
+
+
+
+
+++
+
+
++
+
+
+
+
+
+
++ +
+
+
+
+
+
++
+
+
+
+
+
+ +
+
+
+
+
+
+
+
+
+
+
+
++
+
+
+
+
+
+
+
++
+
+
+
+
+ +
+
+
+
+
+
+
+
++
+
++
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
++
+
++
+
++
+
+
+ +
+
+
+
+
+
+++
+
++
+
+
+
+
+
+
+
+
+
+
++
+
+
+
+++
+
+
+
+
+
++
+
+
+
+
+
++
+
++
+
++ ++++
+
+
+
+
+
+
+
++
+
+
+
+
++
+
+
+
+
++
++
+
+
+
+
++
+
+
+
++
+
+
+
++
+
+
+
+
+
+
+
+
+
+
+
++
+
+
+
+
+
+
+
+
+
++
+
+
+
+
+
+
+
+
+
+++
+
+
++
+
+
+
+
+
+
+
+
+
+ +
++
+
+
+
++
+
+
+
+
+
+
+
++
+
+
+
+
+
+
+
++ +
+
+
+
++
+
+
++
++
+
+
+
+
+
+ +
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
++++
+
+
+
+
+
+
+
+
+ +
+
+
+
+
+
+
+
+
+
+ +
+
+
+ ++
+
+
+
+
+
+
+
+
+
+ +
+
+
+
+
+
+
+
+
+
++
+
++
+
+
+
+
+
+
+
+
+
+++
+
+
+
+
+
+
+
+
+
+
+
+
+ +
+
+
+
++
+
+
+
+
+
+
+
+
++ +
+
+
++
+
+
+
+ ++
+
+
+ +
+
+
++
+
+
++
+
+
+
+
+
++
++
+
+
++
+
++
+
++
++
+
+
+
++
+
+
++
+
+
+
+
+
+
++
+
+
+
+
+
+
+
+
++
+
+
+
+
+
+
+
+
+
+
+
+
+
++
+
+
+
+
+
+
+
+
+
++
+ +
+
+
+
+
++
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
++
+
+
++
+
+
+
+
+ +
+ ++
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+ +
+++
+
+
++++
+
+
+
++
+
+
+
+
+
++
+
+
+
++
+
++
+
+
++
++
+
+
+ ++
+
+
+
+
+
+
++
+
++
+
+
+
+
+
+
+
+
+
+
+
+
+
+
++
+
+ +
+
++
+
+
+ +
+
++
+
+
+
+
+
+
+
+
+
+
+
+
++
+
+
+
+
+
+
+
+
+
+
+
++
+
++
+
+
+
+
+
+
+
+
+
+
++
++
+
+
+
+
+
+
++
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
++
+
+
+
+
+
+
+ +
+
++
+
+
+
+
+
+
++
++
+
++ +
+
+
+
++
+
+
+
++
+
+
+
+
+ +
++
+
+
+
+
+
+
+
+
+
+
++
+
++
+
++
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
++
+
+
++
++
+
+
+
+
+
+
+
++
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
Initial angle
Initialheight(mm)
1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2 2.1
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
θ (radians)
Smoothslot:cfd5 Vs=5.4 m/s
+
++
+
+
++
+
+ +
++++
++
+
+
++
+
+
+
+
+
+
+
+ +
+
++
+
++
+
+
+++
+
+
+ +
++
+
++
+
+
+
+
+++
+
+
++
+
++
+
+
+
++
+
+
+
++ +
+
+
+
+
+
+
+
+
+
+
+ +
++ +
+
+
+
+ +
++
+
+
+
+
+
+
+
+
+
++ +
+
+
++
+
+
+
+
+
+
++
++
+
+
+
+
++
++
+ +
+
+
+
+
+
++
++
++
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
++
+
+
++
+
+
+
+
+
+
+
+
+
++
+
+
+
++
+
++
+
+
+
+
+
+
++
+
+
++
+
+
+
+
+
+
+
+
+ +
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
++
+
+
+
+
+
++
+
+
+
+
+
+
+
+
+
+
+
++
+ +
+
+
+
+
+
+
+
+
+
+
+
+ +
+
+
+
+
+
+
+
+
+
+
+
+
+
+
++
+
+++
+
++
+
+
+
+
+
+
+
+
+
+
+
+
+ +
+
+
+
+
+
++
+
+
+
+
+
++
+
+
++
+
++
+
+
+
+
++
+
+
+
+
+
+
+
+
+
+++
+
+
+
+
+ +
+
++
+
+
+
+
+
+
+
+
+
+
+
+
+
++
+
++
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
++
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
++
+
+
+
++
+++
+
+
+
++
++
+
+
+
+
+
++
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
++
+
+
+
+
++
+
+
+
+
+
+
+
++
+
+
+
+ +
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+ +
+
++
+
+
+
+
++
++
++
+
+
+
+
++
+
+
+
++
+
+
++
+
+
+
++
+
+
+
+
+
+
+
++
+
+
+
+++
+
+
+
+
+
+
+
+
+
+
++
+
+
+
+
++
++
+
+
+
+
+
+++
+
+
+
+
+
+
+
+
+
+
+
+
+
++
++
+
+
+
+
+
++
+
+
+
+
+
+
+
+
+
+
++ +
+
+
++
+
+
+
+
+
+
++ +
+
+
+
+
+
+ +
+
+
+
+
+
++
+
+
+
+
+
+
+
+
+
++
++
+
+
+
+
+
+
+
++
+
+
+
+
+ +
+
+
+
+
+
+
+
++
+
++
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
++ +
+ +
+
++
+
+
++
+
+
+
+
+
++
+
+
++
+
+
+
+
+
+
+
+
+
+
++
+
+
+
+
+++
+
+
+
+
++
+
+
+
+
+
+
++
+
++
+
++ +
+ +++
+
+
+
+
+
+
++
+
+
+
+
++
+
+
+
+
++
++
+
+
+
+
++
+
+
+
++
+
+
+
++
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
++
+
+
+
+
+
+
+
+
+
+
++ +
+
+
++
+
++
+
+
+
+
+
+
+ +
++
+
+
+
+
+
+
+
+
+
+
+
+
+
++
+
+
+
+
+ +
+
+
+
++ +
+
+
+
++
+
+
++
+ +
+
+
+
+
+
++
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+ +++
+
+
+
+
+
+
+
+
++
+
+
+
+
+
+
+
+
+
+ +
+
+
++ +
+
+
+
+
+
+
+
+
+
+ +
+
+
+
+
+
+
+
+
+
++
+
++
+
+
+
+
+
+
+
+
+
++ +
+
+
+
+
+
+
+
+
+
+
+
+
+ +
+
+
+
+ +
+
+
+
+
+
+
+
+
+ ++
+
+
+
+ +
+
+
+
+ ++
+
+
+ +
+
+
+
++
+
+
++
+
+
+
+
+
++
++
+
+
++
+
++
+
++
+ +
+
+
+
++
+
+
++
+
+
+
+
+
++
+
+
+
+
+
+
+
+
+
++
+
+
+
+
+
+
+
+
+
+
+
+
+
++
+
+
+
+
+
+
+
+
+
++
+ +
+
+
+
+
+ +
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
++
+
+
++
+
+
+
+
++
+++
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
++ +
+
+
+
++++
+
+
+
+ +
+
+
+
+
+
++
+
+
+
++
+
++
+
+
++
++
+
+
+++
+
+
+
++
+
+
+
++
+
++
+
+
+
+
+
+
+
+
+
+
+
+
+
+
++
+
++
+
+ +
+
+
++
+
++
+
+
+
+
+
+
+
+
+
+
+
+
++
+
+
+
+
+
+
+
+
+
+
+
+
++
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+ +
++
+
+
+
+
+
+
++
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
++
+
+
+
+
+
+
++
+
++
+
+
+
+
+
+
++
++
+
++ +
+
+
+
++
+
+
+
+
+
++
+
+
+
+
++
++
+
+
+
+
+
+
+
+
+
+
+
+ +
+
++
+
++
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
++
+
++
+
+
++
++
+
+
+
+
+
+
+
+
++
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
Initial angle
Initialheight(mm)
0 0.5 1 1.5 2 2.5 3
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
φ (radians)
Smoothslot:cfd5 Vs=5.4 m/s
Figure 8e (top left) 8f (top right) 8g (bottom left) and 8h (bottom right) Smooth Slot: The effects of
initial height and orientation θ , φ on passage of 3mm rigid fiber for slot velocities 2.4 m/s, 5.4 m/s.
8 TTAAPPPPII JJOOUURRNNAALL MAY 2004
PEER-REVIEWED SPECTROSCOPY
Dissolving pulp is produced in variousgrades to be used in manufacturing
several specialty products like rayon,pharmaceuticals, and photographic films.A number of quality control tests are per-formed on the finished, dried pulp prod-uct in analytical laboratories to monitorpulp quality for every grade manufac-tured.
Some of these laboratory tests involvecomplex and lengthy wet chemistry pro-cedures. Furthermore,most tests are natu-rally destructive to the solid pulp samplessince they entail dissolving the sample inalkaline or acidic solutions. As a result,multiple samples are required for a com-plete analysis of pulp quality.
A novel procedure for testing the fin-ished product is based on multivariateimage analysis (MIA) of near-infrared(NIR), multi-spectral images of pulp sam-ples. Information extracted from theimages is then statistically modeled topredict relevant pulp properties.
NIR SPECTROSCOPYTraditional probe detectors
NIR spectroscopy measures the responseof materials to light in the wavelengthrange of 700–2500 nm. The light isabsorbed and converted to energy mainlyin materials with characteristic functionalgroups such as C–H, O–H, N–H, and C=O.NIR is very sensitive to moisture, whichhas a signal that can sometimes be strongenough to hide the signals of other func-tional groups.
Traditional NIR spectrometers consistof a highly sensitive fiber-optic probedetector that provides precise NIR spec-tra by averaging over a small area of the
933–1787 nm. The back of the NIR camera consisted of an indium–galli-um–arsenide (InGaAs) charge coupleddevice array capable of acquiring intensi-ty images with a resolution of 128 × 128pixels. Consisting of a narrow horizontalentrance slit, the spectrograph dispersedthe entering line of light into 128 hori-zontal lines spanning the spectral rangeof 933–1787 nm.
The NIR camera recorded the 128 dis-persed lines as a two-dimensional intensi-ty image, where the horizontal axis (x)was the spatial dimension and the verti-cal axis was the spectral (l) dimension ofthe sample being imaged [11]. For thisthree-dimensional image, we obtainedthe second spacial dimension (y) by cap-turing multiple lines across a movingsample under the spectrograph–cameraassembly. This setup is ideal for samplesthat are produced on moving webs, suchas pulp and paper samples.
MULTIVARIATE IMAGEREGRESSION OF PULP NIR
IMAGESOverview of MIA
Data from imaging spectrometers can beviewed as multivariate images, wherethe NIR spectrum represents the variabledimension. A multivariate image isdefined as a stack of congruent images[12], where each image is measured for adifferent wavelength, frequency, or ener-gy.
FFiigguurree 11 shows an example of a 512 ×512-pixel multivariate image acquired atfour different wavelengths arranged as a512 × 512 × 4 cube (XX). At each pixellocation in XX, there is high correlation in
sample being tested. This form of NIRspectroscopy has been extensively usedto characterize samples in the pharma-ceutical, food, and chemical industries formany years [1].The pulp and paper indus-try has recently discovered its potentialas a means to achieve the goals of rapidquality testing in a nondestructive man-ner [2, 3]. In the pulp and paper industry,NIR spectroscopy is used to predict qual-ity indicators such as kappa number [4],pulp yield, and cellulose content [5–8].NIR spectroscopy has been well testedand thoroughly documented [9, 10].
NIR imaging spectroscopyA shortcoming of probe-based NIR spec-trometers is that they cannot providesimultaneous multiple-point readingsacross a solid sample. NIR imaging spec-trometers were introduced to compen-sate for this shortcoming.These imagingspectrometers work on the same princi-ples as their traditional counterparts.Thedifference lies in the detector, which inour case is a digital camera with a light-diffracting spectrograph attached. Thecamera captures reflected light from thesample as an intensity image at uniqueNIR wavelengths.The scanned section ofa sample is recorded as a three-dimen-sional, multi-spectral, digital image. Insuch images, the digitized NIR spectrumis defined in the third dimension (wave-length) for each pixel location across thesample.
The NIR imaging spectrometer con-sisted of a light-diffracting prism–grat-ing–prism (PGP) spectrograph attachedbetween the front objective lens and theback of an NIR camera.The unit was sen-sitive to light in the spectral range of
Using near-infrared multivariate image regression to predict pulp properties
MANISH H. BHARATI, JOHN F. MACGREGOR, AND MARC CHAMPAGNE
ABSTRACT: Traditional laboratory testing of dissolving pulp is labor intensive and time consuming.Multivariate image regression (MVIR) offers a testing method that is much more rapid than the analyticalchemistry approach. Pulp properties are predicted from multi-spectral, near-infrared (NIR) images of finishedpulp. A single test sample can be used to predict four indicator variables for the quality of pulp—S10, S18,DCM resin, and intrinsic viscosity. The testing approach provides a framework for studying pulp heterogene-ity through deriving spatial pulp property distribution across the imaged section of a pulp sample.Preliminary test results have been promising for both off-line and at-line studies.
AApppplliiccaattiioonn:: Product quality can be monitored efficiently by modeling data from NIR multi-spectralimages of finished pulp.
VOL. 3: NO. 5 TTAAPPPPII JJOOUURRNNAALL 9
SPECTROSCOPY
pixel intensities because each point of the captured scene isdefined by multiple pixels in the wavelength dimension.
Multivariate images usually contain an enormous amount ofdata (e.g., XX contains 512 × 512 × 4 = 1,048,576 pixel intensi-ties), making the computational analysis intensive. Multivariatestatistical methods like multi-way principal component analysis(MPCA) have proven to be ideal tools for efficiently reducingdata and analyzing multivariate images in a reduced dimension-al latent variable subspace [12]. Figure 1 illustrates the MPCAdecomposition of the three-dimensional image array XX (uponunfolding to a two-dimensional matrix XX) into a linear combi-nation of two scores (tt
1, tt
2) and two loadings (pp
1, pp
2) plus noise
(EE). Details about MPCA decomposition and MIA of XX are pro-vided elsewhere [12, 13].
MIA for feature extraction of NIR multi-spectral
imagesUsing NIR imaging spectrometry, we captured multi-spectralimages of 60 dissolving pulp samples from two main pulpgrades. The dissolving pulp grades were based on a bleachedsoftwood mixture of pine and spruce. The pulp was made bythe sulfite method. For confidentiality reasons, the pulp proper-ty values were scaled to a relative index to preserve the raw val-ues,but the trends of the properties were retained in the resultspresented.
A pulp surface area of 93 mm (W) × 140 mm (L) was cap-tured as intensity images measuring 102 pixels (W) × 448 pix-els (L) in 108 unique wavelengths spanning the 933–1650-nmNIR range. Thus, for purposes of MIA the dimensions of eachmultivariate image were 102 × 448 × 108 pixels. We decom-posed each sample image using MPCA to produce a one-dimen-sional MIA score and loading space.
The MIA scores can be used if one is interested in the spatialvariations of feature pixels throughout the imaged area of thepulp sample. However, our main aim was to derive overall qual-ity data from NIR multi-spectral images of the pulp samples.Wedetermined the pulp quality indicators through wet-chemistrylaboratory testing.
The MIA loadings indicate the NIR spectral variations of pix-els throughout the scanned area of the pulp sample.Thus, theloading vectors provide overall information about chemical fea-tures (related to signatures of various functional groups) in theimaged pulp sample. The first MIA loading vector pp
1of each
pulp sample image was used as its feature vector,which we ana-lyzed further to infer pulp quality. No data scaling was appliedto the multivariate images prior to MPCA decomposition forMIA [12].
Hence, pp1
was a normalized mean NIR spectrum of the45,696 unique NIR spectra over the scanned area of the pulpsample (102 × 448 pixels). This feature vector is similar to apoint reading of a traditional NIR probe. In both cases, the fea-ture vector is an averaged NIR spectrum. The main differencebetween the two techniques lies in the spectral averaging. Asopposed to point averaging in NIR probe instruments, the MIAloading vectors are averaged spectra from multiple pointsacross the imaged area of the sample.
When MPCA is performed on the NIR multi-spectral images,weights of the pp
1loading vector can be plotted against their
respective variables (wavelengths). FFiigguurree 22 shows the pp1
fea-
ture vectors of the 60 spectral images of the 60 pulp samples.This new feature vector set represents a feature space (XX
ffeeaattuurree),
which can be used as regressor vectors to build predictive mod-els for pulp quality.
The pp1
feature vectors in the figure indicate that the higherwavelengths (1400–1650 nm) generally exhibit low NIRreflectance, or high absorbance, whereas the lower wave-lengths (933–1300 nm) exhibit high reflectance.The regions ofhighest varaition among the feature vectors are in the wave-length ranges from 933 to 1100 nm and from 1400 to 1650 nm.
0.065
0.070
0.075
0.080
0.085
0.090
0.095
0.100
0.105
0.110
0.115
933 979 1025 1071 1118 1164 1211 1258 1304 1351 1399 1446 1493 1541 1588 1636
WAVELENGTH, nm
2. Weights of MIA p1
loading vectors of 60 NIR multi-spec-
tral images of pulp.
X
512
512
4PCA
+
t1 * p1'
(512 * 512) x 1
1 x 4
+
t2 * p2'
1 x 4
E4
(512 * 512)
X
512
512
4
Reorganization
512 * 512
X
4
(512 * 512) x 1
1. MPCA decomposition of a 512x512x4 multivariate image
into a reduced dimensional latent variable subspace.
1X
108
448 102
1
60
1081
60
Feature Extraction
p1Loadings
of MIA
Y
OSC+PLS
1
X
feature
3. Scheme of the MVIR strategy for predicting laboratory
data (Y) from NIR multi-spectral pulp images (X).
10 TTAAPPPPII JJOOUURRNNAALL MAY 2004
SPECTROSCOPY
MVIR modeling of MIA feature
space to predict pulp quality The feature vectors extracted from NIRmulti-spectral pulp images can be furtheranalyzed to predict overall pulp qualitydata.We performed this analysis via par-tial least squares (PLS) regression model-ing between the feature space and corre-sponding pulp quality data from labora-tory testing. PLS modeling has been pop-ular in analytical chemistry for multivari-ate calibration of NIR spectra. Oncetrained, these models can be used to pre-dict analyte concentrations from NIRspectra of new samples [14, 15].
The overall MVIR scheme is illustrat-ed in FFiigg.. 33. The MIA feature vectors(XX
ffeeaattuurree) were individually regressed with
four pulp quality variables (YY = S10, S18,DCM resin, and intrinsic viscosity). Fordissolving pulp grades, these four param-eters are the key values for customerswho manufacture rayon, pharmaceuti-cals, photographic films, etc. These vari-ables describe the solubility of hemicel-luloses (S10 and S18), the measure ofextractable materials usingdichloromethane solvent (DCM resin),and the average length of the molecularchains of polymers making up pulpfibers (intrinsic viscosity).
Tembec Inc. previously performed apreliminary analysis to empirically pre-dict pulp properties from NIR spectraobtained through a spectroscope with aprobe detector [16]. PLS regression mod-eling was used to predict pulp propertiesfrom NIR spectra.The NIR spectra wereorthogonal signal corrected (OSC) priorto PLS modeling.We also used OSC of theMIA feature vectors prior to PLS model-ing for predicting pulp properties.
OSC is a spectral filtering technique[17, 18] for removing systematic varia-tions from MIA feature vectors (XX
ffeeaattuurree)
that are unrelated to (i.e. orthogonal to)pulp properties (YY).This type of filteringis particularly important because the NIRspectra are strongly affected by moisturein the pulp. Other systematic variationsin XX
ffeeaattuurreethat are unrelated to YY include
surface effect variations in images of dif-ferent samples and light variationsbetween imaging runs. By subtractingthe OSC components from XX
ffeeaattuurree, we
obtain a filtered predictor matrix, whichcontains the variations of interest thatare related to the variations in YY. (Anexcellent introduction to OSC of NIRspectra is offered by Wold et al. [18].)
We found it optimal to remove twoOSC components, based on the predic-tive ability of the filtered model. Theamount of information removed by OSCcan be more fully appreciated by observ-ing the raw and filtered feature vectors ofthe pulp NIR images.
The graph in FFiigg.. 44aa illustrates featurevectors extracted via MIA for the NIRmulti-spectral images of two pulp sam-ples from the “at-line” imaging study.FFiigguurree 44bb illustrates the signal-correctedfeature vectors after filtering with twoOSC components using YY = intrinsic vis-cosity.
By comparing the magnitudes of thecoefficients in the two feature vectorsbefore and after filtering, we see thatOSC removes most of the variation inXX
ffeeaattuurree. Furthermore, OSC removes the
trends in the feature vector coefficientswith respect to the NIR wavelengthspectrum. Subtle differences betweenthe feature vectors are enhanced in cer-tain wavelengths, indicating higher infor-mation content in specific regions of theNIR spectrum. Such information is relat-ed primarily to the intrinsic viscosityvariations in the dataset.
To develop the empirical models, wethen individually PLS regressed the OSC-filtered feature space with respect to thefour different pulp properties with therespective laboratory-measured pulpproperty data
Regression modeling details
and resultsIn an “at-line”imaging study conducted atthe pulp mill, samples from the end ofeach roll were imaged immediately afterproduction and prior to laboratory test-ing for pulp quality. The following is adescription of the regression modelingand results from this “at-line” imagingstudy.
Sixty pulp samples were imaged in atime sequence over three days at thepulp mill.These samples belonged to twomain grades,Grades 1 and 2.Grade 1 wasfurther divided into two sub-grades, Aand B, because of different specificationson some of the pulp properties.Laboratory tests for the four pulp prop-erties were performed on the pulp sam-ples after imaging.The NIR multi-spectralimages of the pulp samples were decom-posed via MPCA into pp
1loading feature
vectors and were assembled as observa-tions (rows) of XX
ffeeaattuurree(Fig. 3).
Two equal divisions of the completedataset were made (30 samples each) for(a) training the multivariate regressionmodel to fit the YY data (Observations 1,3,…,49 of XX
ffeeaattuurree) and (b) using the devel-
oped model on a validation dataset(Observations 2, 4,…, 50) for predictingthe pulp properties and calculating pre-diction errors. Alternate samples weredivided into training and validation setsmainly to accommodate the analyticallaboratory.The laboratory procedure formeasuring intrinsic viscosity requiredonly 20 min.As a result, this property wasthe only pulp property that was meas-
4. Effect of OSC filtering on pulp image feature vectors. (a) Raw feature vectors of two sample images. (b) Filtered feature
vectors upon removing two OSC components (Y = intrinsic viscosity).
WAVELENGTH, nm
FE
AT
UR
E V
EC
TO
RC
OE
FF
ICIE
NT
S
0.07
0.075
0.08
0.085
0.09
0.095
0.1
0.105
0.11
0.115
933 986 1038 1091 1144 1198 1251 1304 1358 1412 1466 1520 1575 1629
a
-0.0005
-0.0004
-0.0003
-0.0002
-0.0001
0.0000
0.0001
0.0002
0.0003
0.0004
0.0005
0.0006
0.0007
933 986 1038 1091 1144 1198 1251 1304 1358 1412 1466 1520 1575 1629
WAVELENGTH, nm
b
FIL
TE
RE
D F
EA
TU
RE
VE
CT
OR
CO
EF
FIC
IEN
TS
VOL. 3: NO. 5 TTAAPPPPII JJOOUURRNNAALL 11
SPECTROSCOPY
ured for all 60 pulp samples from the at-line study. The otherthree pulp properties (S10, S18, and DCM resin) were measuredfor only half of the dataset (i.e. only for the training dataset)since their wet-chemistry procedures were more tedious andtook much longer. Because three property measurements werenot available in the validation set, PLS model prediction errorsare not available for S10, S18, and DCM resin.
The NIR spectral data in XXffeeaattuurree
were mean-centered withrespect to the 108 wavelengths (columns) prior to modeling.Each of the four pulp property variables in YY was mean-cen-tered and auto-scaled to unit variance. DCM resin values weretransformed by taking the natural logarithm, expressed asln(DCM resin). Beebe et al. [19] and Eriksson et al. [15] provideinsights to different types of variable scaling and variable trans-formations that can be performed to remove nonlinearity andimprove the PLS model fit and predictions.
When the DCM resin values were transferred into ln(DCMresin), the corresponding relative experimental errors (coeffi-cient of variation = ±6.3%) were also stabilized to absoluteerrors (2s experimental error = ±0.13). Separate PLS regressionmodels were developed for each of the four pulp properties in
the training dataset.The corresponding models were then usedto predict pulp property values from feature vectors in the val-idation dataset.The errors between pulp property values fromthe analytical laboratory tests and PLS model predictions fromthe training and validation datasets (for intrinsic viscosity) werethen calculated.
Experimental Error
SAMPLE NUMBER
Lab measurementsOSC+PLS model predictionsfor validation set
S10
GRADE 1a GRADE 1b
Gra
de C
han
ge
27
28
29
30
31
32
33
34
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
GRADE2
6. Time series plot of S10 pulp property experimental data
and OSC + PLS regression model predictions.
Experimental Error
SAMPLE NUMBER
Lab measurementsOSC+PLS model predictionsfor validation set
S18
GRADE 1a GRADE 1b
Gra
de
Ch
an
ge
GRADE2
10.75
11.00
11.25
11.50
11.75
12.00
12.25
12.50
12.75
13.00
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
7. Time series plot of S18 pulp property experimental data
and OSC + PLS regression model predictions.
I. Multivariate regression model results for four pulp property variables.
PULP PLS EXPERIMENT ERROR,PROPERTY COMPONENTS R2
cumQ2
cumRMSEE RMSEP 2 std. dev.
Intrinsic viscosity 3 0.929 0.894 0.289 0.451 0.223S10 1 0.891 0.813 0.545 — 0.600S18 2 0.974 0.964 0.068 — 0.300ln(DCM resin) 2 0.76 0.715 0.596 — 0.126
5. Time series plot of intrinsic viscosity experimental data and OSC + PLS regression model (a) fit to the training dataset
and (b) prediction of the validation dataset.
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59
GRADE 1a
Gra
de
Ch
ang
eGRADE 1b
GRADE2
Experimental Error
a
SAMPLE NUMBER
Lab measurementsOSC+PLS model predictions
INT
RIN
SIC
VIS
CO
SIT
Y
Experimental Error
b
SAMPLE NUMBER
Lab measurementsOSC+PLS model predictions
INT
RIN
SIC
VIS
CO
SIT
Y
2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 56 58 60
GRADE 1a GRADE 1b GR
AD
E 2
Gra
de
Ch
ang
e
Reported Error
Experimental
SAMPLE NUMBER
Lab measurementsOSC+PLS model predictionsfor validation set
In, D
CM
resin
GRADE 1a GRADE 1b
Gra
de
Ch
an
ge
GRADE2
-19
-18
-17
-16
-15
-14
-13
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 5960
8. Time series plot of ln(DCM Resin) pulp property experi-
mental data and OSC + PLS regression model predictions.
12 TTAAPPPPII JJOOUURRNNAALL MAY 2004
SPECTROSCOPY
An overall measure of the quality of aregression model can be determined bycomparing its root mean squared errorof fit (RMSEE) and its root mean squarederror of prediction (RMSEP) with theexperimental errors from laboratorytests of pulp properties. If the RMSEE andRMSEP are similar to experimentalerrors, the regression models can fit andpredict the pulp property at least as wellas the laboratory testing can.The resultsof four OSC + PLS multivariate regressionmodels used for the pulp properties ofinterest are summarized in TTaabbllee II.
In Table I, R2cum
is the cumulative sumof squares of YY explained by the fittedPLS regression model. Q2
cumis the per-
cent sum of squares that can be predict-ed as determined by cross-validation[20]. If Q2
cumis close to R2
cumand both are
close to 1.0, then the model is consid-ered to be highly predictive. RMSEE isdefined as the root mean squared errorof estimation for the fitted data using thePLS model:
(1)
where y^ and y are the predicted andobserved values of the response variable,respectively, and N is the number of fit-ted data points. RMSEP is calculated inthe same manner as RMSEE, except thatit is defined as the root mean squarederror of prediction for the data not usedin the model building stage (i.e., on thevalidation dataset) [21].
FFiigguurree 55 shows the time series plotsof the training and validation datasets ofthe intrinsic viscosity pulp property col-lected during the “at-line” study.As theseplots show,the predictions closely followthe trends of the laboratory test datafrom Grade 1A to Grade 2.The error bararound the “target value” (dash-dot line)gives the 2s experimental error limit forthe laboratory measurements of intrinsicviscosity. The model fit and predictionsseem generally within the limits definedby the error bar. The OSC + PLS modelpredictions not only follow the generaltrends of intrinsic viscosity betweengrades,but the model can also detect anyprocess upsets within a grade (e.g., thetemporary decrease in intrinsic viscosityaround Sample 28 in Grade 1A).
FFiigguurree 66 illustrates a time series plotof the S10 pulp property. PLS model pre-dictions for the validation set have beenoverlaid with the training set laboratorymeasurements as part of a common timeseries plot. Actual laboratory measure-ments for the validation set are not avail-able for the S10 pulp property.As the fig-ure shows, the OSC + PLS model per-forms reasonably well in following theS10 pulp property trends from Grade 1Ato Grade 1B to Grade 2.Furthermore, thismodel can also detect the suddenincrease in the S10 value for Grade 1Aaround Samples 27 and 28. Thus themodel performs well both within andbetween the grades studied.The experi-mental error bar (±0.6) is included toshow the model’s ability to predict thepulp property values for the validationset.
Similar time series plots are shown inFFiiggss.. 77 aanndd 88 for S18 and ln(DCM resin)pulp properties, respectively.The chosenOSC + PLS regression model for S18 alsoseems to perform reasonably well in fol-lowing the pulp property trends fromGrade 1A to Grade 1B to Grade 2.
As Fig. 8 shows, the PLS model pre-dictions for ln(DCM resin) are not asgood as those for the other three pulpproperties. This discrepancy could beattributed to the noisy laboratory meas-urements of the DCM resin pulp proper-ty for Grade 1B and Grade 2.As illustrat-ed in the figure, the laboratory measure-ments (solid diamonds) for these gradesexhibit a bouncing pattern between –15and –18 for Samples 37–60. These inac-curacies may contribute to a bad trainingset for the PLS models,which is reflectedin the poor predictions of ln(DCM resin)for the validation set.
EXTRACTING SPATIAL DISTRIBUTIONS USING
MVIR MODELSThe main objective of the MVIR pulpproperty modeling study was to relatemulti-spectral NIR images of finishedpulp samples to their average properties.We did not use any spatial informationfrom the pulp sample images in theMVIR calibration models. Instead, wetried to extract spatial information of apulp sample through the application ofthe overall MVIR model on subsectionsof its multi-spectral NIR image [22].Doing so would enable us predict thespatial variations in pulp propertiesthroughout the sheet and would there-fore provide a measure of the hetero-geneity of the sample with respect toeach property.This advantage is a poten-tially important one for the two-dimen-sional NIR imaging spectrometer over atraditional NIR instrument, which uses asingle point as a source.
When the multi-spectral NIR image issegmented into various subsections andthe overall MVIR model is applied to pre-dict local pulp properties for each sub-section, one can obtain a spatial distribu-tion of predicted pulp propertiesthroughout a pulp sample. Such a distri-bution can be plotted as a two-dimen-sional histogram of property predictions.The histogram gives an idea of the meanand variance of a pulp property as it isdistributed through a particular sampleimage. The mean property predictiongives an overall measure (similar to theoverall MVIR model predictions), where-as the variance provides a measure ofpulp heterogeneity with respect to theproperty being predicted.
( ) y 2∑ −= NyRMSEE
9. Intrinsic viscosity distribution
across seven subsections of NIR multi-
spectral image of Sample 16 from the
at-line imaging study.
VOL. 3: NO. 5 TTAAPPPPII JJOOUURRNNAALL 13
SPECTROSCOPY
Since the NIR imaging spectrometerscans the pulp sample as a line-scan cam-era, if one were to segment the pulpimage into subsections containing sever-al line scans, the corresponding predic-tion histogram could be used to monitorchanges in pulp properties in an on-linefashion. In a typical monitoring scheme,one could track both the mean and thevariance of the pulp properties.
The idea is illustrated by calculatingthe intrinsic viscosity distribution acrossSample 16 from the “at-line” imagingexperiment.The intrinsic viscosity value(along with 2s experimental errors) forthe sample as measured in the laboratorywas 25.03±0.22, whereas the MVIR pre-diction using the OSC + PLS regressionmodel was 25.14.The multi-spectral NIRimage of the entire pulp sample was seg-mented into seven subsections, witheach subsection covering the full line-scan width of the original pulp sampleimage.
Using the overall OSC + PLS MVIRmodel coefficients within every imagesubsection, we obtained intrinsic viscosi-ty predictions for seven subsections ofthe scanned surface of the pulp sample.The size of image segments chosen willbe a tradeoff between obtaining finerspatial detail with finer segmentationand increased variance of the predictionsin each subsection as result of fewerimage pixels and local variations in light-ing [22]. (The distribution of pulp prop-erty predictions across the same pulpsample image segmented into 48 subsec-tions is illustrated elsewhere [23].)
To investigate the variation of theproperties throughout the sheet, theoverall OSC + PLS MVIR model can beapplied to any image subsection toobtain predictions for that subsection ofthe image. The mean intrinsic viscosityprediction over the seven subsections is24.87, and the standard deviation of thepredictions across the seven subsectionsis 0.549. The physical dimension of thepulp sample area scanned by the imag-ing spectrometer is 140 mm (L) × 93 mm(W). Compared to the width of an indus-trial pulp web, the imaging spectrometercovers less than 1% of its true horizontaldimension. As a result, it would be rea-sonable to assume very little variability inthe spatial distribution of predictedintrinsic viscosity across the seven sub-
sections of the multi-spectral NIR pulpimage.
CONCLUSIONSFFiigguurree 99 shows the distribution of intrin-sic viscosity predictions obtained byapplying the PLS model coefficients toseven subsections. A multivariate imageregression technique was used to predictpulp quality data from NIR multi-spectralimages of pulp samples.The MVIR tech-nique can be used to relate feature infor-mation from multivariate images of fin-ished products to corresponding dataobtained from the quality control lab.Once trained, the MVIR model can thenbe used to infer quality data from prod-uct images in on-line industrial processesequipped with multivariate imaging sen-sors.
The proposed technique has beentested in a feasibility study at a pulp millwith promising results. The empiricalmodeling technique produces adequatepredictions of three out of the four pulpquality indicators studied. With currenttrends in high-speed camera technology,the proposed image regression schemehas the potential for on-line monitoringof pulp quality through NIR multi-spec-tral imaging spectroscopes mounted onpulp machines. TJ
LITERATURE CITED1. Stark, E., and Luchter, K., Appl.
Spectrosc. 22(4): 335(1986).2. Wallbäcks, L., Edlund, U.,
Lindgren, T., and Agnemo, R.,Nordic Pulp Paper Res. J. 2:88(1995).
3. Antti, H., Sjöström, M., andWallbäcks, L., J. Chemometrics 10:591(1996).
4. Birkett, M. D., and Gambino, M.J.,TAPPI J. 72: 193(1989).
5. Schultz, T.P., and Burns, D. A.,TAPPI J. 73: 209(1990).
6. Wright, J.A., Birkett, M.D., andGambino, M.J., TAPPI J. 73:164(1990).
7. Wallbäcks, L., Edlund, U., andNorden, B., TAPPI J. 74: 201(1991).
8. Woitkovich, C.P., McDonough, T.J.,and Malcolm, E.W., TAPPI 1994Pulping Conference Proceedings,TAPPI PRESS, Atlanta, p. 721.
9. Whetsel, K., Appl. Spectrosc. 2(1):1(1968).
10. Handbook of Near-InfraredAnalysis (D.A. Burns and E.W.
Ciurczak, Eds), Marcel Dekker, NewYork, 1992.
11. Herrala, E., and Okkonen, J., Int. J.Pattern Recognition and ArtificialIntelligence 10(1): 43(1996).
12. Geladi, P., and Grahn, H.,Multivariate Image Analysis, JohnWiley and Sons, Chichester, UK,1996.
13. Bharati, M.H., and MacGregor, J.F.,Ind. Eng. Chem. Res., 37:4715(1998).
14. Martens, H. and Næs, T.,Multivariate Calibration, JohnWiley and Sons, Chichester, UK,1989.
15. Eriksson, L., Johansson, E.,Kettaneh-Wold, N., and Wold, S.,Multi- and Megavariate DataAnalysis, Umetrics AB, Umeå,Sweden, 2001.
16. Champagne, M., Meglen, B., Wold,S., and Kettaneh-Wold, N., PulpPaper Can. 41(2001).
17. Svensson, O., Kourti, T., andMacGregor, J.F., J. Chemometrics16: 176(2002).
18. Wold, S., Antti, H., Lindgren, F.,and Öhman, J., ChemometricsIntell. Lab. Syst. 44: 175(1998).
19. Beebe, K.R., Pell, R.J., andSeasholtz, M. B., Chemometrics: APractical Guide, John Wiley andSons, New York, 1998.
20. Wold, S., Technometrics 20:397(1978).
21. Umetrics, SIMCA-P 9.0, UmetricsAB, Box 7960, S-907 19 Umeå,Sweden, 2001.
22. Yu, H., and MacGregor, J.F.,“Multivariate image analysis andregression for prediction of coat-ing content and distribution in theproduction of snack food”, submit-ted to Chemometrics Intell. Lab.Syst., 2002.
23. Bharati, M.H., Multivariate imageanalysis and regression for indus-trial process monitoring and prod-uct quality control, Ph.D. thesis,McMaster University, Hamilton,ON, Canada, 2002.
Received: September 17, 2002Revised: July 21, 2003Accepted: May 10, 2003
This paper is also published on TAPPI’sweb site <www.tappi.org> and
summarized in the May Solutions! forPeople, Processes and Paper magazine
(Vol. 87 No. 5).
14 TTAAPPPPII JJOOUURRNNAALL MAY 2004
SPECTROSCOPY
INSIGHTS FROM THE AUTHORSFor many years now, we have been doing research onprocess control and multivariate statistical methods forextracting information from sets of industrial data. Ourmain area of research is process control as it relates tosensors for monitoring quality. On-line digital imagingsystems represent a powerful new kind of sensor formonitoring and controlling quality in solid productssuch as pulp and paper.
The tremendous potential of NIR imaging for evalu-ating quality was evident to us in this work because wewere able to do better than we expected in determiningthe properties accurately. One problem we had wasgetting pulp samples in a timely enough manner to pre-vent the pulp from showing the effects of aging. Thesolution was obvious. We moved the camera fromMcMaster University to Tembec’s mill site and took themeasurements as the rolls came off the line.
Our next step in developing the method is to getmore detailed spatial information on all the quality vari-
ables throughout the sheets, rather than just the aver-age quality over the image. We also want to developthe imaging system for on-line use. In the meantime,the technique can give mills a new laboratory methodfor monitoring pulp quality that can replace many ofthe tedious methods now used.
Bharati and MacGregor are with the Dept. of ChemicalEngineering, McMaster University, Hamilton, Ontario L8S 4L7,Canada. Champagne (photo not available) is with Tembec Inc.,Temiscaming, Quebec J0Z 3R0, Canada. Email MacGregor [email protected].
Bharati MacGregor
Concentric mixing before the fan pump, if not done proper-ly, can significantly affect the spatial and temporal consis-
tency and chemical uniformity of the stock leaving theapproach flow area, leading to severe MD and CD nonuniformi-ties in the final sheet.TAPPI recently published approach flowguidelines for concentric mixing used for thick stock dilutionbefore the fan pump. However, in view of the importance ofthick stock dilution, a rigorous analysis of concentric mixingwas in order, along with a reexamination of the velocity ratiocriteria.
We completed concentric mixing experiments with velocityratios of up to 6 using hardwood pulp of 1.0%, 1.9%, and 2.9%consistency and water. By increasing the velocity ratio (ratio ofinner:outer jet velocity), the inner jet spread angle is found to belarger and the downstream mixing region uniform.Furthermore, local consistency measurements show a flatteningof the concentration profile with increasing velocity ratio, con-firming mixing improves as velocity ratio increases.
For the fiber stock tested, mixing is significantly dependent onthe stock consistency when the velocity ratio is small (Rv ≅ 1).Thisresult indicates that shear stress and turbulence required to fullydislodge the fiber network are not delivered by the fluid streams.Mixing results from hydrodynamic instabilities and macroscalevariations, which lead to downstream nonuniformities.
At higher velocity ratios, when the flow is turbulent, mixingis significantly affected by the velocity ratio,but there is no clearindication that mixing is affected by the stock consistency.Thesetrends are evidence that once the fiber network strength is over-come by shear stress and turbulence, the mixture behaves as aconventional Newtonian fluid in turbulent flow. Mixing at highvelocity ratio results from microscale turbulence that leads to arelatively uniform downstream mixture. TJ
Giorges is with the Georgia Institute of Technology, Atlanta, GA30332; White is with the Institute of Paper Science and Technology,Atlanta, Georgia 30332; Heindel is with Iowa State University,Ames, Iowa 50011. Email White at [email protected].
Concentric mixing of hardwood pulp and waterAKLILU T.G. GIORGES, DAVID E. WHITE, AND THEODORE J. HEINDEL
Online ExclusiveThe following peer-reviewed paper is summarized in the May 2004 Solutions! magazine (Vol. 87, No. 5). It is available in full, with slight editing, at the end of this document.
AApppplliiccaattiioonn:: This study demonstrates the importance of high velocity ratio in producing a uniform stock mix-ture in concentric thick stock dilution before the fan pump.
PEER-REVIEWED STOCK PREPARATION
PEER-REVIEWED WET-END ADDITIVES
The use of wet-end additives continuesto play an increasingly important role
in papermaking technology. The mainreasons for this include the increasingdiversification of raw materials, increas-ing production rates (which have effec-tively reduced first pass retention of fur-nish), and increasing system closure(further concentrating anionic trash inwhite water systems). Consequently, therapid development of new wet-endchemical technologies and their strate-gies for use have become an even high-er priority among chemical suppliersand papermakers.
Electrostatic interactions play animportant role in the adsorption ofalmost all wet-end additives, such as siz-ing agents, dry strength additives, wetstrength additives, and others [1, 2].Theelectrostatic charge of papermakingmaterials plays a key role in manyimportant wet-end chemistry phenome-na. Subsequently, electrostatic charge,molecular structure, and molecularweight of chemical additives (both ofthe solution type and emulsion type)have been designed to adsorb easilyonto pulp.As with chemical additives, ifthe surface charge of pulp could be con-trolled—if anionic charges could beintroduced onto pulp—then adsorptionof additives onto pulp could be acceler-ated.Thus, there would be an increase infirst pass retention, resulting inincreased efficacy of wet end applica-tions.Therefore, many studies have beenconducted on the introduction of elec-trical charges onto pulp [3-7].
EXPERIMENTALMaterials
Pulp. For our experiment, we beat hard-wood kraft pulp to 450 mL CSF using alaboratory beater and adjusted the pulpto a given electrolyte conductivity withsodium sulfate.The pulp consistency was2.0%.We studied the effects of CMC addi-tion to mixed pulp (study 2) composedof a semibleached softwood kraft pulp(southern bleached kraft pulp; SBKP)(570 mL CSF), a semibleached thermome-chanical pulp (TMP) (85 mL CSF), and adeinked pulp (DIP) (155 mL CSF). Themixed pulp was comprised of 10% semi-bleached softwood kraft pulp, 30% TMP,and 60% DIP.
CMC. We used CMC samples with dif-ferent degrees of substitution (D.S.), butwith the same degree of polymerization(D.P.), to study their influence on theadsorption of CMC onto pulp. All theCMC samples used in this work werekindly offered by Daiichikogyo-Seiyaku,Japan. TTaabbllee II lists the degrees of substi-tution and polymerization.
CMC adsorptionFirst, CMC was dissolved in water [0.5 %(w/w)].Then it was added to pulp slurrythat had its electrical conductivity adjust-ed to a designated level.This was stirredfor 1 min under a designated tempera-ture. Throughout the investigation, theamount of CMC addition that was studiedwas between 0.1%-0.3% (on dry pulp).
AnalysesThe amount of CMC adsorbed onto thepulp was determined by colloid titration.
Since 1994, researchers have studiedpractical methods of ensuring car-boxymethyl-cellulose (CMC) adsorptionto increase anionic charge of pulp. Thesubsequent properties of CMC-treatedpulp have also been investigated.Because of the anionic nature of bothcarboxyl groups of CMC and negativelycharged pulp, electrostatic repulsionposes a significant challenge to over-come in ensuring effective adsorption.
Initially, it was determined that CMCcould be adsorbed onto pulp throughprotonation of CMC by adjusting pHusing mineral acid (after CMC and pulpwere combined). A decrease in zetapotential of the pulp resulted.Handsheet studies of CMC-treated pulpindicated improved efficacy of wet-endadditives [8]. However, to adsorb almostall of the CMC onto the pulp, the pulpslurry pH must be adjusted to less thanpH 4. Consequently, the method is notpractical because it requires a largeamount of mineral acid that may cor-rode the system.
Nevertheless, subsequent researchresolved the problems of low pHrequirements for maximum adsorptionof CMC onto pulp fiber. The methodinvolved the use of CMC with lowdegrees of substitution (D.S.) in con-junction with adjustment of electricalconductivity (electrolyte concentra-tion) of the pulp slurry. Consequently,CMC adsorption was increased, result-ing in improved efficacy of wet-endadditives [9].
Advanced wet-end system with carboxymethyl-cellulose
MASASUKE WATANABE, TOMOHISA GONDO, AND OSAMU KITAO
ABSTRACT: We developed a novel and practical method to greatly enhance the efficacy of wet-endchemicals. It involves the adsorption of carboxymethyl-cellulose (CMC) onto pulp. Investigation showed thatpaper made from CMC-treated pulp required up to 30% to 50% less wet-end additives than usual to obtainthe desired quality when compared to paper made from non CMC-treated pulp. Furthermore, the study iden-tified both the degree of substitution of CMC and electrical conductivity of the pulp slurry as key variablescritical to CMC adsorption onto pulp. Improved fiber charge distribution, which facilitates improved wet-endadditive distribution, is believed to be the main reason for the synergy seen when CMC is used in conjunc-tion with wet end applications. Results of laboratory investigations and development of CMC applicationwere effectively reproduced in mill trials.
AApppplliiccaattiioonn:: Treating pulp fiber with CMC could reduce the dosage levels of wet-end additives, resultingin a cleaner mill system, improved quality, and increased economic efficiency.
VOL. 3: NO. 5 TTAAPPPPII JJOOUURRNNAALL 15
16 TTAAPPPPII JJOOUURRNNAALL MAY 2004
WET-END ADDITIVES
We used potassium polyvinyl sulfate(PVSK) as the anionic titrant and polydiallyldimethyl-ammonium chloride(pDADMAC) as its cationic counterpart.The zeta potential of the pulp slurry wasmeasured by Zetasizer 2000 (MALVERNInstrument Ltd.). The rosin sizing agentcontents in the paper were determinedby pyrolysis-gas chromatography massspectroscopy (PyGC-MS) combined withonline methylation, according to pub-
lished procedures (Yano et al) [10].Thedry strength additives and wet strengthadditives contents in the paper weredetermined by nitrogen analyses usingTN-10 (Mitsubishi-Kasei Co., Ltd.).
Handsheet preparation and
measurementStudy 1 (effect of the CMC treated pulp1). We added 0.1% (on dry pulp) of CMCto bleached hardwood kraft pulp (pulpconsistency was 2.0%, 450 mL CSF, 40°C).We then placed this CMC-treated pulp incontact with 2% (on dry pulp) of alumand a designated amount of emulsionrosin and dry strength additives such asPAM or wet strength additives such aspolyamide-amine epichlorohydrin resin.After addition of the additives,handsheetshaving 60g/m2 bone-dry basis weightwere prepared with a sheet machineaccording to ISO 5269-1:1979 and driedat 100°C using a rotary drum dryer.
Study 2 (effect of the CMC treatedpulp 2). A mixed pulp comprised of 10%semibleached softwood kraft pulp, 30%TMP,and 60% DIP was dispersed in water(pulp consistency was 2.0%) and then0.1% (on dry pulp) of CMC was added.We added 2% of alum to the pulp, fol-lowed by a designated amount of emul-sion rosin and dry strength additives.Wethen prepared 40 g/m2 bone-dry basisweight handsheets from the mixturewith a sheet-machine and dried them at100°C using a rotary drum dryer.
Throughout the investigation, themoisture of the handsheets was keptconstant under a relative humidity of50% at 23°C. Quality tests on the hand-sheets were carried out under the sameconditions.The dry tensile strength wasdetermined according to ISO 1924-1:1992. The wet tensile strength wasdetermined according to ISO 3781:1983.The sizing degree was determinedaccording to TAPPI Useful Methods UM596 (5 µL water drop test).
RESULTS AND DISCUSSIONLaboratory results
Method of CMC adsorption. In general, itis difficult for CMC to adsorb onto pulpbecause of electrical repulsion betweenboth the negatively-charged groups onthe pulp and carboxyl groups of CMC.Therefore, it is necessary to reduce theelectrical repulsion in order to allowthem to come close enough to adsorb toeach other. Throughout the study intothe effect of the CMC adsorption ontopulp, the following parameters con-cerned with electrical repulsion werestudied; i.e. the electrical conductivity ofpulp slurry and the D.S. of CMC.
FFiigguurree 11 shows that the higher theelectrical conductivity of the pulp slurry,the greater the amount of CMC adsorbedonto the pulp. Figure 1 also shows thatCMC with lower D.S. was easier toadsorb onto pulp.
FFiigguurree 22 shows that temperatureswithin the range of typical mill condition(20°C~40°C) only have a slight influenceon the CMC adsorption.We also studiedthe effect of pulp type on CMC adsorp-tion. However, although the adsorptionamount was influenced by pulp type,CMC could be adsorbed onto all kinds ofpulp (TTaabbllee IIII).
All the results are interpreted in termsof the reduction of the electrical repul-sion between the CMC and the pulp.CMC with lower D.S. is advantageous toadsorption because it has a smalleramount of negative charge. As a result,the CMC can approach the pulp moreeasily. Furthermore, soluble mineral ionsreduce the electrical repulsion byscreening the negative charges of CMCand the pulp, therefore the CMC can getcloser more easily. Then the CMC isadsorbed onto the pulp by hydrogenbonding between their cellulose chains.In addition, the CMC has a high affinitywith the pulp because it has more pure
II. The effect of pulp type and CMC type on zeta-potential of pulp added (% on
dry pulp).
CMC
ADDED, ZP, mV
PULP D.S. D.P. % on dry pulp Pre* Post*
Hardwood kraft pulp 0.46 230 0.1 -14.5 -19.1Hardwood kraft pulp 0.46 300 0.1 -14.5 -20.5Hardwood kraft pulp 0.46 400 0.1 -14.5 -21.8Softwood kraft pulp 0.46 230 0.1 -15.0 -19.9Deinked pulp 0.46 230 0.1 -18.2 -21.0Thermo mechanical pulp 0.46 230 0.1 -7.1 -16.4*Pre denotes zeta-potential of pulp; Post denotes zeta-potential of pulp after addition of CMC.
I. Characterization of CMC.
CMC sample D.S.1 D.P.2
A 0.46 230B 0.50 230C 0.65 230
1Degree of Substitution 2Degree of Polymerization
CM
C A
DS
OR
PT
ION
, %
100
80
60
40
20
00 1 2 3 4ELECTRICAL CONDUCTIVITY
OF PULP SLURRY, mS/cm
D.S. = 0.46D.S. = 0.50D.S. = 0.65
1. The relationship between the elec-
trical conductivity of pulp slurry and
CMC adsorption. Treatment condi-
tions were as follows: pulp: bleached
kraft pulp (450 mL CSF), 40°C, electri-
cal conductivity of 0.7 mS/cm. CMC:
addition level of 0.1% (on dry pulp).
CM
C A
DS
OR
PT
ION
, %
100
80
60
40
20
010 20 30 40 50TEMPERATURE OF PULP SLURRY, ˚C
D.S. = 0.46D.S. = 0.50D.S. = 0.65
2. The relationship between the tem-
perature of pulp slurry and CMC
adsorption. Treatment conditions
were as follows: pulp: bleached kraft
pulp (450 mL CSF), 40°C, electrical
conductivity of 0.7 mS/cm. CMC:
addition level of 0.1% (on dry pulp).
VOL. 3: NO. 5 TTAAPPPPII JJOOUURRNNAALL 17
WET-END ADDITIVES
cellulose units than CMC with high D.S.Therefore, CMC with lower D.S. canadsorb onto pulp more easily than thatwith a higher D.S.
Research was undertaken into theelectrical conductivity of raw materialsunder practical mill conditions.We foundthat the electrical conductivity depend-ed upon the method of white waterrecovery and on the recovery rate. Itranges from 0.4 mS/cm to 2.5 mS/cm.Although the adsorption level was slight-ly influenced by the type of pulp and theconcentration of the pulp slurry, almostall the CMC (D.S.=0.46, added level 0.1%on dry pulp) adsorbed onto the pulpwhere the electrical conductivity wasmore than 0.6 mS/cm (Fig. 1).
On the other hand, when the electri-cal conductivity was less than 0.4mS/cm, the rate of CMC adsorptiondecreased. It is easy to control the elec-trical conductivity of the raw materialsby adjusting the method of white waterrecovery and the recovery rate so thatalmost all the CMC can be adsorbed.
CMC adsorption can also be affectedby the zeta potential of the pulp slurry.FFiigguurree 33 shows how the zeta potential ofCMC-treated pulp decreases drasticallyas the amount of CMC added increases.Therefore, the results indicate that CMC-treated pulp will enhance the efficacy ofwet-end chemicals. Although the highermolecular weight of the CMC the greaterthe amount of CMC that is adsorbed ontothe pulp (Table II), the adsorption behav-ior is governed more by the D.S. of CMCthan by its molecular weight.Furthermore, a solution of CMC withhigh molecular weight has high viscosityand so is difficult to handle.This makes it
inappropriate for practical use.Effect of the CMC-treated pulp
1 (enhancement of the efficacy
of wet-end chemicals)We added a designated amount of rosinsizing agent, dry strength additives suchas polyacrylamide (PAM) and/or wetstrength additives such as polyamide-amine epichlorohydrin (PAE) resin toCMC-treated pulp. We then made hand-sheets from the mixture to evaluate thepaper’s performance. FFiigguurreess 44--66 showthe properties of the handsheets (2%alum added in each experiment). Theproperties of the handsheets were muchimproved by the addition of CMC. Notethat the amount of improvement inchemical efficacy was much greater than
the increase in the retention of wet-endchemical.
The results were in part attributed toan increase in the first pass retention ofwet-end additives by means of theincrease in surface anionic sites causedby the addition of CMC. Improved fibercharge distribution, which facilitatesimproved wet-end additive distribution,is believed to be the main reason for thesynergy seen when CMC is used in con-junction with wet end applications.Thatis to say, the enhancement of the efficacyof the wet-end chemicals resulted from amore even distribution of an increasednumber of anionic sites on the pulp, as aresult of CMC treatment. FFiigguurree 77, whichshows the effect of CMC, strongly sup-ports this hypothesis.
ZE
TA
PO
TE
NT
IAL O
F P
ULP
, m
V -10
-15
-20
-250 0.1 0.2 0.3 0.4
CMC ADDED, % on dry pulp
3. The effect of adding CMC on the
zeta potential of pulp. Treatment con-
ditions were as follows: pulp:
bleached kraft pulp (450 mL CSF),
40°C, electrical conductivity of 0.7
mS/cm. CMC: D.S. of 0.46.
SIZ
ING
DE
GR
EE
, s
40
30
20
10
00 0.1 0.2 0.3 0.4 0.5
ROSIN SIZING AGENT RETAINED IN HANDSHEET, %
Without CMCWith CMC
4. The relationship between the
amount of sizing agent in handsheet
and the sizing degrees. Handsheets
were prepared with 2% alum and
0.3% anionic PAM. Addition levels of
rosin sizing agent were 0%, 0.1%,
0.2%, 0.5%.
BR
EA
KIN
G L
EN
GT
H,
km 6.5
6.0
5.5
5.00 0.1 0.2 0.3 0.4 0.5
DRY STRENGTH ADDITIVE RETAINED IN HANDSHEET, %
Without CMCWith CMC
5. The relationship between the con-
tent of dry strength additive in the
handsheet and the tensile index.
Handsheets were prepared with 2%
alum. Addition levels of dry strength
additive were 0%, 0.2%, 0.5%.
WE
T T
EN
SIL
E I
ND
EX
, k
N/m
0.30
0.25
0.20
0.15
0.100 0.2 0.4 0.6
WET STRENGTH ADDITIVE RETAINED IN HANDSHEET, %
Without CMCWith CMC
6. The relationship between the
amount of wet strength additive in
the handsheet and the tensile index.
Handsheets were prepared with 2%
alum and 0.3% rosin sizing agent.
Addition levels of wet strength addi-
tive were 0%, 0.3%, 0.5%.
SIZ
ING
DE
GR
EE
, s
40
30
20
10
00 0.2 0.4 0.6ADDITION OF SIZING AGENT,
% on dry pulp
Without CMCWith CMC, pre-added*
With CMC, post-added*
7. The effect of dosing point on sizing
degree. Handsheets were prepared
with 2% alum, 0.3% anionic PAM and
0.3% rosin sizing agent. *Pre-added
denotes CMC added prior to addition
of wet-end additives. Post-added
denotes CMC added after the addition
of wet-end additives.
18 TTAAPPPPII JJOOUURRNNAALL MAY 2004
WET-END ADDITIVES
The most significant result of thisinvestigation was that handsheets madefrom CMC-treated pulp required as muchas 30% to 50% less wet-end additivesthan usual to provide desired qualitieswhen compared to handsheets madefrom non CMC-treated pulp. The opti-mum addition level of CMC was 0.1% (ondry pulp), although the effectiveness ofwet-end additives was increased by theaddition of the CMC (FFiigg.. 88).This repre-sents the most favorable balancebetween the cost of CMC and the savingsmade on the cost of additives.Furthermore, almost all the CMC can beadsorbed onto the pulp, even at higheraddition levels than 0.1% (for example,0.2% or 0.3%).Therefore, there is no fearabout non-adsorbed CMC contaminatingthe mill system.
Effect of the CMC-treated pulp
2 (CMC added to mixed pulp)Next, we studied the effect of CMC addi-tion to a mixed pulp. The mixed pulpwas comprised of 10% semibleachedsoftwood kraft pulp, 30% TMP, and 60%
DIP. As in Study 1, the sizing degree ofthe handsheets made from CMC-treatedmixed pulp with emulsion rosin sizeagent also increased. However, when thesame quantity of CMC is only added tothe TMP, before mixing with the otherkinds of pulp, the improvement in sizingagent efficacy was greater than when the CMC was added to the mixed pulp(FFiigg.. 99).
The zeta potentials of these pulps(FFiigg.. 1100 aa) were as follows: semibleachedsoftwood kraft pulp (SBKP): –15 mV,TMP: –7 mV, deinked pulp: –18 mV. Thezeta potential of TMP was much higherthan that of the softwood kraft pulp andde-inked pulp. However, the zeta poten-tial of TMP was changed by the additionof the CMC from –7 mV to –5.5 mV.As aresult, these zeta potentials were equal-ized, as shown in Fig. 10 b.Therefore, themajor improvement in the efficacy of thesizing agent in the case of CMC pre-added to TMP resulted from the equaliza-tion of the surface charges (zeta poten-tials) of these pulps.That is to say, if thereare various kinds of pulp in a mix, wet-end chemicals adsorb preferentially ontothe pulp that has more anionic surfacecharge.This produces an uneven adsorp-tion of wet-end additives in the pulpfibers.This, in turn, means that the sizingagent will be unevenly distributed acrossthe sheet. Therefore, the increase inanionic charge of TMP by the CMC
addition minimized the surface chargedifference among the different pulps,leading to a more uniform adsorption ofsizing agent. As a result, the efficacy ofthe sizing agent was greatly increased.
Figure 9 also shows that the enhance-ment in sizing agent efficacy, where thesame quantity of CMC was added only tosoftwood kraft pulp or to the deinkedpulp before mixing, was much smallerthan where CMC was added only to TMP.In these cases, these zeta potentials werenot equalized, as shown in Figs. 10 c and10 d. These results strongly support thepreviously cited hypothesis.
Furthermore, there may be a differen-tial in surface charge between fibers ofthe same kind of pulp. If this is the case,CMC addition may decrease this surfacecharge difference between pulp fibers,and increase the number of surfaceanionic sites on the pulp and improvecharge distribution on the surface of thepulp.
Mill trialsTreatment of pulp fiber with CMC couldreduce the dosage levels of wet-end addi-tives. Therefore, the use of CMC-treatedpulp was expected to result in a cleanermill system, improved quality, andincreased economic efficiency. We havecarried out several mill trials and, as aresult, CMC is currently added to someproducts.
10. The zeta potential of pulp. Treatment conditions were as follows: tempera-
ture of 40°C, CMC addition level of 0.1% (on dry pulp). *CMC was added only to
this kind of pulp and before mixing with the other kinds of pulp.
ZE
TA
PO
TE
NT
IAL
, m
V -25
-20
-15
-10
-5
0SBKP DIP TMP
a. Initial pulp
ZE
TA
PO
TE
NT
IAL
, m
V -25
-20
-15
-10
-5
0SBKP DIP TMP
b. Added to TMP*
After addition of CMCInitial pulp
ZE
TA
PO
TE
NT
IAL, m
V
-25
-20
-15
-10
-5
0SBKP DIP TMP
c. Added to SBKP*
ZE
TA
PO
TE
NT
IAL
, m
V -25
-20
-15
-10
-5
0SBKP DIP TMP
d. Added to DIP*
SIZ
ING
DE
GR
EE
, s
60
40
20
00 0.1 0.2 0.3 0.4
CMC ADDED, % on dry pulp
8. The relationship between the addi-
tion of CMC and sizing degree.
Handsheets were prepared with 2%
alum, 0.3% anionic PAM, and 0.3%
rosin sizing agent.
SIZ
ING
DE
GR
EE
, s
140
120
100
80
60
40
20
00 0.2 0.4 0.6ROSINSIZING AGENT ADDED,
% on dry pulp
*CMC was added only to this kind of paperprior to mixing with the other kinds of paper.
Without CMCAdded to mixed pulpAdded to TMP*
Added to SBKP*
Added to DIP*
9. The effect of addition of CMC to
the mixed pulp.
VOL. 3: NO. 5 TTAAPPPPII JJOOUURRNNAALL 19
Mill trial 1 (kraft paper)Kraft paper requires high levels of tensilestrength, so it is necessary to add a lot ofdry strength additives such as PAM. CMCwas added to the pulp chest and theamount of PAM application was reduced.As with the laboratory results, the kraftpaper prepared with the 0.1% (on drypulp) CMC-treated pulp required only33% of PAM amount to produce paperwith the same tensile index as papermade from non CMC-treated pulp. CMCaddition also led to a reduction of 15% inthe addition level of the sizing agent asshown in TTaabbllee IIIIII.
Mill trial 2 (printing paper with
titanium dioxide)Titanium dioxide has great light scatter-ing power; therefore it is often used bypapermakers. However, its small particlesize and high specific gravity lead to areduction in first pass retention. In addi-tion, deposition of unretained titaniumdioxide in the white water recovery sys-tem significantly reduces paper machineperformance.
Addition of CMC in the mixing chestincreased the first pass retention of tita-nium dioxide from 40% to 70%.Consequently, it led to a 15% reduction inthe dosage level of expensive titaniumdioxide (TTaabbllee IIVV). In this case, CMC wasshown to increase pulp anionicity. Thisprovided more adsorption sites for thetitanium dioxide and improved the effec-tiveness of the cationic additives, both ofwhich increased the first pass retentionof titanium dioxide.
CONCLUSIONSWe developed a novel and practicalmethod to greatly enhance the efficacyof wet-end chemicals. It involves theadsorption of CMC onto pulp. Ourresearch has shown that the CMC can beadsorbed onto the pulp by controllingthe electrical conductivity and D.S.of theCMC. Temperatures within the range oftypical mill condition (20°C~40°C) influ-enced the adsorption of CMC onto thepulp only to a slight extent.The higherthe molecular weight of the CMC, the
greater the amountof CMC that isadsorbed onto thepulp; however, the adsorption behavioris more governed by the D.S. of CMCthan by its molecular weight.
The use of CMC-treated pulp couldgreatly enhance the efficacy of wet-endchemicals.The results occurred not onlybecause of an increase in the first passretention of wet-end additives, butthrough an increase in surface anionicsites of the CMC-treated pulp. We alsoattributed the improved efficacy of thewet-end chemicals to an increased num-ber and a more uniform distribution ofanionic sites on the pulp. Furthermore,the adsorption of CMC leads to a reduc-tion in the differential in surface chargeamong pulp fibers.
The mill trial results also showed thatCMC-treated pulp leads to a reduction inthe addition levels of wet-end additives.Furthermore, CMC treatment leads to anincrease in the first pass retention offiller dramatically.Therefore, it is expect-ed that more applications for CMC inpapermaking will be found.
LITERATURE CITED1. Isogai, A., Kitaoka, C., Onabe, F., J.
Pulp Paper Sci. 23(5): 215(1997).2. Wågberg, L., and Bjorklund, M.,
Nordic Pulp Paper Res. J. 8(1):
53(1993).3. Kitaoka, T., Isogai, A., and Onabe,
F., Nordic Pulp Paper Res. J. 14(4):279(1999).
4. Kitaoka, T., Isogai, A., and Onabe,F., Nordic Pulp Paper Res. J. 15 (3):177(2000).
5. Laine, J., Lindström, T., Nodmark,G., Risinger, G., Nordic Pulp PaperRes. J. 15(5): 520(2000).
6. Laine, J., Lindström, T., Nodmark,G., Risinger, G., Nordic Pulp PaperRes. J. 17(1): 50(2002).
7. Laine, J., Lindström, T., Nodmark,G., and Risinger, G., Nordic PulpPaper Res. J. 17 (1): 57(2002).
8. Japanese Patent Publication No. 9-291490 (1997)
9. Japanese Patent Publication No.2001-262498
10. Ishida, Y., Ohtani, H., Kato, T.,Tsuge, S., Yano, T., TAPPI J. 77(3):177(1994).
Received: May 3, 2003Accepted: December 8, 2003
This paper is also published on TAPPI’s web site <www.tappi.org> andsummarized in the May Solutions! forPeople, Processes and Paper magazine
(Vol. 87 No. 5).
INSIGHTS FROM THE AUTHORSElectrostatic interactions play an important role in determining pulp slur-ry properties. If the surface charge of pulp was controlled—that is to say,if the pulp surface was more anionized—the adsorption of wet-end addi-tives would be accelerated and more wet-end additives would beadsorbed onto the pulp.That is why we chosethis topic.
Watanabe, Gondo, and Kitaoare with the Pulp and PaperResearch Laboratory, OjiPaper Co., Ltd., 1-10-6Shinonome, Koto-ku, Tokyo135-8558, Japan. EmailGondo at [email protected]
III. The results of mill trial 1.
CONTROL TEST
CMC (%) — 0.1Alum (%) 2 2Sizing agent (%) 0.065 0.055PAM (%) 0.3 0.1
Grammage (g/m2) 77.9 77.6Tensile (MD/CD) (kN/m) 5.42/2.59 5.42/2.43Tear (MD/CD) (mN) 1058/1089 1150/1159Internal bond (mN) 3000 2980Sizing degree (s) 26 31
IV. The results of mill trial 2.
CONTROL TEST 1 TEST 2
CMC (%) — 0.1 0.1PAM (%) 0.4 0.4 0.4Cationic starch (%) 1 1 1Sizing agent (%) 0.3 0.3 0.3Alum (%) 1.4 1.4 1.4Titanium dioxide (%) 8.8 8.8 7.5
Grammage (g/m2) 95.1 95.7 96Ash (%) 7.2 8.2 7.1Opacity (%) 93.6 94.5 93.8First pass retention
Total (%) 86.2 92.9 93.5Titanium dioxide (%) 41.3 73.4 73.0
Watanabe Gondo Kitao
20 TTAAPPPPII JJOOUURRNNAALL MAY 2004
PEER-REVIEWED COATING
In dry surface treatment (DST) ofpaper, electrostatically charged pow-
der particles coat the substrate. Such par-ticles attach to paper mainly by electro-static forces and form a layer that is thenfixed in a nip between heated rolls. Thefinal fixing step achieves adhesion to thesubstrate and smoothes the surface.Because the process uses no water, theproperties of the coated paper developquickly and require less space comparedto conventional coating methods basedon water dispersion. Previous studies [1,2] have reported on laboratory-scaleprocess conditions and the adaptabilityof the process, but surface openness ofthe coated paper was problematic.Preparation of coating powder by spray-drying water-based dispersions of coat-ing components resulted in overly largecoating particle agglomerates that couldnot be smoothed after fixing. As withconventional coating layers, the size ofthe coating pigment and particle agglom-eration affect the properties of the DSTcoating layer and the final paper quali-ties. When large agglomerates remain inthe surface, it becomes coarse [3-6].Theaim of this work was to build on the workof previous studies by using reinforcedfixing conditions and a different powderpreparation method, which considerablyreduced the size of agglomerates.We also
corona electrodes causes ion-particle col-lisions.These further charge the particles,while the field directs charged particlesonto the paper surface (FFiigg.. 11) [1, 2].Weused 95-kV negative voltage, with a maxi-mum current of 100 µA (Nordson, SureCoat). Compressed air transported thecoating powder to the electrode systemwith mass flows of 15 (± 2.5) g/min and31 (± 3.0) g/min. Coating powders wereapplied to a 45 g/m2 groundwood-con-taining base paper using an electrodelength of 0.6 m and a gap of 0.3 m.We didnot change the charge potential of thebase paper before electrostatic deposi-tion.The coatings were fixed in a heatedroll nip of a laboratory calender at 6 msand 12 ms dwell times (35 m/min and 17m/min, respectively) to a coat weight of 6g/m2. Roll surface temperatures wereadjusted from 120ºC to 200ºC. The rollsurface materials were hard-chrome forfreeze-dried and Teflon-based cover forspray-dried materials. The backing rollwas a polymer roll with a hardness of 90ShD. Linear load was adjusted between 5kN/m and 70 kN/m.
We used standard procedures to meas-ure final paper properties: Bendtsenroughness (SCAN-P 21:67), Bendtsen airpermeability (SCAN-P 60:87), PPS-S10surface smoothness (ISO 8791-4), Huntergloss (TAPPI T480), IGT surface strength
reduced the coating binder content to alevel equal to the amount used in con-ventional pigmented coatings.
EXPERIMENTALCoating colors were first prepared aswater-based dispersions containing 10 or20 pph (parts per hundred calculated ondry pigment, by weight) styrene-butadi-ene (SB) latex, and 100 pph ground calci-um carbonate (GCC, CaCO
3).The coating
dispersions were freeze-dried [7] to asolids content of 98%, then embedded inliquid nitrogen and ground.We used twospray-dried [8] coating powders to distin-guish the effect of powder preparation(i.e., drying method) on coating struc-ture. One consisted of 30 pph polymericpigment, 70 pph of GCC, 10 pph SB latex(latex A), and 20 pph of an anionic pota-to-based starch. The other consisted of100 pph of GCC and 10 pph SB latex(latex A). TTaabbllee II presents the coatingcomponents and compositions used inour experiments.
We used a laboratory-scale applicationunit to attach dry coating powder parti-cles to paper substrates [1]. In thismethod, high voltage electrodes (coronaelectrodes) generate large quantities ofcharged ions to charge the coating parti-cles. The strong electric field formedbetween the grounding electrode and the
Coating layer formation in dry surface treatment of paper substrates
JUHA MAIJALA, KAISA PUTKISTO, AND JOHAN GRÖN
ABSTRACT: Dry surface paper coating can potentially eliminate drying after coating and produce a finalsurface quality equal to that achieved with conventional coating processes. This work reports on recentexperiments involving pigmented coating of paper substrates with a dry surface treatment process. In theprocess, powdered coating, exposed to a corona charge and an electric field, is directed and guided ontopaper. In a subsequent step, the dry coating powder is anchored to the paper surface by thermomechanicaltreatment in a nip formed between heated rolls, forming the final coating layer.
In dry surface treatment, coating is applied without wetting the paper substrate, thus eliminating the pos-sibility for fiber swelling and surface roughening. Coating penetration into the paper is minimal, as a distinctboundary forms between the coating layer and paper. Sufficient cohesion and adhesion to the paper surfaceare needed to obtain an appropriate coating layer. The greatest challenge is to reduce surface openness andcreate a structure similar to that obtained with conventional coating methods.
In this study, we investigated factors affecting coating structure. The focus was on the influence of theapplied pressure and temperature on fixation of coating powder onto paper. We achieved adequate surfacestrength and were able to improve paper surface properties.
AApppplliiccaattiioonn:: This report provides insights into the experimental technique of dry coating and its possibleapplications.
VOL. 3: NO. 5 TTAAPPPPII JJOOUURRNNAALL 21
COATING
with medium viscosity oil (SCAN-P 63:90), and Cobb-Unger oilabsorption (modified SCAN-P 37:77, 6 s absorption time). Thecoating surface was analyzed using an environmental scanningelectron microscope (ESEM), i.e., an ElectroScan 2020 (USA)with an Oxford Tetra solid state backscatter electron detector(BSE). We quantified the size distribution of particle agglomer-ates for freeze and spray-dried powders with a Malvern Series2600c droplet and particle size laser according to methodENE38M37.As a reference, we used a commercial blade-coatedwood-containing paper, with a basis weight of 54 g/m2 and acoat weight of 6.5 g/m2/side.Measurements were performed onspecimens that had been conditioned according to SCAN-P 2:75at 23 ± 1ºC and 50 ± 2 % relative humidity.
RESULTS AND DISCUSSIONTTaabbllee IIII presents the experimental conditions during the trials.For surface strength measurements of paper, the focus was onhigh temperature regions (200ºC) to ensure sufficient film for-mation in the coating layer. The reference paper contained 8pph of latex and 7 pph of starch for comparison with DSTpapers containing a lower amount of binder (10 pph of latex).
Coatings containing latex B gave similar paper coating sur-face strength to those of the reference paper (FFiigg.. 22).Dwell timeand fixing pressure slightly affected the surface strength of thecoating containing latex A, while the coating with latex Bbehaved more sensitively. This was because of differences inthermoplastic properties of the binders. The high amount oflatex B gave surface strength equal to that of the referencepaper with 12 ms dwell time, but with 6 ms dwell time, lesslatex B provided sufficient surface strength (Fig. 2, right).In fixing, three forces dominate:
1. an adhesive force between the hot roll and the coatinglayer (F
RC),
2. a cohesive force in the coating layer (FC), and
3. an adhesive force (FCP
) between the coating layer and thepaper.
If FC
is less than FRC
,delamination of the coating layer will occur[9]. Wikström, Carlsson and Salminen [10] have studied theeffect of high roll surface temperatures on deposits formationin the calendering process. Generally, the surface strengthshould increase as fixing dwell time increases.However, accord-ing to their study, the reduced deposit formation on the calen-der hot roll surface and the increased surface strength resultingfrom it could be the effect of the reduced nip dwell time (Fig. 2, right, 10 pph of latex B).These authors reported that, forthe case of conventional coatings at elevated temperatures,
1. Schematic illustration of a dry surface treatment unit
used in the experiments.
Unwind
Groundingelectrode
Rewind
Negative (corona) electrodes
High voltagesource
Powder feeder nozzle andcharging electrode
Aircompressor
Powder fluidization withcompressed air
Thermomechanical treatment (fixing)
I. Coating components and compositions used in the experiments.
AVERAGE GLASS TRANSITION COATINGCOMPONENT PARTICLE SIZE, µm TEMPERATURE, °C COMPOSITIONS, pph
Latex A, carboxylated latex based on S/B copolymer, 0.21 43 10/20Dow Chemical Co., Dow Finland
Latex B, carboxylated latex based on S/B copolymer, 0.14 20 10/20Dow Chemical Co., Dow Finland
Ground CaCO3 (GCC), Hydrocarb 90, Omya Oy, Finland 0.65 — 70/100Polymer pigment, modified S/A copolymer, H-N10, 0.45 101 0/30
Dynea Chemicals Oy, FinlandStarch, oxidized low viscosity grade, Raisamyl 302E, 10-100 — 0/20
Raisio Chemicals Oy, Finland (granule size)
II. Experimental conditions in dry surface treatment of paper.
DEPOSITION CHARGING DEPOSITION COND. DWELL ROLL SURFACE LINEAR LOAD BASEPAPER
VOLTAGE, kV CURRENT, µA RH-% /Temp. °C TIME, ms Temp. °C kN/m MOISTURE CONT., %
95 100 50/23 6/12 120/160/200 5/20/35/50/70 6
III. Base paper properties for reference paper and dry surface treated papers.
SMOOTHNESS, ROUGHNESS, AIR PERMEABILITY, OIL ABSORPTION,PPS-s10, µm Bendtsen, ml/min Bendtsen, ml/min GLOSS, Hunter, % Cobb-Unger, g/m2
5.4 210 100 7 30
22 TTAAPPPPII JJOOUURRNNAALL MAY 2004
COATING
adhesion of coating layer to the calendersurface (F
RC) exceeds the cohesive
strength of the coating layer (FC) (stick-
ing).They also reported that an SB latexwith a high ability to flow in the lowtemperature regions (~20ºC-50ºC) and alow ability to flow in the high tempera-ture regions (>100ºC) has improved suit-ability for calendering at high tempera-tures (roll surface temperatures ~200ºCor more).
LePoutre and Hiraharu [11] statedthat, with a binder level sufficient to pro-vide a cohesive structure, void fractiondetermines the strength of the structure.For coating method with coating materi-al agglomerates, coating layer porositydeteriorates; the coating becomes morefragile, and inferior surface strength maybe expected.
TTaabbllee IIIIII presents the base paper val-ues for reference and DST papers before
surface treatment. Overall, paper proper-ties were considerably changed whenthe base paper was dry surface treated(TTaabbllee IIVV). The surface became moreclosed and smoother. The propertieswere similar to those of the referencepaper. In some situations, DST improvedcoating layer surface properties.
In this case, the DST coatings con-taining 10 pph of latex were evaluatedfor their paper characteristics.Furthermore, differences between thecoatings containing latex A and B wererather marginal. Latex B producedrougher coating surfaces, although thecoating layer structure was more closed.However, coatings with 10 pph of latex Aproduced higher average surfacestrength, and therefore were furtherstudied for their paper characteristics.
When we examined the effect of rollsurface temperature on the paper prop-erties, we found that the ability of thebinder to form a film at elevated temper-atures increased, and paper gloss andsmoothness improved (Table IV). Asbridging improved in the coating layersurface, the smoothness of the surfacesconsiderably improved, taking into con-sideration that the sheet passed throughonly one nip. Compared with the refer-ence paper, the air permeability of DSTpaper was higher because of a moreporous structure, as shown by its high oilabsorption values. This suggests insuffi-cient thermal conductivity and reducedbinder film formation in the coating layerbecause of the larger void fraction. Thelarge difference in air permeabilitybetween the reference and the DSTpapers results from the two-side coatingof the reference paper compared to theone-side coated DST paper. High fixingtemperature normally leads to high glossvalues because latex coalesces into afilm, causing the voids to shrink and thusresulting in an increased number ofreflecting areas [6, 12, 13].
Raising the linear load compressedthe coating layer, which reduced air per-meability and oil absorption values(TTaabbllee VV). PPS-S10 smoothness was alsoimproved, but macroroughness(Bendtsen) deteriorated as the linearload increased. Macroroughness mightbe caused by the coating layer sticking tothe hot roll surface, or by the fibershapes showing through the coating
2. Surface strength (IGT) of the dry surface treated papers with a fixing temper-
ature of 200°C and freeze-dried coating powders. Reference paper surface
strength 0.35 m/s (dashed line).
SU
RF
AC
E S
TR
EN
GT
H,
m/s 0.5
0.4
0.3
0.2
0.1
012/20 12/70 6/20 6/70
DWELL TIME, ms/LINEAR LOAD kN/m
Latex A 10 pphLatex A 20 pph
SU
RF
AC
E S
TR
EN
GT
H,
m/s 0.5
0.4
0.3
0.2
0.1
012/20 12/70 6/20 6/70
DWELL TIME, ms/LINEAR LOAD kN/m
Latex B 10 pphLatex B 20 pph
IV. Effect of fixing temperature on dry surface treated paper properties at a
dwell time of 12 ms and a linear load of 20 kN/m. The freeze-dried coating
powder consisted of 100 pph CaCO3
and 10 pph of latex A. SD equals Standard
Deviation.
REFERENCEROLL SURFACE DRY SURFACE TREATED PAPER PAPER
TEMPERATURE, °C 120 160 200 —
PPS-s10 smoothness, 2.70 (0.25) 2.00 (0.11) 1.50 (0.10) 2.5µm, SD
Bendtsen roughness, 43.0 (5.0) 26.5 (1.3) 17.3 (2.5) 50.0mL/min, SD
Bendtsen air permeability, 62.5 (2.9) 52.5 (2.9) 26.3 (2.5) 5.0mL/min, SD
Hunter gloss, %, SD 13.8 (0.7) 26.2 (1.1) 34.0 (1.4) 27.5Cobb-Unger oil absorption, 23.5 (0.5) 30.4 (3.6) 27.7 (2.8) 15.5
g/m2, SD
V. Effect of fixing pressure on dry surface treated paper properties at a dwell
time of 12 ms and a fixing temperature of 200°C. The freeze-dried coating pow-
der consisted of 100 pph CaCO3
and 10 pph of latex A. SD equals Standard
Deviation.
REFERENCE
DRY SURFACE TREATED PAPER PAPER
LINEAR LOAD (kN/m) 5 20 35 50 —
PPS-s10 smoothness, 1.65 (0.15) 1.50 (0.10) 1.25 (0.04) 1.30 (0.13) 2.5µm, SD
Bendtsen roughness, 17.0 (1.8) 17.3 (2.5) 20.3 (3.8) 24.0 (2.2) 50.0mL/min, SD
Bendtsen air permeability, 38.8 (8.5) 26.3 (2.5) 17.5 (2.9) 16.3 (2.5) 5.0mL/min, SD
Hunter gloss, %, SD 24.8 (1.9) 34.0 (1.4) 43.2 (0.9) 39.5 (1.2) 27.5Cobb-Unger oil absorption, 30.1 (4.6) 27.7 (2.8) 22.1 (1.6) 14.4 (1.6) 15.5
g/m2, SD
VOL. 3: NO. 5 TTAAPPPPII JJOOUURRNNAALL 23
COATING
layer (FFiigg.. 33). If fibers are compared withcoating components, we see that thefibers form non-specular reflecting areas[13]. The effect of too high linear loadcan also be seen in gloss. Gloss reached apeak value and subsequently declined asthe linear load increased (Table V).Particle packing was enhanced throughincreased pressure, as seen with theimproved air permeability and oilabsorption values. However, vigorousthermomechanical treatment couldcause excessive compaction of theupper coating layer, resulting in a closedcoating layer, although the inner struc-tures could remain rather open.
ESEM micrographs show how theraised fixing temperature increased latexfilm formation and decreased surfaceopenness, with a subsequent improve-ment in surface smoothness (Fig. 3).At afixing temperature of 120ºC, the surfacehas high macroroughness (resolutionlevel 0.1-1.0 mm), but its microrough-ness (resolution level 1-100 µm) is low(Fig. 3, left column). Agglomerates aredetectable and the surface is rathercoarse (Fig. 3, upper row). At 160ºC thesurface is visually smooth and less open,although agglomerates are still present(Fig. 3, middle column and upper row).However, at 200ºC, compression is moreintense and fibers are detectable under-neath the coating layer. The surface iseven smoother and agglomerates are lev-elled out, although they are to someextent still noticeable. In microrough-ness, no evident change occurred (Fig. 3,lower row).
A comparison of ESEM micrographsof dry surface treated paper consisting offreeze-dried coating powder (FFiigg.. 44, left)with the blade-coated reference paper(Fig.4,middle) shows no major visual dif-ference in surface structure. Surfaceopenness, resulting from the dryingmethod and mentioned earlier, has beenconsiderably reduced,which can be seenin the upper right micrograph of Fig. 4.This is most probably the result ofreduced particle agglomerate size, whichgives a more homogeneous surface.When preparing the powder throughfreeze-drying instead of using spray-dry-ing, the agglomerate size reduced (Fig. 4,top row, left).This is also seen from theparticle size distributions in TTaabbllee VVII,where the spray-dried powder has larger
agglomerate sizes. The cross-sectionmicrographs (Fig. 4, lower row) show adifference in coating penetrationbetween the reference and DST, wherethe reference paper has more unevencoating penetration. If coating layers con-taining either freeze-dried or spray-driedcoating powders are compared, we seethat the freeze-dried materials produce adenser and more uniform cross-direc-tional coating layer structure (Fig. 4,lower row). Because there is practicallyno coating penetration occurring in DSTpapers, there is the possibility of higherpaper bulk compared to water-basedconventionally coated papers.
CONCLUSIONSDry surface treated paper forms coatingsurface structures similar to those ofwater-based dispersion coatings withsimilar coating raw materials.The surfacetreatment, with application and surfacesmoothing, is more or less integrated intoone compact process step. Hotter fixingtemperatures and increased roll nip lin-ear load improves the surface smooth-ness. However, defects such as fibersshowing through the coating layer andpolymer sticking to hot rolls are still aproblem that requires attention.The small particle and agglomerate sizeachieved with freeze-dried coating powder improves the smoothness of the
3. ESEM micrographs describing the effect of fixing temperatures (left to right
column: 120, 160, and 200°C) at a dwell time of 12 ms and a linear load of 20
kN/m. The freeze-dried coating powder consisted of 100 pph CaCO3
and 10 pph
of latex A. The bar of the micrographs in the upper row equals 500 µm and in
the lower row 20 µm.
4. Surface and cross-section ESEM micrographs of dry surface treated papers
(left column: freeze-dried coating powder, 200°C/20 kN/m; right column: spray-
dried coating powder, 200°C/25 kN/m) and blade-coated reference paper (mid-
dle). The bar on the micrographs equals 20µm.
24 TTAAPPPPII JJOOUURRNNAALL MAY 2004
COATING
coating layer and decreases surfaceopenness. Surface strength is equivalent,but improvements in out-of-plane coat-ing structure and material preparationare still needed. We also need to furtherinvestigate the viscoelastic behavior ofpolymer binders and learn more abouttheir effect on structure and strength ofthe coating layer. TJ
ACKNOWLEDGEMENTThe authors gratefully acknowledge theNational Technology Agency of Finland(TEKES), Metso Paper, Inc., and DyneaChemicals Oy for their financial support.
LITERATURE CITED1. Maijala, J., Putkisto, K., Grön, J.,
“Effect of coating powder compo-sition and process conditions ondry surface treated paper proper-
ties,” TAPPI 2002 CoatingConference Proceedings, TAPPIPRESS, Atlanta, Georgia, USA, p.501.
2. Maijala, J., Putkisto, K., Nyberg,T.R., “A novel dry coating methodfor paper,” TAPPI 2001 CoatingConference Proceedings, TAPPIPRESS, Atlanta, p. 287.
3. Järnström, L., Wikström, M.,Righdahl, M., Nordic Pulp PaperRes. J. 15(2): 88(2000).
4. Gane, P. A. C., Wochen.Papierfabrik. 129(4): 176(2001).
5. Preston, J., Nutbeem, C., Parsons,J., Jones, A., Paper Tech. 41:33(2001).
6. Knappich, R., Burri, P., Sandås, S.,Lohmüller, G., Wochen.Papierfabrik. 128(10): 649(2000).
7. Athanasios, I. L., in Handbook ofIndustrial Drying, (A. S. Mujumdar,
Ed.), Marcel Dekker, New York,1995, pp. 295-325.
8. Hasan, M., and Mujumdar, A. S., inHandbook of Industrial Drying (A.S. Mujumdar, Ed.) Marcel Dekker,New York, 1995, pp. 763-764.
9. Satoh, T., Kawanishi, T., Shimizu,R., Kinjo, N., J. Imaging Sci. 35(6):373(1991).
10. Wikström, M., Carlsson, R.,Salminen, P., “Influence of latexbinder viscoelastic properties onhigh temperature calenderingrunnability,” 20th PTS CoatingSymposium Proceedings, R. Sangland H. Runge (Eds.), München,Germany, 2001, pp. 42-41.
11. LePoutre, P., and Hiraharu, T., J.Appl. Polym. Sci. 37(7):2077(1989).
12. Al-Turaif, H., and LePoutre, P.,Prog. Org. Coat. 38(1): 43(2000).
13. Sipi, K.M., and Oittinen, P.T., J.Imaging Sci. Tech. 44(5):442(2000).
Received: August 27, 2002Accepted: January 2, 2004
This paper is also published on TAPPI’s web site <www.tappi.org> andsummarized in the May Solutions! forPeople, Processes and Paper magazine
(Vol. 87 No. 5).
VI. Size distribution of the particle agglomerates for freeze-dried (left column)
and spray-dried (middle and right column) DST coating powders.
SPRAY-DRIED
PARTICLE FREEZE-DRIED SPRAY-DRIED 70 pph GCC/30 pph
SIZED 100 pph GCC/ 100 pph GCC/ polymer pigment/10 pph
DISTRIBUTION 10 pph latex A 10 pph latex A latex A/20 pph starch
D(10%), µm 0.8 3.3 9.2D(50%), µm 2.7 11.9 21.7D(90%), µm 12.6 26.6 45.2
INSIGHTS FROM THE AUTHORSThere is keen interest today in developing high effi-ciency, all online paper surface treatment concepts. Tosome degree, the rewetting of the paper web reducesthe efficiency of the surface treatment section andpaper quality. The dry surface treatment technique is anattempt to tackle several issues: (1) combined coatingand calendering, (2) elimination of the after-drying sec-tion through dry powder application, and (3) no rewet-ting of the web, giving a higher quality potential.
Dry pigment coating of paper had not been studiedpreviously. This research is a part of a continuum. Ourgroup has been studying this method since about 1997.
The most difficult aspect of this research was toachieve sufficient repeatability and functionality of thecoating powder in respect to electrostatic phenomenaand thermomechanical fixation. Our goal was a down-to-earth understanding of the process.
We discovered it is possible to produce dry surfacetreated papers with surface properties matching con-ventionally produced coated papers. This gave birth toa sincere enthusiasm to achieve a commercial processproducing printable coated papers.
This is a new technique, which now needs to be ver-ified in large-scale pilot conditions (i.e. coating powderproduction and application technique). It will openinteresting opportunities in respect to rebuild situa-tions, where in many cases the space available restrictsthe options for upgrading.
Our next step will focus on the dry-state preparationof coating powders and on the short time-scale func-tionality in respect to deposition and powder handling.
Maijala and Putkisto are with Tampere University of Technology,Paper Machine Automation, Korkeakoulunkatu 3, 33720Tampere, Finland; Johan Grön is with Metso Paper Technology,Wärtsilänkatu 100, 04400 Järvenpää, Finland. Email Maijala [email protected], Putkisto at [email protected], or Grön [email protected].
Maijala Putkisto Grön
PEER-REVIEWED EMISSIONS
Sulfur present in coal and residual oil ispredominantly oxidized to form sulfur
dioxide (SO2) during combustion.A small
fraction of this SO2
further oxidizes toform sulfur trioxide (SO
3) and may be
emitted as sulfuric acid (H2SO
4).
Combination boilers at many pulp millsroutinely co-fire coal or residual oil withbark. Previous studies show that a signifi-cant portion of the SO
2generated in com-
bination boilers is captured in the woodresidue ashes.Models correlating SO
2cap-
ture to fuel composition have also beendeveloped [1]. However, the effects of co-firing on H
2SO
4emissions is not well doc-
umented. The purpose of this study wasto quantify the emissions of H
2SO
4from
combination boilers and develop modelscorrelating the acid gas emissions to boil-er operating parameters and fuel compo-sition. This study also investigated theeffectiveness of venturi scrubbers forH
2SO
4capture.
BACKGROUND AND LITERATURE REVIEW
SO2, SO
3, and H
2SO
4formation
in boilersThe mechanisms of SO
2, SO
3, and H
2SO
4
formation in fossil fuel-fired boilers havebeen the subjects of several studies.A typ-ical coal or oil-fired boiler can be dividedinto three reaction zones: the combustionzone, the post-flame zone, and the catalyt-ically active zone (FFiigg.. 11). Sulfur dioxide,which is the primary product of fossil fuelsulfur combustion, is formed in the com-bustion zone. Some of this SO
2is further
oxidized to form SO3
in the post-flame
perature decreases to 200°C, nearly 99%of the SO
3would combine with water to
form H2SO
4[7].
The U.S. Environmental ProtectionAgency (EPA) estimates in its emissionfactor compilation (AP-42) that 0.7% ofthe sulfur in coal and 1% of the sulfur inresidual oil is potentially emitted as SO
3
and gaseous sulfate (H2SO
4) from boilers.
It is also estimated that 95% of the fuelsulfur is emitted as SO
2[8, 9]. SO
3forma-
tion has been attributed to the abovemechanisms both in stoker-fired and pul-verized coal-fired boilers [4,10].Althoughthe reaction mechanisms themselves arenot affected, the extent of these reactionsmay be affected by the method of coal fir-ing. However, these impacts have notbeen investigated and quantified.
The reactions described above couldalso take place in a combination boilerthat burns wood residues along with fos-sil fuels. Wood residue ashes contain sig-nificant amounts of calcium, magnesium,and potassium,plus moderate amounts ofmanganese, aluminum, and iron, presentas their respective oxides [11, 12]. Thereactions in the post-flame zone of thecombination boiler would be similar tothose observed in fossil fuel-fired boilers,with alkali metal sulfates being the pre-dominant end products. However, unlikein fossil fuel-fired boilers, the reactions inthe catalytically active zone would occurbecause of the catalytic nature of woodresidue ash. Wood residue ash is knownto have small particle sizes and contains asignificant amount of carbon [13].Thesecharacteristics of wood residue ash make
zone.At the prevailing high temperaturesof 1450°C to 1550°C in the post-flamezone, the oxidation reactions are inde-pendent of the oxygen content in the fur-nace and result in super-equilibrium con-centrations of SO
3[2, 3].The SO
3formed
in the post-flame zone can further reactwith the oxides of sodium, potassium,and calcium found in the ash to form sta-ble sulfates.Additional oxidation of SO
2to
SO3
also occurs in the catalytically activezone,where the temperatures range from975°C at the superheater to about 400°Cat the economizer. The degree of oxida-tion of SO
2in this zone depends on the
SO2
level (hence, fuel sulfur), percent O2,
and boiler tube fly ash deposits. Fly ashdeposits in coal-fired boilers containoxides of silica,aluminum,and iron,whilefly ash deposits in oil-fired boilers con-tain oxides of vanadium. These metaloxides act as catalysts in the oxidation ofSO
2to SO
3[2, 4, 5].
The SO3
formed in the catalyticallyactive zone is hygroscopic in nature andabsorbs moisture until its moisture con-tent is in equilibrium with the moisturein the flue gas [6].The extent of this reac-tion depends on the concentration ofSO
3, the flue gas temperature, and the
moisture content of the gas stream. Asindicated in TTaabbllee II, at a given flue gasmoisture content, the amount of sulfuricacid formed increases as the flue gas tem-perature decreases. For instance, at tem-peratures around 260°C and a moisturecontent of 8%, prevalent around theeconomizer outlet, nearly 88% of the SO
3
would be converted to H2SO
4.As the tem-
Sulfuric acid emissions from combination boilers
VIPIN K. VARMA AND ASHOK K. JAIN
ABSTRACT: Four coal-bark boilers and one residual oil-bark boiler were tested for sulfuric acid (H2SO4)emissions. Emissions were found to depend on the bark-to-sulfur ratios in the fuel and on the stack oxygen,a surrogate for the oxygen level in the superheater and economizer sections. An emission model represent-ing this two-fold dependence has been developed for use as a predictive tool. With a few exceptions, H2SO4emissions measured during this study were considerably lower than the U.S. Environmental ProtectionAgency’s emission factor compilation (AP-42) estimates for coal and residual oil firing. At the two boilersequipped with venturi scrubbers, sulfuric acid testing was carried out both at the inlet and outlet of the scrub-bers. The efficiency of H2SO4 removal across the scrubbers ranged from 25.7% to 85.2%.
AApppplliiccaattiioonn:: This paper provides a better tool for estimating sulfuric acid emissions from sources such ascombination boilers burning wood residues, or when the widely-used AP-42 emission factor estimates arenot appropriate.
VOL. 3: NO. 5 TTAAPPPPII JJOOUURRNNAALL 25
26 TTAAPPPPII JJOOUURRNNAALL MAY 2004
EMISSIONS
it very similar to activated carbon, aneffective catalyst for the oxidation of SO
2
to SO3. Activated carbon adsorbs signifi-
cant quantities of SO2
and oxidizes it toSO
3through reactions occurring on the
carbon surface [14, 15]. Thus, woodresidue ash,having the properties of acti-vated carbon, could act as a catalyst inthe oxidation of SO
2to SO
3, subsequent-
ly forming sulfuric acid in the super-heater and economizer sections of thecombination boiler.
SO2
and H2SO
4capture in
boilersPublished research suggests that fuel sul-fur is captured by the alkali metal oxidesin the fuel ashes.According to an Electric
Power ResearchInstitute (EPRI)study, SO
2is con-
verted to form sta-ble sulfates of sodi-um, potassium, andcalcium because ofinteractions withalkali metal oxidesin the ash [2].Thisretention of sulfurthrough the forma-tion of metal sul-fates causes signifi-cant reductions inSO
2emissions.
Studies conducted on coal-fired boilershave indicated that sulfur retention is afunction of alkali metal oxide content inthe ash, which in turn is directly relatedto the potential for forming metal sul-fates [16]. For pulverized coal-fired boil-ers with electrostatic precipitators, SO
2
removal through retention in fly ash wasestimated to range from zero for a coalwith very low alkalinity, to around 40%for a 0.5% sulfur coal with 35% alkalinemetal content in the ash [17].
Wood residue ashes also contain sig-nificant quantities of calcium, potassium,and magnesium oxides that contribute toSO
2capture [12]. SO
2capture efficien-
cies in eight combination boilers firingcoal, oil, and wood residues ranged fromzero to 80% and were found to be direct-ly proportional to the bark-to-sulfur ratioin the fuel being burned [1].The bark-to-sulfur ratio in the fuel indicates thecapacity of bark to capture SO
2through
interactions with alkali metal oxides,thereby forming metal sulfates. SO
2emis-
sion reductions from a residual oil-barkcombination boiler equipped with a drygravel bed scrubber ranged from 12% to55% according to an EPA study [18].Analyses of the particulate catch fromoil-fired, coal-fired, and combination boil-ers have shown that metal sulfatesformed as a result of SO
2capture
accounted for 25% to 75% of the totalcatches [19-21].
It is logical to expect a similar reduc-tion in sulfuric acid emissions from boil-ers firing sulfur-containing fuels.However, there is only limited evidencein the literature suggesting SO
3/H
2SO
4
capture by alkali metal oxides in the cat-alytically active zone. The EPRI studyinvestigated the effect of adding a pul-
verized magnesium oxide additive to theresidual oil being burned in the boiler.Atan additive concentration of 1000 ppm,sulfuric acid emissions were 50% lowerthan the base case with no magnesiumadditive [2]. The researchers concludedthat the magnesium in the residual oiladditive could have promoted the forma-tion of magnesium sulfates in the catalyt-ically active zone of the boiler. However,the same magnesium additive whenadded at the outlet of the economizer, inan attempt to reduce plume opacity, didnot affect the sulfuric acid emissions.This would suggest that alkali metalinteractions with sulfuric acid occur inthe superheater and economizer sec-tions, at temperatures higher than thoseobserved at the economizer outlet.Measured declines in SO
3concentration
have also been observed in coal-firedboilers when the flue gases passedthrough the superheater and economizersections.These declines were also attrib-uted to interactions with fly ash particles[22].
H2SO
4capture in control
devicesDry particulate control devices, such aselectrostatic precipitators (ESPs), are pri-marily used to remove fly ash particulatefrom flue gases and are not expected tocapture “free” sulfuric acid [23]. Whensubjected to packed bed scrubbers, sul-furic acid, which exists in the gas phaseat temperatures around 200°C, self-nucle-ates to form a fine sub-micron mist.Thisfine sulfuric acid mist may pass througha packed bed scrubber without beingcaptured, as these scrubbers are notdesigned to capture fine particulate anddo not allow for capture by inertialimpaction or Brownian diffusion [24].On the other hand, venturi scrubbersused for both particulate and gaseouspollutant control could be effective incapturing sulfuric acid mist. However,studies have not been conducted to veri-fy or quantify this phenomenon.
TESTING AND ANALYTICALPROCEDURES
Emission test methodsA modified method based on EPAMethod 8 and NCASI Method 8A wasdeveloped and used for H
2SO
4/SO
2sam-
pling [25, 26]. FFiigguurree 22 shows the exper-imental setup. This modified method is
1 = 80% isopropyl alcohol2 = filter to capture breakthrough3 = blank impinger4 and 5 = 3% peroxide
Heated Probe, 400 ˚F
Heated Filter,
480 ˚F
Two Sulfuric Acid/SO2 Trains
1 4
2
3 5
2. Experimental setup for the simul-
taneous sampling of acid gases.
I. SO3
conversion to H2SO
4vapor at
various flue gas temperatures.
TEMPERATURE, SO3
CONVERSION
°C TO H2SO
4, %
430 3.85370 14.3320 47.54290 70.54260 87.50200 98.86175 99.74
Process Steam
Tubular Air Heater
Pulv. Coalor
Residual OilOverfire
Air
Superheater
SteamDrum
Economizer
Underfire Air
AirIn
Flue Gases
Post-FlameZone
Catalytically ActiveZones
Co
nv
ecti
ve
Pa
ss
CombustionZone
To Ash Handling SystemCoal In
1. Sketch illustrating various regions of a fossil fuel-fired
boiler.
VOL. 3: NO. 5 TTAAPPPPII JJOOUURRNNAALL 27
EMISSIONS
similar to Method 8 in that it uses theprinciple of selective solvent absorption(SSA) and captures SO
3/H
2SO
4in 80% iso-
propyl alcohol (IPA) and the SO2
in 3%hydrogen peroxide. However, the draw-back of using Method 8 without a filteron combination boilers is the potentialfor positive biases in the sulfate measure-ments because particulate matter is cap-tured in the IPA along with SO
3/H
2SO
4.
Method 8A, developed and tested byNCASI as an alternative to Method 8,usesa heated quartz filter for capturing par-ticulates, thereby eliminating the poten-tial for interference from particulate sul-fate [26]. The modified method used inthis study is similar to Method 8A in thatit uses a heated quartz filter for capturingparticulates. Sampling and sample recov-ery procedures are discussed in detailelsewhere [27].
CO2
and O2
concentrations in thestack were measured using a Bacharachor Fyrite combustion gas analyzer.Promulgated EPA test methods wereused to measure stack velocities andmoisture content.
Fuel sampling and analysisWe recorded fuel firing rates during allthe tests. Samples of coal, oil, and barkwere collected from the testing facilities.Coal samples were collected from thepulverizers or the coal feeders and barksamples were collected from the barkconveyors. Oil samples were taken fromthe barrels feeding the oil guns. A com-posite sample was later sent out for ele-mental analysis.
Analytical proceduresThe impinger samples collected duringtesting were analyzed for sulfate usingsuppressed ion chromatography (IC).Weused an anion exchange column, togeth-er with a carbonate/bicarbonate eluent,for isocratic separation and quantifica-tion of the anions of interest.
SOURCES TESTEDFour coal-bark boilers and one residualoil-bark boiler were tested during thisstudy. Two of the four coal-bark boilerswere equipped with dry ESPs and twowere equipped with venturi scrubbers.The residual oil-bark boiler wasequipped with an ESP. TTaabbllee IIII summa-rizes the sources tested and the sulfurcontent in the fossil fuel burned.The sul-fur content in the bark ranged from0.02% to 0.04% on a wet basis.
Testing was carried out at the outletof the dry particulate control devices atMills A through C as the control deviceswere not expected to have any effect onthe emissions of “free” or “unassociated”acid gases. Testing was carried out bothat the inlet and outlet of the venturiscrubbers at Mills D and E.
RESULTS AND DISCUSSIONSulfuric acid emissions
TTaabbllee IIIIII shows results relative to sulfuricacid, along with percent of heat inputsfrom fossil fuel and bark, bark-to-sulfurratios in the fuel,and stack oxygen levels.In the case of Mill A, natural gas account-ed for approximately 20%-25% of thetotal heat input,with the remainder com-ing from coal and bark. Only the contri-butions from coal and bark are shown asa percent of the total heat input in TableIII. The bark-to-sulfur ratio in the fuel,expressed as metric tons bark as firedper kilogram sulfur, has been used as ameasure of the capacity of bark to poten-tially interact with the sulfuric acid.Stackoxygen levels were measured during theruns and were found to reflect the oxy-gen levels in the post-combustion zoneof the boiler.
Two-fold dependence of
H2SO
4emissions
Bark-to-sulfur ratios ranged from 0.07 to0.31 metric tons bark/kg S in the coal-bark boilers and from 0.41 to 0.54 metrictons bark/kg S in the oil/bark boiler.H
2SO
4emissions ranged from 1.0-7.9
kg/1012 J for coal-bark boilers and from1.05 to 2.28 kg/1012 J for the oil-bark boil-er. The emissions seemed to depend onthe bark-to-sulfur ratio in the fuel and onstack oxygen, a surrogate for the oxygenlevel in the superheater and economizersections. The trends observed at Mill Aindicated the possibility of this two-folddependence. Tests 2 through 4 fromstudy No. 1 (0.086 to 0.108 metric tonsbark/kg sulfur) resulted in H
2SO
4emis-
sions ranging from 1.0-1.15 kg/1012 J.Theemissions reduced as the bark-to-sulfurratio increased, suggesting increasedinteractions with bark as its availabilityincreased. However, test 1 in study No. 1and all the tests in study No. 2 were car-ried out at higher stack oxygen levels.Stack oxygen levels ranged from 7.2% to10.2% during these tests, while it rangedfrom 5% to 5.5% during tests 2,3 and 4 ofstudy No. 1. Mill data indicated that thecorresponding post-combustion oxygenlevels were also higher and ranged from6% to 8.4%. The H
2SO
4emission during
test 1 of study No. 1 was higher thanthose observed in tests 2 through 4 inspite of having a higher bark-to-sulfurratio,suggesting that the elevated oxygenlevel in test 1 may have played a role indetermining the emissions. Additionally,H
2SO
4emissions from study No. 2 were
considerably higher those observed dur-ing tests 2 through 4 of study No. 1, inspite of the comparable or, in most cases,higher bark-to-sulfur ratios.These resultsalso suggest that post-combustion zoneoxygen may be playing a role in deter-mining the sulfuric acid emissions fromthe boiler.
Similar indications of the dependenceon oxygen were observed at Mill C.H
2SO
4emissions during tests 1 and 2
were ~ 65% higher than those observedduring the other tests, in spite of thehigher bark-to-sulfur ratios. The stackoxygen levels during tests 1 and 2 were9% and 10%, respectively, while theyranged from 6.8% to 7.5% during theother tests.With the mechanisms of SO
3
formation in fossil fuel boilers in mind,these results suggest that there may beincreased potential for sulfuric acid for-mation at higher levels of boiler oxygen.If true, this would corroborate the theo-ry that sulfuric acid emissions are gov-erned by the reactions occurring in thecatalytically active zone of the boiler.
II. Summary of testing program.
HEAT SULFUR % HEATMILL INPUT FOSSIL OTHER IN FOSSIL FROM CONTROLCODE (109 J/h) FUEL FUEL FUEL, % FOSSIL FUEL DEVICE
A 823 P. Coal Bark, NG 1.27b 40 to 66 ESPB 696 R. Oil Bark 2.44c 23 to 28 ESPC 580 P. Coal Bark 2.31c 30 to 35 ESPD 370 Coala Bark 1.60b 56 to 73 VenturiE 802 P. Coal Bark 1.40b 58 to 74 Venturi
astoker firing; bdry basis; cas fired; ESP-electrostatic precipitator; Venturi-venturi scrubber; P. Coal-pulverized coal; R. Oil-residual oil; NG-natural gas; NA-not available
28 TTAAPPPPII JJOOUURRNNAALL MAY 2004
EMISSIONS
Emission reductions from com-
bination boilers According to AP-42 estimates,0.7% of thesulfur in coal and 1% of the sulfur inresidual oil is potentially emitted as SO
3
and gaseous H2SO
4from boilers [8, 9].
The sulfuric acid emission data in TableIII indicate that 0.14% to 0.71% of sulfurin coal was emitted as sulfuric acid fromcoal-bark boilers and 0.26% to 0.49% ofsulfur in oil was emitted as sulfuric acidfrom the oil-bark boiler.
Emission reductions have been calcu-lated by comparing the AP-42 factors of0.7% for coal and 1% for residual oilagainst the emissions expressed as % sul-fur.These emission reductions representthe sulfuric acid captured in the boilerdue to interactions with alkali metaloxides. With the exception of two testsconducted at Mill D, sulfuric acid emis-
sions were significantly lower than theAP-42 estimates.These results are not sur-prising given that bark contains alkalimetal oxides that could potentially cap-ture the acid gases.
Empirical model for sulfuric
acid emissions The results obtained during this studysuggest sulfuric acid emissions dependon stack oxygen content which is a sur-rogate for post-combustion zone oxygencontent, and the bark-to-sulfur ratio inthe fuel. Selected emission data, encom-passing the range of bark-to-sulfur ratiosand oxygen levels observed during thisstudy, were fitted using a nonlinearmodel of the type given below:
EE == AA∗∗OO ++ BB∗∗OO22 ++ CC∗∗XX
where:
E = sulfuric acid emissions, kg/1012 JO = stack oxygen, %X = metric tons bark/kg sulfur in fuel.
The emission data used for model devel-opment are shown in Table III. Theparameters A, B, and C were determinedto be 0.162, 0.0195, and –1.431 respec-tively.
The above empirical model can beused to estimate sulfuric acid emissionsfrom combination boilers burning fossilfuels. However, the significance of themodel assumptions and their potentialramifications need to be consideredbefore usng the model as a predictivetool. The model used here assumes thatstack oxygen levels and bark-to-sulfurratios are the only variables affecting sul-furic acid emissions.The reliability of themodel depends largely on the validity of
III. Sulfuric acid emissions from combination boilers.
MILL CODE TEST HEAT INPUT TONS BARK/ STACK EMISSIONSa EMITTED EMISSION
(FUEL TYPE) NO. FUEL, % BARK, % kg SULFUR O2, % (KG/1012 J) AS H
2SO
4REDUCTIONSb, %
AA ((CCooaall))cc 1 40.1 40.7 0.258 7.5 1.59d 0.31 55.5Study No. 1 2 60.0 21.9 0.092 5.5 1.07 0.14 79.9
3 56.9 24.3 0.108 5.5 0.99d 0.14 80.44 65.5 22.3 0.086 5.0 1.15 0.14 80.2
AA ((CCooaall))cc 1 52.9 26.7 0.128 7.2 2.39d 0.36 49.1Study No. 2 2 49.7 27.0 0.138 7.2 2.13 0.34 51.8
3 56.8 25.9 0.116 9.0 2.61 0.36 48.34 54.0 27.1 0.128 9.0 2.67 0.39 44.45 53.6 28.5 0.134 9.3 2.57 0.38 46.16 44.4 25.2 0.144 9.5 2.50 0.44 36.77 52.8 24.0 0.116 10.2 2.96d 0.44 36.9
CC ((CCooaall)) 1 31.0 69.0 0.300 10.0 3.30d 0.45 36.42 29.8 70.2 0.318 9.0 3.67d 0.52 26.43 34.9 65.1 0.252 7.5 2.14 0.26 63.44 33.7 66.3 0.266 6.8 1.99 0.25 64.75 34.2 65.8 0.260 7.0 1.32 0.16 76.96 34.1 65.9 0.260 7.5 1.54 0.19 73.0
DD ((CCooaall)) 1 56.3 43.7 0.162 12.0 4.75 0.55 22.42 58.8 41.2 0.146 11.0 4.15 0.46 34.33 73.1 26.8 0.076 10.0 7.94 0.71 —4 63.4 36.6 0.120 11.0 6.87 0.70 —
EE ((CCooaall)) 1 62.4 32.7 0.124 11.5 3.68 0.44 37.12 68.1 27.7 0.096 9.5 3.70 0.40 42.93 57.6 37.3 0.154 9.0 3.45d 0.44 37.14 68.5 26.4 0.092 10.5 3.53 0.38 45.75 73.7 21.6 0.070 9.5 4.02 0.40 42.9
BB ((RR.. OOiill)) 1 26.7 73.3 0.454 8.5 2.28 0.49 52.02 27.4 70.9 0.426 8.5 2.15d 0.44 56.03 23.1 74.9 0.536 7.9 1.05 0.26 74.04 28.0 70.0 0.414 8.0 1.44 0.29 71.0
ato convert from kg/1012 J to lb/109 BTU multiply by 2.32; bcompared to AP-42 estimatescnatural gas accounted for the remainder of the heat input at Mill A; ddata used for model development
VOL. 3: NO. 5 TTAAPPPPII JJOOUURRNNAALL 29
EMISSIONS
these assumptions. For instance, the fol-lowing conditions could change thedynamics of the reactions occurring inthe catalytically active zone, therebyaffecting the sulfuric acid emissions:
1. If the stack-post combustion oxygenlevels deviate significantly from thelevels observed during this study,their effect on sulfuric acid forma-tion or emissions may change andmay not be represented by the abovemodel.
2. The effect of fly ash deposits has notbeen considered in this model.Fireside tubes heavily coated with flyash deposits could contribute toincreased SO
3formation through cat-
alytic oxidation, thereby making it avariable in the analysis.
3. If there are wide variations in theamount of alkali metal oxides in barkbetween loads, the bark-to-sulfurratio will no longer be a good meas-ure of the capacity of ash to capturesulfuric acid.This would affect thepredictive capability of the model.
FFiigguurree 33 is a plot of the actual vs. pre-dicted sulfuric acid emissions.The empir-ical model can be used to predict sulfuricacid emissions for most of the cases,withthe exception of two of the tests con-ducted at Mill D.The two outliers in Fig.3 correspond to the tests where sulfuricacid emissions were approximatelyequal to the AP-42 estimates, as statedearlier. The high sulfuric acid emissionsmay have been caused by one of the fol-lowing two scenarios :
A reduction in the alkali metal oxidecontent between loads may havereduced the level of interaction withSO
3/H
2SO
4, thereby resulting in increased
emissions.
An increase in the activity level of thefly ash deposits or variations in the coalsulfur content between loads may alsohave caused increased emissions.
The empirical model developed heredoes not incorporate these dynamicchanges in boiler operation and hencecould not predict the emissions.
Impact of venturi scrubbers on
sulfuric acid emissions TTaabbllee IIVV provides the sulfuric acid emis-sions at the inlet and outlet of the venturiscrubbers at Mills D and E.The emissionsat the scrubber inlet represent theuncontrolled emissions and incorporateonly the interactions with alkali metaloxides occurring inside the boiler. Theemissions at the scrubber outlet, whencompared against the emissions at thescrubber inlet, provide the efficiency ofsulfuric acid removal across the venturiscrubber. The efficiency of H
2SO
4
removal ranged from 64.5% to 85.2% onthe venturi scrubber at Mill D and from25.7% to 62.4% at Mill E.
Sulfuric acid is expected to self-nucle-ate and form a fine mist (aerosol) whenquenched in the venturi scrubber. Theresults in Table IV indicate that venturiscrubbers were effective in partiallyremoving this sulfuric acid mist. Theseresults are not surprising given that ven-turi scrubbers have been proven to beeffective in capturing fine particulatematter [28]. Published studies indicatethat at a given particle diameter, the par-ticulate removal efficiency across theventuri scrubber increases with anincrease in the pressure drop across theventuri. Hesketh correlated particle col-lection efficiencies at specific particlediameters for a rectangular venturiscrubber operated over a range of pres-sure drops [29]. For instance, the scrub-
ber used during that study captured 40%of the particles, with a mean aerodynam-ic particle diameter of 1 mm at a pres-sure drop of 178 mm H
2O. The capture
efficiency for the same particulateincreased to 70% when the pressuredrop increased to 254 mm H
2O.
Reliable measurements of the venturipressure drops were not available on thescrubbers at Mill D and Mill E. However,they were expected to range from 152-191 mm H
2O at Mill D and approximate-
ly 140 mm H2O at Mill E.The higher sul-
furic acid removal efficiencies at Mill Ecould have been a result of the associat-ed higher venturi pressure drops. Thecapture efficiencies could presumably behigher on high-energy venturi scrubbersoperated at higher pressure drops.
SUMMARY AND CONCLUSIONS
H2SO
4 emissions ranged from 1.0-7.9
kg/1012 J for coal-bark boilers and from1.05-2.28 kg/1012 J for the oil-bark boiler.With a few exceptions, the sulfuric acidemissions were significantly lower thanAP-42 emission factors for coal and resid-ual oil. The sulfuric acid emission data
PR
ED
ICT
ED
EM
MIS
ION
S, k
g/1
012J 10
5
00 2 4 6 8 10ACTUAL EMMISIONS, kg/1012J
Used for model developmentNot used for model development
Tests 3 and 4 Mill D
3. Actual vs. predicted sulfuric acid
emissions.
IV. Sulfuric acid capture across venturi scrubbers.
SULFURIC ACID SULFURIC ACID % H2SO
4
TEST NO. HEAT INPUT EMISSIONS – INLETa EMISSIONS – OUTLETa REMOVAL ACROSS
MMiillll DD COAL, % BARK, % (kg/1012 J) (kg/1012 J) SCRUBBER
1 56.3 43.7 4.75 1.69 64.52 58.8 41.2 4.15 1.16 71.93 73.1 26.8 7.94 1.17 85.24 63.4 36.6 6.87 1.34 80.5
MMiillll EE1 62.4 32.7 3.68 1.99 45.92 68.1 27.7 3.70 2.31 37.53 57.6 37.3 3.45 2.03 41.24 68.5 26.4 3.53 2.62 25.75 73.7 21.6 4.02 1.51 62.4
a to convert from kg/1012 J to lb/109 BTU multiply by 2.32
30 TTAAPPPPII JJOOUURRNNAALL MAY 2004
EMISSIONS
generated during this study were used todevelop a model correlating the emis-sions to fuel composition and stack oxy-gen (related to post-combustion oxy-gen).This model can be used as a predic-tive tool to estimate the sulfuric acidemissions from combination boilers.Sulfuric acid removal efficiencies acrossventuri scrubbers ranged from 25.7% to85.2%.
Sulfuric acid emissions from the com-bination boilers were in most caseslower than the individual AP-42 emissionfactors for coal and residual oil. Theseresults highlight the beneficial impactsof co-firing. TJ
LITERATURE CITED1. “Sulfur capture in combination
bark boilers,” Technical Bulletin No.640, NCASI, New York, 1992.
2. “Sulfate Formation in Oil-FiredPower Plant Plumes – Vol. 1,”Brookhaven National Laboratoryfor Electric Power ResearchInstitute (EPRI), New York, NewYork, USA, 1983.
3. Hunter, S.C., “Sulfur oxides emis-sions from boilers, turbines, andindustrial combustion equipment,”Proceedings from the Workshopon Primary Sulfate Emissionsfrom Combustion Sources, 1978.p. 13.
4. Szalach, P. J., "New York StateElectric & Gas Corporation's expe-rience with sulfur trioxide and sul-furic acid mist," Proceedings ofthe 1998 Conference onFormation, Distribution, Impact,and Fate of Sulfur Trioxide inUtility Flue Gas Streams; availableat <http://www.netl.doe.gov>.
5. Levy, E. K., "Effect of boiler opera-tions on sulfuric acid emissions,"Proceedings of the 1998Conference on Formation,Distribution, Impact, and Fate ofSulfur Trioxide in Utility Flue GasStreams; available at<http://www.netl.doe.gov>.
6. Hardman, R., Stacy, R., andDismukus, E., "Estimating sulfuricacid aerosol emissions from coal-fired power plants," Proceedingsof the 1998 Conference onFormation, Distribution, Impact,and Fate of Sulfur Trioxide inUtility Flue Gas Streams; availableat <http://www.netl.doe.gov>.
7. JANAF – Thermochemical Tables,Journal of Physical and ChemicalReference Data, AmericanChemical Society and AmericanInstitute of Physics for theNational Bureau of Standards,Washington, DC, 1985.
8. Compilation of air pollutant emis-sion factors - Chapter 1 -Bituminous and sub-bituminouscoal combustion (AP-42), U. S.Environmental Protection Agency,Office of Air Quality Planning andStandards, Research Triangle Park,North Carolian, 1995.
9. Compilation of air pollutant emis-sion factors - Chapter 3 – Fuel oilcombustion (AP-42), U. S.Environmental Protection Agency,Office of Air Quality Planning andStandards, Research Triangle Park,1995.
10. Homolya, J. B., Barnes, H. H., andCheney, J. L., "A characterizationof gaseous sulfur emissions fromoil and coal-fired boilers,"Proceedings of the 4th NationalConference on Energy andEnvironment, Cincinnati, OH., USEPA, 1976.
11. Nordin, A., Fuel 74(4): 615(1995).12. “Information on the sulfur content
in bark and its distribution to SO2emissions when burned as a fuel,”Atmospheric Quality ImprovementTechnical Bulletin No. 96, NCASI,New York, 1978.
13. Campbell, A. G., TAPPI J. 73(9):141(1990).
14. Hartman, M., Polek, J. R.,Coughlin, R. W., “Removal of sul-fur dioxide from flue gas by sorp-tion and catalytic reaction on car-bon,” Chemical EngineeringProgress Symposium Series67(115): 7(1982).
15. Siedlewski, J., Intl. Chem. Eng. Vol.5: 297(1965).
16. “Sulfur retention in coal ash,” EPA-600/7-78-153b, U. S. EnvironmentalProtection Agency, ResearchTriangle Park, 1978.
17. Davis, W. J., and Fiedler, M. H.,JAPCA 32(4): 395(1982).
18. Cheney, J. L., Winberry, W. T.,Pleasant, J. M., Grotecloss, G. B.,Analytical Letters 12(A2):155(1979).
19. Jaworowski, R. J., “Condensedsulfur trioxide... particulate orvapor?” APCA 23(9): 791(1973).
20. Crocker, B. B., Chem. Eng. Prog.71(3): 83(1975).
21. Royals, A. L. III., and Webb, C. A.Jr., “Particulate emissions testingof bark-oil fired power boilers,”TAPPI 1972 EnvironmentalConference Proceedings, TAPPIPRESS, Atlanta, Georgia, p. 127.
22. Jones, C, and Ellison, W., Power(July/August): 73(1998).
23. Farber, P. S., “Selecting systems tocontrol emissions,” J. Environ.Protection (December): 10(1992).
24. Schifftner, K. C., “CorrectingProblems in the Operation of WetScrubbers,” Proceedings of 77th
Annual Meeting of the AirPollution Control Association,1984, p. 84.
25. "EPA Method 8 - Determination ofSulfuric Acid Mist and SulfurDioxide Emissions from StationarySources," 40 CFR Part 60,Protection of Environment,Appendix A, 1994, U. S.Government Printing Office,Washington, D.C.
26. “Method 8A – Determination ofSulfuric Acid Vapor or Mist andSulfur Dioxide Emissions fromKraft Recovery Furnaces” methodsmanual, NCASI, Research TrianglePark, 1997.
27. "Emissions of Sulfuric,Hydrochloric, and HydrofluoricAcids from Combination BarkBoilers," Technical Bulletin No.837, National Council for Air andStream Improvement, NCASI,Research Triangle Park, NC, 2001.
28. Pilat, M, and Noll, K., "WetScrubbers," in Air PollutionEngineering Manual, 2nd Edition,John Wiley and Sons, New York,NY, p. 73, 2000.
29. Hesketh, H., "Wet Scrubbers," inAir Pollution Engineering Manual,1st Edition, Van NostrandReinhold, New York, NY, p. 80,1992.
Received : July 2, 2002Revised : September 15, 2003Accepted : September 26, 2003
This paper is also published on TAPPI’sweb site <www.tappi.org> and summa-rized in the May Solutions! for People,Processes and Paper magazine (Vol. 87
No. 5).
VOL. 3: NO. 5 TTAAPPPPII JJOOUURRNNAALL 31
EMISSIONS
INSIGHTS FROM THE AUTHORSSeveral previous studies have addressed the mecha-nisms of sulfuric acid formation in fossil fuel fired boil-ers; however, the impact of co-firing on sulfuric acidformation is not well-documented. Findings from thisstudy support information available in literature thatsulfuric acid formation and emissions from combustionare a result of catalytic reactions occurring in the super-heater and economizer sections.
These results also indicate that sulfuric acid emis-sions are reduced when we increased the bark-to-sulfurratio in the fuel being burned. This finding indicatesincreased potential for capture in the wood residue ash.This is similar to the phenomenon observed in the caseof SO2 emissions from combination boilers.
This study attempted to quantify sulfuric acid emis-sions and also investigate the impacts of process vari-ables like bark/fossil fuel mix and boiler oxygen. Themost difficult aspect of this study was to simulate therelevant test conditions needed to answer these ques-tions.
Results from this study indicate that, with the excep-tion of two tests, sulfuric acid emissions were lower
than the AP-42 estimates for sulfuric acid emissionsdue to fossil fuel firing. This further substantiates thebeneficial effects of co-firing coal or residual oil withwood residues. Mills could use the empirical modeldeveloped here as a predictive tool to estimate poten-tial sulfuric acid emissions from combination boilers.
Additional data encompassing a wider range of oper-ating parameters would help improve the applicabilityof the model. Information available in literature indi-cates that alkali metal oxides present in wood residuescapture the sulfur in fossil fuel in the form of sulfates.Future studies could look at the variability in alkalimetal oxide contents in wood residues and attempt tocorrelate that to sulfuric acid emissions.
Varma is senior research engineer andJain is regional manager,NCASI, Southern Region,402 SW 140th Terrace,Newberry, FL, USA. Email Varma at [email protected].
Varma Jain
Kimberly-Clark gives WMU the gift of technologywill be performed at WMU's McCrackenHall Paper Pilot Plant, and will involvefaculty and students from both the paperengineering program and theDepartment of Chemistry at WMU.
"With this gift,Kimberly-Clark makes animportant public statement about thevalue of the paper programs at WMU," saidJohn Bergin, president of the Paper
Technology Foundation and senior vicepresident of Stora Enso,in the press report.He added that K-C's gift "represents aninvestment in the future of our industry."
WMU and K-C have been in negotia-tions regarding the technology transfersince last June, and K-C executives madeseveral visits to the WMU campus to tourthe facilities and meet with faculty. Theschool's strong engineering program inpaper sciences was a deciding factor in K-C's decision to award the patents. "Of spe-cial importance is the existence of theschool's pilot paper facility and the oppor-
Kimberly-Clark Corp., Dallas, Texas,USA, has donated three new tech-
nologies dealing with paper mill sludgetreatment to Western MichiganUniversity's (WMU) Paper TechnologyFoundation. Seven patents are associatedwith the technology transfer. The dona-tion was officially presented on April 1,2004, in Kalamazoo, Michigan, USA,where WMU is located.
"It's a tremendous compliment tohave such an industry leader recognizethe University's faculty for its expertiseand entrust important technology to usso that it can be developed and commer-cialized in a way that will make a real dif-ference in the marketplace," said WMUPresident Judith I. Bailey in a pressaccount available on WMU's website(www.wmich.edu).The gift also includessupporting funds and equipment.
According to WMU, the technologytransfer will lead to three separateresearch areas, all related to the treat-ment of paper fiber and paper process-ing byproducts, to modify fibers forreuse in production of consumer prod-ucts. Most of the research on the patents
tunity it presents in conducting researchwith the donated technologies," CorrineSukiennik, director of Kimberly-ClarkWorldwide's Global Technology TransferProgram is quoted as saying. "In addition,the University's Paper TechnologyFoundation provides the support andguidance required to impact the future ofthe pulp and paper industry."
Dr.Said AbuBakr,Department of PaperEngineering chairperson, said that hefeels the technology gift will provideexciting research opportunities for WMUfaculty, and help the school recruit andsupport graduate students.“We expect itto lead to technical advances for theindustry and the development of addi-tional patents. With the success weexpect, the industry could enjoy theopportunity to add a portfolio of value-added products to its list of possibilities,”AbuBakr noted in the press report. TJ
Editor’s Note: If your school or compa-ny has research-related news you wouldlike to share with the industry, pleasecontact TAPPI JOURNAL Editor JanBottiglieri at 1 847-466-3891 or [email protected].
“We expect it to lead to technicaladvances for the industry and thedevelopment of additional patents.”
—Dr. Said AbuBakr
32 TTAAPPPPII JJOOUURRNNAALL MAY 2004
P.O. Box 105113Atlanta, GA30348, USA
CHARTER SPONSOR PLAN US$1,495 per month (net) witha 24x agreement. Includes:
• Logo on back cover of TAPPI JOURNAL • Logo and link on monthly email announcing the next edi-
tion of TAPPI JOURNAL to print and online subscribers • Logo on landing page of TAPPI JOURNAL on
www.tappi.org • Logo on downloadable .pdf version of TAPPI JOURNAL• Headline, paragraph and link to an article on sponsors’
website following TAPPI JOURNAL article links onwww.tappi.org
Become a TAPPI JOURNAL Sponsor. For more information, contact:
North AmericaEastern US & CanadaJonathan M. Kiger : [email protected] Center Drive, Suite 200 : Atlanta, GA 30338+1 770 395-9850 : Fax +1 770 551-9427
Central USCorinne Cerny : [email protected] Bridgemill Avenue : Canton, GA 30114+1 770 704-9184 : Fax +1 770 704-1804
Western US and Central and Western CanadaRegan Pickett: [email protected] Ashby St., Suite 111: Warrenton, VA 20186+1 540 428-3736: Fax +1 540 428-3739
Europe, Scandinavia, Far East and AustraliaAlan Brett/Murray Brett : [email protected] de las Cuevas 66, buzon 60,03759 Benidoleig (Alicante), Espana+ 34 96 640 4165, Fax +34 96 640 4022
TAPPI HeadquartersSales and Marketing ManagerVince Saputo : [email protected] Technology Parkway S.Norcross, GA 30092+1 770 209-7529 or +1 404 557-2678 USA, Fax +1 540 364-2951
May editions of TAPPI JOURNAL will be available at TAPPITown (Booth #334) at Paper Summit. Also, they will be
available at PulPaper in Helsinki, June 1-3.
Charter Sponsorships available...visit us at TAPPI Town orcontact your sales rep for details.
SUSTAINING SPONSOR PLAN US$995 per month (net)with a 12x agreement. Includes:
• Logo on back cover of TAPPI JOURNAL• Logo on downloadable PDF version of TAPPI JOURNAL• Text link on landing page of TAPPI JOURNAL on
www.tappi.org
Audience: 4,600+ opt-in online subscribers and 700+ print subscribers.
The future of industry research is at stake. You can make a difference. Become a TAPPI JOURNAL Sponsor.Papers presented in TAPPI JOURNAL are extremely important to people on the front line as they impactdaily productivity and develop research for tomorrow’s application. As a Sponsor, your company can makea difference. Sponsors will see to it that people essential to the industry’s success will continue to receivethe best research every month.
PEER-REVIEWED STOCK PREPARATION
Concentric mixing of hardwood pulp and water
Aklilu T. G. Giorges, David E. White, and Theodore J. Heindel
ABSTRACT We completed concentric mixing experiments with velocity ratios of up to 6 using hardwood pulp of 1.0%, 1.9%, and 2.9% consistency and water. By increasing the velocity ratio (ratio of inner:outer jet velocity), we found the inner jet spread angle to be larger and the downstream mixing region uniform. Furthermore, local consistency measurements show a flattening of the concentration profile with increasing velocity ratio, confirming mixing improves as velocity ratio increases. For the fiber stock tested, mixing was significantly dependent on the stock consistency when the velocity ratio is small (Rv ≅ 1). This result indicates that the fluid streams do not deliver the shear stress and turbulence required to fully dislodge the fiber network. Mixing results from hydrodynamic instabilities and macroscale variations, which lead to downstream nonuniformities. At higher velocity ratios when the flow is turbulent, mixing is significantly affected by the velocity ratio, but there is no clear indication that mixing is affected by the stock consistency. These trends are evidence that once the fiber network strength is overcome by shear stress and turbulence, the mixture behaves as a conventional Newtonian fluid in turbulent flow. Mixing at high velocity ratio results from microscale turbulence that leads to a relatively uniform downstream mixture. Application: This study demonstrates the importance of high velocity ratio in producing a uniform stock mixture in concentric thick stock dilution before the fan pump.
Concentric mixing before the fan pump, if not done properly, can significantly affect the spatial and temporal consistency and chemical uniformity of the stock leaving the approach flow area, leading to severe MD and CD nonuniformities in the final sheet. TAPPI recently published approach flow guidelines for concentric mixing used for thick stock dilution before the fan pump [1]. However, in view of the importance of thick stock dilution, it was considered that a rigorous analysis of concentric mixing was in order and the velocity ratio criteria reexamined.
When two fluid streams enter a concentric mixing region at different velocities, a high shear region forms at the interface between the two fluid streams. Instabilities at this interface cause vortex pairing, intertwining, and rollup. As the vortices evolve downstream, the annular stream cascades toward the center while the center jet disintegrates radially, enabling mixing. The degree of mixing can depend on the following [2]: the ratio of inner-to-outer pipe diameter, flow rate or velocity; the ratio of specific gravity between the two fluid streams; the inner and outer pipe Reynolds numbers; the pipe surface roughness; and any secondary pipe flows. When one of the constituents is a fiber suspension, additional parameters related to the fiber properties also affect the mixing process, which is further complicated by fiber flocculation [3-7].
The mixing of a relatively thick fiber stream with a dilute fiber stream or white water is common in the pulp and paper industry. Concentric mixing occurs both in chemical mixing and in the approach flow area immediately ahead of the fan pump, where thick stock (inner pipe) is diluted
2
to the proper headbox consistency with clear accepts, secondary and tertiary screen accepts, deaeration overflow, etc. (outer annulus) [7]. When pulp and water streams mix concentrically, there is effective mixing when the shear stress provides enough energy to disrupt the fiber network between the two fluid streams. However, when the shear stress at the interface is less than the shear needed to disrupt the fiber network, the fiber network may not fracture and mix effectively. In concentric mixers, the shear stress can be managed by varying the flow rates of the streams. In this study, the concentric mixing process is experimentally investigated using hardwood fiber stock and water streams. We compared experimental results with numerical predictions computed using the standard k-ε turbulence model. EXPERIMENTAL EQUIPMENT AND SETUP Concentric mixing involves mixing a “primary” fluid from an inner pipe with diameter d, volumetric flow rate q, and mean fluid velocity u, with a “secondary” fluid in an outer pipe of diameter D, volumetric flow rate Q, and mean fluid velocity v (Fig. 1). The primary fluid can have a specified species concentration Cp and the secondary fluid a species concentration Cs. One of the important operating parameters in the mixing process is the velocity ratio between the primary and secondary fluids (Rv = u/v). The purpose of concentric mixing is to obtain a uniform species concentration within a short pipe distance from the jet nozzle. The jet issuing from the center pipe may be divided into two regions, the potential core region and the entrainment or mixing region (Fig. 1). The characteristics of the potential core are identical to those of the primary fluid stream while the characteristics of the mixing region vary from those of the primary fluid to those of the secondary fluid. In this study, we evaluate the concentric mixing performance of a short fiber-water system. The primary fluid is a bleached hardwood pulp suspension (Table I) at one of three consistencies and is delivered to the mixing region through the center pipe. The secondary fluid is water.
Experimental Equipment The experimental system consists of two transparent concentric pipes in the test section, a primary fluid mixing tank, a secondary fluid supply tank, a discharge tank, a pump, and associated piping, valves, and flow meters (Fig. 2). The test section consists of a transparent inner pipe with inside diameter d = 2.54 cm and a pipe wall thickness of 0.32 cm (Fig. 3) and a transparent outer pipe with inside pipe diameter D = 6.35 cm. The inner pipe protrudes into the outer pipe approximately l = 39.4 cm after the 90o bend (Fig. 3). The outer pipe extends approximately L = 58 cm beyond the inner pipe trailing edge before exiting into the discharge tank. Although L = 58 cm, the actual mixing region captured by high-speed video is approximately 25.4 cm downstream of the inner pipe trailing edge, corresponding to a mixing region of approximately 4D. To measure the mixing efficiency (concentration profile), we introduced four sampling probes into the mixing pipe at a distance 3.9D downstream from the nozzle. The diameters of the sampling tubes are relatively large compared with the fiber size to avoid clogging, and the sampling tube walls are thin enough to minimize their effect on the flow field. The probes have outside diameters of 18 mm and wall thicknesses of 1 mm and are located at center and off-center of the main pipe (Fig. 4). We used high-speed video equipment (Olympus America Motion Analyzer with a frame rate of 1000 frames/sec) to qualitatively assess the mixing process.
3
Mixing Experiments Approach. During the experiments, the mean secondary fluid (water) velocity was held constant at approximately v = 1.3 m/s, while the mean primary fluid (hardwood pulp) velocity was varied between 1.3 m/s to 7.76 m/s, corresponding to a mean velocity ratio range of 1 < Rv < 6 (Table II). Pulp feed consistencies of 0.97%, 1.86%, and 2.91% were tested to address the effect of fiber concentration.
Experimental Analysis
We used probe samples to determine the concentration distribution in the concentric mixer. The uniformity (homogeneity) of the mixture was characterized by determining the second moment of fiber concentration (mixing) M for the pipe cross-sectional area. M
σ= 2
2
c is the square of the
variation coefficient and was previously defined by Gray [8] and Maruyama et al. [9] as,
dA1cc
A1M
2
A∫
−= , (1)
where A is the cross-sectional area of the main pipe, D is the main pipe diameter, c is the mean fiber concentration in the mixer, and c denotes the local fiber concentration in a given sample probe. The mean fiber concentration c can be written as:
oc
Qqqc+
=, (2)
where co, Q, and q represent the initial jet fiber concentration, the outer pipe volumetric flow rate, and the jet (inner pipe) flow rate, respectively. After computing the concentration distribution over the pipe cross section, M is approximated by the sum of the squares of the local fiber concentration difference from the mean value, written as follows:
∑
∑
=
=
−
= 4
1ii
i
4
1i
2i
a
ac
cc
M , (3)
where ai and ci are the area and pulp concentration of each sample, respectively. When the fluid is uniformly mixed, the second moment of mixing approaches zero. Hence, M characterizes mixing quality.
The mixing uniformity range U is defined as the normalized difference between the mean and local consistencies, as follows:
cccU −= (4)
This measure allows quantification of the change in the range of concentration distribution
relative to the fiber mean concentration. It characterizes the mixing quality and the degree of
4
difference of the concentration, compared with ideal streams where mixing would be complete and uniform. By using longer sampling times, we obtained stable and consistent sampling data. The maximum relative uncertainty in U was 1.7%; most of the data were below 1%. For M the maximum relative uncertainty was 4.8%; most of the data were below 3.5%. The measured consistencies themselves generally had uncertainties less than 1%.
EXPERIMENTAL RESULTS
Figure 5 illustrates typical high-speed video stop-action images of the mixing experiments for the highest consistency (2.9%) evaluated. The bright region downstream of the inner pipe nozzle is the mixing region, which increases radially as the fluids move downstream. The outer pipe boundary is clearly identifiable, and the inner pipe can be recognized by its tip, captured on the left-hand side of the image. The total length of the mixing region captured by the images is approximately 4D downstream of the nozzle exit. The white region is the fiber stock, and the mixing process can be seen from the dispersion of the pulp that had been introduced in the center jet. The sampling ports at 3.9D are located near the right-most “+” mark (visible at the very right-hand side of each image in Fig. 5). Effect of Velocity Ratio (R
v)
When the velocity ratio is near isokinetic (Rv = 1), there is a gradual radial increase in the jet directly downstream of the trailing edge of the inner pipe, even though the inner and outer jet mean velocities are nearly identical (Fig. 5a). This agrees with the flow character reported by Dahm et al. [10] for concentric mixers. Although the center jet wall thickness was very small in their case, they concluded that the boundary layer on both sides of the inner pipe introduced a wake. The evolution of the wake instability caused the two fluid streams to intertwine at the interface. In the current experimental geometry, boundary layers are present on both sides of the inner pipe, creating a velocity defect. The inner pipe has a finite thickness, resulting in wake formation.
Through inspection of multiple images of the Rv = 0.9 mixing process, large vortex rings and weave-like coherent structures are seen along the interface between the two fluids. These structures become unstable as the fluid moves downstream, with hydrodynamic instabilities and large-scale turbulent interactions that propagate downstream. This indicates that when the streams are near isokinetic, mixing results primarily from large-scale, low-intensity interactions. As the mean velocity ratio increases, the mixing region changes from snakelike to a uniform cone-like shape (Figs. 5b and 5c versus 5a). The inner jet spread angle increases (the jet spreads faster), the large vortex rings observed in the isokinetic case are no longer apparent, and the mixing intensity and level of entrainment increase. Mixing is transformed from a large-scale, low intensity process to a small-scale, high-intensity process. In addition, the inner jet and downstream mixing regions appear more uniform. This is the result of the increase in center jet flow rate, which increases the local fiber concentration of the mixture (see Eq. 2) and the flow turbulence, augmenting the mixing. It is also the result of the large velocity difference between the inner and outer fluids, which creates the shear region and enhances small-scale turbulent mixing, creating a more uniform mixture. Similar trends were observed at the lower thick stock consistencies investigated. The improvement in the mixing at higher velocity ratios is quantified using the second moment of mixing (M) and the mixing uniformity range (U) (Tables III-V and Fig. 6). As the mixture becomes more uniform, the difference between the samples and the mean mixture
5
concentration approaches zero, as do M and U. The decreases in both M and U at higher Rv are consistent with the qualitative experimental observations discussed above.
Effect of Stock Consistency
For the three thick stock consistencies investigated, the weave-like structure at the center becomes unstable with increasing downstream distance from the nozzle. However, at higher consistency and low velocity ratio, the weave-like structure is stabilized by the fiber suspension and network formation, while mixing quality degrades—particularly when the velocities of the streams are approximately equal. This is indicated from the qualitative decrease in jet spread angle and the increase in M and U for the isokinetic data at higher consistency and low velocity ratio (Fig. 6 and Fig. 7, Tables III and IV). This is a result of the increase in fiber network strength with concentration, thus requiring more energy to disrupt the fiber network and mix with the water in the annulus.
The consistency effect is not apparent when the flow is in the turbulent range (stock mean velocity near or higher than the estimated fully developed turbulent velocity Vturb, discussed below). At Rv of approximately 3 and 6, there are no clear indications that increasing fiber consistency has a consistent and significant effect in hampering the mixing process (Tables III-V; Fig. 6). This seems to indicate that once the shear stress needed to fracture the fiber network is delivered, the fiber stock mixes in a similar manner regardless of the fiber consistency.
NUMERICAL MODEL
Governing Equations The governing equations for conservation of mass, momentum, and concentration for steady, incompressible, turbulent viscous fluid flow with constant fluid properties are:
0xu
i
i =∂∂ (5)
( )
∂∂
+∂∂
µ+µ∂∂+
∂∂−=
∂∂
ρi
j
j
it
jij
ij x
uxu
xxp
xuu (6)
∂∂
σµ
+µ∂∂=
∂∂ρ
jc
t
jjj x
cScxx
cu (7)
All quantities have been time-averaged in the above equations, and ui (i = 1, 2) are the mean local velocity components in the axial and radial directions. Also, ρ is the fluid density, p is the time-averaged pressure, µ is the dynamic viscosity, µt is the eddy (turbulent) viscosity, c is the time-averaged local fiber concentration, Sc is the Schmidt number, and σc is the turbulent Schmidt number (specified as σc = 0.7 in our calculations). The governing equations are discretized and solved using FLUENT© computational fluid dynamics (CFD) software to simulate the mixing process as two turbulent miscible fluids with the same density and viscosity but different concentrations. FLUENT© uses a finite volume method to discretize the governing equations [11]. It was selected because it can be used to model the conservation equations of multiple fluid streams [12]. The eddy viscosity (µt) is specified through the standard k-ε turbulence model available in FLUENT© [12], allowing simulation of turbulent mixing. The standard k-ε model [13] is widely used because of its robustness, computational economy, and reasonable accuracy for a wide range of engineering problems. The basis of the model is that the eddy viscosity is defined by
6
ε
ρ=µ µ
2
tkC (8)
where Cµ is an empirical constant and k and ε are the turbulent kinetic energy and dissipation rates, respectively. These parameters are determined from the following transport equations
ρε−+
∂∂
σµ
+µ∂∂=
∂∂ρ k
jk
t
jjj G
xk
xxku (9)
k
CGk
Cxxx
u2
2k1j
t
jjj
ερ−ε+
∂ε∂
σµ
+µ∂∂=
∂ε∂ρ εε
ε (10)
where C1ε and C2ε are empirical constants, and σk and σε are the turbulent Prandtl numbers for k and ε, respectively. The Gk term represents the production of turbulent kinetic energy and is modeled by [12] 2
tk SG µ= (11) where S is the modulus of the mean rate-of-strain tensor defined by ijijSS2S ≡ (12) with the mean strain rate given by
∂∂
+∂∂
=i
j
j
iij x
uxu
21S (13)
In the standard k-ε model, the following constant values are used as defaults in FLUENT©: Cµ = 0.09, C1ε = 1.44, C2ε = 1.92, σk = 1.0, and σε = 1.3. Comments on the applicability of these values will be given below. NUMERICAL RESULTS The flow conditions are assumed to be axisymmetric to reduce the computational domain from three to two dimensions. The actual computation domain (Fig. 8) encompasses a radial distance of 3.175 cm (D/2) and an axial distance of 44.45 cm (7D) (1D length upstream of the trailing edge of the inner pipe and 6D length downstream). This region is discretized into a numerical computational grid of 36 x 300 nodes, with a slightly higher node density near the inner pipe trailing edge. Because turbulence enhances the mixing process, the turbulence model used to simulate mixing plays a major role in determining realistic predictions. It has been shown that the values of the standard k-ε model constants, C1ε and C2ε, affect the relative concentration of the mixing streams [14, 15]. Giorges and Heindel [16] showed that as C2ε increases, the length of the potential core of the inner jet decreases and the jet spread increases, and as C1ε increases, the potential core of the inner jet increases and the jet spread decreases. The latter trend is seen in Fig. 9, where change in C1ε from 1.7 to 1.88 results in a significant change in the numerically calculated concentration
7
profile. For this study, we identified specific C1ε and C2ε values that provide reasonable qualitative agreement between the numerically predicted and experimentally observed concentration profiles.
We observe experimentally that when the velocity ratio is large, there is no clear indication that the mixing process is dependent on the concentration of the fiber suspension. However, for the isokinetic case, the mixing process is strongly affected by the concentration. Therefore, universal values of C1ε and C2ε are not appropriate for all cases. However, when the flow is turbulent and the velocity ratio is large, C1ε = 1.88 and C2ε. = 2.4 provide reasonable agreement with all experimental data that satisfy these two constraints. Based on the Reynolds number alone, the streams are in the fully developed turbulent range. However, Reynolds number is not an appropriate measure of fiber suspension turbulence. A range of pulps and pipe diameters have been investigated to identify when fiber suspensions become turbulent [2, 17, 18]. Duffy [17] proposed Eq. 14 for the mean velocity Vw (m/s) at the onset of turbulence (also termed the onset of drag reduction) when the flow regime begins to transform from plug flow to transitional flow. (C is the oven-dried consistency in percent.) At this point, plug flow still exists, while a conventional Newtonian liquid (e.g., water) would already be in fully developed turbulence at the same bulk velocity [17]. Hemstrom et al. [19] suggested Eq. 15 for the mean velocity Vturb (m/s) for fully developed turbulent flow (onset of significant plug reduction) for an unbeaten, unbleached kraft pulp suspension:
Vw = 1.22 C1.40 (14) Vturb = 1.8 C1.4 (15)
Using Eqs. 14 and 15, we calculated velocities where the pulp suspension is considered turbulent (Table VI). Under conditions where the thick stock pulp is turbulent, we determined that a common set of constants, C1ε = 1.88 and C2ε. = 2.4, provide reasonable qualitative agreement of the experimentally observed concentration profiles with the numerical simulations. We reached this conclusion by comparing experimental and numerical concentration profiles at 3.9D for runs where the thick stock velocity exceeds the onset of drag reduction (comparing values in Table II versus Table VI) (Figs. 10-12). Under these conditions, we believe the shear stress at the interface is strong enough to overcome the fiber network strength. Thus, the two streams effectively mix as two miscible fluids. There are only small qualitative variations between the experimental and numerical results, with the largest variations near the mixer centerline (Figs. 10-12). (Differences between the experimental and numerical results cannot be fully explained by the uncertainties in the experimental consistency data, which, as mentioned above, were generally less than 1%.) The dashed lines in Figs. 9-12 represent the numerical values for the local concentrations, while the symbols represent the averaged values (either numerical or experimental) over the specific probe areas (Fig. 4). For example, in Fig. 10 at r(m) = 0, the numerical results indicate a point mass fraction of 0.0164, while the average numerical value over 0 ≤ r(m) ≤ 0.009 is 0.0143. The solid lines in Figs. 9-12 represent an approximation of the experimental concentration profiles, but are not regressed lines.
There is good qualitative agreement between the experimental and numerical results for those conditions that satisfy turbulent fiber suspension flow, as shown in Fig. 13. Similar results are found for other experimental conditions. From the experimental images, the white region at the center indicates the potential core and mixing region. The numerically simulated mixing process images show a darker region indicating the annular fluid (water) and a center, lighter region
8
representing the center jet and mixing region between the two extremes. The interface of the two streams cannot be exactly predicted because of the time-averaged and steady state nature of the numerical results, but Fig. 13 gives a reasonable indication of the interface. However, the vortex ring that can be seen at the interface in the experimental image is not observed in the numerical simulations. CONCLUSIONS The experimental and numerical concentric mixing results reveal that increasing the velocity ratio increases the mixing effectiveness. In this case, the turbulence and the shear stress at the interface between the two fluid streams increase, causing the fiber network to break and mix with the water. Comparisons of the second moment of mixing, Μ, and the mixture uniformity range, U, for various velocity ratios, Rv, show that increasing Rv decreases Μ and U, demonstrating mixing process improvement. When the velocity ratio is high, the second moment of mixing is a stronger function of Rv than consistency (where no clear trend exists). That implies that once the shear stress disperses the fiber network, the flow behaves as if two turbulent Newtonian fluid streams are mixing. We believe this is why the numerical simulations are in close agreement with the experimental results when the thick stock velocity exceeds that required for turbulent fiber suspension flow. When the fiber stream is turbulent, the concentric mixing process can be simulated using the standard k-ε model with C1ε = 1.88 and C2ε = 2.4. The results in Fig. 6 clearly indicate the importance of high velocity ratio in producing a uniform stock mixture in concentric thick stock dilution before the fan pump. Hence, industrial concentric mixing system design must allow for use of high velocity ratios in the concentric mixing area as one requirement for a uniform stock mixture. At low velocity ratios, mixing is inadequate, even after a significant downstream distance. TJ ACKNOWLEDGMENTS The authors acknowledge the funding and support of the Member Companies of the Institute of Paper Science and Technology. The authors also thank Adele Garner for her contributions to this work. The authors also acknowledge the use of FLUENT software from Fluent, Inc. in this study.
LITERATURE CITED
1. TAPPI TIP0404-54, Headbox Approach Piping Guidelines, 2000. Available online at <http://www.tappi.org/index.asp?rc=1&pid=6824&ch=8&ip=>.
2. Forney, L.J., “Jet Injection for Optimum Pipeline Mixing,” Encyclopedia of Fluid Mechanics Volume 2 - Dynamics of Single-Fluid Flows and Mixing, N.P. Cheremisinoff, Ed., Gulf Publishing Company, Houston, pp. 660-690 (1986).
3. Duffy, G.G., and Lee, P.F.W., APPITA J. 31(4): 280(1978)
4. Stenuf, T.J., and Unbehend, J.E., “Hydrodynamics of Fiber Suspensions,” Encyclopedia of Fluid Mechanics: Vol. 5 - Slurry Flow Technology, N.P. Cheremisinoff, Ed., Gulf Publishing Company, Houston, pp. 291-308 (1986).
5. Kerekes, R.J., and Schell, C.J., J. Pulp Paper Sci. 18(1): J32(1992).
6. Helmer, R.J.N., Covey, G.H., and Lai, L.C.-Y., APPITA J. 52(3): 197(1999).
7. Norman, B., and Tegengren, A., Paper Tech. 31(1): 42(1990).
8. Gray, J.B., “Turbulent Radial Mixing in Pipes,” Mixing: Theory and Practice, Vol. III, Ch.13. Academic Press, 1986.
9
9. Maruyama, T., Mizushina, T., and Hayashiguchi, S., Kagaku Kogaku Ronbunshu, 8(4): 1(1982).
10. Dahm, W.N., Frieler, C.E., and Tryggvason, G., J. Fluid Mech. 241: 371(1992).
11. Patankar, S.V., Numerical Heat Transfer and Fluid Flow, Hemisphere Publishing Corp., New York, 1980.
12. Fluent Incorporated, “FLUENT 5 Users Guide,” Lebanon, NH, Fluent, Inc., 1998.
13. Jones, W.P., and Launder, B.E., Int. J. Heat Mass Trans. 15: 310(1972).
14. Monclova, L.A., and Forney, L.J., Ind. Eng.Chem.Res. 34(4): 1488(1995).
15. Giorges, A.T., Forney, L.J., and Wang, X., “Numerical Study of Multi-Jet Mixing,” Transactions of the Institute of Chemical Engineers, Part A, Vol. 79, pp. 515-522 (2001).
16. Giorges, A.T., and Heindel, T., “Concentric Mixing of Two Similar Fluids,” AIChE 2000 Annual Meeting, November 12-17, 2000, Los Angeles, CA.
17. Duffy, G.G., “Flow of Medium Consistency Wood Pulp Fiber Suspensions,” 47th APPITA Annual General Conference Proceedings, pp. 507-514 (1993).
18. Duffy, G.G., “How to Determine Pipe Friction Loss for the Design of Stock Piping Systems”, Proceedings TAPPI Engineering Conference, pp. 247 (1979).
19. Hemstrom, G., Moller, K., and Norman, B., TAPPI J. 59(8): 115(1976).
Received: March 11, 2003 Accepted: January 28, 2004 INSIGHTS FROM THE AUTHORS
Why did you choose this topic to research? This project met an industry need, and it provided a challenging scientific problem.
How does this research either complement and support previous research you (or others) have done, or how does it differ from previous research? This project links directly to the Fluid Dynamics/Forming research effort at IPST.
What was the most difficult aspect of this research and how did you address that? The most difficult aspects of this research were the experimental work at the higher consistency levels, and the simulation work needed to agree with experimental results.
What did you personally discover from this research? What was most interesting or surprising about your findings? We confirmed the importance of high velocity ratio, and how dramatically velocity ratio affects thick stock mixing. One very interesting finding was the ability to use the k-e model to simulate turbulent mixing.
How might mills benefit from or use this information? Mills will benefit via improved design of the approach flow system to provide better mixed stock to the machine, improving MD uniformity.
What's the next step? Followup work on softwood mixing will be presented at the TAPPI 2004 Spring Technical Conference, May 2004. White is with the Institute of Paper Science and Technology at Georgia Tech, Atlanta, Georgia 30332, Giorges is with Georgia Institute of Technology, Atlanta, GA 30332; Heindel is with Iowa State University, Ames, Iowa 50011. Email White at [email protected].
10
This paper is summarized in the May 2004 issues of TAPPI JOURNAL (Vol. 3: No. 5) and Solutions! For People, Processes and Paper magazine (Vol. 87: No. 5).
Table I. Hardwood pulp fiber length distribution, curl, kink, and percent fines. (Fiber Quality Analyzer, OpTest Equipment, Inc., Hawkesbury, Ontario, Canada).
Sample 1 Sample 2
Mean length
Arithmetic (mm) 0.48±0.010 0.468±0.010
Length weighted (mm) 0.739 0.730
Weight weighted (mm) 0.926 0.917
Mean curl index
Arithmetic 0.053±0.002 0.054±0.002
Length weighted 0.056 0.058
Percent fines
Arithmetic (%) 13.2 13.6
Length weighted (%) 1.98 2.09
Mean kink
Kink index (1/mm) 1.09 1.09
Total kink angle (º) 17.0 17.3
Kinks per mm (1/mm) 0.63 0.65
11
Table II. Flow conditions used in the hardwood fiber stock-water concentric mixing experiments, with thick stock consistencies of 0.97, 1.86, and 2.91%.
Volumetric Flow RateConsistency Stream
(gal/min) (lit/min)
Mean Velocity
(m/s)
Mean Velocity
Ratio
Primary 10.4 39.4 1.29
Secondary 50.4 191 1.29
1.00
Primary 29.9 113 3.72
Secondary 50.3 190 1.29
2.89
Primary 58.5 221 7.28
0.97%
Secondary 50.0 189 1.28
5.68
Primary 10.6 40.1 1.32
Secondary 50.6 191 1.30
1.02
Primary 30.5 115 3.80
Secondary 50.1 190 1.29
2.95
Primary 56.4 213 7.02
1.86%
Secondary 49.2 186 1.26
5.56
Primary 9.56 36.2 1.19
Secondary 50.8 192 1.30
0.91
Primary 31.5 119 3.92
Secondary 49.8 188 1.28
3.07
Primary 50.2 190 7.64
2.91%
Secondary 50.2 190 1.29
5.92
12
Table III. Concentration distribution at 3.9D downstream for 0.97% hardwood fiber stock mixing.
Velocity ratio = 1.0 and mean consistency after mixing = 0.17%
Uniformity Sample # Consistency (%)
U=(c-c ) / c U Range M
1 0.091 -0.46
2 0.118 -0.30
3 0.369 1.17
4 0.198 0.17
+1.17
to -0.46
0.19
Velocity ratio = 2.88 and mean consistency after mixing = 0.36%
1 0.220 -0.39
2 0.272 -0.25
3 0.669 0.86
4 0.483 0.24
+0.86
to -0.39
0.12
Velocity ratio = 5.69 and mean consistency after mixing = 0.53%
1 0.441 -0.17
2 0.473 -0.11
3 0.684 0.29
4 0.574 0.082
+0.29
to -0.17
0.017
13
Table IV. Concentration distribution at 3.9D downstream for 1.86% hardwood fiber stock mixing.
Velocity ratio = 1.01 and mean consistency after mixing = 0.32%
Uniformity Sample # Consistency (%)
U=(c-c )/ c U Range M
1 0.148 -0.54
2 0.194 -0.39
3 1.148 2.59
4 0.435 0.36
+2.59 to
-0.54
0.61
Velocity ratio = 2.95 and mean consistency after mixing = 0.70%(Sample 2 was contaminated and not used)
1 0.421 -0.40
3 1.211 +0.73
4 0.877 +0.25
+0.73 to
-0.40
0.10
Velocity ratio = 5.57 and mean consistency after mixing = 0.99%
1 0.819 -0.17
2 0.851 -0.14
3 1.310 +0.32
4 0.996 +0.01
+0.32 to
-0.17
0.022
14
Table V. Concentration distribution at 3.9D downstream for 2.91% hardwood fiber stock mixing. (Samples were not obtained at lower velocity ratios because of sampling probe plugging.)
Velocity ratio = 5.92 and mean consistency after mixing = 1.60%
Uniformity Sample # Consistency (%)
U=(c-c )/ c U Range M
1 1.39 -0.13
2 1.45 -0.09
3 2.05 0.28
4 1.68 0.05
+0.28 to
-0.13
0.013
Table VI. The turbulent velocity required for pulp flow.
Consistency Vw = 1.22 C1.40
(m/s)
Vturb = 1.88 C1.4
(m/s)
0.97% 1.2 1.7
1.86% 2.9 4.3
2.91% 5.4 8.0
FIGURES
d D
Lw
potentialcore
mixingregion
secondary flow(Q, v, Cs)
primary flow(q, u, Cp)
interface between theinner and outer fluid
inner jetangle
Figure 1: Schematic representation of the concentric mixing process.
thickstock
city water
constanthead stuffbox
flowmeters
pumptest section discharge
tank
agitatorflow ratecontrolvalves
thick stockbypass
water overflow tosewer
Figure 2: Experimental mixing facility.
Q
q
v
u
d D
secondaryfluid (water)
primary fluid(thick stock)
l L
mixing region
Figure 3: Concentric pipe mixer test section.
Figure 4: Sampling probes cross-sectional view. Units are meters.
(a)
(b)
(c)
Figure 5: Images of 2.91% consistency hardwood fiber pulp and water mixing for various velocity ratios (Rv): (a) Rv = 0.91, (b) Rv = 3.07, and (c) Rv = 5.92. Flow is left to right.
Figure 6: Second moment of mixing (M) vs. velocity ratio (Rv) for 0.97, 1.86, and 2.91% consistency hardwood
pulp.
(a)
(b)
(c)
Figure 7: For Rv = 0.9-1.0, images of hardwood fiber pulp and water mixing at consistencies of (a) 0.97%, (b) 1.86%, and (c) 2.91%
secondaryfluid inlet
outerpipewall
axis ofsymmetry
primaryfluid inlet
inner pipe wallthickness
Figure 8: The axisymmetric computation domain.
(a)
(b) Figure 9: Experimental and numerical concentric mixing concentration profiles at 3.9D downstream for Rv = 5.68 and 0.97% thick stock consistency, standard k-ε model with (a) C1ε =1.7 and C2ε = 2.4 and (b) C1ε =1.88 and C2ε = 2.4.
Figure 10: Concentric mixing concentration profile at 3.9D predictions for Rv = 2.95 and 1.86% consistency by the
standard k-ε model for C1ε = 1.88 and C2ε = 2.4.
Figure 11: Concentric mixing concentration profile at 3.9D predictions for Rv = 5.57and 1.86% consistency by the
standard k-ε model for C1ε = 1.88 and C2ε = 2.4.
0 0.005 0.01 0.015 0.02 0.025 0.03 0.0350
0.005
0.01
0.015
0.02
0.025
r (m)
Mas
s fr
actio
n
Velocity ratio = 5.92
Consistency = 2.91 %
−. Numerical value
Average numerical values
o Experimental values
c1 ε = 1.88 and c
2 ε = 2.4
Figure 12: Concentric mixing concentration profile at 3.9D predictions for Rv = 5.92 and 2.91% consistency by the
standard k-ε model for C1ε = 1.88 and C2ε = 2.4.
(a)
(b)
Figure 13: Comparisons between the experimental and predicted mixing regions for Rv = 5.92 and c = 2.91%: (a) Representative experimental image and (b) standard k-ε model predictions.