TAPPI Journal - February 2016

W W W. TA P P I . O R G

JOURNAL

THE PAP E R AN D PACK AGING IND UST R IE S’ TEC H NIC A L R ES OUR C E

FEBRUARY 2016 | VOL. 15 NO. 2

69

PULPING

Characterizing latency removal in mechanical pulping processes part I: Kinetics Jiyang Gao, D. Mark Martinez, and James A. Olson

WASTEWATER TREATMENT 80

Biochemical methane potential of kraft bleaching effluent and codigestion with other in-mill streams Temesgen Mathewos Fitamo, Olli Dahl, Emma Master, and Torsten Meyer

91

PRINTING

Causes of back-trap mottle in lithographic offset prints on coated papers Gunnar Engström

PAPERMAKING

103

Effect of sheet temperature on web densification during calendering Ashok Ghosh and Peter W. Hart

PROCESS CONTROL

111

Support vector regression and model reference adaptive control for optimum control of nonlinear drying process C. Karthik, K. Valarmathi, and M. Rajalakshmi

WASTEWATER TREATMENT 127

Anaerobic treatment to improve sludge recovery at a deinked fiber pulp and paper mill Aleksandra Raˇciˇc Kozmus, Andreja Žgajnar Gotvajn, Aleksandra Lobnik, Nina Novak, Aljaž Klasinc, and Gregor Drago Zupanˇciˇc

If you had to choose a kaolin supplier based on one quality, which would you choose?

TAILORED SOLUTIONS SERVICE STABILITY LONGEVITY GLOBAL REACH PRODUCT DEVELOPMENT

The good news is you can have them all with Thiele, a privately held company serving papermakers worldwide since January 1, 1947.

Thiele Kaolin Company • Sandersville, GA (877) 544-3322 • thielekaolin.com

Tailoring kaolin to your needs.

February 2016 TABLE OF CONTENTS

WWW.TAPPI.ORG

VOL. 15 NO. 2

65 EDITORIAL

TAPPI JOURNAL’s new online peer review system goes live

Monica Shaw

69 PULPING

JOURNAL

Editor-in-Chief Scott Rosencrance +1 770 429-2753

[email protected]

Associate Editor Brian N. Brogdon +1 678 581-9114

[email protected]

Editorial Director Monica Shaw +1 770 367-9534 Vice President, Operations Eric Fletty

[email protected] [email protected]

PRESS Manager Jana Jensen +1 770 209-7242

[email protected]

Webmaster Trina Heath +1 770 209-7416

[email protected]

Characterizing latency removal in mechanical pulping processes part I: Kinetics

TJ EDITORIAL BOARD

David A. Carlson, Carlson Consulting [email protected], +1 847 323-2685

Jiyang Gao, D. Mark Martinez, and James A. Olson

80 WASTEWATER TREATMENT

Biochemical methane potential of kraft bleaching effluent and codigestion with other in-mill streams

Temesgen Mathewos Fitamo, Olli Dahl, Emma Master, and Torsten Meyer

91 PRINTING

Causes of back-trap mottle in lithographic offset prints on coated papers

Gunnar Engström

103 PAPERMAKING

Effect of sheet temperature on web densification during calendering

Ashok Ghosh and Peter W. Hart

Brian N. Brogdon, FutureBridge Consulting & Training LLC [email protected], +1 678 581-9114

Jere W. Crouse, JWC Consulting [email protected], +1 608 362-4485 Mahendra Doshi, Doshi and Associates [email protected], +1 920 832-9101 Marc Foulger, GL&V USA Inc. [email protected] +1 315-767-2920 Peter W. Hart, MeadWestvaco Corporation [email protected], +1 409 276-3465 Carl J. Houtman, USDA Forest Products Laboratory [email protected], + 1 608 231-9445 John A. Neun, Albany International [email protected], +1 518 275-8139 Steven P. Ottone, SNP Inc. [email protected], +1 919 641-7854 Gaurav Pranami, Imbed Biosciences [email protected], +1 515 230-7005 Arthur Ragauskas, Oak Ridge National Laboratory [email protected] Gerard J.F. Ring, University of Wisconsin—Stevens Point [email protected], +1 715-592-3732 Scott Rosencrance, Kemira Chemicals [email protected], +1 770 429-2753 Ricardo B. Santos, MeadWestvaco Corporation [email protected], +1 540 969-2426 Paul Wiegand, NCASI [email protected], +1 919 941-6417

111 PROCESS CONTROL

Support vector regression and model reference adaptive control for optimum control of nonlinear drying process

C. Karthik, K. Valarmathi, and M. Rajalakshmi

127 WASTEWATER TREATMENT

Anaerobic treatment to improve sludge recovery at a deinked fiber pulp and paper mill

Aleksandra Raˇciˇc Kozmus, Andreja Žgajnar Gotvajn, Aleksandra Lobnik, Nina Novak, Aljaž Klasinc, and Gregor Drago Zupanˇciˇc

»COVER CREDIT: Schematic of experimental setup for latency removal kinetics used by researchers Gao, Martinez, and Olson (p. 69).

Junyong Zhu, USDA Forest Products Laboratory [email protected], +1 608 231-9520

Join TAPPI today!

TAPPI JOURNAL is a free benefit of TAPPI membership, and is only available to members. To join TAPPI, to renew your TAPPI Membership, or to learn about other valuable benefits, visit www.tappi.org. TAPPI, 15 Technology Parkway S., Suite 115, Peachtree Corners, GA 30092, publishes TAPPI JOURNAL monthly. ATTENTION PROSPECTIVE AUTHORS: All papers published are subject to TAPPI JOURNAL’s peer-review process. Not all papers accepted for review will be published. Before submitting, check complete author guidelines at http://www.tappi.org/s_tappi/doc.asp?CID=100&DID=552877. Statements of fact and opinions expressed are those of individual authors. TAPPI assumes no responsibility for such statements and opinions. TAPPI does not intend such statements and opinions or construe them as a solicitation of or suggestion for any agreed-upon course of conduct or concerted action of any sort. Copyright 2016 by TAPPI, all rights reserved. For copyright permission to photocopy pages from this publication for internal or personal use, contact Copyright Clearance Center, Inc. (CCC) via their website: www.copyright.com. If you have questions about the copyright permission request process, please contact CCC by phone at +1 978 750-8400. To obtain copyright permission to use excerpts from this publication in another published work, send your specific request in writing to Editor, TAPPI JOURNAL, 15 Technology Parkway S., Suite 115, Peachtree Corners, GA 30092, USA; or by fax to +1 770 446-6947. Send address changes to TAPPI, 15 Technology Parkway S., Suite 115, Peachtree Corners, GA 30092, USA; Telephone +1 770 446-1400, or FAX +1 770 446-6947. www.tappi.org

MONTH 2007 | TAPPI JOURNAL

63

FEBRUARY 2016 | VOL. 15 NO. 2 | TAPPI JOURNAL

63

Valmet IQ �e smartest way to reach your quality goals

�e Valmet IQ quality management solution for the pulp and paper industry is a comprehensive new product family that helps you to reach your process and end product quality goals. Our products, applications and services cover all your needs, from replacement of individual equipment to building a full quality management solution. Each Valmet IQ solution is always based on our extensive industry knowhow and designed to move your performance forward.

www.valmet.com/iq

Editorial MONICA SHAW | EDITORIAL DIRECTOR [email protected]

TAPPI JOURNAL’s new online peer review system goes live

B

y definition, “peer review” is a “process by which something proposed (as for research or publication) is evaluated by a group of experts in the appropriate field.” Though not always without controversy, it is a widely accepted way to communicate, validate, and advance science in a variety of fields. This important process is at the heart of TAPPI JOURNAL (TJ), and TAPPI has recently implemented a new web-based system — PeerTrack by Allen Press — to expedite and improve this process for authors, editors, and reviewers. After researching various software packages of this type, TAPPI PRESS Manager Jana Jensen identified this one as a good solution for the TJ process. Now, after customization by the vendor to mirror TJ’s workflow, along with testing by Jana, myself, and members of the Editorial Board (Brian Brogdon, Peter Hart, Scott Rosencrance, and J.Y. Zhu), the system is configured and live. Powered by Allen Press’s Editorial Manager technology, PeerTrack helps automate the submission, tracking, and peer review process. It is available via the TAPPI website at http://www.tappi.org/Publications-Standards/TAPPIJournal/Submit/. Using the system, authors submit original and revised manuscripts, Editorial Board Members (EBMs) send manuscripts out for peer review, reviewers conduct reviews and return comments, and EBMs make final decisions. Users are assigned editorial, author, or reviewer roles — or multiple roles. Here are some of PeerTrack’s benefits for users: Authors – Intuitive interface is easy to use and manuscript turnaround is expedited through streamlined internal and external handling and access to status information.

Caption

Reviewers – Delivery of information is reliable, and reviewers can provide their comments and recommendations online. They also receive automatic reminders when feedback is due. EBMs and editorial staff – With PeerTrack, editors can see where manuscripts are at any stage of the reviewing process. Like reviewers, they receive automatic reminders of material that is due and can automatically send reminders to authors and reviewers. An author will need to register (circled in the figure) the first time he or she uses the new system. Reviewers will need to register the first time after receiving an EBM invitation from the new system. However, the new system is pre-populated with the names of many past TJ reviewers and will check to see if any information exists on that user. TAPPI and the TJ staff are very excited about the efficiencies the new PeerTrack system can provide. Again, it is available at http://www.tappi.org/Publications-Standards/TAPPI-Journal/Submit/. TJ FEBRUARY 2016 | VOL. 15 NO. 2 | TAPPI JOURNAL

65

4-6 April 2016

Manchester, UK

Nestle, VTT, Gruppo di X Gruppo srl, Inca Digital, EcoSynthetix and many more confirmed for 2016 conference programme! As the premier event in Europe dedicated to the specialty papers industry, Specialty Papers is the conference where organizations from across the supply chain turn for the most topical content, presented by industry experts. Earning a reputation as a “a rare opportunity for manufacturers and suppliers in the specialty papers business to gather, network, exchange ideas and search for business solutions” (Finch Paper), Specialty Papers 2016 will bring together more than 200 key decision makers, industry experts and end users, all in one room. Use the promotional code TAPPISPEU16 to save 10% off the current registration price. www.specialtypaperconference.com

“ An excellent blend of topics targeted to the specialty papers industry in a small conference setting with time for individual discussions with conference speakers.” Naeenah Paper “Very energizing to learn about new application ideas and growth opportunities.” Dow Chemical Company “A great learning experience that furthered my knowledge base and brought to life further value adds.” Georgia Pacific

#specialtypapers16

Scan the QR code to visit the website

@smitherspira

PEER-REVIEWED

PULPING

Characterizing latency removal in mechanical pulping processes part I: Kinetics JIYANG GAO, D. MARK MARTINEZ,

and

JAMES A. OLSON

ABSTRACT: Latency removal in the mechanical pulping process occurs in a continuous stirred-tank reactor and non-ideal mixing lowers the performance. In order to optimize the latency removal process and reduce the energy consumption in the operation, a kinetic study was carried out. In the study, the phenomenon of latency and knowledge related to latency removal were critically reviewed and discussed. Latency removal was characterized by the change of Canadian Standard Freeness (or freeness), and its dependences on treatment conditions, i.e., disintegration temperature, power input, pulp consistency and time, were determined. Kinetic models of latency removal for secondary refiner thermomechanical (TMP) and bleached chemithermomechanical (BCTMP) pulps have been developed, which were based on the rate of latency elimination characterized by the decrease of freeness. Application: These original models are available for use as a tool to predict the change of freeness in the industrial latency removal process for the purpose of optimizing the latency removal process and reducing the energy consumption in the operation.

L

atency is a familiarized phenomenon in the mechanical pulping industry. Ever since it was defined by Beath et al. [1], studies on latency formation and latency removal, along with the characterization techniques and factors affecting latency removal on different pulps (e.g., thermomechanical pulp [TMP], chemithermomechanical pulp [CTMP]), have been carried out by a number of authors [1-10]. In these studies, latency formation has been found to be a process of fiber deformation and fiber flocculation resulting from the changes of fiber morphology under extreme refining conditions. Latency removal has been described as a process of fiber deflocculation and fiber straightening that undoes all the changes occurring to the fibers during latency formation and restores the fibers to their original straight and isolated state.

Latency formation and characterization techniques Amorphous materials or amorphous regions within semicrystalline materials undergo thermal softening at certain temperatures, during which the materials change from a hard and relatively brittle state into a molten or rubber-like state. The temperature at which the material undergoes thermal softening is called glass transition temperature or thermal softening temperature. Lignin, hemicellulose, and the amorphous regions of cellulose are amorphous polymers that display thermal softening behavior around their glass transition temperatures. Previous investigations on thermal softening of wood components showed that lignin, hemicellulose, and cellulose

underwent glass transition at different temperatures [7,11-16]. When dry, all the polymers softened at high temperatures. Irvine [13] and Salmén and Olsson [16] reported that both dry lignin and dry hemicellulose softened at similar temperatures ranging from 130ºC to 190ºC, while crystalline cellulose has been found softened at above 220ºC [11,14]. When absorbing water, the glass transition of the lignin and hemicellulose were shifted to lower temperatures as a result of the plasticizing effect of water [11,13,17]. Both lignin and hemicellulose have hydroxyl and other polar groups that have the potential to be plasticized by water through the interaction between water molecules and the hydroxyl groups in lignin and hemicellulose. When the water content increases, the softening temperatures of the polymers lower. Under water-saturated conditions, the glass transition temperature of lignin, depending on the wood species, has been reported to be between 60ºC and 90ºC [13,16]. The transition temperature of hemicellulose approaches the freezing point of water. In comparison, the softening temperature of cellulose was not affected. The distinction in thermal softening behavior among lignin, hemicellulose, and cellulose explains how the fiber morphology changes in the refining process and how latency is formed. In the production of TMP, the fibers are subjected to a succession of compressive and shear stresses at high temperatures of 150ºC–170ºC and high consistencies of 25%–45%, which promotes a change in fiber shape as a result of the physical change of cellulose, hemicellulose, and lignin. Consequently, the fibers become extremely deformed and entangled. When the pulp is discharged from the refiner, either FEBRUARY 2016 | VOL. 15 NO. 2 | TAPPI JOURNAL

69

PULPING being cooled in the air or quenched in cold water, the deformations are retained. The most commonly accepted argument is given by Beath et al. [1] and Jones [2] who state that permanent deformation resulted from lignin stiffening is caused by a temperature change. When mechanical pulp is cooled, lignin in the fibers stiffens, leaving the fibers in their deformed shape. Although the elastic crystalline cellulosic microfibrils tend to resist the deformation, the elastic restoring stress in the cellulose is balanced by the stiffness of the deformed but substantially rigid hemicellulose-lignin matrix. This is undoubtedly the case for isolated fibers in the pulp, but for the majority of the fibers that are restrained in the fiber flocs, we believe that the mechanical forces (i.e., bending, friction, and hooking forces) [18] are dominating and preventing the fibers in the fiber flocs from straightening. From the previous discussion, it is clear that latency removal is a combined process of fiber deflocculation and fiber straightening. Industrially, during latency removal, mechanical action breaks up the fiber flocs and separates the fibers. Lignin softening results in the flow of lignin and hemicellulose, which leads to the relaxation of the internal stresses developed when the fibers are cooled. Under these conditions, the now elastic cellulose, which is restrained in a bent configuration, causes the fiber to return to its straightened state before it is released from the wood [1-2,4-5,7,9-11,19-21]. In addition, a number of authors indicate that fiber swelling, created through the uptake of water, is an additional force that aids in the straightening process [22-24]. This is speculated to result from softening of the hemicellulose and the amorphous cellulose in the fiber wall [25], which assists fiber straightening even at room temperature. As a generalized term to describe the changes in different pulp properties, latency or latency removal can be characterized by any pulp or paper property, including freeness, curl index, and dry sheet properties such as tensile strength (or tensile index), bursting strength (or burst index), tearing strength (or tear index), or wet-web properties. Among these, freeness, tensile strength and bursting strength are commonly reported [1-2,4-5,9-10]. A freeness test is essentially a filtration process that measures the rate of drainage of a diluted pulp suspension under standard conditions (TAPPI Standard Method T 227 om-94 “Freeness of pulp [Canadian standard method]”). Freeness has been reported to be influenced by fiber specific surface [26], fines content [19,27-28] and other pulp and fiber properties [28]. Although the introduction of freeness was based entirely on empirical deductions, an explicit theoretical expression was developed later by El-Hosseiny and Yan [29-30] as a function of pulp consistency, pulp viscosity, specific surface, hydrodynamic specific volume, and the average consistency of the pulp pad formed in a freeness tester. Among various pulp properties that are affected by latency, a 100% decrease in freeness has been reported after latency was removed [5,10]. As the most commonly used quality control parameter in mechanical pulping and the papermaking process, and a 70

TAPPI JOURNAL | VOL. 15 NO. 2 | FEBRUARY 2016

useful pulp property in characterizing strength properties of mechanical pulps [30-31], freeness itself is a useful property for the characterization of latency removal.

Latency removal and factors affecting latency removal As the mechanism of latency formation reveals, the way to remove latency is to undo the changes in fiber morphology through dispersing the latent pulp and straightening the fibers. In both laboratory and industrial practice, latency is removed by disintegrating the pulp suspension at low consistency and elevated temperature. In laboratory research, there are two standard methods used to remove latency: disintegrating pulp suspension at a consistency of 1.2% and a temperature of 85ºC in a British Standard Disintegrator for 15 min (TAPPI T 205 sp-02 “Forming handsheets for physical tests of pulp”); and recirculating a 2% consistency pulp suspension at a temperature of 85ºC for 2 min using Domtar Disintegrator (TAPPI T 262 sp-02 “Preparation of mechanical pulps for testing”). In industrial operation, latency is removed by passing the pulp through a latency chest, where high consistency (45%–50%) pulp is diluted to 3%–5% and is agitated at elevated temperature of 80ºC–90ºC for 20–60 min. In addition to typical latency removal at low consistency, a method of removing latency from medium consistency pulps by pumping has also been patented [32]. By pumping the pulps utilizing a centrifugal pump at a consistency of 8%–25%, the fluidization imparts sufficient energy to the pulp to remove the latency. Removing latency of mechanical pulps is dependent on the treatment conditions (i.e., disintegration temperature and energy of the mechanical action) and is affected by factors such as pulp consistency [1-4,9-10]. Disintegration temperature has been reported as a major factor in latency removal [1-2,4], which, as previously discussed, has been found to result from the change of viscoelastic properties, especially the thermal softening behavior of lignin, hemicellulose, and cellulose. In the latency removal operation, latent pulp is disintegrated at a temperature above the lignin softening temperature. As a result, the pulp is deflocculated and the stiffened lignin in the fibers is softened. The restoring stress of the elastic cellulose then returns the deformed fibers to their original straightness. Mechanical action plays an important role in removing latency. Beath et al. [1] examined the effect of mechanical action on latency removal through sampling at different times. Dawson et al. [4] evaluated the effect of mechanical action using a prototype Domtar Disintegrator. The results show that latency removal is a function of mechanical energy that involves both power intensity and time. To remove the same amount of latency, the application of higher power intensity will reduce the time required to complete the treatment. Beath et al. [1] proposed that the effectiveness of mechanical action on latency removal was “ascribed to the breaking up of flocs rather than to any direct effect of mechanical action on the fibers.”

PULPING Latency removal is affected by pulp consistency. Beath et al. [1] reported that for the range from 0.75% to 2%, consistency had no effect on latency removal. For higher consistency up to 15.6%, the higher pulp holding consistency resulted in higher fiber restraint, and less latency was removed. Dawson et al. [4] examined the hot disintegration of refiner pulp at a temperature of 85ºC but with consistencies from 1% to 3% using both a prototype Domtar Disintegrator and a British Standard Disintegrator. They concluded that consistency within this range had no effect on final freeness after latency was removed. Yet, it has been noted that both reported results neglected an important factor in the experiment — power intensity. The examination of power consumption for the disintegration of TMP pulp carried out in this study at consistencies from 1.2% to 3.0% using a British Standard Disintegrator showed that power increased dramatically as the consistency increased. From 1.2% to 3.0%, the power in the disintegration of a 2 L pulp suspension increased from 31 to 112 W. In addition, the effect of consistency on latency removal is through changing the rate of latency releasing rather than changing the final value after latency is completely removed. In consideration of these findings, the experimental data should be analyzed on the basis of constant power or energy input, and the consistency effect-related data reported by Beath et al. [1] and Dawson et al. [4] are not suggested to be directly used without prior analysis. The success in latency removal depends not only on understanding the influence of each factor in latency removal, but also on the determination of the relationship among these factors so that the optimal operation condition is able to be specified. In reality, latency chests in mechanical pulp mills are often designed empirically and graphically. Rather than achieving complete motion, a latency chest is operated in a zone-agitation mode. As a result, good mixing in the agitation zone and poor mixing outside the mixing zone simultaneously exist. The existence of non-ideal mixing (recirculation, bypassing, dead zone, etc.) lowers the performance of the chest. It is then essential to determine the kinetics of latency removal for the purpose of potential optimization of latency removal, as well as energy reduction, in mechanical pulping processes. This intensive and systematic study is not only to determine the change of fiber and pulp properties in latency removal and the mechanisms responsible for these changes, but also to better understand what happens to the fibers during latency removal, as well as to find solutions to energy reduction in latency removal. EXPERIMENTAL

Materials and methods The vessel used for the study is a fully baffled (four baffles with width of 1.9 cm) cylindrical vessel with a diameter of 22.9 cm and height of 20.0 cm. A top entering geometrically scaled Maxflo impeller (Chemineer Inc.; Dayton, OH, USA) was mounted on an instrumented shaft driven by a 0.25 kW

1. Schematic of experimental setup for latency removal kinetics.

variable speed motor with impeller off-bottom clearance maintained at 3.2 cm. The rotational speed was measured to ±1 rpm using a photoelectric sensor, and shaft torque was measured to ±0.002 Nm using an in-line strain gauge. A schematic of the experimental setup (vertical mixer) is presented in Fig. 1. The study was carried out on never-dried secondary high consistency refining TMP and bleached CTMP (BCTMP) pulps. High consistency (45%) Western hemlock TMP pulp was obtained from Howe Sound Pulp & Paper Ltd., and high consistency (25%) SPF (spruce [75%], pine [20%], and fir [5%]) BCTMP pulp was obtained from Quesnel River Pulp. Fiber lengths of each pulp sample were determined using OpTest Fiber Quality Analyzer (FQA) (OpTest Equipment Inc.; Hawkesbury, ON, Canada) after complete latency removal. The values of the mean lengths of TMP pulp are 0.862 mm (arithmetic), 1.947 mm (length weighted), and 2.376 mm (weight weighted), respectively. The mean lengths of BCTMP pulp are 0.894 mm (arithmetic), 1.601 mm (length weighted), and 2.255 mm (weight weighted), respectively. A series of batch experiments were designed. For each experiment, the pulp sample was added to the preheated FEBRUARY 2016 | VOL. 15 NO. 2 | TAPPI JOURNAL

71

PULPING Consistency, %

0.5

1

Power Intensity, 3.72 W/kg Temperature, °C

60

2

E1

5

E2

10

E3

30

e4

1.89 23

40

60

2.64 80

23

40

60

3.72 80

E10 e14

E23 E27

e7

E11 e15

E20 E24 E28

e5

e8

e6

e9

23

40

60

4.57 80

60

40

2

3

4

3.72

3.72

3.72

60

80

40

60

80

60

E36 e40 E44 e33 E37 e41 E45 e48 E51 e54

e57 E60 e63 E66

E12 e16 E18 E21 E25 E29 e31

e34 E38 e42 E46 e49 E52 e55

e58 E61 e64 E67

e13

e35

e59

Time, min

e17 E19 E22 E26 E30 e32

e39

e43

e47

e50

e53

e56

e62

e65

e68

I. Summary of the experimental conditions for the kinetic study of thermomechanical (TMP) pulp.

water to make a 5 L pulp suspension with designed temperature (i.e., 23°C, 40°C, 60°C, or 80°C) and consistency (i.e., 0.5%, 1%, 2%, 3%, or 4%). The suspension was then mixed at designed rotational speed (i.e., 600, 700, 800, or 900 rpm) for a certain length of time (i.e., 2, 5, 10, or 30 min). The power intensity calculated based on the rotational speed and the torque was 1.89, 2.64, 3.72, or 4.57 W/kg. The consistency of the pulp suspension studied was up to 5% . Since the treatments were carried out at low consistencies, the power intensity was calculated based on the weight of the suspension, rather than that of the fiber in the suspension. Detailed experimental conditions for the kinetic study of latency removal on TMP pulp are listed in Table I. Upon the completion of the treatment, the suspension was quenched to 20°C and diluted to 0.3% with cold distilled water. The freeness was measured according to TAPPI T 227 om94. Three replicate freeness tests on each treated pulp sample were carried out in order to determine the experimental errors, and the values reported in this paper are arithmetical averages. RESULTS AND DISCUSSION

Kinetic expression of latency removal of TMP pulp The initial freeness value of the pulp F0 was determined by direct measurement of the original untreated pulp sample. In this study, the initial freeness has been determined as 820 mL. The final freeness value of the pulp Ff was determined as the constant value of the freeness when all the latency was removed. In this study, Ff was determined as 220 mL, the freeness of the pulp sample being treated at a temperature of 80°C, consistency of 3%, and power intensity of 3.72 W/kg for 30 min. The standard deviations were determined for each treated sample, with the highest value determined as 5 mL. Since the values were too small to discern the difference from one another when shown in the figures, no standard deviations were included in any figure illustrating the variation of freeness. 72


2. Latency releasing at different power intensities and temperatures. All the experiments were carried out at a consistency of 1%.

The time dependent latency releasing process is shown in Fig. 2, which was characterized by the decrease of freeness at two different power intensities and four different temperatures. The initial rate drop in freeness was fast, but it slowed significantly as the treatment proceeded, which was more distinct at high temperatures. The sizes of fiber flocs in the originally highly flocculated pulp sample were seen to decrease rapidly in the first few minutes of the treatment, resulting in a rapid decrease in freeness. As the treatment continued, the freeness reached an asymptotic level, the socalled “floor-level” freeness, which is the characteristic freeness value for a pulp after all the latency has been removed. This value is expected to vary from pulp to pulp produced from different wood species and at different refining conditions. A simple model is proposed herein to model latency removal, expressed by the net change of the freeness (Eqs. [1] and [2]):

PULPING (1)

where:

(2) among which k and n are the rate constant and kinetic order of the model, respectively. Normalized freeness is used in the model expression, where F is the measured freeness, F0 is the initial freeness, and Ff is the final freeness, the “floor-level” value to account for the existence of the residual freeness discussed previously. The rate constant k is dependent on disintegration temperature of the treatment T, power intensity of the treatment ε, and pulp consistency Cm. k0, α, β and γ are the overall constant, the exponential function constant, the order to the power-law function, and the order to the powerlaw function, respectively. There are six unknowns (i.e., k, k0, n, α, β and γ ) in the kinetic expression. There are two methods [34] (i.e., integral and differential methods) available to determine the kinetic order of the proposed model of latency removal for a given set of disintegration conditions (energy input, temperature, and consistency). The integral method was used to determine the kinetic order n and rate constant k. When a value is assigned to the kinetic order, the model expression is able to be integrated, giving the solution as Eq. (3a) or (3b):

(3a) (3b)

Theoretically, the kinetic order can be found when a linear plot between ln

and t (when n=1) or

and t (when n≠1) is obtained, from which the rate constant k can be determined. Practically, the best linear relationship can be determined by comparing the R2 values of the linear trendlines. The experimental data was divided into two groups: one for model development (those beginning with capital letter “E” in Table I), and the other for model validation (those beginning with small letter “e” in Table I). Effect of disintegration temperature The change in freeness at different temperatures is plotted in Fig. 3, which shows the difference in the rate of latency re-

3. Decreasing freeness in latency removal at different temperatures. All experiments were carried out at a consistency of 1%.

4. The linear plot between 1/ T and lnkT verified Arrhenius-type temperature dependence of the rate constant.

moval at different temperatures. In the initial phase (first 5 to 8 min), the higher the disintegration temperature was, the faster the rate of latency releasing. The experimental data was then analyzed to determine the unknowns (i.e., the kinetic order n, and the exponential function constant α). It is obvious that when keeping all other experimental conditions constant and only changing the temperature, the rate of change in freeness will only be a function of temperature. For clarification, a subscript T was added to the rate constant to represent the temperature dependence. An Arrhenius-type expression was explored as the formula for the temperature dependence of the rate constant, which was expressed as:

(4) After the kinetic order was determined as 1.5 by the integral method, the corresponding rate constant kT at each temperature was determined. The result was then plotted to evaluate the temperature dependence of the rate constant. As shown in Fig. 4, a linear relationship between lnkT and 1/T was obtained, which verified that the Arrhenius-type expression for the temperature dependence of the rate FEBRUARY 2016 | VOL. 15 NO. 2 | TAPPI JOURNAL

73

PULPING

5. Latency releasing for pulp samples treated at different power intensities. All experiments were carried out at a consistency of 1%.

7. The linear plot between lnε and lnkε verified the power-law dependence of the rate constant on power input.

(5) A linear relationship between lnk ε and lnε was obtained (Fig. 7). In consideration of the results shown in Figs. 6-7, the value of β was determined as 1.

6. Relationship between treatment energy and latency releasing. All experiments were carried out at a consistency of 1% and a temperature of 60°C.

constant was appropriate for the kinetics of latency removal, where the numerical value of α was also determined as the slope of the trendline. Effect of power input The effect of power input on the rate of latency removal is illustrated in Fig. 5, showing that higher power input speeds up the rate of latency removal, which is more distinct in the first 10 min of the treatment during which the majority of the latency is removed. The data was analyzed in terms of treatment energy as well (Fig. 6), which shows that latency is continuously released as the treatment energy increases. The analysis on the effect of power and energy on latency removal suggests that the application of higher power shortens the time required to remove same amount of latency than lower power, which could potentially increase the production rate. Similar analysis as described previously was applied to evaluate the effect of power input on latency removal and to determine the rate constant kε at each power input. A powerlaw formula was proposed for the dependence of the rate constant on power intensity, which was expressed as: 74


Effect of consistency Similar to the change in freeness at different power inputs, the rate of freeness releasing increased slightly as the pulp consistency increased from 0.5% to 4% at a constant power intensity of 3.72 W/kg (Fig. 8), which occurred in the initial phase of the treatment. As the treatment proceeded, the fiber flocs were dispersed until all the fibers were unrestrained from the flocs and straightened, and the freeness reached the “floor-level” value, which remained constant with extended treatment time. Therefore, it is important to identify the right sampling time. If the sample is taken when the latency has been completely removed, the result from the evaluation on the effect of consistency on latency removal would be misleading. This explains how the reported conclusions of no effect of consistency (up to 3%) on latency removal were obtained in the literature [1,4]. A power-law formula was proposed for the dependence of the rate constant on consistency, expressed as:

(6) A linear relationship between lnkCm and lnCm was obtained (Fig. 9), from which the value of γ was determined as 0.6. Kinetic expression With the determination of kinetic order n, the exponential function constant α, and the orders to the power-law function β and γ, we obtained the complete kinetic expression of latency removal as:

PULPING

8. Decreasing freeness in latency removal at different consistencies. All experiments were carried out at a power intensity of 3.72 W/kg.

10. Model validation through comparison of experimentally measured and model predicted values of freeness.

(8)

9. The linear plot between lnCm and lnkCm verified powerlaw dependence of the rate constant on consistency for consistencies up to 4%.

(7) The model was then verified by comparing the predicted freeness values calculated using the model and the measured values for the experimental data group (those beginning with small letter “e” in Table I) that has not been used for model development. As presented in Fig. 10, the data are close to the 45° straight line, which concludes that the developed kinetic model well predicts the process of latency removal, characterized by freeness. In addition to the linear regression, a Matlab program was written to estimate the unknown parameters in the kinetic expression using nonlinear regression. Based on the Nelder-Mead optimization method, the Matlab function fminsearch was applied to find the optimum set of parameters that minimizes the discrepancy between the experimental data and the model’s predictions and brings the model’s predictions as close as possible to the experimental data. A set of parameters was obtained and the model was determined as:

which is in a good agreement with Eq. (7), which was derived by linear regression. In the operation of latency removal, mechanical pulp mills employ higher temperature and consistency (i.e., consistency of 5% and temperature of 85°C) than the temperatures and consistencies evaluated for the model development in this study. To determine if the model is applicable for industrial operation conditions, an experiment was carried out at a consistency of 5%, temperature of 85°C, and power intensity of 9.6 W/kg for 2 min. The predicted freeness (241 mL) using the model was found in a good agreement with the measured freeness (248 mL).

Kinetic expression of latency removal of BCTMP pulp A similar analysis was carried out on never-dried secondary high consistency BCTMP pulp. The effects of disintegration temperature, power intensity, and consistency on latency removal of the pulp are presented in Figs. 11-13. The kinetic order n, the exponential function constant α, and the orders to the power-law function β and γ are determined one after another. The kinetic expression of latency removal of BCTMP pulp obtained by linear regression is:

(9) The experimental data were then analyzed using the same Matlab program written for the TMP pulp, and the expression was determined as: FEBRUARY 2016 | VOL. 15 NO. 2 | TAPPI JOURNAL

75

PULPING

11. Decreasing freeness in latency removal at different temperatures. All experiments were carried out at a consistency of 1%.

13. Decreasing freeness in latency removal at different consistencies. All experiments were carried out at a power intensity of 3.72 W/kg.

12. Latency releasing for pulp samples treated at different power intensities. All experiments were carried out at a consistency of 1%.

(10) The expressions obtained by linear regression and nonlinear regression both work well in the predicting the change of freeness in latency removal (Fig. 14). In comparison with TMP pulp, latency removal of BCTMP pulp is faster. This is not only observed in the experiment but also evidenced by the higher rate constant in the kinetic expression of latency removal of BCTMP pulp. The lignin sulfonation occurring in chemical pre-treatment results in the weakening of the lignin-hemicellulose matrix through weakening the interactions of the lignin-pectin, lignin-protein, and pectin-protein bonds [35], which consequently improves fiber flexibility and fiber swelling as well as lowering the lignin softening temperature [9,36]. As discussed previously, the rate of latency removal is dominantly controlled by fiber deflocculation. Therefore, it is believed that the fiber changes caused by lignin sulfonation result in a weaker fiber network, making the fiber deflocculation process much easier than that of TMP pulp. 76


14. Comparison between the models determined by linear regression and nonlinear regression (Matlab program) in predicting the change of freeness in latency removal of BCTMP pulp.

CONCLUSIONS For latency removal of both TMP and BCTMP pulps at a temperature higher than lignin softening temperature, the release of latency proceeded rapidly in the initial phase followed by the second phase, in which the freeness releasing rate was much slower and approached asymptotically to a “floor level.” The kinetic expressions of latency removal for secondary refiner TMP and BCTMP pulps have been shown to follow a one-and-a-half order and a second order, respectively, involving disintegration temperature, pulp consistency, and power intensity of the treatment. Resulting from lignin sulfonation, complete latency removal of BCTMP pulp requires less time than TMP pulp at the same treatment energy. The analysis reveals that the potential to optimize the current operation of latency removal in pulp mills and energy reduction can be achieved by applying higher power intensity in the operation of latency removal. TJ

PULPING ACKNOWLEDGEMENTS The authors gratefully acknowledge the financial support from the Natural Sciences and Engineering Research Council of Canada through the Collaborative Research and Development program and through the support of our partners BC Hydro, FP Innovations, Catalyst Papers, Howe Sound Pulp and Paper, West Fraser Quesnel River Pulp, Canfor, Andritz, Arkema, Honeywell, WestCan Engineering, Aikawa Fiber Technologies, Ontario Power Authority, and CEATI international. LITERATURE CITED 1. Beath, L.R., Neill, M.T., and Masse, F.A., Pulp Pap. Mag. Can. 67(10): T423(1966). 2. Jones, H.W.H., Pulp Pap. Mag. Can. 67(6): T283(1966). 3. Alfthan, G.V., Pap. Puu 58(4): 572(1976).

6. Barbe, M.C., Seth, R.S., and Page, D.H., Pulp Pap. Can. 85(3): T44(1984). 7. Page, D.H., Barbe, M.C., Seth, R.S., et al., J. Pulp Pap. Sci. 10(3): 74(1984). 8. Harris, G. and Karnis, A., J. Pulp Pap. Sci. 12(4): J100(1986). 9. Htun, M., Engstrand, P., and Salmén, N.L., J. Pulp Pap. Sci. 14(5): 109(1988). 10. Karnis, A., Pap. Puu 75(7): 505(1993). 11. Goring, D.A.I., Pulp Pap. Mag. Can. 64(12): T517(1963). 12. Back, E.L. and Salmén, N.L., Tappi 65(7): 107(1982). 13. Irvine, G.M., Tappi J. 67(5): 118(1984). 14. Salmén, N.L., J. Mater. Sci. 19(9): 3090(1984). 15. Olsson A.-M. and Salmén, N.L. in Wood Water Relations (P. Hoffmeyer, Ed.), International Conference of COST Action E8, European Commission, Brussels, 1997, pp. 269-280. 16. Salmén, N.L. and Olsson, A.-M., J. Pulp Pap. Sci. 24(3): 99(1998).

4. Dawson, J.A., Maclellan, A.D.G., Shallhorn, P.M., et al., Pulp Pap. Can. 79(C): T124(1978).

17. Salmén, N.L. and Back, E.L., Tappi 60(12): 137(1977).

5. Mohlin, U.-B., Tappi 63(3): 83(1980).

18. Kerekes, R.J., Nord. Pulp Pap. Res. J. 21(5): 598(2006).

ABOUT THE AUTHORS Although many studies on latency removal have been carried out over the years, there was no kinetic model developed that allowed papermakers to examine the industrial latency removal process for the purpose of process optimization and energy reduction. The research here was to develop a kinetic model of latency removal and apply it for optimizing the use of energy in the recovery of latent properties of mechanical pulp. It was conducted as part of a larger research program at the University of British Columbia, “Electrical Energy Reduction in Mechanical Pulping and Pulp Processing,” which aims at reducing electrical energy consumption in mechanical pulping by 20% through scientific discovery and the development of new technology while maintaining or improving product quality and production. The most difficult aspect of this research was to determine the expression of the kinetic model developed by linear regression. A number of attempts were made with different expressions, including power law and exponential functions, before the kinetic expression was finalized. The model was validated by comparing the predicted freeness values with the measured freeness values, and we were surprised by the good result in that the model works very well in predicting the change of freeness in latency removal. Industrial latency chests operate under non-ideal mixing conditions, and the effectiveness of the operation relies on the compensation of other factors (e.g., more energy input for agitation, longer mixing time, etc.). The result of this research shows

Gao

Martinez

Olson

that applying higher power intensity has the potential to reduce the mechanical energy consumption and increase the production rate, leading to higher return on investment. This research was conducted using a “batch” reactor rather than a continuous stirred-tank reactor (CSTR). Future studies could look at the design and use of a CSTR system to develop a dynamic model to simulate the industrial latency removal process in a latency chest and validate the model through mill trials. Gao is research and technology advisor, SR & ED Division, Canada Revenue Agency, Surrey, BC, Canada. Martinez is director, Pulp and Paper Center, and professor, Department of Chemical and Biological Engineering, University of British Columbia, Vancouver, BC, Canada. Olson is associate dean for Research and Industrial Partnerships, Faculty of Applied Science, and professor, Department of Mechanical Engineering, University of British Columbia, Vancouver, BC, Canada. Email Gao at [email protected].


77

PULPING 19. Mohlin, U.-B., Sven. Papperstidn. 83(18): 513(1980).

33. Ein-Mozaffari, F., Bennington, C.P.J., and Dumont, G.A., Pulp Pap. Can. 105(5): T41(2004).

20. Mohlin, U.-B. and Alfredsson, C., Nord. Pulp Pap. Res. J. 5(4): 172(1990).

34. Fogler, H.S., Elements of Chemical Reaction Engineering, Prentice Hall Int. Inc., Englewood Cliffs, NJ, USA, 2006, pp. 267-271.

21. Seth, R.S., Pulp Pap. Can. 107(1): T1(2006). 22. Eriksson, I., Haglind, I., Lidbrandt, O., et al., Wood Sci. Technol. 25(2): 135(1991). 23. Cowan, W.F., Tappi J. 78(1): 101(1995).

35. Stevanic, J.S. and Salmen, N.L., J. Pulp Pap. Sci. 34(2): 107(2008). 36. Atack, D., Heitner, C., and Stationwala, M.I., Sven. Papperstidn. 81(5): 164(1978).

24. Retulainen, E., Niskanen, K., and Nilsen, N. in Paper Physics (K. Niskanen, Ed.), Fapet Oy, Helsinki, Finland, 1998, pp. 64-72. 25. Salmén, N.L., “Temperature and water induced softening behaviour of wood fiber based materials,” Ph.D. thesis, Royal Institute of Technology, Stockholm, Sweden, 1982. 26. Forgacs, O.L., Pulp Pap. Mag. Can. 64(C): T89(1963). 27. Clark, J.d’A., Tappi 53(1): 108(1970). 28. Biermann, C.J., Handbook of Pulping and Papermaking, Academic Press, San Diego, 1996, pp.137-142. 29. El-Hosseiny, F. and Yan, J.F., Pulp Pap. Can. 81(6): T113(1980). 30. El-Hosseiny, F. and Yan, J.F., Pulp Pap. Can. 81(6): T116(1980). 31. Paterson, H.A., Pulp Pap. Mag. Can. 37(12): 79(1936). 32. Prough, J.R., Torregrossa, L.O. and Bucklund, A., U.S. pat. 4,596,631 A (Jun. 24, 1986).

PAPERMAKING SCIENCE AND TECHNOLOGY The Best and Most Comprehensive Bookseries Ever Written on Papermaking

2ND EDITION ! O N W AVAIL ABBLE ALSO AVAIL A LE IN EBOOKS

10% Off Printed Books for TAPPI Members Volume 1 Volume 2 Volume 3 Volume 4 Volume 5 Volume 6 Volume 6 Volume 7 Volume 8 Volume 9 Volume 10

Economics of the Pulp and Paper Industry Forest Resources and Sustainable Management Forest Products Chemistry Papermaking Chemistry Mechanical Pulping Chemical Pulping Part 1, Fibre Chemistry and Technology Chemical Pulping Part 2, Recovery of Chemicals and Energy Recycled Fibre and Deinking Papermaking Part 1, Stock Preparation and Wet End Papermaking Part 2, Drying Papermaking Part 3, Finishing

Volume 11 Pigment Coating and Surface Sizing of Paper Volume 12 Paper and Paperboard Converting Volume 13 Print Media - Principles, Processes and Quality Volume 14 Process and Maintenance Management Volume 15 Materials, Corrosion Prevention and Maintenance Volume 16 Paper Physics Volume 17 Pulp and Paper Testing Volume 18 Paper and Board Grades Volume 19 Environmental Management and Control Volume 20 Biorefining of Forest Resources

ORDER TODAY!

http://www.tappi.org/PSandT 78


THE GENESIS

OF YOUR

SOLUTIONS.

Meet Solenis. A new name for a team with decades of experience. You used to know us as Ashland Water Technologies. Today we’re Solenis, the world leader in pulp and paper chemistries. With years of process experience, advanced technology and 3,500 employees worldwide, the Solenis team is ready to deliver the solutions you need. Meet your team at solenis.com.

WASTEWATER TREATMENT

PEER-REVIEWED

Biochemical methane potential of kraft bleaching effluent and codigestion with other in-mill streams TEMESGEN MATHEWOS FITAMO, OLLI DAHL, EMMA MASTER, and TORSTEN MEYER

ABSTRACT: A biochemical methane potential assay was conducted to investigate the anaerobic digestibility of bleaching effluent from hardwood kraft pulping and the potential of codigestion with other effluents from an integrated pulp and paper mill. Four in-mill streams were tested individually and in combination: total bleaching effluent, alkaline bleaching effluent, kraft evaporator condensate, and chemithermomechanical pulping effluent. The total bleaching effluent, consisting of the chlorine dioxide bleaching and alkaline bleaching effluents, exhibited the highest potential for organic matter degradation and methane generation. Chemical oxygen demand (COD) removal ranged from 57%–76%, and methane generation was 220–280 mL/g COD contained in the wastewater, depending on the degree of dilution. When codigestion was tested, the composite consisting of total bleaching effluent, chemithermomechanical pulping effluent, and kraft condensate was most efficient in terms of COD removal (51%) and methane generation (200 mL/g COD contained in the wastewater). The total bleaching effluent is the largest contributor to the overall amount of wastewater at this mill; it contains relatively low concentrations of anaerobic inhibitors such as adsorbable organic halogens (36 mg/L), total sulfur (170 mg/L), and resin and fatty acids (3.2 mg/L). Therefore, the total bleaching effluent from hardwood kraft pulping may be considered for full-scale anaerobic wastewater treatment, either as a singular stream or as part of a composite stream including other in-mill effluents. Application: Bleaching plant effluent from hardwood kraft pulping may be suitable for high-rate anaerobic wastewater treatment.

A

naerobic treatment of pulp and paper mill wastewater is expanding. Within the last decade, the number of anaerobic installations in mills worldwide has doubled, and the treatment capacity in terms of chemical oxygen demand (COD) removal has quadrupled [1]. However, only a few types of in-mill streams typically are being treated, including paper mill wastewater and condensate streams from chemical pulping. Bleaching effluents generated during chemical pulping are commonly perceived as being toxic to the anaerobic process and are therefore excluded from anaerobic wastewater treatment. Bleaching in chemical pulping usually involves chlorine dioxide bleaching, and the generated process water can contain elevated concentrations of organochlorine compounds. Concentrations of chlorinated compounds in wastewater are often measured using the lump parameter adsorbable organic halogens (AOX). Concentrations exceeding approximately 100 mg/L AOX are assumed to notably inhibit anaerobic digestion [1]. Reported concentrations of AOX in effluents generated during elemental chlorine free (ECF) and kraft alkaline bleaching activities range from 2.6–200 mg/L [2-5]. Chlorine dioxide bleaching and alkaline bleaching are usually combined in one bleaching sequence; therefore, effluents from both

80


types can contain notable amounts of organochlorine substances. Numerous experiments on a laboratory scale, pilot scale, or demonstration scale have investigated the anaerobic digestibility of various bleaching effluents. Reported COD removal rates vary from 15%–90%, and specific methane yields range from 0–0.40 m3 kg-1 COD removed [1]. In general, bleach plant effluents require dilution before anaerobic treatment, which may involve mixing with effluent from aerobic treatment or codigestion with less toxic streams. The fraction of bleaching effluent contributing to the composite stream may range from 5%–50% [3,6]. Microbial adaptation can further improve the suitability for anaerobic digestion [6]. The effluents investigated in this study were generated at an integrated pulp and paper mill in Finland. The mill produces kraft softwood and kraft hardwood pulp, chemithermomechanical (CTMP) pulp, paper, and board. Besides total bleaching effluent and alkaline bleaching effluent as part of the former, two other streams were tested in this study: CTMP effluent and kraft evaporator condensate. CTMP effluent can contain high concentrations of anaerobic inhibitors such as sulfur compounds and resin and fatty acids. Furthermore, the concentration of suspended solids often exceeds 500 mg/L,

WASTEWATER TREATMENT which has previously been marked as a threshold value for influents of high-rate anaerobic reactors above which process deterioration occurs [7]. Anaerobic treatment of CTMP effluent is possible and depends on the extent to which anaerobic inhibitors are contained in the wastewater. Reported COD removal rates are 45%–66%, and specific methane yields are 0.18–0.31 m3/kg COD removed [8,9]. Low specific methane yields are usually caused by high levels of sulfur compounds in the stream. Sulfate- or sulfur-reducing bacteria are competing with acetogenic bacteria and methanogenic archaea for use of organic material [10]. Kraft evaporator condensate is usually well digestible, and numerous mills treat condensate streams in anaerobic reactors. Condensates contain very low concentrations of suspended solids and often relatively low concentrations of anaerobic inhibitors such as resin and fatty acids. The organic matter consists largely of easily digestible methanol, ethanol, and acetate. Accordingly, the COD removal rates are reported to be 70%–99%, with methane yields of 0.29–0.35 m3/kg COD removed [2,3,11]. In pulp and paper mill wastewater, anaerobic inhibitors are often present at levels causing notable inhibition to the anaerobic process. Various strategies have been identified to address inhibition, including codigestion of various streams. In a full-scale anaerobic contact reactor in a Swedish pulp mill, where caustic extraction liquor (CEL) and sulfite evaporator condensate were codigested at a COD ratio of 1:1 and at an organic loading rate of 3 kg COD/m3 day, COD reduction rates of 65% were achieved [6]. In that study, AOX concentrations were as high as 25 mg/L and peroxide concentrations were ~50 mg/L. Codigestion with CEL may have other benefits as well. Qiu et al. [2] pointed out that codigestion of CEL and kraft evaporator condensate, the former containing high amounts of alkalinity, has the benefit of saving costs for caustic addition to the anaerobic process. Previous studies indicate that effluents generated during Parameter

Measuring Facility

hardwood pulping are less inhibitory to anaerobic digestion than streams from softwood pulping [12] because softwood pulping often contains elevated concentrations of resin acids and tannins [13]. Larsson et al. [14] used CEL from a kraft pulp mill as the sole substrate to feed a laboratory-scale upflow anaerobic sludge blanket reactor. Stable operation conditions were achieved with total organic carbon removal rates of ~43% when softwood was processed and ~60% when hardwood was processed. The objective of this study was to investigate the potential of anaerobically treating bleaching effluent individually and in combination with other in-mill streams. The benefit of fullscale anaerobic wastewater treatment would be to notably reduce the amount of COD that requires aerobic treatment in the activated sludge system and at the same time recover energy in the form of usable methane. Also, the associated reduction in waste-activated sludge would lead to cost savings in the sludge dewatering plant. EXPERIMENTAL A biochemical methane potential (BMP) assay [15] was conducted to investigate the anaerobic digestibility of wastewater from an integrated pulp and paper mill. The assay included testing each effluent individually and in combination. The effluents were initially selected according to their presumed suitability for anaerobic treatment, which requires sufficiently high COD concentrations and relatively low levels of anaerobic inhibitors.

Sampling and analyses of inorganic compounds Samples were taken from all four effluent streams, frozen, and shipped via courier to the University of Toronto. Samples were stored at -18°C until the day of the BMP assay setup. All samples were initially analyzed in terms of pH, total suspended solids (TSS), volatile suspended solids (VSS), COD, total sulfur,

Unit

Total Bleaching Effluent

Alkaline Bleaching Effluent

CTMP Effluent

Kraft Condensate

COD

University of Toronto

mg/L

1300

2500

3400

520

TSS

Stora Enso Pulp Competence Center

mg/ L

41

12

1600

6

Cl−


mg/ L

609

334

10

1.5

AOX

Stora Enso Pulp Competence Center

mg/ L

36

20

0.02

0.01

SO42−


mg/ L

180

150

950

40

Total sulfur


mg/ L

170

33

390

22

pH


3.7

9.7

7.4

9.0

Methanol

Cassen Testing Laboratories, Toronto

300

284

282

309

mg/ L

COD = chemical oxygen demand; TSS = total suspended solids; sulfate.

Cl−

= chlorine ; AOX = adsorbable organic halogens; SO42− =

I. Chemical properties of the effluents. FEBRUARY 2016 | VOL. 15 NO. 2 | TAPPI JOURNAL

81

WASTEWATER TREATMENT Total Bleaching Effluent

Alkaline Bleaching Effluent

CTMP Effluent

Kraft Condensate

Dehydroabietic acid

0.96

2.2

27

0.6

Sum of levopimaric and neoabietic acids

0.2

0.03

6.9

0.2

Sum of abietic, pimaric, and sandaracopimaric acids

0.5

0.9

12.5

0.6

Sum of isopimaric and palustric acids

0.5

0.2

21.7

0.3

Chlorodehydroabietic acid

0.01

0,1

Nd

Nd

Dichlorodehydroabietic acid

0.03

0,1

0.01

Nd

2.2

3.4

68

1.7

Linoleic acid

0.3

1.9

1.3

Nd

Palmitic acid

0.2

2.3

0.3

Nd

Oleic acid

0.2

1.0

0.7

Nd

Stearic acid

0.4

1.9

Nd

Nd

1.0

7.1

2.2

0.0

Compound, mg/L

Sum of Resin Acids

Sum of Long-Chain Fatty Acids

II. Concentrations of resin acids and long-chain fatty acids in each of the effluents (Nd = Not detectable).

sulfate, chloride, AOX, peroxide, and resin and fatty acids (RFA). TSS was determined by filtration using 934-AH glass fiber filters (Whatman; Maidstone, UK) with a 1.5 µm pore diameter. VSS was determined by the weight loss of a sample (dried at 104°C) during high-temperature combustion (525°C). COD analyses were performed by applying the dichromate method [16] and using a DR-3900 spectrophotometer (Hach Company; Loveland, CO, USA). Sulfate and chloride were analyzed with a Dionex ICS-2100 ion chromatography system (Thermo Scientific; Waltham, MA, USA). Total sulfur was measured using inductively coupled plasma-optical emission spectrometry. AOX analyses were carried out according to SFS-EN 1485 “Water quality.” Table I shows the chemical properties of the effluents.

Extraction and analysis of resin acids and long-chain fatty acids The method to extract and analyze the particulate and dissolved fractions of RFA is described in detail elsewhere [17]. Briefly, the aqueous samples were pH adjusted and directly analyzed using an exactive liquid chromatography-mass spectrometry (LC-MS) analyzer. Table II shows concentrations of resin acids and long-chain fatty acids in each of the effluents. The target resin acids were dehydroabietic acid (DHA) and the nonaromatic resin acids levopimaric acid, neoabietic acid, abietic acid, pimaric acid, sandaracopimaric acid, isopimaric acid, and palustric acid. The analyzed fatty acids were oleic acid, palmitic acid, linoleic acid, and stearic acid. RFA were quantified with five external calibration standards. LC-MS analysis was performed using a reversed phase column (100 mm x 2.1 mm inside diameter) with a 1.9-µm particle 82


size. The addition of surrogate compounds, including DHA -6, 6-d2, dichlorodehydroabietic acid, and margaric acid, to the wastewater before extraction shows recovery rates of 69%– 89%, as well as relative standard deviations of 5–13.

Biochemical methane potential A BMP assay was conducted with granular seed sludge taken from a full-scale anaerobic reactor at a sulfite pulp mill in Canada. All BMP assays included a positive control and a negative control. The positive control included glucose, acetate, sodium propionate, and methanol. The negative control contained only the inoculum but no wastewater. The 160-mL batch serum bottles contained a working volume of 98 mL comprising a defined nutrient medium, inoculum, substrate, and water. The nutrient mixture contained a phosphate buffer, trace minerals, a redox indicator, vitamins, and ferrous sulfide. All samples and controls were prepared and measured in triplicates. The batch cultures were incubated at 37°C and 115 rpm in a shaker throughout the assay period. The amount of produced biogas was measured with a 20-mL syringe coupled with pressure transducer. To determine the amount of methane produced solely from the organic material in the wastewater, the amount of methane produced in the negative control was subtracted. The assay lasted 80 days; however, the results presented in Figs. 1–4 are related to the cumulative biogas production and COD removal rates after 19 days of digestion. No significant biogas production was recorded for the period after that. The digestion time in BMP assay batch tests is much larger than what can be expected in continuous full-scale anaerobic digestion. Batstone et al. [18] compared the degradation


1. Cumulative biogas production for each effluent at different dilutions. The x-axis designation refers to the effluent type and the amount of wastewater COD (mg) that was fed to the serum bottles.

2. COD removal rates of the effluents at different dilutions, as calculated from the amount of generated methane. The x-axis designation refers to the effluent type and the amount of wastewater COD fed (mg) to the serum bottles. FEBRUARY 2016 | VOL. 15 NO. 2 | TAPPI JOURNAL

83


3. Cumulative biogas production from anaerobic digestion of the blends. Eighty mg of wastewater COD was fed to each of the serum bottles.

4. COD removal rates (%) for the blended effluents, as calculated from the amount of generated methane.

extent and the hydrolysis rates of thermally hydrolyzed activated sludge in BMP assays with that of two full-scale sludge digesters in a wastewater treatment plant in Australia. Accordingly, in the BMP assay the degradability extent was nearly representative of full-scale digestion, and the hydrolysis rate was one order of magnitude lower than in full-scale digestion. The BMP assay consisted of two parts. In the first part, the digestibility of each effluent was tested individually and at different degrees of dilution. The second part involved codigestion of various combinations of in-mill streams. The 84


volume ratios chosen in the codigestion part (Table III) represent the approximate ratios of effluent amounts generated at the mill. The biogas composition in terms of methane and carbon dioxide was determined using an HP 5890 series gas chromatography analyzer with a thermal conductivity detector (Hewlett Packard; Palo Alto, CA, USA). A methane content of approximately 70% (±1%) was measured in the biogas from all wastewater substrates. Considering the COD of 1 mol methane (2 mol O2), 22.4 L methane at standard temperature and pressure (STP) conditions (0°C, 100 kPa) is equivalent to 64 g O2 or 64 g

WASTEWATER TREATMENT Volume Ratio

Total Bleaching Effluent (TB)

Alkaline Bleaching Effluent (AB)

CTMP Effluent (CTMP)

Kraft Condensate (KC)

KC-CTMP-TB

1:2:6

41.5 (32)

0

35.7 (11)

2.8 (5.4)

KC-CTMP-AB

1:2:3

0

40.7 (17)

36.5 (11)

2.8 (5.4)

KC-AB

1:3

0

74.8 (31)

0

5.2 (10)

CTMP-AB

2:3

--

42 (17)

38 (11)

--

BMP Assay: Codigestion

III. Amount of wastewater COD that was fed to each serum bottle. The numbers in parentheses refer to the percentage of the serum bottle working volume consisting of wastewater. Total Bleaching Effluent (TB)

Alkaline Bleaching Effluent (AB)

CTMP Effluent (CTMP)

Kraft Condensate (KC)

TB20

20 (16)

0

0

0

TB40

40 (31)

0

0

0

TB60

60 (47)

0

0

0

AB30

0

30 (12)

0

0

AB50

0

50 (20)

0

0

AB80

0

80 (33)

0

0

CTMP30

0

0

30 (9)

0

CTMP50

0

0

50 (15)

0

CTMP80

0

0

80 (24)

0

KC10

0

0

0

10 (19)

KC25

0

0

0

25 (49)

BMP Assay: Individual Effluents

IV. Amount of wastewater COD that was fed to each serum bottle. The numbers in parentheses refer to the percentage of the serum bottle working volume consisting of wastewater.

COD. Thus, 1 L methane is equivalent to 2.86 g COD (64/22.4), or 1 g COD is equivalent to 350 mL methane (STP). The anaerobic digestibility was quantified by means of the percentage degradation of the COD contained in the substrate. Considering a methane content of 70% within the produced biogas, the percentage COD degradation was calculated using Eq. (1) and Eq. (2):

(1)

(2)

RESULTS AND DISCUSSION

Biochemical methane potential of individual streams Adverse effects caused by anaerobic inhibitors can be addressed by means of dilution. To test inhibition and the neces-

sary degree of dilution, in the first part of the BMP assay each effluent was digested at various degrees of dilution. The latter was expressed by the fraction of serum bottle working volume consisting of wastewater. Total bleaching effluent, alkaline bleaching effluent, and CTMP effluent were tested at three different dilutions. However, kraft condensate was only tested at two dilutions because of comparably low COD concentrations. Because the in-mill streams contained very different COD concentrations, the amounts of wastewater added on a volume basis varied largely among the individual effluents (Table III and Table IV). Figures 1 and 2 show the total cumulative biogas production for each effluent at various dilutions and the anaerobic digestibility expressed as the COD removal rate, respectively. Of the individual effluents, the total bleaching effluent showed the highest degree of digestibility. The COD removal ranged between 57% (±5%) and 76% (±14%), depending on the degree of dilution (Figs. 1 and 2; Table II). The observed decreasing COD removal along with an increasing wastewater concentration indicate some degree of anaerobic inhibition. This inhibition may have been caused by somewhat elFEBRUARY 2016 | VOL. 15 NO. 2 | TAPPI JOURNAL

85

WASTEWATER TREATMENT evated concentrations of organochlorine compounds originating from chlorine dioxide bleaching. The AOX concentration in the total bleaching effluent was 36 mg/L, whereas that in the alkaline bleaching effluent was only 20 mg/L (Table I). Nevertheless, relatively large parts of the organic material in the total bleaching effluent appeared to be easily degradable. Approximately 35% of the COD consisted of methanol (1 g of methanol is equivalent to 1.5 g COD) (Table I), which is anaerobically well digestible. Figures 1 and 2 refer to a notably lower digestibility of alkaline bleaching effluent, with average COD removal rates below 40%. Because it is part of the total bleaching effluent, alkaline bleaching effluent obviously contains a relatively large fraction of difficult-to-digest organic material. The digestibility of the CTMP effluent was moderate, with COD removal rates between 43% (±8%) and 52% (±12%). This effluent was the only stream that contained anaerobic inhibitors at crucial levels. The concentrations of sulfate and the sum of the resin acids were 950 mg/L and 68 mg/L, respectively. Both can be associated with anaerobic inhibition [1]; however, the results in the present study do not show any sign of inhibition, at least for the applied dilutions. The fractions of the serum bottle working volume consisting of CTMP effluent were 9%, 15%, and 24%. The results for the kraft condensate can be interpreted only to a very limited extent because of the associated large error (Figs. 1 and 2). As previously mentioned, due to the very low COD concentration in this effluent, only small amounts of COD could be added to the serum bottles. Nevertheless, kraft condensate seemed to be relatively well digested (Figs. 1 and 2). The organic material consisted largely of easily digestible methanol (Table I).

Anaerobic digestion of blended effluents (codigestion test) In the second part of the BMP assay, various effluents were combined to test the potential for codigestion. For better comparison, the total amount of wastewater COD that was fed to each serum bottle was kept at 80 mg, whereas the volume fractions of the various effluents reflected the amounts of wastewater generated at the mill (Table III). Of the four tested combinations of effluents, the highest methane generation potential and COD removal rate was achieved by blending kraft condensate, CTMP, and total bleaching effluent (KC-CTMP-TB) (Figs. 3 and 4). This was the only blend where total bleaching effluent was included (Table III). When total bleaching effluent was replaced with alkaline bleaching effluent, as in KC-CTMP-AB, the digestibility was significantly lower. Again, the anaerobic digestibility of total bleaching effluent appeared to be notably higher than that of alkaline bleaching effluent (Figs. 1–4). It is worth mentioning that the amount of removed COD was probably higher than what was calculated from the produced methane. A notable part of the COD was likely consumed and used for chlorate reduction [19,20]. 86


The effluent combination CTMP-alkaline bleaching (CTMP-AB) appeared to be more digestible than the blend kraft condensate-alkaline bleaching (KC-AB) (Figs. 3 and 4), again confirming the results of the first part of the BMP assay. In the first part, CTMP effluent was also easier to digest than alkaline bleaching effluent. The contribution of the kraft condensate to the blend, in terms of COD, was very small (Table III). The CTMP effluent seems to be a suitable candidate for codigestion with other streams because a notable part of the organic material is well digestible. Furthermore, codigestion and associated dilution can considerably diminish the effects of anaerobic inhibitors contained in CTMP effluent. Similar to bleaching effluent, the COD removal was likely somewhat higher than what was calculated from the amount of methane produced. A part of the COD was likely used by sulfate- or sulfur-reducing bacteria [10]. Sulfate-reducing bacteria are thermodynamically advantaged compared with their competitors. For example, anaerobic treatment of hydrosulfite bleaching effluent in a thermomechanical pulp mill resulted in 66% of the COD to be used for the reduction of sulfur compounds [21]. Also, very low specific methane yields (0.1 and 0.25 m3/kg removed COD) were achieved during anaerobic treatment of CTMP-chemimechanical pulp wastewater containing high concentrations of sulfur compounds [21].

Degradation of organic compounds in bleaching effluent This study has shown that the effluent from acid bleaching (chlorine dioxide bleaching) during hardwood pulping at the mill is more suitable for anaerobic digestion than the alkaline bleaching effluent. As previously mentioned, one reason for the better digestibility of the acid bleaching effluent could be the higher percentage of methanol in the wastewater (Table I). The relatively high methanol content in effluent from the chlorination stage can be explained by the large extent to which methoxyl groups in the residual pulp lignin are split off and released into the wastewater [22]. Another reason for the better digestibility of the acid bleaching effluent could be the difference in molecular size of the organic substances contained in acid and alkaline bleaching effluents. Bryant et al. [23] observed that, during ECF bleaching, the alkaline effluent contained a significantly higher proportion of high molecular weight substances than the acidic effluent. Specifically, the molecular weight of lignin was much lower in the acidic effluent than in the alkaline effluent. In another study Sågfors and Stark [24] were investigating the organic material contained in acidic and alkaline filtrates from various hardwood and softwood processing mills. Whereas the organics in alkaline effluent consisted of 65%–75% high molecular weight material, the organics in acid bleaching effluent consisted of only 20% high molecular weight material. In general, lower molecular weight organics can be easier degraded by means of biological treatment [25].

WASTEWATER TREATMENT Along with the transition from chlorine bleaching to chlorine dioxide (ECF) bleaching, the anaerobic digestibility of bleaching plant effluent has likely improved. The use of chlorine dioxide has substantially lowered the generation of chloroform, chloroacetones [26], and other toxic chlorinated compounds [27]. The characteristics of the organic material in bleaching effluent also depend on the raw material being used for pulping [27]. Bleaching effluent from hardwood pulping, as used in this study, contains less anaerobic inhibitors than bleaching effluent from softwood pulping [12]. CONCLUSIONS The total bleaching effluent investigated in this study may be suitable for anaerobic wastewater treatment. However, digestion at higher wastewater concentrations showed signs of anaerobic inhibition. Therefore, bleaching effluent may be codigested with other in-mill streams. The BMP assay provides only a first glimpse of the potential for full-scale anaerobic wastewater treatment. Therefore, to further assess the digestibility of the investigated effluents, a bench-scale or pilot-scale experiment may be implemented. TJ ACKNOWLEDGEMENTS The authors are thankful for the assistance and contributions of Kajsa-Stina Ohlström and Kalle Ekman from Stora Enso and Minqing Ivy Yang, Angie Tse, and Wendy Han Zhou for their assistance in the laboratory.

LITERATURE CITED 1. Meyer, T. and Edwards, E.A., Water Res. 65: 321(2014). 2. Qiu, R., Ferguson, J.F., and Benjamin, M.M., Water Sci. Technol. 20(1): 107(1988). 3. Cornacchio, L.-A.T., “Anaerobic treatability of pulp and paper effluents,” M.A.Sc. thesis, University of Guelph, Guelph, ON, Canada, 1990. 4. Setiawan, Y., Soetopo, R.S., and Kristaufan, J.P., “Anaerobic treatment for bleaching effluent of pulp and paper mill,” Proc. Int. Semin. Chem., Indonesian Chemical Society, 2008, p. 367. Available [Online] http://chemistry.unpad.ac.id/isc-proceeding/2008/Pdf/ OP/0367-0372%20Yusup%20Setiawan.pdf . 5. Chaparro, T.R. and Pires, E.C., Braz. J. Chem. Eng. 28(4): 625(2011). 6. Nilsson, B. and Strand, O., Water Sci. Technol. 29(5–6): 399(1994). 7. Totzke, D., 2012 Anaerobic Treatment Technology Overview, Applied Technologies Inc., Brookfield, WI, USA, 2012. 8. Welander, T. and Andersson, P.-E., Water Sci. Technol. 17(1): 103(1985). 9. Habets, L.H.A. and de Vegt, A.L., Water Sci. Technol. 24(3–4): 331(1991). 10. Eis, B.J., Ferguson, J.F., and Benjamin, M.M., J. - Water Pollut. Control Fed. 55(11): 1355(1983). 11. Dufresne, R., Liard, A., and Blum, M.S., Water Environ. Res. 73(1): 103(2001). 12. Yang, M.I., Edwards, E.A., and Allen, D.G., Water Sci. Technol. 62(10): 2427(2010).

ABOUT THE AUTHORS We have been collaborating with some pulp and paper mills on anaerobic treatment of various inmill streams. Although a few previous studies support our results, most others do not. The most difficult part of our research was to investigate the extent to which bleach plant effluFitamo ent requires dilution prior to anaerobic digestion. We addressed this challenge by adding various amounts of wastewater, in terms of COD content as well as volume, to the serum bottles. We found the most interesting aspect of this study to be the widely variable anaerobic digestibility of bleach plant effluent. Bleaching effluents are commonly considered as being toxic to anaerobic digestion. However, hardwood kraft pulping obviously does not release as much toxic material into the wastewater as compared to softwood pulping. Bleaching effluent often constitutes the bulk of the wastewater generated at a mill. If bleaching effluent can be included into anaerobic treatment, a

Dahl

Master

Meyer

mill may be able to save costs for wastewater treatment. The next step in this research would involve a pilot-scale project at a mill. Fitamo is a doctoral student in the Department of Environmental Engineering, Technical University of Denmark, Kongens Lyngby, Denmark. Dahl is a professor in the Department of Forest Products Technology, Aalto University (Finland), Espoo, Finland. Master is a professor and Meyer is a research associate in the Department of Chemical Engineering and Applied Chemistry, University of Toronto, Toronto, ON, Canada. Email Meyer at [email protected].


87

WASTEWATER TREATMENT 13. Makris, S.P., “Removal of resin and fatty acids from pulp mill wastewater streams,” Ph.D. dissertation, Georgia Institute of Technology, Atlanta, GA, USA, 2003. 14. Larsson, M., Truong, X.-B., Björn, A., et al., Environ. Technol. 36(9–12): 1489(2015). 15. Owen, W.F., Stuckey, D.C., Healy Jr., J.B., et al., Water Res. 13: 485(1979). 16. Jirka, A.M. and Carter, M.J., Anal. Chem. 47(8): 1397(1975).

25. Bullock, C.M., “The influence of the high molecular weight fraction of bleached kraft mill effluent on the biological activity of activated sludge,” M.A.Sc. thesis, University of British Columbia, Vancouver, BC, Canada, 1994. 26. Gergov, M., Priha, M., Talko, E., et al., Tappi J. 71(12): 175(1988). 27. Dahl, O., “Evaporation of acidic effluent from kraft pulp bleaching, reuse of the condensate and further processing of the concentrate,” Ph.D. thesis, Oulu University, Oulu, Finland, 1999.

17. Meyer, T., Yang, M.I., Allen, D.G., Tran, H.N. and Edwards, E.A. Environ. Eng. Sci. (submitted). 18. Batstone, D.J., Tait, S., and Starrenburg, D., Biotechnol. Bioeng. 102(5): 1513(2009). 19. Malmqvist, A. and Welander, T., Appl. Environ. Microbiol. 57: 2229(1991). 20. Yu, P. and Welander, T., Appl. Microbiol. Biotechnol. 40: 806(1994). 21. Schnell, A., Skog, S., and Sabourin, M.J., “Chemical characterization and biotreatability of alkaline-peroxide mechanical pulping effluents,” Environ. Conf. Exhib., TAPPI PRESS, Atlanta, GA, USA, 1993, p. 187. 22. Pfister , K. and Sjöström, E., Pap. Puu 61(4): 220(1979). 23. Bryant, P., Sundström, G., and Jour, P., “Ultrafiltration of alkaline filtrates to maximize partial closure of ECF bleach plants,” Int. Pulp Bleaching Conf., TAPPI PRESS, Atlanta, 1998, p. 229. 24. Sågfors, P.E. and Starck, B. in Pulp Bleaching —Principles and Practice, TAPPI PRESS, Atlanta, 1996, p. 749.

Don’t miss the only global conference DISCOVER the latest applications and technical advances: focused on the use and production of Materials – flexible and tunable materials, liquid renewable and sustainable nanomaterials. • Smart crystalline behavior, responsive photonics. • Industrial Applications – coatings and barriers for paper, paperboard and food packaging, energy storage, electronic devices, flexible films, and concrete. • Medical Applications – biomedical scaffolding, drug delivery, tissue and wound treatment. • Composites – aerogels, hydrogels, foams, and more. • Large and lab scale production of cellulose nanomaterials – processing, characterization.

NETWORK

2016 International Conference on Nanotechnology for Renewable Materials 13 – 16 June 2016 • Grenoble, France Keynote Speakers Tuesday, 14 June

Bernard de Galembert Innovation and Bioeconomy Director CEPI

Wednesday, 15 June

Christoph Weder

Executive Director and Professor, Polymer Chemistry and Materials Adolphe Merkle Institute

with Researchers, Scientists and Industry Professionals from around the world.

EXPERIENCE

• Over 150 Technical Programs • Mentoring Program • Fundamentals Workshop • Tec21 & Institut Carnot PolyNat Tours

Conference Co-Chairs Alain Dufresne, Chairman

Professor Grenoble Institute of Technology, Grenoble, France

Julien Bras, Co-Chairman

Professor Grenoble Institute of Technology, Grenoble, France

Johan Foster, Co-Chairman

Associate Professor, Virginia Polytechnic Institute and State University

FOR MORE INFORMATION VISIT: WWW.TAPPI.ORG/16NANO 88


NEW! Application

Product Sample Showcase

The whole is more than the sum of its parts Get the best out of your stock preparation

As a full-line supplier, ANDRITZ

keep the entire production process

ing, or dewatering – we have the right

has

your

in view at all times and not just indi-

solution for every task, and we also

process.

vidual parts. This gives our customers

have the process experts. Get the

stock

With

a

clear

focus

preparation

our

on

process-optimized

perfectly coordinated components to

best out of your stock preparation –

engineered wear parts, your system is

meet their individual goals. Whether

not off the shelf, but custom-tailored.

always state-of-the-art. However, we

OTKOHMF QDjMHMF BKD@MHMF RBQDDM-

This is our understanding of service.

www.andritz.com

We accept the challenge!

PEER-REVIEWED

PRINTING

Causes of back-trap mottle in lithographic offset prints on coated papers GUNNAR ENGSTRÖM

ABSTRACT: Back-trap mottle is a common and serious print quality problem in lithographic offset printing of coated papers. It is caused by nonuniform ink retransfer from an already printed surface when it passes through a subsequent printing nip with the print in contact with the rubber blanket in that nip. A nonuniform surface porosity gives rise to mottle. A key parameter in mottling contexts is the coating mass distribution, which must be uniform. Good relationships between mottle and mass distribution have also been reported; the mottle pattern coincides with that of the coating mass distribution. High blade pressures, compressible base papers, and high water pick-up between application and metering, which plasticizes the paper, yield uniform mass distributions, but these parameters might have a detrimental effect on the runnability in blade coating in terms of web breaks. The general opinion has been that nonuniform surface porosity is caused by binder migration and enrichment of binder in the coating surface, more in the high coat weight areas and less in the low coat weight areas. Recent research has suggested that a more probable mechanism is depletion of binder in the coating surface. Nonuniform shrinkage of the pigment matrix (filter cake) formed during the consolidation between the first critical concentration (FCC) and the second critical concentration (SCC) is another possible mechanism. Relevant relaxation times for latex and the time scales for consolidation show that the high coat weight areas shrink more than the low coat weight areas in the coating layer. A recent pilot-scale experiment has shown that the drying strategy did not affect the differences in shrinkage between high and low coat weight areas. The drying strategy has a pronounced impact on mottle. A high evaporation rate at the beginning of the evaporation results in less mottle than a low evaporation rate. The least mottle is obtained if the drying is performed with a gap in the course of evaporation between the FCC and the SCC. Application: Knowledge of the relationship between 1) the processes of forming and drying of the coating layer in blade coating and 2) back-trap mottle of the printed paper is a prerequisite to be able to improve the print quality and avoid complaints.

M

ottle, defined as an undesirable variation in print density in the printed image, is a common and serious print quality problem in lithographic offset prints on coated papers. The two types of mottle are 1) back-trap mottle and 2) wet repellence mottle. Back-trap mottle is caused by a random nonuniform retransfer of ink from an already printed sheet when it passes through a subsequent printing nip with the print in contact with the rubber blanket in that nip. Wet repellence mottle is caused by fountain solution applied to the paper in one of the first printing units not being absorbed before the paper reaches the next printing unit. This can lead to nonuniform ink transfer and a mottled print. This paper focuses on back-trap mottle. The term “mottle” will pertain only to back-trap mottle. Statistics from Europe in 2003 state that 42% of the complaints over print quality on coated woodfree paper were in response to back-trap mottle [1]. Despite much research over the years, the cause of mottle is not fully understood. A generally recognized hypothesis is that mottle results from nonuniform surface porosity. This hypothesis was recently verified by Chinga and Helle [2] and Preston et al. [3], who report excellent relationships between

variations in surface porosity and print mottle in industrial papers. Similar results have been reported in a study by Xiang et al. [4] in which the local ink tack was measured using a micro-tack gauge (diameter 0.5 mm). A good relationship between the mottle and the variation in tack, ascribed to variations in surface porosity, was reported. Dappen [5] coated glass plates in a classical experiment and dried these in an oven (high evaporation rate) and under ambient conditions (low evaporation rate). The binder (starch) gradient in the thickness direction was measured. The result showed that starch migrated upward toward the coating surface and was enriched there. The enrichment was significantly higher with the higher evaporation rate. An interpretation of Dappen’s experiment, which emerged with time, was that binder migration seals the coating surface nonuniformly, and that to obtain an open coating surface with a minimum of porosity variations, binder migration must be avoided. The recommendation, which with time became generally accepted, was that the coating must be dried at a low evaporation rate to avoid print mottle. Ahrheilger [6] later verified this interpretation in the same type of experiment. Experiments on a pilot scale in which the coating was apFEBRUARY 2016 | VOL. 15 NO. 2 | TAPPI JOURNAL

91

PRINTING plied on real paper, which is an absorbent substrate, and dried at evaporation rates identical with those in industrial practice, and in which the coated material was printed on an industrial scale have, however, shown that this recommendation was incorrect [7,8]. These researchers showed that, to reduce the risk for mottle, the evaporation rate must be high. Other studies of drying strategies on a pilot scale have identified a critical phase in the course of evaporation during which the evaporation rate must be low or zero [9-11]. In practice, this strategy is achieved with a gap in the course of evaporation, and by drying in this way, the risk for mottle can be further reduced. The contradiction between the model experiments on a bench-scale and those performed under industrial conditions will be discussed later in the sections on binder migration and binder distribution. For a given type of base paper and coating color, there is usually a good relationship between the mottle and the binder distribution in the coating surface [7,12], but the relationship is not unique [8]. Some recent papers have reported that the mottle does not rate in the same way as the binder distribution [13,14]. The rating was, however, the same as that of the surface porosity nonuniformity. This shows that causes other than binder migration can lead to nonuniform surface porosity. Researchers suggested that different conditions for the formation of the latex film [13] and the shrinkage of the coating layer during consolidation [15,16] in the low and high coat weight areas were the cause. A common feature of these causes is that they are related to the coating mass distribution. We therefore expect that there is a relationship between the mottle and the coating mass distribution. Good relationships have also been reported in other studies [17,18], but as in the case of the binder distribution, the relationships are not unique [8]. A perfect matching between the mottle pattern and the pattern in the coating mass distribution has been reported for a lightweight coated (LWC) paper [19]. Both these observations make the coating mass distribution a key parameter in a mottling context. A uniform mass distribution and smooth surface are desired, and they are a guarantee for a mottle-free print. The coating technique, the conditions during the forming of the coating layer, the base paper, and the coating color all affect the coating mass distribution. In this paper, the research on the forming of the coating layer in blade coating and on print mottle, in which the author was involved during the 1980s and 1990s, is compiled and discussed in the light of findings of a later date. The focus of this paper is on print mottle, which has received much attention in coating research. The papers discussed here are those based on actual data, not far-fetched speculations. The aim is to provide a brief overview of the causes of mottle and of strategies for avoiding mottle. Areas discussed are 1) the formation of the coating layer in blade coating and its effect on the coating mass distribution, 2) binder migration, 3) binder distribution in the coating surface, 4) drying profiles, and 5) shrinkage and the porosity distribution. Added emphasis is given to the 92


1. Schematic illustration of the forming of the coating layer beneath the blade tip.

key role that coating mass distribution plays for the mottle, which is discussed very little in the literature. The effect of calendering will not be discussed in this paper. Readers interested in its role are referred to a previous review paper [20]. FORMATION OF THE COATING LAYER In blade coating, the coating layer is formed beneath the blade tip. Its thickness profile is given by its upper and lower boundary surfaces, respectively, by the blade tip and the base paper (Fig. 1) [21]. Because the blade tip is straight and the base paper is rough, the thickness profile is given by the surface profile of the base paper. But the base paper is subjected to a compressive load beneath the blade tip, caused by the blade pressure to control the coat weight, so the surface profile referred to here is the surface profile in the compressed state. (For the sake of simplicity, the surface profile in Fig. 1 is depicted as sinusoidal.) The surface profile of the base paper in a compressed state thus determines the thickness profile of the wet coating layer. If we assume that the composition of the coating color entering the slit between the blade tip and the base paper is homogenous, which is reasonable, the thickness profile will be identical to the coating mass profile or the coating mass distribution. On a millimeter-length scale, we can also neglect lateral flow in the wet coating layer during the consolidation. This means that the dry coating mass distribution is given directly by the surface profile of the base paper in a compressed state beneath the blade tip on the same length scale. It should be pointed out that the length scale of the mottle that is most disturbing for the human eye is also on the millimeter level [22]. The surface profile of the base paper is dependent not only on the compressive load (blade pressure), but also on its compressibility. Figure 2 shows the effect of blade pressure on coating mass distribution, estimated using a burn-out test. The material referred to in the figure is a base paper for LWC that was coated on a pilot scale with a kaolin clay-based coating color in which the latex content and the solids content were varied at two levels [23]. As is evident in Fig. 2, a high blade pressure yields a uniform mass distribution. High blade pressures were obtained using a coating color with a high solids

PRINTING

2. Standard deviation in coat weight (2-4 mm wavelength band) (mass distribution) as a function of blade pressure.

3. Standard deviation in coat weight (2-4 mm wavelength band) (mass distribution) on a wood-containing (lightweight coated [LWC]) and a woodfree base paper coated under identical conditions with the same kaolin clay-based coating color.

content or a low latex content. The blade pressure is governed by the high shear rate viscosity of the coating color; a high viscosity requires a high blade pressure, and vice versa [24,25]. In the coating of low-grammage substrates, such as base papers for LWC, a high blade pressure is, however, detrimental for runnability because such papers are weak, and the risk for web breaks increases. A low blade pressure is therefore desired for coating LWC papers. The blade pressure thus cannot be increased in the coating of LWC papers to achieve a uniform mass distribution. In the coating of cartonboard, where web breaks are not critical, the situation is different. Here the coating colors are often formulated so that the blade pressure needed for the target coat weight is as high as possible. The advantages of a high blade pressure can thus only be realized in cartonboard coating. Mass distribution becomes, as expected, more uniform on a compressible base paper such as a wood-containing paper (LWC) than on a woodfree base paper (Fig. 3) [8]. Both types of base papers were coated on a pilot scale under identical conditions using the same coating color. The mass distribution

4. Standard deviation in coating thickness (2-4 mm wavelength band) as a function of dewatering of the coating color. The coat weight and the blade pressure were the same in all data points.

on the final coated materials was measured using soft X-ray radiography. The compressive stiffness of the base paper is governed by the stiffness of the fibers and the porosity (density) of the sheet [26]. In this case, the compressibility was controlled by the density, which was lower for the wood-containing base paper. Compressive stiffness is also governed by the aqueous phase (water) uptake from the coating color between the coating color application and the metering with the blade. This uptake plasticizes the base paper and makes it compressible. This was studied in a pilot coating trial in which a number of ground calcium carbonate (GCC)-based coating colors were prepared to the same high shear rate viscosity, and in which the type and amount of starch were the experimental parameters [27]. All colors needed the same blade pressure (i.e., the compressive load on the base paper was identical when forming the coating layers). The dewatering of the coating colors was measured using the Åbo Academy gravimetric water-retention tester (ÅAGW), and the coating thickness distribution in the final coated material was measured using a burn-out test. Figure 4 shows the plot for the standard deviation in coating thickness versus the water dewatering, and that a high ÅAGW-dewatering gave a uniform coating thickness. This shows that high water uptake before the blade plasticizes the base paper and makes it compressible, which leads to a uniform mass distribution. Pressure dewatering in roll application has also been reported to plasticize the base paper and promote coating uniformity [28]. Water uptake before the blade also reduces the strength of the base paper and, if the grammage is low, this can lead to runnability problems in the form of web breaks. High water retention of the coating color is therefore important, particularly in the coating of base papers of low grammage. The positive effect of a low water uptake before the blade on the web break frequency is well illustrated by the short dwell time application (SDTA) coater, for which 30% fewer web breaks are reported than in conventional blade coaters with roll application of the coating color [29]. FEBRUARY 2016 | VOL. 15 NO. 2 | TAPPI JOURNAL

93

PRINTING

5. Standard deviation in print reflectance (2-4 mm wavelength band) (print mottle) as a function of standard deviation in coat weight (2-4 mm wavelength band) (mass distribution) for an LWC paper coated with kaolin clay-based coating colors with different solids and latex contents.

In cartonboard coating, however, a high water uptake before the blade can be beneficial. The Metso long dwell-time coater illustrates this. Using this coater, a very good coating mass distribution can be attained [30]. Excessive dewatering before the blade in cartonboard coating can, however, cause local immobilization and the formation of aggregates in the coating layer, which can get stuck beneath the blade tip and cause blade streaks or scratches. These are considered to be a serious runnability problem and quality defect. Hamada and Kohno [31] were the first authors to report a relationship between back-trap mottle and mass distribution. Later, this relationship was confirmed by several researchers, including Engström [17] (Fig. 5) and Kolseth [18]. In those later studies, the mass distribution was measured using a burnout test. A study in which a base paper for LWC was coated on a pilot scale and the coated material was printed on an industrial scale showed a very good relationship between the standard deviation in coat weight (mass distribution, burn-out test) and the standard deviation in print reflectance (back-trap mottle). Within the whole investigated wavelength range of 0.125-8.0 mm, the pattern in the mass distribution coincided in all details with that in the mottle (Fig. 6) [19]. The standard deviations in print reflectance in the offset print (mottle) and in reflectance in the burn-out test were measured using the same image analysis procedure.

Binder migration Dappen [5] studied the migration of starch in a classical experiment in which glass plates were coated with a kaolin clay/ starch coating color and dried at ambient conditions (low evaporation rate) and in an oven at 105°C (high evaporation rate). In this case, the coat weight on the glass plates was extremely high. The dry coating layer was then gradually scraped off and the scraped off material was collected and analyzed with respect to binder (starch). The starch content 94


6. Standard deviation in coat weight (mass distribution) and standard deviation in reflectance (print mottle) as functions of standard deviation in coat weight (mass distribution) within the wavelength range 0.125-8.0 mm for an LWC paper.

was plotted against the depth in the coating from which the scraped off material originated. From the plots showing the starch gradient in the thickness direction, it was evident that starch had migrated upward toward the surface during drying and that the gradient was significantly steeper and the starch content in the coating surface significantly higher when the coating had been dried in an oven at a high evaporation rate. Starch and latex were compared with each other with respect to migration by Groves et al. [15], who prepared drawdowns of a GCC/starch and a GCC/latex coating color on flat polyester substrates and dried these, as in Dappen’s experiment, under ambient conditions and in an oven at 105°C. The starch and latex gradients in the thickness direction were determined by a grinding procedure and analysis of the removed material. The thickness of the coating layers was ~160 µm. In agreement with Dappen’s results, a steeper gradient and a greater enrichment of starch in the surface was found for the oven-dried material. With latex, however, no gradient whatsoever could be detected in either the material dried under ambient conditions or in the material dried in the oven. Starch and latex thus behave differently during drying. A difference between starch and latex that could explain their different behaviors is that starch is present as individual molecules whereas latex is a particulate material. The kinetics (evaporation rate) in the model experiments described previously are very different from the conditions in the industrial production of coated paper, so a relevant question is whether the results of Dappen [5] and Groves et al. [15] also hold for industrially coated papers. Three studies, which deal with latex, have been published on binder migration in industrially produced papers. Whalen-Shaw and Eby [32] prepared cross sections of the bright and dark areas of a mottled print on a coated paper and analyzed these using scanning electron microscopy (SEM) with the detection of back-scattered electrons. From the captured images, the elemental composition in the thickness direction was assessed visually. The researchers found that latex was enriched downwards in the

PRINTING

7. Standard deviation in print reflectance (2-4 mm wavelength band) (print mottle) as a function of the mean styrene-butadiene (SB) latex content in the coating surface.

coating toward the base paper surface. Kenttä et al. [14] presented a similar study on cross sections of a single-coated LWC paper (not printed) using the same technique, but they used image analysis for the assessment of the captured images. From these a latex gradient was constructed that showed that the coating surface was weakly depleted of latex and that latex had migrated downward toward the interface between the coating layer and the base paper, confirming the observations of Whalen-Shaw and Eby [32]. In the third paper, Vyörykkä and Vourinen [33] presented a study of a double-coated woodfree paper using confocal Raman microscopy to determine the gradient. They observed a sharp boundary between the coating layers with respect to latex content, but no gradient. The latex content was the same throughout each coating layer. A couple of observations from pilot coating experiments show that starch and latex respond differently to the drying conditions. Engström et al. [7] and Yamazaki et al. [34] both investigated the effect on the binder content in the coating surface of time delay between blade and dryer. Yamazaki et al. [34] compared starch with latex in an experiment in which a wood-containing base paper was coated and the binder content in the coating surface was measured using electron spectroscopy for chemical analysis (ESCA). The results showed that the amount of starch in the surface increased when the time delay was reduced (i.e., when the amount of water evaporated from the coating layer increased and the amount absorbed into the base paper was reduced). The latex content in the coating surface remained unaffected by the time delay. Engström et al. [7] studied starch in an experiment in which a base paper for LWC was coated with a kaolin clay-based coating color with starch as binder. They also found that the amount of starch in the surface increased when the time delay was reduced. They reported mottle values for the printed material and showed that the mottle was reduced when the time delay was reduced. In another experiment by Engström et al. [8], a base paper for LWC and a woodfree base paper were coated on a pilot

scale with two kaolin clay-based coating colors, one with starch as the main binder and the other with styrene-butadiene (SB) latex. The experimental parameter in both experiments was the drying profile. The coated material was offsetprinted on an industrial scale and the mottle was assessed using image analysis. In this experiment, the SB latex content in the coating surface was measured using a technique in which the surface was scanned using ultraviolet radiation [35]. Figure 7 shows the results, where the standard deviation in print density (mottle) is plotted against the mean latex content in the coating surface. It is evident that the binder content in the coating surface varied over a wide range, depending on drying strategy, showing that SB latex had migrated during drying. It is also evident that the mottle in the LWC paper was greater than in the woodfree paper, even though the SB latex content in the coating surface was higher in that paper. Both papers also showed a tendency for the mottle to decrease with increasing SB latex content. These observations suggest that the mottle was the result of SB latex migrating away from the coating surface. Migration away from the coating surface and depletion of SB latex must therefore be regarded as a potential cause of nonuniform surface sealing and surface porosity. Figure 7 thus shows that one should question whether binder enrichment in the coating surface causes mottle. The belief that it is indeed the cause is based on bench-scale experiments and stain tests. Such tests must be interpreted with care and humility. Starch and latex thus behave differently during consolidation by dewatering and evaporation. That latex does not migrate, which was shown by the model work of Groves et al. [15] on a non-absorbent polyester substrate, does not hold for industrial papers. Strong results reported by Engström et al. [8], Whalen-Shaw and Eby [32], and Kenttä et al. [14] clearly show that latex migrates in such papers, not to the coating surface but toward the base paper.

Binder distribution Binder redistribution during consolidation is not a problem, as such. It is only when it occurs nonuniformly that it leads to problems with mottle. Because of natural imperfections in the base paper caused by formation, which makes it rough, the coating layer will, as already discussed, exhibit a certain mass distribution. As a consequence, the binder distribution in the coating surface will also be nonuniform. Ahrheilger [6] illustrated the role of coat weight variations in a model experiment in which smooth and etched glass plates were coated and then dried under ambient conditions and in an oven at 105°C. The dry coatings were stained with colored oil. Two observations were made: 1) the stains on the oven-dried coatings on the smooth and the etched glass plates were uniform and brighter than those on coatings dried under ambient conditions and 2) the stain was mottled on the coating on the etched glass plate dried in the oven but uniform on that dried at room temperature. The interpretation was that binder migrates to the FEBRUARY 2016 | VOL. 15 NO. 2 | TAPPI JOURNAL

95

PRINTING

8. Standard deviation in SB latex content (binder distribution) in the coating surface, a) parts and b) relative units as a function of standard deviation in coat weight (mass distribution) for a) different binder contents and b) different base papers.

surface at a high evaporation rate (oven dried) and seals it, and if the coat weight varies (etched glass plates), the surface is sealed nonuniformly. The hypothesis that nonuniform binder distribution is a factor governing print mottle was verified by Arai et al. [12]. Whether or not binder migrates to the surface during consolidation under industrial conditions can be debated. More binder is present in the coating surface in the high coat weight areas than in the low coat weight areas. Whalen-Shaw and Eby [32] showed this by using an SEM to study cross sections of printed papers exhibiting mottle. The bright areas in the print coincided with areas with high surface concentration of binder, and these areas corresponded in turn to the high coat weight areas. The opposite was also the case, in that the dark areas coincided with a low surface concentration of binder and a low coat weight. That the binder content in the coating surface is higher at high coat weights has also been shown by Engström et al. [7] and Kenttä et al. [14]. Figure 8a demonstrates the relationship between the binder distribution in the coating surface and the mass distribution. The figure comes from a pilot coating trial in which the base paper for LWC was coated using a blade and a metered size press with a kaolin clay-based coating color with carboxymethyl cellulose (CMC)/latex as binder at two solids contents [23]. The mass distribution was assessed using a burn-out test. There are two relationships, one for 10 parts of latex and another for 15 parts. Both relationships go in the expected direction: the binder distribution becomes more uniform when the mass distribution is more uniform. But this relationship is not unique, not even for a given coating color. This is shown in Fig. 8b, which is based on data from the pilot coating experiment described previously [8]. The mass distribution here was measured using soft X-ray radiography. The coating layer was formed under identical conditions so that the standard deviation in coat weight should be the same in each base paper. The standard deviation values are based on a differential measurement (coated paper – base paper), which is a difference between two values of the same magni96


9. Standard deviation in SB latex content in the coating surface as a function of the mean SB latex content.

10. Standard deviation in print reflectance (print mottle) as a function of a) standard deviation in SB latex content and b) standard deviation in starch content in the coating surface.

tude. This, and the fact that a mean value for all the base papers was used, makes the values somewhat uncertain. This explains the scatter in Fig. 8. Although the mass distribution was significantly more uniform in the LWC paper, the variation in SB latex content in the coating surface was not affected; it varied in the same range for both papers. In Fig. 9, the variation in SB latex content in the coating surface is plotted against its mean value. We see that the variation increases when the SB latex content in the coating surface is reduced, indicating that the variations result from depletion of latex in the coating surface. Figure 10 shows the relationship between print mottle and binder distribution in the papers produced [8] for CMC/ latex (Fig. 10a) and for starch (Fig. 10b). CMC/latex and starch are not directly comparable with each other because the binder distribution was measured in different ways. In Fig. 10a, the relationship is product-specific; there is one relationship for the LWC paper and another for the woodfree paper. So, the binder distribution does not solely govern the mottle, but for a given system of base paper and coating color it is a good predictor for mottle. A comparison between Figs. 10a and 10b

PRINTING shows that latex and starch behave in a similar manner on the LWC paper, whereas the mottle was significantly higher with starch than with latex on the woodfree paper. Figure 10b, however, only has three data points, two for the LWC paper and one for the woodfree paper, which restricts the reliability. The fact that there is no unique relationship between the binder distribution in the coating surface (shown only for latex) and the mottle indicates that there are other causes of mottle. Xiang et al. [13] have suggested that variations in surface porosity leading to mottle can result from different filmforming mechanisms for latex in the low (sintering) and high (coalescence) coat weight areas. Different degrees of shrinkage in the high and low coat weight areas have been proposed by Groves et al. [15] and by Rajala and Koskinen [16]. This mechanism explains the effect of drying strategy on mottle and is discussed in the section on shrinkage. It also ought to be mentioned that two fairly recent studies link mottle directly to variations in surface porosity [2,3]. The two research groups studied industrially produced and printed coated paper. Chinga and Helle [2] used SEM and secondary electron imaging of back-scattered electrons to characterize the surface porosity and its variation (standard deviation). The size of the closed areas considered was >8000 µm2. Preston et al. [3] used imaging reflectometry in which the refrac-

11. Standard deviation in reflectance (print mottle) as a function of the evaporation rate for coating colors with carboxymethyl cellulose (CMC)/latex and starch as binders on a) a woodfree base paper and b) a wood-containing base paper.

tive index (RI) is mapped. The standard deviation in RI is a measure of variations in surface porosity. Here, the length scale of the standard deviation reported agreed with that in print mottle (1-8 mm) [22]. Both research groups reported a good relationship between the variations in surface porosity and print mottle, but neither of them mentions or discusses any cause of the variations in surface porosity. DRYING PROFILES As already indicated, the drying profile greatly affects mottle. Earlier, the general opinion was that the drying should begin at a low evaporation rate up to the gel point (probably identical with first critical concentration [FCC]). After the gel point, the evaporation could proceed at a higher rate. Much research during recent years has focused on drying profiles. Later studies based on pilot coating trials and printing of the coated material on an industrial scale have shown that the earlier opinion was wrong, and the opposite is true. For the best result with little mottle, drying should be performed at a high evaporation rate from the beginning, until the coating layer is completely dry. This is true for CMC/latex-based and starch-based binder systems (Fig. 11) [8]. An even better result, with even less mottle, is achieved if the drying is performed with a gap or a conditioning phase during the evaporation (Fig. 12). Aschan was the first person to identify the advantage of such a drying profile in a study on a pilot scale of drying profiles for coated board and their effect on mottle [9]. Engström et al. [7] verified this with a gap during the evaporation in a similar study on a base paper for LWC. Engström et al. [7] also measured the binder content (starch) in the coating surface using ESCA. They observed that the paper dried with a gap during the evaporation had a higher binder content in the surface than the papers dried without a gap. This also supports the notion that binder depletion rather than binder enrichment can cause mottle. The critical phase in the course of evaporation was postulated by Norrdahl [10] to coincide with the evaporation between the FCC and the second critical concentration (SCC). Support for this hypothesis was put forward in a paper reporting results from a pilot coating trial on LWC in which drying profiles were modeled and then tested experimentally. Extensive studies of the critical phase in the course of evaporation

12. Schematic illustration of a drying profile with gap or conditioning phase in the evaporation process. Evaporation in hoods Nos. 1, 3, and 4. Hood No. 2 shut off. FEBRUARY 2016 | VOL. 15 NO. 2 | TAPPI JOURNAL

97

PRINTING have later been performed by Kim et al. [11] and by Rajala et al. [36], who verified the hypothesis put forward by Norrdahl [10]. They also confirmed that drying with a gap during evaporation yields the least mottle. Infrared (IR) drying has been reported to have advantages and to yield less mottle than hot air drying. Support for this assertion is weak or nonexistent. No systematic studies on the subject have been reported, but a rather comprehensive comparison between electric IR and hot air was carried out within a confidential research project in which the author was involved and which now, some 30 years later, is open. This comparison was performed with the same dryer layout, and no difference was found between the effects of the two dryer systems. The mottle reacted exactly in the same way with respect to evaporation rate and distance between the blade and the dryer (delay time). The IR dryer is normally installed just in front of the first hot air hood. This reduces the distance between the blade (time delay), which has been reported to reduce mottle for coatings with starch as binder [7], but to begin with IR drying is not always beneficial. This is evident in the paper by Engström et al. [8], where a number of drying profiles, with and without IR, were evaluated. In most cases, IR drying led to more mottle than drying solely with hot air. This finding is in agreement with the experiences Metso has shared with the author. The inconsistency between the early bench-scale experiment using a stain test and those by Engström et al. [8] in which pilot-coated papers were printed on an industrial scale is because, in stain tests, the ink saturates the coating and in many cases also bleeds through the coating layer into the base paper beneath. This is a poor simulation of the situation where the ink penetrates only the outermost regions of the coating layer. Figure 13, from a pilot coating trial in which a base paper for LWC was coated with a kaolin clay-based coating color with starch as binder [17], illustrates this. The experimental parameter was the drying strategy: 1) hot air or 2) hot air combined with IR. The figure shows the relationship between mottle in the print and mottle in the stain test. Clearly, the stain test gave completely wrong information about the risk for mottle in the print. A stain test must be interpreted with great care. (See also the section on binder migration.)

Shrinkage After the forming stage, the coating layer is consolidated through water release to the base paper and through evaporation in the dryer. During this process, the pore structure in the coating layer is developed. We have already discussed the consequences of the drying on the binder redistribution. Now, we shall discuss the effects of coating layer shrinkage during consolidation on the coating structure (porosity) and structure variations. Watanabe and Lepoutre [37] studied the course of consolidation with the help of model experiments on drawdowns on polyester film. They defined two critical concentra98


13. Mottle in full-scale print as a function of mottle in Crodastain.

tions during this process: the FCC and the SCC. At FCC, a pigment matrix is formed. Between FCC and SCC, the pigment matrix is emptied of water and it simultaneously shrinks. At SCC, the final structure is set. Knowing FCC and SCC, the shrinkage can be estimated. The driving force for the shrinkage is the capillary forces developed in the matrix. Coatings based on fine pigments shrink more than those based on coarser pigments because of the finer pore structure. The binder system opposes the shrinkage; coatings with soft latex shrink more than those with stiff latex [38]. Laudone et al. [39] have reported that coatings with starch as binder shrink more than those with latex. Groves et al. [40] have assumed that the matrix is viscoelastic and that its deformation during consolidation is therefore rate dependent. This means that the consolidation rate is faster in the low coat weight areas than in the high coat weight areas. Based on this and on data of typical relaxation times for latex and a time scale relevant for the consolidation under industrial conditions, they estimated the deformation (shrinkage) as a function of the deformation rate. Their results showed that it is reasonable to assume that, under industrial conditions, the high coat weight areas shrink more than those with low coat weight and that this leads to a nonuniform porosity in the coating layer, with the high coat weight areas being less porous than the low coat weight areas. Groves et al. [15] and Rajala and Koskinen [16] have suggested that porosity variations resulting from nonuniform shrinkage can be a cause of mottle. The approach of Groves et al. [40] fits perfectly with the observations of the influence of the drying strategy, high and low evaporation rates and a gap during the evaporation, on the mottle. At a high evaporation rate and with a gap during evaporation, the compression modulus will not be affected by the deformation (shrinkage) rate and will be the same in the high and low coat weight areas (Fig. 14). This will lead to a uniform porosity in the coating layer. When the evaporation rate is low, however, the compression modulus will be in a range where it declines. Here, the compression modulus will be higher in

PRINTING

14. Compression modulus as a function of deformation rate (schematic graph).

the low coat weight areas (higher deformation rate) than in the high coat weight areas (Fig. 14), so that the high coat weight areas will shrink more than the low coat weight areas and the coating layer will exhibit a nonuniform porosity. Direct experimental support for this idea of nonuniform shrinkage was lacking until recently. To test its relevance, Ragnarsson et al. [41] performed a trial in which a base paper for LWC was blade coated on a pilot scale with GCC-based coating colors in which the binder system consisted of either CMC/latex or starch/latex. The aim was to study whether these two binder systems responded differently to the drying strategy in terms of porosity nonuniformity in the coating layer and mottle in the offset print. Drying strategies, including high and low evaporation rates and a gap in the course of evaporation, were studied. The offset printing was performed on an industrial scale and the porosity variations were assessed using a burn-out test on uncalendered material. The standard deviations in print reflectance (mottle) and in reflectance in the burn-out test were measured using the same image analysis procedure. In the burn-out test, the variation in reflectance (standard deviation), σ (R), is, according to the Kubelka-Munk theory, linked to the standard deviation in the product light scattering coefficient x coat weight, σ (s*w). The standard deviation in coat weight (coating mass distribution), σ (w), can be assumed to be constant and independent of the drying strategy, for coating layers formed under identical conditions (speed, blade pressure, coating color, etc.) so that if σ (R) changes when the drying strategy is changed, this change must be the result of a change in the standard deviation in the light scattering coefficient, σ (s), in the coating layer. This standard deviation is linked to the porosity variations in the coating layer [42]. However, assessment of porosity variations using this approach gave no support for the existence of drying-induced porosity variations in the coating layer that had been suggested by Groves et al. [40]. The investigated evaporation strategies did not affect the porosity variations in the coating layer. This does not, however, mean that there were no porosity

variations in the coating layer, but only that they were not affected by the drying strategies studied. The mottle in the print was as expected: the least mottle with a high evaporation rate. This was true for the CMC/latex and for the starch/latex coating colors. Dahlström and Uesaka [43] have measured porosity variations in material (LWC paper) coated on a Helicoater using argon ion beam milling and field emission scanning electron microscopy. They tested uncalendered and calendered material. Only the calendered material exhibited porosity variations. It should be pointed out that the process of consolidation is a significantly slower process in the Helicoater than in the pilot coater and the results of Dahlström and Uesaka [43] must therefore be interpreted with some care. Ragnarsson et al. [41] used the same approach, based on the burn-out test, to estimate porosity variations in the coating layer caused by calendering. Their measurements showed that the drying strategy had a significant effect on the porosity variations in the coating layer. In calendering, the low coat weight areas (peaks) are more compressed than high coat weight areas (valleys). This means that the drying strategy affected the deformation of the coating layer and thus its mechanical properties. To discuss the effect of calendering on print mottle is outside the scope of this paper. Readers interested in this question are referred to papers by Ragnarsson et al. [41] and by Engström [20], the latter a review paper.

CONCLUSIONS Print mottle is caused by nonuniform surface porosity, which leads to nonuniform ink setting and retransfer from an already printed sheet when the sheet passes through the printing nip in a subsequent printing unit with the print in contact with the rubber blanket in that nip. • A nonuniform surface porosity (uncalendered) in bladecoated papers is a natural consequence of the coating mass distribution. This is more or less nonuniform because of the floc structure of the base paper and the filling-in character of the blade coating process. More coating is deposited between the flocs (high coat weight) than on the peaks (low coat weight). A uniform mass distribution and simultaneously a smooth coated paper surface (uncalendered) are a guarantee for a mottle-free print. In blade coating, this makes the coating mass distribution a key parameter. High blade pressures, compressible base papers, and a high water uptake between the coating color application and the metering with the blade, which plasticizes the base paper, yield a uniform mass distribution. • During the process of consolidation, the binder in the wet coating layer is redistributed. The general opinion earlier was that the surface porosity becomes nonuniform by migration of binder up to the coating surface and sealing it — more in the high coat weight areas than in the low coat weight areas. New research based on coated FEBRUARY 2016 | VOL. 15 NO. 2 | TAPPI JOURNAL

99

PRINTING papers produced and printed under industrial conditions has, however, shown that this opinion is probably incorrect. Modern research shows that the coating surface is depleted of binder and that the mottle and the depletion are rated in the same way. The less binder in the coating surface the stronger is the mottle, and vice versa. • For a given coating color and base paper, there is generally a good relationship between the binder distribution in the coating surface and the print mottle, but the relationship is not unique. The mottle is stronger in LWC papers than in woodfree papers, even though the mass distribution of the former is more uniform. This is probably because depletion of binder from the coating surface is stronger on the absorbent LWC paper. There are indications that coating colors with starch as binder are more sensitive to depletion than those with latex. • The drying profile has a significant effect on mottle. Less mottle is obtained if the drying is performed at a high evaporation rate. Even less mottle can be obtained with a break in the course of evaporation between FCC and SCC. Practically, this is achieved by shutting down a dryer hood. For coatings with starch, a short delay between the metering with the blade and the dryer reduces the mottle. • During consolidation, the wet coating layer (filter cake) shrinks between FCC and SCC, when the final structure is formed. The shrinkage is governed by the capillary forces developed in the filter cake during this process. Because the filter cake can be assumed to be viscoelastic ABOUT THE AUTHORS Drying of coated paper and its impact on offset print mottle was a popular research area worldwide when I was employed in 1975 by the Swedish Pulp & Paper Research Institute (now Innventia AB) in Stockholm. And the Swedish Pulp & Paper Research Institute was no exception; the research there on paper coating had a strong focus on drying and mottling. At that time, the experimental part of the research was solely done on a bench scale. In the 1980s, pilot coaters were erected at many places. This opened possibilities to bring the research forward and make it directly transferable to industrial conditions. During the 1980s, I had a close co-operation first with Valmet, Finland, and later with Centre Technique du Papier (CTP), Grenoble, France, in which we produced coated paper on a pilot scale under industrial conditions and then printed the coated paper in lithographic offset on an industrial scale. The research presented in this paper originates mainly from those cooperations and the results obtained are here discussed in the light of research by others of later date. It is of utmost importance to realize the limitations in the experimental conditions (and for that

100


because of the presence of latex, the shrinkage will be rate dependent. Calculations based on typical relaxation times for latex and a relevant time scale for the consolidation under industrial condition have shown that the shrinkage, under these conditions, will be higher in the high coat weight areas than in the low coat weight areas. A recently performed pilot coating trial has shown that the drying strategy did not affect the porosity variations in the coating layer. TJ LITERATURE CITED 1. Gumbel, R., Produktivitätskriterien und papiereigheschaften im Offsetdruck, Int. Papierwirtsch. 6(4): 38(2004). 2. Chinga, G. and Helle, T., J. Pulp Pap. Sci. 29(6): 179(2003). 3. Preston, J., Hiorns, A., Elton, N., et al., TAPPI J. 7(1): 11(2008). 4. Xiang, Y., Bousfield, D., Hassler, J., et al., J. Pulp Pap. Sci. 25(9): 326(1999). 5. Dappen, J.W., Tappi 37(4): 324(1951). 6. Ahrheilger. D., Das Papier 32(10A): V100(1978). 7. Engström, G., Norrdahl, P., and Ström, G., Tappi J. 70(12): 45(1987). 8. Engström, G., Rigdahl, M., Kline, J., et al., Tappi J. 74(5): 171(1991). 9. Aschan, P.-J., TAPPI Coat. Conf., Proc., TAPPI PRESS, Atlanta, GA, USA, 1986, p. 73. 10. Norrdahl, P., TAPPI Coat. Conf., Proc., TAPPI PRESS, Atlanta, 1991, p. 417. 11. Kim, L., Pollack, M., Wittbrodt, E., et al., TAPPI J. 81(8): 153(1998).

sake, also the possibilities). I was lucky to have, more or less, free access to a pilot coater and a printing press. But all are not that happy. The most interesting result from an industrial point of view was that the recommendation, based on bench coatings and stain tests, to begin the drying Engström process with a low evaporation rate, was completely wrong. Our results based on material produced under industrial conditions showed the opposite; begin the drying with a high evaporation rate. For me personally, the research on the forming process of the coating layer and its role for the print quality has been most fortunate. The drying strategies developed within the cooperation with Valmet and CTP are now generally applied by paper coating industry. I am retired since a few years back, and this paper will be my very last one. Engström is professor emeritus, Karlstad University, Karlstad, Sweden. Email [email protected].

PRINTING 12. Arai, T., Yamasaki, T., Suzuki, K., et al., TAPPI Coat. Conf., Proc., TAPPI PRESS, Atlanta, 1988, p. 187. 13. Xiang, Y., Bousfield, D., Coleman, P., et al., TAPPI Adv. Coat. Fundam. Symp., TAPPI PRESS, Atlanta, 2000, p. 45. 14. Kenttä, E., Pöhler, T., and Juvonen, K., Nord. Pulp Pap. Res. J. 21(5): 665(2006).

38. Lee, D.I., TAPPI Coat. Conf., Proc., TAPPI PRESS, Atlanta, 1974, p. 97. 39. Laudone, G.M., Matthews, G.P., and Gane, P.A.C., “Coating shrinkage during evaporation: observation, measurement and modeling a network structure,” TAPPI Spring Adv. Coat. Fundam. Symp., TAPPI PRESS, Atlanta, 2003. 40. Groves, R., Penson, J.E., and Ruggles, C., TAPPI Coat. Conf., Proc., TAPPI PRESS, Atlanta, p. 187.

15. Groves, R., Matthews, P., Heap, J., et al., Trans. Fundam. Res. Symp., 12th, Fundamental Research Society, Bury, Lancashire, England, 2001, p. 1149.

41. Ragnarsson, M., Engström, G., and Järnström, L., Nord. Pulp Pap. Res. J. 28(2): 257(2013).

16. Rajala, P. and Koskinen, T., TAPPI J. 3(4): 19(2004).

42. Lepoutre, P. and Means, G., Tappi 61(11): 85(1978).

17. Engström, G., Tappi J. 77(4): 160(1994).

43. Dahlström, C. and Uesaka, T., Ind. Chem. Res. 51(24): 8246(2012).

18. Kolseth, P., “Coating structure and print quality,” in New Concepts in Surface Treatment, Institute for Surface Chemistry, Stockholm, Sweden, 2004. 19. Engström, G. and Lafaye, J.F., Tappi J. 75(8): 117(1992). 20. Engström, G., Prepr., 2nd Int. Papermaking Environ. Conf., Book A, Tianjin University of Science and Technology, Tianjin, China, 2008, p. 84. 21. Engström, G. and Morin, V., TAPPI J. 79(9): 120(1996). 22. Johansson, P.-Å., 22th IARIGAI Conf., Adv. Print. Sci. Technol. 22, Pentech Press, London, UK, 1993, p. 403. 23. Engström, G., “Forming and consolidation of a coating layer and their effect on offset print mottle,” Ph.D. thesis, Royal Institute of Technology, Stockholm, Sweden, 1993. 24. Hilden, K. and Mustonen, I., Pulp Pap. Can. 99(8): 57(1998). 25. Okomori, K., Yamaguchi, M., Suzuki, M., et al., “Predicting high speed blade coating runnability,” TAPPI Coat. Conf., Proc., TAPPI PRESS, Atlanta, 2002. 26. Provatas, N. and Uesaka, T., J. Pulp Pap. Sci. 29(10): 323(2003). 27. Ragnarsson, M., Järnström, L., and Engström, G., “Effect of the partial replacement of SB-latex with dextrin starch on the thickness distribution of coating layers,” TAPPI Adv. Coat. Fundam. Symp., TAPPI PRESS, Atlanta, 2008. 28. Engström, G., Trans. Fundam. Res. Symp., 13th, Fundamental Research Society, Bury, Lancashire, England, 2005, p. 1011. 29. Korpela, M., Palsanen, S., and Pitkänen, S., Wochenbl. Papierfabr. 114(8): 267(1986). 30. Leino, M.T.K and Veikkola, M.T.A., Coat./Papermakers Conf., TAPPI PRESS, Atlanta, 1998, p. 791. 31. Hamada, T. and Kohno, M., Kampia Gikyoshi 39(5): 477(1985). 32. Whalen-Shaw, M. and Eby, T., Tappi J. 74(12): 188(1991). 33. Vyörykkä, J. and Vourinen, T., Nord. Pulp Paper Res. J. 19(2): 218(2004). 34. Yamazaki, K., Nishioka, T., Hattori, Y., et al., Tappi J. 76(5): 79(1992). 35. Fujiwara, H. and Kline, J.E., Tappi J. 70(12): 97(1987). 36. Rajala, P., Nenad, M., Kiiskinen, H., et al., Appl. Therm. Eng. 24(1718): 2527(2004). 37. Watanabe, J. and Lepoutre, P., J. Appl. Polym. Sci. 27(11): 4207(1982). FEBRUARY 2016 | VOL. 15 NO. 2 | TAPPI JOURNAL

101

/ Higher in solutions and expertise,

lower in eco-impact

PAPER SOLUTIONS Archroma’s Paper Solutions Business provides worldwide expertise in the management of whiteness, coloration, surface treatment, deposit control and strength for all kind of papers, while taking a responsible approach to the environment. Our latest innovations include new food contact approved Cartasol® colorants, novel concentrated optical brightening agents from the Leucophor® range, Cartaspers® deposit control solutions for tissue, packaging and pulp applications, and Cartaguard® KST, a FDA approved grease repellent for food packaging.

100%

committed to a more sustainable paper industry* Higher use of recycled paper. Cleaner backwaters. 0RUHHI¿FLHQWSURGXFWLRQ processes.

We touch and color people’s lives every day, everywhere www.paper.archroma.com

PEER-REVIEWED

PAPERMAKING

Effect of sheet temperature on web densification during calendering ASHOK GHOSH

and

PETER W. HART

ABSTRACT: We used two different experimental methods to study the effects of sheet temperature at a given moisture content on web densification and sheet smoothness. Commercially prepared paperboard obtained from the WestRock Mahrt mill in Cottonton, AL, USA, was used in this study. Laboratory made handsheets prepared from Mahrt high yield kraft pine pulp were also used for portions of this work. Sheets were sealed in moisture-proof bags, heated to various temperatures, and calendered at multiple nip loads. The resulting changes in sheet density were determined. Additionally, a pilot-scale calender was modified to preheat the non-print side of the sheet before entering the nip. The modified sheet run allowed for an increase in the bulk temperature of the board with minimal changes in the moisture content of the print surface. Increasing the bulk temperature increased board density at a given print surface smoothness. We determined that increasing the board temperature by 14°C resulted in about a 1% increase in board density. The results of this study help explain the commercial results obtained with thermal gradient calendering. Application: Readers can use the information from this study to help determine the potential decrease in densification from reducing sheet temperature before calendering, and to determine the potential return on investment for the type of project described.

D

uring the papermaking process, the forming web is frequently compressed between hydraulically loaded steel rolls to impart a smoother surface to enhance printing properties. This process is referred to as calendering. Improvements in surface smoothness resulting from calendering often are accompanied by or result from sheet densification [1]. The effect of operating variables such as furnish type, sheet moisture, calender temperature, calender loading, and time under pressure have been well documented [2-6]. Significant discrepancy exists in the published information on the effect of sheet temperature during the calendering process. Temperature has been reported to produce calculated sheet temperature coefficients that varied by an order of magnitude [6,7]. The calendering work in those studies was done on newsprint. Relatively little information exists on the calendering of paperboards. What is typically reported is the effect of temperature and applied moisture on sheet densification during calendering [3]. For instance, sheet moisture has a dramatic effect on the glass transition temperature (Tg) of various wood components [8]. Increasing the moisture content of a fiber substantially reduces the softening temperature of that fiber, the extent of which depends on the fiber composition (lignin, carbohydrates, etc.). As pilot trials generally involve calendering of sheets at ambient temperature, there tends to be ambiguity in estimating density loss when scaling up to commercial scale where the sheet is

at elevated temperature. Furthermore, there might be value in actively reducing sheet temperature to further reduce densification during thermal gradient calendering. Commercially available calendering concepts take advantage of the sheet moisture content and Tg relationship. The hot-hard calender concept (Voith and Valmet) dampens the surface of the sheet before compressing the web between steel rolls at elevated temperatures. The damp sheet surface and elevated temperature allow surface smoothness targets to be achieved with relatively low calender loadings. As a result of low loads, sheet densification is also reduced. Valmet’s OptiCalender metal belt concept [9] also pre-wets the sheet. Significantly reduced pressure and temperature (as compared with the hot-hard concept) are used by the metal belt calender process. The process uses extended nip retention times realized through the use of a metal belt nip approximately 1 m in length. Another calendering system includes the Andritz’s PrimeCal Y calender, which has a premoisturizing and a preheating zone using a synthetic belt and a heated cylinder. The subsequent calendering is generally done using a hard nip. The current study attempted to isolate temperature effects from changes in paperboard moisture during hard nip calendering. The study also demonstrates an easy method to determine the effect of the sheet temperature on densification during calendering, which may be used to more accurately scale-up pilot data to actual mill production levels. FEBRUARY 2016 | VOL. 15 NO. 2 | TAPPI JOURNAL

103

PAPERMAKING EXPERIMENTAL Paper samples were obtained from a 686 μm, uncoated, lightly calendered board made at the WestRock Mahrt mill in Cottonton, AL, USA. These rolls were specifically made for pilot plant calendering trials. Water was eliminated from the wet stack and the wet stack was unloaded during production of these rolls. An extremely light calender loading was applied to achieve a reasonably flat cross direction caliper profile. Board samples were cut into 12.7 cm × 25.4 cm samples and sealed in moisture-proof bags. Elevated sheet temperatures were obtained by heating the sealed samples for 4 h in an oven set at the desired calendering temperature. Although some sheet moisture is lost to the air inside the sealed bag, each bag is a self-contained system, and the total moisture content of the system remained unchanged during these trials. The moisture-proof bags used in the current work were made of aluminum foil lined with polyethylene and another polymer. The bag was heat sealed with a 205°C heat bar to form a moisture-proof enclosure. When calendered at 120°C, the polymer lining did not fuse to the board. Each paper sample was calendered in the heated, sealed bag with a hard nip pilot calender operated in a manually fed sheet mode at the WestRock pilot plant in Richmond, VA, USA. The calender roll diameters were 31.75 cm and the calender

speed was 91.5 m/min. Several different temperatures and calendar loads were used. After calendering, the samples were allowed to cool in the sealed bags for 4 h in a TAPPI testing room maintained at 50% relative humidity and 23°C. The cooled, calendered samples were then removed from the sealed bags and conditioned overnight before physical testing. Caliper was measured at four positions on each sheet. The average of three sheets was used to obtain each data point. In addition to the laboratory calendering work, a pilot-scale trial was conducted at the Voith pilot facility in Krefeld, Germany. The pilot calender sheet run was modified to allow the evaluation of thermal gradient calendering. The sheet run was modified to allow the reverse side of the sheet to contact two heated rolls before entering the calender nip. By contacting the reverse side of the sheet twice, the overall sheet temperature was increased without substantially altering the moisture content of the print side of the sheet. Figures 1 and 2 show the standard and modified sheet runs, respectively. The densification of unbleached fibers was also evaluated at various basis weights. The unbleached, high-yield kraft pine fiber was obtained from the WestRock mill at Cottonton. The handsheets were made in a 30.48 cm (12 in.) square handsheet mold at three basis weights (146 g/m2, 342 g/m2, and 537 g/m2). Each sheet was trimmed by 2.54 cm (1 in.) from all sides. The sheet was then cut in half to have two

1. Standard sheet run for the Voith pilot calender.

2. Modified sheet run for preheating the reverse side of the sheet on the Voith pilot calender. 104


PAPERMAKING

3. Effect of sheet temperature on sheet densification at various nip loads using mill board.

samples per handsheet of 12.7 cm × 25.4 cm (5 in. × 10 in.) each. The sheet temperature was varied from an ambient temperature of 21°C to 88°C for subsequent calendering. RESULTS AND DISCUSSION

Effect of sheet temperature for mill-produced uncalendered board Sheet temperature was varied from an ambient temperature of 21°C up to about 120°C. Figure 3 shows the effect of sheet temperature on sheet density for three different calender loadings. As the sheet temperature increased, a resulting nearly linear increase in sheet densification occurred until the sheet temperature was increased to above 100°C. The slope of the curve increased with increasing load, indicating that the rate of densification in calendering also increases with increasing calender load. When the sheet temperature exceeded 100°C, the heatsealed bags containing the trial sheets tended to inflate. We speculate that above 100°C, a significant amount of sheet moisture migrated from the board into the air space sealed in the bag. The resulting water loss decreased the moisture content of the sheet, resulting in a significant increase in the Tg of the fiber and a corresponding decrease in the expected amount of densification. Thus, the maximum sheet temperature was limited to 88°C. Figure 4 shows the effects of calender loading at various sheet temperatures. In general, reducing calender load by about 9 kN/m resulted in about a 1% decrease in board density. Similarly, a 13°C reduction in sheet temperature at nearly constant sheet moisture also resulted in about a 1% (average) decrease in board density (Fig. 3). The effect of calendering on sheet smoothness was not measured for the previously mentioned experiments because of negative effects to the sheet surface from the foil bags containing the sheets. Therefore, sheet density was used as a surrogate for important surface properties.

4. Impact of calender loading at various sheet temperatures using mill board.

Thermal gradient calendering using pilot-scale equipment The calendering of board in foil bags showed a significant effect of temperature at a given moisture content on sheet densification. The foil bag results also tended to show that increased sheet moisture could be used to increase board density during calendering. Although this is useful information in the basic understanding of calendering and board densification principles, it is not of great applicability under normal pilot or commercial plant conditions. Increasing the sheet temperature under typical pilot or commercial plant conditions typically results in a decrease in the moisture content of the board, resulting in a corresponding increase in the Tg of the fiber. To evaluate the effect of temperature while maintaining nearly constant sheet moisture, the pilot calender sheet run at the Voith pilot plant in Krefeld was reconfigured so the reverse (non-print) side of the sheet directly contacted two heated rolls before passing through the calender nip. (Figs. 1 and 2 provide details on the sheet runs and calender setup.) Mahrt mill board produced on the No. 1 paper machine was used for this set of trials. Preheating the reverse side of the board just before the calender nip resulted in an increase in the overall sheet temperature, with only minimal changes in the moisture content of the print side of the board. If most of the board is hotter than the print side surface, the resulting calendering should increase densification for a given surface smoothness. Figure 5 shows the thermal gradient calendering results using Parker print-surf (PPS) as the measure of print surface smoothness. For nip loads of less than about 200 kN/m, heating the back side of the board resulted in a 1% increase in sheet densification for a given smoothness (up to ~3.8 PPS smoothness level). As the calendering loads were increased to more than 200 kN/m, sheet densification for a given surface smoothness was similar for the two different heating conditions. FEBRUARY 2016 | VOL. 15 NO. 2 | TAPPI JOURNAL

105

PAPERMAKING

5. Effect of preheating the non-print side of the board on surface smoothness and density using mill board.

6. Calender load curve for mill produced board and handsheet (537 g/m2) at 88°C.

7. Densification at different sheet temperatures for unbleached pine handsheets with 146 g/m2 basis weight.


Effect of basis weight on densification at different sheet temperatures Having determined the effect of sheet temperature on densification during calendering, an attempt was made to determine whether increasing the basis weight significantly changes the densification rate during calendering for different sheet temperatures. To perform this work, laboratory made handsheets were made at three different basis weights (146 g/m2, 342 g/m2, and 537 g/m2). These sheets were calendered in foil bags at different temperatures similar to the mill board experiments described previously. The effect of temperature on differing basis weight boards is important. Any capital expenditures to alter sheet temperature before calendering would have to return desirable results and value (changes in densification and sheet smoothness) across the entire grade range of a paper machine to be considered. The bulk reduction of mill sheets was less than that for the unbleached pine handsheets for a given calender load (Fig. 6). The average initial bulk of the mill sheets was 106



PAPERMAKING 1.62 cm3/g, and that of the handsheets was 3.02 cm3/g. The difference in initial bulk of the handsheets, in part, resulted in differences in densification during subsequent calendering. Nonetheless, the relative effect of basis weight on densification during calendering at different sheet temperatures may still be evaluated. Figures 7, 8, and 9 show the effects of calender loading at various sheet temperatures on the densification of the handsheets. The densification rate for the lowest basis weight (146 g/m2) was lower than that for the higher basis weights (342 g/m2 and 537 g/m2). Furthermore, the densification rate for 342 g/m2 and 537 g/m2 basis weights were similar. This suggests that reducing sheet temperature might result in relatively greater bulk improvement for the heavier weight grades. CONCLUSIONS The effect of temperature, relatively independent of sheet moisture content, was successfully investigated with two different experimental techniques. The first method involved

ABOUT THE AUTHORS We chose this topic to research because density reduction in paperboard is desirable for most, if not all, applications. Board is typically sold by area, and any weight reduction while maintaining certain properties, such as stiffness, directly helps improve the profitability. Significant discrepancy exists in the published information on the effect of sheet temperature during calendering of newsprint. Temperature has been reported to produce calculated sheet temperature coefficients that varied by an order of magnitude. Relatively little information exists on the calendering of paperboards. What is typically reported is the effect of temperature and applied moisture on sheet densification during calendering. For instance, sheet moisture has a dramatic effect on the glass transition temperature of various wood components. Increasing the moisture content of a fiber substantially reduces the softening temperature of that fiber, the extent of which depends on the fiber composition (lignin, carbohydrates, etc.). As pilot trials generally involve calendering of sheets at ambient temperature, there tends to be ambiguity in estimating density loss when scaling up to commercial scale where the sheet is at elevated temperature. The most difficult challenge of this research was to develop a relatively simple method where one could vary the temperature of the paperboard while maintaining sheet moisture. We accomplished this by placing the paperboard samples in sealed plastic pouches. The samples were then heated in an oven at different temperatures and then calendered while

sealing sheets in a moisture-proof, heat-sealable bag, heating the sealed sheets in an oven, and pilot-scale calendering the hot, sealed sheets at various nip loadings. Reducing the sheet temperature by 14°C resulted in a 1% reduction in densification during calendering. The second method evaluated the effect of thermal gradient calendering on another pilot-scale calender. The sheet run was modified so the non-print side of the board was brought into direct contact with two heated thermal rolls just before entering the calender nip. As expected, heating the back side of the sheet resulted in increased densification for a given print side smoothness value. It was important not to heat the print side of the sheet to maintain similar conditions on the print side of the sheet. Heating the print side would have resulted in increased fiber temperature and decreased sheet moisture on the print side, as well as affecting the resulting sheet smoothness. The effect of basis weight on the sheet temperature induced densification during calendering was investigated using handsheets. The results indicated that reducing the sheet temperature may have a greater effect on bulk improve-

enclosed in the sealed pouches. This method is simple enough for anyone with access to a laboratory or pilot calender to replicate using their own samples. Surprisingly, we could express the change in degree of densification with respect to temperature using simple linear regression. With all the variables Ghosh at play in this work, we somewhat expected a complex non-linear response. The only non-linear response was seen when the sheet temperature was increased to the boiling point of water. The study demonstrated an easy method to determine the effect of the sheet temperature on densification during Hart calendering, which may be used to more accurately scale-up pilot data to actual mill production levels. This study was on unbleached board. The study should be extended to board made with bleached and recycled fibers. Ghosh is senior R&D specialist and Hart is director, Fiber Technology and Innovation, WestRock Company, Richmond, VA, USA. Email Ghosh at ashok.ghosh@ westrock.com.


107

PAPERMAKING ment for the heavier basis weight grades. The densification rate as a function of the sheet temperature initially increased with increasing basis weight and remained at similar levels for subsequent increases in basis weights. Therefore, the relative effect of basis weight on the sheet temperature induced densification during calendering will largely depend on the basis weight range under investigation. Reducing the bulk temperature of the sheet before high temperature calendering significantly reduces densification for a given calendered smoothness. Thus, the findings associated with the effect of temperature on surface smoothness and sheet densification during calendering are in agreement with and help explain the results from the available literature on thermal gradient calendering. TJ

4. Guérin, D. and Morin, V., “Fundamentals and practical experiments of heat transfer in calenders,” 2003 TAPPI Spring Tech. Conf. Exhib., TAPPI PRESS, Atlanta, GA, USA. 5. Crotogino, R.H., Tappi J. 65(10): 97(1982). 6. Colley, J. and Peel, J.D., Pap. Technol. 13(5): T166(1972). 7. Kawka, D.W., Crotogino, R.H., and Douglas, W.J.M., TAPPI J. 1(10): 26(2002). 8. Back, E.L. and Salmén, N.L., Tappi J. 65(7): 107(1982). 9. Halmari, E. and M. Viljanmaa, Fiber & Paper 8(3): 46(2006).

ACKNOWLEDGEMENTS The authors thank T.J. Green and S. Reigel for their technical input, and N. Kilgore for testing. LITERATURE CITED 1. Chapman, D.L.T. and Peel, J.D., Pap. Technol. 10(2): T116(1969). 2. Forseth, T. and Helle, T., J. Pulp Pap. Sci. 23(3): J95(1997). 3. Caulfield, D., in Solid Mechanics Advances in Paper Related Industries: Proceedings of National Science Foundation Workshop (R.W. Perkins, R.E. Mark, and J.L. Thorpe, Eds.), Syracuse University, Syracuse, NY, USA, 1990, pp. 50-62.

KEYNOTE SPEAKER

Ken Schmidt Legendary Former Director of Communications Strategy at Harley-Davidson Motor Company

CO-LOCATED PROGRAMS

• Digital Revolution – Pulp and Paper Mill of the Future • TAPPI Young Professionals Leadership Program • Papermaking Tutorial Program • Women’s Leadership Panel Discussion • OpEx Maintenance and Reliability Program

www.PaperCon.org 108


Unpack the power of Maximyze for packaging. ®

Let Buckman help you improve sheet strength and increase productivity. Buckman announces new Maximyze enzymatic technology for recycled packaging. It can significantly improve sheet strength and drainage, so you can increase machine speeds. With a customized Maximyze program you can reduce fiber costs, steam consumption, transportation costs and your environmental footprint too. No wonder it’s an EPA Presidential Green Chemistry Challenge Award winner! Find out more. Contact your Buckman representative or visit buckman.com. Better drainage Production on a recycled linerboard machine was limited due to drainage. Buckman’s Maximyze application improved drainage, so machine speeds could be increased by as much as 100 mpm. Steam use was reduced 8%, and CO2 emissions were reduced by 1806 metric tons per year. Reduced energy A core and tube producer wanted to increase production, have greater flexibility in its fiber selection and reduce energy use. Buckman applied Maximize to the pulper, which conditioned the fiber faster with less refining energy, preserving fiber strength. Speed increased 10%. Refining energy decreased 30%. And tensile strength increased from 20 to 26 kgf/15mm.

© Buckman Laboratories International, Inc. All rights reserved.

Commitment makes the best chemistry.

ALL PAPER IS NOT CREATED EQUAL. Looking good on paper matters. ColorLok® Technology gives paper more vivid colors, bolder blacks, and faster drying times.* Learn more at colorlok.com THE SCIENCE OF WOWING YOUR CUSTOMERS ColorLok Technology turns everyday paper into a canvas of possibilities. Additives injected during the paper-making process enable the paper to ”lock” pigments at the surface. Without this technology, paper absorbs pigments deep into the fibers of the paper.

Without ColorLok Technology, pigment sinks into the paper, dulling appearances.

With ColorLok Technology, additives keep color near the surface for bold results.

A NEW LOOK FOR ECO-FRIENDLY PAPER Even recyclable paper looks better than ever, because ColorLok Technology gives printouts on recyclable paper a brighter, richer appearance. Plus, pigments sit at the surface of the paper, making recycling easier. So yes, you can look good and do good.

VIVID COLORS

BOLDER BLACKS

ECO-FRIENDLY

FASTER DRYING

CONSISTENCY

© 2015 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice. Independent testing by SpencerLab Digital Color Laboratory, commissioned by HP. The full October 2009 report, “Print Quality Analyses – ColorLok Media Evaluation: AiO Printers: Brother MFC-5490CN, Canon PIXMA MX860 & MX7600, Epson WorkForce 600, HP Officejet 6500 and Officejet Pro 8500,” is available for download at spencerlab.com

*

PEER-REVIEWED

PROCESS CONTROL

Support vector regression and model reference adaptive control for optimum control of nonlinear drying process C. KARTHIK, K. VALARMATHI,

and

M. RAJALAKSHMI

ABSTRACT: In this paper, a support vector regression (SVR)-based system identification and model reference adaptive control (MRAC) strategy for stable nonlinear process input-output form is designed. In order to implement the proposed control structure, SVR-based identification methods are clearly addressed. The control of a moisture process on the paper machine illustrates the proposed design procedure and the properties of the SVR-based model identification-adaptive reference model for the nonlinear system. MRAC is widely used in linear system control areas, and neural networks (NN) are often used to extend MRAC to nonlinear areas. Some drawbacks of NN with MRAC are slow speed in learning, weak generalization ability, and a local minima tendency. To overcome this problem, SVR is used instead of NN. With the support vector regressor, a stable controller-parameter adjustment mechanism is constructed by using the model reference adaptive theory. Simulation results show that the proposed approach could reach desired performance. Application: Paper mills that use the SVR identified MRAC for proper reduction of moisture can achieve less power consumption, more productivity, and improved end-product quality.

N

onlinear system identification is a very important study in the control field. Nonlinear system identification has many difficulties compared to a linear system. Among the different nonlinear identification techniques, methods based on the neural network (NN) model are gradually becoming established — not only in academia but also in industrial applications. However, such methods have problems of slow convergence speed and local minima. In the last ten years, NN models and, more recently, support vector regression (SVR) have appeared as two attractive tools for modeling the system, particularly in situations where the development of phenomenological or conventional regression models becomes impractical or not easily manageable. In recent years, SVR [1-3] has provided a statistical learning theory that is preferable to NN due to its many attractive characteristics and their underlying empirical performance. The major difference between NN and SVR lie in the risk minimization principle. The NN applies the empirical risk minimization (ERM) standard to minimize error in the training data, while SVR follows structural risk minimization (SRM). In SVR, a top bound on the generalization error is minimized. Hence, it opposed to the ERM, which minimizes the prediction mistake of the training data. This equips the SVR to generalize the linear relationships. During its training stage, SVR is able to understand the relationship for making good predictions for the new input data set. Although the basis

of the SVR paradigm was established in the mid-1990s, its chemical technology applications, such as obvious error detection [4,5], have appeared recently. The quality of SVR models primarily depends on a proper setting of support vector machine (SVM) parameters. The main issue is how to set these parameters for a given training and testing data set to ensure good generalization performance. MOISTURE PROCESS A typical machine for producing paper or paperboard is about the length of two football fields. The run time speed can be up to 2000 m/min. The machine consists of seven sections of production units. The flow of production of paper from pulping consists of three major sections: wire, press, and dryer (Fig. 1). The first section of the paper machine is called the wet end. This is where the diluted pulp stock first comes into contact with the paper machine and is poured onto the machine from the headbox. A narrow aperture running across the width of the box allows the stock to flow onto the wire section with the fibers distributed evenly over the whole width of the paper machine. The machine is operated by a central control system, which monitors the paper process for various properties such as moisture content, basis weight, color, etc., and provides necessary adjustments. NN can be used effectively for the identification and control of nonlinear dynamical systems [6-11]. FEBRUARY 2016 | VOL. 15 NO. 2 | TAPPI JOURNAL

111

PROCESS CONTROL which use steel rolls, there are also soft calenders consisting of paired rolls where one is made of steel and the other is coated with a soft, plastic material. This produces a better overall glazing effect. At the end of the machine, the paper is wound onto steel cores. Most paper machines use calenders or pope reels. To make a uniform wind at constant circumferential speed, the paper reel is pressed with opposition force.

1. Paper manufacturing process flow.

Dryer section The paper leaves the press section with a dry content of up to 50%−55%. Additional water is removed by vaporization in the press section. In the dryer section, the most common type of paper drying is contact drying on cylinders heated with vapor. Here, the heat energy is transferred from the outside walls of the drying cylinders to the paper surface by direct contact. The dryer section consists of a succession of drying cylinders, and as the paper web is transported over and between these cylinders, the paper alternately makes contact with the upper and the lower side. Drying takes place in different phases. In the short, first phase, only heat is transferred to the paper. There is no vaporization. This takes place during the second phase, when the wet paper starts to convey its humidity to the surrounding air as water in the paper evaporates. In the third phase, the paper surface has already been dried to the maximum extent, and heat transmission into the dry paper stimulates vaporization inside the paper.

End group After the end of the drying process, the paper is often subjected to machine calenders. Besides machine calenders, 112


SYSTEM IDENTIFICATION System identification is a process that can be built from the collection of input-output data by a mathematical model of dynamic systems. The process model provides system parameters as a transfer function in terms of poles, zeros, and delay terms. This plant model is periodically identified, and the changes in its dynamic characteristics are observed. This periodic observation provides the advantages of conventional controller tuning methods, because the plant model is always at nominal operating conditions in any situation [12]. In conventional modeling methods, a model structure will select to calculate the parameters of the model. This structure selection and parameter estimation is done by optimizing an objective function using various techniques. The selection of model structure is nothing but a compromise between model accuracy and model simplicity [13-15]. Auto regressive with exogenous inputs (ARX) is one of the simplest structures for system identification. Alireza discussed that an auto regressive with moving average exogenous inputs (ARMAX) structure is a better choice for identifying the plant model than the simple ARX structure [16]. Plant identification is the first step before an attempt can be made to control the moisture content in the papermaking process. Based on the set point of moisture, the real values of moisture content were measured by an ABB distributed control system (DCS). The actual value depends on the steam pressure of the main dryer group; i.e., if there is a rise in steam pressure, the moisture content of the paper will be decreased. Thus, the model can be identified from the collected data. Table I shows the technical specifications of the paper machine studied here. Figure 2 shows the method implemented to collect the process data for plant identification. Here, steam pressure data were collected from main dryers in the DCS system. The scanner provides actual information related to the manufacturing process. This information is listed in Table I. NEURAL NETWORK IDENTIFICATION FOR MOISTURE PROCESS In black-box identification of nonlinear dynamic systems, selection of model structures becomes a more difficult task. The multi-layer perceptron (MLP) network is most popular for learning nonlinear relationships from a set of data. Two issues primarily arise when selecting the model structure: • Selecting the inputs to the network • Selecting internal network architecture

PROCESS CONTROL Part Name

Details

DCS system AC 450 controller

Operating system: HMI, ABB Analog input: 4−20 MA Analog output: 4−20MA Software: AMPL (A Mathematical Programming Language)

Quality control scanner

Moisture: Infrared (IR) sensor Output: 4−20 MA Brand: Honeywell

Control valve

Size: 6 in., pneumatic actuated Type: air to open

I/P converter

Input: 4−20 MA Output: 0−6 Bar

Dryer

43 cylinders, 5 groups, material-milled steel

Steam pressure

3.5−4.5 bar

Steam temperature

150°C−180°C

Daily production

350 metric tons

Machine speed

700 m/min

In Eqs. (1) and (2), a general linear model structure and predictor form is given:

(1)

(2)

where ϕ (t,θ) is regression vector, 'θ’ is vector containing the adjustable parameters in the NN known as weight, and g is a function realized by the NN.

Model structure Depending on the choice of regression vector, various NN model structures are examined. The neural network ARMAX (NNARMAX) can be represented by the regression vector and predictor in Eqs. (3) and (4), respectively: Regression Vector: T

(3)


113

where:

I. Technical specifications of the paper machine studied.

2. Motion-detection moisture control.

PROCESS CONTROL Predictor:

(4)

where: C = polynomial in the backward shift operator

sensitive zone. Both C and ε are user-defined parameters. Schematic representation of the SVR using ε-insensitive loss function is illustrated in Fig. 3. Applying the Lagrange function and the Karush-KuhnTucker conditions to Eq. (5) yields the following dual optional form:

(7)

Max: In this paper, the NNARMAX model structure was selected since it has a few advantages over other structures. Only the NN auto regressive with exogenous inputs (NNARX) model has a predictor without feedback. The remaining model types all have feedback through the choice of regressors, which in NN terminology means that the networks become recurrent in that future network inputs will depend on present and past network outputs. This might lead to instability in certain areas of the network’s operating range, and it can be very difficult to determine whether or not the predictor is stable. The ARMAX structure was constructed to avoid this by using a linear MA-filter for filtering past residuals. When a particular model structure has been selected, the next choice that must be made is the number of past signals used as regressors, i.e., the model order [7]. To solve the problem of local minima, the NNARMAX model was trained several times and initialization of weights were varied at the beginning each time. IMPLEMENTATION OF SVR FOR NONLINEAR SYSTEM IDENTIFICATION The basic idea in SVR is to map the dataset {(x1,y1), … … (x1, y1)}c Rn × R) into a high dimensional feature space via non-linear mapping, wherein they are linearly correlated with the outputs [17]. The SVR formalism considers the following linear estimation function:

(5)

∈ Rn

where w is weight vector, b ∈ R is a constant, and ∅ (x) denotes a mapping function in the feature space. Based on the principle of structural risk minimization, the SVR learning problem is recast as the optimization problem, which is given below: Min: (6) Subject to:

where C is the regularization constant used to specify the tradeoff between the empirical risk and regularization term. Two positive slack variables, ξ1, and ξ*1 i =1,2,…..,n can be used to measure the deviation from the boundaries of the ε -insensitive zone. That is, they represent the distance from actual values to the corresponding boundary values of ε -in114


Subject to the constraints:

where the Lagrange multipliers ai, a*i satisfy the equality ai, a*i = 0. The Lagrange multipliers are calculated and an optimal desired weight vector of the regression hyper-plan is w = Σli =1(ai, a*i ) ∅ (xi). Hence, the general form of the SVRbased regression function can be written as:

(8)

where K (x, xi) is called the kernel function and the values of it equal the inner product of two vectors, xi and xj, and feature space ∅ (xi ) and ∅ (xj); i.e., K(x, xi) = ∅ (xi ) ∅ (xj ). The function that meets Mercer’s condition can be used as the kernel function [18] and the radial basis function (RBF) is applied in this paper as kernel function, which is defined as:

(9)

where σ 2 is the width of the RBF. In order to design an effective model, the values of two essential model parameters of the C and ε and kernel function parameter σ 2 in SVR have to be optimally chosen in advance. The regularization parameter C determines the tradeoff cost between minimizing the training error and minimizing model complexity, which will reduce the generalization capability when it is set too small or excessive. The parameter ε defines the nonlinear mapping from the input space to some highdimensional feature space that determines the number of support vectors, and σ 2 reflects correlation of the support vector, which also determines both the generalization capability and the prediction accuracy.

Steps in designing the SVR model The design and development of the SVR model for any regression task involves a sequence of steps, as enumerated below: Step 1. Data scaling/preprocessing The input features in the train set and test set need to be scaled

PROCESS CONTROL leads to underfitting [10]. Hence, proper selection of SVR parameters, while designing the SVR model, is highly important to achieve a good mean squared error (MSE) and high generalization ability [21]. In this paper, the grid search (GS) method is used for optimal selection of SVM parameters.

3. Sketch map of the support vector regression (SVR) using ε-insensitive loss function.

properly before applying SVR [14]. This is important, as the kernel values depend on the inner products of the feature vector. Scaling prevents the domination of any feature over the other because of higher numeric values involved and also avoids numerical difficulties during calculation. Then, each attribute is linearly scaled to the range of [0, 1]. Step 2. Design of SVM model The RBF kernel is chosen as a first choice because of its widely accepted accuracy. Further, it is capable of handling nonlinear relationships existing between the class labels and input attributes. The second reason is that the RBF kernel, unlike other kernels, has only one kernel parameter, thereby reducing the complexity of the model [19]. There are two parameters associated with the SVR model designed with RBF kernel–penalty parameter, C, and RBF kernel parameter, γ. The goal is to identify optimal (C, γ) for the classifier to accurately predict the unknown data (test data) [20]. This can be achieved by different techniques, descriptions of which follow in the next section. After designing the SVR model with the chosen kernel and optimal parameters, it is trained with the scaled input-output train set samples. Once the performance of the regression is found satisfactory in the training phase, the model is validated with test samples to determine its overall performance. Step 3. Selection of SVM parameters The SVR model in this work uses RBF as the kernel mapping function. Two major parameters associated with the SVR model that need to be optimized are: 1) penalty parameter (C), and 2) RBF kernel function parameter (γ). Parameter C, which represents the cost of penalty, influences the classification performance [18]. A larger C value gives a good accuracy rate in the train phase, but very poor accuracy in the test phase. Conversely, too small a C value leads to an unsatisfactory performance, making the model useless. Parameter γ, representing the kernel spread, has a strong influence on partitioning outcome in feature space. An excessive C value results in overfitting, while a disproportionately small value

Step 4. Grid search GS is the most common and simple method used to determine the appropriate values for parameters C and γ. The GS method adopts a v-fold cross validation technique[13]. In a v-fold cross validation, the whole training set is divided into v subsets of equal size. Sequentially, one subset is tested using the SVM regression trained on the remaining (v-1) subsets. Thus, each instance of the training set is predicted once and the crossvalidation accuracy is the percentage of data samples that are correctly classified [21]. In this work, GS using fivefold cross validation is used. All pairs of (C, γ) were tried and the one with highest cross-validation accuracy was selected [15]. Exponentially growing sequences of C and γ is a practical method to identify optimal parameters. The sequences of C = {2-5,…..,215} and γ = {2-15,….,25} are used in this SVR experiment. MODEL REFERENCE ADAPTIVE CONTROL The model reference adaptive scheme is based on the observation that the difference between the output of the plant and the output of the reference model (subsequently called plantmodel error) is a measure of the difference between the real and the desired performance [22]. This information is used by the adaptation mechanism to directly adjust the parameters of the controller in real time in order to asymptotically force the plant model error to zero. This scheme used for control applications may be called a model reference adaptive system (MRAS). Nonlinear characteristics of a system validate the use of adaptive controllers such as model reference adaptive control (MRAC) and SVR [17]. The model reference adaptive system gives the desired response to the reference signal [23]. Thus, the response of the command model acts as reference for the process to the normal feedback loop. The adaptive mechanism used for adjusting the parameters in an MRAS can be obtained in two ways: by using a gradient method or by applying Lyapunov stability theory [24]. The mechanism uses the error (e), the nonlinear process output (y), and the input signal (uc) to adjust the parameters of the control system. These parameters are adjusted so as to minimize the error. As this paper discusses, the parameters are varied so that the cost function J (θ) is reduced. Comparison of plant and model output provides error information that can be used for finding the cost function in Eq. (12). This cost function is to optimize the controller output: Cost function: where e is

(10)


115

PROCESS CONTROL MIT rule is where is is

and

(11)

Also, the controller parameters 01 and 02 are updated from and

.

Fuzzy model reference adaptive control Fuzzy systems based on MRAC have been reported by a number of researchers. The major components are the reference model, fuzzy logic controller (FLC), and an adaptation mechanism. The reference model incorporates the desired performance characteristics of the system response. The FLC is implemented in a reference adaptive control system based on the gradient descent algorithm method [25]. The adaptation mechanism is involved to adjust the characteristic of the FLC in response to the error signal e( t) ,which is the output difference

of the reference model and plant. An inverse plant model is designed as an adaptive mechanism for the correction value of ΔU(t) applied to the U(t), which is the output signal. The adaptive mechanism updates the information in order to apply that correction via the FLC [19,20]. The updating algorithm tries to modify the FLC characteristic such that its output is altered by the value ΔU (t). The adaptation is affected by adjusting the value of the output fuzzy sets [9]. In the proposed method, the error that can be measured between the moisture and the output of a reference model is applied to an FLC, which can force the system to act as the model by adjusting the knowledge base of the fuzzy controller or by adding a correction signal to the fuzzy controller output [22]. The structure of fuzzy modified model reference adaptive control (FMRAC) is monozygotic to that of the FLC, as illustrated in Fig. 4. RESULTS AND DISCUSSION The identification process was performed using a conventional and support vector regressor. A total of 2000 data samples were taken that describe the behavior of the moisture

4. Fuzzy-modified model reference adaptive control (FMMRAC) structure.

5. Input and output response. 116


PROCESS CONTROL process, and 1260 data samples were from a single shift. Another 2500 data samples were collected from two shifts a day. The data set was then divided into 70% for training and 30% for testing. For training data of input-output pairs, output was “moisture of a paper reel” and input was “pressure variation in the cylinder.” Figure 5 shows the input and the output response of the moisture data from the selected training and testing data set. Here, the NN regression models were applied to the divided set of data. The model structure selection consists of two subproblems: choosing a regressor structure and choosing network architecture. For all regressor network architecture, five

hidden “tanh” units and one linear output unit were chosen. From Figs. 6-8, the NN structure for different nonlinear identification-based NNs is presented. NN-based nonlinear structure is generally defined in terms of NNARX structure, and the structure is used to identify the nonlinear system. Based on the identification, the correlation function can be plotted as in Figs. 9-11, where NNARX, NNARMAX, and neural network output error (NNOE) structure were analyzed for the empirical process. Figures 9-11 depict the correlation plot for the prediction error by the different models for the nonlinear system. The comparison provides detailed performance analysis of the different NN mod-

6. Network structure for neural network auto regressive with exogenous inputs (NNARX) model.

7. Network structure for neural network output error (NNOE) model.

8. Network structure for neural network auto regressive with moving average exogenous inputs (NNARMAX) model. FEBRUARY 2016 | VOL. 15 NO. 2 | TAPPI JOURNAL

117

PROCESS CONTROL

9. Auto-correlation and cross-correlation of prediction error in NNARX model.

10. Auto-correlation and cross-correlation of prediction error in NNARMAX model

els. The cross-validation procedure is the best procedure to validate models [24]. During the procedure, the estimated NNARX, NNARMAX, and NNOE models are applied to a data set that differs from the one used for estimation. Based on this analysis, a fitness of 93.24% is obtained from the NNARMAX model. Model validation can also be done by applying residual analysis and correlation analysis to the experimental data. The residual analysis gives the auto correlation of the output residuals and the cross correlation between the input to the process [26]. 118


The validation responses obtained from the various NR structures are plotted in Figs. 12-14. In this analysis, the performance of an NNARMAX model provides a minimized MSE value.

Implementation of SVR identification of moisture process The LIBSVM software developed by Chang and Lin has been used for the design and testing of the SVM model [27]. Pro-

PROCESS CONTROL

11. Auto-correlation and cross-correlation of prediction error in NNOE model.

11. Auto-correlation and cross-correlation of prediction error in NNOE model.

grams for parameter selection of SVM by GS techniques are written in the MATLAB environment. The SVR method can be implemented because it appeared the best tool for modeling the nonlinear system [1,3]. In this paper, the SVM model uses a radial basis function (RBF)-type kernel function. The performance of any SVM regression model will strongly depend on the selection of SVM parameters. The penalty parameter (C) and RBF kernel parameter (c) are associated with

SVM [28-33]. The selection of parameters is performed by the GS technique, and the results are shown in Table II. The selection of parameters was done with the fivefold cross validation procedure. In Fig. 15, SVR model response uses the GS parameter selection method. The best values of SVR parameters obtained for a maximum cross validation error of 0.00068879 as seen from Table II are 2γ = 5 and 2C = 10000. The collectFEBRUARY 2016 | VOL. 15 NO. 2 | TAPPI JOURNAL

119

PROCESS CONTROL

12. Validation response of NNARX model.

13. Validation response of NNARMAX model.

ed data from the DCS were plotted as shown in Fig. 5. From the collection, 54% of data is given for input testing and the remaining are given for input training. Input cylinder pressure and output moisture data are given for training and testing of SVR methods. In Fig. 15, the validation response of an SVR is plotted, and the response shows its capability to better reduce MSE than previous methods. A performance comparison of various identification structures in reducing MSE is given in Table III. The obtained model was identified as 120


suitable and given to the adaptive controller for optimized control of the process. Based on the previous comparison made with an NN structure nonlinear identification, the fitness value and the correlation MSE of the NNARMAX model gives better performance. Ultimately, to improve the value of fitness, an advanced NN strategy was implemented ― SVR. Here, the RBF-based kernel function provides better performance than the value of previous comparative methods listed in Table II, which shows the detailed analysis of SVR within

PROCESS CONTROL

14. Validation response of NNOE model.

Γ

C=0.1

C=1

C=10

C=100

C=1000

C=10000

2-3

0.683743

0.683589

0.0680289

0.0679112

0.672649

0.0682098

2-2

0.683728

0.0683454

0.0678852

0.678734

0.0680817

0.0701059

2-1

0.0683573

0.0683362

0.06781

0.067812

0.0678762

0.0699764

20

0.0683367

0.0683029

0.0678166

0.0677027

0.0675849

0.00712111

21

0.0683047

0.068134

0.0677606

0.0677062

0.0676804

0.0672988

22

0.0681362

0.0679116

0.0676907

0.0675335

0.0675084

0.0708155

23

0.067917

0.0678562

0.0677108

0.0678762

0.0680621

0.0687694

24

0.0678477

0.0677686

0.0679467

0.0682526

0.0680976

0.0709433

25

0.0677513

0.0677195

0.0682725

0.0068338

0.0068310

0.00068879

II. Cross validation error.

a prescribed range of γ and C values. The best values of SVR parameters obtained for a maximum cross validation mean squared error of 0.000689, as seen from Table II, are 2γ = 5 and 2C = 10000; i.e., nearly 214. In Table III, various identification structure methods were compared with SVR performance. Here, the performance index was chosen as MSE, and the least value of MSE is obtained from SVR.

Controller implementation Based on the previously made comparison, the best performance was obtained from the SVR model, so it was also applied as a model to the MRAC. The combination of SVR iden-

Method

MSE

NNARX

0.003232

NNARMAX

0.002176

NNOE

0.03154

SVR

0.000689

MSE = mean square error; NNARX = neural network auto regressive with exogenous inputs; NNARMAX = neural network auto regressive with moving average exogenous inputs; NNOE=neural network output error; SVR = support vector regression.

III. Performance comparison of various identification structures. FEBRUARY 2016 | VOL. 15 NO. 2 | TAPPI JOURNAL

121

PROCESS CONTROL

15. Validation response of SVR model.

16. Servo response analysis of different controllers. 122


PROCESS CONTROL

17. Process input for step input of 3 to the command signal.

tification and MRAC procedure provides an SVR-based MRAC (SVR-MRAC) strategy. This may provide an efficient control strategy for the nonlinear process, which is plotted in Fig. 16. Figure 16 demonstrates the performance of FMMRAC with different control methods. The proposed SVRMRAC method exactly follows the reference signal. Here, a varying input signal is given to the system for analyzing the

Tuning Method

Dynamic Performance Specifications

Performance Index

Tr

Ts

M p, %

MSE

MRAC

512

670

0.0

4.2926

MMRAC

11

16.32

1

0.3342

FMMRAC

11

15.221

0. 0

0.04406

Tr = rise time; Ts = settling time; Mp = peak overshoot; MSE = mean square error; MRAC = model reference adaptive control; MMRAC = modified MRAC; FMMRAC = fuzzy-modified MRAC.

IV. Comparison of performance indices of different controllers.

proposed method with servo response, proving the effectiveness of the proposed controller strategy. Input to the MRAC controller is varied and the response was obtained as shown in Fig. 16. The servo response was observed by varying the moisture value in the range of 4.3%−18%, and the dynamic performance of various controllers was identified and compared in Table IV and shown pictorially in Fig. 16. Based on the comparison, the FMMRAC provides good performance with a sufficient performance index and exactly follows the reference signal. So, the combination of SVR-modeled FMMRAC was identified as the optimized controller for the nonlinear process. Figure 17 shows the different responses for reference moisture value. To control the moisture content, MMRAC and FMRAC show the better performance compared to other methods. The efficient collection of data provides a suitable model with the least MSE values from a support vector concept. This can be effectively controlled by a fuzzy modified adaptive system. Figure 18 shows the simulation model of proposed fuzzy modified MRAC. Proportionalintegral-derivative (PID)-tuned MRAC can be embedded with fuzzy knowledge to provide proper control action. This performance is clearly described in Table IV. The PID FEBRUARY 2016 | VOL. 15 NO. 2 | TAPPI JOURNAL

123

PROCESS CONTROL

18. Simulation model of fuzzy modified MRAC (FMMRAC).

controller-tuned system has been removed, which is shown in Figure 18, so it will act as a simple modified MRAC system. The step input is given to the proposed SVR-MRAC system and the response was identified as shown in Fig. 17, where performances were compared. CONCLUSION The nonlinear moisture process was identified as an overdamped second order process using standard NN-based nonlinear identification procedures such as NNARX, NNARMAX, and NNOE. Then, the model accuracy was increased by using an SVR-based system identification procedure. The validation results show that the model obtained is capable of explaining the main dynamic characteristics of the nonlinear moisture process. The proposed modification of the MRAC scheme has provided improved performance in transient, as well as steady state, analysis when compared to MRAC and PID controllers. The application of fuzzy-based modified MRAC has given better results in transient and steady-state performance than those of the earlier methods of MRAC, modified MRAC, and PID controllers. The simulation results described the superiority of the fuzzy-based modified MRAC, which is proposed over the other three methods studied. The proposed modified MRAC and fuzzy-based modified MRAC give improved 124


transient and steady-state performance in the control of nonlinear processes. Future work aims to identify the moisture process by using a multiple-model approach and to design the proposed fuzzy-based modified MRAC for each local model for each fuzzy set. TJ LITERATURE CITED 1. Vapnik, V., The Nature of Statistical Learning Theory, Springer Verilag, New York, 1995. 2. Vapnik, V., Statistical Learning Theory, John Wiley, New York, 1998. 3. Vapnik, V., Golowich, S., and Smola, A., Adv. Neural Inform. Proces. Syst. 9: 281(1996). 4. Agarwal, M., Jade, A.M., Jayaraman, V.K., et al., Chem. Eng. Prog. 98(1): 57(2003). 5. Jack, L.B. and Nandi, A.K., Mech. Sys. Sig. Proc. 16(2002): 373(2002). 6. Narendra, K. and Parthasarathy, K., “Identification and control of dynamical systems using neural networks,” IEEE Trans. Neural Networks 1(1): 4(1990) . 7. Norgaard, M., Ravn, O., Poulsen, N.K., et al., Neural Networks for Modelling and Control of Dynamic Systems, Springer, London, 2000. 8. Medsker, L.R. and Jain, L.C., Recurrent Neural Networks: Design and Applications, CRC Press, Boca Raton, FL, USA, 2000.

PROCESS CONTROL 9. Vieira, J., Dias, F.M., Mota, A., “Artificial neural networks and neuro-fuzzy systems for modelling and controlling real systems: a comparative study,” Engineering Applications of Artificial Intelligence 17(3):265(2004). 10. Jain, A.K., Mao, J., and Mohiuddin, K.K., “Artificial neural networks: A tutorial,” IEEE Computer Special Issue on Neural Computing, 1996, p. 31–44. 11. Qin, H.L. and Li, X.B., “A chaotic search method for global optimization on tent map,” Electric Machines and Control 8(1): 67(2004). 12. Valarmathi, K., Devaraj, D., and Radhakrishnan, T.K., Braz. J. Chem. Eng. 26(1): 99(2009). 13. Karthik, C. and Rajalakshmi, M., “Nonlinear identification of pH process using NNARX model,” CiiT International Journal of Artificial Intelligent Systems and Machine Learning, (0974 – 9543) 4(8): (2012). 14. Karthik, C., and Rajalakshmi, M., “Non Linear Structure Identification of pH Process,” IEEE - ICAESM 2012, Int. Conf. Adv. Eng., Sci. Management, (978-1-4673-0213-5), 2012, p. 45. 15. Åström, K.J. and Hagglund, T., PID Controllers: Theory, Design, and Tuning, Instrument Society of America, Research Triangle Park, NC, USA, 1994.

16. Alireza, Karbasi, “Comparison of NNARX, ANN and ARIMA Techniques to Poultry Retail Price Forecasting,” Master’s thesis, University of Zabol, Zabol, Iran, 2009. 17. Min, J. and Lee, Y., Exp. Syst. Appl. 28(4): 603(2005). 18. Vapnik, V.N., The Nature of Statistical Learning Theory, Springer, New York, 2000. 19. Rajalakshmi, M., Kalyani, S., Jeyadevi, S., et al.,”Nonlinear Identification of pH Process using Support Vector Machine,” IJCA Proc., Int. Conf. Innovations in Intelligent Instrumentation, Optimization and Electrical Sciences (ICIIIOES) (2): 36( 2013). 20. Gretton, A., Doucet, A., Herbrich, R., et al., “Support vector regression for black box system identification,” IEEE Workshop Stat. Signal Process., 11th, 2001. 21. Cristianini, N. and Shawe-Taylor, J., An Introduction to Support Vector Machines and Other Kernel-based Learning Methods, Publishing House of Electronics Industry, Beijing, 2004. 22. Aström, K. and Wittenmark, B., Adaptive Control, Addison-Wesley, Boston, 1989. 23. Reay, D.S. and Dunniganu, M.W., IEE Proc. - Control Theory and Applications 144(6): 605(1997).

ABOUT THE AUTHORS Generally, process industries — like paper mills — have a number of nonlinear process parameters that must be controlled to obtain better performance. Of these nonlinear parameters in the paper industry, basis weight and moisture are complicated to control. Improving the quality of the controller strategy for these nonlinear processes leads to energy reduction, increased production rate, and economic profit. This is the motivation behind the authors to work in this area of research. The application of fuzzy-based modified model reference adaptive control (MRAC) has given better results in transient and steady-state performance than that of the earlier methods of MRAC, modified MRAC, and PID control proposed by the researchers. The proposed work is to identify the dryer model using support vector regression (SVR) to develop a suitable controller. Finally, the obtained SVR model is optimized with the fuzzy modified MRAC (FMMRAC) controller. The most difficult part in the research was the model identification of the dryer process. With the help of the SVR concept, this difficulty was addressed. The most interesting part in the research was the FMMRAC concept. The concept of the fuzzy inference system is key to the reference model where it is compared with the actual SVR model. The FMMRAC ensures the stability of the control system. The FMMRAC provides improved transient and steady-state performance in the control of nonlinear

Karthik

Valarmathi

Rajalakshmi

processes. This controller concept may be implemented in an industrial scanner for optimum controller performance. Development of modern scanners with the concept of FMMRAC will ensure the quality of production with less consumption of energy and more economic profit. The next step in our research is to identify the moisture process by using a multiple-model approach and to design the proposed FMMRAC for each local model of each fuzzy set. The consolidated work will extend further to identification of basis weight, ash content, color, and online monitoring with the development of a modern scanner. Karthik is assistant professor, Kalasalingam University, Tamilnadu, India. Valarmathi is professor, PSR Engineering College, Tamilnadu, India. Rajalakshmi is assistant professor, Kamaraj College of Engineering & Technology,Tamilnadu, India. Email Karthik at [email protected].


125

PROCESS CONTROL 24. Navghare, S.R., Bodhe, G.L., and Shruti,”Design of Adaptive pH Controller using ANFIS”, Int. J. Computer Applications (0975–8887) 33(6): 41(2011).

32. Al-Hiary, Heba, Braik, Malik, Sheta, Alaa, et al., “Identification of a chemical process reactor using soft computing techniques,” IEEE Int. Conf. Fuzzy Syst., 2008, p. 845.

25. Spooner, J.T., Ordonez, R., and Passino, K.M., “Stable direct adaptive control of a class of discrete time nonlinear systems,” Proc. IFAC World Congr., 13th, International Federation of Automatic Control, Laxenburg, Austria, 1996, p. 343.

33. Resendiz-Trejo, Juan Angel, Yu, Wen, and Li, Xiaoou,” Support vector machine for nonlinear system on-line identification,” Int. Conf. Electr. Electron. Eng., 3rd, Institute of Electrical and Electronics Engineers, New York, 2006.

26. Kamalasadan, Sukumar and Ghandakly, Adel A., IEEE Trans. Instrum. Meas. 56(5): 1786(2007). 27. Chang, C., and Lin, C. LIBSVM — A Library for Support Vector Machines. Available [Online] https://www.csie.ntu.edu.tw/~cjlin/ libsvm/ . 28. Kalyani, S. and Swarup, K.S., Int. J. Electrical Power & Energy Systems 44(1): 547(2013). 29. Silva Neto, J.L. and Huy, H. Le, “An improvement fuzzy learning algorithm for motion control applications,” Proc. IEE, Institute of Electrical and Electronics Engineers, New York, 1998, Vol. 1, p. 1. 30. Lin, S., Ying, K., Chen, S., et al., Exp. Syst. Appl. 35(4): 1817(2008). 31. Ljung, Lennart, System Identification - Theory for the User, 2nd edn., PTR Prentice Hall, Upper Saddle River, NJ, USA, 1999.

NETInc

The 2016

Innovative Nonwovens Conference

Innovative Nonwovens Conference May 15-18, 2016 Cincinnati, Ohio The Innovative Nonwovens Conference (NETInc.) is the only technical conference developed by engineers and technologists to help you understand the innovation behind the science and how it impacts the market.

Featured Sessions*

HERTY Innovations Filtration Tutorial Nanofibers in Nonwoven Products Fiber & Polymer Innovations

Special Events

PaperCon/NETInc Awards Dinner 7:00 P.M. Tuesday, May 17

2016 NETInc Keynote Speaker Announced GREG BUNKER

Global Business Director for Adhesives & Functional Materials Dow Chemicals

*Subject to change

For more information visit www.netincevent.org 126


PEER-REVIEWED


Anaerobic treatment to improve sludge recovery at a deinked fiber pulp and paper mill ˇ C ˇ KOZMUS, ANDREJA ŽGAJNAR GOTVAJN, ALEKSANDRA LOBNIK, NINA NOVAK, ALEKSANDRA RACI ˇ C ˇ ALJAŽ KLASINC, AND GREGOR DRAGO ZUPANCI

ABSTRACT: Investigations were conducted to explore the use of anaerobic treatment options that may allow complete recovery of primary and waste activated sludge streams for incineration from a graphic paper mill with an integrated deinked pulp plant. The effective reduction of waste activated sludge quantity (55%–65%) was obtained by the anaerobic treatment of wastewater from the deinked pulp plant or by the anaerobic stabilization of the mill effluent from primary treatment, where chemical oxygen demand (COD) removal efficiency of 67.6% and 64.8%, and biogas potential of 440 and 496 m3/t CODremoved were achieved, respectively. The anaerobic treatment of wastewater and alkali hydrolysis of waste activated sludge decreased the dewatering properties of solids residue in treated effluent resulting in a negative impact on the sludge recovery energy balance. The addition of municipal wastewater to anaerobic treatment in the amount of 18 vol%, representing 10% of the total COD, increased the reduction of waste activated sludge to 78%, with a positive impact on the treatment costs and efficiency. Anaerobic treatment of the studied effluents may enable incineration of all mill sludge with a positive energy balance, a reduction of biological wastewater treatment operational cost by up to 83%, and mill conventional energy demands by up to 2.7%, resulting in total cost savings of up to EUR 5.2 (USD 5.60) per ton of paper. Application: Mills can use the study results to integrate sludge and wastewater management and reduce mill operational costs and environmental impact.

S

ludge management in paper mills producing graphic paper (newsprint, coated and uncoated, and supercalendered) from recycled fibers (RCF; for a complete list of nomenclature/acronyms used in this paper, see Appendix I) is important due to high sludge generation rates of 170–190 dry kg/ton of paper produced [1]. The mill that is the subject of this study produces deinked RCF onsite from waste paper (newspapers, magazines, and paper cuttings). Most of the primary sludge (PS) is generated in the deinking process. The remaining PS and waste activated sludge (WAS) originates from the wastewater treatment plant (WWTP). The majority of the organic load of mill wastewater (WW) is also attributed to the deinking process [1]. Energy recovery from sludge and other waste with high organic content using heat and electricity cogeneration is considered a best available technique for waste management in the pulp and paper industry [1,2]. Reuse of PS with high fiber content in paper production is limited by product quality requirements. Use of sludge by third parties in other industrial sectors (e.g., cement, ceramics, or brick production) is possible by mixing 5%–20% of PS into the feedstock [3,4]. Composting and subsequent land spreading of sludge is greatly limited by

relevant legislation in the region where the mill is located. At many pulp and paper mills, a portion of WAS typically is mixed with PS, thickened, and dewatered [5]. Dry WAS from RCF paper mills has considerably higher caloric value than the PS, but inferior dewatering properties. Because of the increased recycling rate of waste paper in Europe (above 70%) [6,7], more degraded RCF enter the mill with recovered paper; this can increase the organic and solids load on the WWTP. At the study mill, this additional load and the shorter recycled fiber of PS reduces the proportion of WAS that can be added to PS before dewatering while still maitaining the properties required for self-supporting incineration of the sludge. Many mills incinerate WAS without a positive energy balance at waste incineration plants by supplementing conventional fuels. Alternatively, WAS can be dried with excess heat, if economically feasible, or delivered to contractors for external material utilization [1]. The high cost for WAS management inspires interest in implementing technologies to reduce WAS generation and increase the efficiency of mill sludge recovery. Anaerobic treatment (AT) of organic waste has the potential to address some of the economic and the environmental pressures associated with WAS management [8,9]. The use of FEBRUARY 2016 | VOL. 15 NO. 2 | TAPPI JOURNAL

127

WASTEWATER TREATMENT biogas from anaerobic wastewater treatment and primary and biological sludge from the paper mill as an energy source can contribute to the reduction in greenhouse gas emissions and enhance overall energy efficiency at a mill [4]. While AT of WW is common for paper mills using RCF without deinking, fewer applications of AT are observed with deinked-grade effluents (e.g., from tissue and newsprint mills) where aerobic treatment is more common. Applications of anaerobic and aerobic WW treatment for deinked grades have become more popular recently [1]. The advantages of such effluent treatment lie especially in the energy balance, low sludge growth potential, and stable operation of the aerobic post-treatment units [1,10-12]. Current obstacles to more widespread use of anaerobic digestion of WW in the pulp and paper industry include differences in WW composition and the occurrence of anaerobic inhibitors in some in-mill streams [8]. This paper presents the results of different approaches on anaerobic WW treatment from an RCF graphic paper mill with deinking. The main goal of the research was to substantially reduce the WAS quantity to improve sludge characteristics for energy recovery with a positive influence on the mill’s conventional energy demand and operational costs. MATERIALS AND METHODS

Substrates Effluent and WAS discussed in this paper was sourced from an RCF mill in Slovenia. The mill produces 200,000 tons/year of newsprint and coated and uncoated graphic papers. The mill furnish consists mainly of deinked (70%–95%) and

groundwood (10%–20%) pulp, with purchased sulfate pulp (5%–15%). Wastewater is composed of approximately 60%– 70% effluent from a deinking plant and 30%–40% effluent from three paper machines. The mill’s WWTP is equivalent in size to that serving a population of 180,000 and consists of primary treatment in a chemical-mechanical treatment plant (CMTP), followed by an aerobic activated sludge treatment plant (ASTP). The WWTP also services the local community of Krško, Slovenia, which adds approximately 18%–20% load by volume and 10%–14% load by chemical oxygen demand (COD). Solids and high-molecular-mass colloids are removed from WW in the CMTP primary sedimentation tank as PS. After thickening, the PS is mixed with a portion of the excess WAS before being dewatered by a screw press. The rest of the WAS (approximately 50%) is delivered to external contractors. The amount of WAS added to the PS is 12–13 m/m%; this must be limited so as to achieve the target solids content of the dewatered sludge cake (50%) and its caloric value of about 3.2 MJ/kg. To obtain a suitable mixture for energy use, the CMTP sludge is mixed with the deinked pulp (DIP) sludge, which has higher caloric value (about 3.7 MJ/kg). Sludge energy recovery takes place in a grate-fired incinerator with a rated thermal input of 10.5 MW. The incinerator is designed for self-supporting combustion with a minimum fuel caloric value of 3.5 MJ/kg. Natural gas is used for startup and as a support fuel. The paper mill WWTP schematic and solids balances are shown in Fig. 1. Sludge data are indicated in Tables I and II. Table II provides data on CMTP sludge and the sludge mix-

1. Wastewater treatment plant and solids balance of the investigated paper mill. (For a list of nomenclature/acronyms, see Appendix I.) 128


WASTEWATER TREATMENT Dry Matter, %

Loss of Ignition, %d.m

Net Caloric Value, MJ kg-1d.m.

DIP PS — AT influent (experiment 1, sample point 1 in Fig. 1)

53.5±3.1

35.5±0.6

7.6±0.2

DIP PS — AT effluent (experiment 1, sample point 4 in Fig. 1)

16.4±8.3

45.3±3.1

7.1±0.54

WAS

17.2±1.2

70.3±5.5

13.9±1.3

HWAS

10.5±1.5

46.8±2.5

9.7±0.4

27.9

44.7

10.3

DIP PS*

67.0±2.8

36.7±4.1

6.8±0.7

CMTP sludge*

50.0±3.3

48.9±5.1

8.7±0.8

Sludge Sample

Anaerobic biomass

For a list of nomenclature/acronyms, see Appendix I. *Mill data of the existing DIP and chemical-mechanical treatment plant (CMTP) sludge, year average of daily samples.

I. Analyses of waste activated sludge/hydrolyzed waste activated sludge (WAS/HWAS), anaerobic biomass from upflow anaerobic sludge blanket (UASB) reactor, influent/effluent primary sludge (PS) from anaerobic treatment (AT) of deinked pulp (DIP) wastewater (WW) (experiment 1), and analyses of existing mill sludge.

Unit

Aerobic WWTP

AT DIP WW***

AT CMTP Effluent

AT CMTP Effluent + 18 v/v% MWW

td.m./year

32028

32028

32028

32028

td.m./year

6673 (7501)**

6723

6559

6371

GJ/t

3.21 (1.34)**

3.12

3.21

3.76

%

50.0 (32.0)**

49.0

51.0

56.0

WAS quantity

td.m./year

1601

727

559

360

Anaerobic biomass quantity

td.m./year

0

97

100

112

Total BS reduction

%

0

49

59

71

Mass ratio of BS in CMTP sludge

%

11.6 (21.3)**

12.3

10.1

7.4

t/year

4482 (0)**

0

0

0

t/year

61091 (71228)**

61510

60649

59166

GJ/t

3.61 (2.94)**

3.58

3.61

3.73

- Content of dry matter

%

63.3 (55.5)**

63.0

63.6

64.9

- Sludge thermal energy

TJ

220.5 (209.1)**

220.5

219.0

220.5

- Sludge energy versus mill total fuel energy consumption

%

12.6 (12.0)**

12.6

12.5

12.6

Sludge Properties and Quantity DIP PS quantity CMTP sludge - Quantity† - Caloric value - Content of dry matter

WAS quantity for external contractors* Sludge mixture (DIP PS + CMTP sludge) - Total quantity for incineration - Caloric value

For a list of nomenclature/acronyms, see Appendix I; d.m. = dry matter. *WAS content of dry matter in the calculations is 18.5%. **Sludge features, when the entire quantity of WAS is dewatered in mixture with the CMTP PS. ***DIP WW with total suspended solids (TSS) concentration below 120 mg/L. †Quantity of PS is 5900 td.m./year in all cases.

II. Quantity and properties of sludge at paper mill with aerobic and anaerobic wastewater treatment installation. FEBRUARY 2016 | VOL. 15 NO. 2 | TAPPI JOURNAL

129

WASTEWATER TREATMENT ture characteristics that would result if all WAS were dewatered with the CMTP sludge. The caloric value of this sludge mixture would fall below 3.0 MJ/kg, which is insufficient for incineration at the mill. Effluents and WAS for the studies were collected regularly and stored for not more than 2 days in cold storage (T=3±2°C).

WAS pretreatment with hydrolysis Waste activated sludge from the paper mill ASTP was hydrolyzed in a 1.5-L laboratory beaker with 5 M sodium hydroxide (NaOH), similar to Kaluža et al. [13]. The pH of the sludge was increased to 12.0 by adding 5±1 mL NaOH/L of WAS, and then the sludge was heated at 70±3.0°C for 5 h while maintaining continuous stirring. The pH decreased to 9.0 by the end of hydrolysis. The liquid portion of the hydrolyzed WAS (HWAS) was applied as a supplemental feed in the anaerobic WW treatment test. Analyses on WAS/HWAS (Table III) were performed on average daily samples taken from each batch of hydrolyses, conducted during the anaerobic WW treatment test and stored in a freezer. The solid part of WAS/HWAS was analyzed on two average samples after centrifugation (Table I).

Anaerobic treatment of wastewaters The anaerobic treatment of WW was performed with a pilot 12-L upflow anaerobic sludge blanket (UASB) reactor at a temperature of 30°C–35°C and pH of 6.5–7.5. As an inoculum, 6 L of granulated sludge from brewery effluent treatment with total suspended solids (TSS) concentration of 51.97 g/L was used. The experimental setup is shown in Fig. 2, and the UASB reactor design in Fig. 3. The quantity of the generated biogas was measured continuously by a wet tip gas meter at ambient conditions (20°C–25°C, 1 bar). The biogas composition (methane [CH4], carbon dioxide [CO2], and hydrogen sulfide [H2S]) was determined on 12 representative biogas samples, using Agilent 7890 A GC System gas chromatography (Agilent Technologies; Santa Clara, CA, USA) equipped with two thermal conductivity detectors and high-purity helium as a gas carrier. The composition of the gas is expressed in normalized volume percentage. Two different experiments were performed: (1) experi-

Raw WAS

Hydrolyzed WAS

Hydrolysis Efficiency, %

TSS, g/L

10.76

9.45

12.2

VSS, g/L

8.08

6.09

24.6

CODf*, mg O2/L

159

5,761

/

VSS/TSS, g/g

75.1

64.5

14.1

Parameter

For a list of nomenclature/acronyms, see Appendix I. *Soluble COD, determined in filtrate.

III. Efficiency of waste activated sludge alkali hydrolysis. 130


ment 1 for 67 days with WW from DIP (DIP WW) and (2) experiment 2 for 127 days with effluent from CMTP with various additives. For pH adjustment 10% NaOH or 30% hydrochloric acid (HCl) was used. As nutrients, 7.9% urea and 85% phosphoric acid were applied. Daily we sampled AT influent/effluent WW and influent/effluent DIP PS (Fig. 2, sampling points 2,3 and 1,4). COD, pH, and TSS were analyzed in daily WW samples, while BOD5, ammonium (N-NH4), and total phosphorus (TP) were determined in representative average 2-day samples, once per week. The DIP PS samples were centrifuged and frozen; average multiday samples were analyzed. Volatile fatty acids and alkalinity were not determined. The anaerobic biomass was analyzed at the end of experiment. Its growth was calculated from biomass concentration and volume at the beginning and at the end of the test, and from the amount of COD decomposed during the whole test. The process was always stabilized for a minimum 10 hydraulic retention times (HRT) at a low organic loading rate (OLR) 3–5 kg/m3/day before performance was assessed.

Analytical methods TSS and volatile suspended solids (VSS) were determined according to DIN 38409-2:1987 “German standard methods for the examination of water, wastewater and sludge; parameters characterizing effects and substances (group H); determination of filterable matter and the residue on ignition (H2).” COD was determined by the SIST ISO 6060:1996 procedure “Water quality—determination of the chemical oxygen demand.” Biochemical oxygen demand (BOD) was measured as in SIST EN 1899-1:2000 “Water quality—determination of biochemical oxygen demand after n days (BODn)–part 1: dilution and seeding method with allylthiourea addition (ISO 5815: 1989 modified)” by using the WTW OxiTop measuring system (Xylem Analytics; Weilheim, Germany). The soluble COD fraction was determined after filtration of a sample through a black ribbon MN 640 w REF 202 011 (Machery-Nagel GmBH; Duren, Germany). We calculated reductions of TSS, VSS, COD, or BOD5 as the difference between parameters in the influent and effluent of the particular unit. Ammonium (NH4 -N) was determined by cuvette tests using a DR 2800 digital spectrophotometer (Hach Company; Loveland, CO, USA). To determine total phosphorus (TP), samples were digested in a Hach HT 200 S thermostat. Temperature and pH were measured with a Hach Multi HQ40d pH meter with an Intelli CAL™ pH PHC 101 probe. Analyses of sludge (Table I) were performed after the centrifugation of sludge samples at 6,000 min for 10 min. The content of sludge dry matter (d.m.) was defined according to SIST EN 14346:2007 “Characterization of waste—calculation of dry matter by determination of dry residue or water content,” the caloric value according to SIST EN 15400:2011 “Solid recovered fuels—determination of calorific value,” and the loss on ignition according to SIST EN 15169:2007 “Characterization of waste—determination of loss on ignition in waste, sludge and sediments.”


2. Experimental setup for anaerobic treatment of wastewater with sampling points 1–5.

Volatile organic compounds in the WAS were analyzed by gas chromatography/mass spectrometry (GC/MS) performed according to the modified method EPA 625 “Semivolatile organic compounds by isotope dilution GC/MS.” RESULTS AND DISCUSSION

Alkali pretreatment of WAS

3. Design of pilot UASB reactor for anaerobic treatment of wastewater.

Alkali hydrolysis is a commonly investigated pretreatment method, adopted in many commercialized processes, and has often been used in combination with other types of pretreatment [8,15]. Pretreatment is particularly important in the digestion of sludge, since microbial cells are a relatively unfavorable substrate for degradation. Different pretreatment methods (mechanical, thermal, chemical, or biological) are commonly used for cell wall disruption and degradation of particulate organic material to lower-molecular-weight compounds that are decomposed in subsequent biological treatment [9,14]. In our study, NaOH pretreatment at 70°C was introduced to achieve a reduction of WAS solids and to determine the influence of the liquid portion of HWAS on anaerobic biodegradability of the mill WW in a pilot UASB reactor. The efficiency of WAS alkali hydrolysis was measured by the reduction of TSS and VSS concentration and the increase FEBRUARY 2016 | VOL. 15 NO. 2 | TAPPI JOURNAL

131

WASTEWATER TREATMENT of soluble COD (Table III). After hydrolysis, a significant increase of WAS-soluble COD was measured. In addition, TSS and VSS were reduced by 12.2% and 24.6%, respectively. Alkali hydrolysis also contributed to the reduction of the HWAS solid-portion caloric value and its dewatering properties (Table I). The results obtained were significantly worse than in the work of Kaluža et al. [13], who reported more efficient solids reduction (over 80%) of WAS from a mill producing cardboard. The results of previous studies of alkaline pretreatment are conflicting, and there is no correlation between the digestibility of the WAS, pretreatment efficiency, and the type of mill where the sludge is generated [8]. Some researchers found alkaline pretreatment to be very effective [15,17] in increasing the anaerobic sludge degradation and biomethane potential (BMP) in comparison to untreated sludge. However, others [5,16] measured a BMP decrease. The potential presence of substances that may inhibit an-

Parameter/Samples Influent (sampling point 2, Fig. 1)

Unit

aerobic digestion of the liquid portion of the HWAS was explored by performing GC/MS analyses of WAS. We identified vinyl crotonate (C6H8O2) in the solid and liquid phases of WAS. This compound is a vinyl ester of crotonic acid, commonly present in newspaper ink. There are no reference data about inhibitory effects of that compound and additional research is needed to determine its potential inhibition properties.

Anaerobic treatment of WW in a pilot UASB reactor Anaerobic WW treatment (AT) was tested in a pilot UASB reactor (Figs. 2 and 3) using DIP WW (day 1–67, experiment 1). The average values of the entire experiment are shown in Table IV, daily values in Fig. 4, and sludge analyses in Table I. Because of the high TSS concentration of DIP WW (0.5–3.0 g/L), solids were partially removed by a separation sieve to meet a TSS concentration of 0.56 g/L. The eliminated PS from DIP WW has similar properties as DIP PS or CMTP

DIP WW

CMTP Effluent

CMTP Effluent+ 1% HWAS

CMTP Effluent+ Nutrients

CMTP Effluent+ 18% MWW

Days 1–67

Days 68–84

Days 85–132

Days 146–181

Days 182–194

TSS

g/L

0.56±0.17