Dissolving pulp: Multivariate Characterisation and Analysis of Reactivity and Spectroscopic Properties

Dissolving Pulp – Multivariate Characterisation and Analysis of Reactivity and Spectroscopic Properties

by

Kristina Elg Christoffersson

Akademisk avhandling

Som med tillstånd av rektorsämbetet vid Umeå universitet för erhållande av filosofie doktorsexamen vid Teknisk-Naturvetenskapliga fakulteten, fram- lägges till offentlig granskning vid Kemiska institutionen, Umeå universitet, sal KB3A9, KBC, fredagen den 28 januari 2005, kl 10.00.

Fakultetsopponent: Docent Monica Ek, Avdelningen för träkemi och mas-

sateknologi, Fiber- och polymerteknologi, Kungliga tekniska högskolan,

Stockholm

Copyright © 2005 Kristina Elg Christoffersson ISBN: 91-7305-798-3

Printed in Sweden by VMC-KBC

Umeå University, Umeå 2005

Title

Dissolving pulp – Multivariate characterisation and analysis of reactivity and spectroscopic properties

Author

Kristina Elg Christoffersson, Department of Chemistry, Organic Chemistry, Umeå University, SE-901 87, Umeå, Sweden

Abstract

Various chemical properties can be used to characterise dissolving pulp. The quality of the pulp must be carefully controlled to ensure that it meets the requirements for its intended use and the further processes to be applied. If it is to be used to prepare viscose, or other cellulose derivatives, the key prop- erties of the pulp are its accessibility and reactivity. The studies described in this thesis investigated the potential utility of multivariate analysis of chemi- cal and spectral data for determining the properties of dissolving pulp. Dis- solving pulps produced by a two-stage sulfite process, both in the laboratory and a factory were produced pulps for this purpose. The analyses showed that pulp with high reactivity had short cellulose chains, low molecular weight, low polydispersity, low hemicellulose content, high content of ace- tone-extractable compounds, and high surface charge compared to pulp with low reactivity. Important chemical properties of the pulp, such as viscosity and alkali resistance, were successfully predicted from near infrared spectra.

Predicting the reactivity, or the viscose filterability, of the pulp was more complex. Several chemical methods for analyzing the reactivity of the pulp were examined. The influence of the cellulose structure at the supermolecu- lar level on the reactivity of the pulp was explored by multivariate analysis of solid state ¹³ C nuclear magnetic resonance spectra. Structural variables considered included: differences in hydrogen bonding, contents of hemicel- lulose, amorphous cellulose and crystalline cellulose I and II. Pulps with high reactivity have higher contents of cellulose I and amorphous cellulose than pulps with low reactivity, which have higher contents of cellulose II and hemicellulose.

Key words

Dissolving pulp, reactivity, solid state ¹³ C NMR spectroscopy, NIR spectros- copy, multivariate analysis, cellulose, viscose

ISBN: 91-7305-798-3

Contents

Contents ...iv

List of Papers ...vi

Abbreviations... vii

1. Introduction ...9

1.1. Aims ...9

2. Cellulose and dissolving pulp...10

2.1. Cellulose...10

2.2. Reactivity viewed at the molecular and supermolecular levels..11

2.3. Dissolving pulp...14

2.4. Viscose ...15

2.5. Chemical analyses of dissolving pulp ...16

3. Spectroscopy...19

3.1. NMR spectroscopy ...19

3.1.1. Processing NMR spectra...21

3.2. NIR spectroscopy ...21

4. Multivariate analysis...23

4.1. PCA ...23

4.2. PLS ...24

4.3. Pre-treatment of data ...24

4.3.1. Alignment ...25

4.3.2. Normalization ...25

4.3.3. Scaling ...25

4.4. Validation ...26

5. Experimental design in producing and selecting samples ...29

6. Summary and discussion of Papers I-IV...31

6.1. Process variables correlated to chemical properties of laboratory cooked pulp samples...31

6.2. Prediction of properties by NIR spectroscopy...33

6.2.1. Models developed from laboratory-produced pulps ...33

6.2.2. Models developed from factory-produced pulps ...34

6.3. Reactivity - characterisation using chemical properties ...37

6.3.1. Characterisation using chemical properties ...37

6.3.2. Reactivity using Fock’s method ...40

6.4. Quality and reactivity variations – explanations according to differences in cellulose supermolecular structure studied by solid state NMR...41

6.5. Combining NMR and NIR spectral data – interpretation of subspectra ...44

7. Concluding remarks and future perspectives...47

8. Summary in Swedish ( Sammanfattning)...49

9. References ...50

10. Acknowledgements...59

List of Papers

This thesis is based on the papers listed below, which are referred to in the text by the corresponding Roman numerals (I-IV).

I Kristina Elg Christoffersson, Jon Hauksson, Ulf Edlund, Mi- chael Sjöström, Matti Dolk, Characterisation of dissolving pulp using designed process variables, NIR and NMR spec- troscopy, and multivariate analysis, Cellulose, 6:233-249, 1999

II Matti Dolk and Kristina Elg, Characteristics of dissolving pulp revealed by NIR reflectance spectroscopy, 3 ^rd International Symposium Alternative Cellulose - Manufacturing, Forming, Properties, 2-3/9 1998 Rudolstadt, Germany, 1998

III Kristina Elg Christoffersson, Michael Sjöström, Ulf Edlund, Åsa Lindgren and Matti Dolk, Reactivity of dissolving pulp:

characterisation using chemical properties, NMR spectros- copy and multivariate data analysis, Cellulose, 9 (2):159-170, 2002

IV Kristina Elg Christoffersson, Michael Sjöström, Ulf Edlund and Roland Agnemo, Chemical properties and cellulose su- permolecular structure of dissolving pulps with different reac- tivity revealed by solid state NMR, NIR spectra and multivari- ate data analysis. Manuscript

Papers I and III are reproduced with the kind permission of Springer

Science and Business Media, and Paper II with permission from Thüring-

isches Institut für Textil- und Kunststoff-Forschung e.V, TITK.

Abbreviations

GPC gel permeation chromatography

LV latent variable

MVA multivariate analysis

NMMO N-methyl morpholine-N-oxid

NMR nuclear magnetic resonance

NIR near infrared

MSC multiplicative scatter correction

PC principal component

PCA principal component analysis

PLS partial least squares

RMSEP root mean square error of prediction

8

1. Introduction

Dissolving pulp can be used to manufacture various cellulose-based prod- ucts. When producing dissolving pulp for making products such as carboxy methyl cellulose, viscose, cellulose film and sausage skin, determining the pulp quality is essential. The dissolving pulp quality depends both on prop- erties of the raw wood material and the pulp processing. Clearly, for any investigation of the effects of different wood and/or pulp processing factors on pulp quality, identifying the most influential variables is of considerable importance. The methods used to analyze the properties must have good precision and repeatability. In most cases the further processes demand deri- vatization and/or dissolution of the cellulose. Key factors then are the acces- sibility of the cellulose with respect to the chemicals and its reactivity. Qual- ity therefore often means reactivity. Numerous different chemical properties can be used to measure the quality of the pulp. A good tool for identifying the most important factors, and overviewing them in a convenient way, is multivariate analysis. The reactivity of the pulp depends to a great extent on the supermolecular structure of the cellulose, which can be studied spectro- scopically.

1.1. Aims

What characteristics should dissolving pulp have to be suitable for vis- cose production? What is the best way to measure and study the properties of the pulp? To be able to answer the first question one must first answer the second. The principal aim of the work underlying this thesis was to deter- mine the utility of multivariate analyses of chemical and spectroscopic data obtained from dissolving pulp for exploring the reactivity of the pulp. The focus was on dissolving pulp for viscose manufacture. A futher aim was to determine the supermolecular structure of cellulose in pulps of different quality and reactivity.

What you cannot measure, you cannot control;

what you cannot control, you cannot improve.

D.Beecroft (1992)

9

2. Cellulose and dissolving pulp

2.1. Cellulose

Cellulose is the most abundant compound in the cell wall of plants and woods. Payen was the first to determine the elemental composition of cellu- lose, reporting in 1838 that it had the empirical formula C 6 H 10 O 5 (Payen 1842; Krässig 1996). It was also found that during hydrolysis it yields cello- biose and finally glucose. The nature of the bonds between the atoms in the glucose units and between the glucose units were recognized by Haworth, but it was Staudinger who reported a proof of the polymeric nature of the cellulose molecule (Harworth 1928; Staudinger 1932). The chain–like, lin- ear, macromolecular structure proposed by Haworth in the 1920s is today generally accepted, Figure 1. The glucose units are linked together by β-1, 4-glucosidic bonds between carbon C(1) and C(4) of adjacent glucose units.

Cellulose is actually a β-1, 4-polyacetal of cellobiose. The terminal hydroxyl groups at each end of the polymer chain differ in chemical nature. The C(1) hydroxyl is an aldehyde hydrate group with reducing activity and originates from the formation of the pyranose ring through an intramolecular he- miacetal reaction. The hydroxyl on carbon C(4) is an alcoholic hydroxyl and therefore non-reducing. The β-glucosidic bonds are sensitive to hydrolytic attack. At the supermolecular level the cellulose chains are held together by hydrogen bonds. The resulting cellulose aggregates can be ordered as crys- talline cellulose or unordered, amorphous cellulose. Several polymorphs of crystalline cellulose are known; cellulose I, II, III, and IV.

1

4

O

O OH H

OH

4

1

O O

1

4

O

O

4

O

1

O

H O

OH OH

OH O

H

O H

O H

OH OH

OH

n

OH

Figure 1. Cellulose is a β-1, 4-polyacetal of cellobiose.

10

2.2. Reactivity viewed at the molecular and supermolecular levels

The reactivity of cellulose can refer to its capacity to participate in diverse chemical reactions. Each glucose unit in a cellulose polymer has three dif- ferent hydroxyl groups. The two secondary hydroxyl groups on carbons two and three are more reactive than the primary hydroxyl group on carbon six (Krässig 1996). For derivatization reactions it is important to note that reac- tions with the hydroxyl groups on carbons two and three are kinetically fa- vourable, while substitution on carbon six is thermodynamically more stable (Schlotter 1988; Krässig 1996). Examples of the most important cellulose derivates are ethers and esters, such as carboxymethylcellulose, cellulose acetate and cellulose xanthate. In the reactions involved in making these derivatives the reactivity of the cellulose is an essential factor.

When discussing reactivity the accessibility of the cellulose for the chemicals to be used is a crucial factor. The hydrogen bonds between the cellulose polymers and between the microfibrils play an important role in the accessibility. When studying the hydrogen network in cellulose I and II it is obvious that the accessibility is higher in cellulose I than II. Cellulose I has intramolecular hydrogen bonding between the ring oxygen O(5) and the hydroxyl group on C(3) and between the hydroxyl group on C(2) and the primary hydroxyl group on C(6). Intermolecular hydrogen bonding occurs between the primary hydroxyl group on C(6) and the hydroxyl group on C(3’) in parallel chains (Blackwell et al. 1977), Figure 2. Like cellulose I, cellulose II has intramolecular hydrogen bonding between ring oxygen O(5) and the hydroxyl group on C(3). In the 020-plane for center down chains it also forms intramolecular hydrogen bonds between the hydroxyl group on C(2) and the primary hydroxyl group on C(6) and intermolecular hydrogen bonds between the primary hydroxyl group on C(6) and the hydroxyl group on C(3’) in parallel chains in a similar way to cellulose I, Figure 3a. In the 020-plane for center up chains no hydrogen bonds are formed between the hydroxyl group on C(2) and the primary hydroxyl group on C(6), Figure 3b.

Intermolecular hydrogen bonds between C(2) and C(2’) are instead formed in the 110-plane between antiparallel chains, Figure 3c. In cellulose II the sheets are therefore bonded together with hydrogen bonds while the sheets in cellulose I are held together by van der Waals interactions. The average bonding length in cellulose II is 2.72 Å and in cellulose I 2.80 Å. Thus there is a tighter chain packing in cellulose II compared to cellulose I.

Although cellulose I is the polymorph that plants originally produce dur- ing wood formation, both cellulose I and II have been found in pulp. When comparing different pulps, dissolving pulps have been shown to contain more cellulose II than most other pulps (Lennholm and Iversen 1995; Lenn- holm and Iversen 1995). As cellulose II is more thermodynamically stable

11

than cellulose I this may make the dissolving pulps with large proportions of cellulose II more resistant to heating than pulps with large proportions of cellulose I.

Reactivity is greatly affected by hornification. This phenomenon, which reduces reactivity, has been shown to appear when pulp is dried. Hornifica- tion mechanisms are a continuous matter of debate, and several different models have been proposed to explain it (Weise 1998; Fernandes Diniz 2004; Newman 2004).

1 3 2

4

5

O

O O

O

4

3 2

1 5

O O

O O

6

O

6

O

1

1 3 2

4

5

O

O O

O

4

3 2

1 5

O O

O O

6

O

6

O

3 2

4

5

O

O O

O

4

3 2

1 5

O O

O O

6

O

6

O

3 2

4

5

O

O O

O

4

3 2

1 5

O O

O O

6

O

6

O

' ' ' '

' '

' ' ' '

'

'

' ' ' '

' '

' ' ' '

'

'

Figure 2. Cellulose I has intramolecular hydrogen bonding between ring oxygen O(5) and the hydroxyl group on C(3) as well as between the hydroxyl group on C(2) and the primary hydroxyl group on C(6). Intermolecular hydrogen bonding is shown between the primary hydroxyl group on C(6) and the hydroxyl group on C(3’) in parallel chains (Blackwell 1977) The dotted lines indicate to hydrogen bonds.

12

1 3 2 4

5

O

O O

O

6 4

O

3 2

1 5

O O

O O

6

O

1 3 2 4

5

O

O O

O

4

3 2

1 5

O O

O O O

6

O

6

1 3 2 4

5

O

O O

O

6

O

4

3 2

1 5

O O

O O

6

O

1 3 2 4

5

O

O O

O

4

3 2

1 5

O O

O O O

6

O

6

' ' ' '

' ' ' '

' ' '

' ' '

' '

' ' ' ' ' ' '

3a

1 3 2 4

5

O

O O

6

O

4

O

3 2

1 5

O O

O O

6

O

1 3 2 4

5

O

O O

O

4

3 2

1 5

O O

O O

6

O

6

O

1 3 2 4

5

O

O O

6

O

4

O

3 2

1 5

O O

O O

6

O

1 3 2 4

5

O

O O

O

4

3 2

1 5

O O

O O

6

O

6

O '

' ' '

' ' ' '

' ' '

' ' '

' '

' ' ' ' ' ' '

3b

13

1

2 3

4

O

5

O O

6

O

O

1 3 2 4

5

O

O O

O

6

O

4 3 2 1

O

⁵

O

O O

6

O

1

2 3

4

O

5

O O

O

4 3 2 1

O

⁵

O O

6

O

6

O

O

4

3 2

1 5

O O

O O

6

O

1 3 2 4

5

O

O O

O

4

3 2

1 5

O O

O O O

6

O

6

' ' '

' ' ' '

' ' ' '

'

'

' ' '

' ' ' '

' ' ' '

3c

Figure 3. Like cellulose I, cellulose II has intramolecular hydrogen bonding be- tween ring oxygen O(5) and the hydroxyl group on C(3). In the 020-plane for center down chains, it also forms intramolecular hydrogen bonds between the hydroxyl group on C(2) and the primary hydroxyl group on C(6) and intermolecular hydrogen bonds between the primary hydroxyl group on C(6) and the hydroxyl group on C(3’) in parallel chains in a similar way as cellulose I (Figure 3a). In the 020-plane for center up chains no hydrogen bonds are formed between the hydroxyl group on C(2) and the primary hydroxyl group on C(6) (Figure 3b). Intermolecular hydrogen bonds between C(2) and C(2’) are formed instead in the 110-plane between antipar- allel chains (Figure3c), (Blackwell 1977). The dotted lines indicate hydrogen bonds.

2.3. Dissolving pulp

Dissolving pulp is a high-grade cellulose pulp, with low contents of hemicellulose, lignin and resin. This very clean pulp is well suited as a raw material for different kinds of cellulose products, such as staple fibres, films and derivatives (e.g. carboxymethylcellulose, cellulose acetate and various types of regenerated cellulose such as viscose and microcrystalline cellu- lose).

Production of dissolving pulp is today mainly done by acid sulfite and

prehydrolysis kraft processes. The acid sulfite process is the most common

and benefits of this technique include high recovery rates of the inorganic

cooking chemicals and the totally chlorine free (TCF) bleaching. One disad-

vantage is is that it results in pulps with a broad molecular weight distribu-

tion of cellulose (Sixta et al. 2004). Organosolve processes have been sug-

14

gested as alternative pulping methods. The original one only usined a mix- ture of water and ethanol (Kleinert and Salze 1977). New acid pulping proc- esses such as Acetosolve, Formacell and Milox have also shown promising results (Kleinert et al. 1977; Puls et al. 1999; Vila et al. 2004). Comparisons of the pulps produced by these different methods have recently been pre- sented (Fink et al. 2004; Sixta et al. 2004). (Fink 2004; Sixta 2004). In the cited studies the processability of the pulps regarding viscose staple fibre preparation and NMMO treatment were tested. The pulps were characterized in terms of their chemical properties and structure by microscopy, NMR spectroscopy and X-ray scattering. Another area of great interest is the po- tential of biosulfite pulping in dissolving pulp production (Christov et al.

1998; Viikari 2003).

In the work underlying this thesis most of the dissolving pulp samples ex- amined were produced by a two-stage sulfite cooking process using sodium (Na) as base, either in the laboratory (in Paper I) or factory (in Papers II- IV). One Lyocell (in Paper III) and some electron beamed pulp samples from different sulfite pulp mills (in Papers III and IV) were also used.

2.4. Viscose

The manufacture of rayon was first reported by the American Viscose Company, formed by Samuel Courtaulds and Co., Ltd., in 1910 (www.fibersource.com). It was the first man-made fibre and from the begin- ning it was produced by two methods, yielding viscose rayon and cuppram- monium rayon, respectively, using two different chemicals. The viscose process became the common one and the focus of research and development during the twentieth century (Treiber 1987; Schlotter 1988). In the Reaction scheme 1 the reactions involved in the viscose process are shown. Early work on direct dissolution was published by Turbak (Turbak 1977). The cellulose-dissolving potential of amine oxides was reported for the first time in 1939 (Graenacher and Sallman 1939). However, it was not until 1969 that Johnson described the use of cyclic mono(N-methylamine-N-oxide) (NMMO) compounds as solvent-sizes for strengthening paper by partially dissolving the cellulose fibres (Johnson 1969), ushering in a new modle of cellulose fibre production. At present the simple and environmentally- friendly Lyocell NMMO-technique is making an industrial breakthrough (Bochek 1995; Marini 1996; Kosolowski 1997).

15

O OH

O O

OH

OH n

O O

O O

OH n

OH Na

⁺

O O

O O

OH n

OH Na

⁺

O O

O O

OH

OH

* S S

Na

⁺

O n OH

O O

OH

OH n *

NaOH O

₂

(Co)

CS

₂

Xanthation

Dissolving Ripening

Mercerization Pre-ripening

-CS

₂

n=780 n=270

Regeneration H

⁺

Dissolving pulp

Viscose fibers

Reaction scheme 1. The chemical reactions involved in the viscose process. The first step, mercerisation, is the reaction of the cellulose with sodium hydroxide, forming alkali cellulose. In step two, pre-ripening, oxygen in the air participates in depoly- merization reactions which are catalysed by cobalt and reduce the DP-value. The xanthation step generates alkali cellulose xanthate, which is soluble in alkali. This is followed by a ripening step where some of the alkali xanthate groups are removed from hydroxyl positions two and three, or relocated to position six. The cellulose xanthate is then pumped through spinnerets, which have hundreds of tiny holes, into an acid bath where cellulose is regenerated in the form of long filaments. Trithiocar- bonate, one of the by-products formed in the xanthation step gives a yellow colour to the viscose (Schlotter 1988).

2.5. Chemical analyses of dissolving pulp

Numerous chemical properties of dissolving pulp can be measured. In the

work underlying this thesis we used the properties listed in Table 1. Tradi-

tion, the first property sought when producing dissolving pulp is the viscos-

ity. Viscosity is related to the length of the cellulose chains and can be

measured in different solvents (Krässig 1996). Molecular weight distribution

and polydispersity, measured by gel permeation chromatography (GPC),

viscosity measurements or other methods, are important parameters for

characterizing the pulp. The degree of polymerisation (DP) is related to the

molecular weight (M) by the formula DP=M/162, where 162 is the molecu-

lar weight of the anhydroglucose unit. The molecular weight can be ex-

pressed in various ways, for example the number average molecular weight

(DPn), or the weight average molecular weight (DPw), depending on the

method used. The polydispersity index, the ratio DPn/DPw, corresponds to

16

the width of the molecular weight distribution. Alkali resistance indicates the fraction of the pulp that is insoluble in sodium hydroxide at different sodium hydroxide concentrations. In 18% NaOH (R18) the hemicellulose is soluble while in 10% NaOH (R10) both the low molecular weight cellulose and the hemicellulose are dissolved. However, this is not always true, especially for pulps with low DP both the low molecular weight cellulose and the hemicel- lulose are dissolved in 18% NaOH.

Acetone extractives (%) are compounds extracted by acetone. The content of extractives should be as low as possible, although they have positive ef- fects on the viscose process. However, it is better to have low, constant lev- els than uncontrolled, varying levels in order to maintain constant viscose process parameters. The effect of surface active additives (which, like natu- rally occurring resins, influence the pulp reactivity) is interesting to measure.

The content of silicon (SiO ₂ ) has a strong influence on the filterability of the viscose (Samuelson 1948). Metal ions, e.g. magnesium, iron and manganese ions, can interfere with the pulp and viscose processes. It is therefore impor- tant to control the contents of these metal ions. Surface fibre charge and fibre charge are interesting parameters to measure. The ionic groups (e.g.

carboxylgroups bonded to the xylan hemicellulose) influence water sorption, and thus the softening of the pulp. Dissolving pulp has low carboxyl con- tents because the hemicellulose content is low. High hemicellulose contents are bad because the hemicellulose will react with the carbon disulfide and when recovered together with the cellulose it defects the quality of the vis- cose fibres. Density, Kappa, Brightness, Dirt count and Yield are all basic pulp parameters. The viscose filterability was measured in terms of the filter clogging value (Kw) and filter value (kr24, kr30). The viscose filter value is the most commonly used measure of the reactivity of dissolving pulp. Two parameters are used to describe this property, K _w or k _r , the latter being the viscosity-corrected K w value. The method for preparing a test viscose and measuring its filterability are described in (Treiber 1987; Treiber 1987).

A simple way of measuring the reactivity of the dissolving pulp was pre- sented by Fock (Fock 1959). In this method sodium hydroxide and carbon disulfide are stirred together with the pulp sample, forming a solution of cellulose xanthate which is then collected. The cellulose is regenerated as viscose fibres and the yield of cellulose is determined. The residual cellulose can then be calculated. The reactivity according to Fock’s method is some- times expressed as residual cellulose, as in Paper II, or as the cellulose yield, as in Paper IV. In several cases described in the literature, this prop- erty was used to describe the reactivity of dissolving pulp samples produced from different raw materials and by various different pulp processing meth- ods (Roffael 1988; Aboustate et al. 1990).

17

Table 1. Properties used to describe dissolving pulp.

Property Unit

Viscosity ml/g

R18, alkali resistance in 18% sodium hydroxide %

R10, alkali resistance in 10% sodium hydroxide %

R5, alkali resistance in 5% sodium hydroxide %

Acetone extractives (acetone-soluble compounds in pulp) %

Magnesium (Mg) content mg/kg

Content of silicon (Si) expressed as SiO

2

mg/kg

Density kg/m

³

Carboxyl groups mmol/kg

DPw or GPCMw, weight average degree of polymerisation DPn or GPCMn, number average degree of polymerisation Polydispersity index=DPw/DPn

Yield from acid hydrolyses (acidhyd) %

Ash %

Dirt count mm /kg

²

Suface fibre charge µekv/g

Fibre charge mmol/kg

Kappa number

Brightness % ISO

Yield %

Surface active additive (V-388) kg/ton

Kw, viscose filter value

kr24, viscose filter value, 24% dose of CS in viscose preparation, kr is

viscosity corrected Kw

²

kr30, viscose filter values, 30% dose of CS in viscose preparation, kr is

viscosity corrected Kw

²

Residual cellulose or Cellulose yield %

18

3. Spectroscopy

In the previous chapter it was stated that the properties of the dissolving pulp depend to a large degree on the supermolecular structure of the cellu- lose. It is therefore important to be able to study the cellulose molecules spectroscopically. Several spectroscopic techniques can be used, all of which measure the interaction of molecules with electromagnetic radiation at dif- ferent frequencies or wavelengths. In the work presented in this thesis solid state nuclear magnetic resonance (NMR, 10 ⁷ Hz or 10 m) and near infrared (NIR, 10 ¹³ Hz or 10 ^-5 m) spectroscopy were used and therefore these meth- ods are discussed in this chapter.

3.1. NMR spectroscopy

Nuclear magnetic resonance (NMR) spectroscopy uses the fact that vari- ous magnetic nuclei have different magnetic properties. By recording the differences in the magnetic properties it is possible to estimate the positions of the nuclei in the molecule. It is also possible to determine how many dif- ferent environments there are in the molecule. The first studies on the phe- nomena of NMR were done on the simplest nucleus, the proton, and re- ported independently by Bloch and Purcell in 1946 (Bloch et al. 1946; Pur- cell et al. 1946). The first commercially available spectrometer was intro- duced in 1958 and thus initiated the practical use of the technique. The theory of NMR and its application to organic compounds have been exten- sively described in the literature (Günther 1992; Friebolin 1993).

NMR spectroscopy has shown potential for studying solid state polymers such as cellulose. The introduction of cross-polarisation magic angle spin- ning (CP/MAS) opened up new possibilities (Schaefer 1977; Yannoni 1982).

Proton-carbon cross polarizations enhance the sensitivity and the magic an- gle spinning improves the resolution, resulting in well-resolved ¹³ C spectra.

In solution NMR narrow linewidths are generated by the random tumbling of the molecules, resulting in an averaging of the anisotropic spin interac- tions from proton dipolar broadening and carbon chemical shift anisotropy.

In solid states the molecules are more or less fixed, so the averaging of the anisotropic interactions is incomplete, leading to broad linewidths. The sen-

19

sitivity is reduced by the large linewiths and also by the low natural abun- dance and long relaxation times of ¹³ C nuclear spins.

The important distinction between solid-state spectra and liquid-state spectra is, as mentioned above, that molecular motions average so quickly in liquids that all carbons in the same position give identical signals. In solid- state the molecules are fixed so the isotropic chemical shift is sensitive to the environment. This knowledge can be exploited to detect differences in solid state environments in the form of chemical shift differences for a given car- bon site (Vanderhart and Atalla 1984).

The first applications of CP/MAS ¹³ C solid state NMR to cellulose was reported by Attalla and coworkers (Atalla et al. 1980), who presented spectra of cellulose I and II and an amorphous sample, which demonstrated the non- equivalence of adjacent anhydroglucose units and were consistent with con- formational differences between the polymorphs. Later, VanderHart and Atalla found that the anhydrocellobiose repeating units in cellulose I and II showed different conformations and that both polymorphs exist in wood (Vanderhart et al. 1984; Krässig 1996).

When studying different polymorphs of cellulose by solid state NMR spectra various methods can be used to resolve the overlying peaks, one of which (used in many studies) is curve resolution (Wickholm et al. 1998;

Larsson 1999; Larsson et al. 1999; Hult 2002; Hult et al. 2002; Falt et al.

2004). Another method that has been used with success is principal compo- nent analysis (PCA) and partial least squares (PLS) regression (Lennholm et al. 1994; Lennholm et al. 1995; Lindgren et al. 1995; Wormald et al. 1996).

In the studies presented in this thesis PCA and PLS regression were used.

All samples used in the work underlying this thesis were wetted before being packed into the CP/MAS rotor and contained 50 % water when ana- lysed in the spectrometer. This enhances the spectral resolution (Horii et al.

1985; Lennholm 1994; Lennholm et al. 1994) and facilitates packing of the samples into the rotor. A spectrum of dissolving pulp is shown in Figure 4.

0 0,05 0,1 0,15 0,2 0,25 0,3 0,35 0,4 0,45 0,5

30 40 50 60 70 80 90 100 110 120 130

ppm

Figure 4. A solid state 13C NMR CP/MAS spectrum of dissolving pulp with magni- tude phasing.

20

3.1.1. Processing NMR spectra

After recording NMR spectral data the originally collected data are treated with appropriate functions to facilitate their interpretation. Examples of such treatments include zero-filling, baseline correction (DC), exponential multi- plication (line broadening), Fourier transformation (FT) and phasing (Günther 1992). The spectra in Papers I, III and IV were all acquired with 2 K data points and zero-filled to 4K. Exponential line broadening (lb) was applied, prior to Fourier transformation, of 5 Hz and 10 Hz for spectra in Paper I and Paper IV, respectively. Phasing of the Fourier-transformed spectra in all studies in the thesis were done by magnitude phasing. This is not usually the first choice when dealing with high-resolution NMR spectra since it broadens the peaks. However, when the spectra are going to be used in MVA it is important not to insert variations (which may occur with man- ual phasing) into the spectra during the processing. The effects from phasing when using the NMR spectra in MVA were studied for the spectra in Paper IV. The phasing methods compared were magnitude-, manual-, and SVD- phasing. Magnitude- and SVD-phasing gave results that agreed, but differed from results generated by manual phasing.

3.2. NIR spectroscopy

The NIR spectrum is defined as the region of the electromagnetic spec- trum between 700 and 2500 nm. The absorption bands observed in this area arise from overtones of hydrogentic stretching vibrations involving AH _y functional groups or combinations involving stretching and bending modes of vibration of such groups. Herschel discovered the near infrared region 1800 (Mcclure 1994). The first commercial NIR spectrometer appeared in 1954, initiating extensive studies (Osborne 1986), but Norris is acknowl- edged as the inventor of NIR technology. He and his co-worker used NIR spectroscopy for quantifying grain during the 1970s (Mcclure 1994). As the NIR spectra are mostly explained as bands from overtones and combinations of fundamental vibrations involving hydrogenic stretching modes it is possi- ble to predict the positions of an NIR peak from the well documented mid- IR spectra. The attempts to assign the NIR spectra in the studies underlying this thesis, however, were based on chemical assignments of observed se- lected NIR absoption bands. An NIR spectrum of dissolving pulp is shown in Figure 5.

21

The use of vibration spectroscopy, such as mid-infrared, near-infrared and Raman spectroscopy, has a long tradition in cellulose fibre research. Infrared and Raman spectroscopy yield information about the chemical composition, molecular conformation and hydrogen bonding patterns of the cellulose polymorphs (Atalla 1976; Michell 1993; Michell 1993; Schenzel and Fischer 2001; Michell 1995). Besides X-ray and solid state NMR spectroscopy, vi- brational spectroscopy has been an important tool for studying the structure of cellulose in fibres. While mid-infrared and Raman spectroscopy have been used to reveal chemical information about the cellulose structure, NIR spectroscopy in combination with MVA has proven to be useful in quality control of wood and pulp (Wallbacks et al. 1991; Wallbäcks et al. 1991;

Antti et al. 1996; Michell 1995). The strengths of the NIR technique are the easy sample preparation, the robustness of the instrument and the short time for analyses. On the other hand, the spectra are difficult to interpret due to the overlapping bands of overtones and combinations of molecular vibra- tions. In the studies described in Papers I and II we used NIR spectroscopy to develop MVA models for predicting the pulp quality. In Paper III princi- pal properties from PCA of NIR spectra were used as variables in a D- optimal design to select samples for chemical analyses and calibration ob- servations in model building. In Paper IV we studied the complementary information in NIR and NMR spectra obtained from pulp samples. We also tried to assign peaks in the NIR spectra as hydrogen bonded hydroxyl groups in different cellulose polymorphs.

-0,1 0 0,1 0,2 0,3 0,4 0,5

400 900 1400 1900 2400

nm 1/R

Figure 5. An NIR spectrum of dissolving pulp.

22

4. Multivariate analysis

In all studies data analysis plays a crucial role. In most cases the collected data are complex and difficult to overview. In pulp and paper investigations two or three parameters are commonly plotted against each other in order to study their interactions and attempt to detect important trends. A more sim- ple and efficient way is to use principal component analysis (PCA). PCA has been used in a number of different pulp and paper applications (Wallbacks et al. 1991; Broderick et al. 1995; Lennholm and Iversen 1995; Josefsson et al.

2002). MVA has proved to be a valuable tool for analysing spectral data related to the cellulose structure in pulp (Lennholm et al. 1994; Josefsson et al. 2001). Another way to detect the overlapping peaks from different poly- morphs of cellulose is spectral fitting (Hult et al. 2002; Falt et al. 2004). The drawback of this method is that preconceived notions are applied in the analysis is. When using PCA and PLS the variation in the data extracted is not pre-valued during the decomposition of the original dataset. Instead, prior knowledge is first used in the following interpretation step.

4.1. PCA

Principal component analysis is a method used to extract the main varia- tion in a multivariate set of data into a few new latent variables. The data matrix X is reduced from the K dimensional space to an A dimensional sub- space, a hyperplane, described by the principal components, PCs. Mathe- matically PCA is a decomposition of the data matrix X into the product of two matrices; T and P’. T is a score matrix that gives each object its position in the subspace and P’ is a loading matrix that gives the direction of the hy- perplane. The angle between a variable and a PC can be described as the arccos of the loading value of the variable, provided that the sum of the squared loadings for each model dimension is normalized to one. The devia- tion between the variation explained in the X matrix and the variation ex- plained by the PCs, the product TP’, is the residual matrix, E (Eq. 1). PCA is described in more detailed by Jolliffe (Jolliffe 1986)

X=TP’+E Eq. 1

23

4.2. PLS

Regression models built on many variables that depend on each other demands a regression method that can handle this type of data. One method that meets this criterion is partial least squares (PLS) regression. Another advantage of using PLS regression is that it gives the possibility to interpret the data by studying score and loading vectors. Furthermore, PLS generate robust regression models, i.e. the regression parameters do not change much when one sample is removed from them. PLS regression can also handle missing data to some extent.

PLS relates the descriptor matrix X to a response matrix Y by maximiz- ing the covariance between the X-scores T and the Y-scores U and simulta- neously minimizes the residuals E and F (Eqs. 2 and 3). The contribution of each x-variable to the explanation Y is calculated for each component, re- sulting in the weight matrix, W. The corresponding weight matrix for the response matrix is C. The connection between the two score matrixes T and U is called the inner relation (Eq. 4) with the residual H.

X=TP’+E Eq. 2

Y=UC’+F Eq. 3

U=TB+H Eq. 4

When the calibration model is used for prediction one uses the combina- tion of Eqs. 3 and 4 (Eq. 5).

Ŷ=TBC’+F* Eq. 5

PLS is described in numerous books and articles (Martens and Næs 1989;

Wold et al. 2001).

4.3. Pre-treatment of data

In PCA and PLS all data are mean-centered before modelling. Chemical data are scaled. The scaling method most often used for a variable is division by its standard deviation. When using spectral data they are often not scaled in order to avoid amplifying contributions from the noise.

Sometimes it is necessary in multivariate analysis to pre-treat the data.

Examples of common pre-treatments for this purpose are centring, scaling, logarithmic transformation, and orthogonal signal correction (OSC) (Martens et al. 1989; Wold et al. 1989). Useful treatments for NIR spectral data include derivatisation and multiple scatter correction (MSC) (Savitsky and Golay 1964; Norris and Williams 1984; Geladi et al. 1985).

24

In the studies this thesis is based upon, the NMR spectral data were zero- filled, baseline-corrected, Fourier-transformed, magnitude-phased, aligned according to the anomeric carbon of cellulose (the C1 peak at 105 ppm) and normalized by dividing the intensity of each data point by the sum of the intensities of all data points and centred before the multivariate analysis. The NMR spectral data were not scaled before multivariate analysis, since scal- ing to unit variance could give the noise the same apparent importance as the small peaks in the spectra and thus complicate the interpretation.

The NIR spectra were always centred. Sometimes they were also scaled to unit variance and sometimes MSC was applied before the multivariate analysis. The reasons for scaling the NIR spectra were that they consist of several overlapping peaks and that important information partly hidden in the noise could be highlighted by scaling to unit variance. In Paper II or- thogonal signal correction was applied to the NIR spectra before the PLS regression model was calculated.

4.3.1. Alignment

When using NMR spectral data in MVA it is important to consider the alignment of the spectral data. Peak shifts can be affected by inhomogenei- ties in the applied magnetic field, magnetic field drift, temperature variations and variations in the sample matrix. To correct for shift changes it may be necessary to pick a peak in one spectrum and align all other spectra accord- ing to that shift. Different methods of alignment are aviable, and their bene- fits and drawbacks are continuously discussed (Vogels et al. 1996; Forshed 2003; Torgrip et al. 2003). As mentioned in chapter 3, chemical shift differ- ences can give potentially interesting information about the atoms’ environ- ment. This was pointed out by VanderHart and Lennholm (Vanderhart et al.

1984; Lennholm and Iversen 1995) and we used this in Paper III. Caution must be taken not to destroy important information when aligning the data.

4.3.2. Normalization

Normalization of NMR spectra prior to multivariate analysis is most commonly done by dividing each wavelength by the sum of the spectra for each object. This was done for all spectra reported in Papers I, II and IV.

Normalising the spectra by dividing each wavelength by pulp sample weight was tried, but did not show any positive effects.

4.3.3. Scaling

In order to give each variable equal chance to contribute to the model,

scaling by dividing each value by the standard deviation of the variable is

25

done. This gives variables with different variances the same possibility to contribute to the model. Scaling can also be done in more advanced ways by weighting variables differently to make them contribute differently to the model. However, this must be done very carefully to avoid manipulating the data according to preconceived notions.

4.4. Validation

In order to determine the quality of the model, validation is an important component of modelling. In PCA and PLS regression, when calculating the model, the variation in the X and Y data explained by each component and for all components together is computed and expressed as R ² (Eq. 6), which is the fraction of all the variance, not corrected for degrees of freedom. This indicates how well the model describes the data.

R ² = 1- ( )

( ) ²

2

∑ ∑

Υ

− Υ

Υ

− Υ

mean observed

calculated observed

Eq. 6

Cross-validation (CV) resulting in Q ² (Eq. 7) measures the predictive ability of the model (Wold 1978). Q ² thus gives the statistical significance of a component.

Q ² = 1- ( )

( ) ²

2

∑ ∑

Υ

− Υ

Υ

− Υ

mean observed

predicted observed

Eq. 7

Eigenvalues can be used to validate the significance of a PC, especially when modelling large sets of data, while CV is preferred when modelling small sets of data. Another way of investigating the validity of a model is the permutation test (Eriksson et al. 1997). When building calibration models their ability to predict external observations is the most important criterion.

To estimate the accuracy of predictions derived from the model an external test set of objects is used and the residuals of the predicted objects are inves- tigated by calculating the root mean square error of predictions, RMSEP (Eq.8) (Martens et al. 1989).

26

RMSEP = ( )

N Y Y _{observed predicted}

∑ ^{− 2} _{Eq. 8}

Validation of models based on small data sets involves special difficulties that have been discussed by Martens (Martens 1998). However, chemical interpretation can complement all these statistical tools to validate models.

27

28

5. Experimental design in producing and selecting samples

Statistical experimental design or design of experiments is a method for planning series of experiments in such a way that the experimental variables are varied independently of each other in a systematic and efficient way. The investigated region, the experimental domain is decided by the maximum and minimum values of the variables. The benefits of using design when planning experiments is that one obtains orthogonal variables, facilitating the identification of variables that interact with each other, and performing more experiments than necessary can be avoided (Box et al. 1978; Box and Draper 1987).

Another situation where statistical experimental design should be used is in the selection of samples for analyses and for building calibration models.

Often this is done by extracting the largest variation in the data and com- pressing the original number of variables into a few principal properties by PCA (Wold et al. 1986). This concept is often referred to as multivariate design.

In Paper I we used L9 and L8 Taguchi Robust Designs (Phadke 1989) to design the experimental variables when producing dissolving pulps in the laboratory. Unfortunately, interaction effects could not be measured, due to the limited number of experiments in combination with the number of levels investigated for the variables.

In Papers I and III we used the score vectors, the principal properties, from PCA and PLS regression of chemical properties and NIR spectral data as variables in D-optimal design for selecting samples (Mitchell 1974). The method was useful for obtaining representative samples for chemical analy- ses and model building. Recently, new developments in design methods for selection samples have shown promising results (Olsson et al. 2004).

29

30

6. Summary and discussion of Papers I- IV

The studies underlying this thesis focused on issues related to measuring the quality of dissolving pulp. In the first study, presented in Paper I, labo- ratory cooked pulp samples were used. The samples were cooked according to an experimental design in which important process variables were used as design variables. The chemical properties of the samples were studied by MVA and correlated to the designed process variables. Spectroscopic analy- ses by solid state ¹³ C CP/MAS NMR and diffuse reflectance NIR spectros- copy were done on the laboratory-produced samples and the spectral data were analyzed by MVA. In the second study (Paper II) the utility of NIR spectroscopy and MVA to predict the chemical properties of factory- produced pulp samples was examined. These two studies raised questions about the best way to measure the reactivity of dissolving pulp for viscose manufacture. Therefore, the third study, presented in Paper III, examined a chemical approach for analysing reactivity when making viscose in the labo- ratory; measuring the cellulose yield according to Fock’s method (Fock 1959). Correlations between this reactivity measure and other properties were studied, as well as NMR spectral properties. MVA was used in the analyses. In this study the properties of electron beamed pulp samples were also investigated. In the fourth and last study a series of factory-produced samples with different reactivity (according to Fock’s method) were selected and the relationships between their chemical and spectral properties were studied by MVA and presented in Paper IV.

6.1. Process variables correlated to chemical properties of laboratory cooked pulp samples

In the first study process variables were varied according to Taguchi Ro-

bust designs L8 and L9 (Phadke 1989) when producing 17 dissolving pulps

in the laboratory. The data from these experiments also included information

on changes in variables that were not varied according to the design, so data

were analyzed with PLS regression. As mentioned earlier, the viscosity is

the first property sought by a viscose producer. In Paper I filterability was

plotted versus viscosity and alkali resistance versus viscosity. Both these

31

pair of properties showed a moderate correlation, highlighting the complex- ity of characterizing dissolving pulp quality for viscose manufacture. The viscosity and the alkali resistance were modelled from the ten process vari- ables by PLS regression. Interpretation of the PLS weight vectors showed that high alkali resistance was achieved by keeping the chemical charge (Na ₂ O) and the initial pH of the cooking liquor (pH) low and the cooking time in the second stage (Time2) long, Figure 6. To make pulp with low viscosity the chemical charge and initial pH of the cooking liquor should be low and the cooking time in the second stage long. High Na 2 O and pH, and short Time2 give high viscosity. To be able to produce a pulp with the de- manded viscosity and acceptable alkali resistance there is a need for careful optimization of the cooking parameters. The filter clogging value (Kw) was modelled from the ten cooking variables and the PLS weight vectors clearly indicated that to get a low filter clogging value, i.e. high filterability, the chemical charge should be kept high, Figure 7. Also the cooking time during the second stage (Time2), the maximum temperature in the second stage (Temp) and the total time (Time = 100 min heat up time + Time1 + Time2) are important, and all of these variables should be kept low to get a low filter clogging value, i.e. high filterability. The results indicate the complexity of the correlations between the process variables and the properties of the pulp.

Controlling these parameters is challenging, and the use of multivariate analysis is essential.

-0.80 -0.40 0.00 0.40 0.80 -0.80

-0.40 0.00 0.40 0.80

PLS weight vector 1

PLS weig ht ve ct o r 2

Na2O

pH Temp

Time1 Time2

Time

H1 VäVe Vatten Kvarn

R18

Viscosity

Figure 6. PLS weight vectors from latent variables one and two from the PLS-model of viscosity and R18 and 10 cooking variables (Paper I).

-0.80 -0.40 0.00 0.40 0.80 -0.80

-0.40 0.00 0.40 0.80

PLS weight vector 1

PLS weig ht ve ct o r 2

Na2O

pH Temp

Time1

Time2 H1 Time

L/W Water

Mill Kw

Figure 7. PLS weight vectors from latent variables one and two from the PLS-model of the filter clogging value and 10 cooking variables (Pa- per I).

32

6.2. Prediction of properties by NIR spectroscopy

The chemical analyses needed to measure the pulp quality are often time consuming and complex, requiring special equipment and skilled laboratory personnel, which make the measurements expensive. In order to improve the efficiency of the quality control the use of NIR spectroscopy in combination with MVA has proven to be successful (Wallbacks et al. 1991; Antti et al.

1996; Michell and Schimleck 1996; Michell and Schimleck 1998; Marklund 1999; Michell 1995).

6.2.1. Models developed from laboratory-produced pulps

NIR spectra of the laboratory-produced pulp samples were recorded and analyzed by MVA, as presented in Paper I. A score plot with scores from principal component analysis of the NIR spectral data is shown in Figure 8.

Pulps with high alkali resistance are found to the right in the figure and pulps with low alkali resistance are found to the left. This plot shows that the NIR spectra contain information that correlates with the chemical properties of the pulp and the information can be extracted with MVA. PLS models of alkali resistance and the filter clogging value were presented in Paper I and the results indicate that these properties could be calculated by NIR spec- troscopy and MVA.

33

Figure 8. Principal properties, scores, from PCA of NIR spectra of 17 dissolving pulps. Black, white and grey symbols indicate pulps with low, high and intermediate alkali resistance, respectively. For the identity of the pulps see Table 1 in Paper I.

6.2.2. Models developed from factory-produced pulps

The results from the laboratory-produced pulp samples were promising,

so the next step was to find out if it was possible to develop calibration mod-

els for the chemical properties of the pulps from NIR spectra for factory-

produced samples. PLS regression models for viscosity, alkali resistance,

sheet density and filterability of viscose modelled from NIR spectral data

were presented in Paper II. Viscosity could be predicted with an RMSEP of

28 ml/g for samples in the range from 300 ml/g to 570 ml/g, Figure 9a. Al-

kali resistance respectively at 18 and 10% NaOH could be predicted from

NIR spectral data with an RMSEP of 0.5% and 0.9%, respectively, for sam-

ples ranging between 93.4 to 96.3% and 82 to 92%, Figure 9b and 9c. In

comparison, the reported reproducibility for the SCAN method is 1.26 and

34

2.04% for the solubility at 18 and 10% NaOH, respectively (SCAN-2:61).

Sheet density could be predicted with an RMSEP of 22 kg/m ³ for samples in the range between 600 to 900 kg/m ³ Figure 9d. In contrast to the acceptable predictability of the pulp properties viscosity, alkali resistance and sheet density, the attempts to predict the filter value of viscose were unsuccessful.

Possible explanations for this are that the samples had unevenly distributed filterability; that most samples had low filter values, but a few had much higher values; and that the reproducibility of the filter value measurements were low. This emphasizes the importance of good precision and accuracy in the reference analyses when building calibration models. However, charac- terization of the pulp samples by PCA of the NIR spectra seemed to give good estimates of the filterability, Figure 9 in Paper I.

300 350 400 450 500 550 600

300 350 400 450 500 550 600

Measured values

Calculated values

1 comp. used for prediction with PLS Model 'Viscosity'

Viscosity [ ml/g ]

in model not in model

Figure 9a. Predicted/calculated versus measured viscosity from NIR spectra (Paper II).

82 83 84 85 86 87 88 89 90 91 92 93

82 83 84 85 86 87 88 89 90 91 92 93

Measured values

C a lc ul ated val ues

1 comp. used for prediction with PLS Model 'R10'

R10 [ % ]

in model not in model

Figure 9b. Predicted/calculated versus measured alkali resistance (R10) from NIR spectra (Paper II).

35

93.0 93.5 94.0 94.5 95.0 95.5 96.0 93.0

93.5 94.0 94.5 95.0 95.5 96.0

Measured values

Calculated values

1 comp. used for prediction with PLS Model 'R18'

R18 [ % ]

in model not in model

Figure 9c. Predicted/calculated versus measured alkali resistance (R18) from NIR spectra (Paper II).

550 600 650 700 750 800 850 900 950

550 600 650 700 750 800 850 900 950

Measured values

Calculated values

1 comp. used for prediction with PLS Model 'Density'

Density [ kg/m^3 ]

in model not in model

Figure 9d. Predicted/calculated versus measured sheet density from NIR spectra (Paper II).

36

6.3. Reactivity - characterisation using chemical properties

The concept of cellulose reactivity was presented and discussed in chapter 2, as were the different pulp properties measured to characterize pulp qual- ity. When using pulp for producing cellulose derivatives several different methods can be used for estimating the reactivity. For viscose, the filterabil- ity is the property most commonly used (Treiber 1987). In the studies re- ported in Paper I and II we found that when measuring the viscose filter value, or filter clogging value, the measurement uncertainty were substantial.

The laboratory method used to produce the viscose is complex and demands special equipment. In Paper I we also showed that alkali resistance and vis- cosity were both moderately correlated to the filter clogging value. We therefore decided to investigate Fock’s method to measure the cellulose yield (Fock 1959). Roffael found no relationship between viscose filter value and residual cellulose content measured according to Focks method when studying various kinds of pulp (Roffael 1988). However, when restricting the investigated pulp set merely to wood pulps a correlation was found.

6.3.1. Characterisation using chemical properties

Eleven chemical properties plus the filter values at two different CS 2 con- centrations were studied by PCA in the studies reported in Paper III, Figure 10a and 10b (Figure 3 and 4 in Paper III). The results showed that pulps with high filterability (low filter value) had low alkali resistance (R18 and R10), low viscosity and high contents of extractives and magnesium.

Clearly, pulps with relatively short cellulose chains and low molecular weight cellulose are more dissolvable in alkali. The low alkali resistance at 18 % alkali also indicates that pulps with more amorphous cellulose and low proportions of crystalline cellulose are more reactive. This was also demon- strated by NMR spectral data in Paper I. The influence on the reactivity of the extractives content has been discussed by Samuelsson (Samuelsson 1948). Their positive effect can be explained by the influence of the resin on the superficial tension, which facilitates the diffusion of the carbon disulfide.

High content of extractives may also prevent some extent of hornification. A possible reason for the more reactive pulps amongst the samples discussed in Paper I having higher magnesium contents is that some of the extractives form metal complexes. For example, tropolones form strong metal com- plexes with heavy metal ions (Sjöström 1981). The reasons for the variation in the content of extractives and, thus, magnesium were not explored in the

37

study in Paper I. Samples with high alkali resistance had low carboxyl group contents and low polydispersity. This means that pulps with high alkli resis- tance contain less hemicellulose and cellulose chains with low polydisper- sity.

-4 -3 -2 -1 0 1 2 3 4

-5 -4 -3 -2 -1 0 1 2 3 4 5

t[2]

t[1]

6218

1000 1168 1609

1800 1845 2470

4147 4149

4305 4325

8887 8891

3855 4468

4474

4970

5074 5151

Figure 10a. Scores from the PCA of chemical properties (Paper III).

-0,40 -0,30 -0,20 -0,10 0,00 0,10 0,20 0,30 0,40 0,50

-0,40 -0,30 -0,20 -0,10 0,00 0,10 0,20 0,30 0,40

p[2]

p[1]

Visc

R10

R18 Acet

Mg

SiO2

Dens

COOH DPw

DPn Polyd

kr24kr30

Figure 10b. Loadings from the PCA of chemical properties (Pa- perIII).

The filter value and the residual cellulose determined according to Fock’s method were modelled from the chemical properties, and the resulting weight vectors from the PLS models showed consistent results with the PCA, Figure 11 and 12, Figure 6 and 8 in Paper III. The important variables were viscosity, alkali resistance, contents of extractives and magnesium. For the model where the residual cellulose was used as the response variable, Figure 12, the carboxyl group content and the polydispersity also contrib- uted significantly. Low carboxyl group content and polydispersity were as- sociated with low residual cellulose. This means that pulps with high reactiv- ity had low hemicellulose contents and low polydispersity.

38

-0.60 -0.40 -0.20 0.00 0.20 0.40 0.60

Visc R10 R18 Acet Mg SiO2 Dens COOH DPw DPn Polyd y=kr24

w*c[1]

Figure 11. PLS weight vectors from the PLS model of viscose filter value derived from chemical properties (Paper III).

-0.40 -0.20 0.00 0.20 0.40

Visc R10 R18 Acet Mg SiO2 Dens COOH DPw DPn Polyd y=Res.cell

w*c[1]

Figure 12. PLS weight vectors from the PLS model of residual cellulose according to Fock derived from chemical properties (Paper III).

In the study reported in Paper IV the reactivity was measured according to Fock’s method and expressed as cellulose yield for set of pulp samples.

The chemical properties were analysed with MVA. The PCA shows that pulp samples with high cellulose yield, and thus high reactivity according to Fock’s method, had high alkali resistance at 18 % NaOH (R18), high content of extractives, high brightness and high fibre surface charge, Figure 13a and 13b. The pulps with high cellulose yield had also low viscosity, low alkali resistance at 10 % NaOH (R10), low ash content and low weight average degree of polymerisation. Also the dirt counted, fibre charge, number aver- age degree of polymerisation and yield from acid hydrolyses had negative loadings, but these variables were not as important. These findings were interpreted as showing that pulp samples with high reactivity had short cellu- lose chains, high content of low molecular cellulose and low hemicellulose content. The contents of extractives were high and addition of a surface ac- tive compound had a positive effect. The fibre surface charge was higher in more reactive pulp samples. The positive effect of naturally occurring ex- tractives or added surface active compounds has been discussed in Papers III and IV and in chapter 2.5 in this thesis.

39

-6 -4 -2 0 2 4

D2211 D2136 D2423 D2209 eb D7777 D6758 D7497 D7347 D7567 D7675 D7569

t[1]

Figure 13a. Scores from the PCA of chemical properties (Paper IV).

-0,30 -0,20 -0,10 0,00 0,10 0,20 0,30

aex1 Fock7% Fock8% Fock9% sfchrage R5 R10 acidhyd GPCMw GPCMn fcharge V-388 Dirt count brightness Ash SiO2 visk_calc R18_calc

p[1]

-0,30 -0,20 -0,10 0,00 0,10 0,20 0,30

aex1 Fock7% Fock8% Fock9% sfchrage R5 R10 acidhyd GPCMw GPCMn fcharge V-388 Dirt count brightness Ash SiO2 visk_calc R18_calc

p[1]

Figure 13b. Loadings from the PCA of chemical properties (Paper IV).

6.3.2. Reactivity using Fock’s method

As mentioned above, the methods for estimating reactivity by measuring

the residual cellulose or yield according to Fock’s method were investigated

and the results were presented in Paper III. The error of the laboratory

method, expressed as the standard deviation for replicates of samples ana-

lysed at the same time was 4%, (average, 59.8%). The standard deviations

for samples analysed at different times were 10 or 13%, with means of 47.0

and 47.8%, respectively, Table 2 in Paper III. In comparison, the repro-

ducibility of the filterability measurements was much lower: a set of samples

was measured twice, and differences of 2160 and 1450 were found for kr30

and kr24, respectively, between the two measurements. There are important

differences between the methods for measuring filter value and residual cel-

40