Note: Descriptions are shown in the official language in which they were submitted.
CA 02447098 2003-10-28
1
METHOD FOR ESTIMATING AN OPTIMAL DOSAGE OF
BLEACHING AGENT TO BE MIXED WITH WOOD CHIPS
Field of the invention
The present invention relates to the field of pulp and paper process
automation, and more particularly to methods for estimating optimal dosage
of bleaching agent to be mixed with wood chips to be fed to a process for
producing pulp characterized by a required brightness value.
Background of invention
Thermomechanical pulp properties and quality are influenced by two
types of variables: feed material (chips) and process (refiner). Over the
years, many researchers have underscored the impact of the stability of the
refiner operation for the production of constant pulp quality, as mentioned
by Strand, B. C. in "The Effect of Refiner Variation on Pulp Quality",
International Mechanical Pulping Conference, Proceedings, 125-130 (1995).
However, variations of the process itself are mainly related to variations in
the raw material feeding the system as, mentioned by Wood, J. A. in "Chip
Quality Effects in Mechanical Pulping - a Selected Review", 1996 TAPPI
Pulping Conference, Proceedings, 491-497 (1996). In particular, pulp
brightness is considered as an important quality requirement, as discussed by
Dence, C.~ W. et al. in "Pulp Bleaching - Principles and Practice", TAPPI
Press, 457-490 (1996).
An object of the method according to the invention is to model the
relationship between the quality of the chips feeding the process with an
important pulp and paper resulting property, namely pulp brightness. In
particular, the model is used to evaluate the minimum charge of peroxide
required to reach certain level of pulp brightness according to possible chips
properties fluctuations, in order to minimize the cost and environmental
impact of the bleaching operation.
CA 02447098 2003-10-28
Summary of invention
According to the above mentioned object, there is provided a method
for estimating an optimal dosage of bleaching agent to be mixed with wood
chips to be fed to a process for producing pulp characterized by a required
brightness value. The method comprises the steps of: i) estimating a set of
wood chip properties characterizing said wood chips to generate
corresponding wood chip properties data, said set including reflectance-
related properties; ii) providing an initial dosage value of said bleaching
agent; iii) feeding said wood chip properties and said bleaching agent dosage
value at corresponding inputs of a neural network previously trained
according to experimentally obtained data on said wood chip properties and
on dosage of said bleaching agent, to generate predicted brightness value for
pulp to produce from said wood chips; iv) comparing said brightness
predicted value with said required brightness value to generate error data;
v) optimizing said bleaching agent dosage value to minimize said
error data; and vi) repeating said steps iii) to v) with said optimized
bleaching agent dosage value until said brightness predicted value
substantially reaches said required brightness value, to estimate said optimal
bleaching agent dosage.
brief description of the drawings
The method according to the present invention will be described in
detail with reference to the accompanying drawings in which:
Fig. 1 is a graph showing relative importance index of independent
variables according to PLS analysis;
2S Fig. 2 is a graph showing coefficient of correlation for dependent
variables by PLS analysis;
Fig. 3 is a graph representing observed and predicted values for ISO
brightness; and
Fig. 4 is a block diagram of a neural network-based model that can be
used to carry out the method according to the invention.
CA 02447098 2003-10-28
3
Detailed description of the preferred embodiment
In order to def ne the parameters used for the model, two sets of
experiments corresponding to two different blocks were performed.
In the first block, a potential mix of four species, black spruce, balsam
fir, jack pine and white birch, was studied. The last two species were chosen
because they represent a potential source of new resources. The trees have
been selected, cut, barked and chipped in order to obtain standard chips with
known and controlled age. In fall, outdoor stacks of each species of chips
were prepared. During the following 12 months, six samples were selected
in order to conduct the experimental plan for chips aging as described in
table 1.
Test number Spruce% Jack pine% Birch%
Balsam fir%
1 0 0,2 0,4 0,4
2 1 0 0 0
3 0 1 0 0
4 0,6 0 0 0,4
5 0 0,6 0,4 0
6 0,6 0 0,4 0
7 0 0,6 0 0,4
8 0,2 0 0,4 0,4
Following are repetitions
for experimental error
determination
9 1 0 0 0
10 0 1 0 0
Following are additional
tests
11 0 0 1 0
12 0 0 0 1
Table 1
CA 02447098 2003-10-28
In each sample, the experiments for 100% black spruce and 100% balsam fir
were repeated twice in order to evaluate the experimental error (14 runs in
each sample). The six samples allow to evaluate the evolution of the quality,
i.e. degradation, of the chips in time. This degradation is highly dependent
on storage temperature. The first four samples were evaluated at an interval
of three weeks. After that, there has been a longer waiting time. It was
noticed that the winter degradation of each stack was extremely slow.
The second block of experiments was used to investigate the effects
of other important variables regarding pulp quality. This second block of
experiments has been conducted with four variables: species (black spruce,
balsam fir), density (high, low), initial dryness of the chips (fresh, dry),
and
thickness of the chips (0-4 mm, 4-8 mm). Table 2 describes the
experiments for chips aging that were conducted in this second block.
CA 02447098 2003-10-28
S
INITIAL CHIPS
Large chips Small chips
(4-8 mm) 0-4 mm
Spruce at low densityTest no. 1 Test no. 2
Balsam fir at low Test no. 3 Test no.4~
density
Spruce at high densityTests no.5 and 6 Test no.7
Balsam fir at high Test no. 8 Tests no. 9 and
density 10
Aged chips (dryness
of 75%)
Large chips Small chips
(4-8 mm) 0-4 mm
Spruce at low densityTest no. 11 Tests no.l2 and
13
Balsam fir at low Tests no.l4 and Test no. 16
density 15
Spruce at high density! Test no. 17 Test no. 18
Balsam fir at high ~ Test no. 19 Test no. 20
density
Table 2
Refining was conducted on a pilot unit Metso CD-300. Each sample was
washed and refined in two stages. The first one was conducted at a
temperature of 128°C and the second one at atmospheric pressure. For
each
experiment, pulps with a freeness ranging from 200 to 150 mL were selected
for further peroxide bleaching, which fundamental principles are briefly
described next.
CA 02447098 2003-10-28
6
It is generally accepted that the active mechanism in chromophore
elimination with hydrogen peroxide as bleaching agent involves the
perhydroxyl ion OOH-. As taught by Sundholm, J. in "Papermaking Science
and Technology - Mechanical Pulping", Finnish Pulp and Paper Research
Institute, 313-345 (1999), hydrogen peroxide bleaching is therefore
performed in alkaline systems to produce the active ion according to the
following equation:
H202 + OH- -3' OOH- + H20 ( 1 )
The formation of the perhydroxyl anion can be enhanced by increasing the
pH or by increasing the temperature. Hydrogen peroxide readily
decomposes under bleaching conditions according to the following equation:
2 H202 ~ OZ + HZO (2)
Sodium silicate and magnesium silicate are normally added to the bleach
liquor to stabilize peroxide. Transition metals ions like iron, manganese and
copper catalyze peroxide decomposition. In order to prevent that, before
bleaching with peroxide, the pulp was pretreated with 0,2% of DTPA. The
pretreatment of the pulp was done at 60°C, 15 minutes and 3 % of
consistency.
Different concentrations of hydrogen peroxide varying from 1 to 5%
(O.D. basis) were tested for bleaching the different pulp. Table 3 describes
the experimental conditions used for the peroxide bleaching of the pretreated
pulps.
CA 02447098 2003-10-28
7
Parameters P1 P2 P3 P5
Temperature, 70 70 70 70
C
Retention time, 180 180 180 180
min
Consistency, 12 12 12 12
%
Sodium silicate,3,00 3,00 3,00 3,00
%
Magnesium sulfate,0,05 0,05 0,05 0,05
Total Alcali 2,00 1,20 0,X0 0,80
Ratio
Sodium hydroxyde,1,66 2,06 2,36 3,66
%
Hydrogen peroxide,1,00 2,00 3,00 5,00
Table 3
where:
Total Alcali Ratio = % Hydrogen Peroxide -
OH- given by sodium silicate and sodium hydroxide (3)
Bleaching was conducted at 70°C, 180 minutes and 12 % of
consistency.
The bleaching liquor was composed of 3,00% of sodium silicate, 0,05% of
magnesium sulfate, hydrogen peroxide and sodium hydroxide. After the
bleaching step, the pulp was diluted at 1 % of consistency and neutralized
with sodium metabisulfite at pH 5,5. A volume of the bleaching liquor was
kept to measure the residual peroxide by an iodometric dosage. optical
properties such as ISO brightness and color coordinates (L*, a*, b*) have
been measured according to Paptac standard.
Chips of the eighty four (84) runs in block 1 and twenty (20) runs in
block 2 were systematically analyzed using a wood chip optical inspection
apparatus known as CMS-100 chip management system commercially
CA 02447098 2003-10-28
available from the present assignee, Centre de Recherche Industrielle du
Quebec (Ste-foy, Canada), for measuring a number of optical properties as
well as moisture content. Such wood chip inspection apparatus is described
in U. S. Patent no. 6,175,092 B 1 issued on January 16, 2001 to the present
assignee. Such mufti-sensor system includes main and optional auxiliary
sensors able to characterize wood chips online. The main sensors include
artificial vision sensor (an RGB color camera) and near infrared sensor to
measure chip brightness and moisture content. Auxiliary sensors such as a
distance sensor and an air conditions sensor to measure air temperature and
relative humidity may be advantageously used. They provide -information
that extends measurements of the main sensor to stabilize the system (for
example, variations of the camera measuring distance will influence the chip
brightness measurement). The system will work on frozen and non-frozen
wood chips, and it used for predicting bleach charges or dosage based on
chip quality for use as a bleach control method or system. The correlation
between some chip properties and its possible application in bleach control
is discussed by Ding, P. et al. in "Economizing the Bleaching Agent
Consumption by Controlling Wood Chip Brightness", Control System 2002,
Proceedings, June 3-5, Stockholm, Sweden, 205-209 (2002). The most
relevant wood chips properties measurements for the purpose of the present
invention are described next.
A first measurement relates to chip luminance, wherein the brightness
of black is defined as zero and the brightness of white as 150. The RGB
colour camera of the system is calibrated by a color checker made of black
and white paperboard. The wood chip color is between white and black, so
its brightness is between 0 and 1 S0. A second measurement relates to chip
average moisture content. The system includes a near infrared sensor that is
used to measure surface moisture content of wood chips, without any non-
contact therewith. A phenomenological model may be used to calculate the
average moisture content from surface moisture content, as described by
CA 02447098 2003-10-28
9
Ding, F. et al, in "Economizing the Bleaching Agent Consumption by
Controlling Wood Chip Brightness", Cohta°ol System 2002,
Proceedings,
June 3-5, Stockholm, Sweden, 205-209 (2002). Other measurements may be
obtained from various further sensors, generating a large amount of data
categorized in many different variables. According to the present preferred
embodiment, a number of four (4) other measurements are considered,
namely the image "H", "S" and "L" parameters, as well as a chip average
size estimation, which may be obtained using a surface-based, chip size
classifier such as disclosed by Ding, F. et al. in "Wood Chip Physical
Quality Definition and Measurement", IMPC Proceedings, June 2-5,
Quebec, Canada, 367-373 (2003).
The database resulting from the various experiments gives rise to
three {3) types of variables: chip properties coming from the measuring
system, operational parameters of the TMP and bleach processes, and pulp
quality characteristics. Overall, the database used contained a large number
(n=178) of variables distributed over a corresponding number of columns.
Because all (104) runs for both blocks produced pulps which were bleached
at four (4) different peroxide charges, the database also contains four times
(416) runs distributed over a corresponding number of lines. In order to
capture possible system measurements errors, the database contained many
repeated measurements for the same chips, leading a final database
containing a still greater number (SOC) of data lines. In the following
sections, the techniques that are preferably used to screen the columns of
data to a reasonable amount of most relevant variables and to use the lines
for neural network training will be explained. Both techniques are done with
the obj ective of obtaining a good enough pulp brightness model that could
be used in a brightness control strategy.
The data screening to perform the selection of the independent
variables which have an effect on the dependent variables that have been
measured is preferably done using known PLS (Projection on a Latent
CA 02447098 2003-10-28
Structure) modeling. Fig. 1 presents the independent variables that have
been chosen according their relative importance index. As expected, the
parameters which have most impact and are correlated to dependent
variables are the concentration of sodium hydroxide (NaOH) and the
5 concentration of hydrogen peroxide (PEROA). After that, the variables
Co moy (chip size), MDH (average of H), MMLC (average of luminance),
MDS (average of S), MDL (average of L) and l~IS(JRFM (average of the
surface moisture) also contribute to a lesser extent to the bleached pulp
properties response.
10 The correlation coeff dents for each dependent variable are presented
in Fig. 2. The value R2 shows the correlation for the dependent variables. It
is an indication of how well the model can fit the experimental data. The
value Q2 shows the correlation of the interpolated responses, i.e. predictions
not part of the experimental data. The graph shows that the model is
adequate to predict ISO brightness [ISOB] (coefficient of correlation of
0,88), color coordinates L* [LB] (coefficient of correlation of 0,92) and a*
[AB] (coefficient of correlation of 0,90), and residual peroxide [PEROR]
(coefficient of correlation of 0,83). As for the color coordinate b* [BB], the
coefficient of correlation is only 0,60. We also note that chemical properties
such as MEXT (extractives) and ACJR (fatty and resinic acids) are difficult
to correlate (coefficients of correlation of 0,33 and 0,26 are respectively
obtained).
Fig. 3 presents observed and predicted values for the ISO brightness.
These results show that the model is able to predict adequate values for this
optical property. Brightness ranging from 43.79% to 80.2% were measured
on the bleached pulps.
A neural network-based model that can be used to carry out the
method according to the invention will now be described in reference to Fig.
4. The model generated designated at 10 includes a neural network 12 that
was previously trained according go to the experimentally obtained data on
CA 02447098 2003-10-28
11
wood chip properties and on dosage of said bleaching agent as described
above, i.e. over the nine (9) remaining database columns consisting of eight
(8) inputs identified by PLS method as shown in Fig. 2, and one output,
namely pulp brightness as shown in Fig. 3. Such known neural network and
associated training approach are discussed by Laperriere L.et al. in
"Modeling and simulation of pulp and paper quality. After a few
unsuccessful training trials, it was noticed that the input I~Ta~H is always a
ratio of the input H2~2, so it was eliminated from the training set. ~ut of
the available (506) training lines, a selected number (96) were removed
(about 20%) and injected back to the trained network for validation.
Different sets of the removed 20% were tested and gave similar results. The
final configuration was a 7-5-1 neural network (7 inputs, 5 hidden neurons
and 1 output) as designated at 12 in F ig. 4. Training was stopped after an
average absolute mean error of 5% was reached between the neural network
prediction and the training output brightness value for each of the 506 lines.
The value of 5% was chosen by taking two factors into consideration: 1)
reliability of the output measurements: the experimental error related to the
brightness value is about 3%, i.e. X0.5 brightness points in the experimental
span of 43.79 to 80.2 measured brightness9 and 2) reliability of the input
measurements: calibration errors may encourage an increase of the training
error. The training results of the final network, in terms of the connection
weights between each of its constituting neurons, were imported into the
neural network 12 of model 10, in the form of a computer program that can
be implemented in a microcomputer b:y any person skilled in the art using
well known programming tools. Such program is able to simulate brightness
prediction based on the seven (7) chosen inputs, namely reflectance-related
properties of wood chips that are Luminance, M, H, S, L and chip size from
measurement system 14, and bleaching agent dosage (peroxide charge) value
used by the bleaching unit as represented at 1 ~. ~ptionally, unmodeled
disturbances may also be applied to the neural network at input 17.
CA 02447098 2003-10-28
12
In operation, according to the set of wood chip properties
characterizing the wood chips as estimated by the measurement system 14,
corresponding wood chip properties data are fed at respective inputs 18 of
neural network 12, as well as an initial dosage value of the bleaching agent
(peroxide) at further input 20. In turn, the neural network 12 generates at
output 22 thereof, a predicted brightness value for pulp to produce from the
inspected wood chips. Then, the brightness predicted value is compared with
the required brightness value to generate error dada, as indicated at node 24.
In turn, the error data is used by an optimization module 26, which optimizes
the bleaching agent dosage value to minimize the error data. Finally, the
above prediction, comparison and optimization steps are repeated with the
optimized bleaching agent dosage value until the brightness predicted value
substantially reaches the required brightness value, to estimate the optimal
bleaching agent dosage. In other words, the peroxide charge is tuned to
minimize the error, while maintaining constant chip properties, and an
optimization loops is performed in model 10 for several iterations before it
reaches the peroxide charge that meets the required brightness value or
setpoint according to the neural network model prediction. When this
optimal value has been found, it can be sent back to the actual process for
corrective action on a control valve provided on bleaching unit 2~. Such
control strategy assumes that the time taken for the optimization to take
place is less than the frequency at which brightness setpoints will be
modified, which is a reasonable assumption.
Because the brightness prediction is based upon variables that are
algorithmic transformations of camera signals, a first simulation was
designed in which the neural network model was used nn conjunction with
an optimizer that would find the best combination of measurement system
input variables that would give the best achievable brightness. Simulation
results are shown in table 4.
CA 02447098 2003-10-28
13
Variable Min. experimental Max. experimental
value value
Luminance 13.23 57.93
Moisture 19.4 '70.34
Size 16.49 21.39
H 19.29 224.44
S 82.08 192.89
L 9.17 66.85
H202 0.00 5.00
Table 4
There are two main observations from this result. First, for optimal
brightness all independent variables are either at the minimum or maximum
values or their respective span. This means that the hyper surface for which
a miminum was found slants towards an intersection of the constraint hyper
planes corresponding to the maximum or minimum values of each
independent variable. This also means there is a well defined combination
for maximum brightness (the optimal combination was consistently
reproduced for many different simulation trials). Second, we see that five
(5) of the six (6) system measurements give the best pulp brightness when
they are at their higher values, except for the "H" parameter (lowest value).
Turning back to Fig.l, the sensitivity of output properties with respect
1 ~ to some of the chosen inputs is shown. Because this result was obtained
from a PLS model and the use a neural network model is contemplated,
another set of simulations was razn to verify if the sensitivity of the sole
pulp
brightness to each independent variable would be similar. Table 5 shows a
series of tests where each variable was given sinusoidal swings of its value
over its total experimental span as shown in table 4, while maintaining other
variables at their central values, to show brightness sensitivity to the
independent variables.
CA 02447098 2003-10-28
14
Variable Min. brightness Max, brightness% change in brightness
Luminance 63.55 68.02 12.3
Moisture 64.52 67.01 6.8
Size 64.60 67.48 7.9
H 63.81 67.71 10.7
S 63.99 6?.69 10.1
L 65.18 ~ 66,52 3,6
H202 49.57 70.83 58.4
Table 5
It turns out that peroxide has a predominant effect. In fact, the peroxide
charge fixes the brightness level and changes in the chip properties simply
add small variations around the level attained. Every system measurement
variable, when bumped independently within its full span, contributes to
small percentage of change around the brightness level dictated by the
peroxide charge.
In order to illustrate brightness control feature, a f st set of simulation
results is shown in table 6, representing the effect of chip quality on
peroxide
charges to achieve different brightness setpoints.
CA 02447098 2003-10-28
Brightness Peroxyde chargePeroxyde Peroxyde charge
setpoint (/~) (%) charge (%) (%)
average qualitybest best
chips theoretical experirrnental
ChlpS ChlpS
55 0.77 0.0 0.15
60 1.41 0.0 0.76
i
65 2.22 0.35 1.48
70 4.12 1.21 2.92
71 4.99 1.48 3.54
75 Unachievable 4.96 Unachievable
(max 71%) (max 72%)
Table 6
5 All measurement system parameters (chip properties) were maintained at
their average value and brightness setpoint was bumped from 55 to 75 by
increments of (five) 5 points. For these "average" chips, one can see that a
38% increase in peroxide (from 2.22 to 4.12%) is required to increase the
brightness level from 65 to 70 points (13.7%), and that a further 17.6%
10 increase (from 4.12 to 5%) is required to gain only 1 brightness point
(from
70 to the maximum achievable 71) Doing the same thing with the
theoretical best possible chips as per table 4, one can note that no peroxide
is
required until a brightness setpoint close to 65 is desired. Also, a 71
brightness is achievable with only 1.48% peroxide. Finally, further gains in
15 brightness points from 71 to 75 are only obtained at a high peroxide cost
(from 1.48 to 4.96%). 33ecause one cannot assume that chips with such
properties actually exist, the chip properties contained in the above
mentioned database were also used, which returned the best brightness value
CA 02447098 2003-10-28
16
at 72.46. In this case, the better chip properties still reduce peroxide
consumption for the same brightness level, but to a lesser extent.
When using the method according to the invention, the same
brightness setpoint can be achieved at lower bleaching agent charges when
the chip quality increases. The method may be useful to assist chip
management in the mill, or in the c~ntext of internal model control (IMC) or
model predictive control (MPC) strategies. It is to be understood that dosage
of other bleaching agents such as hydrosulfites may also be performed with
the method of the invention.
