You are on page 1of 9

Chemical Engineering and Processing 46 (2007) 314322

Predicting otation efciency using neural networks


J. Labidi
a,
, M.
`
A. P` elach
b
, X. Turon
b
, P. Mutj e
b
a
Department of Chemical and Environmental Engineering, University of the Basque Country,
Plaza Europa 1, 20018 San Sebastian, Spain
b
Department of Chemical Engineering, Escola Politecnica Superior, University of Girona,
Avda. Llus de Santal o s/n, 17071 Girona, Spain
Received 21 May 2005; received in revised form 29 May 2006; accepted 17 June 2006
Available online 5 July 2006
Abstract
This paper presents a model of otation stage using a neural network to predict the efciency and the effect of operational parameters on
the efciency of ink removing. Two methods are used to determine the kinetic parameters of the otation process using particular experimental
conditions: experimental data obtained at a laboratory level, and simulated data by means of a neural network.
Simulated values obtained with a neural network correspond closely to the experimental results. Neural networks are long-range tools for studying
processes when some knowledge of the phenomena that occur in the process is acquired in order to develop models based on the experimental
results. The neural network model accurately reproduces all the effects of operation variables and can be used in a simulation of a deinking plant
to determine the optimal operational conditions.
2006 Elsevier B.V. All rights reserved.
Keywords: Deinking; Flotation; Neural network; Modeling
1. Introduction
Recycled paper is the main source of cellulosic bre for the
paper making industry. According to data from the European
Council for Paper Recovery and Recycling in 2004 the European
pulp and paper industry recycled around 46.5 million tonnes of
used paper, 1.8 million tonnes more than in 2003. This repre-
sents 53.7% of the total amount of paper consumed. Presently, a
third of the recovered paper is treated in mills with the deinking
system.
Deinking plants are very complex due to:
the large diversity of polluting agents present in the paper,
making the deinking processes difcult;
the lack of knowledge, at the industrial level, about the physic-
chemistry of the different processes involved in the deinking
process, such as otation.

Corresponding author. Tel.: +34 9430171 78; fax: +34 9430171 30.
E-mail address: jalel.labidi@ehu.es (J. Labidi).
2. Deinking operation
2.1. Description of the deinking process
The deinking operation is essentially a purication process
in which the ink, in addition to other polluting agents that can be
found in the recovered paper such as clips, staples, rubber, etc.,
are eliminated. The purpose of this process is to remove all the
polluting agents from the recovered paper in order to produce
a pulp that is free of impurities. The most important steps in
the deinking process are: washing, otation, purication and
screening. The effectiveness of these operations depends on the
size of the particles of the polluting agent which needs to be
eliminated.
A simplied ow sheet of a deinking process is presented
in Fig. 1. Several alternatives are used for deinking recycled
paper. These processes are differentiated by the operations used
to eliminate ink. In the washing stage, ink is eliminated using
washers that consume a lot of water. The otation method com-
bines washing and otation. The washers eliminate small ink
particles, whereas the otation cells remove larger ink parti-
cles. Normally, between the two operations ne screening is
0255-2701/$ see front matter 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2006.06.011
J. Labidi et al. / Chemical Engineering and Processing 46 (2007) 314322 315
Fig. 1. Simplied ow sheet of the deinking process.
used to eliminate small particles and reverse cleaners are used
to eliminate particles with a density lower than 1 (bre suspen-
sion density), as for example sticky and plastic particles.
The deinking process includes an alkaline loop, starting from
the pulper and nishing with the otation cells, propitious to the
ink-bre separation. The second section of the process is an acid
loop, which allows the colloidal matter and the soluble matter
to be precipitated and therefore it can be eliminated later.
2.2. Flotation
Flotation technology was applied for the rst time in the min-
ing industry in the 50 s. Flotation was used to separate mineral
particles from rocks. This technology was later adapted to the
pulp and paper industry.
Flotation is a physicalchemical separation process carried
out in otation cells, it is represented in Fig. 2. An airow is
injected into the bottomof the cell forming small bubbles which
are mixed with the pulp suspension and a surfactant that is added
to facilitate foam formation.
The separation principle consists of inuencing the surface of
ink particles based on their hydrophobic or hydrophilic nature.
If the ink is not hydrophobic it is converted by the adsorption
of the surfactants added to the suspension. The hydrophobic
and converted particles join to the air bubbles and are dragged
in aggregate form with an inferior specic density towards the
surface of the suspension where they form a foam layer.
The process of eliminating ink using otation is complex,
since it involves a mixture of components: ink, water, particles,
bres and loads. The otation process works like a separation
Fig. 2. Experimental otation cell.
316 J. Labidi et al. / Chemical Engineering and Processing 46 (2007) 314322
process where ink particles are eliminated and dragged to the
surface with the foam as a rejected product.
The theory that explains the otation of ink particles from a
physico-chemical point of view has evolved over the last years.
Different theories have been developed to explain the mecha-
nisms that take place in the otation process.
The rst theory was published by Schweizer [1] and com-
pleted by Ortner et al. [2]. In this theory it was assumed that the
surfactant molecules and more concretely soap molecules, were
adsorbed into the ink surface by their hydrophilic end and later
they joined the air bubbles.
According to Bechstein [3] before otation the soap precip-
itates jointly with calcium in the form of akes and the ink
particles (hydrophobic) are adsorbed in these akes and oated
away by the air bubbles. Later some authors demonstrated that a
otation process was not possible as dened by Schweizer and
Ortner without a minimumcalciumconcentration in the suspen-
sion and without considering the fact that the surfactants could
also adhere to the hydrophobic surface of the air bubbles.
Ortner [4] has presented a new mechanism in which the ink
particles and air bubbles both adsorb soap molecules into their
surfaces. The calcium ions act like collectors. Stabilising ink
particles, like hydrophilic compounds, helps to free them from
cellulose bres. The problem that this model presents is that the
calcium ions are also adsorbed into the surface of the bres and
ink particles, which means that the surfactants are adsorbed by
the hydrophilic end and not by the hydrophobic end.
Hornfeck [5] and Larsson et al. [6] have proposed a model in
which the surface charge of particles is reduced due to calcium
ions being absorbed, which are involved in the precipitation of
soap molecules. The result is a completely hydrophobic agglom-
erate that can react with the air bubbles. The ink particles are
surrounded with precipitated calcium soap, like in a microen-
capsulate phenomena. Hornfeck adds to the Larsson model the
possibility of absorbing soap into the surface of the air bubbles
through their hydrophobic end. When air bubbles and ink par-
ticles come into contact the calcium soap precipitates and an
agglomeration bubble/particle forms.
The more recent otation model developed by Putz et al. [7],
considers the effect of the synthetic surfactants that show sig-
nicant differences concerning the mechanism that occurs until
the moment when only the soap takes part as a surfactant. The
proposed model can be applied to hydrophobic and hydrophilic
inks. For hydrophobic ink particles (conventional oil-based ink)
adsorption of the surfactant is independent from its ionic char-
acter. But, if the surfactant is anionic or cationic, ions with the
opposite charge are needed to connect the bubble and the ink
particle, break the hydrated layer and combine the components.
For hydrophilic ink particles (water-based ink), the charge of
the surfactant has a signicant inuence on the hydrophobisa-
tion of the ink particle. Due to the supercial anionic charge of
these types of ink, in alkaline conditions in which conventional
deinking is carried out, the only suitable surfactants are cationic
ones.
The high number of otation models that have been devel-
oped illustrates the complexity of the physicalchemical ota-
tion mechanism.
3. Modeling
During the otation process a number of mass transfer-
ence phenomena take place simultaneously [8] and they can
be described by a kinetic model that includes:
the selective transfer of hydrophobic particles together with
the air bubbles from the suspension to the supercial foam;
the non-selective transfer of both hydrophobic and
hydrophilic particles from the suspension to the foam by
hydraulic capture;
the return of particles from the foam towards the suspension;
the transport of particles fromthe foamto the foamcollection
zone of the otation cell.
In practical terms it is very difcult to separate the contri-
bution made by each mass transfer mechanism with respect to
the overall balance of the otation process. With the objective
of modeling the overall otation process based on time, several
expressions analogous to chemical kinetic reactions have been
proposed (Eq. (1)):
dC
dt
= kC
n
C
m
b
(1)
If the airowrate in the otation cell is constant and high enough
to assure a constant concentration of air bubbles, the free surface
to which the particles to be eliminated can adhere stays constant,
then, expression (1) becomes a rst order kinetic expression
[6,9,10]:
dC
dt
= kC (2)
The kinetic constant of otation (k) depends on the nature of
the particles, the temperature, consistency, the chemical agents
added, composition of the suspension, size of the bubbles, angles
of bubble-particle contact and geometry of the otation cell.
Integrating Eq. (2), from t =0 and C=C
0
, leads to a new
expression that provides the particle concentration in suspension
as function of time:
ln
C
0
C
= kt (3)
The efciency of the otation over time (t) can be determined
by the expression:
F
ef
=
C
0
C
C
0
(4)
Pan et al. [11] have proposed the following expression for the
otation efciency:
F
ef
= 1 exp(kt) (5)
The value of k can be determined fromthe expression (5) as long
as the diameter of the ink particles is constant. Modications
have been made to the rst order kinetic equation to introduce
the concentration of particles that cannot be oated. Arbiter [12]
proposes a modied formula which considers the solid particles
J. Labidi et al. / Chemical Engineering and Processing 46 (2007) 314322 317
that have not been eliminated after a prolonged otation time.
The integrated kinetic equation is given below:
ln
C
0
C

C C

= kt (6)
This kinetic equation was used by Serres et al. [13] and Dor-
ris and Pag e [10] among others to calculate the efciency of
otation of ink particles in brous suspensions.
Obtaining and interpreting kinetic constants must be done
with some caution, since there are many parameters that affect
this constant. In the material purication processes using ota-
tion, the constant increases when the airow rate increases until
it reaches a maximum value. From this maximum value any
increase in the ow rate causes the decrease of the kinetic
constant. The kinetic constant should be considered as a pro-
portionality factor that is necessary to compare the otation of
different types of particles under certain conditions.
More complex models considering the hydrodynamics of the
suspension and the size of the air bubbles and ink particles have
been developed [1418]. Most of these models require param-
eter adjustments which make them impractical since in most
deinking processes the ink particle size is unknown and depends
strongly on the raw material used.
To overcome these problems we propose to use an articial
neural network model to predict the efciency of otation. Neu-
ral networks has been used for several applications in pulp and
paper industry specially for process control [1922]. Only lim-
ited number of applications of ANN to deinking processes has
been reported. Furumoto [23] has patented a device for process
control of a deinked process. Measured variable spectral and/or
physical characteristic values of a waste paper suspension feed
are used as input to an ANN model that determines the opti-
mal settings of the process. Verikas et al. [24] have developed a
ANN for colour classication of images of pulp being deinked.
Smith [25] has reported the use of neural network to adjust the
operation parameters and optimize real deinking plant where the
deinking chemical costs were reduced by 32%.
3.1. Experimental study
The studied experimental variables were: the suspension con-
sistency (c), the airow rate injected into the otation cell (q)
and the agitation speed during otation (N). To carry out the
kinetic study, samples were taken in time intervals of: 0, 3, 6, 9
and 12 min of otation.
During the experiments the following parameters were kept
constant:
Disintegration stage:
suspension volume: 20 L;
suspension initial temperature: 50

C;
agitation speed: 1400 rpm;
consistency of disintegration: 10%;
disintegration time: 15 min;
type of paper: coated and printed paper;
surfactant type: Olinor 1040 from Pulcra S.A.;
percentage of chemical reagents (on dry paper bases)
NaOH: 1%; H
2
O
2
: 1%; Na
2
SiO
3
: 2.5%; EDTA: 0.5%.
Flotation stage:
suspension volume: 50 L;
air pressure in the otation: 1 atm.
The effect of the following variables were studied:
Consistency of otation (c): 0.75, 1, and 1.25%.
Airow rate injected during otation (q): 500, 750, and
1000 L/h.
Agitation speed during otation (N): 850, 1150, and
1450 rpm.
Once the experiment was performed, handsheets were made
with the deinked bres for each combination of experimen-
tal variables to determine the brightness (% ISO brightness)
and ERIC (effective residual ink concentration) of the obtained
paper. Considering the increase in brightness in the otation
stage to be rst order kinetic, in which the initial value (at t =0)
corresponds to the disintegration brightness (B
D
) and the maxi-
mumvalue that can be obtained is the brightness of disintegrated
white paper oated in the same conditions (B
BF
), the kinetic
equation is given by the following expression:
ln
B
BF
B
D
B
BF
B
F
= k
brightness
t (7)
The brightness kinetic constant (k
brightness
) is determined by rep-
resenting ln((B
BF
B
D
)/(B
BF
B
F
)) as a function of time. For
the ERIC, the representation of the experimental values obtained
by analysing the effective residual ink concentration as a func-
tion of time corresponds to rst order kinetic elimination, which
depends on the initial effective concentration (C
0
), and the ink
concentration at innite time (C

). The kinetic equation corre-


sponds to the following expression:
ln
C
0
C

C C

= k
ERIC
t (8)
The kinetic constant k
ERIC
is obtained by representing the values
of ln((C
0
C

)/(CC

)) as a function of time.
3.2. Neural network model
The Neural network used in this work is a module of the soft-
ware Cadsim Plus and was described by Laperri` ere and Wasik
[26]. In this module there can be a maximum of two hidden lay-
ers. The maximum number of neurones in each hidden layer is
100.
A model of the otation operation has been developed based
on the obtained experimental value of ERIC (residual effec-
tive ink concentration) and the brightness (% ISO brightness) of
deinked paper for different operation conditions as inputs: air-
ow rate, time, agitation speed and consistency (Table 1). A set
formed by 135 non-prepocessed experimental data was to train
the ANN while other 15 experiments were used to validate the
model.
318 J. Labidi et al. / Chemical Engineering and Processing 46 (2007) 314322
Table 1
Range of variation of parameters used as inputs for the neural network
Inputs Outputs
Consistency (c): 0.75, 1, 1.25%
Airow rate (q): 500, 750, 1000 L/h ISO brightness (%)
Agitation speed (N): 850, 1150, 1540 rpm ERIC (ppm)
Time (t): 0, 3, 6, 9, 12 min
A number of trial were conducted to determine the optimal
topologies of the network used to model the otation stage by
varying the number of hidden layers, the number of neurones in
each hidden layer, the type of function of neurone activation, the
learning rate and momentum. The best results are obtained with
a neural network formed by two hidden layers each of which
is constituted by 100 neurones. The activation function for both
layers that gives the best result is the sigmoidal function. The
learning ratio used has a value of 0.1, whereas the momentum
was xed at 0. The shutdown criterion is the absolute mean error.
Mean errors of 0.5, 3 and 10% were achieved in the otation
model.
The obtained model was implemented in the dynamic sim-
ulation of the otation process developed with Cadsim Plus.
The owsheet of the simulated process is represented in Fig. 3.
The process is formed by a disintegrator where recycled paper,
water, and chemical agents (NaOH, HO
2
, Na
2
SiO
3
, and sur-
factants) are introduced to produce a pulp with a consistency of
10%. The obtained pulp is diluted until the desired consistency is
reached (0.75, 1, 1.25%) before it is introduced into the otation
cell.
4. Results and discussions
The two methods were compared to nd the kinetic otation
model using particular experimental conditions. In this case the
following operation conditions were selected:
consistency, c =1.25%;
ow rate of injected air, q =1000 L/h;
agitation speed, N=850, 1150, and 1450 rpm.
Table 2 represents ERICand brightness values obtained from
experimental studies and a simulation with the neural network.
We only show the values obtained with the neural network
obtained at 0, 3, 6, 9, and 12 min in order to compare them with
the experimental values, however, the continuous evolution of
ERICand brightness as a function of otation time is represented
in Fig. 4. From Table 2 we can appreciate that the ERIC and
brightness values obtained with the neural network model with
a mean error of 0.5% are very close to the experimental values.
With a mean error of 3% and of 10%, the neural network model
gives higher values than the experimental data. To develop the
kinetic model we have used the results given by the model with
a mean error of 0.5%. These values are introduced in Eqs. (7)
and (8) to determine the value of the kinetic otation constant.
Fig. 3. Simulation of the otation operation with Cadsim Plus.
J. Labidi et al. / Chemical Engineering and Processing 46 (2007) 314322 319
Table 2
Experimental data and simulated values at mean error of 0.5, 3 and 10%
N (rpm) T (min) ERIC (ppm) ISO brightness (%)
Experimental data Calculated value with mean error Experimental data Calculated value
with mean error
0.5% 3% 10% 0.5% 3% 10%
850 0 316.6 316.6 315.5 285.3 78.3 78.3 78.4 79.9
3 241.7 240.4 232.8 226.3 82.3 82.1 82.3 82.4
6 146.3 147.1 136.6 149.8 86.4 86.5 87.2 86.0
9 104.8 106.1 103.5 99.0 89.2 89.2 89.5 89.1
12 98.8 99.5 94.7 74.6 90.1 89.9 90.2 91.0
1150 0 316.6 316.6 315.5 264.7 78.3 78.3 78.4 80.8
3 161.6 164.2 178.1 192.4 84.5 84.6 85.2 83.9
6 78.2 78.0 93.0 123.9 90.7 90.4 90.3 87.5
9 63.2 66.1 75.0 86.0 91.7 92.0 91.7 90.0
12 77.7 69.1 71.1 68.7 92.5 92.5 92.0 91.5
1450 0 316.6 316.6 315.5 253.5 78.3 78.3 78.4 81.2
3 140.1 140.6 159.0 159.7 86.5 86.4 86.4 85.5
6 67.9 66.3 76.9 104.1 91.6 91.8 91.5 88.7
9 49.6 54.9 64.6 76.8 93.4 93.2 92.5 90.8
12 56.1 53.8 62.2 64.4 93.8 93.5 92.8 91.9
Figs. 5 and 6 represent the otation kinetic value obtained
from the experimental values of ERIC and brightness [27] up to
t =9 min.
Figs. 7 and 8 represent the otation kinetic value obtained
from the ERIC values and brightness obtained from the neural
network with a mean error of 0.5%. Observing the results it is
possible to appreciate that the kinetic parameters obtained with
experimental data and those obtained with the neural network
(with a mean error of 0.5%) are similar for both the ERIC and
brightness.
Once the kinetic models were compared, we have studied the
inuence of the different experimental variables (consistency,
agitation speed and airow rate) on the otation operation using
the neural network simulation with a mean error of 0.5%. The
study has been made with the data collected at 12 min which
correspond maximum ink elimination (minimum ERIC value).
Figs. 9 and 10 show that when the otation consistency
increases, ERIC otation decreases and therefore ink elimina-
tion is more efcient. It is also possible to see (Fig. 9) that
Fig. 4. Evolution of the ERIC (ppm) and the brightness (%) over time (min).
Fig. 5. Flotation kinetic value obtained with the experimental ERIC data
(c =1.25%, q =1000 L/h and N=850, 1150, 1450 rpm).
Fig. 6. Flotation kinetic value obtained with brightness experimental data
(c =1.25%, q =1000 L/h and N=850, 1150, 1450 rpm).
320 J. Labidi et al. / Chemical Engineering and Processing 46 (2007) 314322
Fig. 7. Flotation kinetic value obtained with simulated ERIC values (c =1.25%,
q =1000 L/h and N=850, 1150, 1450 rpm).
Fig. 8. Flotation kinetic value obtained with simulated brightness values
(c =1.25%, q =1000 L/h and N=850, 1150, 1450 rpm).
keeping the consistency and airowrate constant, increasing the
agitation speed during the otation operation provides better ink
elimination and lower ERIC values.
The ow rate of injected air has less effect on the ERIC
values than other experimental variables (agitation speed and
consistency). As a general tendency the ERIC decreases when
Fig. 9. Evolution of the ERIC value as a function of consistency for different
agitation speeds and a constant airow rate (1000 L/h).
Fig. 10. Evolution of the ERIC value as a function of consistency for different
airow rates and a constant agitation speed (1450 rpm).
the ow rate of air injected into the otation cell is increased
(Fig. 10).
Observing Fig. 11, it is possible to conclude that a high sus-
pension consistency increases the whiteness of the deinked pulp
and therefore the effectiveness of the otation operation. The
same effect can be obtained by increasing the agitation speed.
Better whiteness of the deinked pulp and effectiveness of
the operation are obtained with high suspension consistencies
(Fig. 12). Variations in the whiteness with the injected airow
rate are less signicant than the other experimental variables
(agitation speed and consistency).
Table 3 summarizes the effect of the process parameters on
the ERIC and the brightness values of deinked pulp. The rel-
ative changes of both parameters are calculated as function of
the maximum change of airow rate and agitation speed for dif-
ferent suspension consistency. The table shows that an increase
of 100% of the airow rate allows a very small increase of the
brightness independently from the suspension consistency. The
increase of airow rate improves the ink elimination for (reduce
the ERIC value) for suspension with high consistency. For low
consistency suspension the ink elimination is not improved with
the airow rate. The agitation speed has more signicant effect
Fig. 11. Evolution of brightness as a function of consistency for different agi-
tation speeds and constant airow rate (1000 L/h).
J. Labidi et al. / Chemical Engineering and Processing 46 (2007) 314322 321
Table 3
Evaluation of the effect of the process parameters on ERIC and brightness
q (L/h) N (rpm) ERIC (ppm) ISO brightness (%)
C=0.75% C=1% C=1.25% C=0.75% C=1% C=1.25%
500 850 149.0 124.0 138.4 85.4 87.4 87.1
1000 850 163.1 122.4 99.5 85.5 87.6 89.9
Variation
*
(=100%) 0% 9.4% 1.3% 28.1% 0.1% 0.1% 3.2%
750 850 156.2 112.7 98.7 84.5 87.6 89.2
750 1450 95.7 80.2 61.4 89.0 91.2 93.3
Variation
*
(=0%) 70.6% 38.7% 28.8% 28.8% 5.3% 4.1% 4.5%
*
Variable variation: =((V
2
V
1
)/V
1
) 100, V
i
: value of the variable (ERIC or ISO brightness).
Fig. 12. Evolution of brightness as a function of suspension consistency for
different airow rates and a constant agitation speed (1450 rpm).
of both ERIC and brightness for all suspension constancies. As
general trend an increase of 70% of the value of N reduce by
approximately 35% the ERIC and increase by 4.5% the bright-
ness.
Finally the experimental and simulated yield of the otation
operation is represented in Fig. 13. Under the conditions studied
there is an optimum yield around 1200 rpm depending on the
consistency of the suspension and the ow rate of the injected
air. The combination of low consistency and reduced airow
rate gives the best results, while an increase in the consistency
reduces the yield [28].
Fig. 13. Experimental and simulated yield of the otation at t =12 min.
5. Conclusions
Two methods were used to determine the kinetic parameters
of the otation process using particular experimental conditions.
The rst method was based on experimental data obtained at
a laboratory level, and the second method was based on data
simulated with a neural network. The results obtained by both
methods are very similar. This fact demonstrates that neural net-
works are long-range tools for studying processes, and some
knowledge of the phenomena that occur inthe process is required
in order to develop models based on the experimental results.
The neural network model accurately reproduces all the
effects of operation variables. It is possible to conclude that:
Generally, the increase in the consistency of the oated sus-
pension decreases the ERIC values and improves the bright-
ness of the deinked pulp.
For constant consistency and airow rate, elevated agitation
speeds during the otation stage, provide better ink elimina-
tion results (low ERIC values and high brightness values).
Working with a high airowrate improves ink elimination but
the airow rate applied during the otation stage generates
less signicant variations in the ERIC and brightness values,
in comparison with the values obtained using the agitation
speed or the otation consistency.
The model obtained can be used to determine the optimal
operation conditions for the otation cell and to simulate the
otation operation.
Acknowledgement
The authors wish to thank the Spanish Ministry of Education
and Science for the nancial support of this work under the frame
of the programme Ram on y Cajal.
Appendix A
B
BF
brightness of disintegrated and oated white paper
B
D
disintegration brightness
B
F
brightness of disintegrated and oated paper
c consistency of suspension
322 J. Labidi et al. / Chemical Engineering and Processing 46 (2007) 314322
C concentration of particles
C
b
concentration of bubbles
C
0
concentration of particles at t =0
C

concentration of non-oatable material at innite time


F
ef
otation efciency
k
brightness
kinetic constant of otation obtained with brightness
k
ERIC
kinetic constant of otation obtained with ERIC
K kinetic constant of otation
q airow rate
n and m exponent
N agitation speed during the otation
t time
References
[1] G. Schweizer, Wochenbl Papierfabr. 93 (19) (1965) 823830.
[2] H. Ortner, R.F. Wood, H. Gartemann, Wochenbl Papierfabr. 103 (16) (1975)
597601.
[3] G. Bechstein, Einsatz von chemikalien und deren auswirkungen
auf physikochemische vorg ange beim deinking-otationprozess, Das
osterreichische Papier 12 (4) (1975) 1622.
[4] H. Ortner, Flotation deinking, in: Tappi Monograph on Recycling, 3rd ed.,
Tappi press, Atlanta, 1981.
[5] K. Hornfeck, Wochenbl Papierfabr. 110 (5) (1982) 542554.
[6] A. Larsson, P. Stenius, L.

Odberg, Surface chemistry in otation deinking.
Par I. The oatability of model ink particles, Svensk Papperstidning 18
(1986) 158164.
[7] H.J. Putz, H.J. Schaffrath, L. G ottsching, Deinking of oil and water-borne
printing inks. A new otation deinking model, Pulp Pap Canada 94 (7)
(1993) 1621.
[8] A.R. Laplane, M. Kaya, H.W. Smith, The effect of froth on otation kinet-
ics. A mass transfer approach Miner, Process. Extr. Metall. Rev. 5 (1989)
147.
[9] G.J. Jameson, Physical factors affecting recovery rates in otation, Min.
Sci. Eng. 9 (3) (1977) 103118.
[10] G.M. Dorris, M. Pag e, Deinking of toner-printed papers. Part I. Flotation
kinetics, froth stability and bre entrainment, in: Proceedings of the Third
Research Forum on Recycling, CPPA, Vancouver, 1995, pp. 215224.
[11] R. Pan, F.G. Paulsen, D.A. Johnson, E. Bouseld, E.V. Thompson, Aglobal
model for predicting otation efciency. Part I. Model results and experi-
mental studies, Tappi J. 79 (4) (1996) 177185.
[12] N. Arbiter, Flotation rates and otation efciencies, Trans. AIME 190
(1951) 791796.
[13] A. Serres, P. Colin, A. Lascar, A speck removal index and a speck detach-
ment index for otation deinking, in: Recycling ForumProceedings, CPPA,
Montreal, 1991, pp. 155159.
[14] N. Ahmed, G.J. Jameson, Flotation kinetics, Miner. Process. Extr. Metall.
Rev. 5 (1989) 7799.
[15] F. Bloom, T.J. Heindel, Atheoretical model of otation deinking efciency,
J. Colloid Interface Sci. 190 (1) (1997) 182197.
[16] F. Bloom, T.J. Heindel, Mathematical modelling of the otation deinking
process, Math. Comput. Model. 25 (5) (1997) 1358.
[17] F. Bloom, T.J. Heindel, On the structure of collision and detachment fre-
quencies in otation models, Chem. Eng. Sci. 5 (13) (2002) 24672473.
[18] F. Bloom, T.J. Heindel, Modeling otation separation in a semi-batch pro-
cess, Chem. Eng. Sci. 58 (2) (2003) 353365.
[19] L. Laperri` ere, C. Leduc, C. Daneault, P. Bedard, Chip properties analysis
for predicting bleaching agents requirements for TMP pulps, in: TAPPI
Fall Technical Conference: Engineering, Pulping & PCE&I, Chicago, IL,
United States, 2003, pp. 513521.
[20] S. Karrila, S. Rezak, Review, Developments and Pulp & Paper Research
Applications of Data Reduction with Neural Networks, TAPPI Technology
Summit, Atlanta, GA, United States, 2002, pp. 171186.
[21] L. JinHee, L. HakLae, Application of articial neural networks to diag-
nosis of web breaks in papermaking, Polpu, Chongi Gisul 33 (5) (2001)
4553.
[22] H.C. Aguiar, R. Maciel Filho, Modeling and optimization of pulp and paper
processes using neural networks, Comput. Chem. Eng. 22 (1998) 981984.
[23] H. Furumoto, Device for controlling a deinked cellulose production plant
with state analysers constituted of neural networks for waste paper suspen-
sion, Patent No. WO9508019, 1995.
[24] Verikas F, K. Malmqvist, M. Bacauskiene, L. Bergman, Monitoring the
de-inking process through neural network-based coulor image analysis,
Neural Comput. Appl. 9 (2000) 142151.
[25] B.A. Smith, A tool for process optimization: neural network software, in:
Recycling Symposium, New Orleans, 1996, pp. 163187.
[26] L. Laperri` ere, L. Wasik, Modeling and simulation of pulp and paper quality
characteristics using neural networks, Tappi J. 84 (2001) 115.
[27] M.A. P` elach, Proc es de destintatge del paper per otaci o. Avaluaci o de
lec` acia deliminaci o de tinta, Doctoral thesis, Department of Agri-
cultural Engineering and Agri-food Technology, University of Girona,
1997.
[28] K. Schwinger, B. Dobias, Inuence de lion de calcium sur la perte en
mati` ere breuse lors du d esencrage par le proc ed e de ottation, in: Pro-
ceedings of the CPPA First Research Forum on Recycling, Toronto, 1991.

You might also like