The separation performance and capacity of zigzag air

classifiers at high particle feed rates

Rosenbrand, G.G.

DOI:
10.6100/IR243640

Published: 01/01/1986

Document Version
Publisher’s PDF, also known as Version of Record (includes final page, issue and volume numbers)

Please check the document version of this publication:

• A submitted manuscript is the author's version of the article upon submission and before peer-review. There can be important differences
between the submitted version and the official published version of record. People interested in the research are advised to contact the
author for the final version of the publication, or visit the DOI to the publisher's website.
• The final author version and the galley proof are versions of the publication after peer review.
• The final published version features the final layout of the paper including the volume, issue and page numbers.
Link to publication

Citation for published version (APA):

Rosenbrand, G. G. (1986). The separation performance and capacity of zigzag air classifiers at high particle
feed rates Eindhoven: Technische Hogeschool Eindhoven DOI: 10.6100/IR243640

General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners
and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.

• Users may download and print one copy of any publication from the public portal for the purpose of private study or research.
• You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying the publication in the public portal ?
Take down policy
If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately
and investigate your claim.

Download date: 25. Ian. 2018

THE SEPARATION PERFORMANCE AND CAPACITY
OF ZIGZAG AIR CLASSIFIERS
AT HIGH PARTICLE FEED RATES

G.G. ROSENBRAND
THE SEPARATION PERFORMANCE AND CAPACITY
OF ZIGZAG AIR CLASSIFIERS
AT HIGH PARTICLE FEED RATES

PROEFSCHRIFT

TER VERKRIJGING VAN DE GRAAD VAN DOCTOR IN DE

TECHNISCHE WETENSCHAPPEN AAN DE TECHNISCHE
HOGESCHOOL EINDHOVEN, OP GEZAG VAN DE RECTOR
MAGNIFICUS, PROF. DR. F.N. HOOGE, VOOR EEN
COMMISSIE AANGEWEZEN DOOR HET COLLEGE VAN
DEKANEN IN HET OPENBAAR TE VERDEDIGEN OP
DINSDAG 25 MAART 1986 OM 16.00 UUR

DOOR

GERRIT GERARDUS ROSENBRAND

GEBOREN TE HEEMSKERK

Druk: Dissertatiedrukkerij Wlbro, Helmond.

 --- -- ------:-----

Dit proefschrift is goedgekeurd door de promotoren:

Prof. ir. M. Tels

en
Prof. dr. ing-habil. F. Ebert
co-promotor
Dr. ir. M.M.G. Senden
Aan mijn ouders,
aan Astrid
Curriculum Vitae
Gerard Rosenbrand werd geboren op 30 juli 1956 te Heemskerk. Hij
volgde zijn middelbare schoolopleidng aan het St. Jansenius Lyceum te
Hulst. In 1975 begon hij zijn studie aan de afdeling voor Scheikundige
Technologie van de Technische Hogeschool te Eindhoven. Het
afstudeerwerk werd in de vakgroep voor Fysische Technologie onder
leiding van prof. ir M.Tels verricht. In 1981 slaagde hij met lof
voor het examen van scheikundig ir.
Van 1981 tot tot 1985 was hij werkzaam als wetenschappelijk assistent
in de vakgroep voor Fysische Technologie. Onder leiding van professor
ir. M.Tels werd het hier beschreven onderzoek verricht.
Sinds september 1985 is hij werkzaam aan het Koninklijke/Shell
laboratorium te Amsterdam.
CONTENTS.

1.INTRODUCTION.
1.1 Zigzag air classification. 1
1.2 Models of zigzag air classifier performance. 2
1.3 Scope of this dissertation. 7
References 9

2.MULTISTAGE PERFORMANCE OF ZIGZAG AIR CLASSIFIERS,

2.1 Introduction. 10
2.2 Characterisation of the classifier separation performance. 10
2.2.1 The separation sharpness. 10
2.2.2 Potential classifier throughput capacity. 11
2.3 Experimental set-up. 12
2.3.1 Classifier channel. 13
2.3.2 Partiele feed system. 13
2.3.3 Airflow regulating and measuring devices. 15
2,3.4 Devices for measuring the partiele mass hold-up. 15
2.4 Experimental procedures. 15
2.4.1 Determination of the separation function <PR• 15
2.4.2 Determination of the mean partiele residence time. 16
2.4.3 Determination of the component separation efficiency
Ef. 16
2.5 Results and discussion. 16
2.5.1 The classifier performance for feeds of identical par-
tic1ès. 16
2,5,2 The component separation efficiency Ef. 30
2.6 Conclusions. 36
References, 37
3.MEASUREMENT OF INTERNAL VARIABLES IN ZIGZAG AIR CLASSIFIERS.
3.1 Introduction. 38
3.2 Some principles of the detection of particles in gas-solid
flows. 38
3.2.1 Light transmission methods. 38
3.2.2 Light reflection methods. 39
3.2.3 Other partiele detection methods. 40
3.2.4 Selection of detection techniques. 41
3.2.5 Processing of detector signals. 42
 3.3 Signal processing by means of correlation function 44
calculations.
3.3.1 Theory. 44
3.3.2 Computer simu1ation of data processing by signal
correlation. 48
3.4 Signal processing in tracer detections. 53
3.4.1 Principles of measuring internal variables. 53
3.4.2 Experimental set-up and data processing proce-
dures. 55
3.4.3 Determination of partiele transition probabili-
ties. 57
3.4.4 Determination of partiele velocities. 68
3.4.5 Local partiele flow rates. 69
3.4.6 Local partiele transition times. 70
References. 71
4.PARTICLE BEHAYIOUR INSIDE ZIGZAG AIR CLASSIFIERS
4.1 Introduction. 72
4.2 Local partiele velocities. 72
4.2.1 Introduction. 72
4.2.2 Meesurement results. 72
4.2.3 Analysis of the behaviour of particles in rising
and falling streams. 81
4.2.4 Comparison of the different classifier geometries. 89
4.3 Partiele transition probabilities and loeal partiele flow
rates. · 91
4.3.1 Introduction. 91
4.3.2 Measurement results. 91
4,3.3 Analysis of the partiele transitions at a elassi-
fier stage. 107
4.4 Loeal partiele transition times. 114
4.4.1 Introduetion. 114
4.4.2 Results. 116
Referenees. 122
5.MODELLING OF MULTISTAGE CLASSIFICATION PERFORMANCE AT HIGH
FEED RATES
5.1 Introduction, 123
 5.2 Description of the classification of feeds consisting
of identical particles. 123
5.2.1 Nomenclature. 123
5.2.2 Calculation of the local partiele flow rates and
of the separation function~R' 124
5.2.3 Mean partiele residence time. 128
5.2.4 Verification of the model. 129
5.2.5 Influence of V and R upon the classifier perfor-
mance. 133
5.3 The classification of binary mixtures of model partieles. 147
5.3.1 Effect of partiele thiekness and density. 147
5.3.2 Predietien of the classification result of binary
mixtures from single component classification
funetions. 149
5.4 Some aspects of the design and sealing-up of zigzag air
classifiers. 150
5.4.1 Selection of the channel geometry. 150
5.4.2 Sealing-up of zigzag air classifiers. 150
5.5 Conclusions. 155
References. 156
LIST OF SYMBOLS. 157
APPENDICES. 162
SUMMARY. 184
SAMENVATTING. 187
1 INTRODUCTION.

1.1 Zigzag air classification.

Zigzag air classification is a separation process in which particles are

classified mainly according to their falling behaviour in an air flow.
Figure l.l.a shows a zigzag air classifier. The channel consists of
rectangular sections joined together at an angle ~ to create a zigzag
shape. A dispersion of particles is fed to this channel. The aero-
dynamically "light" particles are carried to the top by the air flow that
is led through the classifier. The "heavy" particles fa11 to the bottom.
In principle many channel configurations are possible. Some of these have
been patented. Differences in contiguration are for instance:

a)
cyclone

top
product

+t +
Air flo"

d1mensions in cm

Fisure 1.1.
a) Cèaretry of the 90 deg. zigzag air classifier.
b) fr:mi.nant IX3rticle streaml.
c) air flow {X;lttern.

-1-
 variations of the angle a /1.1/-/1.3/, /1.5/, /1.9/-/1.12/,
/1.14/-/1.15/.
- variatien of channel depth over channel length /1.2/, /1.9/,
/1.11/-/1.12/, /1.16/.
-extra air supply at stages /1.14/.
-more product exits /1.10/, /1.13/.
- three-dimensional channels and channels with circular cross sections
/1.3/, /1.7/.
- flattened corners and the introduetion of haffles /1.2/, /1.4/, /1.8/.
Areas in which zigzag air classifiers are being applied are for example
the classification of powders (suitable for partiele dimensions above 40
u), .the food industry (e.g. the separation of veins from leaves in pro-
cessing tobacco and tea) and in the separation of useful fractions, e.g.
paper and plasic from municipal solid waste in refuse processing plants
/1.17/-/1.22/.
Our interest originates from this last application. For this reason the
study that is described in this thesis was dedicated entirely to the
classification of relatively large, foil shaped materials. Financial
support was received from the Commission of the European Communities
within the framewerk of an EEC research programme concerning "Recycling
of waste and thermal treatment of waste".

1.2 Models of zigzag air classifier performance.

Figure 1.1.b schematically shows partiele trajectories inside the

classifier channel. In each sectien separate rising and falling partiele
streams occur. At the lower sectien walls particles move downwards while
rising particles move along the opposite higher wall. At the junctions
between each two sections the rising and falling partiele streams come
together. Here, the particles may continue in their original direction or
change streams and continue in the opposite direction. These partiele
trajectories are the result of the characteristic air flow profile inside
the zigzag channel (figure 1.1.c) /1.15/, /1.24/.
A first attempt by Kaiser /1.23/ in 1963 to describe the classification
of particles in a zigzag channel was based upon the concept of particles
carrying out a random walk from stage to stage. After that, little
rnadelling work was done until in 1978 and 1979 the work of Senden and
Tels /1.15/ ~ /1.24/-/1.25/ was published. They investigated the behaviour
of individual particles in zigzag air classifiers at very low partiele

-2-
concentrations and measured the transition probabilities (i.e. the
probability to move to the next higher stage) of such particles. They
found that these probabilities depended upon the "history" of the
particles. Two different types of partiele transition probabilities could
be distinguished:
pf: the probability to rise to the next higher stage for particles
that have entered the stage in a falling stream (fig. l.2.a).
- pr: the rising probability for particles that have entered the stage
in a rising stream (fig. 1.2.b).
From their experiments it foliowed that, apart from the lower two stages,
pf and Pr were independent of the classifier stage at these low
partiele feed rates.
Senden /1.15/, /1.24/ developed aso called "one step memory" model that
described the classifier performance as a function of the values of pf'
p , the number of stages R, the location of the feed stage V and the
r
rising probability pv of the particles entering the classifier channel.

Figure 1.2.
Partiele transitiro probabilitia>.
a) of particlES entering a stage in
a falling str"€ml.
b) of particles entering a stage in
a rising strer:m.

i, 1-J, J+1: sta~e houndary numlwr~

4>R Figure 1.3.

H I deal
-a- The fractiro of bottan product • 4> R •
0.75 separation
as a functiro of .the superficial air

velocity, vf*
0.50 ·------
a) idEal sep9I'atiro (dotted).
Real
separation b) rea1 sep:u-atiro aJIVe,

1.0

-3-
The performance of the entire classifier was characterised by the
fraction of bottorn product, ~R' from a feed of identical particles that
was obtained at a fixed superficial air velocity vf. Figure 1.3 shows
~R as a function of vf for an ideal classifier and for an actual
classifier, A measure of the partiele residence time was obtained by
calculating the mean number of stages passed by a partiele during its
stay in the channel. It foliowed that the measurèd classifier performance
as described by ~R showed excellent agreement with the ~R values
calculated by means of Sendens model for both the standard 90 and 120
deg. classifiers.
The relation between classification efficiency and potential throughput
capacity was illustrated by calculating ~R and the mean numbers of
stages passed by a partiele as functions of pf and pr (figs. 1.4).
The symbols in the figures represent measured combinations of pf and
pr. The efficiency of the classification was defined by the slope of
the function ~R at its half value:

a<PR d pf a<PR d pr
--+---- (1.1)
opf d vf ()pr d vf

d~/dvf will be largeras the separation efficiency impraves (figure

1.3). Senden showed that large values of Cl<PR/Clpf and 3<PR/3pr and
thus high separation efficiencies were obtained for high values of
pf/pr (upper left corner of fig. 1.4.a). This higher classification
efficiency is obtained at the cost of higher partiele residence times
and, consequently, lower potential throughput capacities (figs. 1.4.b and
c). This result is illustrated by figure l.S. Particles that show high
values of both pf and 1-pr so that pf/pr is high, are seen to
have high probabilities to remain at the same stage for a relatively long
time. These particles have long residence times and are subject to a
large number of single stage classifications. Low values of pf/pr
lead to low partiele residence times and thus to a low mean number of
stages passed by a partiele and low classification efficiencies. It
appeared from measurernents that the partiele residence time for the
standard 120 deg. classifier was indeed much lower than that for the
standard 90 deg. classifier due to P/Pr ratios that were lower for
the standard 120 deg. classifier than for the standard 90 deg.

-4-
 1.0
8.)
0 standard <u deg.
... classifier
• standard lal deg.
• classifier
Pf
mt

~)
Lines of constant

R = 10
V

Pv= I

0.0 completely absorbing

harriers
o.o Pr 1.0

1.0 1.0

b) c)

o.o IL-.,--.....-....,--.---..---.-r--r--r--"' 0. 0 "--...--.--.--.----.----,,---.,--..,---.--"

o.o 1,0 0.0 1.0

Figure 1.4.
P[Pr caibinations of the urxlel particles in the P[Pr diagr/3111 (Sendm /1.15/).
a) for <f!R.
b) [ar mb (llH:lTl Ill.1IIber of transitions for the bottem prodJct particles).
c) for iiit (llH:lTl Ill.1IIber of transitions far the top product particles).

Figure 1.5.

•ij/; lbninant [Brticle flows resulting in

long residence times (a)

lj and in short residence ti.Jres (b).

-5-
classifier. The classification sharpness for the standard 90 deg.
classifier was not significantly higher than that for the standard 120
deg. classifier. Senden showed that this was caused by the lower single
stage classification sharpness dpf/dvf and dpr/dvf of the
standard 90 deg. classifier (see equation 1.1) that suppressed the effect
of the larger number of single stage transitions.
The work of Senden and Tels was limited to 1ow partiele concentrations.
Vesilind and Henrikson /1.26/- /I. 27/ investigated the separation of flat
plastic and aluminium particles in a standard 120 deg. classifier at
higher feed rates. They assumed equal values of pf and pr
(pf=p =p.). The fraction of bottorn product thus could bedescribed
r :t
by the Markovian random walk model:

V-1 m (1-p.)
1
1+L rr
m=l i=1
(1.2)
R-1 m (1- p.)
1
l+L rr
m=1 i=1 pi

Here R is the number of classifier stages and V is the location of the

feed stage. The bottorn stage is numbered 0.
Furthermore, pi was taken to be a function of the partiele
concentration at the partiele stage i:

p.:t = erf (k c.:t I Ct) (1.3)

The error function was applied to describe the relationship between the
partiele concentratien Ci at stage i and the partiele transition
probability pi, as this function has the property of varying between 1
for C/Ct=0 and 0 for C/Ct = oo (Ct being the partiele
concentratien at the highest partiele stage).
Mixtures of plastic and aluminium particles were fed at different feed
rates and the concentration of each component was determined at all
stages from photographs. The air velocity was kept constant in these
measurements. The constant k was calculated for each component and for
each feed rate from equations 1.2 and 1.3 from known values of ei, ct
and <PR. Hen,-ikson and Vesilind found that this value of k was
independen:· · (lf model partiele type, stage and of feed rate.

-6-
They furthermore stated that k is independent of the superficial air
velocity and of the classifier geometry. Their argument was that all
these process variables are already accounted for in the resulting
partiele concentrations Ci.
One of the drawbacks of the work of Vesilind and Henriksou is that it
does not offer the possibility of predicting the fractions of top and
bottorn product for a given feed with a known k because it is not possible
to calculate C and C.• In addition, this model is seen to imply that
t 1
the concentratien profile given by the Ci and, consequently, the
transition probabilities pi are independent of both feed rate Qv and
air velocity vf in cases were the average time that it takes for a
partiele to achieve a transition from one stage to the next is constant.
This is difficult to reconcile with the nature of air classification.
Hence, partiele transition times have to be functions of the partiele
concentrations. However, we found that, for the 120 deg. classifier, the
dependenee of the transition times upon both Qv and vf is not
significant (chapter 4 of this dissertation).
A second disadvantage of their approach is the assumption that a
Markovian random walk model is valid for zigzag air classifiers with
different angles. They aasurne that the history of a partiele no langer
plays a role at higher partiele concentrations. In chapter 4 it will be
shown that, while this assumption is true for classifiers with angles of
120 deg., we cannot confirm it for standard width classifiers withangles
of 90 deg.
Finally, the assumption that partiele transition probabilities decrease
at higher concentrations has been found to be too general for the
standard 90 deg. classifier (chapter 4).

1.3 Scope of this dissertation.

The work described in this dissertation concerns a study of the

performance of zigzag air classifiers at high partiele feed rates. The
separation process for foil shaped materials of which the dimensions are
relatively large compared to the channel dimensions of the classifier was
investigated.
One aim of this study was the characterisation of the separation
sharpness and the potential throughput capacity and the determination of
their interrelations. This relation is discussed in chapter 2 on the
basis of experimental results obtained in zigzag classifiers with angles

-7-
of 120 and 90 degrees. Bath the air velocity vf and the partiele feed
rate Qv were varied. The experiments were carried out using both feeds
consisting of identical particles and binary mixtures. Th~channel width
and the location of the feed stage were varied as well.
The multistage behaviour.of zigzag classifiers was found to betheresult
of the partiele behaviour at the individual stages and of the
interactions between these stages.
A second objective was to gain a fundamental insight into the functioning
of zigzag air classifiers. For this purpose a meesurement method was
developed to analyse the partiele behaviour at the individual stages.
Partiele transition probabilities at the stages, local partiele
veloeities and -flow rates are the main so-called internal variables that
characterise this behaviour.
Chapter 3 provides a survey of the different techniques of detecting
particles in gas-salid flows, as well as of various methods to process
the detector signals and to calculate the internal variables. An optical
meesurement methad based on the detection of tracer particles, in
combination with the use of an on-line computer for data acquisition and
processing was developed and will be discussed in this chapter.
The results of the measurements of internal variables are discussed in
chapter 4. Dimensional analysis was used to derive the relationship
between the internal variables, the process conditions and the classifier
geometry.
In chapter 5, a mathematica! model, which is an extension of Senden's one
step memory model, is presented to describe the classification efficiency
and capacity of zigzag air classifiers. The relationship that was found
to express the transition probabilities as a function of vf, Qv' the
number of stages R and the location of the feed stage V are the input
data of the model. The influence of vf, Qv' channel geometry and
location of the feed stage on the classifier performance which are
described qualitatively in chapter 2 are discussed quantitatively in this
chapter. Design rules based upon model calculations are proposed. More
specific design aspectssuch as the influence of the roughness of the
zigzag walls on the partiele transition probabilities are dealt with in a
more qualitative way.

-8-
References.
1.1 A.H.Stebbins, USA Patent 1,650,727, 28-9-1926.
1.2 A.H.Stebbins, USA Patent 1,861,248, 31-5-1930.
1.3 Carey, F., et al., Brit.Pat.Spec. 468,212, 28-6-1937.
1.4 Lever Brothers Co., USA patent 2,351,351, 13-6-1944.
1.5 T. Eder, Oesterreichische Patentschrift 202087, 10-2-1959.
1.6 Alpine AG, Brit.Pat.Spec. 1,014,723 31-12-1965.
1.7 Scientific Separator Inc., USA Patent 3,441,131, 29-4-1969.
1.8 A.E.Hofmann AG, BRD Offenlegungsschrift 2022036, 17-2-1970.
1.9 Buttner-Schilde-Haas AG, BRD Offenlegungsschrift 1920310,
17-12-1970.
1.10 Alpine AG, BRD Auslegeschrift 1482424, 27-5-1971.
1.11 Vista Chemica! and Fiber Products, USA Patent 3,929,628,
30-12-1975.
1.12 University of Utah, USA patent 3,925,198, 9-12-1975.
1.13 Patent ter inzage legging no. 76 01 930, the Netherlands,
25-2-1976.
1.14 Fastov B.N., Valuiskii P.F., Khimicheskoe i Neftyanoe
Mashinostroenie, ~ (1975), p44-45
1.15 Senden, M.M.G., "Stochastic models for individual partiele behavior
in straight and zigzag air classifiers.", dissertation (1979),
Eindhoven Universiy of Technology.
1.16 Worreil W.A., Thesis (1978), Duke Environmental Center, Duke
University Durham N.C.
1.17 Colon,F.J., Kruydenberg, H., Proc. First World Recycling Congress
(1978), Basel, p3.15.i-3.15.ix.
1.18 Roberg H., Schultz, E., Muell und Abfall, ~ (1974), p263-268.
1.19 Hoberg H., Schultz, E., Aufbereitungstechnik, l (1977), p1-5.
1.20 Diaz, L.F., Savage, G.M., Golueke, C.G., "Resource recovery from
municipal solid waste", Vol.l (1982), CRC-press Inc., Boca Raton,
Florida.
1. 21 Al ter, H. , "Materials recovery from municipal \vaste." ( 1983) ,
Marcel Dekker Inc., N.Y.
1.22 Thome-Kozmiensky, K.J., Recycling International (1982), Berlin,
p188-193.
1.23 Kaiser, F., Chem.Ing.Techn., (1963). no.4, p273-282.
1.24 Senden, M.M.G., Tels, M., J.Powder Bulk Solids Technol. , 2
(1978)' pl6 ff.
1.25 Senden, M.M.G., 2nd Symposium Materials and Energy from Refuse
(1982). Antwerp, p13 ff.
1.26 Vesilind, P.A., Henrikson, R.A., Resources and Conservation, 6
(1981), p211-222. -
1.27 Henrikson, R.A., Thesis (1980), Duke Environmental Center, Duke
University, Durham N.C.

-9-
2 MULTISTAGE PERFORMANCE OF ZIGZAG AIR CLASSIFIERS.

2.1 Introduction.

In this chapter the overall performance of zigzag air classifiers with

angles of 90 and 120 deg. will be discussed. Both the separation
sharpness and the throughput capacity determine the classifier
performance. The relation between separation sharpness obtained in a
classifier and throughput capacity is of first importance. The influence
of channel geometry, location of the feed stage, type of model particles
and feed composition upon the performance were investigated.

2.2 Characterisation of the classifier separation performance.

2.2.1 The separation sharpness.

The separation function ~ characterises the sharpness of the

multistage classification of identical particles. ~ is the bottorn
product fraction that is obtained from a given feed. Figure 2.1 shows
examples of graphs of ~ as different functions of the superficial air
velocity vf and constant solids feed rate Qv. Function -a- of fig.
2.1 would occur if the classifier were to act as an ideal separation
unit. Due to disturbing effects such as velocity gradients, turbulences
of the airflow profile, partiele-partiele interactions, particle-wall
interactions and differences in partiele entry conditions the real ~

function shows the shape of curve -b-. The steepness of the dimensionless
~R(vf/vfSO) function is a measure of the classification efficiency.
This efficiency is defined here by:

Figure 2.1.
I deal The froction of bottem product, \•
-a-
0.75 separation
as a tunetion of the superfic:i81 air
velocity, V f"
0.50 ------- A4>
Kcal
a) ideal separation (cbtted).
ncparut îon b) I13!i1 separation azrve.

1.0

-10-
 (2.1)

vfx is the air velocity at which ~R is x/100. The dimensionless

c1assification efficiency makes it possible to compare classifiers in
which different particles are classified.
Ef characterises the separation efficiency of binary mixtures of
particles. Ef is defined by an equation that was proposed for the first
time by Rieterna /2.1/:

(2.2)

w.1., b and w.~, t are the cumulative component weights in resp. the
bottorn (b) and the top (t) product obtained in a classification
experiment, The index (i=1,2) refers to component i. Ef becomes 1 if
both the top product and the bottorn product consist of one pure component
on1y. For impure top and/or bottorn products Ef will be less than 1.
It follows from eq. 2.2 and the definition of the ~R curve that Ef
equals:

(2.3)

in which ~R, 2 and ~R, 1 repreeent the separation functions of

component 2 and 1 respectively.

2.2.2 Potential classifier throughput capacity.

The potential classifier throughput capacity depends upon the maximum

hold-up H that is acceptable. This value is limited by partiele-partiele
interactions that lead to a decreased separation performance or even to
complete blocking of the channel. The throughput is inversely
proportional to the mean residence time of the particles in the channel:

(\, = H !'i (2.4)

-11-
2.3 ~rimental set-up.

The experimental set-up for the determination of the external variables

consisted of the classifier channel, a feed system that was especially
developed to feed the model particles at a desired and sufficiently
constant feed rate into the classifier, a blower to produce an airflow, a
cyclone to separate the top product particles from the airflow, airflow
measuring and regulating devices and devices for measuring the partiele
hold-up,

Table 2.1.
Widths of the different classifier c:hannels used in the exp-<:riments.

-~------------~--~--~----------~
*: for """"""c1.assifier (~
--
...US) anl feed st:ase locatioo: '1=5.

cyclone

Some relevant partiele properties.

oa:le1 particles

s
dinelsions (01?)

2.002.00.0145
2.0'1.5'0.0291
IO!'.Igbt per unit 0~
surface .....
Ia:l
7/IJ
(g{al
g,,,
l___j product
Dl.S

l cross section
D10
120 A- A'

possible locations
of the feed stage

dimensions 1n cm

Figure 2.2.
The stan.dard 120 deg. zigzag air classifier

-12-
2.3.1 Classifier channel.

In the experiments air classifiers with angles of 90 deg. and 120 deg.
were used. Possible locations of the feed stage were stages 3, 5 and 7.
Figure 2.2 shows a standard 120 deg. air classifier and its dimensions.
For this "standerd" geometry all inward protruding edges of the zigzag
walls are in one vertical plane. Air classifiers with both broader and
narrower than standard channels were used too. Table 2.1 shows the widths
of the different classifier geometries.
The bottorn product particles fall into a drum beneath the zigzag channel.
A cyclone separates the top product particles from the airflow. The top
product is collected in a drum beneath the cyclone. A blower draws in the
air through the zigzag channel and the cyclone.

2.3.2 Partiele feed system.

A feed system was developed to feed the flat model particles (see table
2.2) at a desired and sufficiently constant feed rate into the classifier
channel /2.2/. The system consistedof three sections (fig. 2.3):
- storage vessel and discharge device.
Particles are discharged from the vessel by means of a rotating scraper.
The feed rate is regulated by varying the seraper rotstion speed.
- rotating drum.
This druH levels out fluctuations in the partiele discharge rate from the
starage bin.
- rotary valve.
The particles are fed into the classifier channel by means of a rotary
valve. The valve consists of a rotor in a cylindrical rotor housing. The
rotor contains 4 identical compartments. Rubber flaps seal the walls of
the compartments and reduce the air leakage into the classifier channel.
The feed system was tested with three different serapers /2.2/. Fig. 2.4
shows their performance. The cumulative weight of the particles that were
discharged from the rotary valve was measured in a container that was
suspended from a force transducer. This weight was measured at a sampling
rate of 1 Hz. The mean and the standard deviatieris of the partiele feed
rates were calculated from these data for different seraper rotational
speeds. Seraper 3 showed the best performance. Blocking of the partiele
flow did not occur as was the case for the two other scrapers, and the
fluctuations of the feed rate Qv stayed within acceptable limits. This

-13-
 Figure 2.3.
The feed systen tor the flat llDde1 p3rtieles.

1) starage bin
2) discharge device
3) transport hopper
4) rotating plastic strips
5) and 6) cantraves electric
motor with slowdown
7) rotating drum
il) rot<lry valve
Q) container suspended l'rom
force transducer

Figure 2.4.
Qv The perfOI118llCe of the tested ~·
a) (g/s) y 3 a) The average feed rate as a tunetion of
/1 the seraper rotational srx=ecJ, Vertical
5
,/ · lilles sb.i::M ranges of fluctuations in
4
/ average feed rates.
b) seraper 1 (kd.thout pins P) and seraper 2
(kd.th pins)
3
/ / c) serap;r 3
2 /~~
2
________________ ,
y l ---~-- .. -
0.0 0.1 0.2 0.3 0.4 0. 5 fm (Hz)

b) I

p 0 p

P; pin

I I LBJ~' 11 11

-14-
seraper was used in all our experiments.

2.3.3 Airflow regulating and measuring devices.

A valve between the blower and the cyclone regulated the airflow rate
through the classifier channel. This rate was measured by means of an
Annubar flow meter which was also placed in the tube between the cyclone
and the blower. The Annubar indicated a pressure difference between two
messurement points following the Pitot-tube principle. This pressure
difference was measured both through an inductive differential pressure
transducer and a manometer. The difference corresponds to the mass flow
of air which was determined through calibrating the Annubar by a
rotameter befare the actual experiments were carried out.

2.3.4 Devices for measuring the partiele mass hold-up.

Two pneumatic valves were installed, one at the top in the pipe between
the blower and the Annubar, and the other at the bottorn of the channel.
These could be closed simultaneously to collect the partiele hold-up
after the partiele feed had been shut off.

2.4 Experimental procedures.

2.4.1 Determination of the separation function tR.

At the start of each run the feed system storage bin was filled with the
selected model particles. The superficial airflow rate vf and the
partiele flow rate Q were selected and adjusted by means of the
V
rotational speed of the seraper and the differential pressure over the
Annubar. Both Qv and vf were kept constant during any single
experiment. ~R was calculated from the cumulative weights of the
particles that were collected in the top and the bottorn product vessels
at the end of each run. Qv was determined as the sum of these weights
and that of the partiele hold-up H divided by the duration of the
experiments. Each experiment lasted at least 10 times the mean partiele
residence time. Messurement errors due to instationary starting up
conditions could be neglected this way. Experiments were carried out at a
number of different values of Qv and vf to obtain a range of ~R

values between 0 and 1.

-15-
2.4.2 Determination of the meao partiele residence time.

Partiele residence times Twere determined from Hand Qv according to

equation 2.4. The partiele hold-up H was collected by closing the two
pneumatic valves at the same time and shutting off the partiele feed. lts
value foliowed from the weight of the particles that had been caught
between the valves.
Partiele residence times of individual particles at very low Qv were
measured visually using a stopwatch.

2.4.3 Determination of the component separation efficiency Ef"

Experiments were carried out with binary mixtures of the model particles.
vf and Qv were adjusted and measured in the same way as had been done
in the ~R measurements. The weights of the individual components in the
product streams were determined after splitting each product into its
constituents by means of air classification at recovery efficiencies
above 99.5 %. was calculated by means of equation 2.2. Values of H
were also determined at the end of the experiments.

2.5 Results and discussion.

2.5.1 The classifier performance for feeds of identical particles.

Figs. 2.S.a and b show measured $R functions. The curves were

determined for the standerd 90 deg. classifier and for the standard 120
deg. classifier respectively. The 90 deg. classifier had smooth glass
walls. Qv is the parameter in these figures. A remarkable difference in
the performance of these two classifiers is seen to exist. The fraction
of bottorn product, $R' obtained at a fixed vf is lowest for very low
Qv in the 120 deg. classifier. This fraction increases as Qv
increases. On the other hand, ~R at a fixed vf is highest at Qv=O
in the 90 deg. classifier. The fraction of bottorn product decreases with
increasing feed rates. The increase in ~R at increasing Qv for the
standard 120 deg. classifier is explained by the increase in
partiele-partiele interactions that result from the increased hold-up.
Hence, conglomerates of two or more particles can be formed. These
conglomerates have higher falling veloeities than the corresponding
individual particles and thus tend to fall to the bottom. The decrease in
<PR at increasing Qv for the stanéEJ.rd 90 deg. classifier can be

--16-
<I>R <I>R
(-) (-)
0.8 0.8

0.6

0.4

0.2

1.3 1.3 1.5 1.7

Figure 2.5.
The fraction of bott:all product, <PR as a lunetion of the ~ficial air velocity vf
mèasured inside the standsrd 90 deg. classifier (a) and 120 deg. classifier (b).

b)

Main air
channcl

++ +
Air flow
+ ++
Air flow

Figure 2.6.
Air flow pattem inside the standsrd 90 and 120 deg. classifier.

'T o 'ï
(s) 'XI deg. class:ifier m:xlel !>'Tticles Ç (-> m deg. classif:ier R=lO; V=5; W=lO an
liJ ~=10; '1=5; t-~Paraueter Q_, (glsj
1<.1=14.1 an o o
l(J rodel jm'ticles c2
0
0 0
:ll ) 0 "' 0.67
Paraneter Q" (g/s)
a ,j~o\..ç • 1.5
0/~~ ~ "' 0 2.8
//"--· ~A~
v-"CI""ëfl-=~
..,0 ~
10 10

I I
1.3 1.5 1.7 1.3 1.5 1.7

Figure 2.7.
The l11a91'l partiele residence t:ûre Tas a tunetion of the ~ficial air velocity vf
rreasured inside the st:andard 90 deg. classifier (a) and the 120 deg classifier (b).

-17-
 explained by the characteristic airflow profile that has been measured in
this geometry /2.3/ (fig. 2.6). Particles that move downwards are
influenced by the drag force of the circulation flow along the lower
zigzag walls and thus enter the upward directed main airflow channel at a
relatively high, downwarcts directed partiele velocity. At low partiele
concentrations particles move down freely without much contact with these
walls. Because of their high inertia they may easily cross the main
airflow channel and again fall down to the next lower stage. At higher
partiele concentrations particles can be pushed against the lower walls
by each other. This results in an increased particle-wall friction and a
reduced effect of the drag force of the circulation flow. Hence, the
particles enter the rising air stream at a lower downward directed
velocity and may be taken upward by the main airflow more easily. The
standard 120 deg. classifier lacks this circulating airflow. Particles
more or less slide down along the lower zigzag walls. The velocity of
falling particles is not much influenced by the partiele concentration.
Measurements of both partiele veloeities and partiele transition
probabilities (chapter 4) confirm the above explanations.
Qv and therefore the capacity of the classifier cannot be increased
indefinitely as large conglomerates will be formed. The channel gets
blocked when the feed rate becomes too large. Large conglomerates are
formed mainly at the classifier stage immediately below the feed stage.
e Here, local partiele flow rates will be largest. The flow rate of
particles deseending to this stage becomes larger than the flow of
particles f.alling from this stage. Large partiele conglomerates are
therefore created at the lower zigzag wall of this stage. Eventually the
formation of these conglomerates may lead to the blocking of the entire
channel. Partiele conglomerates occurred at lower Qv in the standard 90
deg. classifier than in the standard 120 deg classifiers. Partiele
hold-ups at the individual stages were higher because particles have
greater probabilities to remain at the same stage for a longer time.
Partiele residence times T are therefore also longer in the standard 90
deg. classifier than in the standard 120 deg. classifier (see figs. 2.7.a
and b).
The Qv at which the formation of conglomerates occurs depends upon the
channel geometry and vf. As a general trend it was observed that
conglomerates were formed at lower Qv when the channel was narrower or
when vf was appr. equal to vfSO" Figs. 2.7 show that T reaches its

-18-
maximum then •. At vf=vf 50 the probabilities of the particles leaving
the classifier through either the upper or the lower classifier exit are
equal.
Blocking of the channel occurred in the 90 deg. classifiers and in the
narrow 120 deg. classifier. In the standard and broad 120 deg.
classifiers large conglomerates formed at high Qv. However, these
conglomerates always fell down to the bottorn exit before they could form
bridges in the channel and block it.
The dimensionless classification sharpness nr (equation 2.1) and the
superficial air velocity for which ~R is 0.5 are quantities that
represent relevant information on ~R curves. nr can be considered to
~R(vf/vf
be a measure for the derivative of 50 ) at vf=vf 50 ,
which equals:

d IPR = a\ d Pf + a\ dpr I (2.5)

d(vrfvf:i) vf.:D a pf Pen d (vrfvf:J vf:IJ a pr pr:IJ d (vrfvf:IJ) vf:IJ

with pf 50 = pf(vf 50 ) and Pr50 = Pr(vf 50 ). This equation can be written by:

(2.6)

where

Af and Ar are the amplification factors of the single stage

efficiencies nf' for falling particles and ~· for rising particles
respectively. Senden /2.3/ showed for low Qv that Af and Ar are
high when particles have high probabilities to remain at the same stage
fora longer time (pf/pr > 1). Af and Ar are low in case
pf/pr < 1. The amplification factors furthermore become larger when
the number of stages R becomes higher and when the feed stage is located
at the middle stage of the channel.
The location of the half value of the ~R function, vf , gives
50
information on the air velocity that is needed for the classification.
Hence, the performance of the different classifiers as described by the
~R curves can be compared by comparing nr and vfSO instead of the
~R curves themselves.

-19-
Table 2.3.
Results of the measurements of the classificatîon functions and residence time functions for the
vatious classifier geomet ries.

a) Narrow 90 deg. classifier with steel zigzag walls.

Feed stage: v-5 Feed stage: V..7
M:xle1 particles s: M:xle1 particles D .5 M:xle1 particles s= nodel. particles D1.5
0" ()"Int •r:o 1)
r ':o 0" Q"/nt •r:o 1)
r T"() 0" Q"lnt •r:o nr T"() 0" Qjm •t:o nr ':o
0 0 1.41 9.8 19 0 0 1.95 8.3 18 0 0 1.45 7.3 15 0 0 2.01 7.4 16
0.51 ll 1.43 8.8 {ij 0.84 12 2.00 9,5 76 0.67 14 1.46 6.4 37 1.0 14 2.(]1 6.7 30
1.0 21 1.51 6.2 45 1.5 21 2.(1! 6.1 45

b) Standard 90 deg. classifier with steel zigzag walls.

Feed stage: V.,J Feed stage: v-s Feed st.a,ge: V..7
M:xle1 particles s: M:xle1 particles S: nodel. I»i'ticles Dl.S M:xle1 particles S
n Q"!m n Q"tm n Q"!m
0" ~m vf"' ':o 0" ':o 0" 0" 1)
r vf;JJ r vf"' r T"() yf"' r ':o
0 0 1.57 6.2 28 0 0 1.53 6.1 39 0 0 2.14 6.2 24 0 0 1.49 5.1 28
1.0 21 1.49 6.5 34 1.0 21 1.39 6.7 41 1.5 21 2.08 6.8 41 1.0 21 1.42 5.8 35
1.5 31 1.49 7,3 28 1.5 32 1.41 8.0 32 2.! 29 2.05 6.4 35 1.5 31 1.37 5.5 32
2.7 56 1.51 5.7 21 2.8 56 1.42 7.9 25 2.7 56 1.34 6.3 28
.7 77 1.52 5.0 16 3.8 79 1.35 6.4 27
5.1 106 1.58 4.0 14

c) Standard 90 deg. classifier d) Broad 90 deg. classifier

with glass zigzag walls. with steel zigzag walls.
Feed stage: v-s Feed stage: V..5
M:xle1 particles s: K:xlel particles Dl.5 M:xle1 particles S• M:xle1 particles 01.5
0" Q"lm vf!J) nr "0 0" Q"lm vf;JJ nr T!J) 0" Q"!m vf!J) Tl r T;JJ 0" Q"lm vf;JJ Tl r T!J)
0 0 1.64 5.6 ~ 0 2.35 4,8 17 0 0 1.13 3.1 9 0 0 1.57 2.8
0.67 14 1.00 7.3 40 1.4 20 2.36 6,2 25 o.n 15 1.12 3.4 9 1.4 19 1.59 3.1 6
1.5 31 1.58 9.0 25 2.7 38 2.1) 10.3 23 1.4 "0 1.13 3.6 8 2.2 l) 1.59 3.4 5
2.8 58 1.55 7.7 21 '(+.7 65 2.23 8.2 22 2.9 00 1.17 3.6 8 4.3 90 1.61 3.4 6
4,8 lW 1.54 7.3 15 5.8 lal 1.16 4.3 6 8.6 120 1.62 4.1 6

e) Narrow 120 deg. classifier with steel zigzag walls.

Feed stage: v.s Feed stage: V..7
!ixlel particles S' K:xlel particles 111. 5 M:xle1 particles s: nodel. particles D1 5
0" Q"/" •rn ~r '!"0 0" Q"lm vf!"O nr T!J) 0" Q"!m vf"' nr T;JJ 0" Q"tm vf;JJ ~r ':o
0 0 1.48 9.2 19 0 0 2.11 10.3 19 0 0 !.47 6.1 22 0 0 2.12 9.0 15
0.43 9 1.54 5.5 33 0.90 13 2.13 9.0 14 0.40 8 1.52 7.2 15 1.2 17 2.17 7.2 14
0.74 15 1.57 5.8 al 1.4 al 2.21 5.0 13 1.2 24 1.58 6.1 17 1.7 24 2.18 6.9 14
1.43 30 1.71 4.4 14 2.5 35 2.37 5.4 9

f) Standard 120 deg. classifier with steel zigzag walls.

Feed stage: v.s Feed stage: v.7
!ixlel particles s· M:xle1 particles D1.5 M:xle1 !8rticles s: nodel. particles Dl.5
0" Q"!m •rn \ Q"!m •m Q"!m n
; Q"~m vf!J) ~, '!"0 t"' 0" llr T!"i} 0" vf"' r T:i)
1.57 7.8 23 0 0 2.27 8,3 17 0 0 1.56 6.7 17 0 0 2.22 8.1
O.ffi 14 l.ffi 8.5 21 l.J 18 2.35 6.8 14 0.43 9 1.58 5.4 0.62 9 2.28 5.3 15
1.2 24 1.67 7.3 18 1.9 26 2.41 6.7 14 0.94 :J) 1.(() 6.2 1.4 19 2.31 5.6 14
2.1 44 1.74 5.5 IS 3.7 51 2.47 5.6 13 2.2 46 1.65 5.0 14 3.0 41 2.34 5.4 10
3.6 75 1.79 4.6 11 5.7 00 2.53 5.4 13 7.1 149 1.85 2.5 9 1.1 154 2.49 4.4 8
g) Standard 120 deg. classifier h) Broad 120 deg. classifier
with glass zigzag walls. with glass zigzag walls.
Feed stage: v.s Feed stage: v.s
M:x!el particles S: M:xle1 particles D!.S - particles s• M:x!el particles 1\.s
0" Q"lm vf"' \ ':o 0" Q"!m vf"' n, '!"0 0" Q"tm vf"' nr t!J) 0" Q"/nt vf"'
1)
r ':o
0 0 1.63 6.4 12 0 0 2.37 7.1 ll 0 0 1.52 4.2 7 0 0 2.16 3.3 6
O,(tl 14 1.70 6.1 16 1.0 14 2.53 6.4 9 1.1 22 1.57 4.9 7 1.2 17 2,18 4.1 5
1.5 31 1.77 5.9 11 2.2 33 2.56 6.0 9 2.5 52 1.61 4.8 8 2.1 19 2.21 3.9 7
2.8 59 1.82 5.0 12 4.4 61 2.67 4,7 8 4.6 96 1.65 4.8 7 4.3 ~ 2.28 4.4 5
5.3 lll 1.92 3.5 9 7.9 110 2.75 4.6 9 8.3 173 1.00 4.2 6 7.7 1(11 2.38 4.0 5
2.0 167 2.43 4.0 5

()": partiele feed rate (g/s) 11_.: - clasificatioo s..r,:ress

m: partiele loleight (g) t'5l: mm j8rticle residence t.ine at vf • •r5l

-20-
The partiele residence time curves that were measured in the various
classifier geometries have been summarized by a single quantity , •
50
'so is the value of the function T(vf) at vf=vfSO'
Table 2.3 summarizes the results of the experiments with feeds consisting
of identical particles.
The separation sharpness nr was plotted as a function of the weight of
particles, Qv' that was classified per volume flow of air, Qf (fig.
2.8). Qf was defined by the product of the classifier cross section
area and vfSO'
Although partiele behaviour inside the standard 120 deg. and 90 deg.
classifier is quite different, the values of nr in zigzag air
c1assifiers with angles of 120 deg. and 90 deg. that have corresponding
channel widths (i.e. both have "narrow" or "standerd" or "broad"
'\. Channel width (cm) I feed stage I
a) 90 deg. wall roughness:
10

8 A: 14.1 I 5 I rough
B: 14.1 I 3 I rough
6
C: 14.1 I 7 I rough

4 D: 20 I 5 I rough
E: 10 I 5 I rough
2 F: 10 I 7 I rough
G: 4 .I I 5 I smooth

0.0 0.01. 0.00 0.12 0.16 <J.IOt (kg/m~

'\ Channel width (cm) I feed stage I

(-}
10 b) 90 deg. wall roughness:

8 A: 10 I 5 I rough
B: 10 I 7 I rough
6
C: JO I 5 I smooth
4 D: 15 I 5 I smooth
E: 6 I 5 I rough
.2 F: 5 I 7 I rough

0.0 0.01. 0.00 0.12 0.16 Q.~ (kg/m~

Figure 2.8.
the classification shar{XJ€!SS nr as a function of the solidB-to-feed ratio Qv/Qf in
different dJannel gearetries of classifiers witb angles of 90 deg. (a) and 120 deg. (b)

-21-
channels) do not differ much. Tllis is remarkable as the partiele
residence times and thus the number of times that a partiele is subjected
to a single stage classification at comparable Qv is much higher for
classifiers with angles of 90 degrees than for angles of 120 deg (compare
the 'so values in table 2.3). This can be explained as follows: The
amplification factors Af and Ar are larger for the 90 deg,
classifiers than for the 120 deg. classifiers as the ratio pf/pr is
larger for the 90 deg. classifiers. Particles are thus subjected to a
larger number of transitions and they remain at the same stage for a
longer time.
The single stage sharpness nr' and ~· are however better for the 120
deg. classifiers than for the 90 deg. classifiers as both pf and pr
are more sensitive to changes of vf. This different single stage
partiele behaviour will be discussed in chapter 4.
The potential throughput capacities of the 120 deg. classifiers are
somwhat higher than those of the corresponding 90 deg. classifiers.
The results of table 2.3 will be discussed in more detail in next
section.

a) Effect of partiele feed rate ~ upon the classifier performance.

At increasing partiele feed rates Qv partiele hold-ups inside the
zigzag channel will increase. The partiele-partiele and partic1e-wall
interactions will become more important and influence partiele behaviour.
The consequences for the classifier performance are discussed below.
Figs 2.9 show values of vf 50 /vfSO,O as functions of Qv/mp for
both 90 and 120 deg. classifiers. Q /m is the number of particles
V p
fed per second. vfSO,O is the value of vfSO for very low partiele
feed rates. Parameter of the figures is the channel width. The value of
vf 50 /vfSO,O represents the relative shift of the half value of ~R
at higher Qv. This value will differ more from unity as the classifier
performance is more sensitive to changes in the partiele feed rate. A
number of conclusions can be drawn from these figures.
Qv influences vf/vfSO strongest and therefore the 90 and 120 deg.
classifiers are most sensitive to changes of Qv when the channel width
is narrower than standard. Broader than standard channels show less
sensitivity to changes in Qv. The reason is that the partiele hold-up
at constant Qv decreases at increasing channel width. Measurements of

-22-
 mean partiele residence times (table 2.3) confirm this.
In genera!, an increase in Qv causes a rise of vfSO because
increasing partiele-partiele interactions leads to the formation of
partiele conglomerates that have higher tendencies to fall. The ~R

functions thus shift to higher air velocities. The standard 90 deg.

classifier represents an exception to this general tendency that was
described in fig. 2.5 and explained above.
Figs. 2.10 show that in general nr values are lowest when the channel
is wider than standard and greatest when the channel is narrower than
standard. The separation sharpness is least sensitive to variations of
Qv when the channel is broader. Qv has a relatively larger influence
upon the location and the steepness of the ~R function in narrower
channels. This is due to the partiele-partiele interactions which are
more important in such cases as the partiele concentrations within the
channel are higher.

Figure 2.9.
vrn,o 9) <leg. c.lassif:iers. R=lO; V=S; model particles c2 ' Effect of the partiele feed
(-)
Parameter: channel width W. rateon the ratio vfsrJv50 , 0
1.1
20011 for different w:idths of the
0
90 deg. classifier (a) and
the 120 deg. classifier (b).

0.8
*: glass zigzag will.s. a)
0.7
100

R=lO; V=S; model particles C

2
0.9 Parameter: channel wirlth W.

0.8
*: glass zigzag will.s. b)
0.7
100

-23-
The classification sharpness nr (figs. 2.10) decreases with increasing
Qv in the 120 deg. classifiers. The transition probabilities decrease
as a result of the formation of conglomerates. In chapter 4 and 5 it will
be shown that pf deereases more than Pr• This results in lower
pf/pr ratios at increasing Qv' as each partiele is subject to a
lower number of single stage classifications during its stay in the
channel. Henee, the amplification factors Af and Ar decrease.
Partiele residence times become shorter at increasing Qv and the
elassification sharpness ~ deereases. A quantitative diseussion of the
influence of the local partiele flow rates upon the partiele transition
probabilities at the individual stages and thus upon the classification
efficiency ~ will be given in chapter 4 and 5.

'\-
(s)
8

'Xl deg. c.lassifiers.

2
* glass zigzag walls.
a)
0
0 100

'\- 120 deg. classifiers. R=lO; V;5; model particles:

(s) c2
Parameter: channel width W,
8

4 15011*

2 b)

0
0 100

F:ispte 2.10.
Effect of the partiele feed rate UfXJll the classification sharfX1€SS n,. for different
widths of the 90 deg. classifier (a) and the 120 deg. classifier (bf.
The increase of n
r with increasing Qv values in the standard 90 deg.
classifiers was explained in the foregoing by the increase of pf that
occurs there. p is much less influenced by an increase in the value of
r
Qv as will be shown in chapter 4. This results in higher pf/pr
ratios and thus higher separation sharpness ~ and longer residence
times T. As soon as the formation of conglomerates causes pf to
decrease, nr again decreases.
High residence times were measured inside the narrow classifiers. The
broad classifiers in general show lower values of TSO than the narrow
and standard width classifiers. Thus, the highest potential throughput
capacities will be obtained in the broad classifier channels. In general,
however, one will have to pay for this higher capacity by a lower
classification sharpness nr.
The partiele residence times can both increase and decrease at increasing
Qv. The reason for this is that the partiele feed rate influences both
the durations of the single stage transitions and the mean number of
transitions that a partiele carries out during its stay in the
classifier. In chapter 4 it will be shown that the transition times
generally increase at increasing partiele feed rates. The number of
transitions depends upon the ratio pf/pr. As has been explained above
partiele transition probabilities can be influenced in sueh a way that
this may result in both a deeresse and an increase of the number of
stages that a partiele passes during its stay inside the classifier, This
depends upon the actual classifier geometry.

b) Effect of classifier feed stage.

The classifier separation sharpness ~ is highest and partiele
residenee times TSO are longest when the particles are fed at the
middle stage of the classifier. Locating the feed stage nearer to either
exit reduces nr and TSO (see table 2.3). It follows from the table
that the influence of Qv upon vfSO is also highest when the feed
stage is loeated in the middle. The number of single stage
classifications to which a partiele is subjected is higher for a central
feed loeation than when the feed stage is located closer to one of the
ehannel exits, Senden /2.3/ found for low Q that in the latter case a
V
partiele may leave the classifier channel within a lower average number
of transitions through the exit nearest to the feed stage. Therefore, the
partiele concentration within the channel will also be higher for a

-2S-
Ctllttai feed stage than for the feed stage located near to one of the
exits. The partiele-partiele interactions resulting from this higher
concentrations will increase and influence nr' TSO and
vfSO/vfSO,O accordingly.

c) Effect of the wall roughness.

Zigzag classifiers with both smooth glass zigzag walls and with rougher
steel walls were used in the experiments (table 2.3). It was observed
that this wall roughness influenced both the separation efficiency and
the throughput capacity. The reason for this is that the friction between
the falling particles and the (rough) zigzag walls influences the falling
veloeities of the particles and thus the partiele transition
probabilities pf. The separation sharpness nr and partiele residence
times T are therefore also influenced, For the standard 120 deg.
50
classifier nr in general is higher for rough walls than for smooth
walls. Apparently the partiele transition probability pf is higher for
rough walls as the particles will slide down at a lower speed and thus
can be taken upward by the main airflow easier when they cross this main
airflow. This results in higher pf/pr ratios as the transition
probabilities pr are influenced less by the wall roughness. The
influence of the wall roughness upon nr for the standard 90 deg.
classifier is less significant. Values of TSO measured in the 120 and
90 deg. classifier were longest for rough walls.
It was noticed that the interaction between the particles and the wall
could also -be influenced when falling particles obtained an
electrostatical charge through friction with the glass walls. Because of
this charge particles were attracted to the lower zigzag walls and moved
downwards at a lower speed. Consequently their rising probability pf
increased. Fig. 2.11 shows the influence of increasing electrostatical
charges upon the location of the~R curves. In principle the separation
in zigzag air classifiers of particles that have small differences in
aerodynamic properties but also have different electrostatical properties
may be improved by electrostatically charging one of the cornponents
selectively through wall friction or by means of applying electrical
fields.

d) Effect of partiele properties.

Table 2.4 shows that vfSO,O' which indicates the location of the half

-26-
 ~R
(-)
0.8

0.6

0.4

0.2

0
1.1

Figure 2 .11.
effect of the electrastatic attraction c.atiiXrl by friction betWEaJ the particles and the
walls on the separation tunetion <IIR.

Table 2.4.
Measured values of flr and v fSO.O for the two types of model particles
inside the various 90 and 120 deg. classifiers.

qle classifier searetrY zigzag loE!ll vf"JJ,C vf~,ll vf:O,D

I') I')
r,C r,D
widt:lt fee:! stage material (m/s) (m/s) (-) H
·f~.c

ro 10 5 steel 1.41 1.95 1.38 9.8 8.3

!0 7 st€el 1.45 2.01 1.39 7.3 7.4
14.1
14.1
3
5
st€el
st€el
1.57
1.53
-
2.14
-
t.lll
6.2
6.1 6.2
-
14.1
14.1
7
5
steel
glass
1.49
1.64
-
2.35
-
1.43
5.1
5.6
-
4.8

--
llJ 5 steel 1.13 1.57 1.39 3.1 2.8

lllJ 6 5 st€el 1.48 2.11 1.43 9.2 !0.3

5 7 !.47 2.12 1.44 6,1 9.0
10 s steel 1.57 2.27 1.45 7.8 8.3
10 7 1.56 2.22 1.42 6.7 8.1
10 5 !.63 2.37 !.45 6.4
15 5 :i: 1.52 2.16 1.42 4.2
7.1
3.3

•f"JJ,C' vslue of vf f<>r Wtidl R.O.S (mlel perticles Czl (m/s).

•f"JJ,D' vslue of •r fOr Wtidl R.o.s (mlel particle& D , ) (m/s).

15
nr,C' clln1eNdaû.ess classificatioo ~ <lll.ldel particles s> <-> .
. nr,D: dinensi<nless classificatioo ~ (mlel rmticles D • ) {-).
15

-27-
value of the ~R curve at Qv=O, increases with the weight per unit of
surface of the particles. Values of the ratio of vfSO for the thick
model particles Dl.S to those for the thinner model particles c2 lie
within the range of 1.38-1.45. This is in accordance with the ratios of
the terminal falling veloeities of these particles in stagnant air that
would be expected on the basis of a simple force balance:

(2.8)

In this equation the left hand term represents the drag force of the air.
Cd is the drag coefficient and A~ is the aerodynamic area of the
particle. vfl is the partiele velocity relative to the velocity of the
air. This force counterbalances the weigth of the partiele minus its
buoyancy (right hand side). A is the product of the two largest
p
partiele dimensions, d its thickness, g the gravity constant and p
p p
and pf are the density of the partiele and of the air respectively. The
terminal falling velocity of the particles becornes:

(2.9)

Hence, for particles that differ only with respect to their thickness
d , will be proportional to ld • The variables that govern the
p p
classification as expressed by 4>R of such particles at very low Qv in
a given classifier are v~ 1 and the linear air veloeities v1 • At the
range of Re values (5.10 -104 ) that were applied in our experiments
the shape of this airflow profile is virtually independent of the
superficial air velocity vf /2.4/. Therefore, v1 is linearly
proportional to vf. Thus, ~R will be a function of vf and vfl'
The following dimensionless number determines the 4>R function:

(2.10)

Particles that have different values of dp will thus have equal values
of 4> R at the same value of the ratio v /v n· For particles that
differ a factor 2 in thickness the values of vfl and vfSO will differ
3 factor 12 = 1.41.

-213-
Apart from meesurement inaccuracies the reasons why the ratios of table
2.4 are not exactly 1.41 are:
- The surfaces of the c2-type particles and the D1 , 5-type particles
are not equally smooth. The particles do not have exactly the same
geometry. Thus Cd is not exactly the same,
- The falling behaviour in air is not the only factor that determines
the classification results. The interaction of the particles and the
wall also plays a role. This interaction depends upon the partiele
surface roughness, the classifier wall roughness and upon the contact
between the particles and the wall.
The influence of the wall roughness upon the values of vfSO for the two
types of model particles is seen quite clearly from table 2.4 for the 90
deg,classifiers. Here, the ratio of the vfSO,O values is less than 12
for all classifiers except for the standard 90 deg. classifier with glass
walls. The interaction of the rough walls and the particles thus levels
out part of the difference in free falling behaviour between the light
and the heavy particles. This levelling out of differences in falling
behaviour is much less for the 120 deg. classifiers. Here the ratios of
vfSO,O for the glass classifiers are approximately the same as for the
classifiers with steel zigzag walls.
In spite of these differences we conclude that that a simple force
balance prediets the location of the ~R curve with an acceptable
accuracy for flat particles in a given classifier when data for particles
that have similar shapes are compared.
The steepness of the dimensionless ~R(vf/v ) curve which is
50
expressed bynr should also be independent of the thickness of the
particles for particles of the same density and the same shape. It
follows from table 2.4 that rather large deviations between the results
obtained with model particles c2 and Dl.S are obtained for the 120
deg. classifiers with channel widths of 5 and 6 cm. In general, however,
the deviations are indeed less than 20 %. In this case too we believe
that, apart from meesurement inaccuracies, these small differences are
caused by differences in surface roughness between the two types of model
particles and by the differences in geometry.
In the narrow 120 deg. classifiers the ratio of partiele dimensions to
channel width (5-6 cm) becomes so large that the relatively small
difference in geometry between the model particles c (2x2 cm 2 ) and
2 2
Dl.S (l.Sx2 cm ) relatively strongly influences nr·

-29-
It follows from table 2.4 that the influence of the partiele feed rate on
the values of nr and vfSO for the two types of model particles shows
a good agreement when the flow rates defined as numbers of particles fed
per unit time are the same.
The residence times that were measured in the 120 deg. classifier
geometries for the model particles of type c2 were slightly longer
(less than 30 %) than for model particles D • (see table 2.3). No
1 5
. important difference between the residence times of these two types of
particles was observed in the 90 deg •.classifiers. These differences are
mainly caused by inaccuracies of the measurements.

2.5.2 The component separation efficiency Ef.

Fig. 2.12.a shows component separation efficiencies as functions of
the superficial air velocity vf measured inside the standard 90 deg.
classifier. Measurement results are represented by the symbols. The
broken curves are comp9nent separation functions that were calculated
from measured separation functions ~R(vf) by means of the following
equation:

Ef 1:

(s)
(-)
0.8 ~
1- b)

0.6 3J

0.4 - 20

0.2 10
t 1
1.1 1.3 1.5 1.7 1.9 2.1 "f (JQ/s) 1.1 1.3 1.5 1.7 1.9 2.1 vf (JW's)

Figure 2.12.
Separation efficiencias and 1ll€6l'l particle residence tinEs, T, as a tunetion of the
superficial air velocity vf IreaStJred inside the standard 90 deg. classifier.
(R=lO; V=5; fred caiJfX.Sition: 50% C
2
50% D •
15
J

-30-
 Table 2.5.

Results of Ef measurements and Ef calculations for the different

classifier geometries.

e) Standard 120 deg classifier with steel zigzag wnl1s.

Feed stage: V=S Feed stage: V=7
%S Q" Q"/m0 Efm Efm,c liEf llEf c %Cz Q" Ojmo Efm Efm,c lltt llEf
0 () 1.0 l.U u.N v.tu U ,U l,U <.>."" U,..,
50 0.8 13 1.0 1.0 0.63 o.oo 50 0.6 10 0.99 1.0 0.70 0.70
50 1.6 27 1.0 1.0 0.64 0.68 50 1.0 16 0.\le 1.0 0.70 0.70
50 3.1 51 1.0 1,0 0.68 0.73 50 2,7 45 O.'f/ 1.0 0.71 0.71
50 6.3 !05 0.93 O.'f/ 0.67 0. 73 50 9.7 161 0.80 0.85 0,48 0.68
so 9.5 158 O.'ió O.'f/ 0.65 0.73

f) Standsrd 120 deg. classifier g) braad 120 deg. classifier

b) Standard 90 deg. classifier c) Broad 90 deg. classifier with glass zigzag walls. with glass zigzag walls ..
with stel zigzag walls. with steel zigug walts ..

Feed stage: v..s Feed stage: V..S

I
(.Al
......
%S Q" Q"/mp Efm Efm c llEf llEf c lS Q" Ojm0 Ef Efm c liEf llEf c
I 0 0 1.0 1.0 0.62 0.62 0 0 0.87 0.87 O.lo4 0.44
J) 1.4 21 0.% O.CJ9 0.55 0.59 ;n l.Z 17 0.112 0.87 0.45 0,47
~:~ ~ g:~ g:~ g:~ g·~
20
g:~ g~~ g·~ g:~
al 1.9 211
20 20 4.0 00
20 8.6 127 o. 76 0.88 0.36 0.50
50 1.3 21 0.% O.CJ9 0.61 0.65 50 1.1 18 0.83 0.87 0.42 0.45 ;
50 1.9 32 0.95 0.99 0.00 0.67 so 1.8 31 0.84 0.87 0.47 0.47
so 3.5 00 o.ro O.!S 0.41 0.65 50 3.2 53 0.8J 0.85 0.44 0.47
50 6.5 100 0.83 0.86 0.41 0.48 .5
:; 1,1
1.7 32
~~ U.',H U.W U,/L U,/1
0.% O.!S 0.67 0.72 ~ ~:;"~i g:[; g:~ g:~ ~:: 8J
80
1. 7 31 0.99 1.0 0.67 o. 70
3.2 59 0,99 1.0 0.66 0.70
80
ro
s1
4.5 84
o.ro o.93 o.54
o.93 0.93 0.54
8J 2.8 53 0.95 0.95 0.55 0.70 8J 3.2 00 0.86 0.89 0.45 0.4& ro 5.8 1(17 o.95 o.95 o.61 o.68 80 6.4 121 0.89 0.9J 0.59
ro 4.0 'ió o.ro o.95 o.42 o.68 ro 6.4 121 0.83 0.85 0.38 0.42

d) Narrow 120 deg classifier witn steel zigzag walls

Q": jmticle feed rate (g/s), mp: wight of nrx!el partiele (g).
Feed stage: v..s Feed stage: V=7
Ea,.: lioldnun Ef value (~) (-).
%S Q" Qjm, Efm
Efm c llEf llEf c %C2 Q" Q"/mn Efm Efm c M:fm llEf C
liEf: width of Ef( vf) curve
Efm,c'
liEf ,c:
i'llxilwn Ef val..., (cakulate<l) (-).
width of Eé vf) curve
0 0 1.0 1.0 0.63 0.63 0 0 1.0 1.0 0.65 0.65
20 1.7 10 1.0 1.0 0.64 0.67 at tt=O·S (~) (m/s). at Ec=D.S (calculate<J) (m/s).
20 3.5 20 O.CJ9 1.0 0.61 o.&:~
1.0 1.0 '!·::? u.59
~
1 1.1 18 50 2.1 10 1.0 1.0 0.65 0.64
1.8 31 1.0 1.0 0.55 0.59 50 2.3 22 O.CJ9 1.0 0.63 0.62
8J 1.2 10 0.99 1.0 0.00 0.61
8J 2.8 22 O.CJ9 1.0 0.55 0.57
 (2.11)

In which 4>R,D refers to the heavy component Dl.S and lP R,C to the
light component c • The partiele feed rates of the single component
2
separation functions were equal to the feed rate of that component inside
the binary mixtures. The figure shows that Ef can be calculated rather
accurately from these ,PR values when the partiele feed rates are not
too high. At higher partiele feed rates deviations of the calculated
efficiencies from the measured efficiencies occur, especially at low air
velocities. This is due to the fact that interactions between the two
different kinds of particles have not been taken into account by

Table 2.6.

Cootparison of the results of Ef measurements and Ef calculauons for the different classifier
geometries.

a) 90 deg. classifiers (steel zigzag walls).

W.IO; V..S. 11=10; v-7. W.l4; v-s. \ó.:D; v-s.
:tS Qjmi1_Rfm Rfw Q"/m0 Rfm l!fw fVmv Rfin Rfw '\,/mD Rfm Rfw
;Al ll U,W l.U
; ~:: ~:~ ~ ~:~ ~:: ~ ~::: ~:::
~ 0.9> 0.71 ID O.'XJ O.'XJ
127 O.l:l'> 0.72
:n 12 1.0 1.0
~~ g:: g:: 21
32
ID
o.<J/ 0.94 18
0.96 0.9:> 31
0.92 0.63 53
0.95
0.97
0.94
0.93
1.00
0.94
lal o.97 o.as
8l 12 0.9> 0.~ 14 0.99 ':'·95 21 0.99 1.0 14 0,94 O.'J{)
21 0.99 o.oo 32 0.9> 0.93 31 0.94 0.87
53 0.'1/ 0.79 ID 0.97 o.oo
76 0.92 0.62 121 o.oo 0.9:)

b) 120 deg. classifiers (steel zigzag walls unless mentioned otherwise).

W.S; v.s. W..S; V=7. W.IO; v..s. w.-10: v-7. jw..lO; v.s (glas) W..lS; V.S
IS '\,/mp Rfm Rfw <J"Im0 Rfm Rfw '\,/mD Rfm Rfw '\.fmo Rfm Rfw flv1"~tRfm Rfw Q"/m Rfm Rfw
0
Zl 10 1.0 0.96 14 1.0 0.89 18 1.0 1.0
Zl 0.99 0.00 :n 0.99 0.84 31 0.9} 0.93
57 0.9} 0.10 ~ 0.97 o.as
lal 0.95 0.62 221 0.74 0.64
X) 18 1.0 0.93
31 1.0 0.93
10 1.0
22 1.0
1.0
1.0 ~ ;:g g:~ :~
51 1.0 0.93 45 0.97 1.00 55
g:: i:~ ~ g:: g:~
0.89 0.95
105 0,96 0.92 161 q.94 0.74 107 0.97 O,'XJ
~ 0.78 0.89
8l 10 0.99 0.00 14 1.0 0.99 31 1.0 om
22 0.99 0.96 31 0.9} 0.96 51 0.97 0.93
59 0.99 0.94 84 1.0 0.95
107 o.97 o.<xJ 121 0.99 1.00

Q,: j>li'ticle feed rate (g/s).

mp: -eight of j>li'ticle (g/s).
Rfm: ratio of ~ to calculated llBlCinun \...:WO (-),
Rfin: ratio of~ width to calculated width of\ cune (m/s),

-32-
eq. 2.11. At low vf falling heavy particles will drag light particles
downwards to the bottorn exit. The effect of heavy particles dragging
light particles down is strengest when the feed rate of heavy particles
is high. Thus this phenomenon occurs when both the total feed rate and
the fraction of heavy particles in this feed rate are high.
Light particles may also entrain heavy particles up towards the top exit
as can be seen from the deviation between measurements and calculations
at higher vf (right side of the Ef curve). This phenomenon is however
less important than that of heavy particles dragging down the light ones.
Fig 2.12.b shows the partiele residence times that were obtained in the
same experiments of which the Ef valnes were plotted. Two maxima occur,
one at the air velocity at which the ~R value of the light component is
appr. 0.5 and the other at appr. the vfSO of the heavy component. At
low air veloeities mainly light particles are present inside the
classifier channel, whereas mostly heavy components are present at high
air velocities. In our view this tends to strengthen the credibility of
our assumption that the Ef curve can be considered to be composed of
the <PR curves of the two components, each taken at its partial feed
rate. An important question is to what extent the Ef curves may be
predicted from the <PR curves, Measurements of Ef were therefore
carried out in classifiers of different geometries. Both the feed rates
and the composition of the feed were varied. The measurements were
compared to values calculated from <PR measurements according to
equation 2.11. Table 2.5 contains the results. Both the maximum and the
width at Ef=D.S have been tabulated for the calculated and the measured
Ef curves. Table 2.6 contains the ratios of the maximum Ef value that
was measured to the maximum calculated value, Rfm and the ratio of the
measured width of the Ef curve to the calculated width, Rfw'

a) Effect of feed rate.

It follows from table 2.6 that the agreement between the calculations and
the measurements is quite good at feed rates that are not too high.
Values of Rfm were generally greater than 0.9 and values of Rfw (=
6. E/''1Ef,c) were greater than 0.85 for feed rates below appr. 50
particles per second. At higher feed rates larger deviations may occur.
These deviations are due to the interactions between the two types of
model particles that are more intense at higher Qv.

-33-
b) Influence of the classifier geometry.
Both the channel width and the location of the feed stage were varied in
the experiments. Figs. 2.13 show Rfw as a function of the partiele feed
rat~ with the channel width as parameter. The deviations between

exper~ments and calculations for the 90 deg. classifiers are much smaller
in the broad channel apparatus than in the classifier that has standard
width. The reason for this is that at equal Qv the partiele
concentrations are lower inside the broad channel than in the standard
channel. Consequently, the interactions between the two different types
of particles are less than in the broad 90 deg. classifier.
For the 120 deg. classifiers Rfw does not depend to any significant
degree upon the channel width as can beseen from figure 2.13.b. In these
clasifiers the differences between the partiele residence times for the
different channel widths are much smaller than in the 90 deg.
classifiers. The influence of interactions between the two model partiele
types will therefore d~pend less upon channel width.
The partiele residence times decrease when the feed stage is located near
the bottorn exit or the top exit rather than at the middle of the
classifier. Consequently less partiele-partiele interaction occurs when
the feed stage is mounted closer to either classifier exit. However, the
influence of the location of the feed stage upon the partiele residence
times is so small that changing the feed stage location does not much
influence Rfw"

c) Effect Qf feed composition.

Experiments were carried out with three different feed compositions of
c2 and Dl.S that contained 20, 50 and 80 % c respectively (based
2
on numbers of particles). The feed composition influenced the values of
Rfw (and thus the agreement between the experimental and calculated
results) for the standard and broad 120 deg. classifiers. However, feed
composition had little influence on Rfw in the 90 deg. classifiers.
Figs. 2.14 show Rfw as a function of Qv/mp for the standard 120
deg. and the standard 90 deg. classifier. It is seen that in the 120 deg.
classifier values of Rfw are largest for feed streams that contain
large fractions of the light component c2 • Apparently light particles
(that will mainly be present in the rising partiele flows) have little
influence on the falling motions of the heavy particles. We believe that
few heavy particles have upward directed veloeities and that the

-34-
Rfwr---r---~--~--~~~--.---~
{-) 10 cm Figure 2.13.
I.o~-o---------- ---
::;-==----0 l'anm!ter: chwlel w:1dth 'Ihe ratio of the rreasured and the calculated

0.9 "-\~ 0, 20 cm
w.idth of the Ef curve, Rfw' as a functim
of the partiele feed rate far different channe1
0.8
w.idths of the 90 deg. cla<;sifier (a) and the
0.7 120 deg. classifier (b).
a) t:. 14.1 cm
0.6
90 deg. classifiers

Rfw
H
1.0 --=iJl:::::;-.::.-------------
olltl .• -• "'

-
b)
0.9 "' -·
c -..
0.8
120 deg. classifiers ........
0.7 parameter: channel width ' ""'- ........
0 W=S cm 0
10 • 10 (glass ..atls)
0.6
"'0 15 (20 % C )
1111 15 (80 %) 2
0 40 80 120 160 Qv/mp (Hz)

Rfw
{-)
1.0
Rfw
(-)
1.0
.
']~--------- ~------;;~.~---

0.9
A ~a) e\ 0 R=lO; V•5; W=l4.1 cm.

0.8
0
'à- 0.9

0.8
o\
1:>.
c
furanet:er: feed canpo-
s1twn (% 2)

.. "'\ 0 20

0.7 0.7 - ~0 ~:
o;6 :00 deg. classif:la:s. 0.6 b)
Jl;.lO; V=5; W.10 cm.
Panm!t:er: feed ~tien (%
0.5 o.s
0 40 ID <!.'\ (Hz) 0

FiB!!!:.e 2.14.
The ratio of the rreasured and the calculated w.idth of the Ef curves, Rfw' as a
functim of the feed carpasitim tor different feed rates in the 90 deg. classifier (a)
and the 120 deg. classifier (b).

-35-
veloeities of these particles that move upwards are small. Collisions
with falling particles cannot lead to a significant increase in upward
velocity for these heavy particles. However, the rising velocity of the
light particles may much be decreased. Larger deviations between the
calculated and measured Ef curves occur when the feed contains a higher
fraction of the heavy D1 • particles. In that case the heavy particles
5
drag the light particles down with them. In the standard 90 deg.
c!assifier the rising particles have much higher veloeities as the local
air veloeities inside the upward air flow are much lower. They may
therefore more easily influence the trajectories of the heavy falling
particles.

2.6 Conclusions.
The separation sharpnesses nr that can be reached in the standard 90
and 120 deg. zigzag air classifiers are roughly equal. The capscity of
the standard 120 deg. classifier is however somewhat larger.
In selecting a classifier for a specific separation process the capacity
of a classifier is usually expressed by the the amount of feed that can
3
be processed by 1 m of air, The results are given in figure 2.8. This
figure also shows that the 120 deg. classifiers have higher capacities
than the 90 deg. classifiers.
The separation sharpness decreases and the potential capacity increases
when the classifier channel is broadened or when the feed stage is
located more towards one of the classifier exits.
nr can be influenced by means of influencing the interactions between
the deseending particles and the lower zigzag walls. This can be done
for example by means of the wall roughness.
The classification efficiency Ef for binary feeds can be predicted with
an acceptable precision from the~R curves of the individual
components. This predietien will be better when the hold up of particles
inside the channel will be lower.

-36-
References.
2.1 Rietema,K., Chem.Eng.Sci., l (1957), p89-96.
2.2 v. Wouwe, F., M.Sc. Thesis (1982), Eindhoven University of
Technology.
2.3 Senden, M.M.G., "Stochastic models for individual partiele behavior
in straight and zigzag air classifiers.", dissertation (1979),
Eindhoven Universiy of Technology.
2.4 Kaiser, F., Chem.Ing.Techn., 35 (1963), no.4, p273-282.

-37-
3 HEASUREHENT OF INTERMAL VARIABLES IN ZIGZAG AIR CLASSIFIERS.

3.1 Introduction.

In chapter 1 it has been made clear that suitable detection techniques

have to be applied in the investigation of individual partiele behaviour
at the different stages of zigzag air classifiers. Such a detection is
required to obtain information on important internal variables such as
partiele transition probabilities, -veloeities -flow rates and
-concentrations. The particles that are to be detected are relatively
2
large, flat (2x2 cm ) pieces of cardboard. In this chapter a number of
detection techniques that are known from literature are discussed
tagether with the advantages and drawbacks of each method. The ehoice of
the methods that were selected for our investigations will be motivated.
The experimental procedures for measuring the relevant internal variables
will be diseussed.

3.2 Some principles of the detection of particles in gas-solid flows.

3.2.1 Light transmission methods.

Particles are detected when they cross a beam of light focused on a

photosensitive detector. Two modes of operation are possible:
- a broad beam of light is used (/3.1/ and /3.2/).
A light souree is placed in the focal point of a postive lense, thus
producing a parallel beam of light which is much wider than the sizes
of the particles in the two-phase flow. A secend lense in front of the
photodetector focuses the light on the photosensitive area of the
detector. The decrease in the light intensity registered by the
photodetector is proportional to that area of the photodetector which
is covered by the projection of the particles in the beam (see figure
3.l.a).
a narrow beam is used /3.3/ and /3.4/.
In this situation the beam width ,is narrow compared to the particles.
The detector signal will show a high-low character, depending upon
whether the beam is interrupted by a partiele or not.
The measurement methad that uses .a broad beam can register a number of
particles at one time. Local partiele concentrations may be determined in
this way. The disadvantage of both methods is that loss of information
may occur when a-symmetrically shaped particles rotate or oscillate or

-38-
when particles overlap. The scattering of light by particles may also
cause inaccuracies.

3.2.2 Light reflection methods.

A light souree illuminates particles. These particles are detected when
the light they reflect falls on a photodetector. The intensity of the
reflected light depends upon the material properties of the partiele and
upon its surface structure. Particles that have smooth surfaces reflect
more or less coherent beams whereas light is scattered in various
directions by particles with rough surfaces. Different modes of operation
are again possible. Some of these are:
- light souree and detector are placed with their optica! axes
perpendicular to each other.
The light souree emits a narrow beam of light. If the partiele surface
is completely smooth, only particles that are favourably orientated
will reflect light onto the detector. Light reflection will be more
diffuse for particles with a rough surface. The determination of the
exact partiele location becomes less sensitive to small deviations from
the most desirabie partiele orientation then. At the same time some
inaccuracy is introduced into the determination of the exact partiele
position. In practice one might well consider applying this method to
measure particles with a fixed spacial orientation such as for instanee
the particles that slide down the lower section walls between the
stages of zigzag air classifiers.
-A diffuse light souree is used /3.3/-/3.6/.
Particles with a rough surface scatter light in various directions.
When a diffuse souree is used the intensity of the reflected light will
be quite high over a fairly large spatial angle. A detector is used
which has been provided with a diaphragm. This detector will only
receive radiation which enters along its optica! axis or which shows
but a small deviation from this direction (fig. 3.l.b).
An important advantage of both techniques is the possibility to
discriminate between particles which reflect much light (white particles)
and particles that reflect little light ( black particles). This offers
the means to study white tracer particles in a main stream of identical
black particles and thus enables us to determine partiele transition
probabilities. In applying either method it is also possible to use two
glass fibres, one to transmit the light radiated by the light souree and

-39-
one to conduct tbe reflected light toa pbotodetector /3.7/.

3.2.3 Other partiele detection metbods.

a) Registration of partiele motions by video camera recordinga /3.8/.

Particles are illuminated by a diffuse light source. A video camera
records the partiele motions. The partiele motions can be analysed only
whçn the identity of eacb individual partiele in tbe surrounding swarm
can be recogni~ed unambiguously. This will be the case at low partiele
concentrations or wben wbite tracer particles are recorded in a stream of
black particles that are otherwise identical to tbe white ones. Tbe
accuracy of this method for determining partiele veloeities is limited
due to the low replacement rate of the picture (every 1/25 second a
completely new picture is taken). Tbis metbod is difficult to automate.
Nevertheless. tbis metbod is applicable as an independent check on
automated registration metbods that are based on different principles.

Figure 3.1.
a) ~ a) Light transmission l1li36S!1ffllmt

classifi..er ~<> using a braad beBm of light.

b) Light reflection l1li36S!1ffllmt
using a diffuse light .source.

Î
air fl<JW

b)

t
air flow

-40-
b) Registration of particles by capacitive detectors /3.1/. /3.9/-/3.11/.
The presence of a partiele is registered due to the difference in the
value of the dielectric constant of air and solid respectively. Two
capacitor plates are placed at either side of the flow channel. This
capacitor is part of an LC-circuit. The resonance frequency signal of the
circuit is measured. This frequency signa! depends upon the partiele
volume between the plates, the dielectric constants of air and solid, the
dielectric constant of the wall material and the area of the plates. It
can be shown that the capacity change is linearly proportional to the
partiele volume concentration if only this concentratien varies.
Fairly large plate dimensions are required to measure capacity changes
with sufficient accuracy. The averaging character of this method makes it
impossible to discriminate individual partiele motions and parallel
countercurrent flows in the measuring volume. This limitation is
inacceptable for measurements in air classifiers in which it is essential
that falling and rising particles are registered separately. An important
drawback is also the sensitivity to the moisture content of the air and
of the particles which is due to the relatively high dielectric constant
of water.
c) Ultrasonic messurement /3.2/.
The two-phase flow is homogeneously irradiated by sound waves. Sound is
detected by a piezoelectrical element. The intensity of the detected
sound is a measure of the partiele concentratien between the sound
emitter and the detector. Overlap of particles and partiele rotations
causes errors in data obtained through this measurement.
d) Piezoelectrical measurement /3.12/.
The impact energy of the collisions of particles with a piezoelectrical
element is measured. This method yields the partiele impact velocity for
particles of a given mass. The element has to be placed in the flow which
causes disadvantageous disturbances of the partiele/air flow.
e) Radioactive aeasurement /3.13/-/3.15/.
Particles doped with a.radioactive material are added to the bulk
partiele flow as tracers. These tracers are detected using suitable
detectors. Drawbacks of this method are that special precautions have to
be taken. Moreover, the method requires considerable investment.

3.2.4 Selection of detection techniques.

The various messurement methods have been briefly summarized in table

-41-
3.1. Two methods were selected for use in this investigation: the light
transmission measurement of a narrow beam of light and the light
reflection method. The main reasans for choosing these techniques were:
-Both methods measure local partiele behaviour without disturbing the
flow in the zigzag channel.
-Photosensitive detectors can easily and inexpensively be constructed
using phototransistors in a suitable electronic circuit.
-The light reflection method enables separate detection of tracer
particles which have light reflecting properties that differ from those
of the main stream. This possibility is of great importance in
determining partiele transition probabilities.

3.2.5 Processing of detector signals.

The selected detectors produce electric signals (electric currents or
electric voltages) which have to be processed in order to obtain useful
information on the partiele behaviour inside the classifier channel.
~
summary of the measurement methods discussed in chapter 3.

W""""ticn ~ """' than I dEt.Ector disad1lantages

obtain!d bv <:apeble of ~: I
I detecto,: veloc. coocentr. tr .prob.

Ji&ht traJ19llissioo change in Ji&ht ~ty

4 2 4
(l:>ro>dl:!ght:beeln) of beeln int.errupted by a c + + 3.2.1
group of jmticles

light l:n!OOm",;;ioo change in 1:!ght: in-ty

(narrow beeln) ofbeelnin~by· +2 .4.5 3.2.1
~ partiele

1:!ght: refla::tioo ~tyofl:!ght: p +2,3 +3 + 3.2.2

(focust!d radiation) refla::ted by a partiele

2 3
lightreflecticn ~tyofl:!ght: p + + + 3.2.2
(diffuse radiatioo) refla::t:ed by a partiele

videc partielellllt:ioos~ la!xriousnethod 3.2.3

difficult to automte
2
caracitive chenge of dielectric cm- + rolocal""""""""t 3.2.3
stant ~ by Jmticles sensitive to HDisture

ultr:os:xuc change in """' intalsity 4 4

~ Ay alloorptioos and c l + 3.2.3
refla::tims by part.icles

piemelactrical .......,_ of partiele lf1li +6 +6 ]lT'Oie disturbes 3.2.3

ooJ.l.isi.a1'l fl<>< pattem

radicective radiationEIIlittedby p +3 +3 extra safety 3.2.3

tnoeer part.icles precau!;ioos

t: nethod ia suitabl.e p: preseoce/oo preserx;e of partiele mv: partiele :iJq>tWle

-: nethod ia - suitable c: partiele COilCSltnlti<JI

1: ooly t:be aetiXlds tmt all<>< trac~ are t:alo!n 4: diaturiBn:es OCOJr W'E!l particles are nat S)'lll1letri<:a
to be suitable. IX" .n,n particles .".,..lap

3: .......n.ng t:be -vioor

2: by """"" of COJ.TOlatlm fln:tion calculatioos
of iR:Iiv:idual particles
5: froD fl<>< ratel and veloeities of part.icles
6: infonatial fron I detector
The character of the partiele motions is stochastic in the sense that
partiele motions are subject to all kinds of influences of which no
functional description is known. Such influences are for instanee the
turbulence of the air stream, partiele-partiele interactions and
interaetions between the particles and the wall. Partiele veloeities and
-concentrations are therefore neither constant nor unambiguously
dependent upon time and place. The signals being received and transmitted
by the detector must therefore also be stochastic.
A first step in the signal processing may be the determination of the
signal distribution function in order to obtain the mean value of the
signa! and its variance. In general, however, the signal of a single
detector does not provide suffieient information on the local partiele
concentrations, veloeities and on the transition probabilities at the
stages. Combination of the signals from a number of detectors is required
to measure the local velocity distribution function of the particles
passing these detectors. Partiele streams may be analysed using arrays of
detectors that scan the partiele stream along its full width. Signals
obtained from arrays that measure individual streams may be combined and
processed to obtain information such as the total mass flow of the
stream, the mean partiele velocity and the mean concentration, The
determination of partiele transition probabilities is more difficult.
Partiele streams that enter and leave a stage either to the next higher
stage or to the next lower stage are interconnected by the overall mass
balance across the stage and by the partiele transition probabilities
pf and Pr· Knowing the total size of these partiele flows is
insufficient to reconstruct the individual values of pf and Pr· It is
necessary to register both the exact stream in which particles enter a
stage and the exact stream in which they leave that stage to determine
pf and Pr· This is achieved by regietering a large number of
particles one by one. Two different methods of data processing will be
presented in paragraph 3.3. The first method uses correlation functions
of detector signals. The signals are regarded as being produced by a
collection of particles. No distinction is made between the signals of
the individual particles. The second method individually processes the
signals of tracer particles that can be distinguished individually.

-43-
3.3 Signa! processing by means of correlation function calculations.

3.3.1 'Theory.

Il has often been reported in .. iterature (/3.1/, /3.2/, /3.6/-/3.11/)

that correlation function calculations are powerful tools in determining
partiele veloeities and velocity distribution functions of gas-solid and
solid-liquid flows. Two detectors located at a distance Ld from each
other register the presence of particles at two different positions
inside the flow channel. Suppose that the mean velocity of the particles
along the connecting line of the two detectors is vp m/s. The signal of
the first detector, x(t) at time t will show a certain resemblance to the
detector signal y(t+T) at a time T seconds later. This resemblance
decreases as deviations of the actual partiele veloeities from the mean
velocity increase.
The signals x(t) and y(t+T) may mathematically be compared for different
values of T by means of the so called cross-correlation function <j) (T).
xy
This function is defined by:

T
$ ( T) = lim 1/r f x(t) y(t +1:) dt (3.1)
xy T-+oo 0

The value of <Pxy(T) is a measure for the resemblance between x(t) and
y(t+T). The location of the maximum of this function, Tm' therefore
provides a measure for the most likely transit time of the particles
between the· two detector positions. This transit time Tm is not
necessarily equal to the transit time t pm ' that is calculated from the
mean partiele velocity vp, i.e. tpm' Ld/vp /3.6/,/3.9. A theory
which is derived from dynamic systems may in a number of cases be used to
determine the mean partiele velocity and the velocity distribution
function /3.1/, /3.9/-/3.11/.
If the detectors produce signals that are linearly proportional to the
concent ration of the particles,. the signals x( t) and y( t) can be
considered to be the input and output signals of a linear dynamic process
/3.16/.
The response of the system can then be calculated for every input
function x(t) by means of suitable linear combinations of finite input
variations of which the response functions are known. The Dirac delta
pulse ö(t) is often taken as input signal and its output is called the

-44-
impulse response function h(t). This function has the following physical
meaning (/3.6/, /3.9/, /3.11/) (see fig.3.2):
Suppose that a set of particles is injected at a time t=O into the
two-phase flow just at the location of detector x. This pulse of
particles will pass the detector and will be registered as a very narrow
peak with an area that is lineary proportional to the partiele
concentration. Between the two detectors all particles have different
veloeities and will thus also have different transit times. The transit
time distribution function will be registered at detector y if the
concentration of the peak at detector x equals a value of 1. However, the
signal that is registered at detector y is by definition also equal to
the irnpulse response function h(t) of the system. The impulse response
function of the linear system thus is equal to the transit time
distribution function. The partiele velocity distribution function h(v)
can be calculated from h(t) by means of the following equation:

(3.2)

The output y(t) of an input signal x(t) follows from the the convolution
integral /3.16/:

0
0 -o :~ 0 0 0
0 0 0 0 c
0 0 o ooo o o
0

Q~te:tor Q~te:tor
x(t)
y(t)
x(t)...S(t) y(t}:t>(t)

. Figure 3.2.
Illustration of the relationship bet~+een the :i.mpJlse resp:nse h(t) and the transit tirte
distrimtion function.
A ó jX1lse of p3rticl€S passes detector x (area of rogistered jX1lse = 1). Detector y
registers a si[PBl h(t). h(t) both equs1s the im[:clse resp:nse function and the transit ·
tirre distriOOtion function.

-45-
 00

y(t) = f h(T ) x(t - T) d T (3.3)

0

For stochastic stationary signals the following equation holds:

00

<Î' (t)=fh(T)<P (t-T)dT (3.4)

X'f 0 XX

Taking the Fourier transforms of both the right and the left side of this
equation gives:

4? xy(f) = H(f) 4?xx(f) (3.5)

Thus,

H(f) = 4? (f) I 4? (f) (3.6)

X'f XX

and the inverse Fourier transferm of H(f) gives h(t), the transit time
probability density function. In addition to the fact that the system has
to be linear dynamic, the signals also have to be stationairy and
ergodic. This last requirement means that averaging the signal by means
of repeating the process a large number of times holds the same result as
taking the time average of the signal.
An interesting conclusion can be drawn from the values of both the
cross-correlation and the auto-correlation functions baseline. The
\
auto-correlation function can be defined as the time averaged mean of the
product x(t) x(t + <):

<P (
XX
T) = E((x(t) x(t+T)) (3.7)

If x(t) is divided into two parts of which one is the time average of
x(t), i.e. x and the other part is the fluctuating part x'(t) then the
0
following equation holds:

cPxx(') = E(('U + x'(t)) ('\) + x'(t +1" ))) (3.8)

This equation equals:

(3.9)

-46-
For large T the first term of eq. 3. 9 becomes 0. The val ue of the
baseline then equals the square of the mean partiele concentratien x .
0
Cross and autocorrelation functions may be interpreted in different ways.
In some situations (/3.1/, /3.9/) simply the value of T at which the
m
cross correlation function ~
(T) reaches its maximum can be used to
xy
calculate the mean velocity of the particles by:

(3.10)

As can be seen from fig. 3.3 this methad is applicable only if the
transit times between the detectors show very little variation.
If the condition of a low varianee of the partiele transit time is not
met, a velocity distribution curve may be assumed, f.i. a normal
distributed velocity, of which the parameters (mean velocity and the
standard deviation) are fitted in such a way that the cross correlation
function calculated by eq, 3.2 and 3.4 corresponds optimally tó the
measured cross-correlation function /3.1/, /3.9/-/3.11/. Other methods of
interpretation can be a normal partiele transit time distribution or a
physical model of the partiele transport that consists of a convective
flow tagether with a dispersion coefficient superimposed upon it
/3.9/-/3.11/. The disadvantage of these methods is that a priori
knowledge of the partiele flow is required. No a-priori information is
neerled if a histogram function for v is taken of which the height of
p
each interval is adjusted by means of a least squares fitting method so
that again the best fit is obtained /3.7/.

•xx
o.o
x x
vp 1.2 m/s.
Lp - o.az m.
0.6

0.4

0.2

o.o
0.03 T (s)

Figure 3.3.
The influence of the fluctuatirns of the partiele transit times on the cross correl8tion
functioo ~ ( 1:).
xy

-47-
The most elegant method appears to he that which makes use of equations
3.4 through 3.6 and uses Fourier analysis to obtain the desired velocity
distribution function. However, as was stated above, the process has to
satisfy a number of strict requirements. One method that applies this
technique was found in literature /3.9/. However, Fourier transfarms and
inverse Fourier transferros were realized by electronic frequency
analysers, Capacitive detectors that transmit signals proportional to the
concentratien of particles inside the measurement volume were used there.
The optical methods that were selected for our partiele detection do not
produce signals that strictly satisfy the properties of linear systems.
It was nevertheless decided to investigate the possibility for applying
this elegant method. Possible errors in the interpretation of the
correlation functions due to non-linearity should be estimated on
forehand. Model calculations were therefore carried out to simulate
partiele movements in air classifiers to test the data acquisition, the
processing and the results. Information about the feasibility of the
correlation method in combination with Fourier transfarms carried out by
computer was to be obtained in this way.

3.3.2 Computer simulation of data processing by signal correlation.

In order to get an impression of the possibilities of the detector data

processing by means of cross- and auto-correlation function calculations
and Fourier analysis, the detection of particles by optical detectors
(light transmission) and the processing of the detector data was
simulated on a Burroughs B7700 computer /3.17/.
A one dimensional flow of flat, square particles (area 2*2 cm 2 ) which
is scanned by two detectors at a distance L from each other has been
simulated.
The detectors functioned according to the light transmission method. The
detectors had infinitely narrow beams which were interrupted as soon as
only a small part of a partiele crossed it. The detector signal then
changed to "high". The detector signal was "low" i f no partiele was
crossing. Thus, these detectors did not meet the condition of producing
signals that are linearly proportional to the partiele concentration.
The particles were further supposed to rotate along their axis. The
projection of the partiele area upon a surface perpendicular to the
optical detector axis varied between 0 and 4 cm 2 (which was the real
partiele surface). The area of the projection of each partiele that

-48-
crossed a detector beam was selected at random by the random number
generator of the computer according to the normal distribution. The
computer model offered the possibility of simulating particles that have
a constant velocity between the detectors or veloeities that vary between
the detectors. Particles moved at different veloeities between the
detectors. The velocity of each partiele was also generated by the random
number generator. Particles had normally distributed velocities.
The detectors sampled the particles at a sampling rate of F Hz. The
number of samples taken was n.
The results of sampling were stored, autocorrelated and crosscorrelated.
A typical result of such a calculation is given in fig. 3.4.a These
correlation functions were Fourier transformed. The correlation functions
have a baseline value that is larger than 0. The functions were
transformed into functions with a zero baseline before Fourier
transformation. This was done by substracting the square of the mean
detector signa! from the correlation function.
The Fourier transformation carried out by means of a digital computer
also requires some mathematica! elaboration /3.18/, /3.19/. The Fourier
transfarm of a function f(t) is defined as:

+""
F(f) = f f(t) exp(-j 2 'lTf t) dt (3.11)

Fig. 3.5 gives a graphical representation of this transformation by a

computer. f(t) had to be presented py discrete points before it could be
transformed. The function was multiplied by a series of sampling pulses
in the time domain (fig.3.5.b) and convoluted by a similar series in the
frequency domain. The function to be transformed had to be limited. This
was realized by multiplying the function by a block function in the time
domain (3.5.d) which resulted in a convolution with the Fourier transfarm
of the block in the frequency domain (3.5.e).
Finally, the Fourier transform was also to be represented by a discrete
function. This was done by multiplying the Fourier transfarm by a series
of pulses (3.5.f) and convoluting the function in the time domain by
another pulse train. The final result of all these manipulations which
were undertaken to perform the Fourier integration on a digital computer
is seen in fig. 3.5.g. The maximum of the Fourier transfarm has a value
that is too smal! and

-49-
 .."" nrnJber of j>'lrt><J.Ps: !lal. •xy
h(vp)
t",·
(-) (-) (s/rn) / \ a::t:ml
nunbet of samples: 23Xü I \
0.8
.§ffi'Pling rate mJ Hz.
0.8 o.m> \
I
vp' 2.0 mis. a) \ b)
0.6 0.6 I
avp: 0.5 m/s. O.OY. I
I
\
0.4 0.4
0.002
0.2 0.2 0.00

o.o 0.0
0 8 16 'lA 32 liJ T (ms) :0

Fi89.re 3.4.
a) SimJlated czngs- and auto-o:Jrrelation functions.
b) Càlrulated and act:ual distribution tunetion of nrxlel si.Jrulation.

L t
a) _ _ A\: :;. ._. .~. _,I I(. . .;f" )j-
A (t) p) :~<OI

. . 11111111Ul...,.-
Ï11111 t1... 1_
t -1/r 1/r f

c)_,·~··
r ' J:A
d) _rl
-1/zr 1/zr r

Figwe 3.5.
Gra[ilical reprasentation of the Foorier transformation as carried oot by a digital
COlplter.
Left: the functions fn the t.ir:.. iana:in; right the functions in the freqtle1Cy danain.

-50-
sidelobs occur in the curve that should be monomodal. Fig 3.4.b shows the
results of the determination of the partiele velocity distribution h(v).
H(v) was calculated through Fourier transforming the correlation
functions, calculating the Fourier transfarm of the transit time
distribution function, H(f) (eq. 3.6), reverse transformating H(f) into
the transit time distribution h(t) and calculating the velocity
distribution according to equation 3.2. This velocity distribution
function differed in several respects from the function that was used in
the simulation model. Sidelobs occurred at the low velocity side of the
function. The maximum value was less than the maximum of the original
distribution function. The preserree of side lobs was caused by the
Fourier transform of the block functions (window functions) that were
used to limit the domain of the functions to be Fourier transformed.
Other window functions that caused less pronounced side lobs were known
from literature. These functions are e.g. the so called Hann window and
the Parzen window /3.18/ (see figures 3.6). The results of Fourier
transfarms obtained with thesewindowsare shown in figure 3.7. It is
seen that the fluctuations in the velocity distribution functions at low
veloeities were suppressed by these windows. However, both the location
and the value of the maximum differed considerably from those in the the
original functions.
These differences could be reduced by using higher sampling rates for the
detector data and by taking more sa~ples. However, both the number of
samples that could be taken and the sampling rate depended upon the
capacity and the speed of the computer. A DEC. Minc-11 laboratory
computer was available for this research • A maximum of 15000 16 bit
values could be sampled at a frequency of 4000 1/s during one experiment.
These values were still not high enough when high precision was required.
It was concluded that the metbod of transforming cross- and
auto-correlation functions in order to calculate the velocity
distribution functions could only be carried out with a high accuracy
when a high speed computer sampling rate could be obtained and the core
starage of the computer would be large enough. Because of the limitations
of the available machine it was decided not to apply the correlation
techniques and the Fourier analysis of signals in this investigation. The
results of the rnadelling experiments showed that these difficulties in
calculations caused by the physical limitations in sampling speed

-51-
influenced the results even more strongly than the fact that our
meesurement system was not linear. Another reason not to use these
methods was that the information which was obtained was limited to
partiele velocity distributions, -concentrations and -flow rates.
Partiele transition probabilities could not be determined in this way.
The tracer rnethods which will be discussed in the following paragraphs
were therefore selected to determine the internal variables.

I: Block functlw
Il: funn ~i.ndoo.
III: Pare.en 1-inda...

0 · t (s) T
0
-1
0 3/fo f (llz)

Figure 3.6.
Different windaw functions.
a) 'TiiiE dcmain.
b) Fm:p.H~Cy danain.

h(vp)
,.
. ,-.
h(vp)
(s/0!1) a) b)
' I '
(s/an) '. a:::tm.l
,' \ actual
0.(1))
'' G.(l))

o.w. o.w.

O.!m O.!m

0.00 o.oo

Figure 3.7.
Illustration of the effect of wind:Jw functions on the results of the velocity
di'3t.ribution calculations.
a" 'TiiiE dcmain: block function; fm:p.H~Cy dcmain: Parzoo wind:Jw.
a 'TiiiE danain: block function; frequency dalmn: Hanming windaw.

-52-
3.4 Signal processing in tracer detections.

As the above correlation techniques had proved unsatisfactory it was

decided to investigate methods based on the detection of tracer particles
in order to develop techniques in which both partiele velocities, flow
rates and transition probabilities could be determined. The principles of
these techniques will be discussed in the next paragraphs.

3.4.1 Principles of measuring internal variables.

a) partiele transition probabilities.

Senden /3.8/ showed that partiele transition probabilities, i.e. the
probabilities of particles at a stage to move to one of the two
neighbouring stages, depend among other things upon the direction from
which particles enter the stage. We developed a method for measuring the
partiele transition probabilities. The detection of the particles was
based upon the light reflection technique. White particles were used as
tracers. Apart from their colour these particles were identical to the
black main stream particles. Arrays of detectors were placed at the
boundaries of each stage in such a way that tracer particles that entered
and left a stage could be detected. Fig. 3.8 shows a scheme of
possible detector locations as well as of the numbering of stages, stage
boundaries and classifier sections. Particles that entered a stage in a
rising stream were detected by detector arrays in front of the channel.
Particles that entered a stream from the next higher stage were detected
by detector arrays placed at tne lower zigzag side walls of the channel.
These falling particles more or less slide down the lower zigzag walls of
the stages with their surfaces turned towards this wall. Locating sensors
at these walls thus offered the possibility to detect the deseending
particles efficiently. Rising particles do not move in a more less fixed
orientation but show a tendency to rotate and "wobble". Detectors to scan
these rising particles were placed in front of the channel so that the
light sourees that were used to irradiate the deseending particles could
not interfere with the detectors. Partiele transition probabilities were
now established for a large number of particles by determining the stream
in which a partiele entered a stage and subsequently left that stage.

b} Partiele flow rates.

Partiele flow rates were determined during the same experiment as the

-53-
a)
.-rtnc
atap ttowaduiea
of -ri"'
of ata... b)
10
8
s·

4
5 {Y)

s: s1de detector orrays

0 i, i-1, i Tl: stage boundary nwab€rs f: front detE>ctor arrays

Figure 3,8.
a) NwiJering of stage lxxlrJdaries and sections.
b) !Èfinition of stages, stage boundaries and sections.
c) IÈfinition of the transition probabilities.
d) Location of the detector arrays.

1 s
5 Figure 3.9.
~ r'-'
PhotO"alSi. tive detector.

25 a a: diaphragm (cylinder)
b: phototransistor (inserted in C)

D: photosensitive area of
t- transistor b

~
3.5 d: diameter of diaphragm
L: distance between diaphragm
6 .5 c aperture and phototransistor
~-7
~6 dimensions in mm.

INCIDENT
S:A11 OF LIGHT

P!IOTOSENSIT IVE
AREA
0 IAPHR AGM

54-
partiele transition probabilities through registering the number of
tracer particles that passed the stage boundaries per time unit. The
local rising and falling partiele flows could be calculated from the
ratio of tracer particles to main stream particles and the corresponding
partiele flow rate,

c) Partiele velocities.
Partiele veloeities were determined by means of the tracer methad by
applying two horizontal arrays of detectors that scan either the rising
or the falling partiele stream. The velocity component along the zigzag
wall was measured by registering the time that it took for a tracer
partiele to cover the distance Ld along this wall between the two
d.etectors. Information on both mean partiele velocity and its varianee
was obtained when this was done for a large number of particles.

d) Local partiele residence times.

Local residence times or transition times were determined by measuring
for a large number of tracer particles the time that it took to cover
each one step transition. For this purpose the information that was
sampled to determine partiele transition probabilities also contained the
times at which registrations of particles at the different stage
boundaries took place.

3.4.2 Experimental set-up and data processing procedures.

a) Partiele detection.
White tracer particles were detected in a main stream of identical black
particles when they reflected light onto the photosensitive area of
detectors. Tungsten halide lamps (colour temperature 3400 K) were used as
light sources, The detectors were phototransistors (type BPX 99,
manufacturer AEG Telefunken) fitted with a diaphragm. The diameter of the
diaphragm and its distance from the transistor were chosen such that only
particles that were in one line with the optical axis of the detector or
that showed only a small deviation from this axis were detected. Fig. 3.9
shows a single detector.
Each detector was part of an electronic circuit which produced a voltage
that was "low" if a partiele was detected and "high" if no detection took
place.
Arrays of such detectors were used to scan partiele streams for white

-55-
tracer particles. All detectors of one array were electronically linked
together so that a detection by one or more detectors in the same array
resulted in a single output signal. The level of this output signal
voltage was high when a tracer partiele was detected and low when no
detection occurred. The electronic circuitry is discussed in more detail
.
in appendix 3.1.
Eight arrays were placed horizontally in front of the classifier at the
boundary no's 1-8 (see fig. 3.8) between each two stages. Each array
contained 23 detectors placed at a distance of 6.0 mm. from each other.
Only the detectors of each array that scanned the tracer particles of the
rising stream were used. This number of detectors depended upon the
geometry of the classifier that was being investigated. Light sourees to
illuminate the rising particles were placed also in front of the channel
behind the detector arrays.
The arrays at the lower zigzag walls contained detectors that registered
tracer particles in the.falling stream. Each array contained 24 detectors
placed at a distance of 8 mm. from each other. Each array was placed
halfway between two stages (fig. 3.10). Lamps were placed bebind the
detector arrays to illuminate the falling particles. The lamps were
located and directed in such a way that there was little chance of these
detectors registering the presence of rising particles.

a classifier channel
b front shield
c back shield (dark)
d zigzag channel
e side detector array
f front detector array

FiBUre 3.10.
Location of the front- and side detector arrays

-56-
 -------- -------------------~

Overall partiele residence times and local partiele transition times were
measured during the same experiments. These measurements were carried out
with the same experimental set-up as were the transition probabilities.
The same arrays now placed at a distance of appr. 6 cm from each other,
were used for measuring partiele velocities. These arrays were located
either at the lower zigzag sides to measure falling veloeities or both at
the front of the channel and at the higher zigzag walls to measure rising
veloeities of particles. Fig. 3.11 shows examples of these set-ups.

b) Signal sampling.
Fig. 3.12 presents a block scheme of the signal sampling.The binary
signals that were produced by the detector arrays were sampled by a
Minc-11 on-line laboratory computer. Up to 16 binary signals could be
sampled at the same time. The computer took a sample each time that a
tracer partiele was detected. The time of sampling was registered too.
The array signals were also processed for this purpose by an electronic
circuit. This so-called trigger circuit produced a trigger pulse
whenever a tracer was detected and informed the computer that a sample
had to be taken, Appendices 3.2 and 3,3 contain detailed information
about this triggering circuit and the computer programme for signal
sampling.

3.4.3 Determination of partiele transition probabilities.

a) experimental procedure.
The superficial air velocity inside the channel and the feed rate of the
black main stream particles were adjusted according to the procedures
discussed in chapter 2. One tracer partiele a time was fed into the

4 3
75
f: froot detector array. _
1
s: side detector array. _".".
dinensions in an.
Figure 3.11.
~t set-up for the determination of the velocities of rising (a) and falling
particles (b).

-57-
channel. The next tracer partiele was not fed to the channel until the
last partiele had left it. The 8 detector arrays at the zigzag sides and
7 of the 8 detector arrays in front of the channel were sampled. Thus one
of the front arrays (the one at stage boundary 1) remained unsampled. The
signal of the 16th computer input was changed each time a new tracer
partiele was fed to the classifier. The measurements were stopped as soon
as the required number of samples had been taken, The fraction of the
bottorn product, ~· the partiele hold-up and the partiele feed rate
were determined in the same way as has been described in chapter 2. The
registered data were processed by the computer.

b) data processing.
The laboratory computer both sampled and processed the detector data and
calculated the internal variables. Fig. 3.13 shows these procedures
schematically.
A data file "DATA" was. created by means of the computer programme SAMPL
and stared on a floppy disk memory. The file consisted of a maximum of
7500 pairs of 16 bit numbers. The first number of each pair represented
the sampling time. The second number contained the 15 detector array

I
I
a I
'I
I
I
a I
I
I
I
I
I
---'
The analog signals (AN) of the detector arrays (a) are converted into binary data and
linked tagether by the interfacing circuit (I).
The resulting binary signals (BN) are connected to a triggering circuit (TR). This
circuit produces a pulse when a partiele is detected, The Schmidt trigger of the computer
(ST) receives the pulse, Simultaneously, the signals are sampled by the digital rnpur
(DI) of the computer and the time of sampling is registered,

Figure 3.12.
Block ciisgram C>f the signa] sampling equiprent.

-58-
signals and the 16th bit control signal. These 16 binary signals were
represented as a single 16 bit digit. Programme "SAMPL" is discussed in
appendix 3.3.
The data file was processed further in order to obtain the partiele
transition probabilities at the different stages, the local flow rates,
the local partiele transition times and the overall partiele residence
times. Two different processing techniques were developed.
The first technique of data processing made use of two computer
programmes. The first programme "MATRIX" determined the transitions by
combining pairs of subsequent detections as stored on disk by means of
the programme "SAMPL". From these pairs of detections the corresponding
transitions were identified by the stage boundary number and the stream
(falling or rising) in which a partiele was detected befare a transition
and after. The transitions were stared in matrix ITRANS(k,l,i,j) of which
the elements indicate the times that a transition was characterized by a
partiele passing the ith boundary in the kth stream (1: falling
stream; 2:rising stream) and subsequently passing the jth boundary in
the lth stream (value as k: 1 or 2). These transitions did not
necessarily correspond to the one-step partiele transitions described by
Sendens one-step-memory model /3.8/. The reason for this was that due to
a number of causes particles might not be detected by a detector array.
Some of these causes were: A white tracer partiele was bidden by black
particles when it passed an array. The light it reflected then did not

"SAMPL": on-line sampling programme.

"Matrix": off-line calculation of detecterl pRrticle
trajectories.
"AKANSl": off-line calculation of transition
probabilities and local flow rates from
"!TRANS" file.
"DATA": file that contains detector data.
"!TRANS": file that contains detected part iele
trajectories.
"ALTRNT": Alternative programme that calculates
transition probabilities and
local flow rates.

Figure 3.13.
Block d:isgrnm of data processing.

-59-
strike the detectors. A partiele might be orientated unfavourably with
respect to the detectors.
To increase the accuracy of the metbod in the first processing technique
not all deteetions were aecepted as a mark for a valid transition. The
programme also had to deal with the possibility that particles could be
detected twice when they passed the stage boundary. This could for
instanee be caused by a rotstion of the partiele in front of the
detector. Rising particles might sametimes also be detected by detectors
scanning the falling streams.
Therefore, in this first processing technique the programme had to
consider the possible occurrence of above disturbances. To do so
information was needed on how much time the real individual transitions
took, This information was not a priori available but had to be estimated
from measured data. The interpretation of these data might cause
systematic deviations between the measured transition probabilities and
the aetual values.
Whether a partiele passed onee or twice was determined by measuring the
time between the two detections. If this time was shorter than a minimum
time then the programme considers that the partiele passed the boundary
only once. The value of this minimum time was determined from histograms
that were measured for the transitions of particles moving from the
falling stream to the rising stream and vice versa (fig 3.14.c and d). It
is concluded that the first peak in these histograms is caused by the
same passage of a partiele that is detected both by falling stream and
rising stream detectors.
The fact that rising particles may sometimes be spotted by the detectors
for the falling particles is dealt with as follows (fig 3,14.a): If the
time difference between two falling stream detections is less than tmin s,
then the partiele passed the stage boundary only once in a falling
stream (situation "1"). If the duration of this transition was shorter
than the sum of the two transit i ons times of < pf > + <1-pr > mot i ons
but longer than t . (situation"2"), then it is concluded that a rising
m~n

partiele was spotted by a "falling particles" detector. This detection is

skipped and not used in the transition matrix. No conclusion is drawn as
to whether the first or the second deteetion was wrong. A flow digram of
programme is given in appendix 3.4.a.
The second programme, "AKANSl" calculates partiele transition
probabilities and flow rates from the transitions that were identified by

-60-
 aJ 'l'he shaded detections "1" are
a
caused by particles that are
"1" detected t~ice during one
....8 passage. The area "2" of fig. a
is caused by rising particles
jlO that are detected by the arrav
"a for the falling partic1es.

i 0
0 0.1 0.2 & (s) 0 0.1 0.2 & (s)

aJ c

....'-'~

~
"a Figure 3.14

i 0 -f't<sa.I4-..U..J.+.J.--.
Transition time distribution
functions measured at a single
stage boundary in the broad
0.1 0.2 & (s) 0 0.1 0.2 & (s) 120 deg. classifier.

boundary
4

boundary
2

a b c d e

Five transitions occurred in which a partiele was detected in the falling stream at
stage boundary 4 and subsequently in the rising stream of boundary 2. The falling streams
of boundary 2 and 3 were passed undetected. Hence, ITRANS(l,2,4,2)•5,
First, it is supposed that the particles cover the trajectories along the shortest path
(a). The above information combined with the results of other trajectories "LTRA~S" gives
a first estimate of partiele transition probabilities and detection efficiencies,
Next, the possibility that the particles may cover langer paths is also taken into
consideration, These longer paths are limited to the occurrence of two extra one-step
transitions (figures b-e).
The probability that each of these trajectories is covered is calculated from the earlier
eastimated transition probabilities and detection efficiencies. New transition
probabilities and detection efficiencies are calculated and the values are used in an
iteration procedure until the values of the transition probabilities do not change
anymore.

Figure 1.15.
Exmple of the p:Irticle trajectory reconstruction by meensof progl"Bfll1}3;'! "AKANSl".

-61-
"MATRIX". The programme determines for each type of transition (i,j,k,l)
of which ITRANS(i,j,k,l) gives the number of occurrences the number of
one step transitions <pf .>, <1-pf .>, <p .> and <1-p .> that
,1 ,1 r,1 r,1
are part of this transition. The number of times that a partiele is
detected c.q. not detected is determined for each detector array. First,
it is assumed that particles travel from stage i to stage j by the
shortest route. From this information, a first estimate of the partiele
transition probabilities pf . and p . at the different stages was
,1 r,1
calculated. The programme also calculated an estimate of the detection
efficiencies Ef(pf ,1.), Et,(p
~ r,1.), Ef(l-p r,1.) and Ef(l-pf ,1.) for
each detector array. Here Ef(Pf,i) etc. is the probability that a
partiele is detected after a transition <pf .> etc.
,1
Next, the probabilities that particles might take longer ways between two
subsequent detections are calculated from the so-called zero-order
transition and detection probabilities. Only the trajectories that take
two extra undetected one-step transitions are taken into consideration.
Fig.3.15 shows this procedure in some more detail. New values for the
transition probabilities and the detection efficiencies are obtained, The
calculated values of the detection efficiencies are lower than the zero
order results because extra transitions that were not detected were taken
into consideration, The new values of the transition prob~bilities and
detection efficiencies were used in an iteration process to calculate new
values. The iteration process was stopped when the difference between the
values of the last iteration and the one but last was sufficiently small.
The programme is described in detail in appendix 3.4.b.

An alternative methad "ALTRNT11 was developed to check the results of

the programmes MATRIX and AKANS1 /3.20/. This methad did not reconstruct
all partiele trajectories completely. It used information on three
succeeding sections that a partiele had passed. Four different
possibilities exist for a partiele to enter a section and then leave it
again. These possibilities are illustrated in figure 3.16.
The partiele motions within a single section i were not taken into
consideration. The probability for particles to leave section i in a
rising stream were defined separately for particles entering this section
in a rising and in a falling stream respectively. For rising particles
this probability is defined by:

-62-
 JIRANS(2,i,2)
p =----------
2Z,i JTRANS(2,i,2) + JIRANS(2,i,l)
(3.12.a)

and for falling particles:

JIRANS(l,i,Z)
p = --------- (3.12.b)
lZ,i JTRANS(l,i,Z) + JIRANS(l,i,1)

These probabilities are related to pf and pr in the following way:

p22 . = p
,1 r,1. + (1 pr,1.) pf ,1. pr,1.
+ (1 - pr,1. ) pf ,1. (1 - pr,1. ) npf ,1. pr,1.
+ •••+ pr,1. (pf ,1. (1 - pr,1.)) (3.13)

. n n
P .
22 ,1
= lim LP . (pf.,1
n-- n=O r,1
(1-p
r,1
.)) (3.14)

stage lxudary i+l -

stage lxudary i

a b c d
a: The partiele enters the section in a r1s1ng flow and leaves in a falling flo~. The
number of occurrences of this transition is recorded in JTRANS( 2, i ,I).
b: idem; partiele enters in a rising flow and leaves in a rising flo,..;
registration in JTRANS(2,i,2).
c: idem; partiele enters in a falling flow and leaves in a rising flow;
registration in JTRANS(l,i,2).
d: idem; partiele enters in a falling flow and leaves in a falling flow;
registration in JTRANS(l,i,l).

Figure 3.16.
The 4 different ~~ays in which a pgrticle can pass through section i.

-63-
This is a geometrie progression of which the solution is given by:

(3.15,a)

It can be shown in the same way that

(3.15.b)
1-pf.
,1 (1-pr,1.)

(1 p .) (1- pf .)
r,1 ,1
P21 . =1 - P22 . = ----'-'-------'-- (3.15,c)
,1 ,1 1 Pf . (1- p .)
,1 r,1

_ pr,1. pf.
,1
p12,i = 1 Pll,i- (3.15.d)
1 - pf . (1 - p . )
,1 r,1

The following equations can be derived from 3.15.a and 3.1S.d:

p
f,i -- p (3.16.a)
22,i

p22. -Pl2.
p . = ,1 ,1 (3.16.b)
r,1 1_p 12,i

This method thus offers a way to determine partiele transition

probabilities. No information is required on partiele transitions within
one section. The problem of rising particles detected by falling partiele
detectors does not influence the results of this method. The results of
this method could in principle be influenced by a partiele having entered
and left a section without having been detected. Video recordinga showed
that this occurred mainly when particles passed several stages in a
continuous rising stream. The probability that a partiele enters a
section in a falling stream and subsequently leaves that section in a
rising stream without being detected is practically nil. The same is true
for particles that enter a stage in a rising stream and leave that stage
in a falling stream. Thus this method is not very sensitive to problems
caused by partiele transitions not being detected.

-64-
However, the method has two disadvantages :
The transitions P and P are not defined for the top section
22 21
(section 8) of the air classifier and the transitions P and P are
11 12
not defined for the bottorn section (seetion 1). Thus Pf,l' Pr, 1 and
Pf,S and Pr,S cannot be determined by means of this method. The
seeond disadvantage is that a large number of deteeted transitions are
not used, namely those that occur within the same section. This
influenees the aeeuraey of the results. The aeeuraey of the results of
the two methods is diseussed in the next paragraph.

c) Accuracy of partiele transitions measurements.

An estimate p' of the partiele transition probability p that is

determined by means of programmes MATRIX and AKANSl, is calculated in the
following way:
Consider n particles arriving at a boundary no. i in a strearn k (k ean be
a rising strearn or a falling stream). n particles of this stream were
s
found to leave the stage in a rising stream. Thus, the caleulated rising
probability p' is:

p' = n /n (3.17)
s

The individual results p. of each partiele transition are 1 (the

l
partiele rises) or 0 (the particles falls). The varianee of these
individual results thus is:

2 n 2
cr = 1/n I (p. p') = p' (l-p') (3.18)
p i=l l

beeause pi ean only be 0 or 1.

The individual values of the transitions show a binomial distribution
with a mean value p' and a varianee cr 2 = p'(l-p').
p
For n>lO this distribution ean be approximated by a normal distribution
2 2
with mean p and varianee a' p = crp /n. Thus, the resulting
distribution will beeorne narrower as n becomes larger. An approximation
of the deviation of p frorn p' is •

cr' =
.ycrD
rT:tn =
~·(l-p'). (3.19)
p p n

-65-
Thus the probability that

p' -o' ::~p ;;;p' +o' (3.2.0)

p p

is 0.68 and the probability that

p'- 2o' ;:;; p ::> p' + 2o' (3.21)

p p

is 0.95. From 3.21 it follows that:

p'- p
--:ii2 (3.22)
o'
p

If a requirement is that the probability of p 1 -p < 0.01 should be larger

2
than 0.95 then cr' < 0.000025. As can be seen from equation 3.19 the
p .
standard deviation is largest if p'=O.S. It can be concluded from
equation 3.21 that at least 10,000 partiele transitions have to be
detected to measure the transition probability with a probability of more
than 95 % that the deviation from the actual value is less than 0.01.
In the same way it can be shown that at least 250 motions have to be
observed to measure p with a probability of 0.68 that it contains an
error that is less than 0.02.
The error made by the alternative method (programme ALTRNT) is larger.
This can be shown by means of the following example:
Suppose that 220 particles arrive at stage i in the rising stream. Of
these particles 110 leave the section in the rising stream, thus p .
r,1
will be:

p
r,1
• = 110 I 22.0= o.s (3.23.a)

The number of particles that enter in the falling stream is 260. 130 of
these 260 particles entered the next higher section after passing section
i. The rising probability of falling particles thus is:

Pf ,l. = 1~ I 2ffJ = 0.5 (3.23.b)

-66-
According to equation 3.19 the standard deviation in this case is:

0 (p .) =
vp . p .) =
r ,1.
(1-
r ,1. 0.034 (3.24.a)
r,l. 220

pf 1. (1-pf .)
1
(J (pf . ) =
,1.
' ' = 0.031 (3.:ëi.b)
2ffJ

The following values of the transitions were found by means of the

alternative method:
JTRANS(I, I ,1) 100; JTRANS(l,I,2) 50
JTRANS(2,I,2) 60; JTRANS(2,I,l) 30
P12 , and P22 are calculated by means of equations 3.14. The standard
deviations are calculated from equations 3.24:

pl2 1.(l- p12 1.)

Pl2 ,1. = 0.33; O(PlZ ,1. ) = ' ' =0.038 (3.25.a)
150

and

p22 1.(l- p22 1.)

P .
22 ,l
= 0.67; O(P . )
22 ,1
= ' ' =0.050 (3.5.b)
'X)

The standard deviation of pf ,1.. and of pr,l.. can now be calculated from
the standard deviations of P and P with equation 3.15 and the
12 22
rules of error propagation.

a (pf ,1.. ) = O.C66 and a (pr,1.. ) = 0.())4 (3.26)

The standard deviation of the partiele transition probabilities for

values of pf and pr around 0.5 measured by the alternative method is
approximately twice as high as the deviations of the values pf and pr
obtained with the method that uses the prgrammes MATRIX and AKANSI. The
number of partiele transitions that are used in the alternative method
will become relatively low in the case where particles have high
probabilities of staying within the same section i.e. in situations
where pf > 0.5 and pr < 0.5. The standard deviation in the result of
this methad will then become higher.

-67-
3.4.4 Determination of partiele velocities.
a} experimental procedures.
Partiele veloeities or rather transit times of particles between two
detector arrays were determined in the following manner:
Detector arrays were placed at two different positions along one section
wall of the classifier channel. These detectors scanned two horizontal
parallel planes. Care had to be taken that the rising and falling
partiele streams were analysed separately. Deseending particles were
scanned by means of two detector arrays placed at the lower zigzag walls
of the sections where the partiele veloeities were to be determined.
Rising partiele streams were scanned by detectors placed both in front of
the channel and at the upper zigzag side walls. Only those detectors of
the front arrays that scanned the rising particles were used. Front and
side detector arrays as well as the electronic circuits were identical to
those used in the determination of the partiele transition probabilities.
The air velocity and the feed rate were set at the desired value. White
tracer particles were introduced into the channel one at a time at such a
low feed rate that the probability of two or more particles to pass
between the two detection planes at the same time was negligible. The
detector signals were sampled by the computer tagether with the
corresponding detection times.

b) Data acquisition and processing.

Data acquisition took place in the same way as in the determination of
the transition probabilities, The computer registered the time of each
detection and sampled the detector array signals. Information about which
detector array detects a partiele at what time was obtained in this way.
The partiele transition times were calculated off-line by determining the
time difference between the detections of a partiele by the arrays at a
plane a and the subsequent detections at a plane b. The standard
deviation of the transit time was calculated by means of

n
-
(t -t.)
2
E P 1 (3.27)
i=l n- 1

In this equation n is the number of tracer particles that were observed,

t is the mean transit time and is the transit time of the i-th
p

-68-
partiele that was detected.
The mean transit time tp is related to the partiele velocity vpm
through the detector distance Ld.

V "'L_,u
pn
I tp (3.28)

A histogram of the transit time distribution was obtained by determining

the fraction of measurements of which the transit lies within the
interval t +i t and t +(i+l) t. This was done for all intervals i
0 0
between t and the longest time interval N t that was found to have
0
actually occurred.
The partiele velocity distribution could be calculated from the transit
time distribution by means of equation 3.2.

3.4.5 Local patricle flow rates.

Partiele flow rates were determined simultaneously with the transition
probabilities. The number of tracer particles that were detected in both
the rising and the falling flow of each stage boundary no. was calculated
from the detection efficiencies of the detector arrays and the number of
detections that had been measured at these arrays. The following
equations were used:

Ff . = ( L (ITRAtf)(k,l,i,j) Ef(l,l,j))
,J i>j (3.29.a)
+ . <L. (ITRAtf)(k,l,i,j) Et<Z,l,j))) ~/Ntr
1 J

F . =( L (ITRAtf)(k,2,i,j) E (1,2,j))
r,J i>j r (3.29.b)
+ _I<. (ITRAtf)(k,2,i,j) E/2,2,j))) ~/Ntr
1 J
Ff . is the local flow rate of the deseending partiele stream at stage
,J
boundary j, F . is the flow rate of the rising stream at this
. r,J
boundary. Qv is the main stream partiele feed rate and Ntr is the
number of the tracer particles that were fed into the channel during one
experiment.

-69-
3.4.6 Local partiele transition times.

The transition time of each detected transition ITRANS was measured. Only
those transitions !TRANS that correspond to one step transitions were
taken into account. The relevant transitions were:

ITRANS(l, 2, i,i) -- <p f,i >

ITRANS(l,l,i,i-1) -- < 1-p . >
f,J.
ITRANS(2,2,i,i+1) -- <p r,J.. >
ITRANS(2,l,i,i) -- < 1-p r,J..>

The measured times of each of the transitions were stored in arrays and
could be represented by histograms. Examples of such histograms have been
shown in fig. 3.14.

-70..;.
References.
3.1 Mesch,F., Kipphan,H., Optoelectronics,! (1972), p451-462.
3.2 Mesch,F., Daucher,H.H., Fritsche, R., Mess Technik, ~ (1971),
p163-169.
3.3 Rosenbrand,G.G., M.Sc. Thesis (1981), Eindhoven University of
Technology.
3.4 Tels,M., Rosenbrand,G.G., Senden,M.M.G., EEC contract RUW-019-N,
Progress report no. 1 (1981), Eindhoven University of Technology.
3.5 ibid., Progress report no. 2 (1982).
3.6 Mesch,F., Fritsche,R., Kipphan,H,. Transactions of the ASME,
(1975), p414-420.
3.7 Kro11, M., "Experimentelle Untersuchung der Partikelbewegung bei
hohen Feststoffvolumenkoncentrationen in turbulenten Mehr-
phasenstromungen.", dissertation (1984), Universitat
Kaiserslautern.
3.8 Senden, M.M.G., "Stochastic models for individual partiele behavior
in straight and zigzag air classifiers.", dissertation (1979),
Eindhoven Universiy of Technology.
3.9 Kipphan,H., "Verfahren zur storungsfreien Identifikation
stochastischer Transportkenngrossen von Mehrphasenstromungen.",
dissertation (1975), Universitat Karlsruhe.
3.10 Kipphan,H., Chem.Ing.Techn., 49 (1977), p695-707.
3.11 Kipphan,H., Mesch,F., Regelungstechnik, 2 (1975), p299-304.
3.12 Mann,U., Crosby,E.J., Ind.Eng.Chem.Process Des.Dev., vol.10, 1
(1977) p9-13.
3.13 Kondukov, N.B. et.al. Int.Chem.Eng.,! (1964), p43.
3.14 van den Langenberg-Schenk,G., "The rheology of gas-fluidized
powders as determined in a vertical standpipe.", dissertation
(1983), Eindhoven University of Technology.
3.15 Van Velzen,D., F1amm,H.J., Langenkamp,H., Casile,A.,
Can.J.Chem.Eng., 52 (1974), p150.
3.16 Van der Grinten, P.M.E.M., Lenoir, J.M.H., "Statistische
procesbeheersing" (1973), Prisma, Utrecht.
3.17 Verstegen, N.M.H.M., M.Sc. thesis (1983), Eindhoven University of
Technology.
3.18 Brigham, E.O., "The Fast Fourier transform." (1974), Prentice Hall
Inc, New Jersey.
3.19 Otnes, R.K. Enochson, L.J., "Digital time series analysis." (1972),
Wiley & Sons, New York.
3.20 Gorissen, H.M., M.Sc. thesis (1984), Eindhoven University of
Technology.

-71-
4. PARTieLE BEHAVIOUR INSIDE ZIGZAG AIR CLASSIFIERS.

4.1 Introduction.

The measurements were carried out in the standard 90 deg. classifier, in

the standard 120 deg. classifier and in the broad 120 deg, classifier,
The feed stage was located above stage no.S.
Partiele transition probabilities, local transition times, partiele flow
rates and -velocities were determined as functions of partiele feed rate
Qv and of the superficial air velocity vf using meesurement methods
that were discussed in previous reports.

4.2 Local partiele velocities.

4.2.1 Iutroduction.

In paragraphs 4.2.2. a qualitative discussion of the results of partiele

veloeities measurements in the three classifier geometries is given. In
paragraph 4.2.3 a simplified model is presented to describe the velocity
of deseending and rising particles in zigzag channels, This model will
give a better insight into the quantities that determine these partiele
veloeities e.g. the superficial air velocity and the partiele feed rate.
Dimensional analysis of the problem supplies the dimensionless numbers
that describe the partiele velocities. In paragraph 4.2.4 the results of
the measurements of the partiele veloeities inside the three different
classifier geometries will be compared and explained by means of the
results of the rnadelling and of the dimensional analysis,

4.2.2 Measurement results.

a) 1he standard 90 deg. classifier.

Veloeities of both falling and rising particles were determined, Fig.
4.1.a and b show the exact location of the detectors. Examples of
measured transit time distributions are given in 4.2. Table 4.1
summarizes the results of these measurements. The table contains the mean
transit times t p and the median t pm ,
The partiele velocity v is defined by:
pm

vpm = Ldlfp (4.1)

Ld is the distance between the two detectors measured parallel to the

-72-
 f(tp>
a) (1/s) b)
~=0 g/s; vf=1.61 m/s :.D ~=(J,6J g/s; Vf:), )7 m/S

l5
t p=0.055 s; at p=O.et.S l5 t =O.IX>7 s; a =O.et.7
p tp
t = 0.038; v =1.11 m/s trm=0.052 s; vrm='l.9l mis.
10 rm rm
10

5 5

r<v ...--r--.---r-r--r--.-.....,r--ï"""'T"-. f(tp) ...-.....--.-.--..... -.-.---r--.-.---,

(1/s) (1/s)
c) standard !l) deg. c.l.a!Bifier. d) standard !l) deg. classilier
:.D 40
~=0 g/s; v~ m/s g/s; vf=l.56 m/s
l5 tp=O.COO s 30 tp =0.051 s: atp=0.024
atp=0.031 trm=O.et.6; "rm=l.~l :nis.
10 20

10

0 0
0 100

f(tp) ,.-..,---,--...-..,---,----,...-..,--,---,r--~

(1/s) Staolanl !l) deg. classi.fier'

:.D e)
~-1.99 g/s; vr1.58 m/s
l5 tp=0. 71 s; otp=0.024 s
trm=O.IX>7 s; vrm=0.86 m/s
10

Figure 4.2.
Exalples of h:ist:ogr8iiB of transit time distrib.ztions moosured in the starxJard 90 deg.
cl.assifier.
a-b: of the risi.ng p3rticles acx:arding to fig. 4.1.a.
c-e: of the falling p3rticles acx:Drding to fig. 4.l.b.
R=lOt V"'5f mxie1 p3rticles: C •.
2

-73-
side wall.
v is plotted as function of Q in fig. 4.3 with vf as parameter.
pm v
It is seen from table 4.1 that the transit times of falling particles are
longest when the air velocity vf is 0. Consequently, partiele
veloeities have their lowest value then. Raising vf to the range of
values where the actual classification of particles is achieved leads to
a fast decrease of the transit times compared to the situation without
air flow. In this range of vf the influence of the partiele feed rate
Qv and of the local partiele flow rate Ff which depends upon both
vf and Qv becomes of more importance than the influence of vf is
upon v and t 4.3). Higher local flow rates in general
pm p
correspond to higher partiele transit times and to lower values of
v • For rising particles Q and vf globally influence the transit
pm v
times in the same way ás for falling particles.
The influence of Qv (or better: the local partiele flow) and of vf
upon the partiele veloeities is explained with the aid of the airflow
profile. This airflow has been investigated by Senden /4.1/. He measured
a main rising stream and a reverse flow of air along the lower zigzag
walls (see fig. 4.4). Particles slide downwards along the lower zigzag
walls in the absence of an airflow. In the presence of an airflow,
particles will move downwards on an air cushion which is formed by the
reverse flow. Particles then move faster than in the absence of an

d:imansions in an.

Figure 4.1.
M::asure~rent set-up for the determ:ination of the velocity of rising (a) and falling (b)
[XJrt:icles in the standard 90 deg. classifier.

-74-
 airflow. The.influence of variations in vf is limited due to tr~

narrow range of practical vf values that have been applied for

classification.
Partiele-partiele interactions become more important at higher partiele
feed rates. The increased weight of the downwarcts moving partiele stream
is then no longer fully supported by the air cushion. Particles contact
the wall more easily and their speed is slowed because of the increased
friction.
The velocity of rising particles decreases at higher Qv values.
Particles will hinder each other more at higher concentrations as
partiele-partiele interactions become more important then. These
interactions may result in the formation of partiele conglomerates in
which parts of the surfaces of the particles touch each other. These
conglomerates are "thicker" than the individual particles and thus will
have higher terminal fall velocities. A simple force balance can
illustrate this:

2
1/2 pf vs Cd Ap = dp Ap g (p-o..)
p 't
(4.2.a)

V
v2 d g (p -g,)
= --E...: _p_ (4.2.b)
s c p
d f

V
Fisure 4.3.
po
(m/s)
1
~ ..... a) Stmdard <.n <~eg. c.l.as!rifier - Partiele velocity v
pn
as a tunetion
.1 r;... ....................
....
~ter: vf (m/s) . 0 1.43 of partiele fa?d rate Qv in the
I:J. 1.58
r ... '"t."h - - - --<> o c 1.7l standard 90 classifier.
0.9 r- ..................... -~-. .._
__ ~ ---""':' c2 •
0:1
............ .......

---..o --- R=lOr V=5; llCide1 particles:

0.5
0 1.0 2.0 ~ (g/s)

VJllll I
(m/s) ~ .... b) Stmdard <.n <leg. c.l.as!rifier
1.2 .....
,
~'....
D .... .... ............ _...
10 1-
... ...... -- ... ---
• ~
- ... •• , ___I:J. •• o.~=~==-==~J +++
Air flow

0.8 Parareter: vf (m/s) 0 1.45

I:J. 1.58 Figure 4.4.
risiug particles C 1.71
0.6 1 Air flow pattem in the standard
0 1.0 2.0 ~ ('1/s) 90 deg. classifier.

-75-
Table 4.1.

Partiele velocity measurements inside the standard 90 deg. classifier.

'lle "_.transit tine tp' t:re IIW!d!M. tpn' 800 t:re sta!ldard deviatioo atp of t:re ............d transit time~
!\mctioo f{tp) as functioos of !Te portiele faiVl rate Q" 800 of ti1e superficial air vekclty vf' 'lle lo:al flow rates
of t11e ~ tmticle streton (Ff) om the r i s i n g - (fr) are als> given.

a) fall~ porticles. b) r~ porticles.

~ locatioo be"..,., stage 3 and 4. ~t l.a:atioo bet:wm stage 4 and 5.
Detector ""'-<JP as in f~ 1.l.a. ~tor ""'-<JP as in figure I. I.b.
Detector distaoce Ldet: 6.1 on. [l,te::tor distaoce Ldet: 6.1 nm.
~biel porticles: DIS 'hle1 porticles:DI.S

Q" •r t a Q" t 0
p
t
pn tp Ff •r t
p pn tp Ff
(gis) (m/s) (s) (s) (s) (gis) (gis) (m/s) (s) (s) (s) (gis)
0 0.0 0.008 0.008 0.031 0.0 0 1.44 0.048 0.039 0.032 o.o
0 1.41 0.052 0.0'.7 0.023 0.0 1.61 0.055 0.0380.()1.5 o.o
0 1.!'6 0.051 O.\A6 0.024 0.0 1.71 0.0'.9 0.035 0.010 0.0
0
0.73
!.tB
1.41
0.051
0.();2
0.0'.5
0.059
o.w
0.021
0.0
3.!3
I g.71 1.45
0.61 1.57 O.Q57
O.Q58 O.aï6 0,()1.3
0,052 0.()1.7
2.52
6.1()
0.65 1.$ 0.065 0.051 0.<:00 7.4!. 0.64 1.71 0.052 0.052 0.0'.7 6.75
O.I'B 1.72 O.lli8 0.055 0.021 6.50 1.32 1.45 0.073 o.w. o.or.s 4.95
1.1() 1.45 0.065 0.();2 o.coo 8.00 1.28 1.57 O.OI'B 0.057 O.\A6 15.12
1.71 1,(1) O.Q57 O,Q56 0.022 16.00 1.28 1.71 0.065 0.051 0,()1.4 12.50
1.49 1.72 0.059 0.057 o.coo 9.a\
2.13 1.45 0.(8) 0.077 0.025 13.53
1.99 1.$ 0.071 O.Q57 0.025 18.AJ

Table 4,2,

Partiele velocity measurements inside the standard 120 deg. classifier.

The """"' transit tiae "Ç• !Te liiE!dian, tpn' 800 ti1e sta!ldard devistioo otp of the ............d transit time distributioo
function f(tp) as !in:tloos of !:he partiele fssd rste Q" and of the superficial air velocity •r· 'lle lo:al flow rates
of !:he~ Jl'rticle """"""' <Frl and ti1e of !:he rising strEam (Fr) arealso given.

fall~ tmticles.
~ locatloo bet:wm stage 3 and 4.
Detector set-up as in figure l.S.
Detector distaoce Ld: 7.2 an.
~biel porticles: 11s
Q" vf tp t
pn
atp Ff (\, vf t
p
t
pn
0
tp Ff
(gis) (m/s) (s) (s) (s) (gis) (gis) l;n/s) (s) (s) (s) (gis)
0 0.0 0.072 O.OiO 0.015 0.0 4.49 2.42 0.077 0.071 0.022 5.77
0 2.19 0.072 O.Q57 0,(!21 o.o 4.88 2.65 0.075 0.074 0.020 10.(1)
0 2.34 O.CJ/6 O.OiO O.OZ7 0.0
0 2.44 o.CJ/6 o.tm o.w 0.0 7.84 2.41 0.075 0.072 0.020 12.33
0 2.53 0,075 O.Cli9 0.025 o.o 7.31 2.16 0.073 O.tm 0.020 8.51
0 2.62 0.074 O.tm 0.025 o.o 7.48 2.65 O.IJ/6 0.074 0.020 15.42

3,); 2.17 O.Ql6 O.Cli9 O.O:U. 4.37 15.4 2.14 O.IJ/6 0.072 0.021 16.0'.
!.~ 2.14 0.073 OJJ10 0.025 2.13 15.0 2.39 0.074 0.073 0,020 Al..ai
J,<n 2.1() O..t-16 0.071 Q.a2l 6.52 18.0 2.64 0.002 0.078 0.025 :D.Z>
i 1.75 2.65 0.079 0.()75 O.lm 3.77

-76-
 F1gure 4.5.
~ set-up for the determination of the
velocity of falling partieles in the sta008rd 12JJ
deg. el~BSifier.

s : side detector arrays

d:inensions in an.

f(tp) f(tp)
(1/s) Stan!ard J.a) deg. {1/s) Standard m deg. c.lassifi~
40
a)
o_,..0 g/s; mis 40 b) gis; m/s

:D
t p>().072 atp..O.OlS :D tpJJ,074 s; otpJJ.CJ2S
tpnJJ,aiB s; vp=l.O mis tpnJJ.aiB s; vpnJJ.97 m/s
al al

10 10

0 0
0 40 !D J.a) 1(() tp (IIIS)
0 40

f<v
(1/s)
40 c)
<\,=18 g/s; vf=2.64 mis
:D t; JJ.082 s: atpJJ.02S
t ..0.078 s
pn
al
v JJ.92 mis
p
10

0
0 40 8) J.a) 100 tp {m/s)

Figure 4.6.
Exa!ples of histograms of transit tioE distributions for the rising partieles ~
in the sta008rd 120 deg. el~BSifier.
Bet-up according to fig. 4.5; R=lO; V=5:
l1lXIe1. partieles DlS

-77-
 vs is the difference between the local linear air velocity v and the
1
partiele velocity v • Thus, v is the falling velocity of the
p s
partiele in stagnant air. pf and pp are the densities of the air and
of the particle. g is the gravity constant. A is the partiele area
p
normal to the airflow direction. (A •d ) is the partiele volume and
p p
Cd is the drag coefficient. Cd is a function of the Reynolds number
Re. For the values of Re that were applied in the measurements Cd can
be considered to be a constant.
The effect of increasing partiele veloeities due to higher values of vf
and consequently higher drag forces is weak due to the large fluctuations
of the local linear air veloeities in the main airflow, both as function
of time and of place (see Senden /4.1/). The fluctuations in the transit
times of rising particles are larger than those of falling particles. The
deseending particles all move downwards along the lower zigzag wal! in
more or less the same orientation. The direction of the velocity and the
orientation of the rising particles is not as sharply defined as is the
case for for the falling particles.

...
V

(m/s) '-1.16 mis (vf=O)

1.0 ... - ·o-:•------
a· • e-o.~D•
• .
--•--c-•
il' •
0.8 .
0.6 - Parameter Qv (g/s)
- • 0
.. 1.8
• 4.7
0.4 -
0 7.5 • ++
b 16.1 Air flow
0.2 ;-
Standard 120 deg. classifier. . Figure 4.8.
Air flow pattem in the standard
2.1 2.1 2.2 2.3 2.4 2.5 2.6 vf (m/s) 120 deg. classifier.

Figure 4.7.
Partiele velocity v as a lunetion of the SU[X:Jlrlicial air velocity v f lia1SIJ.t72d in
the starrlard 120 cieff! classifier.
R=lO; V=5; urxie1 particlf'B: D S
1

-78-
b) The standard 120 deg. classifier.
Veloeities of falling particles were determined. Fig 4,5 shows the
detector locations. Examples of transit time distributions are shown in
fig. 1.7. The results have been tabulated in table 4.2 The veloeities are
plotted against vf in fig. 4.7. The partiele feed rate Qv is a
parameter in this figure.
The partiele transit times t and t and thus the partiele velocity
p pm
are practically independent of both vf and of Qv.
The air velocity profile can again be used to explain these results
(fig.4.8). Senden showed that a reverse airflow along the lower zigzag
walls does not occur in the standard 120 de~ classifier. Falling
particles are therefore not significantly influenced by the airflow. They
slide downwarcis along the lower zigzag walls. This motion is hardly
affected by vf or by the local partiele concentration. Partiele
conglomerates thus behave in the same way as individual particles.
Therefore, the friction coefficient has to be independent of the
thickness of the particles. This will be explained in paragraph 4.2.3.

c) The broader tban standard 120 deg. classifier.

The veloeities of falling particles were determined between two detector

arrays that were located at the lower zigzag wall between stage 3 and 4
(see fig. 4.9). The results are summarized in table 4.3. Fig. 4.10.a
gives a plot of the partiele velocity v as a function of the air
pm
velocity. The results were obtained at very low Q (Q =0). The figure
V V
shows that the partiele velocity increases with increasing vf. The
partiele velocity vpm is plotted against Qv in figure 4.10b at a
constant vf (2.11 m/s). This figure shows that v decreases at
pm
increasing Qv • This relation between v pm and vf on the one hand and
Qv and vf on the other haa alao been noticed in the 90 deg.
classifier, This can again be explained by means of the airflow profile
within the channel of the broad 120 deg. classifier (fig. 4.11) /4.2/ -
/4.5/. The circulation flow that occurred in the 90 deg. classifier ia
also to be expected in the broad 120 deg. classifier. Falling particles
can be accellerated by this circulation flow. Thia accelleration
increases with increasing vf. Again at higher partiele feed rates
particles can push each other against the zigzag wall. This reduces the
partiele velocity because of the increased friction with the wall. A

-79-
 Fisure 4.9.
~t.set-up for the determi.nation of the
velocity of falling {Brticles in the broa:J
120 deg. classifier.

s: side detector arrays

ditrensions in an.

VIJl!
(m/s)
lttOO lal deg. c1assifier.
1.3
Falling partkles. _.-.rr-

1.0 1.1
~~~0
Falling ,mticles.
vf=2.ll m/s
0.8 a) 0.9 b)

0 2 4 6 8 <ly (m/s) 0.7

1.0 1.4 1.8 2.2

Fisure 4.10.
Veloeities of falling {Brticles, /lEaS!JI"e(j in the brood 120 deg. classifier.
a) as a tunetion of the SU[Erficial air velocity v f'
b) as a timetion of the [Brticle feed rate Q •
R=lO; V=5; nrxie1 [Brticles: Dl.S' v

Figure 4.11.
Air flow pattem in the broad
120 deg. classifier.

+ + ~à6 cm
AIR

-80-
comparison of this partiele velocity with the veloeities that were
determined inside the broad 120 deg. classifier shows that the earticle
veloeities measured in the broad 120 deg. classifier channel are higher
because of the effect of the circulation flow. The partiele veloeities
measured in the broad and the standard width 120 deg. classifiers show a
better agreement at higher partiele feed rates. The particles in the
broad 120 deg. classifier then also more or less slide downwards along
the 1ower zigzag walls.

4.2.3 Analysis of the behaviour of particles in rising and falling streams.

In this paragraph simplified models are proposed to put the quantitie~

that describe the motions of identical, flat particles in the falling and
rising partiele flows in a functional perspective. Dirnensional analysis
is used to find the dimensionless groups that describe the partiele
velocities. A power product of these dimensionless numbers will be used
in paragraph 4.2.4 to fit the experimental data of paragraph 4.2.2.

Table 4.3.

Partiele velocit;· measurements inside the broad 120 deg. classifier.

Th:! """" transit t.iJie t;;•

tiE ne:l:ian, tpn' ard tiE standard deviatim otp of tiE - tiansit t.iJie distril>Jtioo
1\mctioo f(tp) as a fur.tioo of tiE J>ll1:icla feed rate Q,. ard of tiE ~idal air velocity vf. Th:! local flow rates

-t
of the falling portiele

falling particles.
st....,. (Ff) ard

location - s t a g e 3 ard 4.
IÀ'ltector "'t-up as in figure 1.9.
IÀ'!tector distance Ld: 7.2 cm.
lt:del particles: D S.
1
the of the r:is:i.ng - (Fr) arealso given.

:Q" vf tp t
pn "(s)tp Ff
:(g/s) (""s) (s) (s) (g/s)
0 0.0 O.Oû 0.00 0.022 o.o
0 1.45 0.073 0.067 0.023 o.o
0 1.lD O.CliS o.re<~ o.an o.o
0 2.11 O.Oû O.rt51 0.022 0.0
0 2.23 O.Oû O.C!XI 0.024 o.o
0 2.33 0.001 O.C!XI 0.022 0.0
0 2.64 0.001 O.C!XI 0.021 o.o
2.0 2.11 0.004 O.re<J 0.021 2.53
4.5 2.11 0.007 0.004 0.022 5.))
9.0 2.11 0.073 0.007 O.CQB 9.8)

-81-
a) falling particles.

The following assumptions are made: Falling particles move in an airflow

along the lower sides of the zigzag walls, The particles are flat and
rigid and have dimensions 1 , b and d (1 = b >> d ). The
p p p p p p
partiele density is pp' the density of the air pf. The linear air
velocity is v • The particles all pass through a narrow volume with
1
length L', width Band thickness ~y (see fig. 4.13), The positive x
coordinate is taken downwards along the side wall. The particles only
have velocity components along this coordinate. The particles enter the
volume at a velocity vp=O at point L'o=O. The flat surface of each
partiele makes an angle a with the side wall. The zigzag walls of the
different sections make an angle S with each other.
The following forces act upon the partiele (fig. 4.12): a drag force
caused by the airflow, a force that lifts the partiele and that is also
caused by the airflow, the gravity force, the buoyancy force and a
friction force caused by the contact of the partiele and the wall.
Influences such as pressure gradients due to accellerations of the
airflow in the absence of particles are neglected. The influence of
partiele-partiele interactions is only taken into account in so far as it
leads to the formation of partiele conglomerates. Secondary motions of
the partiele such as rotations and oscillations are not taken into
account.
The different forces that act on a partiele give a translational
accelleration to the particle:

mdv /dt= l:K (4.3)

""""1>
2
The accelleration a of the partiele is proportional to (vp /L') in
the case that the forces ! are independent of the x coordinate: Assume
that a partiele has covered the length L' of the volume above. Assume
furthermore that the partiele velocity at L' is v • The component of
p
the accelleration a in the direction of L is constant (the components of
the forcesKin that direction are constant). The partiele then covers
the distance L' in a time t= Furthermore v =a•t in this
p
situation. Thus:

2 2
a= vp /0 L')or a= c' vp I L (4.4)

-82-
as for L' any characteristic length of the classifier channel may he
taken.
The quantities that determine the value and the direction of the forces
will be discussed below.

- The gravity force ( •

-g
The magnitude of this force is:

K =Ad p _g_=l b d p .&.

-g ppp pppp
(4.5)

In this equation~ is the accelleration due to the gravity field of the

earth. K works downwards in the vertical direction.
-g

- The buoyancy force!b·

~= Ap dp Pf.&. (4.6)

~ is directed upwards in the vertical direction.

- The drag force !ct and the lift force~·

Thin flat bodies in an airflow in general experience both a drag force

and a lift force /4.6/, /4.7/. The drag force has the direction of the
relative air velocity ~-~· The direction of the lift force is
normal to the free airflow (see fig. 4.12). These forces are functions of
the partiele dimensions, the partiele density and of the angle of attack
a of the airflow. This angle is the angle that the free airflow makes
with the partiele plane. These forces can be described by:

(4.7.a)

and

(4.7.b)

Al_ is the partiele surface normal to the relative airflow. C and Cd

1
are functions of a and Re for a given partiele geometry:
p

f( a,Re );
p
cd = f( a,Rep) (4.8.a)

-83-
 (4.8.b)

At the narrow range of Re values applied in our experiments

(l0 -Io ) the effect of R~ upon Cd and c will be small.
3 4
1
The angle of attack a will be a rather complex function of the airflow
profile within the classifier, of the partiele geometry, the density P
p
and the airflow conditions around the particle. Furthermore a and A~

fluctuate as a function of time. Therefore, a and A~were assumed to be

the mean values averaged over a longer period of time.
An important variabie that determines the drag and lift farces is the
linear air velocity of the circulation flow. ~l is a function of
the shape of the flow and of the superficial air velocity. The shape of
the flow inside a zigzag channel will be a function of the geometry of
the channel (which also includes the wall roughness) and of the Rech
number. Thus:

(4.9.a)

The Rech number is defined by:

(4.9.b)
11

The hydraulic diameter dh is defined by /4.8/:

4 BW
(4.9.c)
~ = 2 (B + I.J)
W is the classifier width, (B•W) is the horizontal cross section area of
4 4
the classifier channel. Values of Rech between 10 and 5.10 were
obtained in most of the experiments. At these high Rech numbers the
shape of the airflow will be practically independent of Rech /4.9/.
Thus the linear air velocity v in the circulation stream will be
1
linearly proportional to the superficial air velocity through the
channel, vf' for a given classifier geometry.
- The friction force -w
K•

The contact between the partiele and the zigzag wall causes a friction
force K • The direction of K is in the negative x-direction, For a
-w -w
partiele that slides down along the lower zigzag walls in the absence of
the circulating airflow the friction force is:

K =-fIK ~ (4.10.a)
-w ·"'N x

KN being

~ = ~cos( S/2) (4.10.b)

K has already been defined by 4.5.

-g
In the case of particles rnaving downwards in an airflow the situation
becomes different (see fig. 4.12). The partiele will not be attached to
the side wall along its whole surface. This may lead to a different
friction coefficient f. Furthermore the normal force~ will now depend
upon both the weight of the partiele and upon the lift force ! 1 • Also
the friction force does not attack at the center of gravitation of the
particles as all the other forces are assumed to do. Thus, in principle
the partiele will start to rotate. However, the moment of rotation will
be small as a will in general be small, It is therefore assumed that the
angle a does not influence the friction coefficient f,

The force balance of an individual partiele can be described by means of

equation 4.3. However, the interaction between different particles has an
important influence upon the motions of particles. The following
assumptions are made regarding the interaction between particles that
move at more or less constant velocity in the same direction:
Contact between two particles may lead to the formation of a conglomerate
consisting of the two individual particles. The aerodynamic properties of
this conglomerate are the same as those of a single partiele that has the
the same mean thickness and mean surface as the conglomerate. It is
assumed that conglomerates of more than two particles are not forrned. Now
suppose that the fraction of particles that form a conglomerate during
their stay in a given volume V is y and that the fraction of the surfaces
that overlap each other is x. The flat surface of the conglomerate

-85-
becomes:

A' = (1-x+l) A = (2-x) A (4.ll.a)

con P P

and the mean surface of both conglomerates and individual particles

wit~n the volume:

(1-y) = (1-xy/2) A
p
A (4.ll.b)
con
(1-y/2) (1-y/2)

The mean thickness of a conglomerate becomes:

(2 (1-x) d +x 2d ) A 2
d' = p p p d (4.12.a)
con A' 2-x p
con

and the mean thickness of both conglomerates and individual particles is:

2 (1-x)
(1-y) d + y/2- d (1-y - )
p 2-x P (2-x)
d d (l.l2.b)
con p
(1-y/2) (1-y/2)

The probability that particles may form conglomerates (y) as well as the
size of these (x) depend upon the hold-up of the particles (defined as
numbers of particles) in the classifier volume under consideration and
upon the dimensions of the partiele compared to that volume. This holdup
H' is (fig. 4.13):

F V FL
H' = - (4.13.a)
v Btw v
p p

in which F is the partiele flow rate (Hz). Thus:

d A
con= f (H') and con (4.13.b)
d l A
p p

Above discussion of the physical quantities that play a role in the

description of the particles moving down along the side wall of the
classifier show that it is difficult to solve the differential equation
4.3 as not all fundamental relations between the variables are known.

-86-
Dimensional a~alysis is therefore used to determine the various groups
that describe the partiele velocity.
The partiele velocity can be described by a function of the form:

However,as has been shown above, a number of additional relationships

exist (equations 4.7, 4.8, 4.11-4.13) as wellas a rather complex
relationship for a:

(4.15)

Thus, equation 4.14 reduces to:

v f(Rech; Re ; 1 ; b ; d ; g; p ; p f; vf; L; B; H; f; f3; F) (4.16.a)

p p p p p p

It has been stated above that the effect of Rep and Rech can be
neglected, thus:

(4.16.b)

This relation contains 14 variables that are defined in terms of the

three reference dimensions: length, mass and time. Thus, according to the
Buckingham "pi-theorem" /4.10/ the equation can be reduced to 14-3=11
independent dimensionless groups. The fol1owing set of numbers can be
derived:

2
v /(L g) = f((v /vf); (R /pf); (L F/v ); (B/L); (W/L);
p p p p
(1p/L); (bp/1p); (dp/1p); f;S) (4.17 .a)

For a given classifier geometry and for given model particles, only
2
(v
p
/(L g)), (v p /vf)' and (L F/v p ) will vary with different
Qv and vf. The fo11owing relationship can then be expected:

2
(vp /(g L)) =f((vf/vp); (L F/vp)) (4.17.b)
2
The dimensionless number (v /(g L)) is called the Froude number
p
(Fr). This number represents the ratio of inertia and gravity of the

-87-
 Figure 4.12.
The torcES that act UfXJil a partiele that m::wes
down in an air classifier.
v : cira.ü.ation air flow; Kef weight of
1
particle; K : lift toreer Ki drag force;
1
K : friction force.
w

Figure 4.13.
~ions of the classifier stage
and of the lllXielling volr.me.

Table 4.4.
Results of best fit calculations.

~ constants of the equation:

<2 °3
:e._=cl (::!_)
vfl vfl vfl

with vfl l(g dp P/Of)' F = Ff/(VP PP) for falling partjeles and F = F/(Vp Pp)
were determined by means of multiple linear regression using the data of the partiele velocity
measurements.

vn cl c2 5
c2 c3 5
c3
R s
(m/s) H H H (-) (-) (-) (-)
st. 90 deg. classifier~
-falling particles (table l.I. a): 0.99 0.923 0.090 0.013 -0.027 0.006 0.925 0.0053
-rising particles (table l.l.b): 0.99 0.838 0.19 0.33 -0.033 0.005 0.938 0.0037

st. 120 deg. classif i er.

-falling particles (table 1.2): 1.40 0.679 -0.005 0.004 -0.002 0.002 0.470 0.0023

broad 120 deg. classifier.

-falling particles (table 1.3): 1.40 0.669 0.288 0.084 -0.009 0.003 0.860 0.0027

R: multiple correlation coefficient.

and sc : standard error of constants c and c respectively.
2 1 2
standard error of estimate.

-80-
partic les.

b) rising particles.

It is assumed that rising particles only experience a drag force which is

directed in the direction of the airflow. Furthermore the interaction
between the particles and the wall is negle~ted. the dimensirnLess
equation that describes the problem is of the form:

2
vp /(L g) f((vp/vf); (P./o.);
p'l: (L F/vp); (B/L); (W/1);
(1 /1); (b /1 ); (d /1 ); 8) (4.17.a)
p p p p p

or again for a classifier of given dimensions and for given particles:

2
v /(g L) = f((vf/vp); (L F/vp)) (4.18.b)
p

4.2.4 Comparison of the different classifier geometries.

The relations 4.17.b and 4.18.b of paragraph 4.2.3 have been used to fit
the data of tables 4.1, 4.2, and 4.3. In order to show the effects of
vf and the local partiele flow rate upon vp separately, these
eqations have been transformed into relationships of the form:

(4.19.a)

By means of rearranging the dimensionless numbers as fellows:

vf vp vf LF vp FL
;-=- -;-=-- (4.19b)

formed from the original set (eq, 4.17a and 4.18.a) by means of taking
powerproductsof the original numbers /4.11/.
A power product relationship between the various dimensionless groups of
equation 4.19.b was assumed. The results of the curve fits have been
tabulated in table 4.4.
It follows from this table that the partiele veloeities v of the
p
particles increase with increasing vf for all classifier geometries
with the exception of the standard 120 deg, classifier. It is con~luded

-89-
from both the low multiple correlation coefficient and the high standard
error of the power c of the term (v /vf 1 ) that no clear influence
- 2 p
of vf upon tp exists for the case of the falling particles in the
standard 120 deg. classifier. This geometry lacks the circulating airflow
that occurs in the other two geometries. Falling particles in the
standard 120 • classifier carry out a sliding motion with a constant
friction coefficient f. The equation of motion is:

dv
p
I dt = g (cos( S/2)- f sin( 8/2)) (4.20)

This relation does not contain the weight of the particle. Thus v will
p
be independent of the partiele flow rate which leads to conglomerates
with a larger weight than the individual particles. A necessary condition
for this is that the particles start to slide down from the same starting
point and with the same starting velocity under different operating
conditions of the classifier.
Broadening the standard 120 deg. classifier leads to increased veloeities
of the falling particles that moreover become dependent upon Qv. An
airflow pattern inside the braad 120 deg. classifier that contains a
circulation airflow like the one noticed in the standard 90 deg.
classifier occurs in this geometry. The maximum veloeities observed are
comparable in bath classifiers (1.2 m/s). Increasing QV causes the
particles eventually to contact the lower section wall and to continue
their way down in a sliding motion with wall contact according to eq.
4.20. The value of v then reduces in the braad 120 deg. classifier to
p
the value observed in the standard 120 deg. classifier. The drop in the
value of v in the standard 90 deg. classifier is stronger than in the
p
broad 120 deg. classifier because the sliding velocity with wall friction
is lower here. The dependenee of v upon is weaker in the broad
p
120 deg. classifier than in the standard 90 deg. classifier as can be
concluded from the power c of (F L/vf ) tabulated in table 4.4. This
31
caused by the fact that the weight component perpendicular to the
lo·..;er section wall that has to be supported by the air lift force is
smaller in the broad 120 deg. geometry than in the 90 deg. classifier.

-90-
4.3 Partiele transition probabilities and local partiele flow rates.

4.3.1 Introduetion.
The results of measurements carried out in the three classifier
geometries will be presented and qualitatively explained in section
4.3.2. In section 4.3.3 the variables that determine the values of pf
and pr will be formulated in a more quantitative way. A simplified
model is presented to give a description of the partiele transition
probabilities as function of the dimensionless numbers that contain these
variables. Correlations between the transition probabilities and the
variables are calculated. These correlations will be used in chapter 5 in
which a mathematica! model is presented that describes the separation
process in zigzag air classifiers.

4.3.2 Messurement results.

a) The standard 90 deg. classifier.

- partiele transition probabilities.
Results for the transition probabilities are given in table A4.1. of
appendix 4. Data at stage boundaries no. 6 and S,are notpresentedas the
feed system made accurate measurements at these two stages impossible.
Transition probabilities as functions of vf obtiined at very low
partiele concentrations are shown in fig. 4.14. Parameter is the stage
boundary number. Both pf and pr are highest at boundary no. 1. pf
and pr decrease towards higher stage boundaries.
There are two reasons for the fact that, even at "zero concentration" the
partiele transition probabilities depend upon the stage boundary number:
- development of the air flow profile at the bottorn entrance.
- the feed stage influences the air flow around the feed point.
In his experirnents, Senden also rneasured the same influence of bottorn
entry effects upon pf and Pr·
The drawn lines in fig. 4.14 indicate the mean values of pf and Pr•
These mean values were calculated according to:

) Pf .*Ff .
"i ,1 ,1 and p =
I. pr,1.*Fr,1.
1
(4.21)
I
i
Fr .
,1
r Ii F .
r,1

-91-
 • . • '.l
. Pr I
Pr
~=() g/s. • 1•1
(-) ~
• .tol lol "'1 (-)

s -
1
.t.1 .2
0.8 r- 0.8
• 1
. lol ~ ~

0.6 "'~3
...
1
·~ ~ t" •"sl~t~ ti . 0.6
7 3 .....

~J .e." t" ~7 s . 5 4
As

.
3
a~ . •1 •tztc!l1

~ i'~4~
l3 §3
0.4 i- l • • • • 2 • 2 2 07 '
0.4
2
~3&2 • 5~ 3
03 ril (9 .. 3~4 5 5CDt.
07~
.JÛ
•.
0.2
_ 0
5
os 3os 5
- 0.2
0.0
Stamard ~ deg. classi fier.
I I I
- o.o
2.0 2.1 2.2 2.3 2.4 2.5 vf (m/s)

FiRUre 4..14..
PélTticle transition probabilities pf ( and ) and
superficial air veloei ty in the stafldard 90 deg. claS.si.fier.
R=lO; V=5; nrxJ.e1. p9rticles: Dl.5; Q =0.
Nurrbers refer to the stage boondary~ the drawn lines represent the me:m transition
probabilities accord:ing to equation 4.21.

Pr Pr
(-) fàraneter: ~ (g/s); Stage boundary: 3 (-)
e o
0.8 0 1.4 0.8
Á 2.7 ~c..,. "'adrP o•
0 4.7 0 ""'
o•0 e ee
0.6 Cl •• 0.6
• • o..,.
• 0

0.4

•oe 0
..,.. 0
D
• 4t
•
<à.:.o
---;;....
. _.
"'a
oe
0.4

0.2 0.2

0
2.0 2.1 2.2 2.3 2.4 2.5 vf (m/s)

Figure 4..15.
Partiele transition probabilities I1EélSl.!1"ed at stage boondary 3 of the standard 90 deg.
classifier for different air veloeities v f and feed rates Q •
R=lO; V=5; nrxJ.e1. p9rticles: Dl.s' v

-92-
Figure 4.15 shows the partiele transition probabilities that were
measured for different vf and different Qv at stage boundary no. 3.
It follows from this figure that the values of pf at given process
conditions are always higher than the values of Pr·
Apart from the values found at Qv is 0 g/s, pf increases at
increasing Qv as can be seen from figure 4.15. This dependenee of pf
upon Qv and thus upon the partiele concentration is explained by the
veloeities of the deseending particles which was discussed in section
4.2.2. Increasing local partiele flow rates causes a decrease of the
velocity of the falling particles that enter the main upward air flow at
edges protruding into the channel (fig. 4.16). The partiele inertia will
be lower then and the particles may be taken upward by the rising air
flow more readily. Consequently, pf will increase at increasing Qv.
This result agrees with the increased fraction of bottorn product, ~R'

that has been noticed to occur at increasing Qv and constant vf in

the standard 90 deg. classifiers. (see chapter 2).
p does not depend much upon the partiele feed rate. This can be
r
explained by the trajectories of particles in the upward air flow.
Senden showed that rising particles collide with the upper section wall
at an angle of impact that is almast perpendicular to that wall. The
particles loose almast all their kinetic energy so that this energy is
no longer available for the continuation of the upward motion.
Increasing partiele-partiele interactions at higher feed rates leads to
lower partiele veloeities as has been shown in 4.2.2. However, these
lower partiele veloeities influence neither the partiele trajectories
nor the values of the transition probabilities. Other factors such as
the orientation of the particles after the callision and the
instantaneous scalar value and orientation of the local air velocity
strongly influence the trajectories that the particles cover after this
collision. These factors seem to be relatively insensitive to the
partiele feed rate. The results of the measurements at the other stage
boundaries show that these conclusions also hold for these stages
table A4.1 of the appendix).

- Local partiele flow rates.

The partiele flow rates of falling and rising particles were deterrnined
simultaneously with the partiele transition probabilities. Table A4.1 of

-93-
appendix 4 also eontains these measurement results.
For a given elassifier, loeal partiele flow rates are determined by the
partiele feed rate Qv and by the partiele transition probabilities at
the different stages. In general, partiele flow rates are highest near
the feed loeation and eontinuously deerease towards both elassifier
exits. An of partiele flow rates at the different stages is
given in 4.17. Flow rates at the different stages are both a
funetiOJ. of Qv and of v f" Figs. 4.18 show these flow rates as

Figure 4.16.
Influence of the partiele velocity UfXJf1
the trajectDries of particles
insiele the standard 00 classifier.
Trajeetory b: high entry velocity at a.
Trajeetory c: low entry velocity at a.

Fr
Ws)
25

15 15

10 10

5 5

0 0
0 2 3 4 5 6 7 8

section no ..

Fisure 4.17.
Local partiele flow rates measured in the standard 90 deg. classifier.
functions of vf' Parameter is the partiele feed rate. The local
partiele flow rate increases as the partiele feed rate increases. A
maximum occurs at those air veloeities where the particles do not have a
preferenee for leaving the classifier either through the top exit or
through the bottorn exit.

- Effect of local partiele flow rate upon pf and Pr·

It has been illustrated above that partiele transition probabilities
depend upon the partiele feed rate Qv. The transition probabilities
Pf and pr at a constant vf differ on the various stages due to
local differences in the partiele concentrations. The question is whether
there is a relation between pf and pr on the one hand and the local
flow rates on the other. This aspect will be dealt with in a more
quantitative way in section 4.3.4. Here, plots of pf . and p .
1 .~ r,~-
versus the flow rate Ff . in figure 4.19.a and b show that, apart from
,J.
the values of pf measured at stage boundary no. 1 and of pr at stage
boundaries 1 and 2, the transition probabilities may be described as
functions of the local partiele flow rate for a given value of vf. The
figures also show that the deviations are still rather large. Reasous
for this are:
- the partiele transition probabilities are not completely independent
of the stage boundary number.

Ff F
r

j~--3
(g/s) (g/s)
25 ,A
I
1
\
\:
• Amm!ter: ~
I
25

;,!} I ~ '~-!!
• . -.~ - ;,!}
I I ~

15

10
./ j/~
/" I
I /-/1~
•
8~
'8
1

}--"j;;o>
15

10
••/ ...-:::-::2' ~ (g/s)
5 -o--o-...- • 1.4 5
AA 2.7
Starmrd 'D deg. classif:ier. • [J 4.7
0 0
2.0 2.1 2.2 2.3 2.4 2.5 vf (m/s)

Figure 4.18.
Ux:a1 flow rates vs. the air velocity vf and the feal rate Qv at stage 3 of the
starxiard 90 deg. classifier.

-95-
 Pf,i Pr,i-1
a)
(-) Standard IJ) deg. clamifter. (-)
0.8 ___ .. .!..~----;;r--·--•1 0.8
1
- __..SJS
0.6 o;_ --- 0 2 - 1t 0.6
0
-..P·---0. 03
05 0 2

04
..--2---

0.4 2
2 0.4
1 - - - - - -.. 1 - - -
3 •1 &2 "".ii---2
4 65
5
- -•"3"'•s -A"'-.!.3 -AT -A3 •s- -
0.2 0.2
pf: OBM e vf: 2.15 m/s.
pr: t:.IDI •
0.0 o.o
5 10 15 25 Ff . (g/s)
,1

Pf,i Pr,i-1
(-)
1 - .. - n -1-. . i.. l StaoJard IJ) deg. clag:dfter b) (-)
0.8 0.8
~a_- -~--o3.03-q,4-
__ o2--o; 4
2 0, 5
3 _ ...... - o4 o
0.6 4----
5 0 7
os 0.6

0.4 0.4

0.2 0.2

o.o o.o
5 10 15 25 · Ff ,1• (g/s)

Figure 4.19.
Partiele transition probabilities pJ . and p . vs. the local flow rate Ff .
1
inside the standard 90 deg. cl.assitïM. The ~ refer to stage boorK:IarieS~
R=lO; V=5; irodel p3rticl.es: D •
15
a) vf=2.15 m/s.
b) v[=2.35 m/s.

-96-
- each individual point contains a measurement error.
- The plots only show the dependenee of Ff .• It will be shown in
,1
section 4.3.4 that F . in principle also influences the values of
r,1- 1
. 1•
pf ,1. and pr,1-
Apart from these deviations it appears that pr is not much influenced
by the local partiele flow rates. pf tends to increase at increasing
values of Ff. For values of Ff less than appr. 10 g/s pf' pf is
not much influenced by the partiele flow rate.
These moderate effects of local flow rates on the single stage
transition probabilities cannot he neglected as these small but
statistically significant effects are strongly amplified in combination
with f.i. ten stages.

b) The standard 120 deg. classifier.

- Partiele transition probabilities.

Partiele transition probabilities were determined in dependenee of both

vf and Qv. Table A4.2 of appendix 4 contains the results. Fig. 4.21
contains plots of pf and pr vs. vf measured at different stage
boundaries for very low partiele feed rates Qv. This figure shows that
both pf and pr are highest at the lowest stage boundaries. However,
this influence of stage location upon pf and pr is less than is the
case in the standard 90 deg. classifier.
The values of pf and of pr are almost equal at a given vf and
Q • The direction of the air velocity profile (fig. 4.8) gives a
V
better insight into the partiele behaviour at a single stage. Because of
the larger angle between the different sections in the standard 120 deg.
classifier the air velocity is deflected less than in the standard 90
deg. classifier. A circulation flow along the lower section walls was
not measured /4.1/. Moreover, the local veloeities within the main
upward air flow channel are lower than in the 90 deg. classifier
Figure 4.20.
DiEferences in trajectories of falling
particles insiele the standard 90 deg.
clEBSifier at low (a) and at higher partiele
flow rates (b).

-97-
pf
(-) StmxJard UD deg. dl'lfBifier.
0.8 '\,=Û g/s

0.6 1
2 ibS
~· 3 a'
,.
4
0.4
~-
5
0.2 4
2 0
5

0.2
1.9 2.0 2.1 2.2 2.3 2.4 2.5 2.6 vf (m/s)

Pr
(-) StmxJard la) deg. class:i.f:ier
0.8 '\,=Og/s

0.6
4

0.4
~:
02
2
3

0.2 sg

o.o
1.9 2.0 2.1 2.2 2.3 2.4 2.5 2.6 vf (m/s)

Figure 4.21.
Partiele transition probabilities as a function of the superficial air velocity in the
standard 120 deg. classifier.
R=lO: V=5; mx1el particles: Dl.5; Q =0.
Nwbers reler to the s~ boundaryr
drawn lines reprasent the nmn tlCITlSition
probabilities according to equation 4.21.

-98-
at correspondtng superficial air velocities. However, the width of the
main upward air flow channel is braader in the standard 120 deg.
classifier than in the standard 90 deg. classifier. Falling particles
that enter the upward air flow of the 120 deg. classifier are subjected
to a smaller drag force than in the 90 deg. classifier. The time during
which they are subjected to this drag is, however, langer because they
have to cross an upward air flow channel that is braader than in the
standard 90 deg. classifier. Senden /4.1/ showed by m~ans of partiele
trajectory calculations that individual particles that fall down will
less easily be taken upwards again by the main air flow inside the
standard 120 deg. classifier than in the standard 90 deg. classifier.
Furthermore, the falling particles of the standard 120 deg. classifier
enter the main upward air flow at a larger angle between the surface of
the partiele and the direction of that flow. The drag on the partiele
area normal to that air flow will thus also be smaller than would be the
case in a standard 90 deg. classifier at equal vf. All these effects
lead to the pf values that are lower in the standard 120 deg.
classifier than in the standard 90 deg. classifier at equal values of
V f" u
It has been explained above that particles moving upwards in the 90 deg.
classifier may easily collide with the higher zigzag walls and loose
almast all of their speed. pr values are therefore relatively low and
are insensitive to the value of vf. The loss of kinetic energy of
rising particles inside the 120 deg. classifier is less as these
particles do not collide with the wall under an angle of 90 deg. The
particles have to be forced less to change the direction of their speed
at each stage. This results in a greater dependency of pr on vf that
moreover may reach higher values at increasing vf than were measured
in the standard 90 deg. classifier.
Figs. 4.22 show that both pf and pr measured at stage boundary no.3
tend to decrease at increasing feed rates Q • These results are caused
V
by the fact that partiele-partiele interactions result in conglomerates
of particles that are thicker than the individual ones. The falling
veloeities of these conglomerates are higher than of single particles.
This same phenomenon was also noticed for the other stages (see table
A4.2).

-99-
pf
(-)
SUu:mrd la) deg. ·classifier.
0.8
stage boundary 3

0.6
Paraneter:
•
•
0,.
0
2.2
(g/s).

. •
. ...
0.4
•
0
4.5
9.1 • • • "
••
0.2 • •

1.9 2.0 2.1 2.2 2.3 2.4 2.5 2.6 vf (m/s)

pr I I I I I I I
(-)
SUu:mrd la) deg. c1assifier. -
0.8 1-
stage boundary 3 •• 0
2.2
-
1-
Pa:raneter: 0,. (g/s) •
• 4.5 -
0~6 ~ 0 9.1
• .._ -
1--
• • •
0.4 r-
• •
•... ... lt.
0
• -
1- • e -
0.2 1-
-
0.0
1-
I I I I I I I
-
1.9 2.0 2.1 2.2 2.3 2.4 2.5 2.6 vf (m/s)

FiB!!.re 4.22.
Partiele transition probabilities aeasured at st:.ag::! 3 of the standard 120 deg.
classifier vs. the air velocity vf'
R=lO; V=5; l1lXiel particles: D1• •
5

-100-
- Local partiele flow rates.

Partiele flow rates are shown in fig. 4.23 as a funetion of vf. Again,
these flow rates depend upon both the feed rate and upon the partiele
transition probabilities pf and Pr· Fig. 4.24 gives an example of
the flow rates measured at the different stage boundaries. This figure
shows that a maximum flow rate oecurs near the feed location. The value
and the exact location of this maximum is a function of both the air
velocity and the partiele feed rate. The local partiele flow rates
decrease from this maximum towards both classifier exits.
At a given partiele feed rate and at a given dimensionless air velocity
vf/vfSO' .i.e. at that air velocity where both the standard 90 deg.

Ff
(flls) Standani lal ~. classifi.er.
• F
r
(flls)
15 stage lx>Undary 3
làraleter: (\, (g/s) .....-a
/'
15

2.2eo ..-i--•
10 4.S.AA ~ __.-A 10
9.1.0 ~----
~ /A
_...::C • 5
5
•--;:::-A•-: _ - -0=:-::::-::::a
~---o
0
2.1 2.2 2.3 2.4 2.5 2.6 vf (m/s)

Fi81Jre 4.23.
li:x:;al ~cle flow rates vs. the air velocity v f nmsured at stage 3 of the Standarel
120 deg. classifier.
R=lO; V=5; Jrodel particles: D • •
15

i
St:andard 120 deg. class.
I Fr
Q =4.7 g/s . (g/s)
.----+-.., • I v~=2.69 m/s 10

5 5

0 0
0 2 3 4 5 6 7 8 9
section no.

Fi81Jre 4.24.
li:x:;al partiele flaws nmsured at the various sections of the standarel 120 deg.
classifier.
R=lO; V=5; Jrodel ~cles: DIS

-101-
Pf,i
(-)
' .
0.8 .
.
0.6 .
.
0.4 .
- .
0.2 - .
1-
a)
.
I I I I I

0 5 10 15 20 25 Ff ,1. {g/s)

Pr,i-1 I I I
(-)
- Standard 120 deg. classifier. -
0.8 - vf=2.64 m/s -
t~'
~5 -~0
-
0.6 ~-~ 2... ••• 1
z-•1
~ 7 ()07 0;
F-4 6 5--
~4
- •-
0 -
0.4 - Cf; 0;,- -oa_ - - o2-o3
04 -
-
(1<

0.2 -
0.0
- b)

I I I I I
-
0 5 10 15 20 25 Ff . (g/s)
,1

Figure 4.25.
Partiele transition probabilities vs. the local flow rate Ff . inside the standard 120
deg. classifier. IU1lbers refer to stage boondaries. ,1.
R=lO; V=5; arxJe1 particles: D • •
15

-102-
classifier and the standard 120 deg. classifier have the same fractim.
of bottorn produ~t ~R' the flow rates in the 120 deg. classifier are
lower than the flow rates inside the 90 deg. classifier. This can be
explained by means of the values of pf and pr that were measured.
The particles inside the standard 90 deg. classifier show a greater
preferenee to stay within the same channel sectien (pf > pr) than the
particles inside the 120 deg. classifier (pf "' pr). Therefore both
the local flow rates and the partiele residence time inside the 90
classifier channel will be larger than in the 120 deg. classifier.

- Effect of local partiele flow rate upon pf and Pr·

Fig. 4.25 shows the measured pf . and p . values plotted against

.~ ,r~- 1

the values of Ff,i at vf=2.64 m/s. The values of pf and pr at

stage boundary 1 differ from the values found at the other stages. The
figure shows that both pf and pr decrease with increasing local flow
rates at higher feed rates. This has been explained above by the
formation of partiele conglomerates that have higher falling veloeities
than the individual particles. The variables that influence the partiele
transition probabilities are discussed in more detail in section.4.3.3.
The correlations that describe the relationship between pf and pr on
the one hand and the loeal partiele flow rates and vf on the other
hand are also presented in that seetion.

c) The broader than standard 120 deg. classifier.

- Partiele transition probabilities.

Table A4.3 of the appendix eontains the partiele transition

probabilities pf and pr as well as the local partiele flow rates
that were measured for different vf and Qv. Data on stage 6 and 8
are again not presented.
The transition probabilities pf and pr at very low partiele
concentrations with the stage boundary number as a parameter are plotted
in fig. 4.26. These results show that the influence of entry effects of
the air flow upon pf and pr are less than in the standard 90 and 120
deg. classifiers. Apparently the air flow that enters the channel shows
a greater similarity to the air flow profile that has to be created
within the channel when this channel is broadened.
The values of pr and pf tend to increase at increasing height within

-103-
 "r ,J Pr
(-)
H Broad 120 deg. classifier.
0.8 8road 120 deg. classHier. 0.8
ParamcHPr: Q, (g/s)
Qv~o g/s
0 0
0.6 0.6 A 2.1 ê.
0 4.4
• 9.1
0
0 •
0.4 0,4 § 'i
•
0.2 0.2 .
a) ~ a)
0.0 o.o
1.5 2.0 2.5 vf (•/s) 1.9 2.1 2.3 2.5 vr(•/s)

Pr
I Broad 120 deg. classifier.
..... Pr,i-1 I (-)
0 (-) Parameter: Qv (g/s) .
,
.!>- Broad 120 deg. classifier. 0.8
I
~
0.8 0 0
Qv•O g/s 87 3 A 2.1 0
07 ~ 0.6 0 4.4 0
•

/
~~! 0~
7
0.6 • 9.1
.,
0
(;.
a.
Os 0.4
0.4 i
0
a.· (;.

02 0.2
0.2 b)
b)
o.o I 1.9 2.1 2.3 2.5 vf(•/s)
1.5 2.0

Figure 4.26. Figure 4.27.

Partiele transition probabilities vs. the ~icial air velocity in the Partiele transition probabilities vs. the S!JfX3rficial air velocity at
broed 120 deg. elassifier. stage 3 of the broed 120 deg. elassifier.
Low QvJ Nunbers reler to the stage borndary nuniJer.
the classifier. It was observed during the experiments that particles in
a rising stream may sametimes pass several stages without falling down
again. The probability for such short cuts increases with increasing
height within the classifier. The values of pr at higher stages will
thus also be higher.
The drawn lines of fig. 4.26 are the average values of the transition
probabilities at Qv=O g/s. The value of pf is always lower than of
pr for this broad 120 deg. classifier.
The high values of p compared to the values found in the other two
r
classifiers at the same vf can be explained by the fact that the broad
channel obstructs the upward air flow and partiele flows less than is
the case in the other two classifiers (compare figure 4.11 with 4.4
and 4.8).
Both pr and pf are hardly influenced by increasing partiele feed
rates as 4.27 shows for example. This is in contrast with the
standard 120 deg. elassifier where pf and pr both elearly deerease
with increasing Q • The reason for this is that loeal partiele
V
eoneentrations within the rising partiele streams of the broad 120 deg.
classifier are lower than in the standard 120 deg. classifier at equal
QV• Thus pf and pr will be influenced less by increasing QV • The
occurrence of partiele eonglomerates at higher Q that might cause a
V
deerease of pf are apparently for the greater part eompensated for by
the increase of pf whieh is due to the lower deseending partiele
veloei ties.

- Loeal partiele flow rates.

Measured flow rates as a funetion of the air velocity are given in table
A4.3 of appendix A4. Fig 4.28 shows the loeal partiele flow rates as
functions of stage boundary number. Again, loeal partiele flow rates are
functions of pf' pr and Qv • The values of the partiele flow rates
measured in the broad 120 deg. classifier are lower than the values
found in both the standard 90 and 120 deg. elassifier under
eorresponding process eonditions. The reason for this is that pf-pr
combinations oceur in this classifier in whieh pf < Pr· Thus,
particles will not remain in the same seetion for a long time. Therefore
p~ residence times and local flow rates will be lower than for the

other two classifiers that were investigated.

-105-
Ff F
r
(g/s) (g/s)
8 8

6 6

4 4

2 2

0 0
0 2 3 4 5 6 7 8 9
sect.ion no.

Figure 4.28.
I..tx:a1 partiele flows Ire8SUieCl inside the broad 120 deg. classifier,

Pr,i
(-) Broad 120 dcg. classificr.
0.8 m/s.
o,
0.6

0.4 o-o--?3°4 o
5 2 - ---02- - - - -
o'
0.2
a)

0 2 4 6 8 10 12 14 Ff,i (g/s)

Pr,i-1
(-)
Broad 120 deg. classîfier,
0.8
m/s.
7
0 7 7
o,o!" s
0.6 tta<;~ 'fls 3
o.
- - - - - -o> - -~
02 02o3

0.4

0.2
b)

0 2 4 6 8 10 12 14 Ff,i {g/s)

Figure 4.29.
Partiele transition probabilities Ire8SUieCl as a functioo of the flow
rate of falling particles in the broad 120 deg. c1assifier.

-106-
- Effect of local partiele flow rate upon pf and Pr·
Fig. 4.29 shows measured pf . and p .values plotted against the
.~ r,~-1
values of Ff,i at a constant value vf=2.39 m/s. The figures show
that both pf and pr only very slightly decrease with increasing
local flow rates at higher feed rates. This weak influence of the flow
rates upon pf and pr has been explained above.
The variables that influence the partiele transition probabilities are
discussed in the next section. The correlations that describe the
dependenee between pf and pr on the one hand and dimensionless
numbers that contain the local partiele flow rates and vf on the other
hand are also presented in that section.

4.3.3 Analysis of the partiele transitions at a classifier stage.

a) The dimensionless numbers that describe the partiele transition

probabilities.

In this section a model to define the variables that èan be expected to

influence the partiele trajectories at the individual classifier stages
is presented. The model is not supposed to give an exact fundamental
description of the partiele behaviour.
The variables will be combined to form the dimensionless numbers that
describe the partiele transition probabilities.

- Falling particles.

Assumptions:
A mass flow Ff of particles that have a mean velocity vpf enters·

FiRUre 4.30.
CkaJ;hical representatión of the ur:xfe1 that

r
determ:ines the dinensiooless f5IW1B to
describe the transition probabilitie;.

-107-
stage i (fig. 4.30). A fraction pf of this flow rises after having
passed the stage. A fraction (1-pf) continues to fall. Both the
impulse of the flow that continues to fall and the impulse of the
partiele flow that is deflected changes.
The change of impulse of these flow is caused by the forces that
influence the partiele motions at the stage. These forces are: the
gravity force, the drag force of the air flow and the buoyancy force.
·Furthermore the change of impulse is caused by the impulse flow of the
stream of rising particles that enters the same stage and collides with
the particles of the falling stream and by the loss of impulse of the
particles due to particle-wall interactions.
The partiele behaviour at higher feed rates is influenced by the
formation of partiele conglomerates which have falling veloeities that
differ from those of the individual particles. The influence of the
formation of conglomera~es is taken into account by assuming that the
change of impulse is experienced not by a single partiele but by a
conglomerate of particles, A layer of particles protrucles into the
channel (fig. 4.30). Both the direction and the value of the velocity of
this packet of particles changes. It is assumed that the mean velocity
of this packet of particles is 0 during the time that it is deflected in
the air flow.
The conglomerate has an area A
con
= Ncon Ap (A p is the area of
an individual particle) and a thickness d = n d in which
con con p
d is the thickness of an individual particle.
p
The x coordinate is taken to be positive in the vertical upward
direction. The impulse balance now becomes:

(4.22)

di/dt is the change of impulse of the partiele flow that enters the stage
in a falling stream. di/dt will be constant as the stationary state is
assumed. Kd' Kg and Kb are the drag force, the gravity and the
buoyancy. Kcoll is the force due to r1s1ng particles that collide with
the conglomerate and Kwall is the loss of impulse due to contact
between the particles originating from the downflow and the wall. Only
the components of the impulse balance in the direction of the x-axis are
assumed to influence the transition probabilities. Thus, this equation
~becomes:

-108-
pf ,J.. Ff ,J.. vpr ,J.. t (1-pf ,J.. ) Ff ,J.. vpr ,J.-
. 1 - Ff ,J.. vpf ,J.. =
(4.23)

v is the average linear air velocity that flows past the partiele
1
conglomerate. It has been explained in section 4.2 that v1 is a
function of the Rech number, of the classifier geometry and of vf.
n and N are functions of H', H' being the partiele hold up as
con con
defined in equation 4.13.a.
The terros Kcoll and Kwall need some explanation. A large number of
quantities play a role in the exchange of impulse between falling and
rising particles. The values of the mass flows and of the partiele
veloeities will be of importsnee but also the classifier geometry, the
geometry of the particles, the shape of the air flow and the local air
veloei ties.
Kwall is a complex function of partiele flow, partiele velocity,
channel and partiele geometry, air flow profile and the local linear air
veloeities within this profile.

- Rising particles.

Similar assumptions as for falling particles are made. Senden /4.1/

pointed out that the rising particles loose a great deal of their impulse
when they collide with the upper zigzag wall. The change of impulse is
caused forthermore by the gravity force, the drag force of the air and by
the exchange of impulse with particles that enter the stage in a falling
stream.
The drag force again acts on a conglomerate of particles as do the
gravity force and the buoyancy. K 11 and K are functions of the
co wa 11 ,
same variables as described for falling particles. For elastic collisions
Kcoll for falling particles has to be equal to -Kcoll for rising
particles. The impulse balance in this situation becomes:

(1 - pr,J.-
. 1) Fr,J.- . 1 + pr,J.-
. 1 vpf ,1- . 1 Fr,J.-
. 1 vpr,J.. - Fr,J.-
. 1 vpr,i-1 =
(4.24)

-109-
I<- foliows frcm eqtBtions 4.23 and 4.'1A and by tbe remrks I!Bde above
concerning v1 , Neon' neon' Kcoll and Kwall that pf and pr
can be written by:

Pr,;• (Rech; Cd; vpf ,1. ; Vpr,1• ; Vpf ,1-

. 1; Vpr,1-
• 1;

vf: Ff ,1. ; Fr,1. ; Ff ,1- . 1; 1; B; W; 13; f;

. 1; Fr,1-
pp;p f; lp; bp; dp; g) (4.25.a)

and

pr,1-
·1=(Rech;Cd;vf.;v p ,1 pr,~.;vf.l;v
p ,1- ·1;
pr,1-
vf; Ff ,1.; Fr,1.; Ff.
.~-1
: 1; B; W;S; f;
p • p • 1 • b • d • g) (4.25)
p' f' p' p' p'

The number of variables of equations 4.25 can be considerably reduced as

follows:
Rech determines the shape of the air flow profile. In section 4.2.3 it
was noticed that under our experimental conditions the influence of
Rech can be neglected.
Cd will be constant (see section 4.2.3),
The partiele veloeities are functions of the form (eq, 4.14):

vp,x = f(lp ; bp; dp; g; p p; p f; vf; 1; B; W; f; 13 ; Fx) (4.14)

in which the index x relates v to the partiele flow that is

p,x
considered.
equations 4.25 thus become:

pf ,1. = f(vf' Ff ,1. ; Fr,1. ; Ff ,1-

. 1: Fr,1-
. 1:
1; B; W; 13; f; pp; Pé lp; bp; dp; g) (4.26)
Pr,1·-1 f(vf·,Ff.;F
,1 r,1.;Ff.1;F
,1- .1;
r,1-
1; B; W; 13; f; p ; pf; 1 ; b ; d ; g) (4.27)
p p p p

The following equations fo11ow from a mass balance over a single stage:

Fr,1.=pf.Ff.+p .lFr,~-
"1 (4.28.a)
11 ,1 r,1-
or

-110-
Fr,1· = f(pf ,1·• Prt-1' Ff , i' Fr,1·-1) (4.28.b)

· 1 = (l-pf ,1.) Ff ,1· + (1-Pr,1-

Ff ,1- . 1) Fr,1-
. 1 (4.29.a)
or
Ff ,1-
. 1 = f(pf' pr,1-
. 1' Ff ,1·• Fr,1-
. 1) (4.29.b)

Substituting 4.28, 4.29 and 4.27 in 4.26 resp. 4.28, 4.29, and 4.26 in
4.27 finally leads to the following relationships:

Pf, 1. . 1; L; B; W; 13; f; pp ; pf; 1p ; bp ; dp ;g)

(vf ,1. ; Ff , i; Fr,1- (4.3))

and

. 1; L; B; W; 13; f; pp ; pf; 1p; bp ; dp;g)

. l = ( vf; Ff ,1. ; Fr,1-
pr,1- (4.31)

Both pf and pr are functions of 14 variables defined in terms of

three dimensions. Thus, 14+1-3=12 independent dimensionless numbers
describe the relation between Pf or pr and these variables. The
following groups can he derived:

(4.32)

(4.33)

In these equations vfl is defined by:

(4.34)

In the resting part of this section Ff is a short notation of Ff,i and

Fr for Fr,1-. •
1
The fo1lowing remarks can be made concerning the dimensionless numbers
given above:
B, (B/L), (W/L) are the ratios of the dimensions of a classifier stage.
(1 p /L} represents the ratio of the largest partiele dimension and the

-111-
classifier channel width. The dimensional ratios of the partiele are
represented by dp/1 p and b p/1 p • These groups are constant for a
given classifier geometry and for a given shape of particles.
(pp/pf) is the ratio of the densities of the fluid and the particles.
This number will be of importance if the buoyancy of the particles cannot
be neglected.
(vf/vf 1 ), (Ff/Fr)' and (Ff L/(Vp Pp vf 1 )) only determine
the transition probabilities of given particles in a given classifier.

b) Correlations for pf and pr as functions of vf and of the

local partiele flows.

For a given partiele geometry and for given particles the transition
probabilities are functions of the following dimensionless numbers:

(4.35)

and

(4.36)

A correlation of the form:

(4.37)

and a similar correlation for pr was used to describe the relationship

between the partiele transition probabilities and the various
dimensionless numbers. At ~ low feed rates (Qv=0) pf and pr are
functions of only the first term of eq.4.37:

(4.38)

Both pf and pr are s-shaped functions of vf/vfl with values that

must lie between 0 (low values of vf) and 1 (high values of vf).
Equation 4.38 which was proposed by Trawinski /4.12/ to describe s-shaped
functions meets these requirements. The steepness of the curve depends

-ll2-
upon the parameter n. The parameter c determines the half value point.
1
Power produetsof Ff/Fr and L Ff/(Vp pp vfl) were taken to
deseribe the influence of the flow rates at higher values of Qv.
Table 4.5 eontains the different parameters of eorrelation 4.37 that were
determined from the experimental data by means of a least squares curve
fitting method. Different partiele flow regimes were distinguished for
very low flow rates (Qv=O) where partiele-partiele interaetions do not
influenee the values of pf and pr and for higher partiele flow rates
(Q >0). Different partiele flow regimes also existed at higherfeed
V
rates for falling partiele motions in the standard 90 deg. classifier
(see section 4.3.3). At values of Ff below 10 g/s,pf is not mueh
influeneed by the flow rates whereas, at values of F f above appr. 10
g/s, pf elearly inereases at inereasing Ff.

Table 4.5.
Results of best fit calculations for partiele transition probabilities.

The constants of the equation:

with vfl = /(g dp P/Pf) • 1.40 m/s and Vp pp = 0.072 g were determined by means
of a least squares curve fitting metbod using the data of the partiele transition
probabilities measurements. At Qv•O the constant c is 1 and the constants c 3 and c 4 are 0.
1

cl c2 n c3 c4 Ro s' R s'
0
st. 90 deg. classifier
pf stage I: !.056 0.763 7.80 -0.017 - 0.920 0.037 0.942 0.027
stage 2-8 (!): p.908 0.687 3.70 0.045 0.080 0. 716 0.054 0.802 0.055
stage
p stage
2-8 (2):
1-2:
~:670
• 786
0.687
0.496
3.70 0.135
1.97 0.002
0.075
0.268
0.580
0.811
0.036
0.024
0.814
0.878
0.049
0.027
r stage 3-8: 1.123 0.375 1.66 -0.028 0.158 0.555 0.031 0.696 0.031

st .120 deg. classifier

pf stage 1:
stage 2-8:
!.lil
1.023
0.590
0.564
7.12
7.49
-O.ll9
-0.129 0.178
- 0.963
0.928
0.050
0.040
0.952 0.042
0.933 0.043
p stage 1-2: l.ll2 0.629 5.40 -0.146 0.176 0.801 0.10 0.921 0.036
r stage 3-8: 0.832 0.571 4.81 -0.024 -0.053 0.926 0.046 0.800 0.053

br. 120 deg. classifie

pf stage 1: 1.934 0.660 10.58 -0.010 - 0.973 0.073 0.985 0.048
stage 2-8: 1.298 0.572 4.74 -0.143 0.115 0.691 0.11 0.954 0.054
p stage 1-2: 1.076 0.656 4.80 -0.183 0.514 0.936 0.065 0.967 0.041
r stage 3-8: 1.057 . 0.688 2.60 -0.028 0.004 0.813 0.056 0.788 0.047

R : multiple correlation coefficient at Qv•O. R: multiple correlation coefficient at Qv > 0

0
s' :standard error of estimate at Qv.O. s': standard error of estimate at Q/0.
0
(1) : experimenta1 data at Ff < 10 'g/s. (2) : ex perimental data at Ff > 10 g/s.

-113-
4.4 Local partiele transition times.

4.4.1 Introduction.
Local partiele transition times were determined by registering the
durations of the different single step transitions during the same
experiments in which local flow rates and partiele transition
probabilities were determined, The following transition times are
distinguished:

Table 4.6.

~easured trans1tion times ol. the four one-step transitions.

a) The standard 90 deg.classifieL

H (J t 0
Qv vf 'rf 0 pf tl-pr 1-pr pr pr tl-pf (j l-pf
(g/s) (g) (s) (s) (s). (s) (s) (s) (s) (s)· (s)
0,310 0.122
~::~~ ~::~ ~:~~~ ~::~~ ~:g~ ~:~~i
u
~ J ~:~~ 0.334 0,138
0 0 2.22 0.173 0.133 0.274 0.123 0.138 0,117 0,324 0,141
0 0 2.16 0.163 0.112 0.273 0.103 0.123 0.109 0.323 0.138
0 0 2.26 0.161 0.118 0.275 0.117 0.153 0.160 0.332 0.149
0 0 2.32 0.150 0.108 0.273 0.120 0.120 0.113 0.330 0.150
0 0 2.37 0.149 0.112 0.262 0.119 0.123 0.118 0.305 0.127
0 0 2,43 0.160 0.128 0.255 0.110 0.148 0.143 0.318 0.158
{) 0 2.43 0.156 0.131 0.271 0.138 0.130 0,1!4 o.336 0.155
0 0 2.45 0.152 0.110 0.266 0.132 0.126 0,123 0.309 0.142
0 0 2.52 0.147 0.111 0.251 O.lO'l 0.131 0.120 0.333 0.168

1.68 15.7 2.124 0.202 0,154 0.298 0.116 0.152 0.125 0,381 0,141
1.31 18.2 2.185 0.182 0.125 0.306 0.129 0.139 0.119 0,376 0.140
1.25 38.4 2.293 0.197 0.140 0.314 0.146 0.132 0.101 0,416 0.164
1.40 31.5 2.331 0.188 0.128 0.313 0.135 0.152 0.152 0.416 0.163
1.58 39.0 2.414 0.182 0.124 0.336 0.162 0.143 0.12& 0.427 0.159
1.24 26.8 2.501 0.174 0.136 0.327 0.164 0.153 0.141 0.405 0.169
1.41 31.8 2.439 0.176 0.136 0.316 0.156 0.149 0,141 0.415 0.164

2.87 30.1 2.13& 0.181 0.121 0.311 0.129 0.136 0,102 0.395 0.140

I 2.1;7
2. 70
2.64
48.4
62.2
60.4
2.243
2.317
2.387
0.198
0.212
0.187
0.150
0.159
0,129
0.329
0.340
0.352
0,149
0.135
0.170
0.156
0.166
0.186
0.143
0.162
0.182
0.445 0.150
0.483 0.164
0.460 0.183

I 4.94
4.61
42.8
81.8
2.07
2.18
0.196
0.217
0.121
0.158
0.333 0.147
0.363 0.166
0.170
0.185
0.122
0.179
0.435
0.492
0.140
0.161
I 4,70 94.4 2. 30 0.222 0.166 0.371 0.165 0.160 o. 132 0.495 0.174
I 4.60 85.0 2.41 0.203 0.155 0.373 0.171 0.185 0.!67 0.462 0.165

b) tlic standard 120 <leg. classifier.

0
prt1-pf
s) s
,2 0 o. 57
0 0 2.11 0.364 0.190 0.392 0.164 0.294 0.165 0.443 0.140
0 0 2.14 0.358 0.173 0.410 0.151 0.270 0.161 0.465 0.152
0 0 2.29 0,353 0.184 0.370 0.162 0.229 0.138 0.467 0.155
0 0 2.37 0,337 0.184 0.410 0.146 0.233 0.129 0.479 0.174
0 0 2.43 0.342 0.174 0.446 0.163 0.229 0.157 0.457 0.162
0 G 2,55 0.344 0.186 0.450 0.164 0.233 0.164 0.491 0.173
0 0 2.64 0.321 0.188 0.442 0.159 0.243 0.165 0.499 0.178

2.34 13.0 2,31 0.336 0.189 0.398 0.133 0.240 0.153 0.460 0.147
2.24 15,8 2.43 0.327 0.171 0.425 0.154 0.236 0.154 0.472 0.145
2.22 26.0 2.59 0.311 0.171 0.421 0.149 0.236 0.155 0.483 0.145
210 19.6 2. 71 0.314 0,182 0.425 0.165 0.213 0.136 0.473 0.148:

-llL~-
 b) the standard 120 deg. classifier (continued).

tpf t
pr
(s)

2.42 0.132 0,249 0.173 0.466 0.133

2.58 0.176 0.240 0.194 0.476 0.136
2.69 0.141 0.250 0.175 0.494 0.146

9.09 49.0 2.58 0.307 0.159 0.427 0.157 0.233 0.145 0.492 0.132
9.08 67.5 2.69 0.300 0.188 0.440 0.148 o. 275 0.211 0.491 0.155

c) The broad 120 deg. classifier.

Qv H tpf
Is
0
2.37 1. 79 0.293 0.189 0,418 0,138 0.229 0.164 0.374 0.115
4.11 1. 79 0.275 0.101 0.391 0.112 0.228 0.111 0.373 0.118
9.48 I. 79 0.291 0,103 0.406 0.139 0.250 0.103 0.376 0.125

0 0 2,11 0.257 0.176 0.391 0.140 0.176 0.085 0.400 0.143

1.94 10.3 2.11 0.244 0.136 0.398 0.136 0.195 0,120 0.413 0.153
4.26 21.9 2.11 0.268 0.158 0.383 0.138 0.186 0.107 0.402 0.125
10.58 42,5 2.11 0.273 0.149 0.444 0.151 0.230 0.152 0.406 0.128

0 0 2,27 0.229 0.160 0.369 0.138 0.182 0.107 0.387 0.147

2.21 12.4 2.27 0.278 0.174 0.406 0.125 0,196 0.124 0.403 0.138
4.51 27.0 2.27 0.222 0.145 0.437 0,179 0.196 0.111 0.413 0.145
7,97 45.6 2.27 0,257 0.169 0.426 0.167 0.222 0,140 0.421 0.138

,0 0 2.39 0.247 0.165 0.421 0.150 0.172 0.140 0.389 0.159

: 1.86 1.86 2.39 0.260 0.163 0.413 0,160 0.192 0.135 0.399 0.124
4.63 4.63 2.39 0.261 0.172 0,401 0.136 0.172 0.099 0.423 0.136
8.11 8.11 2.39 0.270 0.181 0.420 0.139 0.200 0,123 0.417 0.122

0 2.55 0.212 0.145 0.400 0.140 0.166 0.118 0.374 0.159

0 0 2.65 0.244 0.161 0.374 0.145 0.160 0.101 0.371 0.152:

2.36 6.6 2.65 0,234 0.153 0.387 0.149 0.155 0.114 0.363 0.118'
4.56 16.0 2.65 0.246 0.189 0.400 0.111 0.168 0.115 0.398 0.130
9.77 36.3 2.65 0.241 0.174 0,430 0,151 0.187 0.147 0.432 0.151

H: ~ holdup (g) ll1side thi! cl.assifier chamel.

portiele feed rate (g/s)
<lv'
tpf: """"duratioo of a~ transition fr<Jll a fal.ling streon t.o a tising streon (s).
t ;>f' meen duration of a partiele transition fran a fal1ing st"""' t.o a falling streem (s).
1
t : meen duration of a~ transition fr<Jll a tising stnem t.o a tising stre<ll\ (s).
pr
t -pr: """"' dilration of a partiele transition fran a tising streem t.o a falling st....." (s).
1
o pf et&.: The staM8rd deviatioos of thi! var:loos transition times (s).
The meen partiele transition times """" ealo.tlatsd by wei~ thi! transit times by thi!
relative"""" flOii at eoch ~.

tpf: the duration of time for a transition from the falling to the
rising stream at stage boundary i (a <pf> motion),
tl-pf: the duration of a <1-pf> motion.
t pr the duration of a <pr> motion,
t : the duration of a <1-p> motion
1-pr r
The values of the transition times that are discussed are averages
measured at all stages during an experiment. In calculating these
averages the transition time at each stage is weighed according to the
frequency of its occurrence.

-115-
4.4.2 Results.
a) the standard 90 deg. classifier.

The values of the four different transition times have been tabulated
in table 4.6.a. 4.31 show the partiele transition times as
function of vf. Parameter is Qv.
The times for the different single stage transitions differ as can be
seen from the data of both the table and the The transition
times t are lowest whereas the times t f are highest.
pr 1-p
Furthermore < pf > motions take less time than <1-pr> motions. These
transition times can be explained as follows:
Particles that remain in a rising stream cover this trajectory in
general without being hindered or even colliding with the wall. They
experience at all times the upward force that is being exerted by
the main upward air flow. Particles that continue moving downwards while
passing through a stage have much more contact with the walls than
particles that move upwards through this stage. They cross the main
upward air flow from the point where the angle between two channel
sections protrudes into the channel. During this crossing the downward
motion of the particles stagnates as they loose a large amount of their
downward directed velocity. The particles may even temporarily be lifted
again and collide with the opposite wall befare they start
falling again.
The difference between the transition times tpf and tl-pr can be
explained as follows: Rising particles collide with the zigzag wall and
loose almost all their kinetic energy. Their velocity at the wall will
thus be practically nil. From the point of collision they fall down
again. particles that rise follow a more or less smooth
path, They may experience the drag of the circulation air flow at the
lower wall when they move down and are dragged upward by the
rising air flow as soon as they have passed the edge of this wall.

-Influence of air velocity and local partiele flow rates.

The influence of both vf and the local partiele flow rates is best
illustrated by calculating best fit curves of the form:

T c2 Hc3 (4.39)
vf

-116-
 tp{
(s)
• I I tl-pr I I
(s)
0.4 1- Standard 90 deg, classifier. . 0.4 .
.o
,.
0

...
0

0.3 1- . 0.3
0
.. • ..
• • • • • •
•• • • • .
• •.
0.2 1- o. ... .... 0

• •• • •• •• •t.• .. •
0.
0 . 0.2 1-
•
0.1 0.1
a) b)
I I I I.

2.1 2.2 2.3 2.4 2.5 vf (a/s) 2.1 2.2 2.3 2.4 2.5 "f (m/s)

I
t
.............., pr tl-pf ·u I

• •"•• •'o••
I I I
I
(s)
I
' (s)
0
0.4 Parameter: Qv (g/s)
0.4
• •• ••
0,3 1-
•... 0
1.4
• • • •• • • •. ••
• 4.7
0
2.7 0.3

0,2 1- .• 0.2 f-
o•• •o•t• •
0
0

0.1 1- •
• A. e A ••• •• •
A

0.1 1-
c)
I I I I
d? I I I I

2.1 2.2 2.3 2.4 2.5 "t (m/s) 2.1 2.2 2.3 2.4 2.5 vf (m/s)

l!ill!!!:.e 4.31.
M=an partiele tronsition tinl:!s of the foor different single stage transiticns IIB8SI.1l'a'l in the standard 00 deg. classifier.
from the data of table 4.6.a. H is the partiele mass hold-up which is
proportional to the mean local partiele flow rates. Table 4.7.a contains
the results. tpf decreases with increasing vf. The influence of
upon the three other transition times t f' t and t is not
1-p pr 1-pr
significant. A reason for the insensitivity of these transition times for
vf may be that the particles cover longer paths at higher vf.
The transition times in all cases increase with increasing partiele
hold-ups inside the channel. This increase is eaused by the larger
partiele-partiele interaetions that occur in that case. As a result of
these interaetions both the veloeities of falling particles and rising
particles become lower (see section 4.2).

b) the standard 120 deg. classifier.

The results of these measurements have been tabulated in table 4.6.b.

Local partiele transition times have been plotted in figs. 4.32 vs vf.
Parameter is again the feed rate Qv. As is the case in the standard 90

Table 4.7.
The înfluence of the partiele hold-up Hand the superficial air velocity vf upon the vsrious one-step
transitions.
correlation:

<2
t • c 1 vf (s}

a) The stMlard \Xl dog. classifier. b) The standard la) dog. classifier.
R R
Cc2 "3 Cc3 < ot cl ":! 0<2 "3 ""3 at
tpf 0.276 -0.53 0.18 0.025 (<).(lJ3 O.fm 0.011 tpf 0.456 -0,37 o.oo -O.OC65 0.002 O.I'B4 O.roJ

tl;>r o.:m-0.18 0.18 0.025 0.003 o.aea 0.017 t

1-pr
<
o.:m O,l'! 0,10 O.OC62 0.002 0.714 0;013
t 0.!31 0.12 0.34 0.022 o.<m 0.672 0.0!6 t
pr
0.456 -O.XI 0.27 -o.oou o.oo; 0.634 0,012

t~-pf 0.348 0.11 0.19 0.036 0.003 0.927 0.024 tl-pf 0.356 0.32 o.a; -0,0017 0.001 0.8JT 0.0!0

c) The breed 120 dog. classiher.

oc oc R
cl "2 "2 3 3 ot
tpf 0.361 -0.44 0.11 Q.0073 O.CXY. 0.721 0.016
tl-pr 0.400 O,Oll O.Cii O,OCiiO 0.002 0.531 0,018
t O.lS2 -0.81 0.12 0.01al O.CXY. 0.883 0.013
pr
tl-ilf o.:m 0,11 o.ca 0.0054 0.002 0.499 0,018

H: porticle holdop {g) inside the classifier channol.

tpf: 111m duratlm of a l"!'ticle transitlm fran a falline at.- toa ~ atnm (s).
t 1-pf: ....., duratlm of a l"''ti<le transitlm fn:m a falline stnan to a falline - (s).
tp3: ...".. duratlm of a porticle transittoo fran a ~ st:rEall toa ~ streEm {s).
't-pr' """"dumtlm of a l"''ticle transition fran a ris!t1g st,_ toa falline otn!Om {s).
oc ard ~: The stMlard errors of the ragresstoo coeffic:iern:s
2 ":! ard "3 {s).
Ot the stMlard error of the est:iJIBte of the transition time lilder conslderatlm {a).

-118-
 tpf tl-pr I I I

(s) (s) ! • ih • gA
0.4 0.4
~-·
!A • D
•• ".e. • •
0.3
DfA QJilo • a A-
0.3 .
0.2 1- a) 0.2 b)

0.1 Stsndard 120 deg. c1assifier. . 0.1

I I
o.o o.o
2.1 2.2 2.3 2.4 2.5 2.6 vf (•/s) 2.1 2.2 2.3 2.4 2.5 2.6 vf (e/s)

I
t • I I I
- r- I I I
•
......
,_ pr
(s}
tl-pf
•• !I.e. • c(t
-•& 8 i!. -
1.0
I 0.4 1- -
(S)
0.41- -Parameter: Qv (g/s)
.
0.3
I- • - 0.31- • 02.2
A

0.2
.... • 0. •&h •
0
0
A 0.21-
0
0
4.5
9.1
-
0.1 0.1 -
0.0
2.1
c)

2.2
I

2.3 2.4
I .
2.5 2.6 "t(•/s)
0.0
2.1
d)
I

2.2 2.3 2.4 2.5

I
2.6 vf (•/s)

Fif1!!.re 4.32.
lt:!an jEtiele t:mnsition t:in1ss of the four ditterrot single stage t:mnsitions oeasured in the standsrd 120 deg. classi.tier.
deg. classifier, partiele transition times are largest for >
transitions and shortest for <pr > transitions. <pf> transitions again
take less time than <l-p > transitions. The transitions in the 120 deg.
r
classifier last longer than those in the 90 deg. classifier. The reason
for this is that the 120 deg. classifier lacks a circulation flow along
the lower zigzag walls and has lower local veloeities within the main
upward air flow.

Effect of air velocity and partiele feed rate.

It appears from figure 4.32 that the transition times t f and t

P pr
decrease at increasing vf. tl-pf and t 1_pr increase when vf is
increased. The partiele feed rate has hardly any influence upon the
transition times. These conlusions are confirmed by the results of the
best fit calculations that are tabulated in table 4.7.b

c) the broad 120 deg. classifier.

Fig. 4.33 shows the partiele transition times for the different one-step
transitions inside the broad 120 deg. classifier. These data have been
tabulated in table 4.6.c. The transition times are in general longer than
the transition times inside the 90 deg. classifier but are shorter than
the transition times in the standard 120 deg. classifier. That the
transition times are shorter for the broad 120 deg. classifier than for
the standard 120 deg. classifier is the result of the circulation air
flow inside the broad channel which accelerates the falling particles.
Furthermore it can be expected that the linear air velocity in the main
stream also increases with increasing channel width. Izumi et. al. /4.2/
- /4.5/calculated the stream lines of laminar fluid flows (Rech=300)
inside corrugated wall channels. They found that the separation zones of
the air flow (see f.i. fig. 4.4) become larger as the channel is
broadened. The ratio of the the main flow cross section to the cross
section area of the channel becomes smaller then. At a constant
superficial velocity vf the linear velocity of the main flow thus
increases when the channel is broadened. The separation zones become
larger when Rech is increased. Thus, when the channel is broadened for
turbulent flows, the linear air velocity also increases at constant vf
if it is assumed that the exchange of mass between the main flow and the
separation zones is small compared to the mass flow through the channel
cross area.

-120-
 I I
tpf tl-pr

'•
(s) Broad 120 deg. classifier. {s) r,j
0.4 0.4 i~ • • • I

0.3 . . 0.3
i 0 • -e
0.2
• I
• I 0.2 .
0.2 0.1
a) b)
o.o o.o
2.1 2.2 2.3 2.4 2.5 vf {m/s) 2.1 2.2 2.3 2.4 2.5 vf (m/s)

I
.....
N
t
pr tl-pf
..... (s) (S)
I
0.4 0.4
j
• I • 1 • !

Parameter: Qv (g/s)
0.3 0.3
eo
0.2 ; •
0

" •
.J • I
0.2
• 2.1
• 4.4
0 8.1

0.1 0.1
c) d)
I I
o.o o.o
2.1 2.2 2.3 2.4 2.5 vf {m/s) 2.1 2.2 2.3 2.4 2.5 vf (m/s)

Fisure 4.33.
M=lan partiele transition times of the foor different single stage transitions nmsured in the breed 120 deg. clli'3Si.fier.
 This influence of higher air veloeities is apparently more important than
the fact that the channel is wider and the particles have to cover
greater distances.

-Effect of air velocity and partiele feed rate,

Fig. 4.33 suggests that the partiele transition times very slowly
decrease with increasing vf and increase with increasing Qv. The
results of the data fits given by equation 4.39 (see table 4.7.c.)
globally confirm these conclusions. The correlation between t and
1-pr
tl-pf on the one hand and vf on the other hand is however not
significant.

Referencès.
4.1 Senden, M.M.G., "Stochastic roodels for individual partiele behavior
in straight and zigzag air classifiers.", dissertation (1979),
Eindhoven Universiy of Technology.
4.2 Izumi,R., Oyokawa,K., Kaga,S., Yamashita,H., Bulletin of the JSME,
24 (1981), no,198, pl425-1433.
4.3 ibid., 24 (1981), no.198, p2099-2107.
4.4 ibid., (1983), no.216, p1125-1433.
4.5 ibid., 26 (1983), no.217, pl147-1153.
4.6 White,F., "Fluid mechanics." (1979) McGraw-Hill, New York,
4.7 Gupta,V., Gupta,S., "Fluid mechanics and its applications." (1984),
Wiley Eastern Ltd., Bombay.
4.8 Perry,R., Chilton,C., "Chemica! engineers' handbook", 4th ed.
(1973), McGraw-Hill, New York.
4.9 Kaiser, F., Chem.Ing,Techn., 35 (1963), no.4, p273-282.
4.10 Buckingham,E., Phys.Rev. A (1914), no, 4, 345-376.
4.11 Isaacson,E., Isaacson,M., "Dimensional methods in engineering and
physics." (1975), Edward Arnold Publ., New York.
4,12 Trawinski,H., Aufbereitungstechnik, ~ (1976), p248-255,

-122-
5 MODELLING OF MULTISTAGE CLASSIFICATION PERFORMANCE AT HIGH FEIID RATES.

5.1 Introduction.

The results of the measurements of the internal variables inside the

three different classifiers have been discussed in chapter 4. The
partiele behaviour at the individual classifier stages was determined in
order to obtain a more fundamental insight into the functioning of zigzag
air classifiers at high partiele feed rates. A mathematica! model that
prediets the overall classifier performance at low partiele feed rates as
a function of the single stage partiele behaviour was developed by Senden
/5.1/ (see chapter 1). This model uses the concept of single stage
transition probabilities. These transition probabilities are not the same
at all stages, not even at low partiele concentrations. The partiele
transition probabilities furthermore depend upon the process conditions
such as the partiele feed rate and upon the superficial air velocity
vf' It was shown in chapter 4 that satisfactory correlations between
the partiele transition probabilities pf and pr and the local flow
rates Ff and Fr can be found, In this chapter, a mathematica! model
will be presented which is an extension of the one-step memory model of
Senden. The capability of the model to describe the multistage classifier
performance at higher feed rates is discussed. A time base is added to
the model in order to make it possible to predict time related process
variables such as partiele residence times and partiele hold-ups. The
influence of the number of stages and the location of the feed stage upon
the classifier performance will be demonstrated by means of model
calculations.

5.2 Description of the classification of feeds consisting of identical

particles.

5.2.1 Nomenclature.

The classifier consists of R+1 stages that are numbered from 0 to R. R

was 10 in the experiments described here. The particles are fed into the
classifier above stage V. Fig. 5.1 schematically shows the numbering of
the stages, the stage boundaries and the sections. The partiele flows at
the various stage boundaries i are represented by Ff . for falling
,1
streams and by F . for rising streams.
r,1

-123-
5.2.2 Caleulation of the local partiele flow rates and of the
separation tunetion t •
1
a) Strueture of the algorithm
An algorithrn is presented below to calculate the fraction of bottorn
product, ~R' as well as the partiele flow rates Ff . and F . at
,1 r ,1
the different channel stages under given stationary process conditions.
The partiele feed rate Qv and the superficial air velocity vf are
assurned to be known. Furtherrnore it was shown in chapter 4 that pf and
pr can be represented as functions of the forrn:

Pf ,1. = f1(vf, Ff ,1. ' Fr,1'-1) (S.l.a)

Pr,1. = f2(vf, Fr, ' Ff ,1+

. 1) (S.l.b)

The local partiele flow rates and ~R are the unknown variables. The
algorithrn determines these variables as follows:
The separation function ~R is defined as the fraction of the feed that
ends up in the bottorn product:

......,.,riq
of staps Figure 5.1

iË
a.-riq of
stage boundariea Nane:!clature.
r,9
9 9

8
:,~=r=

~*~:
7

~~:
5 (V)

~i-·
3

:!,J ,--t-'·:
Ff,O
i" i-1, 1+1: stage boundary n'UJD.bers
0 0

' -124-
 (5.2)

A first estimate of ~R: ~R,l is made. The partiele flow Fr,R that
leaves the top exit is then calculated from 5.2. Furthermore
Fr, R = Fr,r- 1 (see fig.5.1). The algorithm then calculates the flows
Ff . and F . at the different stage boundaries as well as the the
,1. r,l.
bottorn product flow Ff,O from the relationships 5.1 and from the mass
balances at each classifier stage, This calculation metbod will be
discussed below. The algorithm thereby provides a value for the bottorn
exit partiele flow, Ff,O' Both Ff,O and Fr,R- 1 have to satisfy the
overall mass balance:

(5.3.a)

This is the case only when the estirnate of ~R was correct. Thus the
root of the relation:

(5,3.b)

has to be found, This is done by rneans of the metbod of "interval

bisection" /5,2/, Fig. 5.2 shows a flow diagram of the algorithm.

b) Calculation of the local partiele flows witbiD the channel.

On the basis of equations 5.1a and 3.1b, local partiele flow rates Ff .
,1.
and Fr,l.. can be calculated for given Qv and ~R. It follows from
equation (5.2) that

(1 <l)O=F
""R v r,R
=F
r,R-1 (5.4)

A mass balance over the top section above stage R-2 gives the partiele
flows at boundary R-2 (fig. 5.3a):

F
r,R-1 = pr,R-2 Fr,R-2 (5.5)

Ff,R- 1 is 0. Thus Fr,R-2 can be calculated for known values of

Fr, R- 2 = Fr, R- 1 I Pr, R- 2 = Fr, R- 1 I f2(vf' Ff ,R-1)" (5.6)

-125-
 Begin Figure 5.2
Flow diagram of the algorithm for the nrxlel that determi.nes
the fractioo of bottau product, <I!R, the local partiele flow
raus and the local partiele tranSition pmbabilities of
identical particles inside zigzag air classifiers.

Calculate Fr,R-l
top product flow c)

;~~t-·-·-·-j~r~!
i ... ----- -- !

N
0\
I
,...., Calculate
Pf,i' Pr,i' Ff,i' Fr,i
, ------.~P'
i

1
~r
Pr.v
1 9
!
!
I
~}:::::::.
I I 1-...
., IÇ _ r,V-1
1...,._____ t,Y..l I
at stage i I

f,._l! ;_j. . . !
I

d)

Fi81J!e 5.3.
Mass ba1ances of classifier sections:
a: top exit; b: top section;
c: feed stage; d: bottau section.
Ff,R- follows from the overall mass balanee:
2
F (5.7)
f,R-2 = Fr,R-2 F
r,R-1
A mass balanee over the top seetion boundaries i and i+l above the feed
stage V gives (fig. 5.3b):

Fr ,1+
·1=Pr ,1.Fr ,1.+pf.lFf.l
,1+ ,1+
(5.8)

Thus, F . ean be expressed by:

r,1

F .
r,1 (S.9.a)

or

Fr,i+1- fl(vf' Ff,i+l' Fr,i) Ff,i+l

Fr,1. (S.9.b)
f2(vf, Ff,i+l' Fr,i)

Fr,1. is the only unknown variabie of eq. 5.9.b. Fr,1. is solved from
this equation by means of the method of "Succesive substitution" /5.2/,
/5.3/.
Ff . follows from the overall mass balanee of the section above stage
,1
boundary i:

Ff ,1. F . - F
r,1 r,R-1 (5.10)

After Fr,R-2 and Ff,R- 2 will have been obtained from Fr,R- 1 by
means of eq. 5.6 and 5.7, the remaining flowsin the topseetion are
calculated by repeated applieation of equations 5.9 and 5.10 starting at
i=R-2 downwards.
Fig. 5.3.e shows the partiele flows that are of interest near the feed
stage. A fraetion 1-pv of the feed Qv falls to the next lower stage
·boundary whereas a fraetion pv rises. These partiele transition
probabilities were determined in the same experiments to determine the
various pf and pr. The results are tabulated in table 5.2. A mass
balanee over this part of the classifier gives the following equations:

Fr,V =Fr,V-1 pr,V-1 +Ff,V pf,V +0'v pv (5.11)

-127-
Thus

F -F p -p 0
F = r,V f,V f,V v 'v (5.12)
r,V-1
Pr,V-1

and

F
f,V-1 ( 1-pf,V) Ff,V+Fr,V-1(!-pr,V-1)
+ (1-pv) ~· (5.13)

The flows within the bottorn section are ca1cu1ated in a similar way as
was done for the top sectien fig 5.3d):

Fr,~+
·1=P (5.14)
· r,~.Fr,~.+pf.1Ff.1
.~+ .~+

F . = _F...::r:..!,~i+:.:l:__P...::f'-',.;;:i+.:..:1=-Ff~·~i:.:.+::.l_ (5.15)
r,~

and

=F .-F +F (5.16)
r,1 r,V-1 f,V-1

The partiele flow that leaves the classifier is:

(5.17)

As was mentioned before,this value of ,O will generally not satisfy

the overall mass balance (eq. 5.3a). The above procedure must therefore
be repeated until the root of eq. 5.3b is obtained with an acceptable
accuracy.

5.2.3 Mean partiele residence time.

Mean partiele residence times may be calculated if the mean duration of
each transition is known as well as the number of the various transitions
in which each each partiele is involved.
The mean residence time can be calculated from the partiele hold-up
inside the classifier channel:

(5.1&)

-128-
The overall partiele hold-up H follows from the local partiele hold-ups
H. at the individual stage boundaries i:
l

H = LH.l (5.18.b)

H. consists of 4 different contributions, i.e. the hold-up of particles

l
in the falling stream that subsequently rise, H. f , the hold-up in the
1, r
falling stream that continue to fall, Hi,ff' the hold-up of particles
in the rising stream that continue to rise, H. and finally the
1,rr
hold-up of particles in the rising stream that subsequently fall,
H.1,r f•

H.1, f r = Ff ,1. t pf ,1. pf ,1. (5.19a)

H.l, ff = Ff ,l. t1-pf ,l. ( 1-Pf ,l.) (5.19b)
H. F . t (5.1<k)
1,rr =
r,1 pr,l. pr,1.
H.1,rf = Fr,1.t 1-pr,l. (1-pr,1. ) (5.19d)

t f . etc are the transition times of the four different transitions as

p ,l
defined in section 4.4.
The summation of these holdups leads to:

H.1 = H.1, f r + H.1, ff + H.1,rr + H.1,rf (5.20)

and the partiele residence time can be calculated by substituting

equation 5.20 into equations 5.18:

8
T = L(Ff ,l./O
'v • (tpf ,l. pf ,l. + t1-pf ,l. (1-pf ,l.))) +
1
8
L(Fr,i/~· (tpr,i Pr,i + t1-pr,i ( 1-Pr,i))) (5.21)
1

5.2.4 Verification of the model.

The multistage model was verified by calculating ~R and T from the

single stage behaviour as characterized by pf ,1.(vf,Ff ,1.,F r,l-
. )
1
and p .(vf,Ff . ,f .) and comparing the results with the
r,1 ,l+ 1 r,1
values of ~R and T that were measured experimentally under identical
conditions. The experimental method for determining ~R and the method
to determine T from mass hold up measurements has been discussed in

chapter 2.

-129-
The correlations that were used for the calculation of the transition
pro babilities are shown in table 4. 5. Table 5 .La contains the mean
values of the transition times that were measured for different partiele
feed rates Qv.
Fig. 5.4.a shows the measured and the calculated ~ curves for the
standard 90 deg. classifier. The calculated ~R curves agree fairly well
with the values of~R that were measured for higherfeed rates.
Fig. S.4.b shows measured and calculated residence times for the stancle.rd
90 deg. classifier. The calculated residence times were of the same
order of magnitude as the measured values.

~
H
0.8

0.6

0.4

0.2
a)
0.0
2.0 2.1 2.2 2.3 2.4 2.5 •t (a/s)

7
(s)
25 b)

15

10

0
2.0 2.! 2.2 2.3 2.4 2.5 •r (a/s)

Figure 5.4.
M:asured (synixils) and calcula.ted (li.nes) of the fra.ction of bottan product, .PR
(a) and of the amn perticle residence ti..tl'e T (b) for the st:.Bndard 90 deg. classifier.

-130-
For the standard 120 deg. classifier the measured and the calculated
values of ~R agreed wellat the different Qv and vf (fig. S.S.a).
The measured and calculated partiele residence times also correspond with
each other (fig. S.S.b). This agreement is however less good than for the
~R curves because of the higher fluctuations in the measured values of
f.
Fig 5.6.a shows the results for the broad 120 deg. classifier. The
measured and the calculated ~R curves again correspond fairly well. The
measured residence times are however somewhat lower than the calculated
values of Tas can be seen from fig. 5.6.b.

'\
(-)
0.8

0.6

0.4
a)
Parare~} <g-rs)
1 "' 2.2
0.2 2 0 4.5
3 • 9.1

2.2 2.3 2.4 2.5 2.6

(s) b)
10

8
0

6
• 0~
Stamrd lal ~. c.laasifl..-
R=lO; V=S.
no:Jel fitrticles nl..').
2 f'dt'<l>eter: ()" (!lfs) 0 0 0
1 A 2.2
2 tl 4. ')
3 • 9.1
0
2.2 2.3 2.4 2.S 2.6 2. 7 2.8 • f (./s)

Fisure 5.5.
M:Jasured (syubols) and calculated (lines) val_yss of the fraction of bottem product, ipR
(a) and of the na::m {l'lrticle re9id:ll'lce tine T(b) for the standard 120 deg. classifier.

-131-
Table 5.1.

a) Mean transition times of the various partiele transitions used in the calculations to verify the model
(section 5.2.4).

Standard 120 deg, classifier: llroad 120 deg. classifier:

Q: t : t • t : t •
V tpf' pr t1-pf 1 1-pr' Qv: tpf' pr tl-pf 1 1-pr'
(g/s) (s) {s) (s) (s) (g/s) (s) (a) (s) (s)
0 0.347 0,257 0.470 0.422 0 0,240 0.180 0,382 0.240
2.23 0.322 0.231 0.472 0.417 2.08 0.262 0,194 0.390 0.262
4.51 0.321 0.241 0,474 0.406 4.46 0.254 0.190 0.402 0.254
9,04 0.304 0,254 0,492 0,434 I s.o4 0.266 0.218 0,410 0,266

Standard 90 deg. classifier:

Q: t t •
t1-pf 1
V
(g/s)
tpf'
(s)
pr
(s)
' {s)
1-pr'
(s)
0 0.159 0.132 0.323 0.271
1.40 0.186 0.146 0.405 0.3!6
2. 710.195 0.161 0.446 0.333
4. 710.210 0,173 0,471 0,360

b) the durations of transition times that were used in the model calculations {section 5~2~5) ..

classifier: . t • t
tpf' t1-pf 1 pr • 1-pr Qv.t
(s) (s) (s) 1 (s) (g/s)
st. 90 deg. 0.210 0.471 0.17310.360 4,7
st, 120 deg, 0.304 0.492 0.254 !0.434 9.1
br. 120 deg. 0.266 0.410 0.218 i 0.425 8,1

Qv,t: value of Qv at which these transition times were actually measured.

Table 5.2.

Correlations for the rising probability pv of particles entering the classifier at the feed stage as a
function of vf (m/s) and Qv (g/s),
Correlation: ·

Classifier: cl c2 c3 R
St. 90 deg. (Qv•O) 8 17,4
1.3 10 0 0.688
(Qv 0) 4 w-Il 24.8 -0.02 0.762
St.l20 deg. (Q.•O) 0.028 3.15 0 0.979
(Q. 0) 0.216 0.72 -0.16 0.728
Br.l20 deg, (Qv 0) 0.081 -O.ll 0 0.017
(Qv 0) 0.030 1.48 -0.20 0.370

R: multiple correlation coefficient,

-132-
5.2.5 Influence of V and R upon the classifier performance.
Model calculations were carried out to establish the influence of the
location of the feed stage and the number of stages upon the classifier
performance for feed rates of identical particles. The calculations were
carried out for each of the three classifier geometries that were
discussed in chapter 4.
The correlations for pf and pr as functions of vf' Ff,i and of
.
,1.- 1
that were established in chapter 4, were used in the
calculations. These correlations were assumed to be independent of the
location of the feed stage, V, and of the number of stages, R. Different
correlations were applied at stage 1 of the 120 deg. classifiers and at
stage 1 and 2 of the standard 90 deg. classifier as the transition
probabilities of these stages differed from those found at the higher
stages (see table 4.5) •

• ll
(-) parameter: Qv (g/s)

0.8 0 0
6 2.1
Cl 4.5
• 8.1
0.6
Broad 120 deg. classifier ..
0.4 R·lO: V•5.
model particles: o1 • 5 .
0.2
a)
0.0
1.11 2.0 2.2 2.4

(s)
Broad I 20 deg. classifier.
v.s.
particles: D!.S'
Parameter: Qv (g/s)
6
0 0
A 2.1
4 Cl 4 .5'
• 8.1

b)
0
1.8 2.0 2.2 2.4

Figure 5.6.
~ (syniJols) and calculated (li.n€5) valll€€ of the fraction of bottam product, iPR
(a) and of the trean partiele residence t::i.m? 'f (b) for the braad 120 deg. classifier.

-133-
It has been assumed that pv is a function of the partiele feed rate
Qv and of the air velocity vf for given model particles and of a
given type of classifier, Any influences of the local partiele flow rates
near the feed entry point upon pv were neglected. These correlations
have been determined from experimental results. Table 5.2 contains best
fit correlations of the form:

(5.23)

to describe this relationhip.

Table 5.l.b gives the single stage partiele transition times on the basis
of which partiele residence times were calculated. These mean transition
times were taken to be independent of both vf and Qv.
The following quantities were calculated for different Qv: the
superficial air velocity vf 0 for which the fraction of bottorn product
5
~R equals 50 %, the relative classification sharpness nr which is

defined as:

(5.24)

vfx is the superficial air velocity at which a value~ x/100 is

obtained. Furthermore, the partiele residence time , at vf = vfSO
50
and the maximum local partiele flow rates Ff and F at that air
,m r,m
velocity were determined.
The calculations were eerried out for zigzag air classifiers with 6, 10
and 16 stages respectively. The feed stage V was taken in the middle of
the channel (at stage R/2) for each of the three types of classifiers.
For the classifier with 10 stages, calculations were also eerried out for
feed stages located at stage 3 and at stage 7.
The results of these calculations are summarised in figures 5.7. The
separation sharpness nr has been plotted as a function of Qv/Qf'
3
which is solids-to-air ratio (kg/m ). It was assumed in the
calculations that the maximum obtainable throughput of material was
limited by the highest local partiele flow rate that would occur within
the classifier channel, This maximum acceptable flow rate was taken to be
35 g/s. This number was based upon the experience obtained in actual

-1%-
 n.. a)
(-) !X) deg. chmifJa".
m:x:1e1 part:icle!l: o1•
12 5
1: R-10; v.o. 4: ~; v~3.
2: R-10; v=S. 5: R-16; V~
3_ R=lO; V=7.
8 __:.:...----:::;.......,:

0 0.025 o.m 0.075 0.10

I
.......
n.. Starmrd la) deg. clJHnfier.
!rodel part:icle!l: D1.5'
'\-
(-)
lkmd la) deg. chmifJa".
(-) b) c) m:x:1e1 rarticles: D1.5'
w
1: R-10; \63. 4: ~;.V=3.
Y' 2: R-10; V=S. 5: R-16; V~. 8
1: R-10; \63, 4: ~;v.o.
2: R-10; V=S. s: R-16; v:a.
8
3: R=lO; V=7. 3: R-10; V=7.
6 5 6
5
4 2 3 4

2 2

0 0
0 o.m 0.1 0.15 O.D 0.25 ~ (lqifm~ 0 O.ffi 0.1 0.15 O.D 0.25 ~ (lqifm~

FiBUre 5. 7.
Results of I11XIe1 calculations.
The sepgration sharpJess vs. the solids-to-air ratio of classifiers w:ith diffecmt l'U1i:lers
of stagesRand locations of the feed stage V.
Table 5.3.
Results of model ealculations for partiele feeds consisting of identica1 particles.

a) Standard 90 deg. classifier.

R • 10; V • 3. R •10: v~ 5. R • 10: V • 7.

IQ": •ron' ~r: Ff m1 Fr.m' Tso' ()": •tso' ~:
r Ft.m' Fr.m' Tso' ()": vfSO: ~,: Ff.m' F
rm : Tso'
'0 2.442 6.33 0 0 '1A.7 0 2,364 6.19 0 0 25.3 0 2.3:8 5.20 0 0 :1:),6

i! 2.:m 6.91
2.439 7.3) 8.9 8.4 26.0 I 9.4 8.9 26.0 I 2.313 5.38 6.4 5.9 20.5
2.413 6.82 18.4 17.5 26,6 2 2.357 7.14 19.4 18.4 25.1 2 2.3:8 5.27 12.8 !l.8 27.3
2.319 7.78 37.7 35.9 28.3 4 2.2i0 6.76 43.0 41.0 26.6 4 2.237 5.56 3).3 28.3 19.8
2.193 14.'/A 87.4 83.4 32.3 8 2.177 7,16 9;).4 86.4 26.5 8 2.152 6.11 68.0 64.0 19.8

R = 6: V • 3, R • 16; v •8.
()": "tSO' ~· Ff m1 Fr m: Tso' ()"• •t:!!:l ~· Ff.m' Er.m T!:(j

0 2.:314) 3.86 0 0 7.8 0 2.399 10,00 0 0 64.9

1 2.357 4.56 5.7 5.2 8,5 1 2.388 10.85 15.0 14,5 67.7
2 2.561 4.(8 ll.1 10,1 8.2 2 2,329 9.32 27.9 26.9 62.7
4 2.lll 4.93 25,5 23.5 8.4 4 2,'1A1 7.47 54.3 52.3 00.6
8 2.2:1:) 8.0!+ 56.7 52.7 8.6 8 2.150 7.47109.6 105.6 00.6

b) Standard 120 deg. classifier.

R•10; V•S. R • 10; V • 7.

Frm: F
rm:
~ ~:~ ~~
0
6.1 g,7 8.9
9;9
0 2.464 6.96 0
2 2.648 5.73 6.0
0
5.9
0 2.396 5.84
2 2.541 4,71
0
4.3
0
3.3
8.0
8.7
4 2.!m 4.(!} 11.0 10.2 9.0 4 2.103 5.ll 11.4 9.4 4 2.'-ll 4,38 8,3 6.3 8.2
8 2,(!}8 4.34 20.0 18.3 8.2 8 2.776 4.51 21.6 8 2.644 4.C6 16.2
~:~
17.6
16 3.007 3.05 56.4 32.7 74 16 2.852 3.91 «l.8 328 16 2702 371 31.5 i!
R·l6; v-s.
F :
til

~ ~:: ~:~ ~.00

4 2.617 3,«) 7.7
~.1
5,7
;:~
3.3
~ i:: 1
~:~ ~.7
4 2. 774 6.(jJ 16.5
0
7.7
14.5
~:!
'1A.5
8 2.687 3.00 14.6 10.6 3.0 8 2.846 6.C6 31,2 27.2 22.8
16 2. 766 2.58 27.5 19.7 2.8 16 2.928 5,'/A 58.0 51.0 21.1

c) Broad 120 deg. classifier,

R = 10: V • 3. R·10: V= 5. R=10:V=7.

Cl"• •tso' 'V ,Ff.m' Fr.m' 'so' Cl"• •tso' n r : Ff m1 Frm:
'so' Cl"• •tso' nr ' Ff.m' Frm :
'so'
0 2.!!! 4,81
2 2.378 4.C6
4 2.165 3,(9
0
4.8
9.4
~.8
7.4
~:~
6.4
~ t: ::~ 0
5.2
4 2.357 3.82 10.1
0
4.2
8.1
7,4
6.9
6.9
0
2
4
2.211
2.195
2.254
4,0l 0
3.73 4.3
3.54 8.4
0
3.3
6.4
6.3
6.3
6.3
a 2.541 3.31 18.6 14.6 6,4 8 2.43) 3.63 19.7 15.7 6.8 8 2.317 3.34 16.5 12.5 6.2
16 2.639 2.!1} 42.6 34.6 6.3 16 2.510 3.32 38.3 3),3 6.8 16 2.385 3.13 32.3 24.3 6.2

R"' 6; V • 3. R • 16: V • 8.
()": vfSO: '\-'
Ff m' Frm:
'so' ()"• .tSO, 'lr• Ff.srf Fr m1 tso'
0 2.284 3.54 0 0
~i:~~~~
2.9 0 0 17.4
2 2.276 3,00
3.7 2.7 2.7 7.1 6.1 16.4
4 2.303 2.68
7.4 5.4 2.7 4 2.388 5.33 13.8 11.8 16.1
8 2.415 2.$ 14.4 10.4 2.8 8 2.464 4.93 26.9 22.9 15.8
16 2.494 2.31 33.2 25.2 2.9 16 2.541 4.53 52.4 44.4 15.5
()"• port:icle feed rate (g/s).
vfSO: - velocity for ..trlch the freetial of batton produCt tR • 0.5 (m/s).
V ~ sep!nitial ~ <->.
TSO: port:icle residence tillil at vrf50 (s).
Ff .m: l1BXImiD local port:icle fl<>< in ~ stresn (g/s).
Fr,m' l1BXimiD local port:icle fl<>< in risi.r@ stream (g/s).

-136-
experiments with our model particles in the standard 90 deg. classifier.
From the results shown in these figures it is concluded that an increase
in the number of stages leads to higher separation sharpnesses and lower
capacities for both the standard 90 and 120 deg. classifier. Placing the
feed stage at the middle stage of the 120 deg. classifier leads to high
separation efficiencies at the cost of low capacities. Which is the feed
stage best suited for the 90 deg. classifier depends upon the feed rate
that is applied.
The influence of the number of stages and the location of the feed stage
upon the classifier performance will be discussed in some more detail
below.

a) Influence of the number of stages.

The results are presented in table 5.3 and in fig.s 5.8 - 5.10.
Figures 5.8.a-5.10.a show the relative classification sharpness nr as a
function of the number of stages. nr can be considered to be a measure
for the derivative of $R(vf/vf ) at vf=vf , (see section
50 50
2.5.1, eq. 2.5 and 2.6.

1 1
n =A n +A n (5.26)
T f f r r

Af and Ar are the amplification factors of the single stage

efficiencies nf 1 for falling particles and nr 1 for rising particles
respectively. Senden /5.1/ showed for low Q that Af and A are
v r
high, when particles have high probabilities to remain at the same stage
for a longer time (pf/pr > 1). Af and Ar are low in case
pf/pr < 1. The amplification factors furthermore become larger when
the number of stages R increases and when the feed stage is located at
the middle stage of the channel. At Qv=O, the single stage efficiencies
nr 1 and nf 1 are highest for the standard 120 deg. classifier and
lowest for the standard 90 deg. classifier (compare the values of the
parameter n in table 4.5). From table 4.5 it follows furthermore that
nfl iS larger than n I at Q =0.
r v
The increase of n with increasing numbers of stages (fig.
r
5.8.a-5.10.a) is explained by the increased number of single stage
classifications to which the partiele is subjected before it reaches one
of the exits. This causes higher amplification factors Ar and Af and
thus a better separation sharpness.

-137-
 '\:- Stanlard CJ) deg. cla9slfl.eo.-. ~ St:axlard CJ) deg. clamifier.
1
2 a) The ditrensionless c18EiSification
(..:) (\,
m:xle1 J>rrticles: D • liDdel J>rrticles: D • 4 shar[ness.
15 15
Y..f'J2. ' (-) Y..f'J2. ' 0 b) The ratio of the 11li:'JXÎJT1Jf110C81
10
~ter: Q_(g/s).
12
l'araneter: Q_ (g/s). p:!Itiele flaN rate to the p:!Iticle
ft€([ rate.
8 lO
4 c) The ratio of the air velocity vf50
8 to the air velocity vf50.0
6 at Q =0.
a) 6 b)
d) The fm-ncie residmce tilre at
4
vf50.
2
6 10 16 R (-} 6 10 16 R (-)

YED t~
I vEO,O (s) Sblnlard 'll deg. classifier.
..... ~ liDdel J>rrticles: D
U,) (-) 1' 5
(X) V=fl./2.
I
0.975 ~ter: Q_ (g/s).

St:axlard 'll deg. class:ifier.

0.95 25
liDdel J>rrticles: 0
15
Y=fl./2. ' 4
0.925
~er: Q_ (g/s). ''
I ''
''
O.'ll
c)
''
d)
10
I'
I
6 10 16 R (-)
'
'
I

5
6 10 16 RH

Fisure 5.8.
Results of m:x1e1 calailations as functions of the IIUllber of stagEs R in the standard 90 deg. c18EiSifier.
 llr 0 Ff,m
Standard 120 deg. classifier. a) The d:i.nelsionless el85Sificatioo
(-) Standard 120 deg. classifier. Qv model particles D1 5 •
model particles: Dl.S' (-) V=R/2. '
shar.pless.
V=R/2. b) The ratio of the llEXÎill.B1l local
2
3 Parameter: Qv (g/s). partiele flow rate to the partiele
8 Parameter: Qv (g/s). 4
f€Ed rate.
6 2 c) The ratio of the air velocity vf50
to the air velocity vf50,0
4 at Q =0,
a) b) d) The P:rt:iele resi.droce tiJie at
2 0
10 16 K (-) 6 10 16 R (-) vf50'
6

T~
I Standard 120 deg. classifier • (s) Stalldard lal deg. classifier.
......
w vf50,0 model particles: Dl.S' 25 model JErticles: D1 S
16
"'I V=R/2. ~S V=R/2. '
Panm:>ter: ~

~
1.15 (g/s).
10
1.10

1.05 5
c) d)
1.00 -------·---- 0
Parameter: Qv (g/s) •
2
6 lO 16 R (-) 6 10 16 R (-)

Figure 5.9.
Results of ll1Xlel calrulat:ions as functions of the l'lUI'd:>er of stages R in the standard 120 deg. cl85Sifier.
 a) The dimensionless elassification '\
l!m!d m ~~eg. c.1assl..fie<-.
~
sbar{n2ss. (-) ~ l!m!d m deg. c.1assl..fie<-.
m:xtel (mticles: D • (-) nOOel (mtlcles: Dl.S
b) The ratio of the maxi.nun local V~2.
1•5
partiele flow rate to the partiele 5 V~2.
Panm!ter: 0._, (g/s). 3.0
ffed rate. Panm!ter: 0._, (g/s)
c) The ratiC: of the .air velocity vfSO 4 2.5
to the Bl.r velOCity vf50 0
at Q =0. ' 3 2.0
d) The Partiele residence t:iJre at a) b)
2 1.5
6 10 .16R{-} 6 lO 16 R (-)

I
.....
.10-
?
'[~

vG:l,O l!m!d 12) deg.~l.....Ufier. 16 (s) l!m!d lal deg. classifier.

(-)
m:xtel (mticles: D
lO nOOe1 (mtlcles: Dl s·
~2. • V~2. '
1.075 Parareter: 0._,~8 Panm!ter: <lv.
5
1.05 4
3
l.UlS
2
2
1.00 ~~
c) d)
l,
6 10 16 R (-) 6 10 16 R (-)
Fisure 5.10.
Ra:lults of m:xlel calculations as functions of the nuniJer of stages R in the blU:ld 120 deg. elassifier.
Q influenees the n ' and nf' and for this reason the separation
v r
sharpness most when the number of stages is largest. Partiele-partiele
interaetions are largest then.
Figs. S.lO.b - S.lO.b show the influenee of the number of stages on the
ratio of the maximum loeal flow rate of deseending particles to
,m
Qv • The capacity of the classifier depends upon the value Ff ,m si nee
the formation of eonglomerates and eventually the blocking of the ehannel
will occur when Ff ,m exceeds a maximum value.
The maximum flow rate Ff ,m occurs at the stage boundary immediately
below the feed stage in most of the calculations. At a given Q ,
V
partiele flow rates inside the classifier channel increase when the
channel consists of more stages. The ratio Ff ,m /Qv is not much
influenced by Q (less than appr. 25 %).
V
Figs. 5.8.c - 510.c show the influence of the number of stages upon the
ratio vf 50 /vfSO,O , vf 50,0 being the air velocity at which ~R=O.S
at very low feed rates (Q =0). This ratio represents the location of
V
the half value of the ~R function at a Qv compared to the half value at
Qv=O. The more sensitive this ratio is to changes of Qv' the more
sensitive the classifier performance will be to fluctuations in the
partiele feed rate.
The value of vf 50 /vfSO,O differs more from unity as the number of
stages increases and as Q increases. Both an increase of Q and R
V V
lead to an increase of the internal flow rates, These higher internal
flows influence pf and pr more. The classification result that is
characterized by the function~R(~f) will thus also be influenced most
under these conditions.
The partiele hold-up H depends upon the partiele residence times. The
classifier capscity decreases as H becomes too The mean partiele
residence time has been calculated at vf:vf 50 • This residence time
'so increases with an increasing number of stages as a partiele has to
pass more stages (figs.5.8.d - S.lO.d). The influence of Qv upon ,
50
is insignificant.

b) Influence of the 1ocation of the feed stage.

These results are given in figures 5.11-5.13.

Figures 5.ll.a-5.13.a show the separation sharpness nr as a function of
the location of the feed stage. The highest values of nr were found for
a feed stage located at. the middle stage of the channel (V=S) for the

-141-
standard and broad 120 deg. classifier. However, for the standard 90 deg.
classifier n reaches its maximum at V=S only at Q = 2 g/s. At the
r v
other feed rates, the maximum values of nr were found for V=3. It was
mentioned above that, at constant values of pf and pr' the
amplification factors Af and Ar reach their maximum value for V=R/2.
However, the influence of the 1ocation of V upon the amp1ification
factors is seen to dominate for the standard and for the broad 120 deg.
classifier.
At low Qv nr reaches higher values for V=3 than for V=7 for all
investigated classifiers. Higher vf values have to be applied for a
feed stage V=3 than for V=7 to obtain the same fraction of bottorn product
(see table 5.3). This causes a higher value of the ratio pf/pr as
pf increases faster at increasing vf than pr. The mean number of
single stage classifications becomes larger for each partiele during its
stay in the channel which results in larger amplification factors.
At higher values of Qv' the effect of the partiele flow rates Ff and
F upon the single stage efficiencies n I and nf' also becomes
r r
important. It follows from figs. 5.11.a-5.13.a that nr continuously
decreases at increasing Qv for the 120 deg. classifiers. Here
increasing flow rates lead to lower pf/pr ratio's and therefore to a
lower number of single stage classifications. For the standard 90 deg.
classifier, nr may both increase and decrease at higher Qv depending
upon the effect of flow rates upon pf/pr.
The value of n~ for V=3 at a constant Qv can he both higher and lower
than the válue calculated for V=7 for the standard and broad 120 deg.
classifier. The maximum value ofnrat constant Qv in the standard 90
deg. classifier can both he found at stage 3 and at stage 5 depending
upon the actual value of Qv.
Residence times TSO generally show the same dependenee on V and Qv as
nr· It was shown above that an increase of nr is always caused by the
fact that particles are subjected to a larger number of single stage
classifications. The particles then stay longer in the column which leads
to higher residence times.
From figs. S.ll.b-5.13.b it follows that the ratio Ff /Q has the
,m v
highest values when V=S. The average number of stages that a partiele
passes then is 1arger than when the feed stage is located more to the top
or to the bottorn exit. Consequently, the partiele flow rates within the
channel are larger.

-142-
 a) The diiialsionless classificaticn
shaqness. '\- Stallord IJ) deg. claBi.f:iw.
a) ~ Stallord 'D deg. claBi.f:iw. d)
(-)
m:xle1 pnticles: 0 •
15 Q..
b) The ratio of the max:iJrum local R=IO. 4 • (-)
partiele flow rate to the partiele 7 1 10
tred rate. 2
c) The ratio of the air velocity vf50 0 8
6
to the air velocity Vf50 O
at Q =0. ' 6
5
d) The ~tiele rasidalee tinE at ~:Q_. (g/s).
vt5o· 4
4
3 5 7 V(-)
2
3 5 7 V(-)

T~
I vf:n
..... (~
4
e)
vf~.o
"""
w
I (-) 2

--------
0.975 1

0.95 21.
0 - --
SI1Diard 'D deg. claBi.f:iw.
SI1Diard 'D deg. c1assifier. m1e1 Imt:tcles: o •
0.925 m:xlel particles: D Rl=lO. 1' 5
15
Ri=IO. •
l'armeter: Q_. (g/s).
o.cn F'armeter: Q_. (g/s)
c) al

3 5 7 V(-) 3 5 7 V(-)

Figure 5 .11.
Results of IJDdel cala.!lations as functions of the tred stage locaticn V in the standard 00 deg. elassifier.
a) The d:i.Irensionless elassification
shBrp7eSS. Standard 120 deg. classifier.
b) The ratio of the nax:i.m.Jm local model particles: o • •
R=lO. 1 5
partiele flow rate to the partiele 6 2
feed rate.
4
c) The ratio of the air velocity v f50

~~~
5 3.0
to the air velocity v f50
at Q =0. •0 b)
4 2.5
d) The }mtiele resida'!Ce tille at Parameter: Qv (g/s).
vf5o· 3 2.0
3 5 7 V (-) 3 5 7 V (-)

·~ r----------r-.----~------~~
R:lO. jmticl~es:
Standard 120 deg. classifier (s) nxxlel D1 •
moel particles: o • • 10
R=lO. 1 5
1.15

1.10 8
~·-~
1.05
c) d)
1.00 --------------0 6
Parameter: Qv (g/s). 3 5 7 V(-)

3 1 V (-)

Figure 5.12.
l?esul.ts of rn:x1e1 calrulations as functions of the feed stage location V in the standard 120 deg. elassifier.
a) The di.Irensiooless classificatirn \- ~ 1"'""'----.,.---...,.---.,........,
lkaJd Ja) dog. classifi8'.
sharp:JeSS. (-) llroad Ja) ~ nvdel pnticles: n • b)
5 liDdel pJrl:ie},ilil+----- (-) 1•5
b) The ratio of the maximum local R=IO.
Dl.S 16~--------·--,,
partiele flow rate to the partiele 4 2.5
fa:rl rate. 2 ',. 0
8
c) The ratio of the air velocity vf50 3 2.0
to the air velocity vf50,0 Parareter: <ly (g/s) Pararet.er: <ly (g/s).
at Q =0. 2 ~------~------~----~~ 1.5 '
d) The Partiele residroce tim2 at 3 5 7 V(-) 3 5 7 V(-)
vf50'
VED 1"'""'~---~---~~--~--,
BriJlid Ja) deg. c1aesifier•
veQ,O R=lO. liDdel pnticles:
D liDdel pnticies: D.Ls.:. ____
(-) Parareter: 0,. (g/s). 1.5 R=lO. --- •

~
1.075 7.0 Parareter Q._ (g/s). ',,,

l.ffi 6.5
d) ~\ ' 16
1.025 c)
3 5 7 V(-)
----------- 4

3 7 VH

Figure 5.13.
Re9ults of m:xJe1 calculations as functions of the fa:rl stage location V in the broad 120 deg. elassifier.
An exception occurs for the broad 120 deg. at Q =16 g/s (fig. 5.13.c),
V
Here Ff ,m /Q v reaches its maximum value for V=3. This can again be
explained by the higher vf 0 values found for V=3 than for V=5 (see
5
table 5.3). Thus, pf/pr will be larger for V=3 than for V=5 at
constant values of Ff and Fr. The particles are subjected toa larger
number of single stage classifications which causes an increase of the
local flow rates. Apparently, this increase of the flow rates due to
higher pf/pr ratios is more important than the larger flow rates due
to the higher number of stages that are passed when V=5.
For all investigated classifiers, the value of Ff /Q is higher for
,m v
V=3 than for V=7. This is again explained by the higher pf/pr ratios
for V=3 than for V=7
The influence of V and Qv upon the ratios vf/vf 0 and , can be
5 50
explained by the pf/pr ratios in the same way as has been done for
Tl r and Ff ,m /Q v •

c) Comparison of the investigated classifier geometries at R=lO and V=5.

It follows from fig. 5.11-5.13 that~ Ff,m/Qv and , 50 are lowest

for the broad 120 deg. classifier. The ratios pf/p r measured inside
.
this classifier are low (less than 1) and the amplification factors Af
and A are therefore low too. Particles are subjected to a low number
r
of single stage classifications and have short residence times. Local
partiele flow rates are therefore also small. Hence, this classifier has
a relatively high capacity.
The standard 90 deg. classifier and the the standard 120 deg. classifier
have comparable classification efficiencies. However, partiele residence
times are much shorter for the standard 120 deg. classifier. The
particles inside the the 90 deg. classifier are subjected to a large
number of single stage classifications (pf/pr>1). Thus, the
amplification factors Af and Ar are higher than for the 120 deg.
classifiers. The partiele residence times are long. The single stage
efficiencies nr' and nf' are lower for the 90 deg. classifier than
for the 120 deg, classifiers. These lower values compensate the effect of
higher amplification factors (see equation 5.26).
5.3 The classification of binary mixtures of model particles.

5.3.1 Effect of partiele thickness and density.

In chapter 4 it has been shown that for a given channel geometry and for
particles with given dirnensions b and 1 the following equation
p p
deterrnines the transition probabilities:

. 1) = f( vf'lvfl; Ff ,1./Fr,1-
pf ,1. (or pr,1- . 1; Ff ,1. L /(Vp pp vfl)) (5.27)

The dirnensionless group vf/vfl represents the ratio of the actual

superficial air velocity vf to vf 1 • vfl is a measure of the
partiele falling velocity in stagnant air. At very low Qv the
dimensionless separation function ~R(vf/v 50 ) is for our particles
independent of the partiele thickness d for a given classifier
p
geometry. This was also concluded from the results of Senden /5.1/.
Furthermore, we showed in chapter 2 that vf 0 is linearly proportional
5
to vfl' Thus, the performance of a classifier for flat particles
(d « 1 ) with densities p and thickness d can easily be
p p p p
derived from experimental results obtained for other particles for which
only pp and dp are different, ~R(vf/vf 1 ) will be the same for
the two kinds of model particles.
One would expect that at higher values of Qv the dimensionless groups
Ff ,1./F r,1-
. 1 and (L Ff ,1./(Vp p p vf 1 )) must also play a role as
the partiele heldups and therefore partiele-partiele interactions at the
different stages increase. Therefore the ~R curves were calculated for
the model particles c to establish the sensitivity of the
2
dirnensionless correlations for variations in d and p •
p p
The algorithm that was presented in section 5.2.2. was used. The sarne
dimensionless relationships for pf and pr that were measured for the
model particles n • were applied. It was furthermore assumed that Cd
1 5
was the sarne for bath types of particles and that the variation of the
other dimensionless groups (equation 4.32) had no significant influence
on the separation performance. The effect of these groups will be
discussed in section 5.4. The results are presented in figs. 5.14, Values
of nr and of vf 50 /vfSO,O for the two kinds of model particles are
plotted as a function of the partiele feed rate Q /m • The results
V p
show that the correlations predict only a moderate effect of d and
p
p within the variations used in the experirnents. This is in agreement
p

-147-
 a, c: Effect of partiele feed rate on vfD
~
the ratio of the air velocity vfD,O : (-)
vf50 to the air velocity (-) a) 6 b)
vf50,0 at Ov =0. "' Stmdard '» deg. clsssi.f:iB".
'
b, d: Effect of p:li'ticle feed rateon
0.95 s' '·,, 1\.s 4
the sep:li'ation sharpless f)r. '
Stmdard 'Xl deg. clsssi.f:iB".
O.'Xl
v..s.
f;= p:li't~cle th~ckness: 0.0145 an.
2 ~10;

, :[l!3I't~cle th~ckooss: 0.0291 an.

1 5 0.85
R=lO; V=S.
0

0 25 'I) 75 100 o/-p (&) 0 75 100 o/-p (&)

I
...... vfD
J>..
00
I vfD,O
(-)
c)
~10; V=S.

<1.....----- ---
----
Standani lal deg. ~.=. ~
(-)
1.10 6
,..-
-- .. --
/
/
/
1.m /
4
I
I
I
s------------ -
I
Broad lal deg. ~.
1.00 2 R=lO; V..S.
d)
0 25 75 0
25 75

Figure 5.14.
• Sensi.tivity of the results of the nrxle1 calculations to diEferences in p:li'ticle thickooss.
with the results of chapter 2. There it was shown that the separation
sharpness that was measured for the two kinds of model particles. did not
differ significantly when equal partiele feed rates based upon numbers of
particles were applied. Also the residence times for the two model
partiele types showed similar values. One should however be careful to
these results for particles that have great differences in dp /L
and in shape or density compared to the particles that we used. In these
situations the parameters Ff/(V p pp vfl) and Ff .1./Fr.1-
. 1 may
have values that are much higher or lower than the values of our
measurements. However. in the separation of paper and plastic foils from
fractions of household waste. we believe that the range of values of
these parameters is relatively small. In such situations the influence of
the feed rate of given particles on the classifier performance may be
predicted from results obtained for particles with different d and
p
p •
p

5.3.2 Prediction of the classification result of binary mixtures.

fra. single component classification functions.

The classification of binary mixtures has been characterized by means of

the component separation efficiency Ef:

(5.28)

In this equation wl,b and w2 ,b are the bottorn product weights of

component 1 and 2 respectively. ~l,F and w2 ,F are the feed rates of
the two components. It was shown in chapter 2 that Ef can be calculated
from the two $R(vf.Qv/mp) functions of the individual components.
QV/mp is the numbers base partial partiele flow rate of each
individual component.
At very high Qv' the calculated values of Ef differ from the measured
values. This is caused by interactions between the two types of different
model particles. It was shown also that the predictions of for the
standard 120 deg. classifier agreed better with the measured values than
is the case for the standard 90 deg. classifier. This has been explained
on the basis of the lower partiele hold-ups inside the 120 deg.
classifier that causes partiele-partiele interactions to be less

-149-
important in the 120 deg. classifier than in the 90 deg. classifier. In
addition, partiele-partiele interactions are more intense inside the 90
deg. than in the 120 deg. classifier as partiele veloeities are much
higher.

5.4 Some aspects of the design and sealing-up of zigzag air classifiers.

5.4.1 Selection of the channel geometry.

The selection of a classifier geometry includes the channel angle, the
channel width, the location of the feed stage and the number of stages.
The influence of these parameters has been discussed in chapter 2. Figure
5.15 summarizes the results. The classification sharpness Tt has been
plotted as a function of the classifier solids-to-air ratio Q/Qf. A
channel geometry can be selected by taking that classifier geometry which
supplies the desired separation sharpness at a maximum Qv/Qf. The
designer must scale a classifier to meet a specified capacity. This
aspect is dealt with i~ section 5.4.2.
One can obtain a higher separation sharpness by increasing the number of
stages. This aspect has been delt with in section 5.2. These results have
been summarized schematically in figure 5.7. Again, a selection can be
made by choosing that classifier geometry that.supplies the highest
solids-to-air ratio at a desired separation sharpness.
Whether the selected classifier will be the most suitable in a practical
situation depends upon other aspects too. It must be realised that the
classification efficiency for mixtures of particles at high partiele feed
rates depends upon the interaction between the different sorts of
particles to be classified. These influences have been discussed in
sections 2.5.3 and 5.3.
A further aspect is the cost of building and operating the air
classifier. These costs include the operational costs of the blower that
transports the desired amount of air, the partiele feed system, cyclones
to separate the top product from the air as well as the costs of material
and construction of the classifier channel. These material and
construction costs are low compared to the total costs,

5.4.2 Sealing-up of zigzag air classifiers.

In chapter it has been shown that the partiele transition probabilities

pf and pr are functions of a number of dimensionless groups:

-150-
'\- Channel width (cm) I f eed stage /
wall roughness:
JO a) OOä;g.
8 A: 14.1 I 5 I rough
B: 14.1 I 3 I rnugh
6
C: 14.1 I 7 I rough

4 D: 20 I 5 I rough
E: 10 I 5 I rough
2 F: 10 I 7 I rough
G: 4.1 I 5 I smooth

0.0 0.04 o.rn 0.12 0.16 ~ (kg/m3)

'\- Channel width (cm) I feed stage I

H
wall roughness:
10
b) 11Dä;g.
8 A: 10 f 5 I rough
B: 10 I 7 I rough
6
C: 10 I 5 I smooth

4 D: 15I 5 I smooth
E: 6 I 5 I rough
2 F: 5 I 7 I rough

o.o 0.04 o.rn 0.12 0.16 Q"~ (kg/rh

Figure 5.15.
The classification sharp:Jess nr as a funct:ion of the solids-to-feed ratio Q /Qf in
v
different channe1 gearetrias of classifiers w.tth anglas of 90 deg. (a) and 120 deg. (b)

flat. p1ate

10-1

10~ +-~---r--~~---r--~~---r~
w-2 10-1 10° 101

Figure 5.16.
The drag roefficient cd vs. the Reynolds nurri:Jer Re for different partiele
geametries. P

-151-
with

(5.30)

This equation differs from equations 4.32 and 4.33 that have been derived
in chapter 4 as it contains the Re groups. Under the actual process
conditions of our experiments these numbers could be considered to be
constant. It will be discussed below to what extent this assumption
applies to other classifier dimensions and other particles. The
importsnee of the different dimensionless groups upon the classifier
performance and the effect of sealing-up upon the value of each group
will be discussed also. It must be realised that these effects have not
been established experimentally for many conditions. The discussion below
is therefore qualitative.

a) Dimensionless groups that are independent of the partiele properties.

These groups are Rech' S, B/L, and W/L.

Re eh:
The shape of the air flow profile depends upon the classifier geometry as
well as upon the Rech number. In many practical situations air
veloeities of appr. 8 m/s have been applied to classify waste fractions
/5.4/. In channels with a width of 0.5 m Rech values will then be 10-15
times larger than the Re h values which existed in our measurements
4 4 c
(10 5.10 ). At these ranges of high Re numbers it is expected that
the superficial air velocity will no longer influence the shape of the
air flow to a significant degree.
For this reason the role of Rech in sealing up will be relatively small.

f3 and W/L:
The role of the classifier angle f3 upon the classifier performance has
been discussed throughout this dissertation. It has been shown that
must be considered in relation to the channel width W and channel section

-152-
length L. Sealing up should be carried out by keeping the channel
geometry (W/L) and S constant. If this is done the shape of the air flow
profile that existed in our bench scale experiments will be maintained at
other classifier dimensions.

B/L:
The number B/L does not influence the air flow profile to any significant
degree. It should be realised however, that the feed material has to be
fed uniformly over the classifier depth B of the feed stage. This limits
the value of B that can be applied in practical situations to appr. 1 m.
At larger values of B it will become difficult to distribute the feed
homogeneously.

The dimensionless groups that depend upon the partiele properties are:
vf/vfl, ,/Fr,i- 1 ' L/(Vp pp vfl). P/Pf• 1/L•
d /1 ; b /1 ,Re and f.
p p p p p

v/vn:
In the case of of relatively large, foil shaped particles, the shape of
the dimensionless ~R curve is determined mainly by this number at very
1ow Qv. The shape of the ~R function for other flat materials can
thus easily be determined if vfl is known.

. 1 and Ff ,1. L/(V p p p vf 1 ):

Ff ,1./Fr,1-
The dimensionless groups vf/vfl' Ff ./F .
,1 r,1- 1 and
Ff,i L/(Vp PP vf) have been discussed insection 5.3.1. It was
concluded that the ~R curves could be described by functions of
vf/vfl and Qv/ Vp pp) instead of by the more complex forms that
were derived to describe pf ,1. and pr,1-
. : Ff ,1./Fr,1-
. and
1 1

P/ Pf:
This number delermines the buoyancy force. lts value is generally larger
than appr. 800 and can therefore be neglected, This number would become
important in cases where a high density fluid is to be used for
classification instead of air. This is rarely done in actual practice,

-153-
1 /L:
p
This group represents the ratio of the largest partiele dimension to the
channel section.length L. It is of importance as the intensity of
partiele-partiele and of particle-wall interactions is determined by the
value of this group. The partiele-partiele interactions increase as the
hold-up of partiele surface within a unit of channel volume becomes
larger and thus when 1 /L increases. The influence of increasing
p
hold-ups on the $R curves and on the efficiency curves has been shown
in chapter 2. At values of 1p /L » 1 blocking of the channel will occur.
Our experiments were selected to describe the behaviour of relatively
large, foil shaped materials in zigzag air classifiers (lp/L = 0.1 -
0,3). Sealing-up should be done by keeping 1 /L within the samerange
p
of values at which we carried our measurements if one wants to apply the
results given in this dissertation.

d /1 and b /1 :
p p p p
These groups are determined by the shape of the particle. d /1 will
p p
be small for our type of materials: flat, foil shaped particles like
paper and plastic. The shape of the particles determines the actual value
of the drag coefficient Cd (see below) as well as the area A~ that is
normal to the air flow. At low values of d /1 the effect of this
p p
group upon Cd and A~ will be insignificant.
The motions of particles that fall into a vertical air flow depend upon
b /1 • For values of this group of appr. 1, particles will more or
p p
less "flutter" like leaves that fall from a tree while for values of
b /1 >> 1 particles will rotate around their longest axis of
p p
symmetry /5.3/, /5.5/. Therefore the averaged value of the area A~ is
different in these two cases. Thus, the drag force as well as the lift
force differ.

Re
p:
The Re number is important as the drag coefficient and the lift
p
coefficient Cd and c1 are generally functions of Rep. It was argued
in chapter 4 that Cd and c1 can be considered independent of Rep in
the range of Re numbers we used. A typical relation for Cd as a
function of Re for different partiele geometries is given in fig. 5.16
p
/5.3/, /5.4/. Cd values are given for flat objects with their largest

-154-
surface turned towards the air flow as well as for apheres and cylinders.
Cd becomes independent of Rep for Rep > 1200.
The influence of Cd value and of the numbers dp /1 p and bp /1 p
should be incorporated in the value of vfl when particles with
different values of Cd and/or different values of A~ are being
classified.

f:
The friction coefficient f between the particles and the wall mainly
influences falling particles. Increasing f leads to lower partiele
veloeities and therefore to an increased top product flow. By
means of adjusting this parameter, a better separation efficiency may be
obtained for particles that show a small difference in vfl as well as a
difference in wall friction. This is illustrated by the following
example:
Assume that two types of particles have to be classified. One type has a
rough surface, the other type a smooth one. The rough particles have a
somewhat lower falling velocity. However, the difference in falling
behaviour between the two particles (expressed by vfl which is a
measure of the falling velocity) is small so that the particles cannot be
separated at high classification efficiencies in any zigzag air
classifier. A better separation efficiency is obtained by using a
standard 90 deg. classifier with rough walls instead of smooth walls. In
the falling streams, the rough particles will be slowed down more than
the smooth particles because they have a greater friction coefficient.
Consequently the rough particles will be taken upward more easily. This
results in a better classification than when smooth zigzag walls are
used. A 90 deg. classifier is selected as the particle-wall interactions
are more important here than in a 120 deg. classifier. It will be clear
that rough zigzag walls should be avoided in the case where the heavy
particles have rougher walls than the light particles.

5.5 Conclusions.
The separation function ~ for zigzag air classifiers with angles of 90
and 120 deg. can be predicted rather accurately from the dimensionless
correlations that were derived in chapter 4 to determine the transition
probabilities pf and Pr· Also, acceptatle estimates for the partiele
residence times can be calculated for known values of the single stage

-155-
 transition times. A modification of Senden's mathematica! "one-step
memory" model was used for these calculations. This model also makes it
possible to predict the effect of the number of stages and the location
of the feed stage on the classifier performance.
The ~R curves can be used to predict the separation efficiencies of
mixtures of particles.
Qualitative information about the effect of sealing-up upon the
classifier performance is obtained from the dimensionless groups that
have been derived to describe the transition probabilities.

References.
5.1 Senden, M.M.G., "Stochastic models for individual partiele behavior
in straight and zigzag air classifiers.", dissertation (1979),
Eindhoven Universiy of Technology.
5.2 Veltkamp, G.W., Geurts, A.J., "Numerieke methoden I en II.",
Lecture notes (1979), Eindhoven University of Technology.
5.3 Perry,R., Chilton,C., "Chemical el)gineers' handbook", 4th ed.
(1973), McGraw-Hill, New York.
5.4 Diaz, L.F., Savage> G.M., Golueke, C.G., "Resource recovery from
municipal solid wastes." (1982), CRC press, Inc., Boca Raton,
Florida.
5.5 Willmarth, W.W., Hawk, N.E., Harvey, R.L., The physics of fluids,
l (1964), no.2, p197-208.

-156-
LIST OF SYHBOLS.

a accelleration 1 T-2
A aerodynanic area of p:trticle 12
A IreaiJ. area of p:trticles and cooglarerates 12
con
A' area of conglarerate of m:xlel p:trticles 12
con
1
Af amplification factor of the single stage efficiency Tlf
A area (d;' 1 ) of flat m:xlel p:trticle 12
p p p
A amplification factor of the single stage efficiency Tl '
r r
b width of m:xlel p:trticle 1
p
B classifier channel depth 1
cd drag coefficient
3
c. p:trticle concentration at stage i MC
l

s lift coefficient
p:trticle concentration at top stage MC
3
ct
d IreaiJ. thickness of conglarerates and p:trticles 1
con
d' IreaiJ. thickness of conglarerate of m:xlel p:trticles 1
con
hydraulic d:i.aneter 1
tb
d m:xlel partiele thickness 1
p
e tmit vector in x direction 1
x
e tmit vector in y direction 1
y
Ef canponent separation efficiency.
E nmcinrum value of Ef (calculated)
f,c
E DBXinun value of Ef (nmsured)
f,m
f frequency T-l

f friction coefficient
f, f , functions
1
f2
f(t ) transit time distribution function (histogram) T ..~-1
p
F p:trticle flow rate T-l

local flow rate of falling p:trticles MT-l

Ff
Ff ,l. flow rate of falling particles at stage bounda:ry i M'l'-1
F local flow rate of rising p:trticles MT-l
r
F flow rate of rising particles at stage bounda:ry i MT-l
r,i
g gravity accelleration 1 M-Z
h(t) p:trticle transit time distribution T-l
h(v) p:trticle velocity time distribution rc1
H p:trticle OOld-up M

-157-
H' partiele hold-up in class:ifier sectien voltJie el.e!Ent
H(f) Fourier transfonn of transit t:ine distribution :function,
I impulse MLT-1
K force ~>.Urking on partiele M LT-2
buoyancy force MLT-2
force exerted by colliding jmticles MLT-2
drag force MLT-2
lift force M LT-2
Nonral force caused by pa:rticle weight MLT-2
2
friction force MLT
force on jmticles exerted by wa1l MLT-2
length of mxlel particle. L
lengt:h of classifier channel sectien L
L' characteristic lengt:h channel sectien L
elistance beo.een ti>.Q detectors L
mean number of transitions of t:he botton product jl3I'ticles
m IIBSS of jl3I'ticle M
p
mt mean number of transitions of t:he top product particles
ncon ratio of mean t:hickness of conglaiEates and particles to
t:hickness of
Ntr number of tracer fed during an exper:ilrent
N ratio of mean area of conglarerates and particles to area
con
of partiele
prol:xibility of jl3I'ticle to rise to next higher stage
pf mean value of pf averagedover t:he stage boundaries,
pf ,i probability of partiele at stage boundary i to rise to
next higher stage
pf."() value of pf at 4>R=O.S
Pi jl3I'ticle rising probability at stage i
pr probability of rising jl3I'ticle to rise to next higher stage
boundary
pr mean value of pt averagedover t:he stage botmdaries.
pr ,i prol:xibility of rising jl3I'ticle at stage boundary i to rise to
next higher stage boundary
pr."() value of at .PR=O.S
pv rising prol:xibility of pa:rticle t:hat enters classifier in feed

-158-
Pll,i proOObility of ~e to enter and lmve section in
falling stream
P ,i probability of partiele toenter section in falling stream and
12
leave in rising stream.
P ,i probability of partiele toenter section in rising stream and
21
leave in falling stream.
P . probability of partiele to enter and leave section in a
22 ,1
rising stream
~ partiele feed rate.
~ air flow rate
R IlU!lÎJer of classifier stages
Re Reynolds nunber for airflow around partiele
p
Reeh Reynolds IlU!lÎJer for airflow in channel
Rfw ratio of llE6Sl1red width to cal.culated width of Ef curve
Rfm ratio of llE6Sl1red llBXinun value to calculated nmd..mum value
of Ef curve
t tinE T
t transit tinE of partiele covering distance Ld T
p
t rrean transit tinE of partiele covering distance Ld T
p
t
pn nalian of transit tinE distribution function T
t .
ffi1n
min:illun tinE neceassary for a partiele to achieve a transition T
t' transit titre related to rrean partiele velocity v T
pn p
tpf durarionof pftransition T
tl-pf duration of 1-pf transition T
t durarion of p transition T
pr r
tl-pr duration of 1-pr transition T
V location of feed stage
V volme of classifier section e1arent.
x(t) signal of detector x
x'(t) fluctuating part of signal x
~ nmn value of signal x
x fraction of partiele surfaces form that overlap
y( t) signal of detector y
y fraction of particles that form c.ongl..cmmltes
V IlU!lÎJer of feed stage location
V volUie of section elarent

-159-
vf superfic:ial air velocity
vfl partiele velocity defined by eq, 2.9
vf:D superfic:ial air velocity at which <PR=0.5
vfx superfic:ial air velocity at which <l>R=x/100

v local 1.i.nallr air velocity in cl.assifier channel. L 'r1

1
v partiele velocity L
p
vp nam partiele velocity LT-1
vpn partiele velocity defined by 1/tp L T-1
vpf,i velocity of falling partiele at stage OOtmdary i LT-l
L T-l
v . velocity of rising partiele at stage boundary i
pr,1 L T-l
v slip velocity of partiele relative to air flow
s
W cl.assifier channel width L
wi, b CUIIJ.llative YA;dght of canponent i in bottan product M
wi,f ctm.!lative .m.ght of cernpanent i in feed. M
wi,t cunrulative .m.ght of canponent i in top product M

Gr:eek sphlls
a ang1e of attack of air flow at partiele
f3 cl.assifier angle
o dirac function
IJ. Ef width of curve at Ef=0.5
!:!. Ef,c width of calculated Ef curve at Ef"().5
!:!. y heigth' of voltDIE elarent in cl.assifier section
Tlr separation sharpness
1
nr
single stage efficiency (eq. 2.6)
Tlf 1 single stage efficiency (eq. 2.6)
<l> xx auto correlation function of signal x
<P xy cross correlation function of signals x and y
<l> Foorter transfonn of auto correlation function
XX
<l> Foorter transfonn of cross correlation function
xy
<l>R separation function of zigzag air classifier
separation function of canponent i.
density of partiele
density of air
t:.i.tre shift in correlation calculation
nam partiele residence t:iJre

-160-
1" loeà.tion of rrmdnun of cross correlation ftmction
m
1"jJ value of 'C at vfjJ T
ovar standard deviation of of variable

-161-
Appendix 3.1.

Electrooie circuits for detector signa! conversion.

Flectronic circuits \\ere designed to convert amlogue detector signals into binary
signals that could be S8fi1lled by the canputer'. Fach detector (Iflototransistor BPX 99) was
part of a circuit as shown in fig. A3.l.a. An electric current flows Which is proiXJrtional
to the intensity of the incident light when a voltage Vcc is applied over the emitter"
and the collector. This current is tranS!Ii.tted into a detector voltage Udet by rrmns of
aresistor R (2.2k). A voltage C<llljmator (IM 339 quad. volt. canp.) cai1jm'eS Udet toa
reference voltage uref (ree fig. A3.l.b). The value of uref is adjusted by rrmns of a
potentiareter" RIXJt' Uref is chosen in such a way that Uref > Udet when only black
particles or no particles are detected. The C<llljmator produces a ''high" voltage Ucan.
udet exceeds uref when a white tracer is detected. ucan is "low'' then.
Groups of detectors that scan the S3lre partiele stream \\ere l:i.nked tagether by so called
NAND-gates and CR gates. input signals U \\ere canb:ined by on= NAND-gate. The
can
output of this gate, Unand was high when one or trore input signals had a 1ow value. CR
gates canbined three output signals Unand. and produced a lûgh output when on= or trore

inputs \\ere high. The 24 signals caning fran a single array of detectors \\ere canbined
into one binary signal. The value of this signal, U was high when a tracer partiele
array
was detected. In other cases this value is low. 16ofthese signals could be sampled by
the eauputers digital input trodule. Fig. A3.l.c sl1ows a block diagram of the signal
processing by above electrooie drcu:its.

Appendix 3.2.

The computer triggering circuit.

This circuit produces a short pu1se signal Utrigger (pulse length 0.5 ms) when one of
the detector arrays detects a tracer particle. For this purpose, the output signals of the
detector arrays, U , are converted into negative goi.ng pu1ses with a pulselength of
array
0.5 ms by rrmns of so- called llOOOStable llllltivibrators. Eight of these pulses are inputs
of a NAND gate again. The NAND gate produces a IXJsitive going pu1se when a pu1se occurs on
one of the inputs. The output signals of two NAND gates are canb:ined into one by rrmns of
an CR gate to form one trigger signal U . that goes to the caqmt:er. Fig. A3.2
trtgger
gives a block diagram of the trigger circuit.

-162-
 Figure A3.1.
Electronic cira.Jitry for signa] proc.essing.
a) The el.ectronic cira.Jit to produce a voltage Udet which is proportional to the
incidffit radiation, E
e
b) COnversion of the analog detector voltage ude toa binary OUtplt signa] u •
t can
c) Block diagram of the roliJination of the detecor Signals to a single array signal;

incident
light

a)

,_.I b) ucom
0\
w
I

=
BPI 99 LH 339
c) 1 1

Figure A3.2.
I I Block diagram of the trigger cira.Jit.
I I

~
111'1 HH: monostabie multivibrator
uanay

Figure A3.3.
The sampling programre.
Appendix 3.3.
Signal sampling programme.

The prog:r:amre ''SAMIL" coordinates the sarrq:Uing of detector data and storing on floppy
disk, A block diagram of this progranne is presenteel in fig. A3.3
Up to 16 binary input signals llBY he sampled by t:he digital input rrndule (DI) of t:he
canputer. This rrndule converts t:he input signals into a 16 bit digit in t:he following l>ay:

I ••• I are the values of t:he input bits. The value of I is 1 if the voltage of the
1 16
corresponding signal U ...U is high ( > 2.3 V). I is 0 if this is not t:he
array, 1 array, 16
case (U < 2.'!N). The canputer keeps track of t:he t:ine by rreans of t:he rrumber of
array
overflows of the hardl\8re canputer clock that have occurred since the prograill!E has hem
started. The overflow rate is IOJ 1/s.
The canputer programre checks the value of the Sdmidt trigger input of the canputer (Sf).
As soon as the canputer detects the negative going edge of a pu1se on t:he sr, the t:ine is
registered by the number of clock overflows. The DI rrndule at the sarre ti.tre samples the 16
input signals and converts than into an integer according to equation A3.1 • futh the
signal sample and the t:ine are t~y stored into buffer llBIDry locations. The
contents of these buffers are stored on floppy disk in data file "DATA" as soon as the
required nurnber of samples have been taken.

AppenJ:ix 3.4.

Pruguwll'S :fuc data processing ("MAm:X" and "AKAR3'').

-164-
Programme MATRIX

Start of prcgramne MATRIX

( PROGRM ftf\fA"tx)

".,.~too tE, 2. a, .•~Ron 121' .un<" ..... 1110€<<1" ·1

I !TRI!:ff~t2.2~
_o&!t:Alfl FILE01 ,,rtlUO't
11.11 l, TT0<8.2, 2l Ueelarations of arrays.

Auignment of values to array elements.

I DATA lftAtiS/484•01 I
,2 Hf"€ 5
FfJftf'lATt/' P"RO&fi:AfiftA H:Uirl:E" /I"' U v•rH ~ttt dP. ~'t:ll'ldUN11'\ih
Check of data file name
'lltn-t vu 4f bttu:~nsttrliHU'" en~ I"' vervtrk1Msparanett1''i."
11·· lht. u dt nu111 vu dt Gahftlt • U llantttn) 'SI
~AL~ 6ElSHtt'!i,FH.[If,61
CAU COIIICAYiflLUI,'.DAT',Fit[)
Each data file can consist of a maxi- of 40
TYPt 1(1 records. A record is devided into 64 pairs of
·o
". ...
FOR'I'il!TI !' IJlt 4Uvut nat 64 t-ignalaCIInst'f!rc.;
<11tahlir "~ ''~•01:
bes't.nt •St' two 16 blt word>. One bit represents the sample

... ACCEF'l 20,!SM

~='Oftl'lAffiJI
tiae and the other the values of the 16 input
signa Is.
I
IJP'(JUllttJT~L.ttAft(=:f [\.E,l fP("::'{)l!J' ~ACCtSS= tHPtCf',
1 kE CQJt&S1ZE'!!:12$ ,Uft=10 l
GO!G 62
,,•o Pff'E ~s
FUfrPh\Tl//" De hltnaan lS U~>t >:orre<i.. V~r ~f!,fW * The minimum transition times are assigned to the
array e lements TTO( i ,k ,1).
•qeqt>lltnos 11'1.
GOH) 2 ' TTO(i,k,l) is the minimum time of a transition
from stage i, stream k to stage i. stream 1.
•2 DO 70 (~1,8:
fl0fi.1.2J::-Q-,t
k/1 • 1: falling stream; k/1 • 2: rising stream.
TTOC 1.2.1 ):t.2
Tll){ !,1,1 t:l,
TTOtl,2.2l-:"1, The nuobers of the inputs that sample stream k
'0 COHTINUE
!10 20HI [:! ,e of stage i are assigned to IE(k,i).
lEO,lJ=t IE(k,i).O when na input samples that streàm.
IE12, I 1=17-I
JOit'i ,-IJtHUltE
1[(2,1) .. 0
Ass i gnment of start parameters for the fi rs t

I !ZO I•D<•l
HHI'I:l!
THtEO•O.
'I transition:
The first partitie enters from the feed stage
(i•ll) in a falling stream (k•l). These index
values of array ITRANS(k,l,i,j) are assigned

IICI!I••
lqlfD r~'l)
I

RUtt t2'1t tltRit'A't({),l=1,i2J)f

lF (AR'RAY(t) .GE. 327lt&.t ttl:HT=Q
I to indk and lndi.
Reading of first data pair fr<>m floppy disk.
The value of the l6th input signal is assigned to
I ICHB!T. The nlue of this Input changes when a new
f!tttl!2'151\fll) partiele is fed.
R(A(li2'1SAftl tARRU! 1 I • 1=1 .t28J
li'J'SlQfl'~!
tJ tlt--AY11'2"?l .GE. 32768.1 tHSfOP:9
iiO lOOO t'=tSu+ • .·1
tr <! .!0. !SA.. &010 2900
fiJtH2"ll
ftEA0f2'U tAitlêYtU.t:"l, t21H Detections of the last tracer that was fed are
;t~(f DO 2950 t::t27,1, ... 2
US:Hlh1 not processed. Tl!e comouter determines the record
lf fAitlfAYUO .GE. 1 ••!!TOP'O nna. IBS and sample no. !SS 'of the first doteetion of
!F Ol:WSTIJP .EO. llf$l0f'~ GtHO 29,:Q
lBS~ I this particle.
rss=•K+-1!12
H t!< .•E. L?i! GO!O lOlO
US:Jt1
!"!!S=I
SOlO l010
2"0 tONTtNUE
ltOO COiftHU.IE

I
r:-
llext progra- section is torried out for each
!GIO t*'C no 1<:1, !SA"
I of the ISAH data records.
I 1

-165-
 2

Reading of one data record.

DO l6() T•l,t'Zï.Z Calculation of lnout signal values and of

~51 JO 34"0 tfEt.•IS,U.-1
HV::?.••ITH doteetion time. Results are stored In arrays
!J> tURAtlll .iiE. W.ll t10Til .UO ISIT(I,i) ... lBIT(l6,i) and T!ME(l)
UJH JTEL•l. i t+l )/:?1 :1
GOTO HO
'30 UJHTTEL-tl,lJt-1}1'3)"'0
AR-RAY ( l 111\fl()ll:iAJIRlYC H .H'H
Ho cantHut ·
T1llf( Ot1l f2)dRitAYl 1+1 )/100.
:i6o0 COifTflfUE

tSf!)F:n The value of !STOP is made 0 if tlte last sarnp1e

U: (K,IH.US ~llft, ti':,U.HtS ,AI'IO, t.GL!SSq ISTOih:O has been processed,

ISUI'I=-0 lletermlnatlon of the sum of the values of the

!iO l:'O !1'<<1,8
n· tiE<!.lt 1 .nt. o• :stnt=l$1Jft+fitltH:(t,t1J,!I detector si()n&1S 1 ISUM.
!f Hf<',t'J .l'fE. (I) 16t.ll'b1~1Jft+ltlJ(lEt2,11),ll
I?>J f_{i+lffti!Jf

n
!SUM • 2.
The computer checks wether both falling and rlslng
11J(l fJQ 705 !1~1,8
detectors at one stage deteet partiele.
IF IJ(\!,{! I ,l'i€. I) ,ANI), !f!JlHEi1,ttl,P ,Ëtl, 1
.Alf!). 1E(1, !11 ,Jif. 0 ,1\i!fD, The detection of tlte falling partiele Is sklpped if
!81Tt!F!2.J 1 1,!1 .€0. q 60TO '07 this is the case. lf not, both detections are
SOlO 705
''l? l8lTtlEO,lUrU=O sklpped.
lS'Ht=l
1ftc;; •:OH'f11l!)(

n The sample is sklpped if ISUH is not equal to I

Deeision upon new particle/no new particle.

DO 400 11=1,8 No new particle.

[F q1€l1,!11 ,HE. 0)
.AifD, tUtTll(fl,lll,ll .€J), 111 ~t)Ttl190
!f !lE13,l11 .ttE. OJ
.AH[J, ll8t1f!E!2.!1l.l! .EO. 1,1 ) 6010 jO()
SOTQ 400
··~··· !Hltt ,.t
Oetermination of stage boundary no. and partiele
llf!i:J=f1+1
fi.OHt 405 stream of detectlon.
[~(! .'""1 1+1
&Oli) 4<)'5
<10<) I:ONT]!OI!j[

-166-
2 3
Detennination of ti!lle difference OT between last
two detectlons

TtttU~TUtEU)
lltAifS<IHDK • IJDL, JNitl, lMDJ) = Oetermination of a regular transition
I (JR-115tliii,.Illtl.liiBI,U•DJ)+1
The transition is registered In lTIWIS
~50 l•t~c:O=Ut.l(
IHD!O•IHDI
UtK=lffBL
!Nti=INDJ
GOTO <80 indk•l: both detections come from same rlslng
partlcle; znd detection is skipped.
ind1=2 and indk=l: detections come from same
tlRAHSt UttK(I.IHDK • lWDtO, Ilf'Ol l•
1 lTR:AttStll!lKO. H..Dtt.lNIJtO, IHIH l•1 rising particle; lst doteetion Is skipped,
ITUaStUJIKO. HtDl. IfiiiDlO. INOJ I• hst detel"1llined transition !TIWIS is
i !JUIIISHIHC.tltl,tff\HO.INOJH1
llfDt::-=fttlil corrected.
Ttttt:O=TUI€t U
indk•l and indl•l: if OT Is larger than
GOT048~
TTO(i·l,l,2) then one of the detections
COllieS from rising stream. The 2nd

doteetion is skipped.

Assign~~~ent
of start parameters indk and lndi of
y
next transition to be determined.

Exit of partiele through bottom.

Registration of partiele transition to bottom

I lYRAH~f!HfJt:,lND! ,tH(lf,!liOJJ'= exit in !TRAIIS.
llftAHSf{H[IX:, HHit,.lHOf ,1HI'Ul+!

Check for end of processing.

Assigrwent of start parameters indk and indi

of next transition.

DO -6Q fP'"l,3
lf U!€n.tn .ffr.. 1H .oUfl). Oetennination of partiele stream and stage
tiBITntn.Ht,tl .eo* tn &oro 440 boundary of next detection.
IF tl!((2.!l) ~"E· QJ .A8fl.
HB!T<1Et2,11J,!i .til. tn ~)Q!tl '!")O
GOTO <oO
..o IIJIL•l
I•8:J•tHt
6010 •?O
}ft!)t:2
lltDJ-=tht
GOTO "O
••• co•ruuE

Registration of partiele transition from feed

point in !TRANS

fJtiiP.O•flf(Jtt Ass ig-nt of start parameters

!ff!HO:!tfCH
!ffD:k=htfll_
!Hfll-=tHlJ
Tl"€0~fittEtP

2 3

-167-
2 3 Exit of partiele through top

I 1420 r•DL•2
lftO'J=IO
I
r n•••s<r•e•.uoL.r ..n,noJl• J Registtatlon of partiele transition thro~
t lTRAWSCINJIC.ItUit~UIDl,!f8JH1 top exit in !TRANS.

r U' t JSTOP .EO. 0) 6010 -\f~

I Check for end of processing.

r H?:? IitfJK!tl
UUH=l1
I
DO 1460 {1'=1,8 Oetermination of partiele stream and stage
!f ((lt:{l,l1J .te-E. 0) ,AHI!,
1
ilt!TOE!t,!1l,tJ ~EQ. lH GOTO 1440 boundary of next detection.
lF Htt<2,!1J .HL OI ,IUUI,
l 1
1BlifiE(2,l11,1) .EO. tll GOrt} 141'$0
66l0 1460
tU(J IJfDl~l
{WDJ=t1+1
60TO U70

·~·
lNJl'l112
!lfiJjiiJl•!
GOTO f47(l
'460 COHTIIfL'(

I 1470 n•••s•r•><.r•oL.r•DI.l'OJ>•
'flltAtiSt!;t(IK.!tt.Dl,ltfO!.tff!l ..:l+,
l
Registration of partiele transition from feed
peint in !TRANS

I i>:B:!"hliHn&,t! I
ttt&~0"'hH!K
IaUQ,..lHII Assignment of start parameters indk and indi
I1fDK=IJfDi.. of next transition.
llf{lt=t""t!.l
TI1!:(0-=l!f'IE:if)

y
~

CALt CllJtUfiFltU.·· ...u·· .FilE) Storage of matrix elements ITRANS(k,l,l,j) on

OPEifH;I!I fTII!,I'fU€ r~f tLE. TYPE:::F'lffV., ,ACCESS=- !HRE(l' ,tw.COtDS:UEt:.242
J .~;.t~tC•' J
floppy disk memory.
UUTE (J" 1) ~ !1 1 I Tl!iHfS.t;, .L, I.Jl ,L::t, 2J ,«'"1,2}. J-=1, t 11,1"-~1 ,11 l
f.!..USE HJft1f=3J

-168-
 Programme AKANS
Start of progr......
( PRO&RAI't AtcA-MStJ

Deelaratien of arrays.

Assignment of 0 values to array elements

F'R:HH 1001 I
'tOl FOIU\i\ T( t· ~--~------··--·-· .... _ ..... .,_ ___ ~-·-PROGRAftt OKAJtSJ,.FQI~ ..... Cheok M data file""""'· Opening of data file.
t ~---~----·----------------------"'I)
9tf ffP'E tOOO
!000 FOihttHU' PRC&RAfU\A: OKAffS3~ I
The data file contoins the matrix ITRANS(k,l,i,j)
~· llat n dt 1\UA vu dt datahlt 1 <~u. 4 untttrsH ~., that was determine by 0rogr....., MATRIX
CAu. 6ETST!tE5~FïLU.6i
CIILl COifCAHF!lEif.~ .MH~ ,FILEJ
P"fl'hiT t002.H'!LEHCil.I=J.6)
100'2 f:'O"UATl~ Ftln:u•~ ·,ólfll)
OPEtti UllltT:l.ttAtt(=-fllE:. TYPE•'Oll' .ACcu;;;~~ Uli'€Cf'
lflECCRDSllE1!242.PIAUEt-=1.EIHI'=I01')
GOTO 1030
1011 îYP€ 1020
'020 FORIIATI/1' Deze tllelHI~J• U n:at·<taftWt2'1~. 1
ilt~tn 001\leuw.· l
GOTO 999
REA (I <3~ ll (CIC tTIAit$0(,L.1, J) ,L•I,2 ~.K~l ,:::; ,J;tl .1 J), J;t,
CLOS€ tUJHh31
tt. Reading of data file

JlPE: 13 Introduetion of parameters of tbe stop

'3 :-o'fiMTit ht 1., de t~ixtUil to~~t"ita•e atbrll-.ktout tl' •1•
·tt~o;HuHI vu~r dr overf!!an,$k~nsen 'to.o-t.Oh ~~~
criterion of the iteration process to caleulate
ACU:PT 1~,i[Lf' higher order transition probabilities.
Hf'E 14
•..: '"0kl'l~tu·
uat a dt Ai)lu•nl toel-,tn\31'tt aft:>t~tl':faut 111 M
'br:>a'.llH nn ·r dil' o11l'trttnk:atlstn t CO.O•I.Ol1 ·••
ACCEP'T 1~.ttELEf
11) '0it!AHF?.51
!f tRELP .U. 0 .Oii. UUf .EO. 01 GOH) J
PRHH 9910, <iEV .RHH'
vo10 Fl)it!ATU" s~.npcrtbttu,.: {ttt"atHJn1'0'=f'iJ'I/ .-e•~t.l~l!~~"
tn;,uwtll'\U'tilhttd Ut Pf l!'n Pr: "f5.3' ~' '" Rtht.leve flt,tnal(lu!"1qtl@l•1
ttn €f u Er: •f5.l' •· 1 I
I
DetennînaUon of zero-order partiele transition
&0 110 J:-~,9
&0 100 J:-1 ,10 probabi l i ties.
10 •• 1(:1 .. 2
Particles are considered here to cover the
DO 80 L=!,.:?
11 transition !TRAMS aleng the shortest trajectory.

tF (l .lfE. h .(liL K .•L ~1 6CTO t6 Alldition of partiele transitlens from the feed
!f'(CJ .EQ. U .A*D. ll .EO. Hl point to the transitions within the classifier
,, t.(UL tJ .GT. 6ll lTU.NStte~L.t,J)~fTRAI'fSCK.L.l.JHilRAN$0.L,Il,J}
tF 11 .filE. 5 .OR. fl .riE. n 60l0 t1 channel.
~~~ ((J .EO. $) .u&. tl .U. 211 Only tllose perts of tlre transitions beyond staqe
t,l)k, IJ .\.f. 511 IUUSC.,Ltl,Jl=tTRAMSH(.t,l.JH1TRAISt1,1.,11,J)
4, falling stream and stage 5, rising stream are
are added.
Assigrunent of ·the nU!Iéer of transitlens trom stage

·~>:n:-----,
i, strum k to stage j, stream 1 to variable FL
Analysis of partiele transitions at a single
stage.
TRO Ck ,3 .. K. J) ;TR(l (I< ,J .. K, J hfL
E f l~L ,L, J) =E <3·L ,l, J •+FL The nulliber of one step transltions that make part
lf <K .OE. ll liOTO 80 of !TRANS is determined and registered in TRO.
UOC 3-tt ,t., Jt ::liUH J•K .L~J) +H
.tEtf',3-t~, ,1 l :X:Et«.l-+1.1 HF"l The number of stage boundaries that are passed
GOTO 80 undeteeted is registered in array E, whereas the
reghtered passings are registered in XE.
2

-169-
2 4

,, [10 30 HEl•t,J~t
Analysis of partiele transitloos from a ]QW;lr
!kOl 2, 2 ~ l TEL I= TRO{ 2, 2, I!El )+F'l stage to a higher stage.
,, tf HIHL+fl .LT.Ji tE:f2~2.ITEU=X[(2,2~1TEU+fL
COfiT tNlJE
rr (ii:.E0.2l litHO 40
fftOt 1,2,! l=TRO€ 1.1, I Hl'l The nUlliber of one step transitions that make part
~€11 ,2,1 l=:l:E(1 ,2,!l+~t
of !TRANS is dete!'lllined and registered in TRO.
tF' il .Ea. 21 -GOTO t5
" AE\:!",2.J·1 )::X((2,2,J-! :+ft
'kOi ?.t,J)::_TQ012, 1, JH-H
The number of stage boundaries that are passed
• t 2,! .J'::tc?:,t. JhfL
undêtected is registered in array E, whet"eas the
IJIHQ -80 registered passings are registered in XE.
~ 1 ~. :? • .J·l )1\'E !2.2,J-t 1 +-n
fiOTO 80

DO ~0 !it:L=I.J+1,-t
f!HI\ 1 , i ,ITEUo;Tfti)11,1,ITHH~!.
Arialysis of partiele transitions from a higher
!f l!lfl·~ .lH~ J\ XF.:I!,1,!TtU.,.1Et!,l,{f(tl+flr stage to a 1-r stage,
•• ~OHfJI<I\1(
H \~ .tQ. 1160'!0 70
UOi:?,!. l}:T~012, I .I l+fl The number of one step transitions that make part
tft2. f,l }"'JEt2,l,l t+FL
!f t:l .EO. ll GOTO 7~ of !TRANS is determined and registered in TRO.
Xft l,l,J+1 '::XE~ l, t, J+l )+fl The number of stagé boundaries that are passed
rROf 1 ,.?.JI:f!tOt 1, 2, J )•Fl
€1 ~ ,2,J111l((1,?,JH,F"l undetected is registered în array E, 'tf'htt~as the
GO ro so registered passings are registered in XE.
Et l, l,Jt! l'!'E! 1 ,1,H11+fi

y CONTINUS:
COttTifllJt

Arialysis of particles that enter from the feed

fPVO=O.O
!XPVO:::O.O point.
{10 HU l=1.2 tlumber of deseending particles: TXPVO
DO H12 J::1,~
TU~'~JO-=TXPVO+F'LOAH 1 fRAnSr t ,.(. ,11,.}1) number of rising partieles: TPVO
C01i11 HtJ€ Number of particles detected at stoge 4, falling
bO 1{}l J"'t~,lO
lPV~-:TPIJO+F'lOAT <! tR:4HS! 1, t, 11 ,J I) stream: EPVX;
'IJ COIH!HUE Number of undetected partieles: XEPVX
'0' CllrtH•uE,
(ft\!~rrt.l)AH !T9'AtHÎ t t .~ ,11, l b' tlu!OOer of particles detected at stage 5, rlsing
E+''H:"'fli.iAfl ftfii:AifS< 1, t, ~ 1 .~l l
stream: EPV
.([ö>IJ"' TF'VQ-flOAT!l fftAW$( t .2.11.6 t)
t(PV:(;fXPVO·ft.Ofi:Tt! lRMSi 1,1, t 1.5-l) t;umber of undetected particles: XEPV,

fCOlJMl , 0 Calculation of the partiele transition rro·

lf ! tff'VO+Hf'\10: .b!. ~.OI babil ities and doteetion probabil i ties of
P'.'=t~•JO: \ !PIJ()+TtP'.'iJ!
u:· ! r(P'.1 •Xf9'1.') ~GJ, Q,Ql particles that enter the channel.
EHtl"'tP'I/ ((PIJ+Xff'IJ 1
~f ll~P\!J+.(H'J!I:I ,ljf, 1.•·1
r rpvx "'ff'IJ:! i €P 1Jh X€."''·'' 1
flll ! lO ~ "2. 9 Calculation of the partiele transition pro-
!~<I .U. ~i f-(1,2.11"'1,0
!f!ITRt)(1,2.1Ji>U:OO,~.IH ,6T. 0.01
babil i ties and doteetion probabil i ties at the
P 1' , ~ ,1) =TflO 1 1. ~. 0 t different stages.
I HU)l ~,:.I lfHI:Qt t ,1, f!)
1~ '! ,J.!, b: P!';:, 1 ,! ):;1,()
!''"·~;~•';'.•,! ~•U'()I~,!.!H .5l. ~.0:1
y, .:;. •.! i"'H!C1 :•, 1 I 1.'
' U'•}' ~. • ! l. HtO I:.:~ J j ;
"'!'.! ,(!"'l,•Pt l, , !!
~(J • .?.!l"'' ... Pi], .n

-170-
 €fl t ,1, !+1 !::.0.0
l~ (!E.P,',!+-ll+Xfll,l,!+1H .GT. 0,01
tF 11, t .I t 1 J::.[i 1, 1 , l +l ) I tE ll , ~,I+ 1 ~-+X"F. t ', 1 ,1 +1 } )
n't2.t.Jl-::O.O
'F CtE(2,1,11+t:EtZ,t,ll) .GT. 0,0)
H ( 2. 1 • I! :ft:~.'.! i I (u':!. 1 .1 'HE ( 1. I • I) I
Hl1,2.!)..-o,o
ft Uft1.J,l1Uf.'! .,~.IJl ,ST, ~.f'lt
E_F(! ~~·! !1:(( l ,/.,:I! t{1l,/, j l+l~\ 1,,t,J! I
trn.~.f-1 ~~o.o
lf t!Et2,2.!-11+KE!~.~.I-1ll ,6f. 0.~)
z.
Ht1.2,1-11-=Et?.2.l·t )/~Et 2.l-11UE1 :!' ,2 ,1~1 n
con••uE Calculation of higher order transition
probabil ities.
DO n9 T-=l,tO
TA'Of t ,~. t p:O, Start of iteration proceu.
TR:O( t • t ,I )!\0.
TltOt2,1,t):O.
TR'Ot2.~.u:o.
Et 1 ~ t. J)r;(i,
lE(t ,t ,I):O.
El 1.2',1,=0.
lElt ,?.1 ,,.. ••
Et2,:,0-=~. Assig-nt of 0 va lues to array elements to be
Xt!2,l,IJ•O. useo in itaration process.
[(2.2,!)•0.
J:€<2.1.! )~~>0,
co•ruuE
EPVT(J,O
€PVhl),l)
X.EPI/=0.0
([Pl'(Ti),0

D-Q 430 1=2,9

ttll 6:!0 J•l~tO
80 410 f::'li .~
D0600t.o:!,2

I FL:FUJAHPUifStK .l.t.J) 1 I Assigmnent of the nUlliber of transHions from stage

i. stream k to stage j, stream 1 to variablo fl

I lF !H .n. o. · &oTo •oo I

Analysis of partiele transitions at a single
stage.

I i'*t..;; I J~L,l..,.J)
!,.lc\)tri'IK,~w;., Calculation of the ratio's of the different
!f tY. .EO. I.' CICS•C~Sfll.·l-t:'Ut.l ... K.l I l:tPtl-L!.. I'
rr lCRS .EO, 0.01 CRS".I.ö transitions.
;;!!trPil ,3-r..ll•ll ,-EF<k~ l•t: ,!I i*Pil-Y. ,K,! I' CRS: ratio of particles !TRANS that cover the
• '· -EFtl-K.r..n ,.,.(t:,l·K, 1I•EFU-L.L. I,
Ir 1 V. ~EO. ll t!f:C!!•O~-€ffiC.3-K,lli•P(3-l,l,ll transition along the shortest way.
CllP=O~
CliP: ratio that carries out two extra one step
H !l ~lL?l 60fO '~0 trans i ti ons above stage L
lr!!'l"'LO C!IM: ratio that carries out two extra one-step
ff u: *EQ. :n Cll!t:Cil!t•,t2,t,[Jtlt.-(tC!.!~ll)
!:! !Jt:C:tlft•Pt 1.', Hti l, .. (f(1 ,t oll HP! I ,2. t-1 1·~( 1.-EFH ,1, 1-t)) transitlens below stage i.
tP(2,2,Î-11 Cl!: ·~it1o that ~arries out two extra one•stap
IC 't ~EO. 2t C!!ft'~~tl!M•(ft2~2~l-1)
lf l ;". .El). n tliii•Cltrt•U.-EH2,2, !-U td't2~ 1, tl'tEFt;,1. H transitlans at stage i.
••• :~ • I .sr. •J fHlTC 11J0
CUP=!.
!f lk.EQ,II l;l[fl':::ftU,2,!h0,-Ef0,2,Ilt
-: :f-.,C!!P'tP< l.:!, I Htl,-€Fl2,2. tJ HPt2,t, t+t ):ttl.•(Ft2, 1 1 l+t n
•Ptttt,J+lt
H :.l .U. 11 CttP'•CttPtEFO,t,l•ll
H {' .EO. !I l'!tiP'•CltP'tU,-Efn,l.J+tJJ•PU,2,lJtEF0,2,l)
!lU ,0\./llfflii:CRS+ClittlflltC~tP
IF ICOo•T .EO. t.Ol GOTO &00

2
I 4

-171-
 2 I 1 4

!RO Ik ,3-K, I l =TRO ( K, J·K, I I+ (·CRSICOUIH I•FL Calculation of:

Et 3-l,!.. ,Jl~E I 3-L,l,J l+tCRStCOUtH ltfl
TROt r. ,3-K,II=Tii:OIK ,J-K,l J+t Cli/COUHT l•FL
t:he single step transitioos 1RO(k,l,j),
J(IK. J-1(_ ,1 :=HIK ,3-r..ll+l Cli/COUHT ):e:fL the nunber of detections E(k,l,j),
TRO\ 3-K,I'. .1 l=lli:Of 3-K ,K, I l+ICI I/COUIH ltFL the nunber of mi.ssed particles XE(k,l,j).
tE 1 .1-K,IC ,1 l"H I 3-tt ,r., I l+( Cl I/COUHT l'~~FL
TRO( V. ,3-K, I l =TROl K, 3-K, I l+CCI I/COUfH l•FL
E<J-L ,L,J l=E f 3-L,l,J) +\Cl 1/COUNT Jtfl
TRtH 2 ,2, I t=TRO! ~, 2, I l+ICliP/COUJHI•fl.
~E f 2, 2,! l=XE !2, 2, P+ tCJIP!COUIIIT hFL
'ROl 2,1, !+1 I=TROI 2, I. I+! l+ !Cl IP/COUNTI•FL
XE I 2,!, !+1 l::):[t2 ,1, I+li+(CIIP/COUftT >tfl
TROl t ,1,1+1 I=!ROII,I, I+! l+ICIJPICOIJNTI'tFL
TROl 1, 1 .lt=TROC 1,1, t I+ICII~ICOUNTIHL
X[l!,1,fi=Htl,!,II+(C!!PI/COIJ1Hl:t:FI_
TROl 1, ~,I-!) =TROl I, 2.1-1 I+ICI HI/COUHT)tFl
~E' 1 , 2, l-1 l: Xt I I , 2, I- 1 1t ( CI I 1!/COUifT I :tf!.
;kQ: 2.2, !·1 !=TROt2,2, 1-1 JtfCIIft!COUNTI'tfl.
If o: .HE. ll GOTO 185
X~' r, 3-K, I I =~E (K ,3-K ,! •+I r:lf'S/CIJUHT 1'~'
~~l!· ~-r.,K,! J:fRO~J-r.,•,,i ltiCRSn.:IJIJPITIHI
:eEt r ,l-K ,I t=HIK, 3-r. 1 1,' er I tCOUJH l'fFl
TRO!J-1( ,t:, I! =!RO ( 3-K ,t.• I! •<Cl IICOIJII Tll'fl.
I 8S ff IK .l'fE. • ) GOTO 190
TRO\', 2,! 1 -=TR()•. 1 ,2,! •+(CI IPJCOUHfJ·tr!
(E•' ,2,1 J=Xf< ',2, _l+ 1 C! !PtCOt.:friT •tF"!.
GOTO 19'5
~o r RO ~ .'. • , ! '".r"o' 2, •. 1 1 +, c
I !11/CflUH! ~ tF•
xEt 2,1, I !'"lE' :. •,! :••Cf!t!/COIJHJ JtP
i" I' ,!'lf;,1 •6010 19:'
J;f·:'.2,l-1 P'X€1:',2,!-l l+C!IPI'CilUNTffl.
H0\2,1, I I=Hc012,1, I H·CI 111/COUNhF'•.

~:~:::i:~:!~;~:;:;~; ;:~~~;~~~~~r-,r•_
GOTO ~00
9~ Et ~.:.t-! '=( 1 1,~,1-1 hClt!!.lf.OUIHtF"L
IE( 1,1, 1+1 p:.((( 1.! .!+! hCll~ '(û!JNTtfl
TROl! ,2, I t =TROf •, 2, I l't(! !P/(!!UIIITf.:"l_
~!! ,:, ! 1-=E< 1,2,11+CIJP!C0UfHtfl
GOTt• 60û

~~-n----------~ Analysis of partiele transitions from a lower

stage to a higher stage.

CRS=•. CRS: ratio of particles !TRANS that cover the

OC 210 ITEL=l, J-2 transition along the shortest way.
CRS=CkS,P\:.!.2, ITI;:Utt 1,-Hf2,'.!,t;Eu 1 CIIM: ratio· that carries out two extra one step
'10 C.0"'flfrt!JE
·zo LRS .. =CftStP12",2,J-I) transitions below the feed point.
CPS=CR'SH CliP: ratio that carries out two extra one step
!FOt .EO. ll CRSzCRS•ti.-EFI1,2,lll"tP(1,2,I~
IF U. • EO •. 1 I (RS=CRS,!! .-H!2.:!,J-1) !t:P!2,',JI transitlans above the feed point.
•EFC!,I,Jl C(l ): ratio that carries out two extra one-step
•t .EO. 21 CRS=CRS•H<2,2,J-I I
~ F ( CRS • EO. 0. 0 l CI?S= I. 0
transit i ons at stage( i).
:! [" ~ IJ.
Cllf' = 0.
:r .EO. 2 .010:. ! .LË. 2! 61lTO 24~
•I(
'~\i C! !~t=~ll, 1,! Jtll .-EFt!, t, Il 1 tP( • ,"!,l-1 )1"1 ~ .-EFt 1,2 1 f-1 I l"f
~·1:•,2. I-~ Jt( 1,·E"t2,2,!-1l J•CRSH
.• :.. ,[IL IJ C1Ifll=(!:"f(1,-H(2,2,J-1ll•''l2,1,j)"ff'l2,t,JI
~~ 'l . fO- :-' CI Ht=C !lii•E" ( 2, 2, J-! 1
:•~ !F \L .EO. 2 ,QQ, J .SE. 91 GOTO 270
1: I [~=CWSH• 1 1.-Efl Z,2,J·1! ·•P• :!, 2, ~! t1 • .-~!="<?,2, J1 1
•f' I 2,! , J+ I ) '~ ! , -Ef ( 2,! , J+ I I l "fP t I, 1 , .J+! l *ff t 1, : , ,J+ 1 I
[~ \k .EO. 1' Cl !P=CI!P•P(! ,2,! )"ft 1,-Htl,2,! l I
.?70 ~TEL!: __
lF ! J .fl). !OI !TEL ,.,.9
IJ!) :eo !lEl=!,IHt 1
Ct i lEf. I :CQS• ~· ~ 2, I , :! t! I f I 1 • · fl-" \.?,' , ! fEL: l •
~~ !.2. 1 Htl'f!l,-Ht 1 ,2,1fEL~J

"i)IHIIT-CRS~·- l I •+ ~~!! ~­
)~~ ! T f.l-: i.! T f !_,
·. 0""1' ~~ <!"Hf+(' t T€ I •

4
2 6 5

-172-
2 G I l 5

lf <I .61. J-21 GOIO 300 Calculation of:

00 290 ITEl.:!,.J·2 tJ-e single step transitions m:J(k,l,j),
no ll92, !tELt :uot:?, 2 .nfl Hfl tJ-e <l~J!Ulr of detectims E(k,l,j),
X€t2.2,HEL J=lE! 2 ,2,1 f-ElHFl tJ-e nu<iler of missed particles XE(k,l,j).
COIH ffiUE
fk:Ot 2.2,J-I ):ft.0(2,1,J•1 HFl
lf <tOU•I .EO. 0.01 6018 640
!fOt ~U. 2> &ITO JfO
TftOC 1 9 2.Il.,Tt0!1 ,2,1H{C~ICOUHT )•fl
XE( t ,2,1 #=lEt 1 ,2,! J+tCi$/CilUIH•F't
IF tl .U. l l Gfl18 3t5
GUTO 317
l( t:! ,2 .J-1 ):J:[(2, 2,J-1 i+ (CJt$/COUHl JtF'l
fktH 2,1 ,Jl :fR(H2.1 ,Jl+tUSICOUiflltf'l
Ef2't t tJ1=EU.1,Jt•tCR$/COIHttl•FL
IGlO 120
€ t ~,::! ,J•l 1:E <2,:! ,J-1 HtCt$/COUHt lffl'L
IF <Cll• .(t. 0.1 &Offr llO
!R:Ot t ,I,! l=fROi 1 t 1 ,ltHC{IfVCOfJNT )tFl
XEt t, t .l•=XE~ l, 1 ,I HfC.Ilfi/COUttT )tFl
UOi t ,2. I-tl =TROC 1.2,1-1 HCCitft/Ct)UHT HF'l
x;E,, I ,2, f-1 ~=XEO ,2.!-1 )+(tilfi/COUifH•fl
HI'Ol 2,2,1-1 l =TR'O<:!, 2,l-1 HtCil"/CilUtfT H~l
~El~ I 2,1-1 J .. ~il 2,2, I-11 HClli!IICOHift l.rF!.
lF \l .tiJ. I' IJ~H!'.I 32"5
GllTO 32J
tE,2,2,J-I )"=lil 2, 2, ~~~ l HCtfft/Cl1tJf'fT Ht=l
f!t1>( 2,1 ,Jl =lROt 2, 1, J J-+! ~~I I11/COlHfT J •F1.
Ea.!, Jl1!:((2 ,1,J)+(t!l~!ClJUNThf'l
GOTUJJ•
E! 2.:?. 1-1 ):E:[ 2, 2 I J-1 IHC llt~ ''CI)U"fl ~ tkt
;:~ <CUP .Hl. e. J 6tl11l 14-G
J:t: • 2 • 2, J-1 J=X€ { 2 ~? • J-t Ht C! !~/C\J!JffT; tFJ.
1 iOi 2. 2, J J =UO ( 2• 2 • J I Ht I 1P ICOUIIIT ltfl
XF. t 2. 2, H=tt: t 2.2 ~ JHttt ltiCO\UH )*fl
HtUC!.t • .;•! l'=1t0f Z. t ~J'*1 HtCHPlCOtJNlhfï.
~tr?,: .JH I'=Xtl;?, I ,J+' J+tt!l~/COUHll'tR
"f'~\!.; ~J+t )"nih 1,1. J+l Htt;f !PICOIJ*l >•~l
·.' .J+t '""fl1~'. Jtl •tt(,.ftf'/i;I)UIHt•f•
; ~ '!': • f\}, .! I lJo\)! () .}41/
t~tv< 1.2, t; -net 1 ~2,l•HtiJPtCOtHH ,-.Ft
x: te;.~. f ,."~(!? ,;; ~ f)+(Cl[P!COtJtti !11Ft
t4L' !IQ 350 lTH =I, 1T(t.1
IF n":<IT(u .e.o. o. •
IJUil l!iQ
HtOf t,:•,IHLJ"=!R:Oc_~ ,2, tfEU+lC• 1 lt.l)/C0Ufff)*FI.
1Et; .z-,I TEl ,.,XE{ 1 ,;: .. HEl l+!Ct ITEL l/COUffT t•Ft.
TROCZ, f, 1 rn
J "'li0! :.0.1, !Hl tHCt !TEl• •C!J!.U+f)*Fl
tF.C!" ,1,1 r~:u=:cEt2, • ,lfH. J •<Ct lTS: I.! /C'llUH l1'F!.
!!-' n: .€0. h li-tHil H~
Tl(0( 1,2, f 1=1R:0( l,l,llt-ll:;c !H,l UCOttliiH<fl=!.
Hl 1 • 2, I! ='XfO ,2. I Jt'. C! I fEl 1 /CO!Jft!I·•FI.
;t5 !F fl .EO. 11 GOTOl46
GOTO .H8
H~ XE ( 2., 2 ,J-1 l =XE' 0, 2 ,.J~ ti+~'t. t tTF.:L I !CtlllNf )ffl
T&Ot 2. l ,J) =TRO( 2, 1 I j H~ Ct f TEli/(OUfilf !tfl
Et 2, 1 , Jl=E C2. t ~J I+ u; t t fEl• /COtufT ftft.
6010 l'O
~4i E ( 2,2 • J-1 J "(i 2~2,J .. 1 Hi Ct lTEI. I /COUfiff HF!.
!Sé CO•T l*U{ .
•Jt)ft} ó!lU

AnaÏy;ls of partiele transltions frooa ·ahiïiîüïr

I stage to a 1-r stage.
Pil UtS::-•,
~,'il t' :! , t. t Jtll GOfO lfO
!it• ~M fftL~I,J+~.-t Calculation of tlle ratio's of tlle different
(.!I'S.,.f.RS•f"t1,1wilf.Lit11.~~F<!.•~!!E ''
transitions.
•"N" ~~Joo
CRS: ratio of particles !TRANS that cover tlle
l( •· .~0. :!~ C~S=P"I';!.l,~l'(~, .. tFC:!,l,{!Jti~RS transition a long the shortest way.
:F ~~ .Hl • . . Crtll'"C"SY[fl1,1,.;+t)
CliP: ratio. that carrles out two extra one step
.~0. :;, CflS'"-lFS•it.-lF"lf~!,J+1)l<tP•l,~.Jt•€t•~.l.J'
i~ ''.I?S .EO. t,t,l)! CkS•I.O transltlons above stage i.
Cl H'~').

:F " ,l'lf. 2 .UR. ! .tiE. 9J 60HI 4t~

Cllll: ratio that carries out two extra one-step
':!lP.,.PI:?,;?,tit! ~ ,-(f:1;?,~.1 I )tH:!,1, !tl )'ff 1,-~F'! .!, •, ht ~ )!J transitlans below stage i.
•-, • ,: .!+11' r 1,-n:q ,1.1•,'
"cko; ..
Cll: ·rat1o·tmlt carrles out two extra one-step
'' " .Eil. -; • GO TU: ~ "''
trans\ ti ons at stage i.

<l> ' ' • 1~. " f l ' ' , • 'J •• {' ' . )•<óC.(!.!,!-l•t
• .. , ~. J. J-1 J-'

2 6

-173-
 H iJ .U. I; !TEl!"'~'
iJl) '*"1~ !if~::Utt..O,llflt
Ct fft:l H•Ck:S11f'\ I, 2,11EL H( I .-EfU .2,lfEU )•P12; l,llit.••
•< t .-HL2.t, ITEU)

ilf: •~O lTEl-=!,PH•

COlii'(T."COUHl,..C \ HH l
460 '>WT! lfUE.
!~ :COI;IHT .EO. 0.0-, GOHi 600

!f ! t .d. J+2J GOTO 480 4

2 6 :tD PO !f(L-:J,J+2p .. l
liOt 1, t ,!lEL );TRO( 1 ,1,! fEU+fL
Ut I• l, lrt:U=XEl 1,1, lTELi+H
410 COtlliNUE
Calculation of:
••o Hi:O( 1 1 f ,J+1 l>riRO\ l, l ,JH )+Fl
P' (I( ,(0, ! l GOT!l t90 the single step transitims m:J(k,l,j),
rff~t 2, ~,I l"~"lRiH 2,1, I HtCRSICOlHH ltFL t.he ntJnber of detections E(k,l,j),
l.t 1 Z, 1, f' ~XE! 2,1,! HiCII'SIC0UfiHTJ7r. t.he m.mber of mlssed JBii:icles XE(k,l,j).
'r tl... t:Q. 2 I GOT!! 4t5

1c~ t.t:;, f ,J•1lttXE~ 1, i, ..;tl1+ -~:ll"~iC{H!:"f! t•F-1

-:n;n l ,:; , .j )d>I(P 1 ,Z .J I ~tCFISiCÓt:lli tTF'l
~ · ·, .:: ,jl>'(( I ,2,.nHCRS/C0tHfY •t-F'I
tlûH' 500'
497 ~ 1 1 ",J*l,:[t1,1.Jt-1H\C~S/GOUNl;•F•.
~;oe- u 't:n~· .to. i),) GCHl 5to
"QI)\ ,';,;?,: ~1:'Ut1,2, I!+< C!!PICOU!IIfHft
-~i.'::,~ ,lt :XE( 2,2~ J htCt!P/COVliiO•t!.
fliïH ~. t ~ !f!
J:!JtOt 2,1, l+tlt\ C.IH'ICOIJ~T 1*fl
.~[\:; ~ .1•1 1•' i.!!F':(QUIH 1 tfl
,1 ,hl I=.XE(2,
UO:t t,1,l+l .' =l!{Ot
~, 1, !•1 Jtt.C n~·tt;Ot;ff! l·tl=l
Hl'.'. I•t ):JEII,I, t+1l+H'I fP!COU!f!:,H
H !t .EO. 2' Goro ~0~

~ t1. • ,J+: l«")J> 1,1, J+l 1 +t Cl !f'/CilUHHc~!

'RI::' .2,~J"''k()( l,~,Jh!Cl I~/COUfH itfL
f' l *2 ,J l :::( { l , :; , J) +\ C UF" /COUH'! ltF't
GOHl StO
1 ~.' ,JH 1:::(1! ,I,JH Hl("! lPII:OO•OtfC
JF tC!!I'! .EQ. ~.l IJOfll ~2\l
XE t 1 1 t ,Jtl Pl~ t 1, l ,Jt\ JHClii'I/CQt.IHll tft.
UOi 1.l,.JJ;:-TROt 1, • ,JIHClliltCOUHT ~:tfL
X(ll .t.J)=XEt I ,1, JH (C!HIICOtHtf l 'H
fiHH 1,!~J-T J: ik0 ( 1 ,)..J•I i +ti.l !fl;ÜHII'<I '''"i
(f,t t ,2,J-1 >=~E~ 1, 2, J-1 HtC:Il~~t/CQlJHT trFt
i it0{ 2, ~.J-1 1 =HfO (::! ,2, J~ 1 i+ tC l IIUCOOH i) tfL
tL2 • .3.J-1 !+tCllii/COUI!ITlfrl
!F ! K .!.0, ; J 520
üi 2,1,! .i=X€t2, 1,! H<CilA/COU;.T Jtf'l
r IIOt ::,1 ,1 p: HHH l. ~ ,1 l+(t!Jfll/COUfi r 1·•fl.
fll) 5.10 PEL1:!,ITEU
lR()( 1 ~l, I Ht,.Pfkl)( t ,2, t TE!. I+(U flfLUÇI}!jtfi)<tft
1.E ( 1 , 2• ITH î =.x En t2, ITEl H <Cl !f{.u tCOIJfflltFl
HtOC 2, 1, I TEL I =T~ót2,l, t TEU+lÇ (I TH l/(QUHfl•~t.
•E t 2, 1• ITELJ =XE C~,!, HEt. l+fC ( l l(L IIC:lUHT l 'fft
H cK .EO. 1 l &QTO 52~
Ti0( 2,! 1 1 ):::Tk()~2,1 1 I i HCi! ItLHCOUJHJ·,F'•.
XE11,1, I P:l.El 2,1,! l+! Cl HEU/CQUJHt'ffl.
!f 'l .EO. 21 GiHO ~;?o
GalG 527
lit1,! ,Jt-l ){:lE \1 1 t ,J+llHCJ. tH;:.uO)UNT i•F'l
ll!Oi l ,l,J;; fkil~ I, 2,J 1+1 Cl I HL l/CQUHT uo;l
E~i, 2,Jl"'t i!.? .JHt(;{ IlEUitO!JffT 1tFl
GOlD 530
El I 1 1 ,J+l ):.E\ 1 ~ l,J+l )•{(:( 11H 1/COIHI'l )1F'l
t{Ht 1 .w:IJE

-174-
 4
I l
Analysis of particles that enter from the feed
HHJ~o.o
lXP•J~O~O point.
EP'U~ii.~
Ef"\1 X~O~t'!
CRS: ratio of particles !TAAHS tllat cover the
J:Et\.'=0.0 transition along tlle shortest way.
XEPYX=O.O
CRS=P\.I'I'EfPit Cllll: ratio that carries out two extra one step
Cllii=C 1,wP\I}t:(1 ,-EfPVU•P< 1 ,2,SJ1tO,•(F(1 ,2t!SJ ttPC2t2,SI transitions below the feed point.
tEfC2,2,~l
IF «CRS+CII•I ,(0, 0.01 6010 628 CliP: ratio that carrie-s out two extra one step
transitions above the feed point.

I
TPV= TPII-+CR$/ €CRSH; llflll• Tf•VO
EPIJ-=EPV-+Ct$1 ( CttS+C 11ft) tlPVO
llPV•TIPV+C lU/ ( CRS+CilMJ•TPVO llumber of deseending partides: TXPVO
IEPVX•UPVHCI 1•1 !CRS+CII• >• 1PV0 number of rlsing particles: TPVO
TRO{ l ,295) =TBOt 1 , 2,5) +tt lMf(CtS+Cl Itt HlPVO
NU!d>er of particles detected at stage 4. fa11ing
:l[ll ,.?,5t='~Et t .:t,~HCtlM/(CRS+ClH•)UPVO
TRO! 2,2 ,5 )l'Tft0f2 ,2,5 HC t UV(tft$+Ct lft)'tfP'IO stream: EPVX;
Et 2,2 ,St =Ef2, 2,5J•tf Ut/tCRS-+t:l UO •ttvo
1& US=( 1.~PVl•EFPVX
Number of undetected particles: XEPVX
C!lf"=PV•t 1. -E.FP'J l*P( 2,.1.6ht 1. -E:F t:? • 1 .61 J tf>( 1. i ,6l l!umber of particles detected at stage S, rising
UEft1,1 1 6)
lf «CllP+CR$1 ,[Q. O,Ol 6010 ó2~
stream: EPV
r XP!.h TlP V+ i XPIJOtCRS/ CCt l PttJUl) llumber of undetected particles: XEPV.
(PVX:=(fl'VXtT:J;PVOtCRS/ t Ct iPtUS l
H"V:::lPiJ-+ flii"VO•C 1 tP /(Ct tP+CJtSl
XEPV=XEP\I+TXPYO•C I IP/(CIIP+CitS)
TR0<2.,t ,,)=fROt2, 1 ~4)+UPVO•CllP'/tC1 tP+CRS)
XE t 2, t ,t)=X[ C2,1 ,6l+TX:Pt~O•CUP i cC llhCRS I
lRO t t ,1 ,6 I :.lflOC 1 , 1,U+ TXPIJO•C tlf> I (C t!P+CRSI
((I' 1 ,6lt(l1 'I ,61+UPYO•CIJP/ !Cl!P•CRS)

I
Calculat1on of the partiele transition ~ro·
62~ lf (!TPV+HP!J! .sr. t4o'
1 f"UtfPVIOI>V+UP\I'J babilitles and detection probabilities of
lF t<EPVUEfl'IJ; .Gl. O.Oi particles that enter the channel.
1 EF'PIJ•EPIJ/tEtfJH(fi'IJl
tf' UEl> 1JX+lE.PV:o ~GL O.Oi
1 Hf'!JX•f:PIJU!tf'IJX+J:H'\'l:l

I lCOUNT•tCOUHHl I
I
•o uo t,z,, Calculation of the partiele transition pro·
P10,2,ll.,.P0,2,l)
PtCZ,t,It=Pt2,t,H babillties and detection probabillties at the
Eft 11. Jti+I J::Eft1.1 ,I+t;
EF1 ( 1,2tl )II(Ff 1,2, ll different stages.
tr:t<2.1,l)11(F"t2,t,ll
€FH2, 2,1·1 J:::EF<:t,2.1·11
lF tcllt011,2,lltfft'OH,t,ln .f'tE. O.Ol
f'( I ,2 • D=Tit0(1 ,2, f )/
OA:O<I,2,ll+U:0(1,1 ,In
lF ((TA:0<2,1,U+Tit0<2,2,DI .NE. O.Ol
P(2, t ,J)rTR0(2,1, I I/
nA:0<2,1 .ll+tR0(2.,2,11l
Ptl, J,l)=l.-f>(1 ,2,11
P(2,2,!l= I .·P(2, t, U
n ut:n;l;h11+Ut1tt,J-+nl .~E. oJ.O)
EFt1 t t,hl )11{(1' I ,1+1 )!(€(! t 1,IH H~Et 1,1 I ltlll
If UEt2.2 1 I-tHJ:Et2,2,l-11l &N€~ t.~l
Eft2,2,t•1 )!t[t2,2.1 ... 1 U(Et2.2.1 ... 1 HXEf2,2.I-t i!
lf ttE\1.2.ll•XE<1,2.1)) .tJE. O.fH
Ef( t .2, 1 J=Et 1 ,.?,11/(E> t.i,! i+ X[\ L2~ J-1!
If HEt2,1,tHIEt2.t,lJ) .JE. O.CU
Ef(2,1 ,1 )af:(2, 1,1l/CEt2, 1, I )+XE\2' .t ~I l l
COUIIQE
I
00 650 l,2,t Deelsion upon another iteration or oot.
tf rf'llt2'JU .EO. Ot G(HB t-<12
JF (AftS<Pf1,2tO .. Ptt!,Z,lli/f>t1,2.U .GT. Rtlf")
1 6010 135
642 If <P<2. f, 1• .E1l. OI GOTO 643
If (48Stf't2,t,U·P1t2,1,1JJtPt2,t,tl .GT. RfLPl
1 GOTO 11>
643 tF ~Hil,:!,!l .EO. 0) GOTO U_.
tf (AJStH11~~,ll·Ef11!.2,!Jl/Hi1,::!:,tl 4GT. F;ELEFI
! GOTO 135
!\U !~ lf:t'<2 1 !+J) ~EO. Ol 50TO 650
tF €Afi:SIHt2,l 1 l!-EF1i:!,!,!I'I,·(F(~.I.tJ ~llT. REL(f)
1 GIHO 1l$
<llO COHT t HU€

I 1

-175-
 Print out of partiele transition probabilities
557 PR:I~T 9o31 ,ICOUHT
~UI FOR'rtATt Aantal lttratushgt'IU ',13,1/1' 0\IERGAli&S flow rate of tracer particles and detection
IKANS[H;' l probabilities at the different stage boundaries.
PRIHT 9632, iPV,[FPV,EFPVXl
fOIHtATi/' Votding'itJ\lk: Pv1"F6.l,' EttPv);··f6.3,
[f(1-Pvl:'F6 •.H

PSCUIT 9ó4t,ii-1,P11,::' 1 Il,P\2 1 2 1 IJ,

:uc~.'. 1) .EFI1 ,1,1+1 1 ,Ei12,2, 1-1 J ,Ho ,2, I 11
'i640 COHnNUE
'1641 FORnATI' kntllgreno; No.:•,JJ,· Pf:·,F6.3, ... Pr:·,Ftt.l,'
1 [1(!-PrJ:·,fó.J,· E111-Pfl:',f6,J,· EtiPrl:',Fó.3,'
1 EfiPf I :'Fó,Jl
.i655 lBOD=O
I fOP=U
00 3660 ITEL=2,10
1800= IBOD+ ITRA,.S I 1 , l, ITEL, I l + ITRAHS( 1,2, ITÜ, 1 H
ITRAHS\2 1 1 1 I HL, 1l+IlliAHSt2,2, ITEl 1 ll
! IOP=ITOP+ 1TRA~S1 1, I, ITEL,!Ol +ITRUSi 1 ,2,1TEL, lOl+
lTRAHSi 2,1, ITEL ,1 Ol+ITRAIIISi2,2, ITEL, 10!

!f !(!lûf'+lBOfn .bT. 01
YH!=J:"~OA r t IBOD I !fLOAT i ITOP+IBOD I
f'R!NT 3670
UlO F"ORnATU/ 1 ' ltASSABALANS; ·)
f'R:INT ló8U,tlBOO, :TOP.VHf'
·:~lE· FOR•hl~<:, bii~eltJU 1n fiODEtt: · ,!4 1 .', DeeltJes tn fO~:
1 14·,,~, P~:,R: 'F5.3,!//,' lfEELTJESDf.BIEHIII:'/
t· ·----- Gi1+ll ••••••
·----- fit1; ---·--- ----- Hetto -----· l
liU 4100, IlEt..=1 1 10
ITE!.I=IHL-1
DAAl.=Et 1, I, ITEL+!) +XE i I, 1, !TE!.'+! I+
Et2, t.! T[l)+(E\2, I I ITEll
IF" tl!E,l .EIJ. !)! üAAt.:DML+l;(PV
E~2,! I tTEll=DAAL
STIJG:Et 2,2, ITEl-1 l+XEt 2 ,2, llEL-1 ~ t
[11 12, !TEL Hl([ II ,2, liEll
!f tITEL .EO. I I STIJ6=
[I 1,2, ITH l'+lf( 1 ,2, I!EU
IF i ITEL .EO. 6l STIJG=STIJG+TP1J
Ett,2,1Ttll=ST!J6
PRUT 4200, lTEll , DAAL, ST l JG 1 Hl:l~i DAAL-ST IJ&)
4100 COtifT!IfUE
1200 FORPIATI' Kujgren'i:',I3. Dla!stroo~t: ,f8.2,'
1.f8,2, Hetto;trOOII: ',F6,Jl

-176-
APPENDIX 4.

Table A4.l.

Partiele transition probabilities pf and pr and local partiele flow rates Ff and Fr rneasured
inside the standard 90 deg. classifier_.
~umber of stages: R = 10; Location of the feed stage: V.;:Si {1odel particles Dl.s·

vf: 2.039 m/s; ~R' .934. vf: 2.255 m/s; \• .779.

(g/s): (g/s):

7 .616 .Jffl 0 0 7 0 0
5 .522 .245 0 0 5 0 0
4 .464 .315 0 0 4 .01 0
3 .'>21 .293 0 0 3 .:m .Dl ,OI
2 .475 .348 0 0 2 .614 .01 0
j, .648 .353 0 0 I .825 0 0
(l 0 0 0 0 0 0 0 0
~mn Pr' .524 ~ Pf' .624
~i?an pr: .317 ~ Pr' .327

<),: 0 g/s; 'r' 2.(135 m/s; \ ' .923. o": 0 g/s; "r' 2.318 m/s; \= .581.
(g/s): (g/s):

.545 .302 0 0 .:a. .403 0 0

.519 .237 0 0 ) ,559 .29\ 0 0
4 .497 .337 0 0 4 .%7 .3 .01 .OI
3 .54S .273 IJ 0 3 .658 .373 .OI .OI
.)()3 .139 0 0 2 .ffi .411 .OI .01
,737 .35 0 0 1 .002 .3') 0 0
0 0 () 0 0 0 0 0 0 0
~""" ~'r' 5>4 ~Pr' ,fi<\6
)hm pr: •312 ~ pr: .3:ï2

o": 0 g/s; vr: 2.374 m/s; ~R' .491.

(g/s):

.(65 ,291 () 0 7 .655 .27 0 0

.%7 .222 0 0 5 .'S'i8 .338 0 .01
4 .)1)3 .l!'f! 0 0 4 .6 ,))) .OI .01
.')74 .2&! .0! 0 3 .657
.338 .OI .OI
.579 .412 0 0 2 .679
.418 .OI .OI
,74[, .3)1 0 0 I ,891" .445 0 0
0 0 0 () 0 0 0 0 0 0
""'~m pr:
Pr' ."1!4
.316
M9an Pr' .ffi
M9an Pr' .35
o": 0 g/s; vf: 2.222 mis; <I>R: .732. ()": 0 g/s; V f; 2.427 m/s; <I>R: .4Cl3.
(g/s): (g/s):

7 .524 .4(13 0 0 7 .647 .375 0 0

5 .507 .268 0 0 5 .585 :197 0 .01
4 ,536 .248 .Dl .OI 4 .tm .3(13 .OI .OI
3 .619 .314 .OI .01 3 .6133 .337 .01 .OI
2 .631 .421 .01 .OI 2 •738 .413 .01 .OI
I .886 .4rh .01 .01 1 ,f!h7 .435 0 0
0 0 0 0 0 0 0 0 0 0
M9an Pr' ,642 M9an pf: .675
M9an pr: .335 M9an pr: .353

-177-
 q": 0 g/s; vf: 2.425 m/s; pv: .tm; <l>R: .295.
botmarv: Pr' p'
r
Ff (g!.,): Fr (g/s):
9 0 0 0 0
7 ."68 .J% 0 0 7 .625 .31 1.95 2.31
') .6 .Xfl .01 .OI 5 .578 .:m 4.06 4.42
4 .61 .348 .OI .OI 4 .(()J .2"/j 8.82 7.93
.ïll .336 .OI .OI .647 7.93
' .721
.913 .433
.4:12 .OI
0
.01
0
3
2
1 .833
,({>5
.332
.34
l'il+
8.82
7.44
5.3
6.55
4.41
0 0 0 ,o 0 0 0 0 .88 0
>lam Pf' ~Ó77 ~ren Pr' .6'\4
~bm pr: .364 t"am pr: .119

i~,: (I g/s; \·r: 2.4S'3 m/s; jl: .051; <l>R: .TIJ. .677.
V
txlllflrk)ï\':
~'r' p : Ff (g/s); Fr (g/s):
r
q () 0 0 0
7 • "18 .312 0 0 7 .514 .314 l.ffi 2.11
.614 .277 .01 .01 5 .564 .23 6.57 7.03
4 .619 .J:f, .OI .01 4 .tal ,271 12.78 IJ.83
î .fi>7 .3S2 .OI ,OI 3 .(182 .282 1'\.35 14.4
2 •74l .425 .01 .01 2 ,ff) .345 12.35 11.4
I /)14 .!ö2 0 0 1 .8% .353 9.11 8.16
0 I) IJ () 0 0 0 () .95 0
:-hrn pf; .(;](, ~pf: .678
;.nUl pr: ,·v·xl ~pr: .X!S

o..,..: n g/s: \·r: 2. )J') mis; p : . all; <!>R: .lOS •

V
.37.
lxxtndarv: pf: p : Ff (g/s): Fr (g/s):
r
q n 0 I] 0
,({"1.) .4"1> 0 0 7 .624 .3% 3.52 4.52
,(,]<) .11'\ .(11 .OI 5 .591 .295 8.58 9.58
4 .62"3 .:!1Î .OI .01 4 .687 .3 15.17 14.58
.721 .''h~ .(ll .OI 3 .701 13.01 12.43
.II'6 ,4)3 .OI .OI 2 .735 9.54 8.96
,Q'"::ö .!#< 0 {) I .89 5.32 4.73
(l 0 0 () (l () 0 0 .58 0
}ohm Pr= .(1)4 ~ Pr' .691
~hnl (>r: .]74 ~ Pr' .34

i~: l.fil g/s; 'r' 2.l24m/s; py: 0: 4>R: .962. Q",: 1.41 g/s:
bJtJJ!dary:
Pr' p!
r 'r (g/s): Fr (g/s): OOundarv: Jl:
9 0 0 0 .m 9 0 1.
7 .621 .2'\4 .ff) .75 7 .«> .:l6 4.14 5.16
'; .464 .231 1.59 1.65 5 .284 8.96 9.'E
4 .41 .292 4.74 3.13 4 .316 14.67 14.29
3 .51! .244 6.46 4,85 3 .713 .362 13.43 13.01.
2 .52 ,31 6.59 4.<E 2 • 7ffJ .38 9.51 9.13
1 .775 .278 7.19 5.57 1 .923 .392 5.01 4.62
0 0 0 1.62 0 0 0 0 .39 0
~pf; ."h3
!'Urn l'r' .ff!4
~ Pr' .28 flam pr: .346

q": 1.31 g/s: .894. 1.24 f!is:

7 .em .244 1.16 1.3 7 .658 .324 1.28 4.24

5 ."hl .335 2.15 2.29 5 .61<) .283 8.&:J 9.86
4 .495 ,242 5.46 4.28 4 .7 .:re 13.85 13.'\S
3 .547 .295 6,54 5,37 3 .702 .341 ll,fi, 11.38
2 .m .327 6.65 5.48 2 .772 :J94 8.4 8.12
I ,8J4 .311 5.<;6 4.81 1 .93'3 ,412 4.07 3.fl
0 0 0 1.17 0 0 0 () :n 0
flean Pr' ,ff;> !'Urn Pr' ,7(X)
~ffin pr: .302 fum pr: .34

-178-
<ly: 2.87 g/s; ;r' 2.136 m/s; Pv' .m:l; <I>R: .93'>. <ly: 4.94 g/s; vf: 2.073 m/s; Pv' 0; <I>R: .964.
OOurnlary: p: Ff (g/s): Fr (g/s): boundary: Pc' p: Ff (g/s): Fr (g/s):
Pr' r. r
9 0 0 0 .19 9 0 0 0 .18
7 .422 .'lbl .81 I 7 ."Al !.OI 1.18
5 .535 .275 3.79 3.93 5 .252 2.92 3.1
4 .473 .29 8.78 6.1 4.41'\ .175 15.14 10.38
3 .477 .271 9.87 7.19 3.579 .263 20.33 15.56
2 .4'1'J .315 10.:6 7.88 2.515 ,n; 17.2 12.44
I .745 .333 10.51 7.83 I.ffi3 .:m 15.52 10.75
0 0 0 2.68 0 00 0 4.76 0
~ Pf' .545 ~pf: .545
~\,an pr: ,n; ~pr: .276

<ly: 2.67 g/s; o": 4.61 g/s;

7 .625 .292 2,58 2.95 7 7.29

5 .576 .263 7.63 8 5 15.58
4 .554 .254 15.48 4 .634 27.87
3 .677 .298 17.74 3 .102 30.98
2 .ffJ) .n 13.5 11.19 2 .659 21.6
1 .781 .362 10.52 8.21 1 .82 17.15
0 0 0 2.3 0 0 0 3.09
~ pf: .63 ~ pf; .679
~pr:.m ~ Pr' .299

o": 2.7 g/s;

7 .671 .318 7.46 9.05 7 .743 .325 13,53 17.3

5 .611 .287 15.4 16.'1'J 5 .71 .355 17.58 21.35
4 .662 .32 24.42 23.31 4 .694 .346 25.05 24.12
3 .724 .314 23.86 22.75 3 .696 .318 22.1 21.17
2 .721 .377 15.64 14.53 2 .739 .383 16.04 15.11
1 .872 .43 8.67 7.56 I .895 .41 8.87 7.94
0 0 0 1.11 0 0 0 0 .93 0
~\,an pf: .684 ~\,an pf: .724
~mn pr: .342 ~pr: .362

Q": 2.64 g/s; .!2 •

.m • 3'lb .7.87 10.07 .759 .277 17.14 2l.l8

.693 .298 14.33 16.53 .743 .341 19.59 23.64
4 .754 .:Dl 20.89 20.45 4 .743 .342 25.39 24.84
3 .765 .m 15.94 15.49 'J .7:'1) .279 21.96 21.41
2 .728 .36'; 9.48 9.03 2 .792 .401 13.47 12.91
1 .921 .409 5.({, 5.21 1 .915 .378 6.'5 5.9')
0 0 ,45 0 0 0 0 .55 0
~pf: :hln pf: .ra
f>km pr: H::nn pr: .3'>

-179-
Table A4.2.

Partiele transition probabilities pf and pr and local partiele flow rates Ff and F measured
r
inside the standard 120 deg. classifier ~
Number of stages: R JO; Location of the feed stage: V•S; "odel particles
Dl.5'

Q_.: 0 >:fs; vf: 1.95 m/s; <I>R: .9'17. Q": 0 Jl/s; vf: 2.426 m/s; <I>R: .':6.
(g/s): (g/s):

() .6:-{l () 0 7 .439 .529 0 0

.l'ill .183 0 () 5 :m SJ7 0 0
4 .:m .162 0 0 4 .YJ .429 0 0
[) .)Jl
' .227 .346 0 3 ,4rfl 0 0
.lï'l .221 0 0 • 'j}) .5 0 (I
.211 .:m 0 0 I •52 .519 0 0
u () () 0 0 () 0 0 0 0
~hulpf: .201 ~kun pf: .472
thm pr: .248 ~h-m llr: .'ffi

Q,_.: ll g/s; vf: 2.10) m/s; OR: .9'13. o,: 0 g/s; \' f: 2. ')')') mis: <I>R: .2'14.
(g/s): (Als):

.17 .2116 () () ~Sl7 •')12 I) 0

.]4q .337 () n .'ffl :·,, (l 0
4 .265 .331 0 () .48 .473 0 0
.331 .387 () 0 .s~ . '>14 0 0
.TB .YII () 0 .627 .'fJ7 0 0
.324 .SS:î 'I) 0 .bl7 •~""x'1l 0 0
0 0 0 () 0 0 0 () 0 ()

~hm Pr: .182 fhm Pr' .rx-6

~hm pr: .·)-;q /1-t:un pr: .'>36

IJ,_.: 0 g/s; vr: 2.141 r:l/s; .':ll7. ~.: 0 g/s; vf: 2.642 m/s; p: •6ll; <I>R:
V
.~ .
(g/s): boundary:
Pr' p : F[ (~/s): Fr (g/s):
r
9 () 0 0 0
.l'l7 .TIR 0 () 7 .544 .643 0 0
.247 .:w. 0 () 5 .671 .612 0 0
4 .271 .171l 0 0 4 .53 .548 0 0
3 .2% .)')1 n () 3 .628 .587 0 0
2 .327 .:nl 0 (1 2 .681 .007 0 0
l .321 :s7R () () I .722 .813 0 0
0 (I () 0 0 0 0 0 0 0
.628
~""' "r' .297
?-ban pr: .339
l>t>an pf:
l>t>an pr: .ffJI
Q": 0 g/s; vf: 2.285 m/s; <I>R: .&.2. Q": 2.34 g/s; •29; ~R' .942 •
(~/s): (g/s):

7 .4ll .3ï'l 0 0 .3
5 .358 .429 0 0 ) 1.37 1.51
4 .352 .4"1) 0 0 4 3.85 1.65
3 .393 .349 0 0 'l 4.17 1.97
2 .432 .451 0 0 2 4.32 2.12
I .4()', .452 0 0 1 .577 3.48 1.27
0 0 0 0 0 0 0 2.2 0
l>t>an pf: .N'l r-hln Pr' .:u,
l>t>an pr: • 428 M::lan pr: .'ffl

~: 0 g/s; vr: 2.372 m/s; <I>R: •7fY5. .357; oPR: .681.

(g/s): (g/s):

.~) ."al 0 0 .41'i .376 .7S 1.02

.41'l .458 0 0 'i .3:l .436 2.0\ 2.31
4 .3% .424 0 0 4 .326 .352 4.15 2.38
·; .407 .4'>7 0 0 'l .:Jll .358 4.65 2.68
2 .4'i .'i36 0 () 2 .403 .414 4.'ll 3.01
1 .45 .664 0 0 I .4ft. .'ï68 3.69 1.72
0 0 () 0 0 0 0 0 I.<J7 0
;>-l>an pf: .433 ~hm pf: .:lTl
:bm pr: .481 thm pr: .4Zl

-180-
Q.,: 2.22 g/s; Q.,: 4,81 g/s;

7 .fi-{j ,494 1.73 2.59 ."ffl .%1 1.88 2.92

<; .428 .'iY> 2.79 '3,64 .378 .482 4.'-> 5.5
4 .:m .482 4.84 3.47 4 .?if> ."h7 10.00 6.12
'l .417 .T.J4 1.1'> 1,99 3 .40'3 •'l44 11.41 7,64
2 .';]'i 54?S 5.23 3.frl 2 .IIJ7 .446 10.7) 6.82
I ,'i72 .6N 3.2 J.R3 I .525 .rm 7.94 4.17
0 0 0 I :l7 0 0 0 0 1.77 0
rtnn Pr= .4h7 :i:all pf: .418
Mean pr: .lfl3 Mean Pr' .4'i1
Q".: 2.1 g/s; Q.,: 4,(:1) g/s;

.614 .~1 2.64 4.27 7 ."88 • 'i56 4.2 6,8.'\

.562 .574 2.8 4.44 5 .526 .574 5.29 7,94
."ffl .lffl 4.64 4.18 4 .43 .423 10.7 8.ffi
.527 •.'\31 3.89 3.43 3 .478 .404 12.!}) IO,ffi
.573 ,)!}) 3.17 2.71 2 .584 .532 10.07 8.03
.719 ,751 1.65 1.19 1 .629 .62 5.5 3.46
0 0 0 .46 0 0 0 0 2.04 0
Mean pf; .567 Mean pf: .527
Mean pr: .559 Mean pr: .516

Q",: 4.35 g/s;

7 .371 .322 ,82 .97 .m .535 2.93 4.76

'; .273 .396 2.79 2.95 5 .35 .464 7.22 9.05
4 .233 .282 6 •.'\3 2.33 4 .2'll .46 15.53 8.27
'3 .ïm .37 6.39 2.2 3 .372 .Th 19.17 11.91
2 .2'>2 .272 7.36 3.17 2 .403 .467 17.5 10.24
I .191 .489 6.88 2.ffJ I .4ffi .503 !3.59 6.33
0 0 0 4.19 0 0 0 0 7.'21:> 0
~m Pr' .277 :i:ai1 Pr' .395
~ixm pr: .:"f}) MtWl pr: .448
Q",: 4.18 g/s; .904. Q.,: 9.00 g/s; .m: <l>R: .''il.
(g/s):

.429 .468 .ffi 1.26 7 .56 .(ffi 6.0! 10.01

'} .318 .403 3.2'\ 3.65 'j .4'13 .'5'21:> 9.00 !3.00
4 .11 .1lll 7.'\6 1.78 4 ,:Et; .449 18.2 !3.12
'l .29 .377 7.(> 1.82 3 .472 .45') 18.47 13.38
,'l1 .418 7.61 3.87 2 .441 .414 J(J,'lfJ 11.27
.411 .%8 6.41 2.64 I .)')4 .(f.'l 1].41 6.32
0 0 0 '!.78 0 () 0 0 '\,09 0
Mean Pr' .132 ~m pf: .47'5
tbln rr: .417 ~hm pr: .'il6

-181-
Table A4.3.

Partiele transition probabilities pf and pr and lotal partiele flo" rates Ff and Fr measured
inside the broad 120 deg. classi fier~
Number of stages: R = 10; Location of the feed stage: \'=5: ~todel particles Di~S~

<tR: .974, fJ.: 2.37 gis; vf: 1.79 m/s; 1\! .034: <I>R: .968.
(g/s): (gis):

.164 .144 .15 .22

.]:!i () ) .223 ,''lll .51 ,5')
.4'11 0 4 .215 .495 3.(]) .8
,:lf\1 (I 0 3 .144 .2'i 2.83 .54
.114 .!) ll 0 .aJ'l .297 2.73 A1
.17\ ,2(f', 0 0 1 .161 .372 2.73 ,44
tl () 0 (] () 0 () 0 2.29 I)
:rnn ~'r' .l'll ~\!) Pr' .162
:--hm pr; ."H) ~pr: .42

~ ' .7(i.l. 1.94 g/s;

~,: 0 i!-/~; \"f: 1.11 m/s; R' fl.'
(gis):

.484 .673 0 0 7 .445 .631 .6 1.04

.'lil • 'f,f) () (] 5 .373 .437 1.61 2.05
:124 ,')4'; () (\ 4 .332 .64 3.57 2.07
.:lfil .4(~ () I) 3 .329 .416 3.62 2.12
;z·A .12ï 0 () 2 ,307 .539 3.23 1.73
,4/'f', • ~"'>47 () 0 I SlS .481 3.03 1.53
0 n 0 0 n 0 0 0 1.5 0
~htll pf: ,:I*' ~ Pr' .362
:i:'nn f\: • -i) ~ Pr' .539
• 4'}). fJ.: 2.21 gis; vf: 2.27 m/s; Pv' .184; \ ' •'i82 •
(g/,;): boundary: p: Fr (gis): Fr (gis):
"r' r
.92
9 0 0 0
7 ,655 .(,')6 1.(8 2
.4'r2 0 n 5 .295 .514 1.74 2.66
.4ri) .6Yl 0 ()
4 .371 .591 4.23 2.95
'I :m .6!4 0 ()
3 .34') ."iJl 4.03 2.75
2 .:17 .)<:.[) 0 0 2 .1"!! .563 3.67 2,38
I .cfll .5% 0 0 1 ,"iS/ .582 3.ll 1,83
0 () 0 0 0 0 0 0 1.29 0
:hm pf:· ,4/';7 ~mn Pr' .414
~hm pr: .f>ll ~Pr' •'i82
!~,: () g/" \·(: 2.39 m/si ~R: .184. ()": !.eó gis; vf: 2.39 m/s; pV : .047; <tR: .303.
(gis): OOundary: pf: p: Ff (gis): Fr (gis):
r
9 0 0 0 1.3
.64 .727 0 0 7 .'i:!J .(f>7 1.36 2.65
.426 • "85 0 0 5 .4';3 ,fl]6 l.ffi 3.18
4 .478 .OJ2 0 0 4 .47 .ff'A 4.27 3.7
3 .'}39 ."!! 0 0 3 .433 .612 3.34 2.78
2 ,')4 .571 0 0 2 .476 .>16 3.14 2.58
I .~ .615 0 0 ! .743 _~,,3
2.2 1.6.1
0 0 () 0 0 0 (l 0 .'56 0
~mn .534 Pf' ~mn ~'r' .497
~pr: .635 f<nm Pr' .629

fJ.: 0 g/s: "r' 2.65 m/s: <I>R: .016. <l._,: 2.31 g/s; vf: 2.65 m!s; p : .131;
V
<I>R: .015.
(gis): houndary: p : Ff (gis): Fr (gis):
"r' 0
r
0 2.28
9 0
.702 .rm 0 0 7 .676 .71'i 1.41 3.00
.1'>'5 .rm () 0 'l .5'\7 .645
,(if)
2.06
4,22
4.34
4.19
4 •'i44 .6% 0 0 4 .511
1 .Hl4 .712 0 () 3 .649 .738 2.79 2.75
.(01 .(~1 () 0 2 .779 .mi 1.6 1.57
.9')7 .67() () 0 I.922 .795 .44 .4
() 0 () () 0 0 0 0 .03 0
~hm pf: .61f> ~Pr' .'i'll
'bm pr: ,i'(J', ~pr: .717

-182-
q": 4.11 g/s; vf: 1.79 m/s; pv: .C!lS; 4>R: .965. Q": 9.48 g/s; vf: 1. 79 m/s; pv: .042; 4>R: .952.
botmdary: pf: p: Ff (g/s): Fr (g/s): boundary: p : Ff (g/s):
r Pr' r
Fr (g/s):
9 0 0 0 .14 g 0 {l 0 .45
7 .196 .457 .39 .52 7 .1\93 .497 .54 .Cf)
5 -~ .llll 1.1 1.24 5 .343 ,571 .'11 l.:l6
4 ,158 .514 5.18 1.21 4 .KB .337 10.96 1.93
3 .175 .318 5.21 1.24 3 .I :'liJ .:JfB 11.05 2.02
2 .006 .416 4.76 .79 2 .an ,'/,7 10.52 1.5
1 ,173 .403 4.8 .83 1 .123 .491 10,3 1.27
0 0 0 3,97 0 0 0 0 9.03 0
~ám pf: .159 ~Pf' .n1
~ pr: ,446 fbm pr: .455

q": 4.2f> g/s; rz,: 10.58 g/s;

.429 .622 1.5 2.58 7 ,[>43 ,675 1.78 3.57

.358 .52 2.84 3.92
') .383 • '\!! 4.25 6.03
4 .'JOO .:i% 7.18 3.'E 4 ,7fil. ,';03 16.29 7.49
'l .259 .472 15.62 6.83
3 .311 .'>2 7.19 4.01
]4.Ul
2 .251 .519 6.6 3.42 2 .194 .)'}/) 5.28
l .519 .512 6.62 3.44 I .429 .:B') 15.41 6,61
0 0 0 3.18 0 0 () 0 8.il 0
~ám pf: .'AS ~bUl pf: * 'IJ2
~ Pr' .564
fhm pr: .524

q": 4. 51 g/s; (\,: 7.97 g/s:

7 5 .4'Jl ,({'{> 2.35 4.39

5 6.33 .358 .517 4.74 6,79
4 .416 6.35 4 .3 .'1.7 !4.'Jl 9.06
3 .38 .:>Y:i 5.21 3 .27<l .ss 14.22 8.3
2 .Xll .'>54 4,06 2 .28 .'JR') 11.:11 7.4
I .62 .477 4.22 l .SS? ,49'l 13.37 7~4S
0 0 0 0 0 0 0 '>.92 0
~ Pr' .433 fhln ~'r' .368
~mn Pr' .':l:fl fmn Pr' .559
q": 4.63 g/s: Q": 8.11 g/s; V f: 2,39 m/s; P) .Ol2: 4> R' .496.
boundary: ~'r' p: F'f (g/s): Fr (g/s):
r
9 0 0 0 4,Ul
7 .675 ,6!9 2.58 5.44 7 .631 .632 4.56 8.64
5 .442 .588 3.76 6.62 5 .411 .629 6.(6 10.16
4 .434 .671 B.ffi 6.92 4 .403 .567 16,37 12.35
3 ,437 .:>X; 7.Cf) 6.22 3 .322 .574 14.04 10.01
2 .408 .527 6.95 5.18 2 .373 .649 12,49 8.46
I .717 .523 6.25 4,48 1 .675 ,455 12,39 8.36
0 0 0 1.77 0 0 0 0 4.03 0
fám pf: .496 ~ pf: ,45
fám pr: .612 fám Pr' .S'Jl

q": 4.45 g/s; vf: 2.65 m/s; pv: .069; 4> R' .O:'l6.
botmdary: Pr' pr: Ff (g/s): F'r (g/s):
9 0 0 0 4.'19
7 .662 .lUl 2.48 6.77 7 .718 .659 6.78 16.07
5 .592 .618 4.(!2 8.31 5 .599 .624 7.4 16.7
4 .685 8.37 8.21 4 .48 ,7f:ió 14.64 14.43
3 .663 5.2 5.03 3 .53 .695 10.&'> 10.66
2 .fff> .635 2.96 2.8 2 .676 ,(:ffi 7.35 7.14
1 .853 ,sa; 1.00 .93 1 .93 .8 2.93 2.n
0 0 0 .16 0 0 0 0 .2 0
~ pf: ,622 t-ban Pr' .(iJ2
~ pr: .703 ~ pr: .719

-183-
SliMHARY.

This dissertation describes an investigation of the separation

performance and capacity of zigzag air classifiers at high solids feed
rates.
This separation technique is used e.g. in installations for the
separation of household waste.Foil shaped particles of which the
dimensions are relatively large compared to the classifier channel are
classified according to their falling behaviour in the upward air flow
through the channel.
The multistage behaviour of zigzag air classifiers is the result of the
partiele behaviour at the individual stages and of the interactions
between the stages. The investigation comprised both a study of the
overall separation performance, i.e. the separation sharpness and the
capacity in relation to the channel geometry and of the relation between
this multistage performance and the partiele behaviour at the individual
stages.

The separation sharpness and capacity of zigzag air classifiers with

angles of 90 and 120 deg. were measured. Both the air velocity and the
solids feed rate were varied. The experiments were carried out with feeds
consisting of identical particles and with binary mixtures. The location
of the feed stage and the channel width were varied as well.
The 90 and 120 deg. classifiers with standard widths, i.e. the channel
width for which all inward protruding edges are in one vertical plane,
have similar separation shàrpnesses. The influence of increasing solids
rates on the separation characteristics is quite different however. An
increase of the feed rate at constant air velocity causes the fraction of
bottorn product, ~ to increase in the 120 deg. classifier. At
increasing feed rates, ~ first decreases in the 90 deg. classifier.
This difference is caused by the friction between the falling particles
and the wall. This influence is more important in the standard 90 deg.
classifier than in the standard 120 deg. classifier. Partiele residence
times are longer in the standard 90 deg. classifier than in the 120 deg.
classifier.
The separation performance decreases when the channel is wider. Partiele
residence times then decrease. Because of the shorter residence times at
larger channel widths the capacity increases.

-184-
A measurement technique was developed to obtain a fundamental insight
into the functioning of zigzag air classifiers. This technique made it
possible to analyse partiele behaviour at the individual stages.
Partiele transition probabilities, partiele veloeities and -flow rates
are the most important so-called internal variables that describe this
behaviour. An optical method was developed to detect white tracer
particles in a main stream of black but otherwise identical particles. An
on-line computer was used to process the detector data. The internal
variables were measured inside the standard 90 and 120 deg. classifiers
and in the broad 120 deg. classifier by using an experimental set-up that
was based upon the method mentioned above.
The partiele transition probabilities, i.e. the rising probabilities of
particles at a stage, depend upon the stage geometry, the superficial air
velocity, the local partiele flows at the stage and the partiele
properties. The transition probabilities also depend upon the direction
from which particles enter the stage. Dimensionless groups were derived
by means of dimensional analysis to describe the partiele transition
probabilities. Empirical correlations were determined that were based
upon the dimensional analysis. These correlations describe the transition
probabilities of the particles that were used as functions of the
superficial air velocity and of the rising and falling partiele flow
rates to the stage ••
A mathematica! model·was developed which described the separation
sharpness and capacity based on the partiele behaviour at the individual
stages and the interactions between the stages.This model is an extension
of Sendens "one-step-memory" model (1979). The measured dimensionless
correlations for the transition probabilities, the number of stages and
the location of the feed stage are the input data of the model. The
influence of feed rate, air velocity, number of stages and location of
the feed stage on the separation performance and capacity were determined
quantitatively for each of the channel geometries used in measuring the
internal variables.
Finally, some design aspects were discussed. These aspects play a role in
the selection of a suitable channel geometry and in the sealing-up of the
channel to obtain the desired separation sharpness and capacity. The
dimensional analysis and the results of the model calculations were the
starting points for this discussion. More specific aspects of the design,

-185-
such as the wall roughness that can selectively influence the rising
probabilities of particles, are dealt with in a qualitative way.

-186-
SAHENVATI'ING.

Dit proefschrift beschrijft een studie naar de scheidingascherpte en de

capaciteit van zigzag windzifters bij hoge vaste stof beladingen. Deze
scheidingstechniek wordt o.a. toegepast in huisvuilscheidings-
installaties. Folieachtige deeltjes waarvan de afmetingen relatief groot
zijn in vergelijking tot het zifterkanaal worden geclassificeerd op
grond van hun valgedrag in de omhooggerichte luchtstroom door het
kanaaL
Het meertrapsgedrag van zigzag windzifters is het resultaat van het
deeltjesgedrag op de individuele knikken en de interacties tussen de
knikken. Het onderzoek omvatte zowel een studie naar het totale
scheidingsresultaat, d.w.z. de scheidingascherpte en de capaciteit in
relatie tot de kanaalgeometrie, als naar de relatie tussen dit
meertrapsgedrag en het deeltjesgedrag op de individuele knikken.

De scheidingascherpte en capaciteit van zigzag zifters met knikhoeken

van 90° en 120° is bepaald. Zowel de luchtsnelheid als het vaste
stof debiet werden hierbij gevarieerd. De experimenten werden uitgevoerd
met voedingen bestaande uit identieke deeltjes en met binaire mengsels.
Tevens werd de plaats van de voedingsknik en de kanaalbreedte
gevarieerd.
De 90° en 120° zifters met standaard kanaalbreedte, d.i. de
kanaalbreedte waarbij de naar binnen gerichte knikhoeken in een
verticaal vlak liggen, hebben een vergelijkbare scheidingsscherpte. De
invloed van toenemende vaste stof debieten op de scheidinga-
karakteristiek voor identieke deeltjes is echter verschillend. Een
toename van het voedingadebiet bij een constante luchtsnelheid leidt
ertoe dat de fractie bodemprodukt ~R toeneemt in de 120° zifter. In
de 90° zifter neemt ~R in eerste instantie af met een toenemend
deeltjesdebiet. Dit verschil wordt veroorzaakt door de wr1JV1ng tussen
de dalende deeltjes en de wand welke groter is in·de standaard 90°
zifter dan in de 120° zifter. De verblijftijden van de deeltjes zijn
langer in de 90° dan in de 120° zifter.
De scheidingascherpte neemt af wanneer de kanaalbreedte wordt vergroot.
De verblijftijden van de deeltjes worden dan korter. Ten gevolge van
deze kortere verblijftijden neemt de capaciteit toe bij grotere kanaal-
breedtes.

187-
Om een fundamenteel inzicht in de werking van zigzag zifters te
verkrijgen werd een meettechniek ontwikkeld waarmee het deeltjesgedrag
op de individuele knikken kon worden geanalyseerd. Overgangskansen van
de deeltjes op de knikken, deeltjessnelheden en -debieten vormen de
belangrijkste zogenaamde interne variabelen die dit gedrag beschrijven.
Een optische methode werd ontwikkeld voor het detecteren van witte
tracerdeeltjes in een hoofdstroom van zwarte maar overigens identieke
deeltjes. Een on-line computer werd gebruikt om de detector data te
verwerken. M.b.v. een op deze methode gebaseerde meetopstelling werden
de interne variabelen gemeten in de standaard 90° en 120° zifter en
in de brede 120° zifter.
De overgangskansen, d.w.z. de stijgkansen van de deeltjes op een knik,
zijn afhankelijk van de knikgeometrie, de superficiële luchtsnelheid, de
locale deeltjesdebieten op de knik en de deeltjeseigenschappen. Ook zijn
de overgangskansen afhankelijk van de richting waarin de deeltjes de
knik naderen. M.b.v. dimensieanalyse werden de dimensielose groepen
welke de overgangskansen beschrijven afgeleid. Empirische correlaties,
gebaseerd op de dimensieanalyse werden bepaald. Deze correlaties
beschrijven voor elk van de drie kanaalgeometrieën de overgangskansen
van de gebruikte deeltjes als functie van de superticiele luchtsnelheid
en van het deeltjesdebiet van de daal- en stijgstroom naar een knik.
Een mathematisch model werd afgeleid dat, uitgaande van het
deeltjesgedrag op de individuele knikken en de interacties tussen de
knikken, de scheidingsscherpte en de capaciteit beschrijft, Dit model
is een uitbreiding van het "one-step-memory" model van Senden (1979). De
gemeten dimensielose correlaties voor de overgangskansen, het aantal
knikken en de plaats van de voedingsknik vormen de invoergegevens van
dit model. De invloed van voedingsdebiet, luchtsnelheid, aantal knikken
en de plaats van de voedingsknik op de scheidingsscherpte en de
capaciteit werd kwantitatief bepaald voor de kanaalgeometrieen waarvoor
ook de interne variabelen werden gemeten.
Ten slotte werden een aantal ontwerpaspecten welke een rol spelen bij
de keuze van een geschikte kanaalgeometrie en de opschaling van het
kanaal om een gewenste scheidingscherpte en capaciteit te bereiken
toegelicht. De dimensieanalyse en de modelberekeningen vormden hierbij
de uitgangspunten. Meer specifieke aspecten bij het ontwerp, zoals de
ruwheid van de zigzag wanden waarmee de stijgkansen van deeltjes
selectief kunnen worden beïnvloed, werden kwalitiatief behandeld.

-188-
STELLINGEN
behorende bij het proefschrift van G.G. Rosenbrand

De relatie, welke door Vesilind en Henrikson wordt gegeven om het

verband te beschrijven tussen de stijgkans van een deeltje en de
deeltjesconcentratie op een knik in een standaard 120° zifter, is
onjuist.
Vesilind, P.A., Henrikson, R.A., Resources and Conservation, Q
(1981), p211-222.
Dit proefschrift: Hoofdstuk 1.

2 Het regenereren van niet selectieve ionenwisselaars, welke gebruikt

worden voor het verwijderen van zware metalen uit afvalstromen, is uit
milieutechnische overwegingen onverantwoord.
Peters, R.W. et.al., Journal WPCF, 56 (1984), 6, p560

3 De wet chemische afvalstoffen dient zodanig te worden gewijzigd, dat

het strafbaar wordt om, door uitsluitend mengen, chemische afvalstoffen
om te zetten in een produkt, dat niet onder onderhavige wet valt.
Wet chemische afvalstoffen, bijlage III. lijst van uitzonderingen.

4 De methode, welke door Diaz et.al. wordt voorgesteld om

afvalverwerkingsinstallaties te modelleren, is onjuist. Ten onrechte
karakteriseren zij de deeltjesgrootte van de afvalcomponenten doOr
slechts een dimensie.
Diaz, L.F., Savage, G.M., Golueke, C.G., "Resource recovery from
munici pal solid wast es", Vo1.1 ( 1982) , CRC pre ss, Inc. , Boca Ra ton,
Florida.

5 Bij het ontwerpen van zigzag windzifters dient meer aandacht te worden
besteed aan de invloed van de wandruwheid op het scheidingsgedrag.

Dit proefschrift: Hoofdstuk 5.

6 Bij de mechanische verwerking van huisvuil tot compost wordt

onvoldoende aandacht besteed aan het feit dat een belangrijk deel van
de verontreiniging aan zware metalen wordt veroorzaakt door mechanische
degradatie, verbrokkeling en verspreiding van vaste verontreiningen
tijdens de verwerking.

van Rosmalen, G.R.E.M., "Zware metalen in huishoudelijk afval en

compost", rapport Waste Management KHT-THE, (1985).
7 Veel auteurs WlJZen op de analogie welke bestaat tussen bepaalde
meetsystemen voor het bepalen van deeltjessnelheden in een twee-fasen
stroming en een lineair dynamisch systeem. Ten onrechte suggereren zij
hierbij, dat de verwerking van de detectorsignalen d.m.v. kruis- en
autocorrelatie, gevolgd door Fouriertransformatie van deze
correlatiefuncties, practisch bruikbaar zou zijn om de
snelheidsverdelingsfunctie van de deeltjes tussen de twee detectoren te
bepalen.

dit proefschrift: Hoofdstuk 3.

8 De hoge C0 concentratie die Anderson en Tsai meten in hun

2
reactiemengsels bij de partiele oxidatie van methaan tot etheen en
andere olefinen en alifaten m.b.v een aantal metaaloxide katalysatoren,
is niet het gevolg van de lage selectiviteit van deze katalysatoren
maar van de katalytische werking van de reactorwand.

Anderson, J.R., Tsai, P., Applied catalysis, 19 (1985), pl41-152.

Keiler, G.E., Bhasin, M.M., Journal of catalysis (1982), 9-19.

9 De uitbreiding van de sluitingstijden van loketten in allerlei

overheidsgebouwen vanwege arbeidstijdverkorting is in strijd met het
streven van de overheid om, d.m.v. arbeidstijdverkorting, een
herverdeling van werk te realiseren.

25 maart 1986.