Is the fish-hook effect in hydrocyclones a real

phenomenon?
Florent Bourgeois, Arun K. Majumder

To cite this version:

Florent Bourgeois, Arun K. Majumder. Is the fish-hook effect in hydrocyclones a real phenomenon?.
Powder Technology, Elsevier, 2013, vol. 237, pp.367-375. �10.1016/j.powtec.2012.12.017�. �hal-
00876905�

HAL Id: hal-00876905

https://hal.archives-ouvertes.fr/hal-00876905
Submitted on 25 Oct 2013

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.
This is an author-deposited version published in : http://oatao.univ-toulouse.fr/
Eprints ID : 9913

To link to this article : DOI:10.1016/j.powtec.2012.12.017

URL : http://dx.doi.org/10.1016/j.powtec.2012.12.017

To cite this version :

Bourgeois, Florent and Majumder, Arun K. Is the fish-hook effect
in hydrocyclones a real phenomenon? (2013) Powder Technology,
vol. 237 . pp. 367-375. ISSN 0032-5910

Any correspondence concerning this service should be sent to the repository

administrator: [email protected]!
Is the fish-hook effect in hydrocyclones a real phenomenon?
Florent Bourgeois a, c,⁎, Arun K. Majumder b
a
Université de Toulouse, INPT, UPS, LGC (Laboratoire de Génie Chimique), 4 Allée Emile Monso, BP 44362, 31432 Toulouse Cedex 4, France
b
Department of Mining Engineering Indian Institute of Technology, Kharagpur 721 302, India
c
CNRS, Fédération de recherche FERMAT, Toulouse, France

a b s t r a c t

Although the fish-hook effect has been reported by many for a very long time, scientists and practitioners
alike share contradictory opinions about this phenomenon. While some believe that it is of physical origin,
others opine that it is the result of measurement errors. This article investigates the possibility that the
fish-hook effect could indeed be measurement error related. Since all the experimental errors are embedded
in the raw size distribution measurements, the paper first lays down the steps that lead to estimation of the
Keywords:
partition function and confidence bounds, which are seldom reported in hydrocyclone literature, from the
Hydrocyclone errors associated with the experimental size distribution measurements. Using several data sets generated
Fish-hook effect using a 100 mm diameter hydrocyclone operating under controlled dilute to dense regimes, careful analysis
Classification of the partition functions following the developed methodology yields unambiguous evidence that the
Fine particle processing fish-hook effect is a real physical phenomenon. An attempt is also made to reunite some of the major contra-
dictory views behind the existence of the fish-hook based on sound statistical arguments.

1. Introduction up to a critical particle size and after that the recovery increases with
increase in particle sizes.
There are a number of issues associated with the operation of a Although the fish-hook effect has been reported by many for a very
hydrocyclone that remain without clear explanations, often leading long time, researchers and practitioners alike share different opinions
to some level of discord about their cause and effect. This is not the about this phenomenon. There is one school of thought that believes
least surprising given the complexity of the separation that takes the fish-hook effect to be a real physical phenomenon [4–7], whereas
place inside this unit operation. a second one argues that it is measurement error related [8,3].
One issue that has been the subject of much debate over the years is Flintoff et al. [3] opined that this fish-hook effect was due to poor
what is referred to as the fish-hook effect. As the particle size separation experimental procedures and/or agglomeration of fine particles.
in a hydrocyclone is performed using a fluid medium in dynamic condi- Nageswararao [8] summarized that this is a random and sporadic
tions and usually the feed particles have wide size distributions, the occurrence caused by the imprecision of measurement and it does not
separation is never perfect at any particle size. Many criteria have, affect the hydrocyclone performance. A few others [2–4] however,
therefore, been proposed to evaluate the performance of hydrocyclones have also proposed some empirical correlations to predict the fishhook
but the graphical method of representing the recoveries of each particle effect in a hydrocyclone classifier with reasonable accuracy. But these
size in hydrocyclone underflow in relation to their availability in feed as models failed to explain whether this irregular behavior with ultra
a function of their respective sizes (usually in a log-normal scale) is the fine particle sizes in a hydrocyclone is a characteristic phenomenon
most popular one. The data thus plotted is popularly known as partition or not. Majumder et al. [5] have provided a mechanistic argument
curve or Tromp curve. This is normally an ‘S’ shaped curve and a variety supporting the occurrence of fish-hook in all centrifugal separators
of quantitative expressions have been used to describe the shape of the while treating fine and ultra fine particles. The basis of their argument
curve. However, partition curves do not always follow the conventional is that in a centrifugal force field, there is a sudden drop in relatively
‘S’ shaped pattern, rather they have fish-hook patterns as described in coarser particles settling velocities due to Reynolds number restrictions.
literature [1–3]. This means that the recoveries of relatively finer parti- As scientists and practitioners alike share different opinions about
cle sizes are initially higher than the coarser particle sizes in underflow this phenomenon, an attempt has been made here to assess whether
measurement errors could possibly be responsible, on their own ac-
count, for the fish-hook effect. For the sake of clarity, it is emphasized
that the fish-hook effect that is being investigated in this work is that
⁎ Corresponding author at: Université de Toulouse, INPT, UPS, LGC (Laboratoire de
caused by particle size only. This work is not concerned with other
Génie Chimique), 4 Allée Emile Monso, BP 44362, 31432 Toulouse Cedex 4, France. factors that can also affect classification, such as density variations
E-mail address: fl[email protected] (F. Bourgeois). within a particle size class [9].
2. Accounting for particle size distribution measurement error in estimation of the partition function of the hydrocyclone

2.1. Mass balance solution with particle size distribution error

Mass balancing is an important topic in extractive metallurgy. This section recaps on the set of equations that can be used to reconcile the
measured particle size distributions around a hydrocyclone. The purpose is to give utter transparency to the approach that is used in this paper
for propagating the experimental particle size distributions' measurement errors right through to the estimation of the error associated with the
partition function of the hydrocyclone. One side value of this section is to give the practitioner a summarized account of results that can be readily
used to estimate the partition function from measurements. The results that are presented hereafter are largely based on the two-product mass
balancing solution published by Bazin and Hodouin [10]. Notations follow closely those of these authors also so the reader can refer back to their
original work without obfuscating change in notation.
The mass fractions in the feed, underflow and overflow streams are noted f, u and o respectively. If particle size distributions are discretized
into n size classes, X and VX denote the column vector of the measured mass fractions and the corresponding covariance matrix.

f1 varðf 1 Þ covðf 1 ; f 2 Þ … covðf 1 ; f n Þ

2 3 2 3
6 f2 7 6 covðf 1 ; f 2 Þ varðf 2 Þ ⋮ 7
6 ⋮ 7 ⋮ ⋱ ⋮
6 7 6 7
6 7
6 7 6 7
6 fn 7
6 7
6 covðf 1 ; f n Þ
6 ⋯ ⋯ varðf n Þ 7
7
6 u1 7
6 7
6
6 varðu1 Þ covðu1 ; u2 Þ ⋯ covðu1 ; un Þ 7
7
6 u2 7 6 covðu1 ; u2 Þ varðu2 Þ ⋮ 7
X ¼ 6 7V X ¼ 6
6 7 7
6 7⋮ 6
6 ⋮ ⋱ ⋮ 7
7
6u 7
6 n7
6
6 covðu1 ; un Þ ⋯ ⋯ varðun Þ 7
7
6o 7
6 17
6
6 varðo1 Þ covðo1 ; o2 Þ ⋯ covðo1 ; on Þ7
7
6o 7
6 27
6
6 covðo1 ; o2 Þ varðo2 Þ ⋮ 7
7
4 ⋮ 5 4 ⋮ ⋱ ⋮ 5
on covðo1 ; on Þ ⋯ ⋯ varðon Þ
ð1Þ

In this work, the covariance terms between individual mass fractions, namely cov(fi,fj), cov(ui,uj) and cov(oi,oj) are all assumed to be equal to
0, which will lead to minimum variance estimates from the mass balance calculations. This implies that individual mass fractions in the particle
size distributions are assumed to be uncorrelated.
The solid mass split to underflow can be written as:

f −o
' (
du ¼ ð2Þ
u−o

Noting all estimators with the circumflex ^ character, balanced mass fractions must satisfy the following independent constraints:
) *
f^ −d^ u u
^ − 1−d^ u o^ ¼ 0 ð3Þ

∑f^ ¼ ∑u
^ ¼ ∑o^ ¼ 1

This mass balancing problem is nonlinear. However, given a value of d^ u , defined as the estimate of the solid mass flow split to the underflow,
the mass balancing becomes a linear optimization problem. The mass balancing solution used here consists of minimizing the least-squares
) *t ) *
criterion Jjd^ u ¼ X−X^ V −1
X X−X^ subject to the linear constraint AX^ ¼ B where A is a ((n + 3) × 3n) matrix and B is a (n + 3) vector from
Eq. (4).

1 1 ⋯ 1 0 0 ⋯ 0 0 0 ⋯ 0 1
2 3 2 3
60
6 0 ⋯ 0 1 1 ⋯ 1 0 0 ⋯ 0 7 7
617
6 7
60 0 ⋯ 0 0 0 ⋯ 0 1 1 ⋯ 1 7 617
d^ u 0 1−d^ u 0
6 7 6 7
61
A¼6 0 ⋯ 0 0 ⋯ ⋯ 0 7 607
B¼6 7 ð4Þ
d^ 0 1−d^ u
7
60
6 1 ⋯ 0 0 u ⋯ 0 ⋯ 0 7 7
607
6 7
4⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ 5 4⋮5
0 ⋯ ⋯ 1 0 0 ⋯ d^ u 0 ⋯ ⋯ 1−d^ u 0

The Lagrangian solution to the reconciliation problem has the following solution:

t −1
) *
t
X^ jd^ u ¼ X−V X A AV X A ðAX−BÞ ð5Þ
t −1
) *+ ) *
t
var X^ + ^ ¼ V −V A AV A AV
+
du X X X X

Since d^ u can take values in the [0,1] range only, it is a simple matter to scan the [0,1] range and find the value of d^ u that yields the lowest value
of the least-squares criterion.
 Once the best solution to the mass balance problem has been found, the estimated values of the mass fractions in the feed, underflow and
) *
overflow streams can be used to characterize the recovery curve to underflow through its estimate R ^ and variance var R ^ . By definition, the
estimate of the recovery to underflow is given by:
! ! !
^^
^¼Uu u
^ f^−o^ u
^
R ¼ d^ u ¼ ð6Þ
F^ ^
f ^
f u^ −o^ ^f

) *
Propagation of the variance of the estimated mass fractions f^; u ^ of the partition function.
^ in all 3 streams yields the variance var R
^ and o

0 ) *12 0 ) *12
) * ' (2 o^ f^−o^ ^ ^f −u
u ^ ^ ^2 ^ ^ 2 o^ f^−u^
A varðo^ Þ−2u o3 ðf −o^3 Þ cov f^; u u u
^ o^ f^−u^ f^−o^
u o ^ þ 2 ^3 ð 3 Þ cov f^; o^ −2 ð^2 Þð 4 Þ covðu
) * ) * ) *
^ ¼ 2
var R ^ ^ ^
var f þ @ ^ ^ ^ 2 A varðu
^Þ þ @ ^ ^ ; o^ Þ
^ f ðu^ −o^ Þ ^ ^ 2
f ðu −o Þ ^ ^ ^ f ðu −o Þ ^ ^ f ðu −o Þ^ ^ f ðu −o Þ f ðu −o Þ

ð7Þ

The authors used Matlab© to implement these equations and produce the results that are presented and discussed in the paper.

2.2. Particle size distribution error

Errors associated with the estimated performance of the hydrocyclone include the primary sampling of the feed, overflow and underflow
streams, the handling and sub-sampling stages that lead to the samples required for size analysis, and the size analysis itself. In this work, as
we are concerned with the fish-hook effect, which takes place in the fine particle range, we are dealing with particle size distributions measured
by laser diffraction.
The feed to a 100 mm diameter hydrocyclone was assayed using the same sampling procedure on 13 occasions. The experiments to which
these measurements belong have been reported elsewhere [11–13]. They concerned analysis of the flow pattern inside an operating
hydrocyclone. The feed sample was collected at the underflow before every test, by running the hydrocyclone at a sufficiently low mass flow
rate so that the entire feed stream would report to the underflow. By accumulating test conditions, the experimental work led to collecting
13 samples of the same hydrocyclone feed material. It is worth noting that the feed particles are composed of minus 200 μm high purity silica
particles. Details about the experimental set-up and operating conditions can be found in Davailles et al. [13].
Feed size distributions were measured with a Malvern Mastersizer 2000HydroS. Fig. 1 shows the cumulative feed size distributions for the 13
measurements.
&
The mean values of the mass fractions f"i and covariance matrix cov(fi°,fj°) were calculated for every one of the n particle size classes from the
13 feed size distributions measured with the laser size analyser (Fig. 1). The superscript ‘°’ is used to denote that it is a reference particle size
&
distribution that will be used later for analysis of individual data sets. The average values f"i and corresponding 95% confidence intervals are plot-
ted in Fig. 2, under Student's t-distribution assumption with 13-1 = 12 degrees of freedom.
With the experiments used here, it is believed that our measurement conditions meet the requirements for a precise measurement of particle
size distribution by laser particle sizing as discussed by Xu and Di Guida [14]. Firstly, feed sampling and sub-sampling protocol is the same for all
13 samples, so that it can be assumed that the masses of solid and water used for the laser size analyses are invariant. The feed particles are fine
(minus 200 μm) ground quartz particles (SIBELCO Millisil® C6) that assay over 98.8% SiO2. To eliminate the possibility that measured partition
function variations may be affected by size-density variations, particle density was measured experimentally by Helium pycnometry as a func-
tion of particle size. A 100 g feed sample was first microsieved into 6 size fractions: 0 × 5 μm, 5 μm × 10 μm, 10 μm × 20 μm, 20 μm × 40 μm,
40 μm × 63 μm and 63 × 200 μm. Measured density and corresponding standard deviation for each feed size fraction are given in Table 1. The
data confirm the invariance of particle density in the experiments. In addition, Fig. 3 shows an SEM photomicrograph of the silica particles
used in the hydrocyclone tests, which confirms that particle shape is relatively uniform across the particle size range.
Looking at the relative standard deviation (RSD) for the fi°, which is plotted in Fig. 4 as a function of particle size, it appears that the RSD values
are relatively uniformly distributed, with the exception of few extreme RSD values that occur on both ends, when the measured mass fractions
are insignificant, of the order of 10 −8 or so. We can conclude that measurement error for individual mass fractions can be approximated by an
average RSD value. Here, an average value of RSD of 8.4% gives a good description of the experimental RSD.

1
0.9
Cumulative distribution

0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
1 10 100
Particle size (µm)

Fig. 1. Cumulative feed size distributions, measured by laser diffractometry.

 0.07

0.06

0.05

Frequency distribution
0.04

0.03

0.02

0.01

0
1 10 100
-0.01
Particle size (µm)

Fig. 2. Distribution of mass fractions fi° with their 95% confidence interval estimated from analysis of 13 hydrocyclone feed size distribution samples.

Fig. 5 zooms on the data from Fig. 4 in the domain that does not include the highest RSD values. Even though the average value of RSD is 8.4%,
which may give a good description of the experimental RSD, there is still a noticeable variation of RSD as a function of particle size, which we
note RSD(fi°). Moreover, this variation appears quite marked in the 10 μm region where the fish-hook effect is known to occur.
To assess the level of quality of our data, we reviewed comparable data obtained by Xu and Di Guida [14] on laser diffractometry of glass
spheres. The RSD values obtained by the authors, as shown in Table 2, lie in the range of those obtained in this work.
The match between their RSD values and ours indicates that the laser sizing measurement errors controlled the error in our measurements,
which is an indirect confirmation of the quality of the sampling and handling steps used to sample the hydrocyclone feed.
The last issue that should be discussed is that of the calculation of the covariance matrix VX for the feed, underflow and overflow size distri-
butions for one particular experiment, for which fi, ui and oi have been measured. Analysis of the 13 size distributions through Pearson product–
moment correlation showed some degree of correlation between the measured mass fractions. These correlations were deemed low enough
however to assume that the measured mass fractions are only weakly correlated. This justifies the setting of the covariance terms to zero as
per Section 2.2.
The standard deviations of the n size fractions of the 3 streams are scaled directly from the size-dependent relative standard deviation (RSD)
of the reference feed size distribution f° using Eq. (8).

) &*
sðf i Þ sðui Þ sðoi Þ s f i
¼ ¼ ≡ "& for i; j ¼ 1; n ð8Þ
f"i u"i o" i fi

Although it is possible to scale the covariance terms in the same manner, it is recalled here that covariance terms in VX are set equal to zero.

2.3. Illustrative example of partition function estimation

The mass balance calculations that follow are done in 2 ways, in order to strengthen the analysis and assess its sensitivity to the proper
accounting for the errors associated with the particle size distribution measurements:
• The first set of calculations uses the average RSD of 8.4% calculated earlier, which is applied to all the mass fractions in all 3 stream size
distributions irrespective of particle size. 2 & 3 sðf Þ &

• The second set of calculations uses the size dependent RSD as per Eq. (8) and the size-dependent values RSD f i ¼ f i data from Fig. 4 instead.&
i

Fig. 3. SEM photomicrographs of silica particles.

Table 1
Validation of the invariance of particle density with particle size by He-pycnometry.

Size class Mass analyzed Density Standard deviation

(μm) (g) (kg m−3) (kg m−3)

0–5 4.8842 2631.8 2.9

5–10 5.9525 2642.6 1.5
10–20 6.6038 2641.4 0.9
20–40 7.2606 2640.8 2.5
40–63 8.6193 2641.2 0.5
63–200 9.1402 2641.2 0.2

The mass balance calculations were performed with the 2 methods above using a data set obtained with a 100 mm diameter HC100
hydrocyclone from Neyrtec that was fitted with an 18 mm diameter spigot and 33 mm diameter vortex finder. It was operated at a feed mass
flow rate of 2.6 × 10 −3 m 3 s −1 and an operating pressure of 10 5 Pa. The feed to the hydrocyclone was characterized in the previous section.
This illustrative example corresponds to a test with a feed solids concentration of 20 wt.%.
Key results from the mass balance calculations are given in Fig. 6.
Fig. 6a and b show that differences in the mass balance outputs are marginal only between the two calculations, although mass balancing is
clearly more effective with the size dependent RSD(fi°) than with the average RSD(fi°) value of 8.4%. This is best seen in Fig. 6c, which shows a very
clear minimum of the objective function with the second method. It is worth noting that both approaches yield slightly different estimates of the
solids split to underflow in this example, namely 54.72% with the average value of RSD against 52.96% with the size-dependent RSD.
There is no significant difference in prediction of the recovery curve to underflow with the 2 approaches, as seen in Fig. 6a; however careful
analysis shows a slightly narrower confidence interval with the 2nd method, confirming that it clearly is the best approach from a mass
balancing perspective. Nevertheless, for all practical purposes, these calculations establish that mass balancing around a hydrocyclone can be
done reliably with the average RSD value.
This section has also shown that it is a simple enough matter to estimate the confidence interval of the partition function of a hydrocyclone;
hence there is no justification for not estimating it in practice or not reporting it in written work. More importantly, estimation of the error
around the partition function is invaluable information for analyzing the performance of a hydrocyclone. It is this very information that is
used in Section 3 below to investigate the fish-hook effect.

3. Proof of the “existence” of the fish-hook effect interval of the partition function around the dip and critical points,
as defined by [6], excludes the possibility of size distribution measure-
The purpose of the previous section was to present and validate ment errors being the source of the fish-hook effect in this particular
the mass balance calculation that yields estimation of the partition experiment.
function and associated confidence interval, taking into account the Partition functions were calculated using the mass balancing pro-
particle size distribution measurement errors. For the sake of trans- cedure presented above for a test series that covered dilute to dense
parency, it is recalled that the same distribution RSD(fi°) was applied regime with the same hydrocyclone and feed size distribution as
to all 3 streams as per Eq. (8), excluding the possibility that sampling per Section 2. With the hydrocyclone set-up and operating conditions
and preparation errors may be different between the 3 streams. This reported previously, feed solids concentration was varied from 15 wt.%
is an assumption that could only be lifted by sampling all 3 streams to 50 wt.% in 5 wt.% increments. Raw measurements of particle size dis-
several times, which was not done here. However, given that the tributions can be found in Appendix A of Davailles' PhD dissertation [11].
tests were performed in a controlled laboratory environment with The estimated partition functions are plotted in Fig. 7a to h. Feed
recirculation of the underflow and overflow streams to a feed sump, it solids concentration is reported below each figure.
is not expected that sampling errors would differ significantly between Below a feed solids concentration of 10 wt.% and above 40 wt.%,
the streams. the partition function does not appear to exhibit the fish-hook effect
In the case of the illustrative example of Section 2.3., a fish-hook for this particular hydrocyclone setting and operating conditions.
effect was observed (see Fig. 6a). The narrowness of the confidence Between 15 wt.% and 35 wt.%, the fish-hook effect is clearly visible.

160%
16%
140%
14%
120%
12%
RSD (%)

100%
10%
RSD (%)

80%
8%
60%
6%
40%
4%
20%
2%
0%
1 10 100 0%
Particle size (µm) 1 10 100
Particle size (µm)
Fig. 4. RSD (%) values for particle size distribution assays fi° in all the measured size
classes. Fig. 5. RSD (%) values for particle size distribution assays fi° as a function of particle size.
Table 2 Within the range of solids concentration where the fish-hook was
Data calculated from laser diffraction measurements published by Xu and Di Guida visually observed, Table 3 gives the levels of significance of the fish-
(2003, Table 1 p. 147 and Table 2 p. 148).
hook effect as a function of feed solids concentration, which we define
Monomodal glass spheres Bimodal glass spheres as the level of significance of the difference between the estimated
d^ ¼ 65:27μm d^ 1 ¼ 46:47μm d^ 2 ¼ 96:43μm dip and critical point values.
RSD = 7.7% RSD = 13.6% RSD = 8.4% With the 20 wt.% and 25 wt.% feed solids concentration tests, the
fish-hook effect is the most significant at the confidence level of
99.8% and higher. At 15 wt.% and 30 wt.%, the significance of the
To further confirm the significance of the fish-hook effect, the values fish-hook effect is 66% and 62% respectively. At 35 wt.%, its significance
and 95% confidence interval of the dip and critical points are plotted is less than 3%, and above 35 wt.% it was not detected (See Fig. 7f,
in Fig. 8 as a function of feed solids concentration. g and h).

a) 1 1

0.9 0.9

0.8 0.8

0.7 0.7
Recovery to underflow

Recovery to underflow
0.6 0.6

0.5 0.5

0.4 0.4

0.3 0.3

0.2 0.2

0.1 0.1

0 0
1 10 100 1 10 100
particle size (µm) particle size (µm)

b)
1.4

1.2

1.0
Solid split du

0.8

0.6

0.4

0.2

0.0
1 10 100
Particle size (µm)

c)
10000

1000

100

10

1
0 0.2 0.4 0.6 0.8 1
Solid split dˆu

Fig. 6. Comparison between the mass balancing outputs using the average RSD of 8.415% and the size-dependent RSD.
 We have shown in this section that, under some operating condi- 4.2. Appearance of randomness and sporadicity of the fish-hook effect
tions, the fish-hook effect can occur at a significance level that excludes
the possibility of measurement errors being a probable cause. In other By refuting the possibility that measurement errors could be the rea-
words, the fish-hook effect is a real phenomenon whose origin goes son beyond the fish-hook effect, we have now established beyond
back to the physics of particle transport inside the hydrocyclone. doubt that the fish-hook effect is not a random effect. What could
then possibly explain that a number of researchers, like Nagaswarareo
[8] claim it is random and sporadic? One possible way of answering
4. Discussion this is to ask ourselves whether it could be possible for the fish-hook
effect to be present and for the observer not to observe it, giving an
4.1. Origin of the fish-hook impression of randomness and sporadicity.
In this paper, we have demonstrated the significance of the fish-
The conditions that will yield the fish-hook phenomenon to be hook effect based on analysis of the error of the estimate of the recov-
significant can be numerous, and it should not be inferred from the ery curve. It is often the case that recovery curves are calculated with-
data that were presented in the previous section that the fish-hook out any indication of their confidence bounds. Fig. 9 illustrates the
is most significant in the 20–30 wt.% region only. Indeed, many con- idea that it is possible, in principle, to obtain a high degree of variabil-
figurations and operating conditions can lead to the fish-hook effect. ity in fish-hook measurement if one does not account for the confi-
Further discussion on this is outside the scope of this article. dence bounds of the partition function. Starting with the earlier

a) 15wt% b) 20wt%
1 1

0.9 0.9

0.8 0.8

0.7 0.7
Recovery to underflow
Recovery to underflow

0.6 0.6

0.5 0.5

0.4 0.4

0.3 0.3

0.2 0.2

0.1 0.1

0 0
1 10 100 1 10 100
particle size (µm) particle size (µm)

c) 25wt% d) 30wt%
1 1

0.9 0.9

0.8 0.8

0.7 0.7
Recovery to underflow

Recovery to underflow

0.6 0.6

0.5 0.5

0.4 0.4

0.3 0.3

0.2 0.2

0.1 0.1

0 0
1 10 100 1 10 100
particle size (µm) particle size (µm)

Fig. 7. Estimated partition functions for a 100 mm diameter hydrocyclone with feed solids concentration in the range 15 wt.% to 50 wt.% in 5 wt.% increments.
 e) 35wt% f) 40wt%
1 1

0.9 0.9

0.8 0.8

0.7 0.7

Recovery to underflow
Recovery to underflow

0.6 0.6

0.5 0.5

0.4 0.4

0.3 0.3

0.2 0.2

0.1 0.1

0 0
1 10 100 1 10 100
particle size (µm) particle size (µm)

g) 45wt% h) 50wt%
1 1

0.9 0.9

0.8 0.8

0.7 0.7
Recovery to underflow
Recovery to underflow

0.6 0.6

0.5 0.5

0.4 0.4

0.3 0.3

0.2 0.2

0.1 0.1

0 0
1 10 100 1 10 100
particle size (µm) particle size (µm)

Fig. 7 (continued).

partition function whose fish-hook effect is significant at more than 95% confidence interval of the partition function. It is this error asso-
60% level, Fig. 9 shows that it is possible to draw a partition function ciated with measurement of the partition function which the authors
that shows a sharp fish-hook or no fish-hook inside the bounds of the believe can in part explain the appearance of randomness and
sporadicity of the fish-hook effect that has been reported by some
0.30 authors.
Recovery to underflow

This possible explanation for the apparent randomness and

0.25 fish-hook region no fish-hook
sporadicity appears even more when considering that these effects
0.20 could be obtained with a partition function that is measured in the
0.15 dip point (Pd) laboratory, under well controlled operating conditions. If we now
Critical point (Pc)
0.10

0.05 Table 3
Significance of the fish-hook effect as a function of solids concentration in the conditions of
0.00
the experiment.
10% 15% 20% 25% 30% 35% 40% 45% 50%
Feed solids concentration (wt%) Feed solids concentration 15 wt.% 20 wt.% 25 wt.% 30 wt.% 35 wt.%

Significance level 66.23% 99.99% 99.80% 62.30% 2.85%

Fig. 8. Comparison between dip point and critical point measurements.
 1
Notations
Symbols
0.9 du Solid mass flow split to the underflow
f Vector of mass fractions in the feed stream
0.8 u Vector of mass fractions in the underflow stream
o Vector of mass fractions in the overflow stream
0.7 n Number of particle size classes
Recovery to underflow

0.6
Subscripts
0.5 i,j Indices of particle size class

0.4
Superscripts
0.3 ^ Circumflex symbol denotes the estimate of a random variable
t Transpose of a vector or matrix
0.2

0.1 References
[1] R. Del Villar, J.A. Finch, Modelling the cyclone performance with a size dependent
0 entrainment factor, Minerals Engineering 5 (6) (1992) 661–669.
1 10 100 [2] J.A. Finch, Modelling a fish-hook in hydrocyclone selectivity curves, Powder Technology
particle size (µm) 36 (1983) 127–129.
[3] B.C. Flintoff, L.R. Plitt, A.A. Turak, Cyclone modeling – a review of present technology,
CIM Bulletin 69 (1987) 114–123.
Fig. 9. Illustration of the possibility of “missing” the fish-hook effect. [4] W. Kraipech, W. Chen, F.J. Parma, T. Dyakowski, Modelling the fish-hook effect of
the flow within hydrocyclones, International Journal of Mineral Processing 66
(2002) 49–65.
[5] A.K. Majumder, P. Yerriswamy, J.P. Barnwal, The “fish hook” phenomenon in
consider the same measurements in the plant, where operating con- centrifugal separation of fine particles, Minerals Engineering 16 (2003) 1005–1007.
ditions may fluctuate, the possibility of missing the fish-hook effect [6] A.K. Majumder, H. Shah, P. Shukla, J.P. Barnwal, Effect of operating variable on
for a particular measurement can be greater. shape of “fish-hook” curves in hydrocyclones, Minerals Engineering 20 (2007)
204–206.
[7] B. Wang, A.B. Yu, Computational investigation of the mechanisms of particle
5. Conclusions separation and “fish-hook” phenomenon in hydrocyclones, AICHE Journal 56 (7)
(2010) 1703–1715.
[8] K. Nageswararao, A critical analysis of the fish hook effect in hydrocyclone classifiers,
Based on the aforementioned results and discussion the following Chemical Engineering Journal 80 (2000) 251–256.
conclusions may be drawn: [9] S.K. Kawatra, T.C. Eisele, Causes and significance of inflections in hydrocyclone
efficiency curves, in: Advances in Comminution, The Society for Mining, Metallurgy
and Exploration, Littleton, CO, 2006, pp. 131–147.
• Estimation of confidence bounds of the partition function is of [10] C. Bazin, D. Hodouin, Importance of covariance in mass balancing of particle size
significant value for analysis of the performance of a hydrocyclone. distribution data, Minerals Engineering 14 (8) (2001) 851–860.
[11] Davailles, A., 2011. Effet de la concentration en solide sur les performances
• Partition functions estimated from carefully generated experimental
de separation d'un hydrocyclone (simulations numériques et experiences de
data set in a 100 mm diameter hydrocyclone followed by rigorous references), Ph.D. dissertation, The University of Toulouse, 197 pp.
statistical analyses of the data have completely ruled out the possibil- [12] A. Davailles, E. Climent, F. Bourgeois, Fundamental understanding of swirling flow
pattern in hydrocyclones, Separation and Purification Technology 92 (2011)
ity of measurement errors as the cause of the observed fish-hook
152–160.
effect. This conclusion implies that the fish-hook phenomenon is of [13] A. Davailles, E. Climent, F. Bourgeois, A.K. Majumder, Analysis of swirling flow in
physical origin. hydrocyclones operating under dense regime, Minerals Engineering 31 (2012)
• Statistical arguments have also been put forward to explain why in 32–41.
[14] R. Xu, A.A. Di Guida, Comparison of sizing small particles using different technologies,
some situations the fish-hook effect may appear to be random and Powder Technology 132 (2003) 145–153.
sporadic.