Professional Documents
Culture Documents
i=1
,
2
i
n
i=1
,
i
2
n
(1)
where ,
i
is the model prediction for the ith observation and n is the number of
observations. The residual sum of squares is calculated using the following equation:
Residual sum of squares =
n
i=1
(x
i
,
i
)
2
(2)
where x
i
is the experimental value for the ith observation. The sum of squares for
each term in the model is the difference between the residual sum of squares of the
model and the residual sum of squares of the model with the term excluded. The
E-test for the model indicates the level of signicance of the model prediction and
Table 2 The upper and lower limit of the variables
Factors Variable Low High
A Spigot diameter, mm 3 6.5
B Vortex nder diameter, mm 8 14.3
C Pressure, psi 10 30
D Dispersant, gm/kg 0 2
E Feed solids, % 10 27.12
D
o
w
n
l
o
a
d
e
d
b
y
[
T
a
t
a
S
t
e
e
l
]
a
t
2
2
:
3
3
0
5
J
a
n
u
a
r
y
2
0
1
4
HYDROCYCLONE FOR BENEFICIATION OF IRON ORE SLIMES 91
Table 3 ANOVA for grade of iron
Sum of Degrees of Mean E
Source squares freedom square value Prob > E
Model 127.63 6 21.27 41.96 -0.0001 Signicant
Residual 33.97 67 0.5070
Lack of t 24.88 63 0.3949 0.1737 0.9995 Not signicant
Pure error 9.10 4 2.27
the E-test for lack-of-t indicates the level of signicance of the model predicted
data not tting the observed data. For the model, the E-value is calculated using
the equation given below
E-test =
Estimate of model variance
Estimate of residual variance
(3)
and for the lack-of-t, E-value is dened as
E-test =
Estimate of residual-corrected-for-pure-error variance
Estimate of pure error variance
(4)
It was seen that the main effects, i.e., vortex nder diameter and dispersant are more
signicant compared to spigot diameter. Pressure did not seem to have signicant
effect and hence has been neglected. The quadratic effect of the spigot diameter was
more signicant than the vortex nder diameter. The interaction between the spigot
diameter and the dispersant was also found to be signicant.
The model equation obtained by regression analysis can therefore be written as
Fe grade% = 65.65069 1.82754 spigot diameter 0.11487
vortex nder diameter 2.60073 dispersant
+0.13419 spigot diameter
2
+0.019243
vortex nder diameter
2
+0.7392
spigot diameter dispersant (5)
The E-value of 41.95 at >99.99% condence level shows that the model is
signicant. The E-value of 0.17 at 99.95% condence level for lack-of-t implies that
there is no signicant lack of t compared to pure error. The comparison between
the experimental and estimated grade is shown in Figure 2.
The R
2
of 0.789 shows that the t is quite good. It should also be noted that
the error is not primarily due to the model but due to experimental error or noise,
since the lack-of-t shows that it is not signicant.
Similar analysis was carried out for recovery of iron values. The ANOVA
for recovery is shown in Table 4. The main effects, spigot diameter, vortex nder
diameter, pressure, dispersant, solid percent as well as square of dispersant and
interaction of spigot diameter, and vortex nder diameter have most signicant
effect on the recovery. The square of spigot diameter as well as two-parameter
D
o
w
n
l
o
a
d
e
d
b
y
[
T
a
t
a
S
t
e
e
l
]
a
t
2
2
:
3
3
0
5
J
a
n
u
a
r
y
2
0
1
4
92 S. MOHANTY AND B. DAS
Figure 2 Comparison of experimental and predicted grade.
interaction of spigot diameter and solid percent, vortex nder diameter and pressure,
and dispersant and solid percent were less signicant. Three and higher parameter
interaction were not signicant and hence have been excluded. The E-value of 76.83
for the model implies that the model is signicant at >99.99% condence level. The
E-value of 15.07 for lack-of-t at 0.83% condence level implies that the lack-of-t
is quite signicant. This implies that the error in prediction is primarily due to the
model compared to experimental error or noise. The model equation obtained by
regression analysis can therefore be written as
Fe recovery % = 39.17641 +11.89891 spigot diameter 0.16651
vortex nder diameter +0.325334 pressure
+13.08777 dispersant 0.76444 feed solids
0.04704 spigot diameter
2
7.40765 dispersant
2
Table 4 The ANOVA for recovery of iron
Sum of Degrees of Mean E
Source squares freedom square value Prob > E
Model 10603.6 11 963.964 76.83 -0.0001 Signicant
Residual 777.93 62 12.55
Lack of t 774.39 58 13.35 15.07 0.0083 Signicant
Pure error 3.55 4 0.8863
D
o
w
n
l
o
a
d
e
d
b
y
[
T
a
t
a
S
t
e
e
l
]
a
t
2
2
:
3
3
0
5
J
a
n
u
a
r
y
2
0
1
4
HYDROCYCLONE FOR BENEFICIATION OF IRON ORE SLIMES 93
0.58395 spigot diameter vortex nder diameter
0.00017 spigot diameter feed slidos
0.01044 vortex nder diameter pressure
0.17443 dispersant feed solid (6)
Figure 3 shows a comparison between the experimental and estimated
recovery. The R
2
value of 0.9316 implies that the model t is very good. The normal
plots of the difference between the observed and predicted values of grade and
recovery, i.e., the residuals are shown in the Figures 4 and 5, respectively. For a
good model, the residual should be normally distributed, i.e., the points should lie
on a straight line. Hence, it can be seen that the errors for both recovery and grade
are well distributed.
The models were then used to get the best set of design and operating
parameters so as to maximize grade and recovery. Since the recovery decreases
with an increase in grade, the minimum acceptable grade and recovery were xed
along with other criteria for optimization as shown in Table 5. Optimization of the
process parameters were carried out using built-in optimizer of Design Expert, that
uses NelderMead multidimensional pattern search technique. The ten best possible
optimal solutions are shown in Table 6. From Table 6, it can be seen that maximum
grade of 65.44% can be obtained at 55% recovery and maximum recovery of 56.41%
can be obtained with a grade of 64.32% by varying various design and process
parameters.
Figure 3 Comparison of experimental and predicted recovery.
D
o
w
n
l
o
a
d
e
d
b
y
[
T
a
t
a
S
t
e
e
l
]
a
t
2
2
:
3
3
0
5
J
a
n
u
a
r
y
2
0
1
4
94 S. MOHANTY AND B. DAS
Figure 4 Normal plot of residuals for grade.
Figure 5 Normal plot of residuals for recovery.
D
o
w
n
l
o
a
d
e
d
b
y
[
T
a
t
a
S
t
e
e
l
]
a
t
2
2
:
3
3
0
5
J
a
n
u
a
r
y
2
0
1
4
HYDROCYCLONE FOR BENEFICIATION OF IRON ORE SLIMES 95
Table 5 Criteria for parameter optimization of the hydrocyclone
Parameter Minimum Maximum
Spigot diameter, mm 3 6.5
Vortex nder diameter, mm 8 14.3
Pressure, psi 10 25
Dispersant, g/kg 0 2
Feed solids, % 10 25
Fe grade underow, % 60
Fe recovery underow, % 50
Table 6 Iron grade and recovery under the best optimizing conditions
Spigot Vortex nder Feed Fe-grade Fe recovery
diameter, diameter, Pressure, Dispersant, solids, underow, underow,
Number mm mm psi gm/kg % % %
1 6.50 12.58 25.00 2.00 10.01 65.44 55.00
2 6.50 12.51 24.89 2.00 10.18 65.42 55.02
3 6.50 12.37 20.88 2.00 10.00 65.37 55.03
4 6.50 11.98 12.78 2.00 10.00 65.23 55.00
5 6.50 11.78 18.39 2.00 11.79 65.16 55.00
6 6.50 11.65 12.57 2.00 11.27 65.11 55.00
7 6.50 11.50 25.00 2.00 10.00 65.07 59.48
8 6.49 10.97 25.00 1.83 18.85 64.52 54.94
9 6.50 14.23 10.19 1.23 10.00 64.43 55.00
10 6.50 8.77 10.00 2.00 19.96 64.32 56.41
CONCLUSIONS
Iron values from iron ore slimes can be recovered by manipulating the design
and operating variables of the hydrocyclone with proper choice of dispersant. From
the statistical data analysis, it is seen that the effect of spigot diameter, vortex nder
diameter, and the interaction of spigot diameter and dispersant play signicant
role on the grade of iron. The model equation obtained by regression analysis
for grade of iron is signicant at >99.99% condence level. For the recovery of
iron, the effect of spigot diameter, vortex nder diameter, pressure, dispersant,
solid percent, square of dispersant, and interaction of spigot and vortex nder
diameter are most signicant. The equation obtained for recovery of iron by
regression analysis was signicant at 99.99% condence level. The comparison of
experimental and estimated grade and recovery of iron by the model equations are
quite good. Under optimal conditions predicted by the model, an iron recovery
of 59.5% can be obtained with 65.07% grade of iron. The values for the variables
predicted by the model are 11.50mm for vortex nder diameter, 25psi for pressure,
2gms/kg of dispersant, and 10% solids concentration. It will, therefore, be possible
to obtain around 65% Fe with around 60% recovery by using these combinations in
hydrocyclone for iron ore slimes being generated at Indian iron ore mines.
D
o
w
n
l
o
a
d
e
d
b
y
[
T
a
t
a
S
t
e
e
l
]
a
t
2
2
:
3
3
0
5
J
a
n
u
a
r
y
2
0
1
4
96 S. MOHANTY AND B. DAS
ACKNOWLEDGMENT
The authors are thankful to Director, Institute of Minerals and Materials
Technology, Bhubaneswar for his kind permission to publish this paper.
REFERENCES
Anderson, M. J. and Whitcomp, P. J., 1996, Optimize your process-optimization efforts.
Chemical Engineering Progress, 92(12), pp. 5160.
Das, B., Ansari, M. I., and Mishra, D. D., 1993, Effective separation of Barsua iron ore
slimes using hydrocyclone. Mineral & Metallurgical Processing, 10(1), pp. 5255.
Das, B., Prakash, S., Mohapatra, B. K., Bhaumik, S. K., and Narasimahan, K. S.,
1992, Beneciation of iron ore slimes using hydrocyclone. Mineral & Metallurgical
Processing, 9(2), pp. 101103.
Karr, C. L., Donald, A., Stanley, D. A., and McWhorter, B., 2000, Optimization of
hydrocyclone operation using a geno-fuzzy algorithm. Computer Methods in Applied
Mechanics and Engineering, 186(9), pp. 517530.
Kawatra, S. K. and Eisele, T. C., 2006, Causes and signicance of inections in
hydrocyclone efciency curves. Advances in Comminution, Littleton, CO: The Society
for Mining, Metallurgy and Exploration, pp. 131147.
Lynch, A. J. and Rao, T. C., 1968, The operating characteristics of hydrocyclone classiers.
Indian Journal of Technology, 6, pp. 106114.
Montgomery, D. C., 1997, Design and Analysis of Experiments, 4th Ed., New York: Wiley.
Myers, R. H. and Montgomery, D. C., 2002, Response Surface Methodology, 2nd Ed., New
York: Wiley.
Plitt, A. J., 1976, A mathematical model of the hydrocyclone classier. CIM Bulletin,
69(776), pp. 115123.
Srivastava, M. P., Pan, S. K., Prasad, N., and Mishra, B. K., 2001, Characterisation and
processing of iron ore nes of Kiriburu deposits of India. International Journal of
Mineral Processing, 62(2), pp. 93107.
Udaya Bhaskar, K., Rama Murthy, Y., Ravi Raju, M., Tiwari, S., Srivastava, J. K., and
Ramakrishnan, N., 2001, CFD simulation and experimental validation studies on
hydrocyclone. Mineral Engineering, 20(1), pp. 6071.
D
o
w
n
l
o
a
d
e
d
b
y
[
T
a
t
a
S
t
e
e
l
]
a
t
2
2
:
3
3
0
5
J
a
n
u
a
r
y
2
0
1
4