By
SARDAR ALI
Ph.D. Scholar
By
SARDAR ALI
University Registration No. 87-03
SUPERVISOR
CO-SUPERVISOR
JUNE 2007
ii
In the Name of
Allah,
Most Merciful and Compassionate the
Most Gracious and Beneficent
Whose help and guidance I always solicit at
every step, at every moment.
iii
DEDICATED
To my Wife and Children
iv
Thesis entitled
MATHEMATICAL MODELS FOR THE EFFICIENCY OF FLOTATION
PROCESS FOR NORTH WAZIRISTAN COPPER
Submitted by
MR. SARDAR ALI
Accepted by the Division of Science and Technology, University of Education, in
partial fulfillment of the requirements for degree of Doctor of Philosophy in
Mathematics with specialization in Flotation Process.
Director
External Examiner
Supervisor
Member
Member
Date: ___________
ABSTRACT
The objectives of this research are to analyze empirically the effects of different
explanatory variables on recovery and grade of copper from ore found in North
Waziristan and to develop mathematical models for the enrichment of copper in
Pakistan.
This study is based on the primary data from flotation process experiment for
enrichment of copper. Seven variables were studied in experiments. The variable were
type and dosage of collector (X1g/ton) pH (X2), depressant sodium cyanide (X3 g/ton)
sulfidizer Na2S(X4g/ton), frother dosage (X5 g/ton), pulp density (X6 w/v) and
conditioning time (X7 minute) and consists of 31 observations. Flotation process
parameters were studied to concentrate the copper content of chalocopyrite the North
Waziristan copper ore. Mathematical models were developed using various model
selection procedures. Regression parameters were estimated by applying Ordinary
Least Squares (OLS) method for regression analysis and adopted general to simple
modeling procedure. In this study we found that the variables X1, X3, X4 and X6 of
equation (5.57) are statistically significant and concluded that an increase in these
variables there is increase in recovery of copper.
Maximum grade were obtained from equation (6.65) the combined significance
variable X1, X3, and X7.
vi
ACKNOWLEDGEMENTS
All praise and thanks for Almighty Allah, Who has given me power to
complete this report successfully.
I am extremely grateful to Dr. Ghulam M. Mustafa, Vice-Chancellor,
Education University Lahore for his expert guidance, incisive and scholarly advice
and very useful suggestion which were of great help in making this report.
I am also greatly thankful to my supervisor Dr. Mir Asad Ullah, COMSAT,
Abbottabad, for his constant help at each stage, with out which I probably would not
have been able to execute this project with such professional excellence.
Heartedly thanks are due to my Co-supervisor, Prof. Dr. Muhammad Mansoor
Khan, Dean N-W.F.P, University of Engineering & Technology, Peshawar.
My sincere thanks goes to Prof. Dr. Mian Izhar ul Haq, Director, Ph.D
Programme, Education University, Lahore, for his timely help, encouragement and
cooperation.
I am unable to find words for paying thanks to my wife and my children who
were so helpful and extending warm co-operation whenever called upon.
Last but not the least, I would like to thanks Mr. Syed Sajid, Alias (Doctor),
Supervisor, Words Masters, U.O.P, for compiling this stuff in such a short period of
time.
SARDAR ALI
vii
TABLE OF CONTENTS
Abstract
Acknowledgements
vi
List of Tables
List of Figures
xi
CHAPTER NO. 1:
INTRODUCTION
1.1
Introduction
1.2
1.3
1.4.
1.5
1.6
1.7
1.8
1.9
CHAPTER NO.2:
REVIEW OF LITERATURE
10
CHAPTER NO. 3:
EXPERIMENTS
18
3.1
Previous Work
18
3.2
19
3.3
19
3.4
Uses of Copper
20
3.5
World Occurrences
21
3.6
22
3.7
25
3.7.1
25
3.7.2
25
3.7.3
Punjab Province
25
3.7.4
Baluchistan Province
26
3.8
26
3.9
28
viii
3.9.1
Shinkai Area
28
3.9.2
Degan area
28
CHAPTER NO. 4:
METHODOLOGY
29
4.1
30
4.2
Estimation Techniques
32
4.3
33
4.3.1
33
4.3.2
38
4.3.3
Estimation of 2
39
4.3.4
40
4.3.5
41
4.3.6
Studentized Residuals
41
4.3.7
42
4.3.8
42
4.3.9
44
4.4
Collection of Copper Ore Samples and their Analysis for Pilot Scale Studies 45
4.5
46
4.5.1
46
4.5.2
pH value
46
4.5.3
Depressant
47
4.5.4
Sulphidizer (Na2S)
47
4.5.5
Frothers Dosage
47
4.5.6
Frother
47
4.5.7
48
4.5.8
Flotation time
48
51
5.2
General Description:
51
5.3
53
5.3.1
53
5.3.2
55
ix
5.3.3
55
5.3.4
55
5.3.5
56
5.3.6
56
5.3.7
56
5.4
59
5.5
72
5.6
73
5.7
74
5.8
75
5.9
81
5.9.1
85
5.9.2
86
5.10
87
89
90
94
6.1
94
6.2
97
6.3
110
6.4
Forward Selection
110
6.5
Backward Elimination
112
6.6
112
6.7
120
6.8
123
6.8.1
Statistical Significance
123
6.8.2
Sample Size?
123
6.9
124
6.10
125
CHAPTER 7: CONCLUSION
References
Appendix (1-7)
127
129
137
LIST OF TABLES
Table No.
Title
Page
Table 1:
22
Table 2:
23
Table 3:
54
Table 4:
59
Table 5:
82
Table 6:
Analysis of Variance
82
Table 7:
84
Table 8:
85
Table 9:
88
Table 10:
89
Table 11:
Tests for skewness, kurtosis and Jarque bera for four variables
90
Table 12:
Bin Frequency
90
98
Table 14:
111
Table 15:
119
Table
13:
Seven variables
Table 16
120
Table 17
Analysis of Variance
121
xi
LIST OF FIGURES
Figure No.
Figure 1:
Figure 2:
Figure 3:
Figure 4:
Figure 5:
Figure 6:
Figure 7:
Figure 8:
Figure 9:
Figure 10:
Figure 11:
Figure 12:
Figure 13:
Figure 14:
Figure 15:
Figure 16:
Figure 17:
Title
Effect o1f collector (NaPX) on recovery of copper
Effect of pH on recovery of copper
Effect of depressant (NaCN) on recovery of copper
Effect of sulfidizer (Na2S) on recovery of copper
Effect of frother (pine oil) on recovery of copper
Effect of pulp density on recovery of copper
Effect of conditioning time on recovery of copper
(a). Linear, (b) Logarithmic (c) quadratic (d) power (e)
exponential and (f) two straight-line models fitted to the
recovery of copper data from five levels of collector type
and dosage in the flotation process.
(a) Linear, (b) Logarithmic (c) quadratic (d) power (e)
exponential (f) and two straight-line models fitted to the
recovery of copper data from five levels of pH of pulp in
the flotation process.
(a) Linear, (b) Logarithmic (c) quadratic (d) power (e)
exponential (f) and two straight-line models fitted to the
recovery of copper data from five levels of depressant in
the flotation process.
(a) Linear, (b) Logarithmic (c) quadratic (d) power (e)
exponential (f) and two straight-line models fitted to the
recovery of copper data from five levels of sulphidizer in
the flotation process.
(a) Linear, (b) Logarithmic (c) power (d) and exponential
models fitted to the recovery of copper data from three
levels of frother dosage in the flotation process.
(a) Linear, (b) Logarithmic (c) quadratic (d) power (e)
exponential (f) and two straight-line models fitted to the
recovery of copper data from four levels of Pulp density in
the flotation process.
(a) Linear, (b) Logarithmic (c) quadratic (d) power (e)
exponential (f) and two straight-line models fitted to the
recovery of copper data from four levels of conditioning
time in the flotation process.
Effect of sodium cyanide (X3) on the recovery of copper.
Effect of sodium sulphide (X4) on the recovery of copper
Copper recovery (YR) response surface for sodium cyanide
(X3) and sodium sulphide (X4).
Page
57
57
57
57
58
58
58
63
65
66
67
68
69
71
77
77
78
xii
Figure 18:
Figure 19:
Figure 20:
Figure 21:
Figure 22:
Figure 23:
Figure 24:
Figure 25:
Figure 26:
Figure 27:
Figure 28:
Figure 29:
Figure 30:
Figure 31:
Figure 32:
Figure 33:
Figure 34:
Figure 35:
Figure 36:
Figure 37:
Figure 38:
79
83
84
86
89
91
91
96
96
96
96
96
96
97
101
103
104
105
106
107
109
xiii
Figure 39:
Figure 40:
Figure 41:
Figure 42:
Figure 43
Figure 44
Figure 45
Figure 46
114
114
116
117
121
122
122
126
CHAPTER 1
INTRODUCTION
1.1
Introduction
Mineral processing is the art and science of processing ores to separate
Extensive
flotation test work was carried out to investigate effects of various process variables
on recovery (YR) and grade (YG) of copper.
Sulfidizer dosage; depressant levels, frother dosage, pulp density and conditioning
time were investigated in flotation tests. The results of the pilot scale studies showed
that the copper content in the ore can be upgraded from 0.9 % to 22-25 % in a staged
cleaning flotation with recoveries up to 80%. The grade can be further enhanced by
improving the machine efficiency and conducting more research on reagents.
The important information on some flotation results of copper were obtained
from the Department of Mining Engineering, N.W.F.P University of Engineering and
Technology, Peshawar, and used to develop mathematical models for grade and
recovery of copper.
1.2
It is the only course of action available for the improvement in the system
It will help to improve the process of extraction of copper from the copper ore.
The study of mathematical models, simulation and optimization are important
Increase efficiency.
Decision support
Knowledge management
Technology transfer
optimization in control for more than 32 years. All major technological innovations
involve information technology and mathematical modeling, and apparently
computerized mathematical model play an increasingly decisive role within
engineering sciences, i.e. within industrial production, within planning and
economics, within mineral processing. Mathematical modeling activities are aimed at
methodologies enabling one to deal with todays ever increasing quantities
information.
1.3
To utilize the 122 million ton of copper ore of North Waziristan area.
1.4
1.
2.
3.
4.
chalcopyrite is such that it can be efficiently concentrated by the froth flotation from
associated gangue minerals. Flotation process parameters were studied using
chalcopyrite Copper ore of North Waziristan to obtain a copper concentrate suitable
for further metallurgical treatment. The important flotation variables examined were,
collector, depressant, pH, frothers, Sulphidizer (Na2S), pulp density and conditioning
time. By stage wise optimization of flotation variables, copper were upgraded from
0.9% to 10% and 20% in roughing stage and to as high as 22% in a cleaning stage
with recoveries up to 80 to 90% in experimental work done by the Department of
Seven important explanatory variables, e.g. type and dosage of collector (X1gms/ton) PH (X2),
depressant sodium cyanide (X3gms/ton) sulfidizer Na2 S (X4gms/ton) frother dosage (X5gms/ton), pulp
density (X6) and conditioning time (X7 minute).
1.5
conductor and is used in electrical networks, various equipments and weapons. The
United States, the worlds largest consumer (1999), uses between 2.5 and 3.0 million
tons of copper annually. Most wires and electrical equipment are made of pure copper
and considerable alloys of copper such as brass and bronze. The brasses are Cu Zn
alloys (55%-99% Cu, 45%-1% Zn) and the bronzes are Cu Sn Zn (88% Cu, 10%
Sn and 2% Zn). There are also Ni, Al, and steel alloys of Cu; minor special alloys
utilize arsenic; beryllium, cadmium, chromium, cobalt, iron, lead, magnesium,
manganese and silicon. Copper sulphide deposits of North Waziristan vary in grade
from 0.3% to as high as 1.0%. Due to its low grade it cannot be directly subjected to
metallurgical treatment for producing blister copper. Pakistan is still meeting its
requirement through import from other countries. Successful development of
mathematical model will provide optimum parameters for the enrichment of copper in
the final product. In this way it will save cost for further experimentation and time to
achieve similar objectives.
1.6
1.7
The metal has many uses and ranks fifth amongst the metals in tonnage consumed.
God has blessed Pakistan with abundant copper ore and its occurrences have been
reported throughout the country. However, the occurrences at Saindak and Ricodak in
Baluchistan and North Waziristan Agency in NWFP are of more importance. The
survey conducted by Federally administrated tribal areas (FATA) development
corporation has confirmed a minimum of 122 million tons of reserves of copper ore in
Boya-Datta Khel area (about 40km from Miran Shah), having copper contents better
than that found at Saindak at some places and in some layers.
Thus an extensive study of North Waziristan copper ore was carried out by the
Department of Mining Engineering, N.W.F.P, University of Engineering &
Technology, Peshawar, through a research proposal sponsored by Board of Advanced
Studies and Research (BOASAR). The laboratory evaluations of raw ore were made
in Phase-1 of the project and in order to confirm these evaluations, a study of flotation
process by a single stage pilot plant was carried out in Phase-II. These studies have
generated sufficient data for constructing mathematical models for the processes.
1.8
extract copper from raw ore, experimentation by systematic or by hit and trial
procedures takes a lot of time and costs enormous amount of money. The standard
scientific way to improve and increase the efficiency of the flotation process for
enrichment of copper ore is to develop a mathematical model for the process. It
should be remembered that in some cases mathematical modeling is the only course
of action available for the improvement in the system. Once we successfully construct
mathematical model for a process, it can be improved and used in future for any
alteration for improvement in the process. Thus mathematical models to be developed
will reduce the extent of further experimentation to achieve certain desired objectives.
These will help to improve the process of extraction of copper from the copper ore
and will save sizeable amount of money for the country.
1.9
recommendations.
10
CHAPTER 2
REVIEW OF LITERATURE
11
is the minimization of the absolute sum of the square of deviation at a given time
between observed in calculated recovery.
Wills (1986) studied simple nodal sensitivity analysis in complex circuit
analysis were found by using matrices and statistical techniques. The researcher
worked to develop a best-fit material balance model. The method described makes use
of the minimum number of sampled streams and analysis of only one component,
such as metal assay or dilution ratio, on each stream one involved in a unit process.
He found that plant flow sheet was reduced to a series of nodes, where process either
join or separate. Simple nodes have either one input and two outputs (a separator ) or
two inputs and one output ( a junction ).
Munn (1998) investigated that metal recovery in mineral processing plants is
often linearly correlated with feed or concentrate grade, particularly in flotation. This
correlation can be used to analyze the data form plant trials in which two operating
conditions are being compared, such as different reagent regimes or circuit
configurations. The method involves the statistical comparison of the two linear
recovery grade regression lines corresponding to the two operating conditions.
Although not as efficient as a formal experimental design, the method can be used
where such designs are impractical, or in the analysis of historical data.
Khan (1999) studied flotation process parameters to concentrate the copper
content of chalcopyrite, the North Waziristan copper ore, in pilot-scale to obtain a
copper concentrate suitable for further metallurgical treatment. The important
flotation parameters, e.g. type and dosage of collector, dosage of depressant, and
frother and conditioning time for collector were examined. During stepwise
optimization of flotation parameters, the copper content was upgraded from 0.9% to
12
2
reduction of data. A best model was selected based on R adj
t-ratio and the number of
data points considered. The regression co-efficient of the best models were significant
at the P = 0.001 level.
observations.
Sripriya et al. (2002) examined the kinetic model based on time recovery data,
which uses the extra dimension of rate and has been in vogue since time immemorial
13
for scaling up of laboratory data. The air flow number and the froth number were used
as a basis for scale up. The performance of the froth flotation circuit, an efficiency
parameter (co-efficient of separation c s) was used. The yield from the flotation circuit
improved, the froth ash reduced and the rejects ash went up. Various empirical and
kinetic models were evaluated.
Sripriya developed regression equations for predicting the combustible
recovery ash recovery and Ks for combustibles and ash. The effects of three most
important reagents for coal flotation namely sodium meta silicate, collector (kerosene)
and frother were studied using 23 full factorial design. The regression models were
developed using factorial experiment data to quantify the effect of sodium meta
silicate, collector and frother and to predict grade and recovery of combustible
material for different reagent conditions. The addition of sodium meta silicate
increased the recovery without affecting the grade significantly. The MIBC addition
reduce the surface tension at the liquidvapor interface, which results in the
production of finer bubble size distribution and thus improves flotation rates and
recovery values. However, a finer bubble size is tribution also increases water
recovery, which results in a greater recovery of certain able ash bearing particles and
thus degradation of the product grade. The interaction between OH group of MIBC
and hydrated mineral matter improves floatability of high ash coal particles and
degrades the product grade further. The negative effect of kerosene and MIBC
interaction on both grade and recovery could be due to the recovery of high ash coal
particles in preference to low ash coal particles. The highest possible grade of product
is 94.19% combustibles with 25.33% recovery. A product with 91.11% combustibles
14
grade at 95.58% recovery was obtained at 0.1 g/kg sodium silicate, 0.4 g/kg collector
and 0.075 g/kg frother from the coal fines tested.
Ziyadanogullari (2003) worked on flotation of oxidized copper ore obtained
from Ergani Copper Mining Company in Turkey. The ore contained 2.03% copper,
0.15% cobalt and 3.73% sulfur. An effective processing method has not been found to
recover copper and cobalt from this ore, which has been stockpiled for 40-45 years in
a idled plant. It was established that recovery of copper and cobalt from this ore with
hydrometallurgical treatment is not economical, so using flotation to increase the
concentration of copper and cobalt was chosen. When flotation of the oxidized copper
ore was performed under standard operating conditions in the plant, good results were
not obtained. Because of this, the flotation of samples obtained from sulfurized
medium containing different ratios of H2S+ H2O gases was done under the same
conditions. Following flotation, it was seen that copper, cobalt and sulfur present in
the medium were concentrated. In this solution, concentration of copper and cobalt
were found five times higher than normal level.
Elemental sulfur produced by chloride leaching of sulfide ores or concentrates
contains selenium and tellurium usually too high to be used in various industrial or
agricultural uses. The sulfur in the leaching residue can be upgraded to 90% in grade
by froth flotation and the sulfur concentration can be followed by sulfur purification
and selenium and tellurium removal. The sulfur in the leaching is in a form of discrete
particles with a size range of 5 to 10 microns. The sulfur particles tend to agglomerate
in the pulp and hence mechanically entrap gangue minerals. With sodium silicate as
the dispersant as well as the depressant for siliceous material, a sulfur concentrate of
90% in grade and 90% in recovery can be obtained with a single-stage froth flotation.
15
16
17
18
CHAPTER 3
EXPERIMENTS
3.1
Previous Work
The commercial copper deposits occur in variable sizes. However, the ores
containing 0.3% and more copper are deemed feasible for exploitation on commercial
scale. With the state of the art mineral processing techniques the ores with lower
grades can be economically beneficiated as well. Undoubtedly large area of Federally
Administered Tribal Area (FATA) abounds in mineral resources. The survey
conducted by FATA development corporation has confirmed a minimum of 122
million tons of inferred reserves of copper ore varying in depth upto 30m in BoyaDatta Khel area about 40 kms from Miran-Shah. The average content of this copper
ore is 0.3865%. The copper content increases with depth and at places it is 0.90%
which is better than that found at Saindak (Baluchistan). This low grade raw copper is
of little value unless it is enriched to a higher grade concentrate. The Department of
Mining Engineering through a research proposal (Beneficiation of North Waziristan
Copper Ore) sponsored by Board of Advance Studies and Research (BASAR) carried
out laboratory evaluation of raw ore. Based on the encouraging results of phase-I,
further work on the project was considered necessary. In the phase II of the project, it
was proposed to install a flexible single stage pilot plant for flotation process to study
the laboratory results at the pilot scale. The pilot plant, locally fabricated, has been
19
3.2
3.3
20
located at longitude 69o 55/ 06// and latitude 32o 57/ 09// N on the right bank of Tochi
River. It lies at a distance of 19 km from Miranshah, the Agency Headquarters and the
local business center. Miranshah is fairly connected to the down districts. It is
accessible by about 270 km of Peshawar-Bannu-Miranshah mettalled road and also
through Peshawar-Tal-Miranshah road. Peshawar is connected by road, rail and air to
Islamabad, the distance being 167 km, similar communication links are available to
Bannu(District Headquarters), that falls at a distance of 61 km to the east of Miran
Shah. Bannu is also connected by about 141 km length of metalled road to D.I.Khan
(Divisional Headquarters) in the south. Tal and Bannu are connected by metalled
roads. The former is also connected to Kalabagh by a metalled road.
3.4
Uses of Copper
The tremendous growth in the use of copper is indicated by the fact that of the
total world production of copper during the last 100 year, about 80% was mined in the
last 25 years and more than one half of it in the last 12 years copper consumption by
major countries and regions is given in table-1. Annual world production ranges
around 20 million metric tons of metallic copper. In spite of the significant number of
closures in United States and Canada, Western world copper mine production rose
3.8% due to projects that came on stream in 1999. In Chile, the Pelambres project
came on stream, with its main impact to be felt during 2000. Similarly, Collahuasi
started up in late 1998 and reached full capacity in 1999, as did Andina expansion and
Escondidas SX-EW operation. In Australia the Olympic dam expansion started up, as
did the Cuajone expansion in Peru, then Indonesias Batu Hijau mine began to
produce (Enrique, 2000). Copper ranks fifth among the metals in tonnage consumed.
It has a variety of uses and the important one is in electrical supply, use and
21
manufacturing industries due to its good conductivity that gives it an advantage over
most metals. It is extensively used in communications equipment including cables and
television transmitters and receivers. The second important use of copper is in
construction worm, particularly in plumbing and hardware and decorative purposes. It
is also a substance used in non-electrical industrial applications such as alloy with
nickel for tubing used in sea water desalination plants. It is used in heat exchangers,
pollution control, and liquid waste disposal. Automobile radiator cores are made of
copper; it is also used in air conditioners, heaters, gas and oil line, and bearings and
bushings.
Military uses of copper are fourth in rank, and the price usually go up during
periods of military spending. This is one of the incremental uses which can rapidly
increase consumption. Coinage, jewellery, chemicals, pigments, brass and bronze
wares and a multitude of minor uses also demand copper in variable amounts.
Copper is essential for plant growth, if copper content falls below 10ppm in
soils, good growth is not possible. On the other hand, if a large amount of copper is
present in the soil it is toxic to some plants.
3.5
World Occurrences
There are hundreds of copper minerals and dozens of settings for copper
deposits. By far the most important mineral is chalcopyrite and the large portion of
this mineral and of copper production comes from the porphyry deposits.
The term porphyry refers to a rock which has an intergrowth of distinctly large
and small crystals. Porphyries are considered to have intruded as molten rock or
magma from depth of ten to hundreds of kilometers. To form the texture of porphyry,
22
it should have approached to within about 3km of the surface before crystallizing as
rock. The texture of mixed coarse and fine crystals is brought to indicate fairly rapid
cooling. Copper deposits the world over can be classified according to the nature of
the deposits.
3.6
have been revised upward based on new information from official country sources.
Revisions to other countries were based on updated tabulations of resources reported
by companies or individual proprietors. Table 2 shows the world mine production and
reserves of Copper.
1994
1995
1996
1997
1998
1999 (e)
Western Europe
3341
3388
3345
3536
3751
3710
Africa
123
117
115
118
110
115
Japan
1375
1415
1480
1441
1255
1260
Other Asia
1833
1955
2126
2240
2148
2420
Canada
199
190
218
225
245
270
United States
2560
2534
2621
2790
2905
2935
Latin America
503
511
619
734
828
820
Oceania
148
174
170
166
161
160
10082
10283
10691
11250
11403
11690
Total
Annual Growth (%)
23
Other Asia includes China, Taiwan, India South Korea, Thailand, Malaysia,
and Philippines.
North America
There are some greatest concentrations of copper in Arizona and Cordilleran
parts of the United States, Canada and Mexico and includes all the well known North
American porphyry coppers and a host of other famous districts. All the ores are
associated with felsic types of intrusions. There is another very productive copper
province in Montana at Butte. Other areas where copper can be found include the
Appalachian, the fruitful take Superior district, and Cascadian Coast Range belt
extending from Yukon Territories through northern British Columbia to the state of
Washington.
Canada
Copper deposits extend from Manitoba to New Brunswick includes the
Hudson Bay, Sodbury, Noranda, heath Steele, Kidd Creek, and other deposits.
Table 2: World Copper Mine Production (in 1000 tons of Cu) (Enrique, 2000)
Area
1994
1995
1996
1997
1998
1999 (e)
Western Europe
304
323
290
317
305
260
Africa
647
618
585
577
563
491
Asia
674
806
798
810
1054
1034
Australasia
614
598
660
625
730
871
Latin America
2853
3198
3884
4280
4658
5316
North America
2399
2534
2537
2544
2495
2211
Total
7491
8077
8754
9153
9805
10183
(e)
Estimate
24
South America
Andean copper belt is the most renown in this region. It extends from Chile to
Panama and includes large deposits like Chuquickamata, Braden, Potrerillos, El
Salvador, Cerro Colorado, Rio Blanca, Toquepala, Cerro de Pasco deposits, and many
more. The copper deposits found in these areas are normally associated with
monzonitic intrusives.
Central Africa
The central Africa province constitutes the most concentrated copper belt in
the world and includes the most productive mines of Zambia and adjacent Zaire.
These are strata bound deposits where the metals precipitated from sea with the
sediments.
Several new copper porphyries have been discovered in New Zealand, Fiji,
New Hebrides, Buganville, British Solomon Island protectorate, the territory of
Papua, New Guinea, and West Irian. Some of the copper producing deposits is
Penguna, Ok Tedi, Frieda, and the high grade deposits of Carstenz.
Other copper belts include Uralian province of Russia, the outer Japanese
Island arc, Spain Portugal (Rio Tinto), Bor in Yugoslavia, Mansfeld in Germany,
Outokumpu in Finland, and Boliden in Sweden.
In Australia there are various copper centers such as Mount Lyell, Mount
Morgan, Mount Isa, Cobar, Tennant Creek, and Mount Oxide.
25
3.7
Columbia, Panama, New Guinea, Fiji, New Idria, Brazil, Puerto Rico, New
Brunswick, Philippines, Solomon Islands, North western Brazil (John, 1984).
3.7.1 Copper of Occurrences in Pakistan
Several copper occurrences have been reported in Pakistan. They are available
in numerous geological settings and contain a variety of copper minerals. However,
the occurrences at Saindak (Baluchistan) and North Waziristan (FATA) are of much
importance. Minor occurrences have been reported from various other places of the
country.
Investment oriented study on Minerals and Mineral based Industries, Expert
advisory cell, Ministry of Industries & Production, Govt. of Pakistan. April, 2004
3.7.2 Gilgit Agency (Northern Areas)
Copper minerals have been located in quartz veins in the northeastern regions
of the area. Similarly chalcopyrite has been reported in alluvial sands in Indus, Gilgit,
Nagar and Hunza rivers.
3.7.3 Punjab Province
Small occurrences of copper have also been reported in Northern Punjab at
Kattha, Mussa Khel and Nilawahan Gorge in salt range. In these areas oxide copper
minerals are found in sandstone beds with malachite and cuprite as the major copper
minerals. Up till now these findings has no economic value. No detail exploration
work has been carried out to access the potential of these deposits.
26
3.8
the minerals referred to as Fools Gold because of its bright golden color, but it is
brittle, dissolves in acid and has a dark green streak. It is distinguished from pyrite by
ease of scractching, and by copper tests. The color is slightly more yellow than that of
pyrite or is often tarnished in brilliant iridescent hues, which is also called peacock
copper ore. Pyrite will frequently show striated cubes or pyritohedra, whereas
chalcopyrite, if not massive, has the characteristic sphenoidal or disphenoid crystals.
Chalcopyrite is the primary minerals, which by alteration and successive
enrichment with copper produces the series starting with chalcopyrite and going
through bornite (Cu5SFeS4), covellite (CuS), chalcocite (Cu2S), and ending rarely as
native copper (Cu). Its structure is so closely related to that sphalerite that it forms
intergrowths with mineral, and isolated free-growing crystals perched on crystals of
27
sphalerite are all parallel. The same face on all the chalcopyrite gives simultaneous
reflections. (It Sparkles) from the Greek words chalkos, copper and pyrites, strike
fire
Following are the various physical properties of chalcopyrite ore
Composition:
Class:
Sulfides
Group:
Chalcopyrite
Crystal system:
Tetragonal
Fracture:
Hardness:
3.4-4
Specific gravity:
4.2
Luster:
Metallic
Streak:
Dark green
Cleavage:
Color:
Transparency:
Opaque
Associated Minerals:
Chalcopyrite is usually massive, but crystals are also common. Often they are
large and the faces usually are somewhat uneven or may have striations on most
crystal faces. Chalcopyrite is often tarnished in brilliant iridescent hues. Spheroidal
28
crystals are common. Also common are disphenoid crystals, which are like two
opposing wedges that resemble a tetrahedron. Crystals are sometimes twinned and can
also be botryoidal.
On charcoal, chalcopyrite fuses to magnetic black globule, touched with HCl tints
flame with blue flash. Solution with strong nitric acid is green; ammonia precipitates
red iron hydroxide and leaves a blue solution.
3.9
29
CHAPTER 4
METHODOLOGY
Data from the following seven variables of flotation process were used to
develop mathematical models for recovery and enrichment of copper from North
Waziristan copper ore.
S. No.
Name of Variable
Level used
1.
Propylxanthate
2.
pH
3.
Sodium Cyanide
4.
Sodium Sulphide
5.
Frother (Pineoil)
25, 46, 70
6.
Pulp Density
7.
Conditioning time
30
4.1
unknown parameters that will minimize the sum of squares of errors (or residuals)
where errors are defined as the difference between observed values and the
corresponding values predicted or estimated by the fitted model equation.
The parameters values thus determined, will give the least sum of the squares
of errors and are known as least squares estimates. The method of least squares that
gets its name from the minimization of a sum of squared deviations, is attributed to
Gauss (1777-1855) some believed that the method was discovered at the same time by
Legendre (1952-1833).
Laplace (1749-1827) and other mar Kovs name is also mentioned in
connection with its further development this method is used as one of the important
methods of estimating the population parameters.
The best regression line is the one, which minimizes the sum of the squares of
the vertical deviations between the observed values yi and the corresponding values yi
(hat) predicted by the regression model y i o 1 xi ei (4.1)
31
The set of observations (xi, yi), i = 1,2,...n, where yi are the values of y
randomly drawn from a population and xi and fixed values. Then the observed yi may
be expressed in a linear form of the population parameters as
yi xi i
y i o 1 x i ei (4.2)
.. (4.3)
(4.4)
32
Models that are more complex in structure than equation (4.4) may often still
be analyzed by multiple linear regression techniques.
4.2
Estimation Techniques
The following techniques and test statistics were use in this study.
1.
2.
3.
Adjusted R-squared
4.
5.
F-statistics
6.
7.
Correlation Matrix
8.
9.
Histograms
10.
11.
12.
13.
14.
15.
33
4.3
The method of least squares can be used to estimate the regression coefficients
in eq. (4.2) suppose that n>k observations are available, and let yi denote the i-th
observed response and xij denote the i-th observation or level of regressor xj. The data
is given in table 4.1. Assuming that the error term in the model has E() = 0, Var
() = 2 and that the errors are uncorrelated.
Response
Regression
x1, x2, xp
Y1
x11
x12
x1p
Y2
x21
x22
x2p
YN
xn1
xn2
xnp
Assume that the regressor variables x1, x2, .xp, are fixed. (i.e., mathematical
or nonrandom) variables, measured without error. All the simple linear regression
models of our results are valid for the case where the regressors are random variables.
This is certainly important, because when regression data arises from an observational
study, some or most of the regressors will be random variables. When the data result
from a designed experiment.
34
It is more likely that the xs will be fixed variables. When the xs are random
variables it is only necessary that the observations on each regressor be independent
and that the distribution not depend on the regression coefficients (the s) or on 2.
When the testing hypotheses or constructing confidence intervals, Assume that the
conditional distribution of y given x1, x2, .. xp be normal with mean
yi = 0 + 1x1 + 2x2,+ 3x3,-----------+ pxp and variance 2.
jxij, + i (4.5)
j 1
i = 1,2,.n
The least-square function is
S (0, 1, ,p) =
i 1
i 1
j 1
i2 ( y i 0 j x ij ) 2 (4.6)
0 ,1 ,....... p
n
P
2 y i 0 j x ij 0 (4.7)
i 1
j 1
0 ,1 ,....... p
n
P
and
S
j
35
i 1
i 1
i 1
i 1
n 0 1 x i1 2 x i 2 .......... p xip y i
i 1
i 1
i 1
i 1
i 1
0 x i1 1 x i21 2 x i1 x i 2 .......... p x i1 x ip x i1 y i
i 1
i 1
i 1
i 1
i 1
1 x11 ..........x1P
2
1
y2
1 x .........x
.
.
.
2P
, Y , ,
X 21
: :
:
.
.
.
.
.
1 x n1 .........x nP
y n
p
p
36
S() =
i 1
2
i
( y X )( y X )
2 X y 2 X X 0
which become
X X X y (4.10)
Equation (4.10) are the least-squares normal equations. To solve the normal
equations, multiply both sides of (4.10) by the inverse of X/X. Thus the least-square
estimator of is;
( X X ) 1 X y (4.11)
provided that the inverse matrix (X/X)-1 exists. The (X/X)-1 matrix will always exist if
the regressors are linearly dependent, that is, if no column of the X matrix is a linear
combination of the other columns.
37
The matrix form of the normal equation (4.10) is identical to the scalar form
(4.9).
The normal equation can be written as
n
x i1
i 1
.
.
.
.
n
x iP
i 1
xi 2
i 1
n
x
i 1
i1
i 1
i 1
n
n
x i 2 x i1 .......... x iP x IP
i 1
i 1
.
.
.
.
.
.
.
.
n
n
x i 2 x iP ........ x iP x IP
i 1
i 1
xi1
x i1
.
.
.
.
n
x
i 1
i1
x iP
IP
y i
i 1
0
n
xi 1 y i
1
i 1
=
.
.
.
.
P
x iP y i
i 1
y x 0 j x j
j 1
38
y X X ( X X ) 1 X y Hy (4.12)
The n x n matrix H = X (X/X)-1X/ is usually called the hat matrix. It maps the
vector of observed values into a vector of fitted values. The hat matrix and its
properties play a central role in regression analysis.
The difference between the observed value yi and the corresponding fitted
value y i is the residual ei y i y i . The n residuals may be conveniently written in
matrix notation as:
e=y
(4.13a)
There are several other ways to express the vector of residuals e that will
prove useful, including
e y X y Hy ( I H ) y (4.13b)
39
i 1
i 1
SS Re s ( y i y i ) 2 ei2 e e
substituting e = y - X , we have
SSRes = (y - X )/(y-X )
= y/y /X/y y/X + /X/X
= y/y 2 /X/y + /X/X
since X/X = X/y, this last equation becomes
SSRes
40
SS Re s
(4.15)
nk
2 MS Re s (4.16)
In the simple linear regression case, this estimator of 2 is model dependent.
4.3.4 Test for Significance of Regression:
41
If the null hypothesis is true, then SSR/2 follows a i2 distribution, which has
the same number of degrees of freedom as number of regressor variables in the
model. Also SSRes/2 ~ X n2k 1 and that SSRes and SSR are independent. By the
definition of an F statistic.
F0
SS R / p
MS R
SS Re s /(n p 1) MS Re s
E(MSRes) = 2
E(MSR) = 2 +
* X c/ X c *
p 2
42
points which are outliers among the independent variables, the two concepts are
related.
In the case of studentized residuals, large deviations from the regression line
are identified since the residuals from a regression will generally not be independently
distributed (even if the disturbances in the regression model are), it is advisable to
weight the residuals by their standard deviations.
4.3.7 Test Statistic for Skewness
Let r*=(r(1),,r(T)) be the vector of OLS residuals. Since the mean of the
OLS residuals is zero, the test statistic for skewness can be written as:
SK(r*) =
1
T
r (T )
t 1 SER
T
The assumption that the errors all have the same distribution (identical
distributions) also needs to be tested. The basic lesson is this: we must make sure that
our assumptions about the error term are valid. One assumption we have already
discussed earlier is that of homoscadasticity. We now study violations of this
assumption in greater detail.
43
Whenever one of our assumptions fails in a regression model, we say that we have
a misspecified model. There are generally three goals in misspecification analysis:
1.
2.
3.
common distribution N(0,2). The assumption that all errors have the same variance is
called homoscadasticity. What happens when this assumption is violated?
It is
possible that the t-th error (t) has variance 2 (t). When the errors have different
variances, we say that the errors are heteroscadastic. In this situation, the OLS
estimates continue to be unbiased. They are also consistent - this means that as the
sample size increases to infinity, the OLS estimates will converge to the true
parameters. However, the SER for the regression, and the SEs for the parameters
(and therefore the t-statistics) are incorrectly computed and hence misleading.
We have discussed how OLS analysis is damaged by the presence of
heteroscedasticity. Next we consider the issue of how we can detect if
heteroscedasticity is present. In the type of case under discussion, where
heteroscedasticity increases with X(t), it is relatively easy to detect. One simple test is
the Goldfeld-Quandt test. This consists of splitting the sample into two halves, and
estimating the regression separately on both halves. Let SER(1) and SER(2) be the
Standard Error of Regression for the first half and the second half of the data set
respectively. If the ratio SER(1)/SER(2) is close to 1 then the SEs on both halves of
the data set are similar. If the ratio is far from 1 than the two SEs are different (which
44
is what we expect in the case of heteroscedasticity). Next, the issue is: how do we find
the critical values? That is, at what point can we say that the ratio is too far from 1 for
the null hypothesis of equal variances in both halves to be valid?
The Goldfeld-Quandt statistic is based on the ratio of variances (not SEs):
GQ = [SER(2)/SER(1)]2
4.3.9 The t-statistic - Normal Approximation
45
4.4
46
4.5
4.5.1
Collectors, some time called promoters, are organic substance. Collectors are
the organic chemicals, which are able to selectively adsorb onto the mineral surface,
and render the mineral surface hydrophobic. Commercial collectors should ideally
possess the following character ristics:
1.
2.
3.
4.
5.
They provide higher selectivity, being expected to adsorb only one specific
minerals.
Control of solution pH is one of the most widely used methods, for regulating
complex separations in flotation. The depressant action of alkalis results from an
increase in the rate of dissolution or oxidation of the mineral surface. Pulp pH value
plays a significant role in flotation through its influence both on mineral flotability
and reagent function.
47
Sodium sulphide is a major modifier used for the activation of oxide minerals.
It is a salt produced from reaction between strong alkali and weak acid.
Na2S + H2O
The function of frother is to disperse air into fine bubbles and to form a stable
froth. Frothing action is thus due to the ability of the frother to adsorb on the air water
interface because of its surface activity and to reduce the surface tension. Thus
stabilizing the air bubble interface. Bubbles undue merge or breakage is harmful to
flotation through destroying bubble-particle attachment and dropping the collected
valuables back to pulp before the froth carrying them is removed.
Bubbles strength i.e. their stability is required which can be realized by
increasing aeration and frother proper frother. Frother acts entirely in liquid phase and
does not influences, the state of the mineral surface.
48
1.
2.
Decrease the rate at which air bubbles rise in the flotation machine to the
surface of the pulp.
3.
An increase in pulp density the recovery and grade curves show an upward
trend due to hindered setting conditions upto 30%.
4.5.8 Flotation time (X7):
49
cell during a set of experiments. In order to examine the dependencies between the
state of the process and the separated image variables, a set of experiments was
carried out. The behaviour of the froth state variables imply that these image variables
are useful in the control and monitoring of the complex process. In process industry
one of the most frequently use method for separation of valuable substances from the
waste is flotation. Especially in the mining industry flotation is widely used. Flotation
means the use of air bubbles to concentrate small mineral grains from the ore
suspension. Relatively heavy mineral grains attach themselves to the air bubbles due
to surface chemical phenomena and are transported to the froth. Concentrated froth is
collected for further treatments as it flows over the shoulder into the gutter.
Information of the state of a flotation process can be seen from the appearance of the
froth layer on the top of the flotation cell. Operators at the flotation plant shave
applied this information in manual control of the flotation process for ages. They use
the colour, speed and shape information of the froth layer. Development of image
processing methods has made it possible to acquire real-time numerical data of the
froth for control purposes. The possibility of utilizing image information in mineral
flotation has aroused a lot of interest in the mineral engineering community. Up to
now, however, the research has been mainly concentrating on image analysis
problems, i.e. how to extract a certain image feature from the froth images. To really
investigate whether the image data can be utilized in the monitoring and control of
flotation process or not, a set of experiments was designed, carried out and analyzed.
As a result information would be obtained about the appearance of the froth and the
behaviour of a flotation process in different control circumstances. Experiments were
carried out in the zinc flotation circuit of the flotation plant at Pyhsalmi, Finland, in
October 1998.
50
51
CHAPTER 5
MODELS BUILDING
5.1
maximize the efficiency of flotation process for the recovery (YR) and grade (YG) of
copper ore. The response variables YR and YG were regressed on seven variables X1,
X2, . X7 as shown in the following model:
So that Y = 0 + 1X1 + 2X2 +.+7X7 +
(Where Y = YR and Y = YG)
The above is a multiple linear regression model because more than one
regressor is involved when Xi are called the independent variable or response
variables. The adjective linear is employed to indicat that the model is linear in the
parameters 0, 1, 7 not because YR and YG is a linear function of the Xis. An
important objective of regression analysis is to estimate the unknown parameters in
the regression model. This process is also called fitting the model to the data.
5.2
General Description:
52
which generated the data. If the assumptions do not hold, the inferences may not be
reliable. The limitation is often ignored by the applied workers who fail to identify
crucial assumptions or subject them to any kind of empirical testing. In such
circumstances, using statistical procedures may compound the uncertainty. Therefore
developing models by checking some of the assumptions.
To fit a model, to a set of data, one or both of the following methods are employed.
1)
Start with the general model for YR (the dependent variable) that contains all
available independent variables, then simplify the model by eliminating the
independent variables that do not contribute significantly to the variability in
the dependent variable;
2)
53
with a total of 31 different treatments. The experiments were carried out in the
Department of Mining Engineering, NWFP University of Engineering and
Technology, Peshawar.
The data consist of values for grade and recovery as affected by the different
values of seven flotation process variables; collector NaPX (X1), PH(X2), depressantNaCN (X3), sulphadizer Na2S (X4), frother pine oil (X5), pulp density (X6), and
conditioning time (X7). The whole data are made up of seven sub groups. In each sub
group only one of the process variables was varied and others were kept constant.
5.3.1 Effect of variation in collector dosage, NaPX (X1).
The first experiment was conducted to investigate the effect of five levels of
collector, sodium propylxanthate, on the recovery of copper, while keeping all the
other six variables constant. Effect of collector dosage on recovery is given in Figure
1. The trend of recovery presented in Figure 1 shows that with an increase in level of
sodium propylxanthate up to 200 g/ton of feed, there was a corresponding increase in
recovery of copper; with further increase in the level of collector there was a slight
decrease in recovery. This decrease might be due to the nonspecific absorption of
collector by the gangue particles. Therefore, 200g of Prophylxanthate per ton feed is
the optimum level for recovery of copper.
54
Table 3:
X1
X2
X3
X4
X5
X6
X7
YR
50
11
75
30
10
32
100
11
75
30
10
32.7
150
11
75
30
10
38
200
11
75
30
10
41.5
250
11
75
30
10
41
200
10
75
30
10
35
200
10.3
75
30
10
36
200
11
75
30
10
42
200
11.58
75
30
10
45.2
200
12
75
30
10
40
200
11.58
10
75
30
10
49
200
11.58
15
75
30
10
50
200
11.58
20
75
30
10
55
200
11.58
25
75
30
10
63
200
11.58
30
75
30
10
60
200
11.58
25
10
75
30
10
60
200
11.58
25
30
75
30
10
63
200
11.58
25
40
75
30
10
67
200
11.58
25
50
75
30
10
73
200
11.58
25
60
75
30
10
59.4
200
11.58
25
50
25
30
10
70
200
11.58
25
50
46
30
10
74
200
11.58
25
50
70
30
10
71.56
200
11.58
25
50
46
15
10
56
200
11.58
25
50
46
25
10
64
200
11.58
25
50
46
30
10
75
200
11.58
25
50
46
35
10
68
200
11.58
25
50
46
30
10
69.77
200
11.58
25
50
46
30
13
73.3
200
11.58
25
50
46
30
16
68
200
11.58
25
50
46
30
18
64
55
56
with further increase in the level of suplfidizer, there was a drastic decrease in
recovery of copper. This decrease may be attributed to depressive action of sodium
sulphide.
5.3.5 Effect of Variation in Frother Pine Oil (X5)
Four levels of pulp density were studied in the sixth experiment to investigate
the effect of variation in pulp density (X6) on the recovery of copper. The first five
variables were kept at the levels giving highest recoveries in the previous experiments
while variable seven, conditioning time was kept at 10 minute. Figure 6 shows that
with an increase in pulp density the recovery showed an upward trend due to hindered
setting condition up to 30% Pulp density. Recovery decreased with further increase in
pulp density beyond 30%. The recovery showed marked decrease with the highest
pulp density of 35% due to the entrapped fine slime particles.
5.3.7 Effect of conditioning time (X7)
57
the optimum levels which were found in the previous experiments. The conditioning
time was varied between 10 to 18 minutes. Effect of conditioning time on recovery of
copper is presented in Figure 7. It is evident from the Figure 7 that 13 minutes
conditioning time was optimum, since beyond this, recovery markedly decreasing due
to dissolution of copper xanthate ions in the equilibrium system.
(a)
(b)
45
50
% Recovery
% Recovery
40
35
30
25
20
15
45
40
35
30
25
20
10
50
100
150
200
250
300
10
12
13
pH
PropylXanthate (g/ton)
(c)
80
% Recovery
70
% Recovery
11
60
50
40
60
40
20
30
0
0
10
15
20
25
30
35
20
40
60
80
Sodium Cyanide(g/ton)
58
(f)
(e)
80
70
% Recovery
% Recovery
90
70
50
30
10
0
20
40
60
80
60
50
40
30
20
10
Pineoil (g/ton)
% Recovery
80
70
60
50
40
30
20
0
10
15
20
20
30
Pulp Density (%wt/vol)
40
59
5.4.
variable, four to six models were fitted for each of the seven variables. The models
fitted were linear, logarithmic, quadratic, power, exponential and two straight-line for
each of the seven independent variables; quadratic model was not used for variable
five because it will lead to over fitting and will pass through all the three points.
Minitab statistical analysis was used to fit the forty mathematical models in
single predictor variable. The fitted equations and co-efficient of determination are
given in Table 4 below.
Table 4: Mathematical Models Involving One Predictor Variable For Recovery
Of Copper By Flotation.
YR = 0.0536X1 + 29,
R2 = 0.8897
(5.1)
YR = 6.5631Ln(X1) + 5.0807,
R2 = 0.8619
(5.2)
R2 = 0.9053
(5.3)
YR = 15.289X10.1805,
R2 = 0.8678
(5.4)
YR = 29.541e0.0015X1,
R2 = 0.8884
(5.5)
R2 = 0.9503
(5.6)
R2= .5757
(5.7)
YR = 42.465Ln(X2) - 61.994,
R2 = 0.5940
(5.8)
60
R2 = 0.8568
(5.9)
YR = 2.9069X21.0898,
R2 = 0.6161
(5.10)
YR = 13.468e0.0979X2,
R2 = 0.5975
(5.11)
R2 = 0.9868
(5.12)
YR = 0.7X3 + 41.4
R2 = 0.8210
(5.13)
YR = 12.69Ln(X3) + 18.277
R2 = 0.8123
(5.14)
R2 = 0.8330
(5.15)
YR = 28.043X30.2311
R2 = 0.8302
(5.16)
YR = 42.746e0.0127X3
R2 = 0.8360
(5.17)
YR = 37.8 + 0.94X 6X
R2 = 0.9176
(5.18)
YR = 0.0897X4 + 61.07
R2 = 0.0938
(5.19)
YR = 3.2386Ln(X4) + 53.2
R2 = 0.1652
(5.20)
R2 = 0.4242
(5.21)
YR = 54.228X40.048
R2 = 0.1634
(5.22)
61
YR = 61.129e0.0013X4
R2 = 0.0889
(5.23)
R2 = 0.9222
(5.24)
YR = 0.0314X5 + 70.376
R2 = 0.1233
(5.25)
YR = 1.8784Ln(X5) + 64.781
R2 = 0.2327
(5.26)
YR = 65.03X50.0264
R2 = 0.2391
(5.27)
YR = 70.347e0.0004X5
R2 = 0.1283
(5.28)
YR = 0.76X6 + 45.8
R2 = 0.6694
(5.29)
YR = 18.224Ln(X6) + 7.0532
R2 = 0.7168
(5.30)
R2 = 0.7773
(5.31)
YR = 25.854X60.2881
R2 = 0.7542
(5.32)
YR = 47.704e0.012X6
R2 = 0.7040
(5.33)
R2 = 0.9258
(5.34)
YR = -0.7931 X7 + 80.07
R2 = 0.5153
(5.35)
62
YR = -9.8924Ln(X7) + 94.811
R2 = 0.4352
(5.36)
R2 = 0.9533
(5.37)
YR = 100.95X7-0.1463
R2 = 0.4463
(5.38)
YR = 81.159e-0.0117X7
R2 = 0.5269
(5.39)
R2 = 0.9988
(5.40)
Information presented in Figure 8, reveals that all the models gave good fit to
the data. Two straight line model had the highest R2 of 0.9503, followed by quadratic
model, the simple linear regression model also gave good fit as it had the third highest
R2. Two straight line model is best because the X-maximum calculated from the
quadratic model, 448 gram per ton, is much out of the range used in the study. The
recovery of copper increased at the rate of 0.0536 per gram increase in sodium
propylxanthate considering linear model. The equation for two straight lines show that
recovery increased at the rate of 0.0676% per one gram increase in collector up to
200g/ton, there after the recovery remained the same upto 250 g/ton of collector.
63
(a)
(b)
50
% Recovery
% Recovery
50
40
30
y = 0.0536x + 29
R2 = 0.8897
20
40
90
140
190
40
30
y = 6.5631Ln(x) + 5.0807
R2 = 0.8619
20
10
0
240
290
100
PropylXanthate (g/ton)
50
40
40
% Recovery
% Recovery
(d)
50
30
2
10
0
0
100
200
30
20
y = 15.289x0.1805
R2 = 0.8678
10
0
300
ProphylXanthate
50
100
150
200
250
ProphylXanthate
(e)
(f)
50
% Recovery
50
% Recovery
300
PropylXanthate
(c)
20
200
40
y = 29.541e0.0015x
R2 = 0.8884
30
20
40
90
140
190
240
ProphylXanthate (g/ton)
290
40
Y = 27.6 + 0.0676X - 3.5X'
R2 = 0.9503
30
20
40
90
140
190
240
Sodium propylxanthate (g/ton)
Figure-8: (a) LINEAR (b) LOGARITHMIC (c) QUADRATIC (d) POWER (e)
EXPONENTIAL AND (f) TWO STRAIGHT-LINE MODELS FITTED TO THE
RECOVERY OF COPPER DATA FROM FIVE LEVELS OF COLLECTOR
TYPE AND DOSAGE IN THE FLOTATION PROCESS.
300
64
Two straight lines gave the best fit followed by quadratic model gave best fit
followed by power model, and the other three models also gave good fit to the data
for copper recovery as affected by pH of pulp in flotation process (Fig. 9) the X-max
calculated from the quadratic equation show that 11.3 pH of the pulp will result in
maximum recovery, however, the two straight lines model show that pH of 11.6 is the
joining point with the highest recovery.
(a)
(b)
50
40
y = 3.8154x - 2.2376
R2 = 0.5757
30
% Recovery
% Recovery
50
40
y = 42.465Ln(x) - 61.994
R2 = 0.594
30
20
20
9.8
9.8
10.3
10.8
11.3
10.3
11.8
10.8
11.3
11.8
pH
pH
(d)
(c)
50
40
y = -5.1611x2 + 117.21x - 622.2
R2 = 0.8568
30
% Recovery
% Recovery
50
40
y = 2.9069x1.0898
R2 = 0.6161
30
20
20
9.8
9.8
10.3
10.8
pH
11.3
11.8
10.3
10.8
pH
11.3
11.8
65
(f)
(e)
50
45
% R eco very
% Recovery
50
40
35
y = 13.468e0.0979x
R2 = 0.5975
30
40
Y = -33.52 + 6.8165X - 8.2751X'
R2 = 0.9868
30
25
20
20
9.8
10.3
10.8
11.3
pH
11.8
9.8
10.3
10.8
11.3
11.8
pH
Figure-9: (a) LINEAR (b) LOGARITHMIC (c) QUADRATIC (d) POWER (e)
EXPONENTIAL (f) AND TWO STRAIGHT-LINE MODELS FITTED TO THE
RECOVERY OF COPPER DATA FROM FIVE LEVELS OF PH OF PULP IN
THE FLOTATION PROCESS.
The two straight lines model gave best fit. All the other five models gave good
fit to the data for copper recovery as affected by depressant (Fig. 10) the differences
in the R2s of the models are very small. Though the original data points show
maximum recovery at 25, the quadratic X-max is beyond the range used, the
maximum recovery may be obtained around the last 2 data points i.e. 25 and 30 g/ton
of depressant. The two straight lines model gave in figure 10 show a joining point at
25 g/ton of depressant with decrease on both sides thus 25g depressant per ton is
optimum for recovery of copper.
66
(b)
70
70
60
60
% Recovery
% Recovery
(a)
y = 0.7x + 41.4
R2 = 0.821
50
40
50
y = 12.69Ln(x) + 18.277
R2 = 0.8123
40
30
30
8
12
16
20
24
28
32
12
16
24
28
32
Sodium Cyanide(g/ton)
Sodium Cyanide(g/ton)
(c)
(d)
70
% Recovery
70
% Recovery
20
60
50
y = -0.0143x 2 + 1.2714x + 36.4
R2 = 0.8330
40
60
50
y = 28.043x0.2311
R2 = 0.8302
40
30
30
8
12
16
20
24
28
32
12
16
20
24
28
32
Sodium Cyanide(g/ton)
Sodium Cyanide(g/ton)
(e)
(f)
70
% Recovry
% Recovery
70
60
50
y = 42.746e0.0127x
R2 = 0.836
40
60
50
Y = 37.8 + 0.94X - 6X'
R2 = 0.9176
40
30
30
8
12
16
20
24
28
32
12
16
20
24
28
Sodium Cyanide (g/ton)
Sodium Cyanide(g/ton)
Figure-10: (a) LINEAR (b) LOGARITHMIC (c) QUADRATIC (d) POWER (e)
EXPONENTIAL (f) AND TWO STRAIGHT-LINE MODELS FITTED TO THE
RECOVERY OF COPPER DATA FROM FIVE LEVELS OF DEPRESSANT
IN THE FLOTATION PROCESS.
32
67
max calculated from the quadratic equation is 38 g/ton. However, the two straight
lines model show that the joint point at 50 g/ton of sulphidizer will give maximum
recovery of copper.
80
70
60
50
40
30
20
(b)
y = 0.0897x + 61.07
R2 = 0.0938
80
70
60
50
40
30
20
% Recovery
% Recovery
(a)
y = 3.2386Ln(x) + 53.21
R2 = 0.1652
5 10 15 20 25 30 35 40 45 50 55 60 65
5 10 15 20 25 30 35 40 45 50 55 60 65
(d)
80
80
70
70
60
50
40
30
% Recovery
% Recovery
(c)
60
50
y = 54.228x0.0489
R2 = 0.1634
40
30
20
20
5 10 15 20 25 30 35 40 45 50 55 60 65
5 10 15 20 25 30 35 40 45 50 55 60 65
(f)
(e)
80
70
60
50
40
30
20
70
0.0013x
y = 61.129e
R2 = 0.0889
% Recovery
% Recovery
80
60
50
40
30
5 10 15 20 25 30 35 40 45 50 55 60 65
Sodium Sulphide (g/ton)
20
5
10 15 20 25 30 35 40 45 50 55 60 65
Sodium Sulphide (g/ton)
Figure-11: (a) LINEAR (b) LOGARITHMIC (c) QUADRATIC (d) POWER (e)
EXPONENTIAL (f) AND TWO STRAIGHT-LINE MODELS FITTED TO THE
RECOVERY OF COPPER DATA FROM FIVE LEVELS OF SULPHIDIZER
IN THE FLOTATION PROCESS.
68
Power and logarithmic models gave fair fit to the copper recovery data on
frother dosage from flotation process. The other two models gave poor fit which is
given in (Figure-12). It seems that maximum recovery will be obtained when the
dosage of frother (pine oil) is around 50 (vol/wt). Further data is needed in this case,
as R2 is low.
(b)
(a)
75
74
74
73
72
71
y = 0.0314x + 70.376
R2 = 0.1233
70
% Recovery
% Recovery
75
73
72
71
y = 1.8784Ln(x) + 64.781
R2 = 0.2327
70
69
69
20
40
Frother (g/ton)
60
80
20
% Recovery
% Recovery
y = 65.03x0.0264
R2 = 0.2391
75
74
73
72
71
70
69
20
80
y = 70.347e0.0004x
R2 = 0.1283
0
60
(d)
(c)
75
74
73
72
71
70
69
40
Frother (g/ton)
40
Frother (g/ton)
60
80
20
40
60
Frother (g/ton)
Figure 13 show that two straight lines, quadratic and power function gave fit
than the other models in case of copper recovery data as affected by the pulp density
in the flotation process for enrichment of copper ore. Two straight lines give the next
80
69
best fit quadratic equation also gave good fit and X-max was 32 showing that pulp
density of 32 will gave maximum recovery, though the trend of other functions show
that recovery increased with increase in pulp density.
(b)
80
80
70
60
50
40
30
20
% Recovery
% Recovery
(a)
y = 0.76x + 45.8
R2 = 0.6694
70
60
y = 18.224Ln(x) + 7.0532
R2 = 0.7168
50
40
30
20
10
20
30
40
10
15
20
30
35
40
35
40
(d)
80
80
70
60
50
40
30
20
% Recovery
% Recovery
(c)
70
60
y = 25.854x0.2881
R2 = 0.7542
50
40
30
20
10
10
20
30
40
10
20
30
Pulp density(%wt/vol)
40
% R e c ov e ry
y = 47.704e0.012x
R2 = 0.704
10
20
25
30
(f)
(e)
80
70
60
50
40
30
20
15
Pulp density(%wt/vol)
Pulp density(%wt/vol)
% Recovery
25
Pulp density(%wt/vol)
80
70
60
50
40
30
20
10
20
30
Pulp Density (%wt/vol)
40
Figure-13: (a) LINEAR (b) LOGARITHMIC (c) QUADRATIC (d) POWER (e)
EXPONENTIAL (f) AND TWO STRAIGHT-LINE MODELS FITTED TO THE
RECOVERY OF COPPER DATA FROM FOUR LEVELS OF PULP DENSITY
IN THE FLOTATION PROCESS.
70
The coefficient of determination for the six models given in Figure 14 show
that all the models gave good fit to the data on recovery of copper as affected by
conditioning time. The recovery decreased with increase in the flotation time, beyond
13 minutes. Quadratic model gave better fit the X-max from quadratic equation is
about 11 minutes. However the two straight line model gave best fit with R2 = 0.9988;
there was not much effect of conditioning time in the range of 10-12 minutes beyond
13 minutes the recovery decrease.
(b)
(a)
70
60
50
40
y = -0.7931x + 80.07
R2 = 0.5153
30
% Recovery
% Recovery
80
20
8
12
16
80
70
60
50
40
30
20
y = -9.8924Ln(x) + 94.811
R2 = 0.4352
8
20
12
(d)
80
70
70
60
50
20
8
12
16
20
% Recovery
% Recovery
80
30
20
(c)
40
16
60
50
40
y = 100.95x-0.1463
R2 = 0.4463
30
20
8
12
16
20
71
80
70
60
50
40
30
20
% Recovery
Grade
of copper (% )
% Recovery
(e)
y = 81.159e-0.0117x
R2 = 0.5269
12
16
20
80.0
80
70.0
70
60.0
60
50.0
50
40.0
40
30.0
30
20.0
20
88
(f)
Y = 88.29 - 1.8526X
Y = 88.29
+ 9.16X'
- 1.8526X + 9.16X'
R2 = 0.9988 R2 = 0.9988
16
1212
16
Conditioning
Time (minutes)
Conditioning
Time
Figure 14: (a) LINEAR (b) LOGARITHMIC (c) QUADRATIC (d) POWER (e)
EXPONENTIAL (f) AND TWO STRAIGHT-LINE MODELS FITTED TO THE
RECOVERY OF COPPER DATA FROM FOUR LEVELS OF
CONDITIONING TIME IN THE FLOTATION PROCESS.
20
20
72
5.5.
copper ore. The following four strategies were followed for model selection.
1.
2.
3.
4.
5.
6.
Multiple regression model with testing Apt mess of model and checking the
assumptions.
73
5.6
modeling recovery of copper using the seven variables. The equations selected at each
step are given below. The details of each step are given in appendix-1
YR =38.254+1.101 X3
(5.41)
YR=38.493+0.837 X3+0.168 X4
(5.42)
(5.43)
(5.44)
(5.45)
74
However, when no intercept option was used the stepwise first included X2
then X3, X4, X6 and X1 but finally removed X2 as its probability was greater than
alpha to remove and thus the final model from stepwise procedure was the same as
forward selection with out intercept option.
5.7
strategy was used to select best model for recovery of copper ore.
The following equations, were fitted at each step. The details of each step are
given in appendix 2.
YR=-25.20+0.053X1+3.7X2+0.69X3+0.158X4-0.098X5+0.83X6-0.45X7
(5.46)
YR=-29.93+0.053X1+3.7X2+0.69X3+0.155X4-0.080X5+0.79X6
(5.47)
YR=-35.03+0.053X1+3.7X2+0.69X3+0.191X4+0.76X6
(5.48)
YR = 6.132+0.053X1+0.779X3+0.188X4+0.762X6
(5.49)
75
5.8
very difficult to check all these models so the best subset procedure was used to select
two best models involving one, two, three, four, five and six, variables. Using best
subset method (Minitab), we got thirteen models, two in each subset and the full
model for recovery of copper. The summary of best subset models are given below.
Models with one variable are:
M1: YR = 0 + 1X3
M2: YR = 0+ 1X4
Models with two variable are:
M3: YR = 0+ 1X3 + 2 X4
M4: YR = 0 + 1X3 + 2 X5
Models with three variable are:
M5: YR = 0 + 1X1 + 2 X3 + 3 X4
M6: YR = 0 + 1X3 + 2 X4 + 3 X6
Models with four variable are:
M7: YR = 0 + 1X1 + 2 X3 + 3 X4 + 4 X6
M8: YR = 0 + 1X2 + 2 X3 + 3 X4 + 4 X6
M9: YR =1X2 + 2 X3 + 3 X4 + 4 X6
Models with five variables are:
76
M10: YR = 0 + 1X1 + 2 X2 + 3 X3 + 4 X4 + 5 X6
M11: YR = 0 + 1X1 + 2 X3 + 3 X4 + 4 X6 + 5 X7
Models with six variables are:
M12: YR = 0 + 1X1 + 2 X2 + 3 X3 + 4 X4 + 5 X6 + 6 X7
M13: YR = 0 + 1X1 + 2 X2 + 3 X3 + 4 X4 + 5 X5 +6 X6
Models with seven variables are:
M14: YR = + 1X1 + 2 X2 + 3 X3 + 4 X4 + 5 X5 +6 X6 + 7 X7
Among the subset with single predictors.
The regression equations with single predictors for recovery of copper
obtained from least square analysis are as follows:
YR = 41.4 + 0.7 X3 ..(5.50)
YR = 61.07 + 0.089 X4 (5.51)
The R2 show that the first equation explained 85.2% of the variation and the
second equation explained 67.3% variation in the recovery of copper using flotation
process. The coefficients of equation (5.50) are different from coefficients of equation
(5.13), though both have X3 as independent variable. The differences in coefficients of
the two equations for X3, are due to the fact that equation (5.13) is based on the data
from one experiment and equation (5.50) is based on combined data from seven
experiments. Similarly, the differences in coefficient of equation (5.19) and (5.51) are
due to the same reason as above; equation (5.19) is based on data from one
experiment and equation (5.51) is based on data from seven experiments.
70
60
50
40
30
20
10
0
% Recovery
% Recovery
77
10
20
30
Sodium Cyanide (g/ton)
40
64
63
62
61
60
59
58
0
20
40
60
Sodium Sulphide (g/ton)
80
78
Response surfaces were developed for the variables involved in the above two
equations.
YR
X4
X3
The combine response surface for sodium cyanide (X3 g/ton) and depressant
sodium sulphide (X4 g/ton) on the recovery of copper reveals that the maximum peak
of surface shows the estimated maximum recovery of 73.69% with 30 gram per ton of
sodium cyanide and 60 gram per ton of sodium sulphide.
79
YR
X3
X5
Figure-18: COPPER RECOVERY (YR) RESPONSE SURFACE FOR SODIUM
SUPHIDE (X4), AND FROTHER DOSAGE (X5).
The combine response surface for sodium sulphide (X4 g/ton) and frother
dosage (X5 g/ton) on the estimated recovery of copper is given in the Figure 18. The
maximum peak of surface show the maximum recovery of 76.33% with 30 gram per
ton of sodium sulphide and 68-70 grams per ton of frother.
The best subset program picked the following two best regression equations
involving three predictor variables among the fifty-five, 3-variable models in the
subset with 3-predictors:
YR = 29.0 + 0.053 X1 + 0.782 X3 + 0.171 X4 .(5.54)
YR = 15.6 + 0.835 X3 + 0.185 X4 + 0.762 X6 .(5.55)
The two equations (5.54) and (5.55) explained 91.7% and 90.7% of the total
variation in recovery of copper.
80
Among the next subset with 4 predictors, the following two best regression
equations involving four predictor variables were selected by the program.
YR = 6.13 + 0.0539 X1 + 0.779 X3 + 0.188 X4 + 0.762 X6 ..(5.56)
YR = 0.0615 X1 + 0.777 X3 + 0.191 X4 + 0.918 X6 ..(5.57)
YR = - 26.6 + 3.83 X2 + 0.741 X3 + 0.189 X4 + 0.762 X6 ..(5.58)
Equation (5.56) explained 93.2% of the total variation in recovery; This model
all variables are collectively important except intercept. Equation (5.58) explained
92.4% of the variation in the data for recovery of copper, but the intercept and X2 are
not statistically significant so we drop this model. As intercept in equation (5.56) was
not significant a model with no intercept (model equation 5.57) was fitted to the data
which gave very good fit.
The following two best regression equations involving five predictor variables
were selected by the program:
YR = - 35.0 + 0.0534 X1 + 3.74 X2 + 0.688 X3 + 0.191 X4 + 0.762 X6(5.59)
YR = 8.56 + 0.054 X1 + 0.778 X3 + 0.196 X4 + 0.782 X6 - 0.305 X7 ..(5.60)
In both equations (5.59) and (5.60), the inclusion X2 and X7 did not improve
the fit significantly.
The improvement in R2 from equations with five predictor variables
(equations 5.59 and 5.60) over equations with four predictors (equations
5.56,5.57,5.58) are very small and not significant, so the models with four predictors
sufficiently explained the variation in copper recovery.
81
The two best regression equations involving six predictor variables are given
below:
YR = - 32.6 + 0.053 X1 + 3.75 X2 + 0.687 X3 + 0.200 X4 + 0.783 X6 -0.306 X7
..
(5.61)
The
equation
without
intercept
involving
the
variables
sodium
propylxanthate, sodium sulphide, sodium cyanide and pulp density explains almost
99% of the variation in recovery of copper.
5.9
significance of the parameters reduced model was fitted excluding the variables with
probability greater than 0.05 and the detail study was made about assumption of the
full and reduced model.
82
The recovery data was regressed on the seven flotation process independent
variables. The Excel out put is given Tables 5 and 6 are given below.
Regression output of copper recovery on seven independent variables
Table 5: Coefficient Analysis And Model Fitness Statistic For Seven Variables
Predictor
Coef
Constant
SE Coef
Standard Error
3.752,
-25.20 25.81
-0.98
0.339
R-Square
94.6%,
X1
0.053
0.021
2.53
0.019
X2
3.735
2.162
1.73
0.097
Press
680.517,
X3
0.693
0.104
6.67
0.000
Observation
31%
X4
0.158
0.051
3.05
0.006
X5
-0.098 0.069
-1.42
0.168
X6
0.833
3.51
0.002
X7
-0.450 0.412
-1.09
0.287
0.237
Source
DF
SS
MS
Regression
5629.65
804.24
57.14
0.000
Residual Error
23
323.71
14.07
Total
30
5953.35
83
Visual normal test for standard residuals for seven process parameters
1.2
Residuals
1
0.8
0.6
0.4
0.2
0
1
11 13 15 17 19
Number of Observation
21
23
25
27
29
31
The Figure (19) shows visual test for standard residuals of seven variables and
it has little deviation from 45-degree line yet it does not give vital evidence against
the normality.
84
Bin
Frequency
Cumulative %
-2
3.33%
-1
10.00%
14
56.67%
80.00%
100.00%
100.00%
More
100.00%
Histogram
16
120.00%
14
Frequency
80.00%
10
60.00%
8
6
40.00%
Commulative %
100.00%
12
20.00%
2
0
0.00%
-2
-1
More
Bin
Frequency
Cum ulative %
It is obvious from the histogram that the distribution of the error terms is
symmetric but not normal. In this study the co-efficient of skewness for standard
residual is 0.47, which is inside the 96% confidence interval. Thus the data is note
skewed and therefore satisfies one of the normality conditions. Also E.Kurtosis is
1.107, which is inside the 96% confidence interval and hence satisfies normality
condition.
85
Instead of using the two tests separately, one can use a linear combination of
the two. The Jarque Bera test was devised as an optimal test against a certain class
of alternative to the null distribution. The test statistic is:
JB = T{EK2/24 + (SK)^2/6}
In this study the value of Jarque Bera (JB) is 2.747, in the table given below
here are calculated values of different tests and also there critical values calculated by
simulation.
EK=kurtosis of any distribution 3= k-3
Kurtosis is measure of heaviness of tail K for normal distribution.
Skew ness (SK) a non-symmetric distribution is known as skewed distribution.
Table 8:
Test
Lower critical
Upper critical
value
value
Results
Skew ness
-0.47
-.7
.7
Pass
E.Kurtosis
1.1
-.99
1.55
Pass
Jarque-Bera
2.7
N.A
462
Pass
86
-1
-2
-3
-4
From the standard residual plot it is clear that approximately 68% data is
inside the interval [-1.1] and approximately 95% data is in the interval [-2, 2], which
is a sign of normality. If we look it more deeply we can see that the variation in the
first half is different than the second half, which is indicating heteroscedasticity
problem.
5.9.2 Testing for Heteroscedasticity
Let SER(1) and SET(2) be the Standard Error of Regression for the first half
and the second half of the data set respectively. If the ratio SER(1)/SER/2) is close to
1 then the two SEs are different. The Goldfled-Quandt statistic is based on the ratio
of variances (not Ses)
GQ [ SER (2) / SER(1)]2
Now in study Var1 (Variance of first half)=0.654, Var2 (Variance of second
half) =1.298, GQ test=0.254, p-value=0.995. The value of GQ test is 0.28, which is
87
very much different from 1. This can also be seen from the p-value of GQ test.
Standard residuals are normal but are not identically distributed, so it fails to be i.i.d
random variables. Thus R2 is meaningless.
Table-5 shows that R2 of the full model is high and have a low Standard
Error. It means that 94% variation in YR can be explained by the regressors with 7
units with 95% confidence if the residuals are identity independent Normal. But
residuals are not identity independent Normal so these statistics are meaningless.
Regression coefficients, their standard errors, t-values and probability are
given in Table-5.
From Table-5 we have write the following equation
YR = -25.2049 + 0.053 X1 +3.73 X2 + 0.69 X3 + 0.15 X4 -0.09 X5 + 0.833 X6 0.45X7
. (5.64)
T-statistics and probability given in the Table-4 shows that intercept, X2, X5
and X7 are not significant at 5% level so a reduced model excluding these was tried.
88
be tested by different ways. First we check the basic assumptions about the model and
then will compare both the models, for example By F-test etc. Also before discarding
variables we must consult the theory. Note that when we say that we are discarding
one agent it does not mean that that agent is not involved in the process. It simply
means that it is kept constant, usually on its critical value because its variation is not
effecting the dependent variable and its effect is very slight. Using the reduced model
given above, the recovery was regressed on X1, X2 X3 X4 and X6 with no intercept.
The Excel out put of the fitted model is given in the table 6 to 8. The information in
table 9, show that R2 of (5.65) is slightly less and standard error is slightly more then
model (5.65). It is important to examine the optness of the model (5.65).
Table 9: Coefficient Analysis And Model Fitness Statistic For Four Variables
Predictor
Coef
SE Coef
Standard Error
3.9
Constant
0.0
0.0
0.0
0.0
R-Square
99.60%
X1
0.061
0.019
3.16
0.004
(Adjusted) R-Square
99.55%
X3
0.776
0.093
8.34
0.000
Press
526.202,
X4
0.191
0.043
4.44
0.000
Observation
31
X6
0.917
0.118
7.78
0.000
89
Source
DF
SS
MS
Regression
102807
25702
1688.93
0.000
Residual Errors
27
411
15
Total
31
103218
F-test of the P- value suggests that all variables are collectively important. As
compared to the full model the F-value is more showing better fit.
The regression equation is
The t-statistics
The P-value of t-statistic of each variable shows that all variables are
individually important.
Test for normality of residuals
1
0.8
residuals
0.6
0.4
0.2
31
28
25
22
19
16
13
10
0
1
iii.
No. of observations
90
Figure 22 shows normal test for standard residuals of seven variables. Though
it has little deviation from 45 degree line yet it does not give vital evidence against the
normality.
5.10.2 Other tests for normality:
i)
Tests for skewness, kurtosis and Jarque bera for four variables
Table 11:
Test
Calculated
Lower critical
Upper
value
value
critical value
Skewness
-0.6
-.7
.7
Pass
Kurtosis
E-.89
-.99
1.55
Pass
-1.8
N.A
4.62
Pass
Jarque bera
ii)
Histogram
Table-12:
Bin
Frequency
-3
-2
-1
11
More
Result
91
Histogram
11
12
9
Frequency
10
8
4
2
1
0
More
0
-3
-2
-1
Bin
31
29
27
25
23
21
19
17
15
13
11
-0.5
-1
-1.5
-2
-2.5
-3
-3.5
From the standard residual plot it is clear that approximately 68% data is in
side [-1, 1] and 95% data is in the interval [-2, 2] which is the evidence that data is
normal.
92
iv)
Var-1
Var-2
GQ-Test
P-value
0.8
1.21
0.43
0.94
Structural stability:
Chow Test
2.015
P. significance
0.048
93
Using FDIST (3.1,2,27) we get the p-value 0.0641. Which is not significant.
This means that collectively these three regresors X2, X5, X7 have no collective
importance. It has a simpler theoretical structure. By using this model one can
estimate YR within +3.704 with 68% probability and U+ 7.4 with 9.8% probability
result, drop these variables one by one t-statistics suggest that they are not important.
So our final model is
YR = 0.061*X1 + 0.776*X3 + 0.191*X4 + 0.917*X6 ..(5.67)
It is clear from the model that curve passes through the origin it is obvious
from this model that if we increase one unit of X1, YR will increase 0.061 unit keeping
all other variables constant. We can define the other entire coefficient in the same
fashion. These coefficients (slopes) give partial value.
94
CHAPTER 6
MODEL BUILDING FOR GRADE
6.1
PREVIOUS STUDIES
95
maximum grade of copper was noted when frother, the pine oil, was used at the rate
of 46 g/ton. Beyond 46g/ton of pine oil the grade was reduced as shown in Figure 29
graphically. The optimum pulp density (X6) is of great importance, as in general the
more dilute the pulp, the cleaner the separation. The effect of pulp density on the
grade of chalcopyrite ore has been shown in Figure 30. The curve shows that
maximum values of grade was obtained at 30% solids by weight. However, beyond
that level of pulp density, the grade markedly decreased due to the entrapped fine slim
particles. Next test was conducted to find out the effect of conditioning time of
collectors ranging from 10 to 18 minutes on grade. The graph in Figure 31 indicates
that 13 minutes conditioning time was optimum for obtaining better grade of copper.
Conditioning time greater than 13 minutes reduces the grade due to dissolution of
copper xanthate ions in the equilibrium system.
96
(b)
(a)
17
% Grade
% Grade
15
13
11
9
7
5
0
100
200
16
15.5
15
14.5
14
13.5
13
12.5
9.5
300
10
10.5
11
12.5
(d)
(c)
18
17
16.5
16
15.5
15
14.5
14
13.5
17.5
% Grade
% Grade
12
pH
PropylXanthate (g/ton)
17
16.5
16
15.5
15
10
20
30
40
20
80
(f)
18.5
% Grade
18
17.5
17
16.5
16
20
60
(e)
40
% Grade
11.5
40
60
80
Pineoil (g/ton)
19
17
15
13
11
9
7
5
0
10
20
30
40
97
(g)
19
% Grade
17
15
13
11
9
7
5
5
10
15
20
Conditioning time(minutes)
6.2
forty one models were fitted to select suitable models based on F-test, and R2 of the
model and t-test of the model parameters.
The models and their R2 are given in Table 13 and graphically presented in
Figures (1 to 7). Mathematical models 6.6, 6.13, 6.18, 6.24, 6.29, 6.35 and 6.41 were
best for X1, X2, X3, X4, X5, X6 and X7 respectively among the forty one single predictor
models.
98
R2 = 0.65
(6.1)
YG = 2.4687Ln(X1) + 1.3788
R2 = 0.7728
(6.2)
R2 = 0.9017
(6.3)
YG = 5.1995X10.1929
R2 = 0.7965
(6.4)
YG = 10.748e0.0014X1
R2 = 0.6687
(6.5)
R2 = 0.9421
(6.6)
YG = -0.1495X2 + 16.04
R2 = 0.0169
(6.7)
YG = -1.4068Ln(X2) + 17.767
R2 = 0.0124
(6.8)
R2 = 0.6541
(6.9)
YG = 16.534e-0.0128X2
R2 = 0.0248
(6.10)
YG = 19.32X2-0.1236
R2 = 0.0194
(6.11)
R2 = 0.9868
(6.12)
YG = -0.044X3 + 16.38
R2 = 0.1066
(6.13)
YG = -0.5273Ln(X3) + 17.042
R2 = 0.0461
(6.14)
99
R2 = 0.7114
(6.15)
YG = 17.339X3-0.039
R2 = 0.0574
(6.16)
YG = 16.471e-0.0031X3
R2 = 0.1233
(6.17)
R2 = 0.9808
(6.18)
YG = 0.0158X4 + 15.516
R2 = 0.1047
(6.19)
YG = 0.5068Ln(X4) + 14.352
R2 = 0.1459
(6.20)
R2 = 0.2709
(6.21)
YG = 14.469X40.0306
R2 = 0.1460
(6.22)
YG = 15.529e0.0009X4
R2 = 0.1025
(6.23)
R2 = 0.7835
(6.24)
YG = -0.0154X5 + 18.011
R2 = 0.201
(6.25)
YG = -0.4706Ln(X5) + 19.058
R2 = 0.0992
(6.26)
YG = 19.207X5-0.0282
R2 = 0.1067
(6.27)
YG = 18.03e-0.0009X5
R2 = 0.2111
(6.28)
R2 = 0.3567
(6.29)
100
YG = -0.1829X6 + 21.6
R2 = 0.4213
(6.30)
YG = -3.6902Ln(X6) + 28.686
R2 = 0.3196
(6.31)
R2 = 0.9115
(6.32)
YG = 36.157X6-0.2406
R2 = 0.326
(6.33)
YG = 22.761e-0.0119X6
R2 = 0.4276
(6.34)
YG = 17.6 + 0.0171 X 5X
R2 = 0.9973
(6.35)
YG = -0.9143X7 + 28.229
R2 = 0.8491
(6.36)
YG = 11.961Ln(X7) + 46.689
R2 = 0.7890
(6.37)
R2 = 0.9741
(6.38)
YG = 129.66X7-0.8222
R2 = 0.7805
(6.39)
YG = 36.521e-0.063X7
R2 = 0.8436
(6.40)
YG = 32.3473 1.444X 5 X
R2 = 0.9994
(6.41)
The various models with R2 for collector are presented in Figure 32. Though
quadratic model had better fit, all other models also gave good fit based on R2. The Xmax calculated from the quadratic equation was 189 gram per ton which shows that
collector level of about 190 gram per ton will gave the highest grade of copper. The
101
two straight line model give best fit, showing that the joining point at 200 g/ton of
collection will give the highest grade.
(b)
16
15
15
14
14
% Grade
% Grade
(a)
16
13
y = 0.0182x + 10.67
R2 = 0.65
12
13
y = 2.4687Ln(x) + 1.3788
R2 = 0.7728
12
11
11
10
10
40
90
140
190
240
40
PropylXanthate (g/ton)
(c)
16
(d)
15
13
% Grade
14
y = -0.0002x2 + 0.0756x + 7.32
R2 = 0.9017
12
11
14
13
y = 5.1995x0.1929
R2 = 0.7965
12
11
10
40
90
140
190
240
PropylXanthate (g/ton)
10
290
40
90
16
% Grade
15
14
13
y = 10.748e0.0014x
R2 = 0.6687
12
11
16
15
14
13
12
11
10
90
140
190
240
ProphylXanthate (g/ton)
290
40
10
40
140
190
240
PropylXanthate (g/ton)
(f)
(e)
% Grade
290
16
15
% Grade
90
140
190
240
ProphylXanthate (g/ton)
290
90
140
190
Propylxanthate (g/ton)
240
Figure-32: (a) LINEAR (b) LOGARITHMIC (c) QUADRATIC (d) POWER (e)
EXPONENTIAL (f) AND TWO STRAIGHT-LINE MODELS FITTED TO THE
GRADE OF COPPER DATA FROM FIVE LEVELS OF COLLECTOR USE
IN THE FLOTATION PROCESS.
102
The R2s of the different models given in Figure 33 show that only quadratic
model gave good fit. The X-max calculated from the function is 10.9 showing that pH
of 10.9 that will give the highest grade of copper. The two straight lines gave much
better fit with the highest R2, it show that pH of 11.5 will give the highest grade of
copper.
20
19
18
17
16
15
14
13
12
11
10
(b)
% Grade
% Grade
(a)
y = -0.1495x + 16.04
R2 = 0.0169
9.8
10.8
11.8
12.8
20
19
18
17
16
15
14
13
12
11
10
y = -1.4068Ln(x) + 17.767
R2 = 0.0124
9.8
10.8
11.8
pH
(d)
(c)
20
19
18
17
16
15
14
13
2
12 y = -1.7788x + 38.933x - 197.63
2
11
R = 0.6541
10
9.8
10.8
11.8
pH
16
15.5
15
% Grade
% Grade
12.8
pH
14.5
14
13.5
y = 16.534e-0.0128x
R2 = 0.0248
13
12.5
12.8
9.5
10
10.5
11
pH
11.5
12
12.5
103
(f)
16
15
20
19
18
17
16
15
14
13
12
11
10
% Gra de
% Grade
(e)
y = 19.32x -0.1236
R2 = 0.0194
9.8
10.8
14
13
12
11
10
11.8
9.8
12.8
10.3
10.8
11.3
11.8
pH
pH
Figure-33: (a) LINEAR (b) LOGARITHMIC (c) QUADRATIC (d) POWER (e)
EXPONENTIAL (f) AND TWO STRAIGHT-LINE MODELS FITTED TO THE
GRADE OF COPPER DATA FROM FOUR LEVELS OF PH IN THE
FLOTATION PROCESS.
17
16
15
14
13
12
11
10
(b)
% Grade
% Grade
(a)
y = -0.044x + 16.38
R2 = 0.1066
8
12
16
20
24
28
32
17
16
15
14
13
12
11
10
y = -0.5273Ln(x) + 17.042
R2 = 0.0461
0
10
20
30
Sodium Cyanide (g/ton)
40
104
(d)
17
16
15
14
13
12
11
10
% Grade
% Grade
(c)
12
16
20
24
28
32
17
16
15
14
13
12
11
10
y = 17.339x-0.039
R2 = 0.0574
8
12
17
16
15
14
13
12
11
10
y = 16.471e-0.0031x
R2 = 0.1233
12
16
32
(f)
% Gra d e
% Grade
(e)
16
20
24
28
Sodium Cyanide (g/ton)
20
24
28
32
17
16
15
14
13
12
11
10
12
16
20
24
28
Sodium Cyanide (g/ton)
Figure-34: (a) LINEAR (b) LOGARITHMIC (c) QUADRATIC (d) POWER (e)
EXPONENTIAL (f) AND TWO STRAIGHT-LINE MODELS FITTED TO THE
GRADE OF COPPER DATA FROM FOUR LEVELS OF SULFIDIZER IN
THE FLOTATION PROCESS.
Except the two straight line model, none of the models presented in Figure 35
gave good fit to the observed data for grade as affected by levels of depressant. The
original data points show that highest grade of copper was obtained when 50 g per ton
of depressant was used though X-max from quadratic function was about 40 gram per
ton of depressant. The two straight line model also how that X-max is 50 g/ton.
32
105
(b)
% Grade
% Grade
(a)
19
18
17
16
15
14
13
12
11
10
y = 0.0158x + 15.516
R2 = 0.1047
19
18
17
16
15
14
13
12
11
10
y = 0.5068Ln(x) + 14.352
R2 = 0.1459
5 10 15 20 25 30 35 40 45 50 55 60 65
5 10 15 20 25 30 35 40 45 50 55 60 65
(d)
% Grade
% Grade
(c)
19
18
17
16
15
14
13
12
11
10
19
18
17
16
15
14
13
12
11
10
5 10 15 20 25 30 35 40 45 50 55 60 65
0.0306
y = 14.469x
2
R = 0.146
5 10 15 20 25 30 35 40 45 50 55 60 65
Sodium sulphide (g/ton)
Figure-35: (a) LINEAR (b) LOGARITHMIC (c) QUADRATIC (c) POWER (e)
EXPONENTIAL (f) AND TWO STRAIGHT-LINE MODELS FITTED TO THE
GRADE OF COPPER DATA FROM FIVE LEVELS OF DEPRESSANT IN
THE FLOTATION PROCESS.
None of the models given in figure 36 gave good fit to the observed data on
grade of copper as affected by frother dosage. The original data points show that 46
gram per ton of frother gave maximum grade of copper, the two straight lines who
show the same result regarding X-max. The two straight line gave much better fit than
the other models.
19
18
17
16
15
14
13
12
11
10
(b)
% Grade
% Grade
(a)
y = -0.0154x + 18.011
R2 = 0.201
20
30
40
50
60
Frother (g/ton)
70
80
19
18
17
16
15
14
13
12
11
10
y = -0.4706Ln(x) + 19.058
R2 = 0.0992
20
30
40
50
60
Frother (g/ton)
70
80
106
19
18
17
16
15
14
13
12
11
10
(d)
% Grade
% Grade
(c)
y = 19.207x-0.0282
R2 = 0.1067
20
30
40
50
60
70
80
19
18
17
16
15
14
13
12
11
10
y = 18.03e-0.0009x
R2 = 0.2111
20
Frother (g/ton)
30
40
50
60
Frother (g/ton)
70
80
% G rade
(e)
19
18
17
16
15
14
13
12
11
10
20
30
40
50
60
70
80
Frother (g/ton)
Figure-36: (a) LINEAR (b) LOGARITHMIC (c) QUADRATIC (d) POWER (e)
EXPONENTIAL (f) AND TWO STRAIGHT-LINE MODELS FITTED TO THE
GRADE OF COPPER DATA FROM FOUR LEVELS OF FROTHER DOSAGE
IN THE FLOTATION PROCESS.
The model for grade of copper as effected by pulp density are given Figure 37.
Quadratic regression gave best fit followed by linear regression to the data on grade of
copper as affected by pulp density. However two straight lines give excellent fit with
fit with an R2 of 0.9993. X-max from quadratic equations was 21.8 g per ton X-max
from two straight lines is 30 g per ton.
107
19
18
17
16
15
14
13
12
11
10
(b)
% Grade
% Grade
(a)
y = -0.1829x + 21.6
R2 = 0.4213
10
20
30
Pulp density (%wt/vol)
19
18
17
16
15
14
13
12
11
40
y = -3.6902Ln(x) + 28.686
R2 = 0.3196
10
20
30
Pulp density (%wt/vol)
(d)
% Grade
% Grade
(c)
20
19
18
17
16
15
14
13
12
11
10
19
18
17
16
15
14
13
12
11
10
y = 36.157x-0.2406
R2 = 0.326
10
10
20
30
40
-0.0119x
y = 22.761e
2
R = 0.4276
20
30
40
(f)
% Grade
% Grade
(e)
10
20
Pulp density(%wt/vol)
19
18
17
16
15
14
13
12
11
10
40
30
40
19
18
17
16
15
14
13
12
11
10
10
20
30
Pulp Density
Figure-37: (a) LINEAR (b) LOGARITHMIC (c) QUADRATIC (d) POWER (e)
EXPONENTIAL (f) AND TWO STRAIGHT-LINE MODELS FITTED TO THE
GRADE OF COPPER DATA FROM FOUR LEVELS OF PULP DENSITY IN
THE FLOTATION PROCESS.
The models for grade of copper as affected by conditioning time are present in
Figure 38. Coefficients of determination of models presented in Figure 38 show that
40
108
all the models give good fit to data of copper grade as affected by flotation time.
Grade of copper decreased with increase in time of flotation. Two straight line model
gave excellent fit with an R2 of 0.9994 showing that it explained almost all the
variation in data for grade. The graph for two straight line show that there was no
significant difference in grade when conditioning time was 10 to 13 minutes but
conditioning time greater that 13 minute reduce grade of copper.
20
19
18
17
16
15
14
13
12
11
10
(b)
% Grade
% Grade
(a)
y = -0.9143x + 28.229
R2 = 0.8491
12
16
Conditioning time (minute)
20
20
19
18
17
16
15
14
13
12
11
10
y = -11.961Ln(x) + 46.689
R2 = 0.789
8
12
16
Conditioning time (minute)
(d)
20
19
18
17
% Grade
% Grade
(c)
20
19
18
17
16
15
14
13
12
11
10
20
16
15
14
13
y = 129.66x-0.8222
R2 = 0.7805
12
11
12
16
20
10
8
12
16
20
109
20
19
18
17
16
15
14
13
12
11
10
(f)
% Gra de
% Grade
(e)
y = 36.521e-0.063x
R2 = 0.8436
8
12
16
Conditioning time (minute)
20
19
18
17
16
15
14
13
12
11
10
12
16
Conditioning Time (minutes)
Figure-38: (a) LINEAR (b) LOGARITHMIC (c) QUADRATIC (d) POWER (e)
EXPONENTIAL (f) AND TWO STRAIGHT-LINE MODELS FITTED TO THE
GRADE OF COPPER DATA FROM FOUR LEVELS OF FLOTATION TIME
IN THE FLOTATION PROCESS.
20
110
6.3
copper ore. The following four strategies were followed for models selection.
1.
2.
3.
4.
6.4
Forward Selection
The forward stepwise selection procedure was used to select variables for
modeling grade of copper using seven independent variables. The details of steps are
given in appendix 4. The equations selected at each step are given below.
YG = 14.01 + 0.085 X3
(6.42)
(6.43)
(6.44)
(6.45)
111
Table14:
X1
X2
X3
X4
X5
X6
X7
YG
50
11
75
30
10
11
100
11
75
30
10
12.1
150
11
75
30
10
14.7
200
11
75
30
10
15.2
250
11
75
30
10
14
200
10
75
30
10
14.2
200
10.3
75
30
10
14.2
200
11
75
30
10
15.1
200
11.58
75
30
10
15.5
200
12
75
30
10
13
200
11.58
10
75
30
10
15.4
200
11.58
15
75
30
10
15.5
200
11.58
20
75
30
10
16.3
200
11.58
25
75
30
10
16.5
200
11.58
30
75
30
10
13.8
200
11.58
25
10
75
30
10
15.51
200
11.58
25
30
75
30
10
15.78
200
11.58
25
40
75
30
10
16.2
200
11.58
25
50
75
30
10
17.7
200
11.58
25
60
75
30
10
15.4
200
11.58
25
50
25
30
10
17.2
200
11.58
25
50
46
30
10
18.1
200
11.58
25
50
70
30
10
16.56
200
11.58
25
50
46
15
10
17.8
200
11.58
25
50
46
25
10
18.2
200
11.58
25
50
46
30
10
18
200
11.58
25
50
46
35
10
13.2
200
11.58
25
50
46
30
10
17.9
200
11.58
25
50
46
30
13
18.2
200
11.58
25
50
46
30
16
13.7
200
11.58
25
50
46
30
18
11
112
6.5
Backward elimination
Backward elimination procedure or general to simple model building strategy
was used to select model for grade of copper using data combined over all
experiments. The following models were fitted at each step. The details are given in
appendix 5.
YG = 21.58 + 0.0194X1 + 0.07X2 + 0.038X3 + 0.018X4 -0.024X5 - 0.130X6 - 0.60X7.. (6.46)
YG = 22.30 + 0.019X1 + 0.04X3 + 0.018X4-0.024X5-0.130X6 - 0.60X7.. (6.47)
YG = 20.55 + 0.019X1 + 0.039X3 + 0.028X4 - 0.142X6 -0.57X7 (6.48)
YG = 20.43 + 0.0216X1 + 0.041X4 - 0.141X6 - 0.57X7. (6.49)
YG = 16.48 + 0.021X1 + 0.045X4 - 0.60X7 . (6.50)
Equation (6.46), (6.47), (6.48) and (6.49), are not statistically significant these
models are not good fit. While equation (6.50), is a best fit model.
The forward selection and backward elimination selected at different
equations. So we will look at the best subset procedure and the model from that
procedure which correspond to any of the model from forward or backward will be
the approximate model-1.
6.6
variables therefore the best subset procedure was used to select models involving one,
two, three, four, five and six variables. The best subset procedure (Minitab), produced
the following thirteen models two in each subset and the full model for grade of
copper.
113
114
% Grade
20
18
17.5
17
16.5
16
15.5
15
15
10
%Grade
5
0
0
10
20
30
40
20
40
60
80
115
In the subset with single predictor variable, models involving X3 and X4 were
found as better model than others for grade of copper. Based on R2, F-value, tstatistics and P-value (as shown in the output), model having X3 is better than model
having X4.
Grade increased at the rate 0.0854% per one gram increase in depressant.
Sixty gram per ton of sulfidizer gave maximum grade 19.12%. Grade increased at the
rate of 0.0854% per one gram increase in sulphidizer.
Among the twenty-one models in the subset with two predictor variables, the
two best regression equations involving two predictor variables are:
YG = 20.5 + 0.0518 x4 - 0.599 X7
(6.53)
(6.54)
The equation involving X4 and X7 explained 46.1% and the equation involving
X3 and X7 explained 45.1% of the variation in the grade of copper.
Response surfaces were developed for the variables involved in the above two
equations.
116
YG
X3
X7
The combine response surface for sodium cyanide and conditioning time on
the grade of copper is shown in Figure 41. The maximum peak of surface shows the
estimated maximum grade of 16.42% with 28 gram per ton of sodium sulphide and 11
minutes conditioning time.
117
YG
X4
X7
The combine response surface for sodium sulphide and conditioning time on
grade of copper is shown in Figure 42. The maximum peak of surface shows the
estimated maximum grade of 17.36% with 55 gram per ton of sodium sulphide and 10
minutes conditioning time.
The best subset program picked the following two best regression equations
involving three predictor variables among the 55, 3-variable models.
YG = 16.5 + 0.0216 X1 + 0.0448 X4 - 0.598 X7 ... (6.55)
YG = 15.9 + 0.0191 X1 + 0.0859 X3 - 0.525 X7 (6.56)
The two equation explain 58.6% and 54.3% of total variation in grade of
copper.
118
Among the next subset with four predictors, the follow two best regression
equations involving four predictor variables were selected by the program:
YG = 20.4 + 0.0216 X1 + 0.0411 X4 - 0.141 X6 - 0.570 X7
(6.57)
(6.58)
Equation (6.57) explained 62.7% but in this model the contribution X6 is not
important. Equation (6.58) explained 61.1% of the variation in the data, all variables
in this model are collectively important so this is a good fit model for grade of copper.
The following two best regression equation involving five predictor variables
were selected by the program:
YG = 20.5 + 0.0194 X1 + 0.0386 X3 + 0.0281 X4 - 0.142 X6 - 0.567 X7 (6.59)
YG = 23.3 + 0.0193 X1 + 0.0591 X3 - 0.0371 X5 - 0.133 X6 - 0.593 X7 (6.60)
The improvement in R2 from equations with five predictor variables over
equations with four predictor variables is small so the models (6.57) and (6.58) are
better for data on grade of copper.
Both the above models are not significant different then model (6.57) and
(6.58).
The two best regression equations involving six predictor variables are given
below:
YG = 22.3 + 0.019 X1 + 0.039 X3 + 0.018 X4 - 0.023 X5 - 0.130 X6 - 0.602 X7(6.61)
YG= 19.8+ 0.019 X1 +0.068 X2 +0.036 X3 +0.028 X4 -0.142 X6-0.567 X7.. (6.62)
Both the above models are not statistically significant.
119
YG=21.6+0.0194X1+0.066X2+0.0383X3+0.0182X4-0.0236X5-0.130X6-0.602X7 (6.63)
Full model was fitted for the grade of copper the excel out put is given below:
Table: 15: Coefficient Analysis for Grad And Model Fitness Statistic For Seven
Variables
Predictor
Coef
SE Coef
Standard
1.362
Error
Constant
21.577
9.366
2.30
0.031
R-Square
66.0%
X1
0.019390
0.007662
2.53
0.019
R-Square
55.7%
(Adjusted)
X2
0.0658
0.7846
0.08
0.934
X3
0.03834
0.03773
1.02
0.320
X4
0.01824
0.01885
0.97
0.343
X5
-0.02360
0.02505
-0.94
0.356
X6
-0.13009
0.08607
-1.51
0.144
X7
-0.6017
0.1499
-4.02
0.001
Press
129.905
Source
DF
SS
MS
Regression
82.926
11.847
6.39
0.000
Residual Error
23
42.637
1.854
Total
30
125.563
From Statistical Analysis from Table (15), the regressors X2, X3, X4, X5, and X6 are
insignificant but collectively we cannot exclude these all regressors.
120
Subset F (4,23) = 1.998 [0.1296] so we can exclude X2, X4, X5, and X6 on statistical
basis. So our new significant models is YG=15.9+0.019X1+0.0859X3-0.525X7.
Subset F (3,27) = 10.68 [0.0000] all regressors are collectively important.
6.7
contain all available candidates as predictor, then the model was simplify by
discarding the variables that did not contribute to explaining the variability in the
dependent variable.
After working with different full and reduced models the following final
model was selected. The model gave good fit as judged by the adjusted R2 and it has
the required criteria independence, normality etc of residual to some extent:
YG = 15.9 + 0.0191 X1 + 0.0859 X3 - 0.525 X7 (6.64)
The excel output for the model is given in table 15 and 16.
Table 16: OLS estimates for three significant variables
Predictor
Coef
SECoef
S.E
1.458
Constant
15.876
2.135
7.44
0.000
R.Sq
54.3%
X1
0.019072
0.008204
2.32
0.028
R-Sq(adj)
49.2%
X3
0.08593
0.02450
3.51
0.002
Press
X7
-0.5249
0.1502
-3.49
0.002
121
Source
DF
SS
MS
Regression
68.147
22.716
10.68
0.000
Residual Error
27
57.416
2.127
Total
30
125.563
T-state in column 4 of table 15 for each coefficient suggests that each variable
is individually significant.
F statistics given below tells that collectively, all regressors are important
Significance F = 0.0043
Graphical Analysis:
Histogram of Residuals
Figure 43
122
Figure 44
Residuals are heteroscedastic
Residuals are normal. It qualifies the visual test of normality.
1.2
1
0.8
0.6
0.4
0.2
0
0
10
11
12
13
14
123
6.8
A point estimate
A standard error
The point estimate helps us determine economic significance.
Sample Size?
124
detect, the Type I and type II error rates, and design of the study. But unfortunately
we have limited observation.
Few agents who are not included are important for conducting experiment so
their constant values are present that are clear from the intercept.
Usually we need to exercise judgment to decide whether an estimated effect is
economically significant but in our case we can produce high quality copper grade by
just looking the model.
Usually if an estimated effect is economically and statically significant, we
need to weigh our results against those of other researchers but again it is our
misfortune that we dont have any mathematical model to compare the results.
In analysis few times we face a problem that data are unusual. Because you
did things in a slightly different way than did others? If so, is our method one that is
knowledge and time? But in our case data is from control experiment so it cant be
unusual.
6.9
125
126
PROCESS
Recovery /
Grade
If out of the seven inputs only one is varied then its variation will effect the
recovery/Grade YR/YG. Above figure demonstrates a conceptual model of this
phenomena only when input X is varied.
127
CHAPTER-7
CONCLUSION
I understand that in Pakistan this is the first ever attempt to develop
Mathematical models both for recovery (YR) and grade (YG) of copper from the
copper ore.
Mathematical model for recovery
YR = 0.061X1+0.776X3 +0.191X4+0.917X6
Indicate that out of seven variables X1,X2, --, X7 only four of them Propylxanthate
(X1), Sodium Cyanide (X 3), Sodium Sulphide (X4) and pulp density (X6) are significant.
This model not only gives overall picture of the variables but also shows that X6 and
X3 play dominant role.
For Grade (YG) the mathematical model
YG = 0.019X1+0.0858X3-0.525 X7+15.9
Indicate that out of seven variables only three variables Propylxanthate (X1),
Sodium Cyanide (X3) and conditioning time (X7) are significant. Conditioning time
128
129
REFERENCES
Acherman, P. K., Harris, G. H., Klimpel, R.R., Aplan, F. F., 2000. Int. J.
Miner. Process (Elsevier), Vol. 58, No. 1.
Agar, G.E., Chia, J., Requis-c, L., 1998. Flotation rate measurements to
optimize an operating circuits. Minerals Engineering Vol. 11, No. 4,
pp. 347-360.
Agar, G.F, Stralton-Crawly, R., and Bruee, T.J. 1986. Optimizing the design
of flotation circuits. CIM Bulletin 73, pp. 173-181.
Ahmed, N., Jameson, G.J, 1989. Flotation kinetics. Mineral Processing and
Extractive Metallurgy Review 5, pp. 7799.
Aplan, F.F., 1976. Coal flotation. In: Fuerstenau, M.C. (Ed.), FlotationA.M.
Gaudin Memorial, Vol. 2. SME, pp. 12351264.
Arbiter, N., Weiss, N.L., 1970. Design of Flotation Cells and Circuits, Society
of Mining Engineers, Preprint 70-B-89, AIME annual meeting.
Arbiter, N., Harris, C.C., and Yap, R.F.1967, In: Hydrodynamic of Flotation
cells Society of Mining Engineers, Transactions, Vol. 244, pp.134-148.
130
Arnold, B.J., Aplan, F.F., 1989. Application of rate models to pilot and
commercial scale coal froth flotation circuits. Presented at the SME, Annual
Meeting, Las Vegas, NV.
Arnold, B.J., Aplan, F.F., Arnold and Aplan, 1988. A practical view of rate
and residence time in industrial coal froth flotation circuits. In: Processing
Fifth Annual International, Pittsburgh Coal Conference, Pittsburgh, PA. Vol.
12, No. 16, pp. 328-341.
Banerjee, P.K., Choudhury, B.R., Ramu, A., Rao, P.V.T., Sripriya, R., 1998.
Plant trials with alternative to pine oil at West Bokaro. R & D Internal Report,
Tata steel.
Baro, M. and Piga, L., 1998. Mineral Processing Institute CNR Via Bolognola,
7-00138 Rome, Italy Comparison of pb-zn selective collectors using
Statistical Methods.
Dalton, R. F., Price, R., Hermana, E., and Hoffman, B., 1988. Cuprex-New
Chloridebased Hydrometallurgy to Recover Copper from Sulfide Ores.
131
Mining Engineering, Vol. 40, No. 1, pp. 24-28. Vol. 2, No.1 Characterization
and Flotation of Sulfur from Chalcopyrite Concentrate Leaching.
Deglon, D.A., Sawyerr, F., O., Connor, C.T., 1999. A model to relate the
flotation rate constant and the bubble surface area flux in mechanical flotation
cells. Minerals Engineering Vol. 12 No. 6, Pp. 599608.
Dowling, E. C., Klimpel, R.R. and Aplan, F.F., 1985. Model discrimination in
the floation of a porphyry copper ore. Minerals and Metallurgical Processing,
pp.87-101.
Dowling, E. C., Klimpel, R.R. and Aplan, F.F., 1987. Use of kinetic model to
analyze industrial flotation circuits. In: D.R Gaskell et a,:. International
Symposium on Innovative Technology and Reactor Design in Extractive
Metallugy, TMS-AIME, Warrendale, PA, pp.533-552.
Harris, C.C., 1974. Impeller speed, air and power requirements in floatation
machine scale-up. International journal of Mineral processing, pp.51-64.
132
Harris, C.C., Rimmer, H.W., 1966. Study of a two phase model of the flotation
process. Institution of Mining and Metallurgy, pp. 153 161.
Haver, F. P., and Wong, W. W., 1971. Recovery of Copper, Iron and Sulfur
from Chalcopyrite Concentrate Using a Ferric Chloride Leach, Journal of
Metals, Vol. 23, No 2, pp. 25-29.
Haver, F. P., Baker, R. D., and Wong, M. M., 1975. Improvements in Ferric
Chloride Leaching of Chalcopyrite Concentrate, U. S. Bureau of Mines,
RI8007.
133
Khan, et al. 2000. Flotation of copper minerals from North Wazirsitan copper
ore on pilot-scale. Quarterly science vision vol. 6. No. 1, pp. 10-20.
Klimpel, R.R., Mular, A.L., and Bhappu R.B., Editors, 1980. Selection of
Chemical Reagents for Flotataion mineral Processing Plant Design (2nd ed.),
AIME, New York, pp. 930-934.
Kuopanportti, H., Suorsa, T., Dahl, O., Niinimaki, J., 2000. A model of
conditioning in the flotation of a mixture of pyrite and chalcopyrite ores.
International Journal of Mineral Processing Vol. 59 No. 4, pp. 327338.
Leaching
Residue.
Minerals
&
Materials
Journal
of
Mao, L., Yoon, R.H., 1997. Predicting flotation rates using a rate equation
derived from first principles. International Journal of Mineral Processing Vol.
51, pp. 171181.
134
Oliveira, J.F., Saraiva, S.M., Pimenta, J.S., Oliveira, A.P.A., 2001. Kinetics of
pyrochlore flotation from Arax_a mineral deposits. Minerals Engineering Vol.
14 No. 1, pp. 99105.
Rau, M. F., and Fuerstenau, M. C., 1981. Rate Phenomena Involved in the
Dissolution of Chalcopyrite in Chloride-bearing Lixiviants, Metallurgical
Transaction, Vol. 12B, pp. 595-601,
Parker, A. J., Paul R. L., and Power G. C., 1981. Electrochemical Aspects of
Leaching Copper from Chalcopyrife in Ferric and Cupric Salt Solutions,
Aust. J. of Chemistry, Vol. 34, pp. 13-34.
Pitt, J.C., 1968. The development of system for continuous optimal control of
flotation plants. In: Computer Systems Dynamics and Automatic Control in
Basic Industries, I.F.R.C Symposium, Sydney, pp. 165-171.
Polat, M., Chander, S., 2000. First-order flotation kinetics models and
methods for estimation of the true distribution of flotation rate constants.
International Journal of Mineral Processing Vol. 58, No. (14), pp. 145166.
Sripriya, R., P.V.T. Rao, Choudhry, B.R., Rao, M.V.S., and Sensharma, S.,
1997. Increasing the Yield in flotation-scale up of laboratory data and
commercialization. Fuel Scinece and Technology Vol. 16, No. 3, pp. 71-77.
Sripriya, R. P.V.T. Rao , Chaudhry, B.R. 2003 R and D Division Tata steel,
Jamshedpur, Jharkhand 831 007, India.. Optimization of operating variables
of fine coal flotation using a combination of modified flotation parameters and
135
Vera, N.A. The Modeling of froth zone Recovery in Batch and continuously
operated Laboratory flotation cells N.A. the University of Queens Land
Brisbane, QLD Australia.
Wills, B.A., 1986. Camborne school of mines, Red Ruth, Corn Wall, TR 15
3SE, Great Britain Complex circuit mass balancing-A simple, practical,
sensitivity Analysis Method.
Wills, B.A., 1997. Mineral Processing Technology, sixth ed. ButterworthHeinemann, Oxford. pp. 284316.
Xu, M., 1998. Modified flotation rate constant and selectivity index.
Minerals engineering Vol. 11 No. 3, pp. 271-278.
136
Yianatos, J.B., Bergh, L.G., Condori, P., Aguilera, J., 2001. Hydrodynamic
and metallurgical characterization of industrial flotation banks for control
purposes. Minerals Engineering Vol. 14, No. 9, pp. 10331046.
137
APPENDIX-1
General model for Recovery and Grade of copper using E-view
software:
YR = -25.2 + 0.05334*X1 + 3.735*X2 + 0.6932*X3 + 0.1583*X4
(SE) (25.8) (0.0211)
(2.16)
(0.104)
(0.0519)
(0.237)
(0.413)
Since the t-ratio of following regressors are insignificant so we now try to observe
whether we can collectively drop all these variables or not.
Test for excluding:
[0] = Constant
[1] = X2
[2] = X5
[3] = X7
Subset F (4,23) = 1.5485 [0.2216]
So we can collectively drop them
Now our new model will be
YR = + 0.06148*X1 + 0.7765*X3 + 0.1911*X4 + 0.9176*X6 ---------- (b)
(SE)
(0.0194)
(0.0932)
(0.0431)
(0.118)
138
R2 is very high but we know that t-ratio, F-stat and R2 etc give meaningful values if
residuals are identically independent normal. So we check that whether these
variables are identically independent normal or not.
Figure (A)
31
Mean
0.043489
Std. Devn.
3.6404
Skewness
-0.78933
Excess Kurtosis
1.1999
Minimum
-11.304
Maximum
6.2073
139
Figure (B)
Figure (C)
Residuals look independent.
140
Autocorrelation of residuals
Figure (D)
(0.785)
(0.0377)
(0.0188)
(0.0861)
(0.15)
141
(2.13) (0.0082)
(0.0245)
(0.15)
R2 = 0.542911
Subset F (3,27) = 10.68 [0.0000]**
All variables are collectively important.
Figure (E)
142
It looks normal
Normality test for Residuals
Observations
31
Mean
0.00000
Std.Devn.
1.3607
Skewness
-0.41587
Excess Kurtosis
0.62105
Minimum
-3.3897
Maximum
3.1853
Figure (F)
143
(2.06)
(0.00758)
(0.0111)
(0.147)
12.73 [0.000]**
144
31
Mean
0.00000
Std.Devn
1.2952
Skewness
-0.65212
Excess Kurtosis
1.1558
Minimum
-3.8576
Maximum
2.9363
Residuals histogram
Figure (G)
145
Figure (H)
Figure (I)
146
APPENDIX 2
FORWARD SELECTION
Output from forward model selection procedure used for data on copper
recovery by the flotation process with seven variables (Alpha-to-enter: 0.25, N = 31)
Table A
Step
Constant
38.254
38.493
15.650
6.132
-35.026
-29.933
X3
1.101
0.837
0.835
0.779
0.688
0.694
T-Value
12.94
7.52
8.41
8.29
6.55
6.65
P-Value
0.000
0.000
0.000
0.000
0.000
0.000
X4
0.168
0.185
0.188
0.191
0.155
T-Value
3.18
3.90
4.29
4.53
2.97
P-Value
0.004
0.001
0.000
0.000
0.007
X6
0.76
0.76
0.76
0.79
T-Value
2.87
3.12
3.23
3.38
P-Value
0.008
0.004
0.003
0.003
X1
0.054
0.053
053
T-Value
2.44
2.50
2.52
P-Value
0.022
0.019
0.019
X2
3.7
3.7
T-Value
1.71
1.72
P-Value
0.099
0.098
X5
-0.080
T-Value
-1.19
P-Value
0.247
5.50
4.80
4.28
3.94
3.80
3.77
R-Sq
85.25
89.17
91.69
93.24
93.95
94.28
R-Sq(adj)
84.74
88.39
90.77
92.20
92.74
92.85
C-p
35.4
20.8
12.1
7.6
6.6
7.2
1000.40
787.710
672.060
565.64
537.56
632.9
83.20
86.77
88.71
90.50
90.97
89.37
Press
R-Sq(Pred)
147
APPENDIX 3
BACKWARD ELIMINATION
Minitab out from backward elimination model selection procedure used for the
combined data for copper recovery in the flotation process with seven variables
(Alpha-to-Remove: 0.1, N = 31)
Table B
Step
Constant
-25.20
-29.93
-35.03
X1
0.053
0.053
0.053
T-Value
2.53
2.52
2.50
P-Value
0.019
0.019
0.019
X2
3.7
3.7
3.7
T-Value
1.73
1.72
1.71
P-Value
0.097
0.098
0.099
X3
0.69
0.69
0.69
T-Value
6.67
6.65
6.55
P-Value
0.000
0.000
0.000
X4
0.158
0.155
0.191
T-Value
3.05
2.97
4.53
P-Value
0.006
0.007
0.000
X5
-0.098
-0.080
T-Value
-1.42
-1.19
P-Value
0.168
0.247
X6
0.83
0.79
0.76
T-Value
3.51
3.38
3.23
P-Value
0.002
0.003
0.003
X7
-0.45
T-Value
-1.09
P-Value
0.287
148
3.75
3.77
3.80
R-Sq
94.56
94.28
93.95
R-Sq(adj)
92.91
92.85
92.74
8.0
7.2
6.6
680.517
632.917
537.569
88.57
89.37
90.97
C-p
Press
R-Sq(pred)
149
APPENDIX 4
Minitab out from best subset procedure used for the combined data for copper
recovery by the flotation process with seven variables.
i.
Predictor
Coef
SECoef
Constant
38.254
1.691
22.62
0.000
X3
1.101
0.085
12.94
0.000
Source
DF
SS
MS
Regression
5075.0
5075.0
167.55
0.000
Residual Errors
29
878.4
30.3
Total
30
5953.4
ii.
Redictor
Coef
SE Coef
Constant
44.919
2.058
21.83
0.000
X4
0.464
0.060
7.72
0.000
Source
DF
SS
MS
Regression
4004.3
4004.3
59.58
0.000
Residual Errors
29
1949.0
67.2
Total
30
5953.4
150
iii.
Predictor
Coef
SE Coef
Constant
38.493
1.477
26.07
0.000
X3
0.837
0.111
7.52
0.000
X4
0.168
0.052
3.18
0.004
Source
DF
SS
MS
Regression
5308.4
2654.2
115.22
0.000
Residual Errors
28
645.0
23.0
Total
30
5953.4
iv.
Predictor
Coef
SE Coef
Constant
50.675
5.736
8.84
0.000
X3
0.991
0.093
10.61
0.000
X5
-0.164
0.072
-2.25
0.032
Source
DF
SS
MS
Regression
5209.8
2604.9
98.09
0.000
Residual Errors
28
743.6
26.6
Total
30
5953.4
151
v.
Predictor
Coef
SE Coef
Constant
28.983
4.702
6.16
0.000
X1
0.053
0.025
2.12
0.044
X3
0.781
0.108
7.23
0.000
X4
0.170
0.049
3.42
0.002
Source
DF
SS
MS
Regression
5400.2
1800.1
87.87
0.000
Residual
27
553.1
20.5
Total
30
5953.4
vi.
Predictor
Coef
SE Coef
Constant
15.650
8.078
1.94
0.063
X3
0.834
0.099
8.41
0.000
X4
0.185
0.047
3.90
0.001
X6
0.761
0.265
2.87
0.008
152
Source
DF
SS
MS
Regression
5458.8
1819.6
99.34
0.000
Residual Errors
27
494.5
18.3
Total
30
5953.4
vii.
Predictor
Coef
SE Coef
Constant
6.132
8.392
0.73
0.472
X1
0.053
0.022
2.44
0.022
X3
0.779
0.094
8.29
0.000
X4
0.187
0.043
4.29
0.000
X6
0.761
0.244
3.12
0.004
Source
DF
SS
MS
Regression
5550.7
1387.7
89.61
0.000
Residual Errors
26
402.6
15.5
Total
30
5953.4
153
viii.
Predictor
Coef
SE Coef
Constant
X1
0.061
0.019
3.16
0.004
X3
0.776
0.093
8.34
0.000
X4
0.191
0.043
4.44
0.000
X6
0.917
0.118
7.78
0.000
Source
DF
SS
MS
Regression
102807
25702
1688.93
0.000
Residual Errors
27
411
15
Total
31
103218
Predictor
Coef
SE Coef
Constant
-26.57
27.57
-0.96
0.344
X2
3.831
2.398
1.60
0.122
X3
0.741
0.112
6.57
0.000
X4
0.189
0.046
4.09
0.000
X6
0.761
0.258
2.95
0.007
154
Source
DF
SS
MS
Regression
5503.0
1375.8
79.43
0.000
Residual Errors
26
450.3
17.3
Total
30
5953.4
x.
Predictor
Coef
SE Coef
Constant
-35.03
25.38
-1.38
0.180
X1
0.053
0.021
2.50
0.019
X2
3.744
2.188
1.71
0.099
X3
0.688
0.105
6.55
0.000
X4
0.191
0.042
4.53
0.000
X6
0.761
0.235
3.23
0.003
Source
DF
SS
MS
Regression
5592.9
1118.6
77.59
0.000
Residual
25
360.4
14.4
Total
30
5953.4
155
xi.
Predictor
Coef
SE Coef
Constant
8.563
9.118
0.94
0.357
X1
0.053
0.022
2.41
0.023
X3
0.778
0.094
8.20
0.000
X4
0.195
0.045
4.30
0.000
X6
0.782
0.248
3.15
0.004
X7
-0.305
0.423
-0.72
0.478
Source
DF
SS
MS
Regression
5558.9
1111.8
70.47
0.000
Residual Error
25
394.4
15.8
Total
30
5953.4
xii.
Predictor
Coef
SE Coef
Constant
-32.61
25.81
-1.26
0.219
X1
0.053
0.021
2.48
0.021
X2
3.746
2.207
1.70
0.103
X3
0.687
0.106
6.48
0.000
X4
0.199
0.043
4.54
0.000
X6
0.782
0.239
3.27
0.003
X7
-0.306
0.408
-0.75
0.461
156
Source
DF
SS
MS
Regression
5601.18
933.53
63.62
0.000
Residual Error
24
352.17
14.67
Total
30
5953.35
xiii.
Predictor
Coef
SE Coef
Constant
-29.93
25.54
-1.17
0.253
X1
0.053
0.021
2.52
0.019
X2
3.735
2.170
1.72
0.098
X3
0.693
0.104
6.65
0.000
X4
0.154
0.052
2.97
0.007
X5
-0.079
0.067
-1.19
0.247
X6
0.794
0.235
3.38
0.003
Source
DF
SS
MS
Regression
5612.92
935.49
65.95
0.000
Residual Error
24
340.43
14.18
Total
30
5953.35
157
xiv.
Predictor
Coef
SE Coef
Constant
-25.20
25.81
-0.98
0.339
X1
0.053
0.021
2.53
0.019
X2
3.735
2.162
1.73
0.097
X3
0.693
0.104
6.67
0.000
X4
0.158
0.051
3.05
0.006
X5
-0.098
0.069
-1.42
0.168
X6
0.833
0.237
3.51
0.002
X7
-0.450
0.412
-1.09
0.287
Source
DF
SS
MS
Regression
5629.65
804.24
57.14
0.000
Residual Error
23
323.71
14.07
Total
30
5953.35
158
APPENDIX 5
FORWARD SELECTION FOR GRADE
Minitab output from forward model selection procedure used for data on grade
by the flotation process with seven variables. (Alpha to enter 0.1, N=31)
Table - A
Step
Constant
14.01
19.26
15.88
20.16
23.33
X3
0.085
0.104
0.086
0.060
0.059
T-Value
3.05
4.19
3.51
2.31
2.34
P-Value
0.005
0.000
0.002
0.029
0.028
X7
-0.53
-0.52
-0.63
-0.59
T-Value
-3.26
-3.49
-4.21
-4.05
P-Value
0.003
0.002
0.000
0.000
X1
0.0191
0.0193
0.0193
T-Value
2.32
2.50
2.57
P-Value
0.028
0.019
0.017
X5
-0.044
-0.037
T-Value
-2.14
-1.83
P-Value
0.042
0.080
X6
-0.133
T-Value
-1.58
P-Value
0.126
1.81
1.57
1.46
1.37
1.33
R-Sq
24.31
45.12
54.27
61.11
64.66
R-Sq(adj)
21.70
41.20
49.19
55.13
57.59
C-p
24.3
12.2
8.0
5.3
159
APPENDIX 6
BACKWARD ELIMINATION. ALPHA-TO-REMOVE: 0.1
Minitab output from backward elimination model selection procedure used for
the combined data for copper on grade in the flotation process with seven variables.
(Alphatoremove 0.1,N = 31)
Table - A
Step
Constant
X1
T-Value
P-Value
X2
T-Value
P-Value
X3
1
21.58
0.0194
2.53
0.019
0.07
0.08
0.934
0.038
2
22.30
0.0194
2.59
0.016
3
20.55
0.0194
2.59
0.016
4
20.43
0.0216
2.95
0.007
5
16.48
0.0216
2.85
0.008
0.040
0.039
T-Value
1.02
1.25
1.21
P-Value
X4
T-Value
P-Value
X5
T-Value
P-Value
X6
T-Value
P-Value
X7
T-Value
P-Value
S
R-Sq
R-Sq(adj)
C-p
Press
0.320
0.018
0.97
0.343
-0.024
-0.94
0.356
-0.130
-1.51
0.144
-0.60
-4.02
0.001
1.36
66.04
55.71
8.0
129.905
0.223
0.018
0.99
0.334
-0.024
-0.96
0.345
-0.130
-1.54
0.136
-0.60
-4.10
0.000
1.33
66.03
57.54
6.0
124.851
0.237
0.028
1.84
0.078
0.041
3.76
0.001
0.045
4.05
0.000
-0.142
-1.71
0.100
-0.57
-3.99
0.001
1.33
64.72
57.67
4.9
118.041
-0.141
-1.68
0.105
-0.57
-3.98
0.000
1.34
62.65
56.90
4.3
116.114
-0.60
-4.07
0.000
1.39
58.59
53.99
5.1
73.6739
160
APPENDIX 7
Minitab out from best subset procedure used for the combined data copper
grade by the flotation process with seven variables.
I.
Predictor
Coef
SE Coef
Constant
14.0078
0.5563
25.18
0.000
X3
0.08539
0.02798
3.05
0.005
Source
DF
SS
MS
Regression
30.522
30.522
9.31
0.005
Residual Error
29
95.041
3.277
Total
30
125.563
ii.
Predictor
Coef
SE Coef
Constant
14.4922
0.4652
31.15
0.000
X4
0.03741
0.01361
2.75
0.010
161
Source
DF
SS
MS
Regression
25.938
25.938
7.55
0.010
Residual Error
29
99.625
3.435
Total
30
125.563
iii.
Predictor
Coef
SE Coef
Constant
20.472
1.689
12.12
0.000
X4
0.05177
0.01208
4.29
0.000
X7
-0.5994
0.1647
-3.64
0.001
Source
DF
SS
MS
Regression
57.924
28.962
11.99
0.000
Residual Error
28
67.640
2.416
Total
30
125.563
iv.
Predictor
Coef
SE Coef
Constant
19.256
1.681
11.45
0.000
X3
0.10437
0.02494
4.19
0.000
X7
-0.5266
0.1616
-3.26
0.003
162
Source
DF
SS
MS
Regression
56.653
28.327
11.51
0.000
Residual Error
28
68.910
2.461
Total
30
125.563
v.
Predictor
Coef
SE Coef
Constant
16.480
2.058
8.01
0.000
X1
0.021587
0.007575
2.85
0.008
X4
0.04479
0.01106
4.05
0.000
X7
-0.5979
0.1471
-4.07
0.000
Source
DF
SS
MS
Regression
73.563
24.521
12.73
0.000
Residual Error
27
52.000
1.926
Total
30
125.563
vi.
Predictor
Coef
SECoef
Constant
15.876
2.135
7.44
0.000
X1
0.019072
0.008204
2.32
0.028
X3
0.08593
0.02450
3.51
0.002
X7
-0.5249
0.1502
-3.49
0.002
163
Source
DF
SS
MS
Regression
68.147
22.716
10.68
0.000
Residual Error
27
57.416
2.127
Total
30
125.563
vii.
Predictor
Coef
SE Coef
Constant
20.435
3.082
6.63
0.000
X1
0.021614
0.007331
2.95
0.007
X4
0.04112
0.01092
3.76
0.001
X6
-0.14119
0.08395
-1.68
0.105
X7
-0.5700
0.1433
-3.98
0.000
DF
SS
MS
Regression
78.666
19.666
10.90
0.000
Residual Error
26
46.898
1.804
Total
30
125.563
164
viii.
Predictor
Coef
SE Coef
Constant
20.163
2.836
7.11
0.000
X1
0.019256
0.007710
2.50
0.019
X3
0.06011
0.02600
2.31
0.029
X5
-0.04375
0.02046
-2.14
0.042
X7
-0.6264
0.1489
-4.21
0.000
Source
DF
SS
MS
Regression
76.735
19.184
10.22
0.000
Residual Error
26
48.828
1.878
Total
30
125.563
ix.
Five variables best model YG = 20.5 + 0.0194 X1 + 0.0386 X3 + 0.0281 X4 0.142 X6 - 0.567 X7
Predictor
Coef
SE Coef
Constant
20.547
3.056
6.72
0.000
X1
0.019410
0.007490
2.59
0.016
X3
0.03855
0.03182
1.21
0.237
X4
0.02809
0.01526
1.84
0.078
X6
-0.14219
0.08320
-1.71
0.100
X7
-0.5671
0.1420
-3.99
0.001
165
Source
DF
SS
MS
Regression
81.267
16.253
9.17
0.000
Residual Error
25
44.296
1.772
Total
30
125.563
x.
Five variables best model YG = 23.3 + 0.0193 X1 + 0.0591 X3 - 0.0371 X5 0.133 X6 - 0.593 X7
Predictor
Coef
SE Coef
Constant
23.325
3.405
6.85
0.000
X1
0.019264
0.007496
2.57
0.017
X3
0.05909
0.02528
2.34
0.028
X5
-0.03712
0.02032
-1.83
0.080
X6
-0.13327
0.08416
-1.58
0.126
X7
-0.5926
0.1464
-4.05
0.000
Source
DF
SS
MS
Regression
81.187
16.237
9.15
0.000
Residual Error
25
44.377
1.775
Total
30
125.563
166
xi.
Predictor
Coef
SE Coef
Constant
22.300
3.562
6.26
0.000
X1
0.0194
0.007
2.59
0.016
X3
0.039
0.031
1.25
0.223
X4
0.018
0.018
0.99
0.334
X5
-0.02360
0.02453
-0.96
0.345
X6
-0.13009
0.08427
-1.54
0.136
X7
-0.6017
0.1467
-4.10
0.000
Source
DF
SS
MS
Regression
82.913
13.819
7.78
0.000
Residual Error
24
42.650
1.777
Total
30
125.563
xii.
Six variables second best model YG= 19.8+ 0.019 X1 +0.068 X2 +0.036 X3
+0.028 X4 -0.142 X6-0.567 X7
Predictor
Coef
SE Coef
Constant
19.796
9.152
2.16
0.041
X1
0.019400
0.007644
2.54
0.018
X2
0.0682
0.7828
0.09
0.931
X3
0.03690
0.03761
0.98
0.336
X4
0.02816
0.01559
1.81
0.084
X6
-0.14218
0.08491
-1.67
0.107
X7
-0.5671
0.1450
-3.91
0.001
167
Source
DF
SS
MS
Regression
81.281
13.547
7.34
0.000
Residual Error
24
44.282
1.845
Total
30
125.563
xiii.
Predictor
Coef
SE Coef
Constant
21.577
9.366
2.30
0.031
X1
0.019390
0.007662
2.53
0.019
X2
0.0658
0.7846
0.08
0.934
X3
0.03834
0.03773
1.02
0.320
X4
0.01824
0.01885
0.97
0.343
X5
-0.02360
0.02505
-0.94
0.356
X6
-0.13009
0.08607
-1.51
0.144
X7
-0.6017
0.1499
-4.02
0.001
Source
DF
SS
MS
Regression
82.926
11.847
6.39
0.000
Residual Error
23
42.637
1.854
Total
30
125.563
PAPER PUBLISHED
Abstract
Mathematical models were developed to give an insight to see the effect of process
variables propylxanthat (X1 g/ton), pH (X2), sodium sulphide (X3,g/ton) and sodium
cyanide (X4g/ton) on the recovery (YR) of copper.
The optimum recovery (YR) 62.95%, at X4=60g/ton were obtained (6,7)
Introduction
The simulation of mineral processing system design optimization of flotation
parameter and control is used for the least 30 years. (1-5, 9-11, 14, 20-23). Federally
Administrated Tribal Area Development Corporation carried out exploration work
and confirmed 122 million tons of estimated reserves.
The North Waziristan ore is a sulphide ore body, which contains chalcopyrite as the
ore mineral. The ore is of low grade within economic limit (6,8,17) therefore it must
be upgraded before it can be subjected to metallurgical treatment to obtain blister
copper. Extensive floatation test work were carried out to investigate effects of
various process parameters on recovery of copper. Effects of collector type and
dosage; pH, sulfidizer dosage; depressant were investigated during flotation test. The
results of the pilot scale studies showed that the copper content in the ore was
upgraded from 0.9% to 22-25% in a staged cleaning flotation with recoveries up to
80%. The recovery can be further enhanced by improving the machine efficiency and
conducting more research on reagents. (7,19,24).
______________________
*
University of Education Lahore
**
Department of Mining Engineering , N.W.F.P, U.E.T. Peshawar.
***
Department of Mathematics, COMSAT,Abbottabad.
EXPERIMENTAL WORK
Twenty tests were carried out to evaluate the flotation response, using different
dosages and type of collector. Five tests were conducted out each to investigate the
effect of individual parameter such as the collector type dosage NapX, pH, depressant
and sulphidizer on the grade and recovery of final concentrate copper.
Methodology
Applying Regression Analysis for enrichment of copper ore experiments were
conducted to study the effect of the collector type dosage, depressant, sulphidizer and
frother dosage on recovery of North Waziristan copper ore.
The most general type of linear mathematical model can be described with variables
Z1,Z2, .., Zp in the form as follows where stands for variations caused by other
than Z1, Z2 ..,
Y = oZo + 1 Z1 + 2 Z2 + .. p Zp + .. (1*)
Zo = 1 and stands for effects of the regression model
However, it is some times convenient to have a Zo in the model.
The following four mathematical models were used to estimate recovery of copper ore
in the final product based on first order, second order, logarithmic and exponential.
1.
3.
4.
Eq. (1)
Eq. (2)
Eq. (3)
YR = 29.541e0.0015X1, R2 = 0.8884
Eq. (4)
Eq.(5)
Eq. (6)
Eq. (7)
Eq. (8)
Eq. (9)
Eq(11)
Eq(12)
Eq(13)
Eq(14)
YR = 59.467e0.0012X4, R2 = 0.9367
Eq(15)
CONCLUSION
Suitable models for the effect of individual variable X1,X2,X3, and X4 on the recovery
YR, for the enrichment of copper are the equations 3,7,10 and 14 with high value of
R2. These all equations are quadratic one predictor variable. It was concluded with
high degree of confidence that the effect of processes parameters can be predicated by
these equation within the given rang. Maximum copper recovery were obtained when
the value of X1 is 200mg/ton, X2 is 11.58, X3 is 30gm/ton and the value of X4 is 60
gm/ton. Comparing the results of recovery of all above four parameters, the best
model is which gives the maximum recovery among all the parameters with high
value of R2 and is significant at the level of probability. However it will be more
appropriate if further models may be derived to have combined effect of these
parameters on the recovery of copper concentrate in the treatment of copper ore by
flotation process. Optimum copper recovery were obtained when X1 = 200g/ton, X2 =
11.58, X3 = 30 g/ton and X4 = 60 g/ton.
X4 gives the maximum recovery 62.95%. More models will be derived to have
combine effect of these parameters.
figure b
y = 0.0536x1 + 29
R 2 = 0.8897
%R
50
40
R 30
% 20
10
0
0
100
200
300
Dosage gm/ton
y = 6.5631Ln(x1) + 5.0807
R2 = 0.8619
50
40
30
20
10
0
0
50
100
150
200
250
300
Dosage gm/ton
figure c
figure d
50
50
40
40
30
R%
R%
y = 29.541e0.0015x
R2 = 0.8884
R2 = 0.9053
20
10
30
20
10
0
0
50
100
150
200
250
300
100
Dosage gm/ton
200
300
Dosage gm/ton
50
40
50
30
20
30
40
10
0
20
10
0
9.5
10
10.5
11
11.5
12
12.5
figure c
50
40
30
20
10
0
9.5
10
10.5
11
pH
9.5
10
10.5
11
pH
pH
R%
figure b
y = 4.0591x 2 - 5.1084
R2 = 0.5837
R%
R%
figure a
11.5
12
12.5
11.5
12
12.5
figure a
figure b
y = 12.69Lnx3 + 18.277
R2 = 0.8123
R = 0.821
80
R%
R%
60
40
20
0
0
10
20
30
70
60
50
40
30
20
10
0
40
10
20
Na2S gm/ton
70
60
50
40
30
20
10
0
0
10
40
y = 42.746e0.0127x
R2 = 0.836
figure d
R%
R%
figure c
30
Na2S gm/ton
20
30
70
60
50
40
30
20
10
0
0
40
10
20
30
40
Na2S gm/ton
Na2S gm/ton
64
63
62
61
60
59
58
0
20
40
60
64
63
62
61
60
59
58
0
80
20
40
R2 = 0.9938
62
R%
R%
63
61
60
59
20
40
NaCN gm/ton
80
y = 58.467e0.0012x
R2 = 0.9367
figure d
64
60
NaCN gm/ton
NaCN gm/ton
figure c
y = 1.8153Lnx4 + 54.903
R2 = 0.7831
figure b
R2 = 0.9348
R%
R%
figure a
60
80
64
63
62
61
60
59
58
0
20
40
NaCN gm/ton
60
80
REFERENCES
1.
2.
3.
Pitt, 1968. J.C Pitt, the development of system for continuous optimal control
of flotation plants. In: Computer Systems Dynamics and Automatic Control in
Basic Industries, I.F.R.C Symposium, Sydney (1968), pp. 165-171.
4.
Smith and Lewis, 1969. H.W. and C.L. Lewis, Computer control experiments
at lake default. Canadian IMM Bulletin 62 682 (1969), pp. 109-115.
5.
6.
7.
8.
9.
Dowling et al., 1985. E.C Dowling, R.R. Klimpel and F.F Aplan, model
discrimination in the flotation of a porphyry copper ore. Minerals and
metallurgical processing 2 (1985), pp.
10.
E.C Dowling, R.R.K limped, and F.F Aplan Model Discrimination in the
flotation of a porphyry copper ore May 1985, P 87-101.
11.
Van Orden, 1986. P.R. Van Orden, Simulation of continuous flotation cells
using observed residence time distributions. In: P. Somasundaran, Editor,
Agar et al., 1986 G.F, Agar, r. Stralton-Crawly and T.J. Bruee, Optimizing
the design of flotation circuits. CIM Bulletin 73 (1986), pp. 173-181.
13.
Dowling et al., 1987. E.C Dowling, R.R. Klimpel and F.F Aplan, use of
kinetic model to analyze industrial flotation circuits. In: D.R Gaskell et a,:.
International Symposium on Innovative Technology and Reactor Design in
Extractive Metallugy, TMS-AIME, Warrendale, PA(1987), pp.533-552.
14.
15.
Arnold and Aplan, 1988. B.J. Arnold and F.F. Aplan, A practical view of rate
and residence time in industrial coal froth flotation circuits. In: Processing
Fifth Annual International, Pittsburgh Coal Conference, Pittsburgh, PA.
September 12 16(1988), pp.328-341.
16.
Arnold and Aplan, 1989. B.J Arnold and F.F. Application of rate models to
pilot and commercial scale coal froth flotation circuits presented at the, In:
(1989).
17.
18.
Xu, 1998, M. Xu, Modified flotation rate constant and selectivity index.
Minerals engineering 11 (3) (1998), pp. 271-278.
19.
20.
10
22.
23.
M.Baro and L.Piga Mineral Processing Institute CNR Via Bolognola, 7-00138
Rome, Italy Comparison of pb-zn selective collectors using Statistical
Methods.
24.