0%(1)Il 0% ha trovato utile questo documento (1 voto)

104 visualizzazioni191 pagineThe 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.

Nov 15, 2015

© © All Rights Reserved

PDF, TXT o leggi online da Scribd

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.

© All Rights Reserved

0%(1)Il 0% ha trovato utile questo documento (1 voto)

104 visualizzazioni191 pagineThe 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.

© All Rights Reserved

Sei sulla pagina 1di 191

NORTH WAZIRISTAN COPPER

By

SARDAR ALI

Ph.D. Scholar

PAKISTAN

2007

EFFICIENCY OF FLOTATION PROCESS FOR

NORTH WAZIRISTAN COPPER

By

SARDAR ALI

University Registration No. 87-03

fulfillment of the requirements for the award of degree of

Doctor of Philosophy in Mathematics

with specialization in Mathematical Statistics at the Division of Science and

Technology, University of Education Lahore.

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

5.1

51

5.2

General Description:

51

5.3

Data for recovery of copper

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:

recovery of copper by flotation.

59

Table 5:

82

Table 6:

Analysis of Variance

82

Table 7:

84

Table 8:

85

Table 9:

variables

88

Table 10:

89

Table 11:

Tests for skewness, kurtosis and Jarque bera for four variables

90

Table 12:

Bin Frequency

90

grade of copper.

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:

(X4), and frother dosage (X5).

The figure (5.19) shows visual test for standard residuals of

seven variables.

Histogram

Standard Residual Plot

Plot of residuals

Histogram

Standard Residual Plot

Effect of collector (NaPX) on grade of copper

Effect of pH on grade of copper

Effect of depressant (NaCN) on grade of copper

Effect of sulfidizer (Na2S) on grade of copper

Effect of frother (pine oil) on grade of copper

Effect of pulp density on grade of copper

Effect of conditioning time on grade of copper

(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.

(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.

(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.

(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 depressant in the

flotation process.

(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.

(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.

(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.

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

Effect of sodium Sulphide (X4grams/ton) on the grade of

copper.

Copper grade (YG) response surface for sodium cyanide

(X3) conditioning time (X7).

Copper grade (YG) response surface for sodium Sulphid

(X4) conditioning time (X7).

Histogram

Testing for heteroscedasticity

Residuals are normal. It qualifies the visual test of

normality.

Conceptual general model for recovery and grad

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

gravity separation, magnetic separation and flotation) and of doing it for a profit. To

maximize profitability, accurate simulations of mineral processing unit operations

have been actively pursued as minerals processing technology has matured. The

modeling and simulation of mineral processing systems are inherently difficult

because they are multiphase (particles, fluids and air) and because the ore particles are

heterogeneous (size, shape, composition and texture). From the beginning, the rate of

development in mineral processing modeling was controlled by limitations in ones

understanding of the basic sub processes in each unit operation and in the

computational relationship required to solve model1 equations.

Mathematical statistics is an interdisciplinary subject aimed at developing

models and analytical methods for systems containing a substantial element of

random variation, often the motivation for the research is a practical problem

involving the development and analysis of a mathematical statistical model.

occurring in any activity.

statistically (empirically), which can be used for prediction of the phenomenon

(Pekkanen, 1998b). The development of a new process typically involves a lot of

experiments on several scales: Laboratory, bench scale and pilot stages.

This dissertation is mainly focusing on the state of mathematical modeling and

also to formulate mathematical models for grade and recovery of the North Waziristan

copper ore in Pakistan. The North-Waziristan copper ore is chalcopyrite. The ore is

of low grade within economic limit, therefore it must be upgraded before it can be

subjected to metallurgical treatment to obtain blister copper. The experimental work

was undertaken to upgrade the lean copper ore through flotation technique to make it

suitable for further metallurgical treatment to obtain blister copper.

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

Mathematical Models are required because;

and trial

mathematical model

It is the only course of action available for the improvement in the system

used in future for any alteration for improvement in the process

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

Mathematical models have been used in mineral processing system design

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.

suitable for metallurgical treatment

concentrator, saindak (Baluchistan) process of bentonite clay, enrichement of

uranium, purification of soap, stone and fertilizer industries.

1.4

1.

2.

scientific way.

3.

recovery and grade of copper in the research area of Pakistan.

4.

The North Waziristan copper ore is chalcopyrite (CuFeS2). Chemistry of

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).

Pakistan.

To improve and increase the efficiency of the process by scientific way and to

develop a correct mathematical model, so that in future any alteration or change in the

process can be improved by utilizing the mathematical models. Now-a-days every

chemical, mechanical and electrical process is governed by mathematical or statistical

models. These 5.67 and 6.65 models are the first mathematical models, which have

been developed for the mineral industry in Pakistan and of course these will pave the

way to run our mineral based Industry. Using such mathematical models to improve

their products quantity and quality. These models can be utilized specially in glass,

ceramics industry copper concentrator Saindak (Baluchistan) process of bentonite

clay, enrichment of Uranium, purification of soap, stone and fertilizer industries.

1.5

Copper is one of the very essential minerals in modern industry. It is a good

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

The study is based on primary data from flotation process experiments on

Engineering and Technology, Peshawar Pakistan with the assistance of the political

authorities of North Waziristan agency and Federally Administered Tribal Area

Development Corporation (FATA DC). Other relevant information about copper

deposits were also obtained from FATA DC. An inventory of the ore samples was

prepared and each sample was tagged with a number and weighted.

Both the chemical and mineralogical analysis of the samples were carried out

at Department of Mining Engineering Laboratories (MEL) and Mineral Testing

Laboratories, Sarhad Development Authorities (SDA), Peshawar. The mineralogical

investigations include X-Ray Diffraction, X-Ray Fluorescence and ore microscopy.

The chemical constituents were determined by classical and instrumental methods of

analyses. On site the samples were collected by blasting irregularly spaced holes

within the regularly spaced rows for minimum chances of errors. A total of 30 tons of

samples were collected, comprising of six sub samples weighing five tons each from

six different locations. The rows of holes drilled on each location were spaced at an

equal interval of 300 feet. The collected samples were transported to the NWFP,

University of Engineering and Technology Peshawar, through trucks.

1.7

Extraction of copper from the available copper ore in Pakistan is important.

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

For obtaining optimum level of variables for efficient flotation process to

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

Outline or organization of the study is as follows. Chapter one deals with

processing of North Waziristan copper ore, benefits of the present study, main

objective, scope, background and significance of the study. Chapter two presents

review of literature. Chapter three explains experiments, previous work, geology of

North Waziristan copper ore, location and accessibility Waziristan copper ore, and

occurrences of North Waziristan copper ore. Chapter four deals with methodology,

justification of the explanatory variables and flotation process. Chapter five presents

building of mathematical models. Chapter six consist mathematical models for grade.

Final and

recommendations.

10

CHAPTER 2

REVIEW OF LITERATURE

in mineral processing has been approached in various ways depending on the

philosophy of the researcher as well as the expected usage of the model and the

allowable investment in personnel and time. The goal of using these models is to

undertake plant scale tests on batch laboratory evaluations. Flotation tests were

carried out on samples of a US porphyry ore (Pinto Valley, AZ) by Dowling et al.

(1985). The ore was tested using various collector and frother system to produce

different time recovery profiles these were used to calculate flotation rate and ultimate

recovery parameters for each model. The models were then evaluated statistically to

determine the over all fit of the calculated to the observed data and to test the range of

significance of the parameters in each model.

Each flotation model will have an associated error. This error will effect and

can be measured by both the fit to the observed data and the range of statistical

significance of each parameters. Two types of errors were found due to experiment

and due to model. The researcher found the model variance S2r and compared the

optimal model variance to the model variance from a given change in the parameter

being assessed and F-value calculated. To determine the optimal flotation model

parameters, a generalized parameters estimation computer program was used (Klimpl

1980) by Dowling in his study. The criteria used for estimation of parameters value

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

recoveries of over 83%. A flow sheet depicting different products of flotation, for an

industrial concentrator, has also been suggested.

Elzinga E.J. Van J.J.M and Swartjes. F.A. (1999) worked on General purpose

Freundlich isotherms for cadmium, copper and zinc in soils. They have tried to derive

generally applicable isotherm for cd, zn and zn using data from batch sorption

experiments on a wide range of soils and experimental conditions. They used a

linearized logarithmic transformation of the Freundlich sorption equation.

Freundlich derived equations for cd, zn and zn using multiple linear regression

on batch sorption data. The equations were based both total dissolved metal

concentrations and free metal activities in solution. He calculated free metal activities

from total metal concentration talking into account ionic activity. The logarithmic

transformation of the Freundlich constant for cadmium was regressed on the

logarithmic transformations of cation exchange capacity. He used Minitab for

statistically analysis.

They used step-wise forward regression by evaluating the tratios, stepwise

2

could be attributed either to addition of an argument or to

improvement of R adj

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

remains in the sulfur flotation tailings and can be readily recovered by flotation with

different flotation reagents. When amyl xanthate is used, 85% of chalcopyrite can be

recovered with a copper grade of 14.5% in a single-stage froth flotation. The

chalcopyrite flotation concentrate can be sent back to chloride leaching circuits.

Cilek (2004) combined the classical first order kinetic model with a properly

built statistical model based on a factorial experimental design. In order to accurately

predict the rougher flotation efficiency for various flotation conditions, a three-level,

three factor experimental design was used to develop statistical model to predict each

of the kinetic model parameters as a function of the air flow rate, the feed grade and

the froth thickness. The statistical evaluation of the experimental results indicated that

the ultimate recovery, the rate constant and time correlation are not constant, but each

of these kinetic model parameters can be defined as a function of variables

considered. The rate of change in the kinetic parameters due to the feed grade

fluctuation and their effects on the metallurgical performance can accurately be

predicted by using the models developed. To reduce the detrimental feed grade

fluctuations on the metallurgical performance, the operating variables of the flotation

can be manipulated to obtain the desired concentrate grade. Cilek obtained the results

of the statistical evaluation; the rate data were used to build a statistical model

considering the variables.

Among all models the following models, which were built by using piece wise

(or breakpoint) linear regression method were selected.

Km = (0.4273-0.52 f+0.0051Qa+0.617Tf-0.05QaTf) , km2.24

Km= (3.565+0.38f-0.31Tf+0.104Qa+0.003QaTf), km<2.24,R2=0.9431.........(1)

16

bm= (3.695-3.75f-0.08Qa+0.348Tf-0.694fTf+0.097QaTf),b<0.406, R2 = 0.8945.....(2)

RIm= [27.973f-1.337Qa+0.89Tf(25.33-8.64f-Qa)-9.592Qa-0.549], RIm69.3

RIm= [75.284-4.811f+0.015Qa+0.654Tf(9.896+f-Qa)-0.115fQa],Im69.3,R2=0.9022..(3)

topt = [0.538+0.77f-0.112f Tf+0.02Qa(2.2+Tf-2.4f)] topt 2.32

topt=[4.097-0.867f+0.2fTf+0.051Qa(1.47-Tf+1.087f)],topt<2.32, R2=0.9087...........(4)

The high R2 values for all the responses reveal that the experimental data

provide evidence to indicate that the developed models satisfactorily predict the

Kinetic Parameters, where topt RI, k and b are the optimum flotation, time ultimate

recovery, the rate constant and the time correction factor also Tf, f, Qa denotes pulp

level, feed grade and factor respectively.

Barbaro and Piga (1998), adopted statistical approach to evaluate the Pb-Zn

selectivity of various organic collectors of the Mercaptobenzothiazole (MBT) and

aminothiophenol (ATP) types, in the flotation of lead and zinc minerals. Six

replicated tests were performed using each collector in order to obtain an estimate of a

statistical population characterized by an average and a variance. Comparison of these

statistical populations indicated the most selective collectors. The selectivity exhibited

by the collectors was then related to their molecular structure.

Horbstand and Potapov (2004), reported that mathematical simulation have

been used in mineral processing system design, optimization and control for more

than 30 years. Presently a new set of simulation tools based on the physics of the

underlying processes has been developed. Because these models provide accurate

micro scale simulations of equipment and process behavior, these high-fidelity

17

importance to the mineral processing industry.

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

Peshawar.

3.2

North Waziristan area remained unexplored before seventies. It was in 1985,

geological base map of about 2350 sq. km on 1:50,000. Investigation resulted in

delineating certain prospective areas having copper mineralization. Detailed

topographical mapping on scales 1:10,000, 1:1000 and 1:500 were conducted in the

mineralized zones. Petrographic study of the representative rock samples was made

before starting pilot plant operations. Geochemical studies of grid samples had

already been carried out at laboratory scale. Probe core drilling in two copper

mineralized bodies has just been completed in collaboration with the technical

expertise of China. According to the estimates given by the Chinese geologist there

are 80 million tons of confirmed reserves of copper ore having an average copper

content of 0.8%. The grade increases with depth and at some places it is 1%,

sometimes approaching 2 to 5%.

3.3

The tribal belt lies at the Pak-Afghan Border. This belt is divided into seven

Waziristan and South Waziristan and four frontier regions attached to Peshawar,

Kohat, Bannu and Dera Ismail Khan.

Investigation in the southern region have revealed the presence of copper

mineralization at various places in North Waziristan. Boya, an important locality, is

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

Reserves and reserve based estimates for Australia, Chile, China and Poland

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.

Area

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

Few more new discoveries have been made in Australia, Chile, Peru, British

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

Copper mineralization has been reported in Chagai (Saindak), Loralai and

Zhob districts. Both copper sulphide and oxide minerals as well as native copper have

been reported in Chaghi district. Copper minerals in these areas occur mainly as

sulphides, oxides, silicates and carbonates. The minerals are found in disseminated

form in association with quartz, galena, hematite, siderite and monazite (Zaki, 1969).

In chaghi the amount of potential deposits of copper is reported as 729mt (0.64% Cu

and 0.39g/t Au (Malik,2003). Saindak copper deposits comprise of copper carbonate,

chalcopyrite,

volcanic tuffs and shales, and limestone of Eocene age.

3.8

Chalcopyrite looks like, and is easily confused with pyrite and is also one of

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:

and Siderite. Sphalerite and tetrahedrite are a few of the

most Common.

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

The host rocks in this area are normally breccia composed of fragments of

lava flow, ultra basic and intermediate igneous rocks. At some places breccia occurs

with altered andesitic, rhyodacitic, granodioritic, ultrabasic, doleritic and jesperitic

rock fragments. Still at some other places it is a simple breccia. The associated rocks

commonly found are ultrabsic, dolerite, volcanic, andesite, and lava flow. Jerositic is

the gossan type found in the area. The associated minerals are malachite, azurite,

pyrite, chalcopyrite and chalcocite. The extension of mineralized bodies varies from

960-17400 square meters.

3.9.2 Degan area

The host rocks in this area are normally composed of breccia. The associated

rocks commonly found are lava flow, andesite, interbedded limestone, shale, diorite,

and dolerite. Jerositic is the gossan type found in the area, the same as found in the

shinkai area. The associated minerals are malachite, azurite, pyrite, chalcopyrite and

manganese. The extension of mineralized bodies varies from 2604-108800 square

meters.

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

University of Engineering and Technology Peshawar. The least square fitting

procedure is used for data analysis as purely descriptive technique. Computer

algorithms Minitab statistical software and Microsoft Excel were used for developing

best mathematical models for the efficiency of seven process variables.

30

variable on recovery and grade of copper. Simple and multiples regression were used

for developing the models.

The most general type of linear mathematical models can be described with

variables X1, X2, -----, Xp. in the form as follows where stands for variation caused

by other than X1, X2, -------.

Y= o + 1 X1 + 2 X2 + . p Xp +

4.1

The principle of least squares (LS) consists of determining the values for the

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)

yi i xi i

.. (4.3)

The model y = 0 + 1x1 + 2x2,+ 3x3+,.+ pxp

(4.4)

that involves more than one regressor variables. The parameters j, j = 0,1,2,----,p are

called the regression coefficients.

This model describes a hyper plane in the k dimensional space of the

regressor variables xj. The parameters j represents the expected change in the

response y per unit change in xj when all of the remaining regressor variables xi

(i j) are held constant.

For this reason the parameters j, j = 1, 2, ..p, are often called partial

regression coefficients. Multiple linear regression models are often used as empirical

models or approximating functions. That is the true functional relationship between y

and x1, x2, ., xp known, but over certain ranges of the regressor variables the linear

regression model is an adequate approximate to true unknown function.

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.

Observation

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.

yi = 0 + 1xi1 + 2xi2,+ 3xi3 +,----+ pxip + i = 0 +

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)

squares estimations of 0, 1, ..,p must satisfy.

S

0

0 ,1 ,....... p

n

P

2 y i 0 j x ij 0 (4.7)

i 1

j 1

0 ,1 ,....... p

n

P

i 1

j 1

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

These k = (p +1) equations are called the normal equations, one for each of

the unknown regression coefficients. The solution to the normal equations will be the

least-square estimators ( 0 , 1 ,......, p ). It is more convenient to deal with multiple

regression models, if they are expressed in matrix notation. This allows a very

compact display of the model, data, and results. In matrix notation, the model given

by Equation (4.5) is

Y = X +

where

0

1

y1

1 x11 ..........x1P

2

1

y2

1 x .........x

.

.

.

2P

, Y , ,

X 21

: :

:

.

.

.

.

.

1 x n1 .........x nP

y n

p

p

36

levels of the regressor variables. is a k x 1 vector of the regression co-efficient and

To find the vector of least-square estimators. , that minimizes

S() =

i 1

2

i

( y X )( y X )

S() = y y X y y X X X = y y 2 X y X X

Since X y is 1 x 1 matrix and its transpose ( X y )/ = yX is the same

scalar. The least square estimator must satisfy.

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

normal equation (4.9) is obtained. In this display we see that X/X symmetric matrix

and X/y is a k x *1 column vector. The special structure of the X/X matrix. The

diagonal elements of X/X are the sums of squares of the elements in the columns of X,

and the off-diagonal elements are the sums of cross products of the elements in the

columns of X. The elements of X/y are the sums of cross products of the columns of X

and the observations yi.

The fitted regression model corresponding to the levels of the regressors

variables x/ = [1, x1, x2, , xp] is

p

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)

demonstrated. Consider first bias:

E( ) = E[(X/X)-1X/y] = E [(X/X)-1X/(X + )]

= E[(X/X)-1X/X +(X/X)-1X/]=

since E() = 0 and (X/X)-1 X/X = I. Thus, is an unbiased estimator of .

The variance property of is expressed by the covariance matrix.

Cov ( ) = E{[ -E( )][ - E ( )]/}

39

of j and whose (ij)th off-diagonal element is the covariance between i and j.

the covariance matrix of is

Cov ( ) = 2(X/X)-1

Therefore, if we let C = (X/X)-1, the variance of j is 2Cjj and the covariance

between i and j is 2Cjj. The least-square estimator is the best linear unbiased

estimator of (the Gauss-Markov theorem).

4.3.3 Estimation of 2

residual sum of squares

n

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

since p parameters are estimated in the regression model. The residual mean squares

is

MSRes =

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:

relationship between the response y and any of the regressor variables

x1, x2, , xk. This procedure is often thought of as an overall or global test of model

adequacy. The appropriate hypotheses are:

H0: 1 = 1 = .. = p = 0

H1: j 0 for at least one j

Rejection of this null hypothesis implies that at least one of the regressor

x1, x2, ..xp contributes significantly to the model.

The test procedure is a generalization of the analysis of variance used in

simple linear regression. The total sum of squares SST is partitioned into a sum of

squares due to regression, SSR, and a residual sum of squares, SSRes. Thus,

SST = SSR + SSRes

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

also

E(MSRes) = 2

E(MSR) = 2 +

* X c/ X c *

p 2

Efroymson (1960). Stepwise regression is a modification of forward selection in

which at each step all regressors entered into the models are tested.

A regressor added at an earlier step may now be redundant because of the

relation ship between it and regressor now in the equation. If the partial F-statistic for

a variable is less than Fout that variable is dropped from the model. Stepwise

regression requires two cut off values, FIN and Fout. Some analysists prefer to choose

FIN = Fout, although this is not necessary. Frequently choosing FIN > Fout, making it

relatively more difficult to add a regressor than to delete one.

4.3.6 Studentized Residuals

consistent with the rest of other data the hat matrix is used to identify high leverage

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

usual range of variation of SK(r*) under the null hypothesis that the regression errors

are normal. If the observed statistic falls within the usual range of variation, we will

accept the null hypothesis of normal errors. If it falls outside the usual range then we

will reject the null hypothesis.

4.3.8 Testing for Heteroscedasticity

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.

In the standard regression model, we assume that the error terms are i.i.d. with

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

regressor has any effect on the dependent variable or not. Consider the regression

model

Y(t) = 1 + 2 X2 + 3 X3 + + k Xk + (t).. (4.17)

The null hypothesis that 3=0 says that X3 does not belong in the regression

equation. Equivalently, it says that X3 has no effect on the dependent variable Y(t).

This situation arises when we do not know which variables have an effect of Y

and which do not. We often have a list of variables all of which are potential

candidates for explanatory variables for Y. In this case, we are genuinely interested in

the null hypothesis, and wish to find out whether or not X3 affects Y. The t-test

provides a way of doing this. In such situations, it is often the case that if we find out

that X3 is not significant, we take it out of the regression. We might then try some

other variable. There are a number of ways of including and excluding regressors on

the basis of t tests to try to arrive at a particular set of best explanatory variables. Such

procedures are called stepwise regression procedures.

45

4.4

Scale Studies

This research work is based on primary data. The representative samples were

Engineering and Technology, Peshawar, with the assistance of the political authorities

of North Wazirsitan agency and Federally Administrated Tribal Area Development

Corporation (FATA DC). Other relevant information about copper deposits was

obtained from FATA DC. The survey conducted by FATA DC his conformed a

minimum of 122 million tons of inferred reserves of copper ore Boya-Datta Khel area

about 40 kms from Miran Shah. The average content of this copper ore is 0.3865%

varying from 0 to 100 feet. The Copper content increases with depth and at places it is

.90% which is better than that found at Saindak (Baluchistan) (Badshah 1983, 1985,

1996). An inventory of the ore samples was prepared. Each sample was tagged with a

number and weighed.

Both the chemical and mineralogical analyses of the sample were carried out

at Mining Engineering Laboratories (MEL) and Mineral Testing Laboratories (SDA),

Peshawar. The mineralogical investigation include X-Rays Diffraction, X-Rays

Fluorescence and ore microscopy. These chemical constituents were determined by

classical and instrumental methods of analyses. On site the samples were collected by

blasting the irregularly spaced holes within the regularly spaced rows for minimum

chances of errors. A total of 30 tons of sample was collected, comprising of six sub

sample weighing five tons each from six different locations. The rows of holes drilled

on each location were spaced at an equal interval of 300 feet. The collected samples

46

truck.

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

hydrophobacity.

4.5.4 Sulphidizer (Na2S) (X4):

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

solution. The latter is disassociated with the formation of hydrosulphide ion.

4.5.5 Frothers Dosage (X5):

hydrophobicity.

4.5.6 Frothers:

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):

concentrate grade. Flotation time is dependent on mineral floatability, grinding

fineness, reagent scheme and other conditions.

In flotation process the raw ore is ground with water, the thick pulp 30% water

is prepared by adding various reagents having specific purposes, the copper particles

are floated and/recovered materials is called % recovery and % copper content in it is

called % grade.

Mathematically if C is the concentrate, c is the metal weight in the

concentrate, if f is the average feed and F is the feed weight then % recovery =

(Cc/Ff) x 100

Flotation is one of the most important mineral concentration techniques. It is

known that the appearance of the froth in the flotation cells tells much about the state

of the flotation process. A machine vision measurement device was used to compute

dynamical, morphological and colour variables of the froth on the top of a flotation

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

the minerals; these properties can be modified by the use of suitable chemical

treatment. Chemicals that are used in the flotation process can be roughly divided into

three different categories: collectors, frothing agents and regulators. The task of

collector chemicals is to make valuable minerals hydrophobic. The frothing agent is

used for lowering the surface tension of water; this makes froth, which forms on the

top of the flotation cell, viscous and stable enough. The regulator chemicals control

the selectivity of the flotation process. These chemicals are divided into two

subgroups depending on whether they ease flotation of certain minerals, in which case

they are called activators, or make more difficult the flotation of unwanted minerals,

in which case they are called depressants.

51

CHAPTER 5

MODELS BUILDING

5.1

General to simple strategy was used to construct mathematical models to

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:

However, statistical inferences about data depend upon assumptions about the process

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)

variables. The variable highly correlated with dependant variable is used to

develop simple one variable model, the partial correlations of the remaining

independent variables with dependent variable are calculated and the variable

with highest partial correlation is then included in the simple model and so on.

Here we have employed both the methods in our study.

53

5.3

The data given in Table 3 were recorded in a series of seven experiments

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

Flotation Process Variables (X1 to X7)

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

of pulp (X2) on the recovery of copper, while keeping collector level at the optimum

found in experiment one, and all the other five variables constant. Five levels of pH

were used in the experiment. It is evident from Figure 2 that increase in pH from 10 to

11.58 increased recovery, the fourth level of pH studied gave the highest recovery and

thus pH of 11.58 is optimum for better recovery because pH beyond 11.58, decreased

recovery. This decrease is due to the deactivation of NaOH on the copper minerals.

5.3.3 Effect of variation in depressant, NaCN (X3)

depressant NaCN (X3) on the recovery of copper, while keeping all other variables

constant. The effect of depressant on recovery of copper ore is shown in Figure 3. The

figure clearly shows that recovery increased with increase in depressant dosage upto

25g/ton. Levels of depressant higher than 25 g/ton decreased recovery of copper. This

decrease might be due to the deactivation of copper particles in the pulp by sodium

cyanide as complex cyanides. From the figure 3 it is clear that the optimum dosage of

depressant is 25g/ton.

5.3.4 Effect of variation in sulfidizer, Na2S (X4)

sulfadizer (Na2S) on the recovery of copper, while keeping the first three variables at

the optimum levels found in the previous experiments and the other three variables

constant. The effect of sulfidizer on recovery of copper ore is shown in Figure 4. The

curve in Figure 4 shows that recovery was maximum at 50g sulfidizer per ton of feed;

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)

pine oil (X5) on the recovery of copper, while keeping the first four variables at the

optimum levels found in first four experiments, and variables six and seven constant.

The effect of frother dosage investigated in the fourth experiment is shown in Figure

5. Frother imparts stability to the mineral froth and helps in achieving maximum

recovery.

The optimum frother dosage was found to be 46g of pine oil per ton of feed.

5.3.6 Effect of Pulp density (X6)

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)

conditioning time (X7) on the recovery of copper, while keeping all other variables at

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)

(NaPX) ON RECOVERY OF COPPER

OF COPPER

(d)

(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)

(NaCN) ON RECOVERY OF COPPER

ON RECOVERY OF COPPER

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)

OIL) ON RECOVERY OF COPPER

(g)

% Recovery

80

70

60

50

40

30

20

0

10

15

20

TIME ON RECOVERY OF COPPER

20

30

Pulp Density (%wt/vol)

40

RECOVERY OF COPPER

59

5.4.

To develop mathematical models for recovery of copper involving one

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

R2 = 0.9053

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.

straight lines model which gave best fit followed by quadratic model which gave good

fit to the copper recovery data as affected by sulphidizer in the flotation process X-

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

R2 = 0.4242

% 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

R2 = 0.9222

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)

EXPONENTIAL MODELS FITTED TO THE RECOVERY OF COPPER

DATA FROM THREE LEVELS OF FROTHER DOSAGE IN THE

FLOTATION PROCESS.

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)

R2 = 0.7773

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

R2 = 0.9258

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

R2 = 0.9533

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.

reveals that the predicted values are with in the range when not extrapolated. The

most suitable models for the effect of the seven individual variables on recovery for

enrichment of copper are models 5.6,5.12,5.18,5.24,5.27,5.34 and 5.40 for X1, X2, ----, X7 respectively based on R2. In other words, two straight line equation give best fit

for X1,X2, X3, X4 and X7 while power model give good fit for X5, and exponential

give best fit for variable X6.

20

20

72

5.5.

The data from all experiments were used to construct models for recovery of

copper ore. The following four strategies were followed for model selection.

1.

2.

3.

4.

5.

procedure.

6.

Multiple regression model with testing Apt mess of model and checking the

assumptions.

73

5.6

building

The forward stepwise selection procedure was used to select variables for

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)

strategy, the model involving X3 was the first model fitted. The program then selected

X4 and the next model involved X3 and X4. The process was continued till no more

variable met the criteria of entering in the model. The final model had intercept and

five variables, X1, X2, X3, X4, and X6. In five variables model intercept and X2, are

not important so we drop intercept and X2. Thus model involving X1, X3, X4, and X6

without intercept is the best model.The improvement in R2 of model five over model

four is very meager to 0.

When stepwise procedure was used for model selection, it gave the same

result as forward selection.

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

model building

The backward elimination procedure or general to simple model building

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)

the full model with seven variables was fitted first Variable X7 had not much

contribution so it was eliminated first, X5 was eliminated next. In model including five

variables intercept, X2 were not statistically significant so we drop intercept and X2

from this model and we obtained the best fit model as given in the forward selection

procedure.

75

5.8

The total all possible regression models involving seven variable are 128, it is

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

CYANIDE (X3)ON THE RECOVERY OF

COPPER.

64

63

62

61

60

59

58

0

20

40

60

Sodium Sulphide (g/ton)

80

SULPHIDE (X4)ON THE RECOVERY OF

COPPER.

model than other five models. Based on R2, F-value, t-statistics and P-value (as shown

in the output, given in appendix 3) model having X3 is better than model having X4.

The first model show that thirty g/ton of depressant gave maximum recovery 71.3%

and Recovery increased at the rate 1.10% per one-gram increase in depressant. The

second model show that recovery increased at the rate of 0.465% per one gram

increase in sulfidizer.

Among the twenty-one models in the subset with two predictor variables, the

two best regression equation involving two predictor variables are:

YR = 38.5 + 0.837 X3 + 0.168 X4(5.52)

YR = 50.7 + 0.991 X3 - 0.164 X5.(5.53)

The equation (5.52) involving X3 and X4 explained 89.2% and the equation

(5.53) involving X3 and X5 explained 87.5% of the variation in the recovery of

copper.

78

Response surfaces were developed for the variables involved in the above two

equations.

YR

X4

X3

CYANIDE (X3) AND SODIUM SULPHIDE (X4).

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)

equations (5.59) and (5.60) are not statistically significant. The full model is given

below, the improvement in R2 is very small over model (5.61) and (5.62).

YR = - 25.2 + 0.053 X1 + 3.74 X2 + 0.693 X3 + 0.158 X4 -0.0982 X5 + 0.833 X6 0.450 X7.(5.63)

From the information provided by the best subset procedure, it is concluded

that equations with four predictor variables explains sufficient variation in the

recovery of copper ore data from all the experiments. The best equation involves

variables sodium propylxanthate, sodium sulphide, sodium cyanide and pulp density.

The second best equation involves pH, sodium sulphide, sodium cyanide and pulp

density.

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

An other approached followed was that the full model was fitted and based on

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

we use t-statistic. The t-statistic, and its probability show that intercept, X2, X5, and

83

and fitted a reduced model check the assumptions of the full model tests to nosuality

of residuals give some description.

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

Calculated

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

3

2

1

0

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31

-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.

We discard intercept, X2, X5 and X7, which are not significant therefore our

reduced model is

YR = 1X1+ + 3X3+ 4X4+ 6X6+ R (5.65)

Here are true parameters that we want to conjecture and residuals is identity

independent Normal with mean 0 and variance 2.

Note that the stepwise regression procedures is a little bit risky procedure

there is a chance of loosing some valuable information but over all performance can

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

(i)

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

(ii)

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

Another way of checking the normality is as follows:

iii)

2

1.5

1

0.5

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

v.

Structural stability:

Chow Test

2.015

P. significance

0.048

Next natural question is that can we really drop few variables and whether our

new model is better than before.

We can perform F-test to see if the removal of the three theoretically least

important regressors X2, X5, and X7 has made any significant difference.

The later of F-statistic can be computed as follows:

F = [(CRSS-URSS)/2]/[URSS/(T-K)]

= [(382.5-311.05)/2]/[311.05/(31-4] = 3.1004,

where CRSS=Constrained residuals sum of square.

and URSS=Unconstrained residuals sum of square, T=Number of observations=31

K = Number of regressors = 4

This has degrees of freedom 2 and 27

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

collector propylxanthate (X1) on the grade of copper, while keeping all the other

regressors constant. The response of propylxanthate to grade of the chalcopyrite ore is

shown graphically in Figure 25 with increase in dosage of propylxanthate there was a

corresponding increase in grade of copper. Beyond the dosage of 200g/ton of

propylxanthate there was no increase rather decrease in grade occurred as shown in

Figure 25. Next experiment was conducted to investigate the effect of variation in

collector pH (X2) on the grade of copper, while keeping all other variables constant

Figure 26, shows that pH (X2) at 11.58 gives the maximum grade. But pH greater then

11.58 decreased the grade. The third experiment was carried to investigate the effect

of variation in depressant sodium cyanide (X3), on the grade of copper, while keeping

all other variables constant. The results are given in Figure 27. Grade increase within

increase in sodium cyanide up to 25/gon, however, further increase in depressant

caused decrease in grade. Grade was maximum at 50g/ton dosage of sulfidizer (X4)

and beyond that dosage there was a slight decrease in grade of copper as shown in

Figure 28. The next experiment was conducted to investigate the effect of variation in

pine oil (X5) on the grade of copper, while keeping all other regressors constant. The

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

ROPYLXANTHATE ON GRADE OF

COPPER

12.5

OF COPPER

(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

GRADE OF COPPER

(e)

40

ON GRADE OF COPPER

% Grade

11.5

40

60

80

Pineoil (g/ton)

(PINE OIL) ON GRADE OF COPPER

19

17

15

13

11

9

7

5

0

10

20

30

40

ON GRADE OF COPPER

97

(g)

19

% Grade

17

15

13

11

9

7

5

5

10

15

20

Conditioning time(minutes)

TIME ON GRADE OF COPPER

6.2

For developing single variable mathematical models for the grade of copper,

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

grade of copper.

YG = 0.0182X1 + 10.67

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

R2 = 0.9421

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

R2 = 0.9868

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.

give good fit while the other models gave poor fit. The calculated X-max show that 18

gram per ton of sulfidizer will give the highest grade of copper. The two straight lines

model gave excellent fit with an R2 of 0.9808 indicating that it explained 98% of the

variation in data for grade of copper.

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)

R2 = 0.7114

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

R2 = 0.9808

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

R2 = 0.2709

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

R2 = 0.3567

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

R2 = 0.9115

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

R2 = 0.9973

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

R2 = 0.9741

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

R2 = 0.9994

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

The data from all experiments were used to construct models for grade of

copper ore. The following four strategies were followed for models selection.

1.

2.

3.

4.

procedure

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)

variable, equation (6.44) in three-predictor variable, equation (6.45) in four-predictor

variables were selected for the grade of copper by forward selected procedure.

111

Table14:

X1

Flotation Process Variables (X1 To X7).

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

It difficult to fit and test all the possible regression models involving seven

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

M1: YG = + 1X3

M2: YG = + 1X4

Models with two variable are:

M3: YG = + 1X4 + 2 X7

M4: YG = + 1X3 + 2 X7

Models with three variable are:

M5: YG = + 1X1 + 2 X4 + 3 X7

M6: YG = + 1X1 + 2 X3 + 3 X7

Models with four variable are:

M7: YG = + 1X1 + 2 X3 + 3 X5 + 4 X7

M8: YG = + 1X1 + 2 X4 + 3 X6 + 4 X7

Models with five variable are:

M9: YG = + 1X1 + 2 X3 + 3 X4 + 4 X6 + 5 X7

M10: YG = + 1X1 + 2 X3 + 3 X5 + 4 X6 + 5 X7

Models with six variable are:

M11: YG = + 1X1 + 2 X3 + 3 X4 + 4 X5 + 5 X6+6 X7

M12: YG = + 1X1 + 2 X2 + 3 X3 + 4 X4 + 5 X6+6 X7

114

M13: YG = + 1X1 + 2 X2 + 3 X3 + 4 X4 + 5 X5+ 6 X6+7X7

The regression equations for single predictors for recovery of copper obtained

form least square analysis are as follows:

YG = 14.0 + 0.0854 X3. (6.51)

YG = 14.5 + 0.0374 X4. (6.52)

The R2 show that the first equation explained 24.3% of the variation and the

second equation explained 20.7% variation in the grade of copper using flotation

equation process. The coefficient of equation (6.51) are different then coefficient of

(6.13), because the equation (6.13) is based on data from on experiment and equation

(6.51) is based on data from all experiments. Similarly the coefficients of equations

(6.19) and (6.52) are different because one is based on data from a single experiment

having five treatments and the other is based on combined data from seven

experiment having thirty-one treatment.

% Grade

20

18

17.5

17

16.5

16

15.5

15

15

10

%Grade

5

0

0

10

20

30

40

CYANIDE (X3) ON THE GRADE OF

COPPER.

20

40

60

80

SULPHIDE (X4) ON THE GRADE OF

COPPER.

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

CYANIDE (X3) AND CONDITIONING TIME (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

SULPHID (X4) AND CONDITIONING TIME (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

To find out the best fit model for grade, a full model was filled first that

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

For each regressor, a regression produces

A point estimate

A standard error

The point estimate helps us determine economic significance.

row and comes up heads each time, should I suspect that the coin is not a fair one with

equal chance of heads and tails?

In engineering data, we usually dont have the luxury of adding more

observations to help us decide whether a result is due to chance. So we perform a

hypothesis test.

First, statistical significance. The difference is statistically significant, by

definition, if the 95% confidence interval does not overlap zero, or if the p value for

the effect is less than 0.05. Values of 95% or 0.05 are also equivalent to a type one

error rate of 5%: in other words, the rate of false alarms in the absence of any

population effect will be 5%. It has to be 5%, or less preferably, but like most

researchers I opted for 5%.

6.8.2

Sample Size?

significance of outcome measure. We have to specify the smallest effect we want to

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

We have the model

Here we have a linear model with the constraints of upper limits, which are

obvious from the data. We are choosing variable for inclusion solely on the basis of

statistical significance because we are not removing any regressor entirely. All or

their few combinations are necessary for experiments.

125

justifiable. If X1, X4, X7 all are collectively zero even though we can get copper grade

because other four regressors (here working as a constant) are presents.

In the model we have a negative coefficient of regressor X7 conditioning time

theory suggests the maximum values of grade 18.2 g/ton at 13 minutes can be

obtained. However, beyond that, the grade has markedly decreased due to the

entrapped of fine slim particles. That if we increase the optimum pulp density we will

get poor copper grade.

coefficient

If we generate the data from the model then this generated data will be

meaningless, as it cant be compared with any other set of data because no such data

is available.

Though we have tested a lot of assumptions about the validity of the model

still we are uncertain about its results.

To overcome this problem one can randomly removed few observations from

the given set of data and again built model over same assumptions and then predict

the values that he had removed.

By simulation we checked that the coefficients are reasonable.

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

For Grade (YG) the variable X7 gives optimum results at 12 minute. For this fixed

value of X7, YG is further modified and YG = 0.019X1+0.0858X3+9.6

These mathematical models for recovery and grade, are strictly based on the

data provided by the Department of Mining Engineering N.W.F.P University of

Engineering and technology Peshawar. These models may not be valid for another

data if that do not conform to over data.

128

the above mathematical models and conclusions drawn from the models may be

verified.

A proposal amounting to Rs. 200 million was prepared by Prof. Dr.

Muhammad Mansoor Khan, Dean, Faculty of Engineering and was submitted for

approval to the Pakistan Science Foundation for enrichment and production of 99.9%

pure copper of North Waziristan Copper Ore.

129

REFERENCES

Acherman, P. K., Harris, G. H., Klimpel, R.R., Aplan, F. F., 2000. Int. J.

Miner. Process (Elsevier), Vol. 58, No. 1.

design. In: Yarar, B., Dogan, Z.M. (Eds.), NATO ASI Series E, 122. Martinus

Nijhoff Pub., Dordrecht, pp. 269287.

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.

and Chemical Engineering. Mir Publishers, Moscow.

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.

FATA development corporation, vol-11, pp. 2-13.

FATA development corporation, vol-3, pp. 14-18.

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.

interactions of the operating variables. Mineral Processing Division

Department of Mining Engineering, Faculty of Engineering, Suleyman

Demirel University, Isparta, TR 32260, Turkey. Vol. 17, No. 1. Pp. 81-85.

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.

Medium, Minerals Science. Engineering, Vol. 6, No 2, pp. 59-100.

Outlook, Mining Engineering, Vol. 46, No. 6, pp. 563-567.

copper flotation Plan Mining Engineering, pp. 53-57.

Coal Flotation CircuitsTheir Simulation and Control. Elsevier, New York,

vol. 3, pp. 57 70.

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.

simulation Society for mining, Metallurgy, and exploration, INC. vol. 21. No.

2. pp. 57-63.

models of batch and continuous flotation. In: Fuerstenau, D.W. (Ed.),

Flotation-A.M. Gaudin Memorial Volume, AIME, pp. 675724.

using Mathematical Analysis tool Kohmuench U.S.A.

systems. Transactions of the Institution of Mining and Metallurgy Vol. 70, pp.

191 204.

Proceeding of Seminar on prospects in problem of mineral based industry in

Pakistan, Peshawar published by the Department of Mining Engineering,

NWFP University of Engineering and Technology Peshawar.

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.

mineral based industries in Pakistan.

dosage effects in forth flotation, in the mathematical modeling of Metals

Processing Operation. Extractive and Process Metallurgy Fall Meeting., Palm

Springs, CA, p.12.

sulfide minerals having some natural floatability. International Journal of

Mineral Processing Vol. 58 No. (14), pp. 7784.

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.

Concentrate

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.

default. Canadian IMM Bulletin Vol. 62, No. 682, pp. 109115.

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

No.(1-4), pp. 109-127.

techno metrics, Vol. 4, pp. 531-550.

20.

observed residence time distributions. In: P. Somasundaran, Editor, Advances

in Mineral Processing, SME-AIME, Littleton, Co, pp.714-725.

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.

copper deposits of the world in North Waziristan copper ore.

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.

Oxidized Copper ore Journal of Minerals & Materials Characterization &

Engineering, Vol. 4, No. 2, pp 67-73.

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)

(0.069)

(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)

Subset F (4,27) = 1688.9 [0.0000]**

So collectively they are important.

R2 = 99.60

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)

Observations

31

Mean

0.043489

Std. Devn.

3.6404

Skewness

-0.78933

Excess Kurtosis

1.1999

Minimum

-11.304

Maximum

6.2073

Normality test: Chi^2(2) = 4.8027 [0.0906]

So residuals qualify normality test.

139

Figure (B)

Figure (C)

Residuals look independent.

140

Autocorrelation of residuals

Figure (D)

Therefore residuals are identically independent normal.

YG = 21.58 + 0.01939*X1 + 0.06566*X2 + 0.03842*X3 + 0.01821*X4

(SE) (9.37) (0.00766)

(0.785)

(0.0377)

(0.0188)

(0.025)

(0.0861)

(0.15)

[0] = X2

[1] = X3

[2] = X4

[3] = X5

[4] = X6

Subset F (5,23) = 4.4172 [0.0058]**

141

Test for excluding:

[0] = X2

[1] = X4

[2] = X5

[3] = X6

Subset F (4,23) = 1.9908 [0.1296]

So we can exclude these regressors on totally statistical basis

YG = + 15.88 + 0.01907*X1 + 0.08596*X3 - 0.525*X7 ----------- (d)

(SE)

(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

Normality test: Chi^2(2) = 3.1105 [0.2111]

Figure (F)

143

Chi^2(8) = 29.292 [0.0003]** and F-form F (8,18) = 38.580 [0.0000]**

Residuals are heteroscedastic.

Now we take other possible model

Test for excluding:

[0] = X2

[1] = X3

[2] = X5

[3] = X6

Subset F (4,23) =1.2639 [0.3127]

So we can exclude these variables

YG = + 16.48 + 0.02159*X1 + 0.04478*X4 - 0.598*X7 -------- (e)

(SE)

(2.06)

(0.00758)

(0.0111)

(0.147)

F (3,27) =

12.73 [0.000]**

R2 = 0.585837

T-ratio, F-stat and R2 etc give meaningful values if residuals are identically

independent normal. So we check that whether these variables are identical

independent normal or not.

144

Testing for heteroscedasticity using squares

Chi^2(6) = 5.8926 [0.4353] and F-form F (6,20) = 0.78232 [0.5937]

So residuals are homoscedastics

Normality test for Residuals

Observations

31

Mean

0.00000

Std.Devn

1.2952

Skewness

-0.65212

Excess Kurtosis

1.1558

Minimum

-3.8576

Maximum

2.9363

Normality test: Chi^2(2) = 4.8522 [0.0884]

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.

0.777X3 +0.191X4 + 0.918X6

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

Table A: S = 4.16, R-Sq = 92.4%, R-Sq(adj) = 91.3%, Press = 63.69

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.

0.191X4 + 0.762X6

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.

+ 0.826X6-0.305X7

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.

0.200X4 + 0.783X6 0.306X7

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.

+ 0.079X5 0.795X6

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.

0.158X4 - 0.0982X5 +0.833X6-0.450X7

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

Source

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.

- 0.0437 X5 - 0.626 X7

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.

X5 - 0.130 X6 - 0.602 X7

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.

X3+0.0182 X4-0.0236X5 -0.130 X6 - 0.602 X7

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

M. Mansoor Khan

***

Mir Asad ullah

**

FLOTATION PROCESS FOR THE RECOVERY OF NORTH

WAZIRISTAN COPPER.

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.

we get the simple first-order mathematical model with one predictor variable.

Y = o + 1 X + . (2*)

2.

variable:

Y = o + 1 X + 11 X2 +.. (3*)

3.

Y = o + 1 In X1 + 2 In X2 .. (4*)

4.

In Y = o + 1X1 + 2 X2 + In .(5*)

Fifteen mathematical models were developed by using statistical techniques for

recovery of North Waziristan copper.

The following first order and second order mathematical model were derived with one

predictor variable i.e. Linear, logarithmic, polynomial and exponential.

YR = 0.0536X1 + 29, R2 = 0.8897

Eq. (1)

Eq. (2)

Eq. (3)

YR = 29.541e0.0015X1, R2 = 0.8884

Eq. (4)

YR= 4.0591X2 5.1084, R2= 0.5837

Eq.(5)

Eq. (6)

Eq. (7)

YR = 0.7X3 + 41.4, R2 = 0.821

Eq. (8)

Eq. (9)

Eq(10)

YR = 42.746e0.0127X3, R2 = 0.836

Eq(11)

YR = 0.0720X4 + 58.4, R2 = 0.9348

Eq(12)

Eq(13)

Eq(14)

YR = 59.467e0.0012X4, R2 = 0.9367

Eq(15)

Using X1, in equations 1,2,3 and 4 we obtained equation 3 is the appropriate model is

equation 10 which is quadratic.

Using X2 the models 5,6 and 7 we get the suitable fit model 7 which is again quadratic

using X3 in equations 8,9,10 and 11 the comparatively better quadratic model in using

X4 in equations 12,13,14 and 15we obtained the best fit model 14.

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 a

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

y = 45.116Lnx2 - 68.534

R2 = 0.6005

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

R2 = 0.8239

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

y = 0.7x3 + 41.4

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

R = 0.833

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

y = 0.072x4 + 58.424

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.

techno metrics, 4, 1962, (531-550)

2.

Tennessee copper flotation plan Mining Engineering 18 11 (1966), pp. 53-57.

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.

Mineral Coal Flotation Circuits-Their Simulation and control vol.3,. Elsevier,

New York (1981), pp. 51-64.

6.

FATA development corporation, vol-11, pp 2-13.

7.

Proceeding of Seminar on prospects in problem of mineral based industry in

Pakistan, Peshawar published by the Department of Mining Engineering,

NWFP University of Engineering and Technology Peshawar.

8.

FATA development corporation, vol-3, pp 14-18.

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,

12.

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.

collectors dosage effects in forth flotation, in the mathematical modeling of

Metals Processing Operation, 1978. Extractive and Process Metallurgy Fall

Meeting., Palm Springs, CA, Nov. 29-Dec.2 (1987), p.12.

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.

copper deposits of the world in North Waziristan copper ore.

18.

Xu, 1998, M. Xu, Modified flotation rate constant and selectivity index.

Minerals engineering 11 (3) (1998), pp. 271-278.

19.

copper ore on pilot-scale. Quarterly science vision vol 6(1) July-September

2000. pp. 10-20.

20.

Jamshedpur, Jharkhand 831 007, India. Optimization of operating variables

of fine coal flotation using a combination of modified flotation parameters and

10

1-4,Jan 2003 pages 109-127.

21.

considering interactions of the operating variables. Mineral Processing

Division Department of Mining Engineering, Faculty of Engineering,

Suleyman Demirel University, Isparta, TR 32260, Turkey. Volume 17(1)pp

81-85.

22.

simulation Society for mining, Metallurgy, and exploration, INC. vol 21(2)

pp. 57-63.

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.

based industries in Pakistan.