Building Models in Mining

_________________________________________________________________________
Identification of Critical Key Performance Indicators for

Anglo Coal Underground Mining Operations
by
Madinoge Jeanette Maserumule

s25310454
Submitted in partial fulfilment of the requirements for the degree of

BACHELOR OF INDUSTRIAL AND SYSYTEMS ENGINEERING
in the
FACULTY OF ENGINEERING, BUILT ENVIRONMENT AND INFORMATION

TECHNOLOGY
UNIVERSITY OF PRETORIA
October 2009
_________________________________________________________________________
Executive summary
The most essential task of any production company is to be able to manage production with
reliable methods developed using realistic data, thus creating a platform for continuous
improvement using forecasting methods that can predict the future given past data. The
objective of this paper is to study key performance indicators using statistical tools and
develop a model that can be used to identify a combination of key performance indicators
that are critical in terms of reaching production targets.
Literature review was done to
highlight relevant tools, techniques and methods that can be used to develop an analysis
model that will reach the objectives of the project. This should aid Anglo Coal South Africa in
accelerated decision making capabilities utilising proposed solution strategies to achieve
stated business objectives.
ii
Acknowledgements
I wish to express my gratitude to people who through their contribution made this project
possible.
Thank you Mr. Derek Laing from Anglo Coal South Africa, Mr. Zaais Van Zyl, Mr. Gary Smith
and Mrs. Lauren Theron from Mining Consulting Services.
Additionally I would like to acknowledge the contribution of the following people:
Mr Olufemi Adentunji, for the reviews and provision of theoretical guidance. Andile Ndiweni
and Takalani Mudau for guidance and support.
iii
Contents
1. Introduction ..................................................................................................................... 1
1.1Mining Process............................................................................................................. 2
1.2 Pick Six ....................................................................................................................... 3
1.3 Problem Statement...................................................................................................... 6
1.4 Research design and methodology ............................................................................. 7
2. Literature review ............................................................................................................. 8

2.1 Problem formulation .................................................................................................... 8
2.2 Time series analysis .................................................................................................... 8
2.2.1 Smoothing ............................................................................................................ 9
2.3 Multiple regression analysis ........................................................................................ 9
2.3.1 Least squares estimation .................................................................................... 10
2.3.2 Partial ordinary least squares analysis................................................................ 10
2.3.3 Sequential ordinary least square analysis ........................................................... 11
2.4 Principal component analysis .................................................................................... 12
2.5 Linear programming .................................................................................................. 12
2.6 Excel model............................................................................................................... 13
2.7 Conclusion ................................................................................................................ 14
3. Multiple regression analysis ........................................................................................ 15

3.1 Terminology .............................................................................................................. 16
3.2 General structure ...................................................................................................... 16
3.3 Statistical significance ............................................................................................... 21
iv
4. Computational Results ................................................................................................. 24

4.1 Model outputs............................................................................................................ 24
4.2 Solution quality .......................................................................................................... 26
5. Recommendations ........................................................................................................ 27
5.1 Future work ............................................................................................................... 27
List of figure
1.1: Shift breakdown........2
1.2: Components of a shift time.....2
1.3: Mining process, cutting cycle flow chart...........3
1.4: KPI tree...4
1.5: Production process...5
2.1: Least squares estimation......10
3.1: Studentised residual graph of the final model: Vlaaklagte-Simunye Section.......23
vi
List of table
3.1: Regression terminology..................................................................................................16
3.2: Pick six model symbols.......17
4.1: Significant parameter estimates.................................................................................... 25
vii
Acronyms
ACSA
Anglo Coal South Africa
CM
Continuous miner
KPI
Key performance indicator
SAS
Statistical Analysis Software
Definitions
Continuous miner: remote-controlled machine with a large rotating steel drum equipped
with tungsten carbide teeth that scrapes coal from the seam.
Production time:
production unit of a mine, where coal is mined
Shuttle car:
electric powered vehicle with a capacity of 20 tons used to transport

coal from a mined seam to a section conveyor belt.
Sump:
primary cutting motion of moving cutter head of CM into the coal face.
Tram:
the movement process that a CM undergoes to cut coal
viii
Part I
Introduction
Business organisations make use of key performance indicators (KPIs) to measure the
success of production management and to define a suitable set of performance indicators for
monitoring ongoing performance. KPIs are quantifiable measurements
that reflect the
critical success factors of an organisation thus Anglo Coal South Africa (ACSA) has
identified and defined six key performance indicators referred to as the Pick Six to manage
underground coal mining operations.
ACSA is a division of Anglo American and one of South Africas largest coal producers. It
operates four underground and five open cut mines in Mpumalanga, Witbank Coalfield. It is
well known for producing thermal and metallurgical coal to customers in the inland domestic
markets as well as the export market. ACSAs coal demand has grown tremendously
inducing a need to improve production managing methods that can help with managing coal
operations, budgeting processes and forecasting tools to predict and lessen the risk of not
meeting coal demand.
A standard model has been built to assist in managing coal production using the Pick Six
and fixed mine parameters. With the aid of a third-party data capturing and management
company, Mining Consulting Services, data used to manage coal production is captured
daily using data cards that are attached to each operating Continuous miner during a
scheduled shift. A face shift report document is prepared daily using the mining data from
the previous day illustrating actual key performance indicators relative to the benchmarked
KPIs, required production tonnes, production time etc.
The reports are used as
a check tool to measure actual coal against budgeted coal targets and
illustrate graphically the actual coal tonnes per day against the Pick Six values
1.1 Mining Process

Every mine has a three shift daily mining processes that consists of nine hours per shift. An
illustration of the total shift time is as
a shown in Figure 1.1. The travelling in and out time has
been estimated to be an hour long thus it results in a section time of eight hours which is
referred to as face time. The available time daily is 24 hours and for the three shifts 27 hours
is worked due to the three hours shift change-over
change
phases. Figure 1.2 shows a typical shift
breakdown for most of the underground mines.
Day shift
06:00-15:00
1-1:30 hour
travel time
30-40 minutes
pre and post
meeting
Afternoon
shift
Night shift
14:00-23:00
1-1:30 hour
travel time
30-40 minutes
pre and post
meeting
Figure 1.1: Shift breakdown
Figure 1.2:
1. Components of a shift time
22:00
22:00-07:00
1-1:30
1:30 hour
travel time
30--40 minutes
pre and post
meeting
During a mining operation, a continuous miner(CM) trams coal using the cutter head that is
designed to mitigate methane gas level by absorbing in the fumes and using water sprays to
put out fire sparks. The mining process is illustrated graphically in figure 1.3.
Figure 1.3: Mining process, cutting cycle flow chart
Mines are different in terms of
shift patterns
maintenance allowances
pillar centers
size of coal
haulers
daily production targets etc.
1.2 Pick Six

The Pick Six can be categorised into productive and unproductive times. The unproductive
times are used as a monitoring tool that has to be minimised and if possible eliminated as to
enhance the production time. Productive times are those that have a direct proportional
relationship to coal extracted in a shift.
Productive times
Load time
Away time
Tram time per meter cut
Pick Six
Unproductive times
First sump late

Last sump early
Downtime
Figure 1.4: KPI tree
Operational critical success factors for Anglo Coal for underground operations are to;
achieve planned coal production targets in order to meet customer demand
operate mining equipments efficiently by reducing the machine idle time
Critical factors are only achievable if the KPIs benchmarked values are reliable and their
effect on the coal production tonnes is well understood. The three past projects defined and
reviewed the KPIs to ensure that Anglo Coal has a coal managing standard that can be
deployed in all the underground mines. [3][6][11] The KPIs are explained as follows;
First sump late
It is the time difference between the scheduled start of a shift and the first sump that
the CM makes. It decreases the face time available during a shift by the same
amount of minutes the shift started late.
*Measurement unit: Minutes
Load time
The total time a CM takes to fill up a shuttle car or battery loader with coal. It makes
up the production process together with the away time as depicted in figure 1.5. It is
calculated by the following formula per shift;
*Measurement unit: seconds

Away time
It is the time that the CM awaits a shuttle car or battery loader to load it with coal as
shown in figure 1.5. The time begins when the CM stops loading coal onto a shuttle
car and ends when it starts loading coal onto the next shuttle car.
Tram time per meter cut
Tram time per meter cut is the time taken to tram coal over a meter. It is calculated
by summing the total tram time divided over the number of meters cut during a shift.
Production cycle
Load
Away
Production cycle
Load
Away
Production cycle
Load
Away
Production cycle
Load
Away
CM
Time
Figure 1.5: Production process [6]
Downtime
Total non productive time recorded in minutes during a shift that minimises the overall
production time. The breakdowns are categorised as follows;
Engineering breakdowns
Operational breakdowns
Sectional conveyor breakdowns
Trunk conveyor breakdowns
Other

Last sump early
The time difference between the planned end of a shift and the last sump the CM
makes. The KPI time decreases the face time available during a shift.
The Pick Six is used to calculated figures such as the following that can be used to
calculate the actual coal production per shift.
production time
!
"
!#
Average production rate

'
$%&&
( "
*"
" +
, -.+ '
"
1.3 Problem Statement

Each KPI is evaluated individually without any comparison to the other KPIs as to which
one(s) has the highest impact on coal production target. This gives minute results as to how
the KPIs can be collectively utilised to the maximum in order to make better operational
decisions. ACSA requires a combination of KPIs that are significant for reaching coal
production targets.
1.4 Research design and methodology

Past and present similar projects were studied as to adapt the methodology and obtain the
solution strategy using such methods. Such projects have been performed mostly in the
financial, biological, medical environments and all had a common solution strategy which is
to predict the response variable using KPIs, in this instance the coal production tonnes per
shift/day.
The first phase focused on identifying a solution strategy technique that can be used to
identify a set of KPIs. Using KPI information obtained from the Mining Consulting Services,
the best technique was selected for the identification process of critical KPIs.
The second phase focused on the solution strategy formulation based on the selected
technique to determine the KPI combination. Data was extracted from excel and all the
necessary computations were performed as to achieve the desired end goals. The model
was inputted with all the different mines data.
For the third phase, computational results for each mine were studied and compared to the
other mines. In the case of uncertainties, the model outputs were reviewed as to get an indepth understanding of the uncertainties. This was mostly due to the fact that some of the
mines characteristics differed.
Part II
Literature review
The problem should be defined in such a way that each KPI can be analysed both
individually and collectively with the other KPIs. As mentioned before, each KPI is used to
calculate the production time and thus there exists a relationship between the KPIs and the
actual coal tonnes mined. These relationships will not be considered imperative for
technique selection, however they will only be considered where applicable.
2.1 Problem formulation

Looking at the objectives of the study, the problem can be formulated in such a way that the
KPIs will be used or tested as to yield maximum coal production tonnages. The Pick Six
will be classified as independent variables where X={X1,...,X6} denotes the KPI set. The coal
production tonnages will be classified as the dependent variable.
2.2 Time series analysis

Time series analysis entails analysis of a sequence of measurements made at specified time
intervals. The two main goals of time series analysis [14]:
Identifying the nature of the phenomenon represented by the sequence of

observations,
forecasting i.e. predicting future values of the time series variable
Both of these goals require that the pattern of the observed time series data be
identified and more or less formally described. Interpretation and integration can be made
once a pattern has been identified with the other data i.e., use it in the theory of the
investigated phenomenon, e.g., seasonal death rates. Extrapolation of the identified pattern
can be made as to predict future events.
2.2.1 Smoothing
Smoothing method uses the averaging methods and exponential smoothing methods to
reveal clearly the underlying trend, seasonal and cyclic components from data sets. The
methodology behind smoothing methods is taking the averages of the data sets and
analyzing the data using the following;
the mean squared errors
sum of the squared errors
mean of the squared errors
The above mentioned can be derived from a built forecasting model that when analysed, the
Flaw of averages is sometimes checked for in the model. What the Flaw of averages
states is that plans based on the assumption that average conditions will occur are usually
wrong.
2.3 Multiple regression analysis

The basic purpose of a statistical investigation is to make predictions and optimise by
determining what values of the independent variable/s is the dependent value a maximum or
minimum. [5] Multiple regression analysis is a flexible method of data analysis that is
appropriate whenever a quantitative (response) variable is to be examined in relationship to
any predictor variable and regression analysis is most widely used for such functions. The
basic regression formula is as follows;
)
/0 ( /1 21 ( 3 ( / 4 24 ( 5
6789:
Where
y
= dependent variable
= independent variable
E(y) = deterministic component

= random error component
0
= y-intercept of the line, i.e. point at which the line cuts through the y-axis
= slope of the line, i.e. amount of increase in the mean of y for every 1-unit
increase in x.
2.3.1 Least squares estimation

The regression coefficients,
,. . . ,
are estimated using the method of ordinary least
squares of which the coefficients minimises the error sum of squares
This means the distance between the fitted line which consists of the predicted
values and
the actual Y values denoted by asterisks in figure 2.1 is being minimised. This distance is
referred to as the residual or prediction error.
error
For a test to be accurate, certain assumptions have to be satisfied [1]
data is sampled randomly and independently from the population
and the deviations of the predicted
values from the actual Y values are normally
distributed with equal variance for all predicted values of
Figure 2.1:
2. Least squares estimation
2.3.2 Partial ordinary least squares

square analysis
Randall D. Tobias [12] defines partial least square as a method for constructing predictive
models when the factorss are many and highly collinear. Essentially what partial least square
does is show how much a predictor improves over all the other predictors. If it is added over
all the predictors, Pick six,, it will be defined as the increase in the regression sum of
squares. The regression sum of squares is a variation
variati analysis of Y that shows
show the variation
of Y predicted by the predictor variable/s.
10
The total variation is measured by the following formula;

;
'-+
-<+ " -
=6>? ( >@:A
678$:
And the variation that can be predicted is measured as follows;
C #" -- ! -+
-<+ " -
ED ( >@:A
=6>
?
678F:
The residual sum of squares is measured as follows;
C - + ' "" "-+
-<+ " -
=6>?
?
ED :A
>
678G:
Emphasis is put on the fact that the predictive models are used for predicting the responses
and also on trying to understand the underlying relationship between the explanatory
variables.
2.3.3 Sequential ordinary least square analysis

Sequential least square analysis shows how parameter predictions are improved as more
predictor variables are added in a defined order. The same concept of the increase in sum of
squares as explained under partial least square analysis is the same for sequential least
square analysis and equations 2.3, 2.4 and 2.5 were used. It is beneficial to use sequential
least square analysis if there is a natural order of the predictor variables.
From figure 1.3, the only order that can be distinguished is that the load time has to occur
before the away time. For the sumps KPIs, the first sump late has to occur before the last
sump early and all the other KPIs occur in between the two. This will not have any effect on
the integrity of the model if not followed as the KPI measurements are deemed independent.
Downtime can occur without any of the KPIs could occur such as the loading of coal onto a
shuttle car.
11
2.4 Principal component analysis

Principal component analysis (PCA) is a widely used multivariate method that is seen and
described as a complementary statistical method used to run rough preliminary
investigations, to sort out ideas, to put a new light on problem, or to point out aspects which
would not come out in a classical approach.[9] The main objectives of PCA that are in line
with the problem at hand are; dimensionality reduction, feature selection i.e. the selection of
the most useful variables, identification of underlying variables. [9]
In terms of analysing large data subsets, PCA is a great tool due to its condensing of
information. If all mines KPIs are to be analysed for trends, it will aid in enhancing the
chances of producing useful results.
2.5 Linear programming

A mathematical model can serve a number of purposes. The first function is that a
mathematical model can serve in regards to production as a tool for prediction. The types of
models that can be used are the descriptive mathematical models and simulation models. A
descriptive model can be used to predict the consequences of increasing a KPI value, that
is, the effect on coal production tones. With a simulation model, it can be used to predict the
outcome of prescribed set of strategies so as to provide needed input-output information to
solve a decision problem.
Secondly, it can be used as a tool for control or decision-making purposes. Decision
problems regarding controllable aspects of a process influence the operation of the process
by changing or controlling the value of some decision variable. The type of decision models
are modeled as to find the values of the decision variables that
satisfy all constraints simultaneously
achieve the stated objective
The methodology followed in generating a mathematical model fits perfectly with the
methodology planned for developing the solution strategy for the KPI identification process.
However, problems arise when generating the algorithm as known KPI production tonnes
relationships will be used thus the productive KPIs that have a direct relationship to the
production tones will be favored.
12
For an example, the first sump late, last sump early and downtime KPIs are merely
subtracted from the face time unlike the other KPIs that are included in the production time
calculations used to calculate the actual coal produced in a shift/day.
2.6 Excel model

The modern trend in programming applications is graphical interfaces. This means that the
user interacts with a program through a graphical interface e.g. values are entered in text
boxes, options are selected from graphical menus and answers are presented graphically.
Excel macros are an option in terms of their ability to create options that can be selected
from a button etc. The macros are stored as VBA code that uses an ActiveX interface to
cause excels applications to perform actions such as change formulas, create charts etc.
The type of action that the application supports are defined by what is called an object
model.
The model was built by excel designers to provide an interface so that the programming
language can cause the application to do what a user normally would do interactively with a
mouse and keyboard. It includes the following:
list of application objects that can be managed
properties of these objects that can be examined or altered; in the coal production
model the value of the KPIs can be altered.
methods that can be performed on the object
An excel model can be embedded with calculations and measured values to output the
desired end goals. Using this tool all the requirements can be met using defined formulations
with the use of macros to show each KPI effect on the total coal tonnes. Each KPI factor can
be calculated which can easily be viewed on the excel spreadsheet. Further formulation can
be made using the KPI factors to select a combination of KPI that are critical in order to
reach budgeted coal production tones.
13
2.7 Conclusion
From the different techniques studied, it was deemed multiple regression analysis is the
most appropriate as it can yield the accurate required solution; however multiple regression
has been extensively used in medical fields than engineering fields when dealing with
variable selection studies. The other methods give rise to more problems as the solution
formulation gets complicated due to extensive computations that defeats the purpose of the
study.
Due to the availability of softwares that can perform statistical computations,
Statistical Analysis Software (SAS) was the prominent in terms of usage and simplicity
understanding. The partial ordinary least square regression analysis methods will be used as
it is considered to yield accurate solution outcomes.
14
Part III
Multiple regression analysis
Multiple regression analysis is widely used in the psychology, dietary, population and
financial fields for prediction purposes. The relationship between the predictor variables and
response variable will be analysed as to determine which predictor variable/s has the most
impact on the response variable.
The following are the model assumptions as explained by Mendenhall for a regression
model [7]
Assumption 1: The mean of the probability distribution of
is 0 i.e. the average of the errors
over an infinitely long series of experiments is 0 for each setting of the

independent variable x.
Assumption 2: The variance of the probability distribution of
is constant for all settings of
the independent variable x.

Assumption 3: The probability of
is normal
Assumption 4: The errors associated with any two different observations are independent i.e.
the error associated with one y has no effect on the errors associated with
other y values.
A test for residual normality is crucial before any inferences can be made from the model.
The p-value [table 3.1] can only be used if the residual normality test is satisfied, if not
transformations are made and explained before using the p-value for parameter selection.
Test for normality
H0 : Residual come from a normal distribution
H1 : Residual deviate from a normal distribution
Shapiro-Wilk test
n
50
Reject H0 if W Is small
p-value = p(w)< 0,05 : deviate from normal
>0,05: H0 is not rejected
15
A Shapiro-Wilk test is used only when the number of observation is less or equal to 50. The
rule of thumb is used for the confidence level i.e.
= 0,05 unless stated otherwise due to
failure to comply with normality tests. Data transformations are performed on the y values as
to make them nearly satisfy the assumptions, and for the latter reason, to achieve a model
that provides a better approximation to E(y).
3.1 Terminology
The basic terminology presented in part II; section 2.2 will be used with the additional
terminology explained in table 3.1 throughout the rest of the document.
Table 3.1: Regression terminology
Least square:
Squared distance between the data points and the model.

The aim is to minimise the distance as to get a best fit
linear graph.
R-square:
It measures how much of y, response variable is explained

by x, predictor variable
P value:
In statistical hypothesis testing, the p-value is the

probability of obtaining a test statistic at least as extreme
as the one that was actually observed, assuming that the
null hypothesis is true. The fact that p-values are based on
this assumption is crucial to their correct interpretation.
3.2 General structure

The steps applicable for developing a regression model are as listed below. Collection of
data and variable classification was done in Step 1 and 2. From the code written for the
model with the required output formulation for step 3, 4 and 5, the computations were done
simultaneously. Step 6 calculations are for validation. The model was stopped when the ttest was not violated i.e. the p-value was less than 0.005.
16
Regression Analysis steps
Step 1: Collection of data for each experimental unit in the sample

Step 2: Variable classification
Step 3: Estimation of the unknown parameters,
1,
. . .,
6.
Step 4: Estimation of the probability distribution of the random error component
and its
variance
Step 5: Evaluation of the utility of the model
Step 6: Validation of assumptions on
and make model modifications if necessary
Step 7: If the model is deemed adequate, estimate the mean value of Y , identify the critical
KPI and make other inferences.
Regression algorithm
>
J9K L K%M
/0 ( = /? 2? ( 5?
?I1
6$89:
In this formulation, the dependent variable is denoted by Y, which is the production tonnes
and the independent variables by 2? which are the Pick six as described in table 3.1.
Table 3.2 Pick six model symbols
Term
Description
X1
first sump late
X2
load time
X3
away time
X4
tram time per meter cut
X5
Downtime
X6
last sump early
17
SAS algorithm
The model explained here is for one mine section, a couple of other models were run as to
accomplish a general KPI combination that can be used for all Anglo Coal underground
mining processes.(Addendum A and B)
Input data: Simunye Section
options nodate pageno=1;
title' Vlaklaagte: Simunye Section';
data vlakkies;
label y = 'Actual Production tonnes'
x1 = 'Load time sec'
x2 = 'Away time sec'
x3 = 'FS late min'
x4 = 'Tram time min'
x5 = 'Downtime min'
x6 = 'LS early min';
input y x1 x2 x3 x4 x5 x6;
datalines;
59457 63
72
22
304
13
54847 67
70
26
187
13
51350 67
74
25
155
16
60042 67
76
24
142
20
60898 66
71
20
143
51956 70
72
25
148
46085 67
70
25
203
45962 63
84
28
199
58854 67
71
27
167
52279 88
68
22
187
12
56704 66
66
24
201
19
68864 63
70
20
215
11
18
Data points for each mine section were extracted from an excel spreadsheet and defined as
variables as in line 4(input) of the model algorithm. For this algorithm Simunyes data was
tested using 12 data points.
Iteration & Output selection

proc reg data=vlakkies;
model y = x1 x2 x3 x4 x5 x6;
output out = a1 p = pred1 r = res1 student = studres1;
model y = x1 x3 x4 x5 x6;
model y = x1 x3 x4 x5;
model y = x1 x3 x4;
model y = x1 x3 ;
run;
The full model was ran first and further parameter selection steps were made as to obtain
parameter estimates that were significant. For this model, x2 was eliminated first, followed by
x6, x5 then x4 using the outputted p-value. If the p-value for any predictor variable was
significantly large, the variable was eliminated from the model. This was done till the
remaining p-values for each predictor variable fell within the 95% confidence interval.
Check if errors are normally distributed
goptions reset=all;
symbol i=none value=dot cv=green height=0.7;
proc gplot data=a1;
plot studres1*pred1 / vref=0;
run;
In order to use the t-test for the stepwise elimination process, errors have to be normally
distributed with their mean equal to zero as mentioned earlier.
19
Check for correlation between the x variables and y

proc corr data=vlakkies;
var x1 x2 x3 x4 x5 x6;
with y;
run;
This checks for how y and xs are related i.e. for example if there is a negative relationship
between y and one of the x values, take for instance the away time, the expected actual
production tonnes will decrease as the away time increases.
Check for correlation between the x variables
run;
This was used to check for relationships between the xs themselves. If highly correlated
predictor variables are present in the model, the results may be confusing. In some cases
the t test on the individual parameter estimates may be non-significant even though the Ftest for the overall model adequacy is significant and the parameter estimates may have
signs opposite from what is expected. Variation inflation factors aid in determining whether a
serious multicollinearity problem exists.
Data Normality Test
proc univariate data=vlakkies normal;
var y x1 x2 x3 x4 x5 x6;
histogram y x1 x2 x3 x4 x5 x6 /normal;
run;
The data normality test was to check if some KPIs data were normally distributed. The
sample size was too small, i.e. less than 50 datasets thus it was infrequent for normality to
be observed in the datasets.
20
Fitted model for significant KPIs

model y = x1 x3 ;
output out=a p=pred r=res student=studres;
run;
proc print data=a;
run;
goptions reset=all;
symbol i=none value=dot cv=red height=0.7;
proc gplot data=a;
plot studres*pred / vref=0;
title3 'Studentized residual of the final model';
run;
This ensures that the errors are normally distributed as to satisfy the error normality
assumptions mentioned before. From the graph it can be seen that some of the errors do
come from a normal distribution or are normally distributed. (Refer to figure 3.1)
3.3 Statistical significance

A number of the models output will answer and show how reliable the solution strategy is.
The following are some of the important outputs that were studied;
R2: It provides a measure of how well Y can be predicted from the set of x scores i.e. the
amount of the total variation in Y that can be explained by x.
Adjusted R2: Is the same as the R
however it takes into account the number of
parameters being modeled, x. A high R2 portrays the reliability of a model in terms of

how much of the dependent variable is explained by the independent variables
Hypothesis testing: This aid in testing the validity of the estimated value of a parameter
and the utility of the model. By the rule of thumb, the significance level was defined to
be 5% but for the Vlaaklagte, Simunye Section, 20% was used only for testing the
21
parameters. If parameter estimates fell within the significance level, the null
hypothesis was rejected, stating that the parameters were significant from zero and if
not, stepwise regression was used to eliminate the parameters that were not
significant. The model was run until we had only significant parameters in the model.
This can be viewed in the iteration and output step of the SAS algorithm. X6 was
eliminated first followed by x5 and x4 consecutively.
F-test: The test is done on the full model using the F test for all the predictor variables
except for the intercept values as they have no significance on the solution strategy.
The null hypothesis can be rejected if the p-value is less than the
value.
Test for overall model significance

H0 : Model not significant
H1 : Model significant
=0.05
Reject H0 if p-value is less than 0.05
T-test: Same test as the F-test but only done for parameter estimates.
Test for parameter significance
H0 :
H1 :
=0(Not significantly different from zero)

0 (significantly different from zero)
=0.05*
Reject H0 if p-value is less than 0.05

* Only 20% significance level was used for one (Vlaaklagte: Simunye Section) model as to
accommodate the actuality of not having sufficient data points.
The model has to be run in such a way that all the regression analysis assumptions are not
violated using the normality tests for errors. In order to use the p and f-test, normality for
errors must not be violated as the tests can only be used if the assumptions are met. If
normality is violated, then the model did not pass to merit for further considerations.
22
Studentised residual graph of the final model: As explained before, residuals are values
between the observed responses from the predicted responses. The studentised
residual graph plots the scatter points within the 95% confidence level. In order for
the model not to be rejected, the scatter points should not show any trend.
Studentized residual of the final model
Studentized Residual
2
-1
-2
48000
50000
52000
54000
56000
58000
60000
62000
64000
66000
Predicted Value of y
Figure 3.1 Studentised residual graph of the final model: Vlaaklagte, Simunye Section
23
Part IV
Computational Results
4.1 Model outputs
Out of the five mine sections models, 5 x 6 parameter estimate outputs were studied as to
identify the critical KPIs. As mentioned before the parameter estimate,
, is the amount of
increase or decrease in the mean of y for every 1-unit increase in x. Monthly data was used
for all the models except for Greenside as it had weekly data. The five model outputs give
explanation as per data set inputted in each mine sections model and the output is
demonstrated using the general regression equation (4.1).
)
/N ( /1 O1 ( /A OA ( /P OP ( /Q OQ ( /R OR ( /H OH ( 5
6F89:
Vlaaklagte
Simunye Section
)
9FFGG%
G&S8F7G O1 ( %%%8&&GOA
7F9G89FFOP
9&TFU OQ
TG8%UU OR ( FSG8FF OH
Mangwapa Section
)
7SSGGT
99T$O1
GUG8%&$OA
ST8TS OP
77F89%9OQ ( T9&F89%9 OR
F$G8G$F OH
The parameter estimate for Simunye Sections x1 which is the load time can be interpreted
as for every one unit increase in the load time, the coal production tonnes will decrease by
508.425 units and for x6 which is the last sump early can be interpreted as for every one unit
increase in the last sump early, the coal production tonnes will increase by 485.44 units.
The parameter estimates for Vlaaklagte KPIs cannot be used to derive a general set of
combinations as the estimates differ tremendously. The R2 for the Simunye Section model
was 73% and for Mangwapa Section model, 78% which is significantly high, however the
significant parameter estimates for each mine differ.
24
Greenside
George Section
)
7$%T8%$S
98G7%O1
S8$$%OA ( 9F8S$TOP
9%$8T7TOQ
78F%7OR
%8%&&OH
Vumangara Section
)
GUFS8FU7 ( G8%9&O1
$%8%FFOA
9G877$OP
77S8&GF OQ
&8FGF OR
$8S$%OH
The parameter estimates for Greenside differ slightly as compared to the Vlaaklagte
parameter estimates, however the R2 for both models were significantly low, 45% and 48%
for George and Vumagara respectively. If the model is not good enough, no KPI combination
can be derived from such a model.
New Denmard Central
Simunye Section
)
TFG79
9&$8$ O1
U8S$TOA
$9S8GF%OP
GSS%8SF OQ ( F98G7S OR
$%G89&T OH
The R2 for the New Denmark section was 55% and the adjusted R2 has a negative value
which shows that the model contains terms that do not help in predicting the response.
All the parameter estimates where tested for significance and the significant parameter
estimates are as shown in the table 4.1.
Table 4.1: Significant parameter estimates
Section Name
1
Greenside: Vumagara
NDC_Central :Simunye
Vlaaklagte: Simunye
Greenside: George
Vlaaklgate: Mangwapa
1
1
1
1
1
25
From the table, no parameter estimate(s) dominate the others. This can be due to the reality
that all the mines differ due to the following;
shift patterns
maintenance allowances
pillar centers
size of coal
haulers
daily mines section production targets etc.
From the five models, only three KPIs dominated and this cannot be deemed legitimate as
the sample size of the models is minute. It should not be ignored that the five models did not
yield favorable outputs thus impeding further investigation due to the sample size being too
small &/other unknown factors.
4.2 Solution quality

Test for Model and parameter significance
As mentioned before, the t-test and F-test were used for checking model and parameter
significance. The p- value had to be below 20% were measurable changes were made, and
5% for all the other models using the rule of thumb.
26
Part IV
Recommendations
The model outputs did not show any trend that could be used to select a general KPI
combination. The data samples when tested using the normality test did not yield favorable
results thus suggesting the small sample size might be a problem. The parameter estimates
told a different story for each mine section even though some of the models were significant
and had more or less the same data points as the others.
However the model outputs gave ample information on the effect they have on the actual
production tonnes. As mentioned under future work, further regression analysis
methodologies can be carried out as to improve the solution integrity with the addition of a
bigger sample size.
5.1 Future work

One area that was not explored was the combination of some KPIs by having a product of
two KPIs as to check if the KPIs combined will have an impact on the actual production
tonnes. What this would do is to increase the order of the regression model to two or three
degrees but the model would still be categorised as a linear model. The same goes for
having reciprocals or dividing some of the KPIs.
It would be valuable to gather more data as to develop precise and bettered underground
mining KPI regression models that can be used for KPI combination selection and coal
production tonnes forecasting.
Data transformations can also aid in giving a better model fit and smaller prediction errors,
but also give a more reasonable bias structure as to how the KPI combination output would
fit a real life mining scenario. The type of transformation will depend on the theoretical
relationships between the response variable and the predictor variables.
One important lesson learnt is that one should not trust global measures of prediction
quality too much since they may be misleading. The values for all the models developed are
very reasonable, but the general interpretation of the individual output seemed bias.
27
Bibliography
[1] Berger, D.E. (2003). Introduction to Multiple Regression. Claremont Graduate
University. Pages1-6.
[2] Bise, C.J., and Albert, E.K. Comparison of model and simulation techniques for
production analysis in underground coal mines-document summary. Available online
from http://www.onemine.org/(Retrived 23 May 2009)
[3] Denkhaus, M. Eardley, M. Mufamadi, C. and Raphulu, R.(2007).Review of the
Current Approach to Measurement and Reporting of Continuous Miner Relocation
Time as a Key Performance Indicator. Presentation.
[4] Giltow, H.S., Oppenheim, A.J, and Levine, D.M.(2005). Quality management,3rd
edition. Singapore: McGraw Hill.
[5] Johnson, R.A,(2005).Probability and statistics for engineers.7TH edition. Pages 101146 and 226-281. United States of America: Prentice Hall.
[6] Khumalo, S. (2008). Determining the Suitability of the Current Functional Production
KPIs at Greenside Colliery in Managing Production Section Downtime and To
Identify Critical Areas of Lost Time.
[7] Mendenhall, W, Sincich, T. A Second course in Statistics: Regression Analysis. 6th
edition, pages 90-741. New Jersey: Prentice-Hall.
[8] Montgomery, D.C, Peck, E.A, Vinig, G.G. Introduction to linear regression analysis.
Pages 169-217.
[9] Nass,
P.
(1999).
Multivariate
analysis
methods.
Available
online
from
www.eso.org/projects/esomidas/doc/user/.../node210.html (Retrieved on 19 August

2009)
28
[10] Nielsen, B. R, Stapelfeldt, H, Skibsted, L.F.(1997) Early Prediction of the Shelf-life

of Medium-heat Whole Milk Powders Using Stepwise Multiple Regression and
Principal Component Analysis. Royal Veterinary and University.
[11] Prinsloo, B, Steyn, P, and Ndiweni, A.(2006). Defining the standard Anglo Coal front
line performance measurements for Underground CM sections.
[12] Randall, D.T. An introduction to partial least square regression. SAS Institute Inc,
Cary, NC. Pages 1-4.
[13] Savage, L.S. (2003).Decision making with Insight, 1st edition, pages 17-50.Canada:
Thomson Brooks/Cole.
[14] Statsoft
Website.(2009)
Time
Series
Analysis.
Available
online
from
www.statsoft.com/TEXTBOOK/sttimser.html.(Retrieved 10 October 2009)

[15] Tomerat, J.F, Harakal, C. (1996). Multiple Linear Regression Analysis of Blood
Pressure, Hypertrophy, Calcium and Cadmium in Hypertensive and Nonhypertensive States. Temple University School of Medicine.
[16] Winston, L.W, Ventekataraman, M. Introduction to Mathematical Programming, 4th
edition, volume one of Operations Research, pages 49-111. California: Duxbury.
29
Addendum
Addendum A: Sections SAS models
A 1: Vlaaklagte: Mangwapa section
A 2: Vlaaklagte: Mangwapa Section
A 3: Greenside: George Section
A 4: Greenside: Vumagara Section
A 5: New Denmark Colliery: Simunye Section
Addendum B: SAS sections model outputs

B 1: Vlaaklagte: Simunye Section
B 2: Vlaaklagte: Mangwapa Section
B 3: Greenside: George Section
B 4: Greenside: Vumagara Section
B 5: New Denmark Colliery: Simunye Section
30
Addendum A: Sections SAS models
A 1: Vlaaklagte: Simunye Section

title' Vlaklaagte: Simunye Section';
data vlakkies;
x3 = 'FS late min'
x5 = 'Downtime min'
datalines;
59457 63
72
22
3
54847 67
70
26
3
51350 67
74
25
4
60042 67
76
24
4
60898 66
71
20
4
51956 70
72
25
3
46085 67
70
25
3
45962 63
84
28
4
58854 67
71
27
3
52279 88
68
22
3
56704 66
66
24
3
68864 63
70
20
3
;
304
187
155
142
143
148
203
199
167
187
201
215
13
13
16
20
8
7
5
8
6
12
19
11

model
model
model
model
run;
y
y
y
y
=
=
=
=
x1
x1
x1
x1
x3
x3
x3
x3
x4 x5 x6;
x4 x5;
x4;
;
goptions reset=all;
proc gplot data=a1;
run;
with y;
run;
1
run;
proc univariate data=vlakkies normal;
run;
**********Checking errors for normality for the fitted model*************;
model y = x1 x3 ;
run;
proc print data=a;
run;
goptions reset=all;
proc gplot data=a;
run;
A 2: Vlaaklagte- Mangwapa section

title' Vlaklaagte: Mangwapa Section';
data vlakkies1;
x3 = 'FS late min'
x5 = 'Downtime min'
datalines;
90257 67
37
517
1
83561 70
44
474
8
63425 69
47
467
18
77906 64
49
477
22
95465 58
53
485
11
10045956
52
500
9
97995 54
44
552
10
64849 55
48
524
14
93748 57
47
523
13
89527 57
48
543
12
47986 58
45
562
4
69023 58
43
570
12
95342 59
45
553
12
;
proc reg data=vlakkies1;
output out = a1 p = pred1 r =
model
model
model
model
model
y
y
y
y
y
=
=
=
=
=
2
2
3
3
2
2
2
2
2
2
2
2
2
155
147
224
174
139
165
157
208
134
141
229
201
175
3
12
16
17
21
9
14
16
16
10
17
25
16
res1 student = studres1;
x1 x2 x3 x5 x6;
x1 x2 x3 x6;
x1 x3 x6;
x1 x6;
x6;
run;
goptions reset=all;
proc gplot data=a1;
title3 'Studentized residual of the full model';
run;
proc corr data=vlakkies1
with y;
run;
proc corr data=vlakkies1;
run;
3
proc univariate data=vlakkies1 normal;

run;
proc reg data=vlakkies1;
model y = x6 ;
run;
proc print data=a;
run;
goptions reset=all;
proc gplot data=a;
run;
A 3: Greenside: George Section

title'Greenside: George Section';
data greenside1;
x1 = 'FS late min'
x5 = 'Downtime min'
input x1 x2
datalines;
15
50
17
66
4
60
13
69
10
64
25
56
15
60
10
60
18
59
27
60
13
61
19
62
23
61
13
68
9
65
5
64
1
67
15
69
17
42
9
67
22
66
14
64
30
61
2
64
5
74
2
71
10
71
3
70
15
72
15
71
12
65
;
x3 x4 x5 x6 y;
63
57
65
64
62
62
65
61
64
56
58
62
68
65
58
64
61
64
59
58
65
64
57
64
64
60
62
62
56
61
68
3
2
2
2
3
3
2
2
3
3
2
2
2
2
3
3
3
3
3
3
3
3
3
3
3
3
3
2
3
3
3
175
163
125
138
208
69
95
106
50
59
52
199
105
98
154
142
126
159
226
50
118
138
20
81
203
330
195
138
115
178
155
30
16
45
32
26
21
10
26
13
11
11
11
10
22
19
24
27
21
24
23
16
20
19
17
17
22
17
19
20
21
22
2104
1857
1903
1886
1382
1850
2563
2082
2299
2206
1894
1782
2003
2249
1815
1355
1722
1704
1500
2276
2138
1944
1376
1886
1368
1314
1703
1643
1761
1565
1593
proc reg data=greenside1;

model y = x1 x2 x3 x4 x5 x6 ;
output out = a r = res p = pred student = studres;
model
model
model
model
model
run;
y
y
y
y
y
=
=
=
=
=
x2
x2
x2
x3
x5
x3
x3
x3
x5
;
x4 x5 x6 ;
x5 x6 ;
x5 ;
;
goptions reset=all;
symbol i=none value=dot cv=blue height=0.7;
proc gplot data = a;
plot studres*pred / vref = 0;
run;
proc corr data=greenside1;
with y;
run;
proc corr data=greenside1;
run;
proc univariate data=greenside1 normal;
run;
proc reg data = greenside1;
model y = x5;
output out = a1 r = res1 p = pred1 student = studres1;
title1 ' Fitted regression model';
run;
goptions reset=all;
proc gplot data = a1;
run;
A 4: Greenside: Vumagara Section

title'Greenside: Vumagara Section';
data greenside;
x1 = 'FS late min'
x5 = 'Downtime min'
input x1 x2
datalines;
16
61
5
61
9
63
9
66
9
70
15
71
10
70
16
70
13
68
9
69
5
71
9
70
23
69
11
64
6
68
4
68
2
66
4
62
37
60
32
65
3
57
5
55
10
64
15
56
26
69
19
67
16
63
27
62
20
59
13
59
16
59
11
61
10
63
0
64
8
64
3
67
2
68
7
65
6
61
12
60
4
62
16
64
2
66
x3 x4 x5 x6 y;
88
99
98
70
68
68
70
59
67
67
67
64
68
65
65
60
62
57
58
60
62
65
57
65
61
58
58
63
63
68
61
62
59
67
68
63
55
60
63
61
63
67
63
2.5
2.5
2.5
2.5
2.5
2.5
2.5
2.5
2.5
2.5
2.5
2.5
2.5
2.5
2.5
2.5
2.5
2.5
2.5
2
2
2
2
3
3
3
3
2
2
2
2
2
3
3
3
3
3
3
3
2
2
2
2
221
202
169
183
172
228
158
174
273
141
176
42
159
163
165
105
169
101
166
143
144
126
188
149
222
76
188
103
183
135
129
145
240
235
130
215
220
260
40
223
141
99
134
15
16
18
14
21
19
22
7
23
23
19
16
50
28
18
4
10
25
28
14
45
71
19
47
14
22
23
30
22
26
31
26
25
27
20
24
22
30
25
18
18
19
26
1539
1652
1605
1377
1895
1544
1887
1997
1674
1650
1686
1405
1494
1741
1635
2170
1977
2079
1908
1955
2295
2087
1902
2116
1680
2434
2117
3212
2421
2078
2455
2112
2137
1600
1915
1770
1688
1827
1566
2271
2229
1931
1811
7
6
6
26
3
9
15
14
;
69
66
67
64
61
64
64
69
64
60
57
58
60
60
2
2
3
2
3
2
3
138
183
153
173
134
214
131
18
23
11
19
19
20
20
1844
2171
1898
2169
2012
2290
2068
proc reg data=greenside;

model y = x1 x2 x3 x4 x5 x6 ;
model y = x1 x2 x3 x4 x6 ;
model y = x1 x2 x3 x4 ;
model y = x2 x3 x4;
run;
goptions reset=all;
run;
proc corr data=greenside;
with y;
run;
proc corr data=greenside;
run;
proc univariate data=greenside normal;
run;
proc reg data =
model y = x2 x3
output out = a1
title1 ' Fitted
run;
greenside;
x4;
r = res1 p = pred1 student = studres1;
regression model';
goptions reset=all;
run;
A 5: New Denmark Colliery - Simunye Section

title' NDC_Central: Simunye Section';
data ndc;
x3 = 'FS late min'
x5 = 'Downtime min'
datalines;
25628 96
90
26
5
25873 93
89
24
6
24785 88
95
13
7
24747 111
75
8
7
26208 114
76
21
5
28506 113
81
26
4
27930 115
79
21
6
22736 138
93
15
7
20597 116
85
14
6
36238 113
84
15
5
35034 113
84
18
5
;
267
222
236
215
232
214
260
243
159
200
173
26
21
8
8
24
21
9
4
5
6
13
proc reg data=ndc;

model
model
model
model
run;
y
y
y
y
=
=
=
=
x1
x1
x1
x4
x3
x4
x4
x6
x4 x5 x6;
x5 x6 ;
x6;
;
goptions reset=all;
run;
proc corr data=ndc;
with y;
run;
proc corr data=ndc;
run;
proc univariate data=ndc normal;

run;
proc reg data=ndc;
model y = x4 x6;
run;
proc print data = a1;
run;
goptions reset=all;
run;
10
Addendum B: SAS sections model outputs
*Only one model output is included under this section, the rest of the output
models are saved in the attached disc
11
B 1: Vlaaklagte-Simunye Section
!
#
#
$
$
"
"
"
" '"
" "
$
"
/3./20+-+&
//+31
.2 +,03
$$
"
(
8
!:
(
:
"
"
,-+.-+, &.1,-31/
-0+013-33
8
9
9&
9,
99/
9+
&
&
+
/
%
%
"
/01&-,1+
&/.++1 0
5 )
! 6 5 )
* (
&2&,
320&00
32-33.
"
"
--//+
5/312-&-1+
+++233-+.
5&- /2 -- &
5 30-.
50/2+.1.-1/2-,..1
12
32 .10
* 7 7
,..,.
&+&2,&3.&
00.2 31+0
1. 21/&3&
0+-&2& -00
/-2+,3&1
,+1200&--
,2+&
5 2.321/
5&20
5 25 2,.
2,&
323 /&
32 3,
32-, 0
323-&32& 1+
32&&-/
32&-/&
/
!
#
#
$
$
"
"
%
%
&
.
4
"
" '"
" "
"
$
$
"
&+//,+,..
& &--33
-0+013-33
-1--20-/ +
//+31
120 &&.
$$
&
&
,&0+1 ..
&,-0 //+
5 )
! 6 5 )
"
* (
/2++
323&/0
32//+.
32-/1/
"
"
7 7
8
9
9,
&&++3
5,,12./.,,
5 1,/20.3,
"
(
& 1/
&&320003.
/0+2.,33&
/2+
5 2/5,2 1
32333,
32 /.
323
4
;
"
"
9&
9,
"
#
9
9&
9,
99/
9+
&
&
&
&
&
&
&
9-
9/
9+
"
%
//+31
+021,,,,
0&233333
&-233333
,2,,,,,
102/1,,,
2/3333
+/1+2+1//1
-2/0&+/
&2//132-.&,0
--2+. 0.
-2.+,/3
'#8 ! 8!
<!
++0&.1
1 -233333
1+-233333
&11233333
-3233333
&&/
,1233333
-/.+&
+,233333
++233333
&3233333
,233333
-&233333
/233333
"=
"
13
9
+11+11233333
1-233333
&1233333
-233333
,3-233333
&3233333
:
'
4
$$
%
""
&
//+312 ++0
+/1,2/. 0
32&133,,&.
,20/1- 3
21,.&/&,
;
"
>
4
14
" "
"
%
"
&
++0&.1
-,,-,+0&20
32 / &1.-1
-0+013-33
.332/ .30
? "
"
"
//+312 0
//00/2/3
2
"
+/1-,,-,+0,
&&.3&
13.0
" " $
"
"
A"
5;
%5
5%
" 5
55555
&.2&/.-+
+
,.
" " $
55
"
"
>
4
!
3@3
;
%
" "
;5 )
!5 )
* 7 7
*@ 7 7
*@ 7 7
555
B2333
32333/
32333/
55555
32./+0+32 &0/ ,
323,&3/0
32&--000
'#8 ! 8!
"
" $
555555
555555
B ;
*
* ;5 )
* !5 )
320,+1
*32 /33
*32&/33
*32&/33
"
"
//+312 0
+/1,2/.
""5 $5(
"
>
4
!
" " $
5555
%5
5%
" 5
%
" "
;5 )
!5 )
"
"
55555
555555
32 &0/ , 0
323,&3/+..
32&--000-.
'#8 ! 8!
9
<
555555
*
* ;5 )
* !5 )
"
*32 /3
*32&/3
*32&/3
"
#
:
'
4
$$
%
""
&
+021,,,,,,
+2+1//0.&,
&21+1+-1//031
.21//1.30+
;
"
>
4
15
" "
"
%
"
&
1 --2+.+.+.0
.2 -/&1 /+
-. 2+++++0
2.&..+3-.
? "
"
+021,,,,
+0233333
+0233333
"
16
"
+2+1//1
--2+.+.0
&/233333
&2/3333
" " $
"
3@3
"
A"
,/2 -0/&
+
,.
" " $
"
>
4
!
;
%
" "
;5 )
!5 )
* 7 7
*@ 7 7
*@ 7 7
"
555
B2333
32333/
32333/
55555
32+3 .-&
32,1&.,&
32,/&,/+
2. 1 &
555555
B ;
*
* ;5 )
* !5 )
'#8 ! 8!
"
" $
555555
55
5;
%5
5%
" 5
55555
32333
B323 33
B3233/3
B3233/3
"
"
+021,,,,
+2+1//0.
""5 $5(
"
>
4
!
" " $
5555
%5
5%
" 5
%
" "
;5 )
!5 )
"
"
55555
555555
32,1&., /1
32,/&,//+2. 1 &3,.
'#8 ! 8!
9& <!:
555555
*
* ;5 )
* !5 )
"
B323 3
B3233/
B3233/
"
#
:
'
4
$$
%
""
&
0&
-2/0&+-/.203-+/-0/
+&-,1
+2,/31.0 -
? "
;
"
>
4
"
17
"
%
"
" "
"
"
&
1+&32.3.3.3.
-2 /./-+,
&,3
2,&333. 1
0&233333
0 233333
03233333
18
-2/0&+/
&32.3.3.
1233333
,233333
" " $
3@3
"
"
A"
5;
%5
5%
" 5
55555
/-2/-/31
+
,.
" " $
55
"
"
>
4
!
;
%
" "
;5 )
!5 )
* 7 7
*@ 7 7
*@ 7 7
555
321-+0&&
32&/
32 ,31-0
320+-,-/
555555
B ;
*
* ;5 )
* !5 )
B2333
32333/
32333/
55555
'#8 ! 8!
"
" $
555555
323,,/
323,0+
323,1,
323,/
9&
"
"
0&
-2/0&+-+
""5 $5(
"
>
4
!
" " $
5555
%5
5%
" 5
%
" "
;5 )
!5 )
"
"
55555
555555
32&/333333
32 ,31-0 3
320+-,--0/
555555
*
* ;5 )
* !5 )
'#8 ! 8!
9, <(
323,1
323,1
323,/
"
#
:
'
4
&
&&2//1-31+
532&0,+30+
+.132++33,/1
%
""
$$
? "
;
"
>
4
"
" "
"
19
"
"
%
"
&
&11
+2/-/-/-//
5320&, -1
0&
320,1/-1./
&-233333
&-2/3333
&/233333
20
&2//1+2/-/-/
1233333
,2/3333
" " $
"
"
A"
5;
%5
5%
" 5
55555
,&2-.+ /
+
,.
" " $
55
"
"
>
4
!
3@3
;
%
" "
;5 )
!5 )
* 7 7
*@ 7 7
*@ 7 7
555
B ;
*
* ;5 )
* !5 )
B2333
32333/
32333/
55555
32.-- 0/
32 ++++0
323/ 30,
32,3.+&
'#8 ! 8!
"
" $
555555
555555
32//-3
*32 /33
*32&/33
*32&/33
9,
"
"
&&2//1-3.
""5 $5(
"
>
4
!
" " $
5555
%5
5%
" 5
%
" "
;5 )
!5 )
"
#
55555
32 ++++++0
323/ 30,,1
32,3.+ ./1
21
"
555555
*
* ;5 )
* !5 )
555555
*32 /3
*32&/3
*32&/3
'#8 ! 8!
9- <
=
"
#
%
""
:
'
4
&
,2,,,,,,,,
32-.&,+/.+
321 &-3,1,+
-2003.01.
$$
? "
>
4
"
32-.&,0
32&-&-&
233333
233333
" " $
"
A"
>
4
!
5;
%5
5%
" 5
3@3
" " $
55
"
;
%
" "
55555
&,2-/&31
+
,.
"
"
&
-3
32&-&-&-&5 2+/
&2+++++++0
32 -& ,,1
"
" "
"
,2,,,,,,
,2333333
,2333333
"
"
%
"
;5 )
!5 )
* 7 7
*@ 7 7
*@ 7 7
555
32+313.
32- 0-+/
32- -0./
&2,&,/.
22
555555
B2333
32333/
32333/
55555
B ;
*
* ;5 )
* !5 )
555555
32333
B323 33
B3233/3
B3233/3
'#8 ! 8!
"
" $
9-
"
"
,2,,,,,,
32-.&,++
""5 $5(
"
" " $
5555
>
4
!
%5
5%
" 5
"
%
" "
"
55555
555555
32- 0-+-03
32- -0./-1
&2,&,/1..1
;5 )
!5 )
'#8 ! 8!
9/ <
555555
*
* ;5 )
* !5 )
B323 3
B3233/
B3233/
"
#
%
""
:
'
4
&
102/1,,,,
--2+. 01.+
2+3.1-.,+
---&&
&,21&/3,&&
$$
? "
>
4
"
--2+. 0.
..0
+&233333
/32/3333
" " $
"
A"
>
4
!
5;
%5
5%
" 5
3@3
" " $
55
"
;
%
" "
55555
-2/,.0+
+
,.
"
"
&
&&/
..02,/+3+
,20&,,/& .032. +0
&2.3 -31-
"
" "
"
102/1,,
1023333
1023333
"
"
%
"
;5 )
!5 )
* 7 7
*@ 7 7
*@ 7 7
555
321,.&.0
32 .1,.0
3231+-.+
32+- .+1
23
555555
B2333
32333/
32333/
55555
B ;
*
* ;5 )
* !5 )
555555
323&0
*32 /33
32 //,
3230-0
'#8 ! 8!
"
" $
9/
"
"
102/1,,
--2+. 0.
""5 $5(
" " $
"
>
4
!
5555
%5
5%
" 5
"
%
" "
"
55555
555555
32 .1,.0,/
3231+-.//1
32+- .+1 .
;5 )
!5 )
555555
*
* ;5 )
* !5 )
'#8 ! 8!
9+ <
*32 /3
32 //
3230/
=
"
:
'
4
&
2/
-2.+,/3, +
32-+3,,-,&
1/1
-,2 +31.0
%
""
$$
? "
>
4
"
"
5
A"
-2.+,/3
&-2+,+,+
/233333
0233333
)
"
"
" " $
"
"
8
"
&
,1
&-2+,+,+,+
5321+& /-&
&0
2-,&1,..-
"
" "
"
2/3333
2/3333
1233333
"
%
"
$ &
" :
3@3
"
123&+3 .
+
,.
24
55555
* 7 7
*@ 7 7
*@ 7 7
555555
B2333
32333/
32333/
$ &2
" " $
"
55
5;
%5
5%
" 5
>
4
!
"
&
,
/
+
0
1
.
3
&
/.-/0
/-1-0
/ ,/3
+33-&
+31.1
/ ./+
-+31/
-/.+&
/11//&&0.
/+03+11+-
#
"
;
%
" "
;5 )
!5 )
555
55555
32.,+0
32 0+,30
323-,+&0
32,3&1-&
9&
9,
9-
9/
9+
+,
+0
+0
+0
++
03
+0
+,
+0
11
++
+,
0&
03
00+
0
0&
03
10
+1
++
03
&&
&+
&/
&&3
&/
&/
&1
&0
&&
&&3
,
,
,
,
,
,
,
,
,310
//
-&
-,
-1
&3,
..
+0
10
&3
& /
,
,
+
&3
1
0
/
1
+
&
.
B ;
*
* ;5 )
* !5 )
555555
32-/+0
*32 /33
*32&/33
*32&/33
"
+3. 123/
/&& .23/
/-3/-21//1.32+,
+,/0&20/
/,3,02.+
/-3/-21-..3,2,
/3,1,2&+
/&---230
/+&&.2/.
+-/1.2+,
"
5 -+ 23/
&+&02./
5&03-21- / 2,0
5&+0-20/
5 31 2.+
50.+.215,.- 2,
1-03205 +/230
-0-2-&0-2,0
"
532,,0.32/1-//
532/100321./+.
532+0,13
532&,+1
5 20, 01
532..&1+
2.+/&.
532 & 132 3&+0
2 &1/0
Studentized residual of the final model

Studentized Residual
2
-1
-2
48000
50000
52000
54000
56000
58000
Predicted Value of y
25
60000
62000
64000
66000

Building Models in Mining

Diunggah oleh

Informasi Dokumen

Deskripsi Asli:

Hak Cipta

Format Tersedia

Bagikan dokumen Ini

Bagikan atau Tanam Dokumen

Opsi Berbagi

Apakah menurut Anda dokumen ini bermanfaat?

Apakah konten ini tidak pantas?

Hak Cipta:

Format Tersedia

Building Models in Mining

Diunggah oleh

Hak Cipta:

Format Tersedia

_________________________________________________________________________

Identification of Critical Key Performance Indicators for

Madinoge Jeanette Maserumule

Submitted in partial fulfilment of the requirements for the degree of

FACULTY OF ENGINEERING, BUILT ENVIRONMENT AND INFORMATION

Literature review was done to

2. Literature review ............................................................................................................. 8

3. Multiple regression analysis ........................................................................................ 15

4. Computational Results ................................................................................................. 24

Anglo Coal South Africa

Key performance indicator

Statistical Analysis Software

production unit of a mine, where coal is mined

electric powered vehicle with a capacity of 20 tons used to transport

the movement process that a CM undergoes to cut coal

that reflect the

1.1 Mining Process

Figure 1.1: Shift breakdown

Figure 1.3: Mining process, cutting cycle flow chart

Mines are different in terms of

daily production targets etc.

1.2 Pick Six

First sump late

Figure 1.4: KPI tree

achieve planned coal production targets in order to meet customer demand

operate mining equipments efficiently by reducing the machine idle time

*Measurement unit: seconds

Figure 1.5: Production process [6]

Sectional conveyor breakdowns

Trunk conveyor breakdowns

*Measurement unit: Minutes

Average production rate

1.3 Problem Statement

1.4 Research design and methodology

2.1 Problem formulation

2.2 Time series analysis

Identifying the nature of the phenomenon represented by the sequence of

forecasting i.e. predicting future values of the time series variable

the mean squared errors

sum of the squared errors

mean of the squared errors

2.3 Multiple regression analysis

E(y) = deterministic component

2.3.1 Least squares estimation

are estimated using the method of ordinary least

squares of which the coefficients minimises the error sum of squares

data is sampled randomly and independently from the population

and the deviations of the predicted

values from the actual Y values are normally

distributed with equal variance for all predicted values of

2.3.2 Partial ordinary least squares

The total variation is measured by the following formula;

And the variation that can be predicted is measured as follows;

The residual sum of squares is measured as follows;

C - + ' "" "-+

2.3.3 Sequential ordinary least square analysis

2.4 Principal component analysis

2.5 Linear programming

satisfy all constraints simultaneously

achieve the stated objective

2.6 Excel model

list of application objects that can be managed

methods that can be performed on the object