You are on page 1of 15

90 198

A Knowledge-based Decision Support System for


Mining Method SeleCtlon for
Ore Deposits
Sukumar Bandopadhyay, University of Alaska Fairbanks
and
P. Venkatasubramanian, Temple University, Philadelphia
INTRODUCTION
In recent years, research in the field of artificial intelligence (AI) has had many
important successes. Among the most significant of these has been the development of
powerful new computer software known as the "expert systems".
Expert system programs designed to provide expert-level consultative advice in
mineral exploration (Duda et aI, 1981), scientific (Feigenbaum, Buchanan and Lederburg,
1971) and medical (Shortliffe, 1976) problem solving are generally acknowledged

to

be

among the forerunners of this research.


As the decision makers virtually in all fields face a more complex and involved
world within which to operate, the need for some decision support is also becoming
urgent. Thus applications of expert systems continue to spread out reaching problems
that, because of their dimensions or particular aspects, set more demand on the decision
methodologies. To meet these new requirements, many activities which were performed
in the past by the domain expert or engineer must become automated. Furthermore,
previous research (Dawes and Corrigan, 1974; Dawes, 1979) has shown that automating an
expert's decision rules often provide better decision than the expert does. For large
systems it would be very useful (or even necessary) to have formal tools, allowing one for
example, to automatically discover inconsistencies, contradictions or redundancies, or to
identify the possibilities of wrong lines of reasoning in the decision process.

90-0126.KNO

Page 2

Although computers and simulation models have become indispensable tools in


many mine planning endeavors, there is continued reliance on the human expert's ability
to identify and synthesize diverse factors, to form judgments, to evaluate alternatives and
to aid decisions.
Traditionally simulation methods and other analytical techniques have been used to
aid decision making process in mine planning problems. In some sense all these programs
and techniques try to behave expertly in their attempt to perform some well defined set of
tasks. However, some domains are more highly constrained, not easily amenable to
precise scientific formulations, i.e., domains in which experience and subjective judgment
plays an important role.
While the domain of decision-aid is of immense practical value, it is also of
considerable interest in terms of its AI research content. In its most general form, it
involves representing the structure and functions of complex systems, along with some
knowledge about the problems the system is intended to deal with. Inference mechanisms
are needed which can perform completely, even expertly, in domains where system
variables are ill-defined and fuzzy.
Many variables associated with geological, geotechnical, environmental, and other
conditions influence the selection of a mining method for a given mineral deposit. Each of
these variables in turn depends upon other characteristics, for example, geological
variables depend upon the thickness of are body, the grade, etc.; and geotechnical
variables depend on the rock strength, the presence of fractures, etc. Each set of variables
has significant influence on the selection of a method to mine a deposit. In reality, mine
conditions are so varied that an acceptable decision rule cannot be easily written that
covers the selection of a specific mining system or method for all mines or mineral
deposits. The combination of conditions that affect the analysis for one mine cannot

90-0126.KNO

Page 3

necessarily be applied

to

another mine. By changing just only one variable or condition, a

permutation is created that is not applicable to another mining locale.


The selection of mining method for a mineral deposit is thus a decision problem of
the most general sort, where solution is considered to be a highly skilled art. It is
complicated enough to justify development and use of an expert system. Even the best
geological conditions are difficult to cope with if anything less than the most efficient
mining method is used. In addition, subjective judgments are applied to information about
many geological parameters which are inherently descriptive. Impreciseness arises from
the use of descriptive and some-what ill-defined terms. For example, a decision variable
such as ore body thickness is often expressed as "moderately thick". Similarly, the strength
of the hanging wall of an ore body, expressed as "weak". These qualitative expressions of
important variables in the decision process leads to complexity. Human judgment, based
on experience with mineral seams, and geologic conditions, remains the single most
important input of the decision-making technique in the mining method selection.
In this paper, a knowledge-based decision support expert system model for mining
method selection is presented. The expert systems model not only helps select the correct
mining method, it also helps ensure that all important variables have been examined. It
should serve as a check list to be certain that nothing is forgotten. The best of these
systems enable company specialist to maintain knowledge bases that provide mine
planners with valuable engineering support. The knowledge base can be constantly added
to and edited, becoming, in the longer term, a central part in company's information
resources.
An Expert System for Mining Method Selection for Ore Deposits.

Ore deposits are often characterized by extreme complexity, therefore the number
of methods and their variants used in the practice of mining ore deposits is quite

90-0126.KNO

Page

t1

considerable. The diversity of mining conditions and the great number of existing systems

complicate the elaboration of a simple classification of methods for mining ore deposits.
Many researchers believe that the following ten basic mining methods, not including
hydraulic or solution mining, reflect the essence of the methods to be considered in any
selection process.
1.

Open Pit

6.

Room and Pillar

2.

Block Caving

7.

Shrinkage Stoping

3.

Sublevel stoping

8.

Cut and Fill

4.

Sublevel caving

9.

Top Slicing

5.

Longwall

10.

Square-Set Stoping

The major factor in determining the mining method classification is ground


support, which, in turn, depends largely on the geologic characteristics and mechanical
properties of the ore deposits and its host rock. Boshkov and Wright (1973), Morrison

(1976), Tymshare, Inc. (1981), Nicholas (1981) and others have presented schemes for
selecting mining methods. Boshkow and Wright (1,973) listed the mining methods possible
for certain combinations of ore width, plunge of ore, and strength of ore. Morrison (1976)
classified the mining methods into three basic groups, rigid pillar support, controlled
subsidence, and caving; he then used general definitions of ore width, support type, and
strain energy accumulation as the characteristics for determining mining method (Figure
1). Laubscher (1977), on the other hand, developed a detailed rock mechanics
classification from which cavability, feasibility of open stoping or room and pillar mining,
slope angles, and general support requirements could be determined. Tymshare, Inc.

(1981) developed a numerical analysis that determines one of five mining methods, (1)
open pit, (2) natural caving, (3) induced caving, (4) self-supporting, and (5) artificially
supporting, and calculates the tonnage and grade for the type of deposit described. This

(
90-0126.KNO

Page 5

method is meant to be used as a pre-feasibility tool for geologists.


The decision-making process can be treated as the selection of a particular
alternative from a given set of alternatives so as to best satisfy some given goals or
objectives. The problem to be solved is to evaluate alternatives, e.g., to calculate their
utilities. An expert system for decision-making has to establish an appropriate knowledge
base and use it for utility calculation. In addition to this, it has to explain the way the
utility was calculated.
0-3Om (O-100ft)

Room
Pillar

ShrinkoQ.

'0

'0

. S10cinQ

.!
'0
~ e

80
0

'0

...

::l

:I
0

i-

.=Ic

'0

II)

L
.J

II)

+3Om(+100ft)

Figure 1: A Method Selection Scheme (after Morrison, 1976)

90-0126.KNO

Page 6

The explanation of utility calculation is especially interesting because decision

making knowledge is subjectively defined and often imprecise. It offers different


interpretations and has some degree of uncertainty. This kind of knowledge is usually
referred to as "imprecise knowledge" or soft knowledge. The prevalence of imprecision
increases when the domains are socio-economic in nature. Such domains often have to
contend with nebulous terms and reasoning rules.
Consider the statement "weak hanging wall and weak footwall characteristics have
highly significant influence on the selection of the square-set method for mining an ore
deposit". This statement is useful in selecting a mining system. Note, however, that the
statement is far from precise. First, there is uncertainty in the proposition. Second,
several terms in the statement are ill-defined. "Good" footwall characteristics and "highly
significant" are examples of types of imprecision, distinct from uncertainty, which
arise frequently, and will be referred to as fuzziness. One indication of their being
different is that one type can arise independently of the other.

EXPERT SYSTEM ARCHITECTURE


According to a definition generally accepted as true, expert systems are
characterized by the independence of the control structure and the knowledge base.
Nevertheless, there are different classes of expert systems, each of them implementing a
certain strategy. The architecture of the expert system presented here is based on the
model developed by Bohanec et aI, (1983). The selection of the above formal model was
motivated due to the fact that a semantic tree is a natural form for representing decision
knowledge and provides a suitable framework for experts for systematically formalizing
their decision expertise (Duda et aI., 1978). The tree structure facilitates a gradual
aggregation of the

basic variable

values

through

aggregate

variables.
r("..

\..;

90-0126.KNO

Page 7

Using a top-down approach, a semantic tree with multiple nodes and several leaf
variables (Figure 2) has been defined. Much of this knowledge is internalized in a
knowledge base as production rules, which are IF-TIIEN relationships. A standardized set
of knowledge-independent predicate functions and a range of knowledge specific
attributes, objects and associated values form the vocabulary of primitives for constructing

x =IX
Xl

7 7

X 10 I

= fl (X 27

X2 =

X3 )
f2 (X57 X6)

X3 = f3 (X 41 X7

X4= f4(XS 7 X9, Xl0)

/i~
Figure 2: A Semantic Tree

90-0126.KNO

Page 8

rules. A rule premise is always a conjunction clause, and the action part indicates one or

C\

more conclusion that can be drawn if the premises are satisfied, making the rule purely
inferential. When a quesiton is asked of the knowledge base, a knowledge three is
generated. The derivation of the knowledge is a forward process, where as evaluation of
the tree is a backward contraction process -- a pull-back in the structure of facts.
The Knowledge Base and User's Interface

The knowledge of an area of expertise is generally of the three types: facts, rules of
good judgement (heuristics), and evaluations. The crucial problem in the mining method
selection process is the interpretation of the knowledge, such as:
1.

depth of the orebody and character of the overlying rock,

2.

size, shape and dip of the ore body,

3.

mechanical characteristics of the ore and surrounding rock,

4.

ore grade and degree of continuity.

The goal of the evaluation process is obtained in terms of a preferred mining


method and a description of the mining method. In order to achieve this task, production
rules have been developed which lend themselves to symbolic reasoning.
Within the expert system the knowledge is represented by 4-uplets of the type:
(PARAMETER, CONTEXT, VALUE, CF)
The CONTEXT is instantiated by the name of a mining method. Each
PARAMETER corresponds to an attribute of this CONTEXT and the VALUE qualifies
the attribute. Finally the certainty factor (CF) defines the plausibility of the context. The
plausibility is a number belonging to the (-10, + 10) range (where -10 means false and the

+ 10 means true) and where all the possibilities between the absolutely true and absolutely
false are represented by a number between -10 and + 10 inclusive. For example:

90-0126.KNO

Page 9

(Gen-shape, BLOCK CAVING, irregular, -1/10)


signifies that the selection of block caving is not at all probable when the general shape of
the ore body is irregular. Whereas, (ore-thickness, BLOCK CAVING, very thick, 4/10)
signifies that the selection of bock caving as mining method is probable if the ore body is
very thick.
Since the rules are usually judgmental, that is, they make inexact inferences on a
confidence scale, the conclusions are therefore evaluated by certainty factors. Standard
statistical measures are rejected in favor of certainty factors because experience with
human experts shows that experts do not use information in a way compatible with
standard statistical methods (Negoita, 1985). Thus for example, if some basic variable
describes the "footwall characteristics of the overburden as weak", we can specify the
"selection of square-set mining as a primary method with a certainty factor 4/10".
Certainty factors (CF) are a measure of the association between the premise and
action clauses in each rule and indicate how strongly each clause is believed. When a
production rule succeeds because its premise clauses are true, the certainty factors of the
component clauses are combined. The resulting certainty factor is used to modify the
certainty factor in the action clauses. Thus, if the premise is believed only weakly,
conclusions derived from the rule reflect this weak belief. Also, because conclusions of
one rule may be the premise of another, reasoning from premises with less than complete
certainty factor is common place.
For each rule in the system, a CF is assigned by the domain expert. It is based on
the expert's knowledge and experience. The CF that is included in a rule is a component
certainty factor (CF comp), and it describes the credibility of the conclusion, given only the
evidence represented by the preconditions of the rule. The rules are so structured that any

)
90-0126.KNO

Page 10

given rule either adds to belief in a given conclusion or adds to disbelief. Because there
are many rules that relate to any given conclusion, each of which can add to the overall
belief or disbelief in a conclusion, a cumulative certainty factor is used to express the
certainty of the conclusion, at a given point in execution, in light of all of the evidence that
has been considered up to that point.
Inference rules are defined as situation - action pairs. The left member (i.e., the
situation) describes a constraint on each of the certainity factors associated with several
events. When the constraint is satisfied then the right member (Le., the action) of the rule
is triggered. This "action" modifies the certainty factor associated with all the events
belonging to the right member of the rule, following the certainty factor combination law.
For example:

/* Geometry * /

gen-shape (Open-pit, Massive, 3/10).


gen-shape (Block-caving, Massive, 4/10).
ore-thickness (Open-pit, Narrow, 2/10).
ore-thickness (Block-caving, Narrow, -1/10).
o-rack-strength (Open-pit, Weak, 3/10).
o-rock-strength (Block-caving, Weak, 4/10).
o-fracture-spacing (Open-pit, Close, 3/10).
o-fracture-spacing (Block-caving, close, 4/10).
/* Rule Base * /
Start -7 decision (X,A,B,C,D,K), Write (X,K).
Decision (X,A,B,C,D,K), -

Cl
".J'

j* Ore-zone'" /

geometry (X,A,B,K1),
ore-zone (X,C,D,K2),
K = min (K1,K2).

Geometry (X,A,B,C,K1) -

gen shape (X,A,L)


ore-thickness (X,B,M)
Kl = min (L,M).

Ore-zone (X,C,D,K2) -

o-rock-str (X,C,L1)
o-fracture-spacing (X,D,L2)
K2 - min (L1,L2)

The activation of the rules modifies the certainty factor by combining the individual
certainty factors from each parameter group (Figure 3).

(
90-0126.KNO

Page 11

START

LEVEI:1

WRITE(X ,K)

K= MIt-HK1,K2)
MIN(K1,K2 )

LEVEL-2

GEOMETRY
(X,A,B,K1)

K2= MIN(L1,L2)
MIN (L1, L2)
LEVEL-3

K1=MIN(L,M)
MIN( L,M)

ORE ROCK

ORE-THICKNESS
(X,B,M)

STRENGTH

ORE FRACTURE
SPACING
(X,D,L2)

(X,C,L1 )

Figure 3: A Segment of the Mining Method Selection Semantic Tree


User Interface
The output of the mining method expert system is a characterization of each are
deposit in terms of the stability of the ground (hanging wall, footwall, and are zone) and its
influence on the selected mining method. The strength properties of the hangwall,
footwall and the ore zone are characterized by the ratio of the uniaxial strength of the
material to the overburden pressure, the fracture spacing and the fracture shear strength.
The characteristics of the of the are body are defined by the general shape, the are
thickness, the plunge of the are body, and the grade distribution. Information acquired
from the external environment is qualitative and imprecise -- "narrow", "thick", "uniform",
etc. Based on terms of this type, and the sets of heuristic rules, inferences are developed.

90-0126.KNO

Page 12

The determination of the ground stability and characteristics of the ore body is used
to determine the mining method. Figure 4 is an example of the user interface with the
system. The expert system contains 13 parameters, and the geological knowledge is
encoded within 514 produciton rules. Using the resources available at the University of
Alaskaz Fairbanks, this expert system was implemented and studied on a VAX 750
computer, using essentially standard Prolog.

$ pro
C-Prolog version 1.5
1 ?- ['methfux.pro].
methfux.pro consulted 27752 bytes 1.90201

yes
1 ?- start.
Questions on geometry and grade dbn of deposit
General shape
m: Massive
tp : Tabular or Platy
i : Irregular

I: i.
Ore thickness
n: Narrow
i : Intermediate
t: Thick
\It : Very thick

Rock mech characteristics for hanging wall


Rock material strength
w: Weak
m: Moderate
s: Strong

I:w.
Fracture spacing
vc : Very close
c: Close
w:Weak
vw : Very weak
I: c.
Fracture strength
w: Weak
m: Moderate
s: Strong

(,

I:w.
Rock mech characteristics for ore zone

I: i.
Ore plunge
f: Flat
i : Intermediate
s : Steep

Rock material strength


w: Weak
m: Moderate
s: Strong

I: s.
Grade distribution
u: Uniform
g : Gradational
e: Erratic

Fracture spacing
vc: Very close
c: Close
w:Weak
vw : Very weak

I: u.

I:vc.

I:w.

(
90-0126.KNO

Page 13

Fracture strength
w:Weak
m: Moderate
s: Strong

I:w.
Rock mech characteristics for foot wall
Rock material strength
w:Weak
m: Moderate
s: Strong

I:w.
Fracture spacing
vc : Very close
c: Close
w: Weak
vw : Very weak

I: c.
Fracture strength
w: Weak
m: Moderate
s: Strong
I:w.
Mining methods and their correspoinding scores
2

o
-1

-1

o
o
2

o
3

Openpit
Block Caving
Sublevel Stoping
Sublevel Caving
Longwall
Room and Pillar
Shrinkage Stoping
Cut and Fill Stoping
Top Slicing
Square Set Stoping

no

I?1 Exit

[ Prolog execution halted 1

Figure 4: An Example of the User-Interace with the Expert System.

90-0126.KNO

Page 14

CONCLUSION
Expert system models are still evolving, both theoretically and in terms of their
practical applications in mining engineering. It is an useful tool for the domain considered
since analytical models are not amenable. Mining integrates the skill of many engineering
disciplines. Within these disciplines lies experience and expertise found in not other
industry. To capture and widely apply this expertise is the challenge to developing
knowledge base systems. This paper shows how the methodology of expert systems may.be
integrated in a mining method selection process. The integration of expert knowledge in
designing an inference process seems to be advantageous for many technical reasons.
REFERENCES
Bohanec, M., Bratko, 1., and Rajkovic, V., 1988, "An Expert System for Decision Making",
Proceedings of the Joint IFIPWG 8.3 lHASA Conference on Processes and Tools
for Decision Support, H.G. Sol, (efd), North-Holland, Amsterdam, 254 p.
Boshkov, S.H., and Wright, F.D., 1973, "Basic and Parametric Criteria in the Selection,

c.

Design and Development of Underground Mining Systems", SME Mining


Engineering Handbook, Chapter 12.1, Vol. 1, SMEI AlME, p 12.2-12.13.
Dawes, R.M., 1979, "The Robust Beauty of Improper Linear Models in Decision Making",
American Psychologists, Vol. 34, pp. 571-582.
Dawes, R.M., and Corrigan, B., 1974, "Linear Models in Decision Making", Psychological
Bulletin, Vol. 81, p. 95-106.
Duda, R., Gaschnig, J., and Hart, P., 1981, "Model Design in the Prospector Consultant
System for Mineral Exploration", Expert System in the Micro-Electronic Age, D.
Michie (ed), Cambridge Press, p. 154-167.
Duda, R., Hart, P.E., Nilsson, N.J., and Sutherland, G., 1978, "Semantic Network
Representations in Rule-Based Inference Systems", Pattern Directed Inference

(:.
\

90-0126.KNO

Page 15

System, D.A. Waterman and F. Hayes-Roth (ed), Academic Press, New York.
Feigenbaum, E.A., Buchanan, B.G., and Lederberg, J., 1971, "On Generality and Problem
Solving: A Case Study Using the DENDRAL Program", Machine Intelligence,
Vol. 6, P. 165-190.
Laubacher, D.H., 1977, "Geomechanics Classification of Jointed Rock Masses - Mining
Applications", Transactions of the Institute of Mining and Metallugy of South
Africa, Section A, Vol. 6, p. AI-A7.
Morrison, R.K.G., 1976, A Philosophy of Ground Control, McGill University, Canada, p.
125-159.
Negoita, C.V., 1985, Expert Systems and Fuzzy Systems, Benjamin/Cummings.
Nicholas, D.E., 1981, Method Selection - A Numerical Approach, Design and Operation
of Caving and Sublevel Stoping Mines, D.R. Steward (ed), SME/ AIME, New York,
p.39-53.
Tyrnshere, Inc., 1981, Computer Evaluation of Mining Projects, Mining Journal, Vol. 10,
p.11l.

Shortliffe, E.H., 1976, Computer-based Medical Consultation: MYCIN, American


Elsevier, New York.

90-0126.KNO

You might also like