Professional Documents
Culture Documents
Decision Theory in Maintenance Decision Making
Decision Theory in Maintenance Decision Making
net/publication/235320296
CITATIONS READS
76 481
2 authors, including:
George Bohoris
University of Piraeus
39 PUBLICATIONS 368 CITATIONS
SEE PROFILE
All content following this page was uploaded by George Bohoris on 09 April 2015.
Introduction
The basis of decision theory was developed from the theory of games. The
relation between decision theory and games theory can be better perceived
through the similarity of the basic ingredients of each. More specifically, the
decision theory is a particular case of games theory, which, in its simplest form,
is a two-person game. In decision theory nature takes the role of one of the two
players. Besides the similarities, however, there are a number of differences
between games theory and decision theory. More noticeably, in the former,
players are trying to maximize their winnings simultaneously, while in the
latter, nature chooses a state without any particular objective. Furthermore, in
the former players are assumed to be intelligent adversaries, while, in the latter
nature is assumed to make a choice once and for all. In this case, information
about nature’s choice can be obtained more easily.
Since the 1950s, decision theory has been a very active area of research in
several fields such as economics, statistics, psychology, engineering. As a
result, decision theory has become a useful tool to many professionals including
engineers[1]. It is the main purpose of this article to demonstrate a further use
of decision theory in maintenance engineering and management.
In simple terms, decision theory provides a logical framework for solving
real-life problems. It is concerned with the identification of an action which is
expected to provide maximum benefits to the decision maker. Smith[1]
summarized the benefits of decision theory into the following:
(1) The decision maker’s best course of action is derived based on his/her
objectives and knowledge of the problem.
(2) The decision maker communicates to the other specialists the best
course of action and justifies why the chosen action is optimal.
(3) A framework is provided within which the decision maker’s ideas can be
critically evaluated and modified, especially if new information is to be
incorporated or other than common decisions have to be made.
α1 α2
Table I.
Consequences for a θ1 ρ11 ρ12
simple decision θ2 ρ21 ρ22
problem
α1 α2
Table II.
Expected gains for a θ1 U (θ1, α1) U (θ1, α2)
simple decision θ2 U (θ2, α1) U (θ2, α2)
problem
(4) The loss and util ity functions L( ρ ) = L( θ , α ) and U( ρ ) = U( θ , α ) Decision theory
respectively quantify the loss or the gain incurred from each in maintenance
consequence. There is no real difference between these two functions decision making
since L(ρ) = –U(ρ). Normally statisticians prefer to use loss functions,
while decision theorists and engineers usually express the quantification
of ρ in terms of gains or utility. Table I can be also presented as in Table II.
The final expression for the utility function U(θ, α) is obtained from 41
U(ρ) through the conditional probability P(ρ|θ, α) as follows:
U (θ , α ) ≡ U [ P ( ρ |θ , α )] = E [U ( ρ )] ⋅
In other words, U( θ , α ) is the expected value of U( ρ ) and can be
calculated as follows if ρ is continuous:
U (θ , α ) = ∫0+∞U ( ρ )P ( ρ |θ , α )d ρ⋅ (1)
Otherwise, if ρ is discrete:
U (θ , α ) = ∑U ( ρ )P ( ρ |θ , α ) ⋅ (2 )
ρ
(5) The multi-attribute utility theory (MAUT) is closely related to multi-
criteria decision making and allows quantification and aggregation of
multiple objectives even when these objectives are composed of
conflicting attributes, or when they are subjective. More specifically, the
multi-attribute utility theory is related to the case where the consequence
ρ is a vectorial variable, i.e. there is more than one attribute for ρ = (ρ',
ρ", …). For example, in the case of a power-generating company, ρ might
be associated with two conflicting attributes, the interruption time (TI)
of power supply (ρ' = TI), and the cost (C ) to maintain this power supply
( ρ " = C ). In that case if U( ρ ') and U( ρ ") are single-attribute utility
functions then, if ρ' and ρ" are additive independent, the multi-attribute
(additive) utility function is:
U ( ρ ' , ρ ") = k1U ( ρ ' ) + k2U ( ρ ")
where k1 and k2 are scaling constants and k1 + k2 = 1. If, however, ρ' and
ρ" are not mutually independent, then:
U ( ρ ' , ρ ") = k1U ( ρ ' ) + k2U ( ρ ") + k3U ( ρ ' )U ( ρ ")
where k3 is another scaling constant and k1 + k2 + k3 = 1.
In general, the multi-attribute utility theory strikes a trade-off so that
the optimization procedure obtains the best combination of the
consequences (TI and C) by the choice of an appropriate action.
(6) Elicitation and consistency checking. The elicitation of the utility function
involves the understanding and modelling of the preference structure of
the decision maker regarding the consequence ρ. Cultural differences
and varying company policies are very influential here. Normally, the
JQME utility function U(ρ ) is obtained from a procedure which frequently
1,1 involves the following two steps:
● Modelling the multi-attribute utility function.
● Modelling the conditional utility function for each attribute, which
effectively is a measure of the preference for each value of ρ.
42 To ensure that no errors have been committed in the elicitation procedure
a number of consistency checks are available in decision theory. One of
the most popular checks consists of paired comparisons of the various
consequences. That is, the value of the utility function U( ρ ', ρ ") is
compared with the decision maker’s preferences when all the possible
consequences are considered.
(7) Optimization. The purpose of decision theory is to obtain an optimum
solution to a given problem. Once the basic ingredients have been
established, the fundamental question “how to choose the action α so
that the consequence ρ will provide maximum benefits to the decision
maker?” arises. The optimal solution in decision theory can be obtained
by applying a number of procedures according to the characteristics of
the ingredients of the problem. The two main ones are the following:
● The Minimax principle. This is applied when no data or information
is available. The optimization consists of choosing the action αj
which minimizes the maximum loss L( θ i, α j) for any θ i that the
nature might choose. Then, the solution is found from:
min max L(θ ,α . (3 )
α θ
A practical application
CHESF – Companhia Hidro Electrica do São Francisco
CHESF is in the business of power generation and bulk electricity transmission
in the north-east region of Brazil. This region includes nine federation states
and covers an area of about 1,200,000km2. To secure power supply for this
rather large market, CHESF has several power-generating plants, a
transmission system of approximately 15,000km transmission lines, and more
than 70 integrated stations distributed throughout the whole region.
The communications needs of the company, including the operation, Decision theory
maintenance, and management of the electric power supply, are supported by a in maintenance
private telecommunication system owned by CHESF. This system is composed
of approximately 2,000 pieces of equipment also distributed throughout the
decision making
whole region. The practical application discussed in this article (see multi-
attribute utility theory in the “decision theory” section) has been carried out in
CHESF’s telecommunication system. 45
Results
The implementation of the maintenance decision model presented in this article
in CHESF provided a well-defined, easy-to-use framework for the selection of
the optimum action for the maintenance problem considered. Further to that,
the following have been realized in CHESF:
● Preferences concerning typical maintenance decision variables have
been better modelled.
● Prior knowledge about reliability and maintainability has been more
accurately derived and more consistently utilized.
● The confirmation of the results obtained in the optimization procedure
has been made possible through the sensitivity analysis which evaluated
the robustness of the model.
Conclusions
This article has presented a review of some basic decision theory concepts and
discussed their applicability in maintenance. As a result, a maintenance
decision model has been built and the steps for its application in practice have
been detailed. Finally, the real-life use of the model has been demonstrated by
applying it to a maintenance strategy problem in a power generation company.
The numerous benefits realized by CHESF suggest that such a model has a
viable future in maintenance as a reliable framework for maintenance decision
making.
References
1. Smith, J.Q., Decision Analysis – A Bayesian Approach, Chapman & Hall, London, 1988.
2. Berger, J.O., Statistical Decision Theory and Bayesian Analysis, Springer-Verlag, Berlin,
1985.
3. Kadane, J.B., Robustness of Bayesian Analysis, North-Holland, Amsterdam, 1984.
4. Keeney, R.L. and Raiffa, H., Decision with Multiple Objectives; Preferences and ValueTrade-
offs, John Wiley & Sons, Chichester, 1976.
Further reading
Almeida, A.T., Souza, F.M.C. and Bohoris G.A., Preference Elicitation in Reliability Engineering,
IFORS/IFAC Workshop on Support Systems for Decision and Negotiation Process, 24-26 June
1992, Warsaw, pp. 11-16.
Raiffa, H., Decision Analysis, Addison-Wesley, Reading, MA, 1970.