22nd Congress of International Council of the Aeronautical Sciences, Harrogate, UK, 28 August - 1st September, 2000
Paper ICAS 2000-6.2.1
MAINTENANCE OPTIMISATION OF A DIGITAL ENGINE CONTROL SYSTEM WITH LIMIT FAILURE RATE CONSTRAIN
M. Boussemart (1), T. Bickard (1), N. Limnios (2)
(1) Snecma Control Systems, France;
(2) Université de Technologie de Compiègne, France
Keywords: Null
In this paper, we consider error tolerant
systems that remain fail-operational when
affected by some identified faults. The idea is to
use this feature to enhance the maintenance
procedures for safety-critical systems having a
stochastic failure scheme (e.g. electric and
electronic control components) when they are
embedded in a larger system composed of lifelimited
parts requiring periodical overhaul (e.g.
a jet engine).
The current certification objectives require
the manufacturer to show that the system’s
asymptotic failure rate is bounded to a
prescribed value. One major constrain when
optimizing the maintenance cost is to fulfil this
certification objective.
The paper starts with an unambiguous
redefinition of often misused probabilistic terms
such as failure rate and asymptotic failure rate.
Then some theoretical results are given to
compute the associated figure with continuous
and discrete Markov models. These models are
handled using studies about positive matrices,
which calls for the Frobenius spectral analysis.
In a second part, some examples of
electronic control system architecture are given
with some proposed associated failure model.
Distributed architectures are particularly
detailed because they are suspected to be well
adapted to provide extended time limited
dispatch capabilities due to their multiple
reconfiguration capability.
The optimization of economical criteria is
then introduced. Controlled Markov models are
shown to be well suited to solving maintenance
problems. These systems can be described as a
finite state machine. Each state transition is
associated with a decision making : dispatch or
repair. The cost of each alternative is evaluated
considering the original state and the time since
last maintenance action. Further decisions are
oriented by all past actions.
The optimization consists of computing a
matrix linking the decision probability with the
state.
The optimization criterion is the mean
operating cost considered over the up periods.
The rationales of the choice for that economical
criterion are given.
The optimization problem is then turned
into a linear optimization scheme, which is easy
to solve with a simplex algorithm. For our real
problem, facing a too large number of
unknowns, an other approach need to be
developed.
Finally, a complete and easy to figure out
example is given. Our method is applied on a
triple modular redundant computer but also on
a distributed architecture. The missions are
supposed to be of constant duration. The state is
observed at each mission end and the
probability figures are computed, providing a
help in taking the decision to repair or to
dispatch by indicating what is the best action to
minimize the operating cost on the long term.
view full paper