In your publications you inform that off-line diagnostic
systems have 90-95% certainty of a correct defect identification by vibration
signal in rotating machinery. But the probability of an error about 10%, when
there is need in decision whether to proceed the operation of the equipment or
stop it for maintenance, is too large for example for the nuclear power
stations. The requirements, concerning the diagnostic systems that ensure the
safety of nuclear power stations, have a limitation of the level of the error
probability less than 10-4, and sometimes even more severe. Does it mean that
off-line systems are not applicable to control the condition of equipment that
ensures the safety of vital industries?
(The question was asked by the participants of a seminar "VibroAcoustical
Machine and Equipment Diagnostics" held in VAST, Inc.)
The question is answered by
Alexej V. Barkov:
Any off-line diagnostic system and especially the one, that
utilizes indirect information extracted for example from vibration signal, has
limited certainty. But to speak about certainty of any event it is necessary to
detect this event. Concerning the defects and non specified operation condition
of technical equipment all the events can be divided into several classes by its
value, location of appearance and rate of development. The permissible certainty
of defect detection can be significantly dependent on the class to which the
defect belongs.
An off-line diagnostic system is intended to solve two main problems. The first
one is to detect with maximum possible certainty the faults that can lead to
accident. In practice the certainty of such defects detection can be as high as
0.99-0.995 neglecting the probability of false alarm, that can be up to 1-3%. If
the requirement to the certainty of hazardous defects detection is more strict
it is necessary to use on-line control systems and to utilize several diagnostic
signals. The decision to stop the controlled object in this case has to be by
coincidence of several independent symptoms of a dangerous defect. Naturally
that in this case you have to resign the problem of defect identification and
long term condition prediction of the equipment controlled. The second problem
is to detect and identify most defects on the incipient stage of their
development to ensure the long term condition prediction of the controlled
object. Only if you know the type of the defect and its severity, even with the
probability of an error 10-20%, and the maximum typical rate of its development,
it is possible to predict with high certainty the time interval during which
there will be no fail of the controlled equipment. And you are quite right when
you note that the systems with linear interpolation of the results of control
measurements cannot solve the problem of the long term condition prediction. It
is necessary to use nonlinear systems of condition recognition.
It should be noted also that the number of diagnostic parameters control points
for solving the problem of long term machine condition prediction can be
comparatively large, and the methods of control signal processing very
complicated. That is why it is not expedient to solve the diagnostic problems
only with the on-line monitoring systems.
For the equipment that defines the safety of production it is necessary to use
both types of the systems together, optimizing the distribution of the control
points among the on-line and off-line control and diagnostic systems. For
auxiliary equipment and equipment with hot reserve it is expedient from economic
point of view to use one off-line system for a large number (up to 100 and more)
units of diagnosed equipment.
