134
D. Krpelík et al.
So, again, we cannot determine whether H is valid or not for certain based on the
data only. There is always a possibility that a test would conclude incorrectly, but
the probability that would occur may be controlled. If we were given a fixed set
of observation, we are usually able to control just one of the two types of error.
In practice we usually choose to control the type I. error by selecting the test
significance level accordingly. In the quality testing, the two errors represent the
risks to the producer (type I., the whole batch would be rejected wrongfully) and to
the purchaser (type II., a batch would not meet the requirements although the test
concluded that). In the batch testing procedure, we are able to control both error
levels by selecting a proper amount of products to test, since for fixed I. type error,
the type II. error generally decreases with increasing number of observations. The
question that remains is an economical one, what error levels we deem adequate
given that lowering them could be expensive.
4.5 Designing Highly Reliable Systems
Once we are able to construct mathematical models for the occurrence of failure,
the next logical step is to select the design which best suits our needs. Reliability is
a sidekick to performance. Even if one designs a system with the peak performance,
it would be of little use if it would stay in a failure state most of the time. Generally,
we require the failure probability to be as small as possible. But engineering has its
limits and may only take us so far with constructing reliable devices. This section
is to introduce techniques which might be used to improve reliability of critical
systems other than by improving reliability of its components. It is theoretically
possible to construct devices with any required reliability level, but that is usually
impossible in practice due to the possible correlations among the failure times (when
one failure triggers others) or due to possible failure of switching mechanisms.
Nevertheless, the reliability of a system and its lifetime may be increased drastically.
The pay-off is the cost, spatial dimension and complexity.
4.5.1 Redundancy Allocation
One idea leading to an increase in reliability is to identify and fortify critical
components of the systems [4, Ch. 6], [14, Ch. 9]. If there is a component whose
failure will likely lead to the failure of the whole system and whose reliability
cannot be sufficiently improved by changing its design and construction, we might
consider to introduce additional components which would be able to ease the stress
upon this component (share its load) or which could substitute it in the case of a
failure (redundant spares). Both of these may simply be just replicas of the original
component.
D. Krpelík et al.
So, again, we cannot determine whether H is valid or not for certain based on the
data only. There is always a possibility that a test would conclude incorrectly, but
the probability that would occur may be controlled. If we were given a fixed set
of observation, we are usually able to control just one of the two types of error.
In practice we usually choose to control the type I. error by selecting the test
significance level accordingly. In the quality testing, the two errors represent the
risks to the producer (type I., the whole batch would be rejected wrongfully) and to
the purchaser (type II., a batch would not meet the requirements although the test
concluded that). In the batch testing procedure, we are able to control both error
levels by selecting a proper amount of products to test, since for fixed I. type error,
the type II. error generally decreases with increasing number of observations. The
question that remains is an economical one, what error levels we deem adequate
given that lowering them could be expensive.
4.5 Designing Highly Reliable Systems
Once we are able to construct mathematical models for the occurrence of failure,
the next logical step is to select the design which best suits our needs. Reliability is
a sidekick to performance. Even if one designs a system with the peak performance,
it would be of little use if it would stay in a failure state most of the time. Generally,
we require the failure probability to be as small as possible. But engineering has its
limits and may only take us so far with constructing reliable devices. This section
is to introduce techniques which might be used to improve reliability of critical
systems other than by improving reliability of its components. It is theoretically
possible to construct devices with any required reliability level, but that is usually
impossible in practice due to the possible correlations among the failure times (when
one failure triggers others) or due to possible failure of switching mechanisms.
Nevertheless, the reliability of a system and its lifetime may be increased drastically.
The pay-off is the cost, spatial dimension and complexity.
4.5.1 Redundancy Allocation
One idea leading to an increase in reliability is to identify and fortify critical
components of the systems [4, Ch. 6], [14, Ch. 9]. If there is a component whose
failure will likely lead to the failure of the whole system and whose reliability
cannot be sufficiently improved by changing its design and construction, we might
consider to introduce additional components which would be able to ease the stress
upon this component (share its load) or which could substitute it in the case of a
failure (redundant spares). Both of these may simply be just replicas of the original
component.