Geoscience Reference
In-Depth Information
failure
i.e., the variability inter failures
may inform more on the variation on the
evaluation of each failure
the variability intra failure
than the outer information
brought by the variation between specialists.
11.5 Classification Example
A comparison of results of probabilistic composition with other approaches, based
on data of Tay and Lim ( 2006 ), is performed in Sant
'
Anna ( 2012 ). The modes of
failure are classi
five classes of Table 11.4 with
extremes derived from the number of modes of failure examined.
In Sant
ed according to RPP into the
cation by global
probabilistic evaluations with another directly provided by experts is studied.
The data collected by Tay and Lim ( 2006 ) refer to three processes in semi-
conductors production. The
'
Anna ( 2012 ), the combination of this automatic classi
first is the test handler process. It admits failures in
handling integrated circuits for testing through an interface unit. Table 11.5 presents
for this process the initial evaluations according to each factor of risk and the results
of the probabilistic composition. The fourth column presents the probabilities of
being the highest priority according to the three criteria together multiplied by 10 6 .
The
fifth column presents the result of application of the thresholds in Table 11.5 to
discretize the values in the fourth column. The last column presents the minimum
between the values of the preceding column and the values assigned by experts.
For the three processes, the probabilistic classi
cation is, in general, in lower
levels than the experts
cation, but the differences are small. Only once
the experts reduce the probabilistic level, even so only from
'
final classi
,
what would keep the corresponding failure still indicated to provoke corrective
action in the process.
The failure object of reduction by the experts has values 9, 3 and 1 for severity,
frequency and undetectability, respectively. The rank inversion in the experts
very high
to
high
'
global evaluation was found to re
ect a predilection for the application of the
occurrence criterion.
The discretization preserves the variability of the global probabilities, with the
classi
cation of the modes of failure varying from 1 to 5 in all the three cases. The
discrete probabilistic classi
cation is much more spread than that of the global
scores provided by the experts when asked to evaluate the whole set of potential
failures.
Table 11.4 Probabilistic classes
n 3 /3
>n 3 /3
and
>2n 3 /3
and
>n 3
and
>3n 3 /2
2n 3 /3
n 3
3n 3 /2
Very low risk
Low risk
Moderate risk
High risk
Very High risk
Search WWH ::




Custom Search