26
NuclearPlantJournal.com Nuclear Plant Journal, March-April 2016
the classification process such as data
mining, artificial neural network, hybrid
classification that combines supervised
and unsupervised techniques, expert
systems or any combination of these.
A viable option that became available
because of the high processing power of
modern computers is the combination of
two or more classification techniques with
automated refinement of the results using
unification methods such as the voting
system or reclassification proposing the
unique result that represents the best of
all individual results.
WTT is an example of the hybrid
automated analysis that does not need
any sophisticated detection. It attempts
to mimic human analysis process. In the
preparation phase, data mining is used to
extract all relevant logic and relations that
exist in the degradation signals available
in the training data, in the historical data
from different plants and steam generator
models, and in the industry operating
experience data. This logic and relations
are then used as the hidden layers in the
artificial neural network type classification
technique. Instead of extracting signals
using the detection techniques, analysis
relies on “walking through the tube”
(thus WTT) in a similar way that a human
analyst does when analyzing the data.
The system analyzes each segment of the
tube, varying analysis parameters based
on the signal characteristics similar to
what human analysis is doing. Signals
are funneled through the layers of the
network and classified accordingly.
This is just one of the possible
ways to attempt to mimic the manual
analysis process. Prerequisites for any
unconventional automated analysis are
(1) large amount of experience that needs
to be built into the system, (2) controlled
process flow that has built in firewalls
preventing system failures such as dead
ends, infinite loops or non recognition
of the atypical relevant signals due to
the overfitting, and (3) large amount of
representative training data. As these
types of analysis typically include
significantly more processing than the
conventional ones, more computer power
is needed.
Other inspection techniques for
steam generator tubing examination
exist such as ultrasonic inspection and
eddy current flux leakage. They are used
for punctual, specialized expertise and
typically not as an automated analysis
process.
Smith:
Our business is to sustain
and advance nuclear energy, and we
are committed to invest in research
and development for the future of our
industry. From an ultrasonics perspective,
indications have been missed through
manual analysis. We’re focusing research
and development on automated UT
analysis capabilities, mimicking what
we’ve done for steam generators on the
ultrasonics side of things by developing
these new automated technologies.
4.
How is AREVA Maximizing the
steam generator life?
Smith:
We are inspecting steam
generator tubing from the primary side,
examining steam generator tube bundle
and susceptible areas from the secondary
Pillars of...
AIDA
3
.
