Random fluctuation of
monitored condition data characterizes many otherwise straight-forward CBM
applications. In this exercise we use the monitored pressure test data, which
reflects the deterioration of a sealing system, in a nuclear fuel rod
manipulating mechanism. For additional background and details on this
application, you may refer to the document Fuel Handling System.
|
Step |
Explanation |
Required actions |
|
1 |
|
Download the database files from candu.zip or from the OMDEC CD. |
|
2 |
|
Start “EXAKT for Modeling”, File, Open, Navigate to locate
the file candu_WMOD database file (in c:\Program Files\Exakt\tutorial4\ if
extracted from the CD) |
|
3 |
Note the randomness yet increasing nature (generally
rising slope) of the data. Although it is obvious that the item ages in a
fairly linear fashion, how does one make a decision at any given inspection
if the data is so erratic? How do we know if a high reading is due to noise
or to a deteriorating failure mode? The following steps in EXAKT provide a
solution to this problem. |
Activate left (database explorer view) pane, View,
Inspections, OK, Ident drop down list, hit various idents and observe their
corresponding sets of inspection data, reduce the inspections window, close
(X) the inspections window. |
|
4 |
EXAKT provides a way to perform “smoothing
transformations” of the data. In the OutputVarScript window you will see a
small program that transforms the original variable LeakRate into the
transformed variables leakSmooth and leakSmoothAve. EXAKT’s
programming language provides several smoothing functions. Smooth() and
SmoothAve() are smoothing functions that take parameters to adjust the
way in which they transform the variables. |
Database pane, OutputVarScript, X (Note that we have defined 4 new variables from the
original LeakRate and WorkingAge variables: leakSmooth0, leakSmooth, leakSmoothAve0, and leakSmoothAve By reading the comments in this script and by
studying (in the Guide and Manual) the definitions of the various
EXAKT transformation functions such as Smooth(), SmoothAve(), Last() and
NonDecr(), you will soon get to understand how these transformations work.
For now, just continue to step 5) |
|
5 |
The instruction on the right generates the decision
graphs of the model built directly on the original (untransformed) data.
Observe how much randomness there is
in the inspection data. Such randomness may bias the model and may make it
difficult to clearly apply an optimal decision. |
A) Modeling (on menu bar), Select Current Model, CBM
Model: Seals, Submodel: LR_b1, OK, Procedures panel, Decisions, Select Ident:
5EH1, scroll down to last row, shift+8WH4, Report, Close, PageDown or PageUp,
X B) Modeling (on Procedures panel), Weibull PHM,
Select Covariates, (note the variable used for this model LR_b1 is LeakRate),
Cancel |
|
6 |
The model LR_Smooth0 uses a variable that has been
smoothed by the Smooth() function in EXAKT. On the decision graphs, we
observe that we have eliminated the randomness of the previous submodel. But
we have another problem. We observe a drooping artifact[1]
at the end of every history. This causes a poor model and a poor decision
recommendation because the current value of the condition indicator leakSmooth0
is erroneously low! In step 7 we will correct this problem with a further
transformation. |
Repeat Step 5A but select the submodel LR_Smooth0
instead of LR_b1 Repeat Step 5B but note the variable used for this
model LR_Smooth0 is leakSmooth0, Cancel |
|
7 |
The adjusted smoothed variable produces a better
model and a better decision recommendation. Note that the randomness of the
data is further reduced and the drooping artifact has been corrected. |
Repeat Step 5A but this time use the submodel LR_Smooth Repeat Step 5B but this time note that the variable
used in the submodel LR_Smooth is leakSmooth |
|
8 |
Now that we have seen some techniqes for
pre-processing data to eliminate confusing noise, we may look more closely at
the model itself. You may be wondering about the naming convention we adopted
for the model “LR_Smooth_b1”. The “b1” part of the name indicates that we have
fixed Beta, the shape factor, to 1. We will proceed to learn why we did this. |
|
|
9 |
We note, in carrying out the steps on the
right, that this Submodel “LR_Smooth”
uses the transformed variable leakSmooth and that the “Fix shape
factor to 1” checkbox is unchecked. |
Modeling (on Procedures panel), Weibull PHM, Select
Covariates, Cancel |
|
10 |
Upon executing the steps at the right, we note that
the model is rejected by the Kolmogorov-Smirnov test. The test is telling us
that the hypothesis that the model is “good” (fits the data) must be rejected. |
Residual Analysis, Summary Report, scroll down.
(note that the goodness of fit hypothesis is rejected), reduce window,
X Look at the modeling results in the orange framed
"Parameters" window inside the Procedures window. Note the NS (not
significant) indication after Shape = 1.35644. |
|
11 |
EXAKT has told us in step 10 that working age is not
significant. In fact it is highly significant, so much so that it
correlates closely with the LeakRate. Thus EXAKT is really telling us
that the LeakRate itself contains all the information we need, to establish a
good predictive model, and it is telling us that we should remove
WorkingAge as a significant factor from the model by setting Shape to 1. |
Modeling (on menu bar), Select Current Model,
LR_Smooth_b1, Modeling (on Procedures panel), Weibull PHM, (note that the
shape parameter has been fixed to 1 for this submodel), Cancel Residual Analysis, Summary Report, expand and scroll down. (note that the goodness of
fit hypothesis is not rejected), X |
|
12 |
Similar results can be found for models:
LR_SmoothAve0_b1, and LR_SmoothAve_b1. You may go ahead examine these models
using the tecniques you have learned in this exercise |
|
Once you have made
smoothing and other adjustments to the model, you may apply cost data as in Tutorial
1 in order to develop the decsion model. it is ready to be deployed as
an intelligent agent. You no longer have to worry about the erratic and
noisy nature of the data. The compensating algorithm has been built into
the model and will be applied automatically each time a new set of condition
monitoring data is received.