Aerospace and Electronic Systems - May 2019 - 37

Monedero et al.

Figure 9.
Prediction results for the model of abortive incidences (simulated data).

and TIME_SHIFT > 2.54

ENVIRONMENT

and TIME_SHIFT < ¼ 2.75
then 1.
Thus, this rule generated a pattern for a total of 46 execution tests and had a 95.7% probability of success (that
is, 44 of the 46 test executions that included this rule in
the sample set registered an abortive incidence).
Model to predict nonabortive incidences
Once the different models for abortive incidences
were generated and validated, the model relative to nonabortive incidences (located at the second level of priority
with respect to the previous ones) was approached.
For the generation of the model relating to nonabortive
incidences, the same configuration was used: C&RT trees
(which were the best results generated) and the same 14
input parameters. In addition, the partition was made
based on the technological area parameter (AREA).
Once the model was trained and generated, a prediction
model with 90.69% success was obtained. The C&RT tree
resulted in a total of 739 rules with the next distribution by
technological area: 94 for the tree related to MTISC, 231
for MTISY, 188 for MTISA and, finally, 226 in MTISN.
An example of the rule generated for the technological
area MTISA is detailed below:
Rule 127 for 1 (Instances 36; Confidence 0.917)
if Num_SECTIONS > ¼ 8
and VAR_AIM< ¼ 313
and ABORT_10_PREV < ¼ 4
and DAYS_STATION > ¼ 441
and CONCURRENCE < ¼ 1
and STATION in ["SB1" "SC2" "SG1"]
and ATA in [34 99 23 89 90 43 31]
then 1.
The previous rule included 36 execution tests with a
95.7% probability of success (33 of the 36 test executions
included in this rule registered a nonabortive incidence).

FRAMEWORK ENVIRONMENT
For the use of the prediction models by the Airbus DS
Company, we developed an application to be executed on
Microsoft Windows. This application was developed in
the Cþþ language.
MAY 2019

Thus, once the application is executed, the user is shown
in the main window. This window is shown in the following captured image (see Figure 10).
The window of the application has the following sections, with their corresponding controls with which the
user can interact.
In the upper left part (FILTERS), there is a set of filters that make it possible to select one or more tests from
the historical executions whose pattern needs to be analyzed. Specifically, it contains three filters: ATA, MSN,
and TI. These filters were recommended by Airbus DS to
facilitate the selection of the tests.
The data table shown at the bottom of the previous filters, as indicated by its name (HISTORICAL EXECUTIONS), contains the list of test executions carried out in
the past in the test system, filtered by the three fields
described above.
Once the filter has been applied, the user must select a
row in the table for the selection of a certain execution
test. Once done, the parameters of the selected row are
copied to the list of parameters shown below (PATTERN
ANALYZED). In this part, only those parameters used for
prediction are copied. Thus, these parameters make up an
execution pattern, and the corresponding prediction will
be made using these parameters. In addition, the user can
manually modify any of these parameters in the environment to see how that change affects the probability of an
incidence in the test.
On the right side of the PATTERN ANALYZED,
there are two selectable tabs: ABORTIVE and NONABORTIVE, depending on whether the user wants to carry
out a prediction of the abortive incidence or nonabortive
incidences. Once selected, the prediction information is
shown (PREDICTION), as well as possible actions to be
carried out by Airbus DS engineers to improve in terms
of incidences that result from the test execution
(RECOMMENDATION).
The result of the prediction includes the following
information:
- PROBABILITY: The probability in the percentage
that this execution pattern produces an incidence of
the selected type.
- SUPPORT: The number of executions in the history
of the test system with that same pattern. The higher
this number is, the stronger and more supportive the
prediction will be.

IEEE A&E SYSTEMS MAGAZINE

Aerospace and Electronic Systems - May 2019

Table of Contents for the Digital Edition of Aerospace and Electronic Systems - May 2019