IEEE - Aerospace and Electronic Systems - November 2021 - 24

A Software Prototype for Predicting Ground Test Costs on the A400M Aircraft
figures (such as the code for the GTIs) have been hidden due
to a confidentiality agreement between Airbus DS and the
University ofSeville.)
GENERATION OF PREDICTIVE MODELS
For the generation of the different models, the following
phases were carried out:
1) The extraction of data sources: In this first phase,
the extraction of the information available for GTIs
was carried out. For this purpose, a dump of the aircraft
simulation database and the extraction of the
files of the folder tree of this simulation environment
were used as data sources. The simulation
data condense all the information registered prior to
the real test of the aircraft. Specifically, the data
used for the folder tree include both the information
of the test results and the source codes executed for
those tests. The extraction of the sample was dated
May 2017.
2) The generation of the input data table: The automatic
generation of a base table was programmed,
which would be used to study each of the suitable
parameters for predicting GTI costs. This phase will
be described in detail in the section ''Generation of
Input Data Table. "
3) The study of the influence of the parameters on the
different costs: It was necessary to study and validate
mathematically which parameters, of all the
available and potentially suitable ones, are necessary
for the realization and training of the prediction
models. In this way, from the totality of possible
parameters, an analysis of its correlation with the
output variables in the prediction was made, and the
necessary ones were selected to generate the predictions.
This phase will be described in the section
" Study of the Influence of the Parameters on the
Different Costs. "
4) The training ofpredictive models for costs: Once
the previous steps were completed, the models for
the prediction of costs and times in GTIs were
trained, generated, and validated. Finally, this phase
will be described in detail in the section " Training
of Predictive Models for Costs. "
Figure 2 shows a flowchart of the phases involved in
the design of the different predictive models.
GENERATION OF INPUT DATA TABLE
In this step, the extraction and generation of a table that condensed
and registered all the necessary parameters for the
24
Figure 2.
Phases in design process.
different predictions was programmed. For the generation of
the table, it was necessary to process the dump of the aircraft
simulation database as well as the extraction ofthe files ofthe
folder tree ofthe test system.
Among these parameters of the table would be both
the inputs (used to carry out the predictions) and the targets
(values to be predicted). The processing and generation
of the input data table, as well as later the generation
of the prediction models, were programmed with the IBM
Modeler 18 tool [10], which is one of the most used and
powerful tools for database processing as well as performing
a data mining process.
Initially, the generated table included, for each of the
GTIs performed to date, the set of parameters shown in
Table 1. These parameters were chosen according to the
information provided by Airbus DS about the main factors
influencing the GTI costs, and they were all potentially
suitable for the cost predictions. The parameters with the
RES_ prefix are relative to resources used to complete a
GTI. Together, they provide a pattern of the complexity of
that GTI. On the other hand, the parameters with prefix
COS_ are corresponding to costs generated in the completion
of the GTI.
As it is possible to see in the table, the incidences were
distinguished (COS_INCSA and COS_INCSW) according
to their two types [2]: Abortive incidences (incidences due
to serious failures during the course of the testing process,
which cause high extra costs to the company) and
IEEE A&E SYSTEMS MAGAZINE
NOVEMBER 2021
IEEE - Aerospace and Electronic Systems - November 2021

Table of Contents for the Digital Edition of IEEE - Aerospace and Electronic Systems - November 2021
