#	Pagina
attuale pagina	/open-h2020/projects/221570/results.html

Report

Teaser, summary, work performed and final results

Periodic Reporting for period 1 - FIVER (Innovative solution for FMS computed trajectories validation by means of pilot actions emulation, comparison with PANS-OPS criteria and data mining techniques.)

Teaser

Summary

\"On 21 November 2018 Scalian, CGX AERO and Thales launched the FIVER project, funded by the European CleanSky 2 Joint Undertaking program.

Their main objective: meet Thales\' need to strengthen the robustness of its Flight Management System (FMS) in order to limit the environmental impact of aeronautics.
The FMS is a specialized computer system that automates a wide variety of in-flight tasks, reducing the workload on the flight crew. A primary function is in-flight management of the flight plan. Using various sensors (such as GPS and INS often backed up by radio navigation) to determine the aircraft\'s position, the FMS can guide the aircraft along the flight plan.
The overall performance of the FMS, its ability to adapt to a wide variety of conditions (including the diversity of flight profile) is therefore decisive for flight safety, but also for its environmental impact. We aim to allow THALES\'s FMS to optimize flight paths and therefore fuel consumption, while continuing to ensure the safety of the flight and its crew.

The 24-month FIVER project, for \"\"FMS Innovative Validation for Enhanced Robustness\"\", will enable Thales to improve the maturity of its Flight Management System, through innovative massive testing methods based on simulation and artificial intelligence.

To achieve this objective, CGX AERO and SCALIAN work closely together, and continuously involve the expertise provided by THALES. The high-level distribution of the overall objectives within FIVER is as follows:
THALES provides in particular the FMS and its regular updates, the analyses developed internally in the past, the elements necessary to launch a simulation, and the environment on which the solution will be implemented.
CGX AERO must notably create the simulation solution that will allow the FMS to be subjected to massive testing by proposing realistic flight data (avionics data, human behaviour), develop virtual pilot models that meet specific profiles, and produce an analysis tool based on business rules and procedure design criteria, analyses transmitted by THALES, and statistical gaps.
SCALIAN\'s mandate is to develop a Data Science Studio, a virtual environment capable of analyzing the data produced by simulation, using machine or deep learning methods. This environment must be oriented towards business users and offer a certain understanding of model outputs, so that test engineers in particular can interpret the anomalies detected by the models.\"

Work performed

As at 31 October 2019, half of the time dedicated to the FIVER project, the partners are preparing to implement a first demonstrator of the global solution in the THALES environment, scheduled for mid-December.
To achieve this, the two partners have worked closely with THALES and have already proposed the following elements:
CGX AERO has developed three main components: the Test Campaign Generator module, the Virtual Crew module (including two of the 3 Virtual Pilot profiles), and the Flight Analyzer module with continuous improvement (including the upcoming integration of analyses based on reference flights). These components integrated into the data production chain of THALES (which includes the campaign runner and the aircraft manager) form the simulation production line to be distributed on multiple environments and allowing the generation of a high volume of various data to be analyzed. The Test Campaign Generator by allowing selection of parameters to be explored frames the objectives of the tests (non-regression or outstanding trajectoriesâ€¦). The TCG generates scenario using flight plan generation with pilot model tuning and analysis tool selection. The Virtual Crew Module is supporting multi-actor interactions and allows playing the defined scenario including bias through (among others) the Virtual Pilot profile leading to representative flight simulations. In parallel, during simulation and after trajectory data collection, the Flight Analyzer, built as an analysis tool box is able to perform the analysis planned during the test campaign definition. The results are then sent to the Data Studio with SFTP protocol for result gathering and big data analysis. The Flight Analyzer supports multiple type of analysis using existing THALES analysis tool, comparison of flown data with predicted data and procedure design criteria (through protection area construction) with provided CGX AERO analysis tools and reference database.
SCALIAN has set up a working environment based on the containerization of micro-services. These micro-services include a data ingestion service (from the simulation output storage database) which will soon evolve, and a microanalysis service dedicated to filling one of the major functionalities by which THALES wishes to test its FMS. At this stage, the algorithms dedicated to one of these functionalities have been developed and are currently being tested. User interfaces to navigate through the simulation outputs, and analysis reports have been designed and are in production.
The results obtained for each partner are encouraging and call for continued efforts. Indeed, the simulation module manages to simulate some complete flights, from takeoff to landing, and must now expand the range of its flights, as well as the conditions that can be instantiated (type of pilot, aircraft, conditions, etc.).
The detection of anomalies by various methods has shown interesting results but which require complete validation by the business and above all contextualization by avionic and control data. The rest of the project will be strongly dedicated to this functionality.

Final results

The detection of anomalies in complex systems is historically based on the limitations imposed by a rule system. By complementing such systems with a priori free research, based on unsupervised learning algorithms, the FIVER project plans to achieve better anomaly detection performance than conventional methods.
As the first phase of the project was essentially dedicated to the development of functional bricks, it will be up to the second part of the project to define the performance of the global approach proposed in FIVER.
Intermediate results stand in the first production line built with a Virtual Pilot able to simulate a flight using a Perfect or Classical Pilot Model. The data production line allows to start generating massive data when deployed into an adequate environment.