Stimulation Initiative for European Neural Applications
Esprit Project 9811
Helicopter Flight Data Analysis (HFDA)
Company background
Eurocopter is a European company, manufacturing both civilian
and military helicopters. The company is a joint venture between
French Aerospatiale and German MBB. The application has been
designed in 1990 for Aerospatiale Helicopters which
become later Eurocopter in Marseilles.
The problem
Flight recorders provide a huge amount of data which are seldomly
used by the helicopter manufacturers. Most of the time, flight recorder
are used only in case of crash or disfunction of the helicopter.
On the other hand, structural stress of the helicopter on the miscellaneous
organs is derivated from the flight conditions.
Bad flight conditions will lead to a higher stress on the blades and
the body while quiet flight will less agress these part.
The reason for which the recorded flight parameter are not compiled
is because the computer interpretation was not adequat due to highly
nonlinear cross correlation between different signals.
Neural network application
The HFDA has been developped using the NLS (Nestor Learning System)
a software solution. Basic function of HFDA includes viewing signals
like vertical acceleration, altitude, engines RPM and torque, horizontal
speed. Then, signal are selected by the operator and, initially, flight
phases (about 40 differents) are delimited by the operator and the
neural network is teached with the flight category (take off for
example) to which the set of signal belong, for a given time period.
When the HFDA has been trained, the system can start to recognize flight phasis.
HFDA feature dynamic category addition and incremental learning.
Benefits
HFDA analyses one hour of flight data in about 5 minutes on a 486 PC platform
while a high level human expert will spend few hours for the same tedious task.
The correlation with a human expert is close to 90%. The system quickly detects
strange flight configurations and high stress configurations.
Generalization
This application can be generalized to virtually any kind of data analysis like
aircraft flight recorder analysis, truck analysis, locomotive analysis and
industrial process analysis. The availability of hardware neural
networks chips with real time learning capability allows to design low
cost condition monitoring systems to prevent failure.
Contact person
Mr. Carlier, Eurocopter Aeroport International 13700 Marignane.
Fax +33 42 85 86 05