Study of pilot's control actions personality during landing based on neural network models

Аuthors

Evdokimenkov V. N.^1*, Kim R. V.^2**, Vekshina A. B.^3***, Yakimenko V. A.^4****

1. ,
2. ,
3. Moscow Aviation Institute (National Research University), 4, Volokolamskoe shosse, Moscow, А-80, GSP-3, 125993, Russia
4. ,

*e-mail: evdokimenkovvn@mai.ru; vnevdokimenkov@gmail.com
**e-mail: romanvkim@yandex.ru
***e-mail: mopy_ann@mail.ru
****e-mail: whyacehka@gmail.com

Abstract

The paper considers an approach to developing pilot's control actions models during execution of typical flight modes. Implementation of such kind of models for the structures of pilot's backup systems will allow providing the possibility of parallel solution of task complex, aimed at pursuing an objective of a concrete flight mode.

The results of landing trajectories processing are reported. These results indicate that both aircraft trajectory parameters distribution and control parameters distribution during landing performed by various pilot's reveal statistic significant differences. This fact allows us to say that efficiency of upgrading of pilots actions support systems requires provision of possibility to adapt implemented models of pilots control actions, taking into account experience, qualification and peculiarities of control actions of a certain pilot.

We suggest an approach to forming individually adapted models of pilot's control actions based on neural network of a multilayer perceptron type. Such kind of model uses actual parameters of aircraft-pilot system status as input variables, including dynamic and control parameters. The output of neural network model considered in the paper is scalar indicator function, representing the convolution of parameters, characterizing the accuracy of aircraft touchdown to a runway. Parameters of the neural network model are determined as a result of neural network «training» by the data obtained during execution of previous flights. Thus, the suggested model allows predict the accuracy of bringing an aircraft to a runway based on current values characterizing aircraft-pilot system status.

The paper gives the results of pilot's control actions model building-up using the data obtained in the course of landing modes execution by two operators working at hardware-software MiG-AT aircraft simulator. Based on the obtained results we can draw the following conclusions. Adaptation of neural network model of a more experienced pilot is achieved only by updating weight factors of neurons with retention of its structure. With a primary pilot who does not demonstrate aircraft handling stable skill, adaptation of neural network model in the course of his professional activity is provided by changing both the neural network model structure and its parameters. Thus, the obtained results allow claim that individual manner of actions of a concrete pilot reveals either in the structure, or in parameters of neural network model, characterizing him.

Keywords:

artificial neural network, pilots activities model, pilot's decision aid system

References

1. Bazlev D.A., Evdokimenkov V.N, Kim N.V, Krasilshchikov M.N. Vestnik kompyuternykh i informatsionnykh tekhnologii, 2007, no.1, pp. 10-25.

2. Bazlev D.A., Evdokimenkov V.N, Krasilshchikov M.N. Izvestiya RAN. Teoriya i sistemy upravleniya, 2008, no.4, pp. 97-108.

3. Krasilshchikov M.N., Evdokimenkov V.N., Bazlev D.A. Individualno-adaptirovannye bortovye sistemy kontrolya tekhnicheskogo sostoyaniya samoleta i podderzhki upravlyayushchikh deistvii letchika. (Individually-adapted onboard systems of control of technical condition of aircraft and support for pilots control actions), Moscow, MAI, 2011, 438 p.

4. Zavalova N.D., Lomov B.F., Ponomarenko V.A. Obraz v sisteme psikhicheskoi regulacii deyatelnosti (Image in psychical regulation system of activity), Moscow, Nauka, 1986, 174 p.

5. Zavalova N.D., Lomov B.F., Ponomarenko V.A. Voprosy Psikhologii, 1971, no 3, pp. 3-5.

6. Draper N., Smith H. Applied regression analysis (Wiley Series in Probability and Statistics). 3rd edition. NJ, Wiley — Interscience, 1998, 736 p.

7. Haykin S. Neural Networks: A Comprehensive Foundation. 2nd edition. NJ, Prentice Hall, 1998, 842 p.

8. Wasserman P. Neural Computing: Theory and Practice. Scottsdale, Coriolis Group (Sd), 1989, 230 p.