Microelectromechanicheskie sensors - based Compact Strapdown Inertial Navigation System
Control and navigation systems
Аuthors1*, 1**, 2***, 3****
1. Moscow Aviation Institute (National Research University), 4, Volokolamskoe shosse, Moscow, А-80, GSP-3, 125993, Russia
2. Siberian Aeronautical Research Institute named after S.A. Chaplygin, 21, Polzunov St., Novosibirsk, 630051, Russia
3. Main research and testing center for robotics Ministry of Defense of the Russian , 5, ul. Seregina, Moscow, 125167, Russia
Approaches to realization of the neural network (NN) regulator in the automatic control system (ACS) of an unmanned aerial vehicle (UAV) are studied in the paper. UAV flight in the approach and landing modes affected by wind disturbance in the vertical plane is considered. The structure of the NN and modeling of NN regulator training results are presented.
The UAV motion in the longitudinal plane is considered on the assumption of indicator speed stabilization by means of auto-throttle. At a stage of final approach maneuvering the preselected flight altitude is maintained, after crossing the glide path the UAV ACS eliminates the current height mismatch and provides its transition to a mode of holding of desired track coinciding with the direction of the glide path. The appropriate equations of UAV motion are presented in . The UAV control in the vertical plane is realized as follows: under big deviations from the preselected altitude of flight when descending at the preset constant airspeed stabilization the control signal (1) on the elevator drive arrives:
After the deviation from the preselected altitude of flight becomes less than 10 m, the control law switches to the following one:
where — coefficients of the control law, — increments of UAV coordinates, increments of estimates of the UAV pitch and angle of attack caused by vertical wind derived as a result of estimation by means of the optimum Kalman filter . The first control law (1) provides maximum system response speed when eliminating altitude mismatch, and the second control law (2) provides accuracy in maintenance of the desired track.
Transitions from one control law to another can cause loss of UAV stability. To prevent such occasion the presented regulators were realized on the neural network base which will provide the specified system response speed and continuity of control signal feeding the elevator drive, excluding thereby probability of UAV stability loss.
For creation of training data access there were used files of coordinate data included in control laws which are obtained by modeling of operation of the regulators mentioned above under various initial data: initial deviation from the desired altitude ΔH and speed of vertical wind Δαw. The following meanings of initial data were taken: for ΔH=0 m, 50 m, 100 m, 150 m, 200 m, and for Δαw=0 m/s, ±1 m/s, ±2 m/s, ±4 m/s. The NN output control signal feeds the UAV elevator drive.
The program of formation, training and testing of NN operation was developed. Results of modeling of NN training for initial deviation from the desired altitude of ΔH=50 m, and a vertical wind speed of Δαw =-1 m/s were analyzed in the paper. It was shown that on the 1000th epoch of NN training the value of a root mean square error amounted to 4.17 • 10-4 .
Thus, NN solution of the control system developed reproduces a command signal with the adequate accuracy.
Keywords:unmanned aerial vehicle, automatic control system, neural network
- Bitter V. V., Mikhailin D. A. Informatsionno-izmeritelnye i upravlyayushchie sistemy, 2009, no. 7, pp.3-14.
- Lebedev G. N., Efimov A.V., Mikhailin D. A. Vestnik Moskovskogo aviatsionnogo institute, 2012, vol. 19, no.1, pp.12-16.