Flight safety improvements for small size unmanned aerial vehicles
D. Stojcsics
Óbuda University, Doctoral School of Applied Informatics, Budapest, Hungary stojcsics.daniel@nik.uni-obuda.hu,
Abstract—Unmanned aerial vehicles from micro to large scale are a common research topic in the latest years. Small size (under 3 m wingspan and between 1-15 kg all up weight) UAVs and the autopilots usually have no redundancy and have been not prepared for emergency situations. Loss of control, power management, engine or a faulty sensor could cause serious damage. To prevent the disaster many things must be considered and added to the UAV such as redundant manual and autonomous controller, fault detection and isolation subsystem (FDI) or bearing and position estimation. The AERObot autopilot was developed at Obuda University for research purposes using multiple flight safety improvement methods.
I. AUTONOMOUS – MANUAL MODE SWITCH During the test flights manual control is essential especially in early stages of tests. During the tuning of the basic controller channels (airspeed, altitude, bearing) mixed manual and autonomous control surfaces were used. A switching logic (actuator interface) is needed between the actuators and the receiver/autopilot to prevent the UAV from crashing.
A. Simple actuator interface
In manual control mode the actuators and the speed controller are driven by a human pilot via radio controller.
The human pilot can switch between full manual and full autonomous control mode (fig. 1). In the early stages of test flights the autopilot can drive the desired control surface only to setup the chosen control channel. With a simple actuator interface there is no possibility to use partial manual (heterogeneous) control.
B. Heterogeneous actuator interface
The heterogeneous actuator interface is an improved interface where direct remote control channel presets (DRCs) can be defined. The inputs of the interface are the manual and autopilot actuator signals, the auto/heterogeneous/manual switch from the RC receiver and the DRC preset from the autopilot (fig. 1).
Figure 1. Heterogeneous UAV setup
The DRC defines which control surfaces are driven by manual and autonomous mode. A DRC e.g. for the bearing controller is:
• Autonomous control: ailerons and rudder
• Manual control: throttle, elevator, flap
The interface should be independent from the autopilot so the UAV can also be controlled when a software error occurs. The AERObot is using a CPLD circuit for this.
C. Redundant actuator interface
Redundant autopilots in manned flight are essential but in unmanned flight it is not yet a standard. To improve flight safety a redundant autopilot should be integrated into the actuator interface. Not only the autopilot core (microcontroller unit) but the whole autopilot board with the power management and the sensor block (GPS, altimeter, airspeed meter, IMU, etc.) should be at least doubled as well as the RC receivers (using PCM failsafe channel).
In this scenario the first autopilot board is master the other is slave (fig. 2). They can detect the fault of their sensors, controllers with an observer. When a failure occurs in the master, the slave can take over control.
Figure 2. Redundant UAV setup with master-slave architecture The master-slave architecture is better than redundancy less autopilots, but has many weaknesses. Firstly, two equal systems are not enough to select the failed one, secondly the hierarchy based systems do not guarantee the faultless operation of the next system on the list.
Advanced redundant autopilot systems (fig. 3) such as the AERObot are using multiple equal complete autopilot units without hierarchy. The autopilots are connected to the actuator interface.
RC receivers
Autopilot
1. Autopilot
2.
Actuator interface
Actuators
RC receiver
Autopilot
Actuator interface
Figure 3. Advanced redundant UAV system with equal complete autopilot units
All autopilots are getting all received messages from the telemetry but only one can transmit and control the actuators. The autopilots have a PWM signal which tells the interface how good their qualities are (using the observer). The best autopilot (with the lowest error lever) can control the actuators and transmit using the RF modem.
II. EMERGENCY SITUATIONS
When the desired route is longer than the maximum range of the UAV or there is a problem with the propulsion system, the engine power will be reduced or ceased. The crash of the plane is avoidable even in this situation. The airspeed controller of AERObot autopilot is controlling the elevator rather than the throttle (common solution [1]) using the potential energy of the UAV. With this solution the desired airspeed can be kept even without propulsion (with continuously decreasing altitude).
Figures 4 and 5 represents a test flight where the electric propulsion UAV batteries depleted after 9 minutes of flight. Because the batteries were Lithium-polymer batteries, the final discharge session is short; the thrust decreasing period is quick. After this point the electric engine will stop soon. From the beginning of the thrust fall to the final depletion there is around 160 seconds left.
After the depletion the autopilot keeps the desired airspeed while the UAV is driven in a glide slope down to the ground on the desired path. With this solution a controlled emergency landing can be performed without crashing the plane.
III. FAULT TOLERANT CONTROL REALLOCATION Most of the actuators used in small size UAVs are standard model grade RC servos. Even the high quality digital, metal gear servos can easily be damaged during the flight (especially at high speed or at landing). When this happens the UAV will most likely crash.
Loss of a throttle, rudder (vertical control) or one aileron (longitudinal control) is not as crucial as the elevator (lateral control). General planes usually have only one elevator control surface. Without it the autopilot cannot control the airspeed and altitude, unless it is a tailless flying wing. The tailless planes have usually two control surfaces, the elevons only. These control surfaces are the combination of the elevator and the aileron, hence
Figure 4. Airspeed control with depleted engine battery
Figure 5. Test flight path with depleted main battery There are many ways to create a fault tolerant control such as multiple model switching and tuning, model reference adaptive control or control reallocation [2].
The basic idea for the new fault tolerant control reallocation comes from a Hungarian experimental glider, the KM-400 (fig. 6). It was a general-like glider with a
“T” tail with rudder, but without elevator. The lateral control was made by the elevons and flaps on the main wing [3][4].
Since it was a controllable plane, the faulted elevator can be replaced by the ailerons using them as an elevon in the case of a small size UAV.
The RC model servos are commanded through analog PWM signals. It has an internal MCU and a feedback potentiometer. An easy implementation of a fault detection supervision module, a passive FDI subsystem, can be created using a loopback from a standalone potentiometer per control surface to the autopilot. Passive FDI systems will wait until the failure occurs.
Figure 6. KM-400 experimental glider designed by M. Kesselyák
Hardware In the Loop simulations were made to examine the discussed methods in Matlab. A Simulink model with FDI subsystem has been created using AeroSim block set.
The simulated flight was made in a continuous flight path with 4 waypoints (fig. 7). At t=570s, the elevator failed and the autopilot reallocated the controls. The new elevator deflection command became 0 and aileron became an elevon.
The elevon on a tailless plane works the same as an elevator on the lateral axis, so the control parameters were not modified after the error detection and control reallocation. Since the forces on the elevon and the size of the control surface are different from the original elevator, the measured values (altitude and airspeed) are worse than before the failure. However the plane maintained level flight and the flight path was not influenced after the reallocation.
With further research the implemented passive FDI subsystem can be improved to an active FDI with error prediction. In this case the actuator failure (locked-in- place, hard-over or loss of effectiveness) can be prevented by setting the actuator to neutral trim position and reallocating the control.
Figure 7. Elevator failure at t=570s on a continuous 4 waypoint flight route
IV. TAKE-OFF AND LANDING
The take-off and landing are the most dangerous parts of the unmanned flight. In the case of small-size UAVs it is performed from a flat ground, not from a solid airfield, so there is no physical limitation to the take-off and landing position or direction. The only limit is the direction of the local airflow because both maneuvers should be performed against the wind.
The take-off can be performed as a classic take-off from landing gear, a hand launch like model airplanes, from belly skid with a catapult or bungee to accelerate the plane. The take-off is always done with maximum thrust.
Our UAVs with maximum thrust are climbing intensively so there is no need of sophisticated take-off algorithms [5][6] for that. The AERObot autopilot is using simple controllers, which compensates the pitch, roll and yaw angles against the airflows, without letting the airspeed reduced under the stall speed. The roll angle is around 0°, the yaw angle is the take-off direction and the pitch angle is around the climbing angle (<30°). When the UAV reaches the secure H=100m altitude it switches to navigation flight mode (fig. 8).
There are also many ways to land a small-size UAV [7][8]. In case of a flat ground it can land to belly skid or to landing gears. Otherwise it can use a parachute or land into a net. The local airflow direction and speed should be considered in all types of landing (always performed against the wind so there is no need for high airflow compensation during the maneuver). In case of belly landing or landing gears, a position has to be marked where the plane should land. The direction is defined by the local airflow. While taking the last waypoint the UAV turns to the direction of that point. After this from a specified distance (depends on the natural glide angle of the aircraft and the wind speed) the UAV performs the glide slope maneuver with the landing speed and idle throttle (the flight controllers are the same as used for navigated level flight).
The final stage of the landing is the flare maneuver when the altitude reaches 20m. In the flare the nose of the plane is raised, slowing the descent rate for touchdown by the autopilot.
Figure 8. Simulated take off and landing
Altitude
Latitude Longitude
elevator failure Waypoint turns
runway
V. BEARING ESTIMATION
The base sensor for the navigation is the GPS. It provides the position, speed over ground (SOG), bearing and many other parameters used by the navigation algorithms. The maximum position refresh rate of best GPS modules is about 5-10Hz, which is not enough for precision flight control.
The other problem is the refresh rate is not reliable because NMEA sentences and the checksums are often incorrect. It is necessary to estimate the missing, and intermediate positions for the continuous bearing control (100Hz – update frequency of the most actuators). This estimation can be computed using the yaw rate and the GPS bearing from the last valid NMEA sentence. The Z- axis angular rate (provided by the IMU) should be converted to heading turn rate using the UAV bank angle.
With this procedure the actual bearing can be estimated between two valid GPS sentences (1). The heading angle provided by the IMU is often not reliable, because of the magnetic sensor used by the internal sensor fusion algorithms. This sensor is very sensitive to strong electromagnetic fields especially when the UAV has electric propulsion.
(1) where:
θ: previous bearing (measured by GPS) θe: estimated bearing
: heading turn rate
Small size (around 1 m wingspan) and small weight (under 2 kg) UAVs are quite agile so turn rates can reach high levels. For these airframes bearing estimation is essential (fig 9.).
Figure 9. Xeno UAV flight test, purle triangle indicates the estimated bearing
VI. POSITION ESTIMATION
During the navigation, not only the bearing but also the position has to be estimated with a navigation formula (2), well known is nautical terms [9]. Unlike a complete inertial navigation system with minimal error [10][11][12]
this system is using the GPS module as the main position sensor but makes it better.
The estimation (Fig. 10) can be calculated from the estimated SOG, estimated position and bearing (when known position and bearing are not available).
asin sin cos
cos sin cos
2 sin sin
cos , cos sin
sin
where: (2)
Lat1: previous estimated latitude of the UAV Lon1: previous estimated longitude of the UAV
: estimated bearing (in radians, clockwise from north);
: estimated travelled distance R: Earth radius
The travelled distance can be calculated from the estimated bearing and the estimated SOG which correlates with the local airflow [13].
Usually there are 200ms between two valid GPS sentence burst, but the distribution of measured positions are not uniform. Even if one or two is corrupted the refresh time is always under 1000ms.
Obviously there will be some errors, but usually it is low (under 1 m) and not causing any serious false position estimation. When there is no position fix due to a GPS error the bearing and position estimation method acts like an inertial navigation unit and the UAV can safely return home.
Figure 10. Position estimation simulation from telemetry data (downsampled, big dots are the measured, small dots are the estimated
positions) VII. SUMMARY
Usage of an advanced redundant autopilot system such as the AERObot (with multiple equal, complete autopilot units and heterogeneous actuator interface) can improve the flight safety. In manual flight the redundant receivers with different antenna allocations can prevent signal losses, while in autonomous mode the equal autopilot units can prevent software, electronic and sensor based
When the autopilot is controlling the airspeed with the elevator rather than the throttle, it can control the UAV satisfactorily, even with depleted engine battery like a glider as the test flights showed.
Control surface failures can also be handled with a passive FDI, so the UAV remains controllable. Even GPS failure is avoidable if the autopilot has bearing and position estimation algorithms.
With the discussed methods flight safety can be improved for small size unmanned aerial vehicles as the simulations and the test flights have shown.
ACKNOWLEDGMENT
The author gratefully acknowledges the grant provided by the project TÁMOP-4.2.2/B-10/1-2010-0020, Support of the scientific training, workshops, and establish talent management system at the Óbuda University.
REFERENCES
[1] S. Leven, J. Zufferey, D. Floreano, "A minimalist control strategy for small UAVs", in Proc. IROS, 2009, pp. 2873-2878.
[2] G. J. J. Ducard “Fault-tolerant Flight Control and Guidance Systems Practical Methods for Small Unmanned Aerial Vehicles”, Springer-Verlag London Limited, 2009, ISBN 978-1-84882-560-4 [3] G. Jereb “Magyar vitorlázó repülőgépek”, Műszaki Könyvkiadó,
Budapest, 1988, ISBN: 963 10 7126 X
[4] M. Kesselyák.: “An Improved-Performance Control System for Low-Speed Flight” AIAA Paper No. 74-1039. Cambridge. Mass.
USA 1974.
[5] W. Rui, Z. Zhou, S. Yanhang: “Robust Landing Control and Simulation for Flying Wing UAV”, Proceedings of the 26th Chinese Control Conference, July 26-31, 2007, Zhangjiajie, Hunan, China, ISBN: 978-7-81124-055-9, pp. 600 - 604
[6] A. Cho et al.: “Fully Automatic Taxiing, Takeoff and Landing of a UAV using a Single-Antenna GPS Receiver only”, International Conference on Control, Automation and Systems 2007, Oct. 17- 20, 2007 in COEX, Seoul, Korea, ISBN: 978-89-950038-6-2, pp.
821 - 825
[7] A. Molnár, D. Stojcsics: “Fixed-wing small-size UAV navigation methods with HIL simulation for AERObot autopilot”, 9th International Symposium on Intelligent Systems and Informatics, ISBN: 978-1-4577-1975-2, Subotica, Serbia, 8-10 Sept. 2011, pp.
241 - 245
[8] S. Kurnaz, O. Çetin: “Autonomous Navigation and Landing Tasks for Fixed Wing Small Unmanned Aerial Vehicles”, Acta Polytechnica Hungarica Vol. 7, No. 1, 2010, pp. 87-10
[9] F. Ucan, D.T. Altilar: “Navigation and Guidance Planning for Air Vehicles”, 20th IEEE International Conference on Tools with Artificial Intelligence, ICTAI 2008, vol.2, no., pp.534-538 [10] Q. Li, Z. Fang, H. Li: “The Application of Integrated GPS and
Dead Reckoning Positioning” in Automotive Intelligent Navigation System, Journal of Global Positioning Systems (2004) Vol. 3, No. 1-2: 183-190
[11] P. Davidson, J. Hautamäki, J. Collin: “Using Low-Cost MEMS 3D Accelerometers and One Gyro to Assist GPS Based Car Navigation System”, Proceedings of 15th Saint Petersburg International Conference on Integrated Navigation Systems, May 2008
[12] L. Zhao, W. Y. Ochieng, M. A. Quddus and R. B. Noland: “An Extended Kalman Filter algorithm for Integrating GPS and low- cost Dead reckoning system data for vehicle performance and emissions monitoring”, Journal of Navigation, 2003, 56: 257-275 [13] A. Molnár, D. Stojcsics: “New approach of the navigation control
of small size UAVs”, Proceedings of 19th International Workshop on Robotics in Alpe-Adria-Danube Region, IEEE Catalog Number: CFP1075J-CDR, ISBN: 978-1-4244-6884-3, Budapest, Hungary, 2010, pp. 125-129
.
