Learning to Stall: Using in-air pressure data to identify, characterise and control fixed-wing aircraft stall

Abstract

Fixed-wing unmanned aerial vehicles (UAV) are a highly efficient element of the UAV-family. Mainly, due to their need for large and obstacle free clearings for save launch and landing, they are not as popular as one may expect. A much wider range of applications is achieved by using deep-stall landings for fixed-wing UAVs, reducing the needed space massively. This is possible because in deep-stall the drag gets increased massively whereas the lift decreases. Therefore, much steeper flightpath angles are possible, while keeping the horizontal and vertical speed at a sufficiently low level. Deep-stall landing have several advantages compared to other landing approaches: No additional equipment is needed as necessary for parachute or net recovery. The range doesn’t get reduced as it is the case for vertical take-off or landing (VTOL) vehicles. In a previous project ([1]) successful deep-stall landings were performed. However, for precise deep-stall landings a more in depth understanding of the flow field over the wings is needed. Therefore, a new sensing approach is proposed, using 24 differential pressure sensors comparing the pressure above and below the wing at a range of span- and chordwise positions. To keep it as practical as possible, those sensors are mounted within the wing, without the need of any external tubing. This is achieved by developing an airfoil shaped printed circuit board (PCB) accommodating the sensors. Additionally, the PCB provides all necessary peripherals to assure a reliable long range I2C communication. The PCBs are contained by specially developed 3D-printed airfoil-sections, already containing the channels between pressure taps and sensors. Data is collected in a large wind tunnel and with real flight tests. The angle of attack is determined over the whole range of flight states by only using the pressure data and a neural network. This approach gets simplified to only a few sensors and conventional fitting techniques to allow for more practical approaches. The possibility of estimating the angle of slip as well is shown. However, it is much more challenging to simplify the used sensors and evaluation techniques. A new neural network based detector to determine the local flight state is derived, predicting if the flow is attached, separated or in transition. This is opening a wide field of applications by preventing or exploiting local separation for improved flight safety or more precise deep-stall landings. Moreover, the sensor-set detected limits in the calibration processes used in previous works [1] and it shows phenomena as for example the stall hysteresis and local flow separation over the wing. The collected data can be used for future research, paving the way to exact, autonomous deep-stall landings. Furthermore, this work’s findings are not just limited on deepstall manoeuvres. These findings can be applied on other high angle of attack flight states occurring during hovering, perching and some VTOL transitions.