Experimental Wind Field Estimation Gautier Hattenberger Joint work - PowerPoint PPT Presentation

Experimental Wind Field Estimation Gautier Hattenberger Joint work with : Jean-Philippe Condomines and Murat Bronz ENAC UAV Lab, French Civil Aviation University, France gautier.hattenberger@enac.fr ISARRA2016 – Toulouse – mai 2016 Hattenberger et al. (ENAC) Wind Estimation ISARRA2016, Toulouse 1 / 17

Outline Introduction 1 Wind Field Estimation 2 Improvements with Aircraft Model 3 Conclusion and Future Work 4 Hattenberger et al. (ENAC) Wind Estimation ISARRA2016, Toulouse 2 / 17

SkyScanner Project Founded by STAE Foundation, outcome from the Micro Air Vehicle Research Center of Toulouse https://www.laas.fr/projects/skyscanner http://websites.isae.fr/mav-research-center Study and experimentation of a fleet of mini-drones that coordinate to adaptively sample cumulus-type clouds refine aerological models of clouds conceive enduring and agile micro-drones fleet control and trajectory optimization Hattenberger et al. (ENAC) Wind Estimation ISARRA2016, Toulouse 3 / 17

Objectives Within the global scope of the project, some particular objectives: Aircraft performances identification aerodynamic and propulsion performances are required for aircraft control and trajectory planning Wind field estimation real-time estimation of the local wind field for the atmospheric studies but for the trajectory planning as well Hattenberger et al. (ENAC) Wind Estimation ISARRA2016, Toulouse 4 / 17

Outline Introduction 1 Wind Field Estimation 2 Improvements with Aircraft Model 3 Conclusion and Future Work 4 Hattenberger et al. (ENAC) Wind Estimation ISARRA2016, Toulouse 5 / 17

Wind Field Principles Based on the velocity triangle W = ⃗ ⃗ V wind ⃗ v pitot O α β X s ⃗ V ground ⃗ V norm ⃗ V norm Y s Body frame Velocity triangle Z s V g = � � V n + � W Hattenberger et al. (ENAC) Wind Estimation ISARRA2016, Toulouse 6 / 17

Measurement Issues Problem A direct measure of the wind is not possible Hattenberger et al. (ENAC) Wind Estimation ISARRA2016, Toulouse 7 / 17

Measurement Issues Problem A direct measure of the wind is not possible Full airspeed measurement 3D airspeed sensors compatible with mini-UAVs are available but can be fragile and expensive Hattenberger et al. (ENAC) Wind Estimation ISARRA2016, Toulouse 7 / 17

Measurement Issues Problem A direct measure of the wind is not possible Full airspeed measurement 3D airspeed sensors compatible with mini-UAVs are available but can be fragile and expensive No flow sensor Without airspeed measurement, wind-field can still be estimated from GPS/IMU data, but needs special trajectories: flying in circle for the horizontal components gliding for the vertical component Hattenberger et al. (ENAC) Wind Estimation ISARRA2016, Toulouse 7 / 17

Measurement Issues Problem A direct measure of the wind is not possible Full airspeed measurement 3D airspeed sensors compatible with mini-UAVs are available but can be fragile and expensive No flow sensor Without airspeed measurement, wind-field can still be estimated from GPS/IMU data, but needs special trajectories: flying in circle for the horizontal components gliding for the vertical component This solution gives too much constraints on the trajectories, especially the glides Hattenberger et al. (ENAC) Wind Estimation ISARRA2016, Toulouse 7 / 17

Wind Estimation Low-cost sensors solution Wind estimation is done with a non-linear Unscented Kalman Filter (UKF) by fusing at least: GPS velocities accelerometers, gyrometers and magnetometers airspeed norm from Pitot tube Hattenberger et al. (ENAC) Wind Estimation ISARRA2016, Toulouse 8 / 17

Wind Estimation Low-cost sensors solution Wind estimation is done with a non-linear Unscented Kalman Filter (UKF) by fusing at least: GPS velocities accelerometers, gyrometers and magnetometers airspeed norm from Pitot tube Improvements Add an extra angle-of-attack probe in order to improve the estimation of the vertical component  | < v , e 1 > |   y v  q m ∗ v ∗ q − 1 + ν b � v = v × ω m + q − 1 y V m ˙ ∗ A ∗ q m + a m (evolution) m   q − 1 ( S )    = (measurement) ∗ B ∗ q m   y B m ν b = 0 ˙  tan − 1 � < v , e 3 >   � y α < v , e 1 > Hattenberger et al. (ENAC) Wind Estimation ISARRA2016, Toulouse 8 / 17

Wind Estimation Results Estimation of an updraft during a gliding phase Hattenberger et al. (ENAC) Wind Estimation ISARRA2016, Toulouse 9 / 17

Aircraft Instrumentation Aircraft integration commercially available airframe (Mako) GPS, IMU and barometer for position and attitude estimation integration of a Pitot tube and an angle of attack probe on-board data logging on SD card controlled using Paparazzi UAV system http://paparazziuav.org AirSpeed Angle of Attack Sensor Hattenberger et al. (ENAC) Wind Estimation ISARRA2016, Toulouse 10 / 17

Angle of Attack Sensor Homemade with an absolute angular sensor (US Digital, 12 bits resolution, hall effect) and 3D-printed flag Calibration in wind tunnel is required to compensate interaction with the fuselage Hattenberger et al. (ENAC) Wind Estimation ISARRA2016, Toulouse 11 / 17

Angle of Attack Sensor Homemade with an absolute angular sensor (US Digital, 12 bits resolution, hall effect) and 3D-printed flag Calibration in wind tunnel is required to compensate interaction with the fuselage In-flight comparison with a 5-holes probe (Aeroprobe) (in red) variations are coherent but there are an offset and a scaling error Hattenberger et al. (ENAC) Wind Estimation ISARRA2016, Toulouse 11 / 17

Outline Introduction 1 Wind Field Estimation 2 Improvements with Aircraft Model 3 Conclusion and Future Work 4 Hattenberger et al. (ENAC) Wind Estimation ISARRA2016, Toulouse 12 / 17

Improving the Wind Estimation Estimation can be improved by knowing: the aircraft aerodynamic model the propulsion set model Hattenberger et al. (ENAC) Wind Estimation ISARRA2016, Toulouse 13 / 17

Improving the Wind Estimation Estimation can be improved by knowing: the aircraft aerodynamic model the propulsion set model Lift coefficient results 0.8 Angle-of-attack sensor is gliding phase level flight also used for aircraft 0.7 curve fit identification 0.6 0.5 Extracted points comes CL from gliding phases (in 0.4 red) and from level 0.3 cruise flights (in green) 0.2 0.1 -4 -2 0 2 4 6 8 10 12 alpha Hattenberger et al. (ENAC) Wind Estimation ISARRA2016, Toulouse 13 / 17

Motor Test Bench Build an accurate model of the propulsion system Automated measurement procedure in wind tunnel Torque Sensor Thrust Sensor Attached piece to the main shaft V ∞ Thin Wall Bearings Hattenberger et al. (ENAC) Wind Estimation ISARRA2016, Toulouse 14 / 17

Motor Analyses 12 100 0 m/s 10 m/s 5 m/s 13 m/s 10 80 10 m/s 15 m/s 8 13 m/s 15 m/s 60 Aero power [W] 18 m/s 6 Thrust [N] 22 m/s 4 40 2 20 0 0 -2 -4 -20 0 1000 2000 3000 4000 5000 6000 7000 8000 -50 0 50 100 150 200 250 RPM [rev/min] Electrical power [W] In the useful range of airspeed, linear relation between the electrical power input and the resulting propulsive power Hattenberger et al. (ENAC) Wind Estimation ISARRA2016, Toulouse 15 / 17

Outline Introduction 1 Wind Field Estimation 2 Improvements with Aircraft Model 3 Conclusion and Future Work 4 Hattenberger et al. (ENAC) Wind Estimation ISARRA2016, Toulouse 16 / 17

Conclusion and Future Work Conclusion Development of a wind field estimation algorithm Evaluation on real flight data Integration of extra low-cost sensors Aircraft and motor identification Future Work Use the aircraft and propulsion data in the wind estimation filter Integrate the filter on an on-board computer for real-time processing Hattenberger et al. (ENAC) Wind Estimation ISARRA2016, Toulouse 17 / 17