April 28, 2017

The Technology of Airborne Wind Energy – Part II: the Drone

What is the status of the AP-3 and AP-4 development, and how does your technology actually work? In a short series of three articles Michiel Kruijff covers the Frequently Asked Questions for all aspects of our solution.

By: Michiel Kruijff, Product Development Ampyx Power


One of the most striking features of the AP-3 drone is its double fuselage. We have a patented concept in which we control the aircraft on the tether much like a conventional aircraft using control surfaces. The tether is attached midwing, close to the center of mass. The double fuselage provides the necessary clearance for the tether during maneuvers, such as the climb and the landing. An added benefit is that we can divide the propulsion needed for climb after launch over two fuselages. Both propellers are needed for tethered climb, but with a single propeller we can return to base in any condition. Even if the tether would break or disconnect for some reason. The propulsion runs on batteries. They are sized for a single climb and to assist an emergency landing. It is not active during power generation or during nominal (tethered) landing.

The wing is optimized for power output, it is a high lift design that generates an immense lift.

The wing is optimized for power output, it is a high lift design that generates an immense lift equal to twelve times the mass of the aircraft (4200 kg force). During power generation, we operate at an angle of attack (angle between wing chord and apparent wind) close to the maximum allowable (stall limit). The wing profile design provides smooth stall characteristics, and thus aerodynamic safety. Most of the load is carried by a carbon spar. The wing skin is made of fiber-glass and is not load-carrying. The wing can thus sustain significant impact damage, even puncture, without losing its overall integrity (think e.g. of impact of heavy hail or of a small bird).

Our airfoil is designed for maximum power production using an adjoint optimization approach in CFD which we implemented in OpenFoam.


All load carrying structure is designed for infinite lifetime, it is designed to sustain fatigue loading even following superficial damage. The AP-4 drone is designed to land safely after most bird impact and lightning strikes.


The AP-3 horizontal tail is set between the fuselages and is an all-moving tail. This maximizes the control authority and secures sufficient pitch control during landing (nose down/nose up direction), even if the wind would suddenly drop during the final landing maneuver.


The landing gear of AP-3 consists of four retractable legs with shock dampers. It is designed to enable landing on either the platform or on a test field. The AP-4 undercarriage will deviate in that it will be optimized for platform landing only.

The autopilot

The drone avionics in AP-3 and AP-4 are identical. Control of the dynamics is fully automatic in all weather conditions and even for failure recovery. It covers all flight phases: launch, power generation, land, storage and relaunch. The autopilot only requires human supervision at wind park level. It makes use of a triple-redundant on-board computer developed to the most stringent aviation standard. Each computer is connected to a separate and complete set of sensors, including a dual-band GNSS unit (receiving GPS, Glonass, Galileo at L1 and L2), an air data probe (measuring air speed and angle of attack), a three-axis accelerometer and a ring-laser gyro. A ground station transmits GNSS corrections for improved accuracy (RTK). All sensor data are shared between the three computers and integrated using an extended Kalman filter.

Test of the navigation unit in the ESA ESTEC GNSS test facility. In this facility the GNSS signals are simulated as received under actual flight conditions, allowing us to safely quantify availability and accuracy of the navigation solution during our highly dynamic flight conditions.

Each computer has an aviation grade real-time operating system with dual partitions each running an independently developed code. The main control code takes care of nominal behavior and deals with detection and recovery of any single system failure. A back-up landing algorithm, independently developed, runs continuously in parallel to take over should the main code for some reason falter. Redundancy between the computers is achieved by a voting algorithm (best-of-three logic). The algorithms that control the drone are developed in Simulink. We have set up a toolchain to automate much of the configuration management and verification work before a test flight, which includes coverage testing, Monte Carlo simulation (with realistic environmental disturbances and sensor errors selected at random) and hardware-in-the-loop testing.


The drone computers communicate with the platform. In most cases the winch dynamics are demanded by the drone. The launch and land system is mostly controlled by independent control loops that are synchronized with the drone dynamics.


This architecture makes it possible to achieve a safety and reliability level approaching that of commercial aviation. As will be explained in more detail in the final article of this series, we believe tolerance to any single failure is fundamental for commercial viability of utility scale AWE.