Fault-detection and isolation of propeller icing for unmanned aerial vehicles
Background
Before unmanned aerial vehicles can operate completely autonomously in non-segregated airspace, there are many challenges that needs to be tackled related to reducing the operational risk. A major risk is loss of propulsion, which could imply loss of the vehicle and severe damage to third parties. One cause of loss of propulsion is propeller icing, which can cause the propeller to loose 75% of its thrust in less than 2 minutes. Icing is not only limited to arctic environments, and is not limited to the propeller of the vehicle, but is a global problem and also applies for the wings of the aircraft. There are solutions for the icing problem for large aircraft, but they are generally difficult to scale down to a small aircraft. The probability of icing is also larger for UAVs, as icing conditions occur more often at altitudes at which UAVs operate.
Recent studies have shown that it is possible to identify changes in propeller efficiency, which could be caused by e.g. propeller icing, based on available sensor data such as airspeed, motor rotational speed (RPM), battery voltage and current drawn by the motor. These faults can also be separated from other faults, such as mechanical failures in the ball bearing of the motor.
Scope
A challenge with the fault detection is that the change in the propeller efficiency can be unobservable if the motor operates at a constant speed. To mitigate this effect, it is of interest to develop a fault detection and identification algorithm that couples the estimation of the propeller efficiency (the fault detection) with the control of the motor; if the fault detection suspects that there is icing, it can command the motor to change its rotational speed such that the propeller parameters are observable, to determine if there actually was a fault or not.
Proposed tasks
- Familiarize with fault detection and isolation algorithms, especially with respect to detection of icing on propellers.
- Adapt an existing fault detection algorithm implementation to make it run in real-time
- Extend an existing simulation framework (based on Simulink and ArduPilot) to include the above fault detection implementation
- Investigate methods for increasing the observability of the motor/propeller faults. Of particular interest is the coupling of the fault detection with the control of the motor.
- Conclude the work in a written report
Possible extensions for the master thesis include
- implementation on a physical UAV and experimental validation of the method
- improve the accuracy and applicability of the fault detection by improving the underlying model, based on existing data that has been recorded in an icing wind tunnel.
Prerequisites
This is a list of recommended prerequisites, more to signal what it will involve than to be used as a filter on candidates.
- Strong theoretical background and interest in statistics for estimation/sensor fusion, through e.g. TTK4250 Sensor Fusion (which also could be taken in parallel with the project, instead of the two 3.75 credit specialization courses).
- Basic familiarity with models of electro-mechanical systems, from e.g. TTK4130 Modelling and Simulation
- Ideally knowledge and interest in fluid/aerodynamics, through e.g. TEP4100 Fluid Mechanics or TEP4160 Aerodynamics
Contact
Contact supervisors Kristoffer Gryte, Richard Hann and Tor Arne Johansen