Model-based in-flight icing detection for fixed-wing UAVs

Background

Today, a key limitation to the operability of unmanned aerial vehicles(UAVs) is the risk of encountering in-flight icing conditions. Due to their relatively small size, UAVs are highly sensitive to the performance degradation caused by icing, leading to faults ranging from reduced efficiency to total loss of lift. Developing an aerodynamic model that can be used to detect change in the flight performance of a UAV during icing is therefore highly valuable.

Iced leading-edge of a wing

Scope

The aim of this project is to develop a set of aerodynamic models of the Skywalker X8 UAV in nomial and iced conditions. The idea is to fly the UAV with 3D-printed ice shapes attached to the wings, record the flight data and run the data through a system identification algorithm. Having an identified model opens for further possibilities to work on model-based fault detection and isolation algorithms as well as the development of model-based control schemes capable of handling in-flight icing.

NTNU UAVlab X8

The project emphasizes collection of flight data, which brings a unique and exciting opportunty to familiarize with the UAVs of the UAVlab, through in-field experience.

Proposed tasks

The list below presents a list of possible tasks related to the project:

• Literature survey. Review literature on system identification methods. Explore methods used specifically in aerodynamic model identification.
• Take part in the flight tests of the Skywalker X8 with artificial ice shapes. Contribute by designing optimal experiments to collect data such that it can be used to identify aerodynamic parameters with high accuracy.
• Using the recorded flight data, identify the coefficients for the nominal and iced UAV models. System identification includes both finding the optimal model complexity and choosing a suitable parameter identification method, allowing for comparative analysis of the various method such as equation error methods, output error methods, and filter error methods.
• (If time) Using the identified model, develop and implement a performance degradation detection algorithm. The algorithm can be based on multiple model estimation frameworks, hypothesis testing, machine-learning-based classification, or other strategies.

Contact

Contact supervisors Kristoffer Gryte, Richard Hann or Bogdan Løw-Hansen