System identification of fixed-wing UAV

Background

A model is at the core of most cybernetic problems, ranging from estimation and control, to planning and decisionmaking. We often take the models for granted, as they are stated in the problem description in the textbook, but for many real-world problems models needs to be identified before they can be applied.

Unfortunately, obtaining a good models can be challenging. This is particularly true for aerodynamic models of fixed-wing aircraft. One common approach is to perform wind tunnel experiments, where an aircraft is placed in a controlled airstream of a wind tunnel while the forces and moments acting on the aircraft is recorded. The controlled environment of the wind tunnel generally give accurate data, but unfortunately wind tunnels are expensive to build and operate, especially for full 6 DOF models with damping effects. Another approach is computational fluid dynamics (CFD), where the aircraft geometry is drawn on a computer such that the influence of the airstream can be analyzed, e.g. using the Navier-Stokes equations. Setting up such analysis demands expertize and a lot of computational resources, and the realisim of the analysis needs to be verified with data from wind tunnels or flight tests.

A third option is to identify the model based on logged sensor data, such as airspeed, acceleration/force and angular velocity, from flight experiments. This has the advantage that it does not need additional equipment, while having a moderate computational load. This approach, known as system identification, has been applied to larger aircraft for many decades already (see e.g. Jategaonkar (2006) and Klein and Morelli (2006)), but the miniaturization of avionics and popularity of smaller fixed-wing unmanned aircraft over the last decade has ensured continuation of this interest, see e.g. Simmons, McClelland, and Woolsey (2019), Grauer and Morelli (2015), Licitra et al. (2018), Licitra et al. (2019).

Another aspect of data-driven system identification is the possibility to identify models in real time. In addition to the above applications, this also enables diagnostics; has the propeller broken off? are we experiencing dangerous icing on the wings?

Scope

One of the more difficult parts of system identification is to design the experiment such that the collected data is sufficiently rich to extract the information about the model, which is the focus of this project.

Proposed tasks

• Investigate methods for system identification of fixed-wing UAVs
• Design optimal experiments to collect data that can be used to identify aerodynamic parameters with high accuracy
• Benchmark the experimental design by logging data in a simulator and comparing the identified model to that of the simulator
• Conclude the work in a written report

Possible extensions for the master thesis include

• perform the experiments on a real unmanned aircraft
• identify build-up of ice on the wings and/or propeller

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 background in rigid body dynamics, such as TTK4190 Guidance and Control of Vehicles and TTK4130 Modelling and Simulation
• Knowledge of optimization, from e.g. TTK4135 Optimization and Control
• Knowledge of modelling of dynamical systems. One example is TTK4260 Introduction to Multivariate Data Modelling, but also TTK4215 - Adaptive Control
• Ideally knowledge and interest in fluid/aerodynamics, through e.g. TEP4100 Fluid Mechanics or TEP4160 Aerodynamics

Contact

Contact researchers Kristoffer Gryte and Richard Hann

References

Grauer, Jared A, and Eugene A Morelli. 2015. “Generic Global Aerodynamic Model for Aircraft.” Journal of Aircraft 52 (1): 13–20.

Jategaonkar, Ravindra. 2006. Flight Vehicle System Identification: A Time Domain Methodology. Vol. 216. Reston, VA, USA: AIAA.

Klein, Vladislav, and Eugene A Morelli. 2006. Aircraft System Identification: Theory and Practice. American Institute of Aeronautics; Astronautics Reston, Va, USA.

Licitra, Giovanni, Adrian Bürger, Paul Williams, Richard Ruiterkamp, and Moritz Diehl. 2018. “Optimal Input Design for Autonomous Aircraft.” Control Engineering Practice 77: 15–27.

———. 2019. “Aerodynamic Model Identification of an Autonomous Aircraft for Airborne Wind Energy.” Optimal Control Applications and Methods.

Simmons, Benjamin M, Hunter G McClelland, and Craig A Woolsey. 2019. “Nonlinear Model Identification Methodology for Small, Fixed-Wing, Unmanned Aircraft.” Journal of Aircraft 56 (3): 1056–67.