One of the key requirements to bring small Unmanned Aircraft Systems (sUASs) into commercial use is the ability to autonomously monitor system health and act appropriately if malfunctions are detected. Robustness with respect to sensor malfunctions is essential, especially for vehicle navigation. A viable sensor suite usually consists of gyros, accelerometers, a GPS device, and a 3-axis magnetometer (TAM). Measurements of the latter are often affected by distortions of the Earth’s magnetic field, and thus have to be considered carefully. In this thesis, this problem is taken into account and robust mechanisms for proper attitude and velocity estimation in case of time-variant magnetic distortions are provided.

The structure of this work can roughly be divided into three parts. The first part deals with the question of how attitude and velocity of an sUAS can be estimated properly and without loss of generality, even without using TAM measurements. In this case, attitude observability is time-variant and has to be ensured by certain maneuvers. Hence, knowledge of the related uncertainties is vital if robustness is considered. A particle filer (PF) is used to estimate attitude and velocity and a quantitative measure for non-observability is derived from the particle distribution. The derivation of this measure is supported by a rigorous description of the conditions for attitude observability.

The second part describes how TAM measurements can be safely integrated in the estimation process. For that reason, the PF is extended by a recursive maximum likelihood estimator (MLE) to monitor parameter changes in the TAM model. If the likelihood of the current parameter set drops significantly, TAM measurements are disabled and attitude is solely updated from GPS velocities. In the meantime, the MLE tries to recover a proper parameter set. As the method relies on proper attitude information, the proposed measure of non-observability is used as an additional penalty term in the likelihood function.

The third part deals with the practical application of the proposed algorithms. As PFs are computationally very demanding, they are usually not applicable to solve estimation problems on board an sUAS in real-time. Onboard computer systems (OCSs) for sUASs have to be small, lightweight and energy efficient, resulting in limited computational power. To overcome this problem, an OCS has been developed which comprises a field programmable gate array (FPGA) as coprocessor. The parallel computing capabilities of FPGAs are very well suited for PFs and computation speed can be increased dramatically if the properties of such devices are considered carefully.

The results presented in this thesis are supported by simulations and real flight tests with different vehicles. To get a quantitative measure of the quality of attitude and velocity estimates, the FPGA-based OCS has been flight tested on a civil aviation aircraft equipped with a high-precision attitude and heading reference system (AHRS). Further tests have been carried out with a fixed-wing sUAS, where attitude estimates of the proposed PF have been used for attitude control. The MLE for TAM monitoring and re-calibration has been tested in simulations where the model describing the thrust-dependent distortions is extracted from pre-recorded TAM measurements of the sUAS.

Schlagwörter: unmanned aircraft; attitude observability; particle filter; FPGA

Fortschrittsberichte des Instituts für Flugmechanik und Flugregelung