Abstract
D.Ing. (Electrical & Electronic Engineering Science)
The objective of the research performed in this thesis is to address the calibration process
of Fiber-Optic Gyroscopes (FOGs) - a class of gyroscopes that make use of the Sagnac
effect to determine rotational information from laser-light traveling in an optical fiber. The
calibration process has traditionally been a time-consuming and therefore an expensive
one due to the various environmental parameters that can influence the sensor under
operation. Calibration is not a step that can be neglected as it is the process whereby the
residual manufacturing errors in the sensor are characterized. If these measurement errors
are not eliminated, the sensors would result in the host vehicle's assumed position rapidly diverging from its true position. Once the errors are characterized, they can be removed from the sensor output to improve the accuracy of the complete navigation system. The class of the sensor is determined by the amount of residual errors and the smaller the residual errors, the more expensive the sensor. The specific focus of the study is to determine whether it is possible to reduce the
calibration cost of the Fiber-Optic Gyroscopes through the use of innovative calibration
strategies. The use of neural networks are investigated as an alternative to the traditional
calibration strategies which consists of the estimation of the constant error parameters
through stochastic estimation strategies such as Kalman filters. The whole calibration
problem is recast into the well-defined Systems Identification (SID) domain where the
whole calibration problem is considered in terms of the systems identification design steps. The main contributions presented in this study are that the traditional calibration strategy is reviewed by casting the calibration problem into the Systems Identification domain;
that a unified FOG error model is developed that combines a number of seemingly
contradictory error models available in the technical literature;
that computational intelligence techniques are used to perform gyro calibration;
that a novel, non-linear gyro calibration strategy is developed; and that the sensors are calibrated under the simultaneous dynamic excitation of the
full range of multi-dimensional environmental conditions.
In the process of the development of this new calibration strategy the need for a problemspecific Criterion of Fit was observed. Such a Criterion of Fit was therefore developed
and it acted as the core criterium whereby the accuracy of the new calibration strategy
was assessed. One of the most important results obtained from the research presented
in this thesis is that the new strategy significantly outperforms the traditional strategies
and that, with the availability of high-performance embedded computational platforms,
it has potential to be used within an operational environment as the gyro compensation
strategy of choice.