https://www.um.edu.mt/library/oar/handle/123456789/76357 2026-04-23T04:08:10Z https://www.um.edu.mt/library/oar/handle/123456789/101726 Title: EEG signal phase analysis for brain-computer interfacing Abstract: Brain-computer interfaces (BCIs) are devices that provide a direct link between an individual's brain and a computer, thereby allowing the user to communicate with the surrounding environment or to control equipment solely with the use of his mental activity. The application possibilities for BCIs are wide, ranging from communication devices for locked-in patients to brain-controlled gaming tools. The potential uses of BCI systems have led to the establishment of various BCI research groups worldwide working on different aspects to improve the performance and reliability of BCI systems. Most of the developed systems rely on electroencephalography (EEG) as a modality for recording brain activity. This choice is understandable since when compared to other recording modalities, EEG provides a non-invasive, and affordable solution for recording the dynamic electrical activity in the brain. A central issue that determines the performance of EEG-based BCIs is the extraction of reliable features that can adequately represent phenomena in the underlying brain activity and that can be used to distinguish between different mental states. Several feature extraction methods have been developed for this purpose. However, the majority of these methods focus primarily on the amplitude and power characteristics of the EEG signals. On the other hand, the phase relationships between different regions of the brain, which are associated with interactions between different neuronal areas in the brain have been considered to a much lesser extent for task discrimination in BCIs. In this work, two novel feature extraction methods that consider explicit phase information in the EEG data, namely the 'phase-synchronisation'-based common spatial patterns (P-CSP) method and the analytic common spatial patterns (ACSP) method, are proposed. The P-CSP method considers the most discriminative phase synchronisation links in order to separate two classes of data, while the ACSP method considers an analytic representation of the EEG signals to obtain an explicit representation of amplitude and phase components of the EEG signals. The performance of the two methods is analysed through a number of simulation examples and tests on real EEG data, where it is shown that the methods can yield a good classification rate and additionally provide informative spatial patterns on the underlying discriminative phenomena for the considered tasks. Furthermore, the ACSP method is also tested on a six-target phase coded steady state visual evoked potential (SSVEP) setup, where the classification accuracy depends primarily on the discrimination of brain activity in response to visual targets that only differ in their phase. It is shown that the ACSP method outperforms the conventional CSP method and other techniques typically used for such setups, and gives spatial patterns that are more representative of the underlying activity than the conventional CSP technique. Description: PH.D 2012-01-01T00:00:00Z https://www.um.edu.mt/library/oar/handle/123456789/101382 Title: Multiple modelling of EEG data to classify different mental states Abstract: An electroencephalogram EEG gives the possibility of recording the electrical brain activity non-invasively from different locations on the scalp. The EEG data is known to be nonstationary due to continuously changing dynamics characterising the underlying mental states. This work investigates the applicability of the Autoregressive Switching Multiple Model (AR-SMM) framework for the automatic labelling of EEG data, taking into consideration both simulated data to mimic the nature of transient events and get more insight on the sensitivity of the framework to different mental state characteristics, as well as real EEG data presenting different challenges for the modelling of the data through AR-SMMs. Although Autoregressive (AR) models have been applied extensively for EEG data analysis, their combination with a switching framework that can handle better the abruptly changing dynamics of the nonstationary data has not been given much attention. In this work an existing model order identification criterion is modified and validated through Monte Carlo analysis on both univariate and multivariate data showing that it gives more accurate model order estimates. AR features estimated through an EM-based Kalman Smoother (EMKS) were then shown to give features that can reliably distinguish between left and right hand movements and where performance was insensitive to the model order estimate. The second part of this work focuses on the AR-SMM framework using lower bounding approximations for the identification of transitions between different candidate models. Applied to real EEG data the framework showed the capability of identifying multiple transient events in a unified framework which requires very little training data. A novel approach of learning new states in a semi-supervised manner also showed the possibility of using this framework as an analytical tool to obtain further insight on the dynamics of the EEG data. Description: PH.D 2012-01-01T00:00:00Z https://www.um.edu.mt/library/oar/handle/123456789/101211 Title: Computational intelligence methods for dynamic control of mobile robots Abstract: The ever-increasing scale and complexity of modern machines and processes, coupled with a tightening of performance specifications, necessitates control systems with higher levels of intelligence. Towards this aim, the field of intelligent control strives to endow systems with the key abilities of adaptation, learning and autonomy, so that they operate successfully in complex and uncertain environments with minimal human intervention. Despite a substantial body of work that has accumulated in this field of research over the past two decades, there is still ample scope for development. In particular, there is a pressing need for intelligent schemes with enhanced functional adaptability for the control of multivariable nonlinear systems such as autonomous mobile robots. The work presented in this thesis is a step in this direction. More specifically, the first part of the thesis presents novel explicit dual adaptive neural network-based control schemes, for a general stochastic class of multivariable nonlinear systems. These schemes address issues of system complexity such as multivariable nonlinear dynamics, functional uncertainty, unpredictable external disturbances and measurement noise. In contrast to the majority of adaptive controllers, which rely on the certainty equivalence assumption, a dual adaptive scheme seeks to adapt to unknown situations as quickly as possible, and at the same time strives to minimize the errors it makes in the process by taking into consideration the uncertainty of its estimates. Few such controllers have ever been implemented and tested in practical applications, especially within the context of intelligent control, and to the best of the author's knowledge none for the motion control of mobile robots. For this reason, the second part of the thesis deals with the development of novel dual adaptive neuro-control schemes for the dynamic control of nonholonomic wheeled mobile robots with un known or uncertain dynamics. A comprehensive statistical comparative analysis, including Monte Carlo simulations and hypothesis testing, confirms that the proposed dual adaptive schemes are truly effective. In addition, the schemes proposed for mobile robots are also validated experimentally on a real robot designed and built by the author for the purpose of this research, thereby bridging the gap between theory and practice. Description: PH.D 2011-01-01T00:00:00Z https://www.um.edu.mt/library/oar/handle/123456789/101122 Title: Power filters for power factor correction and harmonics suppression employed with non-linear loads under non-sinusoidal conditions Abstract: The thesis is about power filters for power factor correction and harmonic suppression applied to non-linear loads under non-sinusoidal conditions. A non-linear load made up of a variable DC load connected via some power electronic converter connected to a three-phase non-sinusoidal supply is chosen as a typical load. The thesis presents two major methods of compensation. This first method does not involve any modification of either the converter or the load. Several power filters are considered and are connected in an external way to the load. Such filters include Passive Power Filters, Active Power Filters and Hybrid and Combination Power Filters. Two different algorithms are considered for the operation of the Active Power Filter; mainly that of the Instantaneous Power Values Algorithm and the Integral Power Values Algorithm. The global approach is considered for this type of compensation. Maximizing the filter's compensation capabilities is done by matching the characteristic impedance of the load with that of the supply at the point of common coupling. The second method is about modifying the converter in order to incorporate active power filtering techniques. In this case two AC to DC converters are considered, that is the Boost and the Buck-Boost Converter. The single-phase converter is modified to a three-phase system in order to include load balancing, while all the design considerations are taken into account. The thesis is complemented by simulated and practical results. Description: PH.D 2000-01-01T00:00:00Z