Data mining (DM) is the process of finding patterns and relationships in databases. The breakthrough in computer technologies triggered a massive growth in data collected and maintained by organisations. In many applications, these data arrive continuously in large volumes as a sequence of instances known as a data stream. Mining these data is known as stream data mining. Due to the large amount of data arriving in a data stream, each record is normally expected to be processed only once. Moreover, this process can be carried out on different sites in the organisation simultaneously making the problem distributed in nature. Distributed stream data mining poses many challenges to the data mining community including scalability and coping with changes in the underlying concept over time. In this thesis, the author hypothesizes that learning classifier systems (LCSs) - a class of classification algorithms - have the potential to work efficiently in distributed stream data mining. LCSs are an incremental learner, and being evolutionary based they are inherently adaptive. However, they suffer from two main drawbacks that hinder their use as fast data mining algorithms. First, they require a large population size, which slows down the processing of arriving instances. Second, they require a large number of parameter settings, some of them are very sensitive to the nature of the learning problem. As a result, it becomes difficult to choose a right setup for totally unknown problems. The aim of this thesis is to attack these two problems in LCS, with a specific focus on UCS - a supervised evolutionary learning classifier system. UCS is chosen as it has been tested extensively on classification tasks and it is the supervised version of XCS, a state of the art LCS. In this thesis, the architectural design for a distributed stream data mining system will be first introduced. The problems that UCS should face in a distributed data stream task are confirmed through a large number of experiments with UCS and the proposed architectural design. To overcome the problem of large population sizes, the idea of using a Neural Network to represent the action in UCS is proposed. This new system - called NLCS { was validated experimentally using a small fixed population size and has shown a large reduction in the population size needed to learn the underlying concept in the data. An adaptive version of NLCS called ANCS is then introduced. The adaptive version dynamically controls the population size of NLCS. A comprehensive analysis of the behaviour of ANCS revealed interesting patterns in the behaviour of the parameters, which motivated an ensemble version of the algorithm with 9 nodes, each using a different parameter setting. In total they cover all patterns of behaviour noticed in the system. A voting gate is used for the ensemble. The resultant ensemble does not require any parameter setting, and showed better performance on all datasets tested. The thesis concludes with testing the ANCS system in the architectural design for distributed environments proposed earlier. The contributions of the thesis are: (1) reducing the UCS population size by an order of magnitude using a neural representation; (2) introducing a mechanism for adapting the population size; (3) proposing an ensemble method that does not require parameter setting; and primarily (4) showing that the proposed LCS can work efficiently for distributed stream data mining tasks.

本文首次提出應用新的模擬方法：定態(stationary) bootstrap來估計涉險值(Value-at-Risk, 以下簡稱VaR)。VaR是衡量投資組合市場風險(market risk)的工具，由於1990年代以來國際間對市場風險管理的重視，且VaR較傳統風險衡量指標容易瞭解，又考慮整個投資組合資產間相關性降低風險的效果，加上VaR可作為企業內控、主管機關監督、以及投資人評估企業營運狀況等指標，故已廣為實務界及學界所接受。目前幾種主要衡量VaR的方法，包括變異數—共變數法、歷史模擬法、蒙地卡羅模擬法、classical bootstrap法以及壓力測試法等，各有其假設限制及優缺點。其中，classical bootstrap在衡量VaR時，使用的假設比較少，似乎非常適合衡量VaR。但是classical bootstrap會割裂了資料自我相關的性質，較適用於獨立且相同分配的樣本。我們在本文中介紹修正classical bootstrap的方法：移動區塊(moving block) bootstrap以及定態bootstrap，並利用統計模擬的方式證明定態bootstrap適合用於時間序列資料，對於捕捉真實分配的能力很強。接著我們選取11檔上市公司股票建構投資組合，並利用classical bootstrap以及定態bootstrap來衡量1999年共266個交易日的VaR。實證結果支持定態bootstrap能夠正確地衡量VaR，且其結果與classical bootstrap有明顯的不同。定態bootstrap法是個比較合理的衡量VaR方法，因此，我們建議風險管理者可採用定態bootstrap 衡量VaR。

Data mining (DM) is the process of finding patterns and relationships in databases. The breakthrough in computer technologies triggered a massive growth in data collected and maintained by organisations. In many applications, these data arrive continuously in large volumes as a sequence of instances known as a data stream. Mining these data is known as stream data mining. Due to the large amount of data arriving in a data stream, each record is normally expected to be processed only once. Moreover, this process can be carried out on different sites in the organisation simultaneously making the problem distributed in nature. Distributed stream data mining poses many challenges to the data mining community including scalability and coping with changes in the underlying concept over time. In this thesis, the author hypothesizes that learning classifier systems (LCSs) - a class of classification algorithms - have the potential to work efficiently in distributed stream data mining. LCSs are an incremental learner, and being evolutionary based they are inherently adaptive. However, they suffer from two main drawbacks that hinder their use as fast data mining algorithms. First, they require a large population size, which slows down the processing of arriving instances. Second, they require a large number of parameter settings, some of them are very sensitive to the nature of the learning problem. As a result, it becomes difficult to choose a right setup for totally unknown problems. The aim of this thesis is to attack these two problems in LCS, with a specific focus on UCS - a supervised evolutionary learning classifier system. UCS is chosen as it has been tested extensively on classification tasks and it is the supervised version of XCS, a state of the art LCS. In this thesis, the architectural design for a distributed stream data mining system will be first introduced. The problems that UCS should face in a distributed data stream task are confirmed through a large number of experiments with UCS and the proposed architectural design. To overcome the problem of large population sizes, the idea of using a Neural Network to represent the action in UCS is proposed. This new system - called NLCS { was validated experimentally using a small fixed population size and has shown a large reduction in the population size needed to learn the underlying concept in the data. An adaptive version of NLCS called ANCS is then introduced. The adaptive version dynamically controls the population size of NLCS. A comprehensive analysis of the behaviour of ANCS revealed interesting patterns in the behaviour of the parameters, which motivated an ensemble version of the algorithm with 9 nodes, each using a different parameter setting. In total they cover all patterns of behaviour noticed in the system. A voting gate is used for the ensemble. The resultant ensemble does not require any parameter setting, and showed better performance on all datasets tested. The thesis concludes with testing the ANCS system in the architectural design for distributed environments proposed earlier. The contributions of the thesis are: (1) reducing the UCS population size by an order of magnitude using a neural representation; (2) introducing a mechanism for adapting the population size; (3) proposing an ensemble method that does not require parameter setting; and primarily (4) showing that the proposed LCS can work efficiently for distributed stream data mining tasks.

Data mining (DM) is the process of finding patterns and relationships in databases. The breakthrough in computer technologies triggered a massive growth in data collected and maintained by organisations. In many applications, these data arrive continuously in large volumes as a sequence of instances known as a data stream. Mining these data is known as stream data mining. Due to the large amount of data arriving in a data stream, each record is normally expected to be processed only once. Moreover, this process can be carried out on different sites in the organisation simultaneously making the problem distributed in nature. Distributed stream data mining poses many challenges to the data mining community including scalability and coping with changes in the underlying concept over time. In this thesis, the author hypothesizes that learning classifier systems (LCSs) - a class of classification algorithms - have the potential to work efficiently in distributed stream data mining. LCSs are an incremental learner, and being evolutionary based they are inherently adaptive. However, they suffer from two main drawbacks that hinder their use as fast data mining algorithms. First, they require a large population size, which slows down the processing of arriving instances. Second, they require a large number of parameter settings, some of them are very sensitive to the nature of the learning problem. As a result, it becomes difficult to choose a right setup for totally unknown problems. The aim of this thesis is to attack these two problems in LCS, with a specific focus on UCS - a supervised evolutionary learning classifier system. UCS is chosen as it has been tested extensively on classification tasks and it is the supervised version of XCS, a state of the art LCS. In this thesis, the architectural design for a distributed stream data mining system will be first introduced. The problems that UCS should face in a distributed data stream task are confirmed through a large number of experiments with UCS and the proposed architectural design. To overcome the problem of large population sizes, the idea of using a Neural Network to represent the action in UCS is proposed. This new system - called NLCS { was validated experimentally using a small fixed population size and has shown a large reduction in the population size needed to learn the underlying concept in the data. An adaptive version of NLCS called ANCS is then introduced. The adaptive version dynamically controls the population size of NLCS. A comprehensive analysis of the behaviour of ANCS revealed interesting patterns in the behaviour of the parameters, which motivated an ensemble version of the algorithm with 9 nodes, each using a different parameter setting. In total they cover all patterns of behaviour noticed in the system. A voting gate is used for the ensemble. The resultant ensemble does not require any parameter setting, and showed better performance on all datasets tested. The thesis concludes with testing the ANCS system in the architectural design for distributed environments proposed earlier. The contributions of the thesis are: (1) reducing the UCS population size by an order of magnitude using a neural representation; (2) introducing a mechanism for adapting the population size; (3) proposing an ensemble method that does not require parameter setting; and primarily (4) showing that the proposed LCS can work efficiently for distributed stream data mining tasks.

<p>In this thesis, porous graphitic carbon (PGC) has been used as packing material in packed capillary liquid chromatography. The unique chromatographic properties of PGC has been studied in some detail and applied to different analytical challenges using both electrospray ionization-mass spectrometry (ESI-MS) and ultra violet (UV) absorbance detection. </p><p>The crucial importance of disengaging the conductive PGC chromatographic separation media from the high voltage mass spectrometric interface has been shown. In the absence of a grounded point between the column and ESI emitter, a current through the column was present, and changed retention behaviors for 3-O-methyl-DOPA and tyrosine were observed. An alteration of the chromatographic properties was also seen when PGC was chemically oxidized with permanganate, possibly due to an oxidation of the few surface groups present on the PGC material. </p><p>The dynamic adsorption of the chiral selector lasalocid onto the PGC support resulted in a useful and stable chiral stationary phase. Extraordinary enantioselectivity was observed for 1-(1-naphthyl)ethylamine, and enantioseparation was also achieved for other amines, amino acids, acids and alcohols. </p><p>Finally, a new strategy for separation of small biologically active compounds in plasma and brain tissue has been developed. With PGC as stationary phase it was possible to utilize a mobile phase of high content of organic modifier, without the addition of ion-pairing agents, and still selectively separate the analytes. </p>

In this thesis, porous graphitic carbon (PGC) has been used as packing material in packed capillary liquid chromatography. The unique chromatographic properties of PGC has been studied in some detail and applied to different analytical challenges using both electrospray ionization-mass spectrometry (ESI-MS) and ultra violet (UV) absorbance detection. The crucial importance of disengaging the conductive PGC chromatographic separation media from the high voltage mass spectrometric interface has been shown. In the absence of a grounded point between the column and ESI emitter, a current through the column was present, and changed retention behaviors for 3-O-methyl-DOPA and tyrosine were observed. An alteration of the chromatographic properties was also seen when PGC was chemically oxidized with permanganate, possibly due to an oxidation of the few surface groups present on the PGC material. The dynamic adsorption of the chiral selector lasalocid onto the PGC support resulted in a useful and stable chiral stationary phase. Extraordinary enantioselectivity was observed for 1-(1-naphthyl)ethylamine, and enantioseparation was also achieved for other amines, amino acids, acids and alcohols. Finally, a new strategy for separation of small biologically active compounds in plasma and brain tissue has been developed. With PGC as stationary phase it was possible to utilize a mobile phase of high content of organic modifier, without the addition of ion-pairing agents, and still selectively separate the analytes.

New technologies, such as an efficient vapor-compression evaporator, a stationary lime kiln (SLK), and the MixAlco process, compelled us to re-evaluate methods for producing sugar from cane. These technologies allow more water and lime to be used, and they add more value to bagasse. Extracting and preserving the sugars, and lime pretreating the bagasse to enhance biodigestibility, all at the same time in a pile, was demonstrated to be unfeasible; therefore, sugar extraction must occur before lime treating the bagasse. Sugar extraction should occur countercurrently by lixiviation, where liquid moves in stages opposite to the soaked bagasse (megasse), which is conveyed by screw-press conveyors that gently squeeze the fiber in each stage, improving extraction. The performance of a pilot-scale screw-press conveyor was tested for dewatering capabilities and power consumption. The unoptimized equipment decreased megasse moisture from 96 to 89%. Simulation of the process suggested that eight stages are necessary to achieve 98% recovery from typical sugarcane. The cumulative power for the screw-press conveyor system was 17.0±2.1 hp•h/ton dry fiber. Thin raw juice preserved with lime for several months showed no sucrose degradation and no quality deterioration, except for reducing sugar destruction. The lime loading needed for 1-year preservation is 0.20 g Ca(OH)2/g sucrose. Shorter times require less lime. After preservation, the juice was carbonated and filtered, and the resulting sludge pelletized. Due to their high organic content, the pellets were too weak for calcination temperatures used in the SLK. The organics must be decreased prior to pelletization and sodium must be supplemented as a binding agent. Long-term lime pretreatment of bagasse showed two delignification phases: bulk (rapid) and residual (slow). These were modeled by two simultaneous first-order reactions. Treatments with air purging and higher temperatures (50 – 57oC) delignified more effectively, especially during the residual phase, thus yielding higher cellulase-enzyme digestibilities after 2 – 8 weeks of treatment. At temperatures > 60oC, pure oxygen purging is preferred. Fresh bagasse was of better quality than old bagasse. Treatment with NaOH yielded a larger bulk delignification phase than Ca(OH)2. Long-term lime pulping of bagasse was unsuitable for copy-quality paper, but it was appropriate for strawboard and other filler applications.

Today’s buildings are responsible for 40% of the world’s energy use and also a substantial share of the Global Warming Potential (GWP). In Sweden, about 21% of the energy use can be related to the heat losses through the climatic envelope. The “Million Program” (Swedish: Miljonprogrammet) is a common name for about one million housing units, erected between 1965 and 1974 and many of these buildings suffer from poor energy performance. An important aim of this study was to access the possibilities of using Vacuum Insulation Panels (VIPs) in buildings with emphasis on the use of VIPs for improving the thermal efficiency of the “Million Program” buildings. The VIPs have a thermal resistance of about 8-10 times better than conventional insulations and offer unique opportunities to reduce the thickness of the thermal insulation. This thesis is divided into three main subjects. The first subject aims to investigate new alternative VIP cores that may reduce the market price of VIPs. Three newly developed nanoporous silica were tested using different steady-state and transient methods. A new self-designed device, connected to a Transient Plane Source (TPS) instrument was used to determine the thermal conductivity of granular powders at different gaseous pressure combined with different mechanical loads. The conclusion was that the TPS technique is less suitable for conducting thermal conductivity measurements on low-density nanoporous silica powders. However, deviations in the results are minimal for densities above a limit at which the pure conduction becomes dominant compared to heat transfer by radiation. The second subject of this work was to propose a new and robust VIP mounting system, with minimized thermal bridges, for improving the thermal efficiency of the “Million Program” buildings. On the basis of the parametric analysis and dynamic simulations, a new VIP mounting system was proposed and evaluated through full scale measurements in a climatic chamber. The in situ measurements showed that the suggested new VIP technical solution, consisting of 20mm thick VIPs, can improve the thermal transmittance of the wall, up to a level of 56%. An improved thermal transmittance of the wall at centre-of-panel coordinate of 0.118 to 0.132 W m-2K-1 and a measured centre-of-panel thermal conductivity (λcentre-of-panel) of 7 mW m-1K-1 were reached. Furthermore, this thesis includes a new approach to measure the thermal bridge impacts due to the VIP joints and laminates, through conducting infrared thermography investigations. An effective thermal conductivity of 10.9 mW m-1K-1 was measured. The higher measured centre-of-panel and effective thermal conductivities than the published centre-of-panel thermal conductivity of 4.2 mW m-1K-1 from the VIP manufacturer, suggest that the real thermal performance of VIPs, when are mounted in construction, is comparatively worse than of the measured performance in the laboratory. An effective thermal conductivity of 10.9 mW m-1K-1 will, however, provide an excellent thermal performance to the construction. The third subject of this thesis aims to assess the environmental impacts of production and operation of VIP-insulated buildings, since there is a lack of life cycle analysis of whole buildings with vacuum panels. It was concluded that VIPs have a greater environmental impact than conventional insulation, in all categories except Ozone Depilation Potential. The VIPs have a measurable influence on the total Global Warming Potential and Primary Energy use of the buildings when both production and operation are taken into account. However, the environmental effect of using VIPs is positive when compared to the GWP of a standard building (a reduction of 6%) while the PE is increased by 20%. It was concluded that further promotion of VIPs will benefit from reduced energy use or alternative energy sources in the production of VIP cores while the use of alternative cores and recycling of VIP cores may also help reduce the environmental impact. Also, a sensitivity analysis of this study showed that the choice of VIPs has a significant effect on the environmental impacts, allowing for a reduction of the total PE of a building by 12% and the GWP can be reduced as much as 11% when considering both production and operation of 50 yes. Finally, it’s possible to conclude that the VIPs are very competitive alternative for insulating buildings from the Swedish “Million Program”. Nevertheless, further investigations require for minimizing the measurable environmental impacts that acquired in this LCA study for the VIP-insulated buildings. / Dagens byggnader ansvarar för omkring 40% av världens energianvändning och  står också för en väsentlig del av utsläppen av växthusgaser. I Sverige kan ca 21 % av energianvändningen relateras till förluster genom klimatskalet. Miljonprogrammet är ett namn för omkring en miljon bostäder som byggdes mellan 1965 och 1974, och många av dessa byggnader har en dålig energiprestanda efter dagens mått. Huvudsyftet med denna studie har varit att utforska möjligheterna att använda vakuumisoleringspaneler (VIP:ar) i byggnader med viss fokus på tillämpning i Miljonprogrammets byggnader. Med en värmeledningsförmåga som är ca 8 - 10 gånger bättre än för traditionell isolering erbjuder VIP:arna unika möjligheter till förbättrad termisk prestanda med minimal isolerings tjocklek. Denna avhandling hade tre huvudsyften. Det första var att undersöka nya alternativ för kärnmaterial som bland annat kan reducera kostnaden vid produktion av VIP:ar. Tre nyutvecklade nanoporösa kiselpulver har testats med olika stationära och transienta metoder. En inom projektet utvecklad testbädd som kan anslutas till TPS instrument (Transient Plane Source sensor), har använts för att mäta värmeledningsförmågan hos kärnmaterial för VIP:ar, vid varierande gastryck och olika mekaniska laster. Slutsatsen blev att transienta metoder är mindre lämpliga för utföra mätningar av värmeledningsförmåga för nanoporösa kiselpulver låg densitet. Avvikelsen i resultaten är dock minimal för densiteter ovan en gräns då värmeledningen genom fasta material blir dominerande jämfört med värmeöverföring genom strålning. Det andra syftet har varit att föreslå ett nytt monteringssystem för VIP:ar som kan användas för att förbättra energieffektiviteten i byggnader som är typiska för Miljonprogrammet. Genom parametrisk analys och dynamiska simuleringar har vi kommit fram till ett förslag på ett nytt monteringssystem för VIP:ar som har utvärderats genom fullskaleförsök i klimatkammare. Resultaten från fullskaleförsöken visar att den nya tekniska lösningen förbättrar väggens U-värde med upp till 56 %. En förbättrad värmegenomgångskoefficienten för väggen i mitten av en VIP blev mellan 0.118 till 0,132 W m-2K-1 och värmeledningstalet centre-av-panel 7 mW m-1K-1 uppnåddes. Detta arbete innehåller dessutom en ny metod för att mäta köldbryggor i anslutningar med hjälp av infraröd termografi. En effektiv värmeledningsförmåga för 10.9 mW m-1K-1 uppnåddes. Resultaten tyder även på att den verkliga termiska prestandan av VIP:ar i konstruktioner är något sämre än mätvärden för paneler i laboratorium. En effektiv värmeledningsförmåga av 10.9 mW m-1K-1 ger dock väggkonstruktionen en utmärkt termisk prestanda. Det tredje syftet har varit att bedöma miljöpåverkan av en VIP-isolerad byggnad, från produktion till drift, eftersom en livscykelanalys av hela byggnader som är isolerade med vakuumisoleringspaneler inte har gjorts tidigare. Slutsatsen var att VIP:ar har en större miljöpåverkan än traditionell isolering, i alla kategorier förutom ozonnedbrytande potential. VIP:ar har en mätbar påverkan på de totala utsläppen av växthusgaser och primärenergianvändningen i byggnader när både produktion och drift beaktas. Miljöpåverkan av de använda VIP:arna är dock positiv jämfört med GWP av en standardbyggnad (en minskning med 6 %) medan primärenergianvändningen ökade med 20 %. Slutsatsen var att ytterligare användning av VIP:ar gynnas av reducerad energiförbrukning och alternativa energikällor i produktionen av nanoporösa kiselpulver medan användningen av alternativa kärnmaterial och återvinning av VIP kärnor kan hjälpa till att minska miljöpåverkan. En känslighetsanalys visade att valet av VIP:ar har en betydande inverkan på miljöpåverkan, vilket ger möjlighet att reducera den totala användningen av primärenergi i en byggnad med 12 % och utsläppen av växthusgaser kan vara minska, så mycket som 11 % när det gäller både produktion och drift under 50 år. Avslutningsvis är det möjligt att dra slutsatsen att VIP:ar är ett mycket konkurrenskraftigt alternativ för att isolera byggnader som är typiska för Miljonprogrammet. Dock krävs ytterligare undersökningar för att minimera de mätbara miljöeffekter som förvärvats i denna LCA-studie för VIP-isolerade byggnader. / <p>QC 20151109</p> / Simulations of heat and moisture conditions in a retrofit wall construction with Vacuum Insulation Panels / Textural and thermal conductivity properties of a low density mesoporous silica material / A study of the thermal conductivity of granular silica materials for VIPs at different levels of gaseous pressure and external loads / Evaluation of the thermal conductivity of a new nanoporous silica material for VIPs – trends of thermal conductivity versus density / A comparative study of the environmental impact of Swedish residential buildings with vacuum insulation panels / ETICS with VIPs for improving buildings from the Swedish million unit program “Miljonprogrammet”

Το πρόβλημα που αντιμετωπίζει η διατριβή αφορά στη μοντελοποίηση μη-στασίμων τυχαίων ταλαντώσεων επί τη βάσει μετρήσεων του σήματος της ταλάντωσης, μέσω μοντέλων FS-TAR/TARMA. Οι στόχοι της διατριβής περιλαμβάνουν την αποτίμηση της εφαρμοσιμότητας των μεθόδων FS-TAR/TARMA για την μοντελοποίηση και ανάλυση της ταλάντωσης χρονικά μεταβαλλόμενών κατασκευών, καθώς και τη σύγκρισή τους με εναλλακτικές παραμετρικές μεθόδους του πεδίου του χρόνου. Ιδιαίτερη βαρύτητα δίνεται και στην αντιμετώπιση θεμάτων που σχετίζονται με την εκτίμηση μοντέλων FS-ΤAR/TARMA, καθώς και στην θεωρητική ασυμπτωτική ανάλυση των ιδιοτήτων των εκτιμητριών που χρησιμοποιούνται. Η διατριβή αρχικά παρουσιάζει μια συγκριτική ανασκόπηση της βιβλιογραφίας στο θέμα της μοντελοποίησης μη-στασίμων ταλαντώσεων μέσω παραμετρικών μεθόδων του πεδίου του χρόνου, η οποία και επιδεικνύει τα πλεονεκτήματα των μεθόδων FS-TAR/TARMA. Στη συνέχεια αντιμετωπίζεται μια σειρά προβλημάτων που εμφανίζονται κατά την εκτίμηση (των παραμέτρων) και την επιλογή της δομής του μοντέλου. Η αποτελεσματικότητα των μεθόδων FS-TAR/TARMA για την μοντελοποίηση και ανάλυση μη-στάσίμων ταλαντώσεων επιδεικνύεται και πειραματικά μέσω εφαρμογής στην οποία πραγματοποιείται επιτυχής εξαγωγή των δυναμικών χαρακτηριστικών μιας εργαστηριακής χρονικά μεταβαλλόμενης κατασκευής. Στη συνέχεια, η διατριβή εστιάζει στην αναζήτηση ακριβέστερων εκτιμητριών, καθώς και στην ασυμπτωτική ανάλυση των ιδιοτήτων των εκτιμητριών «γενικών» (όχι αναγκαστικά περιοδικά μεταβαλλόμενων) μοντέλων FS-TAR/TARMA. Συγκεκριμένα, εξετάζονται οι περιπτώσεις των εκτιμητριών σταθμισμένων ελαχίστων τετραγώνων [Weighted Least Squares (WLS)], μέγιστης πιθανοφάνειας [Maximum Likelihood (ML)], καθώς και μια εκτιμήτρια πολλαπλών σταδίων [Multi Stage (MS)], η οποία αναπτύσσεται στην παρούσα διατριβή και είναι ασυμπτωτικά ισοδύναμη με την εκτιμήτρια ML ενώ ταυτόχρονα χαρακτηρίζεται από μειωμένη υπολογιστική πολυπλοκότητα. Στη διατριβή αποδεικνύεται η συνέπεια (consistency) των εκτιμητριών αυτών και εξάγεται η ασυμπτωτική κατανομή (asymptotic distribution) τους. Παράλληλα, αναπτύσσεται μια συνεπής εκτιμήτρια του ασυμπτωτικού πίνακα συνδιασποράς και μια μέθοδος για τον έλεγχο εγκυρότητας των μοντέλων FS-TAR/TARMA. Η ορθότητα των αποτελεσμάτων της ασυμπτωτικής ανάλυσης επιβεβαιώνεται μέσω μελετών Monte Carlo. / The thesis studies the problem of non-stationary random vibration modeling and analysis based on available measurements of the vibration signal via Functional Series Time-dependent AutoRegressive / AutoRegressive Moving Average (FS-TAR/ TARMA) models. The aims of the thesis include the assessment of the applicability of FS-TAR/TARMA methods for the modeling and analysis of non-stationary random vibration, as well as their comparison with alternative time-domain parametric methods. In addition, significant attention has been paid to the FS-TAR/TARMA estimation problem and to the theoretical asymptotic analysis of the estimators. A critical overview and comparison of time-domain, parametric, non-stationary random vibration modeling and analysis methods is firstly presented, where the high potential of FS-TAR/TARMA methods is demonstrated. In the following, a number of issues concerning the FS-TAR/TARMA model (parameter) estimation and model structure selection are considered. The effectiveness of the FS-TARMA methods for non-stationary random vibration modeling and analysis is experimentally demonstrated, through their application for the recovery of the dynamical characteristics of a time-varying bridge-like laboratory structure. In the sequel, the thesis focuses on the asymptotic analysis of “general” (that is not necessarily periodically evolving) FS-TAR/TARMA estimators. In particular, the Weighted Least Squares (WLS) and Maximum Likelihood (ML) estimators are both investigated, while a Multi Stage (MS) estimator, that approximates the ML estimator at reduced complexity, is developed. The consistency of the considered estimators is established and their asymptotic distribution is extracted. Furthermore, a consistent estimator of the asymptotic covariance matrix is formulated and an FS-TAR/TARMA model validation method is proposed. The validity of the theoretical asymptotic analysis results is assessed through several Monte Carlo studies.

Cette recherche a pour but d’évaluer le rôle de la vision et de la proprioception pour la perception et le contrôle de l’orientation spatiale de la main chez l’humain. L’orientation spatiale de la main est une composante importante des mouvements d’atteinte et de saisie. Toutefois, peu d’attention a été portée à l’étude de l’orientation spatiale de la main dans la littérature. À notre connaissance, cette étude est la première à évaluer spécifiquement l’influence des informations sensorielles et de l’expérience visuelle pour la perception et le contrôle en temps réel de l'orientation spatiale de la main pendant le mouvement d’atteinte naturel vers une cible stationnaire. Le premier objectif était d’étudier la contribution de la vision et de la proprioception dans des tâches de perception et de mouvement d’orientation de la main. Dans la tâche de perception (orientation-matching task), les sujets devaient passivement ou activement aligner une poignée de forme rectangulaire avec une cible fixée dans différentes orientations. Les rotations de l’avant-bras et du poignet étaient soit imposées par l’expérimentateur, soit effectuées par les sujets. Dans la tâche de mouvement d’orientation et d’atteinte simultanées (letter posting task 1), les sujets ont réalisé des mouvements d’atteinte et de rotation simultanées de la main afin d’insérer la poignée rectangulaire dans une fente fixée dans les mêmes orientations. Les tâches ont été réalisées dans différentes conditions sensorielles où l’information visuelle de la cible et de la main était manipulée. Dans la tâche perceptive, une augmentation des erreurs d’orientation de la main a été observée avec le retrait des informations visuelles concernant la cible et/ou ou la main. Lorsque la vision de la main n’était pas permise, il a généralement été observé que les erreurs d’orientation de la main augmentaient avec le degré de rotation nécessaire pour aligner la main et la cible. Dans la tâche de mouvement d’orientation et d’atteinte simultanées, les erreurs ont également augmenté avec le retrait des informations visuelles. Toutefois, les patrons d’erreurs étaient différents de ceux observés dans la tâche de perception, et les erreurs d’orientation n’ont pas augmenté avec le degré de rotation nécessaire pour insérer la poignée dans la fente. En absence de vision de la main, il a été observé que les erreurs d’orientation étaient plus petites dans la tâche de mouvement que de perception, suggérant l’implication de la proprioception pour le contrôle de l’orientation spatiale de la main lors des mouvements d’orientation et d’atteinte simultanées. Le deuxième objectif de cette recherche était d’étudier l’influence de la vision et de la proprioception dans le contrôle en temps réel de l’orientation spatiale de la main. Dans une tâche d’orientation de la main suivie d’une atteinte manuelle (letter posting task 2), les sujets devaient d’abord aligner l’orientation de la même poignée avec la fente fixée dans les mêmes orientations, puis réaliser un mouvement d’atteinte sans modifier l’orientation initiale de la main. Une augmentation des erreurs initiales et finales a été observée avec le retrait des informations visuelles. Malgré la consigne de ne pas changer l’orientation initiale de la main, une diminution des erreurs d’orientation a généralement été observée suite au mouvement d’atteinte, dans toutes les conditions sensorielles testées. Cette tendance n’a pas été observée lorsqu’aucune cible explicite n’était présentée et que les sujets devaient conserver l’orientation de départ de la main pendant le mouvement d’atteinte (mouvement intransitif; letter-posting task 3). La diminution des erreurs pendant l’atteinte manuelle transitive vers une cible explicite (letter-posting task 2), malgré la consigne de ne pas changer l’orientation de la main pendant le mouvement, suggère un mécanisme de corrections automatiques pour le contrôle en temps réel de l’orientation spatiale de la main pendant le mouvement d’atteinte naturel vers une cible stationnaire. Le troisième objectif de cette recherche était d’évaluer la contribution de l’expérience visuelle pour la perception et le contrôle de l’orientation spatiale de la main. Des sujets aveugles ont été testés dans les mêmes tâches de perception et de mouvement. De manière générale, les sujets aveugles ont présenté les mêmes tendances que les sujets voyants testés dans la condition proprioceptive (sans vision), suggérant que l’expérience visuelle n’est pas nécessaire pour le développement d’un mécanisme de correction en temps réel de l’orientation spatiale de la main basé sur la proprioception. / The goal of this research was to study the contribution of vision and proprioception to the perception and control of hand orientation in human subjects. Spatial orientation of the hand is an important component of reaching and grasping movements. However, not much attention has been given to spatial hand orientation in the literature. To our knowledge, this study is the first to specifically investigate the influence of sensory information for the perception and on-line control of hand orientation during natural reaching movement to stationary targets. The first objective of this research was to study the contribution of vision and proprioception in perceptual orientation-matching and motor letter posting tasks. In the perceptual orientation-matching task, subjects attempted to passively or actively align a match handle, to a target that was fixed in different orientations. In the passive perceptual task, passive rotations of the forearm and wrist were imposed by the experimenter; whereas in the active perceptual task, the rotations were actively executed by the subjects. In letter posting task 1, subjects simultaneously reached and rotated the right hand to insert a match handle into a target slot fixed in the same orientations. The tasks were performed in different sensory conditions where the visual information about the target and the hand was manipulated. In the perceptual orientation-matching task, augmentation of hand orientation errors was observed with the withdrawal of visual information related to either the target and/or the hand. When full vision was not allowed, hand orientation errors were larger overall when larger rotations of the wrist were required to match the target, whether the rotations were made actively by the subject or were imposed passively by the experimenter. In letter posting task 1, augmentation of hand orientation errors was also observed with the withdrawal of visual information related to either the target and/or the hand. However, errors patterns were different from those observed in the perceptual task, and hand orientation errors were not larger for larger target orientations. Without vision of the hand, final hand orientation errors were smaller overall in letter-posting task 1 than in the orientation-matching task. This suggests the implication of the proprioceptive information for the control of spatial hand orientation during reach-and-orient movements. The second objective of this research was to study the influence of vision and proprioception in on-line control of spatial hand orientation. In letter posting task 2, subjects first aligned their hand to the angle of the target and then reached to it with the instruction not to change their initial hand orientation. The augmentation of initial and final errors was observed with the withdrawal of vision. Although subjects were instructed to not change their hand orientation, in all sensory condition tested, hand orientation changed overall during reaching in a way that reduced the initial orientation errors. This trend did not occur when there was no explicitly defined target toward which the subjects reached (letter-posting task 3; intransitive movement). The reduction in hand orientation errors during transitive reach in letter-posting task 2, even when told not to change it, suggests the engagement of an automatic error correction mechanism for hand orientation during natural reaching movements toward stationary targets. The third objective of this research was to investigate the contribution of visual experience to the perception and control of spatial orientation of the hand. Blind subjects were tested in the same perceptual and motor tasks. Overall, no differences were observed between performance of blind subjects and normally-sighted subjects tested without vision (proprioceptive condition), suggesting that prior visual experience is not necessary for the development of an on-line error correction mechanism for hand orientation guided by proprioceptive inputs.