Audiosignaalin digitaalinen käsittely aktiivisessa meluntorjunnassa

Meluntorjuntaan on perinteisesti käytetty passiivisia menetelmiä. Monissa sovelluksissa melua voidaan vaimentaa myös aktiivisella meluntorjunnalla. Tässä työssä tutkitaan aktiivisen meluntorjunnan signaalinkäsittelyä sekä signaalinkäsittelyyn soveltuvia laitteistoja. Lisäksi selvitetään ANC-järjestelmien (Active Noise Control) toimintaanja signaalinkäsittelyyn vaikuttavia tekijöitä. Tutkinnassa rajoitutaan yksikanavaisiin järjestelmiin. Esimerkkisovelluksena käytetään ulkotilan melunvaimennukseen soveltuvaa järjestelmää. Esimerkkijärjestelmään suunniteltiin signaalinkäsittelylaitteisto, jonka soveltuvuutta järjestelmän signaalinkäsittelyyn selvitettiin suorituskykymittauksin ja kokeellisin mittauksin.Lisäksi pohditaan signaalinkäsittelyn toteuttamista tutkittavassa järjestelmässä ja sovelluskohteessa.

Rasvaisten jätevesien puhdistus

Rasvaisten jätevesien puhdistus on sitä tuottaville yrityksille kallista ja hankalaa. Nykyisten päästövaatimusten saavuttaminen on perinteisillä jätevedenkäsittelymenetelmillä vaikeaa tai lähes mahdotonta, riippuen käsiteltävän jäteveden ominaisuuksista. Rasvaisen jäteveden käsittelyssä on käytetty mm. laskeutusta, flotaatiota, hydrosykloneja, pisaroitusta, suodatusta sekä biologista käsittelyä. Lisäksi happohydrolyysiä voidaan soveltaa edellä mainittujen menetelmien esikäsittelynä. Useita näitä erotusmenetelmiä voidaan myös tehostaa kemikaalien lisäyksellä. Työn kirjallisuusosassa on käsitelty rasvaisten jätevesien ja emulsioiden kalvosuodatusta ja kalvojen käyttöä pisaroituselementtinä. Tavanomaisessa kalvosuodatuksessa tarkoituksena on erottaa kalvoa läpäisemätön rasvajae ja permeoituva vesijae toisistaan, kun taas pisaroittamisen tarkoituksena on saada dispergoituneen faasin pisarakoko kasvamaan joko kalvon pinnalla tai sen huokosissa. Pisarakoon kasvaessa emulsion stabiilisuus heikkenee ja faasit voidaan helpommin erottaa toisistaan. Työn kokeellisessa osassa tavoitteena oli tutkia kalvosuodatuksen ja erilaisten kalvojen toimivuutta esteröintilaitoksen rasvaisten jätevesien käsittelyssä. Tutkimuksessa käytettiin MW- (GE Osmonics), C30F- (Nadir Filtration), Teflon Typar- (Tetratex) sekä JX-kalvoa (Osmonics Desal). Haastetta työhön syntyi tutkittujen jätevesien ominaisuuksien suuresta vaihtelusta sitä tuottavan laitoksen panostyylisestä toiminnasta johtuen. Lisäksi tutkittiin, onko syöttöliuoksen pH-säädöllä ja laskeutuksella ennen suodatusta merkittävää etua itse kalvosuodatusprosessiin. Kalvotekniikkaa voidaan tämän tutkimuksen perusteella soveltaa myös esteröintilaitoksen rasvaisten jätevesien suodatukseen, ja erityistä etua saadaan jäteveden pH-säädöllä (pH 3) ja laskeutuksella ennen varsinaista suodatusta. Tällaiseen käsittelyyn soveltuu tutkituista kalvoista parhaiten hydrofiilinen regeneroidusta selluloosasta valmistettu C30F-kalvo, jonka etuna on vähäinen foulaantuminen muihin tutkittuihin kalvoihin verrattuna.

Inhimillisten virheiden tarkastelu ydinvoimalaitosten käyttökokemusten hyödyntämisessä

Tässä työssä on kuvattu ydinvoimalaitosten käyttökokemusten tutkimusta keskittyen erityisesti inhimillisten toimintojen tarkasteluun. Työssä on kerrottu kansainvälisistä vaatimuksista ja järjestöistä sekä yleisesti käyttökokemusten tutkimuksessa käytössä olevista menetelmistä keskittyen perussyyanalyysimenetelmiin. Suomen osalta työssä on käsitelty lainsäädännön asettamia velvoitteita ja muita vaatimuksia, jotka ydinvoima-alalla koostuvat lähinnä Säteilyturvakeskuksen YVL-ohjeista. Viranomaisena toimivan Säteilyturvakeskuksen, alan tutkimusta suorittavan Valtion teknillisen tutkimuskeskuksen ja Teollisuuden Voima Oy:n käyttökokemusten tutkimiseen liittyvät organisaatiot ja menettelytavat on esitelty. Fortum Power and Heat Oy:n omistaman ja käyttämän Loviisan ydinvoimalaitoksen käyttökokemusten hyödyntäminen on käsitelty tarkemmin. Loviisan voimalaitoksen organisaatio ja käyttökokemusten sekä inhimillisten virheiden käsittelymenetelmiä on esitelty ja analysoitu. Työn alkuvaiheessa Loviisan voimalaitoksella inhimillisistä virheistä kerätystä tiedosta koottu tietokanta järjesteltiin kuntoon. Järjestelyn jälkeen tietoa analysoitiin ja analysoinnin tulokset on esitetty tässä työssä. Sekä järjestelyn että analysoinnin aikana havaitut kehityskohteet kirjattiin muistiin. Pienet toimenpiteet suoritettiin heti ja suuremmat kirjattiin tämän työn toimenpide-ehdotuksiin. Kehittämiskeinoja on ehdotettu virheiden luokittelumenetelmään ja käyttökokemusten käsittelymenetelmiin.

Studies of Solar Activity with Emphasis on Quasi-Periodic Oscillations

In this thesis mainly long quasi-periodic solar oscillations in various solar atmospheric structures are discussed, based on data obtained at several wavelengths, focussing, however, mainly on radio frequencies. Sunspot (Articles II and III) and quiet Sun area (QSA) (Article I) oscillations are investigated along with quasi-periodic pulsations (QPP) in a flaring event with wide-range radio spectra (Article IV). Various oscillation periods are detected; 3–15, 35–70 and 90 minutes (QSA), 10-60 and 80-130 minutes (in sunspots at various radio frequencies), 3-5, 10-23, 220-240, 340 and 470 minutes (in sunspots at photosphere) and 8-12 and 15-17 seconds (in a solar flare at radio frequencies). Some of the oscillation periods are detected for the first time, while some of them have been confirmed earlier by other research groups. Solar oscillations can provide more information on the nature of various solar structures. This thesis presents the physical mechanisms of some solar structure oscillations. Two different theoretical approaches are chosen; magnetohydrodynamics (MHD) and the shallow sunspot model. These two theories can explain a wide range of solar oscillations from a few seconds up to some hours. Various wave modes in loop structures cause solar oscillations (<45 minutes) both in sunspots and quiet Sun areas. Periods lasting more than 45 minutes in the sunspots (and a fraction of the shorter periods) are related to sunspot oscillations as a whole. Sometimes similar oscillation periods are detected both in sunspot area variations and respectively in magnetic field strength changes. This result supports a concept that these oscillations are related to sunspot oscillations as a whole. In addition, a theory behind QPPs at radio frequencies in solar flares is presented. The thesis also covers solar instrumentation and data sources. Additionally, the data processing methods are presented. As the majority of the investigations in this thesis focus on radio frequencies, also the most typical radio emission mechanisms are presented. The main structures of the Sun, which are related to solar oscillations, are also presented. Two separate projects are included in this thesis. Solar cyclicity is studied using the extensively large solar radio map archieve from Metsähovi Radio Observatory (MRO) at 37 GHz, between 1978 and 2011 (Article V) covering two full solar cycles. Also, some new solar instrumentation (Article VI) was developed during this thesis.

Variable speed drive in fan system monitoring

Fan systems are responsible for approximately 10% of the electricity consumption in industrial and municipal sectors, and it has been found that there is energy-saving potential in these systems. To this end, variable speed drives (VSDs) are used to enhance the efficiency of fan systems. Usually, fan system operation is optimized based on measurements of the system, but there are seldom readily installed meters in the system that can be used for the purpose. Thus, sensorless methods are needed for the optimization of fan system operation. In this thesis, methods for the fan operating point estimation with a variable speed drive are studied and discussed. These methods can be used for the energy efficient control of the fan system without additional measurements. The operation of these methods is validated by laboratory measurements and data from an industrial fan system. In addition to their energy consumption, condition monitoring of fan systems is a key issue as fans are an integral part of various production processes. Fan system condition monitoring is usually carried out with vibration measurements, which again increase the system complexity. However, variable speed drives can already be used for pumping system condition monitoring. Therefore, it would add to the usability of a variablespeed- driven fan system if the variable speed drive could be used as a condition monitoring device. In this thesis, sensorless detection methods for three lifetime-reducing phenomena are suggested: these are detection of the fan contamination build-up, the correct rotational direction, and the fan surge. The methods use the variable speed drive monitoring and control options for the detection along with simple signal processing methods, such as power spectrum density estimates. The methods have been validated by laboratory measurements. The key finding of this doctoral thesis is that a variable speed drive can be used on its own as a monitoring and control device for the fan system energy efficiency, and it can also be used in the detection of certain lifetime-reducing phenomena.

Ion mobility spectrometry in liquid analysis

Ion mobility spectrometry (IMS) is a straightforward, low cost method for fast and sensitive determination of organic and inorganic analytes. Originally this portable technique was applied to the determination of gas phase compounds in security and military use. Nowadays, IMS has received increasing attention in environmental and biological analysis, and in food quality determination. This thesis consists of literature review of suitable sample preparation and introduction methods for liquid matrices applicable to IMS from its early development stages to date. Thermal desorption, solid phase microextraction (SPME) and membrane extraction were examined in experimental investigations of hazardous aquatic pollutants and potential pollutants. Also the effect of different natural waters on the extraction efficiency was studied, and the utilised IMS data processing methods are discussed. Parameters such as extraction and desorption temperatures, extraction time, SPME fibre depth, SPME fibre type and salt addition were examined for the studied sample preparation and introduction methods. The observed critical parameters were extracting material and temperature. The extraction methods showed time and cost effectiveness because sampling could be performed in single step procedures and from different natural water matrices within a few minutes. Based on these experimental and theoretical studies, the most suitable method to test in the automated monitoring system is membrane extraction. In future an IMS based early warning system for monitoring water pollutants could ensure the safe supply of drinking water. IMS can also be utilised for monitoring natural waters in cases of environmental leakage or chemical accidents. When combined with sophisticated sample introduction methods, IMS possesses the potential for both on-line and on-site identification of analytes in different water matrices.

Ionic liquid mediated biomass deconstruction: from analysis challenges to fermentable sugars

Ionic liquids, ILs, have recently been studied with accelerating interest to be used for a deconstruction/fractionation, dissolution or pretreatment processing method of lignocellulosic biomass. ILs are usually utilized combined with heat. Regarding lignocellulosic recalcitrance toward fractionation and IL utilization, most of the studies concern IL utilization in the biomass fermentation process prior to the enzymatic hydrolysis step. It has been demonstrated that IL-pretreatment gives more efficient hydrolysis of the biomass polysaccharides than enzymatic hydrolysis alone. Both cellulose (especially cellulose) and lignin are very resistant towards fractionation and even dissolution methods. As an example, it can be mentioned that softwood, hardwood and grass-type plant species have different types of lignin structures leading to the fact that softwood lignin (guaiacyl lignin dominates) is the most difficult to solubilize or chemically disrupt. In addition to the known conventional biomass processing methods, several ILs have also been found to efficiently dissolve either cellulose and/or wood samples – different ILs are suitable for different purposes. An IL treatment of wood usually results in non-fibrous pulp, where lignin is not efficiently separated and wood components are selectively precipitated, as cellulose is not soluble or degradable in ionic liquids under mild conditions. Nevertheless, new ILs capable of rather good fractionation performance have recently emerged. The capability of the IL to dissolve or deconstruct wood or cellulose depends on several factors, (e.g. sample origin, the particle size of the biomass, mechanical treatments as pulverization, initial biomassto-IL ratio, water content of the biomass, possible impurities of IL, reaction conditions, temperature etc). The aim of this study was to obtain (fermentable) saccharides and other valuable chemicals from wood by a combined heat and IL-treatment. Thermal treatments alone contribute to the degradation of polysaccharides (e.g. 150 °C alone is said to cause the degradation of polysaccharides), thus temperatures below that should be used, if the research interest lies on the IL effectiveness. On the other hand, the efficiency of the IL-treatment can also be enhanced to combine other treatment methods, (e.g. microwave heating). The samples of spruce, pine and birch sawdust were treated with either 1-Ethyl-3-methylimidazolium chloride, Emim Cl, or 1-Ethyl-3-methylimidazolium acetate, Emim Ac, (or with ionized water for comparison) at various temperatures (where focus was between 80 and 120 °C). The samples were withdrawn at fixed time intervals (the main interest treatment time area lied between 0 and 100 hours). Double experiments were executed. The selected mono- and disaccharides, as well as their known degradation products, 5-hydroxymethylfurfural, 5-HMF, and furfural were analyzed with capillary electrophoresis, CE, and high-performance liquid chromatography, HPLC. Initially, even GC and GC-MS were utilized. Galactose, glucose, mannose and xylose were the main monosaccharides that were present in the wood samples exposed to ILs at elevated temperatures; in addition, furfural and 5-HMF were detected; moreover, the quantitative amount of the two latter ones were naturally increasing in line with the heating time or the IL:wood ratio.

Laser scribing of thin-film PV modules

This study is a literature review on laser scribing in monolithically interconnected thin-film PV modules, focusing on efficiency of modules based on absorber materials CIGS, CdTe and a-Si. In thin-film PV module manufacturing scribing is used to interconnect individual cells monolithically by P1, P2 and P3 scribes. Laser scribing has several advantages compared to mechanical scribing for this purpose. However, laser scribing of thin-films can be a challenging process and may induce efficiency reducing defects. Some of these defects can be avoided by improving optimisation or processing methods.

Cost Reduction of Permanent Magnet Synchronous Machines

Electrical machine drives are the most electrical energy-consuming systems worldwide. The largest proportion of drives is found in industrial applications. There are, however many other applications that are also based on the use of electrical machines, because they have a relatively high efficiency, a low noise level, and do not produce local pollution. Electrical machines can be classified into several categories. One of the most commonly used electrical machine types (especially in the industry) is induction motors, also known as asynchronous machines. They have a mature production process and a robust rotor construction. However, in the world pursuing higher energy efficiency with reasonable investments not every application receives the advantage of using this type of motor drives. The main drawback of induction motors is the fact that they need slipcaused and thus loss-generating current in the rotor, and additional stator current for magnetic field production along with the torque-producing current. This can reduce the electric motor drive efficiency, especially in low-speed, low-power applications. Often, when high torque density is required together with low losses, it is desirable to apply permanent magnet technology, because in this case there is no need to use current to produce the basic excitation of the machine. This promotes the effectiveness of copper use in the stator, and further, there is no rotor current in these machines. Again, if permanent magnets with a high remanent flux density are used, the air gap flux density can be higher than in conventional induction motors. These advantages have raised the popularity of PMSMs in some challenging applications, such as hybrid electric vehicles (HEV), wind turbines, and home appliances. Usually, a correctly designed PMSM has a higher efficiency and consequently lower losses than its induction machine counterparts. Therefore, the use of these electrical machines reduces the energy consumption of the whole system to some extent, which can provide good motivation to apply permanent magnet technology to electrical machines. However, the cost of high performance rare earth permanent magnets in these machines may not be affordable in many industrial applications, because the tight competition between the manufacturers dictates the rules of low-cost and highly robust solutions, where asynchronous machines seem to be more feasible at the moment. Two main electromagnetic components of an electrical machine are the stator and the rotor. In the case of a conventional radial flux PMSM, the stator contains magnetic circuit lamination and stator winding, and the rotor consists of rotor steel (laminated or solid) and permanent magnets. The lamination itself does not significantly influence the total cost of the machine, even though it can considerably increase the construction complexity, as it requires a special assembly arrangement. However, thin metal sheet processing methods are very effective and economically feasible. Therefore, the cost of the machine is mainly affected by the stator winding and the permanent magnets. The work proposed in this doctoral dissertation comprises a description and analysis of two approaches of PMSM cost reduction: one on the rotor side and the other on the stator side. The first approach on the rotor side includes the use of low-cost and abundant ferrite magnets together with a tooth-coil winding topology and an outer rotor construction. The second approach on the stator side exploits the use of a modular stator structure instead of a monolithic one. PMSMs with the proposed structures were thoroughly analysed by finite element method based tools (FEM). It was found out that by implementing the described principles, some favourable characteristics of the machine (mainly concerning the machine size) will inevitable be compromised. However, the main target of the proposed approaches is not to compete with conventional rare earth PMSMs, but to reduce the price at which they can be implemented in industrial applications, keeping their dimensions at the same level or lower than those of a typical electrical machine used in the industry at the moment. The measurement results of the prototypes show that the main performance characteristics of these machines are at an acceptable level. It is shown that with certain specific actions it is possible to achieve a desirable efficiency level of the machine with the proposed cost reduction methods.

Mineraalien rikastusprosessin vesijalanjälki

Mineraalien rikastamiseen käytetään useita fysikaalisia ja kemiallisia menetelmiä. Prosessi sisältää malmin hienonnuksen, rikastuksen ja lopuksi vedenpoistamisen rikastelietteestä. Malmin rikastamiseen käytetään muun muassa vaahdotusta, liuotusta, magneettista rikastusta ja tiheyseroihin perustuvia rikastusmenetelmiä. Rikastuslietteestä voidaan poistaa vettä sakeuttamalla ja suodattamalla. Rikastusprosessin ympäristövaikutuksia voidaan arvioida laskemalla tuotteen vesijalanjälki, joka kertoo valmistamiseen kulutetun veden määrän. Tässä kirjallisuustyössä esiteltiin mineraalien käsittelymenetelmiä sekä prosessijätevesien puhdistusmenetelmiä. Kirjallisuuslähteiden pohjalta selvitettiin Pyhäsalmen kaivoksella valmistetun kuparianodin vesijalanjälki sekä esitettiin menetelmiä, joilla prosessiin tarvittavan raakaveden kulutusta voitaisiin vähentää. Pyhäsalmella kuparirikasteesta valmistetun kuparianodin vesijalanjälki on 240 litraa H2O ekvivalenttia tuotettua tonnia kohden. Pyhäsalmen prosessin raakaveden kulutusta voidaan vähentää lisäämällä sisäistä vedenkierrätystä. Kalsiumsulfaatin saostuminen putkiin ja pumppuihin on ilmentynyt ongelmaksi vedenkierrätyksen lisäämisessä. Kalsiumsulfaattia voidaan erottaa vedestä membraaneihin, ioninvaihtoon ja sähkökemiaan perustuvilla tekniikoilla. Vaihtoehdossa, jossa johdetaan kaikista kolmesta vaahdotuksesta saatavat rikastuslietteen ja rikastushiekan sakeutuksien ylitteet sekä suodatuksien suodosvedet samaan vedenkäsittelyyn voidaan kattaa arviolta noin 65 % koko veden tarpeesta. Raakavettä säästetään vuodessa 3,4 Mm^3 ja samalla rikastushiekka-altaiden tarvittava koko pienenee, joka vähentää ympäristöriskejä.

Level Set Methods in Image Processing

This thesis gives an overview of the use of the level set methods in the field of image science. The similar fast marching method is discussed for comparison, also the narrow band and the particle level set methods are introduced. The level set method is a numerical scheme for representing, deforming and recovering structures in an arbitrary dimensions. It approximates and tracks the moving interfaces, dynamic curves and surfaces. The level set method does not define how and why some boundary is advancing the way it is but simply represents and tracks the boundary. The principal idea of the level set method is to represent the N dimensional boundary in the N+l dimensions. This gives the generality to represent even the complex boundaries. The level set methods can be powerful tools to represent dynamic boundaries, but they can require lot of computing power. Specially the basic level set method have considerable computational burden. This burden can be alleviated with more sophisticated versions of the level set algorithm like the narrow band level set method or with the programmable hardware implementation. Also the parallel approach can be used in suitable applications. It is concluded that these methods can be used in a quite broad range of image applications, like computer vision and graphics, scientific visualization and also to solve problems in computational physics. Level set methods and methods derived and inspired by it will be in the front line of image processing also in the future.

Application of capillary electrophoretic methods for determining carbohydrates and aliphatic carboxylic acids formed during wood processing

Knowledge of the behaviour of cellulose, hemicelluloses, and lignin during wood and pulp processing is essential for understanding and controlling the processes. Determination of monosaccharide composition gives information about the structural polysaccharide composition of wood material and helps when determining the quality of fibrous products. In addition, monitoring of the acidic degradation products gives information of the extent of degradation of lignin and polysaccharides. This work describes two capillary electrophoretic methods developed for the analysis of monosaccharides and for the determination of aliphatic carboxylic acids from alkaline oxidation solutions of lignin and wood. Capillary electrophoresis (CE), in its many variants is an alternative separation technique to chromatographic methods. In capillary zone electrophoresis (CZE) the fused silica capillary is filled with an electrolyte solution. An applied voltage generates a field across the capillary. The movement of the ions under electric field is based on the charge and hydrodynamic radius of ions. Carbohydrates contain hydroxyl groups that are ionised only in strongly alkaline conditions. After ionisation, the structures are suitable for electrophoretic analysis and identification through either indirect UV detection or electrochemical detection. The current work presents a new capillary zone electrophoretic method, relying on in-capillary reaction and direct UV detection at the wavelength of 270 nm. The method has been used for the simultaneous separation of neutral carbohydrates, including mono- and disaccharides and sugar alcohols. The in-capillary reaction produces negatively charged and UV-absorbing compounds. The optimised method was applied to real samples. The methodology is fast since no other sample preparation, except dilution, is required. A new method for aliphatic carboxylic acids in highly alkaline process liquids was developed. The goal was to develop a method for the simultaneous analysis of the dicarboxylic acids, hydroxy acids and volatile acids that are oxidation and degradation products of lignin and wood polysaccharides. The CZE method was applied to three process cases. First, the fate of lignin under alkaline oxidation conditions was monitored by determining the level of carboxylic acids from process solutions. In the second application, the degradation of spruce wood using alkaline and catalysed alkaline oxidation were compared by determining carboxylic acids from the process solutions. In addition, the effectiveness of membrane filtration and preparative liquid chromatography in the enrichment of hydroxy acids from black liquor was evaluated, by analysing the effluents with capillary electrophoresis.

Paper Surface Image Processing for Roughness Quality Measurement

Diplomityössä on käsitelty paperin pinnankarkeuden mittausta, joka on keskeisimpiä ongelmia paperimateriaalien tutkimuksessa. Paperiteollisuudessa käytettävät mittausmenetelmät sisältävät monia haittapuolia kuten esimerkiksi epätarkkuus ja yhteensopimattomuus sileiden papereiden mittauksissa, sekä suuret vaatimukset laboratorio-olosuhteille ja menetelmien hitaus. Työssä on tutkittu optiseen sirontaan perustuvia menetelmiä pinnankarkeuden määrittämisessä. Konenäköä ja kuvan-käsittelytekniikoita tutkittiin karkeilla paperipinnoilla. Tutkimuksessa käytetyt algoritmit on tehty Matlab® ohjelmalle. Saadut tulokset osoittavat mahdollisuuden pinnankarkeuden mittaamiseen kuvauksen avulla. Parhaimman tuloksen perinteisen ja kuvausmenetelmän välillä antoi fraktaaliulottuvuuteen perustuva menetelmä.

Methods of Design and Testing of Real Time Systems

Teollusuussovelluksissa vaaditaan nykyisin yhä useammin reaaliaikaista tiedon käsittelyä. Luotettavuus on yksi tärkeimmistä reaaliaikaiseen tiedonkäsittelyyn kykenevän järjestelmän ominaisuuksista. Sen saavuttamiseksi on sekä laitteisto, että ohjelmisto testattava. Tämän työn päätavoitteena on laitteiston testaaminen ja laitteiston testattavuus, koska luotettava laitteistoalusta on perusta tulevaisuuden reaaliaikajärjestelmille. Diplomityössä esitetään digitaaliseen signaalinkäsittelyyn soveltuvan prosessorikortin suunnittelu. Prosessorikortti on tarkoitettu sähkökoneiden ennakoivaa kunnonvalvontaa varten. Uusimmat DFT (Desing for Testability) menetelmät esitellään ja niitä sovelletaan prosessorikortin sunnittelussa yhdessä vanhempien menetelmien kanssa. Kokemukset ja huomiot menetelmien soveltuvuudesta raportoidaan työn lopussa. Työn tavoitteena on kehittää osakomponentti web -pohjaiseen valvontajärjestelmään, jota on kehitetty Sähkötekniikan osastolla Lappeenrannan teknillisellä korkeakoululla.

New kernel functions and learning methods for text and data mining

Recent advances in machine learning methods enable increasingly the automatic construction of various types of computer assisted methods that have been difficult or laborious to program by human experts. The tasks for which this kind of tools are needed arise in many areas, here especially in the fields of bioinformatics and natural language processing. The machine learning methods may not work satisfactorily if they are not appropriately tailored to the task in question. However, their learning performance can often be improved by taking advantage of deeper insight of the application domain or the learning problem at hand. This thesis considers developing kernel-based learning algorithms incorporating this kind of prior knowledge of the task in question in an advantageous way. Moreover, computationally efficient algorithms for training the learning machines for specific tasks are presented. In the context of kernel-based learning methods, the incorporation of prior knowledge is often done by designing appropriate kernel functions. Another well-known way is to develop cost functions that fit to the task under consideration. For disambiguation tasks in natural language, we develop kernel functions that take account of the positional information and the mutual similarities of words. It is shown that the use of this information significantly improves the disambiguation performance of the learning machine. Further, we design a new cost function that is better suitable for the task of information retrieval and for more general ranking problems than the cost functions designed for regression and classification. We also consider other applications of the kernel-based learning algorithms such as text categorization, and pattern recognition in differential display. We develop computationally efficient algorithms for training the considered learning machines with the proposed kernel functions. We also design a fast cross-validation algorithm for regularized least-squares type of learning algorithm. Further, an efficient version of the regularized least-squares algorithm that can be used together with the new cost function for preference learning and ranking tasks is proposed. In summary, we demonstrate that the incorporation of prior knowledge is possible and beneficial, and novel advanced kernels and cost functions can be used in algorithms efficiently.