Kaikki aineistot
Lisää
Extreme Learning Machine (ELM) and Minimal Learning Machine (MLM) are nonlinear and scalable machine learning techniques with randomly generated basis. Both techniques share a step where a matrix of weights for the linear combination of the basis is recovered. In MLM, the kernel in this step corresponds to distance calculations between the training data and a set of reference points, whereas in ELM transformation with a sigmoidal activation function is most commonly used. MLM then needs additional interpolation step to estimate the actual distance-regression based output. A natural combination of these two techniques is proposed here, i.e., to use a distance-based kernel characteristic in MLM in ELM. The experimental results show promising potential of the proposed technique.Extreme Learning Machine (ELM) and Minimal Learning Machine (MLM) are nonlinear and scalable machine learning techniques with randomly generated basis. Both techniques share a step where a matrix of weights for the linear combination of the basis is recovered. In MLM, the kernel in this step corresponds to distance calculations between the training data and a set of reference points, whereas in ELM transformation with a sigmoidal activation function is most commonly used. MLM then needs additional interpolation step to estimate the actual distance-regression based output. A natural combination of these two techniques is proposed here, i.e., to use a distance-based kernel characteristic in MLM in ELM. The experimental results show promising potential of the proposed technique.
Industrial, electrical power generation, and transportation systems, to name but a few, rely heavily on power electronics to control and convert electrical power. Each of these systems, when encountering an unexpected failure, can cause significant financial losses, or even an emergency. A condition monitoring system would help to alleviate these concerns, but for the time being, there is no generally accepted and widely adopted method for power electronics. Acoustic emission is used as a failure precursor in many applications, but it has not been studied in power electronics so far. In this doctoral dissertation, observations of acoustic emission in power semiconductor components are presented. The acoustic emissions are caused by the switching operation and failure of power transistors. Three types of acoustic emission are observed. Furthermore, aspects related to the measurement and detection of acoustic phenomena are discussed. These include sensor performance and mechanical construction of experimental setups. The results presented in this dissertation are the outset of a research program where it will be determined whether an acoustic-emission-based condition monitoring method can be developed.
Societal impact as a phenomenon is closely connected to scientific practices, thus its expressions have been part of science for a long while. However, as a concept and as an object of assessment societal impact of research has become common during the 21st century due to emphasis’ in science policy and changes in research funding. This master’s thesis focuses on statements concerning societal impact of research in funded grant proposal research plans. Societal impact of research is defined, in the context of the thesis, as contributions that research processes and outputs make to society. Impact is understood as effects formed by flows of knowledge that emerge and manifest in interactions and artefacts related to social practices. The approach is based in practice theory which emphasises the diverse, contingent, and connected nature of social phenomena and entails that the analysed research plans are created, maintained, and changed in interconnected practices of research, research funding and science policy. The research data consists of all grant proposal research plans from the years 2006 and 2017 funded by the Academy of Finland’s Research Council for Culture and Society. The research plans are from Academy Projects, Academy Research Fellow projects, and postdoctoral research projects. The research plans are analysed using deductive content analysis and quantification to examine (dis)similarities, variations and patterns in the statements concerning societal impact. Observations are categorized as practices and targets of societal impact. Practices are further categorized into two impact dimensions which each contain two dimension specific impact forms. Targets of societal impact are categorized separately to general and occupation related categories. The results indicate that plans to promote knowledge flows are typical and varied among social scientists that are present in the data. Knowledge flows are slightly more likely to be planed through interaction than artefacts. The most typical interaction related practices were observed in the form of information dissemination, discussions, education, and training. As for practices related to artefacts the most typical ones concerned the form of media artefacts meaning primarily the production of popular, occupational, and open access publications. Practices were most often directed towards public sector and second most often towards private sector. No systematic differences were observed between different project types. The years 2006 and 2017, on the other hand, differed among each category used to examine societal impact. On average the prevalence and length of statements concerning societal impact increase as well as frequency and variety of practices, targets, forms, and dimensions of societal impact. The results represent similarities, varieties and systematic changes in grant proposal writing practices. They are also interpreted to indicate impact statements typical to social scientists as well as changes in research plan contents that seem to correspond with the 20th century impact agenda of science policy.
As the power specifications of power electronic devices has increased, efficiency has become one of their most important features. At high power levels, even relatively low losses are significant and cause unnecessary energy costs and need to remove heat from the devices. For these reasons customers demand devices that operate at high efficiency, so manufacturers pursue to make such devices. Simulation models are valuable tools in the device design process. Efficiency optimization requires that the losses can be modeled, so that the effect of component choises, control methods and main circuit topologies on efficiency can be estimated. In this thesis we take a look at the different insulated gate bipolar transistor (IGBT) models, and their applicability for modeling the losses present in IGBTs is assessed. Further, a model is compared against measurements, and the requirements directed at the models are considered.
Opinnäytetyön tavoitteena oli suunnitella Stora Enso Oy:n Veitsiluodon paperitehtaalle uudet ohjaukset logiikan avulla sellujauhimille. Sellujauhimilla on ilmennyt paljon vikoja tietämättä täysin mistä ne johtuvat. Logiikan avulla on mahdollista seurata trendikäyriä ja saada selkeämmät ohjaukset, joita voi seurata reaaliaikaisesti. Trendeissä voidaan esimerkiksi seurata hälytysten aiheuttamia jauhimen laukaisuja tai terämoottorin ohjauksia. Logiikkaan pääsee käsiksi sähköverstaalta ja paikan päältä. Nykyiset kontaktori- ja releohjaukset ovat jo vanhat ja erittäin vaikeat vikoja korjatessa. Suunnitelmaan tehtiin uudet selkeämmät piirikaaviot, logiikkakeskuksen layoutin, komponenttien valintoja, logiikan ohjelman ja uusia tarvittavia kaapeleita. Dokumentointi tehtiin CADS-ohjelmistolla ja logiikkakaaviot TIA Portal-ohjelmistolla. Työn tuloksena syntyi tarpeelliset piirikaaviot, logiikkaohjelmat ja laitesuunnittelut sellujauhimien automatisoinnin uudistamiseksi. Työssä otettiin myös huomioon energiatehokkuus ja jauhatuksen merkitys prosessissa.
The extreme learning machine (ELM) and the minimal learning machine (MLM) are nonlinear and scalable machine learning techniques with a randomly generated basis. Both techniques start with a step in which a matrix of weights for the linear combination of the basis is recovered. In the MLM, the feature mapping in this step corresponds to distance calculations between the training data and a set of reference points, whereas in the ELM, a transformation using a radial or sigmoidal activation function is commonly used. Computation of the model output, for prediction or classification purposes, is straightforward with the ELM after the first step. In the original MLM, one needs to solve an additional multilateration problem for the estimation of the distance-regression based output. A natural combination of these two techniques is proposed and experimented here: to use the distance-based basis characteristic in the MLM in the learning framework of the regularized ELM. In other words, we conduct ridge regression using a distance-based basis. The experimental results characterize the basic features of the proposed technique and surprisingly, indicate that overlearning with the distance-based basis is in practice avoided in classification problems. This makes the model selection for the proposed method trivial, at the expense of computational costs.
The classical Taylor’s formula is an elementary tool in mathematical analysis and function approximation. Its role in the optimization theory, whose data-driven variants have a central role in machine learning training algorithms, is well-known. However, utilization of Taylor’s formula in the derivation of new machine learning methods is not common and the purpose of this article is to introduce such use cases. Both a feedforward neural network and a recently introduced distance-based method are used as data-driven models. We demonstrate and assess the proposed techniques empirically both in unsupervised and supervised learning scenarios.
A combination of Extreme Learning Machine (ELM) and Minimal Learning Machine (MLM)—to use a distance-based basis from MLM in the ridge regression like learning framework of ELM—was proposed in [8]. In the further experiments with the technique [9], it was concluded that in multilabel classification one can obtain a good validation error level without overlearning simply by using the whole training data for constructing the basis. Here, we consider possibilities to reduce the complexity of the resulting machine learning model, referred as the Extreme Minimal Leaning Machine (EMLM), by using a bidirectional sampling strategy: To sample both the feature space and the space of observations in order to identify a simpler EMLM without sacrificing its generalization performance.
In this thesis a device was designed, which performs Clarke's transformation to a three phase measurement signal in real time. In this context, a three phase measurement signal is defined as a signal representing the phase currents or voltages measured from a three phase system. The output is a two phase signal which can be examined with an oscilloscope. By using the oscilloscope in the xy mode, the current or voltage circle is drawn on the display. The device can be used in testing variable frequency drives, for instance, to allow immediate observation of changes in the output of the drive caused by changes in the drive's parameters.