| 1. |
- Alonso, Pedro, 1986-
(författare)
-
Faster and More Resource-Efficient Intent Classification
- 2020
-
Licentiatavhandling (övrigt vetenskapligt/konstnärligt)abstract
- Intent classification is known to be a complex problem in Natural Language Processing (NLP) research. This problem represents one of the stepping stones to obtain machines that can understand our language. Several different models recently appeared to tackle the problem. The solution has become reachable with deep learning models. However, they have not achieved the goal yet.Nevertheless, the energy and computational resources of these modern models (especially deep learning ones) are very high. The utilization of energy and computational resources should be kept at a minimum to deploy them on resource-constrained devices efficiently.Furthermore, these resource savings will help to minimize the environmental impact of NLP.This thesis considers two main questions.First, which deep learning model is optimal for intent classification?Which model can more accurately infer a written piece of text (here inference equals to hate-speech) in a short text environment. Second, can we make intent classification models to be simpler and more resource-efficient than deep learning?.Concerning the first question, the work here shows that intent classification in written language is still a complex problem for modern models.However, deep learning has shown successful results in every area it has been applied.The work here shows the optimal model that was used in short texts.The second question shows that we can achieve results similar to the deep learning models by more straightforward solutions.To show that, when combining classical machine learning models, pre-processing techniques, and a hyperdimensional computing approach.This thesis presents a research done for a more resource-efficient machine learning approach to intent classification. It does this by first showing a high baseline using tweets filled with hate-speech and one of the best deep learning models available now (RoBERTa, as an example). Next, by showing the steps taken to arrive at the final model with hyperdimensional computing, which minimizes the required resources.This model can help make intent classification faster and more resource-efficient by trading a few performance points to achieve such resource-saving.Here, a hyperdimensional computing model is proposed. The model is inspired by hyperdimensional computing and its called ``hyperembed,'' which shows the capabilities of the hyperdimensional computing paradigm.When considering resource-efficiency, the models proposed were tested on intent classification on short texts, tweets (for hate-speech where intents are to offend or not to), and questions posed to Chatbots.In summary, the work proposed here covers two aspects. First, the deep learning models have an advantage in performance when there are sufficient data. They, however, tend to fail when the amount of available data is not sufficient. In contrast to the deep learning models, the proposed models work well even on small datasets.Second, the deep learning models require substantial resources to train and run them while the models proposed here aim at trading off the computational resources spend to obtaining and running the model against the classification performance of the model.
|
|
| 2. |
- Brännvall, Rickard, 1975-
(författare)
-
Machine learning based control of small-scale autonomous data centers
- 2020
-
Licentiatavhandling (övrigt vetenskapligt/konstnärligt)abstract
- The low-latency requirements of 5G are expected to increase the demand for distributeddata storage and computing capabilities in the form of small-scale data centers (DC)located at the edge, near the interface between mobile and wired networks. These edgeDC will likely be of modular and standardized designs, although configurations, localresource constraints, environments and load profiles will vary and thereby increase theDC infrastructure diversity. Autonomy and energy efficiency are key objectives for thedesign, configuration and control of such data centers. Edge DCs are (by definition)decentralized and should continue operating without human intervention in the presenceof disturbances, such as intermittent power failures, failing components and overheating.Automatic control is also required for efficient use of renewable energy, batteries and theavailable communication, computing and data storage capacity.These objectives demand data-driven models of the internal thermal and electricprocesses of an autonomous edge DC, since the resources required to manually defineand optimize the models for each DC would be prohibitive. In this thesis machinelearning methods that are implemented in a modular design are evaluated for thermalcontrol of such modular DCs. Experiments with small server clusters are presented, whichwere performed in order to investigate what parameters that are important in the designof advanced control strategies for autonomous edge DC. Furthermore, recent transferlearning results are discussed to understand how to develop data driven models thatcan be deployed to modular DC in varying configurations and environmental contextswithout training from scratch.The first study demonstrates how a data driven thermal model for a small clusterof servers can be calibrated to sensor data and used for constructing a model predictivecontroller for the server cooling fan. The experimental investigations of cooling fancontrol continues in the next study which explores operational sweet-spots and energyefficient holistic control strategies. The machine learning based controller from the firststudy is then re-purposed to maintain environmental conditions in an exhaust chamberfavourable for drying apples, as part of a practical study how excess heat produced bycomputation can be used in the food processing industry. A fourth study describes theRISE EDGE lab - a test bed for small data centers - built with the intention to exploreand evaluate related technologies for micro-grids with renewable energy and batteries,5G connectivity and coolant storage. Finally the last work presented develops the modelfrom the first study towards an application for thermal based load balancing.
|
|
| 3. |
- Burman, Anton
(författare)
-
Hydraulic Modelling of Dynamics in Regulated Rivers
- 2020
-
Licentiatavhandling (övrigt vetenskapligt/konstnärligt)abstract
- The Nordic countries hold a significant portion of the European hydropower production. One advantage of hydropower is its ability to store water in reservoirs in times when the energy demand is low. The readjustment of energy production to renewable energy sources, as required by the Paris agreement, like wind power and solar power is likely going to change the role of Nordic hydropower production. Wind power and solar power are both dependent on the current weather conditions, in times when the weather is not favourable, hydropower can be used to stabilize the electricity grid. Since weather can change rapidly so will the discharge from the hydropower plants, causing hydropeaking events. Hydropeaking rapidly changes the flow conditions in proximity to the power plant. Such changes can be detrimental to the downstream habitats in and along the river. The study reach in this work is the bypass reach in Stornorrfors in the Ume River. The open-source hydrodynamic solver Delft3D is used to numerically model the flow in the study reach. To validate the simulations water level measurements have been used. The aim of the thesis is to investigate inherent damping properties in the river reach that can be used to mitigate the influence of hydropeaking scenarios. The influence of parameters such as upstream closing time, manning number distribution and hydropeaking frequency have been investigated. It is shown that the closing time drastically affects the dynamics of the wetted area. The water surface elevation exhibits a hysteresis like behaviour. Inherent damping increases with the downstream coordinate. The frequency of the flow changes affects the areas upstream more than downstream. As a result, potential habitats in the downstream parts of the reach could become more stable if more frequent flow changes occur.
|
|
| 4. |
- Cao, Zhejian, 1991-
(författare)
-
Structured Carbon-Alkaline Earth Metal Halides Composites for Ammonia Storage
- 2020
-
Licentiatavhandling (övrigt vetenskapligt/konstnärligt)abstract
- NOx (NO, NO2) is one of the most harmful air-pollutants from exhaust, resulting in series of environmental problems as well as severe healthy issues for human beings. Selective catalytic reduction (SCR) system is a common approach to eliminate NOx onboard by using ammonia as a reductant. However, ammonia storage unit has been one of the restriction factors for the NOx conversion efficiency because of insufficient ammonia dosing rate and the corrosive and hazardous nature of ammonia. Thus, a reliable ammonia storage and delivery system is of high scientific and commercial desire.In this thesis, novel composites were fabricated and studied based on MgCl2 and SrCl2, two commercial alkaline earth metal halides (AEMH) for ammonia storage. In order to reduce the melting issue and enhance the kinetics of the ammonia sorption, carbon materials, graphite (Gt) and graphene nanoplatelets aggregates (GNA) were added to MgCl2 at 1 wt.%, 10 wt.% and 20 wt.%. With ball milling and hydraulic pressing, the aforementioned carbon-MgCl2 composites were structured into pellets for various characterization. With real-time recording in the tube furnace at 1073 K, we observed that with 20% carbon additives, the pelletized composites maintained their structure with 95% mass retention, while the pure MgCl2 completely melted and disintegrated. According to the SEM images, carbon materials separated MgCl2 so that the molten MgCl2 cannot form large droplet to spread out. Furthermore, the 20 wt.% GNA-80 wt.% MgCl2 (GNA20) composites demonstrated enhanced kinetics in both absorption and desorption of ammonia, which is 83% faster in ammonia absorption and 73% faster in desorption in the first two minutes compared to the pure MgCl2. The BET surface area and mercury intrusion porosimetry results explains the kinetic elevation by the GNA by introducing extra reaction surface and nanopores as the diffusion path for ammonia. The enhancement of both structural stabilityand kinetics make the GNA20 composite a robust ammonia carrier.During the chemical absorption process, SrCl2 uptakes 8 ammonia molecules resulting in 4 times volume expansion. This dramatic expansion and shrinkage during the absorption and desorption will destroy the structure and disintegrate the SrCl2 into powder, which could bring the dust explosion risk for many applications. Based on the carbon-salts composites, a novel porous SrCl2 structure is designed and fabricated with graphene oxide as skeleton by freeze casting process. Porous SrCl2 structure is feasible for various geometries with different molds at a wide SrCl2 load from 0 wt.% to 96 wt.%. The ammonia capacity of the porous SrCl2 is linear proportional to the SrCl2 load. During the ammonia absorption and desorption cycles, the graphene oxide skeleton could self-adjust along with the volume swing to within its flexibility. This porous SrCl2 demonstrates excellent tolerance of volume swing and enhanced kinetics as a promising ammonia storage material. Our approach and results may cast light on the obstacles of structuring self-expansion and shrinkage materials as well as on enhancing the gas sorption properties.
|
|
| 5. |
- Decrozant-Triquenaux, Justine, 1994-
(författare)
-
High Temperature Tribology of Aluminium : Effect of Lubrication and Surface Engineering on Friction and Material Transfer
- 2020
-
Licentiatavhandling (övrigt vetenskapligt/konstnärligt)abstract
- Lightweight design for automotive applications has been pursued for several decades and continues to increase. The main driving forces are new and increasingly stringent emission regulations as well as the increasing popularity of electric, or hybrid, vehicles where the increased weight of the batteries need to be compensated by light weight structures. It is also critical to maintain or improve passenger safety while creating components and structures with lower weight. Materials exhibiting a high strength-to-weight ratio, such as high strength aluminium alloys, are highly interesting to realise the next generation of lightweight vehicle structures.The high strength aluminium alloys include the 5XXX, 6XXX and 7XXX series. In order to increase their formability and minimise springback induced during forming at room temperature, these alloys are preferentially formed at elevated temperatures. Different forming processes such as warm forming and hot stamping (e.g. hot forming and quenching) have been developed to enable forming of components with high geometrical complexity and mechanical properties. However, hot forming of aluminium alloys leads to a challenging tribological interface. Aluminium alloys are ductile and reactive metals, prone to severe adhesion (also termed as seizure or galling) when sliding against a harder metallic counter surface. Aluminium transfer to the forming dies affect the tool lifetime and impacts the quality of the formed component which leads to significant maintenance costs and reduced productivity. These are the main limitations that hinder the implementation of hot aluminium forming for mass production.Lubrication as well as surface engineering strategies are potential methods to control friction and wear in the hot aluminium-tool steel interface. Solid lubricants such as graphite and hexagonal boron nitride (hBN) have been studied for aluminium forming. Polymer-based lubricants are also increasingly evaluated for high-temperature applications. Surface engineering techniques includes both the control of the tool topography and the use of protective coatings. Surface roughness has been observed as a crucial parameter in the initiation of aluminium transfer to the counter surface. PVD and CVD thin coatings are increasingly studied as ways to alleviate galling. Among others, CrN and DLC coatings are known to reduce adhesion when sliding against aluminium. Despite the research efforts in this field, there is still lack of systematic studies where synergistic effects of lubrication, surface topography and coatings are explored in the context of hot aluminium forming.The aim of this research is to enhance the understanding of the tribological behaviour of aluminium sliding against tool steel at elevated temperatures. The effect of tool steel composition, surface roughness (as-received and post-polished), and PVD surface coating composition (CrTiN, CrAlN, CrN and DLC ta-C) has been evaluated under dry and lubricated conditions (hBN-based and polymer-based).High temperature tribological tests were carried out in a reciprocating sliding flat-on-flat configuration. In dry conditions, the aluminium-tool steel tribosystem is characterised by severe adhesive wear and high friction. Effective control of friction and wear was found to be highly dependent on the ability of the lubricant to remain in the contact zone. The combined use of a polymer-based lubricant with post-polished surface topography on a PVD coated tool led to the best improvements in terms of frictional stability and reduced material transfer. This was mainly attributed to prevention of direct contact between the tool material and aluminium together with minimised mechanically initiated material transfer. Post-polished uncoated tool steels resulted in the development of a protective tribolayer in the contact and together with flattening of the aluminium surface, led to friction and wear reduction. In case of post-polished PVD coatings, the lubricant entrapment in the contact zone as well as the development of mechanically mixed layers on the aluminium lowered friction and wear.
|
|
| 6. |
- El-Tawil, Asmaa
(författare)
-
Bio-coal as an alternative reducing agent in the blast furnace
- 2020
-
Licentiatavhandling (övrigt vetenskapligt/konstnärligt)abstract
- The steel industry is aiming to reduce CO2 emissions by different means; in the short-term, by replacing fossil coal with highly reactive carbonaceous material like bio-coal (pretreated biomass) and, in the longer term, by using hydrogen. The use of bio-coal as part of top charged briquettes also containing iron oxide has the potential to lower the thermal reserve zone temperature of the Blast furnace (BF) and, due to improved gas efficiency, thereby give a high replacement ratio to coke.In order to select a suitable bio-coal to be contained in agglomerates with iron oxide, the current study aims at investigating the devolatilization behavior and related kinetics of different types of bio-coals. In addition, the aim is to investigate the self-reduction behavior of bio-coal-containing iron ore composite under inert condition and simulated blast furnace thermal profile.In the BF the temperature of the top-charged material will increase rather quickly during the descent in the upper part. Ideally, all the carbon and hydrogen contained in the top-charged bio-coal should contribute to the reduction. The devolatilization of bio-coal is thus important to understand and to compare between different types of bio-coal.To explore the devolatilization behavior for different materials, a thermogravimetric analyzer equipped with a quadrupole mass spectrometer was used to monitor the weight loss and off-gases during non-isothermal tests for bio-coals having different contents of volatile matter. The samples were heated in an inert atmosphere up to 1200°C at three different heating rates: 5, 10 and 15°C/min. The thermogravimetric data were evaluated by using the Kissinger–Akahira–Sonuse (KAS) iso-conversational model and the activation energy was determined as a function of the conversion degree. Bio-coals with both low and high content of volatile matter can produce reducing gases that can contribute to the reduction of iron oxide in bio-agglomerates. Bio-coals containing a higher content of catalyzing components such as CaO and K2O will enhance the devolatilization and release of volatile matter at a lower temperature. The self–reduction of composites was investigated by thermogravimetric analyses in argon atmosphere up to 1100°C and evolved gases were monitored by means of quadrupole mass spectroscopy. Composites with and without 10% bio-coal and sufficient coke breeze to keep the C/O molar ratio equal to one were mixed and Portland cement was used as a binder. To explore the effect of added bio-coals, interrupted vertical tube furnace tests were conducted in a nitrogen atmosphere at temperatures selected based on thermogravimetric results, using a similar thermal profile as for the thermogravimetric analyzer. The variation between fixed carbon, volatile matter contents and ash composition for different types of bio-coal influences the reduction of iron oxide.The results showed that the self-reduction proceeds more rapidly in the bio-coal-containing composite and that the volatile matter could have contributed to the reduction. The self-reduction of bio-coal-containing composites started at 500°C, while it started at 740°C with coke as the only carbon source. The hematite was successfully reduced to metallic iron at 850°C with bio-coal present as a reducing agent, but not until 1100°C when using coke.Use of bio-coal with high content of volatile matter but low content of catalyzing elements as potassium, sodium and calcium in bio-agglomerates for the BF can be recommended because it enhances the self-reduction of iron oxide, e.g., wustite was detected by XRD analysis in samples treated up to 680°C. Bio-coal with low content of volatile matter, low alkalis, low phosphorous and high content of fixed carbon will also be suitable to use in the BF.
|
|
| 7. |
- Fischer, Robert, 1966-
(författare)
-
Investigation into sustainable energy systems in Nordic municipalities
- 2020
-
Licentiatavhandling (övrigt vetenskapligt/konstnärligt)abstract
- Municipal energy systems in Nordic environments face multiple challenges: the cold climate, large-scale industries, a high share of electric heating and long distances drive energy consumption. While actions on the demand side minimize energy use, decarbonization efforts in mining, industries, the heating and the transport sector can increase the consumption of electricity and biofuels. Continued growth of intermittent wind and solar power increases supply, but the planned phase out of Swedish nuclear power will pose challenges to the reliability of the electricity system in the Nordic countries. Bottlenecks in the transmission and distribution grids may restrict a potential growth of electricity use in urban areas, limit new intermittent supply, peak electricity import and export. Environmental concerns may limit growth of biomass use. Local authorities are committed in contributing to national goals on mitigating climate change, while considering their own objectives for economic development, increased energy self-sufficiency and affordable energy costs.Given these circumstances, this thesis investigates existing technical and economic potentials of renewable energy (RE) resources in the Nordic countries with a focus on the northern counties of Finland, Norway and Sweden. The research further aims to provide sets of optimal solutions for sustainable Nordic municipal energy systems, where the interaction between major energy sectors are studied, considering multiple objectives of minimizing annual energy system costs and reducing carbon emissions as well as analyzing impacts on peak electricity import and export.This research formulates an integrated municipal energy system as a multi-objective optimization problem (MOOP), which is solved by interfacing the energy system simulation tool EnergyPLAN with a multi-objective evolutionary algorithm (MOEA) implemented in Matlab. In a first step, the integration or coupling of electricity and heating sectors is studied, and in a second step, the study inquires the impacts of an increasingly decarbonized transport sector on the energy system. Sensitivity analysis on key economic parameters and on different grid emission factors is performed. Piteå (Norrbotten County, Sweden) is a typical Nordic municipality, which serves as a case study for this research.The research concludes that significant techno-economic potentials exist for the investigated resources. Optimization results show that CO2 emissions of a Nordic municipal energy system can be reduced by about 60% without a considerable increase in total energy system costs and that peak electricity import can be reduced by up to 38%. The outlook onto 2030 shows that the transport sector could be composed of high electrification shares and biofuels. Technology choices for optimal solutions are highly sensitive to electricity prices, discount rates and grid emission factors.The inquiries of this research provide important insights about carbon mitigation strategies for integrated energy sectors within a perspective on Nordic municipalities. Future work will refine the transport model, develop and apply a framework for multi-criteria decision analysis (MCDA) enabling local decision makers to determine a technically and economically sound pathway based on the optimal alternatives provided, and analyze the existing policy framework affecting energy planning of local authorities.
|
|
| 8. |
|
|
| 9. |
|
|
| 10. |
- Hannl, Thomas Karl, M.Sc. 1993-
(författare)
-
Phosphorus recovery from sewage sludge fluidized bed gasification processes
- 2020
-
Licentiatavhandling (övrigt vetenskapligt/konstnärligt)abstract
- One of the most sustainable pathways of sewage sludge treatment in recent years has been thermal conversion. The benefits of thermal treatment of sewage sludge are the recovery of energy or valuable chemical products, the destruction of harmful organic compounds, the separation of heavy metals from the P-rich coarse ash fraction, and the decreased and sanitized ash volume. The ashes created by these thermal conversion processes of sewage sludge are often rich in P that is mostly present in minerals with low plant-availability such as apatite. Due to the enrichment of P in the created ashes, a variety of post-processing steps have been developed to recover P from sewage sludge ashes. One proven way for the sus-tainable recovery of P from such ashes is thermal post-processing with alkaline salts, e.g., Na2SO4 or K2CO3, which was able to transform less plant-available phosphates in the sewage sludge into more plant-available alkali-bearing phos-phates. Based on these results, one could facilitate creating these phosphates with enhanced plant-availability by providing the chemical potential to form them already during the thermal conversion process of sewage sludge. This thesis aims to increase the current knowledge about the ash transformation processes of P and to suggest suitable process parameters for the alteration of the phosphate speciation in sewage sludge ashes by co-conversion with alkaline-rich agricultural residues. More specifically, the possibility of incorporating K derived from agricultural residues in phosphate structures derived from sewage sludge was evaluated with respect to the influence of the process temperature, the conver-sion atmosphere, and the fuel mixture. The studied parameters were chosen to generate knowledge relevant for fluidized bed gasification processes, with a spe-cial focus on dual fluidized bed (DFB) gasification systems.The applicability of feldspar bed materials in fluidized bed gasification systems was investigated to enable the substitution of the commonly used olivine, which often contains heavy metals (potentially contaminating recovered ashes), and quartz, which is very reactive towards fuel-derived K and potentially leads to bed material fragmentation and bed agglomeration (Paper I & II). Subsequently, the thermodynamic potential for the alteration of the P-species in sewage sludge ash during co-combustion and co-gasification processes with agricultural residues was investigated (Paper III). Thereafter, an experimental evaluation of the ash transformation chemistry in thermal conversion processes of sewage sludge with different types of alkali-rich agricultural residues in temperatures relevant for flu-idized bed technology was conducted (Paper IV & V).The methodology employed was chosen with respect to the state of technology of the specific investigated process. Paper I & II applied SEM, EDS, XRD, and thermodynamic equilibrium modeling for bed material samples derived from an industrial indirect gasifier. Paper III applied thermodynamic equilibrium calcula-tions to theoretically evaluate ash compositions resulting from co-conversion of sewage sludge and agricultural residues. Paper IV & V employed SEM, EDS, ICP-AES/MS, XRD, and thermochemical modeling on ash samples derived from single pellet lab-scale experiments.The results obtained by analysis of bed material from indirect wood gasification showed the difference in interaction mechanism for K-feldspar and Na-feldspar, most notably the enhanced disintegration of Na-feldspar by K originating from the fuel (Paper I & II). Thermodynamic models employed for fuel mixtures of sewage sludge and agricultural residues showed the thermodynamic preference for the formation of the desired alkali-bearing phosphates (Paper III). Experi-ments conducted with these fuel mixtures (Paper IV & V) supported the theo-retical findings, and the influence of temperature and process conditions could be obtained. However, practical investigations also showed that attainment of the desired ash composition is subject to significant restrictions.Derived from the elaborated results and discussions, it was possible to assess the critical process and fuel parameters for the development of up-scaled gasification processes focusing on the conversion of sewage sludge with the aim of creating improved phosphate formation in the ash. The selection of a suitable bed material in fluidized bed conversion and the transformation mechanisms defining the ash chemistry were found to be of vital importance for future applications. The pur-suit of the predefined aims in reference to P-recovery from sewage sludge has led to a multitude of suggestions for suitable process parameters that must be ad-dressed in future bench- and pilot-scale experimental runs.
|
|
