Motivated by A. Z. Petrov's classification of four-dimensional Lorentzian metrics, we provide an algebraic classification of the isometry-isotropy pairs of four-dimensional pseudo-Riemannian metrics admitting local slices with five-dimensional isometries contained in the Lorentz algebra. A purely Lie algebraic approach is applied with emphasis on the use of Lie theoretic invariants to distinguish invariant algebra-subalgebra pairs. This method yields an algorithm for identifying isometry-isotropy pairs subject to the aforementioned constraints.

Enhanced biodegradation using carbon donor and microbial addition is being considered as a possible remediation technique for a trichloroethylene (TCE) contaminated area in Sunset, Utah, west of the source area on Hill Air Force Base. As a precursor to any in situ remediation attempts, several laboratory treatability experiments are being conducted, including the construction of microcosms and flow-through columns. Nine large-scale flow-through columns were built using site groundwater and aquifer material. Bromide tracer tests were conducted to establish and understand the hydraulic conditions within the columns prior to the commencement of the TCE biodegradation experiments. Four predictive models were created to show potential degradation scenarios in the columns and in the field using microcosm data for various system treatments. Treatments selected for modeling indicated that carbon addition alone is insufficient in stimulating dechlorination of TCE. Microbial amendments will be necessary in the column systems when the TCE dechlorination experiments begin.

京都大学 / 0048 / 新制・課程博士 / 博士(理学) / 甲第20602号 / 理博第4317号 / 新制||理||1620(附属図書館) / 京都大学大学院理学研究科地球惑星科学専攻 / (主査)教授 山路 敦, 教授 山 明, 准教授 伊藤 正一 / 学位規則第4条第1項該当 / Doctor of Science / Kyoto University / DGAM

No description available.

The aim of explainable machine learning is to aid humans in understanding how exactly complex machine learning models work. Machine learning models have offered great performance in various areas. However, the mechanisms behind how the model works and how decisions are being made remain unknown. This specific constraint increases the user’s hesitation to trust the results of the model and even to improve their performance further. Counterfactual explanation is one method to offer explainability in machine learning by indicating what would have happened if the input of a model was modified in a specific way. Fraud is the action of acquiring something from someone else in a dishonest manner. Companies’ and organizations’ vulnerability to malicious actions has been increasing due to the development of digitalization. Machine learning applications have been successfully put in place to tackle fraudulent actions. However, the severity of the impact of fraudulent actions has highlighted the need for further scientific exploration of the topic. The current research will attempt to do so by studying counterfactual explanations related to fraud detection. Latent-CF is a method for counterfactual generation that utilizes an autoencoder and gradient descent in its latent space. LatentCF++ is an extension of Latent-CF. It utilizes a classifier and an autoencoder. The aim is to perturb the encoded latent representation through a gradient descent optimization for counterfactual generation so that the initially undesired class is then classified with the desired prediction. Compared to Latent-CF, LatentCF++ uses Adam optimization and adds further constraints to ensure that the generated counterfactual’s class probability surpasses the set decision boundary. The research question the current thesis addresses is: “To what extent can LatentCF++ provide reliable counterfactual explanations in fraud detection?”. In order to provide an answer to this question, the study is applying an experiment to implement a new application of LatentCF++. The current experiment utilizes a onedimensional convolutional neural network as a classifier and a deep autoencoder for counterfactual generation in fraud data. This study reports satisfying results regarding counterfactual explanation production of LatentCF++ on fraud detection. The classification is quite accurate, while the reconstruction loss of the deep autoencoder employed is very low. The validity of the counterfactual examples produced is lower than the original study while the proximity is lower. Compared to baseline models, k-nearest neighbors outperform LatentCF++ in terms of validity and Feature Gradient Descent in terms of proximity.

The Two Element in Series (TEIS) and Two Zone models stand out as powerful tools that enable deeper understanding of compressor stage designs after they have been tested. The insights gained from these investigations have aided in improving new stage designs. Up to now, it has only been possible to use the TEIS and Two Zone models for analysis of test data due to the inability to predict the four required input parameters for untested machines. Empirical models for the TEIS and Two Zone model input parameters, ETAa5, ETAb5, Chi5, and DELTA5p, for two different types of vaned diffusers, channel and cascade, are proposed. These models were developed with frozen impeller modeling. This is the first time that modeling the TEIS and Two Zone input parameters has been attempted for vaned diffusers and impeller-diffuser coupling was not considered in this initial investigation. The centrifugal compressor experimental data used in the model building was obtained from Concepts NREC, an industry sponsor. Each dataset provided was evaluated for quality and reliability and only the data deemed reliable were used in the model building databases. The empirical models presented are built solely on this higher quality data. Seven models are proposed for use in predicting the TEIS and Two Zone model input parameters ETAa5, ETAb5, Chi5, and DELTA5p. Models for ETAa5, ETAb5, and DELTA5p are specific to the type of vane present in the diffuser, while the model for Chi5 is common to both diffuser types. These are the first models ever built for the TEIS and Two Zone model inputs applied to channel and cascade diffusers and become a benchmark for future studies. The work with these models is not complete, however. The databases are not of a size that data could be withheld from empirical model building for the express purpose of validation. Instead the model performance is evaluated by applying all of the models, simultaneously, to the database from which they were built. The determination of the effectiveness of the combined modeling is based on the average error across the entire speedline. The models proved to be effective and a contributing step to employing such models for use in future compressor design.

The Two-Element In Series (TEIS) and Two-Zone models have been used successfully for over twenty years to model test data for radial flow compressors and pumps. The models can also be used to predict the performance of new machines provided that the model inputs can be accurately specified. Unfortunately, use of the TEIS and Two-Zone models as a predictive tool has been limited because an accurate and broadly applicable method of predicting the modeling parameters, etaA, etaB, chi and d2p does not exist. Empirical models have been developed to predict the TEIS and Two-Zone modeling parameters based on a large database of centrifugal pump and compressor test results. These test data were provided by ConceptsNREC and have been collected over the past 40 years. The database consists of a wide range of machines including some that were designed and tested by ConceptsNREC and others from the open literature. Only cases with a vaneless diffuser or volute have been included in the analysis to avoid any possible impeller-diffuser interactions. From the database, models for all of the TEIS and Two-Zone parameters have been derived using basic regression techniques. Three different models are proposed for each of the two TEIS modeling parameters, etaA and etaB. One model for pumps, another for compressors, and a combined model applicable for all machines is given. For the Two-zone parameters, chi and d2p, a single set of models was developed to represent the design point performance and another showing how chi and d2p vary off-design. The combined models for etaA and EtaB are 30% and 70% more accurate than the current state-of-the-art models, respectively. The new models account for the variance in chi and d2p at off-design flow conditions and further refine the accuracy of the overall prediction by correctly modeling the loss mechanisms in the impeller passage. Validation work has shown that the set of models that predict etaA, etaB, chi and d2p can be solved to consistently produce sensible results and yield a reasonable "blind" prediction of the performance of a wide range of radial compressors and pumps. These models constitute the first broadly applicable method for predicting the required TEIS and Two-Zone variables and are sufficiently accurate to provide initial performance estimates of new impeller designs

The mechanism of flame propagation in fuel beds of wildland fires is important to understand and quantify fire spread rates. Fires spread by radiative and convective heating and often require direct flame contact to achieve ignition. The flame interface in an advancing fire is unsteady and turbulent, making study of intermittent flames in complex fuels difficult. This thesis applies the one-dimensional turbulence (ODT) model to a study of flame propagation by simulating a lab-scale fire representative of the flame interface in a fuel bed and incorporating solid fuel particles into the ODT code. The ODT model is able to resolve individual flames (a unique property of this model) and provide realistic turbulent statistics. ODT solves diffusion-reaction equations on a line-of-sight that is advanced either in time or in one spatial direction (perpendicular to the line-of-sight). Turbulent advection is modeled through stochastic domain mapping processes. A vertical wall fire, in which ethylene fuel is slowly fed through a porous ceramic, is modeled to investigate an unsteady turbulent flame front in a controlled environment. Simulations of this configuration are performed using a spatial formulation of the ODT model, where the ODT line is perpendicular to the wall and is advanced up the wall. Simulations include radiation and soot effects and are compared to experimental temperature data taken over a range of fuel flow rates. Flame structure, velocities, and temperature statistics are reported. The ODT model is shown to capture the evolution of the flame and describe the intermittent properties at the flame edge, though temperature fluctuations are somewhat over predicted. A solid particle devolatilization model was included in the ODT code to study the convective heating of unburnt solid fuels through direct flame contact. Here the particles are treated as sweet gum hardwood and a single-reaction, first order decomposition model is used to simulate the devolatilization rates. Only preliminary results were presented for a simple case, but this extension of the ODT model presents new opportunities for future research.

For a topological space X, the fundamental group can be topologized as a quotient of the path space with the compact-open topology. For one-dimensional or planar Peano continua, the fundamental group with this topology is a topological group if and only if it is semilocally simply connected. In particular, we demonstrate that the group operation is not continuous in this setting.

Detection of explosive gas leaks such as hydrogen (H2) becomes key element in the wake of counter-terrorism threats, introduction of hydrogen powered vehicles and use of hydrogen as a fuel for space explorations. In recent years, a significant interest has developed on metal oxide nanostructured sensors for the detection of hydrogen gas. Gas sensors properties such as sensitivity, selectivity and response time can be enhanced by tailoring the size, the shape, the structure and the surface of the nanostructures. Sensor properties (sensitivity, selectivity and response time) are largely modulated by operating temperature of the device. Issues like instability of nanostructures at high temperature, risk of hydrogen explosion and high energy consumption are driving the research towards detection of hydrogen at low temperatures. At low temperatures adsorption of O2- species on the sensor surface instead of O- (since O- species reacts easily with hydrogen) result in need of higher activation energy for hydrogen and adsorbed species interaction. This makes hydrogen detection at room temperature a challenging task. Higher surface area to volume ratio (resulting higher reaction sites), enhanced electronic properties by varying size, shape and doping foreign impurities (by modulating space charge region) makes nanocrystalline materials ideal candidate for room temperature gas sensing applications. In the present work various morphologies of nanostructured tin oxide (SnO2) and indium (In) doped SnO2 and titanium oxide (titania, TiO2) were synthesized using sol-gel, hydrothermal, thermal evaporation techniques and successfully integrated with the micro-electromechanical devices H2 at ppm-level (as low as 100ppm) has been successfully detected at room temperature using the SnO2 nanoparticles, SnO2 (nanowires) and TiO2 (nanotubes) based MEMS sensors. While sensor based on indium doped tin oxide showed the highest sensitivity (S =Ra/Rg= 80000) and minimal response time (10sec.). Highly porous SnO2 nanoparticles thin film (synthesized using template assisted) showed response time of about 25 seconds and sensitivity 4. The one dimensional tin oxide nanostructures (nanowires) based sensor showed a sensitivity of 4 and response time of 20 sec. Effect of aspect ratio of the nanowires on diffusion of hydrogen molecules in the tin oxide nanowires, effect of catalyst adsorption on nanowire surface and corresponding effect on sensor properties has been studied in detail. Nanotubes of TiO2 prepared using hydrothermal synthesis showed a sensitivity 30 with response time as low as 20 seconds where as, TiO2 nanotubes synthesized using anodization showed poor sensitivity. The difference is mainly attributed to the issues related to integration of the anodized nanotubes with the MEMS devices. The effect of MEMS device architecture modulation, such as, finger spacing, number and length of fingers and electrode materials were studied. It has been found that faster sensor response (~ 10 sec) was observed for smaller finger spacing. A diffusion model is proposed for elucidating the effect of inter-electrode distance variation on conductance change of a nano-micro integrated hydrogen sensor for room temperature operation. Both theoretical and experimental results showed a faster response upon exposure to hydrogen when sensor electrode gap was smaller. Also, a linear increase in the sensor sensitivity from 500 to 80000 was observed on increasing the electrode spacing from 2 to 20 μm. The improvement in sensitivity is attributed to the higher reactive sites available for the gaseous species to react on the sensor surface. This phenomenon also correlated to surface adsorbed oxygen vacancies (O-) and the rate of change of surface adsorbed oxygen vacancies. This dissertation studied in detail dimensionality aspects of materials as well as device in detecting hydrogen at room temperature.