Transfer signal without wire has been widely accepted after the introduction of cellular technology and WiFi technology, hence the power cable is the last wire that has yet to be eliminated. Inductive power transfer (IPT) has drawn substantial interest in both academia and industry due to its advantages including convenience, nonexistence of cable and connector, no electric shock issue, ability to work under some extreme environment, and so on. After performing thorough literature review of IPT systems, two major drawbacks including low power efficiency and coil displacement sensitivity are identified as the main obstacles that have to be solved in order for these systems to reach full functionality and compete with existing wired solutions. To address the limitations and design challenges in the IPT systems, a detailed electric circuit modeling of individual part of the IPT DC-DC stage is performed. Several resonant DC-AC inverters and output AC-DC rectifiers are compared based on their performance and feasibility in inductive charging applications. Different equivalent circuit models for the loosely coupled transformer (LCT) are derived which allows for better understanding on how power is distributed among the circuit components. Five compensation networks to improve the power transfer efficiency are evaluated and their suitable application occasions are identified. With comprehensive circuit model analysis, the influence of the resonant compensation tank parameters has been investigated carefully for efficient power transfer. A novel tuning network parameters design methodology is proposed based on multiple given requirement such as battery charging profile, geometry constraints and operating frequency range, with the aim of avoiding bifurcation phenomenon during the whole charging process and achieving decent efficiency. A 4-kW hardware prototype based on the proposed design approach is built and tested under different gap and load conditions. Peak IPT system DC-DC efficiencies of 98% and 96.6% are achieved with 4-cm and 8-cm air gap conditions, which is comparable to the conventional plug-in type or wired charging systems for EVs. A long-hour test with real EV batteries is conducted to verify the wireless signal transmission and CC/CV mode seamless transition during the whole charging profile without bifurcation. To reduce the IPT system sensitivity to the gap variation or misalignment, a novel LCT design approach without additional complexity for the system is proposed. With the aid of FEA simulation software, the influence of coil relative position and geometry parameters on the flux distribution and coupling coefficient of the transmitter and receiver is studied from an electromagnetic perspective. An asymmetrical LCT based on the proposed design method is built to compare with a traditional symmetrical LCT. With fixed 10-mm gap and 0 to 40-mm misalignment variation, the coupling coefficient for the symmetrical LCT drops from 0.354 to 0.107, and the corresponding efficiency decrease is 16.6%. The operating frequency variation is nearly 100 kHz to maintain same input/output condition. When employing the proposed asymmetrical LCT, the coupling coefficient changes between 0.312 and 0.273, and the maximum efficiency deviation is kept within 0.67% over the entire 40-mm misalignment range. Moreover, the required frequency range to achieve same operation condition is less than 10 kHz. Lastly, some design considerations to further improve the IPT system efficiency are proposed on the basis of the designed asymmetrical LCT geometry. For given circuit specifications and LCT coupling conditions, determination of the optimal primary winding turns number could help achieve minimal winding loss and core loss. For lower output power, the optimal primary winding turns number tends to be larger compared to that for higher output power IPT system. Two asymmetrical LCT with similar dimension but different number of turns are built and tested with a 100-W hardware prototype for laptop inductive charging. The proposed efficiency improvement methodology is validated by the winding loss and core loss from experimental results. / Ph. D.

In recent years, there is a clear trend toward increasing the demand for electric power in high-power applications. High-power converters are making major impacts on these high-power applications. Recent breakthroughs in Silicon Carbide (SiC) materials and fabrication techniques have led to the development of high-voltage, high-frequency power devices, which are at the heart of high-power converters. SiC metal-oxide semiconductor field-effect transistors (MOSFETs) have advantages over silicon (Si) devices due to their higher breakdown voltage, higher thermal capability, and lower on-state resistance. However, their fast switching frequency and high blocking voltage bring challenges to the gate-drive circuit design. The gate driver of SiC-MOSFETs requires a power supply that provides a high-voltage, high-density design, a low input-output capacitance (CI/O) transformer design, good voltage regulation, as well as good resilience to faults to enable safe and fast operation. In this thesis, a power supply that supplies multiple gate drivers for 10 kV SiC MOSFETs is presented. A transformer design approach with a single turn at the primary side is proposed. A 20 kV insulation is achieved by the primary HV cable insulation across a toroid transformer core. The CI/O is designed less than 2 pF to mitigate the Common-Mode (CM) noise. A circuit topology analysis is performed and the inductor/capacitor/capacitor/inductor (LCCL) – inductor/capacitor (LC) circuit is selected. This circuit allows Zero-Voltage Switching (ZVS) at full operation range. A Resonant-Current-Bus (RCB) is built at the transformer primary side to achieve load-independence. / Master of Science / Wide-bandgap semiconductor devices have attracted widespread attention due to their superior performance compared to their silicon devices counterpart. To utilize its full benefits, this thesis presents a complete design and optimization of a gate-drive power supply that supplies multiple gate drivers for high-voltage, high-speed semiconductor devices. Four objectives, including high density at high voltage, good noise mitigation, fair voltage regulation, resilience to faults have been achieved. During the design procedure, different topology candidates are introduced and compared, after which a resonant topology is selected. The wide-bandgap semiconductor devices are utilized to reduce the size and losses. Hardware assembly is shown and experimental testing results are provided in the end to verify the design.

This thesis presents a novel Resonant Switched-Capacitor Converter with Multiple Resonant Frequencies, abbreviated as MRSCC for both high density and efficiency non-isolated large step-down Intermediate Bus Converter (IBC). Conventional Resonant Switched-Capacitor Converter (RSCC) proposed by Shoyama and its high voltage conversion ratio derivation such as Switched-Tank Converter (STC) by Jiang and li employ half sinusoidal-current charge transfer method between capacitors to achieve high efficiency and density operation by adding a small resonant inductor in series to pure switched-capacitor converter's (SCC) flying capacitor. By operating switching frequency to be the same as its resonant frequency, RSCC achieves zero-current turn off operation, however, this cause RSCC and its derivation suffer from component variation issue for high-volume adoption. Derived from RSCC, MRSCC adds additional high frequency resonant component, operates only during its dead-time, by adding small capacitor in parallel to RSCC's resonant inductor. By operating switching frequency higher than its main resonant frequency, MRSCC utilizes double chopped half-sinusoidal current charge transfer method between capacitors to further improve efficiency. In addition, operating switching frequency consistently higher than its resonant frequency, MRSCC provides high immunity towards component variation, making it and its derivation viable for high-volume adoption. / MS / Following the recent trend, most internet services are moving towards cloud computing. Large data applications and growing popularity of cloud computing require hyperscale data centers and it will continue to grow rapidly in the next few years to keep up with the demand [4]. These cutting-edge data centers will require higher performance multi-core CPU and GPU installations which translates to higher power consumption. From 10MWatts of power, typical data centers deliver only half of this power to the computing load which includes processors, memory and drives. Unfortunately, the rest goes to losses in power conversion, distribution and cooling [5]. Industry members look into increasing backplane voltage from 12V to 48V in order to reduce distribution loss. This thesis proposes a novel Resonant Switched-Capacitor Converter using Multiple Resonant Frequencies to accommodate this increase of backplane voltage.

There is a growing interest in electric transportation, and the number of electric vehicles (EVs) is increasing. The resulting increase in EV charging power demand has an adverse impact on the existing power grids, especially the distribution transformers. The repeated and continued overloading caused by EV charging can significantly reduce their operational life. This dissertation aims to comprehensively study the adverse impacts of EV charging on distribution transformers and provide robust and practical solutions to mitigate it. A typical North American secondary distribution system with different EV penetration levels and four realistic residential EV charging scenarios are used for the analyses. The IEEE Standard C57.91-2011 is used to quantify transformer life under different scenarios and to validate the efficacy of the proposed overloading mitigation strategies. It is observed that EV charging can have a significant impact on the life of distribution transformers. To mitigate the impact of EV charging on the distribution transformer, first, a practical solution based on reactive power compensation is proposed. The method is based on reducing the over- all transformer losses by providing a component of the residential reactive power demand through non-unity power factor operation of the EV charger. A centralized recursive control structure is proposed to compute and communicate the required reactive power values to the individual EVs. It is shown that the proposed technique increases the distribution transformer’s life by an average of nearly 47% in all four scenarios considered. Moreover, the proposed controller’s structure makes it effective even on low-bandwidth, high-latency communication networks. To verify this, the proposed controller’s stability under communication delays and its robustness against potential communication failures is also validated. This research also studies potential concerns about the charger’s reliability by non-unity power factor operation. Accordingly, an alternative overloading mitigation strategy is also proposed based on fixed charging current magnitude. This second method is shown to be more effective in reducing transformer overloading at the cost of a marginal decrease in the charging rate. Lastly, a high-level overview of the existing vehicle-to-grid communication standards is presented to provide a better context for practical implementation and identify potential challenges.

An ever present trend in the power conversion industry is to get higher performance at a lower cost. In a computer server system, the front-end converter, supplying the load subsystems, is typically a multiple output power supply. The power supply unit is custom designed and its output voltages are fully regulated, so it is not very efficient or cost effective. Most of the load systems in this application are supplied by point-of-load converters (POLs). By leaving the output voltage regulation aspect to POLs, the front-end converter does not need to be a fully regulated, multiple output converter. It can be replaced by a dc-dc transformer (DCX), which is a semi-regulated or unregulated, single output dc-dc converter. A DCX can be made using a modular design to simplify expansion of the system capacity. To realize this concept, the DCX block must have a current sharing feature. The current sharing method for a resonant DCX is discussed in this work. To simplify the system architecture, the current sharing method is based on the droop method, which requires no communication between paralleled units. With this method, the current sharing error is inversely proportional to the droop voltage. In traditional DCX implementations, the droop voltage depends on the resistive voltage drops in the power stage, which is not sufficient to achieve the desired current sharing error. The resonant converter has the inherent characteristic that its conversion gain depends on the load current, so the virtual droop resistance can realized by the resonant tank and the droop voltage can be obtained without incurring conduction loss. An LLC resonant converter is investigated for its droop characteristic. The study shows the required droop voltage is achievable at very high switching frequency. To lower the switching frequency, a notch filter is introduced into the LLC resonant tank to increase the sensitivity of the conversion gain versus the operating frequency. The design of the multi-element resonant tank is discussed. Depending soly on the resonant tank, the droop characteristic is largely varied with the component tolerance in the resonant tank. The current sharing error becomes unacceptable. The active droop control is imposed to make the output regulation characteristic insensitive to the component tolerance. The proposed resonant DCX has simpler circuit structure than the fully regulated resonant converter. Finally simulation and experimental results are presented to verify this concept. / Master of Science

This thesis focuses on the design issues of the multiple-output boost full-bridge converter, which is constructed by cascading the boost regulator with the inductor-less full-bridge converter. The design of the boost regulator has been proposed briefly with component selection and compensator design. After that, the inductor-less full-bridge converter is analyzed extensively. In the first place, the operation principle of the inductor-less full-bridge converter is introduced. Later, the effect of parasitic resistance and inductance is analyzed in an L-R series circuit model as step-response, which relates the drop of output voltage to the load current. Then, the effects of the dc blocking capacitor for the unbalanced load condition and unbalanced duty cycle are tackled. The theoretical results are compared with the experimental results and the simulation results to verify the relationship between the output voltage drop and load current. The overall efficiency of the converter is tested under various conditions. The design of the planar transformer is critical to limit the profile of the converter and the leakage phenomenon. A planar transformer fit for the inductor-less full-bridge converter is designed and analyzed in 3D FEA software. An N-port transformer model is proposed to implement the inductance matrix into the leakage inductance matrix for circuit analysis. Based on this N-port model several measurements to extract the parameters in this model are proposed, where only the impedance analyzer is needed. Finally, the effects of trace layout and encapsulation on breakdown voltage in PCB are summarized from experimental results. / Master of Science

This thesis proposes one comparison methodology for single and three-phase high power high frequency transformers in power conversion systems. The objective is to compare the volume of the transformers. And single and three-phase Dual Active Bridge Converter (DAB1 and DAB3) topologies with single and three-phase isolating transformers are selected for the transformer comparison. Design optimization of power transformer has been studied and simplified models have been built for the single and three-phase transformer design optimization in this work, including assumptions for core shapes, materials, winding structures and thermal model. Two design methods have been proposed according to different design constraints, named T – B Method and J – B Method separately. T – B Method is based on feature of the core, which has the major limits of maximum flux density and temperature rise. The flux density should not reach the saturation value of the core, and temperature rise should meet specifications in different applications to assure the performance of the core (permeability, saturation flux density, and core loss) and the insulation of the wire. And J – B Method starts from the comparison of area product in conventional design method. The relationship between area product of transformer cores and the flux and current of the transformer in design is analyzed. There is specified relationship between area product of single and three-phase transformers if flux and current densities are specified for both. Thus J – B Method is proposed with the design constraints of specified current and flux density. Both design methods include both single and three-phase transformer design. One example case for single and three-phase transformer comparison is selected as high power high frequency DAB conversion system. Operation principles are studied for both DAB1 and DAB3 based on previous work. And transformer design based on the T – B and J – B Methods are carried out and transformer volumes are compared. And results show that three-phase transformer has little benefit in volume or thermal than single-phase transformer, when they are utilized in single-phase DAB and three-phase DAB converters separately. Scaled-down single and three-phase DAB systems have been built and volume and thermal tests have been carried out. / Master of Science

Passwords are common figures in everyone’s everyday life. One person has in average80 accounts for which they are supposed to use different passwords. Remembering allthese passwords is difficult and leads to people reusing, or reusing with slight modification,passwords on many accounts. Studies on memory show that information relating tosomething personal is more easily remembered. This is likely the reason as to why manypeople use passwords relating to either self, relatives, lovers, friends, or pets. Hackers will most often use either brute force or dictionary attacks to crack a password.These techniques can be quite time consuming so using machine learning could bea faster and easier approach. Segmenting someone’s previous passwords into meaningfulunits often reveals personal information about the creator and can thus be used as a basisfor password guessing. This report focuses on evaluating different sizes of the GPT-SW3model, which uses a transformer architecture, on password segmentation. The purposeis to find out if the GPT-SW3 model is suitable to use as a password segmenter and byextension if it can be used for password guessing. As training data, a list of passwords collected from a security breach on a platformcalled RockYou was used. The passwords were segmented by the author to provide themodel with a correct answer to learn from. The evaluation metric, Exact Match, checksif the model’s prediction is the same as that of the author. There were no positive resultswhen training GPT-SW3, most likely because of technical limitations. As the results arerather insufficient, future studies are required to prove or disprove the assumptions thisthesis is based on.

Recent advances including the Transformer architecture have revolutionized the Natural Language Processing community by providing immense performance improvements across many tasks, including the development of Large Language Models (LLMs). LLMs show enormous promise as few-shot learners, common-sense knowledge repositories, conversational agents, writing assistants, and coding tools, and are gaining widespread traction in commercial industry. However, LLMs are expensive and time-consuming to train, requiring many passes over terabytes of data for the largest models. In this paper, we present Superstilling, a method for reducing the sample complexity of language model training by distilling the knowledge from a previously-trained model (the teacher) into a new, larger model (the student). This method does not require conformity between the architectures of the two models, and can be applied even when the weights and training data of the teacher model are not available, for example in federated learning scenarios. We apply Superstilling to train models of various sizes and show this method can decrease sample complexity by more than 10\% on models with over 160M parameters. We also show that in certain scenarios, Superstilling can be used to speed up training despite the need to run the teacher and student models simultaneously.

In this work, we first analyze the behavior of the acoustic wave from the theoretical point of view using a simplified 1-dimensional model. The model was developed based on the conservation of mass, the conservation of momentum, and the state equation; in addition, the fluid medium obeys Stokes assumption and it is homogeneous, adiabatic and isentropic. Experiment and simulation results show consistency to theoretical calculation. The second part of this thesis focuses on the PD signal analysis from an on-site PD measurement of the in-house design fiber optic sensors (by Virginia Tech, Center for Photonics Technology). Several commercial piezoelectric transducers (PZTs) were also used to compare the measurement results. The signal analysis employs the application of wavelet-based denoising technique to remove the noises, which mainly came from vibration, EMI, and light sources, embedded in the PD signal. The denoising technique includes the discrete wavelet transform (DWT) decomposition, thresh-holding of wavelet coefficients, and signal recovery by inverse discrete wavelet transform. Several approaches were compared to determine the optimal mother wavelet. The threshold limits are selected to remove the maximum Gaussian noises for each level of wavelet coefficients. The results indicate that this method could extract the PD spike from the noisy measurement effectively. The frequency of the PD pulse is also analyzed; it is shown that the frequencies lie in the range of 70 kHz to 250 kHz. In addition, with the assumed acoustic wave propagation delay between PD source and sensors, it was found that all PD activities occur in the first and third quadrant in reference to the applied sinusoidal transformer voltage. / Master of Science