Theses/Dissertations - Electrical and Computer Engineering

Permanent URI for this collectionhttps://hdl.handle.net/2104/4811

Browse

Now showing 1 - 20 of 111

Power distribution feeder response to the asymmetric saturation of substation transformers caused by significant high-side DC currents.
(May 2023) Weldy, Christopher D., 1984-; Grady, William Mack.
High-altitude electromagnetic-pulse and geomagnetic disturbances can lead to asymmetric saturation of utility transformer cores by causing significant DC current to flow through the transformer windings. Transformer secondary voltage and current waveforms are distorted by core saturation and this distortion can be amplified at points on power distribution feeders where circuit topologies create series or parallel resonances. The performance of distribution feeders during asymmetric core saturation is explored in this work using harmonic powerflow simulation. A harmonic voltage source model is developed to represent steady-state transformer secondary terminal response during DC current flow in the primary windings. The model performance is compared to field test measurement data and then used to simulate the energization of realistic electric power distribution feeder models. The response of the feeder models is compared to industry standards of power quality.
Characterization and lifetime prediction of roller and journal bearings in Si and SiC motor drive applications.
(May 2023) Chen, Hellen W., 1997-; Jouanne, Annette von.
In the electrified propulsion system, a variable frequency drive (VFD) controls an electric motor. In many cases, this electric motor is an induction motor, which has been called the workhorse of industry. With semiconductor switching device advancements enabling higher power density inverter drives, mitigation techniques are being explored to remedy common issues that arise such as damaging bearing currents to ensure smooth implementation into motor drive systems. As higher power applications are enabled, the components within motors (including bearings) must have the capacity to handle the potential problems that arise from higher slew rates of advanced wide band gap (WBG) semiconductor switching devices in inverter drives. This thesis explores the behaviors of roller and journal bearings under electrical discharge conditions that are common in advanced motor drive applications, comparing the results with some known characteristics of ball bearings under similar conditions.
Wide bandgap inverter duty bearings : development and lifetime prediction of conductive grease bearings with integrated graphene nanoparticles.
(May 2023) Stephens, Madeline Anne, 1996-; Jouanne, Annette von.
As society pushes toward a sustainable, electrified future, the superior qualities of wide bandgap (WBG) devices will enable the next generation of power electronics that allow for greater system efficiencies and reduced form factor when compared to conventional silicon devices. However, the integration of WBG devices into motor drive systems exacerbates the existing application issues surrounding pulse width modulated (PWM) variable frequency drives (VFDs) that lead to system degradation and costly premature failure. A significant application issue of WBG devices in motor drive systems is the premature failure of motor bearings due to an increase in vibration caused by electric discharge machining (EDM) resulting from inverter-induced, high frequency bearing currents. This dissertation presents a novel method to reduce bearing currents and mitigate EDM damage by increasing the conductivity of motor bearings via conductive bearing lubricant. A survey of commercially available conductive greases is conducted, and the possibility of increasing lubricant conductivity through the integration of graphene nanoparticles is explored. Novel conductive greases are developed from base materials, so that the effects of grease formulations can be precisely controlled and studied. Through extensive experimental validation, statistical analysis, and rheological characterization, the most promising conductive grease formulations are selected, and through the accelerated aging of conductive grease bearings, their lifetimes are predicted. The work presented in this dissertation includes key contributions to the development of WBG inverter duty bearings and conductive bearing grease, which contributes significant value by extending the bearing lifetimes in PWM inverter driven induction motor applications and preventing costly premature bearing failure.
Forecast ergodicity and instantaneous active information.
(May 2023) Amigo Galan, Glauco A., 1984-; Marks, Robert J., II (Robert Jackson), 1950-
This dissertation introduces two new information theory tools for machine learning. Forecast ergodicity models prediction using algorithmic information theory, stating that future data can be forecasted if it has the same structure as past data. The novelty of this approach is that it is model-free, based only on data. The other tool is instantaneous active information. It measures the active information at each step of iterative searches that are assisted by oracles. The instantaneous active information characterizes the difficulty of each subsequent step of a search. Knowing the difficulty of each search step helps to make real-time decisions about conflicting goals. As a demonstration, this search analysis is applied to a phased array optimization problem where there exists a tradeoff between the time devoted to parameter optimization and signal transmission.
Design and development of autonomous electric vehicles capable of following an EPA drive cycle on a dynamometer testbed and navigating an on-road obstacle course.
(May 2023) Brown, Ezekiel B., 1998-; Jouanne, Annette von.
This dissertation presents the background, research and further advancement of the design and development of an autonomous electric vehicle that can follow an Environmental Protection Agency (EPA) drive cycle on a dynamometer testbed. Design, development, and test data collected, analyzed, presented, and referred to in this dissertation stems from an all-electric Chevy Bolt and an electric converted Chevy Tahoe. The work done on the Chevy Bolt enables the vehicle with the capability to autonomously follow an EPA drive cycle on a dynamometer and the work done on the Chevy Tahoe advances on the autonomous acceleration system, incorporating autonomous steering and sensory integration, enabling the vehicle to follow a path autonomously. The programmable throttle and the programmable brake research on the all-electric Bolt are discussed followed by the implementation of a programmable acceleration system, programmable steering, and the development of a sensory system on an all-electric converted Chevy Tahoe.
Outlier selection methods for improved bearings-only geolocation.
(December 2022) Barnett, Trevor J., 1993-; Thompson, Michael Wayne.
Bearings-only location estimation is a problem that has widespread applications. While the concept of bearings-only location estimation is not new, there are still many problems inherent to the process. There is high demand for a process that can reduce bias, remove outliers, and more accurately estimate emitter location using bearing data, exclusive of range. This thesis applies outlier removal methods, the parameterization of which are characterized within, to create more reliable data sets from noisy data which likely contain a significant percentage of outliers. Secondly, it uses statistical estimation and resampling to create a small, plausible ellipse representing the location of the emitter being tracked. In this thesis simulation studies show these methods to be a significant improvement over the standard implementation of Cartesian Pseudo-Linear Estimation with the presence of outlier data.
Ab-initio models of quantum dot cellular automata molecules.
(December 2022) Liza, Nishat Tasnim, 1993-; Blair, Enrique Pacis.
Molecular quantum-dot cellular automata (QCA) is a charge based, low-power, energy-efficient alternative to transistor-based, general-purpose computation. In molecular QCA, redox centers of a mixed-valence (MV) molecule function as coupled quantum dots, and localized charge states of the molecule encode binary information useful for classical computing. Molecular QCA promises ultra-high device densities, THz-scale switching speeds and room temperature readout. While the fundamental principle of molecular QCA have been tested and established, major challenges must be overcome to successfully implement molecular QCA. This work applies ab-initio techniques in the design and modeling of candidate MV molecules for QCA. Here, we study and characterize ∼ 1-nm-scale MV QCA molecules using first principle calculations. The structural and electronic properties of QCA molecules are calculated utilizing Hartree-Fock, Post-Hartree-Fock and Density functional theory (DFT) methods. Asymmetric, cationic, MV molecules are designed for spectroscopic state readout of QCA devices at room temperature. Tip-enhanced Raman spectroscopy is proposed to detect the state of QCA devices in a circuit if the QCA molecules have slightly dissimilar quantum dots. Clocked zwitterionic three-dot QCA molecules with built-in counterions at the center of the molecules are modeled. The choice and design of the central linkers of these molecules determines number of mobile charges in the molecules for encoding the device states on the three quantum dots. These molecules show different device responses to applied clocking electric field based on different central linkers designed and used, similar to the complementary responses of PMOS and NMOS transistors to gated voltage control. Counterion effects on QCA candidate molecules are also explored in terms of electron transfer parameters. The complete active space self consistent field (CASSCF) method is used to calculate electron transfer (ET) matrix element and inner-sphere reorganization energy of the molecules in the presence of nearby counterions. Results demonstrate that randomly placed externel counterions may degrade device states by causing mobile charge to localize in undesirable ways on the QCA molecule. New zwitterionic molecules with a built-in counterion are proposed to eliminate unpredictable effects of external counterions in QCA circuits. Novel organometallic zwitterionic QCA molecules with ferrocene dots are designed and proposed for synthesis. The chemical stability of these ferrocene based molecules are evaluated by theoretical calculations. The synthesis of these stable zwitterionic molecules by collaborating experimental chemists is in progress and may open a new path to realize molecular QCA computing. A new machine-learning-based DFT functional, DM21 is investigated and benchmarked against traditional methods by comparing the calculated ET matrix elements of several QCA molecules. Preliminary results calculated from DM21 functional did not show significant improvements in accuracy and computational cost. Modification and improvement of the neural network used in the development of the functional, as well as the underlying code is proposed which might open new path to computationally inexpensive QCA calculations.
Brillouin and transverse mode instabilities in fiber amplifiers for high-energy laser systems.
(December 2022) Young, Joshua T., 1995-; Hu, Jonathan.
High-energy fiber lasers have developed a lot of interest due to their applications in industry, medicine, and defense. Recent advances enabled an explosive growth in operating power to the scale of multi-kilowatts. However, nonlinear effects such as the Brillouin instability (BI) and the transverse mode instability (TMI) impose limits on the power of high-energy fiber amplifiers. This dissertation focuses on the theoretical modeling of these nonlinear effects. The goal is to model these effects and further propose new mitigation techniques to increase the operating powers and advance the techniques for high-energy fiber amplifiers. The nonlinear effects in optical fibers often involve mode coupling. Avoided crossings occur when two modes are strongly coupled to each other and share similar propagation constants. We start with a tutorial to study avoided crossings in one-dimensional slab waveguides in both index guiding and antiresonant waveguides. We use simple one-dimensional slab waveguides as examples to illustrate the physics and properties of avoided crossings in more complicated specialty optical fibers. We study the TMI in an Yb-doped fiber amplifier in the presence of a single higher-order mode (HOM). Current modeling techniques for TMI require that the longitudinal discretization be substantially smaller than the beat length between the fundamental mode and HOM. We formulate the phase-matched model for TMI, which only considers the phase-matched terms that contribute to the coupling between the fundamental mode and HOMs. By doing so, the number of sections in the longitudinal discretization may be greatly decreased, which leads to a large computational win with no loss of accuracy. The BI may be modeled as a three-wave mixing process where two optical modes interact with a resonant acoustic mode. We consider phase modulation of the input pump as a suppression technique for BI. We show that piecewise parabolic phase waveforms like sawtooth and triangle phase may provide larger power thresholds compared to that of the more commonly used pseudorandom bitstream (PRBS) modulation. Because of the nearly rectangular spectrum associated with piecewise parabolic phase modulation, these modulation schemes are better fitted for power scaling such as spectral beam combining. Recently, our piecewise parabolic phase idea that was published was experimentally demonstrated. We further consider a single computational model that models BI and TMI together. A multi-time-scale approach must be used since these nonlinear effects evolve over drastically different time scales. Both BI and TMI depend differently on the core diameter of the fiber. At and under the pump power threshold for the combined BI-TMI model, the pump power threshold closely follows that of the individual BI and TMI models. However, BI may trigger TMI when strong BI leads to stochastic oscillations in the fundamental mode amplitude. This feature cannot be predicted by modeling either BI or TMI alone. At the end, we discuss the future prospects for high-energy laser fiber amplifiers and give a summary.
A deep convolutional neural network approach for biomedical applications.
(December 2022) Nguyen, Hanh Hong, 1991-; Schubert, Keith Evan.
Deep learning is a subset of machine learning that uses multi layer neural networks to perform desired tasks by using trained models. Neural networks are nonlinear mapping systems whose structure and function are loosely modeled on the physical structure of the nervous systems in humans and animals. In deep learning, convolutional neural networks (CNNs) have been used to analyze visual tasks for more than 40 years. Since the mid-2000s, they have revolutionized image processing and analysis. The goal of this dissertation is designing a deep CNN approach for biomedical applications, including automation of the process of colon polyps classification as well as single particle identification in radiation therapy.
Models and designs of clocked molecular quantum-dot cellular automata circuits.
(2022-03-30) Cong, Peizhong, 1991-; Blair, Enrique Pacis.
Quantum-dot cellular automata (QCA) is a general-purpose, low-energy, high-efficiency, general purpose computing paradigm. QCA computation may overcome challenges such as heat dissipation and the clock speed bottleneck facing the extreme scaling of CMOS. A molecular implementation of QCA promises nanometer-scale device sizes, ultra high device densities, and ∼THz clock speed all at room temperature. We model and design molecular QCA circuits immersed in applied electric fields to address various technical challenges in the path to realizing molecular QCA. We explore the use of molecular QCA circuits and field-generating electrodes for both the write-in and read-out of classical bits, two problems central to interfacing conventional semiconductor logic with molecular QCA logic. We design and model novel circuits which may support the input and output of classical bits with the help of an applied electric field. While electric fields are useful for input and output circuits, they may disrupt information-processing circuits. Therefore, we also explore the extent to which molecular circuits (logic and interconnects) tolerate applied electric fields. We demonstrate that molecular QCA circuits can tolerate significant unwanted electric fields, well beyond those fields required for bit input or output.
Sensing human physiological motion using millimeter wave radar.
(2022-03-28) Bresnahan, Drew G., 1994-; Li, Yang, 1982-
For the past 100 years, radio detection and ranging technology, or radar, has been developed to detect the presence and motion of a wide array of objects, vehicles, and living beings. With the advent of small-scale, portable, affordable radar sensors, biological radar sensing has opened a new field of opportunity for healthcare applications such as vital signs detection, fall detection, and gait analysis. Telehealth technology is becoming a popular method to replace in-person office consultations for simple medical visits, reducing costs significantly. However, telehealth communication currently lacks the sensors required to provide the clinician a direct line of information about the patient’s physical condition. Radar can measure a variety of physiological motions for basic health checkups and become another diagnostic tool in the doctor’s arsenal. Furthermore, radar can act as a preventive safety device by detecting human drowsiness or distraction. This dissertation details the procedures and analysis of employing a single millimeter-wave radar unit to measure human vital signs, head movements, deep tendon reflex motion, and hip movements.
The impact of conventional and wide bandgap semiconductor PWM inverter EDM bearing currents on motor bearings.
(2022-02-21) Collin, Ryan, 1995-; Jouanne, Annette von.
The pulse width modulated (PWM) ac motor drive has revolutionized the industrial sector by drastically reducing energy consumption and improving the efficiency of industrial processes. To continue the transition towards an electrified society, power density in electric motor drives must continue to increase. To this end, motor drives of the future will contain wide bandgap (WBG) semiconductor devices such as silicon carbide (SiC) and gallium nitride (GaN). The introduction of these WBG devices into motor drives comes with exacerbated application issues due to faster dv/dt device turn on/off times and relatively higher switching frequencies for the same device cooling requirements as conventional silicon (Si) devices. A key issue produced by these PWM motor drives is premature bearing failure due to an increase in vibration from the damage caused by electric discharge machining (EDM) bearing currents. This dissertation presents detailed discharge circuit modeling of motor bearing currents, investigates the impact of discharge energy on the damage produced on the bearing race and ball surfaces through the use of a finite element (FE) model, presents a novel technique to statistically characterize EDM amplitudes over various motor operating conditions, and uses this statistical approach along with experimental evidence to predict a bearing’s lifetime. The work presented here is a key contribution to developing intricate predictive maintenance models for motor bearings, which are of great value to industrial operators because of the enormous costs associated with motor down time.
Stochastic computing and stochastic resonance demonstrated in custom analog neuromorphic hardware.
(2022-01-21) Coker, Cameron, 1997-; Koziol, Scott M.
Stochastic computing offers an alternative computing method to standard systems. Stochastic resonance is a means of leveraging noise to improve system performance. This thesis applies both concepts to spiking analog neurons. The general usefulness of stochastic resonance is tested while the principles of stochastic resonance are applied to determine the viability of a stochastic spiking neural network.
Enabling and directing real-time cognitive radar transmitter optimization.
(2021-10-27) Egbert, Austin S., 1995-; Baylis, Charles Passant, 1979-
As the available wireless spectrum grows more crowded with increased usage from high bandwidth telecommunications applications, it becomes infeasible for many other users of wireless spectrum to continue operating with static, inflexible methods. Among these users are radar systems, which have historically been allocated large sections of bandwidth. In order to adapt and coexist with new technology in a dynamically managed environment, next generation radars must be able to adjust their spectral configuration in real time. The research presented in this dissertation provides a framework that can be used for determining transmission constraints over both spatial direction and signal frequency. While existing research has demonstrated how to optimize radar transmitters using adjustable amplifier matching networks, such optimizations have not been able to complete quickly enough for use in real-time adaptation. To accelerate these optimizations, this dissertation presents a faster method for evaluating the performance of transmit amplifiers using a software-defined radio (SDR) and a load-pull extrapolation method using deep learning image completion techniques. Additionally, the accelerated optimization technique has been adapted for use with the pulse-to-pulse waveform agility paradigm of cognitive radars. Finally, the impact on Doppler detection accuracy of modifying the radar transmit chain during a coherent radar processing interval is analyzed, along with techniques for correcting the resulting distortions.
Neural network watchdog for out-of-distribution input mitigation.
(2021-10-19) Bui, Justin M., 1988-; Marks, Robert J., II (Robert Jackson), 1950-
Neural networks have often been described as black boxes. The prevalence of publicly available neural networks and the application of transfer learning has allowed for the development of systems with minimal understanding of the data distribution. For example, a generic neural network trained to differentiate between kittens and puppies will classify a picture of a kumquat as either a kitten or a puppy, despite the kumquat residing outside the known data distribution. The neural network watchdog is a technique which screens trained classifier and regression machine input candidates to determine the distribution validity, and allows for methods of out-of-distribution removal with minimal performance impact.
Asynchronous image reconstruction.
(2021-12-13) Cai, Ritchie Zongqi, 1982-; Schubert, Keith Evan.
Algebraic Reconstruction Technique, also known as ART, is the go-to method for medical image reconstruction. For more than 50 years, since the original paper about ART is published, various reconstruction methods based on ART emerged for faster runtime performance. These methods are designed to suit different types of hardware. However, none of these methods is ART equivalent. In this work, I introduce a new implementation method that is ART equivalent, has a very fast runtime performance, and is very scalable on today and future hardware. It opens a brand new door to how we should implement ART in the future.
Reducing the size and power of stochastic computing neural networks through training.
(2021-08-01) Carrano, Matthew J., 1996-; Koziol, Scott M.
It has been demonstrated that stochastic computing (SC) has the ability to reduce the size and power requirements of artificial neural network (ANN) circuits [1]. There are two prevailing SC neuron topologies: multiplexer (MUX) and approximate parallel counter (APC) based [2]. Both topologies contain an activation module with a state parameter that affects the respective output function as well as the size and power requirements. This thesis explores altering this state parameter and the network training process in order to reduce the size and power of each neuron without incurring significant accuracy loss. As part of this exploration, a stochastic artificial neural network (SANN) is created in Verilog and implemented on a Field Programmable Gate Array (FPGA). Additionally, a SANN simulator is built in MATLAB to assist in rapid prototyping. Both simulation and hardware results demonstrate that the size/power utilized by SANNs can be reduced without significant accuracy loss.
Essential elements of proton computed tomography for practical applications.
(2021-07-30) Schultze, Blake Edward, 1983-; Schubert, Keith Evan.
The work presented in this dissertation all pertains to developments of proton computed tomography (pCT) and the elements essential to its viability as a clinical imaging modality. This includes methodological and implementational developments for reducing reconstruction time and improving pCT image quality, each advancing pCT towards clinical viability. The corresponding methods are presented in the chronological order of their development. Hull-detection, a method for differentiating voxels internal and external to an object, is presented first. Hull-detection was specifically developed for pCT as a preferable means for obtaining a binary image of the object, a preconditioning step often referred to as object detection. The concept of hull-detection, similar to the way a sculptor chisels away portions of material to produce the desired sculpture, is that voxels along the paths of protons that completely miss the object can be carved away to reveal the object hull. However, this neglects to account for the ramifications of uncertainties in the data, which was accounted for in different ways. Several hull-detection algorithms were developed and compared to the classic object detection method based on thresholding the filtered backprojection image. The second topic presented is efficiently implementing the most-likely path (MLP) formalism for pCT. This formalism was developed to more accurately approximate proton paths within an object, increasing the achievable spatial resolution. Computing the MLP is, by far, the most computationally expensive task performed during image reconstruction, making it the biggest hurdle to achieving clinically viable image reconstruction times (below 10 minutes). A computationally efficient implementation of the MLP was developed by simplifying the associated equations and incorporating several software design principles to reduce the number of compute operations and improve numerical stability. The final topic presented is the incorporation of recent advancements of total variation superiorization (TVS) into pCT. A fixed parameter version of TVS was initially incorporated into the feasibility-seeking algorithms of pCT, which included a step verifying successful TV reduction. Presented here is the modern version of TVS applied to pCT, with user-control of parameters, removal of the verification step, and additional option to perform repeated perturbations.
Study of pumping conditions on output of optically pumped interband cascade lasers.
(2021-08-08) Stephens, Kyler A., 1997-; Olafsen, Linda J.
This thesis presents the results of optically pumping two interband cascade laser (ICL) semiconductor samples and measuring both their output power and spectral data. The samples were prepared by the Naval Research Laboratory using molecular beam epitaxy. These samples were pumped with a Nd:YAG laser in conjunction with a optical parametric oscillator that allowed for tunable pump beam wavelength. The peak output power from the samples was plotted versus pump intensity to create light-light curves. A novelty of this work was that the ICLs were optically pumped with a range of pump wavelengths from 1800 nm to 1950 nm. Additionally, the temperature of the sample was varied from 80 K to 220 K. The temperature dependence of threshold pump intensity as well as slope efficiency of ICL semiconductor samples is discussed. Also, how the spectral output of the samples change with temperature and pump wavelength is presented, particularly with respect to the effect of apply a graphene monolayer top contact to the ICL surface.
Improving the quality of proton CT scan images by utilizing straggling power.
(2021-04-27) Reeser, Jordan Lee, 1996-; Schubert, Keith Evan.
Proton therapy has grown in interest within the medical community for its promise in cancer treatment. Proton CT (pCT) scans are an essential part of this process to map treatment areas, but a common problem is the lack of precision along the artifact boundaries within the image. It was hypothesized that combining two forms of collected data, stopping power and scattering power, may result in a sharper image. Initial testing harnessed typical filtering techniques to improve the combined image, but did not result in sharpened boundaries. Further testing harnessed knowledge of the data itself, such as respective contained information, and medical imaging techniques to improve the combined image. Qualitative analysis by inspection was verified quantitatively by calculating the FWHM. These results showed an average improvement of 0.3967 mm for the combined image compared to the stopping power image.

Browse

Recent Submissions