Theses and Dissertations
This collection includes both ASU Theses and Dissertations, submitted by graduate students, and the Barrett, Honors College theses submitted by undergraduate students.
Displaying 1 - 4 of 4
Filtering by
- Creators: Kwon, Beomjin
Description
A Compact Linear Fresnel Reflector (CLFR) is a simple, cost-effective, and scalable option for generating solar power by concentrating the sun rays. To make a most feasible application, design parameters of the CLFR, such as solar concentrator design parameters, receiver design parameters, heat transfer, power block parameters, etc., should be optimized to achieve optimum efficiency. Many researchers have carried out modeling and optimization of CLFR with various numerical or analytical methods. However, often computational time and cost are significant in these existing approaches. This research attempts to address this issue by proposing a novel computational approach with the help of increased computational efficiency and machine learning. The approach consists of two parts: the algorithm and the machine learning model. The algorithm has been created to fulfill the requirement of the Monte Carlo Ray tracing method for CLFR collector simulation, which is a simplified version of the conventional ray-tracing method. For various configurations of the CLFR system, optical losses and optical efficiency are calculated by employing these design parameters, such as the number of mirrors, mirror length, mirror width, space between adjacent mirrors, and orientation angle of the CLFR system. Further, to reduce the computational time, a machine learning method is used to predict the optical efficiency for the various configurations of the CLFR system. This entire method is validated using an existing approach (SolTrace) for the optical losses and optical efficiency of a CLFR system. It is observed that the program requires 6.63 CPU-hours of computational time are required by the program to calculate efficiency. In contrast, the novel machine learning approach took only seconds to predict the optical efficiency with great accuracy. Therefore, this method can be used to optimize a CLFR system based on the location and land configuration with reduced computational time. This will be beneficial for CLFR to be a potential candidate for concentrating solar power option.
ContributorsLunagariya, Shyam (Author) / Phelan, Patrick (Thesis advisor) / Kwon, Beomjin (Committee member) / Zhuang, Houlong (Committee member) / Arizona State University (Publisher)
Created2021
Description
Laser powder bed fusion (LPBF) additive manufacturing (AM) has received widespread attention due to its ability to produce parts with complicated design and better surface finish compared to other additive techniques. LPBF uses a laser heat source to melt layers of powder particles and manufactures a part based on the CAD design. This process can benefit significantly through computational modeling. The objective of this thesis was to understand the thermal transport, and fluid flow phenomena of the process, and to optimize the main process parameters such as laser power and scan speed through a combination of computational, experimental, and statistical analysis. A multi-physics model was built using to model temperature profile, bead geometry and elemental evaporation in powder bed process using a non-gaussian interaction between laser heat source and metallic powder. Owing to the scarcity of thermo-physical properties of metallic powders in literature, thermal conductivity, diffusivity, and heat capacity was experimentally tested up to a temperature of 1400 degrees C. The values were used in the computational model, which improved the results significantly. The computational work was also used to assess the impact of fluid flow around melt pool. Dimensional analysis was conducted to determine heat transport mode at various laser power/scan speed combinations. Convective heat flow proved to be the dominant form of heat transfer at higher energy input due to violent flow of the fluid around the molten region, which can also create keyhole effect. The last part of the thesis focused on gaining useful information about several features of the bead area such as contact angle, porosity, voids and melt pool that were obtained using several combinations of laser power and scan speed. These features were quantified using process learning, which was then used to conduct a full factorial design that allows to estimate the effect of the process parameters on the output features. Both single and multi-response analysis are applied to analyze the output response. It was observed that laser power has more influential effect on all the features. Multi response analysis showed 150 W laser power and 200 mm/s produced bead with best possible features.
ContributorsAhsan, Faiyaz (Author) / Ladani, Leila (Thesis advisor) / Razmi, Jafar (Committee member) / Kwon, Beomjin (Committee member) / Nian, Qiong (Committee member) / Zhuang, Houlong (Committee member) / Arizona State University (Publisher)
Created2021
Description
Progressive miniaturization in electronics demands advanced materials with excellent energy conversion and transport properties. Opportunities exist in novel material morphologies such as hierarchical structures, multi-functional composites and nanoscale architectures which may offer mechanical, thermal and electronic properties tailored to a wide range of applications (e.g., aerospace, robotics, biomedical etc.). However, the manufacturing capabilities have always posed a grand challenge in realizing the advanced material morphologies. Furthermore, the multi-scale modeling of complex material architectures has been extremely challenging owing to the limitations in computation methodologies and lack of understanding in nano-/micro-meter scale physics. To address these challenges, this work considers the morphology effect on carbon nanotube (CNT)-based composites, CNT fibers and thermoelectric (TE) materials. First, this work reports additively manufacturable TE morphologies and analyzes the thermo-electric transport behavior. This research introduces innovative honeycomb TE architectures that showed ~26% efficiency increase and ~25% density reduction compared to conventional rectangular TE architectures. Moreover, this work presents 3D printable compositionally segmented TE architecture which provides record-high efficiencies (up to 8.7%) over wide temperature ranges if the composition and aspect ratio of multiple TE materials are optimized within a single TE device. Next, this research proposes computationally efficient two-dimensional (2D) finite element model (FEM) to study the electrical and thermal properties in CNT based composites by simultaneously considering the stochastic CNT distributions, CNT fractions (upto 80%) and interfacial resistances. The FEM allows to estimate the theoretical maximum possible conductivities with corresponding interfacial resistances if the CNT morphologies are carefully controlled, along with appreciable insight into the energy transport physics. Then, this work proposes a data-driven surrogate model based on convolutional neural networks to rapidly approximate the composite conductivities in a second with accuracy > 98%, compared to FEM taking >100 minutes per simulation. Finally, this research presents a pseudo 2D FEM to approximate the electrical and thermal properties in CNT fibers at various CNT aspect ratios (up to 10,000) by simultaneously considering CNT-CNT interfacial effects along with the stochastic distribution of inter-bundle voids.
ContributorsEjaz, Faizan (Author) / Kwon, Beomjin (Thesis advisor) / Zhuang, Houlong (Committee member) / Song, Kenan (Committee member) / Wang, Robert (Committee member) / Kang, Wonmo (Committee member) / Arizona State University (Publisher)
Created2023
Description
This dissertation contributes to uncertainty-aware neural networks using multi-modality data, with a focus on industrial and aviation applications. Drawing from seminal works in recent years that have significantly advanced the field, this dissertation develops techniques for incorporating uncertainty estimation and leveraging multi-modality information into neural networks for tasks such as fault detection and environmental perception. The escalating complexity of data in engineering contexts demands models that predict accurately and quantify uncertainty in these predictions. The methods proposed in this document utilize various techniques, including Bayesian Deep Learning, multi-task regularization and feature fusion, and efficient use of unlabeled data. Popular methods of uncertainty quantification are analyzed empirically to derive important insights on their use in real world engineering problems. The primary objective is to develop and refine Bayesian neural network models for enhanced predictive accuracy and decision support in engineering. This involves exploring novel architectures, regularization methods, and data fusion techniques. Significant attention is given to data handling challenges in deep learning, particularly in the context of quality inspection systems. The research integrates deep learning with vision systems for engineering risk assessment and decision support tasks, and introduces two novel benchmark datasets designed for semantic segmentation and classification tasks. Additionally, the dissertation delves into RGB-Depth data fusion for pipeline defect detection and the use of semi-supervised learning algorithms for manufacturing inspection tasks with imaging data. The dissertation contributes to bridging the gap between advanced statistical methods and practical engineering applications.
ContributorsRathnakumar, Rahul (Author) / Liu, Yongming (Thesis advisor) / Yan, Hao (Committee member) / Jayasuriya, Suren (Committee member) / Zhuang, Houlong (Committee member) / Kwon, Beomjin (Committee member) / Arizona State University (Publisher)
Created2024