<p dir="ltr">The study builds and evaluates three advanced neural network models—ResNet-34, EfficientNet B0, and SqueezeNet—for defect detection and classification of silicon wafer map images. The study evaluates the neural network model in two cases, binary and multi-class classifications. The binary classification, which is crucial for promptly determining whether a wafer map is defective, EfficientNet-B0 led with the highest test accuracy of 94.62% and an average accuracy of 93.2%. Similarly, in multi-class classification, necessary for pinpointing specific defect causes early in the manufacturing process, EfficientNet-B0 achieved the top test accuracy of 84.22% with an average accuracy of 84.07%. Further enhancements in the study resulted from strategic pruning of EfficientNet-B0, specifically the removal of Residual Block 2 after convolutional layer visualization revealed minimal impact on accuracy, with a reduction of just 1.33%. These modifications not only refined the learning process but also reduced the model size by 33%, thereby increasing computational efficiency. The integration of Grad-CAM++ visualizations ensured the model focused on pertinent features, thus boosting the transparency and reliability of the defect detection process. The results underscore the potential of advanced neural networks to significantly enhance the accuracy and efficiency of semiconductor manufacturing.</p>

Developing software that meets the customers’ requirements, expectations, and quality standards is a challenging task for all software organizations. As modern software becomes more and more complex, so do the defects of the software. The aim of this study was to develop a simplified and consistent defect taxonomy that could be executable and usable for smaller enterprises or organizations that want to implement a simplified taxonomy. The aim of this study was also to find characteristic defects that exist in Small- and Medium-sized Enterprises (SMEs) by using the taxonomy. A manual defect classification was done on bug reports collected from three organizations with the same company size as SMEs. An agreement analysis was also conducted in this study to investigate the consistency of the taxonomy. This was done by letting different people classify a subset of the bug reports collected for this study using the proposed taxonomy. Furthermore, in this study, I also investigated how executable and usable the taxonomy would be for smaller enterprises. This was done through four interviews and a survey with seven respondents. The result of the defect classification indicates that Program anomaly (58%), GUI (17%) and Configuration (13%) are three of the most common defect types that exist in SMEs. The result of the defect classification indicates that SMEs have a problem with defects breaking features that worked correctly before, where 19% of all bug reports used in this study were classified as Regression. The survey result indicates that taxonomy is easy to use. However, the result of the different classifications showed that the use of the taxonomy is not consistent between different classifiers. The results of the interviews and the survey indicate that the taxonomy would be executable for smaller enterprises. However, to also be fully usable, the proposed taxonomy needs to be adapted to the particular enterprise, and requires the right competence to propose appropriate measures for the particular enterprise. / Att utveckla mjukvara som möter kundernas krav, förväntningar och som håller en hög kvalité är en utmanande uppgift som alla företag står inför. Den moderna mjukvaran blir allt mer komplex, vilket i sin tur även bidrar till fler och mer komplexa mjukvarufel. Den här studien har som mål att skapa en ny förenklad och konsekvent defekttaxonomi som kan vara genomförbar och användbar för mindre företag och organisationer som vill använda sig av en enklare taxonomi. Målet med det här examensarbetet är också att kartlägga karaktäristiska mjukvarufel i små och medelstora företag med hjälp av den nyutvecklade taxonomin. För att lyckas med detta klassificerades felrapporter från tre organisationer som är av samma storlek som små och medelstora företag. Den här studien analyserade även hur konsekvent taxonomin är genom att låta olika människor klassificera en delmängd av studiens felrapporter. Den här studien undersöker även hur genomförbar och användbar defekttaxonomin är för mindre företag. Detta genomfördes genom fyra intervjuer och en enkätundersökning med sju respondenter. Resultatet av den här studien indikerar att små och medelstora företag främst har problem med mjukvarufel som ger upphov till oväntat och oönskat beteende kopplat till mjukvarans funktionalitet (58%). Den här studien indikerar även att gränssnittsdefekter (17%) och konfigureingsproblem (13%) också är vanliga mjukvarufel inom dessa organisationer. Den här studien har även uppmärksammat att mindre företag kan ha problem med mjukvarufel som tar sönder funktionalitet som fungerat korrekt tidigare, även kallad regressioner, där 19% av alla felrapporter blev klassificerade som det. Resultatet av enkätundersökningen indikerar att taxonomin är enkel att använda. Å andra sidan, de olika klassifikationerna visade att användningen av taxonomin inte var konsekvent. Resultatet av intervjuerna och enkätundersökningen indikerar att taxonomin är genomförbar för mindre företag. Å andra sidan, för att defekttaxonomin ska vara användbar behöver den anpassas efter det särskilda företaget. Den kräver även den rätta kompetensen för att ta fram lämpliga åtgärden för det särskilda företaget.

Master of Agribusiness / Department of Agricultural Economics / Ted C. Schroeder / The CNH Wichita Product Center has had a chronic leak problem with the Skid Steer Loaders. The objective of this project was to analyze the manufacturing plant leak data and make improvements to correct the issue. The objective is twofold: 1) to make process or design improvements on current products produced in the plant and 2) to make recommendations for future designs to prevent such leak issues from reoccurring. The manufacturing data had to be transformed into usable form and then it was analyzed mostly by utilizing Pareto Charts. The highest six problem leak points were chosen from the manufacturing data. Process changes were implemented on these particular leak joints and the results were analyzed using two proportions hypothesis tests. The process changes reduced the leak rate by an average percent reduction of 86 percent. The process changes implemented will also be applied to other similar joints, and results documented in the future. The future design recommendations made from the analyzed data included the increased use of o-ring face seal connections at certain locations and where possible, reducing the number of joints per machine.

Many large composite structures are manufactured using sandwich laminates to achieve high specific bending strength and stiffness. Examples include wind turbine blades, where self-weight becomes increasingly important as blade size increases. Resin infusion of three-dimensional sandwich laminates can result in complex resin flow paths, and subsequent defect formation, which are difficult to predict. The core material used for sandwich construction and its interaction with liquid resins may also influence the formation of defects, and in the case of balsa this effect can be used to reduce defect severity. In order to evaluate the effect of cored sandwich laminate construction on the formation of defects, this thesis concentrates on the characterisation of commonly used core materials and their interaction with liquid resin under high vacuum conditions. It also considers two numerical flow-modelling packages which are shown to be effective at the prediction of flow front convergence for monolithic laminate, but over-estimate defect severity when modelling air- permeable cored laminates. For balsa core, experiments indicate that the available pore space can act as sink for trapped air, which can aid the reduction of defects where multiple flow fronts converge due to the complexity of flow in sandwich laminates. Empirical data for air absorption and desorption rates in balsa core were obtained using a custom-designed experiment. Using these data a theoretical model was developed that can indicate available pore space, which can inform optimum processing conditions, such as time under vacuum. The diffusion coefficients obtained for air absorption and desorption in balsa are very similar, and lie in the middle of published ranges for hard woods at around 2 x 10 -7 m2/s. The methodology developed for this research project represents actual behaviour of air absorption/desorption during resin infusion, whilst other techniques do not, merely measuring diffusion of air through a sample not allowing for finite pore space. In consequence, infusion strategies can be planned more precisely because core/resin interaction is better understood. Knit line defect formation could be predicted with greater accuracy with suitably modified flow-modelling programs.

The displacement damage induced by primary recoils of fusion neutrons in tungsten and alloys has been studied with self-ion irradiations, followed by damage characterization with electron microscopy. Tungsten and alloys (≤ 5 wt.% Re, Ta, V) were implanted with 2 MeV W+ ions over a dose range of 3.3×1017 - 2.5×1019 W+m-2 at temperatures ranging from 300 to 750°C. Dislocation loops of b = ½<111> (> 60%) and b = <100> were identified, and that ½<111> loops were found more thermally stable. Among loops that were large enough for nature determination, at least 50% were found to be of interstitial type, with larger fractions in high-temperature and high-dose conditions. The diameter of loops did not exceed 20 nm, with the majority being ≤ 5 nm. The loop number density varied between 1022 and 1023 m-3. The effects of ion dose, irradiation temperature, composition and grain orientation on damage microstructure were investigated. In-situ irradiations (150 keV W+ ions) were carried out as a complement to the bulk implantations. Qualitative trends in loop size, geometry and nature with irradiation dose and temperature were similar to bulk irradiated specimens. Also, the dynamics of defects and their effects on the damage evolution were explored. In-situ annealing of irradiated thin foils was performed to investigate the thermal stability of radiation damage in tungsten. The majority of microstructure transformations were completed within 15 min of annealing. However, extended durations did favour the increase of loop size and the fraction of ½<111> loops.

Traumatic skeletal muscle injuries that include loss of large amounts of muscle mass are becoming more common in today’s warfare. Traditional treatments often do not prevent long term functional impairments. Using a decellularized extracellular matrix (ECM) as scaffolding to replace lost muscle tissue allows for transmission of force through the injury site, and provides a suitable microenvironment receptive to myofiber growth. Seeding the ECM with progenitor cells improves cellular content in the defect area. Exercise exposes the muscle to improved blood flow as well as higher than normal loading. This results in increased blood vessel density as well as higher levels of cellular content, and near complete restoration of function. / text

The optimization of powder x-ray diffraction (PXRD) for the study of cheese crystals was the focus of this study. A survey was conducted of various manifestations of calcium lactate crystals on the rindless surface and within mechanical openings of Cheddar cheese using PXRD. The diffraction reference card database contained a card that was entitled calcium lactate pentahydrate and corresponded to some of the crystalline material found on the cheeses. Diffractions patterns generated from other samples of crystalline material revealed the existence of an unknown crystal that resembled and behaved similarly to calcium lactate pentahydrate, but did not match the reference card. The existence of two enantiomeric variants of calcium lactate pentahydrate had been firmly established; an experiment was thus designed to determine if the unknown diffraction pattern represented one enantiomeric form, and if the ambiguously named reference card represented the other. This experiment demonstrated that the existing reference card corresponded to calcium DL-lactate pentahydrate and that the unknown diffraction pattern was generated from calcium L-lactate pentahydrate. This study resulted in the proposal of a new reference card for calcium L-lactate pentahydrate and the proposed renaming of the existing card to calcium DL-lactate pentahydrate. This discovery allows the rapid identification of both forms of calcium lactate that form in and on cheese. In order to conduct the survey and experiment that are described above, the PXRD method needed to be adjusted for use with cheese crystals. Samples of cheese crystals pose a particular challenge because they are often composed of high proportions of moisture, fat, protein, and other amorphous material; these all disrupt the efficient diffraction of crystals and thus needed to be removed or minimized. The removal of water from samples is a particular challenge because some cheese crystals contain water of hydration that may be driven off in the process, thereby destroying the crystals. A protocol for the preparation of cheese samples for PXRD was consequently developed.

Myocardial perfusion imaging (MPI) is an effective technique used to study the left ventricular ejection function (LVEF), myocardial perfusion, wall motion, and wall thickening. Positron emission tomography (PET) and single photon emission computed tomography (SPECT) are two modalities that can be used to quantify the left global and regional perfusion at rest and stress. While PET and SPECT rely on similar principles to produce images, important differences in instrumentation and experimental applications are dictated by inherent differences in their respective physics of radioactive decay. With a sensitivity > 90% in combination with a high specificity, PET is today the best available nuclear imaging technique for the diagnosis of coronary artery disease (CAD). The short half-life of the perfusion tracers in combination with highly sophisticated hard- and software enables rapid PET studies with high patient throughput. Rubidium-82 (82Rb) is a PET perfusion imaging agent that has a shot half-life of 76 seconds which enables multiple sequential data acquisitions in a short duration of time. It also reduces the number of false-positive SPECT scans and artifacts from soft tissue attenuation due to the routine application of attenuation correction. However 82Rb PET imaging is under-utilized clinically due to difficulty optimizing the imaging parameters. The major challenge of 82Rb imaging is determining when to begin the image acquisition post infusion, as imaging too early results in images with high background (low contrast), and imaging too late results in noisy images due to low count statistics. 82Rb rest/stress dynamic and gated data from 16 patients were available for analysis. The FWHM of the 82Rb infusion, LV cavity and LV myocardial uptake in time activity curves were generated and compared to isolate the dominant parameter in determining image quality. The measured and actual infusion-time correlated only at rest (r = 0.93, P = 0.006). Splitting-time at rest and stress correlated (r = 0.74, P = 0.09). But the study was not able to identify a single dominant parameter that would determine the image quality due to the unpredictable nature of hemodynamics during the vasodilatory induced cardiovascular stress. First pass radionuclide angiography (FPRNA) is the gold standard for quantification of ejection fraction. We examined the quantification of the ejection function (LVEF) to determine whether the gated 82Rb PET data, using quantitative gated SPECT (QGS), would accurately predict changes in the chamber volume and correlated the results with those obtained from FPRNA technique. There was a good correlation between the resting FPRNA data and resting gated 82Rb QGS data (r= 0.81, P=0.0005) showing that this method can be applied to 82Rb PET.99mTc SPECT was considered the gold standard for this study, as it is the most widely used technique for myocardial perfusion imaging. The under-perfused area of the myocardium is defined as defect. 99mTc agents, 18F-FDG, and 82Rb can all be used for cardiac imaging 1-7. However, count rates, energy and camera differences can yield image differences that are independent of the actual biological distribution. We examined whether PET with an 82Rb-labeled tracer would provide information on defect size similar to that provided by 99mTc SPECT, using a cardiac phantom in which the true defect size is known. Since 82Rb has such a short half-life (76 seconds), filling and imaging a phantom was going be a great challenge. Hence 124I which is a high-energy radioisotope like 82Rb, was used in this phantom study as a surrogate for 82Rb. Static cardiac phantom studies with 99mTc, 18F and 124I (surrogate for 82Rb) were conducted. The percent defect sizes were measured and compared with the true defect size. Our results demonstrated that at 45% threshold, the measured defect size was representative of true defect size for 99mTc SPECT data. Using this threshold as the standard, we smoothed the 18F and 124I PET data until the measured defect size for PET was representative of the true defect size. An optimal filter cutoff frequency (Butterworth filter, cutoff = 0.80 cycles/pixel, order=5 at 45% threshold for 124I or 82Rb) was found for the PET data within the range of values studied, and this frequency was higher than the clinical norm for SPECT data. Our results also illustrated that the measured SPECT defect size varied greatly depending on the thresholds used to define a defect, whereas measure PET defect size was relatively constant over the range of cutoffs tested7. The optimal cutoff may depend on defect size, patient variability, and noise level. When assessing myocardial defect size, physical properties need to be taken into consideration, particularly when comparing images obtained using different nuclides (i.e. 82Rb or 99mTc agent perfusion and 18F FDG viability).

Bridges are a vitally important part of modern infrastructure. Their condition needs to be monitored on a continuous basis in order to ensure their safety and functionality. Teams of engineers visually inspect more than half a million bridges per year in the US and the EU. There is clear evidence to suggest that they are not able to meet all bridge inspection guideline requirements. In addition, the format and storage of inspection reports varies considerably across authorities because of the lack of standardisation. The availability of a comprehensive and open digital representation of the data involved in and required for bridge inspection is an indispensable necessity for exploiting the full potential of modern digital technologies like big data exploration, artificial intelligence and database technologies. A thorough understanding of bridge inspection information requirements for reinforced concrete bridges is needed as basis for overcoming the stated problem. This work starts with a bridge inspection guideline analysis, from which an information model and a candidate binding to Industry Foundation Classes (IFC) is developed. The resulting bridge model can fully store inspection information in a standardised way which makes it easily shareable and comparable between users and standards. Then, two inspection stages for locating and classifying visual concrete defects are devised, implemented and benchmarked to support the bridge inspection process: In a first stage, healthy concrete surfaces are located and disregarded for further inspection. In a second hierarchical classification stage, each of the remaining potentially unhealthy surface areas is classified into a specific defect type in accordance with bridge inspection guidelines. The first stage achieves a search space reduction for a subsequent defect type classification of over 90% with a risk of missing a defect patch of less than 10%. The second stage identifies the correct defect type to a potentially unhealthy surface area with a probability of 85%. A prototypical implementation serves as a proof of concept. This work closes the gap between requirements arising from established inspection guidelines, the demand for holistic data models which has recently become known as "digital twin", and methods for automatically identifying and measuring specific defect classes on small scale images. It is of great significance for bridge inspectors, bridge owners and authorities as they now have more suitable data models at hand to store, view and manage maintenance information on bridges including defect location and defect types which are being retrieved automatically. With these developments, a foundation is available for a complete revision of bridge inspection processes on a modern, digital basis.

abstract: In this project, current-voltage (I-V) and Deep Level Transient Spectroscopy (DLTS) measurements are used to (a) characterize the electrical properties of Nb/p-type Si Schottky barriers, (b) identify the concentration and physical character of the electrically active defects present in the depletion region, and (c) use thermal processing to reduce the concentration or eliminate the defects. Barrier height determinations using temperature-dependent I-V measurements indicate that the barrier height decreases from 0.50 eV to 0.48 eV for anneals above 200 C. The electrically-active defect concentration measured using DLTS (deep level transient spectroscopy) drops markedly after anneals at 250 C. A significant increase in leakage currents is almost always observed in near-ideal devices upon annealing. In contrast, non-ideal devices dominated by leakage currents annealed at 150 C to 250 C exhibit a significant decrease in such currents. / Dissertation/Thesis / Masters Thesis Materials Science and Engineering 2018