Investigation of neural network algorithm for in-situ X-ray tomography
Visualitza/Obre
Estadístiques de LA Referencia / Recolecta
Inclou dades d'ús des de 2022
Cita com:
hdl:2117/356555
Tipus de documentProjecte Final de Màster Oficial
Data2021-11-10
Condicions d'accésAccés obert
Tots els drets reservats. Aquesta obra està protegida pels drets de propietat intel·lectual i
industrial corresponents. Sense perjudici de les exempcions legals existents, queda prohibida la seva
reproducció, distribució, comunicació pública o transformació sense l'autorització del titular dels drets
Abstract
X-ray tomography is an important non-destructive technique for materials structure analysis. In certain applications, especially during in-situ experiments, the constraints posed by the experimental conditions limit the image quality obtainable from the limited data acquired. Commonly used direct image reconstruction algorithms tend to produce images with insufficient accuracy when fed with limited data, while the more accurate iterative algorithms introduce the challenge of high computational cost. A proposed alternative is the use of machine learning to
improve the image quality of direct algorithms. The Mixed-Scale Dense convolutional neural network algorithm (M-SDNet) was therefore utilized in this study to quantitatively investigate its effect in improving the image quality of image reconstructions using direct algorithms, for in-situ tomography. Results are shown for the effect of number of projections, threshold values, and resolution, for data acquired in laboratory conditions. The cavities present in the studied sample were the focus of the quantitative analysis, where parameters like number of cavities, sphericity, and volume fraction were tracked across the output images from using the M-SDNet algorithm. Two different training strategies of M-SDNet; segmentation training and regression training, were compared with the segmentation training proving to better at reproducing cavities in the output images. The reduction on the number of projections and the required scan time suggest that the Mixed-Scale Dense networks are able to significantly improve the accuracy of image reconstructions, and thus suitable to overcome the experimental constraints during in-situ tomography
MatèriesTomography -- Industrial applications -- Software, Image processing -- Digital techniques -- Mathematical models, Materials -- Materials -- Computer simulation, Tomografia -- Aplicacions industrials -- Programari, Imatges -- Processament -- Tècniques digitals -- Models matemàtics, Materials -- Anàlisi -- Simulació per ordinador
TitulacióMÀSTER UNIVERSITARI EN ENGINYERIA NUCLEAR (Pla 2012)
Col·leccions
Fitxers | Descripció | Mida | Format | Visualitza |
---|---|---|---|---|
master-thesis-report-justice-nwade.pdf | 3,924Mb | Visualitza/Obre |