IDENTIFICATION OF FISSION GAS VOIDS Ryan Collette
Introduction The Reduced Enrichment of Research and Test Reactor (RERTR) program aims to convert fuels from high to low enrichment in order to meet non-proliferation goals High atom density requirements in test reactors has forced the development of new fuel types Extensive research into uranium-molybdenum plate-type fuels Microstructural characterization of these fuels can provide information as to their behavior under various irradiation conditions
Fission Gas Bubbles Release of gaseous fission products (Xenon, Krypton) into fuel matrix during irradiation can lead to fuel swelling and possible delamination of the cladding Porosity determinations (bubble count, distribution, volume fraction) can aid in the evaluation of fuel performance Current methods involve hand counts and visual inspection of images An automated methodology could allow for more precise correlations Similar studies are limited, but cells provide a valid parallel
Cell Segmentation Methods Thresholding Global thresholding Local adaptive thresholding Edge detection Linking procedure Feature matching Region growing Watershed Pre-processing Notch filtering Noise removal Feature smoothing Grayscale morphology Post-processing Binary morphology Data extraction
Image Considerations Images provided by Idaho National Laboratory Samples milled with a focused ion beam (FIB) and imaged with a scanning electron microscope (SEM) FIB milling creates a curtaining effect in the image
Sample FIB-SEM images
Preprocessing Removing the curtaining Frequency domain filtering High frequencies correlate to edges in image Low frequencies represent areas with more constant gray levels Notch filter using FFT (Fast Fourier Transform) Basic algorithm Convert spatial image to frequency domain Shift spectrum for visualization Identify frequencies corresponding to uniform edges Attenuate those frequencies (zero out with a mask) Return to spatial domain
Notch filtration example
Noise reduction Histogram equalization and contrast adjustment reveals significant peppering between void events Techniques tested: Global averaging filter Median/Gaussian blur out small voids and leave halos Wiener filter Effective, but ultimately not enough information about the noise Anisotropic diffusion Deals with aliasing well, but orientation capability irrelevant Bilateral filtering Simple edge preserving smoothing but suits the needs of this project
Bilateral Algorithm Two Gaussians operate on localized pixel neighborhood Domain Filter (Spatial) Range Filter (Intensity) Pixels with very different intensity values are weighted less even though they may have close proximity to central pixel Behaves similarly to standard domain filter in smooth regions At edges, similarity function will ignore pixels on opposite side of step edge and average similar intensity pixels in the vicinity
Thresholding Otsu Global Uneven illumination in some images makes this unrealistic Local adaptive Threshold varies on per pixel basis based on image characteristics Sauvola method Threshold computed using the mean and standard deviation of pixel intensities in window around target pixel
Edge detection Sobel, Prewitt, Roberts Identify only edges we want Rarely captures the full contour of the void Hysteresis edge linking algorithm improves connectivity, but less prominent voids are lost In conclusion, edge detection methods are not precise enough to warrant consideration over thresholding methods
Post-processing Binary morphological operations Openings and closing to combine or divide connected components Filling of holes within objects Clearing of objects touching borders Regionprops Object count - Size distribution - Total void volume fraction Porosity: 3.8% Hand count: 294 events Matlab: 286 events
Sample data correlations File name Count Hand Count Deviation (%) Mean Area (µm^2) Area Fraction (%) MZ 50C_XS_Site 1 41 46 10.8 0.020 1.03 MZ 50C_XS_Site 2 65 58 12.1 0.028 1.57 MZ 50C_XS_Site 3 93 87 6.9 0.026 2.16 MZ 50C_XS_Site 4 107 104 2.9 0.024 2.33 MZ 50C_XS_Site 5 76 68 11.7 0.027 2.40 MZ 50C_XS_Site 6 279 244 14.3 0.016 2.82 MZ 50C_XS_Site 7 329 313 5.1 0.014 2.56 MZ 50C_XS_Site 8 284 267 6.4 0.014 2.40
Issues and possible improvements Identifying voids as singular events remains problematic due to the sharp gradients created by solid fission products Method susceptible to false positive identifications based on background texture Possible feature detection system Lower magnification images identified to a higher degree of confidence Verification necessary to assure accuracy of segmentation Hand counts subjective. What is a gas void and what isn t? Synthetic images? Sample preparation and imaging conditions need to be consistent to make the process repeatable