SRF2015 - List of Authors (Iriks, D.C.)

Paper	Title	Page
TUPB082	Automatic Surface Defect Detection and Sizing for Superconducting Radio Frequency Cavity Using Haar Cascades	788
	G.V. Eremeev JLab, Newport News, Virginia, USA D.C. Iriks Santa Rosa Junior College, Santa Rosa, USA
	Serious albeit tiny surface defects can remain on the surface of superconducting radio frequency (SRF) cavities after polishing and cleaning. These defects reduce the efficiency of cavities and often limit the maximum attainable fields. We applied a Haar cascade artificial vision technique for automated identification, counting, and sizing of defects induced on niobium surface by Nb-H precipitates formed at cryogenic temperatures. The defects were counted and sized by a computer program and also counted and measured manually to estimate detection rate and accuracy of sizing. The overall detection rate was 53%, and the overall false positive rate was 29%. The technique that was used to automatically size the features was found to oversize the features, but oversize them consistently, resulting in a size histogram that represents the defect size distribution on the sample. After scaling the histogram data, the average defect area was found to be 90 square micrometers with the standard deviation of 70 square micrometers.
Export •	reference for this paper to ※ BibTeX, ※ LaTeX, ※ Text, ※ RIS/RefMan, ※ EndNote (xml)

Paper

Title

Page

Automatic Surface Defect Detection and Sizing for Superconducting Radio Frequency Cavity Using Haar Cascades

788

G.V. Eremeev
JLab, Newport News, Virginia, USA
D.C. Iriks
Santa Rosa Junior College, Santa Rosa, USA

Serious albeit tiny surface defects can remain on the surface of superconducting radio frequency (SRF) cavities after polishing and cleaning. These defects reduce the efficiency of cavities and often limit the maximum attainable fields. We applied a Haar cascade artificial vision technique for automated identification, counting, and sizing of defects induced on niobium surface by Nb-H precipitates formed at cryogenic temperatures. The defects were counted and sized by a computer program and also counted and measured manually to estimate detection rate and accuracy of sizing. The overall detection rate was 53%, and the overall false positive rate was 29%. The technique that was used to automatically size the features was found to oversize the features, but oversize them consistently, resulting in a size histogram that represents the defect size distribution on the sample. After scaling the histogram data, the average defect area was found to be 90 square micrometers with the standard deviation of 70 square micrometers.

Export •

reference for this paper to ※ BibTeX, ※ LaTeX, ※ Text, ※ RIS/RefMan, ※ EndNote (xml)