Conventional and Machine Learning (ML)-based image analysis techniques for superconducting wires and cables

• Algirdas Baskys - algirdas.baskys@cern.ch

Room: 30/7-018 - Kjell Johnsen Auditorium Image analysis is a powerful technique that can be applied to various aspects of superconducting material research. The value of an automated image analysis approach is especially apparent when the data volume is high and a systematic approach is needed. The images/micrographs can reveal the nature and cause of defects in cables and wires as well as provide a proxy for other material properties that are not possible or more difficult to measure directly, given that structure-to-property relationships are known (or can be established). I'll discuss how conventional image analysis methods excel when image color/brightness/edge contrast directly encodes features and where their limitations lie. In cases where feature morphology and adjacent background are a factor in pixel classification, the machine learning-based segmentation approach is often the better choice. The seminar will feature several case studies showcasing image analysis with unique challenges and approaches, including quality control of Rutherford cables and measurements of nanoprecipitate and grain sizes in Nb3Sn wires with artificial pinning centers. Ongoing work in cracked subelement detection within reacted Nb3Sn wires will also be presented.