PR-DT1020-1 - 2020 Digital Transformations, October

Inclusion Classification by Computer Vision and Machine Learning

N. Gao, M. Abdulsalam, M. Potter, G. Casuccio, et al.

Download PDF

View Sample Page

This paper describes the use of computer vision and machine-learning methods to classify non-metallic inclusions in steel based on backscattered electron (BSE) scanning electron microscope (SEM) images obtained during automated inclusion analysis. The use of automated inclusion analysis has produced major contributions to both control of inclusions during steel processing and a mechanistic understanding of inclusion evolution.13 Automated analysis utilizes an SEM equipped with a BSE detector and energy-dispersive x-ray spectroscopy (EDS). Thousands of features can be observed in times on the order of hours, yielding representations of the variable distributions. BSE images provide quantitative information on inclusion amount, size, shape and location, whereas EDS spectra provide information on chemical composition. BSE images also contain information about inclusion composition, since the production of backscattered electrons increases with atomic number. The objective of this work was to create a system that relates BSE images to EDS composition measurements. This required conversion of the BSE images into a numerical representation so that they could be interpreted by a computer.

Keywords: digital transformation, industry 4.0