new statistical methods ( also called “ clean steel measurements ”) 8 . By leveraging the vast amounts of metallographic structure inspection data along with these image processing technologies , advanced deep learning techniques can be developed to inspect the image samples and detect non-metallic inclusions . With experienced human inspectors to help train these AImodels , productivity should increase , so long as the inspector is able train the models with relative ease and not be burdened with understanding a new set of tools to perform the training .
Currently , national regulations require human inspection and prohibit the use of AI to perform inspection by itself . As the technology matures , and AI models are proven to be as reliable as experience human inspectors , such regulations could change . In fact , the steel industry has begun developing standards for automatic inclusion inspections 9 . However , for the present , it is not expected to eliminate human inspectors ; rather , human inspectors are considered high value assets and critical to the measurement process .
8
9
ASTM : ASTM E768 - 99 ( 2018 ), Standard Guide for Preparing and Evaluating Specimens for Automatic Inclusion Assessment of Steel . s . l . : ASTM , 2018 .
76 November 2021