| Abstract Scope |
The δ of STS 300 series steels has a high solubility of γ grain boundary embrittlement elements (p, s), and suppresses the growth of grain boundaries, thereby reducing the occurrence of linear defects. So, obtaining sufficient δ fraction can suppress surface cracks that occur during the rolling process, and securing δ fraction is a key factor in component design. This study measures the shape change of steel grades that transform into austenite after delta solidification based on in-situ images. Image data on the growth process of acicular microstructure and vermicular microstructure are analyzed in real time using an artificial neural network. This artificial neural network structure shows good performance in image recognition for weakly-supervised data, and by applying a conditional random field, the resolution of the output image is increased and the IOU (Intersection over Union) performance is improved. |