Metallographic Spheroidization Rate Classification by Using Deep Learning

Abstract

In the steel manufacturing process, spheroidizing annealing is a crucial heat treatment step primarily aimed at improving the ductility and machinability of the material. Currently, the determination of the spheroidization rate in metals mainly relies on manual inspection through a microscope. These methods are time-consuming and subject to inconsistent subjective judgments. To overcome these challenges, this paper proposes a deep learning method for classifying metallographic spheroidization rates using an improved YOLOv8 model, referred to as YOLOv8-DFFN. This model integrates channel attention (CA) and vital feature fusion (VFF) techniques, effectively increasing the classification accuracy for different spheroidization levels. Experimental results show that the YOLOv8-DFFN model achieves a mean average precision (mAP) of 98.17% across metallographic datasets of various alloy compositions. This represents an improvement of 1.42% over the baseline model. Additionally, the YOLOv8-DFFN model surpasses the performance of the original YOLOv8 algorithm. This innovative technology is expected not only to enhance production efficiency and material quality but also to significantly reduce inspection costs and human resource investment. It will contribute to the continuous innovation and advancement of the metal processing industry.