Metallurgical microstructure classification using CNN: A comprehensive study on heat treatment analysis for steel

Abstract

Predicting metallurgical microstructures is challenging due to the intricate features that emerge during manufacturing. This research utilizes Convolutional Neural Networks (CNNs) to recognize microstructure images of mild steel alloys, representing diverse microstructures. The study focuses on forecasting steel microstructure properties, specifically identifying and predicting Pearlite (P) and Bainite (B) microstructures after steel heat treatment, using a CNN approach. This comprehensive workflow leads to robust microstructure classification, achieving a model training accuracy rate of 98.95 %. While its final validation accuracy of 98.8 %, highlighting its capability to generate precise predictions across both classes seamlessly. Valuable insights from confusion matrices highlight the model’s performance in distinguishing between “Bainite” and “Pearlite” microstructures. Additionally, the study underscores the significance of CNNs in metallurgical microstructure analysis, with insightful confusion matrices revealing the model’s capability to differentiate between microstructures. This is useful for innovative microstructure analysis in materials science, emphasizing the transformative potential of AI in enhancing our understanding of processing-microstructure relationships in industrial applications.