Kinetic Oxidation Analysis in AISI 1045 Steel Using Infrared Thermography and Convolutional Neural Networks

TL;DR

This paper introduces a new method using infrared thermography and deep learning to study oxide layers on steel surfaces, improving monitoring in steelmaking.

Contribution

The novel integration of infrared thermography and CNNs for non-destructive oxide layer analysis in steel processes.

Findings

The CNN model achieved 96.40% accuracy in identifying oxide layers on steel surfaces.

The method quantified oxide evolution kinetics and activation energy within known ranges.

The approach enables large-scale non-destructive monitoring, enhancing industrial process control.

Abstract

This study presents a pioneering approach, integrating infrared thermography and deep learning to analyse surface oxide layers on AISI 1045 steel, addressing the critical need for advanced monitoring in steelmaking processes. Using thermography for observation and semantic segmentation for accurate identification, 50 tests between 200 and 700 °C were analysed in a Joule-controlled heating system to study the formation and thickening of oxide layers on steel surfaces. A convolutional neural network (CNN), specifically SegNet, was trained for semantic segmentation, facilitating detailed analysis. The model achieved an overall accuracy of 96.40% in identifying the presence of oxide. By quantifying pixelation changes, relationships in oxide evolution kinetics were obtained, and by quantifying the activation energy in isothermal cases, the magnitude is in the range reported by other works. The approach also highlighted the potential for non-destructive monitoring and control on a large scale without compromising personnel safety. This potential could improve industrial process control, predict surface quality or provide data relevant to sub-processes.

Figures