Classification of Steel Microstructure Image Using CNN

Abstract

The purpose of this study is to create a computerized system that can automatically evaluate microstructure images of steel materials, specifically focusing on ferrite, using a Convolutional Neural Network (CNN) model. Steel materials play a crucial role in our everyday lives, and their mechanical properties and reliability are determined by their microstructure, which is influenced by heat treatment and processing. It is essential to ensure the quality of steel, as problems can arise if the microstructure and mechanical properties are not adequately assessed before shipping. To accomplish this, the study involved preparing four different steel specimens with varying material properties and heating conditions, which were then photographed using a digital camera. The proposed CNN model was tested and validated to accurately classify the ferrite substances, and it was found that even a simple CNN structure could achieve high accuracy in image classification. The implementation of this system will alleviate the burden of human visual inspection. The paper provides detailed information on the preparation of the steel specimens, the method used to capture the images, the structure of the proposed CNN model, the experimental conditions, the validation methods employed, and the results obtained.