부산대학교 도서관

The influence of MnS inclusions on steel properties is highly noticeable. For instance, higher severity levels of inclusions are associated with lower mechanical properties and a higher risk of failure in service. Manual inclusions classification methods are the most used in laboratories and metallurgical sector industries because of their low cost, while automatic methods have high operating costs, which makes their use more restrict. Neural network models, on the other hand, are extremely advantageous for several applications. The present study is motivated by the use of a neural network model for classifying inclusions in steels. The aim is to achieve the highest possible accuracy in classifying the MnS inclusions severities (0, 0.5, 1.0, 1.5, 2.0, 2.5, 3.0, 3.5, 4.0, and 4.5) using optical images captured from steel specimens. The results have showed that the classification of MnS severities was very sensitive to the database number of images. A 98% training accuracy was obtained by increasing the number of images from 3,156 to 4,136, mostly adding images for some severity levels. However, validation and test results were not satisfactory. As such, a severity re-categorization of the database was able to enhance the true positive values, with an error of 8%. In general, the neural network represented speed in decision making, proving to be a potential tool for classifying steel inclusions.