Automated Detection of Non-metallic Inclusion Clusters in Aluminum-deoxidized Steel

This study presents a method to automatically identify inclusion clusters within a sample. Utilizing the output of scanning electron microscopy's automated feature analysis along with Density-based Spatial Clustering of Application with Noise, an unsupervised machine learning algorithm, inclusion clusters are identified based on their spatial position. The analysis was initially conducted on two samples known by manual analysis to be clustered and non-clustered to evaluate the applicability of this technique. A serial-sectioning analysis was performed to obtain a 3D representation of the inclusion distribution. The 2D and 3D results were consistent. To evaluate the area effected by clusters, the convex hull area was utilized rather than the total inclusion area in a cluster. The analysis was then applied to a series of samples from three aluminum-alloyed heats to investigate cluster evolution throughout the secondary steelmaking process. Several distinct types of clusters were identified. Agglomerated globular alumina inclusion clusters were observed after tapping, which then evolved to non-globular inclusion clusters. The same types of clusters were also observed for spinel inclusions, but they were not as pervasive as alumina inclusions. In addition, clustering of small micro-inclusions around a large macro-inclusion was occasionally observed.