# A Machine Vision-Enhanced Framework for Tracking Inclusion Evolution and Enabling Intelligent Cleanliness Control in Industrial-Scale HSLA Steels

## TL;DR

This paper introduces a machine vision system to track and control non-metallic inclusions in HSLA steel during industrial production, improving steel purity and quality.

## Contribution

The novel framework integrates motion control, optical imaging, and laser spectral analysis for automated inclusion tracking and intelligent cleanliness control in large-scale HSLA steel production.

## Key findings

- Type D inclusions decreased significantly during electroslag remelting, from over 8000 in the electrode to 4000–7000 in the ingot.
- Type C silicate inclusions increased fourfold in the ingot tail due to solidification segregation and flotation dynamics.
- Forging reduced all inclusion types to extremely low levels in the billet tail, nearly eliminating Types A, B, and C.

## Abstract

The quantity, size, and distribution of non-metallic inclusions in High-Strength Low-Alloy (HSLA) steel critically influence its service performance. Conventional detection methods often fail to adequately characterize extreme inclusion distributions in large-section components. This study developed an integrated full-process inclusion analysis system combining high-precision motion control, parallel optical imaging, and laser spectral analysis technologies to achieve rapid and automated identification and compositional analysis of inclusions in meter-scale samples. Through systematic investigation across the industrial process chain—from a dia. 740 mm consumable electrode to a dia. 810 mm electroslag remelting (ESR) ingot and finally to a dia. 400 mm forged billet—key process-specific insights were obtained. The results revealed the effective removal of Type D (globular oxides) inclusions during ESR, with their counts reducing from over 8000 in the electrode to approximately 4000–7000 in the ingot. Concurrently, the mechanism underlying the pronounced enrichment of Type C (silicates) in the ingot tail was elucidated, showing a nearly fourfold increase to 1767 compared to the ingot head, attributed to terminal solidification segregation and flotation dynamics. Subsequent forging further demonstrated exceptional refinement and dispersion of all inclusion types. The billet tail achieved exceptionally high purity, with counts of all inclusion types dropping to extremely low levels (e.g., Types A, B, and C were nearly eliminated), representing a reduction of approximately one order of magnitude. Based on these findings, enhanced process strategies were proposed, including shallow molten pool control, slag system optimization, and multi-dimensional quality monitoring. An intelligent analysis framework integrating a YOLOv11 detection model with spectral feedback was also established. This work provides crucial process knowledge and technological support for achieving the quality control objective of “known and controllable defects” in HSLA steel.