Monitoring Multiple Operational Statuses of Blast Furnace via Multifeature Fusion From Burden Surface Video Images

Accurate monitoring of various operational statuses within blast furnaces (BFOS) is crucial for optimizing furnace regulation and ensuring stable operation. However, existing methods struggle to directly perceive and accurately monitor multiple types of BFOS. Therefore, a novel BFOS monitoring method is proposed, integrating multifeature extraction and fusion from burden surface video (BSV) images. First, a novel industrial endoscope (NIE) is developed to acquire abundant BSV data, and an unsupervised method based on multiscale texture fuzzy C-means (MTF-FCM) is established to accurately obtain the high-temperature gas flow (HGF) area within the BSV images. Second, a novel BFOS monitoring network (BOSMNet) is designed to recognize various BFOS, including hanging, collapsing, high stockline (HS), low stockline (LS), oblique stockline (OS), abnormal gas flow (AGF), and normal statuses. To achieve superior monitoring performance, BOSMNet integrates three key modules for multifeature extraction and fusion from BSV images: the multilevel feature adaptive extraction and fusion module (MFAEFM), the handcrafted features of high-temperature gas image extraction module (HHFEM), and the sequential feature of BSV extraction module (SFBEM). Extensive experiments and ablation studies have validated the effectiveness of these modules and demonstrated that BOSMNet can achieve accurate, online monitoring of multiple BFOS, providing significant assurance for smooth furnace operation.