Fully Actuated System-Based Deep Learning Method for Blast Furnace Fault Diagnosis

The complexity and variability of blast furnace operations pose substantial challenges for fault detection in the metallurgical industry. To improve the reliability and precision of blast furnace condition fault diagnosis, this study innovatively proposes a novel framework that integrates the Fully Actuated System-Based Deep Learning Method for Fault Diagnosis (FAFD). First, a Multivariate Time-Series Encoder (MTE) module is applied to extract critical features across multiple temporal scales to enhance the representation of time-frequency signals. Then, a Fully Actuated Feature Extractor (FAFE) module is developed based on the fully actuated method, effectively leveraging multi-source sensor data and optimizing fault recognition. Experimental validation on real-world datasets from an industrial smelting plant demonstrates that the proposed method surpasses state-of-the-art techniques in precision, recall, and other key metrics, showcasing superior generalization capability and industrial adaptability.