Blast Furnace Wall Bonding Intelligent Diagnostic Model Based On CNN-BiLSTM-Attention

In the complex production environment of blast furnace ironmaking, wall sticking, as a common operational fault, poses a constant threat to both production safety and efficiency. With the ongoing intelligent transformation of the steel industry, traditional diagnostic methods reliant on manual experience, due to their inefficiency and subjectivity, can no longer meet the demands of modern production. In response, we have innovatively developed an intelligent diagnostic model for furnace wall bonding based on CNN-BiLSTM-Attention, leveraging a multimodal deep learning architecture to achieve accurate fault detection. This model adopts a hierarchical feature extraction strategy: first, the CNN (Convolutional Neural Network) is used to capture local features from time series data; then, the BiLSTM (Bidirectional Long Short-Term Memory) model analyzes temporal dependencies; finally, the attention mechanism automatically focuses on key feature nodes. Through the sigmoid function in the fully connected layer, the system outputs an intuitive prediction of the likelihood of sticking, enabling intelligent discrimination of furnace wall status. Experimental validation shows that the model achieves an AUC of 0.87 in the ROC curve, with a diagnostic accuracy of 88.4% and a root mean square error significantly better than traditional methods. This study provides the steel industry with an innovative intelligent monitoring solution, which, through data-driven decision support, not only effectively prevents production accidents but also propels the ironmaking process toward higher efficiency, intelligence, and sustainability, marking a new phase in the deep integration of traditional heavy industry and artificial intelligence technologies.