Intelligent diagnosis model of blast furnace wall bonding based on decision tree-deep neural network

The condition of the blast furnace is fundamental to ensuring the stability of the entire production process in the iron and steel industry. Traditional methods of monitoring are often prone to human error and are influenced by subjective judgment. As a result, the integration of automated and data-driven technologies has become indispensable in improving both the efficiency and quality of ironmaking. Abnormal furnace states, such as sliding scale, wall bonding, hanging, collapsing and hearth accumulation. To enhance the speed and reliability of blast furnace operations, it is crucial to accurately identify and address the underlying causes of these anomalies. This article introduces an intelligent diagnostic model designed to detect abnormal furnace conditions by combining decision trees and deep neural networks. Diagnostic features are selected based on expert-driven rules using a decision tree, while a deep neural network classification model is employed for diagnosing the furnace conditions. The model is trained using 80% of the available data in chronological sequence, with the remaining 20% reserved as a test set for optimisation. Upon evaluation, the model achieves an accuracy of 81%. A comparative analysis of four diagnostic models is conducted, utilising precision, recall, accuracy and F1-score as metrics, with the proposed model demonstrating superior performance across all criteria. The results demonstrate that the model effectively captures furnace conditions, with diagnostic outcomes closely aligning with actual anomalies, thereby supporting stable furnace operations. After the implementation of the blast furnace abnormal condition intelligent diagnosis system, production efficiency was significantly improved, response times for abnormal condition handling were greatly shortened, and unnecessary downtime was effectively avoided. The fuel ratio was reduced by nearly 2.6 kg·t − 1 , the utilisation coefficient increased by ∼4%, and the number of significant fluctuations was significantly decreased. This system has provided automation and data support for the ironmaking industry, contributing to the advancement of the sector towards greater intelligence.