Judgment Model of a Copper-Converter End Point Based on a Target Detection Algorithm

The end-point determination of the slag- and copper-making periods in copper converting directly affects the quality of copper, stability of the furnace condition, and efficiency of converting. However, existing end-point determination methods such as manual experience, sampling analysis and detection, and flame image recognition have problems such as poor accuracy, high cost, and lagging judgment. In this paper, target detection is used to determine the end point of copper converting. By integrating the efficient-channel attention (ECA) mechanism and Res2Net module into the You-Only-Look-Once-v5s (YOLOv5s) target detection algorithm and using Alpha Intersection-Over-Union to replace the Intersection-Over-Union loss function, we constructed a copper-converting end-point determination model based on the YOLOv5s-Res2Net-ECA collaboration. The results show that the high-temperature melt cooling image data from 11 time points during the blowing process of ten copper-converting furnaces were used as samples for training. Then, end-point determination prediction was carried out. Compared with YOLOv5s, the average accuracy of the constructed model was improved by 4.94%, reaching 89.25%. The average accuracy of the end-point determination of the slag-making and copper-making periods was 97.4% and 98.4%, respectively. The results lay a foundation for the intelligent determination of the end point of copper converting.