A Deep Learning–Based Diagnosis Model Driven by Tuyere Images Big Data for Iron‐Making Blast Furnaces
Qiang Li, Zijia Wang, Shuai Wang, Mingming Li, Lei Hong, Zongshu Zou
Abstract
Deep learning, applied to traditional industries, makes many challenges and previously impossible tasks turn practicable. However, progress is still limited in the steelmaking industry. This study aims to establish a deep convolutional neural network–based diagnosis model for blast furnaces (BF) driven by tuyere image big data, which is not effectively exploited but contains valuable information. The model is characterized by having the ability to learn value features of abnormal events hidden in images and acts as a high‐dimensional complex function of abnormal events correlated to image features. The implementation framework of the model includes two stages: offline training based on the labeled historical database and online real‐time diagnosis. Specifically, the image data are first de‐noised and enhanced via the median filter and Gamma transformation algorithms. Then, when the model is offline trained, a preferable clustering algorithm combined with the manual determination is applied to labeling image data. Finally, the built model is examined for accuracy and performance. It is shown in the results that the model has an accuracy of 98% and 88.58% for diagnosing results in training and validating datasets, respectively. Thereby, the established model may be applied to the online diagnosis of an industrial BF using real‐time images.