Deep learning for blast furnaces: Skip-dense layers deep learning model to predict the remaining time to close tap-holes for blast furnaces

Abstract

Manufacturing steel requires extremely challenging industrial processes. In particular, predicting the exact time instance of opening and closing tap-holes in a blast furnace has a great influence on steel production efficiency and operating cost, in addition to human safety. However, currently predicting the time to open and close tap-holes of the blast furnace still highly relies on manual human expertise and labor. Also, most of the prior research is limited to indirectly model the level of liquids in the hearth, using complex mathematical models or classical machine learning approaches. In this paper, we use a data-driven deep learning method to more accurately predict the remaining time to close each tap-hole in a blast furnace and develop an AI-enabled automated advisory system to reduce manual human efforts as well as operation cost. We develop a multivariate time series forecasting algorithm using Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) models to more accurately predict the opening and closing time for the Pohang Iron and Steel Company (POSCO) blast furnace. In particular, we use and validate data from one of the largest operating furnaces in the world to develop our system. Our proposed Skip-dense CNN (S-CNN) model achieves more than 90% accuracy within ±30 minutes tolerance, compared to other LSTM baseline models. Our S-CNN model has been successfully deployed at a large-scale blast furnace of POSCO since January 2018 and has achieved similar accuracy. And we even exceeded the reported human performance in a real operational environment.