A novel intelligent monitoring method for the closing time of the taphole of blast furnace based on two-stage classification

https://doi.org/10.1016/j.engappai.2023.105849Get rights and content

Abstract

Determining the taphole closing time is an essential task in the blast furnace ironmaking process because the closing time directly affects the efficiency of iron production and the stability of the blast furnace. However, at present, the taphole closing time in most ironmaking plants is judged by on-site workers based on experience, which lacks scientific guidance. To determine the taphole closing time intelligently and accurately, a novel monitoring method is proposed, which innovatively simplifies the monitoring problem of the absolute taphole closing time into a two-stage classification problem of relative tapping state. In the first stage, a classification algorithm SE-ResNeXt, which only takes the molten iron flow image data as the input data, is used to preliminarily determine the current molten iron flow state in the time dimension during tapping. When it is recognized that the molten iron flow is in the last tapping state in the first stage, the second stage is carried out. In the second stage, a novel multimodal data fusion network SENeXt-Decoder consisting of a novel image feature extraction module, a novel fusion module and a multi-head attention decoder is proposed to obtain the exact taphole closing time, which fuses the molten iron flow image data and blast furnace operating state data. The comparison experiment with the actual taphole closing time on site shows that the absolute monitoring error of this method is within 120 s, and the relative monitoring error is within 1.2%, which better meets the factory’s demand for error accuracy.

Introduction

The steel industry is a significant basic industrial sector that provides steel raw materials for many industries and supports the development of the national economy and national defense construction. Blast furnace (BF) ironmaking is an important process in the production of iron and steel (Pan et al., 2018, Zhou et al., 2021), and its products are liquid iron and slag. In the BF ironmaking process, raw materials such as sinter and coke will be put into the BF from the top of the BF, and then a series of complex redox reactions will occur under the action of high-temperature gas in the furnace. Finally, the produced molten iron and slag drip down into the BF hearth, where they wait to be discharged from the furnace (Jiang et al., 2021).

The process of discharging these liquid molten iron and slag is called tapping, as shown in Fig. 1. The purpose of tapping is to discharge slag and molten iron in time and prevent the interference of high liquid levels in the hearth to the stable operation of the BF, which is of crucial importance to prevent catastrophic events in the BF operation (Chen, 2001). At the end of tapping, to prevent the gas in the furnace from escaping with the slag and molten iron, the taphole must be closed with the mud gun machine in time. If the taphole is closed too early, the molten iron and slag in the furnace will not be discharged completely. If the taphole is closed too late, the gas in the furnace will eject, which would cause potential safety hazards. Therefore, the timing of closing the taphole is very important.

Currently, monitoring the time to open and close the tapholes of the blast furnace still highly relies on manual human expertise and labor in most ironworks (Kim et al., 2019), which is too subjective and less intelligent. According to existing studies, the ideas for determining the blocking time of taphole can be divided into four categories: based on the predicted liquid level in the hearth, based on electromotive force signal analysis, based on BF operating state data (Stand for some operating variables of BF), based on molten iron flow image at the taphole. The method based on the predicted liquid level in the hearth needs to use the mass balance equation to build a mathematical model. A disadvantage of this type of method is that it requires too many unrealistic mathematical assumptions (Brännbacka and Saxén, 2001, Brännbacka and Saxen, 2004, Upadhyay and Kundu, 2013, Agrawal et al., 2016, Roche et al., 2018, Roche et al., 2019). The method based on electromotive force signal analysis needs to install an EMF probe on the blast furnace hearth and then process the EMF signal. The disadvantage of this method is that the EMF signal is susceptible to other electrical signals and usually exhibits strong drift, which makes it unsuitable for measuring absolute liquid levels in furnaces (Brännbacka and Saxen, 2004). The existing method based on BF operating state data is oriented to the unique ultra-large blast furnace, so the method is poor versatility (Kim et al., 2019). The existing method based on molten iron flow image requires manual feature extraction, which has limited feature extraction ability and weak anti-interference ability. In addition, the molten iron flow image can often only reflect the rough information of the tapping process, and it is difficult to reflect the taphole closing time precisely.

The existing studies provide some meaningful solutions to the monitoring problem of the closing time of taphole, but their limitations are also very apparent. In this paper, a method is proposed to utilize both molten iron flow images and blast furnace operating state data. With the development of artificial intelligence and the improvement of computing power, deep neural networks can better automatically learn the hierarchical feature representation of the image and show satisfactory performance in learning representations of industrial process operating state data with multiple levels of abstraction (Ragab et al., 2021). A lot of deep neural networks are designed to solve vision classification or monitoring problems in various engineering fields, such as the field of vision include (Tang et al., 2022a, Hu et al., 2017, Dai et al., 2021), the field of monitoring include (Sun and Ge, 2021, Khanafer and Shirmohammadi, 2020, Chen and Pollino, 2012). These methods also provide a new idea for the realization of intelligent closing operation of the taphole in the ironmaking process. Therefore, this paper proposes a novel monitoring method for the closing time of the taphole based on two-stage classification, in which the SE-ResNeXt network utilizes molten iron images to accurately obtain the relative time state of tapping in the first stage and a novel deep neural network SENeXt-Decoder is designed to use the fusion data of molten iron flow image data and blast furnace operating state data to obtain the exact closing time of the taphole in the second stage. The specific contributions of this paper are as follows:

(1) The monitoring problem of absolute taphole closing time is transformed into a two-stage classification problem of relative taphole closing time, and a novel two-stage classification strategy for monitoring the closing time of the taphole is proposed to achieve high monitoring accuracy.

(2) For the first time, a solution to monitoring the closing time of taphole by fusing molten iron flow image data and BF operating state data is proposed, which provides a new idea for the intelligent determination of the closing time of taphole.

(3) A novel deep learning classification network SENeXt-Decoder for multimodal data fusion is proposed to obtain the exact closing time of the taphole, which can fuse molten iron flow image data and BF operating state data and achieve high-precision classification.

The structure of this paper is as follows. Section 2 describes the related work on the taphole closing time monitoring problem. Section 3 describes the tapping process of BF ironmaking and problem formulation. Section 4 details the monitoring method based on two-stage classification. Section 5 analyzes the experimental results. Section 6 summarizes the conclusions.

Section snippets

Related work

According to existing studies, the ideas for determining the blocking time of taphole can be divided into four categories: based on the predicted liquid level in the hearth, based on electromotive force signal analysis, based on BF operating state data, based on molten iron flow image at the taphole.

The method based on the predicted liquid level in the hearth is modeled using the molten iron and slag mass balance equation. By estimating the production rate and outflow rate of molten iron and

Blast furnace tapping process description

Fig. 1 shows the three-dimensional simulation diagram of the BF tapping process. When the amount of molten iron and slag in the BF reaches a certain amount, a taphole is opened with a drilling machine to make the molten iron and slag flow into the main pipeline of molten iron, and then the molten iron is transferred to the steelmaking process (Pan et al., 2020). When the tapping process is completed, the taphole needs to be blocked in time by the mud gun machine to prevent gas leakage from the

Overall structure

To determine the closing time of the taphole intelligently and accurately, this paper proposes a novel monitoring method. The novel monitoring method framework is shown in Fig. 4. In the first stage, the whole tapping cycle is divided into n time segments as n categories. With the start of the tapping process, the molten iron flow image data of the taphole is collected and preprocessed in real-time. Then the preprocessed images are classified by the classification algorithm, if the

Experiment and results

To verify the effectiveness of the method proposed in this paper, we did a series of experiments on the method.

Conclusion

Monitoring the closing time of taphole intelligently and accurately is of great significance to ensure the safety of BF ironmaking and improve iron production efficiency. This paper proposes a novel intelligent monitoring method for the closing time of the taphole of BF based on two-stage classification, which simplifies the monitoring problem of absolutely closing time of taphole into a two-stage classification problem of relative tapping state and realizes high-precision monitoring of the

CRediT authorship contribution statement

Zhaohui Jiang: Visualization, Supervision. Jinzong Dong: Writing – original draft, Data curation, Software. Dong Pan: Conceptualization, Methodology, Validation. Tianyu Wang: Investigation. Weihua Gui: Supervision.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgments

This work was supported in part by the National Major Scientific Research Equipment of China under Grant No. 61927803, the Changsha Natural Science Foundation project under Grant No. kq2202075 and in part by the Foundation for Innovative Research Groups of the National Natural Science Foundation of China Grant No. 61621062.

References (34)

  • DosovitskiyA. et al.

    An image is worth 16x16 words: Transformers for image recognition at scale

    (2020)
  • GattaC. et al.

    Ace: An automatic color equalization algorithm

  • GeerdesM. et al.

    Modern Blast Furnace Ironmaking

    (2020)
  • GomesF.S. et al.

    Multistep forecasting models of the liquid level in a blast furnace hearth

    IEEE Trans. Autom. Sci. Eng.

    (2016)
  • GomesF.S. et al.

    A new prediction model for liquid level in blast furnaces based on time series analysis

  • He, K., Zhang, X., Ren, S., Sun, J., 2016. Deep residual learning for image recognition. In: Proc. IEEE Conf. Comput....
  • Howard, A., Sandler, M., Chu, G., Chen, L.-C., Chen, B., Tan, M., Wang, W., Zhu, Y., Pang, R., Vasudevan, V., et al.,...
  • View full text