EP3730630A1

EP3730630A1 - Furnace condition control apparatus and method

Info

Publication number: EP3730630A1
Application number: EP18891914.6A
Authority: EP
Inventors: Kyung-Lyong HAN; Jin-Hwi LEE; Sang-Han Son; Gi-Wan SON
Original assignee: Posco Co Ltd
Current assignee: Posco Holdings Inc
Priority date: 2017-12-19
Filing date: 2018-12-18
Publication date: 2020-10-28
Anticipated expiration: 2038-12-18
Also published as: KR20190074132A; EP3730630B1; KR102075210B1; EP3730630A4; CN111492070A; JP7050934B2; JP2021507115A; WO2019124931A1

Abstract

The present invention relates to a furnace condition control apparatus and method for guiding preemptive actions to stably maintain the condition of a blast furnace, by using various operational data and sensor data generated from the blast furnace. The furnace condition control apparatus according to an embodiment of the present invention may comprise: a first sensor part for imaging at least one of temperature data and pressure data of a blast furnace according to measurement positions; a second sensor part for detecting unstructured data of the blast furnace; and an action guidance part having an artificial intelligence algorithm for outputting an action guidance for operating the blast furnace, on the basis of the imaged temperature data or pressure data from the first sensor part and the unstructured data from the second sensor part. The furnace condition control method according to an embodiment of the present invention may comprise the steps of: collecting, by a data preprocessing part, unstructured data of at least one of the feed material condition, the tuyere condition, and the taphole condition of a blast furnace, and imaging temperature data and pressure data of the blast furnace according to measurement positions; receiving the data preprocessed by the data preprocessing part and outputting an action guidance for operating the blast furnace, by an artificial intelligence algorithm; determining whether the artificial intelligence algorithm requires relearning, according to whether an operator employs the action guidance; and determining whether to replace the artificial intelligence algorithm, according to whether the corresponding artificial intelligence algorithm achieves the relearning.

Description

[Technical Field]

The present disclosure relates to furnace condition control apparatus and method for controlling conditions of a blast furnace.

[Background Art]

A blast furnace process is a typical process, which is mainly dependent on the experience and intuition of operators to manually perform operations, among iron making processes.
A blast furnace is a facility for charging iron ore and coke in an upper portion of the blast furnace and blowing hot air through a tuyere to produce molten iron through a taphole using internal oxidation and reduction reactions . Due to a high temperature and high pressure in the blast furnace, a measurement cannot be performed through a sensor. Therefore, a situation of the blast furnace is indirectly predicted through a thermometer, a pressure gage, and the like, mounted on an external wall of the blast furnace, and operators perform operations based on the prediction of the situation of the blast furnace.
There are several indices indicating current states, for example, conditions of a blast furnace. Among the indices, three representative indices are furnace heat, air permeability, and circumferential balance. The furnace heat is an index obtained by manually measuring a temperature of molten iron, coming out through a taphole, to predict an internal temperature of the blast furnace. The air permeability is an index to indirectly infer a state of hot air, flowing from a lower portion to an upper portion in the blast furnace, with an air permeability index, or the like, through measurement of a pressure gage on an external wall. The circumferential balance is an index on a state in which there is no significant difference in pressure and temperature in a circumferential direction of a circular blast furnace, for example, balance is maintained.
Operators take actions to maintain the above-described three indices at desired values. Representative examples of the actions are control of a pulverized coal injection (PCI) rate, control of hot air volume, control of the amount of oxygen contained in hot air, control of a ratio of charged iron ore and coke, control of distribution of coke having a large grain size entering a central portion, and the like.
In the current blast furnace operations, operators basically determine the state of a blast furnace with their experience and intuition, operation standards, and the like, using information able to be obtained from structured data such as a measured value of a thermometer, a pressure gage, or the like, and unstructured data such as CCTV, and take operation actions based on the determination.
However, for more stable control of conditions of the blast furnace, it is important to predict a future condition of the blast furnace through a current condition and a current action and to perform an operation based on the prediction.
Such a prior art will be easily understood with reference to Korean Patent Laid-Open Publication No. 10-1995-0014631 .

[Disclosure]

[Technical Problem]

An aspect of the present disclosure is to provide a furnace control apparatus and method for guiding proactive actions to stably maintain a furnace condition using various operations, occurring in a blast furnace, and sensor data.

[Technical Solution]

According to an aspect of the present disclosure, a furnace condition control apparatus includes a first sensor unit configured to image at least one of temperature data and pressure data of a blast furnace depending on a measured location, a second sensor unit configured to detect unstructured data of the blast furnace, and an action guidance unit having an artificial intelligence algorithm outputting action guidance regarding a blast furnace operation based on imaged temperature or pressure data from the first sensor unit and unstructured data from the second sensor unit.
According to another aspect of the present disclosure, a furnace condition control method includes collecting, by a data preprocessing unit, at least one of a charging material state, a tuyere state, and a taphole state of a blast furnace as unstructured data and imaging temperature data and pressure data of the blast furnace depending on a measured location, receiving, by an artificial intelligence algorithm, preprocessed data to output action guidance regarding a blast furnace operation, determining relearning of the artificial intelligence algorithm depending on whether an operator employs the action guidance, and determining replacement of the artificial intelligence algorithm depending on whether or not to perform relearning of a corresponding artificial intelligence algorithm.

[Advantageous Effects]

As set forth above, according to an example embodiment of the present disclosure, stable production in a blast furnace may be achieved, efficiency of the blast furnace may be improved, a condition of the blast furnace may be controlled to maintain constant performance, and operations may be automated and standardized.

[Description of Drawings]

FIG. 1 is a schematic block diagram of a furnace condition control apparatus according an example embodiment of the present disclosure.
FIG. 2 illustrates a concept of artificial intelligence (AI) applied to a furnace condition control apparatus according an example embodiment of the present disclosure.
FIG. 3 illustrates a schematic operation flow of a furnace condition control method according an example embodiment of the present disclosure.
FIG. 4 illustrates an example of a GUI of a furnace condition control apparatus according to an example embodiment of the present disclosure.
FIG. 5 illustrates imaged data of a thermometer and a pressure gage applied to a furnace condition control apparatus according an example embodiment of the present disclosure.

[Best Mode for Invention]

Hereinafter, example embodiments of the present disclosure will be described in detail with reference to the accompanying drawings so that the present disclosure can be easily realized by those skilled in the art. Further, when it is determined that the detailed description of the related functions and constructions would obscure the gist of the present disclosure, the description thereof will be omitted. In addition, like reference numerals refer to elements having like functions and operations throughout the drawings.
FIG. 1 is a schematic block diagram of a furnace condition control apparatus according an example embodiment of the present disclosure.
Referring to FIG. 1, a furnace condition control apparatus 100 according to an example embodiment may include a first sensor unit 110, a second sensor unit 120, and an action guidance unit 130.
The first sensor unit 110 may image at least one of temperature data and pressure data of a blast furnace depending on a measured position.
The first sensor unit 110 may include a temperature sensor unit 111, a pressure sensor unit 112, and a data processing unit 113.
The temperature sensor unit 111 may include a plurality of temperature sensors mounted in the blast furnace. The plurality of temperature sensors may detect temperatures of the blast furnace in mounted locations thereof, respectively.
The pressure sensor unit 111 may include a plurality of pressure sensors mounted in the blast furnace. The plurality of temperature sensors may detect pressures of the blast furnace in mounted locations thereof, respectively.
The data processing unit 113 may map data on the temperature, detected by each of the plurality of temperature sensors of the temperature sensor unit 111, to the detected location and may image the mapped data. Similarly, the data processing unit 113 may map data on the pressure, detected by each of the plurality of pressure sensors of the pressure sensor unit 112, to the detected location and may image the mapped data. In addition, the data processing unit 113 may map data on the temperature and data on the pressure, detected by each of the plurality of temperature sensors and each of the plurality of pressure sensors of the temperature sensor unit 111, to the detected locations and may image the mapped data.
Due to characteristics of the blast furnace, there may be an interrelationship between location-dependent temperature and pressure. Accordingly, when the location-dependent temperature and pressure are imaged to generate information on even the interrelationship to be used as input data of a deep learning algorithm, it is advantageous for analysis of a state of the blast furnace, which may be a main factor to improving performance for action guidance.
The data processing unit 113 may map data on a detected temperature or pressure to a detected location and may two-dimensionally image the mapped data.
FIG. 5 illustrates imaged data of a thermometer and a pressure gage applied to a furnace condition control apparatus according an example embodiment of the present disclosure.
Referring to FIG. 5 together with FIG. 1, an example of imaging sensor data of the blast furnace, that is, detected data of the temperature sensor unit 111 and the pressure sensor unit 112 may be illustrated.
In FIG. 5, a left image shows a heatmap drawn under an assumption that a plurality of temperature sensors are distributed on a surface of a cylindrical blast furnace and then cut and expanded at a zero degree. For example, a horizontal direction of the drawing is an angle at which the temperature sensors are distributed. In addition, a height-dependent distribution of the temperature sensors corresponded to a height in the drawing. As a result, each black dot expresses a temperature sensor. As can be seen from left and center images, temperatures values of the blast furnace change every moment while having an organic interrelationship.
In the case of a pressure sensor illustrated in a right image in FIG. 5, four directional values are expressed. A directional pressure gage may be divided into four colored lines. A horizontal axis denotes a pressure value, and a vertical axis denotes a height location of the pressure sensor. In the present disclosure, an imaging technology illustrated in the drawing is used to efficiently input such a required location information relationship to artificial intelligence.
The second sensor unit 120 may measure at least one of a state of a charging material, a state of a tuyere, and a state of a taphole of the blast furnace to detect unstructured data.
The present disclosure may propose optimal action guidance for determining a furnace state based on current state data of the blast furnace through a deep learning-based algorithm and maintaining a normal furnace condition. Since the deep learning-based algorithm is a data-driven algorithm, a large amount of data, capable of representing a condition well, is necessary.
Therefore, operators structure data, which is not used in a control operation using a computer because it is not structured while being contents used as a basis for determining operations of a blast furnace with naked eyes, and apply the structured data to the present disclosure.
First data is data generated by measuring grain sizes of charged iron ore and coke. The first data is related to air permeability.
Second is data used as numerical data on conditions of a combustion zone of a tuyere. The combustion zone of the tuyere is the only facility allowed to observe the inside of the blast furnace and blowing hot air. In this case, pulverized coal is blown together, and the combustion zone serves to monitor a combustion state of the pulverized coal or a fuel and a raw material falling from an internal wall of the blast furnace without melting.
Third is a measuring device for measuring a state of a taphole, and measurement of a temperature of molten iron is an especially important factor. In the case of basic operations of the blast furnace, a temperature of molten iron tapped from the blast furnace is manually measured once every one or two hours. Since a measurement location is also spaced apart from a taphole by a predetermined distance and the degree of measurement taken by a person is also not constant, disturbance is considerably included in a measured value. This value is important data related to furnace heat.
To this end, the second sensor unit 120 may include a charging material state measuring device 121, a tuyere state measuring device 122, and a taphole state measuring device 123.
The charging material state measuring device 121 is disposed on a conveyor belt, along which a fuel and a raw material charged in the blast furnace pass, to measure at least one of a grain size of a charging material, a grain size distribution, and a humidity state of the blast furnace and to convert measured unstructured data into structured data and transfer the structured data to the action guidance unit 130.
The tuyere state measuring device 122 may measure at least one of a pulverized coal injection state and a raw ore falling state of the blast furnace through a plurality of tuyere cameras, and may convert measured unstructured data into structured data and transfer the structured data to the action guidance unit 130.
The taphole state measuring device 123 may measure a temperature of the molten iron tapped from the blast furnace in real time, and may measure the amount of the tapped molten iron with an angle, a thickness, or the like, of a branch of the molten iron. Then, the taphole state measuring device 123 may convert measured unstructured data into structures data and may transfer the structured data to the action guidance unit 130.
The action guidance unit 130 may output action guidance regarding a blast furnace operation, based on imaged temperature or pressure data from the first sensor unit 110 and unstructured data from the second sensor unit 120.
FIG. 2 illustrates a concept of artificial intelligence (AI) applied to a furnace condition control apparatus according an example embodiment of the present disclosure.
Referring to FIG. 2 together with FIG. 1, the action guidance unit 130 may include a learning unit 131, a control unit 132, and a reinforcement learning unit 133.
FIG. 3 illustrates a schematic operation flow of a furnace condition control method according an example embodiment of the present disclosure.
Referring to FIG. 3 together with FIGS. 1 and 2, the learning unit 131 may include an action guidance on-line algorithm. The action guidance on-line algorithm learns based on two-dimensionally imaged temperature data and pressure data (S10, S11) from the first sensor unit 110 and structured data of a charging material state, a tuyere state, and a taphole state of the blast furnace (S10, S12) form the second sensor unit 120, and may generate action guidance regarding the blast furnace operation (S20 and S20)
The action guidance on-line algorithm may include a deep learning-based algorithm to learn input data X to generate action guidance C .
The control unit 132 may output the action guidance C of the learning unit, and whether an operator accepts the action guidance may be feedbacked to the reinforcement learning unit 133 (S40) .
The reinforcement learning unit 133 may include an action guidance off-line algorithm including a deep learning-based algorithm. The action guidance off-line algorithm may receive action guidance, unaccepted by an operation, to reinforce algorithm learning.
The control unit 132 may determine relearning of the action guidance on-line algorithm and whether or not to replace the action guidance on-line algorithm with an action guidance off-line algorithm of the reinforcement learning unit 133.
For example, when unstructured data, generated in the blast furnace and structured due to importance for an operation, and existing structured data are collected and then input to an artificial intelligence system using deep learning, the deep learning algorithm proposes guidance for an action that an operation should take for stable furnace condition control based on a learned model. The operator determine whether or not to accept such action guidance, and the deep learning algorithm uses the determination as a feedback and utilize the feedback in an algorithm for improving performance. In addition, after a predetermined period of time or when characteristics of the input data are changed more than a predetermined reference, an artificial intelligence algorithm, appropriate to a current blast furnace condition, is maintained through relearning to optimize performance.
More specifically, data collected by summing unstructured data, structured and then input, and structured data, directly input, is preprocessed and then input to the deep learning-based action guidance algorithm. In this case, the algorithm proposes action guidance based on its own model. The operator determines whether the proposed action guidance is appropriate to a blast furnace operation, and then accepts or rejects the proposed action guidance. An operation using a first algorithm is performed through such a repeated loop.
In addition, when a parallel furnace condition control algorithm is present off-line, the operator receives a result of whether or not to accept artificial intelligence action guidance as a feedback value (S60) to perform on-line learning or reinforcement learning. For example, the deep learning-based guidance off-line algorithm compensates for an action guidance value feedbacked and input depending on whether a previous operator accepts action guidance (S50) to be used in algorithm reinforcement. A reinforcement learning part is basically present in a deep learning-based action guidance off-line algorithm. In the case in which the deep learning-based action guidance off-line algorithm make an erroneous determination, the reinforcement learning part reflects and uses the erroneous determination to improve algorithm performance. In addition, when a compensation value is reduced to a predetermined level or less or a difference characteristics of data and learned characteristics is increased by a predetermined level or more, a determination is made as to whether relearning is required. When the relearning is required, it is performed (S70).
When algorithm replacement is required as a result of the relearning (S80), the deep learning-based action guidance on-line algorithm is replaced with a newly learned action guidance off-line algorithm. Thus, an algorithm corresponding to a condition of the blast furnace may be maintained and a furnace condition control apparatus, having performance improved as an operation is performed, may be implemented.
FIG. 4 illustrates an example of a graphic user interface (GUI) of a furnace condition control apparatus according to an example embodiment of the present disclosure.
Referring to FIG. 4 together with FIG. 1, the action guidance unit 130 may propose an action regarding a blast furnace operation such as air volume, oxygen, pulverized coal, charging fuel/raw material cost, a center coke distribution, and the like. For example, an action guidance value required to control the air volume may be confirmed through the illustrated GUI, and a trend of related data may be confirmed. In addition, an operation may be manually performed, as necessary.
As described above, according to the present disclosure, an action of an operator required to maintain a stable furnace condition may be guided to achieve stable production of a blast furnace. Thus, efficiency of the blast furnace may be improved. In addition, a furnace condition control system, maintaining constant performance using a method of maintaining an algorithm able to response to operating conditions and blast furnace conditions varying depending on time, may be implemented. Furthermore, operations may be automated and standardized to reduce the load of the operator and to change tacit knowledge, such as know-how, experience, and the like, of the operator, into spredable and shareable explicit knowledge.
According to an example embodiment, stable production of a blast furnace may be achieved, efficiency of the blast furnace may be improved, furnace conditions may be controlled to maintain constant performance, and operations may be automated and standardized.
While the example embodiments have been illustrated and described above, it will be apparent to those skilled in the art that modifications and variations could be made without departing from the scope of the present invention as defined by the appended claims.

Claims

A furnace condition control apparatus comprising:
a first sensor unit configured to image at least one of temperature data and pressure data of a blast furnace depending on a measured location;

a second sensor unit configured to detect unstructured data of the blast furnace; and]

an action guidance unit having an artificial intelligence algorithm outputting action guidance regarding a blast furnace operation based on imaged temperature or pressure data from the first sensor unit and unstructured data from the second sensor unit.
The furnace condition control apparatus of claim 1, wherein the first sensor unit comprises:
a temperature sensor unit including a plurality of temperature sensors configured to measure temperatures of respective locations of the blast furnace;

a pressure sensor unit including a plurality of pressure sensors configured to pressures of respective locations of the blast furnace; and

a data preprocessing unit configured to match a measured temperature of the temperature sensor unit and a measured pressure of the pressure sensor unit with a measured location and to image a matching result.
The furnace condition control apparatus of claim 2, wherein the data preprocessing unit matches a measured temperature and a measured pressure with a measured location to two-dimensionally image a matching result.
The furnace condition control apparatus of claim 1, wherein the second sensor unit measures at least one of a charging material state, a tuyere state, and a taphole sate of the blast furnace.
The furnace condition control apparatus of claim 4, wherein the second sensor unit comprises:
a charging material state measuring device configured to measure at least one of a grain size of a charging material, a grain size distribution, and a humidity state of the blast furnace;

a tuyere state measuring device configured to measure at least one of a pulverized coal injection state and a raw ore falling state of the blast furnace; and

a taphole state measuring device configured to measure at least one of a temperature of molten iron and an amount of tapped molten iron.
The furnace condition control apparatus of claim 1, wherein the second sensor unit converts collected unstructured data into structure data and transfers the structured data to the action guidance unit.
The furnace condition control apparatus of claim 1, wherein the action guidance unit comprises:
a learning unit configured to learn based on data collected from the first sensor unit and the second sensor unit and having an action guidance on-line algorithm generating action guidance regarding a blast furnace operation;

a reinforcement learning unit configured to reinforce algorithm learning depending on whether an operator accepts the action guidance; and

a control unit configured to output the action guidance of the learning unit and determine relearning of the action guidance on-line algorithm and whether or not to replace the action guidance on-line algorithm with an action guidance off-line algorithm of the reinforcement learning unit.
A furnace condition control method comprising:
collecting, by a data preprocessing unit, at least one of a charging material state, a tuyere state, and a taphole state of a blast furnace as unstructured data and imaging temperature data and pressure data of the blast furnace depending on a measured location;

receiving, by an artificial intelligence algorithm, preprocessed data to output action guidance regarding a blast furnace operation;

determining relearning of the artificial intelligence algorithm depending on whether an operator employs the action guidance; and

determining replacement of the artificial intelligence algorithm depending on whether or not to perform relearning of a corresponding artificial intelligence algorithm.