JP7264132B2 - Blast Furnace Status Determining Device, Blast Furnace Operating Method, and Hot Metal Manufacturing Method - Google Patents

Description

The present disclosure relates to a blast furnace state determination device, a method of operating a blast furnace, and a method of manufacturing hot metal.

In recent years, blast furnaces may be operated at low coke ratio (CR) and high pulverized coal injection (PCI) flow rates. Under such an operating environment, it is important to quickly and accurately grasp the furnace conditions of the blast furnace. Since the blast furnace is cylindrical, if there is unevenness in the insertion of raw materials in the circumferential direction or in the reaction in the circumferential direction, the tapping condition may vary in the circumferential direction, and operating conditions may deteriorate. . Therefore, it is important to detect deviation in the circumferential direction early and accurately.

One method of detecting the deviation in the circumferential direction is to measure the temperature distribution directly above the raw material charging surface. It is known that when at least one of raw material charging bias and reaction bias occurs, it appears as this temperature distribution bias. Conventionally, the temperature distribution directly above the raw material charging surface was often measured with a thermocouple inserted into the furnace. The thermocouples are usually inserted from about four directions so as not to interfere with the raw material insertion, and measure the temperature at a plurality of points in each radial direction. However, since it is limited to about four directions, it was difficult to grasp the entire temperature distribution. Also, durable thermocouples often have slow response times. Therefore, it was difficult to detect the temperature change of the gas corresponding to the charging of the raw material by the raw material chute rotating at high speed.

Here, recent technological improvements have made it possible to collect the temperature distribution immediately above the raw material charging surface at high sampling intervals. As a result, it has become possible to visualize the temperature distribution immediately above the raw material charging surface almost in real time, and to grasp abnormalities such as deviations in the temperature distribution. However, it is difficult for humans to constantly monitor the visualization information such as the temperature distribution screen. Moreover, when a human judges an abnormality, the judgment criteria become personal, and it is difficult to always make an appropriate judgment. On the other hand, Patent Literature 1 discloses a device for judging the condition of a blast furnace that makes an appropriate judgment by indexing variations in temperature.

JP 2018-165399 A

Here, when an abnormality in the state of the blast furnace is determined, it is preferable to control the operation of the blast furnace in order to resolve the abnormality. However, the content of blast furnace control for resolving the abnormality differs depending on the state of the disturbance in the temperature distribution. Therefore, there is a need for a technique that not only detects anomalies but also determines and classifies the state of disturbance in temperature distribution without human intervention.

An object of the present disclosure, which has been made to solve the above problems, is to provide a blast furnace state determination device that automatically determines and classifies the state of disturbance in the temperature distribution of the blast furnace. Another object of the present disclosure is to eliminate the disturbance of the temperature distribution of the blast furnace based on the determination and classification results of the blast furnace furnace condition determination device, and the blast furnace operation method capable of increasing the automation rate of the operation. An object of the present invention is to provide a method for producing hot metal.

A blast furnace condition determination device according to an embodiment of the present disclosure includes:
A blast furnace furnace condition determination device for determining the furnace condition of a blast furnace,
A temperature distribution measuring unit that measures the temperature distribution directly above the raw material charging surface in the blast furnace;
a storage unit that stores a learning model trained to output a plurality of classes for the temperature distribution pattern based on the measured temperature distribution data;
a determination unit that determines, using the stored learning model, which of the plurality of classes the temperature distribution measured by the temperature distribution measurement unit belongs to.

A blast furnace operating method according to an embodiment of the present disclosure includes:
The operating conditions are changed according to the furnace conditions of the blast furnace determined by the blast furnace condition determination device.

A method for producing hot metal according to an embodiment of the present disclosure includes:
Molten iron is produced using a blast furnace operated by the above-described blast furnace operating method.

Advantageous Effects of Invention According to the present disclosure, it is possible to provide a blast furnace state determination device that automatically determines and classifies the state of disturbance in the temperature distribution of a blast furnace. In addition, according to the present disclosure, a blast furnace operation method and molten iron that can eliminate the disturbance of the temperature distribution of the blast furnace and increase the automation rate of the operation based on the determination and classification results of the blast furnace state determination device. A manufacturing method can be provided.

FIG. 1 is a diagram showing a configuration example of a blast furnace state determination device according to an embodiment of the present disclosure. FIG. 2 is a diagram showing the flow of learning, determination, and display performed by the blast furnace state determination device according to the embodiment of the present disclosure. FIG. 3 is a diagram showing the flow of learning using a convolutional neural network. FIG. 4 is a diagram illustrating determination results. FIG. 5 is a diagram showing a configuration example of a blast furnace state determination device according to a modification.

An embodiment of a blast furnace state determination device 1 for determining the state of a blast furnace, a method of operating a blast furnace, and a method of manufacturing hot metal to which the present disclosure is applied will be described below.

(Equipment configuration)
As shown in FIG. 1 , a blast furnace condition determination device 1 according to the present embodiment includes a temperature distribution measurement unit 2 , a storage unit 3 , a determination unit 5 and a display unit 6 . In addition, the blast furnace condition determining apparatus 1 is configured to be able to communicate with the learning model generating apparatus 10, and inputs/outputs a learning model and learning data to/from the learning model generating apparatus 10, which will be described later. In this embodiment, the learning model generation device 10 includes a learning model generation unit 4 .

The temperature distribution measuring unit 2 measures the temperature distribution of the space immediately above the raw material charging surface, which is the uppermost surface of the layer composed of raw materials charged from the top of the blast furnace, inside the blast furnace that produces pig iron using iron ore as the raw material. to measure The temperature distribution measuring unit 2 may be various sensors or devices that rapidly (for example, within 10 seconds) measure the temperature distribution directly above the raw material charging surface in the blast furnace. In this embodiment, the temperature distribution measuring unit 2 is a temperature measuring device that utilizes the temperature dependence of the speed of sound. The temperature distribution measuring unit 2, for example, has eight transceivers installed on the circumference of the top of the blast furnace, transmits and receives sound waves toward the raw material charging surface, and measures the temperature based on the time from transmission to reception of the sound waves. Measure.

The storage unit 3 stores a learning model trained to output a plurality of classes for the temperature distribution pattern based on the measured temperature distribution data. The learning model is generated (learned) by the learning model generation unit 4 and stored in the storage unit 3 after learning has been completed. The storage unit 3 also stores temperature distribution data measured by the temperature distribution measurement unit 2 . The temperature distribution measurement unit 2 measures the temperature distribution just above the raw material charging surface in the blast furnace in real time, and the storage unit 3 accumulates the temperature distribution data. When the learning model generation unit 4 generates a learning model, the temperature distribution data accumulated in the storage unit 3 is used as learning data. Further, when the determining unit 5 determines the state of the blast furnace, the determining unit 5 acquires real-time temperature distribution data from the temperature distribution measuring unit 2 via the storage unit 3 . Further, when the determination unit 5 determines the state of the blast furnace, the learning model stored in the storage unit 3 is acquired by the determination unit 5 .

Based on the temperature distribution data stored in the storage unit 3, the learning model generation unit 4 generates a learning model that outputs a plurality of classes of temperature distribution patterns. A class is a set or group into which patterns of temperature distribution are classified. Details of the classes used in this embodiment will be described later.

The determination unit 5 uses the learning model (learned model) stored in the storage unit 3 to determine the state of the blast furnace. In this embodiment, in order to detect an abnormality in the furnace condition of the blast furnace, the determination unit 5 determines which of a plurality of classes the temperature distribution measured by the temperature distribution measurement unit 2 belongs to. In this embodiment, the plurality of classes includes at least two classes into which abnormal temperature distribution patterns are classified. In other words, when acquiring data of an abnormal temperature distribution, the determination unit 5 does not simply determine that it is abnormal, but determines whether it is classified into one of a plurality of classes according to the pattern. Here, the abnormal temperature distribution pattern in the present embodiment includes at least a temperature distribution pattern that can be caused by biased charging of raw materials and biased reaction in the circumferential direction of the blast furnace.

The display unit 6 presents the determination result of the determination unit 5 to the operator as an image. The display unit 6 may further have a function of outputting sound to issue an alarm to the operator along with the image. Operators are, for example, workers in a blast furnace run. The blast furnace state determined by the blast furnace state determination device 1 can be used to change the operating conditions in the blast furnace operating method. The operator, who has learned of the abnormal furnace conditions from the image displayed on the display unit 6, changes the operating conditions of the blast furnace so as to suppress the bias in raw material charging and reaction in the circumferential direction of the blast furnace. Altering the operating conditions of the blast furnace may include, for example, altering the coke ratio. Altering the operating conditions of the blast furnace may include, for example, altering the blast flow rate. In addition, for example, when a plurality of burden control patterns are set in advance, such as how to charge the raw material in the circumferential direction of the blast furnace, changing the operating conditions of the blast furnace is equivalent to changing the burden control pattern. may contain. Also, the operation of the blast furnace described above can be carried out as part of a production method for producing molten iron. In a blast furnace, raw iron ore is melted and reduced to become pig iron, which is tapped as molten pig iron. Impurities such as sulfur and phosphorus are removed from hot metal tapped from a blast furnace in a hot metal pretreatment process. Further scouring takes place in a converter to remove carbon.

The storage unit 3 , the determination unit 5 and the display unit 6 of the blast furnace condition determination device 1 may be realized by a computer that acquires temperature distribution data measured by the temperature distribution measurement unit 2 . A computer includes, for example, a memory (storage device), a CPU (processing device), a hard disk drive (HDD), and a display control unit that controls a display device such as a display. An operating system (OS) and application programs for performing various processes can be stored in a hard disk drive, and read from the hard disk drive into memory when executed by the CPU. If necessary, the CPU controls the display control section to display a required image on the display. In addition, data in the process of being processed is stored in the memory, and if necessary, is stored in the HDD. Various functions are realized by organically cooperating hardware such as a CPU and memory with an OS and necessary application programs. The storage unit 3 may be realized by, for example, a memory and a hard disk drive. The determination unit 5 may be realized by, for example, a CPU. Also, the display unit 6 may be realized by, for example, a display control unit and a display.

The learning model generation unit 4 of the learning model generation device 10 may be realized by a computer separate from the blast furnace state determination device 1 . The learning model generator 4 may be realized by, for example, a CPU. In this embodiment, the trained model is stored in the storage unit 3 as described above, but as another example, it may be stored in the memory or hard disk drive of the learning model generation device 10 . In this case, the judging section 5 may access the memory or hard disk drive of the learning model generating device 10 to read out the learned model when judging the state of the blast furnace. Also, in the present embodiment, the learning data is stored in the storage unit 3 as described above, but as another example, it may be stored in the memory or hard disk drive of the learning model generation device 10 . In this case, the learning model generation unit 4 may access the memory or hard disk drive of the learning model generation device 10 to read the learning data.

Here, the configuration of the blast furnace state determination device 1 in FIG. 1 is an example, and some of the components may not be included. Moreover, the blast furnace furnace state determination device 1 may include other components. For example, the blast furnace condition determination device 1 may have a configuration in which the display unit 6 is omitted. At this time, the blast furnace state determination device 1 may include a communication unit that outputs the determination result of the determination unit 5 to the operator's terminal device via the network.

(Method for judging blast furnace conditions)
FIG. 2 is a diagram showing a rough flow of a blast furnace state determination method executed by the blast furnace state determination device 1. As shown in FIG. The blast furnace condition determination method is divided into learning, determination, and display processes. In FIG. 2, online indicates that the treatment is performed as part of the blast furnace operation. Conversely, off-line indicates that the treatment is carried out separately from the operation of the blast furnace.

The blast furnace condition determination device 1 off-line causes the learning model generation device 10 to generate a learning model. For example, triggered by a command from the blast furnace state determination device 1, the learning model generation unit 4 first acquires data selected from the temperature distribution data accumulated in the storage unit 3 as labeled image data. The learning model generator 4 generates a learning model using labeled image data (step S1). Details of the learning model and learning will be described later. As another example, the learning model generating device 10 may start generating a learning model without waiting for a command from the blast furnace condition determining device 1 . That is, as long as the learned model is stored in the storage unit 3 before the determination unit 5 determines the reactor state, the learning model generation device 10 may take the lead in executing step S1.

The determination unit 5 reads out the stored learning model from the storage unit 3 . The determination unit 5 also acquires real-time temperature distribution data from the temperature distribution measurement unit 2 via the storage unit 3 . The determination unit 5 inputs real-time temperature distribution image data to the learning model to determine the furnace condition of the blast furnace online (step S2). In the present embodiment, the determining unit 5 determines whether or not the real-time temperature distribution is classified into an abnormal temperature distribution pattern class as the determination of the furnace condition.

After that, the display unit 6 displays an image of the determination result of the determination unit 5 (step S3). The image is not limited to a specific format as long as the operator can grasp whether the image is classified into the abnormal temperature distribution pattern class. An operator who has learned of an abnormality in the furnace situation from the image may change the operating conditions of the blast furnace.

(learning model)
It is known that when raw materials are normally charged, that is, when there is no deviation in raw material charging, the temperature distribution immediately above the raw material charging surface is a concentric distribution. It is also known that even when there is no reaction bias, the distribution is concentric. Conversely, if there is an imbalance in the raw materials or the unbalanced feeding of the raw materials due to variations in the material descending speed, the temperature distribution will be disturbed and will not be concentric. Therefore, the disturbance of the temperature distribution is imaged and classified, and the blast furnace furnace condition determination device 1 determines which disturbance corresponds to the image of the real-time temperature distribution, so that the furnace condition does not require human intervention. Automatic judgment is possible. In this embodiment, the learning model used for determination is generated by learning data having a disturbed temperature distribution selected from the temperature distribution data stored in the storage unit 3 as learning data.

The temperature distribution data selected as learning data includes data having an abnormal temperature distribution pattern as described below when imaged. It is known that the abnormal temperature distribution patterns described below can cause anomalies in blast furnace operation.

The first anomalous temperature distribution pattern has the hot spot off-center on the feed surface. Since the blast furnace is cylindrical, the shape of the raw material charging surface is a circle, but the hot part is a pattern that does not include the center. Aeration is important for blast furnaces in order to stabilize their operations. Therefore, the raw materials are distributed so that the gas can easily pass through the center of the blast furnace. In the normal case, the temperature distribution immediately above the raw material charging surface has a high temperature at the center. A first abnormal temperature distribution pattern can be caused by a distribution bias in raw material charging. For example, it is possible to normalize the operation of the blast furnace by modifying the distribution of the raw material charge. Hereinafter, the class to which the temperature distribution pattern classified as the first abnormal temperature distribution pattern belongs is referred to as the first class.

In the second abnormal temperature distribution pattern, the high temperature portion extends from the center to the periphery of the raw material charging surface. The high-temperature portion includes the center of the raw material charging surface, but the high-temperature portion continues to the periphery (circumferential portion) of the raw material charging surface, and the temperature distribution is not concentric. As described above, since the temperature in the center is normally high, the temperature in the middle (between the center and the periphery) is relatively low. The second abnormal temperature distribution pattern is believed to be caused by at least one of raw material charging bias and reaction bias. Hereinafter, the class to which the temperature distribution pattern classified as the second abnormal temperature distribution pattern belongs is referred to as the second class.

The temperature distribution data selected as training data is "labeled" according to the above. For example, temperature distribution image data classified into the first class may be labeled with "1", and temperature distribution image data classified into the second class may be labeled with "2".

(study)
FIG. 3 is a diagram showing the flow of learning performed by the learning model generation unit 4. As shown in FIG. The learning model generation unit 4 acquires the learning data, creates input data, and causes learning to generate a learning model. In the present embodiment, the learning model generation unit 4 learns using a convolutional neural network (hereinafter referred to as CNN), which is one method of deep learning. CNN is a method specialized for image processing that enables image recognition. Here, the learning method of the learning model generating unit 4 is not limited to CNN, and may be another method capable of identifying pattern images.

Here, the learning model generation unit 4 multi-values the temperature distribution selected as the learning data using one or more threshold values and then uses it as the image data. In this embodiment, the learning model generator 4 uses binarized image data. In the example of the image shown as the learning data in FIG. 3, the portion with the temperature equal to or higher than the threshold corresponds to the high temperature portion and is displayed in black. The threshold is 150° C. as an example, but is not limited to this. The threshold may be set to a temperature between the temperature of the center and the middle part of the raw material charging surface of a blast furnace in normal operation.

The learning model generator 4 divides the above image data into appropriate sizes (pixel sizes), averages each divided data, and uses the data as input data for the CNN. The learning model generation unit 4 may increase the size of the partitioned images within a range in which the temperature distribution pattern can be grasped. As an example, the learning model generation unit 4 may divide the image data into 10 pieces in each of the vertical and horizontal directions. Conversely, when the performance of the CPU that realizes the learning model generation unit 4 is sufficiently high and the learning by CNN can be executed within a realistic time, the learning model generation unit 4 generates image data may be used as input data in units of pixels without dividing.

The above input data corresponds to the input layer of the CNN shown in FIG. In addition, the feature amount corresponds to the intermediate layer. Also, the output layer corresponds to the determination of which of the plurality of classes the temperature distribution pattern belongs to. The learning model generation unit 4 generates a learning model by such a method. The learning model generation unit 4 may store the generated learning model in the storage unit 3 . Here, the image data used as learning data is the temperature distribution of the raw material charging surface of a blast furnace having a centrosymmetric structure. Therefore, data obtained by rotating the image data around the center of the raw material charging surface can also be used for learning. In other words, the learning model generating unit 4 can increase the number of input data by using the original image data and the image data obtained by rotating the original image data, and perform efficient learning.

(Determination of Abnormality)
The determination unit 5 uses the learning model generated by the learning model generation unit 4 to determine the state of the blast furnace. When judging the state of the blast furnace, the judging section 5 may read out the learning model from the storage section 3 . The determination unit 5 also acquires real-time temperature distribution data from the temperature distribution measurement unit 2 via the storage unit 3 . Similar to the learning model generation unit 4, the determination unit 5 converts the obtained temperature distribution into multiple values using one or more threshold values, and then uses it as image data. Further, the determination unit 5 divides the image data into appropriate sizes in the same manner as the learning model generation unit 4, averages each divided data, and inputs the data to the learning model. A judgment result is obtained as an output of the learning model. In this embodiment, the output of the learning model is a plurality of classes of abnormal temperature distribution patterns ("1" corresponding to the first class and "2" corresponding to the second class). For example, if the real-time temperature distribution is a pattern in which the hot part is off-center, the learning model will output a "1". Further, for example, when the real-time temperature distribution is a pattern in which the high-temperature portion is connected from the center to the periphery of the raw material charging surface, the learning model outputs "2".

The display unit 6 presents the determination result of the determination unit 5 to the operator as an image. The display unit 6 may display, for example, an image of the temperature distribution and the corresponding abnormal temperature distribution pattern. For example, when the learning model outputs "1", the display unit 6 may display "first class" together with the temperature distribution image. Here, the display unit 6 may display "third class" together with the image of the temperature distribution for patterns that are determined not to correspond to the first class or the second class.

The operator may operate the blast furnace based on the determination result. The operator performs an operation to change the charge control pattern, for example, in the case of the first class (pattern in which the hot part is off-center), when a different charge control pattern can be used. For example, in cases other than the first class, the operator may change the operating conditions so as to stabilize future operations. Changes in operating conditions that will stabilize future operations include, for example, increasing the coke ratio, decreasing the PCI ratio, and decreasing the blast flow rate.

The present disclosure will be specifically described below using examples. The temperature distribution measuring unit 2 is a temperature measuring device that utilizes the temperature dependence of the speed of sound described above, and is provided in a blast furnace that produces pig iron. The sampling period for measuring the temperature distribution is 10 sec.

In this embodiment, the first abnormal temperature distribution pattern (the high temperature part does not include the center) and the second abnormal temperature distribution pattern (the high temperature part includes the center of the raw material charging surface, The high-temperature part is connected to the periphery of the raw material charging surface, and the temperature distribution is not concentric).

First, CNN was used to learn these abnormal temperature distribution patterns. About 50 samples corresponding to these abnormal temperature distribution patterns were selected by humans from the temperature distribution data accumulated in the storage unit 3 as learning data. Learning was performed after increasing the number of samples by rotating the image data of the selected temperature distribution data by 30 degrees. The combined number of original samples and samples obtained by spinning is about 7000. Also, in this embodiment, in order to make the abnormal pattern easier to understand, the temperature is binarized with a threshold value. The binarization threshold was set to 150°C. However, the binarization operation is not an essential process of the present disclosure. That is, binarization may be omitted.

For the learning model obtained in the above learning (CNN), the judgment accuracy was verified using data different from the learning data. As a result of the verification, a determination accuracy rate of 90.3% was obtained. FIG. 4 exemplifies the judgment result in this verification.

It was also confirmed that it is possible to distinguish (classify into a third class) a temperature distribution pattern that does not fit into either the first class or the second class.

As described above, the blast furnace state determination device 1 according to the present disclosure can automatically determine and classify the state of disturbance in the temperature distribution of the blast furnace. In addition, in the blast furnace operation method and the hot metal production method, it is possible to eliminate the disturbance of the temperature distribution of the blast furnace based on the determination and classification results of the blast furnace state determination device 1 and increase the automation rate of the operation. .

Although the present disclosure has been described with reference to drawings and examples, it should be noted that various variations and modifications will be readily apparent to those skilled in the art based on the present disclosure. Therefore, it should be noted that these variations and modifications are included within the scope of this disclosure. For example, functions included in each means and each step can be rearranged so as not to be logically inconsistent, and multiple means and steps can be combined into one or divided. .

In the learning of the above embodiment, image data in which the high temperature portion does not include the center was selected as the first abnormal temperature distribution pattern. That is, image data was selected based on whether the high temperature portion included the center point of the raw material charging surface. Here, the image data may be selected based on whether or not the center portion having a certain size overlaps the high temperature portion instead of the center point.

Further, as shown in FIG. 5 , the blast furnace furnace state determination device 1 may further include a learning model generation unit 4 . That is, the blast furnace condition determination device 1 may include a temperature distribution measurement unit 2 , a storage unit 3 , a learning model generation unit 4 , a determination unit 5 and a display unit 6 .

Reference Signs List 1 blast furnace condition determination device 2 temperature distribution measurement unit 3 storage unit 4 learning model generation unit 5 determination unit 6 display unit 10 learning model generation device

Claims (7)

A blast furnace furnace condition determination device for determining the furnace condition of a blast furnace,
A temperature distribution measuring unit that measures the temperature distribution directly above the raw material charging surface in the blast furnace;
a storage unit that stores a learning model trained to output a plurality of classes for the temperature distribution pattern based on the measured temperature distribution data;
a determination unit that uses the stored learning model to determine which of the plurality of classes the temperature distribution measured by the temperature distribution measurement unit belongs to,
The learning model outputs a plurality of classes of abnormal temperature distribution patterns . A blast furnace furnace condition determination device for determining the furnace condition of a blast furnace,
A temperature distribution measuring unit that measures the temperature distribution directly above the raw material charging surface in the blast furnace;
a storage unit that stores a learning model trained to output a plurality of classes for the temperature distribution pattern based on the measured temperature distribution data;
a determination unit that uses the stored learning model to determine which of the plurality of classes the temperature distribution measured by the temperature distribution measurement unit belongs to,
The plurality of classes are a class of the abnormal temperature distribution pattern in which the high temperature portion is off the center of the raw material charging surface, and an abnormal temperature distribution in which the high temperature portion extends from the center to the periphery of the raw material charging surface. and a class of patterns of blast furnace conditions. The storage unit stores data of the measured temperature distribution,
3. The blast furnace state determination apparatus according to claim 1 , further comprising a learning model generation unit that generates a learning model outputting a plurality of classes of the temperature distribution pattern based on the stored data. 4. The apparatus for judging the state of a blast furnace according to claim 3 , wherein said learning model generation unit generates said learning model using a convolutional neural network. 5. The apparatus for determining the state of blast furnace conditions according to claim 3, wherein said learning model generation unit and said determination unit multi-value said temperature distribution using one or more threshold values. A method of operating a blast furnace, comprising changing operating conditions according to the state of the blast furnace determined by the device for determining the state of blast furnace conditions according to any one of claims 1 to 5 . A method for producing molten iron using a blast furnace operated by the method for operating a blast furnace according to claim 6 .

Images