Method for estimating state of coke oven lid parts and method for generating state estimation model

Abstract

To provide a method for estimating a state of coke oven lid parts capable of simultaneously identifying positions of lid parts where different abnormalities are occurring and each abnormality even when the different abnormalities occur simultaneously in a plurality of lid parts of a coke oven.SOLUTION: A method for estimating a state of coke oven lid parts comprises: a monitoring image division step of dividing a region of the coke oven included in monitoring image data into a plurality of divided image data based on positions of a plurality of lid parts in the monitoring image data; and an estimation step of inputting the divided image data in the monitoring image data as input data to a state estimation model, and outputting state information indicating any one of a normal state, a moving machine present state, a gas leakage state, and an ignition state in a state of the lid part included in the divided image data to estimate the state of the lid part.SELECTED DRAWING: Figure 4

Description

[Technical Field]

[0001] The present invention relates to a method for estimating the state of a furnace lid of a coke oven, which makes it possible to quickly grasp the state of gas leakage or ignition in the furnace lid of a coke oven, and a method for generating a state estimation model. [Background technology]

[0002] Coke, which is used as a raw material for blast furnaces, is produced by heating and carbonizing coal in the carbonization chamber of a coke oven. After carbonization of the coal in the carbonization chamber, the carbonized coal (coke cake) is removed from the oven mouth with the oven lid removed.

[0003] However, if the sealing insulation at the kiln mouth is damaged by heat and the coke chamber loses its airtightness, an abnormality occurs in the gas flow in the coke chamber, and a flame occurs due to gas leakage from the kiln mouth. When a flame occurs, work such as stopping the operation of the kiln to extinguish the flame or make repairs is required, resulting in problems such as a decrease in production rate. For this reason, technology for quickly identifying abnormalities in coke oven equipment has been researched for some time.

[0004] Patent Document 1 discloses a method for extracting pixel-by-pixel brightness value information from an image obtained by a surveillance camera, identifying a surrounding light area based on the brightness value, and identifying the presence or absence of a flame. Patent Document 2 discloses a method for detecting a gas concentration using a gas sensor whose degree of discoloration changes based on the gas concentration, etc., and detecting the gas concentration from a detection current obtained by a light detection sensor that measures the degree of discoloration of the gas sensor. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Japanese Patent Application Publication No. 2019-96265 [Patent Document 2] Japanese Patent Application Publication No. 11-14546 Summary of the Invention [Problem to be solved by the invention]

[0006] However, while the method disclosed in Patent Document 1 can identify the presence or absence of a flame, it is difficult to simultaneously identify the occurrence of a flame and a gas leak, which is a precursor to the occurrence of a flame. Furthermore, when using the method disclosed in Patent Document 2, it is necessary to install gas sensors near multiple furnace lids where gas leaks occur, which creates a cost problem for the equipment. Furthermore, in aging coke ovens, gas leaks and flames often occur simultaneously in multiple locations, and a technology that can detect these simultaneously is needed.

[0007] The present invention has been made in consideration of the above circumstances, and its purpose is to provide a method for estimating the state of a coke oven furnace lid, which is capable of simultaneously determining the positions of multiple furnace lids in which different abnormalities are occurring and each of the abnormalities, even if different abnormalities are occurring simultaneously in multiple furnace lids of the coke oven. [Means for solving the problem]

[0008] [1] A method for estimating the state of a coke oven lid, comprising: a monitoring image division process for dividing a region of a coke oven included in monitoring image data into a plurality of divided image data based on the positions of a plurality of lid sections; and an estimation process for inputting the divided image data in the monitoring image data as input data into a state estimation model, and outputting state information indicating one of the states of the lid included in the divided image data, namely, a normal state, a state with a moving machine, a gas leak state, and an ignition state, thereby estimating the state of the lid. [2] A method for generating a state estimation model, which involves training a machine learning model using multiple data sets as training data, each set consisting of divided image data in which the area of ​​a coke oven contained in monitoring image data is divided based on the positions of multiple furnace lid sections, and state information contained in the divided image data that indicates the state of the furnace lid section, which is either a normal state, a state with a mobile unit present, a gas leak state, or an ignition state, to generate a state estimation model that uses the divided image data as input and the state information in the divided image data as output. [Effects of the Invention]

[0009] According to the present invention, even if different abnormalities occur simultaneously in multiple furnace lid sections of a coke oven, it is possible to simultaneously grasp the positions of the multiple furnace lid sections in which the different abnormalities occur and each of the abnormalities. [Brief explanation of the drawings]

[0010] [Figure 1] FIG. 1 is a perspective view showing an example of a coke oven. [Figure 2] FIG. 1 is a schematic configuration diagram illustrating an example of a system. [Figure 3] FIG. 10 is a schematic diagram showing an example in which monitoring image data is divided into a plurality of divided image data. [Figure 4] 10 is an image showing the results of estimating the state of the furnace roof by applying a state estimation model to monitoring image data. DETAILED DESCRIPTION OF THE INVENTION

[0011] Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. The following embodiments show preferred examples of the present invention, and the present invention is not limited to these examples.

[0012] <Coke oven and furnace cover> FIG. 1 shows the configuration of a coke oven 10 according to this embodiment. FIG. 1 is a perspective view showing an example of a coke oven 10. First, the coke oven 10 will be described using FIG. 1. The coke oven 10 has a heat storage section 12 consisting of a plurality of heat storage chambers arranged side by side, and a plurality of coke chambers 14 and combustion chambers 16 provided above the heat storage section 12. The coke chambers 14 and the combustion chambers 16 are arranged adjacent to each other alternately. A coal loading car 18 travels above the coke chambers 14 and the combustion chambers 16 in the longitudinal direction L of the coke oven 10. A plurality of coal loading holes (not shown) are formed in the upper wall of the coke oven 14 along the transverse direction S of the coke oven 10, and coal, which is the raw material for coke, is loaded into the coke oven 14 through the loading holes. Oven openings 14a are provided on both sides of the coke oven 14, and the oven openings 14a are covered and blocked by removable oven lids 15. An extruder 20 is arranged on one kiln opening 14a side of the carbonization chamber 14, and a guide vehicle, or moving machine 22, is arranged on the other kiln opening 14a side. The extruder 20 and the moving machine 22 travel along the longitudinal direction L of the furnace.

[0013] In the carbonization chamber 14, the coal is carbonized to produce a coke cake. To carbonize the coal, fuel gas is supplied from each heat storage chamber of the thermal storage unit 12 to the combustion chamber 16 and burned. The combustion heat is transferred to the adjacent carbonization chamber 14, heating the interior of the carbonization chamber 14. This increases the temperature of the carbonization chamber 14 and carbonizes the coal. When the carbonization of the coal is complete, the furnace lid 15 is removed, and the extrusion ram of the extruder 20 is inserted into the carbonization chamber 14. The coke cake obtained by the carbonization of the coal is pushed out of the carbonization chamber 14 by inserting the extrusion ram and received by the moving machine 22 on the opposite side of the extruder 20. A fire truck 24 is disposed below the moving machine 22 and is capable of traveling in front of the thermal storage unit 12 along the furnace longitudinal direction L. The fire truck 24 receives the coke cake from the moving machine 22. The fire truck 24 transports the coke cake to a predetermined location.

[0014] When coal is loaded into the coke chamber 14 and gas is generated from within the coke chamber, if the sealing insulation at the kiln mouth has deteriorated, a flame will break out from the kiln mouth. At this time, gas will first leak from the furnace cover 15. For this reason, in the coke oven 10, it is necessary to instantly grasp the occurrence of gas leakage and fire, as well as the location of the gas leakage and fire.

[0015] <System configuration> Next, the configuration of a system 30 according to this embodiment will be described with reference to Fig. 2. Fig. 2 is a schematic diagram showing an example of the system 30. The system 30 includes an imaging device 31, a transmitter 32, a receiver 33, a processing device 34, and an output device 35.

[0016] The imaging device 31 is positioned so that all of the lid portions 15 of the coke oven 10 shown in FIG. 1 are within its field of view. The imaging device 31 continuously captures images of all of the lid portions 15 of the coke oven 10 and generates monitoring image data. The transmitter 32 transmits the monitoring image data generated by the imaging device 31 to the receiver 33. The receiver 33 transfers the received monitoring image data to the processing device 34. The processing device 34 may be a monitoring personal computer that monitors the lid portions 15. The processing device 34 estimates the condition of the lid portions 15 of the coke oven 10 based on the monitoring image data. The output device 35 outputs the condition of the lid portions 15 estimated by the processing device 34. The output device 35 may be a display with a display unit, etc.

[0017] Here, the monitored image dividing step and the estimation step will be described with respect to the coke oven furnace hood state estimation method according to the present invention.

[0018] <Monitoring image division process> The monitoring image division process according to the present invention will be described with reference to FIG. 3. FIG. 3 is a schematic diagram showing an example of dividing monitoring image data into a plurality of divided image data N. FIG. 3 is based on monitoring image data generated by imaging the coke oven 10 with an imaging device 31 so that all of the oven hoods 15 are included in the field of view. The imaging device 31 may capture an image of the coke oven 10 at a distance of about 70 m from the imaging device 31. The processing device 34 then divides the area T of the coke oven 10 included in the monitoring image data into a plurality of divided image data N, as indicated by "1" to "8" in the figure, based on the positions of the plurality of oven hoods 15. Note that although the image shown in FIG. 3 is approximately the same size as the area T of the coke oven 10, it is also possible to capture an image including both the coke oven 10 and peripheral equipment and confirm the area T of the coke oven 10 included in the captured image to generate the image shown in FIG. 3. Furthermore, the division into divided image data N based on the positions of multiple furnace lid sections 15 may be performed based on the position of a single furnace lid section 15 so that only a single furnace lid section 15 is included in the divided image data N, or based on the positions of multiple furnace lid sections 15 so that multiple furnace lid sections 15 are included in a single divided image data N. Furthermore, although the multiple divided image data N shown in Figure 3 is shown in a format in which the image is divided into multiple sections in the horizontal direction, the acquired image may also be divided into multiple sections in the vertical direction, or the image may be divided into multiple sections in a matrix format in the vertical and horizontal directions.

[0019] <Estimated process> The processing device 34 inputs the divided image data N in the monitoring image data as input data into the state estimation model, and outputs state information indicating one of the states of the furnace cover section 15 contained in the divided image data N, namely, normal state, state with mobile unit present, gas leak state, or ignition state, thereby estimating the state of the furnace cover section 15.

[0020] Here, we will explain the state of the furnace lid section 15 in the divided image data N. In the furnace lid section 15 of the coke oven 10, the state of the furnace lid section 15 changes depending on the situation inside the coking chamber 14 that is being closed. In this embodiment, the state of the furnace lid section 15 is assumed to be in four states: normal state, state with moving machine, gas leak state, and ignition state. The normal state refers to a state in which there is no gas leak or flame generation from the furnace lid section 15. The state with moving machine refers to a state in which the moving machine 22 has moved (is positioned) to the furnace lid section 15 so as to cover the furnace lid section 15. The gas leak state refers to a state in which gas is leaking from the furnace lid section 15. The ignition state refers to a state in which flame is generated in the furnace lid section 15.

[0021] In this way, the processing device 34 can estimate the state of the furnace hood portion 15 in each area divided into the monitoring image, so that abnormalities in multiple furnace hood portions 15 where different abnormalities have occurred can be detected simultaneously.

[0022] The processing device 34 may estimate the state of the furnace hood section 15 in each of the divided image data N divided into a plurality of pieces in the monitoring image dividing step, and may superimpose a display showing the estimated state information at each position (for example, "1" to "8" shown in FIG. 3) in the divided monitoring image data. That is, an image is generated in which the state of the furnace hood section 15 estimated for each divided image data N is visually displayed at each position in the monitoring image data.

[0023] The processing device 34 transmits the generated image data to the output device 35, and causes the output device 35 to display the image data. By displaying the image data in this manner, it is possible to simultaneously grasp the furnace lid parts 15 that are in a "gas leak state" and the furnace lid parts 15 that are in an "ignition state" in the multiple divided image data N of the coke oven 10, as shown in Fig. 4. Furthermore, as shown in Fig. 4, it is also possible to grasp the respective positions of the furnace lid parts 15 that are in a "gas leak state" and the furnace lid parts 15 that are in an "ignition state".

[0024] When the image data is displayed on the output device 35, as shown in Fig. 4, for a furnace lid 15 that is in an "ignition state" and requires immediate attention in the coke oven 10, the word "ignition" may be displayed on a display or the like near the furnace lid 15 to proactively alert a manager or the like. Then, based on the word "ignition" displayed on the output device 35 and the position where the word is displayed, water can be quickly sprayed on the furnace lid 15 that is in an "ignition state," enabling prompt firefighting activities.

[0025] Furthermore, during machine learning as a state estimation model (machine learning model), 16,000 images of surveillance image data captured in the past were used as training data. Then, in the estimation process, 1,200 images of surveillance image data were used as test data. As a result, for each divided image data N in the surveillance image data, the estimation rate for correct estimation was 88.8% for "normal state (state with mobile device)", 83.5% for "gas leak state", and 83.4% for "fire state", with the average estimation rate for these three states being approximately 85%.

[0026] Regarding the monitoring image data, the estimation process in the processing device 34 classifies the estimated patterns of the state of the multiple furnace lid sections 15 of the coke oven 10 into the following four patterns. From the perspective of monitoring the occurrence of abnormalities in the furnace lid sections 15 of the coke oven 10, the four estimated patterns are described, including the divided image data N to which the "normal state" is assigned and the divided image data N to which the "mobile machine present state" is assigned.

[0027] The first is a "normal (mobile device)" estimated pattern. The "normal (mobile device)" estimated pattern is a pattern in which all divided image data N included in the monitoring image data are in the "normal state (mobile device present state)" state.

[0028] The second is the "normal (mobile device) + gas leak" estimation pattern. The "normal (mobile device) + gas leak" estimation pattern is a pattern that includes a "gas leak state" in addition to a "normal state (state with mobile device)" for all divided image data N included in the surveillance image data. In this case, for all divided image data N included in the surveillance image data, multiple divided image data N also include a state that is estimated to be a "gas leak state."

[0029] The third is the "normal (mobile device) + firing" estimated pattern. The "normal (mobile device) + firing" estimated pattern is a pattern that includes an "firing state" in addition to a "normal state (mobile device present state)" for all divided image data N included in the surveillance image data. In this case, all divided image data N included in the surveillance image data also include a state in which multiple divided image data N are estimated to be in an "firing state."

[0030] The fourth is a "normal (mobile device) + gas leak + ignition" estimated pattern. The "normal (mobile device) + gas leak + ignition" estimated pattern is a pattern that includes, in addition to the "normal state (mobile device present state)", a "gas leak state" and an "ignition state" for all divided image data N included in the surveillance image data. In this case, for all divided image data N included in the surveillance image data, it also includes a state in which multiple divided image data N are estimated to be a "gas leak state" and a state in which multiple divided image data N are estimated to be an "ignition state".

[0031] Next, a method for generating a state estimation model according to the present invention will be described.

[0032] The processing device 34 prepares divided image data N by dividing the area of ​​the coke oven included in the monitoring image data captured in the past based on the positions of the multiple furnace hood sections 15. It also prepares status information indicating the status of the furnace hood section 15 included in the prepared divided image data N, which is either a normal status, a status with a moving machine, a gas leak status, or an ignition status. Then, it prepares multiple datasets as training data, each pair consisting of the divided image data N and the status information indicating the status of the furnace hood section 15 included in the divided image data N, and trains a machine learning model using the training data. Then, it generates a state estimation model that uses the divided image data N as input and the status information in the divided image data N as output. It is preferable to prepare 1,000 or more datasets as training data.

[0033] The status information may be data accumulated in association with the divided image data N in the monitoring image data captured in the past. Alternatively, the status information may be information added by a worker or the like to the divided image data in the newly acquired monitoring image data through visual inspection.

[0034] In the model training process, the dataset may be generated by machine learning (deep learning) while adjusting weighting coefficients. As an introduction model of the neural network used in the machine learning, the InceptioRes NetV2 model may be used from the viewpoint of improving the accuracy of image recognition technology.

[0035] As described above, the coke oven lid state estimation method and state estimation model generation method according to the present invention make it possible to simultaneously grasp the positions of the lids 15 in which different abnormalities occur and each of the abnormalities, even when different abnormalities occur simultaneously in multiple lids 15 of a coke oven 10. Furthermore, since the state of the lids 15 is estimated instantaneously using the state estimation model, the states of the multiple lids 15 can be grasped quickly, and ultimately, measures can be taken promptly to address a lid 15 in a gas leak state or a fire state. Furthermore, since the system 30 can be configured using an imaging device 31 whose field of view includes multiple lids 15 of the coke oven 10, the monitoring system configuration can be simplified. [Explanation of symbols]

[0036] 10. Coke oven 12 Heat storage section 14 Carbonization chamber 14a Kiln mouth 15 Furnace lid part 16 Combustion chamber 18 Coaling car 20 Extruder 22 Mobile Device 24 Fire truck 30 systems 31 Imaging device 32 Transmitter 33 Receiver 34 Processing equipment 35 Output Device L furnace longitudinal direction S Furnace short direction N divided image data T Coke oven area

Claims

1. a monitoring image dividing step of dividing a region of the coke oven included in the monitoring image data into a plurality of divided image data based on the positions of a plurality of furnace hoods; an estimation process of inputting the divided image data in the monitoring image data as input data into a state estimation model, outputting state information indicating one of the states of the furnace hood section included in the divided image data, namely, a normal state, a state with a moving machine, a gas leak state, and an ignition state, and estimating the state of the furnace hood section; A method for estimating the state of a furnace lid of a coke oven, comprising:

2. a machine learning model is trained using as training data a plurality of data sets each including a set of divided image data obtained by dividing an area of ​​a coke oven included in monitoring image data based on the positions of a plurality of furnace lid parts, and state information indicating a state of the furnace lid part included in the divided image data, the state being one of a normal state, a state with a mobile device, a gas leak state, and an ignition state; A state estimation model generation method for generating a state estimation model that receives the divided image data as an input and outputs the state information in the divided image data.

Images