Flame detecting device, iron-making facility, and flame detecting method

Abstract

The present disclosure has been made in view of the problems of the related art, and an object of the present disclosure is to provide a flame detection device capable of reducing erroneous determination of flame detection.SOLUTION: The flame detecting device 1 of the present disclosure is a flame detecting device for an ironmaking-related facility, and includes the video image acquiring unit 10 that acquires a video image of a detected region, the partial video image detecting unit 32 that detects a partial video image having a predetermined feature from within a frame of the acquired video image, the motion vector detecting unit 33 that detects a motion vector of the partial video image between frames of the video image, and the flame state determining unit 36 that determines the state of a flame in the detected region by using the detected motion vector.SELECTED DRAWING: Figure 4

Description

[Technical Field]

[0001] The present invention relates to a flame detection device, iron-making related equipment, and a flame detection method. [Background technology]

[0002] A flame detection device mounted on a coal loading car for loading coal into a coke oven is known. For example, Patent Document 1 describes a coal loading car for loading coal through a coal loading port, and the coal loading car is equipped with an optical flame detection device that has the coal loading port in its field of view. The optical flame detection device monitors a flame at the coal loading port and emits an output signal when a flame that meets predetermined conditions is detected. The coal loading car is configured to notify an operator of the detection of a flame when the optical flame detection device detects a flame. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2000-273455 Summary of the Invention [Problem to be solved by the invention]

[0004] In steelmaking-related equipment with hot parts, high temperatures and flames can cause fire damage. For example, in coke ovens, the heat and flames from the coke oven can cause fire damage to moving machinery. In coke ovens, for example, improper attachment of the charging lid to the charging port can create a gap between the lid and the charging port, allowing flames to rise from the gap and burn the coal loading car. For this reason, it is desirable for steelmaking-related equipment to be able to detect flames and prevent damage to the equipment.

[0005] In the technology described in Patent Document 1, a flame detection device is installed on a coal loading car as a countermeasure against fire accidents. This flame detection device detects infrared rays emitted by a flame to determine whether or not there is a flame. Configurations that use infrared rays to detect flames can erroneously detect a flame even when there is no flame due to infrared rays emitted by sunlight or high-temperature parts such as coke oven riser pipes, resulting in a problem of lack of reliability. For these reasons, the technology described in Patent Document 1 has room for improvement in terms of reducing erroneous flame detection.

[0006] The present invention has been made in view of the above problems, and one of its objects is to provide a flame detection device that can reduce erroneous flame detection determinations. [Means for solving the problem]

[0007] In order to solve the above problems, one embodiment of the flame detection device of the present invention is a flame detection device for steelmaking-related equipment, and includes an image acquisition unit that acquires an image of the detection area, a partial image detection unit that detects a partial image having a predetermined characteristic from within a frame of the acquired image, a motion vector detection unit that detects a motion vector of the partial image between frames of the image, and a flame state determination unit that determines the state of the flame in the detection area using the detected motion vector.

[0008] Any combination of the above components, or mutual substitution of the components or expressions of the present invention between methods, systems, etc., are also valid aspects of the present invention. [Effects of the Invention]

[0009] According to the present invention, it is possible to provide a flame detection device that can reduce erroneous flame detection determinations. [Brief explanation of the drawings]

[0010] [Figure 1] 1 is a front view showing an example of a coke oven to which a flame detection device according to an embodiment is applied. [Figure 2] FIG. 2 is a plan view showing the relationship between the flame detection device of FIG. 1 and a coke oven. [Figure 3] 2 is a side view showing the relationship between the flame detection device of FIG. 1 and a detection area. FIG. [Figure 4] FIG. 2 is a block diagram illustrating the flame detection device of FIG. 1. [Figure 5] FIG. 2 is a diagram schematically illustrating motion vectors of a video image. [Figure 6] 10 is a diagram plotting an example of the relationship between the area of ​​a partial image of a video and the amount of movement. [Figure 7] FIG. 10 is a diagram showing an example of each RGB histogram of a flame image. [Figure 8] 2 is a flowchart showing an example of the operation of a coal loading car equipped with the flame detection device of FIG. 1.

[0011] The present invention will be described below based on preferred embodiments with reference to the drawings. In the embodiments and modifications, identical or equivalent components and members are designated by the same reference numerals, and redundant explanations will be omitted where appropriate. The dimensions of the members in the drawings are enlarged or reduced as appropriate for ease of understanding. Some members that are not important for explaining the embodiments will be omitted from the drawings.

[0012] Furthermore, terms including ordinal numbers such as first and second are used to describe various components, but these terms are used only to distinguish one component from another and do not limit the components.

[0013] First, the background to the development of the flame detection device 1 according to an embodiment of the present invention will be described. The flame detection device 1 can be applied mainly to steel-making related facilities. In steel-making related facilities such as blast furnaces and coke ovens, flames can occur unintentionally on their surfaces. Because the occurrence of a flame can damage the facilities, it is desirable to detect the occurrence of a flame early and take appropriate measures. One possible way to detect a flame is to use an infrared sensor. However, because steel-making related facilities have many high-temperature parts, it was found that detection methods using infrared sensors erroneously identify infrared rays emitted from the high-temperature parts as flames.

[0014] Therefore, the present inventors compared flames with infrared rays from high-temperature parts and found that flames move more rapidly than infrared rays from high-temperature parts. Flame detection device 1 was developed based on this finding, and is capable of suppressing the influence of infrared rays from high-temperature parts and reducing erroneous flame detection in steelmaking-related equipment that has many high-temperature parts. Below, as an example, flame detection device 1 applied to a coke oven, which is steelmaking-related equipment, will be described, but flame detection device 1 can be applied not only to coke ovens but also to steelmaking-related equipment in general.

[0015] [Embodiment] A flame detection device 1 according to an embodiment of the present invention will now be described with reference to the drawings. Fig. 1 is a front view showing an example of a coke oven 40 equipped with a flame detection device 1 according to an embodiment. Fig. 2 is a plan view showing the relationship between the flame detection device 1 and the coke oven 40. Fig. 3 is a side view showing the relationship between the flame detection device 1 and a detection area 6. The detection area 6 will be described later.

[0016] The following explanation will be mainly based on the XYZ Cartesian coordinate system. The X direction corresponds to the left-right direction on the paper in FIG. 1. The Y direction corresponds to the up-down direction on the paper in FIG. 1. The Z direction corresponds to the direction perpendicular to the paper in FIG. 1. The Y direction and Z direction are both orthogonal to the X direction. These directional notations do not limit the orientation of the flame detection device 1, and the flame detection device 1 can be used in any orientation depending on the application.

[0017] The coke oven 40 has multiple (100 in this example) carbonization chambers 47 arranged in the Z direction. Figure 1 shows one of the multiple carbonization chambers 47. A coal loading car 100 has an electric motor (not shown) for movement, and moves in the Z direction above the carbonization chambers 47, loading coal into each carbonization chamber 47. The Z direction is an example of the first direction. In addition to the coal loading car 100, the coke oven 40 also has a pusher, a guide car, a bucket car, and a train (all not shown) that pulls the bucket car.

[0018] The ceiling 45 of each carbonization chamber 47 is provided with multiple (five in this example) charging ports 46 spaced apart in the X direction. The charging ports 46 are openings for charging raw coal and have removable lids 48. The number of charging ports 46 is not limited to five and may be, for example, four. The coal loading car 100 has five coal loading sections 5 corresponding to the five charging ports 46. The coal loading car 100 may load coal into the multiple charging ports 46 simultaneously from multiple coal loading sections 5, but in this embodiment, coal is loaded sequentially into the multiple charging ports 46. The coal loading section 5 has a coal feeding sleeve 51, a fire protection hood 52, and a lifting magnet device 54.

[0019] The sleeve 51 is a cylindrical member that guides a coal charging hopper (not shown) that receives coal and charges it into the coking chamber 47, and is provided so as to be able to move up and down. The hood 52 is a cylindrical member that covers the charging port 46 and the sleeve 51 when charging coal into the coking chamber 47, and is provided so as to be able to move up and down independently of the sleeve 51. FIG. 1 shows the hood 52 in a lowered state. FIG. 3 shows the hood 52 in a raised state. The hood 52 has fire and dust prevention functions. The lifting magnet device 54 is provided so as to be able to move back and forth in the space within the hood 52 through an opening formed in the side of the hood 52. The lifting magnet device 54 includes an electromagnet (not shown) called a lifting magnet, and can attract the lid 48 using the magnetic force of the electromagnet, remove the lid 48 from the charging port 46, and attach the lid 48 to the charging port 46. Coal is charged into the coal charging section 5 with the lid 48 removed, and after charging is completed, the lid 48 is attached to the charging port 46.

[0020] The flame detection device 1 is provided on a coal loading car 100 that loads coal while moving on top of the ceiling 45 of the coke oven 40. In this example, the coal loading car 100 has a first floor and a second floor, and the flame detection device 1 is located under the floor of the second floor or on the ceiling of the first floor. The coal loading car body is not shown in Figures 2 and 3. The flame detection device 1 determines the state of the flame in the detection area 6 where the flame is to be detected. In this example, the detection area 6 is an area that includes the loading entrance 46, which is a location where flames are likely to occur.

[0021] The flame detection device 1 includes an image acquisition unit 10 and an information processing unit 3. The image acquisition unit 10 includes a known imaging means such as a camera, acquires an image of the detection area 6 captured by the imaging means, and transmits it to the information processing unit 3. The information processing unit 3 processes the image acquired by the image acquisition unit 10. The flame detection device 1 has five image acquisition units 10 that individually capture images of the detection areas 6 set for each of the five charging ports 46. The flame detection device 1 may be configured to acquire images of all of the detection areas 6 using a number of image acquisition units 10 that is fewer than the number of charging ports 46.

[0022] As shown in FIG. 3, the image acquisition unit 10 captures an image of the detection area 6 from diagonally above at a position away from the detection area 6 in the first direction, i.e., the Z direction. Reference numeral 13 denotes the field of view of the image acquisition unit 10. The image acquisition unit 10 captures an image of the detection area 6 set around the loading slot 46 viewed from diagonally above, and acquires an image of the detection area 6 (hereinafter referred to as "image F1"). By acquiring the image F1 of the detection area 6 viewed from diagonally above, the image acquisition unit 10 can detect the inclination of the lid 48 of the loading slot 46 relative to the ceiling 45. From the image F1, it is also possible to detect positional deviations of the lid 48 in the X, Y, and Z directions relative to the loading slot 46.

[0023] As shown in Fig. 3, the irradiation unit 12 is disposed outside the hood 52. The irradiation unit 12 irradiates the lid 48 with light from an obliquely upward direction when the image acquisition unit 10 acquires an image F1 of the detection area 6. Reference numeral 14 indicates the irradiation range of the light emitted by the irradiation unit 12. The flame detection device 1 does not necessarily require the inclusion of the irradiation unit 12, but in this example it is provided mainly to detect the inclination of the lid 48.

[0024] The information processing unit 3 will be described with reference to Figure 4. Each functional block of the information processing unit 3 shown in Figure 4 can be realized in terms of hardware by elements such as a computer processor, CPU, and memory, electronic circuits, and mechanical devices, and in terms of software by a computer program, etc., but here, functional blocks realized by cooperation of these elements are depicted. Therefore, it will be understood by those skilled in the art that these functional blocks can be realized in various ways by combining hardware and software.

[0025] The information processing unit 3 includes an input unit 30, an image adjustment unit 31, a partial image detection unit 32, a motion vector detection unit 33, an area specification unit 34, a color specification unit 35, a flame state determination unit 36, an ellipse estimation unit 37, a lid state determination unit 38, a retry counter 39, a notification unit 42, a model generation unit 43, and a memory unit 41. These functional blocks can exchange data with each other via an internal data bus 44.

[0026] The input unit 30 receives an image F1 of the detection area 6 acquired by the image acquisition unit 10. The image adjustment unit 31 adjusts the input image F1 and converts it into an adjusted image F2 (hereinafter, sometimes simply referred to as "image F2"). The image F2 is also an image of the detection area 6, and is provided to other functional blocks within the information processing unit 3. The image adjustment unit 31 in this example can perform image adjustments on the input image F1, including trimming, scale adjustment, exposure adjustment (brightness adjustment), saturation adjustment, hue adjustment, contrast adjustment, etc. The input images F1 and F2 may be moving images.

[0027] Video F2 includes partial images (hereinafter referred to as "partial images P") that are a collection of pixels that are distinguished from their surroundings by their brightness. Flame detection device 1 determines whether each partial image P of video F2 is a flame using a motion vector. Partial image detection unit 32 detects partial images P that have predetermined characteristics from within the frames of acquired video F2. In other words, partial images P are partial images that are a collection of pixels that are distinguished from their surroundings by their brightness. In this example, partial images P are blocks made up of multiple pixels.

[0028] 5 is a diagram schematically showing motion vectors of video F2. The motion vector detection unit 33 detects a motion vector V of the partial video P between multiple frames of video F2. The motion vector V may be determined by optical flow analysis. In the embodiment, the motion vector V of video F2 is determined as a vector of movement from a frame of video F2 at a certain moment to a frame of another moment.

[0029] In this example, the motion vector V is a vector on a two-dimensional plane of the partial video P, and includes information on the direction of movement on the plane (hereinafter referred to as the "movement direction") and the magnitude of the movement on the plane (hereinafter referred to as the "movement amount M"). In Figure 5, white circles arranged in a matrix pattern on the top, bottom, left, and right indicate the positions of observation points in the base frame. In this figure, the movement direction of the motion vector V of the observation point is indicated by the direction of the arrow, and the movement amount M is indicated by the length of the arrow. In other words, the longer the arrow, the greater the movement amount M of that observation point. The movement amount M of the motion vector V of the partial video P may be the movement amount M of an observation point included in that partial video P.

[0030] The detection area 6 may contain both embers and flames, and it has been difficult to distinguish between the two using flame detection methods using infrared sensors. The inventors discovered that while images of flames move frequently, images of embers show almost no movement. Based on this discovery, the inventors focused on the amount of movement M of the motion vector V of the partial image P and devised a technology that uses this amount of movement M to reduce the influence of embers and detect flames with high accuracy. In other words, if the amount of movement M of the partial image P is less than a threshold, it is determined that the image is not a flame, and if the amount of movement M is equal to or greater than the threshold, it is determined that the image is a flame. This threshold can be set through experiments or simulations based on the desired flame detection accuracy.

[0031] It is desirable to further reduce the false positives of flame detection. This was investigated from the perspective of being able to distinguish flame images from ember images with higher accuracy. As a result, it was found that the area A of a partial flame image P is often larger than the area A of a partial ember image P.

[0032] Please refer to Figure 6. Figure 6 is a diagram showing an example of the relationship between the area A and the amount of movement M of a partial image P of image F2. This diagram is a plot of the relationship between the area A and the amount of movement M of a partial image P for a large number of images F2. The area A is shown as a relative value on the horizontal axis, and the amount of movement M is shown as a relative value on the vertical axis. The white circles in this diagram represent partial images P that are known to be flames, and their area A and amount of movement M are larger than those of the other partial images P. The squares in this diagram represent partial images P that are known to be stationary embers, and their area A and amount of movement M are smaller than those of the white circle partial images P. The black circles in this diagram represent partial images P that are known to be free of flames, and their area A and amount of movement M are even smaller than those of the square partial images P.

[0033] In this embodiment, the flame detection device 1 further includes an area specification unit 34 that specifies the area A of the partial image P, and the flame state determination unit 36 ​​determines the state of the flame by referring to the area A of the partial image P. For example, a separation line S1, which can be considered a two-dimensional threshold, is set in advance, and if the area A and amount of movement M of the partial image P are smaller than the separation line S1, the partial image P is determined to not be an image of a flame, and if the area A and amount of movement M are equal to or greater than the separation line S1, the partial image P is determined to be an image of a flame. The separation line S1 may be a straight line, a curved line, or a bending line. FIG. 6 shows an example of the separation line S1.

[0034] The separation line S1 can be generated by a learning model so as to distinguish between a first group of data on the area and movement amount of a partial image P that has been determined to be an image of a flame and a second group of data on the area and movement amount of a partial image P that has been determined to be an image of embers rather than an image of a flame, based on past measurement data on the area and movement amount of the partial image P. The learning model K is an AI model generated by machine learning based on previously acquired reference information and information on the presence or absence of a flame corresponding to the reference information. The learning model K can be generated using a known machine learning method, such as a support vector machine, a neural network (including deep learning), or a random forest. The learning model K is stored in the storage unit 41.

[0035] In other words, the flame state determination unit 36 ​​determines the state of the flame using a learning model K obtained by machine learning, using as learning data the data on the motion vector V of the partial image P relating to the image of the flame and the image of the remaining remains, data on the area of ​​the partial image P, and data on the state of the flame in the detection area 6, which have all been acquired in advance.

[0036] Please refer to Figure 7. Figure 7 is a diagram showing a histogram of RGB luminance (brightness) of a partial image P that is known to be a flame. In this diagram, the horizontal axis represents luminance, with an upper limit of 255. The vertical axis represents the number of pixels as a frequency. In this specification, the frequencies of R, G, and B are referred to as the R value, G value, and B value. The example in Figure 7 shows an image engulfed in flames, and among the R value, G value, and B value, the frequency of the R value is significantly higher near a brightness of 255.

[0037] From the perspective of being able to distinguish flame images with higher accuracy, it was found that partial images P of the flame and embers have a flame color, while other partial images P do not have a flame color. Color specification unit 35 specifies whether the color of partial image P is a flame color, and provides the result of specification to flame state determination unit 36. In this example, color specification unit 35 uses the RGB histogram of partial image P to specify a divided value R obtained by dividing the R value at brightnesses of 254 and 255 by the average value of the G and B values.

[0038] As an example, the color specification unit 35 specifies that the partial image P is not a flame image when the division value R is less than a predetermined threshold, and specifies that the partial image P is a flame image when the division value R is equal to or greater than the threshold. The inventor's research suggests that flame color specification within the visible light color gamut provides higher flame specification accuracy than flame color specification including color gamuts other than visible light. The threshold value of the color specification unit 35 can be set by experiment or simulation based on the desired flame detection accuracy.

[0039] Flame state determination unit 36 ​​determines the state of the flame by referring to the color, which is the color information of partial image P. In this embodiment, flame state determination unit 36 ​​determines that partial image P is an image of a flame if the area and amount of movement of partial image P are equal to or greater than separation line S1 and the color of partial image P is flame-colored. Flame state determination unit 36 ​​determines that partial image P is not an image of a flame if the area or amount of movement of partial image P is less than separation line S1 or if the color of partial image P is not flame-colored.

[0040] The ellipse estimation unit 37 estimates an ellipse that approximates the outline of the charging port 46 and the outline of the lid 48 and calculates ellipse parameters. These ellipse parameters can be used to identify the installation posture of the lid 48. The lid state determination unit 38 uses the judgment result of the flame state determination unit 36 ​​and the estimation result of the ellipse estimation unit 37 to comprehensively determine the installation state of the lid 48 relative to the charging port 46. The retry counter 39 monitors the number of retry operations, which will be described later. The notification unit 42 notifies the outside when the number of retry operations exceeds an upper limit. The memory unit 41 stores images F1 and F2 and intermediate processing information based on these. The memory unit 41 also stores learning models F and K. The model generation unit 43 is capable of generating learning models F and K.

[0041] Next, an example of the operation of the coal loading car 100 equipped with the flame detection device 1 will be described with reference to Fig. 8. Fig. 8 is a flowchart showing process S110 of an example of the operation of the coal loading car 100 equipped with the flame detection device 1. Process S110 is a lid state determination process that determines whether the lid 48 is properly attached to the charging opening 46, and includes a flame state determination process.

[0042] First, an overview of the coal loading process will be described. As described above, the coke oven 40 has multiple (100 in this example) coking chambers 47 arranged in the Z direction. The coal loading car 100 moves in the Z direction, stops over each coking chamber 47, and loads coal into that coking chamber 47. Each coking chamber 47 has five loading ports 46 spaced apart in the X direction. As described above, the coal loading car 100 has a coal loading section 5, an image acquisition section 10, and an irradiation section 12. Each coal loading section 5 has a sleeve 51, a hood 52, and a lifting magnet device 54. The lifting magnet device 54 uses controllable magnetic attraction to remove and attach the lid 48.

[0043] When loading coal, the coal loading car 100 moves in the Z direction and stops above the target carbonization chamber 47, lowers the hood 52 to remove the lid 48 from the loading port 46, lowers the sleeve 51 to load coal through the loading port 46, and then raises the sleeve 51 and hood 52 to attach the lid 48.

[0044] Once the lid 48 is attached, the coal loading car 100 acquires an image F1 of the detection area 6 using the image acquisition unit 10. The image F1 includes an image of the lid 48 and the area around the loading entrance 46. In the information processing unit 3 of the coal loading car 100, the flame state determination unit 36 ​​uses the image F2 converted from the image F1 to determine the flame state, the ellipse estimation unit 37 estimates an ellipse related to the lid 48, and the lid state determination unit 38 determines the lid state. If the flame state is good and the lid posture is good, the information processing unit 3 determines that the lid 48 is attached in a good state.

[0045] Here, the lid 48 being in a good state of attachment means that no flame is detected and the lid posture is not defective. When the lid 48 is determined to be in a good state of attachment, the coal loading car 100 moves to the coal tower (not shown) to replenish coal, and then performs the same operation to load coal into the next coking chamber 47.

[0046] The coal loading car 100 performs a retry operation if the installation condition is determined to be poor. In the retry operation, the lid 48 is reattached and detached, and an image F1 of the detection area 6 is acquired and converted into an image F2 using the same process as above, and the lid position and flame condition are re-determined from the image F2. If the installation condition is still determined to be poor after repeating the retry operation a predetermined number of times, the coal loading car 100 notifies the outside of the abnormality.

[0047] The details of process S110 will be described. Process S110 starts in a state where the lid 48 is attached after the coal has been charged, and the area around the lid 48 and the charging opening 46 has been blown away with air. First, in step S111, the image acquisition unit 10 acquires an image F1 of the detection area 6 (step S111). The image F1 acquired in this step is transmitted to the information processing unit 3.

[0048] Next, the information processing unit 3 adjusts the input image F1 using the image adjustment unit 31 (step S112). In this step, the image adjustment unit 31 performs image adjustments on the input image F1, including trimming, scale adjustment, exposure adjustment (brightness adjustment), saturation adjustment, hue adjustment, contrast adjustment, etc., and converts it into an image F2. The image F2 is stored in the storage unit 41.

[0049] The image F2 is provided to the flame state determination process and the ellipse estimation unit 37. The flame state determination process determines whether or not a flame is detected in the image of the detection area 6. If a flame is detected in the flame state determination process, it is predicted that the position of the lid 48 is irregular and that there is a gap between it and the charging port 46.

[0050] In the flame state determination process, the information processing unit 3 detects a partial image P having a predetermined characteristic from within the frame of image F2 using the partial image detection unit 32 (step S113). In this step, the partial image detection unit 32 detects a cluster of multiple pixels that can be distinguished from the surroundings by brightness as a partial image P that includes at least one of a flame and embers.

[0051] Next, the information processing section 3 causes the motion vector detection section 33 to detect a motion vector V of the partial video P between a plurality of frames of the video F2 (step S114).

[0052] Next, the information processing unit 3 specifies the area of ​​the partial video P using the area specifying unit 34 (step S115). As a method for specifying the area of ​​the partial video P, for example, the area can be specified based on the number of observation points included in the partial video P.

[0053] Next, the information processing unit 3 determines whether the color of the partial image P is flame-colored by using the color specifying unit 35 (step S116). In this step, the color specifying unit 35 determines whether the color of the partial image P is flame-colored based on the magnitude of the divided value R using the threshold value as a reference.

[0054] Next, the information processing unit 3 determines whether the flame state of the partial image P is good or not using the flame state determination unit 36 ​​(step S117). In this step, the flame state determination unit 36 ​​determines that the partial image P is an image of a flame if the area and movement amount of the partial image P are equal to or greater than the separation line S1 and the color of the partial image P is a flame color. The memory unit 41 stores the determination result of whether a flame has been detected.

[0055] If an image of a flame is detected in the image of the detection area 6 in step S117 and the flame condition is determined to be poor (N in step S117), this indicates a poor installation condition with a gap between the lid 48 and the loading port 46, and the process proceeds to step S120 without determining the lid condition and proceeds to a retry operation.

[0056] If no flame image is detected in the image of the detection area 6 and the flame condition is determined to be good (Y in step S117), the information processing unit 3 estimates an ellipse relating to the lid 48 and the opening of the loading port 46 using the ellipse estimation unit 37 (step S118).

[0057] An example of ellipse estimation will be described. In this example, the ellipse estimation unit 37 identifies the outline of the lid 48 and the outline of the charging port 46, which is the outline of the inner edge of the opening of the charging port 46, from the image F2. When viewed from diagonally above, the outlines of the lid 48 and the charging port 46 can be approximated as ellipses. Therefore, the ellipse estimation unit 37 identifies the ellipse of the lid 48 and the ellipse of the charging port 46 from the outlines of the lid 48 and the charging port 46, and converts the identified ellipses into mathematically expressible parameters (hereinafter referred to as "ellipse parameters"). In other words, the outlines of the lid 48 and the charging port 46 can be identified using these ellipse parameters. From these two ellipse parameters, the positional deviation in the X, Y, and Z directions of the outline of the lid 48 relative to the outline of the charging port 46 can be calculated.

[0058] After executing step S118, the information processing unit 3 determines whether the state of the lid 48 is good or bad using the lid state determination unit 38 (step S119). In this embodiment, the lid state determination unit 38 determines whether the lid posture is good or bad using a learning model (hereinafter referred to as "learning model F") that determines the lid state from the ellipse parameters identified by the ellipse estimation unit 37. The learning model F outputs a determination result of whether the lid posture is good or bad by inputting the above-mentioned ellipse parameters.

[0059] Learning model F is an AI model generated by machine learning based on previously acquired reference information and information on the quality of the installation state of the lid 48 corresponding to the reference information. In this example, the reference information is the ellipse parameters of the outline of the lid 48 and the ellipse parameters of the outline of the loading port 46. Learning model F tends to have higher identification accuracy by collecting information on a large number of cases in advance.

[0060] The learning model F can be generated using a known machine learning method such as a support vector machine, a neural network (including deep learning), or a random forest. The learning model F is stored in the memory unit 41. The learning model F may be generated based on actual measurement data collected in the past for other coke ovens of the same type. In the embodiment, the learning model F can be generated by the model generation unit 43 using a data set in which reference video collected for the specific target coke oven 40 is used as input data and pass / fail information is used as output data.

[0061] If the condition of the lid 48 is determined to be poor (N in step S119), the information processing unit 3 adds 1 to the count value of the retry counter 39 and determines whether the count value after the increment is equal to or less than an upper limit value (e.g., 2) (step S120). By limiting the number of retries, it is possible to prevent the operation time from exceeding the coal supply cycle time. Therefore, the upper limit value of the count value may be set according to the coal supply cycle time. The count value of the retry counter 39 is reset to zero at the start of process S110.

[0062] If the count value exceeds the upper limit, it is possible that there is some serious malfunction in the lid 48 or the loading port 46. Therefore, if the count value exceeds the upper limit (N in step S120), the alarm unit 42 of the information processing unit 3 notifies the operator that the count value has exceeded the upper limit (step S122). There are no limitations on the manner of the notification, but the alarm device 61 may emit light or sound to notify the operator, or information regarding the notification may be transmitted to an external information processing device (computer, etc.) via communication means. Once step S122 is executed, process S110 ends.

[0063] If the count value is equal to or less than the upper limit (Y in step S120), the coal loading car 100 executes a retry operation in which the lid 48 is removed from the charging port 46 and then reattached (step S121). In this step, when the removed lid 48 is reattached, the lid 48 and the area around the charging port 46 are blown away with air. After executing step S121, the process jumps to the beginning of step S111, and steps S111 to S120 are executed again.

[0064] If it is determined that the condition of the lid 48 is good (Y in step S119), the coal loading car 100 executes termination processing (step S123). In this termination processing, predetermined processing such as applying mortar around the lid 48 and to the charging port 46 is performed. After step S123 is executed, process S110 ends. Process S110 may be executed sequentially for multiple coking chambers 47. Process S110 is an example, and various modifications are possible.

[0065] The following describes the features of the flame detection device 1 of the embodiment. The flame detection device 1 of the embodiment is a flame detection device for steelmaking-related facilities, and includes an image acquisition unit 10 that acquires an image of a detection area 6, a partial image detection unit 32 that detects a partial image P having a predetermined characteristic from within a frame of the acquired image, a motion vector detection unit 33 that detects a motion vector V of the partial image P between frames of the image, and a flame state determination unit 36 ​​that determines the state of the flame in the detection area 6 using the detected motion vector V.

[0066] This configuration uses the motion vector V to determine the state of the flame in the detection area 6, reducing erroneous determinations due to partial images with little or no movement that are unlikely to be a flame. In particular, it reduces erroneous determinations due to infrared rays emitted from high-temperature areas of steelmaking-related equipment.

[0067] As an example, the flame detection device 1 further includes an area specification unit 34 that specifies the area of ​​the partial image P, and the flame state determination unit 36 ​​determines the state of the flame by referring to the area of ​​the specified partial image P. In this case, by referring to the area, it is possible to reduce erroneous determinations due to small partial images that are unlikely to be flames.

[0068] As an example, the flame detection device 1 further includes a color specification unit 35 that specifies the color of the partial image P, and the flame state determination unit 36 ​​determines the state of the flame by referring to the color of the partial image P specified by the color specification unit 35. In this case, by referring to the color of the partial image P, it is possible to reduce erroneous determinations based on partial images with a color that is unlikely to represent a flame.

[0069] As an example, the flame state determination unit 36 ​​determines the state of the flame using a learning model K obtained by machine learning using data on the motion vector V of the partial video P and data on the area of ​​the partial video P, both of which have been acquired in advance. In this case, the determination accuracy can be improved compared to determination using a simple threshold value.

[0070] As an example, the flame detection device 1 is installed on a coal loading car 100 that loads coal while moving in a first direction on the top surface of the coke oven 40, and the image acquisition unit 10 captures an image of the detection area 6 from diagonally above at a position away from the detection area 6 in the first direction. In this case, it is possible to reduce erroneous judgments caused by infrared rays emitted from high-temperature parts such as the riser pipe of the coke oven 40 and erroneous judgments caused by unburned coal. Furthermore, because the flame detection device 1 moves with the coal loading car 100, it is possible to simultaneously install the lid 48 and detect the flame, thereby improving work efficiency. Furthermore, because it is possible to make a judgment when the lid is installed, it is possible to minimize delays in the coal loading work that accompany the judgment.

[0071] Several exemplary embodiments of the present invention have been described in detail above. Each of the above-described embodiments merely illustrates a specific example of how the present invention may be implemented. The content of each embodiment does not limit the technical scope of the present invention, and many design modifications, such as changes, additions, and deletions of components, are possible within the scope of the invention as defined in the claims. In each of the above-described embodiments, content that allows such design modifications is described with the notation "in the embodiment" or "in the embodiment," but design modifications are also permitted even in content that does not have such notation.

[0072] The following describes modified examples. In the drawings and descriptions of the modified examples, the same or equivalent components and members as those in the embodiment are denoted by the same reference numerals. Explanations that overlap with the embodiment will be omitted as appropriate, and the description will focus on the configurations that differ from the embodiment.

[0073] (Variation) In the above description, an example has been shown in which image adjustments including trimming, scale adjustment, exposure adjustment (brightness adjustment), saturation adjustment, hue adjustment, contrast adjustment, etc. are applied to the input image F1, but the image adjustments are not limited to these. The image adjustments may include only some of these, or may include adjustments other than these. Note that it is not essential to apply image adjustments to the input image F1.

[0074] In the above description, an example has been shown in which one image acquisition unit 10 acquires an image of one detection area 6, but this is not limiting. An image of the detection area 6 may be acquired by one image acquisition unit 10, or an image of one detection area 6 may be acquired by multiple image acquisition units 10.

[0075] In the above description, an example has been given in which the learning model K is a model obtained by machine learning using data on the motion vector V of the partial video P and data on the area of ​​the partial video P as learning data, but this is not limiting. The learning model K may also be generated by machine learning using data on the motion vector V of the partial video P, data on the area of ​​the partial video P, and data on the color of the partial video P as learning data. In this case, by inputting the movement amount, area, and color of the partial video P into the learning model K, it is possible to obtain a result as to whether the partial video P is a flame or not.

[0076] In the above description, an example using an RGB histogram was shown, but this is not limiting. Three primary colors other than RGB may also be used. For example, the color identification unit may identify whether a partial image is a flame color using the color temperature of the partial image.

[0077] In the above description, an example has been shown in which light is emitted from the irradiating unit 12 when acquiring the image F1, but this is not limiting. When acquiring an image for flame detection, the irradiating unit 12 may be turned off. In this case, when acquiring an image for lid state determination, the irradiating unit 12 may be turned on.

[0078] Each of the above-described modifications provides the same functions and effects as the embodiment.

[0079] Any combination of the components and modifications of the above-described embodiments is also useful as an embodiment of the present invention. A new embodiment resulting from the combination has the combined effects of the respective embodiments and modifications. [Explanation of symbols]

[0080] 1 Flame detection device, 3 Information processing unit, 5 Coal loading unit, 6 Detection area, 10 Image acquisition unit, 31 Image adjustment unit, 32 Partial image detection unit, 33 Motion vector detection unit, 34 Area identification unit, 35 Color identification unit, 36 Flame state determination unit, 37 Ellipse estimation unit, 38 Lid state determination unit, 40 Coke oven, 48 Lid, 100 Coal loading car.

Claims

1. A flame detection device for steelmaking-related facilities, an image acquisition unit that acquires an image of the detection area; a partial image detection unit that detects a partial image having a predetermined characteristic from within a frame of the acquired image; a motion vector detection unit that detects a motion vector of the partial image between frames of the image; a flame state determination unit that determines the state of the flame in the detection area using the detected motion vector.

2. further comprising an area specification unit that specifies an area of ​​the partial image; The flame detection device according to claim 1 , wherein the flame state determination unit determines the state of the flame by referring to an area of ​​the identified partial image.

3. a color specification unit that specifies a color of the partial image, The flame detection device according to claim 1 , wherein the flame state determination unit determines the state of the flame by referring to the color of the partial image identified by the color identification unit.

4. 3. The flame detection device according to claim 2, wherein the flame state determination unit determines the state of the flame using a learning model obtained by machine learning using, as learning data, data on the motion vector of the partial image, data on the area of ​​the partial image, and data on the flame state in the detection area, each of which has been acquired in advance.

5. The flame detection device is provided on a coal loading car that loads coal while moving in a first direction on an upper surface of a coke oven, The flame detection device according to claim 1 , wherein the image acquisition unit captures an image of the detection area from a position away from the detection area in the first direction, obliquely from above.

6. A steelmaking facility equipped with the flame detection device according to claim 1.

7. A flame detection method for steelmaking-related facilities, comprising: Acquire an image of the detected area, Detecting a partial image having a predetermined characteristic from within the frame of the acquired image; Detecting a motion vector of the partial image between frames of the image; A flame detection method for determining a state of the flame in the detection area using the detected motion vector.

Images