Electric arc furnace equipment

Abstract

The electric arc furnace equipment includes an electric arc furnace, a camera, and an information processing device. The electric arc furnace uses the heat generated by the arc discharge between the scrap and the electrode to melt the scrap. The camera photographs the scrap put into the electric arc furnace. The information processing device inputs the configuration and density state of the scrap based on the image data obtained from the camera and the control actual data after the start of melting related to the electrode into an arc interruption prediction model, and predicts the occurrence of arc interruption caused by the short circuit between the scrap and the electrode due to the collapse of the melting scrap.

Description

Technical Field

[0001] The present invention relates to an electric arc furnace device that melts scrap by using heat generated by arc discharge between steel sheets (scrap) and electrodes. Background Art

[0002] An electric furnace supplies electric power to heat and melt scrap inserted into the furnace. In electric furnaces, there are: resistance furnaces that heat scrap by Joule heat, induction furnaces that heat scrap by inducing current in the scrap through electromagnetic induction and heating it by its own resistance, and electric arc furnaces that utilize heat generated by electric arcs and electrodes.

[0003] In particular, an electric arc furnace has the following advantages: it can use low-cost iron filings as the main raw material, it can be easily operated or stopped and the equipment operation is flexible, and it can achieve relatively large-scale production with less investment than a blast furnace.

[0004] An electric arc furnace generates an arc between the scrap and the electrodes or between the electrodes, and melts the scrap by its heat. By operating the electric power supplied to the electrodes and the distance between the electrodes and the scrap, an arc can be generated. Patent Document 1 discloses the following: impedance constant control is performed to stably generate an arc, and the lifting position of the electrodes is controlled.

[0005] From the time of charging the scrap until the melted scrap (molten steel) is discharged, usually 2 to 3 times of scrap additional charging and subsequent composition adjustment and other processes are performed. This series of operations is called feeding. In the additional charging of scrap, the operator of the electric arc furnace imagines the furnace interior situation based on empirical rules and charges a predetermined amount of scrap at a predetermined timing.

[0006] In addition, in order to stably generate an arc, sometimes the operator raises and lowers the electrodes in addition to automatic control. During the melting operation, sometimes the operator greatly separates the electrodes from the scrap due to certain important factors. For example, when the electrodes are close to the scrap and the current value exceeds the allowable range. In the case where the electrode is not raised even when the current value exceeds the allowable range, it will become an overcurrent and a short circuit will occur between the electrodes and the scrap. When a short circuit occurs, usually the circuit breaker will trip. Here, this phenomenon is called arc interruption.

[0007] Once the circuit breaker trips, its recovery takes time, so the operation stops and the production volume will decrease. The operator performs the lifting and lowering operation of the electrodes based on empirical rules to prevent this situation.

[0008] Prior Art Documents

[0009] Patent Documents

[0010] Patent Document 1: Japanese Patent Application Laid-Open No. 2006-85936

[0011] Patent Document 2: Japanese Patent Application Laid-Open No. 2018-28421

[0012] Patent Document 3: Japanese Patent Application Laid-Open No. 2008-116066 Summary of the Invention

[0013] Problems to be Solved by the Invention

[0014] As one of the important factors for the current value exceeding the allowable range, the unclear input state of the scrap input can be cited. The input amount of the scrap can be determined according to the operating conditions, but the input state of the scrap such as the density state and the offset of the scrap at the time of input cannot be operated. In addition, after the scrap is input, the lid is closed to maintain the temperature inside the furnace and prevent the molten steel from scattering to the periphery. Therefore, the operator cannot grasp the state inside the furnace during melting.

[0015] Depending on the input state of the scrap, sometimes a part of the scrap collapses during melting and contacts the electrode. In this case, an arc interruption is generated due to an accidental short circuit, so it cannot be prevented under the current situation. Thus, a method for grasping the input state of the scrap, the state inside the furnace during melting, or predicting and avoiding the occurrence of abnormal phenomena such as the collapse of the scrap is required.

[0016] As a method for grasping the state inside the furnace, for example, Patent Document 2 discloses a method of predicting the length of the electrode consumed due to melting and appropriately obtaining the distance between the scrap and the electrode. However, this is only a method for predicting the length of the electrode and cannot prevent the operation stop caused by an important abnormal factor such as the collapse of the scrap.

[0017] In addition, as a method for grasping the state inside the furnace, for example, Patent Document 3 discloses an operation method of providing thermometers at various places on the furnace wall and grasping the state inside the furnace based on the temperature outside the furnace. Accordingly, the state inside the furnace can be visualized according to the melting point of the steel type of the scrap input and the temperature of the entire furnace. However, by only obtaining the scrap state near the outer periphery of the furnace and the temperature history information of the molten scrap, it is not always possible to grasp the state of the scrap during melting. Therefore, it is difficult to prevent the operation stop caused by an important abnormal factor such as the collapse of the scrap.

[0018] The present invention has been made to solve the above-mentioned problems. An object of the present invention is to provide an electric arc furnace device capable of predicting the occurrence of the disconnection (arc interruption) of a circuit breaker caused by a short circuit due to the collapse of the scrap during melting.

[0019] Means for Solving the Problems

[0020] The first aspect relates to an electric arc furnace device.

[0021] The electric arc furnace device includes an electric arc furnace, a camera, and an information processing device.

[0022] The above-mentioned electric arc furnace uses the heat generated by the arc discharge between the scrap and the electrode to melt the above-mentioned scrap.

[0023] The above-mentioned camera photographs the above-mentioned scrap put into the above-mentioned electric arc furnace.

[0024] The above-mentioned information processing device includes a memory and a processor.

[0025] The above-mentioned memory stores information including image data at the time of scrap input obtained from the above-mentioned camera, control actual data after the start of melting related to the above-mentioned electrode, and an arc interruption prediction model.

[0026] The above-mentioned processor processes the above-mentioned information stored in the above-mentioned memory.

[0027] The above-mentioned processor is configured to input the configuration and density state of the above-mentioned scrap based on the above-mentioned image data and the above-mentioned control actual data into the above-mentioned arc interruption prediction model, and perform prediction processing for predicting the occurrence of arc interruption caused by the short circuit between the above-mentioned scrap and the above-mentioned electrode.

[0028] The second aspect further has the following features based on the first aspect.

[0029] The configuration and density state of the above-mentioned scrap are based on the brightness index of each pixel of the above-mentioned image data.

[0030] The above-mentioned control actual data includes operation time series data and operation attribute data.

[0031] The above-mentioned operation time series data includes current value, voltage value, and lifting speed value related to the above-mentioned electrode at each sampling moment.

[0032] The above-mentioned operation attribute data includes the input time and input amount of the above-mentioned scrap.

[0033] The third aspect further has the following features based on the second aspect.

[0034] The above-mentioned memory stores learning data.

[0035] The above-mentioned learning data includes the above-mentioned image data, the above-mentioned operation time series data, the above-mentioned operation attribute data, and an arc interruption occurrence label related to the scrap discharged from the above-mentioned electric arc furnace in the past. The above-mentioned arc interruption occurrence label indicates the occurrence of arc interruption associated with the sampling moment when the current value is higher than the short circuit related value for each sampling moment of the above-mentioned operation time series data.

[0036] The above-mentioned processor is configured to perform learning processing for constructing the above-mentioned arc break prediction model based on the above-mentioned learning data.

[0037] The fourth aspect further has the following features on the basis of any one of the first to third aspects.

[0038] The above-mentioned camera photographs the above-mentioned waste material input into the above-mentioned electric arc furnace three-dimensionally.

[0039] The above-mentioned image data includes a brightness index and a height index related to each pixel.

[0040] The arrangement and density state of the above-mentioned waste material are based on the brightness index and height index of each pixel of the above-mentioned image data.

[0041] The fifth aspect further has the following features on the basis of any one of the first to fourth aspects.

[0042] The above-mentioned processor is configured to perform short-circuit prevention processing for outputting an ascending instruction of the above-mentioned electrode when it is predicted by the above-mentioned prediction processing that an arc break will occur.

[0043] Advantages of the Invention

[0044] According to the present invention, the electric arc furnace device obtains information on the arrangement and density state of the waste material at the time of waste material input based on the image data obtained from the camera. The electric arc furnace device can predict the occurrence of an arc break according to this information and the current control actual data after the start of melting. Accordingly, the electric arc furnace device can predict the occurrence of the disconnection (arc break) of the circuit breaker caused by a short circuit due to the collapse of the waste material during melting. Brief Description of the Drawings

[0045] Figure 1 It is a diagram for explaining a configuration example of the electric arc furnace device according to Embodiment 1.

[0046] Figure 2 It is a diagram showing the imaging range of the waste material imaging camera according to Embodiment 1.

[0047] Figure 3 It is a block diagram exemplifying the outline of the functions of the arc break prediction device according to Embodiment 1.

[0048] Figure 4 It is a diagram showing the outline of the operation time series data and operation attribute data according to Embodiment 1.

[0049] Figure 5 It is a flowchart for explaining the steps of obtaining waste material image data in the camera image acquisition unit according to Embodiment 1.

[0050] Figure 6This is a diagram showing an example of the waste image data of Embodiment 1.

[0051] Figure 7 This is a diagram showing an outline of the storage of the brightness index of the waste image data of Embodiment 1.

[0052] Figure 8 This is a flowchart for explaining the steps of constructing the arc break prediction model in the arc break learning unit of Embodiment 1.

[0053] Figure 9 This is a flowchart for explaining the steps of assigning the arc break label in the arc break learning unit of Embodiment 1.

[0054] Figure 10 This is a diagram for explaining the steps of assigning the arc break label in the arc break learning unit of Embodiment 1.

[0055] Figure 11 This is a diagram showing an outline of the input value of Embodiment 1.

[0056] Figure 12 This is a flowchart for explaining the arc break prediction steps in the arc break prediction unit of Embodiment 1.

[0057] Figure 13 This is a diagram showing an outline of the brightness index and height index of the waste image data of Embodiment 2.

[0058] Figure 14 This is a schematic block diagram showing the functions of the arc break prediction device according to Example Embodiment 3.

[0059] Figure 15 This is a conceptual diagram showing a hardware configuration example of the processing circuit included in the arc break prediction device. Detailed Embodiment

[0060] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. In addition, the same reference numerals are assigned to the elements common to the respective drawings, and duplicate explanations are omitted.

[0061] Embodiment 1.

[0062] 1. Electric Arc Furnace

[0063] Figure 1 This is a diagram for explaining a configuration example of the electric arc furnace device 1 of Embodiment 1.

[0064] The electric arc furnace melts steel sheets (waste 2) containing various alloys input into the furnace to form molten steel 3. The electric arc furnace includes: a furnace body 4 for inputting waste 2 and storing the molten steel 3 after melting, a lid 5 for preventing heat dissipation during melting and the scattering of the molten steel 3 to the surroundings, and a screw hole 6 for discharging the molten steel 3.

[0065] When melting, the inside of the furnace is at a high temperature to melt the steel sheets (scrap 2). Therefore, refractory bricks are pasted on the inner surfaces of the furnace body 4 and the cover body 5. A hole is provided in the cover body 5, and the electrode 7 for generating an electric arc is inserted into the furnace body 4. Generally, the electrode 7 is formed of carbon and multiple electrodes are provided.

[0066] Electric power is supplied to these electrodes 7 to generate an electric arc between the electrodes 7 and the scrap 2 put into the interior. The scrap 2 is melted by the heat to obtain molten steel 3. In order to obtain a distance suitable for generating an electric arc between the electrode 7 and the scrap 2, the electrode 7 is lifted and lowered by an electrode lifting device 8. The electric power supplied to the electrode 7 and the lifting position of the electrode 7 are controlled by a control device 9. The output device 10 includes, for example, a power source and a circuit breaker.

[0067] At this time, command values such as the current and voltage for obtaining the electric power supplied to the electrode 7 and the lifting speed for obtaining the lifting position of the electrode 7 are set based on information such as the steel type of the scrap 2 and the input amount of the scrap 2. Although automatic operation can be performed based on these settings, as described above, since the generation of the electric arc is unstable, usually manual operation is applied by the operator each time. This manual operation is mainly applied to the lifting speed for changing the lifting position of the electrode 7.

[0068] In addition, the actual information during the actual melting operation is sometimes taken into the control device 9 and used for feedback control and the like.

[0069] Based on the molten steel 3 produced by one charge from the initial input of the scrap 2 until the molten steel 3 after melting is discharged, a melting number is set and used for product management and quality management.

[0070] The electric arc furnace equipment 1 of the present embodiment further includes a camera (scrap camera 20) for photographing the inside of the furnace body 4 and an information processing device (arc interruption prediction device 30) for predicting arc interruption.

[0071] 2. Scrap camera

[0072] The scrap camera 20 photographs the input state of the scrap 2 every time the scrap 2 is input. The imaging range of the scrap camera 20 is as Figure 2 shown. The scrap camera 20 is preferably arranged directly above the furnace body 4 so that the entire inside of the furnace can be photographed as Figure 2 shown. During melting, the inside of the furnace body 4 becomes at a high temperature, and its radiant heat is also transferred to the outside. Therefore, the scrap camera 20 preferably has a water cooling mechanism. Moreover, since the steam and gas generated by melting rise directly upward, the scrap camera 20 is preferably arranged at a distance from the furnace body 4 so as not to be affected by them.

[0073] In addition, when photographing the input state of the scrap 2, it is preferable to adjust the contrast, brightness, exposure, etc. so that the scrap 2 can be visually confirmed appropriately. These adjustments can be processed by the scrap photographing camera 20 or by the camera image acquisition unit 32 described later.

[0074] 3. Arc interruption prediction device

[0075] Figure 3 It is a block diagram outlining the functions of the arc interruption prediction device 30 of Example Embodiment 1. Figure 3 The configuration shown is an example and is not limited thereto. The arc interruption prediction device 30 predicts arc interruption based on the input state of the scrap 2 and the operation data during melting obtained each time the scrap 2 is input, and outputs the result.

[0076] (Operation data collection unit)

[0077] The operation data collection unit 31 collects control actual data. The control actual data includes operation time series data and operation attribute data.

[0078] The operation time series data is information on the control output during melting. The operation time series data includes, for example, the set and actual current values, actual voltage values, lifting speed reference values of each electrode 7, and a flag signal indicating whether the actual current value has reached the overcurrent level. In addition, the operation time series data also includes a flag signal indicating whether there is a manual operation by the operator, its lifting speed command value, and a flag signal indicating the on / off of the circuit breaker accompanying the occurrence of a fault such as arc interruption. The operation time series data is data of each sampling point (sampling time) that has acquired these J signals at a certain sampling interval Δt.

[0079] The operation attribute data is information accompanying other melting operations. The operation attribute data includes, for example, the input time and input amount of the scrap 2, the consumption state of the electrode 7, the temperature inside the furnace after melting, the steel type, composition of the input scrap 2, and the amount of elements input during composition adjustment. The operation attribute data is information that summarizes the melting conditions.

[0080] The information collected by the operation data collection unit 31 is, for example, Figure 4 As shown, the operation time series data and operation attribute data are stored in the database 33 in association with the melting number and the start time of each process.

[0081] (Camera image acquisition unit)

[0082] The camera image acquisition unit 32 acquires image data (imaging data) obtained by photographing the input state of the scrap 2 (the arrangement and density state of the scrap 2) from the scrap photographing camera 20. Then, the camera image acquisition unit 32 generates scrap image data for use by the arc-breaking learning unit 34 and the arc-breaking prediction unit 35.

[0083] Figure 5 It is a flowchart for explaining the generation process of the scrap image data in the camera image acquisition unit 32.

[0084] In step S100, the camera image acquisition unit 32 determines whether it is the timing when the scrap 2 is input into the furnace body 4. If it is not the timing when the scrap 2 is input, the camera image acquisition unit 32 waits for the scrap input timing.

[0085] On the other hand, if it is the timing when the scrap 2 is input, in step S110, the camera image acquisition unit 32 acquires the image data (imaging data) at the time of scrap input from the scrap photographing camera 20.

[0086] In step S120, the camera image acquisition unit 32 grayscales the imaging data. The brightness representing the density state at the time of scrap 2 input is obtained through grayscaling. As a method of grayscaling, for example, there are conversion methods as follows.

[0087] [Equation 1]

[0088]

[0089] [Equation 2]

[0090] I(x, y) = [r xy g xy b xy (2)

[0091] [Equation 3]

[0092]

[0093] [Equation 4]

[0094] I′(x, y) = 0.299·r xy + 0.587·g xy + 0.114·b xy (4)

[0095] Here, img is a numerical array of the color space of the captured data. w is the number of pixels in the horizontal direction of the image. h is the number of pixels in the vertical direction of the image. I is the pixel value at the horizontal pixel position x and the vertical pixel position y. img_gray is the numerical array after grayscale conversion. I’(x, y) is the pixel value after grayscale conversion at the horizontal pixel position x and the vertical pixel position y.

[0096] As an example, the color space of I(x, y) is described in the RGB space, but it is not limited thereto. In addition, each coefficient for calculating I’(x, y) is determined by ITU-R BT.601 (Studio encoding parameters of digital television for standard 4:3 and widescreen 16:9 aspect ratios International Telecommunication Union), an international standard related to the conversion between analog signals and digital signals, and other standards can also be complied with.

[0097] Depending on the device environment, the waste 2 may not always be visually confirmed. Therefore, after the waste is put in and before the lid 5 is closed, after obtaining multiple captured data, the grayscale image (Equation (5)) obtained through the above processing can be obtained based on the average value at each pixel position.

[0098] [Equation 5]

[0099]

[0100] Here, N cap is the number of captured data obtained after the waste is put in and before the lid 5 is closed.

[0101] Next, in step S130, the camera image acquisition unit 32 changes the number of pixels of the grayscale data.

[0102] After smoothing the grayscale image, the camera image acquisition unit 32 adjusts it to a specified number of pixels (data number) used in the broken arc learning unit 34 and the broken arc prediction unit 35 to obtain waste image data. For example, the Gaussian filter smooths the image by weighting the nearby pixel values with the Gaussian distribution g (Equation (6)).

[0103]

[0104] [Equation 6]

[0105] This is an example that can also perform smoothing based on average filtering and median filtering. Adjust the number of pixels for the smoothed image. The number of pixels can be adjusted through the following steps (Equation (7)).

[0106] [Equation 7]

[0107]

[0108] Here, img_resize is the numerical array of the smoothed image. x is the horizontal pixel position of the smoothed image. y is the vertical pixel position of the smoothed image. α is the magnification. The symbol [] represents rounding. R is the pixel value (brightness index) of the smoothed image. w’ is the number of horizontal pixels of the smoothed image. h’ is the number of vertical pixels of the smoothed image.

[0109] The brightness of the pixels is used as an indicator representing the input state of the scrap 2. The configuration and density state of the scrap 2 are represented based on the brightness index of each pixel of the scrap image data. Figure 6 This is a diagram showing an example of the scrap image data.

[0110] In step S140, the camera image acquisition unit 32 stores the scrap image data in the database 33. The numerical array of the scrap image data is stored in the database 33 as a brightness index. The scrap image data is saved Figure 7 in association with the melting number and the photographing time as such.

[0111] (Arc break learning unit)

[0112] Next, the arc break learning unit 34 will be described.

[0113] The arc break learning unit 34 constructs an arc break prediction model based on the learning data and evaluation data stored in the database 33. The learning data and evaluation data include scrap image data, operation time series data, operation attribute data, and arc break generation labels related to the scrap 2 discharged from the electric arc furnace in the past. The arc break generation label indicates the generation of an arc break associated with the sampling time when the current value is higher than the short-circuit correlation value for each sampling time of the operation time series data. Refer to Figure 8 to describe the construction steps of the arc break prediction model.

[0114] Figure 8 This is a flowchart for explaining the construction steps of the arc break prediction model in the arc break learning unit 34.

[0115] In step S200, the arc break learning unit 34 determines the total number of melting numbers stored in the database 33, that is, the number of charges M charge is more than the number of times M required for model learning Learning .

[0116] At M charge More than M Learning In the case of, in step S210, the arc-breaking learning unit 34 obtains operation time series data, operation attribute data, and scrap image data from the database 33 based on the melting number and date and time. On the other hand, at M charge Is M Learning In the following cases, the number of data used for learning the arc-breaking prediction model is insufficient. Therefore, the arc-breaking learning unit 34 executes the process of step S200 again after the current charging is completed.

[0117] In step S220, the arc-breaking learning unit 34 obtains an arc-breaking label. Here, refer to Figure 9 The step of assigning the arc-breaking label in step S220 will be described. Figure 9 It is a flowchart for explaining the step of assigning the arc-breaking label in the arc-breaking learning unit 34. Figure 9 The process is executed for each operation time series data stored in the database 33.

[0118] In Figure 9 In step S300 of, the arc-breaking learning unit 34 obtains operation time series data.

[0119] In step S310, the arc-breaking learning unit 34 determines whether the actual current value I of the data at the sampling point n of the obtained operation time series data Act(n) Is less than the specified current value I ARC . In I Act(n) Is less than I ARC In this case, the process of step S320 is executed.

[0120] In step S320, the arc-breaking learning unit 34 determines whether the on-off signal S of the circuit breaker of the data at the sampling point n VCB(n) Is off. In S VCB(n) Is off, the process of step S330 is executed.

[0121] In step S330, the arc-breaking learning unit 34 determines whether the signal S indicating whether there is a manual operation by the operator MAN(n) Is off. In S MAN(n) Is off, the process of step S340 is executed. That is, when all the determination conditions in steps S310 to S330 are satisfied, the process of step S340 is executed.

[0122] In step S340, the arc-breaking learning unit 34 obtains the specified number of sampling points N before and after the sampling point n OC The number of sampling points (n - N OC , n - N OC+1, n - N OC +2, ……, n - 1, n, n + 1, …, n + N OC -2, n + N OC -1, n + N OC ) of the operation time series data.

[0123] In step S350, the arc-breaking learning unit 34 determines whether there is a signal S indicating whether the actual current value has reached the level of overcurrent (short-circuit related value) in the operation time series data obtained in step S340 Overcurrent The moment when it becomes ON. When there is S Overcurrent In the case of the moment when it becomes ON, the arc-breaking learning unit 34 applies a mark to the arc-breaking label L ARC Apply a mark (L ARC = 1).

[0124] The specified number of sampling points N at the sampling point n OC Is a predetermined number of points that specifies a shorter range within the melting process of the operation time series data. Then, at this time, it is necessary to capture this omen before the arc-breaking occurs. Therefore, after tracing the number of sampling points N aim Calculated by Equation (8) based on the desired moment T aim For arc-breaking prediction and the sampling interval Δt, an arc-breaking label is assigned at the sampling point obtained by tracing that amount. That is, the arc-breaking learning unit 34 applies a mark to the arc-breaking label L aim At the sampling point n - N ARC(n-Naim) Apply a mark.

[0125] [Equation 8]

[0126] N aim = T aim / Δt (8)

[0127] That is, when all the determination conditions of steps S310 to S350 are satisfied, in step S360, the arc-breaking learning unit 34 applies a mark to the arc-breaking label L ARC(n-Naim) Apply a mark (L ARC(n-Naim) = 1). The arc-breaking label to which the mark has been applied is called the arc-breaking generation label. On the other hand, when any of the determination conditions of steps S310 to S350 is not satisfied, in step S370, the arc-breaking learning unit 34 does not apply a mark to the arc-breaking label L ARC(n-Naim) Apply a mark (L ARC(n-Naim) = 0).

[0128] The above method of assigning the arc-breaking label is an example. The arc-breaking label can also be generated according to other conditions and steps, or can be generated by the operator from a certain input terminal. The arc-breaking label is stored in the database 33 in association with the melting number and sampling point of the operation time series data.

[0129] Return Figure 8 The construction steps of the broken arc prediction model will be further described.

[0130] In step S220, through the above Figure 9 processing, the broken arc labels stored in the database 33 are obtained.

[0131] In step S230, the broken arc learning unit 34 determines whether the total number N Label of broken arc labels is more than the specified number N Label ’ of broken arc generation times. The total number N Label of broken arc labels is the total number of broken arcs generated at each sampling point of the operation time series data of the charge quantity M charge quantity.

[0132] When N Label is more than N Label ’, in step S240, the broken arc learning unit 34 obtains learning data from the database 33. The learning data is a part of the group of waste image data, operation time series data, operation attribute data, and broken arc labels stored in the database 33. On the other hand, when N Label is N Label ’ or less, the amount of data used in the learning of the broken arc prediction model is insufficient. Therefore, the broken arc learning unit 34 executes the process of step S210 again after the current charge is completed.

[0133] In step S250, the broken arc learning unit 34 uses the obtained learning data to learn the broken arc prediction model described later.

[0134] In step S260, the broken arc learning unit 34 obtains evaluation data from the database 33. The evaluation data is a part of the data group formed by the group of waste image data, operation time series data, operation attribute data, and broken arc labels stored in the database 33, excluding the learning data.

[0135] In step S270, the broken arc learning unit 34 uses the evaluation data to evaluate the broken arc prediction model described later.

[0136] In step S280, the broken arc learning unit 34 determines whether to apply the broken arc prediction model. When the evaluation value is equal to or higher than the specified value, the broken arc prediction model is saved in the database 33. On the other hand, when the evaluation value is lower than the specified value, the accuracy of the learned broken arc prediction model is insufficient. Therefore, the broken arc learning unit 34 executes the process of step S210 again after the current charge is completed.

[0137] In step S290, the broken arc learning unit 34 updates or saves the broken arc prediction model. When there is an already learned broken arc prediction model, the broken arc learning unit 34 can update it or save it separately. The broken arc prediction model is stored in the database 33.

[0138] Specifically, the learning of the broken arc prediction model in step S250 and the evaluation of the broken arc prediction model in step S270 are described.

[0139] As an example, the broken arc prediction model can be represented by a linear regression model as follows.

[0140] [Equation 9]

[0141]

[0142] [Equation 10]

[0143] I pred = Xβ + ε (10)

[0144]

[0145] [Equation 11]

[0146] β = ( X TX) -1 X T I pred (11)

[0147] [Equation 12]

[0148]

[0149] Here, L pred is the broken arc prediction, I pred is the predicted current value, X is the explanatory variable (input value) of the prediction model, β is the coefficient, ε is the disturbance term, N L is the total number of learning data, and p is the number of explanatory variables. In Equation (9), the case where the broken arc prediction L pred i = 1 at the sampling point i means that it is predicted that an arc break will occur at the time i + N aim after the number of sampling points N aim shown in Equation (8).

[0150] The broken arc prediction model is constructed by determining β and ε in a way that satisfies Equation (10) and Equation (11) using the learning data. Then, the evaluation data and Equation (12) are used to evaluate the broken arc prediction model. The broken arc prediction model is preferably constructed such that Equation (12) becomes minimum for the learning data and the evaluation data. Here, N Vis the total number of data used in the evaluation. In the above step S280, the evaluation value based on Equation (12) can be used.

[0151] Taking the linear regression model as an example, a non-linear model such as machine learning can also be used to construct an arc break prediction model. For example, as an arc break model, an ensemble learning model based on decision trees, that is, a classifier based on XGBoost (Extreme Gradient Boosting), the same Random Forest, etc. can also be applied. Here, as an example, the concept of XGBoost is shown below. In XGBoost, the predicted value is obtained by aggregating the results of multiple parallel decision trees. For the input value X i the obtained predicted value y pred i can be obtained by the following formula.

[0152] [Equation 13]

[0153]

[0154]

[0155]

[0156] Here, X i is the input value, K is the number of trees, T is the number of leaves, ω is the weight of the leaf, and F is the regression tree space. At the initial stage of model construction, the number of trees K = 1. When the difference between the obtained predicted value and the target value is minimized, a model with good accuracy is obtained. The accuracy of the model is evaluated by calculating a loss function L(φ) as follows.

[0157] [Equation 14]

[0158]

[0159]

[0160] Here, y i is the target value, l is the residual between the predicted value and the target value, Ω is the regularization term, and γ and λ are parameters. The model that minimizes the loss function L(φ) is obtained. To generate the model that minimizes the loss function L(φ), GradientBoosting is used to increase the decision trees to minimize L(φ). In the case of using such a classifier, the predicted value y pred i is obtained as the arc break prediction L pred . Alternatively, in the case of a regression model for predicting the current value, the predicted value y pred i can also be obtained as the predicted current value I pred, and determine the broken arc according to the conditions such as formula (9).

[0161] In order to output the broken arc prediction L at the sampling point i pred i The explanatory variables input to the broken arc prediction model can be obtained as follows, for example.

[0162] [Number 15]

[0163]

[0164] Here, R w’h’ is the brightness index of the scrap image data, P K is the operation attribute data, O i (j) is the j component (j = 1, 2,..., J) of the operation time series data of the sampling point i. The operation time series data can use only the component of the sampling point i as the explanatory variable, or can add the components of the sampling points of a specified sampling point N Buff quantity of sampling points to the explanatory variable. N Buff is Figure 11 the number of points in a short range within the melting process as shown. Obtain the operation time series data corresponding to the time data associated with the brightness index R w’h’ of the scrap image data and the operation attribute data P K That is, when the broken arc prediction model is a linear regression model, for example, p (the number of explanatory variables) in formula (10) is expressed as follows.

[0165] [Number 16]

[0166]

[0167] (Broken arc prediction unit)

[0168] Next, the broken arc prediction unit 35 will be described.

[0169] The broken arc prediction unit 35 inputs the configuration and density state of the scrap 2 based on the scrap image data and the control actual data into the broken arc prediction model, and predicts the occurrence of a broken arc caused by the short circuit between the scrap 2 and the electrode 7.

[0170] Figure 12 is a flowchart for explaining the broken arc prediction steps in the broken arc prediction unit 35.

[0171] In step S400, the broken arc prediction unit 35 determines whether it is the timing when the scrap 2 is input into the furnace body 4. If it is not the timing when the scrap 2 is input, the broken arc prediction unit 35 waits for the scrap input timing.

[0172] On the other hand, if it is the timing when the scrap 2 is input, by the aboveFigure 5 In the same process, waste image data is generated by the camera image acquisition unit 32 before the start of melting.

[0173] In step S410, the arc break prediction unit 35 acquires the latest waste image data. The waste image data can be directly acquired from the camera image acquisition unit 32, or can be acquired after being stored in the database 33.

[0174] In step S420, the arc break prediction unit 35 determines whether melting has started. If melting has not started, the arc break prediction unit 35 waits for the start of melting.

[0175] After the start of melting, in step S430, the arc break prediction unit 35 acquires the operation time series data and operation attribute data in the current charge. The operation time series data collected by the operation data collection unit 31 can be directly and continuously acquired from the operation data collection unit 31 according to the sampling period, or can be acquired after being stored in the database 33.

[0176] In step S440, using the waste image data, operation time series data, and operation attribute data as input values, an arc break prediction model is used to predict arc break. The input values of the arc break prediction model are explanatory variables (Equation (15)) as in the case of model construction. Therefore, the sampling point quantity from the latest sampling point n Now is traced back by N Buff points as the operation time series data. The input values (explanatory variables) of the arc break prediction model obtained here are input to the arc break prediction model stored in the database 33 to predict arc break.

[0177] In step S450, the arc break prediction unit 35 outputs the arc break prediction result related to the latest sampling point n Now For example, the arc break prediction result is stored in the database 33. Here, when the arc break prediction L pred becomes 1, the arc break prediction output unit 36 reports the generation prediction of arc break through images, sounds, vibrations, etc. The operator who receives the report can perform an operation to raise the electrode 7 to avoid arc break. Thus, the operation stop time due to the tripping of the circuit breaker can be reduced.

[0178] In step S460, the arc break prediction unit 35 determines whether melting has ended. If melting has not ended, the arc break prediction unit 35 returns to the process of step S430 again and predicts arc break based on the latest control actual data.

[0179] 4. Effects

[0180] As described above, the arc interruption prediction device 30 according to the present embodiment can predict the occurrence of arc interruption (arc interruption) of the circuit breaker caused by a short circuit due to the collapse of the scrap during melting. In addition, by reporting the prediction of the occurrence of arc interruption, the operator can raise the electrode 7 to prevent a short circuit caused by the contact between the scrap 2 and the electrode 7. Therefore, it is possible to avoid the operation stop due to the tripping of the circuit breaker.

[0181] 5. Modification

[0182] However, in the electric arc furnace apparatus 1 of the above-described Embodiment 1, the arc interruption prediction unit 35 predicts the occurrence of arc interruption using the arc interruption prediction model constructed in the arc interruption learning unit 34. However, in the case where an appropriate arc interruption prediction model already exists, the arc interruption prediction unit 35 may also use this prediction model to predict the occurrence of arc interruption. This also applies to the following embodiments.

[0183] In addition, in Figure 3 the arc interruption prediction output unit 36 is arranged outside the arc interruption prediction device 30, but may also be included in the arc interruption prediction device 30.

[0184] Embodiment 2.

[0185] Next, Embodiment 2 of the present invention will be described with reference to Figure 13 The description of the above-described embodiment will be omitted.

[0186] In Embodiment 2, as an index indicating the input state of the scrap 2, the height is used in addition to the brightness of the pixels. The arrangement and density state of the scrap 2 are represented based on the brightness index and height index of each pixel of the scrap image data. Thus, as an explanatory variable of the arc interruption prediction model, the input state of the scrap 2 is given more accurately, and it is possible to predict with high accuracy whether arc interruption occurs.

[0187] The scrap camera 20 of Embodiment 2 can photograph an object three-dimensionally. The arrangement of the scrap camera 20 is the same as that of the above-described Embodiment 1. The imaging data of the scrap camera 20 includes height information when the scrap is input.

[0188] As the type of camera capable of three-dimensional photography, for example, there are a stereo camera method, a ToF (Time of Flight) method, a projector method, etc. In the stereo camera method, multiple cameras are used, and the distance between the object and the stereo camera is derived based on the respective setting intervals, focal lengths, and imaging differences (parallax). In the ToF method, the distance between the object and the camera is derived based on the speed at which the pulsed light and continuous light irradiated from the same camera are reflected by the object. In the projector method, the height difference of the object is derived based on the deformation of the stripe pattern irradiated onto the object.

[0189] The three-dimensional photography method shown here is a method for explaining its principle and can be carried out in any way. Additionally, even if it is a method not shown here, as long as it can perform three-dimensional photography. As long as it has the structure of the furnace body 4 and the information on the location where the waste photography camera 20 is installed, it is possible to obtain the height information of the waste 2 in the furnace based on the three-dimensional camera after the waste is input and the height information constituted by the height difference of the waste 2.

[0190] The above-mentioned luminance value array img based on the input state of the waste resize and the height value array img of the input state of the waste height For example, they can be obtained as follows.

[0191] [Equation 17]

[0192]

[0193]

[0194] Here, H is the height value (height index) among the horizontal pixel number w' and the vertical pixel number h'. The camera image acquisition unit 32 stores the waste image data containing this information in the database 33 in the form shown Figure 13 as shown.

[0195] Based on the waste image data obtained in this way, a broken arc prediction model is constructed in the same way as in Embodiment 1, and this prediction model is used to predict the broken arc. At this time, the construction of the broken arc prediction model and the input values (explanatory variables) used in the prediction are configured as follows.

[0196] [Equation 18]

[0197]

[0198] As described above, according to the electric arc furnace device 1 of Embodiment 2, the information including the luminance index and the height index is used as the explanatory variable of the broken arc prediction model. By adding the height index, compared with Embodiment 1, the generation of the broken arc can be predicted with better accuracy.

[0199] Embodiment 3.

[0200] Next, Embodiment 3 will be described with reference to Figure 14 This. The description repeated with the above embodiments will be omitted.

[0201] Embodiment 3 is characterized in that, in addition to the arc break prediction output unit 36 that reports a prediction of an arc break in the case where an arc break is predicted, it further has a short-circuit prevention output unit 37. The short-circuit prevention output unit 37 automatically operates the lifting of the electrode in the case where an arc break is predicted, thereby preventing the circuit breaker from being cut off due to a short circuit associated with the contact between the waste material and the electrode.

[0202] The input value in Embodiment 3 may also be any of the ways in Embodiment 1 or Embodiment 2. In the arc break prediction unit 35, when the arc break prediction L pred becomes 1, an upward command value for the electrode 7 is output in the short-circuit prevention output unit 37. The command value is output to the control device 9 to raise the electrode 7 to a predetermined avoidance height. Therefore, the cutting off of the circuit breaker can be prevented without waiting for the operation of the operator.

[0203] In addition, in Figure 14 , the short-circuit prevention output unit 37 is arranged outside the arc break prediction device 30, but may also be included in the arc break prediction device 30.

[0204] (Hardware configuration example)

[0205] Figure 15 is a conceptual diagram showing a hardware configuration example of the processing circuit included in the arc break prediction device 30 of each of the above embodiments. Figure 3 and Figure 14 show a part of the functions of each part in the arc break prediction device 30, and each function is implemented by a processing circuit. As one mode, the processing circuit includes at least one processor 91 and at least one memory 92. As another mode, the processing circuit includes at least one dedicated hardware 93.

[0206] When the processing circuit includes the processor 91 and the memory 92, each function is implemented by software, firmware, or a combination of software and firmware. At least one of the software and the firmware is described as a program. At least one of the software and the firmware is stored in the memory 92. The processor 91 reads and executes the program and various data stored in the memory 92, thereby implementing each function. The memory 92 includes a main storage device and an auxiliary storage device. The memory 92 stores various data saved in the database 33.

[0207] When the processing circuit includes the dedicated hardware 93, the processing circuit is, for example, a single circuit, a composite circuit, a programmed processor, or a combination of these. Each function is implemented by the processing circuit.

[0208] As described above, the embodiments of the present invention have been described, but the present invention is not limited to the above embodiments, and various modifications can be made and implemented within the scope not departing from the gist of the present invention. When numerical values such as the number, quantity, amount, range, etc. of each element are mentioned in the above embodiments, unless otherwise specifically stated or unless the numerical value can be clearly determined in principle, the present invention is not limited to the mentioned numerical values. In addition, the structures and the like described in the above embodiments are not essential in the present invention unless otherwise specifically stated or clearly determined herein in principle.

[0209] Description of Symbols

[0210] 1 Electric arc furnace equipment

[0211] 2 Scrap

[0212] 3 Molten steel

[0213] 4 Furnace body

[0214] 5 Cover

[0215] 6 Screw hole

[0216] 7 Electrode

[0217] 8 Electrode lifting device

[0218] 9 Control device

[0219] 10 Output device

[0220] 20 Scrap camera

[0221] 30 Arcing prediction device

[0222] 31 Operation data collection unit

[0223] 32 Camera image acquisition unit

[0224] 33 Database

[0225] 34 Arcing learning unit

[0226] 35 Arcing prediction unit

[0227] 36 Arcing forecast output unit

[0228] 37 Short - circuit prevention output unit

[0229] 91 Processor

[0230] 92 Memory

[0231] 93 Hardware

Claims

1. An electric arc furnace device that uses the heat generated by the arc discharge between the scrap and the electrode to melt the above-mentioned scrap, characterized in that, Comprising: a camera that photographs the above-mentioned waste material introduced into the electric arc furnace; and an information processing device including a memory and a processor, the above-mentioned memory stores information including image data at the time of waste material input obtained from the above-mentioned camera, control actual data after the start of melting related to the above-mentioned electrode, and an arc interruption prediction model, and the above-mentioned processor processes the above-mentioned information stored in the above-mentioned memory, the above-mentioned processor is configured to input the configuration and density state of the above-mentioned waste material based on the above-mentioned image data and the above-mentioned control actual data into the above-mentioned arc interruption prediction model, and perform prediction processing for predicting the occurrence of arc interruption caused by a short circuit between the above-mentioned waste material and the above-mentioned electrode.

2. The electric arc furnace device according to claim 1, characterized in that, The configuration and density state of the above-mentioned waste material are based on the brightness index of each pixel of the above-mentioned image data, the above-mentioned control actual data includes operation time series data and operation attribute data, the above-mentioned operation time series data includes current value, voltage value, and lifting speed value related to the above-mentioned electrode at each sampling moment, the above-mentioned operation attribute data includes the input time and input amount of the above-mentioned waste material.

3. The electric arc furnace device according to claim 2, characterized in that, The above-mentioned memory stores learning data, the above-mentioned learning data includes the above-mentioned image data, the above-mentioned operation time series data, the above-mentioned operation attribute data, and an arc interruption occurrence label related to the waste material discharged from the above-mentioned electric arc furnace in the past, and the above-mentioned arc interruption occurrence label indicates the occurrence of arc interruption associated with the sampling moment when the current value is higher than the short circuit related value for each sampling moment of the above-mentioned operation time series data, the above-mentioned processor is configured to perform learning processing for constructing the above-mentioned arc interruption prediction model based on the above-mentioned learning data.

4. The electric arc furnace device according to any one of claims 1 to 3, characterized in that, The above-mentioned camera photographs the above-mentioned waste material introduced into the above-mentioned electric arc furnace three-dimensionally, the above-mentioned image data includes a brightness index and a height index related to each pixel, the configuration and density state of the above-mentioned waste material are based on the brightness index and height index of each pixel of the above-mentioned image data.

5. The electric arc furnace device according to any one of claims 1 to 3, characterized in that, The above-mentioned processor is configured to perform short circuit prevention processing for outputting an ascending instruction for the above-mentioned electrode when it is predicted by the above-mentioned prediction processing that an arc interruption occurs.

Images