Method for creating cast slab quality prediction model, quality prediction model creation device, and quality control method

Abstract

A method for creating a cast slab quality prediction model according to the present invention includes: an imaging step for capturing an image of a molten metal surface inside a mold of a continuous casting machine; a determination step for determining the operation mode of the continuous casting machine by using the image captured in the imaging step; an extraction step for extracting time-series image data in cast-slab units by subjecting the image captured in the imaging step to image processing; a calculation step for calculating a feature quantity for the time-series image data extracted in the extraction step; and a creation step for creating, for each operation mode of the continuous casting machine, a quality prediction model having the time-series image data feature quantity as an input variable and the quality of the cast slab corresponding to the time-series image data as an output variable, by using the operation mode of the continuous casting machine when the time-series image data extracted in the extraction step is obtained, the feature quantity of the time-series image data, and an actual value of the quality of the cast slab corresponding to the time-series image data.

Description

Cast slab quality prediction model creation method, quality prediction model creation device, and quality control method

[0001] The present invention relates to a method for creating a quality prediction model for a slab, a quality prediction model creating device, and a quality control method.

[0002] One of the qualities required for cast slabs produced by continuous casting machines is that they have few defects caused by impurities such as bubbles and inclusions mixed near the surface of the slab. In a continuous casting machine, molten metal poured into a mold through an immersion nozzle begins to solidify in a shell-like shape from the mold wall (hereinafter, the shell-like solidified steel is referred to as the solidified shell), and the thickness of the solidified shell increases as casting progresses. Bubbles, inclusions, and other particles are suspended in the molten metal poured into the mold. If these bubbles, inclusions, and other particles are trapped in the solidified shell and solidification continues, the above-mentioned defects occur. Against this background, methods have been proposed for capturing images of the molten metal surface in the mold (hereinafter, referred to as molten metal surface images) and controlling or monitoring the operating status of a continuous casting machine based on the molten metal surface images (see, for example, Patent Documents 1 to 6).

[0003] JP-A-2-235556 JP-A-5-77015 JP-A-8-117943 JP-A-10-34305 Patent No. 7056401 Patent No. 7337297

[0004] Generally, the appearance of a molten metal surface image changes significantly depending on the operating mode of the continuous casting machine. However, all conventional methods utilize molten metal surface images without taking into account the operating mode of the continuous casting machine. Therefore, conventional methods may not be able to accurately control the quality of a cast slab when the operating mode of the continuous casting machine changes. More specifically, when operator work such as molten steel collection or powder scattering occurs, the molten metal surface image becomes different from the normal molten metal surface image. Furthermore, at the end of casting, the immersion nozzle is withdrawn, making the molten metal surface image brighter than the normal molten metal surface image. Therefore, when controlling or monitoring the operating state of a continuous casting machine based on molten metal surface images, it is necessary to consider such changes in the molten metal surface image that accompany changes in the operating mode, otherwise the quality of the cast slab may not be accurately controlled.

[0005] The present invention has been made to solve the above-mentioned problems, and an object of the present invention is to provide a slab quality prediction model creation method and a quality prediction model creation device that are capable of creating a quality prediction model that accurately predicts slab quality. Another object of the present invention is to provide a slab quality control method that is capable of accurately controlling slab quality.

[0006] The method for creating a quality prediction model of a slab according to the present invention is a method for creating a quality prediction model that predicts the quality of a slab produced in a continuous casting machine, and includes the following steps: a photographing step of photographing images of the surface of molten metal in a mold of the continuous casting machine; a determination step of determining the operation mode of the continuous casting machine using the images photographed in the photographing step; a cutting step of extracting time-series image data on a slab-by-slab basis by performing image processing on the images photographed in the photographing step; a calculation step of calculating feature quantities of the time-series image data extracted in the cutting step; and a creation step of creating, for each operation mode of the continuous casting machine, a quality prediction model in which the feature quantities of the time-series image data are input variables and the quality of the slab corresponding to the time-series image data is an output variable, using the operation mode of the continuous casting machine when the time-series image data extracted in the cutting step was obtained, the feature quantities of the time-series image data, and actual values ​​of the quality of the slab corresponding to the time-series image data.

[0007] The operation mode of the continuous casting machine is preferably determined in accordance with the operating mode of the mold.

[0008] The method may further include a step of changing the feature values ​​included in the quality prediction model created in the creating step in accordance with the contribution rate of each feature value included in the quality prediction model to the quality prediction of the slab.

[0009] The input variables of the quality prediction model preferably include information on the determination result of the operation mode of the continuous casting machine in the determination step.

[0010] The slab quality prediction model creation device of the present invention is a slab quality prediction model creation device that creates a quality prediction model that predicts the quality of a slab produced in a continuous casting machine, and includes: a determination means that determines the operation mode of the continuous casting machine using images of the surface of molten metal in the mold of the continuous casting machine; an extraction means that extracts time-series image data on a slab-by-slab basis by applying image processing to the images; a calculation means that calculates feature quantities of the time-series image data extracted by the extraction means; and a creation step that uses the operation mode of the continuous casting machine when the time-series image data extracted by the extraction means was obtained, the feature quantities of the time-series image data, and the actual quality values ​​of the slab corresponding to the time-series image data to create a quality prediction model for each operation mode of the continuous casting machine, with the feature quantities of the time-series image data as input variables and the quality of the slab corresponding to the time-series image data as output variables.

[0011] The slab quality control method according to the present invention is a slab quality control method for controlling the quality of a slab using a quality prediction model created by the slab quality prediction model creation method according to the present invention, and includes: a determination step of determining the operating mode of the continuous casting machine using an image of the surface of the molten metal in the mold of the continuous casting machine; a prediction step of predicting the quality of the slab by inputting feature quantities of time-series image data obtained from the image into a quality prediction model corresponding to the operating mode of the continuous casting machine determined in the determination step; and a step of controlling the operating conditions of the continuous casting machine based on the quality of the slab predicted in the prediction step.

[0012] The slab quality prediction model creation method and quality prediction model creation device according to the present invention make it possible to create a quality prediction model that accurately predicts the quality of a slab, and the slab quality control method according to the present invention makes it possible to accurately control the quality of a slab.

[0013] Fig. 1 is a block diagram showing the configuration of a slab quality control system according to one embodiment of the present invention. Fig. 2 is a flowchart showing the flow of a quality prediction model creation process according to one embodiment of the present invention. Fig. 3 is a flowchart showing the flow of a quality control process according to one embodiment of the present invention. Fig. 4 is a diagram showing an example of time-series changes in brightness of a molten metal surface image. Fig. 5 is a diagram for explaining the quality prediction model creation process of the example.

[0014] Hereinafter, a cast quality prediction model creation method, a quality prediction model creation device, and a quality control method according to one embodiment of the present invention will be described with reference to the drawings.

[0015] [System Configuration] First, with reference to FIG. 1, the configuration of a cast slab quality control system according to one embodiment of the present invention will be described.

[0016] Fig. 1 is a block diagram showing the configuration of a slab quality control system according to one embodiment of the present invention. As shown in Fig. 1, the slab quality control system 1 according to one embodiment of the present invention is a system for controlling the quality of a slab produced by a continuous casting machine A, and includes an imaging device 2, a quality prediction model creation device 3, and a control device 4.

[0017] The imaging device 2 is constituted by a known imaging device and is installed in the vicinity of the mold of the continuous casting machine A. The imaging device 2 takes an image of the surface of the molten metal in the mold (hereinafter referred to as a molten metal surface image) and inputs the data of the taken molten metal surface image into the quality prediction model creation device 3 and the control device 4. The installation position of the imaging device 2 may be any position as long as it is a position where the molten metal surface image can be taken.

[0018] The quality prediction model creation device 3 is composed of an information processing device such as a computer, and creates a quality prediction model that predicts the quality of the cast piece produced in the continuous casting machine A by executing the quality prediction model creation process described below.

[0019] The control device 4 is composed of an information processing device such as a computer, and controls the operating conditions of the continuous casting machine A and controls the quality of the cast pieces produced by the continuous casting machine A by executing the quality control processing described below.

[0020] [Quality Prediction Model Creation Processing] Next, with reference to FIG. 2, a quality prediction model creation processing according to an embodiment of the present invention will be described.

[0021] 2 is a flowchart showing the flow of a quality prediction model creation process according to one embodiment of the present invention. The flowchart shown in FIG. 2 starts when an execution command for the quality prediction model creation process is input to the quality prediction model creation device 3, and the quality prediction model creation process proceeds to step S1. Note that the operation of the quality prediction model creation device 3 described below is realized by an arithmetic processing device such as a CPU within an information processing device constituting the quality prediction model creation device 3 executing a computer program.

[0022] In the process of step S1, the quality prediction model creation device 3 acquires data of images of the molten metal surface (hereinafter referred to as image data) at predetermined time intervals from the image capture device 2. This completes the process of step S1, and the quality prediction model creation process proceeds to the process of step S2.

[0023] In step S2, the quality prediction model creation device 3 determines the operation mode of the continuous casting machine A in chronological order based on the image data acquired in step S1. Specifically, the quality prediction model creation device 3 determines the operation mode of the continuous casting machine A in chronological order by comparing the molten metal surface image acquired in step S1 with past molten metal surface images for which the operation mode has been identified, or by inputting the molten metal surface image acquired in step S1 into a machine learning model in which the molten metal surface image is an input variable and the operation mode of the continuous casting machine A is an output variable. Examples of operation modes of the continuous casting machine include normal, beginning of casting, end of casting, operator operation, with automatic charging device, without automatic charging device, and others, and can be set arbitrarily depending on the operating mode of the mold. This completes step S2, and the quality prediction model creation process proceeds to step S3.

[0024] In step S3, the quality prediction model creation device 3 performs image processing in time series on the imaging data acquired in step S1. Examples of image processing to be performed include cutting out image data of the mold region in the imaging data, grayscaling, binarizing brightness using a reference brightness threshold, binarizing using a color space value threshold, RGB decomposition, etc. Any image processing may be performed as long as it processes the imaging data and extracts quantitative image data in time series. This completes step S3, and the quality prediction model creation process proceeds to step S4.

[0025] In step S4, the quality prediction model creation device 3 extracts time-series image data for each slab (strand) from the time-series image data obtained in step S3 for each operation mode of the continuous casting machine A determined in step S2. The image data and the slab can be associated with each other by using a tracking technique that uses, for example, the casting speed. This completes step S4, and the quality prediction model creation process proceeds to step S5.

[0026] In step S5, the quality prediction model creation device 3 calculates, for each slab, feature quantities of the time-series image data extracted in step S4. Examples of feature quantities to be calculated include the average brightness value, maximum value, minimum value, standard deviation, slope, and time-series frequency intensity (FFT value of brightness). This process makes it possible to calculate feature quantities of the time-series image data for each operation mode of the continuous casting machine A and for each slab. This completes step S5, and the quality prediction model creation process proceeds to step S6.

[0027] In step S6, the quality prediction model creation device 3 uses the feature values ​​calculated in step S5 and the actual slab quality values ​​at the time the corresponding image data was acquired to create a quality prediction model for each operation mode of the continuous casting machine A, with the feature values ​​of the time-series image data as input variables and the slab quality as output variables. Examples of the predicted slab quality include the incidence rate of defects that will become apparent in subsequent processes, such as slab defects (cracks, fissures, etc.) and inclusion defects. When creating the quality prediction model, the quality prediction model creation device 3 may use operational data of the continuous casting machine A, such as mold temperature and casting speed, in addition to information obtained from the imaging data. The quality prediction model creation device 3 may also change the image processing content and feature values ​​used to create the quality prediction model depending on the prediction accuracy of the created quality prediction model. The quality prediction model creation device 3 may also create a quality prediction model only for a specific operation mode (e.g., normal operation, the presence or absence of an automatic casting machine, etc.). Information regarding the determination result of the operation mode of the continuous casting machine A in step S2 may also be included as an input variable of the quality prediction model. For example, when the operation mode is determined by machine learning or the like, the probability of that operation mode being determined can also be calculated, and the calculated probability can be included in the input variables of the quality prediction model. By including information on the determination result of the operation mode of continuous casting machine A in the input variables of the quality prediction model, the accuracy of quality prediction can be improved. This completes the processing of step S6, and the series of quality prediction model creation processes ends.

[0028] [Quality Control Processing] Next, the quality control processing according to one embodiment of the present invention will be described with reference to FIG.

[0029] 3 is a flowchart showing the flow of quality control processing according to one embodiment of the present invention. The flowchart shown in FIG. 3 starts when an instruction to execute the quality control processing is input to the control device 4, and the quality control processing proceeds to step S11. The operation of the control device 4 shown below is realized by an arithmetic processing unit such as a CPU within the information processing device that constitutes the control device 4 executing a computer program.

[0030] In the process of step S11, the control device 4 acquires data of images of the molten metal surface (hereinafter referred to as image data) at predetermined time intervals from the image capturing device 2. This completes the process of step S11, and the quality control process proceeds to the process of step S12.

[0031] In the process of step S12, the control device 4 determines the operation mode of the continuous casting machine A in chronological order based on the image data acquired in the process of step S11, by the same process as the process of step S2 in the quality prediction model creation process. This completes the process of step S12, and the quality control process proceeds to the process of step S13.

[0032] In the process of step S13, the control device 4 performs image processing on the imaging data acquired in the process of step S11 by the same process as the process of step S3 in the quality prediction model creation process. This completes the process of step S13, and the quality control process proceeds to the process of step S14.

[0033] In the process of step S14, the control device 4 extracts time-series imaging data in units of slabs from the time-series image data obtained by the process of step S13. This completes the process of step S14, and the quality control process proceeds to the process of step S15.

[0034] In the process of step S15, the control device 4 calculates the feature quantities of the time-series image data extracted in the process of step S14. This completes the process of step S15, and the quality control process proceeds to the process of step S16.

[0035] In the processing of step S16, the control device 4 identifies the operation mode of the continuous casting machine A when the time-series image data for which feature quantities have been calculated was acquired using the processing results of step S12, and acquires quality prediction model data corresponding to the identified operation mode of the continuous casting machine A from the quality prediction model creation device 3. The control device 4 then inputs the feature quantities of the time-series image data into the quality prediction model, thereby predicting the quality of the slab corresponding to the time-series image data. This completes the processing of step S16, and the quality control processing proceeds to processing of step S17.

[0036] In step S17, the operator modifies the feature values ​​included in the quality prediction model and the operational status of the continuous casting machine A based on the slab quality predicted in step S16. For example, if it is determined that the standard deviation of the number of pixels with an R (red) value of 200 or greater contributes significantly to the quality prediction, the operator improves the prediction accuracy by adding the standard deviations of the number of pixels with an R (red) value of 205 or greater and 210 or greater to the feature values ​​of the quality prediction model. Furthermore, if there is a large discrepancy between the predicted quality value and the actual quality value, the operator determines that the prediction accuracy of the quality prediction model is low and modifies the image processing and feature values ​​used to create the quality prediction model. Furthermore, if the predicted quality value does not satisfy the required conditions, the operator modifies the operational status of the continuous casting machine A. This completes step S17, and the series of quality control processes ends.

[0037] Example 1 In this example, the defect occurrence rate of a slab was estimated taking into account the operation mode of a continuous casting machine. Figure 4(a) shows the time-series change in the brightness of a molten metal surface image. In this example, the operation mode of a continuous casting machine was determined from the molten metal surface image using a machine learning model that was previously created by machine learning and that classifies input molten metal surface images into molten metal surface images under normal conditions (Figure 4(b)) and molten metal surface images obtained when operator intervention occurred (Figure 4(b)). In addition, a quality prediction model was created using past data, which indicates that if the standard deviation of the brightness in the mold region is equal to or greater than a predetermined value, there has been a significant change in the molten metal surface condition, and the likelihood of an inclusion defect involving powder is high.

[0038] Table 1 shows the average values ​​and standard deviations of luminance when "operator work" is considered (normal operation only) and when it is not (normal operation and operator work included) based on the results of the operation mode determination. As shown in Table 1, when "operator work" is not considered, the standard deviation of luminance is larger than when "operator work" is considered, due to the significant influence of changes in luminance during operator work. This confirms that good quality determination results can be obtained by determining the operation mode of a continuous casting machine and using only the normal operation mode as the quality determination standard (adoption).

[0039]

[0040] Example 2 In this example, a quality prediction model was created taking into account the presence or absence of an automatic charging device, as shown in Figure 5. Because the appearance of molten metal surface images differs significantly between cases with and without an automatic charging device, the same quality prediction model cannot predict slab quality. Therefore, the operating mode of the continuous casting machine was first determined by image classification, and the presence or absence of an automatic charging device was classified. In this example, a machine learning model that had previously trained on images with and without an automatic charging device was used to determine the probability of the image being one of the two. Next, the resulting images were subjected to image processing. The image processing content may be the same regardless of the operating mode, or it may be varied depending on the operating mode. In this example, grayscaling and RGB decomposition (calculation of RGB values ​​(0 to 255) for each pixel) were used to calculate brightness, and the number of pixels with an R (red) value above a threshold (e.g., 200) was calculated to obtain flame brightness above a predetermined value, and the total brightness values ​​of the left and right sides were calculated to use information on the image location. Next, time-series image data was extracted for each slab from the image data obtained by image processing.

[0041] Next, feature quantities of the extracted image data were calculated. In this example, the average value, maximum value, and standard deviation of brightness were aggregated into single-point data to be used as feature quantities. Next, the quality of the slab was predicted using the calculated feature quantities. Here, all of the created feature quantities may be used, or they may be selected based on domain knowledge (know-how). In this example, the defect occurrence rate was predicted based on a multiple regression model using all of the feature quantities. This makes it possible to predict the occurrence of defects based on images, regardless of whether an automatic inserter is used or not. Furthermore, depending on the model used to estimate the quality, it is possible to determine feature quantities that contributed to classifying defect occurrence.

[0042] In the multiple regression model used in this study, the coefficient of each variable is equivalent to the prediction contribution rate. It is also possible to add new features based on these features. In this study, the feature based on the number of pixels with an R (red) value of 200 or greater was found to be effective, so to further improve prediction accuracy, the number of pixels with an R (red) value of 205 or greater and the number of pixels with an R (red) value of 210 or greater were added as features. Furthermore, the construction of the quality estimation model revealed that when brightness is high without an automatic dispenser, the defect rate is high. Therefore, operational action was taken to add more powder when the automatic dispenser was not in use, thereby stabilizing quality.

[0043] Although the present invention has been described above as an embodiment, the present invention is not limited to the descriptions and drawings that form part of the disclosure of the present invention. In other words, other embodiments, examples, and operational techniques that can be made by those skilled in the art based on the present invention are all included in the scope of the present invention.

[0044] According to the present invention, it is possible to provide a slab quality prediction model creation method and a quality prediction model creation device capable of creating a quality prediction model that accurately predicts slab quality, and also to provide a slab quality control method that accurately controls slab quality.

[0045] 1. Quality control system for cast slab 2. Imaging device 3. Quality prediction model creation device 4. Control device A. Continuous casting machine

Claims

1. A method for creating a quality prediction model for predicting the quality of a slab produced by a continuous casting machine, the method comprising: a photographing step of photographing images of the surface of molten metal in a mold of the continuous casting machine; a determination step of determining the operating mode of the continuous casting machine using the images photographed in the photographing step; a cutting step of cutting out time-series image data for each slab by performing image processing on the images photographed in the photographing step; a calculation step of calculating feature quantities of the time-series image data cut out in the cutting step; and a creation step of creating, for each operating mode of the continuous casting machine, a quality prediction model in which the feature quantities of the time-series image data are input variables and the quality of the slab corresponding to the time-series image data is an output variable, using the operating mode of the continuous casting machine when the time-series image data cut out in the cutting step was obtained, the feature quantities of the time-series image data, and actual values ​​of the quality of the slab corresponding to the time-series image data.

2. The method for creating a quality prediction model for a slab according to claim 1, wherein the operation mode of the continuous casting machine is determined according to the operating mode of the mold.

3. A method for creating a quality prediction model of a slab as described in claim 1 or 2, further comprising a step of changing the feature quantities to be included in the quality prediction model created in the creation step according to the contribution rate of each feature quantity included in the quality prediction model to the quality prediction of the slab.

4. A method for creating a quality prediction model for a slab as set forth in any one of claims 1 to 3, wherein the input variables of the quality prediction model include information relating to the determination result of the operating mode of the continuous casting machine in the determination step.

5. A cast slab quality prediction model creation device that creates a quality prediction model that predicts the quality of a cast slab produced by a continuous casting machine, comprising: a determination means for determining the operation mode of the continuous casting machine using images of the surface of molten metal in a mold of the continuous casting machine; an extraction means for extracting time-series image data for each cast slab by performing image processing on the images; a calculation means for calculating feature quantities of the time-series image data extracted by the extraction means; and a creation step for creating, for each operation mode of the continuous casting machine, a quality prediction model in which the feature quantities of the time-series image data are input variables and the quality of the cast slab corresponding to the time-series image data are output variables, using the operation mode of the continuous casting machine when the time-series image data extracted by the extraction means was obtained, the feature quantities of the time-series image data, and actual values ​​of the quality of the cast slab corresponding to the time-series image data.

6. A method for controlling the quality of a slab using a quality prediction model created by the method for creating a slab quality prediction model as set forth in claim 1, comprising: a determination step of determining the operating mode of the continuous casting machine using an image of the surface of the molten metal in the mold of the continuous casting machine; a prediction step of predicting the quality of the slab by inputting feature quantities of time-series image data obtained from the image into a quality prediction model corresponding to the operating mode of the continuous casting machine determined in the determination step; and a step of controlling the operating conditions of the continuous casting machine based on the quality of the slab predicted in the prediction step.

Images