Steel refining control system and method based on machine vision

Abstract

The invention provides a steel refining control system and method based on machine vision, and the system comprises a refining furnace, and is characterized in that the system also comprises an image collection module, an analysis control module and a judgment module which run on a computer and are connected with the refining control system, and the image collection module, the analysis control module and the judgment module are mutually connected; the refining furnace is used for refining molten steel in the furnace, and refining information in the refining process is in electric connection or / and signal connection with the refining furnace control system, so that data interaction is realized; the image acquisition module is used for acquiring refining furnace mouth flame image information; the analysis control module is used for calculating and analyzing the flame image information acquired by the image acquisition module; and the judgment module is used for accurately and effectively judging the carbon content and the temperature of the molten steel tapping end point of the refining furnace, and finally controlling the refining furnace to refine qualified molten steel and discharge the molten steel, so that cost increase and resource waste caused by remelting due to misjudgment are avoided.

Description

Technical Field

[0001] This invention relates to the field of machine vision, and more particularly to a machine vision-based steel refining control system and method. Background Technology

[0002] Endpoint control is a crucial operational technique in the final stage of argon-oxygen refining of low-carbon ferrochrome alloys. Its purpose is to control the carbon content and temperature of the molten iron at the refining furnace mouth, ensuring that both simultaneously meet the tapping requirements. Currently, the argon-oxygen refining process for low-carbon ferrochrome alloys in my country commonly employs an empirical method, namely the manual flame observation method. This method is highly random, heavily reliant on the skill level of the smelters, and results in inconsistent accuracy in determining the carbon content and temperature of the molten iron at the endpoint. This often leads to remelting, increasing production costs and wasting resources.

[0003] A review of publicly available technical solutions, such as CN101845531B, entitled "A Control System and Method for Carbon and Temperature of Molten Steel at the End of Converter Smelting," reveals that this system includes hardware devices and a software processing module. The hardware devices include a converter auxiliary lance measuring device and a converter exhaust gas analysis device. The software processing module runs on a computer and includes a converter production process data acquisition module, a converter smelting end-point carbon content calculation module, a converter smelting end-point temperature calculation module, and an information display module. The hardware devices and software modules are connected to a converter production process database via Ethernet to achieve data interaction. This invention can control the carbon content and temperature of molten steel at the end of converter smelting. For example, the invention disclosed in CN212688117U, entitled "A Dynamic Control System for End-Point Carbon in the Entire Converter Smelting Process Based on Gas Analysis," uses a gas detection probe at the converter site to collect and analyze the mole fraction of gases in the air before each smelting process, ensuring the accuracy of the basic gas analysis data. It constructs a dynamic model for decarburization through gas analysis throughout the converter smelting process and a dynamic end-point carbon time prediction model. The constructed system and dynamic mathematical model overcome the shortcomings of existing technologies, enabling dynamic control of end-point carbon in the entire converter smelting process. It is widely applicable to various types of newly built, expanded, or renovated converter steelmaking plants. However, the above solutions are all characterized by high cost, large investment, and limited applicability to small and medium-sized smelting equipment. Therefore, improving the accuracy of iron tapping end-point prediction while reducing production costs and establishing a system and method universally applicable to various types of smelting equipment has become an urgent problem to be solved. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of current methods by proposing a machine vision-based steel refining control system and method.

[0005] The present invention provides a machine vision-based steel refining control system, including a refining furnace, characterized in that: it further includes an image acquisition module, an analysis and control module, and a judgment module running on a computer and connected to the refining furnace control system, and the image acquisition module, the analysis and control module, and the judgment module are interconnected. The refining furnace is used to refine the molten steel in the furnace, and the refining information during the refining process is electrically connected or / and signal connected to the refining furnace control system to realize data interaction.

[0006] The image acquisition module is used to acquire image information of the flame at the furnace opening of the refining furnace, and includes: an optical image sensor camera, an image acquisition card, and a buffer register; wherein:

[0007] An optical image sensor camera is used to capture the flame at the furnace opening of a refining furnace and convert the acquired flame into first image information;

[0008] The image acquisition card is used to process the acquired and converted first image information to obtain the second image information;

[0009] The buffer register is used to store the second image information;

[0010] The analysis and control module is used to calculate and analyze the second image information acquired and processed by the image acquisition module. It includes: an image preprocessing unit, a feature extraction unit, and a similarity calculation unit; wherein:

[0011] The image preprocessing unit is used to perform noise reduction and standardization preprocessing on the second image information acquired and processed by the image acquisition module;

[0012] The feature extraction unit is used to extract information feature data of image pixels, including brightness and brightness variation, contrast and contrast variation, and structure and structure variation, from the preprocessed second image information.

[0013] The similarity calculation unit is used to calculate the similarity by using a weighted fusion algorithm on the extracted information feature data and then comparing it with the template image information obtained from historical data experience template images.

[0014] The judgment module is used to compare the similarity obtained by the analysis and control module with the set threshold. After the comparison is completed, the end point of the molten steel tapping from the refining furnace is judged so as to control the refining furnace to refine qualified molten steel and tap it out.

[0015] The refining furnace is a conventional steelmaking equipment, along with a conventional refining monitoring device and a conventional refining furnace control system.

[0016] The similarity of the analysis and control modules is calculated using the following formula:

[0017] S=[α1×a1(x,y)]+[α2×a2(x,y)]+[β1×b1(x,y)]+[β2×b2(x,y)]+[γ1×c1(x,y)]+[γ2×c2(x,y)];

[0018] In the formula, S represents the similarity between the second image information and the template image information; a1, α2, β1, β2, γ1, and γ2 are weighting factors; a1(x, y) is the brightness similarity function; a2(x, y) is the brightness change similarity function; b1(x, y) is the contrast similarity function; b2(x, y) is the contrast change similarity function; c1(x, y) is the structural similarity function; and c2(x, y) is the structural change similarity function. The formulas for each function are as follows:

[0019]

[0020] In the formula, u x u is the average pixel value of the second image. y k is the average pixel value of the template image. x k is the average pixel change in the second image. y t is the average pixel change of the template image; x t is the standard deviation of the second image pixels; y r is the standard deviation of the template image pixels; x r is the standard deviation of the pixel variation in the second image. y t is the standard deviation of the pixel variation in the template image; xy r is the covariance of the pixels in the second image and the template image. xy Let c be the covariance of the changes in the second image and the template image; c1, c2, c3, c4, c5, and c6 are stability coefficients to avoid zero denominators in the corresponding formulas, and their specific calculation formulas are as follows:

[0021]

[0022]

[0023] In the formula, N is the total number of pixels in the second image and the template image, and x i (1≤i≤N) represents the pixel values ​​of each point in the second image arranged in order, y i (1≤i≤N) represents the pixel values ​​of each point in the template image arranged in order.

[0024] The template image information feature data and the second image information feature data obtained from the historical data experience template image are: the brightness, contrast, and structure of the flame image pixels, as well as the changes in brightness, contrast, and structure of the flame image pixels.

[0025] The present invention provides a method for steel refining based on a machine vision-based steel refining control system, comprising the following steps:

[0026] S1: At the end of the refining process in the refining furnace, the flame at the furnace opening is captured by the optical image sensor camera in the image acquisition module. After the captured flame is converted into first image information, it is sent to the image acquisition card for sampling and preprocessing to obtain second image information. The second image information is then stored in the buffer register.

[0027] S2: The second image information from step S1 is preprocessed by the image preprocessing unit in the analysis control module to perform denoising and standardization preprocessing.

[0028] S3: By analyzing the feature extraction unit in the control module, extract the brightness, contrast and structure of the image pixels, as well as the feature data of the brightness change, contrast change and structure change of the image pixels from the second image information after preprocessing in step S2.

[0029] S4: By analyzing the similarity calculation unit in the control module, the feature data in the second image information of step S3 is weighted and fused with the following functions from the template image information obtained from historical data experience template images: pixel brightness similarity function and brightness change similarity function, contrast similarity function and contrast change similarity function, structural similarity function and structural change similarity function. The calculation formulas for each function are as follows:

[0030]

[0031]

[0032] In the formula, u x u is the average pixel value of the second image. y k is the average pixel value of the template image. x k is the average pixel change in the second image. y t is the average pixel change of the template image; x t is the standard deviation of the second image pixels; y r is the standard deviation of the template image pixels; x r is the standard deviation of the pixel variation in the second image. y t is the standard deviation of the pixel variation in the template image; xy r is the covariance of the pixels in the second image and the template image. xy Let c be the covariance of pixel changes between the second image and the template image; c1, c2, c3, c4, c5, and c6 are stability coefficients to avoid zero denominators in the corresponding formulas; the specific calculation formulas are as follows:

[0033]

[0034] In the formula, N is the total number of pixels in the second image and the template image, and x i (1≤i≤N) represents the pixel values ​​of each point in the second image arranged in order, y i (1≤i≤N) represents the pixel values ​​of each point in the template image arranged in order;

[0035] S5: Based on the function values ​​in step S4, calculate the similarity between the second image and the template image using the following formula;

[0036] S=[α1×a1(x,y)]+[α2×a2(x,y)]+[β1×b1(x,y)]+[β2×b2(x,y)]+[γ1×c1(x,y)]+[γ2×c2(x,y)];

[0037] In the formula, S is the similarity between the second image information and the template image information, α1, α2, β1, β2, γ1 and γ2 are weighting factors, α1(x,y) is the brightness similarity function, α2(x,y) is the brightness change similarity function, b1(x,y) is the contrast similarity function, b2(x,y) is the contrast change similarity function, c1(x,y) is the structural similarity function, and c2(x,y) is the structural change similarity function.

[0038] S6: After comparing the similarity of step S5 with the threshold, the endpoint of the molten steel tapping from the refining furnace is determined in order to control the refining furnace to produce qualified molten steel and tap it out.

[0039] The final refining stage in step S1 refers to the period after refining in the refining furnace, when top and / or bottom blowing is carried out for twenty minutes, which is the final refining stage.

[0040] The threshold in step S6 is a dynamic reference value obtained through learning and optimization from historical data, and it is determined through the following steps:

[0041] S61) Template Image Selection and Parameter Setting: Five flame images at the optimal tapping endpoint of the corresponding steel grade successfully refined from the system's historical data were selected as template images. All relevant parameters were set based on conventional historical data learning and experimental optimization, including:

[0042] The weighting factors are set as follows:

[0043] α1=0.2, α2=0.15, β1=0.15, β2=0.1, γ1=0.25, γ2=0.15;

[0044] The stability coefficient is set as follows:

[0045] c1 = c2 = (0.01 × L) 2c3 = c4 = (0.03 × L) 2 c5 = c6 = c3 / 2

[0046] In the formula: L = 255;

[0047] S62) Calculate the similarity of the template images themselves: Calculate the similarity between the five template images in step S61) to form a similarity set value, which is used to represent the degree of similarity between the final state images;

[0048] S63) Statistical analysis: Calculate the mean and variance of the similarity set values ​​from step S62), or estimate them using confidence intervals to obtain a threshold. This threshold can be set as the average value of the similarity set values ​​minus 2-3 times the standard deviation, or a conservative value that can guarantee a high hit rate can be selected based on experimental data, such as the lower 5% quantile of the similarity set.

[0049] S64) Verification and Adjustment: Apply the threshold from step S63) to another set of known historical data for verification and fine-tuning, ultimately determining an optimized threshold that ensures both accuracy and a certain degree of fault tolerance: S thresold The optimized threshold S threshold It can be pre-stored in the system's judgment module and supports empirical fine-tuning by engineers based on different steel grades, furnace conditions, or production stages.

[0050] The determination of the endpoint of molten steel tapping from the refining furnace in step S6 is as follows:

[0051] When the similarity is greater than or equal to the preset optimization threshold, that is: S ≥ S threshold If the current furnace flame image is found to be highly similar to the optimal tapping endpoint template image, indicating that the tapping endpoint has been reached, the judgment module will send a tapping command to the refining furnace control system to execute the tapping operation or proceed to the next process.

[0052] When the similarity is less than the optimization threshold, i.e., S < S threshold If the current flame state at the furnace opening is determined to be not yet in line with the optimal iron tapping endpoint template image requirements, i.e., the steel tapping endpoint has not been reached, the judgment module sends a command to the refining furnace control system to stop tapping steel and continue refining. The analysis and control module continues to calculate and analyze the next frame of flame image information acquired by the image acquisition module. This process continues until the steel tapping endpoint requirements are met, at which point the steel tapping operation is performed or the process moves to the next step.

[0053] The beneficial effects achieved by this invention are as follows: the refining state inside the refining furnace can be easily understood through the flame image of the furnace opening. Specifically, the similarity between the two images is obtained by comparing the collected flame image information of the refining furnace opening with the template image information features obtained from historical data experience template images. The image information features include the brightness, contrast, and structure of the flame image pixels, as well as the changes in brightness, contrast, and structure of the flame image pixels. By accurately judging the refining state inside the refining furnace through these image information features, the steel tapping endpoint of the refining furnace can be accurately and effectively judged, and the refining furnace can be controlled to produce qualified molten steel and tap it out of the furnace, avoiding the increased cost and waste of resources caused by remelting due to misjudgment. Attached Figure Description

[0054] Figure 1 This is a schematic diagram of the system of the present invention;

[0055] Figure 2 This is a diagram showing the module structure of the system;

[0056] Figure 3 This is a flowchart of the present invention. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of this invention clearer and more complete, the invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. Other systems, methods, and / or features of this embodiment will become apparent to those skilled in the art after reviewing the following detailed description; all such additional systems, methods, features, and advantages are intended to be included within this specification; included within the scope of the invention, and protected by the appended claims; further features of the disclosed embodiments are described in the following detailed description, and these features will become apparent from the following detailed description.

[0058] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or component referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present patent. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0059] Example 1

[0060] like Figure 1 , Figure 2 As shown, the steel refining control system based on machine vision provided in this embodiment 1 includes a refining furnace, and also includes an image acquisition module, an analysis and control module, and a judgment module running on a computer and connected to the refining furnace control system. The image acquisition module, the analysis and control module, and the judgment module are interconnected. The refining furnace is used to refine the molten steel in the furnace, and the refining information during the refining process is electrically connected or / and signal connected to the refining furnace control system to realize data interaction.

[0061] The image acquisition module is used to acquire image information of the flame at the furnace opening of the refining furnace, and includes: an optical image sensor camera, an image acquisition card, and a buffer register; wherein:

[0062] An optical image sensor camera is used to capture the flame at the furnace opening of a refining furnace and convert the acquired flame into first image information;

[0063] The image acquisition card is used to process the acquired and converted first image information to obtain the second image information;

[0064] The buffer register is used to store the second image information;

[0065] The analysis and control module is used to calculate and analyze the second image information acquired and processed by the image acquisition module. It includes: an image preprocessing unit, a feature extraction unit, and a similarity calculation unit; wherein:

[0066] The image preprocessing unit is used to perform noise reduction and standardization preprocessing on the second image information acquired and processed by the image acquisition module;

[0067] The feature extraction unit is used to extract information feature data of image pixels, including brightness and brightness variation, contrast and contrast variation, and structure and structure variation, from the preprocessed second image information.

[0068] The similarity calculation unit is used to calculate the similarity by using a weighted fusion algorithm on the extracted information feature data and then comparing it with the template image information obtained from historical data experience template images.

[0069] The judgment module is used to compare the similarity obtained by the analysis and control module with the set threshold. After the comparison is completed, the end point of the molten steel tapping from the refining furnace is judged so as to control the refining furnace to refine qualified molten steel and tap it out.

[0070] The refining furnace is a conventional steelmaking equipment, along with a conventional refining monitoring device and a conventional refining furnace control system.

[0071] The similarity of the analysis and control modules is calculated using the following formula:

[0072] S=[α1×a1(x,y)]+[α2×a2(x,y)]+[β1×b1(x,y)]+[β2×b2(x,y)]+[γ1×c1(x,y)]+[γ2×c2(x,y)];

[0073] In the formula, S represents the similarity between the second image information and the template image information; α1, α2, β1, β2, γ1, and γ2 are weighting factors; a1(x, y) is the brightness similarity function; α2(x, y) is the brightness change similarity function; b1(x, y) is the contrast similarity function; b2(x, y) is the contrast change similarity function; c1(x, y) is the structural similarity function; and c2(x, y) is the structural change similarity function. The formulas for each function are as follows:

[0074]

[0075] In the formula, u x u is the average pixel value of the second image. y k is the average pixel value of the template image. x k is the average pixel change in the second image. y t is the average pixel change of the template image; x t is the standard deviation of the second image pixels; y r is the standard deviation of the template image pixels; x r is the standard deviation of the pixel variation in the second image. y t is the standard deviation of the pixel variation in the template image; xy r is the covariance of the pixels in the second image and the template image. xy Let c be the covariance of the changes in the second image and the template image; c1, c2, c3, c4, c5, and c6 are stability coefficients to avoid zero denominators in the corresponding formulas, and their specific calculation formulas are as follows:

[0076]

[0077]

[0078] In the formula, N is the total number of pixels in the second image and the template image, and x i (1≤i≤N) represents the pixel values ​​of each point in the second image arranged in order, y i (1≤i≤N) represents the pixel values ​​of each point in the template image arranged in order.

[0079] The template image information feature data and the second image information feature data obtained from the historical data experience template image are: the brightness, contrast, and structure of the flame image pixels, as well as the changes in brightness, contrast, and structure of the flame image pixels.

[0080] Example 2

[0081] The present invention provides a method for steel refining based on the machine vision-based steel refining control system, comprising the following steps:

[0082] S1: At the end of the refining process in the refining furnace, the flame at the furnace opening is captured by the optical image sensor camera in the image acquisition module. After the captured flame is converted into first image information, it is sent to the image acquisition card for sampling and preprocessing to form second image information. The second image information is then stored in the buffer register.

[0083] S2: The second image information from step S1 is preprocessed by the image preprocessing unit in the analysis control module to perform denoising and standardization preprocessing.

[0084] S3: By analyzing the feature extraction unit in the control module, extract the feature data of image pixels, including brightness and brightness change, contrast and contrast change, and structure and structure change, from the second image information after preprocessing in step S2.

[0085] S4: By analyzing the similarity calculation unit in the control module, the feature data from step S3 is used to calculate the second image information and the template image information obtained from historical data experience template images, including: pixel brightness similarity function and brightness change similarity function, contrast similarity function and contrast change similarity function, structural similarity function and structural change similarity function; the calculation formulas for each function are as follows:

[0086]

[0087] In the formula, u x u is the average pixel value of the second image. y k is the average pixel value of the template image. x κ is the average pixel change in the second image. y t is the average pixel change of the template image; x t is the standard deviation of the second image pixels; y r is the standard deviation of the template image pixels; x r is the standard deviation of the pixel variation in the second image. y t is the standard deviation of the pixel variation in the template image; xy r is the covariance of the pixels in the second image and the template image. xy Let c be the covariance of pixel changes between the second image and the template image; c1, c2, c3, c4, c5, and c6 are stability coefficients to avoid zero denominators in the corresponding formulas; the specific calculation formulas are as follows:

[0088]

[0089]

[0090] In the formula, N is the total number of pixels in the second image and the template image, and x i (1≤i ≤ N) represents the pixel values ​​of each point in the second image arranged in order, y i (1≤i≤N) represents the pixel values ​​of each point in the template image arranged in order;

[0091] S5: Based on the function values ​​in step S4, calculate the similarity between the second image and the template image using the following formula: S=[α1×a1(x,y)]+[α2×a2(x,y)]+[β1×b1(x,y)]+[β2×b2(x,y)]+[γ1×c1(x,y)]+[γ2×c2(x,y)];

[0092] In the formula, S is the similarity between the second image information and the template image information, α1, α2, β1, β2, γ1 and γ2 are weighting factors, α1(x,y) is the brightness similarity function, a2(x,y) is the brightness change similarity function, b1(x,y) is the contrast similarity function, b2(x,y) is the contrast change similarity function, c1(x,y) is the structural similarity function, and c2(x,y) is the structural change similarity function.

[0093] S6: After comparing the similarity of step S5 with the threshold, the endpoint of the molten steel tapping from the refining furnace is determined in order to control the refining furnace to produce qualified molten steel and tap it out.

[0094] The threshold in step S6 is a dynamic reference value obtained through learning and optimization from historical data, and it is determined through the following steps:

[0095] S61) Template Image Selection and Parameter Setting: Five flame images at the optimal tapping endpoint of the corresponding steel grade successfully refined from the system's historical data were selected as template images. All relevant parameters were set based on conventional historical data learning and experimental optimization, including:

[0096] The weighting factors are set as follows:

[0097] α1=0.2, α2=0.15, β1=0.15, β2=0.1, γ1=0.25, γ2=0.15;

[0098] The stability coefficient is set as follows:

[0099] c1 = c2 = (0.01 × L) 2 c3 = c4 = (0.03 × L) 2 c5 = c6 = c3 / 2

[0100] In the formula: L = 255;

[0101] S62) Calculate the similarity of the template images themselves: Calculate the similarity between the five template images in step S61) to form a similarity set value, which is used to represent the degree of similarity between the final state images;

[0102] S63) Statistical analysis: Calculate the mean and variance of the similarity set values ​​from step S62), or estimate them using confidence intervals to obtain a threshold. This threshold can be set as the average value of the similarity set values ​​minus 2-3 times the standard deviation, or a conservative value that can guarantee a high hit rate can be selected based on experimental data, such as the lower 5% quantile of the similarity set.

[0103] S64) Verification and Adjustment: Apply the threshold from step S63) to another set of known historical data for verification and fine-tuning, ultimately determining an optimized threshold that ensures both accuracy and a certain degree of fault tolerance: S threshold The optimized threshold S threshold It can be pre-stored in the system's judgment module and supports empirical fine-tuning by engineers based on different steel grades, furnace conditions, or production stages.

[0104] The determination of the endpoint of molten steel tapping from the refining furnace in step S6 is as follows:

[0105] When the similarity is greater than or equal to the preset optimization threshold, that is: S ≥ S threshold If the current furnace flame image is found to be highly similar to the optimal tapping endpoint template image, indicating that the tapping endpoint has been reached, the judgment module will send a tapping command to the refining furnace control system to execute the tapping operation or proceed to the next process.

[0106] When the similarity is less than the optimization threshold, i.e., S < S threshold If the current flame state at the furnace opening is determined to be not yet in line with the optimal iron tapping endpoint template image requirements, i.e., the steel tapping endpoint has not been reached, the judgment module sends a command to the refining furnace control system to stop tapping steel and continue refining. The analysis and control module continues to calculate and analyze the next frame of flame image information acquired by the image acquisition module. This process continues until the steel tapping endpoint requirements are met, at which point the steel tapping operation is performed or the process moves to the next step.

[0107] Example 3

[0108] This embodiment 3 uses a 120-ton refining furnace at the Kunming Iron & Steel Co., Ltd. of Wuhan Iron and Steel Group as an application scenario to describe a complete smelting cycle of refining SPHC steel. The invention is explained in detail below, including the following steps:

[0109] Step (1) System presets and initialization:

[0110] Template Image Selection: From the historical data of the refining furnace control system, select the flame images at the five best tapping endpoints of the successfully refined SPHC steel as template images.

[0111] Threshold setting: Based on historical data learning, the system sets the similarity threshold for this SPHC steel grade as: S threshold =0.85;

[0112] The weighting factors were optimized and set as follows based on experiments:

[0113] α1=0.2, α2=0.15, β1=0.15, β2=0.1, γ1=0.25, γ2=0.15;

[0114] The stability coefficient is set as follows:

[0115] c1 = c2 = (0.01 × L) 2 c3 = c4 = (0.03 × L) 2 c5 = c6 = c3 / 2

[0116] Where: L = 255 (the maximum value of an 8-bit pixel);

[0117] Step (2) Image acquisition and preprocessing

[0118] At the end of the refining process in the refining furnace, that is, after the refining process in the refining furnace, when the top and bottom combined blowing is carried out for 20 minutes, the optical image sensor camera in the image acquisition module acquires the flame at the furnace mouth of the refining furnace at a frequency of 1 frame per second. After the acquired flame optical signal is converted into the first image information, it is sent to the image acquisition card to sample one frame of image (i.e. the first image information), perform Gaussian filtering and noise reduction preprocessing, and then uniformly scale it to 640x480 pixels to form the second image information (i.e. the current image). The second image information is then stored in the buffer register, and its pixel values ​​are shown in the table below.

[0119] pixel point sequence number i ) Template image Pixel value ( y_i )]]> <![CDATA[Second Image Pixel value ( x_i )]]> 1 200 195 2 210 215 3 190 198 4 205 208

[0120] Step (3) Image information feature data extraction and similarity calculation

[0121] By analyzing the feature extraction unit in the control module, image pixel values ​​are extracted from the template image and the second image respectively, and the following calculations are performed:

[0122] (a) Calculate the mean pixel value between the template image and the second image: u x u y :

[0123] u y= (200 + 210 + 190 + 205) / 4 = 201.25

[0124] u x = (195 + 215 + 198 + 208) / 4 = 204.0

[0125] Calculate pixel change (one-section difference):

[0126] Template image pixel changes: (210-200)=10, (190-210)=-20, (205-190)=15

[0127] Second image pixel change: (215-195)=20, (198-215)=-17, (208-198)=10

[0128] Calculate the mean pixel change of two images: k x k y :

[0129] k y = (10-20+15) / 3≈1.667

[0130] k x = (20-17+10) / 3≈4.333

[0131] Calculate the standard deviation of pixels in two images: t x , t y :

[0132]

[0133] Calculate the standard deviation of pixel variation in two images: r x r y :

[0134]

[0135] Calculate the covariance of pixels in two images: t xy r xy :

[0136] t xy = [(195-204)(200-201.25)+(215-204)(210-201.25)+(198-204)(190-201.25)+(208-204)(205-201.25)] / 4≈67.75

[0137] r xy= [(20-4.333)(10-1.667)+(-17-4.333)(-20-1.667)+(10-4.333)(15-1.667)] / 3≈242.89

[0138] (b) Calculate the values ​​of each function

[0139] Brightness similarity function:

[0140]

[0141] Brightness change similarity function:

[0142]

[0143] Contrast similarity function:

[0144]

[0145] Contrast change similarity function:

[0146]

[0147] Structural similarity function:

[0148]

[0149] Structural change similarity function:

[0150]

[0151] (c) Calculate the similarity S between the two images.

[0152] S=(0.2*0.999)+(0.15*0.742)+(0.15*0.997)+(0.1*0.993)+(0.25*0.997)+(0.15*0.987)

[0153] S=0.1998+0.1113+0.1496+0.0993+0.2493+0.1481

[0154] S≈0.957

[0155] Step (4) Endpoint Judgment and Execution

[0156] Based on the similarity calculated in step (3) and the set threshold, the following judgments and actions are performed after comparison by the judgment module:

[0157] Judgment: The similarity S≈0.957 is compared with the preset optimization threshold S. threshold After comparing with 0.85,

[0158] Because the similarity is greater than or equal to the preset optimization threshold, i.e., 0.957 ≥ 0.85, it is determined that the current furnace flame image is highly similar to the optimal steel tapping endpoint template image, and the steel tapping endpoint has been reached. The judgment module sends a "steel tapping is possible" command to the refining furnace control system to execute the steel tapping operation, thereby completing this refining process.

[0159] Example 4

[0160] This embodiment 4 uses a 120-ton refining furnace at the Kunming Iron & Steel Co., Ltd. of Wuhan Iron and Steel Group as an application scenario to describe a complete smelting cycle of refining SPHC steel. The invention is explained in detail below, including the following steps:

[0161] Step (1) System presets and initialization:

[0162] Template Image Selection: From the historical data of the refining furnace control system, select the flame images at the five best tapping endpoints of the successfully refined SPHC steel as template images.

[0163] Threshold setting: Based on historical data learning, the system sets the similarity threshold for this SPHC steel grade as: S threshold =0.85;

[0164] The weighting factors were optimized and set as follows based on experiments:

[0165] α1=0.2, α2=0.15, β1=0.15, β2=0.1, γ1=0.75, γ2=0.15;

[0166] The stability coefficient is set as follows:

[0167] c1 = c2 = (0.01 × L) 2 c3 = c4 = (0.03 × L) 2 c5 = c6 = c3 / 2

[0168] Where: L = 255 (the maximum value of an 8-bit pixel);

[0169] Step (2) Image acquisition and preprocessing

[0170] At the end of the refining process in the refining furnace, that is, after the refining process in the refining furnace, when the top and bottom combined blowing is carried out for 20 minutes, the flame at the furnace mouth of the refining furnace is captured by the optical image sensor camera in the image acquisition module at a frequency of 1 frame per second. After the acquired flame optical signal is converted into the first image information, it is sent to the image acquisition card to sample one frame of image (i.e. the first image information), perform Gaussian filtering and noise reduction preprocessing, and then uniformly scale it to 640x480 pixels to form the second image information (i.e. the current image). The second image information data is stored in the buffer register, and its pixel values ​​are shown in the table below.

[0171] <![CDATA[Pixel point serial number ( i )]]> <![CDATA[Template Image Pixel value ( y_i )]]> <![CDATA[Second image Pixels Value( x_i )]]> 1 210 170 2 175 160 3 190 180 4 185 175

[0172] Step (3) Image information feature data extraction and similarity calculation

[0173] By analyzing the feature extraction unit in the control module, image pixel values ​​are extracted from the template image and the second image respectively, and the following calculations are performed:

[0174] (a) Calculate the mean pixel value between the template image and the second image: u x u y :

[0175] u y = (210 + 175 + 190 + 185)⁴ = 190

[0176] u x = (170 + 160 + 180 + 175) / 4 = 171.25

[0177] Calculate pixel change (one-section difference):

[0178] Template image pixel changes:

[0179] (175-210)=-35, (190-175)=15, (185-190)=-5

[0180] The pixel change in the second image: (160-170) = -10, (180-160) = 20, (175-180) = -5. Calculate the average pixel change of the two images: k x k y :

[0181] k y = (-35 + 15 - 5) / 3 ≈ -8.33

[0182] k x = (-10 + 20 - 5) / 3 ≈ 1.67

[0183] Calculate the standard deviation of pixels in two images: t x , ty :

[0184]

[0185] Calculate the standard deviation of pixel variation in two images: r x r y :

[0186]

[0187] Calculate the covariance of pixels in two images: t xy r xy :

[0188] t xy = [(170-171.25)(210-190)+(160-171.25)(175-190)+(180-171.25)(190-190)+(175-171.25)(185-190)] / 4≈31.25

[0189] r xy = [(-10-1.67)(-35+8.33)+(20-1.67)(15+8.33)+(-5-1.67)(-5+8.33)] / 3≈283.89

[0190] (b) Calculate the values ​​of each function

[0191] Brightness similarity function:

[0192]

[0193] Brightness change similarity function:

[0194]

[0195] Contrast similarity function:

[0196]

[0197] Contrast change similarity function:

[0198]

[0199] Structural similarity function:

[0200]

[0201] Structural change similarity function:

[0202]

[0203] (c) Calculate the similarity S between the two images.

[0204] S=(0.2*0.9946)+(0.15*0.271)+(0.15*0.9996)+(0.1*0.9992)+(0.25*0.4889)+(0.15*1.0478)

[0205] S=0.19892+0.040651+0.14994+0.09992+0.122225+0.15717

[0206] S≈0.7888

[0207] Step (4) Endpoint Judgment and Execution

[0208] Based on the similarity calculated in step (3) and the set threshold, the following judgments and actions are performed after comparison by the judgment module:

[0209] Judgment: The similarity S≈0.7888 is compared with the preset optimization threshold S. threshold =0.85 After comparison, since the obtained similarity is less than the preset optimization threshold, it is determined that the current furnace flame state has not reached the optimal iron tapping endpoint template image requirement, that is: the steel tapping endpoint has not been reached. The judgment module sends a steel tapping command to the refining furnace control system to continue refining. The analysis and control module continues to extract image information feature data and calculate similarity in step (3) and perform endpoint judgment and execution in step (4) on the next frame flame image information collected by the image acquisition module. Finally, after the steel tapping endpoint requirement is reached, the steel tapping operation is performed to complete this refining process.

[0210] The content disclosed above is only a preferred and feasible embodiment of the present invention, and is not intended to limit the scope of protection of the present invention. Therefore, all equivalent technical changes made based on the content of the present invention specification and drawings are included within the scope of protection of the present invention. Furthermore, the elements therein can be updated as technology develops.

Claims

1. A machine vision-based steel refining control system, comprising a refining furnace, characterized in that: It also includes an image acquisition module, an analysis and control module, and a judgment module that run on a computer and are connected to the refining furnace control system. The image acquisition module, the analysis and control module, and the judgment module are interconnected. The refining furnace is used to refine the molten steel in the furnace. The refining information during the refining process is electrically connected or / and signal connected to the refining furnace control system to realize data interaction. The image acquisition module is used to acquire image information of the flame at the furnace opening of the refining furnace, and includes: an optical image sensor camera, an image acquisition card, and a buffer register; wherein: An optical image sensor camera is used to capture the flame at the furnace opening of a refining furnace and convert the acquired flame into first image information; The image acquisition card is used to process the acquired and converted first image information to obtain the second image information; The buffer register is used to store the second image information; The analysis and control module is used to calculate and analyze the second image information acquired and processed by the image acquisition module. It includes: an image preprocessing unit, a feature extraction unit, and a similarity calculation unit; wherein: The image preprocessing unit is used to perform noise reduction and standardization preprocessing on the second image information acquired and processed by the image acquisition module; The feature extraction unit is used to extract information feature data of image pixels, such as brightness and brightness variation, contrast and contrast variation, and structure and structure variation, from the preprocessed second image information. The similarity calculation unit is used to calculate the similarity by using a weighted fusion algorithm on the extracted information feature data and then comparing it with the template image information obtained from historical data experience template images. The judgment module is used to compare the similarity obtained by the analysis and control module with the set threshold. After the comparison is completed, the end point of the molten steel tapping from the refining furnace is judged so as to control the refining furnace to refine qualified molten steel and tap it out.

2. The steel refining control system based on machine vision according to claim 1, characterized in that... The similarity of the analysis and control modules is calculated using the following formula: S=[α1×α1(x,y)]+[α2×α2(x,y)]+[β1×b1(x,y)]+[β2×b2(x,y)]+[γ1×c1(x,y)]+[γ2×c2(x,y)]; In the formula, S represents the similarity between the second image information and the template image information; α1, α2, β1, β2, γ1, and γ2 are weighting factors; a1(x, y) is the brightness similarity function; a2(x, y) is the brightness change similarity function; b1(x, y) is the contrast similarity function; b2(x, y) is the contrast change similarity function; c1(x, y) is the structural similarity function; and c2(x, y) is the structural change similarity function. The formulas for each function are as follows: In the formula, u x u is the average pixel value of the second image. y k is the average pixel value of the template image. x k is the average pixel change in the second image. y t is the average pixel change of the template image; x k is the standard deviation of the second image pixels; y r is the standard deviation of the template image pixels; x r is the standard deviation of the pixel variation in the second image. y t is the standard deviation of the pixel variation in the template image; xy r is the covariance of the pixels in the second image and the template image. xy Let c be the covariance of pixel changes between the second image and the template image; c1, c2, c3, c4, c5, and c6 are stability coefficients to avoid zero denominators in the corresponding formulas; the specific calculation formulas are as follows: In the formula, N is the total number of pixels in the second image and the template image, and x i (1≤i≤N) represents the pixel values ​​of each point in the second image arranged in order, y i (1≤i≤N) represents the pixel values ​​of each point in the template image arranged in order.

3. The steel refining control system based on machine vision according to claim 2, characterized in that, The template image information feature data and the second image information feature data obtained from the historical data experience template image are: the brightness, contrast, and structure of the flame image pixels, as well as the changes in brightness, contrast, and structure of the flame image pixels.

4. A method for steel refining using a machine vision-based steel refining control system as described in claim 1, characterized in that... Includes the following steps: S1: At the end of the refining process in the refining furnace, the flame at the furnace opening is captured by the optical image sensor camera in the image acquisition module. After the captured flame is converted into first image information, it is sent to the image acquisition card for sampling and preprocessing to obtain second image information. The second image information is then stored in the buffer register. S2: The second image information from step S1 is preprocessed by the image preprocessing unit in the analysis control module to perform denoising and standardization preprocessing. S3: By analyzing the feature extraction unit in the control module, extract the brightness, contrast and structure of the image pixels, as well as the feature data of the brightness change, contrast change and structure change of the image pixels from the second image information after preprocessing in step S2. S4: By analyzing the similarity calculation unit in the control module, the feature data in the second image information of step S3 is weighted and fused with the following functions from the template image information obtained from historical data experience template images: brightness similarity function and brightness change similarity function, contrast similarity function and contrast change similarity function, structure similarity function and structure change similarity function. The calculation formulas for each function are as follows: In the formula, u x u is the average pixel value of the second image. y k is the average pixel value of the template image. x k is the average pixel change in the second image. y t is the average pixel change of the template image; x k is the standard deviation of the second image pixels; y r is the standard deviation of the template image pixels; x r is the standard deviation of the pixel variation in the second image. y t is the standard deviation of the pixel variation in the template image; xy r is the covariance of the pixels in the second image and the template image. xy Let c be the covariance of pixel changes between the second image and the template image; c1, c2, c3, c4, c5, and c6 are stability coefficients to avoid zero denominators in the corresponding formulas, and their specific calculation formulas are as follows: In the formula, N is the total number of pixels in the second image and the template image, and x i (1≤i≤N) represents the pixel values ​​of each point in the second image arranged in order, y i (1≤i≤N) represents the pixel values ​​of each point in the template image arranged in order; S5: Based on the function values ​​in step S4, calculate the similarity between the second image and the template image using the following formula; S=[α1+a1(x, y)]+[α2×a2(x, y)]+[β1×b1(x, y)]+[β2×b2(x, y)]+[γ1×c1(x, y)]+[γ2×c2(x, y)]; In the formula, S is the similarity between the second image information and the template image information, α1, α2, β1, β2, γ1 and γ2 are weighting factors, a1(x,y) is the brightness similarity function, a2(x,y) is the brightness change similarity function, b1(x,y) is the contrast similarity function, b2(x,y) is the contrast change similarity function, c1(x,y) is the structural similarity function, and c2(x,y) is the structural change similarity function. S6: After comparing the similarity of step S5 with the threshold, the endpoint of the molten steel tapping from the refining furnace is determined in order to control the refining furnace to produce qualified molten steel and tap it out.

5. The method according to claim 4, characterized in that... The threshold in step S6 is a dynamic reference value obtained through learning and optimization from historical data, and it is determined through the following steps: S61) Template Image Selection and Parameter Setting: Five flame images at the optimal tapping endpoint of the corresponding steel grade successfully refined from the system's historical data were selected as template images. All relevant parameters were set based on conventional historical data learning and experimental optimization, including: The weighting factors are set as follows: α1=0.2, α2=0.15, β1=0.15, β2=0.1, γ1=0.25, γ2=0.15; The stability coefficient is set as follows: c1=c2=(0.01×L) 2 ,c3=c4=(0.03×L) 2 ,c5=c6=c3 / 2, In the formula: L = 255; S62) Calculate the similarity of the template images themselves: Calculate the similarity between the five template images in step S61) to form a similarity set value, which is used to represent the degree of similarity between the final state images; S63) Statistical analysis: Calculate the mean and variance of the similarity set values ​​from step S62), or estimate them using confidence intervals to obtain a threshold. This threshold can be set as the average value of the similarity set values ​​minus 2-3 times the standard deviation, or a conservative value that can guarantee a high hit rate can be selected based on experimental data, such as the lower 5% quantile of the similarity set. S64) Verification and Adjustment: Apply the threshold from step S63) to another set of known historical data for verification and fine-tuning, ultimately determining an optimized threshold S that ensures both accuracy and a certain degree of fault tolerance. threshold The optimized threshold S threshold It can be pre-stored in the system's judgment module and supports empirical fine-tuning by engineers based on different steel grades, furnace conditions, or production stages.

6. The method according to claim 4, characterized in that... The determination of the endpoint of molten steel tapping from the refining furnace in step S6 is as follows: When the similarity is greater than or equal to the optimization threshold, that is: S ≥ S threshold If the current furnace flame image is found to be highly similar to the optimal tapping endpoint template image, and the tapping endpoint has been reached, the judgment module will send a tapping command to the refining furnace control system to execute the tapping operation or proceed to the next process. When the similarity is less than the optimization threshold, i.e., S < S threshold If the current flame state at the furnace opening is determined to be not yet in line with the optimal iron tapping endpoint template image requirements, i.e., the steel tapping endpoint has not been reached, the judgment module sends a command to the refining furnace control system to stop tapping steel and continue refining. The analysis and control module continues to calculate and analyze the next frame of flame image information acquired by the image acquisition module. This process continues until the steel tapping endpoint requirements are met, at which point the steel tapping operation is performed or the process moves to the next step.

Images