Metal smelting reduction endpoint judgment system based on deep learning

Through a metal smelting reduction end point judgment system based on deep learning, image processing and deep learning technology are used to automatically identify highlights in the furnace, which solves the problem that smelting reduction end point judgment in the existing technology depends on manual experience, and improves the accuracy and robustness of judgments reduces cost and error.

CN119048481APending Publication Date: 2024-11-29KUNMING UNIV OF SCI & TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411199369.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-29
Publication Date
2024-11-29

AI Technical Summary

Technical Problem

In the existing metal smelting process, the judgment of the end point of smelting reduction relies on manual experience, the accuracy is not high, which increases employment pressure and errors.

Method used

A metal smelting reduction end point judgment system based on deep learning is designed, including image acquisition module, color space conversion module, feature highlight extraction module, highlight counting module and restore end point judgment module. Grayscale is carried out by monitoring the images in the furnace in real time. and binary processing, using deep learning to identify the number and distribution of bright spots, determine whether the end point of smelting and restoration is reached, and a countdown alarm is issued when the end point is reached.

Benefits of technology

It realizes automated judgment, reduces artificial errors, improves the quality of smelting output, reduces the loss of sulfur dioxide and oxygen ventilation, reduces employee employment costs, and improves the accuracy and robustness of judgments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119048481A_ABST
    Figure CN119048481A_ABST
Patent Text Reader

Abstract

The invention discloses a metal smelting reduction endpoint judgment system based on deep learning, and relates to the field of converter smelting intellectualization. A color space conversion module; a feature bright spot extraction module; a bright spot counting module; and a restoration end point judgment module. And after the current smelting RGB image data in the converter is obtained, whether the converter reaches the smelting reduction end point or not is automatically judged, errors caused by human experience judgment and staff operation are avoided, and the smelting output quality can be effectively controlled. The operation indexes are further quantified, so that the loss of sulfur dioxide and oxygen ventilatory capacity can be actually reduced, and the labor cost of staff is also reduced.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of intelligent metal smelting, and in particular relates to a metal smelting reduction endpoint judgment system based on deep learning. Background Art

[0002] Intelligent metal smelting refers to the use of automation and intelligent technology to achieve automation and intelligent control of the metal smelting process and improve production efficiency and quality. In the process of metal smelting, the control and monitoring of the smelting process are multifaceted, among which the judgment of the metal smelting reduction end point is particularly important. The judgment of each moment of smelting in the smelting process often affects the output and quality. Under this circumstance, it is particularly important for manufacturing enterprises to take a preventive approach to product quality control rather than a post-remedy management approach. It is particularly important to apply effective prediction and judgment methods in the production process to assist employees in judging the end point of smelting reduction, form a systematic and standardized model for the entire production cycle, help managers to perceive the production status in a refined manner, comprehensively improve product quality, increase economic benefits, and how to adapt to the laws of products in the entire production cycle and further improve product quality are all worthy of systematic exploration. Therefore, in the research on intelligent metal smelting technology, the judgment of the end point of reduction is an important link. In the existing process of judging the end point of smelting reduction, the method of artificial experience judgment is mainly used.

[0003] In the actual metal smelting reduction endpoint judgment scenario, the reduction endpoint moment can only be judged by manual experience, which increases the pressure on the workers. This method is heavily dependent on the personal experience level of the operator and has low accuracy. Therefore, an accurate and fast metal smelting reduction endpoint judgment system is very important for the field of metal smelting intelligence. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides a metal smelting reduction endpoint judgment system based on deep learning, comprising:

[0005] An image acquisition module is used to acquire the image inside the furnace during the current metal smelting process;

[0006] A color space conversion module, connected to the image acquisition module, for graying and binarizing the image in the furnace to obtain an image to be identified;

[0007] A feature bright spot extraction module is connected to the color space conversion module and is used to identify the image to be identified by using a connected region search function and count the bright spots within a preset brightness threshold to obtain the number of bright spots in the furnace;

[0008] A bright spot counting module, connected to the characteristic bright spot extraction module, is used to determine the number of bright spots in the furnace, record the number of bright spot areas, and record the number of successes;

[0009] The restoration end point judgment module is connected to the bright spot counting module and is used to judge whether the restoration end point state has been reached. When it is judged that the end point has been reached, a countdown is performed and an alarm is sounded at the end of the countdown.

[0010] Preferably, the image acquisition module includes a high-speed camera and a memory;

[0011] The high-speed camera is used to monitor the working medium state of the smelting metal whose reduction endpoint is to be determined in real time, and obtain a real-time RGB image of the working medium in the furnace;

[0012] The memory is used to store the real-time RGB image of the working medium in the furnace for reading by the color space conversion module.

[0013] Preferably, the color space conversion module includes an RGB image grayscale processing unit and an image binarization unit;

[0014] The RGB image grayscale processing unit is used to grayscale the acquired real-time RGB image of the working medium in the furnace in real time to generate a grayscale image after grayscale processing;

[0015] The image binarization unit is used to perform binarization processing on the image edge of the grayscale image to obtain the image to be identified.

[0016] Preferably, the process of the RGB image grayscale processing unit performing grayscale processing on the acquired real-time RGB image of the working medium in the furnace in real time includes:

[0017] In the process of image color space conversion, the threshold range for identifying bright spots is obtained based on the empirical judgment process of the reduction endpoint of specific metal smelting and coupled with deep learning of the historical furnace endpoint image data. Based on the threshold range of the bright spots to be identified, the grayscale processing method is selected for grayscale processing to generate a grayscale image after grayscale processing.

[0018] Preferably, the grayscale processing method includes component method, maximum value method, average value method and weighted average method;

[0019] The formula expression of the component method is: F1(i, j) = R(i, j) F2(i, j) = G(i, j) F3(i, j) = B(i, j);

[0020] The formula expression of the maximum value method is: F(i, j)=Max(R(i, j), G(i, j), B(i, j));

[0021] The formula expression of the average value method is: F(i, j)=(R(i, j), G(i, j), B(i, j)) / 3;

[0022] The formula of the weighted average method is: F(i, j)=0.30R(i, j)+0.59G(i, j)+0.11B(i, j).

[0023] Preferably, the feature bright spot extraction module uses a connected region search function to identify the image to be identified, and counts bright spots within a preset brightness threshold. The process of obtaining the number of bright spots in the furnace includes:

[0024] In the image to be identified, the regionprops function is used to find and obtain the high-point connected area reflecting the real-time RGB image of the working fluid in the furnace;

[0025] The number of characteristic bright spots and the distribution of characteristic bright spots are obtained based on the connected area of ​​the highlight spots.

[0026] Preferably, the bright spot counting module comprises the following steps in determining the number of bright spots in the furnace:

[0027] The identified bright spot image data is annotated using an annotation tool, and the identified bright spots are marked and counted in the real-time RGB image of the working fluid in the furnace using a circular frame.

[0028] Preferably, the restoration endpoint judgment module includes a bright spot marking times judgment unit and a restoration endpoint alarm countdown unit;

[0029] The bright spot marking times judging unit is used to judge whether the count times of the marked bright spot meet the restoration end point condition;

[0030] The reduction endpoint alarm countdown unit is used to start the countdown and alarm the end of reduction when the real-time image of metal smelting meets the reduction endpoint condition.

[0031] Preferably, the reduction endpoint judgment module, in the process of judging whether the smelting reduction endpoint has been reached, couples the deep learning of real-time image data with the reduction endpoint experience judgment process of the specific metal smelting to derive the number of bright spots at the reduction endpoint and the required countdown duration, counts the number of successful markings of the bright spots, and starts the countdown when the number of markings meets the reduction endpoint conditions, and sounds an alarm of "reduction end" when the countdown ends.

[0032] Compared with the prior art, the present invention has the following advantages and technical effects:

[0033] The metal smelting reduction endpoint judgment system based on deep learning of the present invention automatically judges whether the current furnace has reached the smelting reduction endpoint after obtaining the RGB image data of the current furnace smelting, thereby avoiding errors in human experience judgment and employee operation, and can effectively control the smelting output quality. Further quantitative operation indicators can actually reduce the loss of sulfur dioxide and oxygen ventilation, and also reduce the employment cost of employees.

[0034] The metal smelting reduction endpoint judgment system proposed in the present invention is used for the actual production data of a domestic smelting plant. Through experimental analysis, on different smelting metals produced by different types of smelting furnaces, it is verified that the proposed metal smelting reduction endpoint judgment system based on deep learning has high accuracy and robustness. Through deep learning of different smelting process reduction endpoint image data and reduction endpoint empirical judgment, and comparing the endpoint judgment situation, the accuracy of the present invention is improved compared with other methods. In addition, the present invention can reduce the impact of data quality differences and different production environments. Relying on the data collected independently, data correction can be performed more accurately, further verifying that the system proposed in the present invention has a good judgment accuracy rate, and can still make high-precision judgments and predictions in response to different situations. And in the process of continuous realization of the automation process, it can reduce personnel maintenance, use more scientific and reasonable means to achieve intelligence, and reduce various costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0036] Figure 1 A schematic diagram of the system structure of an embodiment of the present invention;

[0037] Figure 2 is a system working flow diagram of an embodiment of the present invention;

[0038] Figure 3 A three-dimensional grayscale value map of the restoration end point ROI area according to an embodiment of the present invention;

[0039] Figure 4 This is a bright spot marking diagram of the restoration endpoint according to an embodiment of the present invention. DETAILED DESCRIPTION

[0040] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0041] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0042] like Figure 1 As shown, in this embodiment, a metal smelting reduction endpoint judgment system based on deep learning is provided, comprising:

[0043] An image acquisition module is used to acquire the image inside the furnace during the current metal smelting process;

[0044] A color space conversion module is connected to the image acquisition module and is used to grayscale and binarize the image in the furnace to obtain an image to be identified;

[0045] The feature bright spot extraction module is connected to the color space conversion module and is used to identify the image to be identified by using the connected region search function, and count the bright spots within the preset brightness threshold to obtain the number of bright spots in the furnace;

[0046] The bright spot counting module is connected to the characteristic bright spot extraction module, and is used to judge the number of bright spots in the furnace, record the number of bright spot areas, and record the number of successes;

[0047] The restoration end point judgment module is connected to the bright spot counting module and is used to judge whether the restoration end point state has been reached. When it is judged that the end point has been reached, a countdown is performed and an alarm is sounded at the end of the countdown.

[0048] Preferably, the image acquisition module includes a high-speed camera and a memory;

[0049] The high-speed camera is used to monitor the working fluid state of the smelting metal whose reduction endpoint is to be determined in real time, and obtain the real-time RGB image of the working fluid in the furnace;

[0050] The memory is used to store the real-time RGB image of the working fluid in the furnace for reading by the color space conversion module.

[0051] Preferably, the color space conversion module includes an RGB image grayscale processing unit and an image binarization unit;

[0052] The RGB image grayscale processing unit is used to grayscale the acquired real-time RGB image of the working medium in the furnace in real time to generate a grayscale image after grayscale processing;

[0053] The image binarization unit is used to perform binarization processing on the image edge of the grayscale image to obtain the image to be recognized.

[0054] Preferably, the process of graying the acquired real-time RGB image of the working medium in the furnace in real time by the RGB image graying processing unit includes:

[0055] In the process of image color space conversion, the threshold range for identifying bright spots is obtained based on the empirical judgment process of the reduction endpoint of specific metal smelting and coupled with deep learning of the historical furnace endpoint image data. Based on the threshold range of the bright spots to be identified, the grayscale processing method is selected for grayscale processing to generate a grayscale image after grayscale processing.

[0056] Preferably, the grayscale processing method includes a component method, a maximum value method, an average value method, and a weighted average method;

[0057] The formula expression of the component method is: F1(i, j) = R(i, j) F2(i, j) = G(i, j) F3(i, j) = B(i, j);

[0058] The formula expression of the maximum value method is: F(i, j) = Max(R(i, j), G(i, j), B(i, j));

[0059] The formula expression of the average value method is: F(i, j) = (R(i, j), G(i, j), B(i, j)) / 3;

[0060] The formula of the weighted average method is: F(i, j) = 0.30R(i, j) + 0.59G(i, j) + 0.11B(i, j).

[0061] Preferably, the feature bright spot extraction module uses a connected region search function to identify the image to be identified, and counts bright spots within a preset brightness threshold. The process of obtaining the number of bright spots in the furnace includes:

[0062] In the image to be identified, the regionprops function is used to find and obtain the high-light point connected area that reflects the real-time RGB image of the working fluid in the furnace;

[0063] The number and distribution of characteristic highlights are obtained based on the connected areas of the highlight points.

[0064] Preferably, the bright spot counting module comprises the following steps in determining the number of bright spots in the furnace:

[0065] The identified bright spot image data is annotated using an annotation tool, and the identified bright spots are marked and counted in the real-time RGB image of the working fluid in the furnace using a circular frame.

[0066] Preferably, the restoration endpoint judgment module includes a bright spot marking times judgment unit and a restoration endpoint alarm countdown unit;

[0067] The bright spot marking times judging unit is used to judge whether the count times of the marked bright spot meet the restoration end point condition;

[0068] The reduction endpoint alarm countdown unit is used to start the countdown and alarm the end of reduction when the real-time image of metal smelting meets the reduction endpoint conditions.

[0069] Preferably, when judging whether the smelting reduction endpoint has been reached, the reduction endpoint judgment module couples the deep learning of real-time image data with the reduction endpoint experience judgment process of the specific metal smelting to derive the number of bright spots at the reduction endpoint and the required countdown duration, and counts the number of successful bright spot markings. When the number of markings meets the reduction endpoint conditions, the countdown starts, and an alarm of "reduction end" is sounded at the end of the countdown.

[0070] The present invention provides a metal smelting reduction endpoint judgment system based on deep learning. Based on the deep learning of reduction endpoint image data and reduction endpoint experience judgment in the metal smelting furnace in the early stage, the system includes an image acquisition module, a color space conversion module, a feature bright spot extraction module, a bright spot counting module, and a reduction endpoint judgment module. The image acquisition module and the color space conversion module extract and store the image data in the current furnace, and perform color space conversion on the acquired image data to extract key feature images, ensure the integrity of the image data, and enhance its recognizability; the feature bright spot extraction module, the bright spot counting module, and the reduction endpoint judgment module read and identify the above-mentioned acquired feature images in sequence based on the early deep learning, and finally obtain a feature image that meets the reduction endpoint feature and alarm, thereby avoiding errors in human experience judgment and employee operation. Further quantified operation indicators can achieve stable judgment of the system and reduce employee employment costs.

[0071] like Figure 2 As shown, this embodiment also discloses a method for determining the end point of metal smelting reduction based on deep learning, which mainly includes:

[0072] Obtain the RGB color images of the reduction endpoint of several historical furnaces and the empirical judgment method of the reduction endpoint for deep learning to derive the image processing threshold;

[0073] like Figure 3 As shown, in this embodiment, the grayscale value three-dimensional visualization processing is performed on the RGB color image of the historical furnace smelting reduction endpoint, and professionals use the obtained visualized grayscale value image combined with the manual experience judgment method to further obtain the smelting reduction endpoint characteristics.

[0074] In this step, first ensure that the acquired image is the smelting reduction endpoint image, and the detailed process of implementing the artificial experience judgment method; if the acquired RGB image does not correspond to the smelting reduction endpoint, it is necessary to reconfirm the smelting reduction endpoint image judged by artificial experience.

[0075] Then, based on the coupling results of deep learning of the above image data and empirical judgment results, the image features of the smelting reduction endpoint are obtained.

[0076] Obtain the image of the furnace during the current metal smelting, perform grayscale processing and binarization processing on the image of the furnace to obtain the image to be recognized;

[0077] Use the connected region search function to identify the image to be identified, and count the bright spots within the preset brightness threshold to obtain the number of bright spots in the furnace; judge the number of bright spots in the furnace, and record the number of bright spot areas and the number of successes;

[0078] Determine whether the restoration end state has been reached, and when it is determined that the end state has been reached, start a countdown and alarm at the end of the countdown.

[0079] like Figure 4 As shown, according to the image features of the smelting reduction endpoint, a labeling tool is used to mark the bright spots within the threshold range, and the number of markings is counted until the reduction endpoint features are met.

[0080] Figure 3 For the corresponding Figure 4 In the embodiment of the invention, the three-dimensional gray value map of the end point ROI area is restored, and it is not difficult to see that Figure 3 The highlighted position of the gray value in the Z direction of the ROI area corresponds to the embodiment of the present invention Figure 4 According to the obtained bright spot annotated image and the empirical judgment method, professionals can adjust the thresholds in the image processing and recognition process in the color space conversion module, the feature bright spot extraction module, and the bright spot counting module.

[0081] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A metal smelting reduction endpoint judgment system based on deep learning, characterized in that: include: An image acquisition module is used to acquire the image inside the furnace during the current metal smelting process; A color space conversion module, connected to the image acquisition module, for graying and binarizing the image in the furnace to obtain an image to be identified; A feature bright spot extraction module is connected to the color space conversion module and is used to identify the image to be identified by using a connected region search function and count the bright spots within a preset brightness threshold to obtain the number of bright spots in the furnace; A bright spot counting module, connected to the characteristic bright spot extraction module, is used to determine the number of bright spots in the furnace, record the number of bright spot areas, and record the number of successes; The restoration end point judgment module is connected to the bright spot counting module and is used to judge whether the restoration end point state has been reached. When it is judged that the end point has been reached, a countdown is performed and an alarm is sounded at the end of the countdown.

2. The metal smelting reduction endpoint judgment system based on deep learning according to claim 1 is characterized in that: The image acquisition module includes a high-speed camera and a memory; The high-speed camera is used to monitor the working medium state of the smelting metal whose reduction endpoint is to be determined in real time, and obtain a real-time RGB image of the working medium in the furnace; The memory is used to store the real-time RGB image of the working medium in the furnace for reading by the color space conversion module.

3. The metal smelting reduction endpoint judgment system based on deep learning according to claim 1 is characterized in that: The color space conversion module includes an RGB image grayscale processing unit and an image binarization unit; The RGB image grayscale processing unit is used to grayscale the acquired real-time RGB image of the working medium in the furnace in real time to generate a grayscale image after grayscale processing; The image binarization unit is used to perform binarization processing on the image edge of the grayscale image to obtain the image to be identified.

4. The metal smelting reduction endpoint judgment system based on deep learning according to claim 3 is characterized in that: The process of the RGB image grayscale processing unit performing grayscale processing on the acquired real-time RGB image of the working medium in the furnace in real time includes: In the process of image color space conversion, the threshold range for identifying bright spots is obtained based on the empirical judgment process of the reduction endpoint of specific metal smelting and coupled with deep learning of the historical furnace endpoint image data. Based on the threshold range of the bright spots to be identified, the grayscale processing method is selected for grayscale processing to generate a grayscale image after grayscale processing.

5. The metal smelting reduction endpoint judgment system based on deep learning according to claim 4 is characterized in that: The grayscale processing method includes component method, maximum value method, average value method and weighted average method; The formula expression of the component method is: F1(i, j) = R(i, j) F2(i, j) = G(i, j) F3(i, j) = B(i, j); The formula expression of the maximum value method is: F(i, j)=Max(R(i, j), G(i, j), B(i, j)); The formula expression of the average value method is: F(i, j)=(R(i, j), G(i, j), B(i, j)) / 3; The formula of the weighted average method is: F(i, j)=0.30R(i, j)+0.59G(i, j)+0.11B(i, j).

6. The metal smelting reduction endpoint judgment system based on deep learning according to claim 1 is characterized in that: The feature bright spot extraction module uses a connected region search function to identify the image to be identified, and counts bright spots within a preset brightness threshold. The process of obtaining the number of bright spots in the furnace includes: In the image to be identified, the regionprops function is used to find and obtain the high-point connected area reflecting the real-time RGB image of the working fluid in the furnace; The number of characteristic bright spots and the distribution of characteristic bright spots are obtained based on the connected area of ​​the highlight spots.

7. The metal smelting reduction endpoint judgment system based on deep learning according to claim 1 is characterized in that: The process of the bright spot counting module determining the number of bright spots in the furnace includes: The identified bright spot image data is annotated using an annotation tool, and the identified bright spots are marked and counted in the real-time RGB image of the working fluid in the furnace using a circular frame.

8. The metal smelting reduction endpoint judgment system based on deep learning according to claim 1 is characterized in that: The restoration endpoint judgment module includes a bright spot mark times judgment unit and a restoration endpoint alarm countdown unit; The bright spot marking times judging unit is used to judge whether the count times of the marked bright spot meet the restoration end point condition; The reduction endpoint alarm countdown unit is used to start the countdown and alarm the end of reduction when the real-time image of metal smelting meets the reduction endpoint condition.

9. The metal smelting reduction endpoint judgment system based on deep learning according to claim 1 is characterized in that: In the process of judging whether the smelting reduction endpoint has been reached, the reduction endpoint judgment module couples the deep learning of real-time image data with the reduction endpoint experience judgment process of specific metal smelting to obtain the number of bright spots at the reduction endpoint and the required countdown time, counts the number of successful bright spot markings, starts the countdown when the number of markings meets the reduction endpoint condition, and alarms "reduction end" when the countdown ends.