A method, system, medium and device for detecting tilting flow slag in a converter desulfurization ladle
By installing industrial imaging devices at the converter desulfurization station and combining them with target deep learning algorithms, accurate monitoring of slag spillage from the converter desulfurization ladle was achieved. This solved the problem of inaccurate manual operation, reduced costs, and improved production safety and product quality.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BEIJING SHOUGANG AUTOMATION INFORMATION TECH
- Filing Date
- 2025-08-15
- Publication Date
- 2026-07-21
AI Technical Summary
In existing technologies, monitoring of slag spillage from desulfurized steel ladles in converters relies on manual operation, which lacks precision and leads to inaccurate judgment of the steel-slag interface, affecting product quality and increasing production costs. Furthermore, multi-sensor monitoring systems are complex and costly.
An industrial imaging device is installed at the converter desulfurization station to collect continuous frame streams of panoramic images in real time. The target deep learning algorithm is used to identify and analyze three types of targets and their location area related data. Combined with a hierarchical strategy, decision-making is carried out to achieve accurate monitoring of ladle tipping and slag flow.
It enables accurate monitoring of ladle tipping and slag flow, reduces hardware costs and system complexity, improves production safety and product quality, and has millisecond-level tipping angle control accuracy.
Smart Images
Figure CN121259710B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of automation control technology in the metallurgical industry, and in particular to a method, system, medium, and equipment for detecting tipping slag in a converter desulfurization ladle. Background Technology
[0002] In the steelmaking process, converter desulfurization is a key step in improving the quality of molten steel.
[0003] In traditional converter desulfurization processes, the ladle tilting and slag removal process relies primarily on manual operation and experience. Operators typically judge the interface between molten steel and slag by visual inspection. When molten steel or slag overflows, the operator must notify the control room via walkie-talkie, and the control room operators will then stop tilting. Due to factors such as operator experience, ambient lighting, and safety equipment limitations, manual judgment of the molten steel-slag interface lacks precision. This often results in stopping tilting too early, leading to insufficient slag removal, or stopping too late, causing significant molten steel loss. This affects product quality, wastes resources, and increases production costs.
[0004] Most existing improvement solutions employ multi-sensor monitoring systems, which are not only complex and extremely difficult to deploy and maintain, but also increase the number of potential failure points and implementation costs.
[0005] Therefore, how to achieve accurate monitoring of ladle tipping and slag flow through simple deployment is an urgent problem to be solved. Summary of the Invention
[0006] Considering the shortcomings of existing multi-sensor monitoring systems, how to accurately monitor ladle tipping and slag flow through simple deployment is a pressing issue. This invention provides a method, system, medium, and equipment for detecting tipping and slag flow in a converter desulfurization ladle. During the ladle tipping process, an industrial imaging device installed at the converter desulfurization station only needs to collect a continuous frame stream of panoramic images covering the entire ladle tipping area in real time. Then, a target deep learning algorithm is used to process the continuous frame stream of panoramic images to identify three types of targets and their location-related data in each frame. The changes of the three types of targets and their location-related data in the continuous frame stream of panoramic images are analyzed to determine the slag flow trend of the steel slag in the molten steel flow. Based on the slag flow trend of the steel slag in the molten steel flow, a hierarchical strategy is adopted for corresponding decision-making. It can be seen that this solution only requires the simple deployment of an industrial imaging device, combined with a target deep learning algorithm, to achieve accurate monitoring of ladle tipping and slag flow, and to achieve millisecond-level precise control of the converter tipping angle.
[0007] To address the aforementioned technical problems, a first aspect of the present invention discloses a method for detecting tipping slag in a converter desulfurization ladle, the method comprising:
[0008] During the ladle tilting process, an industrial camera installed at the converter desulfurization station is used to collect a continuous stream of panoramic images of the entire ladle tilting area in real time.
[0009] A target deep learning algorithm is used to process a continuous frame stream of a panoramic image to identify three types of targets and their location region-related data contained in each frame; wherein, the three types of targets are: the edge of the package, the molten steel flow, and the steel slag;
[0010] By analyzing the changes in the three types of targets and their location regions in the continuous frame stream of the panoramic image, the slag flow trend of the steel slag in the molten steel flow is determined.
[0011] Based on the slag flow trend of the steel slag in the molten steel flow, a grading strategy is adopted to make corresponding decisions.
[0012] Optionally, after processing the continuous frame stream of the panoramic image using a target deep learning algorithm to identify the three types of targets and their location region-related data contained in each frame image, the method further includes:
[0013] The packet edge position change rate is determined based on the packet edge and its position region related data within a set number of adjacent frames;
[0014] When the rate of change of the package edge position reaches a preset threshold, the industrial imaging device is controlled to enter the high-frequency detection mode.
[0015] Optionally, the step of using a target deep learning algorithm to process the continuous frame stream of the panoramic image and identify the three types of targets and their location region-related data contained in each frame specifically includes:
[0016] In the target deep learning algorithm, feature information is extracted from each frame of the panoramic image in the continuous frame stream; wherein, the feature information includes, but is not limited to, texture, color, and brightness; the feature information of each frame is processed to output the three types of targets and their location region-related data contained in each frame; wherein, the target and its location region-related data includes one or more of the following: target location, target region, category label, and confidence score; the target region is the overall region obtained with reference to the corresponding target location.
[0017] Optionally, the analysis of the changes in the three types of targets and their location area related data in the continuous frame stream of the panoramic image determines the slag flow trend of the steel slag in the molten steel flow;
[0018] In each frame of the image, a first relative distance of the slag region relative to the lower edge of the ladle is determined, and a second relative distance of the molten steel flow region relative to the lower edge of the ladle is determined.
[0019] Analyze the first relative distance and the second relative distance presented in the image frames within each cycle to determine the slag flow trend within each cycle.
[0020] Optionally, the analysis of the first relative distance and the second relative distance presented by the image frames within each period determines the slag flow trend within each period:
[0021] If no slag is detected in the image frame or only the molten steel flow is detected in each cycle, then the slag flow trend is: no slag flow.
[0022] If the image frame detects an increasing trend in the slag region within each cycle, or if the first relative distance is less than a set distance threshold, then the slag flow trend is: slag is about to flow.
[0023] If the image frame detects that the slag region exhibits a flow trend relative to the lower edge of the baffle in each cycle, or if the slag region reaches the lower edge of the baffle, then the slag flow trend is: slag flow begins.
[0024] Optionally, the step of making corresponding decisions based on the slag flow trend exhibited by the steel slag in the molten steel flow using a grading strategy specifically includes:
[0025] If no slag flows out, the ladle is flipped at a first preset flipping speed; wherein, the first preset flipping speed is a set angle under normal flipping conditions;
[0026] If slag is about to flow out, reduce the first preset flipping speed to the second preset flipping speed;
[0027] If slag begins to flow, control the ladle to stop tipping at the set stop time.
[0028] Optionally, the stopping point is the moment when the slag region begins to show a flow trend, or the moment when the slag region reaches the lower edge of the baffle.
[0029] A second aspect of the present invention discloses a detection system for tipping slag in a converter desulfurization ladle, the system comprising:
[0030] The sensing module is used to collect a continuous stream of panoramic images of the entire ladle tilting area in real time using an industrial camera installed at the converter desulfurization station during the ladle tilting process.
[0031] The first processing module is used to process the continuous frame stream of the panoramic image using a target deep learning algorithm, and to identify three types of targets and their location region-related data contained in each frame image; wherein, the three types of targets are: the edge of the package, the molten steel flow, and the steel slag;
[0032] The second processing module is used to analyze the changes of the three types of targets and their location area related data in the continuous frame stream of the panoramic image, and to determine the slag flow trend of the steel slag in the molten steel flow.
[0033] The decision-making module is used to make corresponding decisions based on the slag flow trend of the steel slag in the molten steel flow, using a graded strategy.
[0034] A third aspect of the present invention discloses a computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the steps of the above-described method.
[0035] A fourth aspect of the present invention discloses a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the above-described method.
[0036] Through one or more technical solutions of the present invention, the present invention has the following beneficial effects or advantages:
[0037] This invention provides a method, system, medium, and equipment for detecting slag overflow during ladle tilting in a converter desulfurization process. During ladle tilting, an industrial imaging device installed at the converter desulfurization station only needs to collect a continuous frame stream of panoramic images covering the entire ladle tilting area in real time. Then, a target deep learning algorithm is used to process the continuous frame stream of the panoramic images, identifying three types of targets and their location-related data in each frame. The changes in the three types of targets and their location-related data in the continuous frame stream of the panoramic images are analyzed to determine the slag overflow trend exhibited by the slag in the molten steel flow. Based on the slag overflow trend exhibited by the slag in the molten steel flow, a hierarchical strategy is adopted for corresponding decision-making. Therefore, this solution only requires the simple deployment of an industrial imaging device combined with a target deep learning algorithm to achieve accurate monitoring of slag overflow during ladle tilting and to achieve millisecond-level precise control of the converter tilting angle.
[0038] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description
[0039] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:
[0040] Figure 1A flowchart of a method for detecting tipping slag in a converter desulfurization ladle according to an embodiment of the present invention is shown;
[0041] Figure 2 A schematic diagram of a detection system for tipping slag in a converter desulfurization ladle according to an embodiment of the present invention is shown. Detailed Implementation
[0042] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this invention will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.
[0043] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this invention will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.
[0044] Firstly, such as Figure 1 As shown, the method for detecting tipping slag in a converter desulfurization ladle provided in this embodiment of the invention includes at least the following steps:
[0045] S101, during the ladle tilting process, uses an industrial camera installed at the converter desulfurization station to collect a continuous frame stream of panoramic images of the entire ladle tilting area in real time.
[0046] Specifically, the industrial imaging device employs a high-performance industrial camera, installed at the converter desulfurization station. The mounting bracket features a vibration-damping design to eliminate the impact of on-site vibrations on image quality. The industrial imaging device is required to withstand high temperatures and employs a special protective design to ensure reliable operation in high-temperature, high-dust environments. For example, its operating temperature range is -10℃ to 60℃, and with a specially designed cooling shield, it can operate stably for extended periods in high-temperature environments. The camera has a resolution of at least 4K and a frame rate of at least 60fps, equipped with a high-quality industrial-grade lens (adjustable focal length 12-50mm) and an automatic aperture control system to ensure image clarity and stability under different lighting conditions. The angle of the industrial imaging device is precisely calculated and optically optimized, observing the entire ladle tilting area from a 30° to 45° downward angle to ensure that the device can simultaneously capture the ladle edge state and the molten steel flow state. Because this application uses only a single industrial camera as the sensing device, the system structure is extremely simple, significantly reducing hardware costs and system complexity.
[0047] Each frame in the panoramic continuous frame stream contains the ladle tipping area, clearly showing the ladle edge, molten steel, and slag produced by the reaction.
[0048] S102 uses a target deep learning algorithm to process the continuous frame stream of the panoramic image, and identifies the three types of targets and their location region-related data contained in each frame image.
[0049] In the specific implementation process, industrial-grade computers are used to support the target deep learning algorithm. For example, an Intel i7 processor and 32GB of memory are configured, and an NVIDIA RTX 3060 graphics card is installed for AI inference calculations to ensure the real-time performance of the target deep learning algorithm. The storage system uses a 1TB industrial-grade SSD (Solid State Drive) to ensure the stability and durability of data recording.
[0050] The target deep learning algorithm is built using an optimized YOLO (You Only Look Once) deep learning framework, specifically designed for real-time target detection in high-temperature metallurgical scenarios. For example, it uses an optimized YOLO v8 version as the basic architecture and is trained through transfer learning.
[0051] The architecture of a target deep learning algorithm includes: an input layer, an image preprocessing layer, a backbone network, a neck network, a postprocessing layer, and an output layer.
[0052] The input layer is used to input each frame of the continuous frame stream of the panoramic image to the image preprocessing layer.
[0053] The image preprocessing layer performs scaling and pixel enhancement operations on each frame of the image. For example, it scales the image to 640×640 pixels dynamically to balance speed and accuracy; and randomly adjusts brightness and contrast for pixel enhancement. Of course, other image processing operations can also be performed, such as stitching, rotation, and flipping.
[0054] The backbone network extracts feature information from each frame of the continuous frame stream of the panoramic image. For example, a modified version of CSPDarknet53 is used to enhance edge feature extraction capabilities. The extracted feature information includes, but is not limited to, texture, color, and brightness. Specifically, the edge has a relatively fixed, clearly defined arc shape, a dark color, and a rough texture, serving as a dynamic reference in the image. Molten steel flow has extremely high brightness, an orange-yellow or golden hue, a relatively smooth surface, and exhibits continuous, directional fluid dynamics. Steel slag has significantly lower brightness than molten steel, a dark red or dark gray color, a rough and uneven surface, and often exhibits a lumpy, flocculent, or foamy form, forming a decisive difference from the smooth surface of molten steel.
[0055] In practice, the backbone network uses deep convolutions to extract multi-level features. Different levels contain different features; for example, lower-level features include edge details, while higher-level features contain semantic information. Furthermore, attention mechanisms can be added to deep convolutions to enhance feature responses in high-temperature regions; or depthwise separable convolutions can be used to reduce computational cost.
[0056] The neck network adopts a PANet structure. Specifically, multi-level features are fed into the PANet structure, and the feature representation is enhanced through bidirectional fusion of the Enhanced Feature Pyramid Network (FPN) and the Path Aggregation Network (PAN), thereby highlighting the differences between the molten steel flow, steel slag, and the rim. The fused multi-level feature map is then processed to predict the initial results: three types of targets and their location region-related data; wherein, the three types of targets are: the rim, the molten steel flow, and the steel slag; the data related to each type of target and its location region includes one or more of the following: target location, target region, category label, and confidence score; the target region is the overall region obtained by referring to the corresponding target location.
[0057] For example, if the target is the edge of a molten steel slag, the relevant data for the edge's location area would be: edge position (represented by coordinates), edge, and confidence score of the edge. If the target is slag, the relevant data for the slag's location area would be: slag position (i.e., the center of the slag area), slag area, slag, and confidence score of the slag. If the target is molten steel flow, the relevant data for the molten steel flow's location area would be: molten steel flow position (i.e., the center of the molten steel flow area), molten steel flow area, molten steel flow, and confidence score of the molten steel flow.
[0058] The post-processing layer employs WBF (Weighted Box Fusion) + threshold filtering to perform redundancy removal and confidence filtering on the data related to the three types of targets and their location regions, outputting accurate structured detection data to improve detection stability.
[0059] The training process of the object detection algorithm is described below:
[0060] First, continuous frame stream image samples of the converter production scene are acquired using industrial cameras, covering different lighting conditions (e.g., strong light, weak light), different shooting angles, and different steel types (e.g., low carbon steel, high alloy steel), ensuring that the samples contain the core feature information of three types of targets:
[0061] Texture characteristics: coarse texture of the rim, smooth texture of molten steel, agglomerated and flocculent texture of steel slag. Color characteristics: dark gray of the rim, orange-yellow and golden-yellow of molten steel, dark red and dark gray of steel slag. Brightness characteristics: extremely high brightness of molten steel, low brightness of steel slag (significantly lower than that of molten steel), and medium brightness of the rim.
[0062] Collect at least 1000+ image samples and divide them into a training set (for model feature learning) and a test set (to verify generalization ability) in an 8:2 ratio.
[0063] For each frame of image annotation, for three types of targets (wall edge, molten steel flow, and steel slag), the target position (center coordinates) and target region (boundary box range, reflecting the overall region) are annotated, generating a YOLO format annotation file containing class_id (category label), x_center (target position x), y_center (target position y), width (region width), and height (region height).
[0064] It is worth noting that by establishing a coordinate system with the ladle opening as the origin, the target position coordinates (x, y) of the ladle edge, slag, and molten steel flow, as well as the respective regions of the slag and molten steel flow, can be obtained.
[0065] Pre-trained model initialization.
[0066] The training set is input into the framework corresponding to the target deep learning algorithm for training. During training, the AdamW optimizer can be used to balance the stability and convergence speed of feature learning, ensuring that the model can accurately capture the subtle feature differences between the three types of targets (such as the brightness gradient between steel slag and molten steel). The loss function used can be: classification loss function (BCEWithLogitsLoss): used to optimize the accuracy of class label prediction, distinguishing between molten steel (high brightness golden yellow) and steel slag (low brightness dark red) based on color and brightness features; regression loss function (CIoU Loss): used to optimize the accuracy of target location and region, combining texture features (such as the curved contour of the rim) to improve the bounding box fitting accuracy; confidence loss function (FocalLoss): used to solve the imbalance of positive and negative samples (such as the large proportion of molten steel region and the small proportion of steel slag region), strengthen the feature learning of small targets (such as small pieces of steel slag), and ensure that their confidence scores are reliable.
[0067] Through multiple iterations, the mapping relationship between features and targets is fully learned, and a set of structured data is finally stably output, including: target location coordinates and region bounding box coordinates [x,y,width,height], category labels (wall edge, molten steel flow, slag), and confidence scores, providing reliable structured data support for subsequent slag state judgment and real-time control.
[0068] Of course, in order to improve the accuracy of recognition, the target deep learning algorithm will be continuously optimized using the collected historical data, and the training samples will be expanded through incremental learning to adapt to different steel grades and process parameters.
[0069] In the process of using a target deep learning algorithm to process a continuous frame stream of a panoramic image and identify the three types of targets and their location region-related data contained in each frame, the image frames are first preprocessed, and then feature information is extracted from each frame in the continuous frame stream of the panoramic image. The feature information includes, but is not limited to, texture, color, and brightness. The feature information of each frame is processed to output the three types of targets and their location region-related data contained in each frame. Each type of target and its location region-related data includes one or more of the following: target location, target region, category label, and confidence score. The target region is the overall region obtained by referring to the corresponding target location.
[0070] After processing the continuous frame stream of the panoramic image using a target deep learning algorithm to identify the three types of targets and their location region data contained in each frame, the rate of change of the envelope position is determined based on the envelope edge and its location region data within a set number of adjacent frames. This rate of change quantifies the intensity of the envelope edge's movement; a sudden increase in the rate of change indicates that the envelope edge has begun to move significantly, breaking its stability. Therefore, when the rate of change of the envelope edge position reaches a preset threshold, the industrial imaging device is controlled to enter a high-frequency detection mode, increasing the imaging frequency of the industrial imaging device and also increasing the detection frequency of the target deep learning algorithm. This allows for targeted adjustments based on actual working conditions, ensuring production safety on the production line.
[0071] S103, Analyze the changes of the three types of targets and their location area related data in the continuous frame stream of the panoramic image to determine the slag flow trend of the steel slag in the molten steel flow.
[0072] Since each frame contains three types of targets and their location area related data, by analyzing the three types of targets and their location area related data in each frame of the panoramic continuous frame stream according to the shooting sequence, the changing trends of steel slag, molten steel flow, and rim can be determined, and thus the slag flow trend of steel slag in the molten steel flow can be obtained.
[0073] Of course, in order to improve control accuracy, analysis and decision-making can be carried out on a cycle, with one cycle controlled within 80ms.
[0074] During ladle tilting, the molten steel flow and / or slag gradually approach the lower edge of the ladle. If the molten steel flow and / or slag reaches the lower edge of the ladle, or exceeds the lower edge of the ladle, slag flow occurs. Therefore, to determine whether slag flow exists, in each image frame, a first relative distance of the slag region relative to the lower edge of the ladle and a second relative distance of the molten steel flow region relative to the lower edge of the ladle are determined. By analyzing the first and second relative distances presented in the image frames within each cycle, the slag flow trend within each cycle is determined.
[0075] Specifically, determine the closest distance of the steel slag area relative to the lower edge of the ladle lip, and take this closest distance as the first relative distance D1. Determine the closest distance of the molten steel flow area relative to the lower edge of the ladle lip, and take this closest distance as the second relative distance D2. If D1 > D2, it indicates that the steel slag is at the downstream position of the molten steel flow and is closer to the ladle lip. If D1 < D2, it indicates that the steel slag is at the upstream position of the molten steel flow and is farther from the ladle lip.
[0076] Analyze the first relative distance D1 and the second relative distance D2 presented in the image frames of each cycle, and the change situations of the first relative distance D1 and the second relative distance D2 can be obtained, so as to infer the slag flow trend in each cycle.
[0077] If the steel slag is not detected in the image frames of each cycle or only the molten steel flow is detected, then the slag flow trend is: no slag flow;
[0078] If it is detected in the image frames of each cycle that the steel slag area shows an increasing trend, or the first relative distance is less than the set distance threshold, then the slag flow trend is: about to have slag flow.
[0079] Specifically, if the area range of the steel slag area becomes larger, it indicates that the steel slag area has an increasing trend and may have the risk of moving to a more downstream or edge position. If the first relative distance D1 is less than the set distance threshold, it indicates that the steel slag area has already been within the range of warning for slag flow. The above several situations all indicate that there may be a risk of "slag overflow" for the steel slag.
[0080] If it is detected in the image frames of each cycle that the steel slag area shows a flowing trend relative to the lower edge of the ladle lip, or the steel slag area reaches the lower edge of the ladle lip, then the slag flow trend is: starting to have slag flow. Specifically, if the first relative distance D1 continuously decreases, it indicates that the steel slag area moves to a more downstream or edge position. If the first relative distance D1 is 0, it indicates that the steel slag area reaches the lower edge of the ladle lip. The above several situations all indicate that the steel slag may have already overflowed.
[0081] S104, according to the slag flow trend presented by the steel slag in the molten steel flow, adopt a hierarchical strategy to make corresponding decisions.
[0082] Among them, the decision-making object is the first-level PLC system, and the response time is controlled within 50 ms, so that the response decision of this application can reach the millisecond level.
[0083] In the process of making corresponding decisions by adopting the hierarchical strategy, it specifically includes:
[0084] If there is no slag flow, flip the ladle at the first preset flipping speed; among them, the first preset flipping speed is the set angle in the normal flipping state;
[0085] If slag is about to flow, the first preset turning speed is reduced to the second preset turning speed to prevent slag from flowing in advance.
[0086] If slag flow begins, the ladle is controlled to stop tipping at a predetermined stop time. This predetermined stop time is the moment when the slag area begins to show a flow trend.
[0087] For example, if molten steel is detected flowing below the edge of the ladle, it will be tilted normally at the first preset tilting speed: 1.5° / s.
[0088] If sporadic "steel slag" targets are detected and show an increasing trend, or if the first relative distance is less than the set distance threshold, an early warning state is activated. The first preset overturning speed is reduced to the second preset overturning speed, for example, the overturning speed is reduced to a safe value of 0.5° / s, in order to buy time for a precise stop.
[0089] If it is confirmed that the "steel slag" has formed and is showing a flowing trend, or if the first relative distance is less than a set distance threshold, the starting moment when the steel slag area shows a flowing trend, or the moment when the steel slag area reaches the lower edge of the ladle, is determined as the stop time point, i.e., the optimal stop tipping time point. The ladle is controlled to stop tipping at the set stop time point, and an audible and visual alarm is triggered simultaneously. The core function of the optimal stop tipping time point is to achieve the best balance between the two mutually restrictive objectives of "minimizing molten steel loss" and "maximizing steel slag removal," thereby maximizing the molten steel yield while ensuring product quality.
[0090] Of course, the entire process of this solution is completed automatically without human intervention, which significantly improves the safety and effectiveness of the desulfurization and slag removal process.
[0091] An emergency stop is automatically triggered in the event of communication interruption or system malfunction.
[0092] In addition, to improve recognition accuracy, camera calibration and algorithm testing are automatically performed every shift to ensure long-term stable operation. Furthermore, new data is collected regularly to update the model, thereby continuously improving recognition accuracy.
[0093] In addition, complete image data of each tilting process is automatically recorded, including time-series images, target detection results, control commands, etc., and a database is established for storage to support subsequent process optimization and quality traceability.
[0094] Compared with the prior art, the present invention also has the following significant advantages:
[0095] Minimalist architecture: Using only a single industrial camera as the sensing device, the system structure is extremely simple, significantly reducing hardware costs and system complexity.
[0096] Easy to deploy: The installation and debugging process of a single industrial camera is simple, it is highly applicable, and it is easy to promote and apply in various converter production lines.
[0097] Accurate identification: The target deep learning algorithm based on the YOLO architecture can simultaneously identify the ladle edge, molten steel, and slag interface with an accuracy of over 95%.
[0098] High real-time performance: The target deep learning algorithm is optimized for fast processing speed, with a system response time of less than 80 milliseconds, meeting the requirements of real-time control.
[0099] Low cost: Compared with multi-sensor solutions, hardware costs are reduced by more than 70% and maintenance costs by more than 50%.
[0100] Highly adaptable: Through continuous learning, it can adapt to different production conditions and steel properties, and has a strong ability to generalize.
[0101] Easy to integrate: Seamlessly integrates with Level 1 PLC systems without requiring large-scale modifications to existing equipment.
[0102] Good scalability: Based on the same architecture, it can be extended to identify targets and states in other metallurgical processes through model training.
[0103] Secondly, based on the same inventive concept as the detection method for overturned slag in a converter desulfurization ladle provided in the first aspect of the embodiment, this embodiment of the invention also provides a detection system for overturned slag in a converter desulfurization ladle, see below. Figure 2 The system includes:
[0104] The sensing module 201 is used to collect a continuous frame stream of panoramic images of the entire ladle tilting area in real time using an industrial camera installed at the converter desulfurization station during the ladle tilting process.
[0105] The first processing module 202 is used to process the continuous frame stream of the panoramic image using a target deep learning algorithm, and identify three types of targets and their location region-related data contained in each frame image; wherein, the three types of targets are: the edge of the package, the molten steel flow, and the steel slag;
[0106] The second processing module 203 is used to analyze the changes of the three types of targets and their location area related data in the continuous frame stream of the panoramic image, and to determine the slag flow trend of the steel slag in the molten steel flow.
[0107] The decision module 204 is used to make corresponding decisions based on the slag flow trend of the steel slag in the molten steel flow, using a graded strategy.
[0108] It should be noted that the specific operation of each module in the converter desulfurization ladle detection system provided in the embodiments of the present invention has been described in detail in the method embodiments provided in the first aspect above. The specific implementation process can be referred to the method embodiments provided in the first aspect above, and will not be described in detail here.
[0109] Thirdly, based on the same inventive concept as the detection method for overturned slag in a converter desulfurization ladle provided in the first aspect embodiment, this embodiment of the invention also discloses a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of any of the methods described above.
[0110] Fourthly, based on the same inventive concept as the detection method for overturned slag in a converter desulfurization ladle provided in the first aspect embodiment, this embodiment of the invention also discloses a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of any of the methods described above.
[0111] Through one or more embodiments of the present invention, the present invention has the following beneficial effects or advantages:
[0112] This invention provides a method, system, medium, and equipment for detecting slag overflow during ladle tilting in a converter desulfurization process. During ladle tilting, an industrial imaging device installed at the converter desulfurization station only needs to collect a continuous frame stream of panoramic images covering the entire ladle tilting area in real time. Then, a target deep learning algorithm is used to process the continuous frame stream of the panoramic images, identifying three types of targets and their location-related data in each frame. The changes in the three types of targets and their location-related data in the continuous frame stream of the panoramic images are analyzed to determine the slag overflow trend exhibited by the slag in the molten steel flow. Based on the slag overflow trend exhibited by the slag in the molten steel flow, a hierarchical strategy is adopted for corresponding decision-making. Therefore, this solution only requires the simple deployment of an industrial imaging device combined with a target deep learning algorithm to achieve accurate monitoring of slag overflow during ladle tilting and to achieve millisecond-level precise control of the converter tilting angle.
[0113] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.
[0114] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.
Claims
1. A method for detecting tipping slag in a converter desulfurization ladle, characterized in that, The method includes: During the ladle tilting process, an industrial camera installed at the converter desulfurization station is used to collect a continuous stream of panoramic images of the entire ladle tilting area in real time. A target deep learning algorithm is used to process a continuous frame stream of a panoramic image to identify three types of targets and their location region-related data contained in each frame; the three types of targets are: the edge of the package, the molten steel flow, and the steel slag. The changes in the three types of targets and their location regions are analyzed in the continuous frame stream of the panoramic image to determine the slag flow trend of the steel slag in the molten steel flow. Specifically, in each frame, a first relative distance between the steel slag region and the lower edge of the molten steel rim is determined. If no steel slag is detected in each frame of the image cycle, or only the molten steel flow is detected, the slag flow trend is: no slag flow. If the steel slag region is detected to be increasing in each frame of the image cycle, or the first relative distance is less than a set distance threshold, the slag flow trend is: slag flow is imminent. If the steel slag region is detected to be flowing relative to the lower edge of the molten steel rim in each frame of the image cycle, or the steel slag region reaches the lower edge of the molten steel rim, the slag flow trend is: slag flow has begun. Based on the slag flow trend of the steel slag in the molten steel flow, a graded strategy is adopted to make corresponding decisions; wherein, if no slag flow occurs, the ladle is flipped at a first preset flipping speed; wherein, the first preset flipping speed is a set angle under normal flipping conditions; if slag flow is about to occur, the first preset flipping speed is reduced to a second preset flipping speed; if slag flow begins, the ladle is controlled to stop flipping at a set stop time point.
2. The method as described in claim 1, characterized in that, After processing the continuous frame stream of the panoramic image using a target deep learning algorithm to identify the three types of targets and their location region-related data contained in each frame, the method further includes: The packet edge position change rate is determined based on the packet edge and its position region related data within a set number of adjacent frames; When the rate of change of the package edge position reaches a preset threshold, the industrial imaging device is controlled to enter the high-frequency detection mode.
3. The method as described in claim 1, characterized in that, The process of using a target deep learning algorithm to process continuous frame streams of panoramic images to identify three types of targets and their location region-related data contained in each frame specifically includes: In the target deep learning algorithm, feature information is extracted from each frame of the panoramic image in the continuous frame stream; wherein, the feature information includes, but is not limited to, texture, color, and brightness; the feature information of each frame is processed to output the three types of targets and their location region-related data contained in each frame; wherein, the target and its location region-related data includes one or more of the following: target location, target region, category label, and confidence score; the target region is the overall region obtained with reference to the corresponding target location.
4. The method as described in claim 1, characterized in that, The stop time point is the moment when the slag region begins to show a flow trend, or the moment when the slag region reaches the lower edge of the baffle.
5. A detection system for tipping slag in a converter desulfurization ladle, characterized in that, The system includes: The sensing module is used to collect a continuous stream of panoramic images of the entire ladle tilting area in real time using an industrial camera installed at the converter desulfurization station during the ladle tilting process. The first processing module is used to process the continuous frame stream of the panoramic image using a target deep learning algorithm, and to identify three types of targets and their location region-related data contained in each frame image; wherein, the three types of targets are: the edge of the package, the molten steel flow, and the steel slag; The second processing module is used to analyze the changes of three types of targets and their location area related data in the continuous frame stream of the panoramic image, and to determine the slag flow trend of the steel slag in the molten steel flow; wherein, in each frame, a first relative distance of the steel slag area relative to the lower edge of the rim is determined; if the steel slag is not detected or only the molten steel flow is detected in each period of the image frame, the slag flow trend is: no slag flow; if the steel slag area is detected to be increasing in each period of the image frame, or the first relative distance is less than a set distance threshold, the slag flow trend is: slag flow is imminent; if the steel slag area is detected to be flowing relative to the lower edge of the rim in each period of the image frame, or the steel slag area reaches the lower edge of the rim, the slag flow trend is: slag flow has begun. The decision module is used to make corresponding decisions based on the slag flow trend of the steel slag in the molten steel flow using a graded strategy; wherein, if no slag flow occurs, the ladle is flipped at a first preset flipping speed; wherein, the first preset flipping speed is a set angle under normal flipping conditions; if slag flow is about to occur, the first preset flipping speed is reduced to a second preset flipping speed; if slag flow begins, the ladle is controlled to stop flipping at a set stop time point.
6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the method according to any one of claims 1-4.
7. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method according to any one of claims 1-4.