Converter unmanned tapping monitoring method and device based on artificial intelligence

Abstract

The invention provides a converter unmanned tapping monitoring method and device based on artificial intelligence. The method comprises the steps that historical image data of the converter tapping process are obtained; performing labeling and feature extraction on the historical image data; establishing a database on the basis of the extracted features, and training a tapping monitoring model by using the database, so that the tapping monitoring model learns, identifies and judges normal tapping, abnormal tapping, starting and ending of molten steel, and the position relationship between the normal converter and the molten steel tank and between the abnormal converter and the molten steel tank; and the trained tapping monitoring model is deployed in an actual production environment, the unmanned tapping process is monitored in real time, and the tapping process and an alarm prompt are provided. Various states in the tapping process are monitored in real time through the manual model, normal / abnormal tapping, the position relation between the molten steel tank and the converter, the rotating speed of the converter and the like are recognized, and an alarm is given out when abnormal conditions occur, so that safety accidents are avoided, and the safety of unmanned operation is guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of artificial intelligence and steelmaking, and in particular to an artificial intelligence-based converter unmanned steel tapping monitoring method and device. Background Art

[0002] At present, different automatic steel-tapping technologies have been applied in some factories. There are currently a variety of control technologies for automatic steel-tapping. For example, the patent application with publication number CN113088615A discloses a converter steel-tapping control method and system based on video recognition technology, which includes the following steps: analyzing the historical data of manual steel-tapping and self-learning to generate several steel-tapping inclination angle time tables, and automatically selecting the corresponding inclination angle time according to the current furnace conditions; judging in real time the slag overflow from the furnace mouth, slag overflow from the bottom of the furnace, low molten steel level and slag at the tapping mouth during the steel-tapping process based on the video at the converter mouth and under the converter furnace; and automatically generating instructions for lifting the converter upward or rotating it downward. For another example, the patent application with publication number CN 118685584A discloses a converter intelligent steel-tapping control method, device, medium and electronic equipment. The converter intelligent steel-tapping control method includes: real-time detection and acquisition of the furnace mouth slag line distance; controlling the rotation and tilting of the converter to maintain the furnace mouth slag line distance at a preset distance so that the static pressure of the steel-tapping port is the maximum value; confirming the position of the stable landing point of the steel flow at the steel-tapping port, and controlling the movement trajectory of the ladle car according to the position of the stable landing point of the steel flow so that the center position of the ladle car coincides with the position of the stable landing point of the steel flow, and controlling the setting position of the alloy chute above the ladle car so that the landing point position of the material flow flowing out of the alloy chute coincides with the position of the stable landing point of the steel flow.

[0003] However, the above-mentioned automated control technologies are unable to monitor the entire automatic steel-making process, and are unable to identify various abnormal situations during the automatic steel-making process, such as slag overflow from the converter mouth, slag overflow from the molten steel tank under the furnace, high or low liquid levels in the molten steel tank, and slag at the steel-making port, and are unable to provide alarms and early warning prompts for these abnormal situations. Summary of the invention

[0004] In view of this, the present invention provides an artificial intelligence-based converter unmanned steel tapping monitoring method and device to solve at least one of the above-mentioned problems.

[0005] In order to achieve the above object, the present invention adopts the following scheme:

[0006] According to a first aspect of the present invention, there is provided a method for monitoring unmanned steel tapping of a converter based on artificial intelligence, the method comprising: acquiring historical image data of the steel tapping process of the converter, the historical image data comprising normal steel tapping image data and abnormal steel tapping image data; labeling and extracting features of the historical image data, the features comprising the degree of molten steel splashing, slag and steel overflow at the turning junction, the relative position of the molten steel tank and the converter, the clearance of the liquid level in the molten steel tank, the time when the molten steel starts to flow out, and the time when the molten steel ends to flow out; establishing a database based on the extracted features, and using the database to train a steel tapping monitoring model, so that the steel tapping monitoring model learns to identify and judge normal steel tapping, abnormal steel tapping, the start and end of molten steel outflow, and the positional relationship between the normal and abnormal converter and the molten steel tank; deploying the trained steel tapping monitoring model into an actual production environment, monitoring the unmanned steel tapping process in real time, and providing the steel tapping process and alarm prompts.

[0007] As an embodiment of the present invention, the prompts of the steel tapping process in the above method include: the start stage of steel tapping and the end stage of steel tapping; the alarm prompts include: severe molten steel splashing, slag overflow from the converter mouth, molten steel overflow from the switchback, excessive deviation of the position of the molten steel tank, and abnormal clearance of the molten steel level in the molten steel tank.

[0008] As an embodiment of the present invention, the above method also includes: based on the normal steel tapping image data in the historical image data, using a single-classification support vector machine algorithm to train an abnormality recognition model to identify abnormal steel tapping conditions that do not appear in the abnormal steel tapping image data.

[0009] As an embodiment of the present invention, the single-class support vector machine algorithm in the above method adopts a radial basis kernel function, and optimizes the kernel parameter γ and the penalty parameter ν by a grid search or cross-validation method to determine the optimal anomaly recognition model.

[0010] As an embodiment of the present invention, the above method also includes: using sensors to obtain the weight, position, temperature and vibration data of the molten steel during the converter steel tapping process, and based on the molten steel weight, position, temperature and vibration data and the monitoring results of the steel tapping monitoring model, conducting a comprehensive analysis of abnormal conditions in the steel tapping process.

[0011] As an embodiment of the present invention, the above method also includes: issuing an early warning based on the positional relationship between the molten steel ladle and the converter identified by the steel tapping monitoring model, including: issuing an early warning signal if the distance between the mouth of the molten steel ladle and the converter mouth exceeds a set threshold.

[0012] As an embodiment of the present invention, after acquiring the historical image data of the converter steel tapping process in the above method, the method further comprises: performing data enhancement operation on the abnormal steel tapping image data in the historical image data to increase the amount of abnormal steel tapping image data.

[0013] As an embodiment of the present invention, the data enhancement operation in the above method includes: performing an image-based enhancement operation on the abnormal steel tapping image data: including rotating and translating, scaling, adding noise, blurring and occluding operations on the abnormal steel tapping image data to produce new abnormal steel tapping image data; performing a video-based enhancement operation on the abnormal steel tapping image data: playing the video clip to which the abnormal steel tapping image data belongs at a variable speed and then producing new abnormal steel tapping image data; performing a feature-based enhancement operation on the abnormal steel tapping image data: producing new abnormal steel tapping image data based on the SMOTE method.

[0014] As an embodiment of the present invention, the above method for producing new abnormal steel tapping image data based on the SMOTE algorithm includes: extracting enhanced features from the abnormal steel tapping image data, the enhanced features including: color histogram, texture features, edge features, shape features, key point features; using the vector of the enhanced features as input to the SMOTE algorithm to generate a feature vector of a new abnormal steel tapping image sample; and inputting the feature vector into a generative model trained by a generative adversarial network to convert it into a new abnormal steel tapping image.

[0015] According to a second aspect of the present invention, there is provided an unmanned steel-tapping monitoring device for a converter based on artificial intelligence, the device comprising: a historical data acquisition unit, for acquiring historical image data of the steel-tapping process of the converter, the historical image data comprising normal steel-tapping image data and abnormal steel-tapping image data; a feature extraction unit, for annotating and extracting features from the historical image data, the features comprising the degree of molten steel splashing, slag and steel overflow at the turning junction, the relative position of the molten steel tank and the converter, the clearance of the liquid level in the molten steel tank, the time when the molten steel starts to flow out, and the time when the molten steel ends to flow out; a model training unit, for establishing a database based on the extracted features, and using the database to train a steel-tapping monitoring model, so that the steel-tapping monitoring model learns to identify and judge normal steel-tapping, abnormal steel-tapping, the start and end of molten steel outflow, and the positional relationship between the normal and abnormal converter and the molten steel tank; a model deployment unit, for deploying the trained steel-tapping monitoring model into an actual production environment, for real-time monitoring of the unmanned steel-tapping process, and providing the steel-tapping process and alarm prompts.

[0016] As an embodiment of the present invention, the prompts of the above-mentioned steel tapping process include: the start stage of steel tapping and the end stage of steel tapping; the alarm prompts include: severe molten steel splashing, slag overflow from the converter mouth, molten steel overflow from the switchback, excessive deviation of the position of the molten steel tank, and abnormal clearance of the molten steel level in the molten steel tank.

[0017] As an embodiment of the present invention, the above-mentioned device also includes: an abnormality recognition unit, based on the normal steel tapping image data in the historical image data, using a single-classification support vector machine algorithm to train an abnormality recognition model to identify abnormal steel tapping conditions that do not appear in the abnormal steel tapping image data.

[0018] As an embodiment of the present invention, the above-mentioned single-classification support vector machine algorithm adopts a radial basis kernel function, and optimizes the kernel parameter γ and the penalty parameter ν by a grid search or cross-validation method to determine the optimal anomaly recognition model.

[0019] As an embodiment of the present invention, the above-mentioned device also includes: a sensor unit, which is used to use sensors to obtain the weight, position, temperature and vibration data of the molten steel during the steel tapping process of the converter, and based on the molten steel weight, position, temperature and vibration data and the monitoring results of the steel tapping monitoring model, conduct a comprehensive analysis of abnormal conditions in the steel tapping process.

[0020] As an embodiment of the present invention, the above-mentioned device also includes: an early warning unit, which is used to issue an early warning based on the positional relationship between the molten steel ladle and the converter identified by the steel tapping monitoring model, including: if the distance between the mouth of the molten steel ladle and the converter mouth exceeds a set threshold, a early warning signal is issued.

[0021] As an embodiment of the present invention, the above-mentioned device further includes: a data enhancement unit, which is used to perform data enhancement operation on the abnormal steel tapping image data in the historical image data to increase the amount of abnormal steel tapping image data.

[0022] As an embodiment of the present invention, the data enhancement operation includes: performing an image-based enhancement operation on the abnormal steel tapping image data: including rotating and translating, scaling, adding noise, blurring and occluding operations on the abnormal steel tapping image data to produce new abnormal steel tapping image data; performing a video-based enhancement operation on the abnormal steel tapping image data: playing the video clip to which the abnormal steel tapping image data belongs at a variable speed and then producing new abnormal steel tapping image data; performing a feature-based enhancement operation on the abnormal steel tapping image data: producing new abnormal steel tapping image data based on the SMOTE method.

[0023] As an embodiment of the present invention, producing new abnormal steel tapping image data based on the SMOTE algorithm includes: extracting enhanced features from the abnormal steel tapping image data, the enhanced features including: color histogram, texture features, edge features, shape features, key point features; using the vector of the enhanced features as input to the SMOTE algorithm to generate a feature vector of a new abnormal steel tapping image sample; and inputting the feature vector into a generative model trained by a generative adversarial network to convert it into a new abnormal steel tapping image.

[0024] According to a third aspect of the present invention, there is provided an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0025] According to a fourth aspect of the present invention, there is provided a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of the above method when executed by a processor.

[0026] According to a fifth aspect of the present invention, there is provided a computer program product, comprising a computer program / instructions, which implement the steps of the above method when executed by a processor.

[0027] It can be seen from the above technical solutions that the artificial intelligence-based converter unmanned steel-tapping monitoring method and device provided by the present invention monitors various states in the steel-tapping process in real time through an artificial model, identifies normal / abnormal steel-tapping, the positional relationship between the steel ladle and the converter, the converter rotation speed, etc., and issues an alarm when an abnormal situation occurs, thereby avoiding safety accidents and ensuring the safety of unmanned operation. It fills the gap of the existing unmanned steel-tapping system lacking a complete monitoring system. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0029] Figure 1 It is a flow chart of an artificial intelligence-based converter unmanned steel tapping monitoring method provided in an embodiment of the present application;

[0030] Figure 2 It is a flow chart of an unmanned converter steel tapping monitoring method based on artificial intelligence provided by another embodiment of the present application;

[0031] Figure 3 It is a flow chart of an unmanned converter steel tapping monitoring method based on artificial intelligence provided by another embodiment of the present application;

[0032] Figure 4 is a flowchart of a data enhancement operation provided in an embodiment of the present application;

[0033] Figure 5 It is a schematic diagram of a process for producing new abnormal steel tapping image data based on the SMOTE method provided in an embodiment of the present application;

[0034] Figure 6 This is a schematic diagram of the structure of an unmanned steel-tapping monitoring device for a converter based on artificial intelligence provided in an embodiment of the present application.

[0035] Figure 7 It is a schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0036] To make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings. Here, the exemplary embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0037] like Figure 1 The figure shows a flow chart of an artificial intelligence-based converter unmanned steel tapping monitoring method provided in an embodiment of the present application, the method comprising the following steps:

[0038] Step S101: Acquire historical image data of a converter steel tapping process, wherein the historical image data includes normal steel tapping image data and abnormal steel tapping image data.

[0039] The historical image data here mainly uses a large number of videos of the converter steel-tapping process recorded by high-definition cameras, and then extracts image data from the video. The historical image data collected here should include two categories: normal steel-tapping image data and abnormal steel-tapping image data. The normal steel-tapping image data includes the picture of molten steel falling smoothly into the ladle with only a small amount of steel flowers splashing; the abnormal steel-tapping image data includes the picture of abnormal conditions such as molten steel splashing, molten steel overflowing, slag overflowing from the furnace mouth, too low molten steel level clearance in the ladle, and abnormal relative position of the converter and the ladle.

[0040] In order to ensure the training effect of the subsequent steel-tapping monitoring model, this step requires the collection of sufficient data samples, covering all possible normal and abnormal situations. At the same time, the collected image data is stored according to normal and abnormal classification for subsequent labeling and feature extraction.

[0041] Step S102: annotate and extract features from the historical image data, wherein the features include the degree of molten steel splashing, slag and steel overflow at the turning point, the relative position of the molten steel tank and the converter, the liquid level clearance in the molten steel tank, the time when the molten steel starts to flow out, and the time when the molten steel stops flowing out.

[0042] Here, the annotation of historical image data can be done manually or by a semi-automatic annotation method that uses a pre-trained model for preliminary annotation and then manual assistance correction. Because there is no mature artificial intelligence application in the field of converter steel tapping, and for "normal steel tapping" and "abnormal steel tapping" requires certain professional knowledge and judgment, therefore, using manual annotation or manual assisted annotation to annotate the training data is a reliable method to ensure the quality of annotation. In addition to the labels of "normal steel tapping" and "abnormal steel tapping", you can also annotate "converter molten steel begins to flow out", "converter molten steel ends", "molten steel splashes", "furnace mouth overflows slag", "molten steel overflows", "molten steel level clearance in molten steel tank is normal", "molten steel level clearance in molten steel tank is too low", "normal relative position of converter and molten steel tank" and "abnormal relative position of converter and molten steel tank".

[0043] Step S103: Establish a database based on the extracted features, and use the database to train a steel tapping monitoring model, so that the steel tapping monitoring model learns to identify and judge normal steel tapping, abnormal steel tapping, the beginning and end of molten steel outflow, and the normal and abnormal positional relationship between the converter and the molten steel tank.

[0044] The extracted feature data and corresponding image labels are stored in the database to form a training data set, and then a suitable AI model, such as a convolutional neural network (CNN), a recurrent neural network (RNN), or a Transformer, is selected for training the steel-tapping monitoring model. During the model training process, the performance of the model is continuously optimized by adjusting model parameters, increasing training data, and other methods to improve the accuracy of identifying abnormal situations.

[0045] Through the above training, the steel-tapping monitoring model of the present application can learn the association between different features and different steel-tapping situations. In addition to giving the judgment of "normal steel-tapping / abnormal steel-tapping", it can also give various situations corresponding to normal steel-tapping or abnormal steel-tapping. For example, when it is detected that the molten steel splashes violently and exceeds the preset threshold, it can correspond to "abnormal steel-tapping-molten steel splashing"; when a large amount of slag overflowing from the furnace mouth is identified, it can correspond to "abnormal steel-tapping-slag overflow from the furnace mouth"; when the molten steel liquid level exceeds the safety threshold and may cause the molten steel to overflow, it can correspond to "abnormal steel-tapping-risk of molten steel overflow"; or when there is a deviation in the relative position of the converter and the ladle, which may cause the molten steel not to fall into the ladle, it corresponds to "abnormal steel-tapping-abnormal position" and so on.

[0046] Step S104: deploy the trained steel-tapping monitoring model into the actual production environment, monitor the unmanned steel-tapping process in real time and provide the steel-tapping process and alarm prompts.

[0047] The trained steel-tapping monitoring model is deployed to the hardware equipment in the actual production environment, such as industrial computers or embedded devices. At the same time, the image data of the steel-tapping process is collected in real time using high-definition cameras and input into the deployed AI model for analysis. Finally, the key parameters of the steel-tapping process can be displayed in real time on the system interface, such as the flow rate of molten steel, the degree of molten steel splashing, the state of the furnace mouth, the height of the molten steel level in the ladle, the relative position of the converter and the ladle, and the judgment result of "normal steel-tapping / abnormal steel-tapping", etc., and an alarm prompt is issued for the identified abnormal situation, which is convenient for operators to monitor and deal with in time.

[0048] Preferably, the prompts of the steel-tapping process include: the beginning stage of steel-tapping and the end stage of steel-tapping; and the alarm prompts include: severe splashing of molten steel, slag overflow from the converter mouth, molten steel overflow from the switch mouth, excessive deviation of the position of the molten steel tank, and abnormal clearance of the molten steel level in the molten steel tank. When the steel-tapping monitoring model detects an abnormal situation, it immediately sends out an audible and visual alarm signal, and displays information such as the abnormal type, hazard level, and handling suggestions to remind the operator to take timely measures. For example, if severe splashing of molten steel is detected, an alarm of "severe splashing of molten steel" is issued, and it is recommended to "adjust the converter angle / stop tapping".

[0049] It can be seen from the above technical solutions that the artificial intelligence-based converter unmanned steel-tapping monitoring method provided by the present invention monitors various states in the steel-tapping process in real time through an artificial model, identifies normal / abnormal steel-tapping, the positional relationship between the steel ladle and the converter, the converter rotation speed, etc., and issues an alarm when an abnormal situation occurs, thereby avoiding safety accidents and ensuring the safety of unmanned operation. This fills the gap that the existing unmanned steel-tapping system lacks a complete monitoring system.

[0050] The above-mentioned converter unmanned steel-tapping monitoring method can detect abnormal conditions in the historical image data during the steel-tapping process when the amount of training data is sufficient. However, when it is difficult to collect enough abnormal image data samples, the above-mentioned steel-tapping monitoring model may miss the abnormal conditions. In addition, since the converter steel-tapping process itself has a wide variety of abnormal conditions, in order to increase the reliability of the final monitoring results, such as Figure 2 As shown, the artificial intelligence-based converter unmanned steel tapping monitoring method of the present application may include the following steps:

[0051] Step S201: Acquire historical image data of a converter steel tapping process, wherein the historical image data includes normal steel tapping image data and abnormal steel tapping image data.

[0052] Step S202: annotate and extract features from the historical image data, wherein the features include the degree of molten steel splashing, slag and steel overflow at the turning point, the relative position of the molten steel tank and the converter, the liquid level clearance in the molten steel tank, the time when the molten steel starts to flow out, and the time when the molten steel stops flowing out.

[0053] Step S203: Establish a database based on the extracted features, and use the database to train a steel tapping monitoring model, so that the steel tapping monitoring model learns to identify and judge normal steel tapping, abnormal steel tapping, the beginning and end of molten steel outflow, and the normal and abnormal positional relationship between the converter and the molten steel tank.

[0054] The above steps S201-S203 are the same as steps S101-S103 in the aforementioned embodiment, and will not be further described here.

[0055] Step S204: Based on the normal steel-tapping image data in the historical image data, a single-class support vector machine algorithm (One-Class SVM) is used to train an abnormality recognition model to identify abnormal steel-tapping situations that do not appear in the abnormal steel-tapping image data.

[0056] This step is performed simultaneously with S202 and S203. The difference from step S202 and step S203 is that this step only uses the normal steel tapping image data in the historical image data. The purpose of this step is to identify abnormal steel tapping situations that have not appeared in the abnormal steel tapping image data.

[0057] The single-class support vector machine is an anomaly detection algorithm. It does not identify anomalies by learning abnormal samples, but by learning the characteristics of normal steel-tapping data to build a model that describes the normal steel-tapping state. This model can be regarded as a boundary surrounding normal data points. During the training process, the single-class support vector machine algorithm will try to find a hyperplane that includes as many normal steel-tapping data points as possible, while minimizing the area outside the hyperplane. This hyperplane defines the boundary of "normal" steel-tapping. When new steel-tapping data arrives, if the data point falls within the hyperplane (that is, within the boundary), it is considered to be normal steel-tapping; if the data point falls outside the hyperplane, it is considered to be abnormal steel-tapping, even if this abnormal situation has never appeared in the training data. Therefore, the single-class support vector machine can be trained to identify abnormal situations based only on normal steel-tapping data.

[0058] Further preferably, the above-mentioned single-classification support vector machine algorithm adopts a radial basis kernel function, and optimizes the kernel parameter γ and the penalty parameter ν by a grid search or cross-validation method to determine the optimal anomaly recognition model.

[0059] Single-class support vector machines can use different kernel functions to map data to high-dimensional space. The RBF kernel function is a commonly used kernel function that can effectively process nonlinearly separable data. The reason for selecting the RBF kernel function in this embodiment is that it has good generalization ability and flexibility and can adapt to various data distributions. The formula of the RBF kernel function is as follows:

[0060] K(x,y)=exp(-γ*||xy||2)

[0061] Where x and y are two data points, ||xy||2 is the square of their Euclidean distance, and γ is the kernel parameter that controls the width of the RBF kernel function.

[0062] Kernel parameter γ (Gamma): The γ value determines the influence range of the RBF kernel function. A smaller γ value indicates a larger influence range, and the model will be smoother and prone to underfitting; a larger γ value indicates a smaller influence range, and the model will be more complex and prone to overfitting. Therefore, choosing an appropriate γ value is crucial.

[0063] Penalty parameter ν (Nu): ν controls the sensitivity of the model to outliers. It indicates the upper limit of the proportion of training data that is allowed to be considered as outliers, and also indicates the lower limit of the proportion of support vectors. A smaller ν value means that the model is more sensitive to outliers and is prone to misclassifying normal data points as abnormal; a larger ν value means that the model is more tolerant to outliers and is prone to misclassifying abnormal data points as normal.

[0064] The grid search in this step is a hyperparameter optimization method that finds the best parameter value by trying different parameter combinations. For OC-SVM, you can use grid search to try different γ and ν values ​​and select the best combination based on the performance of the model.

[0065] Cross-validation is a model evaluation method that divides the data set into multiple subsets, uses one of the subsets as the test set and the remaining subsets as the training set in turn, and the final performance indicator is the average of multiple test results. Cross-validation can more accurately evaluate the generalization ability of the model and avoid overfitting.

[0066] Specifically, the above optimization process is as follows:

[0067] 1) Select RBF kernel function: First, determine to use RBF kernel function as the kernel function of One-Class SVM.

[0068] 2) Define parameter range: Define a reasonable search range for γ and ν. For example, you can set the range of γ to [0.001, 1000] and the range of ν to [0.01, 0.5].

[0069] 3) Grid search or cross validation: Use grid search or cross validation methods to try different combinations of γ and ν within the defined parameter range. For each combination, train an OC-SVM model and evaluate its performance using cross validation.

[0070] 4) Selecting the best parameters: Select the best performing values ​​of γ and ν and use these values ​​to train the final anomaly recognition model.

[0071] Through this optimization method, the optimal kernel parameter γ and penalty parameter ν can be found, thereby improving the accuracy and generalization ability of the One-ClassSVM model, enabling it to more effectively identify abnormal steel tapping situations that do not appear in the abnormal steel tapping image data.

[0072] Step S205: deploy the trained steel-tapping monitoring model and the abnormality recognition model into the actual production environment, perform real-time monitoring of the unmanned steel-tapping process, and provide steel-tapping process and alarm prompts.

[0073] Different from the above step S204, this step uses both the steel-tapping monitoring model and the abnormality recognition model to monitor the unmanned steel-tapping process in real time and provide the steel-tapping process and alarm prompts. This involves the comprehensive use of two models. In this embodiment, the master-slave collaboration mode can be used for monitoring:

[0074] Main model (steel tapping monitoring model): This model is trained based on normal and abnormal steel tapping data, and is mainly responsible for identifying known normal and abnormal steel tapping conditions, such as molten steel splashing, abnormal relative position between converter and ladle, molten steel overflow, slag overflow from furnace mouth, and too low clearance of molten steel level in ladle. It can classify the current steel tapping situation and determine whether an alarm is needed based on the preset threshold. For example, if the degree of molten steel splashing exceeds the threshold, an alarm of "abnormal steel tapping-severe splashing" is issued; if the relative position deviation between converter and ladle is too large, an alarm of "abnormal steel tapping-abnormal position" is issued; if molten steel overflow or slag overflow from furnace mouth is detected, an alarm of "abnormal steel tapping-molten steel / slag overflow" is issued; if the clearance of molten steel level in ladle is too low, an alarm of "abnormal steel tapping-low liquid level" is issued.

[0075] Auxiliary model (abnormal identification model - OC-SVM): This model is trained only on normal steel tapping data and is mainly responsible for identifying unknown abnormal situations, that is, those abnormal situations that have not appeared in the training data. It learns the normal steel tapping pattern and determines whether the current steel tapping situation deviates from the normal pattern. If the deviation exceeds a certain threshold, an "unknown abnormality" alarm is issued.

[0076] Collaborative work: The two models analyze the real-time camera images at the same time and perform collaborative analysis according to the following three situations:

[0077] Case 1: The main model identifies a known abnormality. If the main model identifies a known abnormality, such as "abnormal steel tapping - severe spatter", it will directly issue a corresponding alarm and take corresponding measures according to the preset strategy, such as stopping the steel tapping operation. At this time, the output of the auxiliary model can be used as a reference, but it does not play a decisive role.

[0078] Case 2: The primary model does not identify an abnormality, but the auxiliary model does. If the primary model believes that the current steel tapping situation is normal, but the auxiliary model identifies a large deviation from the normal pattern, an "unknown abnormality" alarm is issued to prompt the operator to pay attention and conduct further inspections. This can serve as an additional safety guarantee to prevent unknown risks.

[0079] Case 3: Both models do not identify an abnormality. If both models believe that the current steel tapping situation is normal, the steel tapping operation will continue.

[0080] From the above, it can be seen that this collaborative working mode can combine the advantages of the two models to improve the reliability and safety of the system. The main model is responsible for identifying known anomalies, and the auxiliary model is responsible for identifying unknown anomalies, thereby covering a more comprehensive range of abnormal situations, realizing comprehensive monitoring of the unmanned steelmaking process, and ultimately promoting the popularization and application of converter unmanned steelmaking technology.

[0081] The above embodiment discusses the method of using the steel-tapping monitoring model and the abnormality recognition model for monitoring. However, relying solely on image information may be limited by some environmental factors, such as lighting changes, camera occlusion, etc. Therefore, in order to improve the reliability and robustness of the system, the embodiment of the present application further considers the solution of multi-model fusion, combining the image recognition model (steel-tapping monitoring model) with the sensor-based model. Specifically, Figure 3 As shown, the unmanned converter steel tapping monitoring method based on artificial intelligence in this embodiment may include the following steps:

[0082] Step S301: Acquire historical image data of the converter steel tapping process, wherein the historical image data includes normal steel tapping image data and abnormal steel tapping image data.

[0083] Step S302: annotate and extract features from the historical image data, wherein the features include the degree of molten steel splashing, slag and steel overflow at the turning point, the relative position of the molten steel tank and the converter, the liquid level clearance in the molten steel tank, the time when the molten steel starts to flow out, and the time when the molten steel stops flowing out.

[0084] Step S303: Establish a database based on the extracted features, and use the database to train a steel tapping monitoring model, so that the steel tapping monitoring model learns to identify and judge normal steel tapping, abnormal steel tapping, the beginning and end of molten steel outflow, and the normal and abnormal positional relationship between the converter and the molten steel tank.

[0085] The above steps S301-S303 are the same as steps S101-S103 in the above embodiment, and will not be further described here.

[0086] Step S304: deploy the trained steel-tapping monitoring model and the abnormality recognition model into the actual production environment, and use sensors to obtain the weight, position, temperature and vibration data of the molten steel in the converter steel-tapping process, and based on the molten steel weight, position, temperature and vibration data and the monitoring results of the steel-tapping monitoring model, conduct a comprehensive analysis of the abnormal conditions in the steel-tapping process, and according to the analysis results, conduct real-time monitoring of the unmanned steel-tapping process and provide steel-tapping process and alarm prompts.

[0087] Unlike step S104, this step also uses sensor data for comprehensive analysis when actually monitoring the steel-making process. Image data can intuitively reflect the macroscopic situation in the steel-making process, such as molten steel splashing, positional relationship, etc. Sensor data can provide more detailed and quantitative information, such as weight change, temperature change, vibration amplitude, etc. Some abnormal conditions may not be easy to detect in image data, but can be reflected by sensor data. For example, abnormal vibration inside the converter may indicate damage or failure of certain components, even if there is no obvious abnormal performance in the image. For another example, the abnormality of the weight change rate of molten steel can be used to detect problems such as too fast or too slow molten steel pouring speed early. By comprehensively analyzing image data and sensor data, these potential abnormalities can be detected early, and timely measures can be taken to avoid accidents. Therefore, this embodiment combines the two for comprehensive analysis, which can more comprehensively and accurately judge abnormal conditions in the steel-making process and reduce false alarms and missed alarms. In addition, combined with sensor data, it can provide richer alarm information, such as "the temperature of the molten steel tank is too high", "abnormal vibration", "abnormal weight change of molten steel", etc., to help operators judge the type and severity of the abnormality more quickly and accurately, and take corresponding measures.

[0088] The sensor data here include the weight, position, temperature and vibration data of the molten steel during the steel-making process. The corresponding sensors are weight sensors, position sensors, temperature sensors and vibration sensors. The weight sensor can monitor the weight change of the molten steel tank in real time, so as to more accurately judge the outflow status and end time of the molten steel, and avoid the image recognition model from misjudging that it is still outflowing after the molten steel has run out. The position sensor can accurately measure the relative position of the molten steel tank and the converter, and make up for the error in position judgment of image recognition, especially when the camera's viewing angle is limited or the picture is blurred. The temperature sensor can monitor the temperature of the molten steel, and can determine whether the molten steel has reached the predetermined temperature, and whether there are abnormal conditions such as overheating. The vibration sensor can monitor the vibration of the molten steel tank and the converter, and can determine whether there are abnormal conditions such as splashing and spilling of molten steel, and serve as a supplement and verification of the image recognition model.

[0089] The comprehensive analysis method in this step can be implemented in a variety of ways, such as setting a threshold, weighted average or logical judgment. Setting a threshold means setting a threshold for the output results of the sensor data and the steel-out monitoring model respectively. If any data source exceeds the threshold, an alarm will be triggered. For example, if the steel-out monitoring model identifies that the molten steel splashes seriously, or the sensor detects that the converter vibrates violently, an alarm will be triggered and the steel-out operation will be stopped. Weighted average means that different weights are assigned to the output results of the sensor data and the steel-out monitoring model according to experience or experimental results, and then a weighted average is performed to obtain a comprehensive score, and whether to trigger an alarm is determined according to the level of the comprehensive score. Logical judgment means designing more complex logical judgment rules based on the output results of the sensor data and the steel-out monitoring model. For example, if the steel-out monitoring model identifies that the position of the ladle is abnormal, and the sensor detects that the weight of the molten steel suddenly drops, it is judged that the ladle is tilted too much, an alarm is triggered and the steel-out operation is stopped immediately. For another example, if the steel-out monitoring model identifies that the slag overflows from the furnace mouth, and the sensor detects that the temperature of the molten steel rises abnormally, it is judged that the slag reflux causes the temperature to rise, an alarm is triggered and corresponding processing is performed.

[0090] As can be seen from the above, the artificial intelligence-based converter unmanned steel-tapping monitoring method provided by the present invention can more comprehensively identify and judge various abnormal situations, including molten steel splashing, abnormal position relationship, excessive speed, abnormal weight, abnormal temperature, and abnormal vibration, etc., through the fusion analysis of AI model and sensor data. Once an abnormality is found, the system will immediately issue an alarm to prompt the operator to take measures to avoid accidents and ensure production safety.

[0091] The artificial intelligence-based converter unmanned steel-tapping monitoring methods in the above-mentioned embodiments can only provide an alarm function after an abnormality occurs. However, if an early warning function can be provided during the monitoring process, the safety of production can be further improved, accidents can be further avoided, and the safety of production can be further guaranteed. Figure 1 and Figure 3 In the corresponding embodiment, when the steel tapping monitoring model is used for actual monitoring, the following steps may also be included: issuing an early warning based on the positional relationship between the molten steel ladle and the converter identified by the steel tapping monitoring model, including: if the distance between the mouth of the molten steel ladle and the converter mouth exceeds a set threshold, issuing an early warning signal.

[0092] Through the above steps, this embodiment can give an early warning before the abnormality occurs, thereby avoiding the occurrence of the abnormality, and thus directly avoiding the occurrence of related accidents, so that the unmanned steel-making technology of the converter can be promoted and applied.

[0093] The core of the above-described method for unmanned steel-tapping monitoring of a converter based on artificial intelligence is to enable the AI ​​model to identify and judge abnormal conditions in the steel-tapping process by training a large number of normal and abnormal steel-tapping images. However, in reality, since abnormal steel-tapping situations are relatively rare, in order to improve the accuracy of the model in identifying abnormal situations, this embodiment further adopts data enhancement technology. That is, after obtaining the historical image data of the converter steel-tapping process in the above steps S101, S201 or S301, the method of this embodiment may also include the following steps: performing data enhancement operations on the abnormal steel-tapping image data in the historical image data to increase the amount of abnormal steel-tapping image data.

[0094] Preferably, Figure 4 As shown, the data enhancement operations here include:

[0095] Step S401: performing image-based enhancement operations on the abnormal steel tapping image data: including rotating and translating, scaling, adding noise, blurring and blocking operations on the abnormal steel tapping image data to generate new abnormal steel tapping image data. Wherein:

[0096] Rotation and translation: The abnormal steel-tapping image is rotated and translated at a small angle, which can simulate the steel-tapping scenes under different viewing angles.

[0097] Zoom: Zoom the image, which can simulate the steel tapping scene at different distances.

[0098] Adding noise: adding Gaussian noise, salt and pepper noise, etc., which can simulate image interference in actual production environments.

[0099] Brightness and contrast adjustment: adjust the brightness and contrast of the image, which can simulate the steel-making scenes under different lighting conditions.

[0100] Blur: Blur the image, which can simulate camera focus inaccuracy or image shaking.

[0101] Occlusion: It simulates the situation where molten steel splashes or other objects block the camera's view, which can partially block the image.

[0102] Step S402: performing a video-based enhancement operation on the abnormal steel tapping image data: playing the video segment to which the abnormal steel tapping image data belongs at a variable speed and then generating new abnormal steel tapping image data.

[0103] Here, by playing the video clips of abnormal steel tapping at variable speed, the change of steel tapping speed can be simulated. In addition, different abnormal steel tapping video clips can be combined to generate new abnormal samples.

[0104] Step S403: performing feature-based enhancement operation on the abnormal steel tapping image data: generating new abnormal steel tapping image data based on the SMOTE method.

[0105] Preferably, Figure 5 As shown, this step may further include the following sub-steps:

[0106] Step S4031: extracting enhanced features from the abnormal steel tapping image data, wherein the enhanced features include: color histogram, texture features (such as gray level co-occurrence matrix GLCM), edge features (such as Canny edge detection), shape features (such as Hu moment), and key point features (such as SIFT, SURF).

[0107] Step S4032: using the enhanced feature vector as an input of the SMOTE algorithm to generate a new feature vector of the abnormal steel tapping image sample.

[0108] After feature extraction, the feature vector of the abnormal steel-tapping sample is used as the input of the SMOTE algorithm. The basic steps of the SMOTE algorithm are as follows: for each minority class sample (i.e., abnormal steel-tapping sample), find its k nearest neighbors (k is usually 5); randomly select a nearest neighbor sample; randomly select a point on the line between the minority class sample and the selected nearest neighbor sample as a new minority class sample.

[0109] By repeating the above steps, a specified number of feature vectors of new abnormal steel tapping image samples can be generated.

[0110] Suppose two features are extracted: the angle and speed of the molten steel splash. The feature vector of an abnormal steel tapping sample is (30°, 10m / s). The feature vector of one of its nearest neighbor samples is (40°, 12m / s). The SMOTE algorithm can randomly select a point on the line between the two feature vectors, such as (33°, 10.6m / s), as the feature vector of the new abnormal steel tapping sample.

[0111] Step S4033: input the feature vector into a generative model trained by a generative adversarial network to convert it into a new abnormal steel tapping image.

[0112] It can be seen that by combining feature extraction and SMOTE algorithm, new abnormal steel-tapping samples can be effectively generated, thereby improving the accuracy and reliability of the converter unmanned steel-tapping monitoring system.

[0113] like Figure 6The structure diagram of an artificial intelligence-based converter unmanned steel-tapping monitoring device provided in an embodiment of the present application is shown, and the device includes: a historical data acquisition unit 610, a feature extraction unit 620, a model training unit 630 and a model deployment unit 640, which are connected in sequence. Among them:

[0114] The historical data acquisition unit 610 is used to acquire historical image data of the converter steel tapping process, wherein the historical image data includes normal steel tapping image data and abnormal steel tapping image data.

[0115] The feature extraction unit 620 is used to annotate and extract features from the historical image data, wherein the features include the degree of molten steel splashing, slag and steel overflow at the turning point, the relative position of the molten steel tank and the converter, the liquid level clearance in the molten steel tank, the time when the molten steel starts to flow out, and the time when the molten steel stops flowing out.

[0116] The model training unit 630 is used to establish a database based on the extracted features, and use the database to train a steel-tapping monitoring model, so that the steel-tapping monitoring model learns to identify and judge normal steel-tapping, abnormal steel-tapping, the beginning and end of molten steel outflow, and the normal and abnormal positional relationship between the converter and the molten steel tank.

[0117] The model deployment unit 640 is used to deploy the trained steel-tapping monitoring model into the actual production environment, to monitor the unmanned steel-tapping process in real time and to provide steel-tapping process and alarm prompts.

[0118] Preferably, the prompts of the above-mentioned steel-tapping process include: the start stage and the end stage of steel-tapping; the alarm prompts include: severe molten steel splashing, slag overflow from the converter mouth, molten steel overflow from the switchback, excessive deviation of the position of the molten steel tank, and abnormal clearance of the molten steel level in the molten steel tank.

[0119] Preferably, the above-mentioned device also includes: an abnormality recognition unit, based on the normal steel tapping image data in the historical image data, using a single-classification support vector machine algorithm to train an abnormality recognition model to identify abnormal steel tapping conditions that do not appear in the abnormal steel tapping image data.

[0120] Preferably, the above-mentioned single-classification support vector machine algorithm adopts a radial basis kernel function, and optimizes the kernel parameter γ and the penalty parameter ν by a grid search or cross-validation method to determine the optimal anomaly recognition model.

[0121] Preferably, the above-mentioned device also includes: a sensor unit, which is used to use sensors to obtain the weight, position, temperature and vibration data of the molten steel during the steel tapping process of the converter, and based on the molten steel weight, position, temperature and vibration data and the monitoring results of the steel tapping monitoring model, conduct a comprehensive analysis of abnormal conditions in the steel tapping process.

[0122] Preferably, the above-mentioned device also includes: an early warning unit, which is used to issue an early warning based on the positional relationship between the molten steel ladle and the converter identified by the steel tapping monitoring model, including: if the distance between the mouth of the molten steel ladle and the converter mouth exceeds a set threshold, a early warning signal is issued.

[0123] Preferably, the above-mentioned device further comprises: a data enhancement unit, which is used to perform data enhancement operation on the abnormal steel tapping image data in the historical image data to increase the amount of abnormal steel tapping image data.

[0124] Preferably, the above-mentioned data enhancement operation includes: performing an image-based enhancement operation on the abnormal steel tapping image data: including rotating and translating, scaling, adding noise, blurring and occluding operations on the abnormal steel tapping image data to produce new abnormal steel tapping image data; performing a video-based enhancement operation on the abnormal steel tapping image data: playing the video clip to which the abnormal steel tapping image data belongs at a variable speed and then producing new abnormal steel tapping image data; performing a feature-based enhancement operation on the abnormal steel tapping image data: producing new abnormal steel tapping image data based on the SMOTE method.

[0125] Preferably, the above-mentioned production of new abnormal steel tapping image data based on the SMOTE algorithm includes: extracting enhanced features from the abnormal steel tapping image data, the enhanced features including: color histogram, texture features, edge features, shape features, key point features; using the vector of the enhanced features as input to the SMOTE algorithm to generate a feature vector of a new abnormal steel tapping image sample; inputting the feature vector into a generative model trained by a generative adversarial network to convert it into a new abnormal steel tapping image.

[0126] The detailed description of each of the above units can be found in the corresponding description in the aforementioned method embodiment, which will not be further elaborated here.

[0127] It can be seen from the above technical solutions that the artificial intelligence-based converter unmanned steel-tapping monitoring device provided by the present invention monitors various states in the steel-tapping process in real time through an artificial model, identifies normal / abnormal steel-tapping, the positional relationship between the steel ladle and the converter, the converter rotation speed, etc., and issues an alarm when an abnormal situation occurs, thereby avoiding safety accidents and ensuring the safety of unmanned operation. It fills the gap that the existing unmanned steel-tapping system lacks a complete monitoring system.

[0128] Figure 7 is a schematic diagram of an electronic device provided by an embodiment of the present invention. Figure 5The electronic device shown is a general data processing device, which includes a general computer hardware structure, which at least includes a processor 801 and a memory 802. The processor 801 and the memory 802 are connected via a bus 803. The memory 802 is suitable for storing one or more instructions or programs executable by the processor 801. The one or more instructions or programs are executed by the processor 801 to implement the steps in the above-mentioned artificial intelligence-based converter unmanned steel tapping monitoring method.

[0129] The processor 801 may be an independent microprocessor or a collection of one or more microprocessors. Thus, the processor 801 executes the commands stored in the memory 802 to execute the method flow of the embodiment of the present invention as described above to realize the processing of data and the control of other devices. The bus 803 connects the above-mentioned multiple components together, and at the same time connects the above-mentioned components to the display controller 804 and the display device and the input / output (I / O) device 805. The input / output (I / O) device 805 may be a mouse, a keyboard, a modem, a network interface, a touch input device, a somatosensory input device, a printer, and other devices known in the art. Typically, the input / output (I / O) device 805 is connected to the system via an input / output (I / O) controller 806.

[0130] The memory 802 may store software components such as an operating system, a communication module, an interaction module, and an application program. Each of the modules and applications described above corresponds to a set of executable program instructions that implement one or more functions and methods described in the embodiments of the invention.

[0131] An embodiment of the present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned artificial intelligence-based converter unmanned steel tapping monitoring method are implemented.

[0132] An embodiment of the present invention also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the above-mentioned artificial intelligence-based converter unmanned steel tapping monitoring method.

[0133] In summary, the artificial intelligence-based converter unmanned steel-tapping monitoring method and device provided by the present invention monitors various states in the steel-tapping process in real time through an artificial model, identifies normal / abnormal steel-tapping, the positional relationship between the steel ladle and the converter, the converter rotation speed, etc., and issues an alarm when an abnormal situation occurs, thereby avoiding safety accidents and ensuring the safety of unmanned operation. This fills the gap that the existing unmanned steel-tapping system lacks a complete monitoring system.

[0134] The preferred embodiments of the present invention have been described above with reference to the accompanying drawings. Many features and advantages of these embodiments are clear from this detailed description, and therefore the claims are intended to cover all such features and advantages of these embodiments that fall within their true spirit and scope. In addition, since many modifications and changes are readily apparent to those skilled in the art, the embodiments of the present invention are not intended to be limited to the precise structure and operation illustrated and described, but all suitable modifications and equivalents that fall within their scope may be included.

[0135] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0136] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0137] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0138] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0139] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A converter unmanned steel tapping monitoring method based on artificial intelligence, characterized in that: The method comprises: Acquire historical image data of the converter steel tapping process, wherein the historical image data includes normal steel tapping image data and abnormal steel tapping image data; Annotating and extracting features from the historical image data, wherein the features include the degree of molten steel splashing, slag and steel overflow at the turning point, the relative position of the molten steel tank and the converter, the liquid level clearance in the molten steel tank, the time when the molten steel starts to flow out, and the time when the molten steel ends to flow out; A database is established based on the extracted features, and a steel tapping monitoring model is trained using the database, so that the steel tapping monitoring model learns to identify and judge normal steel tapping, abnormal steel tapping, the start and end of molten steel outflow, and the normal and abnormal positional relationship between the converter and the molten steel tank; The trained steel-tapping monitoring model is deployed in the actual production environment to monitor the unmanned steel-tapping process in real time and provide steel-tapping process and alarm prompts.

2. The artificial intelligence-based converter unmanned steel tapping monitoring method according to claim 1, characterized in that: The prompts of the steel-tapping process include: the beginning stage of steel-tapping and the end stage of steel-tapping; the alarm prompts include: severe molten steel splashing, slag overflow from the converter mouth, molten steel overflow from the switchback, excessive deviation of the position of the molten steel tank, and abnormal clearance of the molten steel level in the molten steel tank.

3. The artificial intelligence-based converter unmanned steel-tapping monitoring method according to claim 1, characterized in that: The method further includes: based on the normal steel-tapping image data in the historical image data, using a single-classification support vector machine algorithm to train an abnormality recognition model to recognize abnormal steel-tapping situations that do not appear in the abnormal steel-tapping image data.

4. The artificial intelligence-based converter unmanned steel tapping monitoring method according to claim 3, characterized in that: The single-class support vector machine algorithm adopts a radial basis kernel function and optimizes the kernel parameter γ and the penalty parameter ν by a grid search or cross-validation method to determine the optimal anomaly recognition model.

5. The artificial intelligence-based converter unmanned steel tapping monitoring method according to claim 1, characterized in that: The method also includes: using sensors to obtain the weight, position, temperature and vibration data of molten steel during the converter steel tapping process, and based on the molten steel weight, position, temperature and vibration data and the monitoring results of the steel tapping monitoring model, comprehensively analyzing abnormal conditions in the steel tapping process.

6. The artificial intelligence-based converter unmanned steel tapping monitoring method according to claim 1, characterized in that: The method further includes: according to the positional relationship between the molten steel ladle and the converter identified by the steel tapping monitoring model, including: if the distance between the mouth of the molten steel ladle and the converter mouth exceeds a set threshold, issuing an early warning signal.

7. The artificial intelligence-based converter unmanned steel tapping monitoring method according to claim 1, characterized in that: After acquiring the historical image data of the converter steel tapping process, the method further includes: A data enhancement operation is performed on the abnormal steel tapping image data in the historical image data to increase the amount of the abnormal steel tapping image data.

8. The artificial intelligence-based converter unmanned steel tapping monitoring method according to claim 7, characterized in that: The data enhancement operation includes: Performing image-based enhancement operations on the abnormal steel tapping image data: including rotating and translating, scaling, adding noise, blurring and blocking operations on the abnormal steel tapping image data to generate new abnormal steel tapping image data; Performing a video-based enhancement operation on the abnormal steel tapping image data: playing the video clip to which the abnormal steel tapping image data belongs at a variable speed and then generating new abnormal steel tapping image data; Performing feature-based enhancement operations on the abnormal steel tapping image data: generating new abnormal steel tapping image data based on the SMOTE method.

9. The artificial intelligence-based converter unmanned steel tapping monitoring method according to claim 8, characterized in that: The new abnormal steel tapping image data generated based on the SMOTE algorithm includes: Extracting enhanced features from the abnormal steel tapping image data, wherein the enhanced features include: color histogram, texture features, edge features, shape features, and key point features; Using the enhanced feature vector as an input of the SMOTE algorithm to generate a feature vector of a new abnormal steel tapping image sample; The feature vector is input into a generative model trained by a generative adversarial network to be converted into a new abnormal steel tapping image.

10. An unmanned converter steel-tapping monitoring device based on artificial intelligence, characterized in that: The device comprises: A historical data acquisition unit, used to acquire historical image data of the converter steel tapping process, wherein the historical image data includes normal steel tapping image data and abnormal steel tapping image data; A feature extraction unit is used to annotate and extract features from the historical image data, wherein the features include the degree of molten steel splashing, slag and steel overflow at the turning point, the relative position of the molten steel tank and the converter, the liquid level clearance in the molten steel tank, the time when the molten steel starts to flow out, and the time when the molten steel ends to flow out; A model training unit, for establishing a database based on the extracted features, and using the database to train a steel-tapping monitoring model, so that the steel-tapping monitoring model learns to identify and judge normal steel-tapping, abnormal steel-tapping, the start and end of molten steel outflow, and the normal and abnormal positional relationship between the converter and the molten steel tank; The model deployment unit is used to deploy the trained steel-tapping monitoring model into the actual production environment, to monitor the unmanned steel-tapping process in real time and to provide steel-tapping process and alarm prompts.

11. An electronic 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 computer program, the steps of the method according to any one of claims 1 to 9 are implemented.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 9 are implemented.

13. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 9 are implemented.

Images