Control method for continuous casting process of stainless steel

Abstract

The invention relates to a control method for a continuous casting process of stainless steel, which comprises the following steps of: acquiring first information of defects of a continuous casting billet and time sequence parameters of a casting process of the continuous casting billet, and establishing a correlation database of each defect and the corresponding time sequence parameters according to the first information and the time sequence parameters; according to the correlation database, obtaining a model in which each defect is correlated with the time sequence parameters; the model comprises the following second information: an influence weight of a time sequence parameter corresponding to each defect on the defect, a cause path of each defect and a confidence coefficient corresponding to the cause path; and controlling the continuous casting process of the stainless steel according to the model to reduce the generation of defects. The method is used for continuous casting production control, and the purposes of producing high-quality stainless steel continuous casting billets, reducing the grinding ratio of the continuous casting billets and reducing the stainless steel production cost can be achieved.

Description

Technical Field

[0001] This invention belongs to the field of stainless steel continuous casting technology, and specifically relates to a method for controlling the continuous casting process of stainless steel. Background Technology

[0002] Stainless steel is a high-alloy steel containing a certain amount of chromium or other metals such as chromium, nickel, and molybdenum, and its corrosion resistance is significantly superior to that of ordinary carbon steel. Stainless steel is widely used in daily life and industrial manufacturing fields such as kitchenware, decoration, transportation, and chemical manufacturing.

[0003] Stainless steel products typically use their stainless steel surface directly, without further painting or spraying, thus requiring high surface quality. Because stainless steel contains high levels of valuable metals such as chromium, nickel, and molybdenum, both raw material and manufacturing costs are high. Furthermore, continuously cast billets have more surface defects than ordinary carbon steel billets. Therefore, before rolling, the surface of the continuously cast billet usually needs to be ground, typically removing a certain depth of the outermost layer of the stainless steel billet using a grinding wheel, usually about 1mm thick. However, if the surface quality of the continuously cast billet is good, it can be rolled without grinding, which effectively reduces the cost of stainless steel products and improves production efficiency. Summary of the Invention

[0004] In order to solve all or some of the above problems, the present invention aims to provide a method for controlling the continuous casting process of stainless steel.

[0005] According to one aspect of the present invention, a method for controlling a continuous casting process of stainless steel is provided, comprising:

[0006] The first information on defects in continuously cast billets and the timing parameters of the casting process of continuously cast billets are obtained. The first information includes the type of defect, the characteristics of each defect, and the location distribution of each defect.

[0007] Based on the first information and the timing parameters, establish a relational database for each defect and its corresponding timing parameters;

[0008] Based on the relevant relational database, a model is obtained that correlates each defect with the time series parameters; the model includes the following second information: the influence weight of the time series parameters corresponding to each defect on the defect, the cause path of each defect, and the confidence level corresponding to the cause path;

[0009] The continuous casting process of stainless steel is controlled according to the model to reduce the generation of defects.

[0010] Furthermore, before obtaining the first information on defects in the continuously cast billet and the timing parameters of the continuous casting process, the method further includes:

[0011] Collect monitoring videos of the surface quality of continuously cast billets.

[0012] Furthermore, the first information on the defects in the continuously cast billet specifically includes:

[0013] The monitoring video is processed to obtain the first information about defects in the continuously cast billet.

[0014] Furthermore, the specific steps of processing the monitoring video to obtain the first information about the defects in the continuously cast billet are as follows:

[0015] Image analysis is performed on the surveillance video to obtain each type of defect and its corresponding location.

[0016] Based on each defect corresponding to the surveillance video, the characteristics of the defect are obtained.

[0017] Furthermore, the step of establishing a relational database for each defect and its corresponding time series parameters based on the first information and the time series parameters specifically involves:

[0018] Based on the first information and the time series parameters, a database of relationships between each defect and its corresponding time series parameters is established using big data analysis methods.

[0019] Furthermore, the specific steps of obtaining the model that correlates each defect with the time-series parameters based on the relevant relational database are as follows:

[0020] Based on the aforementioned relational database, the influence weight of the time series parameters corresponding to each defect on that defect is obtained;

[0021] Based on the process rules of stainless steel continuous casting, the cause path of each defect and the confidence level corresponding to the cause path are obtained.

[0022] Based on the confidence level and the influence weight, a model is obtained that correlates each defect with the time series parameters.

[0023] Furthermore, the step of obtaining the cause path of each defect and the confidence level corresponding to the cause path according to the process rules of the stainless steel continuous casting process specifically involves: obtaining the process rules of the stainless steel continuous casting process, and obtaining the cause path of each defect and the confidence level corresponding to the cause path according to the process rules.

[0024] Furthermore, the specific steps of controlling the continuous casting process of stainless steel according to the model to reduce the generation of defects are as follows:

[0025] Monitor the continuous casting process of stainless steel; issue alarm information after detecting defects; and adjust the corresponding timing parameters after receiving the alarm information and observe whether the corresponding defects are eliminated.

[0026] Furthermore, after controlling the continuous casting process of stainless steel according to the model to reduce the generation of defects, the method further includes:

[0027] The model is modified based on the defects detected in the continuous casting process of stainless steel.

[0028] As can be seen from the above technical solution, the method for controlling the continuous casting process of stainless steel provided by the present invention has the following beneficial effects:

[0029] This invention establishes a database of the correlation between each type of defect and its corresponding timing parameters. Then, the surface defects are detected by the online surface quality monitoring system for continuous casting billets, which is used to guide the real-time adjustment of continuous casting timing parameters.

[0030] This invention adjusts the continuous casting sequence parameters and observes whether the surface defects of the billet are eliminated. Based on the results, a model for good billet surface quality can be established and used for continuous casting production control to achieve the goals of producing high-quality stainless steel continuous casting billets, reducing the billet grinding ratio, and lowering the production cost of stainless steel.

[0031] This invention can monitor defects generated during the continuous casting process, promptly detect problems such as uncontrolled timing parameters or improper parameter matching during the continuous casting process, and provide early warning information through the model so as to dynamically adjust the corresponding continuous casting timing parameters. By continuously iterating the model data, the accuracy and hit rate of the model are improved, thereby establishing a continuous casting parameter optimization control model based on actual production conditions, and producing high-quality stainless steel continuous casting billets. Attached Figure Description

[0032] Figure 1 This is a flowchart of a control method for a continuous casting process of stainless steel according to an embodiment of the present invention;

[0033] Figure 2 This is a schematic diagram of a control method for a continuous casting process of stainless steel according to an embodiment of the present invention;

[0034] Figure 3 This is a schematic diagram of the defect corresponding to Example 1;

[0035] Figure 4 This is a schematic diagram of the defect corresponding to Example 2. Detailed Implementation

[0036] To fully understand the purpose, features, and effects of this invention, the following detailed embodiments are provided. Except as described below, the process methods of this invention employ conventional methods or apparatus in the art. Unless otherwise specified, the terms and expressions used below have the meanings commonly understood by those skilled in the art.

[0037] When a range of values ​​is disclosed herein, the range is considered continuous and includes the minimum and maximum values ​​of the range, as well as every value between the minimum and maximum values. Furthermore, when the range refers to integers, it includes every integer between the minimum and maximum values ​​of the range. Additionally, when multiple ranges are provided to describe a feature or characteristic, the ranges may be combined. In other words, unless otherwise specified, all ranges disclosed herein should be understood to include any and all subranges to which they are incorporated.

[0038] 304 and 430 are the steel grades with the best overall performance and manufacturing cost among chromium-nickel austenitic and ferritic stainless steels, respectively, and are also the most produced steel grades currently. Solving the surface quality problem of 304 and 430 continuously cast billets essentially solves the surface quality problem of continuously cast billets for conventional stainless steel varieties. The surface quality of stainless steel continuously cast billets is highly correlated with key control parameters of the continuous casting process and the precision of the continuous casting machine. Therefore, under the premise of ensuring the good condition of the main equipment of the continuous casting machine, dynamically and reasonably adjusting key control parameters such as continuous casting vibration parameters, casting speed, casting temperature, and selection of protective slag is the key to obtaining good surface quality of stainless steel continuously cast billets. Adjusting the continuous casting control parameters can also intervene and adjust the surface quality of the continuously cast billets to a certain extent.

[0039] The idea behind this invention is to establish a correspondence between surface defects of continuously cast billets of the two most typical steel grades, 304 and 430, and the timing parameters of continuous casting operation; to dynamically adjust the timing parameters of continuous casting by online monitoring of surface defects of the billets; and thereby achieve the goal of producing high-quality stainless steel continuously cast billets, reducing the billet grinding ratio, reducing billet grinding losses, and reducing the production cost of stainless steel.

[0040] The probability P of any type of defect occurring on the surface of 304 and 430 continuously cast billets can be expressed by the following formula:

[0041] P=∑(k i *i*F i +(C I +Q i )*R i -τ i *S i ) Formula 1

[0042] In formula 1, k i F represents the coefficient of influence of factor i, where factor i is a timing parameter of the continuous casting process, such as casting speed, tundish temperature, and vibration parameter combinations. i C represents the confidence level of factor i. i Q represents the weights of factor i in a CNN (convolutional neural network). i R represents the temporal feature weights of factor i in an LSTM (Long Short-Term Memory) network. iτ represents the path weight of factor i. i S represents the tolerance correction coefficient for factor i. i This represents the tolerance value of factor i.

[0043] According to the above formula, as Figures 1-2 As shown, the control method for the continuous casting process of stainless steel according to an embodiment of the present invention includes the following steps:

[0044] Step S001: Obtain the first information on defects in the continuously cast billet and the timing parameters of the continuous casting billet casting process. The first information includes the type of defect, the characteristics of each defect, and the location distribution of each defect. The timing parameters include at least one of the following: casting speed, casting temperature, cooling intensity, vibration parameters, nozzle insertion depth, and inlet / outlet water temperature difference. The characteristics of each defect are the length, width, and depth of each defect.

[0045] Step S002: Based on the first information and time series parameters, establish a relational database for each type of defect and its corresponding time series parameters;

[0046] Step S003: Based on the relevant relational database, obtain the model that correlates each defect with the time series parameters; the model includes the following second information: the influence weight of the time series parameters corresponding to each defect on the defect, the cause path of each defect, and the confidence level corresponding to the cause path;

[0047] Step S004: Control the continuous casting process of stainless steel according to the model to reduce the generation of defects.

[0048] Specifically, before obtaining the first information on the defects of the continuously cast billet in step S001, the method of this embodiment of the invention further includes: acquiring a monitoring video of the surface quality of the continuously cast billet.

[0049] In practice, the surface quality monitoring video of the continuously cast billets is collected through an online monitoring system for the surface quality of 304 and 430 stainless steel continuously cast billets.

[0050] Step S001, obtaining the first information about the defects in the continuously cast billet, specifically involves processing the monitoring video to obtain the first information about the defects in the continuously cast billet.

[0051] The first information about defects in continuously cast billets is obtained by processing the monitoring video: image analysis is performed on the monitoring video to obtain each type of defect and its corresponding location; based on each type of defect in the monitoring video, the characteristics of the defect are obtained.

[0052] Specifically, for monitoring videos, for example, starting from a set time T0, images are captured every t seconds, and then image analysis is performed on the captured images to obtain information such as the types, characteristics, and distribution locations of typical defects in continuously cast billets. In practice, this information can be recorded in a table as shown in Table 1:

[0053] Table 1: Table of Image Analysis

[0054]

[0055] To address surface defects in continuously cast billets, 20-megapixel high-definition images (cracks, dents, inclusions, scale, etc.) acquired by a video acquisition system are used for feature extraction using a ResNet-34 pre-trained model. Dimensionality reduction is achieved through convolutional layers (Kernel = 3×3, Stride = 2) and max pooling (Pooling = 2×2), resulting in a 256-dimensional feature vector.

[0056] After the above processing, first information on defects in the continuously cast billet, including the type of defect, the characteristics of each defect, and the location distribution of each defect, is obtained. Step S001 of this embodiment of the invention further includes obtaining the timing parameters of the continuous casting billet casting process. These timing parameters include at least one of the following: casting speed, casting temperature, cooling intensity, vibration parameters, nozzle insertion depth, and inlet / outlet water temperature difference, as well as information on the corresponding continuous casting steel grade.

[0057] Specifically, step S002, which establishes a relational database of each defect and its corresponding time series parameters based on the first information and time series parameters, involves using big data analysis methods to establish a relational database of each defect and its corresponding time series parameters.

[0058] Specifically, by using big data analysis methods such as CNN (Convolutional Neural Network), LSTM (Long Short-Term Memory Network), and CART (Classification and Regression), we can achieve dual analysis and connection of time-series parameters and primary information, and establish a database of correlations between different surface defects and key control parameters.

[0059] Step S003, based on the relevant relational database, obtains the specific model of the correlation between each defect and the time series parameters as follows:

[0060] Based on the relevant relational database, the influence weight of the time series parameters corresponding to each type of defect is obtained; based on the process rules of stainless steel continuous casting, the causal path of each type of defect and the confidence level corresponding to the causal path are obtained; based on the confidence level and influence weight, the model of the relationship between each type of defect and the time series parameters is obtained.

[0061] Based on the process rules of stainless steel continuous casting, the cause path of each defect and the corresponding confidence level are obtained. Specifically, the process rules of stainless steel continuous casting are obtained, and based on the process rules, the cause path of each defect and the corresponding confidence level are obtained.

[0062] The model obtained in this embodiment has the following functions:

[0063] Defect morphology analysis function: Through gradient weighted class activation mapping (Grad-CAM) technology, the initiation position of cracks (such as 3-8 meters from the head of the billet and 100-300 mm from both sides of the billet) can be visualized, thereby locating the high-incidence and easy-to-occur areas of defects, so as to modify the corresponding timing parameters in time before defects occur.

[0064] Enhanced fusion and fitting output functionality through key control point attention: A cross-attention layer is designed to dynamically allocate weights between LSTM output and CNN features, thereby improving the accuracy of defect identification. For example, when the liquid surface fluctuates violently (fluctuation amplitude > 3mm), the weight of the CNN features decreases to 0.3, while the weight of the LSTM temporal features increases to 0.5.

[0065] Output layer optimization function: The Softmax function is used to output the probability of key defects (such as cracks, inclusions, scabs, depressions, and porosity). The threshold adjustment strategy is dynamically changed according to the steel grade (e.g., the crack warning threshold for 430 series stainless steel is set to 0.5, and the crack warning threshold for 304 stainless steel is set to 0.3).

[0066] The model in step S003 was obtained by a rule-based decision-making engine designed by continuous casting process experts, based on a relevant relational database. Specifically:

[0067] For example, process experts can integrate 30+ process rules to form a multi-layered decision logic.

[0068] For example, regarding crack defects, based on their cause path, the corresponding determination path is:

[0069] If the liquid level fluctuation is greater than ±3 mm and the superheat is greater than 40℃, then a crack warning will be triggered (confidence level +0.4).

[0070] If the crystallizer vibration spectrum is abnormal (frequency > 120 Hz), then increase the confidence level by 0.3.

[0071] If the overall confidence level is greater than 0.75, then it is determined to be a crack defect and an early warning is issued.

[0072] The model for this invention can adaptively and dynamically adjust the threshold based on the production conditions: for example, during the tundish change operation, the model’s tolerance for pull speed fluctuations is relaxed to ±30%, and the tolerance for tundish temperature fluctuations is increased to ±10%.

[0073] The model in this embodiment establishes the interrelationship between process parameters: a process parameter correlation network with 300+ nodes is constructed, the causal relationship between parameters is quantified, and the weights of the corresponding paths are set; for example, the path weight of "insufficient cooling water flow in the crystallizer → increased temperature gradient at the solidification front → increased risk of cracking" is 0.82.

[0074] The model in this embodiment of the invention employs multi-dimensional data analysis technology, combined with a process expert knowledge base and artificial intelligence algorithms, to locate the causes of defects. The core method includes:

[0075] CART decision tree algorithm:

[0076] By constructing defect cause paths using Classification and Regression Tree (CART), the correlation between process parameters and defects is quantified. For example, for "slag entrapment defect", the algorithm identifies "pulling speed fluctuation > 0.4 m / min", "crystallizer liquid level fluctuation > ±3 mm" and "sprue alignment deviation > 2 mm" as key root causes. The key root causes have relatively large weights, and the corresponding path probabilities reach 93%.

[0077] Bayesian networks and causal inference:

[0078] Based on the Bayesian probability model, the probability of longitudinal crack defects caused by the coupling effect of multiple factors, such as "carbon content greater than 0.06%, superheat > 45℃ and tensile speed greater than 1.25m / s", is 76%.

[0079] Therefore, the model in this embodiment supports reverse reasoning, that is, inferring the most likely abnormal combination of process parameters from the defect results.

[0080] Step S004: Controlling the continuous casting process of stainless steel according to the model to reduce the generation of defects specifically involves:

[0081] Monitor the continuous casting process of stainless steel; issue alarm information after detecting defects; adjust the corresponding timing parameters after receiving alarm information and observe whether the corresponding defects are eliminated.

[0082] The purpose of issuing alarm information in this embodiment is to prompt operators to adjust parameters to prevent the continuous occurrence of defects.

[0083] After controlling the continuous casting process of stainless steel according to the model to reduce the generation of defects, the method of this embodiment of the invention further includes: correcting the model according to the defects of the continuous casting process of stainless steel obtained by monitoring.

[0084] By monitoring the surface quality of continuously cast billets, adjusting key timing parameters, and improving the surface quality of continuously cast billets, a self-feedback and self-learning mechanism is established for the model. This continuously corrects and optimizes the model, improving its control accuracy, response speed, and application level. Applying the model to continuous casting production control can achieve the goals of producing high-quality stainless steel continuously cast billets, reducing the billet grinding ratio, and lowering stainless steel production costs.

[0085] The embodiments of the present invention have the following advantages:

[0086] By using big data analysis methods such as CNN (Convolutional Neural Network), LSTM (Long Short-Term Memory Network), and CART (Classification and Regression), we can achieve dual analysis and connection between continuous casting control timing parameters and slab surface defect characteristics, establish a database of correlations between different surface defects and key timing parameters, and then detect surface defects through the online surface quality monitoring system for continuous casting slabs to guide real-time adjustment of continuous casting timing parameters.

[0087] By adjusting the continuous casting timing parameters, we can observe whether the surface defects of the billet are eliminated or reduced, establish a model to achieve good billet surface quality, and use it for continuous casting production control to achieve the goal of producing high-quality stainless steel continuous casting billets, reducing the billet grinding ratio, and reducing stainless steel production costs.

[0088] This invention can monitor surface quality defects in continuous casting online, promptly detect out-of-control or improperly matched key timing parameters during the continuous casting process, issue early warning information through the model, and dynamically adjust the corresponding continuous casting timing parameters in manual / automatic modes. By continuously iterating the model data, the accuracy and hit rate of the model are improved, thereby establishing a continuous casting parameter optimization control model based on actual production conditions, and producing high-quality stainless steel continuous casting billets.

[0089] The following is a specific example:

[0090] Example 1: During a casting of 304 stainless steel, video monitoring revealed intermittent scabbing defects on the south side of the continuously cast billet, 150mm from the narrow edge, 6 minutes after the second heat was started. Figure 3 As shown, the model yielded the following conclusions: the casting temperature was 5℃ lower than the control threshold, which may have caused poor melting of the protective slag. Therefore, the solution was to increase the casting speed to 0.05m / s and increase the insertion depth of the crystallizer nozzle by 5mm. After adjusting the parameters for 3 minutes, video monitoring showed that the scabbing defect disappeared and the surface of the continuously cast billet returned to normal.

[0091] Example 2: During a certain casting batch of 430, video monitoring 10 minutes after the first heat started revealed intermittent depressions / longitudinal cracks on the north side of the continuously cast billet, 200-300mm from the narrow edge, and in the middle of the billet. Figure 4As shown, the model suggests that there may be reasons such as poor lubrication of the mold mold flux, possible slag ring, or excessively high continuous casting speed. It is recommended to reduce the continuous casting speed by 0.05 m / s. Therefore, the casting speed was reduced by 0.05 m / s, and the slag ring in the mold mold was checked and removed in time. After 5 minutes, the video monitoring of the continuous casting billet showed that the billet depression was eliminated and the surface quality of the continuous casting billet returned to normal.

[0092] The present invention has been disclosed above with reference to preferred embodiments. However, those skilled in the art should understand that these embodiments are merely illustrative of the invention and should not be construed as limiting its scope. It should be noted that any variations and substitutions equivalent to these embodiments should be considered to be covered within the scope of the claims. Therefore, the scope of protection of the present invention should be determined by the scope defined in the claims.

Claims

1. A method of controlling a continuous casting process of stainless steel, characterized by, include: The first information on defects in continuously cast billets and the timing parameters of the casting process of continuously cast billets are obtained. The first information includes the type of defect, the characteristics of each defect, and the location distribution of each defect. Based on the first information and the timing parameters, establish a relational database for each defect and its corresponding timing parameters; Based on the relevant relational database, a model is obtained that correlates each defect with the time series parameters; the model includes the following second information: the influence weight of the time series parameters corresponding to each defect on that defect, the cause path of each defect, and the confidence level corresponding to the cause path; The continuous casting process of stainless steel is controlled according to the model to reduce the generation of defects.

2. The control method according to claim 1, characterized by, Before acquiring the first information on defects in the continuously cast billet and the timing parameters of the continuous casting process, the method further includes: Collect monitoring videos of the surface quality of continuously cast billets.

3. The control method according to claim 2, characterized by, The first information obtained regarding defects in the continuously cast billet specifically refers to: The monitoring video is processed to obtain the first information about defects in the continuously cast billet.

4. The control method according to claim 3, characterized by The specific steps for processing the monitoring video to obtain the first information about the defects in the continuously cast billet are as follows: Image analysis is performed on the surveillance video to obtain each type of defect and its corresponding location. Based on each defect corresponding to the surveillance video, the characteristics of the defect are obtained.

5. The control method according to claim 1, characterized by, The step of establishing a relational database for each defect and its corresponding time series parameters based on the first information and the time series parameters specifically involves: Based on the first information and the time series parameters, a database of relationships between each defect and its corresponding time series parameters is established using big data analysis methods.

6. The control method according to claim 1, characterized by The specific steps for obtaining the model that correlates each defect with the time-series parameters based on the relevant relational database are as follows: Based on the aforementioned relational database, the influence weight of the time series parameters corresponding to each defect on that defect is obtained; Based on the process rules of stainless steel continuous casting, the cause path of each defect and the confidence level corresponding to the cause path are obtained. Based on the confidence level and the influence weight, a model is obtained that correlates each defect with the time series parameters.

7. The control method according to claim 6, characterized by The step of obtaining the cause path of each defect and the confidence level corresponding to the cause path according to the process rules of the stainless steel continuous casting process is as follows: obtain the process rules of the stainless steel continuous casting process, and obtain the cause path of each defect and the confidence level corresponding to the cause path according to the process rules.

8. The control method according to claim 1, characterized by, The specific steps of controlling the continuous casting process of stainless steel according to the model to reduce the generation of defects are as follows: Monitor the continuous casting process of stainless steel; issue alarm information after detecting defects; and adjust the corresponding timing parameters after receiving the alarm information and observe whether the corresponding defects are eliminated.

9. The control method according to claim 8, characterized by, After controlling the continuous casting process of stainless steel according to the model to reduce the generation of defects, the method further includes: The model is modified based on the defects detected in the continuous casting process of stainless steel.

Images