A stainless steel cleanliness and cold-rolled sheet surface quality prediction method based on evaluation model

Abstract

The application discloses a stainless steel cleanliness prediction method, comprising the following steps: obtaining steady-state parameters and their influence coefficients in a production process, calculating the steady-state cleanliness index of each steel furnace; determining the influence coefficient of non-steady-state factors according to whether the non-steady-state factors occur in the production cycle of each product unit and the influence degree; and calculating the cleanliness index of each product unit.

Description

Technical Field

[0001] This invention belongs to the field of intelligent manufacturing in iron and steel metallurgy, and relates to a method for predicting the cleanliness of stainless steel and the surface quality of cold-rolled thin plates. More specifically, this invention relates to a method for predicting the cleanliness of stainless steel and the surface quality of cold-rolled thin plates based on an evaluation model. Background Technology

[0002] The cleanliness of molten steel is primarily evaluated through total oxygen content determination and inclusion statistical analysis. However, due to the simplistic evaluation methods, it is difficult to comprehensively and accurately reflect the impact of large inclusions on product quality. In actual production, these methods also suffer from a certain degree of lag and are difficult to implement for every furnace, thus hindering the ability to proactively predict steel cleanliness and product quality based on on-site processes and operations. Consequently, feedback on product quality information is delayed. Furthermore, the concept of "good" or "bad" cleanliness is somewhat vague; current qualitative or quantitative methods struggle to accurately define the cleanliness level of molten steel, failing to provide guidance for production processes.

[0003] To address the above issues, various methods for predicting cleanliness or product quality have been developed both domestically and internationally. These methods mainly include: calculations based on large amounts of detection and analysis data, predictions based on thermodynamic and kinetic models, and predictions based on steady-state process parameters and neural network models.

[0004] Among them, invention patent applications with patent numbers 201910920199.2 and 202010796468.1 are mainly based on calculations using a large amount of detection and analysis data. Through extensive statistical analysis of inclusions in samples, and based on inclusion particle size data obtained from metallographic observation, the maximum possible inclusion size in steel is predicted using extreme value theory. This method has the following problems: ① It requires a large amount of metallographic observation data, placing high demands on equipment and hardware conditions, and the analysis process is complex and lengthy; ② It does not establish a connection with process parameters and cannot reflect the influence of process parameters on cleanliness.

[0005] Patent application number 200680037136.3 provides quality prediction and quality control for continuously cast slabs. Patent application number 201710470968.4 provides a method for predicting the size of MnS inclusions in continuously cast steel slabs. Wu Ling, in his doctoral dissertation "Prediction Model of Oxide Inclusions in Steel Based on GANN and Irreversible Thermodynamics," provides a model prediction method. The Technical University of Freiberg in Germany and the University of Science and Technology Beijing have developed a simulation model for the evolution of inclusions in ladle metallurgy. All of the above methods belong to the cleanliness or product quality prediction methods based on thermodynamic and kinetic models. They are mainly based on the current basic theories and research results of thermodynamics and kinetics. Therefore, they are limited by the current understanding of thermodynamics and kinetics, require high cognitive ability from developers, and the models cannot update data autonomously in a timely manner.

[0006] Furthermore, by using neural network models and deep learning principles, a relationship can be established between parameters and cleanliness or product quality, enabling real-time prediction when production data is input again. This type of method has mature models, wide application areas, and high accuracy; it demonstrates great flexibility and adaptability when dealing with large amounts of raw data that cannot be described by rules or formulas. However, this method also has significant drawbacks: ① It requires a comprehensive analysis of influencing factors to identify the main factors. If any factors are omitted, the accuracy decreases. ② Under non-steady-state or abnormal conditions, the model's accuracy declines. Summary of the Invention

[0007] In view of the above problems, the present invention provides a method for predicting the cleanliness of stainless steel and the surface quality of cold-rolled sheet based on an evaluation model. This method mainly provides a method for predicting the cleanliness of stainless steel and the surface quality of cold-rolled sheet.

[0008] Specifically, the present invention is achieved through the following technical solution:

[0009] A method for predicting the cleanliness of stainless steel, comprising:

[0010] Step S1: Obtain the steady-state parameters and their influence coefficients during the production process. Calculate the steady-state cleanliness index of each process for each heat of steel according to formula (1) and calculate the sum of the steady-state cleanliness indices y for each process:

[0011]

[0012] in,

[0013] y p It is the steady-state cleanliness index of the p-th process in each heat of steel;

[0014] x i These are the normalized data for the i-th steady-state parameter, and their range is 0 ≤ x. i ≤1;

[0015] α i It is the influence coefficient of the i-th steady-state parameter, and its range is 1≤α. i ≤10;

[0016] m is a positive integer;

[0017] Step S2: Identify n non-steady-state factors and determine the influence coefficient of each non-steady-state factor based on whether each non-steady-state factor occurs and the degree of influence during the production cycle of each product unit.

[0018] Step S3: Calculate the cleanliness index of each product unit according to formula (2):

[0019]

[0020] in,

[0021] k is the cleanliness index of each product unit;

[0022] β j It is the influence coefficient of the j-th unsteady-state factor, and its range is 0 ≤ β. j ≤1;

[0023] y is the steady-state cleanliness index of each heat of steel;

[0024] n is a positive integer.

[0025] Optionally, in step S1, the influence coefficient α of the i-th steady-state parameter i =b i1 +b i2 +b i3 +b i4 +b i5 ;

[0026] in,

[0027] b i1 Let b be the effect of the i-th steady-state parameter on cleanliness. If the effect of the i-th steady-state parameter on cleanliness is indirect, then b i1 =0, if the i-th steady-state parameter has a direct impact on cleanliness, then b i1 =5;

[0028] b i2 Let b be the process in which the i-th steady-state parameter is located. If the process in which the i-th steady-state parameter is located is an AOD process, then b i2 =1, if the process containing the i-th steady-state parameter is an LTS process or an LF process, then b i2 =2, if the process containing the i-th steady-state parameter is the CCM process, then b i2 =3;

[0029] b i3 Let b be the parameter type of the i-th steady-state parameter. If the i-th steady-state parameter is a quantized accurate value unaffected by human factors, then b... i3 =1, if the i-th steady-state parameter is a non-quantized accurate value or is affected by human factors, then b i3 =-1, if the i-th steady-state parameter is otherwise, then b i3 =0;

[0030] b i4 Let b be the parameter category of the i-th steady-state parameter. If the i-th steady-state parameter is an effect parameter, then b i4 =1, if the i-th steady-state parameter is a control parameter, then b i4 =0;

[0031] b i5 Let b be the variable case of the i-th steady-state parameter. If the i-th steady-state parameter is immutable, has no compensation measures, or is the only parameter, then b i5 =1, if the i-th steady-state parameter is fine-tunable or affects other parameters together, then b i5 =0, if the i-th steady-state parameter is changeable or repeats with other parameters, then b i5 =-1.

[0032] Optionally, in step S1, the i-th steady-state parameter is the basicity of the refining slag, and its normalized data x is obtained using the following method. i :

[0033] Raw data x is obtained through chemical analysis of slag samples, judgment based on slag sample color, or material balance calculations. i0 ;

[0034] According to x i0 The normalized data x is obtained by using the corresponding algorithm for the type. i .

[0035] Optionally, if x i0 The type is continuous values, and its normalized data x can be obtained using any of the following algorithms. i :

[0036] |x i0 -x iμ | / (x imax -x imin ), x imax -x i0 / (x imax -x imin ), x i0 -x imin / (x imax -x imin ), x i0 / x imax ,1-x i0 / x imax Empirical formulas or fitting formulas

[0037] Where, x iμ It is the average value of the i-th steady-state parameter, x imax It is the maximum value among the historical valid data of the i-th steady-state parameter, x imin It is the minimum value among the historical valid data of the i-th steady-state parameter;

[0038] If x i0 The data type is discrete, and its normalized data x is obtained using the assignment method. i .

[0039] Optionally, in step S2, if the j-th unsteady factor does not occur, then β j =1, otherwise β j <1.

[0040] Optionally, the stainless steel is stainless steel with the grade 430.

[0041] As can be seen from the above technical solution, the prediction method of the present invention has at least the following beneficial effects:

[0042] The prediction method of this invention, by developing a cleanliness evaluation index, comprehensively reflects the type, size, quantity, and degree of harm of inclusions in steel to the steel matrix, thereby achieving real-time prediction of cleanliness and overall quality; it establishes an organic link between cleanliness, product quality, and process parameters, providing quantitative grading and circulation suggestions for products, thereby reducing rolling surface defects; and it provides a reference for quality monitoring, timely problem detection, and rapid focus on the causes of problems during the production process.

[0043] Compared to simple qualitative analysis based on process parameters or test data, this invention proposes an evaluation method that combines qualitative and quantitative approaches, and comprehensively considers both steady-state and unsteady-state factors.

[0044] Compared with prediction methods based on a large amount of detection and analysis data, the present invention has the following advantages: ① It does not depend on equipment and hardware conditions and can obtain prediction data online in real time; ② It can establish the relationship between cleanliness, product quality and process parameters, and can trace the root cause of the influence of process parameters during data analysis.

[0045] Compared to predictions based on thermodynamic and kinetic models, this invention has the following advantages: it does not require a high level of cognitive ability from developers and allows for timely data updates to optimize the model;

[0046] Compared with neural network models and deep learning principles, this invention has the following advantages: it is not limited by model algorithms and individual influencing factors, and it takes into account both steady-state and non-steady-state factors, and has strong adaptability and compatibility with actual production. Attached Figure Description

[0047] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0048] Figure 1 A schematic diagram illustrating the principle of the technical solution of the present invention is shown.

[0049] Figure 2The relationship between the cleanliness index and the TO content in the intermediate package for 12 batches of Example 1 is shown.

[0050] Figure 3 The data shows the surface defects of cold-rolled sheets with a thickness of less than 0.6 mm corresponding to different ranges of billet cleanliness index. Detailed Implementation

[0051] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this invention will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.

[0052] To address the problem of predicting steel cleanliness and product quality, the inventors of this invention, through research, established a cleanliness index evaluation system based on process parameters. This system reflects the type, size, quantity, and degree of harm to the steel matrix caused by inclusions in the steel, thereby determining the surface inclusion defects and overall quality of the product. Figure 1 As shown, it includes two parts: ① evaluation and prediction of cleanliness (endogenous deoxygenation products) under steady state; ② evaluation and prediction of cleanliness (foreign or large inclusions) under non-steady state or abnormal conditions.

[0053] This invention proposes a method for predicting the cleanliness of stainless steel, comprising:

[0054] Step S1: Calculate the steady-state cleanliness index for each heat of steel. This specifically includes:

[0055] Step S101: Determine steady-state influencing factors. Based on historical experience or data analysis, screen out steady-state parameters that affect cleanliness or product quality from the product production cycle.

[0056] Step S102: Determine the influence coefficient of steady-state parameters, which is mainly determined by comprehensively considering the influence of steady-state parameters on inclusion particle size and quantity (TO content in steel) and the influence on the generation of harmful inclusions.

[0057] Specifically, the influence coefficient α of the i-th steady-state parameter i According to α i =b i1 +b i2 +b i3 +b i4 +b i5 To calculate.

[0058] in:

[0059] b i1Let b be the effect of the i-th steady-state parameter on cleanliness. If the effect of the i-th steady-state parameter on cleanliness is indirect, then b i1 =0, if the i-th steady-state parameter has a direct impact on cleanliness, then b i1 =5;

[0060] b i2 Let b be the process in which the i-th steady-state parameter is located. If the process in which the i-th steady-state parameter is located is an AOD process, then b i2 =1, if the process containing the i-th steady-state parameter is an LTS process or an LF process, then b i2 =2, if the process containing the i-th steady-state parameter is the CCM process, then b i2 =3;

[0061] b i3 Let b be the parameter type of the i-th steady-state parameter. If the i-th steady-state parameter is a quantized accurate value unaffected by human factors, then b... i3 =1, if the i-th steady-state parameter is a non-quantized accurate value or is affected by human factors, then b i3 =-1, if the i-th steady-state parameter is otherwise, then b i3 =0;

[0062] b i4 Let b be the parameter category of the i-th steady-state parameter. If the i-th steady-state parameter is an effect parameter, then b i4 =1, if the i-th steady-state parameter is a control parameter, then b i4 =0;

[0063] b i5 Let b be the variable case of the i-th steady-state parameter. If the i-th steady-state parameter is immutable, has no compensation measures, or is the only parameter, then b i5 =1, if the i-th steady-state parameter is fine-tunable or affects other parameters together, then b i5 =0, if the i-th steady-state parameter is changeable or repeats with other parameters, then b i5 =-1.

[0064] The steps for determining the influence coefficient are summarized as follows:

[0065]

[0066] Preferably, the influence coefficient α i Historical process parameters and quality information can be obtained through in-depth mining using the analytic hierarchy process (AHP) or numerical analysis methods, while a deep learning model can be established to continuously update the data.

[0067] Step S103: Collect the data corresponding to the steady-state parameters and normalize each data item.

[0068] For example, if the i-th steady-state parameter is the basicity of refining slag, its normalized data x can be obtained using the following method. i Raw data x is obtained through chemical analysis of slag samples, judgment based on slag sample color, or material balance calculations. i0 According to x i0 The normalized data x is obtained by using the corresponding algorithm for the type. i .

[0069] The normalization method follows these rules:

[0070]

[0071] Step S104: Calculate the steady-state cleanliness index of each process for each heat of steel according to formula (1), and calculate the sum of the steady-state cleanliness indices y for each process:

[0072]

[0073] in,

[0074] y p It is the steady-state cleanliness index of the p-th process in each heat of steel;

[0075] x i These are the normalized data for the i-th steady-state parameter, and their range is 0 ≤ x. i ≤1;

[0076] α i It is the influence coefficient of the i-th steady-state parameter, and its range is 1≤α. i ≤10;

[0077] m is a positive integer, indicating that the p-th process has m steady-state parameters.

[0078] Specifically, the sum of the steady-state cleanliness indices y for each process in the production of a heat of steel is obtained by weighted summation of the cleanliness indices of each process in the production process of that heat of steel. For example, the production process of stainless steel of grade 430 can be divided into AOD process, LF process and continuous casting process. The weights of the steady-state cleanliness indices of each process are 4.29, 10.71 and 17.39 respectively. Then y = 4.29 × y1 + 10.71 × y2 + 17.39 × y3, where y1 is the steady-state cleanliness index of AOD process, y2 is the steady-state cleanliness index of LF process and y3 is the steady-state cleanliness index of continuous casting process.

[0079] Step S2: Determine the non-steady-state factors and their influence coefficients in the production cycle of each product unit. The product unit may be, for example, a billet or a steel coil.

[0080] Specifically, it includes:

[0081] Step S201: Identify the non-steady-state factors and record n ​​non-steady-state factors occurring in each unit of the production cycle, using the smallest product unit (slab, coil, etc.) as the unit. Non-steady-state factors refer to parameters related to cleanliness that exceed the normal range or fluctuate irregularly over time, such as continuous casting start-up, ladle change, and stop-casting, as well as fluctuations in the liquid level in the crystallizer, casting speed fluctuations, and changes in stopper rod position. Recording methods can be based on certain marking rules, or can be numerical values, arrays, or curves.

[0082] Step S202: Based on the impact of large inclusion formation and product quality in historical data, determine the influence coefficient β corresponding to the j-th unsteady-state factor. j If the process has no non-steady-state conditions (i.e., the j-th non-steady-state factor does not occur), then β j =1; otherwise β j <1, and the greater the influence, the higher the β. j The smaller the value.

[0083] Preferably, the influence coefficient β j Historical process parameters and quality information can be obtained through in-depth mining using the analytic hierarchy process (AHP) or numerical analysis methods, while a deep learning model can be established to continuously update the data.

[0084] Step S3: Calculate the cleanliness index of each product unit according to formula (2):

[0085]

[0086] in,

[0087] k is the cleanliness index of each product unit;

[0088] β j It is the influence coefficient of the j-th unsteady-state factor, and its range is 0 ≤ β. j ≤1;

[0089] y is the steady-state cleanliness index of each heat of steel;

[0090] n is a positive integer.

[0091] The surface defects of each product unit can be determined based on its cleanliness index. For example, Figure 3 The data shows the surface defects of cold-rolled sheets with a thickness of less than 0.6 mm corresponding to different cleanliness index ranges of the cast billet. As the cleanliness index decreases, the surface defect rate of the cold-rolled sheet increases. When the cleanliness index of the cast billet is >600, the defect rate approaches 0. Therefore, the surface defects of the cast billet during subsequent rolling can be predicted in real time using the cleanliness index.

[0092] Example

[0093] The present invention is further illustrated below by way of embodiments, but the invention is not limited to the scope of the embodiments described herein. Experimental methods in the following embodiments, unless otherwise specified, were selected according to conventional methods and conditions.

[0094] Example 1

[0095] This embodiment focuses on 430 stainless steel, and its production process is: hot metal pretreatment → 180t AOD → LTS → LF → continuous casting. Taking heat number 3394 as an example, the specific implementation process of this embodiment is as follows:

[0096] (1) Calculate the steady-state cleanliness index of the steel from this furnace.

[0097] Based on historical experience, 19 steady-state parameters that affect cleanliness or product quality were selected from the product manufacturing cycle.

[0098] The influence coefficients of these 19 steady-state parameters are determined as follows:

[0099]

[0100] Normalization is performed:

[0101]

[0102] Substituting the above data into formula (1), the cleanliness index y1 for the AOD process is calculated to be 49.5, y2 for the LF process is 28, and y3 for the continuous casting process is 26. Then, the cleanliness index for this heat is calculated as y = 4.29 × y1 + 10.71 × y2 + 17.39 × y3 = 964.20

[0103] (2) Determine the non-steady-state factors and their influence coefficients

[0104] Using cast billets as the product unit, this furnace batch consisted of 11 cast billets, numbered 00 to 10. Two unsteady-state factors occurring during the production cycle of each cast billet were recorded. After analyzing and summarizing the historical data using numerical analysis methods, the inventors assigned values ​​to the influence coefficients of each unsteady-state factor, as shown in the table below.

[0105] Calculate the cleanliness index of each billet according to formula (2):

[0106] The calculation process and results are shown in the table below:

[0107]

[0108] As can be seen from this embodiment, the prediction method of the present invention can realize the real-time collection and calculation of production site data. The cleanliness index k is calculated for each steel furnace and each billet when it is cut and sprayed. According to the inventor's statistics, the following valid data were obtained from 2020 to 2021: the steady-state cleanliness index y of 6822 furnaces and the cleanliness index k of 57219 billets.

[0109] Repeat the implementation process of this embodiment. For example... Figure 2 As shown, the cleanliness index corresponds to the TO content in the ladle for a total of 12 heats. As the cleanliness index increases, the TO content in the ladle decreases. The TO content in the ladle reflects the overall situation of inclusions or cleanliness in the steel to a certain extent. Therefore, the cleanliness index reflects and predicts the cleanliness of molten steel.

[0110] Comparative Example 1

[0111] The steel grade and production process used in this comparative example are the same as those in Example 1.

[0112] The implementation method of this comparative example is as follows: Oxygen content in the ladle is analyzed using an online sampling and rapid detection system, with three samples analyzed per heat of steel. The oxygen content is used to evaluate and predict cleanliness and the quality of the cold-rolled sheet.

[0113] However, due to the time delay in analysis, the results cannot be obtained immediately, generally by 1 to 48 hours, and cannot reflect the cleanliness status in a timely manner.

[0114] Secondly, by statistically analyzing the correlation between TO content and inclusion defects on the surface of cold-rolled steel sheets, the results show that although there is a certain correlation under specific short-term conditions, the TO content analysis has errors due to the large number of influencing factors on the inclusion defects on the surface of cold-rolled steel sheets and does not consider the effects of unsteady processes. When the amount of data is large, the TO content alone cannot fully reflect the inclusion defects on the surface of cold-rolled steel sheets, let alone make predictions.

[0115] A comparison of Example 1 and Comparative Example 1 shows that the prediction method of the present invention has advantages such as low latency and high accuracy, and can realize real-time prediction of cleanliness and rolling surface condition.

[0116] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any substitutions, modifications, combinations, changes, simplifications, etc., made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. A method for predicting the cleanliness of stainless steel, characterized in that, include: Step S1: Obtain the steady-state parameters and their influence coefficients during the production process. Calculate the steady-state cleanliness index of each process for each heat of steel according to formula (1) and calculate the sum of the steady-state cleanliness indices y for each process: in, y p It is the steady-state cleanliness index of the p-th process in each heat of steel; x i These are the normalized data for the i-th steady-state parameter, and their range is 0 ≤ x. i ≤1; α i It is the influence coefficient of the i-th steady-state parameter, and its range is 1≤α. i ≤10; m is a positive integer; Step S2: Identify n non-steady-state factors and determine the influence coefficient of each non-steady-state factor based on whether each non-steady-state factor occurs and the degree of influence during the production cycle of each product unit. Step S3: Calculate the cleanliness index of each product unit according to formula (2): in, k is the cleanliness index of each product unit; β j It is the influence coefficient of the j-th unsteady-state factor, and its range is 0 ≤ β. j ≤1; y is the steady-state cleanliness index of each heat of steel; n is a positive integer.

2. The prediction method according to claim 1, characterized in that, In step S1, the influence coefficient α of the i-th steady-state parameter i =b i1 +b i2 +b i3 +b i4 +b i5 ; in, b i1 Let b be the effect of the i-th steady-state parameter on cleanliness. If the effect of the i-th steady-state parameter on cleanliness is indirect, then b i1 =0, if the i-th steady-state parameter has a direct impact on cleanliness, then b i1 =5; b i2 Let b be the process in which the i-th steady-state parameter is located. If the process in which the i-th steady-state parameter is located is an AOD process, then b i2 =1, if the process containing the i-th steady-state parameter is an LTS process or an LF process, then b i2 =2, if the process containing the i-th steady-state parameter is the CCM process, then b i2 =3; b i3 Let b be the parameter type of the i-th steady-state parameter. If the i-th steady-state parameter is a quantized accurate value unaffected by human factors, then b... i3 =1, if the i-th steady-state parameter is a non-quantized accurate value or is affected by human factors, then b i3 =-1, if the i-th steady-state parameter is otherwise, then b i3 =0; b i4 Let b be the parameter category of the i-th steady-state parameter. If the i-th steady-state parameter is an effect parameter, then b i4 =1, if the i-th steady-state parameter is a control parameter, then b i4 =0; b i5 Let b be the variable case of the i-th steady-state parameter. If the i-th steady-state parameter is immutable, has no compensation measures, or is the only parameter, then b i5 =1, if the i-th steady-state parameter is fine-tunable or affects other parameters together, then b i5 =0, if the i-th steady-state parameter is changeable or repeats with other parameters, then b i5 =-1.

3. The prediction method according to claim 1, characterized in that, In step S1, the i-th steady-state parameter is the basicity of the refining slag, and its normalized data x is obtained using the following method. i : Raw data x is obtained through chemical analysis of slag samples, judgment based on slag sample color, or material balance calculations. i0 ; According to x i0 The normalized data x is obtained by using the corresponding algorithm for the type. i .

4. The prediction method according to claim 3, characterized in that, If x i0 The type is continuous values, and its normalized data x can be obtained using any of the following algorithms. i : |x i0 -x iμ | / (x imax -x imin ), x imax -x i0 / (x imax -x imin ), x i0 -x imin / (x imax -x imin ), x i0 / x imax ,1-x i0 / x imax Empirical formulas or fitting formulas Where, x iμ It is the average value of the i-th steady-state parameter, x imax It is the maximum value among the historical valid data of the i-th steady-state parameter, x imin It is the minimum value among the historical valid data of the i-th steady-state parameter; If x i0 The data type is discrete, and its normalized data x is obtained using the assignment method. i .

5. The prediction method according to claim 1, characterized in that, In step S2, if the j-th unsteady factor does not occur, then β j =1, otherwise β j <1.

6. The prediction method according to any one of claims 1 to 5, characterized in that, The stainless steel in question is grade 430 stainless steel.

Images