Hot-rolled strip steel surface quality evaluation method based on superior and inferior solution distance algorithm
Through the advantages and disadvantages solution distance algorithm and Spearman correlation analysis, combined with linear regression model, the surface quality evaluation method of hot-rolled strip is constructed, which solves the rough and single consideration problems of the traditional evaluation method, and realizes multi-level evaluation and efficient evaluation , adapt to different customer needs, improve the accuracy of evaluation and production performance.
Patent Information
- Application Number
- CN202410404015.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-04-05
- Publication Date
- 2024-11-22
AI Technical Summary
The quality evaluation method of traditional hot-rolled strip is rough, and it has not achieved comprehensive multi-level evaluation. Due to the single consideration of production equipment parameters and product parameters, it is difficult to meet the needs of different customers. The accuracy of manual inspection is limited, so it is impossible to effectively evaluate defects. and cause.
The advantages and disadvantages solution distance algorithm is used combined with Spearman's correlation analysis, and the data is divided into grids for evaluation. The linear regression model is used for comprehensive evaluation to construct an automatic evaluation model for surface quality of hot-rolled strip to ensure the rationality of the evaluation. and accuracy.
Multi-grade evaluation of the surface quality of hot-rolled strip is achieved, evaluation efficiency and production performance are improved, evaluation time is reduced, evaluation accuracy and reliability are enhanced, and evaluation is adapted to different customer needs.
Smart Images

Figure CN119005767A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to a method for evaluating the quality of steel continuous casting hot-rolled strip steel, and specifically relates to the technical field of automatic surface quality evaluation model construction and reliability verification of hot-rolled strip steel. Background Art
[0002] Hot-rolled strip is one of the main products of the steel industry. Its product quality mainly includes indicators such as size, plate shape, surface quality, and performance. Among them, surface quality has the greatest impact on the quality of the final product due to its complex formation mechanism and high frequency of occurrence. It is one of the most important and difficult to control quality indicators. Comprehensive quality evaluation of the entire roll of hot-rolled strip combined with various surface quality defects is conducive to more efficient evaluation of the quality grade of the steel strip.
[0003] The traditional comprehensive quality evaluation of hot-rolled strip steel is to compare the quality index of hot-rolled strip steel with the specified values of relevant indexes of national standards or enterprise standards to judge whether it is qualified. The evaluation result is only "qualified product" or "unqualified product". The main evaluation method at present is based on the real-time production data related to the controllable parameters such as dimensional tolerance, curvature, heating temperature, time, speed, etc. in the production process including billet preparation, heating, rough rolling, and finishing rolling, that is, production equipment parameters, combined with the parameters such as length, width, defect size, and position of the product, that is, product parameters, and the experience of experts in the field of hot-rolled strip steel and national standards are used to evaluate the surface quality of hot-rolled strip steel. The specific implementation plan is to summarize the corresponding production equipment parameters and product parameters of qualified products by experts in this field and combine them with national standards to form a rule database. When making grade assessment, the expert experience and national standards stored in the rule database are compared with the corresponding batch of products to judge whether the production equipment parameters and product parameters are within the range of the rule data. The evaluation results of this evaluation method are rough and fail to achieve all-round multi-level evaluation. The quality assessment scheme that only considers production needs cannot meet the needs of subsequent production and manufacturing. In the face of different customer needs, there is a lack of targeted quality assessment of hot-rolled strip. At the same time, the classification accuracy of the surface detector is difficult to reach the ideal state, and is greatly restricted by the human level factor, and it is impossible to correspond defects with their causes. Summary of the invention
[0004] The superior-inferior solution distance algorithm is mainly adopted. Through the preliminary correlation analysis, the parameters related to the surface quality assessment of hot-rolled strip can be fully selected to achieve the characteristics of wide initialization dimension, with the advantages of variable parameter dimensions and strong evaluation reliability.
[0005] A method for evaluating the surface quality of hot-rolled strip steel based on a superior-inferior solution distance algorithm, the specific steps are as follows:
[0006] (1) Data collection is carried out through on-site sensors, metering instruments, MES and REP, etc., to obtain relevant production equipment parameters and product parameters of the corresponding hot-rolled strip. The defect data in the product parameters include warping, wrinkling and scarring, and the production equipment parameters include rolling temperature, rolling pressure, steel plate width, steel plate length and steel plate thickness;
[0007] (2) Data transmission and storage, that is, exporting the production equipment parameters and product parameters related to the hot-rolled strip stored in the on-site storage database according to the hot-rolled strip number to the database corresponding to the hot-rolled strip;
[0008] (3) Performing data cleaning to obtain a data set that is neatly stored according to the hot-rolled strip number;
[0009] (4) The Spearman method is used for data processing to perform correlation analysis on the defect information in the production equipment parameters and product parameters stored in the database. For a sample with a sample size of n, n original data are converted into graded data, and the correlation coefficient is:
[0010]
[0011] where x i and i Represent the positions of x and y respectively, and Respectively represent the average rank;
[0012] (5) Obtain the relative relationship between the defect information in the production equipment parameters and the product parameters, select the production equipment parameters and product parameters related to the surface quality of the hot-rolled strip as the initial data of the good-bad solution distance algorithm, and the initial data is used as the data set X = {x1, x2, ..., x n}, according to the thermal value of the relative relationship between the production equipment parameters and the defect data in the product parameters as the index weight of each parameter, the index weight is initialized to obtain the initialization matrix ω={ω1,ω2,...,ω m};
[0013] (6) Divide the above-obtained data into grids with equal spacing according to the position relationship. Different grids have different numbers of defects, and each grid is evaluated separately;
[0014] (7) The weight attributes of indicators are made homogeneous, and cost-based indicators are transformed into benefit-based indicators. The larger the value, the higher the evaluation. The extremely small indicator indicates that the expected indicator value takes the minimum value.
[0015]
[0016] M is the maximum value of the index x. The interval index is the expected index value that preferably falls in the interval [a, b].
[0017]
[0018] Among them, [a, b] is the index x in the stable interval, that is, the index x value is in the stable interval, [a * ,b * ] is the maximum tolerance interval, that is, the indicator value is acceptable but not in the stable interval;
[0019] (8) According to the parameter type, the forward type is set to obtain the forward index, and the normalized initial matrix is constructed at the same time. Figure 1 From the initial data obtained, we can see that there are n evaluation objects in total, each object has m evaluation indicators, and the normalized initial matrix is constructed as follows:
[0020]
[0021] (9) Perform vector normalization on the normalized initial matrix, that is, each column element of the normalized initial matrix is divided by the norm of the current column vector, that is, the cosine distance metric is used.
[0022]
[0023] From the above, we get the standardized matrix Z:
[0024]
[0025] (10) According to the obtained standardized matrix Z, determine the optimal solution and the worst solution. The optimal solution Z + The worst solution Z is composed of the maximum value of each column element in Z. - It is composed of the minimum value of each column element in Z:
[0026]
[0027]
[0028] (11) Calculate the distance between each defect in a single grid and the optimal solution and the worst solution:
[0029]
[0030] where w j is the weight of the jth attribute, and the indicator weight is obtained based on Spearman correlation analysis. Represent the best solution and the worst solution respectively; ij represents the standardized value of the jth evaluation index of the i-th element, represents the sum of the Euclidean distances between the i-th element and the optimal solution, Represents the sum of the Euclidean distances between the i-th element and the inferior solution.
[0031] (12) Calculate the distance C between each evaluation object and the optimal solution i :
[0032]
[0033] c i →1 indicates that the evaluation object is better;
[0034] (13) According to the obtained c i and the number of defects corresponding to a single grid, and calculate the average score of each grid to obtain the evaluation score of each grid. The evaluation score reflects the evaluation level;
[0035] (14) Perform the same process on the n grids of the entire hot-rolled steel strip and obtain the evaluation scores of the n grids respectively;
[0036] (15) Using the linear regression model, the evaluation scores of n grids and the historical evaluation data of hot-rolled strip steel are used as the training data set and the test set. The model parameter β to be learned is solved through continuous iteration. The linear regression model is:
[0037] y=β0+β1x1+β2x2+…+β n x n (12)
[0038] The dependent variable y represents the evaluation score of the whole coil of hot-rolled steel strip, and the independent variables x1, x2, ..., x n represents the evaluation scores of n grids, β0, β1, β2, ..., β n is the parameter to be learned;
[0039] In order to obtain accurate parameters to be learned and verify the accuracy of the model, the corresponding loss function is selected as:
[0040]
[0041] Where Q represents the error value, y i Indicates the actual evaluation score of the whole coil of hot-rolled strip. It indicates the prediction evaluation score of the whole hot-rolled strip. Represents the optimal learning parameter. When the loss function takes the smallest value Q, the parameter to be learned β is the optimal learning parameter x1, x2, ..., x n represents the evaluation scores of n grids, Q is a measure of the gap between the model's predicted evaluation scores and the actual evaluation scores. By minimizing the loss function, the optimal parameters to be learned can be found to minimize the gap between the predicted evaluation scores and the actual evaluation scores.
[0042] The advantages are:
[0043] 1. In the process of data processing, the production equipment parameters and product parameters are fully considered, and the rationality of the surface quality assessment of hot-rolled strip is ensured through data cleaning and processing.
[0044] 2. Divide all the information data of hot-rolled strip into n grids with equal spacing according to the position relationship, and the number of defects in different grids is different. Evaluate the n grids separately, and finally combine the evaluation information of the n grids to make a comprehensive evaluation of the overall grid.
[0045] 3. In the quality evaluation of hot-rolled strip, fixed logical rules are not used to verify the information of each defect of the hot-rolled strip. Instead, all the defect information of the hot-rolled strip is combined to make an overall evaluation, which effectively improves the efficiency of hot-rolled strip quality evaluation, reduces the time-consuming hot-rolled strip quality evaluation, improves production performance, and is beneficial to the long-term production operation of the enterprise. BRIEF DESCRIPTION OF THE DRAWINGS
[0046] Figure 1 This is a diagram showing the relevant parameters after data correlation analysis.
[0047] Figure 2 Flow chart of the hot rolled strip quality evaluation steps.
[0048] Figure 3 A graph showing the error interval between the actual result and the label. DETAILED DESCRIPTION
[0049] To facilitate technical personnel to understand and implement the present invention, the method is further described in detail below in conjunction with the accompanying drawings and related implementation principles.
[0050] Data collection is to obtain the relevant production equipment parameters and product parameters of the corresponding hot-rolled strip through on-site sensors, metering instruments, MES and REP. The defect data in the product parameters include warping, wrinkling, and stuttering. The production equipment parameters include rolling temperature, rolling pressure, steel plate width, steel plate length, and steel plate thickness; data transmission and storage is to export the production equipment parameters and product parameters related to the hot-rolled strip stored in the on-site database according to the hot-rolled strip number to the corresponding database of the hot-rolled strip; the purpose of data cleaning is to review and verify the data, with the aim of deleting duplicate information, correcting existing errors, and providing data consistency. Through data cleaning, a relatively regular data set stored by the hot-rolled strip number can be obtained. The specific parameters include the hot-rolled strip number, length and width, defect coordinates, camera posture and other information.
[0051] In the parameter selection stage, because the parameter types are large in scale, Spearman correlation analysis is used for parameter screening, and n parameter types related to the surface quality of hot-rolled strip are selected as the initial data for the subsequent good and bad solution distance algorithm. The n original data are converted into graded data. At the same time, since the connection between variables is insignificant, the corresponding elements of the two observed variables are subtracted to obtain a difference. The Spearman correlation coefficient calculation formula is:
[0052]
[0053] where d i For X i and Y i The specific level data between each parameter is as follows Figure 1 shown.
[0054] The relative relationship between the defect data in the production equipment parameters and the product parameters is obtained, that is, the correlation heat map is obtained. According to the relative relationship, the production equipment parameters and product parameters related to the surface quality of the hot-rolled strip are selected as the initial data of the good and bad solution distance algorithm. The initial data is used as the data set X = {x1, x2, ..., x n}, according to the thermal value of the relative relationship between the production equipment parameters and the defect data in the product parameters as the index weight of each parameter, the index weight is initialized to obtain the initialization matrix ω={ω1,ω2,...,ω m}.
[0055] The data obtained above are divided into each grid at equal intervals according to the position relationship. The number of defects in different grids is different, and each grid is evaluated separately.
[0056] Figure 2 It includes the construction of production equipment parameters and product parameters. The same-direction judgment of indicator weight attributes is divided into extremely small indicators and interval indicators. At the same time, it should be emphasized that attention should be paid to different parameters during the conversion process. If the data of one dimension is as large as possible, and the data of another dimension is as small as possible, it will cause scale confusion. Since the applicable object is steel quality evaluation information, the initial data collected by on-site sensors, measuring instruments, MES and REP should be within the range specified by production and manufacturing technical indicators. The parameters related to the quality of hot-rolled strip itself should be small to be more conducive to the evaluation of steel surface quality. At the same time, regarding the position relationship of defects, since the defect position has a great influence on the quality evaluation of hot-rolled strip, the position is considered to be a corresponding interval conversion. Therefore, the relevant conversion expression relationship involved is as follows:
[0057]
[0058] M is the maximum value of the index x, and the interval conversion expression relationship is:
[0059]
[0060] Where [a, b] is the index x in the stable interval, that is, the index x value is in the stable interval; [a * ,b * ] is the maximum tolerance interval, that is, the indicator value is acceptable but not in the stable interval.
[0061] According to the parameter type, the positive index is obtained by setting the positive index type and constructing the normalized initial matrix. According to the initial data obtained, there are n objects to be evaluated, each object has m indicators, and the normalized initial matrix is constructed:
[0062]
[0063] Normalize the normalized initial matrix by dividing each column element of the normalized initial matrix by the norm of the current column vector, that is, use the cosine distance metric
[0064]
[0065] Get the standard matrix Z:
[0066]
[0067] According to the obtained standardized matrix Z, the optimal solution and the worst solution are determined. The optimal solution Z + The worst solution Z is composed of the maximum value of each column element in Z. - It is composed of the minimum value of each column element in Z:
[0068]
[0069]
[0070] Calculate the distance between each defect in a single grid and the optimal and worst solutions
[0071]
[0072] where w j is the weight of the jth attribute, and the indicator weight is obtained based on Spearman correlation analysis. Represent the best solution and the worst solution respectively; ij represents the standardized value of the jth evaluation index of the i-th element, represents the sum of the Euclidean distances between the i-th element and the optimal solution, Represents the sum of the Euclidean distances between the i-th element and the inferior solution. Calculate the distance C between each evaluation object and the optimal solution i :
[0073]
[0074] c i →1 indicates that the better the evaluation object is, the numerator can also be set to Indicates the closeness to the worst solution. At this time, c i →0 indicates that the evaluation object is better.
[0075] According to the obtained c i The number of defects corresponding to each grid is calculated, and the average score of each grid is calculated to obtain the evaluation score of each grid. The evaluation score reflects the evaluation level. The same treatment is performed on the n grids of the entire hot-rolled strip to obtain the evaluation scores of the n grids.
[0076] like Figure 3 As shown in the figure, the quality of the whole hot-rolled strip is evaluated. The linear regression model is used, and the evaluation scores of n grids and the historical evaluation data of hot-rolled strip are used as the training data set and the test set. The parameter β to be learned in the model is solved through continuous iteration. The linear regression model is expressed as:
[0077] y=β0+β1x1+β2x2+…+β n x n (12)
[0078] The dependent variable y represents the evaluation score of the whole coil of hot-rolled strip steel, and the independent variables x1, x 2… 、x n represents the evaluation scores of n grids, β0, β1, β 2… , β n is the parameter to be learned;
[0079] In order to obtain accurate parameters to be learned and verify the accuracy of the model, the corresponding loss function is selected as:
[0080]
[0081] Where Q represents the error value, y i Indicates the actual evaluation score of the whole coil of hot-rolled strip. It indicates the prediction evaluation score of the whole hot-rolled strip. Represents the optimal learning parameter. When the loss function takes the smallest value Q, the parameter to be learned β is the optimal learning parameter x1, x 2… 、x n represents the evaluation scores of n grids, Q is a measure of the gap between the model's predicted evaluation scores and the actual evaluation scores. By minimizing the loss function, the optimal parameters to be learned can be found to minimize the gap between the predicted evaluation scores and the actual evaluation scores.
[0082] The following system equipment is mainly used in the design and implementation of this method:
[0083] 1) The software system uses Anaconda, Python 3.7, and PyCharm 2021.3.3.
[0084] 2) The library functions involved include PyTorch, pandas, matplotlib, graphviz, seaborn, numpy, etc.
[0085] 3) Use Python language to implement the corresponding model.
Claims
1. A method for evaluating the surface quality of hot-rolled strip steel based on a superior-inferior solution distance algorithm, characterized in that: The specific steps include: (1) Collect relevant production equipment parameters and product parameters of hot-rolled strip steel through on-site sensors, metering instruments, MES and REP. Defect data in product parameters include warping, wrinkling and scarring. Production equipment parameters include rolling temperature, rolling pressure, steel plate width, steel plate length and steel plate thickness. (2) Data transmission and storage, that is, exporting the production equipment parameters and product parameters related to the hot-rolled strip stored in the on-site storage database according to the hot-rolled strip number to the database corresponding to the hot-rolled strip; (3) Performing data cleaning to obtain a data set that is neatly stored according to the hot-rolled strip number; (4) The Spearman method is used for data processing to perform correlation analysis on the defect data in the production equipment parameters and product parameters stored in the database. For a sample with a sample size of n, n original data are converted into graded data, and the correlation coefficient is: where x i and i Represent the positions of x and y respectively, and Respectively represent the average rank; (5) Obtain the relative relationship between the defect data in the production equipment parameters and the product parameters, select the production equipment parameters and product parameters related to the surface quality of the hot-rolled strip as the initial data of the good-bad solution distance algorithm, and the initial data is used as the data set X = {x1, x2, ..., x n }, according to the thermal value of the relative relationship between the production equipment parameters and the defect data in the product parameters as the index weight of each parameter, the index weight is initialized to obtain the initialization matrix ω={ω1,ω2,...,ω m }; (6) Divide the above-obtained data into grids with equal spacing according to the position relationship. Different grids have different numbers of defects, and each grid is evaluated separately; (7) The weight attributes of indicators are made homogeneous, and cost-based indicators are transformed into benefit-based indicators. The larger the value, the higher the evaluation. The extremely small indicator indicates that the expected indicator value takes the minimum value. M is the maximum value of the index x. The interval index is the expected index value that preferably falls in the interval [a, b]. Among them, [a, b] is the index x in the stable interval, that is, the index x value is in the stable interval, [a * ,b * ] is the maximum tolerance interval, that is, the indicator value is acceptable but not in the stable interval; (8) According to the parameter type, the positive index is obtained by setting the positive index type and constructing the normalized initial matrix. According to the initial data obtained, there are n objects to be evaluated, each object has m indicators, and the normalized initial matrix is constructed: (9) Perform vector normalization on the normalized initial matrix, that is, each column element of the normalized initial matrix is divided by the norm of the current column vector, that is, the cosine distance metric is used. Get the normalized matrix Z: (10) According to the obtained standardized matrix Z, determine the optimal solution and the worst solution. The optimal solution Z + The worst solution Z is composed of the maximum value of each column element in Z. - It is composed of the minimum value of each column element in Z: (11) Calculate the distance between each defect in a single grid and the optimal solution and the worst solution: where w j is the weight of the jth attribute, and the indicator weight is obtained based on Spearman correlation analysis. Represent the best solution and the worst solution respectively; ij represents the standardized value of the jth evaluation index of the i-th element, represents the sum of the Euclidean distances between the i-th element and the optimal solution, represents the sum of the Euclidean distances between the i-th element and the inferior solution; (12) Calculate the distance C between each evaluation object and the optimal solution i : c i →1 indicates that the evaluation object is better; (13) According to the obtained c i and the number of defects corresponding to a single grid, and calculate the average score of each grid to obtain the evaluation score of each grid. The evaluation score reflects the evaluation level; (14) Perform the same process on the n grids of the entire hot-rolled steel strip and obtain the evaluation scores of the n grids respectively; (15) Using the linear regression model, the evaluation scores of n grids and the historical evaluation data of hot-rolled strip steel are used as the training data set and the test set. The model parameter β to be learned is solved through continuous iteration. The linear regression model is: y=β0+β1x1+β2x2+…+β n x n (12), The dependent variable y represents the evaluation score of the whole coil of hot-rolled strip steel, and the independent variables x1, x 2… 、x n represents the evaluation scores of n grids, β0, β1, β 2… , β n is the parameter to be learned; In order to obtain accurate parameters to be learned and verify the accuracy of the model, the corresponding loss function is selected as: Where Q represents the error value, y i Indicates the actual evaluation score of the whole coil of hot-rolled strip. It indicates the prediction evaluation score of the whole hot-rolled strip. Represents the optimal learning parameter. When the loss function takes the smallest value Q, the parameter to be learned β is the optimal learning parameter x1, x 2… 、x n represents the evaluation scores of n grids, Q is a measure of the gap between the model's predicted evaluation scores and the actual evaluation scores. By minimizing the loss function, the optimal learning parameters can be found to minimize the gap between the predicted evaluation scores and the actual evaluation scores.
