A performance prediction method for cold finishing shear cutter based on Pearson correlation coefficient

Abstract

The present application belongs to the technical field of cold rolling processing, and particularly relates to a performance prediction method of a cold rolling finishing shear cutter based on a Pearson correlation coefficient. The present application comprises five steps of sample collection and data set division, key process parameter screening based on a training set and the Pearson correlation coefficient, multivariate linear regression model construction based on the training set and a key process parameter set, model verification and adaptive iterative optimization based on a verification set, and cutter performance online prediction and model dynamic updating, realizes accurate and stable prediction of the performance of the cold rolling finishing shear cutter, effectively adapts to various production conditions such as cold rolling strip steel specification switching and process parameter adjustment, avoids data leakage problems caused by full sample screening from the root, effectively eliminates process parameters weakly related or falsely related to the performance of the cutter, simplifies the model input dimension, significantly improves the production efficiency and product qualification rate of the cold rolling finishing process, and reduces the operation and maintenance and production costs of the production line.

Description

Technical Field

[0001] This invention belongs to the field of cold rolling processing technology, specifically relating to a performance prediction method for cold rolling finishing shearing tools based on Pearson correlation coefficient. Background Technology

[0002] Cold-rolled strip steel is a core raw material for automobile manufacturing, home appliance production, and equipment processing. Finishing shearing is a critical process in the cold-rolled strip steel production line, directly determining the dimensional accuracy and surface quality of the final strip steel product. As the core component in the shearing process, the wear condition, edge morphology changes, and remaining service life of the cold-rolled finishing shearing cutter directly affect the shearing quality of the strip steel and are closely related to the production line's continuity, maintenance costs, and production efficiency. Therefore, accurate and real-time prediction of the performance of the cold-rolled finishing shearing cutter is a core element for achieving intelligent operation and maintenance of the cold-rolled finishing process, ensuring production stability, and improving product quality.

[0003] Currently, performance prediction technologies for metal cutting tools are mainly divided into two categories: prediction methods based on physical models and empirical formulas, and intelligent prediction methods based on data-driven approaches. Among these, prediction methods based on physical models and empirical formulas largely rely on ideal cutting condition assumptions and a large amount of empirical parameters from the field, making them only suitable for specific tool specifications and fixed production conditions. However, cold rolling finishing shearing processes are characterized by frequent changes in strip specifications, large fluctuations in shearing loads, and complex and variable boundary conditions. Such methods struggle to adapt to the diverse production conditions on-site, exhibiting weak generalization ability and making it difficult to guarantee prediction accuracy in practical applications.

[0004] Data-driven intelligent prediction methods often employ complex black-box models such as neural networks and deep learning. While these methods can fit the nonlinear relationship between process parameters and tool performance to some extent, they have significant application drawbacks: First, the model structure is complex, and the parameters lack clear physical meaning, making it difficult for on-site process personnel to understand, debug, and maintain, resulting in extremely difficult engineering implementation. Second, such models require massive amounts of high-quality labeled samples for training, but it is difficult to obtain sufficient and compliant tool lifecycle sample data in cold rolling production sites, making the models prone to overfitting and resulting in poor prediction stability in actual production. Third, the models have high computational complexity and stringent hardware computing power requirements, making it impossible to achieve real-time online prediction in conventional PLCs and edge computing devices in industrial sites, thus failing to meet the real-time requirements of continuous production lines.

[0005] Some technical solutions employ correlation coefficient analysis for process parameter screening. However, these solutions often use the entire sample set for feature screening before dividing the training and validation sets, inevitably leading to data leakage. This results in excellent model performance during training but a significant drop in prediction accuracy during actual field applications. Furthermore, feature screening relies solely on correlation coefficient thresholds, lacking necessary statistical significance verification. This can easily introduce process parameters that are spuriously correlated with tool performance, further reducing the model's stability and reliability. In addition, most existing technologies are designed for general turning and milling tools, failing to specifically address the unique characteristics of cold rolling finishing shearing tools, such as intermittent shearing, high-tonnage loads, and frequent fluctuations in operating conditions. Moreover, they lack a complete closed loop of "feature screening - model construction - model verification - iterative optimization - online updates," making it impossible for the models to adapt to long-term changes in production line equipment status and operating conditions. After long-term operation, prediction accuracy continuously declines, failing to meet the long-term stable application requirements of cold rolling production sites. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a performance prediction method for cold rolling finishing shearing tools based on Pearson correlation coefficient.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A performance prediction method for cold rolling finishing shearing tools based on Pearson correlation coefficient includes the following steps: S1, Partitioning the dataset; Obtain the process parameter set and performance index set of the cold rolling finishing shearing tool, and form a paired sample from the process parameter set and performance index set. A total of [number] samples were collected. Group paired samples; The paired samples are divided into non-overlapping training and validation sets according to a preset ratio, with the training set having a sample size of [missing information]. The validation set sample size is ,satisfy ; S2. Screening of key process parameters without data leakage based on the training set and Pearson correlation coefficient; For the training set samples, for the first The first group of training samples Process parameters With the Target performance metrics in training samples Pearson correlation coefficients were calculated to obtain the correlation coefficient values ​​between each process parameter and the target performance index. Set the correlation coefficient threshold. ,Will Furthermore, the process parameters that pass the significance test are identified as critical process parameters and are recorded in the selected critical process parameter set, which contains N parameters. Total number of process parameters ; S3. Construction of a multiple linear regression model based on the training set and key process parameter set; Using key process parameters as independent variables and target performance indicators as dependent variables, the regression coefficients of the multiple linear regression equation are solved using the least squares method based on the training set samples. The model was optimized through significance tests of regression coefficients and multicollinearity tests, resulting in the mathematical expression of the multiple linear regression model; where... For the intercept term, For the first Partial regression coefficients of key process parameters, ; S4. Model validation and adaptive iterative optimization based on validation sets; The key process parameters from the validation set are substituted into the mathematical expression of the multiple linear regression model to calculate the predicted performance index values. These values ​​are then compared with the measured performance index values ​​from the validation set to calculate the prediction accuracy evaluation index. A dual verification is performed using both the model fit index and the prediction accuracy evaluation index. If both indices meet the preset conditions, the current model is determined as a tool performance prediction model suitable for online prediction; otherwise, the correlation coefficient threshold is adjusted. Return to step S2 to re-screen key process parameters and remodel until the preset conditions are met or the maximum number of iterations is reached. S5. Online prediction of tool performance and dynamic model updating; The system collects the current values ​​of key process parameters during the cold rolling finishing shearing process in real time, substitutes them into the mathematical expression of the tool performance prediction model, outputs the predicted value of the current tool performance index, and outputs graded tool status prompts based on the prediction results; at the same time, it triggers the rolling dynamic update of the model based on the measured data in the production process.

[0008] In step S1, the process parameters of the cold-rolled finishing shearing tool include at least five of the following: shearing tonnage, shearing gap, shearing speed, strip thickness, strip width, strip material hardness, cumulative shearing cycles, initial cutting edge radius, tool surface hardness, and tool coating thickness; the performance indicators include at least one of the following: tool wear, change in cutting edge radius, and remaining tool life; and the number of paired sample groups. It is a positive integer greater than or equal to 30; the proportion of the training set samples is 70% to 80% of the total number of samples collected.

[0009] In step S2, the Pearson correlation coefficient The calculation and significance test process are as follows: S21: Calculate the sample mean of the j-th process parameter in the training set. and the sample mean of the target performance index The calculation formula is: ; In the formula: The number of samples in the training set; For the training set The first group of samples Measured values ​​of the process parameters; For the training set Measured values ​​of the target performance indicators for the group of samples; S22: Calculate the first... in the training set Covariance between process parameters and target performance indicators The calculation formula is:

[0010] The formula uses an unbiased covariance calculation method with degrees of freedom correction to avoid covariance estimation bias under small samples and to maintain consistency with the degrees of freedom in the subsequent standard deviation calculation. S23: Calculate the standard deviation of the j-th process parameter in the training set. and the standard deviation of the target performance index The calculation formula is: ; The formula uses the unbiased standard deviation calculation method with degrees of freedom correction to ensure the unbiasedness of the correlation coefficient calculation and eliminate statistical bias caused by small samples; S24: Calculate the Pearson correlation coefficient The calculation formula is: ; In the formula: The value range is [-1, 1], positive values ​​indicate positive correlation, negative values ​​indicate negative correlation, and the closer the absolute value is to 1, the stronger the linear correlation. S25: Pearson correlation coefficient Significance The verification and calculation formula is as follows: ; In the formula: The statistic follows a set of degrees of freedom. of Distribution; setting significance level When the calculation is obtained At that time, the judgment of the first The linear correlation between the process parameters and the target performance indicators is significant; among them For degrees of freedom The significance level α corresponding to the two-sided significance level Distribution critical value; S26: For all Repeat steps S21 to S25 for each process parameter to obtain the correlation coefficient sequence. The corresponding significance test results will simultaneously satisfy... And through significance The process parameters being tested are sorted from largest to smallest according to the absolute value of their correlation coefficients, and then identified as the key process parameters.

[0011] The correlation coefficient threshold The initial value range is The correlation coefficient threshold The lower limit is 0.3; when the model verification in step S4 does not meet the preset conditions, the correlation coefficient threshold is gradually reduced by a preset step size. To introduce more key process parameters that pass the significance test; the preset step size adopts an adaptive adjustment method: the initial iteration stage uses a step size of 0.05, and when the difference between the prediction accuracy evaluation index and the preset condition is less than 10%, it switches to a step size of 0.02, taking into account both iteration efficiency and optimization accuracy.

[0012] The process of establishing and optimizing the multiple linear regression model in step S3 is as follows: S31: Construct a system containing N key process parameters and The training samples consist of a data matrix X and a target vector Y, where: ; In the formula: Matrix X is The matrix has all 1s in the first column, corresponding to the intercept term. ; Indicates the first training set The value of the kth critical process parameter in the sample group. , ; Y is A target vector, whose elements are the measured values ​​of the target performance indicators of the training set samples; S32: Solving the regression coefficient vector using the least squares method The calculation formula is: ; In the formula: Let X be the transpose of matrix X. For matrix The inverse matrix; verification is required before solving. It is a full-rank matrix, ensuring that the regression coefficients have a unique solution; S33: Perform a significance t-test on the obtained partial regression coefficients. The calculation formula is as follows: ; In the formula: Partial regression coefficients The standard error is due to The square root of the product of the k-th diagonal element of the matrix and the mean square of the model residuals is obtained; Set significance level ,when When the partial regression coefficient of the k-th key process parameter is determined to be significant, parameters that fail the significance test are removed, and the regression coefficients are recalculated to ensure that all independent variables in the model have significant explanatory power for the target performance index. S34: Perform multicollinearity tests on key process parameters, using the variance inflation factor (VIF) as the evaluation index. The calculation formula is as follows: ; In the formula: The coefficient of determination is denoted by ... When the parameter is determined to have no serious multicollinearity, it is determined that the parameter does not have severe multicollinearity; if it does exist... The parameters are removed, and the parameter with the smallest absolute value of the correlation coefficient is eliminated. S31 to S33 are executed again to eliminate the impact of multicollinearity on the stability of the model. S35: The mathematical expression for constructing the final optimized multiple linear regression model is: ; In the formula: The predicted value of the target performance index. These are the values ​​of N key process parameters after significance and multicollinearity tests. These are the regression coefficients obtained after optimizing S32 to S34.

[0013] In step S35, a goodness-of-performance evaluation index is fitted according to the multiple linear regression model, wherein the goodness-of-performance evaluation index includes the coefficient of determination. With Adjusted Determination Coefficient The calculation formulas are as follows: ; ; In the formula: To make the first The target performance index prediction value is obtained by substituting the key process parameters of the training samples into the mathematical expression of S35. The sample mean of the target performance index of the training set calculated in step S21; The coefficient of determination The value of is between 0 and 1, with a value closer to 1 indicating a higher degree of fit between the model and the training set data; the adjustment of the determination coefficient... The value ranges from 0 to 1, and the closer the value is to 1, the stronger the actual fitting ability of the model.

[0014] The adjustment determination coefficient As the core goodness-of-fit evaluation index for model validation in step S4, the overall significance F-test result of the model is also used as an auxiliary validation index. The formula for calculating the overall significance of the model using the F-test is as follows: ; In the formula: The F-statistic follows a set of degrees of freedom. The F-distribution; setting the significance level. ,when When the overall linear relationship of the model is significant, the adjusted coefficient of determination is determined. If the correlation coefficient threshold is less than the preset determination coefficient threshold, or if the model fails the F-test, the model reconstruction process is triggered regardless of whether the prediction accuracy evaluation index in S4 meets the preset conditions. The model reconstruction process includes: adjusting the correlation coefficient threshold. Then re-execute S2 to S3.

[0015] The specific process of model validation in step S4 is as follows: S41: For each sample in the validation set, substitute its key process parameters into the mathematical expression of the multiple linear regression model constructed in S3 to calculate the corresponding predictive performance index value. ; S42: Calculate the prediction accuracy evaluation metrics for the validation set, including Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and Nash Efficiency Coefficient (NSE). The formulas for these metrics are as follows: ； ; ； ; In the formula: The number of samples in the validation set; To verify the first set Predictive performance index values ​​for the group of samples; To verify the first set The measured performance index values ​​of the group of samples; To verify the mean of the measured values ​​of the target performance index of the set; Among them, MAE reflects the absolute error level between the predicted value and the measured value, RMSE is more sensitive to large error samples, MAPE reflects the relative error level, and NSE evaluates the model's ability to fit the trend of the measured data. The value range is (-∞, 1], and the closer it is to 1, the better the model's prediction performance. S43: Determine whether the prediction accuracy evaluation index meets the preset accuracy conditions; the preset accuracy conditions are: mean absolute percentage error (MAPE) ≤ 15%, or the ratio of MAE to the mean of the measured values ​​of the validation set ≤ 0.15, or the ratio of RMSE to the mean of the measured values ​​of the validation set ≤ 0.2, or the Nash efficiency coefficient (NSE) ≥ 0.75; if any of the above conditions are met and the goodness-of-fit index meets the requirements, the model is deemed to have passed the validation. When model validation fails to meet the preset conditions, the following adaptive iterative optimization process is executed: S431: Verify the current correlation coefficient threshold If the condition is not met, the iteration will terminate and a model optimization failure message will be output. If the condition is met, proceed to the next step. S432: Update the correlation coefficient threshold according to the adaptive step size. , in: This is the current iteration step size; S433: Based on the updated correlation coefficient threshold, re-execute S2 to screen key process parameters and obtain an updated set of key process parameters to ensure that the selected parameters meet both the threshold requirement and the significance test requirement. S434: Based on the updated set of key process parameters, re-execute S3 to establish and optimize the multiple linear regression model, and complete the regression coefficient solution, significance test and multicollinearity test; S435: Re-execute S41 to S43 to validate the updated model; repeat S431 to S435 until the model passes validation, or the number of iterations reaches the preset maximum number of iterations, or the correlation coefficient threshold drops to the lower limit.

[0016] The process of applying the tool performance prediction model for online prediction in step S5 is as follows: S51: Real-time values ​​of the N key process parameters under current production conditions are collected via pressure sensors, displacement sensors, encoders, hardness testers, and PLC control systems installed on the cold rolling finishing shearing production line. ; S52: Preprocess the acquired real-time process parameters, including based on... Outlier removal and min-max normalization are used to eliminate the impact of data noise and dimensional differences on the prediction results. S53: Substitute the pre-processed real-time values ​​of the N key process parameters into the mathematical expression of the multiple linear regression model established in S35: ; The predicted values ​​of the tool performance indicators at the current moment are calculated. ; S54: Based on the predicted value The output will display a graded tool status message within the specified value range. When the target performance index is the amount of tool wear or the change in the radius of the cutting edge: if the predicted value is less than 60% of the preset threshold, the output tool is in normal condition; if the predicted value is between 60% and 80% of the preset threshold, the output tool is slightly worn and should be monitored; if the predicted value is between 80% and 100% of the preset threshold, the output tool is severely worn and should be prepared for replacement; if the predicted value reaches or exceeds the preset threshold, the output tool has reached its wear limit and should be replaced immediately. When the target performance indicator is the remaining tool life: if the predicted value is greater than 200% of the preset threshold, the output tool status is normal; if the predicted value is between 100% and 200% of the preset threshold, the remaining tool life is sufficient for normal use; if the predicted value is between 50% and 100% of the preset threshold, the remaining tool life is insufficient, and a replacement plan is formulated; if the predicted value is lower than the preset threshold, the output tool has reached its service life and an emergency warning message is issued for immediate replacement. S55: The real-time collected process parameters, predicted performance index values, and subsequent measured tool performance index values ​​are paired and stored to form new samples, providing data support for dynamic model updates.

[0017] The model dynamic update mechanism in step S5 is as follows: S56: Set two model update trigger conditions; an update will be triggered if either condition is met: Periodic update trigger: After each tool change, all newly added paired samples accumulated during the tool usage cycle are added to the original sample set to form an expanded sample set; when the number of new samples in the expanded sample set reaches 20% to 50% of the number of samples in the original sample set, a full update is triggered. Emergency update trigger: When the relative prediction error of 10 consecutive production samples exceeds 20%, the model is determined to have a working condition drift, and an incremental update is immediately triggered. S57: Execute the corresponding update method for different triggering conditions: Full update: Based on the expanded sample set, S1 to S4 are re-executed to complete the re-partitioning of the dataset, re-screening of key process parameters, model reconstruction and verification, and obtain the updated tool performance prediction model; Incremental Update: Based on the new samples, the regression coefficients of the existing model are incrementally corrected using the weighted least squares method. The weight coefficients are positively correlated with the collection time of the new samples; the newer the samples, the higher the weight. The calculation formula is as follows: ; In the formula: This represents the regression coefficient vector of the current model; This is the vector of regression coefficients after incremental updates; , The data matrix and target vector for the newly added samples; This is the data matrix of the original training set; W is a diagonal weight matrix, where the diagonal elements are the weight values ​​of the corresponding newly added samples; S58: Perform backtesting on the updated model, using historical validation set samples to test the prediction accuracy of the updated model, ensuring that the accuracy is not lower than that of the original model; S59: Replace the currently running online model with the updated tool performance prediction model that has been verified through backtracking, complete the rolling update of the model, and archive the updated model parameters and verification results.

[0018] Beneficial effects: 1. This invention, through a dedicated design based on the working conditions and tool failure characteristics of the cold rolling finishing shearing process, constructs a complete technical closed loop from sample collection, key parameter screening, model construction and optimization, model verification and iteration to online prediction and dynamic updating. This enables accurate and stable prediction of the performance of cold rolling finishing shearing tools, and can effectively adapt to the changing production conditions such as strip specification switching and process parameter adjustment in the cold rolling field, solving the problems of insufficient targeting and poor adaptability of existing technologies.

[0019] 2. This invention adopts a key process parameter screening method based on the Pearson correlation coefficient of the training set samples combined with statistical significance test, which avoids the data leakage problem caused by full sample screening from the root. At the same time, it effectively eliminates process parameters that are weakly or spuriously correlated with tool performance. This ensures that the screened key parameters have a strong linear correlation with tool performance, simplifies the model input dimension, and greatly improves the model's generalization ability and prediction stability in field applications.

[0020] 3. This invention uses a multiple linear regression model to replace the complex black-box deep learning model. At the same time, it combines the significance test of regression coefficients and the multicollinearity test to complete the model optimization. The final model has a simple structure and the regression parameters have clear physical meanings. On-site process personnel can intuitively understand the influence of each process parameter on tool performance, which is convenient for on-site debugging, maintenance and process optimization. At the same time, the model has a small amount of computation and low hardware computing power requirements. It can be directly deployed in conventional PLCs and edge computing devices in industrial sites to realize real-time online prediction of tool performance, which greatly reduces the threshold for engineering implementation of the technology.

[0021] 4. This invention designs a model verification mechanism that verifies both goodness of fit and prediction accuracy, and is equipped with an adaptive iterative optimization process. When the model accuracy is not up to standard, it can automatically adjust the feature selection threshold to complete the reselection of key parameters and model reconstruction optimization. It ensures the reliability of the model from two dimensions: training fit effect and actual prediction ability, and solves the problems of insufficient model verification and insufficient accuracy in actual application of existing technologies.

[0022] 5. This invention sets up a dynamic model update mechanism with dual trigger modes, which can complete the rolling update and optimization of the model based on the actual operating data accumulated during the production process. This enables the model to adapt to long-term changes in the status and operating conditions of production line equipment and maintain stable prediction accuracy. This solves the problems of accuracy decay and inability to adapt to operating condition drift after long-term operation of existing technology models.

[0023] 6. This invention provides intuitive tool maintenance guidance to on-site production personnel through a graded tool status prompt and early warning mechanism, which facilitates the advance formulation of tool replacement plans, effectively avoids unplanned downtime accidents, reduces strip shearing quality defects caused by excessive tool wear, significantly improves the production efficiency and product qualification rate of the cold rolling finishing process, and reduces production line maintenance and production costs.

[0024] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0028] Example 1: like Figure 1 As shown, a performance prediction method for cold rolling finishing shearing tools based on Pearson correlation coefficient includes the following steps: S1. Sample collection and dataset partitioning; This step provides basic data support for model construction. The specific implementation method is as follows: First, obtain the set of process parameters and performance indicators for the cold-rolled finishing shearing tool. The process parameters for the cold-rolled finishing shearing tool include at least five of the following: shearing tonnage, shearing gap, shearing speed, strip thickness, strip width, strip material hardness, cumulative shearing cycles, initial cutting edge radius, tool surface hardness, and tool coating thickness. The performance indicators include at least one of the following: tool wear, change in cutting edge radius, and remaining tool life. The process parameters are collected as follows: the shearing tonnage is collected in real time by the pressure sensor matched with the main hydraulic cylinder of the shearing machine; the shearing gap is collected by the displacement sensor at the cutter shaft; the shearing speed is collected by the encoder of the main shaft of the shearing machine; the strip thickness, strip width, and strip material hardness are obtained in real time by the online thickness gauge, width gauge, and hardness tester of the production line; the cumulative shearing times of the tool are counted and statistically analyzed in real time by the production line PLC system; and the initial cutting edge radius, tool surface hardness, and tool coating thickness are entered into the system after being tested by metallographic microscope, Vickers hardness tester, and coating thickness gauge during the incoming inspection process. The performance indicators are collected as follows: the tool wear amount and the change in the blunt radius of the tool cutting edge are measured offline when the machine is stopped by the laser profilometer and tool image detection system that are equipped with the production line. The remaining tool life is calculated based on the correspondence between the measured wear amount and the preset tool failure threshold. Each set of synchronously acquired process parameters and the corresponding set of measured performance indicators constitute a paired sample, which are collected in total. Paired samples, number of paired sample groups It is a positive integer greater than or equal to 30, satisfying the minimum sample size requirement of mathematical statistics; The paired samples are divided into non-overlapping training and validation sets according to a preset ratio, with the training set having a sample size of [missing information]. The validation set sample size is ,satisfy The training set accounts for 70%-80% of the total samples, and the validation set accounts for 20%-30% of the total samples.

[0029] S2. Screening of key process parameters without data leakage based on the training set and Pearson correlation coefficient; This step performs feature selection solely based on training set samples, completely avoiding validation set data leakage. It also eliminates redundant parameters that are weakly or spuriously correlated with tool performance, simplifying the model input dimension. The specific implementation method is as follows: For the training set samples, for the first The first group of training samples Process parameters With the Target performance metrics in training samples Pearson correlation coefficients were calculated to obtain the correlation coefficient values ​​between each process parameter and the target performance index. Set the correlation coefficient threshold. ,Will Furthermore, the process parameters that pass the significance test are identified as critical process parameters and are recorded in the selected critical process parameter set, which contains N parameters. Total number of process parameters ; Among them, the Pearson correlation coefficient The calculation and significance test process are as follows: S21: Sample mean calculation; Calculate the sample mean of the j-th process parameter in the training set. and the sample mean of the target performance index The calculation formula is: ; In the formula: The number of samples in the training set; For the training set The first group of samples Measured values ​​of the process parameters; For the training set Measured values ​​of the target performance indicators for the group of samples; S22: Covariance calculation; Calculate the first in the training set Covariance between process parameters and target performance indicators The calculation formula is:

[0030] The formula uses an unbiased covariance calculation method with degrees of freedom correction to avoid covariance estimation bias under small samples and to maintain consistency with the degrees of freedom in the subsequent standard deviation calculation. S23: Standard deviation calculation; Calculate the standard deviation of the j-th process parameter in the training set. and the standard deviation of the target performance index The calculation formula is: ; The formula uses the unbiased standard deviation calculation method with degrees of freedom correction to ensure the unbiasedness of the correlation coefficient calculation and eliminate statistical bias caused by small samples; S24: Calculation of Pearson correlation coefficient; Calculate the Pearson correlation coefficient The calculation formula is: ; In the formula: The value range is [-1, 1], positive values ​​indicate positive correlation, negative values ​​indicate negative correlation, and the closer the absolute value is to 1, the stronger the linear correlation. S25: Significance t-test; Pearson correlation coefficient Significance The verification and calculation formula is as follows: ; In the formula: The statistic follows a set of degrees of freedom. of Distribution; setting significance level When the calculation is obtained At that time, the judgment of the first The linear correlation between the process parameters and the target performance indicators is significant; among them For degrees of freedom The significance level α corresponding to the two-sided significance level Distribution critical value; S26: Determination of key process parameters; For all Repeat steps S21 to S25 for each process parameter to obtain the correlation coefficient sequence. The corresponding significance test results will simultaneously satisfy... And through significance The process parameters being tested are sorted from largest to smallest according to the absolute value of their correlation coefficients, and then identified as the key process parameters.

[0031] In the specific implementation of this step, the correlation coefficient threshold The initial value range is The correlation coefficient threshold The lower limit is set at 0.3 to avoid introducing weakly correlated parameters that could lead to model overfitting. When subsequent model validation fails to meet the preset conditions, the correlation coefficient threshold is gradually reduced with a preset step size. To introduce more key process parameters that pass the significance test; the preset step size adopts an adaptive adjustment method: a large step size of 0.05 is used in the initial iteration stage, and when the difference between the prediction accuracy evaluation index and the preset condition is less than 10%, it is switched to a small step size of 0.02, taking into account both iteration efficiency and optimization accuracy.

[0032] S3. Construction of a multiple linear regression model based on the training set and key process parameter set; This step, based on the selected key process parameters, constructs a multiple linear regression model with clear physical meaning and high stability. The specific implementation method is as follows: Using key process parameters as independent variables and target performance indicators as dependent variables, the regression coefficients of the multiple linear regression equation are solved using the least squares method based on the training set samples. The model was optimized through significance tests of regression coefficients and multicollinearity tests, resulting in the mathematical expression of the multiple linear regression model; where... For the intercept term, For the first Partial regression coefficients of key process parameters, ; The process of establishing and optimizing the multiple linear regression model is as follows: S31: Construction of data matrix and target vector; Construct a system containing N key process parameters and The training samples consist of a data matrix X and a target vector Y, where: ; In the formula: Matrix X is The matrix has all 1s in the first column, corresponding to the intercept term. ; Indicates the first training set The value of the kth critical process parameter in the sample group. , ; Y is A target vector, whose elements are the measured values ​​of the target performance indicators of the training set samples; S32: Solving for regression coefficients; Solving the regression coefficient vector using the least squares method The calculation formula is: ; In the formula: Let X be the transpose of matrix X. For matrix The inverse matrix; verification is required before solving. It is a full-rank matrix, ensuring that the regression coefficients have a unique solution; S33: t-test for the significance of regression coefficients; The significance of the partial regression coefficients obtained is tested using a t-test, and the calculation formula is as follows: ; In the formula: Partial regression coefficients The standard error is due to The square root of the product of the k-th diagonal element of the matrix and the mean square of the model residuals is obtained; Set significance level ,when When the partial regression coefficient of the k-th key process parameter is determined to be significant, parameters that fail the significance test are removed, and the regression coefficients are recalculated to ensure that all independent variables in the model have significant explanatory power for the target performance index. S34: Multicollinearity test; Multicollinearity tests were performed on key process parameters, and the variance inflation factor (VIF) was used as the evaluation index. The calculation formula is as follows: ; In the formula: The coefficient of determination is denoted by ... When the parameter is determined to have no serious multicollinearity, it is determined that the parameter does not have severe multicollinearity; if it does exist... The parameters are removed, and the parameter with the smallest absolute value of the correlation coefficient is eliminated. S31 to S33 are executed again to eliminate the impact of multicollinearity on the stability of the model. S35: Model expression determined; The mathematical expression for constructing the final optimized multiple linear regression model is as follows: ; In the formula: The predicted value of the target performance index. These are the values ​​of N key process parameters after significance and multicollinearity tests. These are the regression coefficients obtained after optimizing S32 to S34.

[0033] S36: Goodness-of-fit evaluation; After the model is built, the goodness-of-fit evaluation index of the multiple linear regression model is calculated, including the coefficient of determination. With Adjusted Determination Coefficient The calculation formulas are as follows: ; ; In the formula: To make the first The target performance index prediction value is obtained by substituting the key process parameters of the training samples into the mathematical expression of S35. The sample mean of the target performance index of the training set calculated in step S21; The coefficient of determination The value of is between 0 and 1, with a value closer to 1 indicating a higher degree of fit between the model and the training set data; the adjustment of the determination coefficient... The number of independent variables and sample size were adjusted for degrees of freedom, eliminating ordinary... The artificially high bias that increases with the number of independent variables can more objectively evaluate the goodness of fit of a multivariate model. Its value ranges from 0 to 1, and the closer the value is to 1, the stronger the actual fitting ability of the model.

[0034] The adjustment determination coefficient calculated in this step As the core goodness-of-fit evaluation index for subsequent model validation, the overall significance F-test results of the model are also used as auxiliary validation indexes. The formula for calculating the overall significance of the model using the F-test is as follows: ; In the formula: The F-statistic follows a set of degrees of freedom. The F-distribution; setting the significance level. ,when When the model is determined to have a significant overall linear relationship, it is considered to be true.

[0035] When the adjustment determination coefficient If the correlation coefficient threshold is less than the preset determination coefficient threshold, or if the model fails the F-test, the model reconstruction process is triggered regardless of whether the prediction accuracy evaluation index in S4 meets the preset conditions. The model reconstruction process includes: adjusting the correlation coefficient threshold. Then re-execute S2 to S3.

[0036] S4. Model validation and adaptive iterative optimization based on validation sets; This step verifies the model's generalization ability using a validation set, and simultaneously ensures that the model meets the accuracy requirements of field applications through adaptive iterative optimization. The specific implementation method is as follows: The key process parameters from the validation set are substituted into the mathematical expression of the multiple linear regression model to calculate the predicted performance index values. These values ​​are then compared with the measured performance index values ​​from the validation set to calculate the prediction accuracy evaluation index. A dual verification is performed using both the model fit index and the prediction accuracy evaluation index. If both indices meet the preset conditions, the current model is determined as a tool performance prediction model suitable for online prediction; otherwise, the correlation coefficient threshold is adjusted. Return to step S2 to re-screen key process parameters and remodel until the preset conditions are met or the maximum number of iterations is reached. The specific process of model validation is as follows: S41: Validation set prediction calculation; For each sample in the validation set, its key process parameters are substituted into the mathematical expression of the multiple linear regression model constructed in S3 to calculate the corresponding predictive performance index value. ; S42: Calculation of prediction accuracy evaluation index; The prediction accuracy evaluation metrics for the validation set are calculated, including Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and Nash Efficiency Coefficient (NSE). The formulas for these calculations are as follows: ； ; ； ; In the formula: The number of samples in the validation set; To verify the first set Predictive performance index values ​​for the group of samples; To verify the first set The measured performance index values ​​of the group of samples; To verify the mean of the measured values ​​of the target performance index of the set; Among them, MAE reflects the absolute error level between the predicted value and the measured value, RMSE is more sensitive to large error samples, MAPE reflects the relative error level, and NSE evaluates the model's ability to fit the trend of the measured data. The value range is (-∞, 1], and the closer it is to 1, the better the model's prediction performance. S43: Precision condition check; Determine whether the prediction accuracy evaluation index meets the preset accuracy conditions; the preset accuracy conditions are: mean absolute percentage error (MAPE) ≤ 15%, or the ratio of MAE to the mean of the measured values ​​of the validation set ≤ 0.15, or the ratio of RMSE to the mean of the measured values ​​of the validation set ≤ 0.2, or the Nash efficiency coefficient (NSE) ≥ 0.75; if any of the above conditions are met and the goodness-of-fit index meets the requirements, the model is deemed to have passed the validation. When model validation fails to meet the preset conditions, the following adaptive iterative optimization process is executed: S431: Iteration condition verification; Validate the current correlation coefficient threshold If the condition is not met, the iteration will terminate and a model optimization failure message will be output. If the condition is met, proceed to the next step. S432: Threshold update; Update the correlation coefficient threshold according to the adaptive step size. , in: This is the current iteration step size; S433: Key Parameter Rescreening: Based on the updated correlation coefficient threshold, S2 is re-executed to screen key process parameters, resulting in an updated set of key process parameters, ensuring that the selected parameters simultaneously meet the threshold requirements and significance test requirements. S434: Model Reconstruction; Based on the updated set of key process parameters, S3 was re-executed to establish and optimize the multiple linear regression model, and the regression coefficients were solved, significance tests and multicollinearity tests were completed. S435: Iterative verification; Re-execute S41 to S43 to validate the updated model; repeat S431 to S435 until the model passes validation, or the number of iterations reaches the preset maximum number of iterations, or the correlation coefficient threshold drops to the lower limit.

[0037] S5. Online prediction of tool performance and dynamic model updating; This step enables the industrial application of the model, while ensuring the long-term accuracy and stability of the model through a dynamic update mechanism. The specific implementation method is as follows: The system collects the current values ​​of key process parameters during the cold rolling finishing shearing process in real time, substitutes them into the mathematical expression of the tool performance prediction model, outputs the predicted value of the current tool performance index, and outputs graded tool status prompt information based on the prediction results; at the same time, it triggers the rolling dynamic update of the model based on the measured data in the production process. The process of online prediction by the tool performance prediction model is as follows: S51: Real-time data acquisition; The system uses pressure sensors, displacement sensors, encoders, hardness testers, and a PLC control system installed on the cold rolling finishing shearing production line to collect real-time values ​​of the N key process parameters under the current production conditions. ; S52: Data preprocessing; The collected real-time process parameters are preprocessed, including based on Outlier removal and min-max normalization are used to eliminate the impact of data noise and dimensional differences on the prediction results. S53: Online prediction calculation; Substitute the preprocessed real-time values ​​of N key process parameters into the mathematical expression of the multiple linear regression model established in S35: ; The predicted values ​​of the tool performance indicators at the current moment are calculated. ; S54: Hierarchical status indication; According to the predicted value The output will display a graded tool status message within the specified value range. When the target performance index is the amount of tool wear or the change in the radius of the cutting edge: if the predicted value is less than 60% of the preset threshold, the output tool is in normal condition; if the predicted value is between 60% and 80% of the preset threshold, the output tool is slightly worn and should be monitored; if the predicted value is between 80% and 100% of the preset threshold, the output tool is severely worn and should be prepared for replacement; if the predicted value reaches or exceeds the preset threshold, the output tool has reached its wear limit and should be replaced immediately. When the target performance indicator is the remaining tool life: if the predicted value is greater than 200% of the preset threshold, the output tool status is normal; if the predicted value is between 100% and 200% of the preset threshold, the remaining tool life is sufficient for normal use; if the predicted value is between 50% and 100% of the preset threshold, the remaining tool life is insufficient, and a replacement plan is formulated; if the predicted value is lower than the preset threshold, the output tool has reached its service life and an emergency warning message is issued for immediate replacement. S55: Sample storage; The real-time collected process parameters, predicted performance index values, and subsequent measured tool performance index values ​​are paired and stored to form new samples, providing data support for dynamic model updates.

[0038] Furthermore, the model dynamic update in step S5 is carried out as follows: S56: Update trigger condition settings: Two model update trigger conditions are set; an update is triggered when either condition is met: Periodic update trigger: After each tool change, all newly added paired samples accumulated during the tool usage cycle are added to the original sample set to form an expanded sample set; when the number of new samples in the expanded sample set reaches 20% to 50% of the number of samples in the original sample set, a full update is triggered. Emergency update trigger: When the relative prediction error of 10 consecutive production samples exceeds 20%, the model is determined to have a working condition drift, and an incremental update is immediately triggered. S57: Model update execution: Execute the corresponding update method for different triggering conditions: Full update: Based on the expanded sample set, S1 to S4 are re-executed to complete the re-partitioning of the dataset, re-screening of key process parameters, model reconstruction and verification, and obtain the updated tool performance prediction model; Incremental Update: Based on the new samples, the regression coefficients of the existing model are incrementally corrected using the weighted least squares method. The weight coefficients are positively correlated with the collection time of the new samples; the newer the samples, the higher the weight. The calculation formula is as follows: ; In the formula: This represents the regression coefficient vector of the current model; This is the vector of regression coefficients after incremental updates; , The data matrix and target vector for the newly added samples; This is the data matrix of the original training set; W is a diagonal weight matrix, where the diagonal elements are the weight values ​​of the corresponding newly added samples; S58: Backtracking Verification: The updated model is backtested, and the prediction accuracy of the updated model is tested using historical validation set samples to ensure that the accuracy is not lower than that of the original model. S59: Model Replacement: Replace the currently running online model with the updated tool performance prediction model that has been validated through backtracking, complete the rolling update of the model, and archive the updated model parameters and validation results.

[0039] Example 2: Reference Figure 1 As shown, this embodiment uses a disc shear tool from a 1700mm cold rolling finishing line in a domestic steel mill as the prediction object. The target performance index is the wear amount on the back face of the tool, and the preset tool wear failure threshold is 0.3mm. The specific implementation process is as follows: Step 1: Sample Collection and Dataset Division Production data for the entire lifecycle of the cutting tools for three consecutive cycles were collected from the production line. Ten process parameters were selected: shearing tonnage X1, shearing gap X2, shearing speed X3, strip thickness X4, strip width X5, strip material hardness X6, cumulative shearing times of the cutting tool X7, initial cutting edge radius of the cutting tool X8, surface hardness of the cutting tool X9, and coating thickness of the cutting tool X10. That is, the total number of process parameters J=10. The target performance index is the tool flank wear Y. Offline measurement is performed every 2000 shearing cycles using a laser profilometer. Process parameters are paired with wear values ​​one-to-one, and a total of valid paired samples are collected. Groups that meet the sample size requirement of ≥30.

[0040] The training and validation sets are divided in a 75%:25% ratio, with the training set having the largest sample size. Group, validation set sample size The two groups of samples have no data overlap. Step 2: Screening of key process parameters: Set the initial correlation coefficient threshold significance level Based on 90 training set samples, the Pearson correlation coefficient and t-test value of each process parameter and tool wear were calculated, as well as the critical value of the t-distribution. The calculation results are shown in the table below: Table 1 Screening Table for Key Process Parameters

[0041] After screening, the key process parameters that simultaneously meet the threshold requirements and significance tests are: X7 cumulative shearing times, X4 strip thickness, X1 shearing tonnage, and X6 strip material hardness, for a total of N=4 parameters, which are sorted from largest to smallest by the absolute value of their correlation coefficients.

[0042] Step 3: Construction and optimization of the multiple linear regression model: Based on four key process parameters, a 90×5 dimensional training data matrix X (with all 1s in the first column) and a 90×1 dimensional target vector Y are constructed. The regression coefficients are solved using the least squares method to obtain the initial regression coefficient vector: ; The corresponding mathematical expression for the model is: ; A significance t-test was performed on the regression coefficients, and the |t| values ​​of all four partial regression coefficients were greater than 1. All parameters were significant; the variance inflation factor (VIF) for each parameter was calculated and the results were 1.23, 1.18, 1.25, and 1.09, respectively, all less than 10, indicating no serious multicollinearity and no need to remove parameters.

[0043] Calculate the goodness of fit of the model: coefficient of determination Adjust the coefficient of determination The preset determination coefficient threshold is 0.75, which meets the requirements; the overall F-test value of the model is F=178.6, which is greater than... The overall linear relationship of the model is significant.

[0044] Step 4: Model Validation and Iterative Optimization Substituting four key process parameters from 30 validation set samples into the model, the predicted wear amount was calculated and compared with the measured values ​​to obtain the prediction accuracy evaluation index: The mean absolute error (MAE) is 0.0082 mm, and the ratio of MAE to the measured mean of the validation set is 0.12, ≤0.15. The root mean square error (RMSE) is 0.0095 mm, and the ratio of RMSE to the measured mean of the validation set is 0.14, which is ≤0.2. Mean absolute percentage error (MAPE) = 11.8%, ≤15%; The Nash efficiency coefficient NSE = 0.87, ≥ 0.75.

[0045] All accuracy indicators met the preset conditions, and the adjustment of the determination coefficients met the requirements. The model passed the verification and was determined to be a tool wear prediction model that can be used online.

[0046] Step 5: Online Prediction and Dynamic Model Updates The validated model is deployed to the production line edge computing gateway, connecting in real time with the production line PLC and sensor system. During online operation, key process parameters at a specific moment are collected in real time, including shearing tonnage. strip thickness Hardness of strip steel material Total number of cuts by the blade Second-rate.

[0047] Substituting the parameters into the model, the predicted value of the current tool wear is calculated:

[0048]

[0049] The preset wear failure threshold is 0.3mm. When the predicted value is in the range of 80% to 100% of the threshold, the system outputs a graded prompt message of "severe tool wear, prepare for replacement". On-site personnel formulate a tool replacement plan based on the prompt and complete the tool replacement during the planned downtime, thus avoiding unplanned downtime.

[0050] After the completion of this tool's full lifecycle, a total of 36 new valid paired samples were added, reaching 30% of the original sample size of 120, triggering a full update of the cycle. Based on the 156 expanded samples, the dataset was re-divided, key parameters were screened, and the model was built and validated to obtain the updated prediction model. After backtracking verification showed that the accuracy was better than the original model, the online running model was replaced, completing a rolling update.

[0051] This embodiment verifies that the method of the present invention can accurately predict the wear state of cold rolling finishing shearing tools. The prediction accuracy meets the needs of industrial sites, can effectively guide on-site tool maintenance, reduce unplanned downtime, and improve strip shearing quality and production line efficiency.

[0052] Where there is no conflict, those skilled in the art can combine the relevant technical features in the above examples according to the actual situation to achieve the corresponding technical effects. Specific details of various combinations are not elaborated here; for components not specifically described, the prior art is used.

[0053] Furthermore, the use of terms such as "first" and "second" in this invention is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features.

[0054] The above description is merely a preferred embodiment of the present invention. The present invention is not limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein. Any simple modifications, equivalent variations, and alterations made to the above embodiments based on the technical essence of the present invention shall still fall within the scope of the present invention.

Claims

1. A performance prediction method for cold rolling finishing shearing tools based on Pearson correlation coefficient, characterized in that: Includes the following steps, S1. Sample collection and dataset partitioning; The process parameter set and performance index set of cold rolling finishing shearing tools were obtained, and the process parameter set and performance index set were combined into a pair of samples, and a total of samples were collected. Group paired samples; The paired samples are divided into non-overlapping training and validation sets according to a preset ratio, with the training set having a sample size of [missing information]. The size of the validation set is ,satisfy ; S2. Screening of key process parameters without data leakage based on the training set and Pearson correlation coefficient; For the training set samples, for the first The first group of training samples Process parameters With the Target performance metrics in training samples Pearson correlation coefficients were calculated to obtain the correlation coefficient values ​​between each process parameter and the target performance index. ; Set correlation coefficient threshold ,Will Furthermore, the process parameters that pass the significance test are identified as critical process parameters and are recorded in the selected critical process parameter set, which contains N parameters. Total number of process parameters ; S3. Construction of a multiple linear regression model based on the training set and key process parameter set; Using key process parameters as independent variables and target performance indicators as dependent variables, the regression coefficients of the multiple linear regression equation are solved using the least squares method based on the training set samples. The model was optimized through significance tests of regression coefficients and multicollinearity tests, resulting in the mathematical expression of the multiple linear regression model; where... For the intercept term, For the first Partial regression coefficients of key process parameters, ; S4. Model validation and adaptive iterative optimization based on validation sets; The key process parameters from the validation set are substituted into the mathematical expression of the multiple linear regression model to calculate the predicted performance index values. These values ​​are then compared with the measured performance index values ​​from the validation set to calculate the prediction accuracy evaluation index. A dual verification is performed using both the model fit index and the prediction accuracy evaluation index. If both indices meet the preset conditions, the current model is determined as a tool performance prediction model suitable for online prediction; otherwise, the correlation coefficient threshold is adjusted. Return to step S2 to re-screen key process parameters and remodel until the preset conditions are met or the maximum number of iterations is reached. S5. Online prediction of tool performance and dynamic model updating; The system collects the current values ​​of key process parameters during the cold rolling finishing shearing process in real time, substitutes them into the mathematical expression of the tool performance prediction model, outputs the predicted value of the current tool performance index, and outputs graded tool status prompts based on the prediction results; at the same time, it triggers the rolling dynamic update of the model based on the measured data in the production process.

2. The performance prediction method for cold rolling finishing shearing tools based on Pearson correlation coefficient as described in claim 1, characterized in that: In step S1, the process parameters of the cold-rolled finishing shearing tool include at least five of the following: shearing tonnage, shearing gap, shearing speed, strip thickness, strip width, strip material hardness, cumulative shearing cycles, initial cutting edge radius, tool surface hardness, and tool coating thickness; the performance indicators include at least one of the following: tool wear, change in cutting edge radius, and remaining tool life; and the number of paired sample groups. It is a positive integer greater than or equal to 30; the proportion of the training set samples is 70% to 80% of the total number of samples collected.

3. The performance prediction method for cold rolling finishing shearing tools based on Pearson correlation coefficient as described in claim 1 or 2, characterized in that: In step S2, the Pearson correlation coefficient The calculation and significance test process are as follows: S21: Calculate the sample mean of the j-th process parameter in the training set. and the sample mean of the target performance index The calculation formula is: ； In the formula: The number of samples in the training set; For the training set The first group of samples Measured values ​​of the process parameters; For the training set Measured values ​​of the target performance indicators for the group of samples; S22: Calculate the first... in the training set Covariance between process parameters and target performance indicators The calculation formula is: The formula uses an unbiased covariance calculation method with degrees of freedom correction to avoid covariance estimation bias under small samples and to maintain consistency with the degrees of freedom in the subsequent standard deviation calculation. S23: Calculate the standard deviation of the j-th process parameter in the training set. and the standard deviation of the target performance index The calculation formula is: ； The formula uses the unbiased standard deviation calculation method with degrees of freedom correction to ensure the unbiasedness of the correlation coefficient calculation and eliminate statistical bias caused by small samples; S24: Calculate the Pearson correlation coefficient The calculation formula is: ； In the formula: The value range is [-1, 1], positive values ​​indicate positive correlation, negative values ​​indicate negative correlation, and the closer the absolute value is to 1, the stronger the linear correlation. S25: Pearson correlation coefficient Significance The verification and calculation formula is as follows: ； In the formula: The statistic follows a set of degrees of freedom. of Distribution; setting significance level When the calculation is obtained At that time, the judgment of the first The linear correlation between the process parameters and the target performance indicators is significant; among them For degrees of freedom The significance level α corresponding to the two-sided significance level Distribution critical value; S26: For all Repeat steps S21 to S25 for each process parameter to obtain the correlation coefficient sequence. With the corresponding significance test results; will simultaneously satisfy And through significance The process parameters being tested are sorted from largest to smallest according to the absolute value of their correlation coefficients, and then identified as the key process parameters.

4. The performance prediction method for cold rolling finishing shearing tools based on Pearson correlation coefficient as described in claim 3, characterized in that: The correlation coefficient threshold The initial value range is The correlation coefficient threshold The lower limit is 0.3; when the model verification in step S4 does not meet the preset conditions, the correlation coefficient threshold is gradually reduced by a preset step size. To introduce more key process parameters that pass the significance test; the preset step size adopts an adaptive adjustment method: the initial iteration stage uses a step size of 0.05, and when the difference between the prediction accuracy evaluation index and the preset condition is less than 10%, it switches to a step size of 0.02, taking into account both iteration efficiency and optimization accuracy.

5. The performance prediction method for cold rolling finishing shearing tools based on Pearson correlation coefficient as described in claim 4, characterized in that: The process of establishing and optimizing the multiple linear regression model in step S3 is as follows: S31: Construct a system containing N key process parameters and The training samples consist of a data matrix X and a target vector Y, where: ； In the formula: Matrix X is The matrix has all 1s in the first column, corresponding to the intercept term. ; Indicates the first training set The value of the kth critical process parameter in the sample group. , ; Y is A target vector, whose elements are the measured values ​​of the target performance indicators of the training set samples; S32: Solving the regression coefficient vector using the least squares method The calculation formula is: ； In the formula: Let X be the transpose of matrix X. For matrix The inverse matrix; verification is required before solving. It is a full-rank matrix, ensuring that the regression coefficients have a unique solution; S33: Perform a significance t-test on the obtained partial regression coefficients. The calculation formula is as follows: ； In the formula: Partial regression coefficients The standard error is due to The square root of the product of the k-th diagonal element of the matrix and the mean square of the model residuals is obtained; Set significance level ,when When the partial regression coefficient of the k-th key process parameter is determined to be significant, parameters that fail the significance test are removed, and the regression coefficients are recalculated to ensure that all independent variables in the model have significant explanatory power for the target performance index. S34: Perform multicollinearity tests on key process parameters, using the variance inflation factor (VIF) as the evaluation index. The calculation formula is as follows: ； In the formula: The coefficient of determination is the linear regression model constructed with the k-th key process parameter as the dependent variable and the remaining N-1 key process parameters as independent variables; when When the parameter is determined to have no severe multicollinearity, it is determined that the parameter does not have severe multicollinearity; if it does exist... The parameters are removed, and the parameter with the smallest absolute value of the correlation coefficient is eliminated. S31 to S33 are executed again to eliminate the impact of multicollinearity on the stability of the model. S35: The mathematical expression for constructing the final optimized multiple linear regression model is: ； In the formula: The predicted value of the target performance index. These are the values ​​of N key process parameters after significance and multicollinearity tests. These are the regression coefficients obtained after optimizing S32 to S34.

6. The performance prediction method for cold rolling finishing shearing tools based on Pearson correlation coefficient as described in claim 5, characterized in that: In step S35, a goodness-of-performance evaluation index is fitted according to the multiple linear regression model, wherein the goodness-of-performance evaluation index includes the coefficient of determination. With Adjusted Determination Coefficient The calculation formulas are as follows: ； ； In the formula: To make the first The target performance index prediction value is obtained by substituting the key process parameters of the training samples into the mathematical expression of S35. The sample mean of the target performance index of the training set calculated in step S21; The coefficient of determination The value of is between 0 and 1, with a value closer to 1 indicating a higher degree of fit between the model and the training set data; the adjustment of the determination coefficient... The value ranges from 0 to 1, and the closer the value is to 1, the stronger the actual fitting ability of the model.

7. The performance prediction method for cold rolling finishing shearing tools based on Pearson correlation coefficient as described in claim 6, characterized in that: The adjustment determination coefficient As the core goodness-of-fit evaluation index for model validation in step S4, the overall significance F-test result of the model is also used as an auxiliary validation index. The formula for calculating the overall significance of the model using the F-test is as follows: ； In the formula: The F-statistic follows a set of degrees of freedom. The F-distribution; setting the significance level. ,when When the overall linear relationship of the determination model is significant, the adjusted coefficient of determination is determined. If the correlation coefficient threshold is less than the preset determination coefficient threshold, or if the model fails the F-test, the model reconstruction process is triggered regardless of whether the prediction accuracy evaluation index in S4 meets the preset conditions. The model reconstruction process includes: adjusting the correlation coefficient threshold. Then re-execute S2 to S3.

8. A performance prediction method for cold rolling finishing shearing tools based on Pearson correlation coefficient as described in any one of claims 1, 2, 4-7, characterized in that: The specific process of model validation in step S4 is as follows: S41: For each sample in the validation set, substitute its key process parameters into the mathematical expression of the multiple linear regression model constructed in S3 to calculate the corresponding predictive performance index value. ; S42: Calculate the prediction accuracy evaluation metrics for the validation set, including Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and Nash Efficiency Coefficient (NSE). The formulas for these metrics are as follows: ； ； ； ； In the formula: The number of samples in the validation set; To verify the first set Predictive performance index values ​​for the group of samples; To verify the first set The measured performance index values ​​of the sample group; To verify the mean of the measured values ​​of the target performance index of the set; Among them, MAE reflects the absolute error level between the predicted value and the measured value, RMSE is more sensitive to large error samples, MAPE reflects the relative error level, and NSE evaluates the model's ability to fit the trend of the measured data. The value range is (-∞, 1], and the closer it is to 1, the better the model's prediction performance. S43: Determine whether the prediction accuracy evaluation index meets the preset accuracy conditions; the preset accuracy conditions are: Mean Absolute Percentage Error (MAPE) ≤ 15%, or the ratio of MAE to the mean of the measured values ​​in the validation set ≤ 0.15, or the ratio of RMSE to the mean of the measured values ​​in the validation set ≤ 0.2, or the Nash efficiency coefficient (NSE) ≥ 0.75; if any of the above conditions are met and the goodness-of-fit index meets the requirements, the model is deemed to have passed the validation. If the model validation does not meet the preset conditions, the following adaptive iterative optimization process is executed: S431: Verify the current correlation coefficient threshold If the condition is not met, the iteration will terminate and a model optimization failure message will be output. If the condition is met, proceed to the next step. S432: Update the correlation coefficient threshold according to the adaptive step size. , in: This is the current iteration step size; S433: Based on the updated correlation coefficient threshold, re-execute S2 to screen key process parameters and obtain an updated set of key process parameters to ensure that the selected parameters meet both the threshold requirement and the significance test requirement. S434: Based on the updated set of key process parameters, re-execute S3 to establish and optimize the multiple linear regression model, and complete the regression coefficient solution, significance test and multicollinearity test; S435: Re-execute S41 to S43 to validate the updated model; repeat S431 to S435 until the model passes validation, or the number of iterations reaches the preset maximum number of iterations, or the correlation coefficient threshold drops to the lower limit.

9. A performance prediction method for cold rolling finishing shearing tools based on Pearson correlation coefficient as described in any one of claims 1, 2, 4-7, characterized in that: The process of applying the tool performance prediction model for online prediction in step S5 is as follows: S51: Real-time values ​​of the N key process parameters under current production conditions are collected via pressure sensors, displacement sensors, encoders, hardness testers, and PLC control systems installed on the cold rolling finishing shearing production line. ; S52: Preprocess the acquired real-time process parameters, including based on... Outlier removal and min-max normalization are used to eliminate the impact of data noise and dimensional differences on the prediction results. S53: Substitute the pre-processed real-time values ​​of the N key process parameters into the mathematical expression of the multiple linear regression model established in S35: ； The predicted values ​​of the tool performance indicators at the current moment are calculated. ; S54: Based on the predicted value The output will display a graded tool status message within the specified value range. When the target performance index is the amount of tool wear or the change in the radius of the cutting edge: if the predicted value is less than 60% of the preset threshold, the output tool status is normal. If the predicted value is between 60% and 80% of the preset threshold, the output tool will show slight wear; please monitor it carefully. If the predicted value is between 80% and 100% of the preset threshold, the output tool is severely worn and ready for replacement; if the predicted value reaches or exceeds the preset threshold, the output tool has reached its wear limit and must be replaced immediately. When the target performance index is the remaining tool life: if the predicted value is greater than 200% of the preset threshold, the tool status is output as normal. If the predicted value is between 100% and 200% of the preset threshold, the remaining tool life is sufficient for normal use. If the predicted value is between 50% and 100% of the preset threshold, the remaining tool life is insufficient, and a replacement plan is formulated. If the predicted value is lower than the preset threshold, an emergency warning message will be output indicating that the tool has reached the end of its service life and should be replaced immediately. S55: The real-time collected process parameters, predicted performance index values, and subsequent measured tool performance index values ​​are paired and stored to form new samples, providing data support for dynamic model updates.

10. A performance prediction method for cold rolling finishing shearing tools based on Pearson correlation coefficient as described in any one of claims 1, 2, 4-7, characterized in that: The model dynamic update mechanism in step S5 is as follows: S56: Set two model update trigger conditions; an update will be triggered if either condition is met: Periodic update trigger: After each tool change, all newly added paired samples accumulated during the tool usage cycle are added to the original sample set to form an expanded sample set; when the number of new samples in the expanded sample set reaches 20% to 50% of the number of samples in the original sample set, a full update is triggered. Emergency update trigger: When the relative prediction error of 10 consecutive production samples exceeds 20%, the model is determined to have a working condition drift, and an incremental update is immediately triggered. S57: Execute the corresponding update method for different triggering conditions: Full update: Based on the expanded sample set, S1 to S4 are re-executed to complete the re-partitioning of the dataset, re-screening of key process parameters, model reconstruction and verification, and obtain the updated tool performance prediction model; Incremental Update: Based on the new samples, the regression coefficients of the existing model are incrementally corrected using the weighted least squares method. The weight coefficients are positively correlated with the collection time of the new samples; the newer the samples, the higher the weight. The calculation formula is as follows: ； In the formula: This represents the regression coefficient vector of the current model; This is the vector of regression coefficients after incremental updates; , The data matrix and target vector for the newly added samples; This is the data matrix of the original training set; W is a diagonal weight matrix, where the diagonal elements are the weight values ​​of the corresponding newly added samples; S58: Perform backtesting on the updated model, using historical validation set samples to test the prediction accuracy of the updated model, ensuring that the accuracy is not lower than that of the original model; S59: Replace the currently running online model with the updated tool performance prediction model that has been verified through backtracking, complete the rolling update of the model, and archive the updated model parameters and verification results.

Images