Steel plate thickness measurement data optimization acquisition method

Abstract

The invention discloses a steel plate thickness measurement data optimization acquisition method, which comprises the following steps: acquiring temperature data, and constructing a temperature distribution sequence and a temperature fitting curve; constructing a single-peak data set based on the temperature fitting curve, splitting the single-peak data set into a local single-peak data set and an adjacent single-peak data set, and calculating a neighborhood difference index; according to a neighborhood difference index, a density representative ratio vector is corrected, a temperature compensation coefficient is adjusted according to the corrected density representative ratio vector, then steel plate thickness measurement data is optimized, the stability under a standard working condition is guaranteed through a layered compensation mechanism and a static reference, real-time noise interference is avoided, and the accuracy of thickness measurement is improved. Dynamic correction can quickly respond to temperature abrupt change and local abnormity, and the adaptability to complex scenes is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of thickness measurement, in particular to a steel plate thickness data optimization collection method. BACKGROUND

[0002] In the field of steel manufacturing, steel plate thickness measurement is a key link to ensure product quality and production process stability. Accurate steel plate thickness data is of great significance for subsequent processing, use and quality control. However, in the actual production process, due to the influence of various factors, it is not easy to accurately obtain the thickness of the steel plate. Temperature is an important factor affecting the measurement of the thickness of the steel plate. Due to the process links such as rolling and cooling, the surface temperature distribution of the steel plate is often uneven. Changes in temperature will cause changes in the physical properties of the steel plate material, thereby affecting the accuracy of the thickness measurement.

[0003] For example, Chinese patent application No. 202311368950.5 discloses a steel plate thickness data optimization collection method. Temperature data of each measurement position of the steel plate to be measured is collected to form a temperature compensation matrix, a temperature distribution sequence of each element is constructed, a temperature fitting curve is fitted, and a unimodal feature sequence of each element unimodal data set is constructed in combination with the data distribution on the temperature fitting curve of each element. The unimodal variation degree and the density consistency index of each unimodal data set of the element are calculated. The density deviation rate of each element is obtained in combination with the LOF value of the density consistency index of each element. Then the density representative ratio vector of each element is obtained. The temperature compensation coefficient of each element is obtained according to the density representative ratio vector of each element, and the optimized steel plate thickness of each measurement position of the steel plate to be measured is obtained based on this. The optimization of the steel plate thickness data is completed by collecting the steel plate thickness data of the steel plate to be measured. The steel plate thickness data has high precision.

[0004] In the existing patent technology, the steel plate thickness measurement method does not fully consider the influence of temperature distribution on the measurement result, and only performs data processing based on a single measurement point or a simple temperature compensation method. This cannot accurately reflect the real temperature condition of the steel plate surface, resulting in a large error in the thickness data. The traditional thickness compensation method often only focuses on the data of a single production line, lacks collaborative optimization between multiple production lines, and cannot realize the consistency and stability of global production. SUMMARY

[0005] The present application provides a steel plate thickness data optimization collection method, which uses a hierarchical compensation mechanism to ensure stability under standard working conditions through static reference, avoids real-time noise interference, dynamically corrects temperature mutations and local abnormalities, and improves adaptability in complex scenarios.

[0006] The present application provides a steel plate thickness data optimization collection method, which includes: S11, collect temperature data, construct a temperature distribution sequence and a temperature fitting curve; construct a unimodal data set based on the temperature fitting curve, split the unimodal data set into a local unimodal data set and an adjacent unimodal data set, and calculate a neighborhood difference index; S12, correct the density representative ratio vector according to the neighborhood difference index, adjust the temperature compensation coefficient according to the corrected density representative ratio vector, and further optimize the steel plate thickness data.

[0007] Preferably, the data of the window where the unimodal data set is currently located is extracted as the local unimodal data set, the mode and median features of the local unimodal data set are extracted, the current window where the local unimodal data set is located is taken as the center, the adjacent windows around the current window are taken as the adjacent windows, the unimodal data sets corresponding to the adjacent windows are extracted as the adjacent unimodal data sets, and the mode and median of the adjacent unimodal data sets are extracted.

[0008] Preferably, the density representative ratio vector is corrected according to the neighborhood difference index, and the correction formula is: , wherein, is the corrected density representative ratio vector; is the original density representative ratio vector; e is the base number of natural logarithm; is a spatial continuity adjustment factor; is the neighborhood difference index.

[0009] Preferably, the formula for adjusting the temperature compensation coefficient according to the corrected density representative ratio vector is: , wherein, is the optimized compensation coefficient; is the material thermal expansion coefficient; is the nominal thickness of the steel plate; is the local temperature value of the current window; is a pre-set reference temperature value; is the corrected density representative ratio vector; e is the base number of natural logarithm.

[0010] Preferably, S201, split the adjusted compensation coefficient obtained in step S104 into a basic compensation value and a dynamic adjustment value; S202, calculate a comprehensive temperature compensation coefficient according to the basic compensation value and the dynamic adjustment value.

[0011] Preferably, the formula for calculating the comprehensive temperature compensation coefficient according to the basic compensation value and the dynamic adjustment value is: , wherein, is the comprehensive temperature compensation coefficient, is a dynamic weight factor, ; and ; is a dynamic adjustment value.

[0012] Preferably, the adjustment of the temperature compensation coefficient is an adjustment of single-point local compensation, and in addition to the compensation of single points, a multi-line collaborative compensation network is constructed.

[0013] Preferably, S301, collecting a plurality of line data, and constructing a multi-line collaborative compensation network; S302, according to the plurality of line data, collaboratively compensating the plurality of lines to obtain a global standard density, and calculating a cross-line dynamic adjustment value according to the global standard density; S303, calculating a collaborative thickness correction value according to the cross-line dynamic adjustment value and a multi-factor collaborative weight.

[0014] Preferably, the cross-line dynamic adjustment value is calculated according to the global standard density, and the formula is: , wherein, is a cross-line dynamic adjustment value; is a density-thickness conversion factor; is the density of the i-th line currently measured; is a global standard density; is an attenuation factor, is a parameter related to the material properties; is the difference between the average temperature of the i-th line and the adjacent lines; is the maximum value of the temperature difference between all lines; is the difference between the average rolling speed of the i-th line and the adjacent lines; is the maximum value of the rolling speed difference between all lines; is the difference between the average cooling water flow of the i-th line and the adjacent lines; is the maximum value of the cooling water flow difference between all lines; is the sensitivity coefficient of rolling speed; is the sensitivity coefficient of cooling water flow.

[0015] Preferably, the formula for calculating the collaborative thickness correction value according to the cross-line dynamic adjustment value and the multi-factor collaborative weight is: , wherein, is a collaborative thickness correction value; is an original thickness value; is an adjustment coefficient; is a multi-factor collaborative weight; is a cross-line dynamic adjustment value; is a basic compensation value; is a collaborative compensation scaling factor; a coordination weight of the production line i and a neighborhood production line j; a synchronization compensation amount of the neighborhood production line j.

[0016] One or more technical solutions provided in the present application have at least the following technical effects or advantages: after introducing neighborhood differences, if the current window and the adjacent window distribution difference is significant, it is determined as a boundary area, by reducing the modified density representative ratio vector, avoiding overcompensation of non-continuous areas; local windows may have abnormal unimodal characteristics due to sensor noise or transient interference, neighborhood data provides spatial consistency verification, if the current window characteristics and the adjacent window difference are too large, it is determined as noise interference, by reducing the modified density representative ratio vector to reduce its contribution to the compensation coefficient; in the natural temperature transition zone, the neighborhood difference is small, the modified density representative ratio vector is approximately equal to the original density representative ratio vector, and the original compensation strength is maintained; in the mutation zone, the compensation is automatically inhibited, by fusing the spatial continuity characteristics of local and adjacent unimodal data sets, the boundary overcompensation and noise sensitivity problems are effectively solved, and the thickness measurement accuracy is significantly improved; Through the hierarchical compensation mechanism, the static reference guarantees the stability under standard working conditions, avoids real-time noise interference, and dynamic correction can quickly respond to temperature mutations and local abnormalities, improve the adaptability of complex scenes, and multi-factor weight distribution enables ΔTx and LOFx to quantify temperature differences and abnormality levels, dynamically adjust compensation weights, avoid single model limitations, and solve the precision deficiency problem of embodiment one in extreme temperature mutation and sensor abnormality scenarios by introducing a basic compensation value-dynamic adjustment value hierarchical model, realizing a thickness measurement error reduction of more than 50% under all working conditions, and significantly improving the quality control capability of hot rolling production lines; By constructing a cross-multi-production line coordination compensation network, data sharing and strategy linkage among multiple production lines are realized, the influence of single production line error on the global quality is reduced, and the consistency and stability of the overall production are improved, by fusing multi-production line data to generate a globally consistent thickness correction value, the thickness difference of products produced by each production line can be significantly reduced, and the consistency of product thickness is improved, combined with rolling speed, cooling water flow, material parameters and other multi-factors, real-time linkage optimization of compensation strategy is realized, through the LOF index and production line grouping strategy, abnormal production lines are isolated to avoid the spread of false data, and single machine optimization realizes local precision improvement with low risk and low cost; multi-production line coordination realizes global optimization through data-driven, builds industry competition barriers, covers the whole industry chain demand from low-end to high-end, from local to global, and promotes the hierarchical upgrading of the steel industry; The active disturbance identification non-linear interval is identified by actively injecting errors, the compensation consistency of the model under positive and negative extreme errors is improved by confrontation training, the cross-validation error is reduced to ±0.05mm, the range utilization rate is improved from 70% to 95%, the over-sampling of the sensor in the non-critical interval is avoided, the error is actively amplified, the compensation model is reversely optimized, and the error is converted into a 'probe' of system optimization by means of multi-line confrontation training and dynamic stability control, the precision is improved, the anti-interference is strengthened, and the global consistency is broken through by temperature-error double-drive compensation and active disturbance confrontation training, the environmental adaptation and error control are deeply coupled, a technical paradigm upgrade from'single-point optimization' to'system immunity' is provided for the steel industry, and the technical paradigm upgrade is especially suitable for the fields of aerospace, new energy and other ultra-precision manufacturing. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 A flowchart of a steel plate thickness measurement data optimization method according to the present application is shown in the figure. Figure 2 A flowchart of further correction of the adjusted compensation coefficient by static and dynamic methods according to the present application is shown in the figure. Figure 3 A flowchart of constructing a multi-line collaborative compensation network according to an embodiment of the present application is shown in the figure. Figure 4 A flowchart of actively injecting errors and reversely optimizing a compensation model according to an embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0018] In order to facilitate the understanding of the present application, the present application will be described more fully below with reference to the accompanying drawings; the preferred embodiments of the present application are shown in the drawings, but the present application can be implemented in many different forms and is not limited to the embodiments described herein; on the contrary, the purpose of providing these embodiments is to make the disclosure of the present application more thorough and comprehensive.

[0019] It should be noted that the terms "vertical", "horizontal", "up", "down", "left", "right" and similar expressions used herein are only for the purpose of illustration and do not represent the only implementation.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs; the terms used herein in the specification of the present application are only for the purpose of describing the specific embodiments and are not intended to limit the present application; the term "and / or" used herein includes any and all combinations of one or more related listed items.

[0021] Embodiment one: Figure 1is a flowchart of a steel plate thickness data optimized acquisition method according to an embodiment of the present application, comprising: S101, collecting temperature data, constructing a temperature distribution sequence and a temperature fitting curve; Specifically, the surface of the steel plate to be measured is evenly divided into a plurality of grid blocks. The size of each grid block is determined according to the actual size of the steel plate and the measurement accuracy requirement. The center point of each grid block is determined as a temperature measurement position to ensure that these points can uniformly cover the entire surface of the steel plate. A temperature sensor is used to sequentially measure the temperature at each measurement position, and the temperature value of each measurement position is recorded. The collected temperature data is checked to remove any missing values or obviously abnormal values. The cleaned temperature data is arranged in the order of the measurement positions to form a temperature compensation matrix.

[0022] For each element (i.e. each measurement position) in the temperature compensation matrix, a 3x3 local window is set around it, all temperature values in the window are extracted, and these values are sorted by size to form a temperature distribution sequence. The number of occurrences of each temperature value in the sorted temperature sequence is counted. Using the temperature value as the independent variable and the frequency as the dependent variable, a fitting method polynomial fitting is used to fit the temperature fitting curve.

[0023] S102, constructing a unimodal data set according to the temperature fitting curve of the measurement position, splitting the unimodal data set into a local unimodal data set and an adjacent unimodal data set, and calculating a neighborhood difference index; Further, for the temperature fitting curve of each measurement position, the slope change of the curve is identified by using mathematical analysis method, and the local trough in the temperature fitting curve, i.e. the minimum value point, is identified as the key breakpoint of the segmentation curve. The minimum value point appears at the position where the slope of the curve changes from negative to positive, i.e. the point where the second derivative is greater than zero. The identified minimum value point is used as the breakpoint of the segmentation curve. These breakpoints divide the temperature fitting curve into a plurality of continuous intervals. For each breakpoint interval, all temperature values in the interval are extracted to form a sub-data set. This sub-data set is a unimodal data set, which represents a local peak in the temperature distribution. The data of the current window of the unimodal data set is extracted as a local unimodal data set. The mode and median features of the local unimodal data set are extracted. The current window is centered, and the upper, lower, left, right and diagonal of the current window are adjacent windows. For example, in a 3x3 grid division, each window usually has 8 adjacent windows. For each unimodal data set, the unimodal data sets corresponding to all adjacent windows are extracted as adjacent unimodal data sets. The mode and median of the adjacent unimodal data sets are extracted. The neighborhood difference is calculated according to the mode and median of the local unimodal data set and the adjacent unimodal data set. The formula is: + , wherein, is the neighborhood difference indicator, representing the temperature distribution difference degree between the current window and the adjacent window; is the mode of the local unimodal data set of the current window, i.e., the temperature value with the highest frequency in the window; is the mode of the unimodal data set of a window adjacent to the current window; is the median of the local unimodal data set of the current window, i.e., the temperature value located in the middle position after sorting all temperature values in the window by size; is the median of the unimodal data set of a window adjacent to the current window; and is the weight coefficient, used to adjust the relative importance of the mode difference and the median difference in the neighborhood difference indicator, according to experiments, a is set to 0.6 and β is set to 0.4, indicating that the weight of the mode difference in the neighborhood difference is higher than that of the median difference, and the neighborhood difference indicator is calculated.

[0024] S103, according to the neighborhood difference indicator, the density representative ratio vector is corrected; Specifically, a spatial continuity adjustment factor is introduced to control the influence of spatial continuity on the correction degree of the density representative ratio vector, the spatial continuity adjustment factor is optimized by using a data set with a known thickness and temperature distribution, and the optimal spatial continuity adjustment factor is found, the density representative ratio vector is corrected according to the calculated neighborhood difference indicator and the spatial continuity adjustment factor, and the correction formula is: , wherein, is the corrected density representative ratio vector, which can more accurately reflect the spatial continuity of the temperature distribution, thereby suppressing the compensation amplitude of the boundary mutation region; is the original density representative ratio vector, which serves as the basis for correction, provides the basic density or temperature feature information of the current window; e is the base of natural logarithm, approximately equal to 2.71828, e is the base of the exponential function, used to achieve the effect of exponential decay; is the spatial continuity adjustment factor, which determines the influence of spatial continuity on the correction degree, a larger λ value will result in a stronger correction effect, i.e., more significantly suppressing the compensation amplitude of the boundary mutation region, while a smaller λ value will result in a weaker correction effect; is the neighborhood difference indicator, the size of directly determines the value of the exponential decay factor, when is large, indicating that the temperature distribution is discontinuous in the region, there is a boundary mutation, at this time the exponential decay factor will be small, thereby reducing the value of the corrected density representative ratio vector; on the contrary, when is small, indicating that the temperature distribution is relatively continuous in the region, at this time the exponential decay factor will be large, and the influence on the corrected density representative ratio vector will be small.

[0025] S104, adjusting the temperature compensation coefficient according to the corrected density representative ratio vector, and further optimizing the steel plate thickness measurement data; Further, according to the corrected density representative ratio vector, an optimized compensation coefficient is generated, and the formula is: , wherein, is the optimized compensation coefficient; is the material thermal expansion coefficient; is the nominal thickness of the steel plate; is the local temperature value of the current window; is the preset reference temperature value; is the corrected density representative ratio vector; e is the base of natural logarithm, and the optimized compensation coefficient is composed of two parts: local temperature error and exponential decay term (e ), the local temperature error reflects the difference between the current window and the reference temperature, which is the basis for temperature compensation, and the exponential decay term adjusts the strength of compensation according to the value of the corrected density representative ratio vector When is large (that is, the neighborhood difference is large, and the temperature distribution is discontinuous), the exponential decay term will be small, thereby reducing the compensation coefficient and avoiding excessive compensation for the boundary mutation area.

[0026] Based on the adjusted compensation coefficient, the steel plate thickness measurement data is optimized, and the formula is: , wherein, is the optimized thickness value, is the original thickness value; is the compensation strength coefficient, which is determined by the calibration experiment, determines the influence degree of the optimized compensation coefficient on the thickness value, by adjusting the value of , the accuracy of thickness calculation is optimized; represents the optimized compensation coefficient, and the optimized thickness value is composed of the original thickness value and the compensation term, and the compensation term adjusts the original thickness value according to the value of the optimized compensation coefficient and the compensation strength coefficient, so as to obtain more accurate thickness measurement results, and by determining the appropriate γ value through the calibration experiment, the accuracy and reliability of the thickness calculation can be ensured.

[0027] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages: by introducing the neighborhood difference, if the current window and the adjacent window have significant distribution difference (such as Δx>θ), it is determined as a boundary area, and by adjusting the compensation strength coefficient The modified density representative ratio vector is reduced to avoid overcompensation of non-continuous areas; local windows may have abnormal unimodal characteristics due to sensor noise or transient interference, and neighborhood data provides spatial consistency verification. If the current window characteristics differ too much from the adjacent windows (such as Δx abnormal), it is determined to be noise interference, and the contribution of the modified density representative ratio vector to the compensation coefficient is reduced by reducing the modified density representative ratio vector; in the natural temperature transition zone (such as the gradual temperature zone in the rolling direction), the neighborhood difference is small, and the modified density representative ratio vector is approximately equal to the original density representative ratio vector, maintaining the original compensation strength; in the mutation zone, the compensation is automatically inhibited, and by fusing the spatial continuity characteristics of the local and adjacent unimodal data sets, the problems of boundary overcompensation and noise sensitivity are effectively solved, and the thickness measurement accuracy is significantly improved.

[0028] Embodiment two: based on the scheme in embodiment one, this embodiment introduces a basic compensation value and a dynamic adjustment value, and further calculates and corrects the adjusted compensation coefficient in embodiment one in a static and dynamic manner, as shown in Figure 2 .

[0029] S201, the adjusted compensation coefficient obtained in step S104 is split into a basic compensation value and a dynamic adjustment value; Further, by collecting historical data under standard working conditions, including temperature, rolling speed, thickness measurement value and corresponding compensation coefficient, the collected historical data is used to train a linear regression model to predict the compensation coefficient under standard working conditions. For a given rolling condition (under standard working conditions), the trained linear regression model is used to calculate the basic compensation value, and the formula is: , wherein, is the density of the steel plate to be measured under standard temperature; is the normal density of the standard steel plate; is a decay factor, the basic compensation value reflects the amount of compensation needed to correct the effect of temperature on thickness measurement under ideal conditions; during rolling, real-time data such as temperature, rolling speed, etc. are collected, the temperature change rate is calculated according to the collected temperature data, which reflects the speed of temperature change over time, the collected temperature, rolling speed, and time stamp data are recorded and stored in the database, the collected real-time data is cleaned to improve the training effect and accuracy of the model, the features related to the compensation coefficient are extracted from the preprocessed data, and these features are used as the input of the model, according to the characteristics and requirements of the rolling process, a neural network model is selected as the dynamic adjustment model, historical data (including data under standard and non-standard working conditions) are used to train the dynamic adjustment model, the parameters of the model are adjusted to minimize the prediction error, the validation set data is used to verify the trained dynamic adjustment model, and the accuracy and generalization ability of the model are evaluated, according to the verification result, the model is optimized, such as adjusting the model structure, adding or reducing features, the real-time collected temperature change rate and rolling speed feature data are input into the trained dynamic adjustment model, the model is used to predict the input feature data, and the dynamic adjustment value is obtained, the formula for calculating the dynamic adjustment value is: wherein, is a dynamic adjustment value; is a local temperature value of the current window; is a pre-set reference temperature value; is a corrected density representative ratio vector; e is the base of natural logarithm; is a difference between the current measurement point and the neighborhood average temperature, is the temperature of the current measurement point, is the neighborhood average temperature; is a maximum allowed error, the dynamic adjustment value represents the additional compensation amount needed due to the difference between the real-time working condition and the standard working condition.

[0030] S202, calculating a comprehensive temperature compensation coefficient according to the basic compensation value and the dynamic adjustment value; Specifically, the comprehensive temperature compensation coefficient is calculated based on the basic compensation value and the dynamic adjustment value, and the formula is: wherein, is a comprehensive temperature compensation coefficient, is a dynamic weight factor, wherein, is used to allocate the weights of the basic compensation value and the dynamic adjustment value in the calculation of the final compensation coefficient, and the value is between 0 and 1, reflecting the proportion of the dynamic adjustment value in the final compensation; e is the base of natural logarithm; k is a sensitivity coefficient, which is set to 0.05, which determines the sensitivity of the dynamic weight factor to temperature difference and local abnormality degree.​ This represents the temperature difference between the current measurement point and the average temperature of its neighborhood. The LOF value is a density consistency index used to reflect the degree of local anomaly. The larger the LOF value, the more likely the current measurement point is to be an anomaly in the density distribution. ; To dynamically adjust the value, the steel plate thickness measurement data is corrected based on the comprehensive temperature compensation coefficient. The formula is as follows: ,in, This is the corrected thickness value; This is the original thickness value; This is the compensation coefficient scaling factor; This is the comprehensive temperature compensation coefficient.

[0031] The technical solutions in the above embodiments of this application have at least the following technical effects or advantages: Through the hierarchical compensation mechanism, the static benchmark ensures stability under standard working conditions and avoids real-time noise interference; the dynamic correction can quickly respond to temperature changes and local anomalies, improve the adaptability to complex scenarios; the multi-factor weight allocation enables ΔTx and LOFx to quantify temperature differences and anomalies; the compensation weight is dynamically adjusted to avoid the limitations of a single model; by introducing a hierarchical model of basic compensation value and dynamic adjustment value, the problem of insufficient accuracy in extreme temperature changes and sensor anomalies in Embodiment 1 is solved, and the thickness measurement error under all working conditions is reduced by more than 50%, significantly improving the quality control capability of the hot rolling production line.

[0032] Example 3: Based on the single-point local collaborative compensation in Example 1 and the dynamic hierarchical compensation in Example 2, this example constructs a cross-production line collaborative optimization framework. It integrates real-time data (rolling speed v, cooling water flow rate Q, material parameter κ) from multiple rolling production lines with a temperature compensation model. Through a three-level mechanism of data sharing, dynamic collaboration, and global optimization, it achieves dynamic linkage of compensation strategies between production lines, reducing the impact of single-point errors on overall quality and improving the thickness consistency and process stability of multi-production line collaborative production. Figure 3 As shown.

[0033] S301 collects data from multiple production lines and constructs a multi-production-line collaborative compensation network; Further, the construction of the multi-line collaborative compensation network includes building a data interconnection layer and line correlation modeling; for building the data interconnection layer, the MindSphere platform is selected, the thickness data, temperature data, rolling speed, flow data and material parameters on each line are collected through the set device, the device is connected to the MindSphere platform in a wireless manner, the data acquisition frequency is set according to the process requirement, the real-time data collected is stored in a time series database, for line correlation modeling, based on the collected line data, data cleaning and normalization processing are performed, the rolling speed change rate and temperature fluctuation rate are extracted, the rolling speed difference, cooling water flow difference and material parameter difference are calculated, the rolling speed difference between lines i and j is calculated = , is the rolling speed of line i, is the rolling speed of line j, the cooling water flow difference is calculated = , wherein is the cooling water flow of line i, is the cooling water flow of line j, the material parameter difference is calculated = , wherein is the parameter difference of line i, is the parameter difference of line j, the Pearson correlation coefficients between the rolling speed, the cooling water flow and the material parameters are calculated using the Pearson correlation coefficient, the Pearson correlation coefficient is a prior art, which will not be described here, the comprehensive correlation coefficient is calculated using the weighted average method, and the formula is: , wherein is the comprehensive correlation coefficient, , and are the Pearson correlation coefficients between the rolling speed, the cooling water flow and the material parameters, , and are weight coefficients, an n*n matrix C is created, wherein n is the number of lines, the calculated comprehensive correlation coefficients are filled in the corresponding positions of the matrix to construct the correlation matrix.

[0034] According to the process requirement and the actual production situation, a threshold value is set for distinguishing the correlation of lines in the correlation matrix, the lines with a comprehensive correlation coefficient greater than the preset threshold value are divided into a strong correlation group, and the lines with a comprehensive correlation coefficient less than or equal to the preset threshold value are divided into a weak correlation group, a joint compensation model is constructed for the strong correlation group to realize the collaborative adjustment of the rolling speed, the cooling water flow and other parameters, and the lines in the weak correlation group are independently optimized to reduce mutual interference.

[0035] S302, according to the plurality of production line data, the plurality of production lines are compensated, and the global standard density is obtained, and the cross-production line dynamic adjustment value is calculated according to the global standard density; Specifically, in order to avoid the deviation of single production line data from misleading the compensation strategy, a standard density benchmark is introduced, which is derived based on the historical data of multiple production lines and can comprehensively reflect the production characteristics of each production line under standard conditions. According to N production lines, for each production line, collect the density data at standard temperature, add the density of all N production lines participating in the calculation at standard temperature to obtain the total, and then divide the total by the total number of production lines N. The result is the global standard density.

[0036] According to the global standard density and the process parameters such as temperature, rolling speed and cooling water flow rate between the production lines, the cross-production line dynamic adjustment value is calculated, and the formula is: , wherein, is the cross-production line dynamic adjustment value; is the density currently measured by the i-th production line; is the global standard density; is a decay factor, is a parameter related to material properties and production process, etc.; is the difference between the average temperature of the i-th production line and the adjacent production line; is the maximum temperature difference between all production lines; is the difference between the average rolling speed of the i-th production line and the adjacent production line; is the maximum rolling speed difference between all production lines; is the difference between the average cooling water flow rate of the i-th production line and the adjacent production line; is the maximum cooling water flow rate difference between all production lines; is the sensitivity coefficient of rolling speed; is the sensitivity coefficient of cooling water flow rate; The dynamic weight factor in step S202 is expanded, and the expanded dynamic weight factor introduces material parameter difference, forms a multi-factor coordination weight, calculates the difference between the average temperature of the i-th production line and the adjacent production line, the difference between the average material parameter of the i-th production line and the adjacent production line, and the local outlier factor of the i-th production line, introduces the adjustment coefficient and the sensitivity coefficient of the material parameter difference, multiplies the difference between the average temperature of the i-th production line and the adjacent production line by the local outlier factor of the i-th production line to obtain the temperature difference related quantity, multiplies the average material parameter of the i-th production line and the adjacent production line by the sensitivity coefficient of the material parameter difference to obtain the material parameter difference quantity, adds the temperature difference related quantity and the material parameter difference quantity to obtain the comprehensive difference influence value, multiplies the comprehensive difference influence value by the adjustment coefficient, takes the negative exponent of the multiplication result, and adds 1 to the negative exponent of the multiplication result to obtain the multi-factor coordination weight.

[0037] S303, calculate a collaborative thickness correction value according to the cross-line dynamic adjustment value and the multi-factor collaborative weight; Further, in the multi-line production scenario, due to the influence of various factors (such as equipment difference, environmental factors, material characteristics, etc.) on each line during production, the product thickness fluctuates to a certain extent. In order to improve product quality, the production data of multiple lines are fused to generate a globally consistent collaborative thickness correction value. The formula for calculating the collaborative thickness correction value is: +(1- ) , wherein, is the collaborative thickness correction value; is the original thickness value; is the adjustment coefficient; is the multi-factor collaborative weight; is the cross-line dynamic adjustment value; is the basic compensation value; is the collaborative compensation scaling factor; is the collaborative weight of line i and neighboring line j, calculated based on ; is the synchronization compensation amount of neighboring line j.

[0038] The technical solutions in the embodiments of the present application have at least the following technical effects or advantages: by constructing a cross-line collaborative compensation network, data sharing and strategy linkage between multiple lines are realized, the influence of single-line error on global quality is reduced, the consistency and stability of overall production are improved, a globally consistent thickness correction value is generated by fusing multi-line data, which can significantly reduce the difference in product thickness between lines, improve the consistency of product thickness, combine rolling speed, cooling water flow, material parameters, etc. Multi-factor, realize real-time linkage optimization of compensation strategy, isolate abnormal lines through LOF index and line grouping strategy, avoid the spread of false data, and realize local precision improvement with low risk and low cost through single machine optimization; Multi-line collaboration through global optimization driven by data, build industry competition barriers, cover from low-end to high-end, from local to global, meet the demand of the whole industry chain, and promote the hierarchical upgrading of the steel industry.

[0039] Embodiment four: based on the multi-line collaborative optimization of embodiment three, on the basis of embodiment three, the multiple lines are divided into low temperature zone and high temperature zone, the error characteristics are divided into stable group and fluctuation group, the temperature characteristics and error characteristics are combined, the error is actively injected, and the compensation model is optimized in reverse, as shown in Figure 4 .

[0040] S401, temperature zoning and error characteristic zoning are performed on each line; Further, by collecting the ambient temperature data of all production lines at different time points, collecting and analyzing the ambient temperature data in the past period of time, understanding the regularity and range of temperature change, setting the partition threshold according to the historical data analysis result and the production demand, traversing all collected ambient temperature data, dividing the temperature less than the partition threshold into the low temperature zone, and dividing the temperature greater than or equal to the partition threshold into the high temperature zone; collecting the error data of all production lines at different time points, and calculating the error standard deviation of each time point, setting the error characteristic partition threshold, traversing all calculated error standard deviation data, dividing the partition with error standard deviation less than the error characteristic partition threshold into the stable zone, and dividing the partition with error standard deviation greater than or equal to the error characteristic partition threshold into the fluctuation zone.

[0041] S402, combining the temperature partition and the error characteristic partition; Specifically, in order to balance the influence of different ambient temperatures on the production process, the low temperature zone and the high temperature zone are paired, the temperature difference between the two is used for complementation, and the potential risks caused by a single temperature condition are reduced, in order to reduce the influence of error fluctuation on product quality in the production process, the stable zone and the fluctuation zone are cross-paired, the data of different error characteristics are combined, the overall error level tends to be stable, and the reliability and consistency of the production process are improved, and the specific combination is low temperature stable zone and high temperature fluctuation zone, high temperature stable zone and low temperature fluctuation zone, for the low temperature stable zone and the high temperature fluctuation zone, the data belonging to the stable group is selected from the low temperature group, and the data belonging to the fluctuation group is selected from the high temperature group. Pairing, for the high temperature stable zone and the low temperature fluctuation zone, the data belonging to the stable group is selected from the high temperature group, and the data belonging to the fluctuation group is selected from the low temperature group.

[0042] S403, constructing a double-drive compensation model according to the combined temperature partition, error characteristic partition and active injection error, and obtaining a double-drive compensation coefficient based on the double-drive compensation model; Further, a machine learning algorithm is used to generate a model, an error is automatically injected every certain period of time, the error disturbance amplitude of each injection is ±0.15mm, the full range of the production system is ensured, the disturbance time lasts for several minutes each time, the system response data is recorded during the disturbance, the response data includes the compensation amount and the error change, a positive disturbance group and a negative disturbance group are formed by injecting errors on the production line, for the positive disturbance group, +0.2mm error is injected in production line A, at the same time of injecting the error, the model is used to compensate production line A, and the model parameters are adjusted according to the compensation effect, the model is optimized, and the key data in the compensation process are recorded, such as the compensation amount, the error change, etc., for the negative disturbance group, -0.2mm error is injected in production line B, similarly, the model is used to compensate production line B, and the model parameters are adjusted according to the compensation effect, the key data in the compensation process are recorded, the positive disturbance group and the negative disturbance group are weighted and averaged to construct a double-drive compensation model, the positive disturbance group enables the model to learn the compensation strategy for positive error, improves the compensation performance of the model under the condition of positive error, and the negative disturbance group enables the model to learn the compensation strategy for negative error, enhances the adaptability of the model under the condition of negative error, and the compensation is segmented by the double-drive compensation model;

[0043] wherein, is a double-drive compensation coefficient, is a disturbance error; is a basic error, and the thickness data of the steel plate is further optimized according to the obtained compensation coefficient.

[0044] The technical solutions in the embodiments of the present application have at least the following technical effects or advantages: non-linear intervals are identified by actively injecting errors for active disturbance, compensation consistency of the model under positive and negative extreme errors is improved by adversarial training, error reduction is reduced to ±0.05mm by cross-validation, the range utilization rate is improved from 70% to 95%, over-sampling of the sensor in the non-critical interval is avoided, the compensation model is optimized by actively amplifying the error, and the error is converted into a "probe" for system optimization by means of multi-line adversarial training and dynamic stability control, precision leap, anti-interference strengthening and global consistency breakthrough are achieved by temperature-error double-drive compensation and active disturbance adversarial training, environmental adaptation and error control are deeply coupled, a technical paradigm upgrade from "single-point optimization" to "system immunity" is provided for the steel industry, and it is especially suitable for aerospace, new energy and other ultra-precision manufacturing fields.

[0045] The above merely describes the preferred embodiments of the present application and is not used to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. An optimized data acquisition method for steel plate thickness measurement, characterized in that, include: S11. Collect temperature data and construct a temperature distribution sequence and a temperature fitting curve. Based on the temperature fitting curve, construct a single-peak dataset, split the single-peak dataset into a local single-peak dataset and an adjacent single-peak dataset, and calculate the neighborhood difference index. S12, based on the neighborhood difference index, corrects the density representative ratio vector, and adjusts the temperature compensation coefficient based on the corrected density representative ratio vector, thereby optimizing the steel plate thickness measurement data.

2. The method for optimizing the acquisition of steel plate thickness measurement data as described in claim 1, characterized in that, The data in the current window of the unimodal dataset is extracted as the local unimodal dataset. The mode and median features of the local unimodal dataset are extracted. The current window of the local unimodal dataset is taken as the center, and the surrounding windows are the adjacent windows. The unimodal datasets corresponding to the adjacent windows are extracted as the adjacent unimodal datasets. The mode and median of the adjacent unimodal datasets are extracted.

3. The method for optimizing the acquisition of steel plate thickness measurement data as described in claim 2, characterized in that, The density-representing ratio vector is corrected based on the neighborhood difference index using the following formula: ,in, The corrected density represents the ratio vector; The original density represents the ratio vector; e is the base of the natural logarithm; It serves as a spatial continuity adjustment factor. This is an indicator of neighborhood differences.

4. The method for optimizing the acquisition of steel plate thickness measurement data as described in claim 1, characterized in that, The formula for adjusting the temperature compensation coefficient based on the corrected density-representation ratio vector is as follows: ,in, To optimize the compensation coefficient; The coefficient of thermal expansion of the material; This refers to the nominal thickness of the steel plate. This represents the local temperature value of the current window. The preset reference temperature value; is the corrected density-representation ratio vector; e is the base of the natural logarithm.

5. The method for optimizing the acquisition of steel plate thickness measurement data as described in claim 1, characterized in that, S201, the adjusted compensation coefficient obtained in step S104 is split into basic compensation value and dynamic adjustment value; S202, calculate the comprehensive temperature compensation coefficient based on the basic compensation value and the dynamic adjustment value.

6. The method for optimizing the acquisition of steel plate thickness measurement data as described in claim 5, characterized in that, The comprehensive temperature compensation coefficient is calculated based on the basic compensation value and the dynamic adjustment value, using the following formula: ,in, For the comprehensive temperature compensation coefficient, As a dynamic weighting factor, ; ; This is a dynamically adjusted value.

7. The method for optimizing the acquisition of steel plate thickness measurement data as described in claim 1, characterized in that, The temperature compensation coefficient is adjusted to adjust the local compensation at a single point. In addition to compensating at a single point, a multi-production line collaborative compensation network is also constructed.

8. The method for optimizing the acquisition of steel plate thickness measurement data as described in claim 7, characterized in that, S301 collects data from multiple production lines and constructs a multi-production-line collaborative compensation network; S302, based on data from multiple production lines, performs collaborative compensation on multiple production lines to obtain a global standard density, and calculates cross-production line dynamic adjustment values ​​based on the global standard density; S303, calculates the collaborative thickness correction value based on the cross-production line dynamic adjustment value and the multi-factor collaborative weight.

9. The method for optimizing the acquisition of steel plate thickness measurement data as described in claim 8, characterized in that, The dynamic adjustment value across production lines is calculated based on the global standard density, using the following formula: ,in, This is a dynamic adjustment value across production lines; Density-thickness conversion factor; The density currently measured for the i-th production line; The global standard density; As a decay factor, These are parameters related to material properties; The difference in average temperature between the i-th production line and its neighboring production lines; This represents the maximum temperature difference between all production lines. The difference between the average rolling speed of the i-th production line and the neighboring production lines; This represents the maximum difference in rolling speed among all production lines; The difference in average cooling water flow rate between the i-th production line and its neighboring production lines; This represents the maximum difference in cooling water flow rate among all production lines. The sensitivity coefficient is the rolling speed. This is the sensitivity coefficient for cooling water flow rate.

10. The method for optimizing the acquisition of steel plate thickness measurement data as described in claim 8, characterized in that, The formula for calculating the synergy thickness correction value based on the cross-production line dynamic adjustment value and the multi-factor synergy weight is as follows: +(1- ) ,in, This is a collaborative thickness correction value; This is the original thickness value; This is the adjustment coefficient; Multi-factor collaborative weighting; This is a dynamic adjustment value across production lines; This is the basic compensation value; For collaborative compensation scaling factor; Let be the collaborative weight between production line i and its neighboring production line j; This represents the synchronization compensation amount for the neighboring production line j.

Images