Joint monitoring and diagnosis method of stand in hot strip finishing rolling process based on PLS algorithm

Abstract

The present invention discloses a method for joint monitoring and diagnosis of sub-racks in the hot finishing rolling process of plates and strips based on a PLS algorithm, which relates to the technical field of quality control of the hot-rolled plate and strip finishing rolling process, and includes the following steps: according to the hot-rolled plate and strip finishing production stage, the process variables related to the thickness quality of each finishing rolling rack rolling process are selected, and the production data sub-block samples of each rack are constructed, and the original data are standardized; the cumulative contribution rate of each rack sub-block is obtained by using a PRESS test, and the number of principal elements is determined; the data sub-block model of each rack is constructed by using a PLS algorithm, and the sub-block statistics are calculated to obtain the monitoring data status information of each rack production process; the sub-block variable contribution diagram of the sub-rack data is constructed, and the quality-related variable abnormality reasoning is performed to obtain the abnormal root variable. The present invention improves the deficiency of root diagnosis errors caused by quality inheritance in the rolling process of multiple racks, and achieves the goal of accurate tracing of quality abnormalities in the rolling process of multiple racks and long-term stable production of the rolling process.

Description

Technical Field

[0001] The present invention relates to the technical field of quality control of hot-rolled plate and strip finishing rolling process, and in particular to a method for joint monitoring and diagnosis of plate and strip hot-rolling process sub-stands based on a PLS algorithm, taking into account the quality genetic characteristics of a hot-rolling multi-stand rolling process. Background Art

[0002] Monitoring and diagnosis of quality-related fault variables in the multi-stand hot finishing rolling process is an important part of its intelligent control and steady-state operation. By monitoring the fluctuation of process variables and quality abnormality signals, on-site workers can timely grasp the equipment operation status information and quality control status. Monitoring and diagnosis can improve the misjudgment caused by manual inspection, effectively improve the utilization rate of production data on industrial sites, and mine the deep-level characteristics of data.

[0003] The monitoring and diagnosis of quality-related variables is an important guarantee for the production process of hot-rolled plates and strips. Fast diagnosis speed and high recognition accuracy are two important indicators of the diagnosis process. In the traditional methods of monitoring and diagnosis of quality-related variables in the plate and strip production process, methods to solve the strong coupling and nonlinearity between data variables and related methods for multivariate quality diagnosis have been widely used. However, due to the complexity and changeability of the process in the finishing rolling stage and the coordination of each rack, most of the current monitoring and diagnosis methods focus on a certain variable, ignoring the component modules of a single process, so that the genetic characteristics of the quality of each processing module in a single process are also ignored, resulting in the misjudgment of the root cause variables affecting quality abnormalities in the diagnosis process. Due to the inability to control quality from the root, on-site production often fails to achieve long-term stable production. Summary of the invention

[0004] In view of this, the present invention provides a joint monitoring and diagnosis method for sub-racks in the hot finishing rolling process of plates and strips based on the PLS algorithm, so as to realize the root cause diagnosis of quality-related faults in the existing process industry and reduce the misjudgment of the root cause variables affecting quality abnormalities in the diagnosis process.

[0005] To this end, the present invention provides the following technical solutions:

[0006] The present invention provides a method for joint monitoring and diagnosis of sub-stands in a plate and strip hot finishing rolling process based on a PLS algorithm, comprising the following steps:

[0007] According to the hot-rolled strip finishing production process, the production data related to thickness quality is collected, the collected data is standardized, and the data is divided into sub-blocks according to each sub-rack;

[0008] Determine the cumulative contribution rate of each rack variable according to the forecast error square sum cross-validation model, and obtain the number of principal elements of each data sub-block monitoring variable;

[0009] Based on the number of principal components of the monitoring variables of each data sub-block, the partial least squares method (PLS) is used to establish the data sub-block model of each rack, and the principal component vector of the variable is extracted;

[0010] Calculating monitoring statistics of each data sub-block according to the data sub-block model, and determining monitoring process control limits of each rack;

[0011] According to the monitoring statistics of each data sub-block and the monitoring process control limit of each rack, the stability of the process variable monitoring of each rack is obtained, and the status information of whether the production process is stable is obtained;

[0012] According to the projection size of each rack data submodule on the principal component of the variable, the contribution value of each rack process variable is calculated to obtain the contribution diagram of each variable to the quality-related faults. The cause of the abnormal quality state in the rack is determined by combining the contribution diagram and the state information of whether the production process is stable;

[0013] Based on the Granger causality idea, the causal topological structure of the continuous rolling process is constructed, and the genetic topological structure of each rack sub-monitoring module is established. According to the genetic characteristics of quality among racks and the maximum contributing variable of the contribution graph of each rack, the root cause of the fault is inferred.

[0014] Furthermore, the cumulative contribution rate of each rack variable is determined according to the forecast error square sum cross-validation model, and the number of principal components of the monitoring variables of each data sub-block is obtained, including:

[0015] Given the number of pivot elements in each rack, A i , and then calculate the prediction error after modeling and testing; divide the sample data into two groups, n2 samples are used for regression modeling, and n1 samples are used for testing;

[0016] According to the sampling test, the total prediction error is:

[0017]

[0018] Where y i is the i-th sample point among n1 test sample points; The predicted value of the i-th sample point of the test sample;

[0019] Calculate the prediction error S of n1 sample points for A-1 components A-1 , then the cross validity of the Ath principal component of the dependent variable is:

[0020]

[0021] when When , the added Ath principal component will improve the model accuracy, otherwise the added component will be discarded.

[0022] Furthermore, based on the number of principal components of the monitoring variables of each data sub-block, a partial least squares method (PLS) is used to establish a data sub-block model for each rack, including:

[0023] The process variables of each hot finishing rolling stand are used as independent variables as the input matrix X of the model, and the thickness quality of the finishing rolling outlet is used as the dependent variable as the model output matrix Y to obtain the external model:

[0024]

[0025] Among them, T is the process variable score matrix, P is the process variable load vector; U is the quality variable score matrix; Q is the quality variable load vector; E is the process variable residual matrix; F is the quality variable residual matrix; A is the number of principal components of the PLS model; t j ,u j The principal components extracted from process variables and quality variables respectively.

[0026] Further, calculating monitoring statistics of each data sub-block according to the data sub-block model includes:

[0027] Calculate the score vector t of the variable data in the i-th rack production process new,i and residual

[0028] t new,i =R T x new,i ;

[0029]

[0030] Among them, x new,i is a new sample of the variable data related to the thickness quality of the plate strip to be monitored, R is the projection direction matrix, R = W (P T W) -1 ;P T R=R T P=W T W=I A ; R T is the transpose of the projection direction matrix, I is the unit matrix; I A is the unit matrix corresponding to the number of principal elements; W is the weight matrix of the input matrix;

[0031] Calculate T based on the score vector and residual 2 And the Q statistic:

[0032]

[0033]

[0034] Where ∧ is the score matrix covariance in the training model, is the transpose of the score vector, is the transpose of the new sample.

[0035] Further, the calculating of the control limit of each rack includes:

[0036] Statistics T 2 and Q follow F distribution and standard normal distribution respectively, calculate the control limits of the two variables and

[0037]

[0038] in, is the statistic T 2 The control limits, is the control limit of the statistic Q, n is the number of sample points, and α is the control limit confidence level; S is the variance of the Q statistic, and μ is the mean of the Q statistic; F A,n-A；α and The distribution of data.

[0039] Furthermore, according to the projection size of each rack data submodule on the principal component of the variable, the contribution value of each rack process variable is calculated to obtain the contribution diagram of each variable to the quality-related faults, including:

[0040] Calculate the contribution of the j-th variable of the out-of-limit sample point i to the squared prediction error;

[0041]

[0042] In the formula, is the total contribution value; x ij is the detection value of the jth variable in the i-th sample point; is the predicted value of the jth variable in the i-th sample point;

[0043] Adding expectation factor:

[0044]

[0045] Based on the contribution value of each variable of the out-of-limit sample point to the squared prediction error after adding the expected factor, a contribution graph is formed.

[0046] The present invention has the following beneficial effects:

[0047] The present invention provides a method for joint monitoring and diagnosis of the sub-frames of the plate and strip finishing rolling process based on the PLS algorithm, including: obtaining industrial data of the plate and strip rolling process, dividing the data sub-blocks of each frame of the rolling process according to the process system of each frame of the finishing rolling process, modeling and analyzing each data sub-block according to the PRESS cross-check and PLS algorithm, constructing model statistics, and obtaining the monitoring statistical information and monitoring results of the quality-related variables of each sub-block. According to the contribution graph method, the contribution degree of each sub-block variable to the quality anomaly is analyzed to obtain the main variables causing the quality anomaly. According to the genetic characteristics of the quality variables among multiple frames, the diagnosis results of each sub-block are combined to infer the main reasons for the abnormality of each frame variable, obtain the root abnormal variables and the passive abnormal variables caused by the genetic characteristics, and realize the root diagnosis of the quality-related faults of the multi-frame process production process according to the reasoning of each sub-block. The technical solution provided by the present invention improves the deficiency of the root diagnosis error caused by the quality inheritance of the multi-frame rolling process, and realizes the goal of accurate tracing of the quality anomaly of the multi-frame rolling process and long-term stable production of the rolling process. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0049] Figure 1 Schematic diagram of a single rack quality monitoring and diagnosis framework based on PLS in an embodiment of the present invention;

[0050] Figure 2 It is a schematic diagram of the PLS algorithm in an embodiment of the present invention;

[0051] Figure 3 It is a structural diagram of a hot strip finishing mill in an embodiment of the present invention;

[0052] Figure 4 A statistical diagram of longitudinal thickness deviation of hot-rolled strip in an embodiment of the present invention;

[0053] Figure 5 It is a statistical diagram of F1 rack T2 statistics in an embodiment of the present invention;

[0054] Figure 6 This is a statistical diagram of SPE statistics for the F1 rack in an embodiment of the present invention;

[0055] Figure 7 This is a statistical diagram of T2 statistics of the F2 rack in an embodiment of the present invention;

[0056] Figure 8This is a statistical diagram of SPE statistics for the F2 rack in an embodiment of the present invention;

[0057] Fig. 9 This is a statistical diagram of T2 statistics of the F3 rack in an embodiment of the present invention;

[0058] Fig.10 This is a statistical diagram of SPE statistics for the F3 rack in an embodiment of the present invention;

[0059] Fig.11 This is a statistical diagram of T2 statistics of the F4 rack in an embodiment of the present invention;

[0060] Fig.12 This is a statistical diagram of SPE statistics for the F4 rack in an embodiment of the present invention;

[0061] Fig.13 This is a statistical diagram of T2 statistics of the F5 rack in an embodiment of the present invention;

[0062] Fig.14 This is a statistical diagram of SPE statistics of the F5 rack in an embodiment of the present invention;

[0063] Fig.15 This is a statistical diagram of T2 statistics of the F6 rack in an embodiment of the present invention;

[0064] Fig.16 This is a statistical diagram of SPE statistics for the F6 rack in an embodiment of the present invention;

[0065] Fig.17 This is a statistical diagram of T2 statistics of the F7 rack in an embodiment of the present invention;

[0066] Fig.18 This is a statistical diagram of SPE statistics for the F7 rack in an embodiment of the present invention;

[0067] Fig.19 It is a statistical diagram of the contribution value of the F1 rack variables in an embodiment of the present invention;

[0068] Fig. 20 This is a statistical diagram of the contribution value of the F2 rack variables in an embodiment of the present invention;

[0069] Fig.21 This is a statistical diagram of the contribution value of the F3 rack variables in an embodiment of the present invention;

[0070] Fig. 22 This is a statistical diagram of the contribution value of the F4 rack variables in an embodiment of the present invention;

[0071] Fig.23 This is a statistical diagram of the contribution value of the F5 rack variables in an embodiment of the present invention;

[0072] Fig.24 This is a statistical diagram of the contribution value of the F6 rack variables in an embodiment of the present invention;

[0073] Fig.25 This is a statistical diagram of the contribution value of the F7 rack variables in an embodiment of the present invention;

[0074] Fig.26 The overall monitoring T in the embodiment of the present invention 2 Statistics chart;

[0075] Fig. 27 It is a statistical diagram of overall monitoring SPE statistics in an embodiment of the present invention;

[0076] Fig.28 A statistical diagram of the total contribution values ​​of quality-related variables in an embodiment of the present invention;

[0077] Fig.29 A genetic topology map is screened for quality-related abnormal variables in an embodiment of the present invention. DETAILED DESCRIPTION

[0078] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0079] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0080] The plate and strip rolling process is a multi-stand joint rolling process. There is data transmission between the stands, and there is information flow and mass flow transmission in each rolling process. In the quality control and diagnosis stage of the rolling production process, most of them are reflected in the overall monitoring and diagnosis of the rolling stage, ignoring that the information flow and mass flow of the rolling process are also in a state of transmission between multiple stands. The abnormal fluctuation of variables in the rolling process may cause the amplification of abnormal signals, resulting in the neglect of the original abnormal variables in the diagnosis process and misjudgment.

[0081] In view of the above problems, in order to realize the root cause diagnosis of quality-related abnormalities in multi-stand hot finishing rolling process, considering the genetic characteristics of variables between processes, the present invention proposes a joint monitoring and diagnosis method for the hot finishing rolling process of plates and strips based on PLS algorithm. Squares, partial least squares method, is a new type of multivariate statistical data analysis method. By projecting the high-dimensional data space of independent variables and dependent variables into the corresponding low-dimensional space, the mutually orthogonal eigenvectors of independent variables and dependent variables are obtained respectively, and then the univariate linear regression relationship between the eigenvectors of independent variables and dependent variables is established. It can not only overcome the problem of collinearity, but also emphasize the explanatory and predictive role of independent variables on dependent variables when selecting eigenvectors, remove the influence of unhelpful noise on regression, and make the model contain the least number of variables) The data sub-blocks composed of the thickness-related variables of each rack in the hot finishing stage are modeled, and the fluctuation of the quality-related monitoring signal is obtained by monitoring statistics. The contribution graph is used to build the contribution histogram of each sub-block variable to obtain the key variables that cause abnormal quality fluctuations. Finally, the root causes of quality-related faults are identified by combining the joint monitoring strategy of sub-racks, and the root causes of quality-related abnormalities in the hot finishing process of multiple racks are diagnosed, which improves the accuracy of diagnosis.

[0082] like Figure 1 As shown, a method for joint monitoring and diagnosis of sub-racks in a hot finishing rolling process of a plate and strip based on a PLS algorithm is provided in an embodiment of the present invention, comprising:

[0083] C1 is the data layer. According to the hot-rolled strip finishing production stage, the thickness quality-related process variables of each finishing stand are selected to construct the production data sub-block samples of each stand, and the original data is standardized at the initial stage of the data.

[0084] In this embodiment, the hot-rolled strip finishing production stage flow chart is as follows: Figure 3 As shown in the figure, the rolling stage is composed of a seven-stand four-roller finishing mill. Plate and strip temperature measuring devices are arranged at the entrance and exit of the finishing rolling mill. The entrance temperature is used as the finishing entrance temperature variable, and the exit temperature of the last stand is used as the final rolling temperature variable. The ibaAnalyzer offline software is used to extract relevant data from the data storage in the actual production process, and the sub-blocks are divided according to each stand. The actual measured data of the plate and strip thickness during the rolling process is shown in Figure 4 As shown, it can be seen that the difference in head thickness of the plate and strip exceeds the limit and the head thickness of the strip is too thick.

[0085] In this embodiment, the data of the finishing rolling stage of the 1580 hot rolling production line is used as the training set and the test set. The data in the table is divided into data sub-blocks of each rack according to the process variables of each rack, and the original data is standardized by the mean and standard deviation of the data to eliminate the influence of its dimension on the results of the monitoring and diagnosis process. The names of the data variables are shown in Table 1:

[0086] Table 1

[0087]

[0088] In this embodiment, 44 process variables are collected, including 43 process variables and 1 quality variable.

[0089] C2 is the feature extraction layer. The PRESS (Predicted Error Sum of Squares) cross-validation model is used to verify the cumulative contribution rate of each rack. The number of principal components that can explain 85% of the feature information in the original data is determined based on the cumulative contribution rate.

[0090] In a specific implementation, the step of determining the number of pivot elements includes:

[0091] S201, given the number of main elements of each rack A i (i=1…7), calculate the prediction error after modeling and testing; divide the sample data into two groups, n2 samples are used for regression modeling, and n1 samples are used for testing.

[0092] S202. According to the sampling test, the total prediction error is:

[0093]

[0094] In the formula, y i is the i-th sample point among n1 test sample points; The predicted value for the i-th sample point of the test sample.

[0095] S203, use the above method to calculate the prediction error S when the number of principal components is A-1 A-1 , the cross validity of the Ath principal component of the dependent variable is:

[0096]

[0097] S204, when When , the added Ath principal component will improve the model accuracy, otherwise the added component will be discarded.

[0098] This method requires first giving the number of principal components of each rack, and then checking one by one whether the number of principal components is appropriate. When the cycle meets the test conditions, the number of principal components A is the number of principal components used in the final monitoring modeling.

[0099] According to steps S201 to S204, the number of principal components selected in this embodiment is shown in Table 2:

[0100] Table 2

[0101]

[0102] C3 is the statistical monitoring layer, which uses the partial least squares method to build the data sub-block model of each rack, calculate the sub-block statistics, and obtain the production process monitoring data status information of each rack;

[0103] S301, using partial least squares method to establish each rack data sub-block model; Figure 2 shown.

[0104] The external model can be obtained by taking the process variables of each stand as the independent variables as the input matrix X of the model and taking the thickness quality of the finishing rolling outlet as the dependent variable as the output matrix Y of the model;

[0105] The PLS model is formed based on the above variables:

[0106]

[0107] Among them, T is the process variable score matrix, P is the process variable load vector; U is the quality variable score matrix; Q is the quality variable load vector; E is the process variable residual matrix; F is the quality variable residual matrix; A is the number of principal components of the PLS model; t j ,u j The principal component vectors extracted from process variables and quality variables respectively. When extracting principal components, PLS should not only consider the data information that can best summarize the independent variable space, but also consider the explanatory effect of the principal component of the independent variable on the change of the dependent variable.

[0108] S302, calculating monitoring statistics of each data sub-block according to the score vector and the residual;

[0109] The new input sample is x new , training model independent variable projection direction matrix R = [r1, r2, ... r A ]. Satisfying T = XR, the principal component matrix can be obtained. Then the projection direction matrix is:

[0110] R=W(P T W) -1 (4)

[0111] P T R=R T P=W T W=I A (5)

[0112] Compute the score vector and residual:

[0113] t new =R T x new (6)

[0114]

[0115] Calculate T 2 And the Q statistic:

[0116]

[0117]

[0118] Where ∧ is the covariance of the score matrix in the training model.

[0119] S303, calculating the control limit of each rack monitoring process;

[0120] Statistical Process T 2 and Q follow F distribution and standard normal distribution respectively. Calculate the control limits of the two variables:

[0121]

[0122] S304. Obtain the stability of the process variable monitoring of each rack according to the monitoring statistics of each data sub-block, and obtain status information on whether the production process is stable.

[0123] When the system is running stably and no faults occur, the relationship between the statistic and the control limit is:

[0124]

[0125] When the statistic and the control limit do not satisfy the above relationship, it indicates that there is an abnormality in the production process.

[0126] Figure 5-Figure 18 The monitoring statistics of the partial least squares method for each stand in this embodiment show that most sample points of the plate and strip are below the control limit during the rolling process, while the head of the strip exceeds the limit to varying degrees. It can be seen that with the progress of rolling, the abnormal over-limit of the head of the plate and strip gradually decreases, and the over-quality of the head of the plate and strip is corrected to a certain extent during the rolling process. Therefore, the initial positioning of the over-thickness phenomenon of the head of the plate and strip during the monitoring process is consistent with the actual measurement results.

[0127] C4 is the diagnostic reasoning layer. First, according to the projection size of each rack data sub-module on the principal component of the variable, the contribution value of each rack process variable is calculated to obtain the contribution diagram of each variable to quality-related faults. The cause of the abnormal quality state in the rack is determined by combining the contribution diagram and the state information of whether the production process is stable. Then, based on the Granger causality idea, the causal topological structure of the continuous rolling process is constructed, and the genetic topological structure of each rack sub-monitoring module is established. According to the genetic characteristics of quality between each rack and the maximum contributing variable of the contribution diagram of each rack, the root cause of the fault is inferred.

[0128] Among them, the contribution graph is composed of the contribution value of each variable of the over-limit sample point to the squared prediction error. The calculation formula for the contribution value of the j-th variable of the over-limit sample point i to the squared prediction error is as follows:

[0129]

[0130] In the formula, is the total contribution value; x ij is the detection value of the jth variable in the i-th sample point; is the predicted value of the jth variable in the ith sample point.

[0131] Adding expectation factors to make it more consistent with diagnostic requirements;

[0132]

[0133] According to the contribution graph, the variable with the larger contribution value is most likely to cause abnormal fluctuations in quality-related variables, thereby finding the root cause of the abnormality in quality-related variables.

[0134] Figure 19-Figure 25 It is a histogram of the variable contribution values ​​in this embodiment. It can be seen from the contribution diagram that F1 frame variable 1 (rolling speed), F2 frame variable 2 (front tension), F3, F4, F5 frame variable 4 (entrance thickness), F6 frame variable 1 (rolling speed), F7 frame variable 6 (bending roll force) are important influencing variables in each sub-block.

[0135] By combining the diagnostic results of each sub-block, the main cause of the abnormal variables of each rack is inferred, and the root abnormal variables and passive abnormal variables caused by genetic characteristics are obtained. According to the reasoning of each sub-block, the root cause diagnosis of quality-related faults in the multi-rack process production process is realized.

[0136] Figure 26-28 For the overall monitoring and contribution diagram of the rolling stage, it can be seen through monitoring that the statistical monitoring of quality-related variables in this embodiment is mostly below the control limit, and F6 (bending roll force) becomes the main factor for the head deviation. However, the abnormal head thickness between the F1-F7 frames exceeds the limit, which is caused by the abnormal head of the finishing rolling material. Therefore, the diagnostic results of the overall monitoring and diagnosis process obviously ignore the genetic characteristics of the quality of the multi-frame rolling process.

[0137] In the above embodiment, each sub-block monitoring model is constructed to calculate the production process monitoring statistics such as Figure 4-Figure 18 As shown in the figure, it can be seen that the strip head exceeds the limit during the finishing rolling process. Construct a contribution diagram of each variable to the strip head abnormality, as shown in Figure 19-Figure 25As shown in the figure, the inlet temperature of F1 stand contributes the most to the abnormal strip head, and the inlet thickness variables or roll gap values ​​of F2-F7 stands make a large contribution. Due to the genetic characteristics of the quality of the continuous rolling process, the analysis of the causal topological relationship shows that the thickness tolerance of the hot-rolled strip head is caused by the low rolling inlet temperature, and the head thickness abnormality is transmitted during the rolling process, causing the subsequent stand inlet thickness or roll gap value detection abnormality. It is concluded that the main reason affecting the thickness tolerance of the strip head is the abnormal inlet temperature of F1 stand, such as Fig.29 shown.

[0138] Comprehensive monitoring and diagnosis of each stand sub-block show that the strip head tolerance during the rolling process did not cause abnormal fluctuations in thickness-related variables of each stand, while some sample points in the rear stand appeared out of tolerance in the quality-independent SPE statistics. Although it had no direct impact on the final strip quality, the abnormal fluctuation as a potential variable should also be taken seriously. Therefore, the technical solution provided in this embodiment achieves the goal of monitoring and diagnosing quality-related variables in a multi-stand hot rolling process taking into account the genetic characteristics of quality.

[0139] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for joint monitoring and diagnosis of the sub-stands in the hot finishing rolling process of a plate and strip based on a PLS algorithm, characterized in that: The following steps are involved: According to the hot-rolled strip finishing production process, the production data related to thickness quality is collected, the collected data is standardized, and the data is divided into sub-blocks according to each sub-rack; Determine the cumulative contribution rate of each rack variable according to the forecast error square sum cross-validation model, and obtain the number of principal elements of each data sub-block monitoring variable; Based on the number of principal components of the monitoring variables of each data sub-block, the partial least squares method (PLS) is used to establish the data sub-block model of each rack, and the principal component vector of the variable is extracted; Calculating monitoring statistics of each data sub-block according to the data sub-block model, and determining monitoring process control limits of each rack; According to the monitoring statistics of each data sub-block and the monitoring process control limit of each rack, the stability of the process variable monitoring of each rack is obtained, and the status information of whether the production process is stable is obtained; According to the projection size of each rack data submodule on the principal component of the variable, the contribution value of each rack process variable is calculated to obtain the contribution diagram of each variable to the quality-related faults. The cause of the abnormal quality state in the rack is determined by combining the contribution diagram and the state information of whether the production process is stable; Based on the Granger causality concept, the causal topology of the rolling process is constructed, and the genetic topology of each rack monitoring module is established. According to the genetic characteristics of quality between racks and the maximum contributing variable of the contribution graph of each rack, the root cause of the fault is inferred; Among them, based on the number of principal components of the monitoring variables of each data sub-block, the partial least squares method PLS is used to establish the data sub-block model of each rack, including: The process variables of each hot finishing rolling stand are used as independent variables as the input matrix X of the model, and the thickness quality of the finishing rolling outlet is used as the dependent variable as the model output matrix Y to obtain the external model: Among them, T is the process variable score matrix, P is the process variable load vector; U is the quality variable score matrix; Q is the quality variable load vector; E is the process variable residual matrix; F is the quality variable residual matrix; A is the number of principal components of the PLS model; t j ,u j The principal components extracted from process variables and quality variables respectively.

2. The method for joint monitoring and diagnosis of the sub-stands in the hot finishing rolling process of a plate and strip based on the PLS algorithm according to claim 1, characterized in that: The cumulative contribution rate of each rack variable is determined according to the forecast error square sum cross-validation model, and the number of principal elements of the monitoring variables of each data sub-block is obtained, including: Given the number of pivot elements in each rack, A i , and then calculate the prediction error after modeling and testing; divide the sample data into two groups, n2 samples are used for regression modeling, and n1 samples are used for testing; According to the sampling test, the total prediction error is: Where y i is the i-th sample point among n1 test sample points; The predicted value of the i-th sample point of the test sample; Calculate the prediction error S of n1 sample points for A-1 components A-1 , then the cross validity of the Ath principal component of the dependent variable is: when When , the added Ath principal component will improve the model accuracy, otherwise the Ath principal component will be discarded.

3. The method for joint monitoring and diagnosis of the sub-stands in the hot finishing rolling process of a plate and strip based on the PLS algorithm as claimed in claim 2, characterized in that: Calculating monitoring statistics of each data sub-block according to the data sub-block model includes: Calculate the score vector t of the variable data in the i-th rack production process new,i and residual Among them, x new,i is a new sample of the variable data related to the thickness quality of the plate strip to be monitored, R is the projection direction matrix, R = W (P T W) -1 ;P T R=R T P=W T W=I A ; R T is the transpose of the projection direction matrix, I is the unit matrix; I A is the unit matrix corresponding to the number of principal elements; W is the weight matrix of the input matrix; Calculate T based on the score vector and residual 2 And the Q statistic: Where ∧ is the score matrix covariance in the training model, is the transpose of the score vector, is the transpose of the new sample.

4. The method for joint monitoring and diagnosis of the sub-stands in the hot finishing rolling process of a plate and strip based on the PLS algorithm as claimed in claim 3, characterized in that: The calculation of the control limits of each rack includes: Statistics T 2 and Q follow F distribution and standard normal distribution respectively, calculate the control limits of the two variables and in, is the statistic T 2 The control limits, is the control limit of the statistic Q, n is the number of sample points, and α is the control limit confidence level; S is the variance of the Q statistic, and μ is the mean of the Q statistic; F A,n-A；α and The distribution of data.

5. The method for joint monitoring and diagnosis of the sub-stands in the hot finishing rolling process of a plate and strip based on the PLS algorithm according to claim 1, characterized in that: According to the projection size of each rack data submodule on the principal component of the variable, the contribution value of each rack process variable is calculated to obtain the contribution diagram of each variable to the quality-related faults, including: Calculate the contribution of the j-th variable of the out-of-limit sample point i to the squared prediction error; In the formula, is the total contribution value; x ij is the detection value of the jth variable in the i-th sample point; is the predicted value of the jth variable in the i-th sample point; Adding expectation factor: Based on the contribution value of each variable of the out-of-limit sample point to the squared prediction error after adding the expected factor, a contribution graph is formed.

Images