Hot rolling production line surface defect process monitoring method based on rank test non-parameter control chart

Abstract

The invention relates to a hot rolling production line surface defect process monitoring method based on a rank test non-parameter control chart, which comprises the following steps of: acquiring part number statistical data with different sampling time intervals, different batch sizes and different defect types in a hot rolling defect statistical table and a hot rolling production line production plan table; calculating to obtain a high-dimensional rank matrix of the sample after the proportion number of each defect is counted; obtaining an asymptotic rank transformation statistic after asymptotic rank transformation and removal of negligible asymptotic terms, and further establishing a control chart; and calculating the average operation chain length of the control chart of the real-time hot rolling production line surface defect data, and combining with the control chart with the sliding window to realize the state monitoring of the hot rolling production line surface defect process. According to the method, the problem that the mean value and the variance cannot be monitored under the condition that data distribution and correlation are unknown can be solved. And the method has a better monitoring effect in the process monitoring of the actual hot rolling production surface data.

Description

Technical Field

[0001] The present invention relates to a technology in the field of hot rolling production control, specifically a method for monitoring the surface defect process of a hot rolling production line based on a rank test non-parametric control chart. Background Art

[0002] Modern steel production often requires multiple processing steps. Generally, the final rolled steel products need to go through processes such as ironmaking, steelmaking, continuous casting, hot continuous rolling, and cold rolling. Among them, the most important link in slab rolling is the hot continuous rolling process stage. Therefore, by monitoring the hot rolling process through slab defect data, abnormal situations in the hot rolling production process can be effectively prevented and controlled, and the stability of the hot rolling production process can be improved. Applying control charts to the production of rolled steel can effectively monitor the stability of the hot rolling production process and promptly detect abnormal situations that occur during the production process. Conducting quality control over the production process also means improving the stability of the products, thereby enhancing the economic benefits of hot rolling production. Summary of the Invention

[0003] Aiming at the problems in the prior art that the data distribution is unknown and the number of controlled samples is small during the monitoring process, the present invention proposes a method for monitoring the surface defect process of a hot rolling production line based on a rank test non-parametric control chart. In view of the data characteristics of the hot rolling surface defect data, such as inconsistent sampling time intervals, inconsistent batch sizes, non-normal distribution, and non-Poisson distribution, through high-dimensional rank test non-parametric process monitoring based on the proportional method, it can be applicable to different data distributions. Compared with traditional monitoring methods, the monitoring variable is changed from the traditional mean and variance to its own high-dimensional rank test statistic. Because the high-dimensional data order represents the degree of dispersion of a set of data, it can overcome the problem that the mean and variance cannot achieve monitoring when the data distribution and correlation are unknown. It has a better monitoring effect in the actual process monitoring of hot rolling production surface data.

[0004] The present invention is realized through the following technical solutions:

[0005] The present invention relates to a method for monitoring the surface defect process of a hot rolling production line based on a rank test non-parametric control chart. By collecting the statistical data of the number of parts with different sampling time intervals, batch sizes, and defect types in the hot rolling defect statistical table and the hot rolling production line production plan form, and calculating the proportion numbers of each defect, the high-dimensional rank matrix of the sample is obtained; after asymptotic rank transformation and removing negligible asymptotic terms, the asymptotic rank transformation statistic is obtained, and then a control chart is established; then, by calculating the average run length of the control chart of the real-time hot rolling production line surface defect data, combined with the control chart with a sliding window, the state monitoring of the hot rolling production line surface defect process is realized.

[0006] The present invention relates to a system for implementing the above method, including: a control chart unit, a statistical unit, and a process monitoring unit, wherein: the control chart unit uses controlled data and set control chart constants to calculate the control limits of the control chart to construct the control chart; the statistical unit uses the input discrete data to be monitored, converts it into a proportional number through calculation, calculates the run sum and the asymptotic rank transformation, and finally outputs the control chart test statistic; the process control unit realizes the monitoring of the surface process defects of the hot rolling production line by judging the relationship between the control chart statistic and the control limit. Technical effects

[0007] Considering the characteristics of the data in the surface defect data set of the hot rolling production line, such as inconsistent sampling time intervals, inconsistent batch sizes, non-normal distribution, and non-Poisson distribution of the data, the present invention uses the asymptotic rank transformation statistic based on the proportional method to cope with the above data characteristics, combines the control chart, and calculates the corresponding control limits according to the controlled data, so as to construct a control chart with a sliding window to monitor the process.

[0008] Compared with the existing multivariate exponentially weighted moving average control chart (MEWMA) method, taking the average run length (ARL) as the evaluation index, the first statistic after the abnormal data appears in the present invention alarms, but the MEWMA control chart alarms at the 23rd statistic after the abnormal data appears, and its alarm time is longer. In addition, the MEWMA control chart also has false alarms under the controlled state, and the data fluctuations are large under the controlled state. The monitoring of the present invention is more stable under the controlled state and has excellent monitoring capabilities. Description of the drawings

[0009] Figure 1 is the flow chart of the present invention;

[0010] Figure 2 is the calculation flow chart of the rank test control quantity;

[0011] Figure 3 is the Q-Q chart of 6 kinds of defect data;

[0012] In the figure: a is edge crack, b is iron red skin, c is dot indentation, d is roll break, e is roll mark, f is scratch;

[0013] Figure 4 is the heat map of the correlation of hot rolling defect data;

[0014] Figure 5 is the schematic diagram of the control chart monitoring result of the embodiment on the hot rolling production surface defect data set;

[0015] In the figure: (a) the control chart of the present invention; (b) the MEWMA control chart. Detailed implementation manners

[0016] Such asFigure 1 As shown in the figure, this embodiment relates to a process monitoring method for surface defects in a hot rolling production line based on a non-parametric control chart using rank test for surface defect data of a certain iron and steel company's hot rolling production line, including:

[0017] Step 1: Obtain the statistical data of the number of parts with different sampling time intervals, batch sizes, and defect types from the hot rolling defect statistical table and the hot rolling production line production plan form, and use the normalized proportional method to transform the discrete hot rolling defect data according to the batch and defect type, so as to obtain the proportion of the p-th type of defect in the total P defects in the same batch. Where: X ip represents the number of the p-th type of defect at the i-th moment, p ∈ [1, P], and Z ip represents the proportion of the p-th type of defect in the total P defects in the same batch.

[0018] When the process is in a stable operation state, it is generally considered that the proportion number Z ip obeys a uniform distribution, that is, when it does not conform to the uniform distribution, it is considered that the process is abnormal.

[0019] Step 2: Calculate the rank of the obtained proportion number Z ip Assume that Z ij =(Z ij1 ,......, Z ijp ) T are the same and independently distributed observed values, i = 1, 2; j = 1,......, n i , use n = n 1 +n 2 to represent the total sample size, N = p(n 1 +n 2 ) is all the observed values of p dimensions corresponding to the total sample size n, then the rank matrix R ijk of all N observed values Z ijk The element r ijk in it represents the order of the Z ijk element in the k-th column, 1 ≤ r ijk ≤ n, the rank matrix The pn i observed values Z ijk of the i-th group in the high-dimensional rank matrix The element in it represents the order of the element in the k-th column, the high-dimensional rank matrix

[0020] Step 3: For the high-dimensional rank matrix of the sample Solve its asymptotic rank transformation Y ij =(Y ij1 ,......,Y ijp ), where: X T ijk =H(X ijk ), the average of the marginal distribution functions Therefore, the asymptotic rank transformation Y ij and the rank of the sample satisfy

[0021] Step 4: For the asymptotic rank transformation Y ij , further remove the negligible asymptotic terms to obtain the asymptotic rank transformation statistic

[0022] Step 5: Calculate and determine whether the asymptotic rank transformation statistic based on the non - parametric hypothesis is greater than the approximate α - test Z α , specifically: where: Z α is the (1 - α) quantile of the standard normal distribution.

[0023] Step 6: Set a sliding window for the asymptotic rank transformation statistic based on the non - parametric hypothesis, establish a control chart, and accumulate the information contained in the historical data to achieve the goal of efficient and rapid alarm.

[0024] The control chart mentioned above is a unilateral exponentially weighted moving average (EWMA) control chart, and its statistic is: where k 0 =0, λ is the smoothing coefficient, λ = 0.05, 0.1.

[0025] Step 7: Calculate the control limits of the control chart where: the smoothing coefficient λ, the sliding window D, and the control limit constant L.

[0026] Step 8: Calculate the average run length of the control chart for the real - time surface defect data of the hot - rolling production line, and combine the control chart with a sliding window built in Step 7 to realize the state monitoring of the surface defect process of the hot - rolling production line.

[0027] In this embodiment, further application verification was carried out on the measured data set of surface defects on the hot-rolling production line of a certain steel company in China. The specific application analysis results are as follows: The data set records the surface defect data of hot-rolling production from March 28, 2022 to March 30, 2022. Some of the original data are shown in Table 1. The production process record table shows that there was a problem with abnormal process parameters of the hot-rolling production line on March 30, 2022.

[0028] Table 1 Statistical Table of Defects on the Hot-Rolling Production Line

[0029] By observation, it is found that the sampling time intervals of the defect data are not consistent. This is because when the hot-rolling image sensor identifies slab defects, it collects and statistics in batches. Such a sampling method will cause the defect data to show large numerical fluctuations under stable operating conditions. If only the mutation (sudden increase / sudden decrease) of the data is used as the basis for judging the abnormal operating state, such monitoring results are inaccurate.

[0030] By observation, it is found that the production batches under different orders are inconsistent, and at the same time, orders of multiple users will be borne within a short time interval.

[0031] Perform a Poisson distribution hypothesis test on the hot-rolling defect data. The results are shown in Table 2. Since p << 0.5, these discrete data do not follow the Poisson distribution.

[0032] Table 2 Poisson Distribution Test Results of False Defects

[0033] Perform a normal distribution hypothesis test on the hot-rolling defect data. The results are as Figure 3 shown. It can be found that even under the condition of a large amount of data, the hot-rolling data still does not follow the normal distribution.

[0034] Perform a correlation test on the hot-rolling defect data. The results are as Figure 4 shown. According to the Pearson coefficient, it can be found that there is a certain correlation between individual variables.

[0035] To sum up, these surface defect data have data challenges such as inconsistent sampling time intervals, inconsistent batch sizes, non-normal distribution, non-Poisson distribution, and correlation between variables. Using traditional control charts cannot effectively monitor the discrete hot-rolling defect data.

[0036] Select the hot-rolling defect data among them as the direct data input of the present invention.

[0037] To verify the efficiency and accuracy of the hot-rolled production line surface defect process monitoring method based on the rank test non-parametric control chart, the present invention is compared and analyzed with the existing high-dimensional non-parametric process monitoring method (MEWMA) control chart: According to the production process record form, the data to be monitored consists of 180 stable operation data and 120 abnormal data at continuous time. The operation data is input into the EWMA control chart and MEWMA control chart of the present invention in sequence, and the results are as Figure 5 shown.

[0038] As Figure 5 shown, using the control chart of the present invention to monitor this data set, it can be found that: the first 180 sample points belong to the hot-rolled data of stable operation. At this time, the statistical quantities of the control chart of the present invention are all within the control limits; the latter 120 sample points belong to the hot-rolled data under abnormal process parameters. At this time, the statistical quantities of the control chart of the present invention all exceed the control limits, and the control chart judges that the process is abnormal and alarms. Using the MEWMA control chart to monitor this data set, it can be found that: compared with the EWMA control chart of the present invention, the MEWMA method issues an alarm at the 203rd point, and the alarm time is longer; at the same time, the MEWMA control chart also has false alarm situations under the controlled state, and the data fluctuations are larger under the controlled state. The control chart of the present invention is more stable in the monitoring under the controlled state.

[0039] The application results of the above embodiments on the surface defect data data set of the hot-rolled production line of a steel company show that compared with the current traditional high-dimensional non-parametric process monitoring method, the proposed method can monitor the production process more stably, more effectively monitor the abnormal state in the hot-rolled production process, and alarm quickly.

[0040] In summary, the present invention is applicable to the hot-rolled production line. The surface defect process monitoring of the hot-rolled production line is an important link in the quality control of the hot-rolled steel production process. The present invention is based on the high-dimensional rank test non-parametric process monitoring method based on the proportional method, and solves the process monitoring under the data difficulties of inconsistent sampling time intervals, inconsistent batch sizes, non-normal distribution of data, and non-Poisson distribution.

[0041] The above specific implementation can be locally adjusted in different ways by those skilled in the art without departing from the principles and purposes of the present invention. The protection scope of the present invention is subject to the claims and is not limited by the above specific implementation. All implementation solutions within its scope are subject to the present invention.

Claims

1. A method for monitoring the surface defect process of a hot rolling production line based on a rank - test non - parametric control chart, characterized in that, by collecting the statistical data of the number of parts with different sampling time intervals, different batch sizes and different defect types in the hot rolling defect statistical table and the hot rolling production line production plan form, and calculating the proportion of each defect, a high - dimensional rank matrix of the sample is obtained; after asymptotic rank transformation and removing negligible asymptotic terms, an asymptotic rank transformation statistic is obtained, and then a control chart is established; furthermore, by calculating the average run length of the control chart for the real - time surface defect data of the hot rolling production line, combined with the control chart with a sliding window, the state monitoring of the surface defect process of the hot rolling production line is realized.

2. The method for monitoring the surface defect process of a hot rolling production line based on a rank - test non - parametric control chart according to claim 1, characterized in that specifically, it includes: Step 1: Obtain the statistical data of the number of parts with different sampling time intervals, batch sizes, and defect types from the hot-rolled defect statistics table and the hot-rolled production line production plan form. According to the batch and defect type, use the normalized proportional method to transform the hot-rolled defect discrete data, so as to obtain the proportion of the p-th defect in the total P defects in the same batch. Where: X ip represents the number of the p-th defect at the i-th moment, p ∈ [1, P], Z ip represents the proportion of the p-th defect in the total P defects in the same batch; Step 2: For the proportional number Z obtained after transformation ip perform rank calculation. Assume Z ij =(Z ij1 ,......, Z ijp ) T are identical and independently distributed observations, where i = 1, 2; j = 1,......, n i , and use n = n 1 + n 2 to represent the total sample size. Let N = p(n 1 + n 2 ) be all the observations of p dimensions corresponding to the total sample size n. Then, for all N observations Z ijk , the element r ijk in the rank matrix R ijk ijk represents the order of the Z ijk element in the k-th column, where 1 ≤ r ijk ≤ n. For the high-dimensional rank matrix of the pn i observations Z ijk in the i-th group, the element in the high-dimensional rank matrix represents the order of the element in the k-th column, and the high-dimensional rank matrix Step 3: For the high-dimensional rank matrix of the sample Solve its asymptotic rank transformation Y ij =(Y ij1 ,......, Y ijp ) T , where: Y ijk =H(X ijk ), the average value of the marginal distribution function Therefore, the asymptotic rank transformation Y ij and the rank of the sample satisfy Step 4: For the asymptotic rank transformation Y ij , further remove the negligible asymptotic terms to obtain the asymptotic rank transformation statistic Step 5: Calculate and determine the asymptotic rank transformation statistic based on non-parametric assumptions whether it is greater than the approximate α-test Z α , specifically: where: Z α is the (1 - α)-quantile of the standard normal distribution; Step 6: Asymptotic rank transformation statistic based on non-parametric hypothesis Set a sliding window, establish a control chart, and accumulate the information contained in historical data to achieve the goal of efficient and rapid alarm; Step 7: Calculate the control limits of the control chart Where: smoothing coefficient λ, sliding window D, control limit constant L; Step 8: Calculate the average run length of the control chart for the real - time surface defect data of the hot rolling production line, and combine with the control chart with a sliding window built in Step 7 to realize the state monitoring of the surface defect process of the hot rolling production line.

3. The method for monitoring the surface defect process of a hot rolling production line based on a rank - test non - parametric control chart according to claim 2, characterized in that, When the process is in a stable operating state, the proportional number Z is usually considered ip to follow a uniform distribution, that is, when it does not conform to the uniform distribution, the process is considered abnormal.

4. The method for monitoring the surface defect process of a hot rolling production line based on a rank - test non - parametric control chart according to claim 2, characterized in that, The control chart mentioned above is a one-sided exponentially weighted moving average (EWMA) control chart, and its statistic is: where k 0 = 0, λ is the smoothing coefficient, and λ = 0.05, 0.

1.

5. A system for monitoring the surface defect process of a hot rolling production line based on a rank - test non - parametric control chart for implementing any one of the methods described in claims 1 - 4, characterized in that, it includes: a control chart unit, a statistical unit and a process monitoring unit, where: the control chart unit uses the controlled data and set control chart constants to calculate the control limits of the control chart to build the control chart, the statistical unit uses the input discrete data to be monitored, converts it into a proportion number through calculation and then calculates the order sum and asymptotic rank transformation, and finally outputs the control chart test statistic, and the process control unit realizes the monitoring of the surface process defects of the hot rolling production line by judging the relationship between the control chart statistic and the control limit.

Images