A spatiotemporal root cause diagnosis method for quality anomalies in hot rolling process

Abstract

The invention discloses a method for diagnosing the spatiotemporal root causes of abnormal quality in a hot rolling process, including: establishing a hot rolling process history database; aligning data samples according to the percentage of steel rolling position, and dividing the entire hot rolling process into multiple levels and sub-blocks; constructing fault detection models corresponding to each sub-block respectively, and training each fault detection model based on the history database to achieve real-time detection of quality-related faults in each sub-block; when a fault occurs, locating the spatial position of the abnormal quality according to the fault detection results corresponding to each sub-block, and determining the time position of the fault by reverse mapping the percentage of steel rolling position of each rack; calculating the size of the disproportionation coefficient of each process variable in the faulty rack, locating the fault root cause variable, determining the root cause variable causing the fault, locating the equipment where the fault occurs, and achieving spatiotemporal root cause diagnosis of the fault. The invention can better guide enterprise production and further improve the economic benefits of the enterprise.

Description

Technical Field

[0001] The present invention relates to the technical field of fault monitoring and diagnosis related to process industry quality, and in particular to a temporal and spatial root cause diagnosis method for abnormal quality of hot rolling process, which can realize real-time fault detection and temporal and spatial root cause diagnosis on industrial site. Background Art

[0002] The development goal of modern process industries is to be green, intelligent, and efficient. Typical process industries include petrochemicals, steel, nonferrous metals, etc. The above process industries are characterized by long processes, multiple levels, and multiple working conditions. If any process fails, the failure will continue to spread and evolve among multiple system levels with the flow of information, material flow, and energy flow, further affecting the stability of on-site equipment operation and the stability of industrial product quality. Since product quality and real-time working conditions cannot be monitored online, fault detection and root cause diagnosis in process industries have become a difficult point.

[0003] Fault detection and root cause diagnosis technology in process industries has become a research hotspot for scholars. However, many research results do not take the final product quality into consideration. In actual industrial processes, not all faults will cause product quality abnormalities. On-site operators pay more attention to faults that can cause product quality abnormalities and take corresponding optimization and control strategies in a timely manner. Therefore, it is very important to study quality-related fault detection and root cause diagnosis technology.

[0004] The hot rolling process is a typical modern process industry, which includes multiple processes such as heating furnace, roughing mill, hot output roller and flying shear, finishing mill, laminar cooling, and curling mill. Among them, the stable and safe operation of the roughing mill and the finishing mill is crucial to the quality of the finished strip steel. In order to ensure the safety and efficiency of the production of finished strip steel, the hot rolling process needs to be monitored. Once a fault occurs, a comprehensive fault diagnosis of the entire hot rolling process is required. However, after detecting the occurrence of a fault, the traditional fault diagnosis method can only roughly locate the variables that cause the fault, and the variables located are often multiple. Such root cause diagnosis methods cannot meet the actual production needs well. The main problems to be solved are as follows:

[0005] First, the hot rolling process has the characteristics of full process and multi-level, and the various processes are coupled with each other, which makes the traditional centralized fault detection and root cause diagnosis methods difficult to apply. In order to meet the needs of the actual production process and facilitate on-site staff to monitor and maintain production equipment, it is necessary to combine the process knowledge of the hot rolling process and propose a decentralized fault detection and root cause diagnosis method.

[0006] Second, the failures that occur during hot rolling do not always cause quality fluctuations in the final steel plate. However, traditional fault detection methods can only monitor whether abnormal working conditions occur. Once abnormal working conditions occur, in order to ensure production safety, on-site operators often take measures to stop production and eliminate hidden dangers. However, this approach does not meet the development goals of green and efficient modern process industries, and is not conducive to improving the economic benefits of enterprises. Therefore, it is necessary to focus on the production characteristics of the hot rolling process and the needs of enterprises, and to associate process variables with quality variables, and propose a quality-related fault detection and root cause diagnosis method.

[0007] Third, in general, traditional root cause diagnosis methods can only locate multiple process variables. Such root cause diagnosis methods are often too rough and cannot provide clear guidance to on-site operators. At the same time, traditional diagnostic methods locate faults at the equipment level, but not at the product level, that is, it is impossible to determine where the quality fluctuations of the industrial products obtained occur. It is not difficult to find that the diagnostic results obtained by traditional root cause diagnosis methods are not comprehensive enough. For this reason, it is necessary to design a novel spatiotemporal diagnosis method for the root cause of faults, which can detect the time location of the fault from a temporal point of view. From a spatial point of view, it is necessary not only to accurately locate the production unit where the fault occurs, that is, to realize the root cause diagnosis of the fault at the equipment level, but also to locate the spatial location where the quality fluctuations of the industrial products occur, that is, to realize the root cause diagnosis of the fault at the product level.

[0008] From the above, it can be seen that the significance of studying more refined and comprehensive fault root cause diagnosis methods is very important, that is, it is necessary to conduct spatiotemporal diagnosis of quality anomalies in the hot rolling process, and "time and space" includes two dimensions: time and space. From the time dimension, the fault diagnosis of the hot rolling process needs to locate the exact time when the fault occurs; from the space dimension, the fault diagnosis of the hot rolling process needs to locate not only the root variable that causes the fault, but also the specific position in the length direction of the finished strip where the quality anomaly exists. This fault diagnosis from the two dimensions of time and space can better guide the production of enterprises and further improve the economic benefits of enterprises. Summary of the invention

[0009] The invention provides a spatiotemporal root cause diagnosis method for abnormal quality in a hot rolling process, so as to solve the technical problem that the existing root cause diagnosis method for faults lacks refinement and comprehensiveness.

[0010] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0011] A method for diagnosing the spatiotemporal root causes of quality anomalies in a hot rolling process comprises the following steps:

[0012] Obtain historical data samples of hot rolling process and establish a historical database of hot rolling process;

[0013] The acquired data samples are aligned according to the percentage of rolling position, and the entire hot rolling process is divided into multiple levels and sub-blocks; each sub-block corresponds to a stand in the rolling process;

[0014] Constructing a fault detection model corresponding to each sub-block respectively, and training each constructed fault detection model based on the hot rolling process history database, and using the trained fault detection model to achieve real-time detection of quality-related faults of each sub-block;

[0015] When a fault occurs, the spatial position of the quality anomaly is located according to the fault detection results corresponding to each sub-block, and the time position of the fault is determined by reverse mapping the rolling position percentage of each rack;

[0016] Calculate the size of the disproportionation coefficient of each process variable in the faulty rack, locate the root cause variable of the fault, determine the root cause variable causing the fault, locate the faulty equipment, and realize the spatiotemporal root cause diagnosis of the fault.

[0017] Furthermore, the entire hot rolling process is divided into multiple levels and sub-blocks, including:

[0018] From a vertical perspective, the entire hot rolling process is divided into product layer, full process layer, subsystem layer and equipment layer. From a horizontal perspective, the entire hot rolling process is divided into the first sub-block, the second sub-block, the third sub-block, the fourth sub-block, the fifth sub-block, the sixth sub-block, the seventh sub-block and the eighth sub-block. Among them, the first sub-block corresponds to the stand in the rough rolling process, and the second to eighth sub-blocks correspond to the first to seventh stands in the finishing rolling process respectively.

[0019] Furthermore, the data samples acquired are aligned according to the percentage of rolling position, including: positive alignment of data from time dimension to space dimension, and negative alignment of data from space dimension to time dimension;

[0020] The implementation process of data alignment from the time dimension to the space dimension is as follows:

[0021] Read the process data of a slab rolling process in chronological order and store it in the corresponding vector;

[0022] The vectors corresponding to each rack are normalized, and the length of the steel plate rolled at each moment is divided by the total length of the steel plate rolled by the rack to obtain the percentage of the steel rolling position of each rack;

[0023] The obtained steel rolling position percentage data of each rack is expanded to construct an index column, the index column is constructed in the first column, and the second column is the steel rolling position percentage data of each rack;

[0024] Obtain the data length of the rolling position percentage data of each stand, and compare to obtain the longest data length;

[0025] Rewrite and splice the index columns of the steel rolling position percentage of each rack after the index column is constructed, and rewrite according to the steel rolling percentage position based on the steel rolling position percentage of the longest data length; when the steel rolling percentage position is not greater than the reference steel rolling percentage position, write the index of the same row, otherwise take the index of the next row; splice in sequence to form the index matrix of each rack data and outlet temperature and actual thickness;

[0026] The process data is re-read according to the constructed index matrix to obtain the percentage data of the rolling position of each rack after data alignment, and the different data are rewritten into column vectors with the number of rows being the long data length;

[0027] For the data anti-alignment from the space dimension to the time dimension, the time information is obtained through the reverse index of the rolling position percentage of each rack to achieve the data anti-alignment from the space dimension to the time dimension.

[0028] Furthermore, the variables of the hot rolling process include process variables and quality variables;

[0029] The fault detection models corresponding to the sub-blocks are constructed respectively, and the constructed fault detection models are trained based on the hot rolling process history database, and the trained fault detection models are used to realize real-time detection of quality-related faults of each sub-block, including:

[0030] The maximum information coefficient is used to replace the mutual information in the maximum relevance minimum redundancy algorithm to obtain an improved maximum relevance minimum redundancy algorithm.

[0031] For the process variables and quality variables of the hot rolling process corresponding to each sub-block, the quality-related process variables corresponding to each sub-block are screened out by using the improved maximum relevance minimum redundancy algorithm;

[0032] The kernel function is introduced into the canonical correlation analysis to obtain the kernel canonical correlation analysis model;

[0033] The screened quality-related process variables and corresponding quality variables corresponding to each sub-block are sent to the core canonical correlation analysis model to perform quality-related process monitoring.

[0034] Furthermore, the quality-related process variables and corresponding quality variables screened out from each sub-block are sent to the nuclear canonical correlation analysis model to perform quality-related process monitoring, including:

[0035] The process variables screened out from each sub-block are mapped into a high-dimensional space through nonlinear mapping;

[0036] The mapped process variables are used to conduct canonical correlation analysis with the corresponding quality variables to determine the statistics and control limits corresponding to each sub-block and conduct quality-related process monitoring.

[0037] Furthermore, the process variables screened by each sub-block are mapped to a high-dimensional space through nonlinear mapping, including:

[0038] The screened quality-related process variables are mapped into a feature space using a kernel function; wherein the kernel canonical correlation analysis model combines a polynomial kernel with a radial basis function kernel.

[0039] Furthermore, the spatial position of the quality anomaly is located according to the fault detection results corresponding to each sub-block, and the time position of the fault occurrence is determined by reverse mapping the rolling position percentage of each rack, including:

[0040] According to the fault detection results corresponding to each sub-block, the spatial position of the quality anomaly is located, and the timestamp corresponding to the fault is reversely indexed through the rolling position percentage of the faulty frame when the fault occurs, the time position of the fault is determined, and the length of the steel plate rolled by each frame at this time is summed up. Combined with the rolling position percentage of the faulty frame when the fault occurs, it is obtained where the quality fluctuation of the entire steel plate occurs.

[0041] Furthermore, the calculation formula of the disproportionation coefficient is:

[0042]

[0043] Where difference(i) represents the disproportionation coefficient of the i-th process variable, n is the number of sample points, represents the fault dataset, represents a normal data set;

[0044] The larger the disproportionation coefficient of the process variable in the faulty rack is, the more likely this process variable is to be the root variable causing the current fault.

[0045] The beneficial effects brought about by the technical solution provided by the present invention include at least:

[0046] 1. The present invention solves the drawbacks of the traditional centralized method. The present invention adopts a decentralized method for the roughing and finishing mills in the hot rolling process, dividing the entire process into multiple levels and sub-blocks. This multi-level and multi-sub-block fault detection and root cause diagnosis method facilitates engineers to promptly determine the precise location of the fault and perform targeted maintenance on on-site production equipment, better meeting the production needs of enterprises;

[0047] 2. The present invention can better explore the relationship between process variables and quality variables, realize the fault detection related to the quality of hot rolling process, and can better guide the production practice of enterprises and improve product quality.

[0048] 3. Compared with the traditional root cause diagnosis method that can only locate the fault at the equipment level, the spatiotemporal diagnosis method proposed in the present invention not only realizes the fault location in the time dimension, but also realizes the fault location in the space dimension. Among them, the fault location in the space dimension includes two aspects. On the one hand, the fault location at the equipment level is completed, that is, the fault is located at the production unit where the fault occurs. On the other hand, the fault location at the product level is completed, that is, the detailed location of the product quality fluctuation occurs. Such a spatiotemporal diagnosis method for the root cause of the fault is conducive to the enterprise to locate the root cause of the fault in time and take corresponding optimization and control measures. At the same time, timely detection of whether the products have quality problems can avoid defective products from entering the market and damaging the corporate image. Therefore, it is of great help to improve the economic value and brand value of the enterprise. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0050] Figure 1 Schematic diagram of the connotation of the spatiotemporal root cause diagnosis method of quality abnormality in hot rolling process provided by an embodiment of the present invention;

[0051] Figure 2 It is a schematic diagram of the execution flow of the spatiotemporal root cause diagnosis method of quality abnormality in hot rolling process provided by an embodiment of the present invention;

[0052] Figure 3 It is a schematic diagram of a specific execution flow of the spatiotemporal root cause diagnosis method for quality anomalies in a hot rolling process provided by an embodiment of the present invention;

[0053] Figure 4 It is a schematic diagram of the principle of the hot rolling process data alignment method provided by an embodiment of the present invention;

[0054] Figure 5Schematic diagram of fault detection results of the spatiotemporal root cause diagnosis method for abnormal quality in hot rolling process provided by an embodiment of the present invention; wherein (a) indicates that the fault starts from the third stand in the finishing process, and the third stand fails in the interval [16.7%, 50%]; (b) indicates that the fault starts from the fourth stand in the finishing process, and the fourth stand fails in the interval [16.7%, 50%]; (c) is the fault detection result of the stand in the roughing process and the fault detection results of the first, second, fifth, sixth and seventh stands in the finishing process;

[0055] Figure 6 It is a schematic diagram of the spatiotemporal root cause diagnosis results of the spatiotemporal root cause diagnosis method for abnormal quality in the hot rolling process provided by an embodiment of the present invention; wherein (a) is the time dimension monitoring result of the third rack, (b) is the time dimension monitoring result of the fourth rack, and (c) is a schematic diagram of the disproportionation coefficient of each variable of the third rack. DETAILED DESCRIPTION

[0056] In order to make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0057] In view of the problem that existing fault root cause diagnosis methods lack refinement and comprehensiveness, this embodiment provides a spatiotemporal root cause diagnosis method for quality anomalies in hot rolling process driven by knowledge and data, so as to solve the three problems existing in the prior art mentioned in the background technology. The specific solutions are as follows.

[0058] In order to solve the first problem, this embodiment proposes a multi-level and multi-sub-block fault detection and root cause diagnosis strategy, which solves the drawbacks of the traditional centralized method. This embodiment adopts a decentralized approach for the rough rolling and finishing rolling racks in the hot rolling process. From a vertical perspective, four levels are formed: product layer, full process layer, subsystem layer, and equipment layer. At the same time, from a horizontal perspective, eight sub-blocks from the first subsystem to the eighth subsystem are formed. This multi-level and multi-sub-block fault detection and root cause diagnosis method makes it easy for engineers to promptly determine the exact location of the fault and to perform targeted maintenance of on-site production equipment, better meeting the production needs of the enterprise. The various production processes in the hot rolling process, such as Figure 1 shown.

[0059] In order to solve the second problem, this embodiment proposes a quality-related fault detection method based on improved maximum relevance minimum redundancy-kernel canonical correlation analysis. Considering that the data of industrial processes are strongly nonlinear, the idea of ​​kernel function is introduced to improve the canonical correlation analysis. At the same time, the maximum information coefficient is used instead of the traditional mutual information to improve the maximum relevance minimum redundancy algorithm, which can better explore the relationship between process variables and quality variables. Kernel canonical correlation analysis links process variables with quality variables, and uses quality variables to decompose process variables into quality-related input subspaces and quality-irrelevant input subspaces. Statistics and control limits are designed in the quality-related input subspaces to monitor whether quality-related faults occur in the hot rolling process. In this way, quality-related fault detection in the hot rolling process is realized, which can better guide enterprise production practices and improve product quality.

[0060] In order to solve the third problem, this embodiment proposes a novel spatiotemporal diagnosis method for the root cause of the fault, that is, data alignment is performed according to the percentage of the rolling position of each rack. This data alignment method establishes a relationship between the sampling point, the timestamp and the rolling length of each rack, and finally the identification of the root cause variable of the fault is realized by comparing the size of the disproportionation coefficient of the process variable in the faulty rack. Compared with the traditional root cause diagnosis method that can only realize the fault location of the equipment layer, the spatiotemporal diagnosis method proposed in the present invention not only realizes the fault location of the time dimension, but also realizes the fault location of the space dimension. The fault location of the space dimension includes two aspects. On the one hand, the fault location of the equipment layer is completed, that is, the production unit where the fault occurs is located. On the other hand, the fault location of the product layer is completed, that is, the detailed location of the product quality fluctuation occurs. Such a spatiotemporal diagnosis method for the root cause of the fault is conducive to the enterprise to locate the root cause of the fault in time and take corresponding optimization and control measures. At the same time, timely detection of whether the products produced have quality problems can avoid the flow of defective products into the market and damage the corporate image. Therefore, it is of great help to improve the economic value and brand value of the enterprise.

[0061] Based on the above, the core idea of ​​the spatiotemporal root cause diagnosis method of quality anomalies in the hot rolling process of this embodiment is to monitor the hot rolling process and diagnose the root causes of faults from the time dimension and the space dimension. The general process is as follows: First, the real-time data is sensed and collected by sensors at the industrial site. Then, the data of each rack is spatially aligned according to the length direction by the percentage of the rolling position. Next, a quality-related abnormal space monitoring model is established for each rack. Once the occurrence of quality anomalies is identified, the corresponding fault root variable identification method is further combined to determine the root cause variable causing the fault. Finally, the time position of the fault is reversely indexed based on the spatial position of the fault, and the length of the rolled strip of each rack is summed to lock in where the quality fluctuation of the entire strip begins. Thereby, the spatiotemporal diagnosis of quality anomalies in the hot rolling process is realized, providing valuable guidance for the optimization and regulation of enterprises. The schematic diagram of the connotation of this method is shown in the following figure Figure 1 As shown, the execution steps are as follows Figure 2 and Figure 3 As shown, the following steps are included:

[0062] S1, obtaining historical data samples of hot rolling process and establishing a historical database of hot rolling process;

[0063] It should be noted that the rolling data of a batch of hot rolling process is divided into two parts, one part is rough rolling data, and the other part is finishing rolling data. Specifically, in this embodiment, the sampling period is 0.01s, and the rolling data is measured by sensors at the industrial site.

[0064] S2, aligning the acquired data samples according to the percentage of rolling position, and dividing the entire hot rolling process into multiple levels and sub-blocks; wherein each sub-block corresponds to a stand in the rolling process;

[0065] It should be noted that the traditional data alignment method uses time as the horizontal axis, but this data alignment method has some disadvantages for the root cause diagnosis of faults. Taking into account the actual production scenario of the hot rolling process, that is, the same part of the steel plate will pass through the roughing stand and the finishing stand in sequence at different sampling times. In other words, at the same sampling time, the rolling positions corresponding to the roughing stand and the finishing stand are not the same positions of the steel plate. At the same time, the lengths of the steel plates rolled by the roughing and finishing stands are also different. In order to better diagnose the root causes of faults in the hot rolling process, this embodiment aligns the data of the roughing and finishing rolling processes with the length of the steel plate. In order to make the length of the horizontal axis consistent, the data of the roughing and finishing rolling processes are aligned using the percentage of the rolled steel length (spatial position), such as Figure 4 shown.

[0066] The specific data alignment algorithm is implemented as follows:

[0067] ⑴ Read the rough rolling and finishing rolling process data of a slab in chronological order and store them in the corresponding vectors respectively;

[0068] (2) Normalize the vectors corresponding to each stand, and divide the length of the steel plate rolled at each moment by the total length of the steel plate rolled by the stand, and obtain a value range of [0%, 100%], that is, the percentage of the rolling position of each stand;

[0069] ⑶ Construct an index column for the percentage data of steel rolling position of each rack obtained in the above two steps. Construct an index column in the first column, and the second column is the percentage data of steel rolling position of each rack;

[0070] (4) Obtain the data length of the rolling position percentage data of each rack, and compare to obtain the longest data length, which will be used as a benchmark to expand the data of the remaining racks;

[0071] ⑸ Rewrite and splice the index columns of the steel rolling position percentage of each rack after the index column is constructed, and use the steel rolling position percentage of the longest data length as the benchmark, and rewrite according to the steel rolling percentage position. The specific operation is as follows: when the steel rolling percentage position is not greater than the benchmark steel rolling percentage position, write the index of the same row, otherwise take the index of the next row. Splice in sequence to form the index matrix of each rack data and outlet temperature and actual thickness;

[0072] (6) Reread the rough rolling and finishing rolling data according to the constructed index matrix to obtain the rolling position percentage data of each stand after data alignment. The different data are rewritten into column vectors with the number of rows being the long data length;

[0073] ⑺The above six steps realize the data alignment of rough rolling and finishing rolling data from the time dimension to the space dimension. In order to realize the reverse alignment of data from the space dimension to the time dimension, the time information can be obtained by reverse indexing the spatial information (the percentage of the rolling position of each rack).

[0074] Furthermore, after achieving the positive and negative alignment of the spatiotemporal dimensions of the rough rolling and finishing process data, taking into account the actual production needs of the hot rolling process, this embodiment divides the hot rolling process into four levels and eight sub-blocks according to the process characteristics and expert knowledge and combined with the production needs of the enterprise. From a vertical perspective, the four levels include the product layer, the full process layer, the subsystem layer, and the equipment layer. From a horizontal perspective, the eight sub-blocks include the first sub-block to the eighth sub-block, where the first sub-block corresponds to the rack in the rough rolling process, and the second sub-block to the eighth sub-block correspond to the first rack to the seventh rack in the finishing process, respectively.

[0075] S3, constructing a fault detection model corresponding to each sub-block respectively, and training each constructed fault detection model based on the hot rolling process history database, and using the trained fault detection model to achieve real-time detection of quality-related faults of each sub-block;

[0076] It should be noted that the variables in the hot continuous rolling process include two parts, namely process variables and quality variables. Among them, the process variables include rolling force, bending roll force, shifting roll force, rolling force difference, tension between stands, roll gap, power, roll speed, current, etc., and the quality variables include thickness, temperature, crown, flatness, width, etc. The process variables included from the rough rolling mill stand block to the seventh stand block of the finishing mill are represented by X1, X2......X8 respectively. Similarly, the quality variables are represented by Y1, Y2......Y8. In this regard, this embodiment uses the ImRMR-KCCA method to establish corresponding fault detection models in each subsystem, and based on the historical database of the hot continuous rolling process established in S1, trains the models of each system, determines the corresponding statistics and control limits, and realizes the process monitoring of each subsystem respectively. The specific implementation process is as follows:

[0077] S31, assume that the process variable X = [X1, X2, X3, X4, X5, X6, X7, X8] ∈ R N×m , and the quality variable Y = [Y1, Y2, Y3, Y4, Y5, Y6, Y7, Y8] ∈ R N×n , in order to better reveal and quantify the relationship between process variables and quality-related variables, this embodiment uses the improved maximum relevance minimum redundancy algorithm (ImRMR) to screen the process variables related to quality. Considering that the maximum information coefficient (MIC) has characteristics such as generality, balance, and symmetry, this embodiment uses the maximum information coefficient to replace the mutual information (MI) in the maximum relevance minimum redundancy algorithm to further improve the performance of the algorithm. The change range of the maximum information coefficient is from 0 to 1. For any two variables x i and x j , their maximum information coefficient is

[0078]

[0079] In the above formula, a and b are the number of grids divided in the x i , x j coordinate axis directions, and need to satisfy a * b < C, where the size of C is usually set to the 0.6th power of the data volume. I(x i , x j ) represents the mutual information between x i , x j .

[0080] The calculation formula for the maximum correlation degree between the process variable and the quality-related variable in each subsystem is:

[0081] maxD(X B , Y B ),

[0082] In the above formula, B = {1, 2, 3,…, 8}.

[0083] The calculation formula for the minimum redundancy between process variables in each subsystem is:

[0084] minR(X B ),

[0085] In the above formula, B = {1, 2, 3,…, 8}.

[0086] The calculation formula combining the two in each subsystem is:

[0087] maxα(X B ,Y B ),

[0088] In the above formula, B = {1, 2, 3,…, 8}.

[0089] Will As the threshold of α, where MIC mean is the average of the maximum information coefficients of process variables and quality variables.

[0090] The quality-related process variables of rough rolling and finishing rolling are screened out by the improved maximum relevance minimum redundancy algorithm. The quality variable is still Y = [Y1, Y2, Y3, Y4, Y5, Y6, Y7, Y8] ∈ R N×n .

[0091] S32, the quality-related process variables screened out by each subsystem and the corresponding quality-related variables are sent to the kernel canonical correlation analysis (KCCA) for quality-related process monitoring. Considering that industrial process data has strong nonlinearity, this embodiment introduces the kernel function into the canonical correlation analysis, that is, the process variables screened out by each subsystem are mapped to a high-dimensional space through nonlinear mapping, and then the mapped process variables and quality-related variables are used for canonical correlation analysis. The specific steps are as follows:

[0092] First, the kernel function is used to screen out the quality-related process variables X′ B Mapped into a feature space F, there is a nonlinear mapping Make However, this nonlinear mapping function It is difficult to calculate directly, so we need to define a positive semidefinite matrix K instead. satisfy The K matrix can be calculated by the kernel function. express and The dot product between K (i,j) represents the kernel function, They represent the i-th variable and the j-th variable in the B sub-block respectively.

[0093] In order to ensure better generalization capability, this embodiment combines the polynomial kernel and the radial basis function kernel together in the following manner:

[0094]

[0095] The former is a polynomial kernel, and the latter is a radial basis function kernel, where ω is the weight coefficient, ω∈[0,1]; σ represents the width parameter of the radial basis function.

[0096] In order to concentrate the mapped process variables and eliminate the mean effect in the feature space, the process variables are first preprocessed as follows:

[0097]

[0098] in,

[0099] K (0) represents the processed kernel function, and N represents the number of samples;

[0100] The quality-related process variables X′ will be screened out B Mapped to X through the kernel function * B , the quality-related variable is still Y B Then the mapped process variable is The quality variable is Y = [Y1, Y2, Y3, Y4, Y5, Y6, Y7, Y8] ∈ R N×n .

[0101] Canonical correlation analysis (CCA) is a multivariate statistical analysis method that uses the correlation between comprehensive variable pairs to reflect the overall correlation between two groups of indicators. Its basic principle is: in order to study the correlation between two groups of variables in general, two representative variables G and Q (linear combinations of the variables in the two variable groups) are extracted from the two groups of variables, and the correlation between these two variables is used to reflect the overall correlation between the two groups of variables. That is, in order to study the correlation between variable X * B With variable Y B The correlation between them introduces two new sets of variables G = A T X * B and Q = C T Y B, by studying the correlation between G and Q, we can get X * B and Y B The correlation between them, B = {1,2,3,…,8}. A represents the correlation between X * B The transformation matrix of Y B The transformation matrix of .

[0102] The Pearson correlation coefficient is a measurement tool used to calculate the correlation between variables. Its range is -1 to 1. Positive values ​​represent positive correlation, 0 represents no correlation, and negative values ​​represent negative correlation. The specific formula is as follows

[0103]

[0104] in, Represents X * B The mean of Represents Y B The mean of Represents X * B The standard deviation of Represents Y B The standard deviation of

[0105] In order to make X * B and Y B The Pearson correlation coefficient between them is the largest, so we only need to maximize the Pearson correlation coefficient between G and Q, so

[0106]

[0107] The CCA algorithm becomes a solution to an optimization problem, namely:

[0108] J = max(ρ) = max(A T cov(X * B ,Y B )C)

[0109] Among them A T cov(X * B ,X * B )A=C T cov(Y B ,Y B )C=1 is a constraint condition.

[0110] Let A = cov(X * B ,X *B ) -1 / 2 u,C=cov(Y B ,Y B ) -1 / 2 v, substituting it into the constraints, we get:

[0111] u T cov(X * B ,X * B ) -1 / 2 cov(X * B ,X * B )cov(X * B ,X * B ) -1 / 2 u=1

[0112] v T cov(Y B ,Y B ) -1 / 2 cov(Y B ,Y B )cov(Y B ,Y B ) -1 / 2 v=1

[0113] Further get: u T u=v T v=1.

[0114] Then the optimization objective function J can be converted to:

[0115]

[0116] where u T u=v T v=1 is a new constraint. * B ,X * B ) -1 / 2 cov(X * B ,Y B )cov(Y B ,Y B ) -1 / 2 Perform SVD decomposition to obtain

[0117] cov(X * B ,X * B ) -1 / 2cov(X * B ,Y B )cov(Y B ,Y B ) -1 / 2 =ΓΛΔ T

[0118] Finally, we can calculate the values ​​of A and B respectively.

[0119] A=cov(X * B ,X * B ) -1 / 2 Γ(:,1:κ)

[0120] C=cov(Y B ,Y B ) -1 / 2 Δ(:,1:κ)

[0121] Where rank(ΓΛΔ T )=κ.

[0122] From this we can see that:

[0123] G=cov(X * B ,X * B ) -1 / 2 Γ(:,1:κ) T X * B

[0124] Q=Bcov(Y B ,Y B ) -1 / 2 Δ(:,1:κ) T Y B

[0125] Construction T 2 Statistics are used to monitor faults that occur in the input subspace and are related to the output (quality-related faults), specifically T 2 The calculation formula of the statistic is as follows

[0126] T 2 =X * B T AA T X * B

[0127] Indicates T 2 The control limit of the statistic follows the F distribution and is calculated as

[0128]

[0129] where F 1-α (κ,N-κ) represents the critical value of the F distribution, where the degrees of freedom are κ and N-κ and the confidence level is 1-α.

[0130] When no quality-related failures occur, That is T 2 The statistic is below the control limit. When a quality-related failure occurs,

[0131] The above part realizes the fault detection related to the quality of the hot rolling process. Next, it is necessary to realize the spatiotemporal root cause diagnosis of the faults in the hot rolling process.

[0132] S4, when a fault occurs, the spatial position of the quality anomaly is located according to the fault detection results corresponding to each sub-block, and the time position of the fault is determined by reverse mapping the rolling position percentage of each stand;

[0133] It should be noted that, in this embodiment, when constructing the hot rolling process history database, a data alignment method based on the percentage of rolling position is adopted, such as Figure 4 As shown, the process data of each rack rolling in a batch is aligned according to the percentage of the rolling position. This data alignment method establishes a connection between the timestamp (time dimension) and the length of the rolled steel plate (space dimension). Based on this, once the quality-related fault detection method based on ImRMR-KCCA described in S3 above detects the occurrence of a fault, the timestamp corresponding to the occurrence of the fault can be reversely indexed by the percentage of the rolling position of the faulty rack when the fault occurs, and the length of the steel plate rolled by each rack at this time is summed. It can be obtained where (length) the quality fluctuation of the entire steel plate occurs, which is convenient for on-site operators to conduct targeted inspections on the quality of the steel plate, greatly saving the manpower, material and financial resources of the enterprise.

[0134] S5, calculate the size of the disproportionation coefficient of each process variable in the faulty rack, locate the root cause variable of the fault by comparing the size of the disproportionation coefficient of the process variable in the faulty rack, determine the root cause variable causing the fault, further locate the faulty equipment, and then realize the spatiotemporal root cause diagnosis of the fault.

[0135] It should be noted that after monitoring the occurrence of quality abnormality-related faults, it is necessary to locate the root variable that causes the fault. To this end, it is necessary to calculate the disproportionation coefficient of each process variable in the faulty rack to perform fault separation, that is, to compare the similarity between the real-time data of the fault and the normal data.

[0136] The calculation formula for the disproportionation coefficient of the process variable in the faulty rack is:

[0137]

[0138] Where n is the number of sample points. The larger the disproportionation coefficient of the process variable in the faulty rack, the more likely this process variable is to be the root variable causing the fault. The root variable causing the fault can be located by comparing the disproportionation coefficients of each process variable.

[0139] According to the above hot rolling process quality related fault detection and spatiotemporal root cause diagnosis model and the actual production data of the hot rolling process, the results obtained are as follows: Figure 5 and Figure 6 shown.

[0140] Next, we will analyze in detail Figure 5 The experimental results show that Figure 5 As shown in (c) in FIG. 1 , this figure shows the fault detection results of the rough rolling process stand and the fault detection results of the first, second, fifth, sixth, and seventh stands in the finishing process. Figure 5 As shown in (a) and (b), the fault started in the third stand during the finishing process. The third stand failed in the interval [16.7%, 50%]. Through the 16.7% rolling position percentage, the corresponding time and space dimension information can be obtained by reverse indexing, namely, Time position =f(Space position ), where Space position Indicates the spatial location point. From the time point of view, using the above formula, it is calculated that the fault occurred at 1:07:22.22. By establishing the time dimension monitoring model of the third rack, it is not difficult to find that the fault did occur at this time point. Figure 6 As shown in (a), "0" represents normal and "1" represents fault. From a spatial perspective, the length of the steel plate rolled by the seven stands in the finishing rolling process before this time point is calculated by accumulating and summing up, that is, length i It represents the length of the steel plate that has been rolled by the i-th stand at a certain point in time. Using the above formula, it is calculated that the fault occurs at 136912.56mm of the steel plate. From this position onwards, the quality of the steel plate will fluctuate.

[0141] At the same time, since the fault of the third stand in the finishing rolling process propagates downward and causes the fourth stand in the finishing rolling process to fail, the fourth stand can use the same method to reversely index the time information of the fault based on the percentage of the rolling position (spatial information) when the fault occurs and the above formula. That is, from a time point of view, the fault of the fourth stand occurred at 1:07:25.02, and the time when the fault of the fourth stand occurred was obviously later than the time when the fault of the third stand occurred, which is in line with the process characteristics of the hot rolling process. At the same time, by establishing the time dimension monitoring model of the fourth stand, it is obvious that the fault did occur at this point in time, such as Figure 6 As shown in (b), "0" represents normal and "1" represents fault. Since the third rack has already failed, that is, the steel plate has already experienced quality fluctuations at 1:07:22.22, there is no need to analyze and discuss the fault space dimension at the product level for the lagging fourth rack.

[0142] contrast Figure 5 The information provided by each sub-graph in the figure shows that according to the fault detection results in the previous part, the fault originates from the third stand during the finishing rolling process, and the root cause variables of the fault of the third stand need to be separated. By calculating the disproportionate coefficients of the four process variables of the third stand rolling force, bending roll force, roll gap, and roll speed, the disproportionate coefficients of the third stand rolling force are 0.513, the disproportionate coefficients of the third stand bending roll force are 0.0179, the disproportionate coefficients of the third stand roll gap are 0.7848, and the disproportionate coefficients of the third stand roll speed are 0.33261. The disproportionate coefficient of the third stand roll gap is the largest. The results are as follows Figure 6 By comparing the disproportionation coefficients of the above four variables, it can be concluded that the root variable causing the failure is the third stand of the finishing process.

[0143] In summary, this embodiment provides a spatiotemporal root cause diagnosis method for quality anomalies in the hot rolling process driven by knowledge and data. This method not only realizes fault location in the time dimension, but also realizes fault location in the space dimension. The fault location in the space dimension includes two aspects. On the one hand, the fault location at the equipment level is completed, that is, the production unit where the fault occurs is located. On the other hand, the fault location at the product level is completed, that is, the detailed location of the product quality fluctuation occurs. Such a spatiotemporal root cause diagnosis method is conducive to enterprises to locate the root cause of the fault in a timely manner and take corresponding optimization and control measures. At the same time, timely detection of whether the products produced have quality problems can avoid defective products from entering the market and damaging the corporate image. Therefore, it is of great help to improve the economic value and brand value of the enterprise.

[0144] In addition, it should be noted that the present invention can be provided as a method, an apparatus or a computer program product. Therefore, the embodiments of the present invention can take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, the embodiments of the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code.

[0145] The embodiments of the present invention are described with reference to the flowcharts and / or block diagrams of the methods, terminal devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, an embedded processor, or other programmable data processing terminal device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0146] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device, so that a series of operation steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable terminal device provide for implementing the process in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0147] It should also be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "include", "comprises" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal device. In the absence of further restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, article or terminal device including the elements.

[0148] Finally, it should be noted that the above is a preferred embodiment of the present invention. It should be pointed out that although the preferred embodiment of the present invention has been described, for those skilled in the art, once the basic creative concept of the present invention is known, several improvements and modifications can be made without departing from the principles of the present invention, and these improvements and modifications should also be regarded as the protection scope of the present invention. Therefore, the attached claims are intended to be interpreted as including the preferred embodiment and all changes and modifications that fall within the scope of the embodiments of the present invention.

Claims

1. A method for diagnosing the spatiotemporal root causes of quality anomalies in hot rolling process, characterized in that: include: Obtain historical data samples of hot rolling process and establish a historical database of hot rolling process; The acquired data samples are aligned according to the percentage of rolling position, and the entire hot rolling process is divided into multiple levels and sub-blocks; each sub-block corresponds to a stand in the rolling process; Constructing a fault detection model corresponding to each sub-block respectively, and training each constructed fault detection model based on the hot rolling process history database, and using the trained fault detection model to achieve real-time detection of quality-related faults of each sub-block; When a fault occurs, the spatial position of the quality anomaly is located according to the fault detection results corresponding to each sub-block, and the time position of the fault is determined by reverse mapping the rolling position percentage of each rack; Calculate the disproportionation coefficient of each process variable in the faulty rack, locate the root cause variable of the fault, determine the root cause variable causing the fault, locate the faulty equipment, and realize the spatiotemporal root cause diagnosis of the fault; The variables of hot rolling process include process variables and quality variables; The fault detection models corresponding to the sub-blocks are constructed respectively, and the constructed fault detection models are trained based on the hot rolling process history database, and the trained fault detection models are used to realize real-time detection of quality-related faults of each sub-block, including: The maximum information coefficient is used to replace the mutual information in the maximum relevance minimum redundancy algorithm to obtain an improved maximum relevance minimum redundancy algorithm. For the process variables and quality variables of the hot rolling process corresponding to each sub-block, the quality-related process variables corresponding to each sub-block are screened out by using the improved maximum relevance minimum redundancy algorithm; The kernel function is introduced into the canonical correlation analysis to obtain the kernel canonical correlation analysis model; The quality-related process variables and corresponding quality variables selected from each sub-block are sent to the core canonical correlation analysis model to perform quality-related process monitoring; The calculation formula of the disproportionation coefficient is: Where difference(i) represents the disproportionation coefficient of the i-th process variable, n is the number of sample points, represents the fault dataset, represents a normal data set; The larger the disproportionation coefficient of the process variable in the faulty rack is, the more likely this process variable is to be the root variable causing the current fault.

2. The method for diagnosing the temporal and spatial root causes of quality anomalies in hot rolling process according to claim 1, characterized in that: The entire hot rolling process is divided into multiple levels and sub-blocks, including: From a vertical perspective, the entire hot rolling process is divided into product layer, full process layer, subsystem layer and equipment layer. From a horizontal perspective, the entire hot rolling process is divided into the first sub-block, the second sub-block, the third sub-block, the fourth sub-block, the fifth sub-block, the sixth sub-block, the seventh sub-block and the eighth sub-block. Among them, the first sub-block corresponds to the stand in the rough rolling process, and the second to eighth sub-blocks correspond to the first to seventh stands in the finishing rolling process respectively.

3. The method for diagnosing the temporal and spatial root causes of quality anomalies in hot rolling process according to claim 1, characterized in that: The data samples acquired are aligned according to the percentage of rolling position, including: positive alignment of data from time dimension to space dimension, and negative alignment of data from space dimension to time dimension; The implementation process of data alignment from the time dimension to the space dimension is as follows: Read the process data of a slab rolling process in chronological order and store it in the corresponding vector; The vectors corresponding to each rack are normalized, and the length of the steel plate rolled at each moment is divided by the total length of the steel plate rolled by the rack to obtain the percentage of the steel rolling position of each rack; The obtained steel rolling position percentage data of each rack is expanded to construct an index column, the index column is constructed in the first column, and the second column is the steel rolling position percentage data of each rack; Obtain the data length of the rolling position percentage data of each stand, and compare to obtain the longest data length; Rewrite and splice the index columns of the steel rolling position percentage of each rack after the index column is constructed, and rewrite according to the steel rolling percentage position based on the steel rolling position percentage of the longest data length; when the steel rolling percentage position is not greater than the reference steel rolling percentage position, write the index of the same row, otherwise take the index of the next row; splice in sequence to form the index matrix of each rack data and outlet temperature and actual thickness; The process data is re-read according to the constructed index matrix to obtain the percentage data of the rolling position of each rack after data alignment, and the different data are rewritten into column vectors with the number of rows being the long data length; For the data anti-alignment from the space dimension to the time dimension, the time information is obtained through the reverse index of the rolling position percentage of each rack to achieve the data anti-alignment from the space dimension to the time dimension.

4. The method for diagnosing the temporal and spatial root causes of quality anomalies in hot rolling process according to claim 1, characterized in that: The process variables and corresponding quality variables screened out from each sub-block are sent to the core canonical correlation analysis model to perform quality-related process monitoring, including: The process variables screened out from each sub-block are mapped into a high-dimensional space through nonlinear mapping; The mapped process variables are used to conduct canonical correlation analysis with the corresponding quality variables to determine the statistics and control limits corresponding to each sub-block and conduct quality-related process monitoring.

5. The method for diagnosing the temporal and spatial root causes of quality anomalies in hot rolling process according to claim 4, characterized in that: The process variables screened by each sub-block are mapped to a high-dimensional space through nonlinear mapping, including: The screened quality-related process variables are mapped into a feature space using a kernel function; wherein the kernel canonical correlation analysis model combines a polynomial kernel with a radial basis function kernel.

6. The method for diagnosing the temporal and spatial root causes of quality anomalies in hot rolling process according to claim 1, characterized in that: The method locates the spatial position of the quality anomaly according to the fault detection results corresponding to each sub-block, and determines the time position of the fault by reverse mapping the rolling position percentage of each rack, including: According to the fault detection results corresponding to each sub-block, the spatial position of the quality anomaly is located, and the timestamp corresponding to the fault is reversely indexed through the rolling position percentage of the faulty frame when the fault occurs, the time position of the fault is determined, and the length of the steel plate rolled by each frame at this time is summed up. Combined with the rolling position percentage of the faulty frame when the fault occurs, it is obtained where the quality fluctuation of the entire steel plate occurs.

Images