A Control Chart Pattern Recognition Method Based on Weighted Ordered Patterns and Ensemble Classifiers

Abstract

The control chart pattern recognition method based on weighted ordered patterns and ensemble classifiers includes: Step 1: Collecting partial data of the controlled process in actual manufacturing, and generating eight abnormal control chart patterns (CCPs) using the Monte Carlo method; Step 2: For CCP observation data of a given length, reconstructing it into a symbolic phase space according to a specific parameter combination; Step 3: Obtaining all ordered pattern types based on the embedding dimension parameter, and calculating the ordered pattern features of the identified CCPs; Step 4: Generating weighted ordered pattern (WOP) features, calculating the amplitude features of the identified CCPs, and weighting the ordered pattern features from Step 3 to obtain WOP features; Step 5: Changing the time delay parameter to obtain WOPs corresponding to different time delay parameters, constructing a homogeneous classifier with the same number of time delay parameters, and performing individual classifier classification; Step 6: Based on a democratic voting method, voting on the recognition results of multiple classifiers in Step 5 to obtain the comprehensive diagnostic result of the identified CCPs.

Description

Technical Field

[0001] This invention relates to a method for identifying abnormal states in industrial manufacturing processes, and more specifically to a method for using process data or product quality data from the manufacturing process, combined with weighted ordered pattern features and an integrated classifier method, to perform control chart pattern recognition in order to determine whether the monitored manufacturing process is under control and the type of out-of-control mode, thereby achieving quality control. Background Technology

[0002] In industrial environments, the stability of the manufacturing process is a key factor in ensuring product quality. Instability and abnormal patterns in the manufacturing process can cause processing defects, resulting in rough surfaces, workpiece dimensional errors, and product or machine structural deformation. This can affect processing efficiency and product quality, causing serious internal and external failures, and ultimately significantly increasing production costs.

[0003] Statistical Process Control (SPC) is an effective method for detecting process changes and is used to monitor and control product quality in manufacturing processes. Control charts are one of the commonly used statistical tools in SPC. Traditional control chart methods are only used to identify whether a process exceeds limits, but cannot identify abnormal patterns in the process. When a manufacturing process is only affected by random causes, the process is considered controlled and will appear as a normal pattern on the control chart. However, unstable processes will produce unnatural abnormal control chart patterns (CCPs), such as upward trends or cyclical patterns. When dealing with continuously monitored variables, these abnormal patterns are associated with definite causes. By analyzing and identifying the corresponding patterns, the causes of abnormalities in the manufacturing process can be diagnosed. Early detection of abnormal patterns and their causes can enable timely prevention of problems and process improvements, thereby reducing production losses.

[0004] Control chart pattern recognition (CCPR) methods utilize machine learning models for anomaly detection and quality control in manufacturing processes. Model input includes both raw data and features. Raw data often results in excessively high model dimensionality and computational cost, and its recognition accuracy is lower than that of feature-based methods. Feature-based methods mainly include statistical features, shape features, and wavelet features. However, these traditional features typically only focus on the magnitude and overall information of a time series of data, ignoring the structural, sequential relationships, and detailed fluctuation information between data points. Therefore, there is a need in this field for a method that can effectively improve the accuracy of control chart pattern recognition. Summary of the Invention

[0005] To address the aforementioned issues, embodiments of the present invention provide a control chart pattern recognition method based on weighted ordered pattern (WOP) features and an ensemble classifier for monitoring abnormal states in the manufacturing process.

[0006] According to an embodiment of the present invention, a control chart pattern recognition method based on weighted ordered patterns and an ensemble classifier is provided, characterized by the following steps: Step 1: Data acquisition and generation: Collect partial data of the controlled process in the actual manufacturing process, and generate eight abnormal control chart patterns (CCPs) using the Monte Carlo method based on this partial data; Step 2: Construct symbolic phase space: For CCP observation data of a given length, reconstruct it into a symbolic phase space according to a specific parameter combination; Step 3: Extract ordered pattern features: Obtain all ordered pattern types according to the embedding dimension parameters, and calculate the ordered pattern of the identified CCPs. Features; Step 4: Generate Weighted Ordered Pattern (WOP) Features: Calculate the amplitude features of the identified CCP and weight the ordered pattern features from Step 3 to obtain WOP features; Step 5: Generate Multi-Delay Weighted Ordered Pattern (WOP) Features and Ensemble Classifier Method: By changing the size of the delay parameter, obtain WOPs corresponding to different delay parameters, and construct a homogeneous classifier with the same number of delay parameters for individual classifier classification; Step 6: Comprehensive Diagnosis Based on Democratic Voting: Using a democratic voting method, vote on the identification results of multiple classifiers from Step 5 to finally obtain the comprehensive diagnostic result of the identified CCP.

[0007] In an optional implementation, the actual manufacturing process described in step one includes machining, assembly, and break-in processes.

[0008] In an optional implementation, some of the data from the controlled process described in step one includes the normal dimensional data of the product in the machining process and the normal coating thickness data of the product in the spraying process.

[0009] In an optional implementation, the CCPs of the eight modes mentioned in step one refer to the CCPs corresponding to the eight modes, including normal (NOR) mode, cyclic (CYC) mode, uptrend (UT) mode, downtrend (DT) mode, upstep (US) mode, downstep (DS) mode, stratification (STR) mode, and system (SYS) mode. The NOR mode indicates that the process is only affected by random disturbances and is in a controlled state. The other seven modes are abnormal modes, that is, the manufacturing process is in an uncontrolled state, corresponding to specific abnormal causes.

[0010] In an optional implementation, the generation of eight CCP modes using the Monte Carlo method in step one is based on data distribution and obtained through Monte Carlo simulation, specifically including:

[0011] Based on partial data collected from the normal mode of the controlled process in actual manufacturing, the mean μ and standard deviation σ of this partial data are calculated. The expressions for each simulation CCP are as follows:

[0012] y t =μ+x t +d t

[0013] Among them, y t Let represent the monitored values ​​of the manufacturing process at time t, μ represent the mean of the controlled process, and x represent the measured values. t =r t ×σ represents random disturbances during the process, which follow a normal distribution. t ~N(0,σ),r t Let d be a random variable at time t that follows a standard normal distribution. t Used to simulate specific abnormal patterns, where d t =0 indicates a NOR mode subject only to random interference.

[0014] In an optional implementation, the parameter combination mentioned in step two mainly refers to the two parameters m and τ used to construct the ordered pattern features. m is called the embedding dimension parameter, i.e., the data dimension of each ordered pattern. τ is called the time delay parameter, i.e., the sampling interval when constructing the ordered pattern. τ can reflect the inherent scale information of the time series to a certain extent, and can reflect the fluctuation structure of the time series at different scales. The range of τ is affected by the parameter m and the CCP data length N. To ensure that at least one complete ordered pattern is extracted at a specific scale, the maximum value of τ is τ. max Satisfy the following formula:

[0015] N = 1 + (m - 1)τ max

[0016] Wherein, the range of τ should satisfy τ∈[1, τ max ], round indicates rounding down;

[0017] The symbolic phase space mentioned in step two is a method that uses parameters m and τ to convert CCP data of length N into a symbolic phase space represented by multiple ordered patterns. The specific operation is as follows:

[0018] Given a CCP data sequence of length N, Y = {y1, y2, ..., y...} N Based on parameters m and τ, the sequence is divided into n subsequences of length m.

[0019]

[0020] Where y1, y2, ..., y N For data points in CCP;

[0021] subsequence The original phase space matrix Sm The i-th subsequence in the data, i = 1, 2, ..., n, and n is determined by parameters m and τ and data length N, satisfying n = N - (m - 1)τ;

[0022] The ordered pattern consists of an m-dimensional tuple [r1, r2, ..., rm] containing m positive integers [1, 2, ..., m], where each number appears only once. j ,...,r m The ] indicates that the subsequence is sorted in ascending order based on the size of its elements, and the subsequence is... Convert to the corresponding ordered mode

[0023]

[0024] In the formula, r j Let the integers be positive integers [1, m], j = 1, 2, ..., m, r j Representing a subsequence The positions of the data points after sorting. Representing a subsequence The ordered pattern corresponding to a specific embedding dimension parameter m and a time delay parameter τ, when the subsequence When there are equivalent elements in the data, Let the ordering relationship between the two satisfy r l <r l+1 , where r l This represents the position of a numerical point in a subsequence, l = 1, 2, ..., m-1; these ordered patterns The set of symbols is called the symbolic phase space matrix of the CCP, denoted as π. m .

[0025] In an optional implementation, the ordered patterns described in step three are all possible data fluctuation patterns obtained based on the embedding dimension parameter m. Each ordered pattern consists of m positive integers from 1 to m, and each positive integer appears only once. Based on the sorting order of the m positive integers, parameter m corresponds to a total of m! ordered patterns. The set of all types of ordered patterns is denoted as […]. Each ordered pattern is denoted as Π m (k), where k represents the type index of the ordered pattern, k = 1, 2, ..., m!;

[0026] The ordered pattern feature mentioned in step three refers to the symbol phase space matrix π of the statistical CCP. m All types of ordered patterns Π m The probability of (k) occurring is used as a feature vector, also known as an ordered pattern distribution. Its calculation steps are as follows:

[0027] According to step two, the original phase space matrix S of CCP m Each row vector in the matrix is ​​mapped to the symbolic phase space matrix π. m The only ordered pattern in S m There are n row vectors, corresponding to n ordered patterns, and different row vectors may correspond to the same ordered pattern;

[0028] Statistical ordered pattern set Π m Each type of ordered pattern Π m The probability of (k) occurring is used as an ordered pattern feature, and the calculation formula is as follows:

[0029]

[0030] In the formula, m is the embedding dimension parameter, τ is the time delay parameter, and n = N - (m - 1)τ represents the original phase space S. m The number of neutron sequences, i.e., the number of row vectors. S represents m The row vector of the i-th row, i = 1, 2, ..., n, I A(u) The indicator function for set A is used to statistically analyze the original phase space S. m The number of ordered patterns is specified in the table, represented as:

[0031]

[0032] in, Representing a subsequence Corresponding ordered mode With ordered pattern set Π m Π in m (k) are the same, where k represents the index of the ordered pattern, k = 1, 2, ..., m! Representing a subsequence Corresponding ordered mode Belongs to the ordered pattern set Π m ;

[0033] The above steps are used to calculate the identified CCP data within the ordered pattern set Π. m The probability of all types of ordered patterns appearing in the CCP can be used to obtain the ordered pattern feature OP(m, τ) corresponding to the CCP. OP(m, τ) is a vector containing the probability of each type of ordered pattern.

[0034]

[0035] In the formula, Let represent the probability of the k-th ordered pattern feature under parameters m and τ, where k = 1, 2, ..., m!.

[0036] In an optional implementation, the weighted ordered pattern (WOP) feature mentioned in step four is obtained by weighting the ordered pattern feature OP(m, τ) with the amplitude information of the CCP data to obtain a new feature WOP that simultaneously contains the amplitude information of the CCP data and the order information between the data. The calculation formula is as follows:

[0037]

[0038] In the formula, WOP(m, τ) represents the weighted ordered pattern (WOP) feature of CCP, which is weighted by the magnitude feature of CCP, and w represents the magnitude feature of CCP. Let w represent the eigenvalues ​​of the k-th class of WOP features under parameters m and τ, where k = 1, 2, ..., m!, and w is represented by the variance of the CCP data, as shown below:

[0039]

[0040]

[0041] In the formula, y represents the mean of all data points in CCP. q Let q represent the data points in the CCP, where q = 1, 2, ..., N, and N represents the total number of samples in the CCP.

[0042] In an optional implementation, the multi-delay WOP features mentioned in step five are multiple sets of WOP features obtained by changing the time delay parameter τ. When the embedding dimension parameter m is determined, the process information represented by CCP can be observed from different scales by changing different time delay parameters τ.

[0043] In an optional implementation, the ensemble classifier method described in step five utilizes multiple individual classifiers to diagnose the identified CCP. The ensemble classifier is divided into homogeneous and heterogeneous classifiers. A homogeneous classifier means all individual classifiers belong to the same category, while a heterogeneous classifier means all individual classifiers belong to different categories. In the ensemble classifier method, each individual classifier derives a corresponding recognition result based on its own input data or features. The homogeneous classifier method is used for pattern recognition. For each WOP feature in the multi-delay WOP feature set, one individual classifier is matched for diagnosis, resulting in a total of τ. max Each individual classifier, and each CCP, will obtain τ. max The recognition result V h h = 1, 2, ..., τ max .

[0044] The method provided by the above embodiments of the present invention utilizes product quality or process data collected during partially controlled manufacturing processes, combined with control chart pattern recognition technology, to establish a monitoring model for abnormal states in the product manufacturing process. The proposed WOP feature is characterized by simple computation, strong noise resistance, and high sensitivity. Employing a time series analysis method based on ordered patterns, the sequential relationships and inherent fluctuations between CCP data can be identified, reflecting the underlying information of the data. Weighting the ordered pattern features using data amplitude yields WOP features that integrate CCP amplitude and sequence information under specific parameters. Based on WOP features with multiple time delay parameters, the CCP signal is analyzed using time delay parameters, fully extracting observation information at different scales from the CCP data. An ensemble classifier method based on multiple time delay parameters and a democratic voting method are proposed to fuse and diagnose the identification results of multiple sets of WOP features, which helps improve the accuracy of the model. The proposed method can achieve early anomaly identification in the manufacturing process, assisting on-site personnel in taking preventative measures and reducing the generation of defective products.

[0045] The embodiments of the present invention include at least the following advantages. The present invention employs a product manufacturing process anomaly monitoring method based on multi-delay WOP features. It unfolds and analyzes CCP data at different scales, extracting the sequential relationships and structural features between CCP data under different delay parameters. This method is characterized by strong noise resistance, simple computation, and sensitivity to minute changes in CCPs. Since WOP features with different delay parameters have varying sensitivities to different types of CCPs, the concept of ensemble learning is used to assign corresponding individual classifiers to WOP features with different delay parameters to obtain multiple identification results. These multiple identification results are then fused for diagnosis, further improving the accuracy of CCP diagnosis. Furthermore, the proposed method is applicable to manufacturing process data of different lengths and sample sizes. It can accurately identify CCP types even with small sample conditions, exhibiting strong transferability and strong applicability in various industrial scenarios. The method has a certain degree of openness in practical applications, suitable for real-time status monitoring of product quality data or manufacturing process data, as well as early warning of anomalies. Attached Figure Description

[0046] The foregoing features of the present invention can be more readily understood in conjunction with the accompanying drawings and the following detailed description, wherein:

[0047] Figure 1 A flowchart of a control graph pattern recognition method based on ordered patterns and an ensemble classifier according to an embodiment of the present invention is shown.

[0048] Figure 2 A flowchart of a comprehensive diagnostic step based on an ensemble classifier according to an embodiment of the present invention is shown;

[0049] Figure 3 Eight CCPs are shown in an example of applying the method according to an embodiment of the present invention, obtained from controlled data of the thickness of a galvanized metal sheet in a coating process.

[0050] Figure 4 An example of an ordered pattern feature map (where m = 4, τ = 15) calculated by eight CCPs is shown in an example of applying the method according to an embodiment of the present invention;

[0051] Figure 5 The following is an example of a WOP feature map based on multiple delay parameters calculated using the UT mode (where m = 4, τ = 1-21) obtained by applying the method according to an embodiment of the present invention.

[0052] Figure 6 The illustration shows the diagnostic rates of 21 homogeneous individual classifiers and the overall diagnostic rate obtained from democratic voting in an example of applying the method according to an embodiment of the present invention.

[0053] Figure 7 The diagram illustrates a confusion matrix of identification results for all unidentified samples after a comprehensive diagnosis following democratic voting, in an example of applying the method according to an embodiment of the present invention.

[0054] The serial numbers, symbols, and codes in the diagram are explained as follows:

[0055] Y: Represents a control chart pattern data based on the manufacturing process.

[0056] N: Represents the length of the data

[0057] m: A parameter representing the ordered pattern feature, indicating the embedding dimension.

[0058] τ: A parameter representing the ordered pattern characteristic, indicating the time delay.

[0059] τ max : Indicates the maximum possible time delay parameter value.

[0060] w: represents the weighting values ​​needed to calculate the weighted ordered pattern.

[0061] The probability values ​​of the k-th ordered pattern features in the CCP data corresponding to parameters m and τ

[0062] The k-th weighted ordered pattern features of CCP data corresponding to parameters m and τ

[0063] m! -dim: Indicates that the WOP feature dimension is m!.

[0064] CCP: Represents the control chart pattern obtained from the manufacturing process.

[0065] WOP: Represents the weighted ordered pattern feature of the calculated control chart pattern.

[0066] V1: Represents the recognition result of the first individual classifier when parameter τ = 1.

[0067] V2: Represents the recognition result of the second individual classifier when parameter τ = 2.

[0068] The parameter τ = τ max τ max The recognition results of the individual classifier

[0069] NOR: Indicates normal mode in CCP.

[0070] CYC: Represents the cyclic mode in CCP.

[0071] UT: Indicates an upward trend pattern in CCP.

[0072] DT: Indicates a downward trend pattern in CCP.

[0073] US: Indicates the ascending step mode in CCP.

[0074] DS: Represents the descent step mode in CCP.

[0075] STR: Represents the hierarchical pattern in CCP.

[0076] SYS: Represents the system mode in CCP. Detailed Implementation

[0077] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey its scope to those skilled in the art. It should be noted that, unless otherwise stated, the technical or scientific terms used in this invention should be understood in their ordinary meaning as understood by those skilled in the art. Specific embodiments of the invention are described in detail in conjunction with the accompanying drawings, but the invention is not limited to these specific embodiments.

[0078] Weighted Ordered Pattern (WOP) features reflect the structural order and amplitude information of a time series under a single time delay parameter, while multi-time delay WOP features integrate useful features of time series under multiple time delay parameters, deconstructing the original time series at different scales to more comprehensively characterize the state changes of the manufacturing process. Multi-time delay WOP features directly apply to the original time series without data transformation, exhibiting theoretical simplicity and strong noise resistance. Employing an ensemble classifier-based approach, utilizing the same number of homogeneous classifiers corresponding to the time delay parameters for fusion diagnosis significantly improves the ability to identify abnormal patterns.

[0079] Therefore, process or quality data generated during product manufacturing is recorded, and control charts are generated from data within a fixed-length window to obtain WOP features under specific parameter combinations. Changing the time delay parameters allows for the acquisition of WOP features at different observation scales, generating multi-time-delay WOP features. Based on this, an ensemble classifier method is used, employing a homogeneous classifier with the same number of time delay parameters. The WOP features under each time delay parameter are used as input to the corresponding classifier, resulting in multiple recognition results. Finally, a democratic voting-based fusion diagnosis is performed to ultimately achieve Control Chart Pattern Recognition (CCPR). Based on this, the embodiments of the present invention provide a control chart pattern recognition method based on weighted ordered patterns and an ensemble classifier. According to the embodiments of the present invention, the control chart pattern recognition method based on weighted ordered patterns and an ensemble classifier reconstructs the original data in the control chart into a phase space matrix using specific parameter combinations and a phase space reconstruction method. It then converts the data into a symbolic phase space matrix based on the ordering relationship of the values ​​in each row vector of the phase space matrix and counts the number of various ordered patterns in the matrix. Based on this, the amplitude features of the original data in the control chart are used for weighting to obtain WOP features containing both data amplitude and order information. WOP features under different time delay parameters are calculated to obtain multi-time delay WOP features. Since the representational ability of control chart patterns varies significantly across different observation scales, multiple time delay parameters can expand the original data at different scales to extract richer manufacturing process information. An ensemble classifier recognition method based on multiple time delay parameters is proposed. The WOP features under each experimental parameter are input into the corresponding classifier, and a recognition result is output. A democratic voting-based method is used to fuse the outputs of multiple classifiers according to the "majority rule" principle for diagnosis. Based on the diagnostic results, it is possible to determine whether the current manufacturing process is under control and the causes of abnormalities (uncontrolled processes), providing a technical basis for subsequent early warning, preventative measures, and avoiding defective products.

[0080] The following describes in detail, with reference to the accompanying drawings, a control graph pattern recognition method based on ordered patterns and an ensemble classifier according to an embodiment of the present invention. Figure 1A flowchart of a control graph pattern recognition method based on ordered patterns and an ensemble classifier, according to an embodiment of the present invention, is shown. Figure 2 A flowchart illustrating a comprehensive diagnostic process based on an ensemble classifier, according to an embodiment of the present invention, is shown. (Refer to...) Figure 1-2 According to an embodiment of the present invention, a control chart pattern recognition method based on weighted ordered patterns and an ensemble classifier is proposed, comprising the following steps: Step 1: Data acquisition and generation: Collect partial data of the controlled process in actual manufacturing, and generate control chart pattern recognition (CCP) of 8 patterns using the Monte Carlo method based on this partial data; Step 2: Construct symbolic phase space: For CCP observation data of a given length, reconstruct it into a symbolic phase space according to a specific parameter combination; Step 3: Extract ordered pattern features: Obtain all ordered pattern types according to the embedding dimension parameter, and calculate the ordered pattern features of the identified CCP; Step 4: Generate Weighted Ordered Pattern (WOP) Features: Calculate the amplitude features of the identified CCP and weight the ordered pattern features from Step 3 to obtain WOP features; Step 5: Generate Weighted Ordered Pattern (WOP) Features Based on Multiple Delay Parameters and an Ensemble Classifier Method: By changing the size of the delay parameters, obtain WOPs corresponding to different delay parameters, and construct a homogeneous classifier with the same number of delay parameters for individual classifier classification; Step 6: Comprehensive Diagnosis Based on Democratic Voting: Using a democratic voting method, vote on the identification results of multiple classifiers from Step 5 to finally obtain the comprehensive diagnostic result of the identified CCP.

[0081] Optionally, the actual manufacturing process described in step one includes production activities related to product manufacturing and from which process or quality data can be obtained, such as, but not limited to, machining processes, assembly processes, and break-in processes.

[0082] Optionally, some of the data from the controlled process described in step one includes data collected when the product is in a controlled state during the manufacturing process, indicating that the product has no quality problems. For example, normal dimensional data of the product in the machining process, normal coating thickness data of the product in the spraying process, etc.

[0083] Optionally, the eight CCP modes mentioned in step one refer to the CCPs corresponding to the eight modes: Normal (NOR), Cyclic (CYC), Uptrend (UT), Downtrend (DT), Upward Step (US), Downward Step (DS), Stratified (STR), and System (SYS). The NOR mode indicates that the process is only affected by random disturbances and is under control. The other seven modes are abnormal modes, meaning the manufacturing process is in an uncontrolled state, corresponding to specific abnormal causes. For example, trend modes are often related to equipment wear and tear, personnel fatigue, etc.; step modes are often related to material replacement, personnel turnover, equipment maintenance, etc. A CCP of length N is represented as a one-dimensional time series of length N.

[0084] Optionally, the generation of eight CCP models using the Monte Carlo method described in step one involves generating CCPs related to various anomalies using Monte Carlo simulation. Since collecting all possible anomaly data on-site is time-consuming, difficult, and costly, Monte Carlo simulation is often used to obtain various CCPs based on data distribution.

[0085] Based on the obtained data from the normal mode, the mean μ and standard deviation σ of this data are calculated. The expressions for each simulation CCP are as follows:

[0086] y t =μ+x t +d t

[0087] Among them, y t Let x represent the monitored values ​​of the manufacturing process at time t, and μ represent the mean of the controlled process. t =r t ×σ represents random disturbances during the process, which follow a normal distribution. t ~N(0,σ),r t Let d be a random variable at time t that follows a standard normal distribution. t Used to simulate specific abnormal patterns, where d t =0 indicates a NOR mode subject only to random disturbances. Table 1 summarizes the relevant formulas and parameter ranges for the eight CCPs obtained from the above equation.

[0088] Table 1. Formulas and parameter settings for generating CCP simulation data

[0089]

[0090]

[0091] Where N represents the number of sample points in the CCP. v represents the step position, and its specific expression is as follows:

[0092]

[0093] Based on the above formula, eight types of basic CCP data can be generated. When generating CCPs, except for the mean μ and standard deviation σ, which are determined by the actual process, the other parameters are randomly selected within a certain range to simulate the randomness of the actual process as much as possible.

[0094] Optionally, the parameter combination described in step two refers to the two parameters m and τ used to construct the ordered pattern features. m is called the embedding dimension parameter, i.e., the data dimension of each ordered pattern. In an optional implementation, m can be set in the range [3, 6]. τ is called the time delay parameter, i.e., the sampling interval when constructing the ordered pattern. τ can reflect the inherent scale information of the time series to a certain extent, and can reflect the fluctuation structure of the time series at different scales. Furthermore, the range of τ is affected by the parameter m and the CCP data length N. To ensure that at least one complete ordered pattern is extracted at a specific scale, the maximum value of τ is τ. max The following formula should be satisfied:

[0095] N = 1 + (m - 1)τ max

[0096] Therefore, the range of τ should satisfy τ∈[1, τ] max ], round indicates rounding down to the nearest integer.

[0097] Optionally, the symbolic phase space mentioned in step two refers to the method of converting CCP data of length N into a symbolic phase space represented by multiple ordered patterns using parameters m and τ. The specific operation is as follows:

[0098] Given a CCP data sequence of length N, Y = {y1, y2, ..., y...} N Based on parameters m and τ, the sequence is divided into n subsequences of length m. y1, y2, ..., y N For data points in CCP. Subsequence The original phase space matrix S m The i-th subsequence in the array, i = 1, 2, ..., n. In this implementation, n is determined by parameters m and τ and the data length N, satisfying n = N - (m - 1)τ.

[0099] The ordered pattern consists of an m-dimensional tuple [r1, r2, ..., rm] containing m positive integers [1, 2, ..., m], where each number appears only once. j ,...,r m The subsequence is sorted in ascending order based on the size of its elements. Can be converted into the corresponding ordered pattern

[0100]

[0101] In the formula, r j Belonging to positive integers [1, m], j = 1, 2, ..., m, r j Representing a subsequence The position of the data points after sorting. Representing a subsequence The ordered patterns corresponding to specific embedding dimension parameter m and time delay parameter τ. To ensure the uniqueness of ordered patterns corresponding to equal-value sequences, i.e. Given a sequence containing m positive integers [1, 2, ..., m], where each number appears only once. When the subsequence... When there are equivalent elements in the middle Let the ordering relationship between the two satisfy r l <r l+1 Among them, r l This represents the position of a numerical point in a subsequence, where l = 1, 2, ..., m-1.

[0102] For example, for subsequences The elements in the sequence satisfy y2≤y3≤y1≤y4, that is, r1=2, r2=3, r3=1, r4=4, therefore it is an ordered pattern. as follows:

[0103]

[0104] Subsequences in the original phase space of CCP Convert to the corresponding ordered pattern The set of these ordered patterns is called the symbolic phase space matrix of the CCP, denoted as π. m .

[0105] Step two above converts the data in the CCP into a symbolic phase space based on ordered patterns, which facilitates the subsequent calculation of ordered pattern features.

[0106] Optionally, the ordered patterns described in step three are all possible data fluctuation patterns obtained based on the embedding dimension parameter m. Each ordered pattern consists of m positive integers from 1 to m, and each positive integer appears only once. Based on the sorting method of the m positive integers, parameter m corresponds to m! ordered patterns, and the set of all types of ordered patterns is denoted as [equation missing]. Each ordered pattern is denoted as Π m(k), where k represents the type index of the ordered pattern, k = 1, 2, ..., m!. For example, when m = 4, a total of m! = 24 ordered patterns are generated, k = 1, 2, ..., 24.

[0107] Optionally, the ordered pattern feature mentioned in step three is the symbol phase space matrix π of the statistical CCP. m All types of ordered patterns Π m The probability of (k) occurring is used as a feature vector, also known as an ordered pattern distribution. Its calculation steps are as follows:

[0108] According to step two, the original phase space matrix S of CCP m Each row vector in the matrix is ​​mapped to the symbolic phase space matrix π. m The only ordered pattern in it. S m There are n row vectors, corresponding to n ordered patterns, and different row vectors may have the same ordered pattern.

[0109] Statistical ordered pattern set Π m Each type of ordered pattern Π m The probability of (k) occurring is used as an ordered pattern feature, and the calculation formula is as follows:

[0110]

[0111] In the formula, m is the embedding dimension parameter, and τ is the time delay parameter. n = N - (m - 1)τ represents the original phase space S. m The number of neutron sequences, i.e., the number of row vectors. S represents m Let i be the row vector of the i-th row in the array, where i = 1, 2, ..., n. A(u) The indicator function for set A is used to statistically analyze the original phase space S. m The number of ordered patterns is specified in the table, represented as:

[0112]

[0113] in, Representing a subsequence Corresponding ordered mode With ordered pattern set Π m Π in m (k) are the same, where k represents the sequence number of the ordered pattern, k = 1, 2, ..., m!. Representing a subsequence Corresponding ordered mode Belongs to the ordered pattern set Π m .

[0114] The above steps are used to calculate the identified CCP data within the ordered pattern set Π. mThe probability of each type of ordered pattern occurring in the CCP can be used to obtain the ordered pattern feature OP(m, τ) corresponding to that CCP. OP(m, τ) is a vector containing the probability of each type of ordered pattern.

[0115]

[0116] In the formula, Let represent the probability of the k-th ordered pattern feature under parameters m and τ, where k = 1, 2, ..., m!.

[0117] Optionally, the weighted ordered pattern (WOP) feature mentioned in step four mainly refers to using the amplitude information of CCP data to weight the ordered pattern feature OP(m, τ) to obtain a new feature WOP that simultaneously contains both CCP data amplitude information and data order information. Its calculation formula is as follows:

[0118]

[0119] In the formula, WOP(m, τ) represents the weighted ordered pattern (WOP) feature of CCP, which is weighted by the overall ordered pattern feature OP(m, τ) using the magnitude feature of CCP. w represents the magnitude feature of CCP. Let represent the eigenvalues ​​of the k-th class of WOP features under parameters m and τ, where k = 1, 2, ..., m!. In this implementation, w is represented by the variance of the CCP data, as shown below:

[0120]

[0121]

[0122] In the formula, y represents the mean of all data points in CCP. q Let q represent the data points in the CCP, where q = 1, 2, ..., N. N represents the total number of samples in the CCP.

[0123] Optionally, the WOP features based on multiple time delay parameters described in step five are multiple sets of WOP features obtained by changing the time delay parameter τ. When the embedding dimension parameter m is determined, the process information represented by the CCP can be observed from different scales by changing different time delay parameters τ. For example, for a system pattern CCP data of length N=30 [0, 1, 0, 1, 0, 1, ..., 0, 1, 0, 1], when m=3, a total of m! = 6 ordered patterns are generated, namely Π 3 (1) = [1, 2, 3], Π 3 (2) = [1, 3, 2], Π 3 (3) = [2, 1, 3], Π 3(4) = [2, 3, 1], Π 3 (5) = [3, 1, 2], Π 3 (6) = [3, 2, 1]. When τ = 1, there are only two subsequences: [0, 1, 0] and [1, 0, 1], which correspond to the ordered pattern [0, 1, 0] → Π respectively. 3 (2) = [1, 3, 2], [1, 0, 1] → Π 3 (3) = [2, 1, 3], with the ordered pattern feature being [0, 0.5, 0.5, 0, 0, 0]. However, when τ = 2, only two subsequences exist: [0, 0, 0] and [1, 1, 1], both corresponding to the ordered pattern Π. 3 (1) = [1, 2, 3], and the ordered pattern features are [1, 0, 0, 0, 0, 0].

[0124] It is evident that the CCP information observed under different time delay parameters τ varies significantly. Therefore, extracting WOP features of CCP from different scales can further refine the manufacturing process information. Based on the previous discussion, the range of τ satisfies τ∈[1, τ...]. max Therefore, the WOP characteristics based on multiple time delay parameters are determined by τ. max It consists of WOP features.

[0125] Optionally, the ensemble classifier method described in step five is a method for diagnosing the identified CCP using multiple individual classifiers. Individual classifiers are typically represented by a machine learning classification model, such as a neural network or support vector machine. Ensemble classifiers are divided into homogeneous classifiers and heterogeneous classifiers. A homogeneous classifier means that all individual classifiers belong to the same class, while a heterogeneous classifier means that not all individual classifiers belong to the same class. In the ensemble classifier method, each individual classifier derives a corresponding recognition result based on its own input data or features. In this embodiment, a homogeneous classifier method is used for pattern recognition. For each WOP feature based on multiple time delay parameters, an individual classifier is matched for diagnosis, thus a total of τ max Each individual classifier. In this step, each CCP will obtain τ. max The recognition result V h h = 1, 2, ..., τ max . τ max It is the maximum value of the delay parameter τ.

[0126] Optionally, the democratic voting method described in step six is ​​a method of fusion diagnosis based on the matching criterion of "majority rule". The final identification result of the CCP is V. ensemble By τ max The result that receives the most votes among all possible outcomes will be determined.

[0127] The method provided by the above embodiments of the present invention utilizes product quality or process data collected during partially controlled manufacturing processes, combined with control chart pattern recognition technology, to establish a monitoring model for abnormal states in the product manufacturing process. The proposed WOP feature is characterized by simple computation, strong noise resistance, and high sensitivity. Employing a time series analysis method based on ordered patterns, the sequential relationships and inherent fluctuations between CCP data can be identified, reflecting the underlying information of the data. Weighting the ordered pattern features using data amplitude yields WOP features that integrate CCP amplitude and sequence information under specific parameters. Based on WOP features with multiple time delay parameters, the CCP signal is analyzed using time delay parameters, fully extracting observation information at different scales from the CCP data. An ensemble classifier method based on multiple time delay parameters and a democratic voting method are proposed to fuse and diagnose the identification results of multiple sets of WOP features, which helps improve the accuracy of the model. The proposed method can achieve early anomaly identification in the manufacturing process, assisting on-site personnel in taking preventative measures and reducing the generation of defective products.

[0128] The following detailed description, with reference to the accompanying drawings, illustrates an exemplary embodiment of a control graph pattern recognition method based on ordered patterns and an ensemble classifier according to an invention.

[0129] Figure 3 Eight CCPs are shown in one example of applying the method according to an embodiment of the present invention, obtained from controlled data of the thickness of a galvanized metal sheet in a coating process. Figure 4 An example of an ordered pattern feature map (m=4, τ=15) calculated by eight CCPs is shown, applied to a method according to an embodiment of the present invention. Figure 5 The following is an example of a WOP feature map (m=4, τ=1-21) calculated using the UT mode in an example of applying the method according to an embodiment of the present invention. Figure 6 The illustration shows the diagnostic rates of 21 homogeneous individual classifiers and the overall diagnostic rate obtained from democratic voting in an example of applying the method according to an embodiment of the present invention. Figure 7 The diagram illustrates a confusion matrix of identification results for all unidentified samples after a comprehensive diagnosis using a democratic voting method, as shown in an example of applying the method according to an embodiment of the present invention.

[0130] Reference Figure 3-7 In an exemplary embodiment of the present invention, coating thickness data monitored during a galvanized metal sheet coating process is used as the basis for analysis, as coating thickness directly affects the corrosion resistance of the metal sheet. In the experiment, the data acquisition window size is set to N = 64, the mean μ of the controlled process is 60, and the standard deviation μ = 0.1.

[0131] Figure 1A flowchart of a control graph pattern recognition method based on ordered patterns and an ensemble classifier, according to an embodiment of the present invention, is shown. Figure 2 A flowchart of a comprehensive diagnostic process based on an ensemble classifier, according to an embodiment of the present invention, is shown. The following references... Figure 1 and Figure 2 This paper describes a control chart pattern recognition method based on an ordered pattern and an ensemble classifier, according to one embodiment of the present invention. The method includes the following steps:

[0132] Step 1: Data Acquisition and Generation: Based on the data from the controlled process in the actual coating process of the galvanized metal sheet, the mean μ = 60 and the standard deviation μ = 0.1 were obtained. According to the mean, standard deviation, and parameter settings in Table 1, 500 samples were generated for each CCP type using the Monte Carlo simulation method, for a total of 8 CCP types. 400 samples were used for training, and 100 samples were used for testing. The training sample set consisted of 400 × 8 = 3200 samples, and the test sample set consisted of 100 × 8 = 800 samples. Figure 3 Eight CCP diagrams were generated based on simulations of partially controlled data from this process.

[0133] Step Two: Constructing the Symbol Phase Space: Based on the experimental parameters, each CCP sample is reconstructed into a phase space. In this implementation, the data length of the CCP sample is N = 64, the embedding dimension parameter is m = 4, and there are a total of m! = 24 ordered patterns. According to the formula... τ was calculated max =21, that is, τ∈[1,21] and is a positive integer. Fixing the parameters m and τ can yield the corresponding symbolic phase space.

[0134] Step 3: Extract ordered pattern features: Based on the parameters m and τ from Step 2, the ordered pattern features of CCP can be extracted. Figure 4 The diagram shows the ordered pattern features extracted from 8 CCPs when m=4 and τ=15. The parameter τ∈[1,21], thus 21 ordered pattern features at different observation scales can be obtained.

[0135] Step 4: Generate Weighted Ordered Pattern (WOP) Features: To further improve the sensitivity of features to different patterns, the features are weighted from the perspective of data amplitude. The standard deviation of each CCP sample is used as the weight w to weight the ordered pattern features. The extracted WOP features integrate the order information and amplitude information of the CCP data.

[0136] Step 5: Generating WOP features based on multiple time delay parameters and an ensemble classifier method: By changing the size of the time delay parameter, WOP features corresponding to different observation scales are obtained. Since the parameter τ∈[1,21], 21 types of WOP features can be obtained, collectively referred to as WOP features based on multiple time delay parameters. In this embodiment, the multiple time delay WOP features of each CCP are 21×24 dimensional. Figure 5 A 3D histogram of multi-delay WOP features for a certain UT mode CCP data sample is presented. It can be seen that as the delay parameter τ increases, the corresponding WOP feature's representation ability of the UT mode strengthens; that is, the WOP feature corresponding to a larger parameter τ has a stronger ability to identify the UT mode.

[0137] For each delay parameter τ, a classifier is assigned to the WOP features, and all classifiers belong to the same class, i.e., homogeneous classifiers. In this embodiment, a support vector machine is used as the homogeneous classifier. The input of each classifier is the corresponding WOP feature vector, and the output is the type of CCP identified. Figure 6 The presentation shows the recognition results of 21 individual classifiers corresponding to 21 time delay parameters τ. The classifier achieves the highest recognition accuracy when τ = 15, V... 15 =94.625%.

[0138] Step Six: Comprehensive Diagnosis Based on Democratic Voting: Each CCP is identified according to the above steps. A democratic voting method is used, and the identification results of the 21 individual classifiers obtained in Step Five are comprehensively diagnosed based on the "majority rule" principle. The result is V. ensemble =97.375%, which is 2.75% higher than the recognition method using a single classifier. Figure 7 The confusion matrix diagram shows the identification results of 800 CCP samples to be tested in the test sample set.

[0139] Using a segment of data collected during the actual spraying process as the CCP sample to be tested, Table 2 shows the judgment results of 21 individual classifiers for this CCP. The identification results indicate that 14 classifiers identified it as DS, 5 as DT, 1 as SYS, and 1 as CYC. Based on the majority rule principle, the CCP type was ultimately determined to be DS, i.e., descent step mode.

[0140] Table 2 Description of Vibration Signal Set

[0141]

[0142] The occurrence of the DS mode described above may be due to malfunction of the spraying equipment at some point, resulting in the coating failing to adhere properly to the metal plate, or due to operator replacement and potential improper operation. Upon verification, the anomaly was confirmed to be caused by malfunction of the spraying equipment.

[0143] Therefore, based on the identification results, the proposed method can not only promptly identify and alarm abnormalities in the manufacturing process, but also obtain the abnormality pattern type and infer the possible causes of the abnormality. This provides reasonable suggestions for on-site personnel to conduct inspections, shutdowns, or maintenance of the manufacturing process or system, avoiding manufacturing system failures or product manufacturing process abnormalities caused by untimely detection of abnormalities or unclear causes. This is conducive to improving the qualification rate of the manufacturing system and ensuring product quality.

[0144] This invention proposes a control chart pattern recognition method based on weighted ordered patterns and an ensemble classifier. This method utilizes quality or process data generated during product manufacturing to promptly detect anomalies and infer their causes. Ordered pattern features are highly sensitive to abnormal fluctuations in CCP data, and are computationally simple and noise-resistant. Multi-delayed (WOP) features integrate the sequential relationships and amplitude information between data points, and analyze CCP data from different scales, resulting in richer manufacturing process information. The ensemble classifier method based on democratic voting, by combining the recognition results of multiple classifiers for fusion diagnosis, significantly improves the model's accuracy in CCP identification. The proposed model is not only applicable to anomaly monitoring in the metal spraying process in the example, but also applicable to the production and processing of other products or components. This invention has good scalability and provides a valuable reference for other skilled personnel in this field.

[0145] It should be noted that the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0146] Furthermore, the foregoing only describes some embodiments, and changes, modifications, additions, and / or variations can be made without departing from the scope and spirit of the disclosed embodiments. These embodiments are illustrative and not restrictive. Moreover, the described embodiments relate to those currently considered most practical and preferred, and should be understood as not being limited to the disclosed embodiments, but rather intended to cover different modifications and equivalent arrangements included within the spirit and scope of that embodiment. Furthermore, the various embodiments described above can be used in conjunction with other embodiments; for example, an aspect of one embodiment can be combined with an aspect of another embodiment to implement yet another embodiment. Additionally, individual features or components of any given component can constitute another embodiment.

[0147] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions 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, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A control chart pattern recognition method based on weighted ordered patterns and ensemble classifiers, characterized by The method comprises the following steps: Step one: data collection and data generation: collecting part of the data of the controlled process in the actual manufacturing process, and generating 8 modes of CCPs by using the Monte Carlo method according to the part of the data; Step two: constructing symbolic phase space: for CCP observation data of a given length, the CCP observation data is reconstructed into symbolic phase space according to parameter combination; Step three: extracting ordered pattern features: obtaining all ordered pattern types according to the embedding dimension parameter, and calculating the ordered pattern features of the identified CCP observation data; Step four: generating WOP features: calculating the amplitude features of the identified CCP observation data, and weighting the ordered pattern features in step three to obtain WOP features; Step five: generating multi-delay-based WOP features and integrating through the integrated classifier method: by changing the size of the delay parameter, WOP corresponding to different delay parameters is obtained, and as many homogeneous classifiers as the number of delay parameters are constructed for individual classifier classification; Step six: comprehensive diagnosis based on democratic voting: using the democratic voting method, the identification results of the multiple classifiers in step five are voted, and finally the comprehensive diagnosis result of the identified CCP is obtained; The actual manufacturing process in step one comprises a machining process, an assembly process and a running-in process. The 8 modes of CCPs in step one comprise CCPs corresponding to NOR mode, CYC mode, UT mode, DT mode, US mode, DS mode, STR mode and SYS mode, wherein the NOR mode represents a process affected only by random interference and in a controlled state, and the remaining 7 modes are abnormal modes and the manufacturing process is in an uncontrolled state.

2. The control chart pattern recognition method based on weighted ordered patterns and integrated classifiers according to claim 1, wherein: The part of the data of the controlled process in step one comprises size data of a product in a machining process and spraying thickness data of a product in a spraying process.

3. The control chart pattern recognition method based on weighted ordered patterns and integrated classifiers as claimed in claim 1, wherein, The 8 modes of CCPs in step one are obtained by using the Monte Carlo simulation method according to data distribution, and specifically comprise: According to the part of the data of the controlled process in the actual manufacturing process in the NOR mode, the mean μ and the standard deviation σ of the part of the data are calculated, and the expression of each simulation CCP is as follows: y t = μ + x t + d t where y t represents the observation of the monitoring variable of the manufacturing process at time t, μ represents the mean of the controlled process, x t = r t x σ is the random disturbance in the process, which is normally distributed x t ~ N(0, σ), r t is a random variable at time t, which is normally distributed, d t is used to simulate abnormal patterns, where d t = 0 represents the NOR mode only subject to random disturbance.

4. The control chart pattern recognition method based on weighted ordered patterns and integrated classifiers as claimed in claim 3, wherein : The parameter combination in step two refers to two parameters m and τ used for constructing the ordered pattern feature, where m is referred to as an embedding dimension parameter, i.e., the data dimension of each ordered pattern, and τ is referred to as a time delay parameter, i.e., a time delay parameter, representing the sampling interval when constructing the ordered pattern. τ reflects the inherent scale information of the time series, as well as the fluctuation structure of the time series at different scales. The range of τ is affected by the parameters m and the length N of the CCP observation data. In order to ensure that at least one complete ordered pattern is extracted, the maximum value τ of τ is max satisfies the following formula: N = 1 + (m - 1)τ max wherein the range of τ should satisfy τ ∈ [1, τ max ], round denotes rounding down; The symbolic phase space in step two is a method of converting CCP observation data of length N into symbolic phase space expressed by multiple ordered patterns by using parameters m and τ, and the specific operation is as follows: The CCP observation data sequence Y = {yl, y2,..., yN} of length N is divided into n subsequences of length m according to the parameters m and τ N} according to the parameters m and τ where y1, y2,..., y N are data points in the CCP observation data; subsequence for the i-th subsequence in the original phase space matrix S m i = 1, 2, …, n, and n is determined by the parameters m and τ and the length N of the CCP observation data, satisfying n = N - (m - 1)τ; The ordered pattern consists of an m-dimensional tuple [r1, r2, ..., rm] containing m positive integers [1, 2, ..., m], where each number appears only once. j ,…,r m The ] indicates that the subsequence is sorted in ascending order based on the size of its elements, and the subsequence is... Convert to the corresponding ordered mode where r j is a positive integer [1, m], j = 1, 2, …, m, r j represents the position of a data point in the sub-sequence , represents the position of a data point in the sub-sequence corresponding to the embedding dimension parameter m and the time delay parameter τ, when there are equal elements in the sub-sequence , the ordering relationship of the two satisfies r l < r l+1 < r l , where r m represents the position of a data point in the sub-sequence, l = 1, 2, …, m-1; the set of these ordered patterns is called the symbolic phase space matrix of CCP, denoted as π m .

5. The control chart pattern recognition method based on weighted ordered patterns and integrated classifiers according to claim 4, wherein: The ordered pattern in step three is all possible data fluctuation patterns obtained according to the embedding dimension parameter m, each ordered pattern is composed of m positive integers from 1 to m, and each positive integer appears only once, according to the ordering mode of the m positive integers, the parameter m corresponds to m! ordered patterns, and the set of all types of ordered patterns is denoted as Each ordered pattern is denoted as Π m (k), k represents the type serial number of the ordered pattern, k = 1, 2, …, m!; The ordered pattern features described in step three are the symbolic phase space matrices π of the CCP observation data m All types of ordered patterns Π m (k) The probability of occurrence as a feature vector, also known as the ordered pattern distribution, is calculated as follows: According to step two, the original phase space matrix S of the CCP observation data m Each row vector in the matrix is ​​mapped to the symbolic phase space matrix π. m The only ordered pattern in S m There are n row vectors, corresponding to n ordered patterns, and different row vectors may correspond to the same ordered pattern; Statistical ordered pattern set Π m Each type of ordered pattern Π m The probability of (k) occurring is used as an ordered pattern feature, and the calculation formula is as follows: where m is the embedding dimension parameter, τ is the time delay parameter, and n = N - (m - 1)τ represents the number of row vectors in the original phase space S m the number of neutron sequences, i.e., the number of row vectors, denotes S m the row vector in the i-th row, i = 1, 2, …, n, I A(u) denotes the indicator function of the set A, which is used to count the original phase space S m denotes the number of specified ordered patterns, which is denoted as: wherein, subsequence corresponding ordered pattern belongs to the ordered pattern set m Π m (k) is the same as Π m (k), k denotes the serial number of the ordered pattern, k = 1, 2, …, m!, and subsequence corresponding ordered pattern belongs to the ordered pattern set m ; The probability of the identified CCP observation data appearing in all types of ordered patterns in the ordered pattern set Π m is calculated to obtain the ordered pattern feature OP(m, τ) corresponding to the CCP observation data. OP(m, τ) is a vector containing the probability of each type of ordered pattern: wherein, denotes the probability of the kth ordered pattern feature at parameters m and τ, k = 1, 2,..., m!.

6. The control chart pattern recognition method based on weighted ordered patterns and integrated classifiers as claimed in claim 5, wherein : The WOP feature in step four is to weight the ordered pattern feature OP(m, τ) using the amplitude information of CCP observation data to obtain a new feature WOP containing both the amplitude information and the sequential information of CCP observation data, and the calculation formula is as follows: wherein WOP(m,τ) represents a WOP feature of the CCP observation data, the ordered pattern feature OP(m,τ) is weighted as a whole by the amplitude feature of the CCP observation data, and O represents the amplitude feature of the CCP observation data, represents an eigenvalue of the kth WOP feature at parameters m and τ, k = 1, 2, …, m!, and w is represented by the variance of the CCP observation data as follows: wherein denotes the mean of all data points in the CCP observation data, y q denotes a data point in the CCP observation data, q = 1, 2,..., N.

7. The control chart pattern recognition method based on weighted ordered patterns and integrated classifiers as claimed in claim 6, wherein : The multi-delay WOP feature in step five is obtained by changing the delay parameter τ, and when the embedding dimension parameter m is determined, the process information represented by CCP observation data is observed from different scales by changing different delay parameters τ.

8. The control chart pattern recognition method based on weighted ordered patterns and integrated classifiers as claimed in claim 7, wherein : The integration by the ensemble classifier method in step five is to diagnose the identified CCP observation data by using multiple individual classifiers. The ensemble classifier is divided into homogeneous classifiers and heterogeneous classifiers. The homogeneous classifier refers to all individual classifiers being of the same type. The heterogeneous classifier refers to all individual classifiers not being of the same type. In the ensemble classifier method, each individual classifier will obtain a corresponding recognition result according to respective input data or features. For each WOP feature in the multi-delay WOP feature, a homogeneous classifier is matched for diagnosis. There are τ max individual classifiers. Each CCP obtains τ max recognition results V h , h = 1, 2, …, τ max .

Images