A method for monitoring welding quality of resistance spot welding data multi-neighbor feature analysis

Abstract

The application discloses a welding quality monitoring method of resistance spot welding data multi-neighbor feature analysis, which collects process data when the resistance spot welding is normal and implements standardization processing; obtains a regression coefficient matrix of the standardized data matrix; obtains a multi-neighbor relationship matrix through the regression coefficient matrix, further obtains a generalized eigenvalue problem and solves it; obtains a feature transformation matrix according to the solving result, further obtains a score matrix, a covariance matrix and a monitoring index matrix; obtains control upper limits of two monitoring indexes according to the monitoring index matrix; implements standardization processing on the latest process data; obtains a score vector and an error vector of the standardized data vector according to the feature transformation matrix, further obtains the two monitoring indexes; determines whether the welding quality is normal or not by judging whether the two monitoring indexes are within the control upper limits; and the method has the advantage that the reliability of the welding quality monitoring can be effectively improved.

Description

Technical Field

[0001] This invention relates to an intelligent resistance spot welding quality monitoring technology, and more particularly to a welding quality monitoring method based on multi-nearest neighbor feature analysis of resistance spot welding data. Background Technology

[0002] Resistance spot welding boasts advantages such as high production efficiency and adaptability to automated operations, making it a crucial sheet metal processing technology frequently employed in large-scale industries like automotive and aerospace. However, the short weld nugget formation time and the influence of various factors on the welding process, coupled with varying degrees of deterministic and stochastic interactions among these factors, necessitate real-time monitoring of the resistance spot welding process, presenting significant challenges to its implementation. With the increasing application of resistance welding in various industrial sectors, the ever-growing demands for welded joints, and the considerations of energy and production efficiency in large-scale production environments, spot welding parameter measurement and quality monitoring and control technologies are receiving growing attention. Because resistance spot welding is a dynamic process, particularly due to its instantaneous nature and the invisibility of weld nugget formation, controlling the quality of the spot welding process presents considerable difficulties. With the gradual maturation of data-driven intelligent methods, research incorporating data mining techniques into the field of spot welding quality monitoring is also on the rise.

[0003] In existing data-driven fault detection methods, multivariate statistical process monitoring is the mainstream technique, such as Principal Component Analysis (PCA), Locality Preserving Projections (LPP), and Neighborhood Preserving Embedding (NPE). From the perspective of data point spatial dispersion, PCA extracts the variance information of the original data; that is, PCA tries to distribute the original data as widely as possible during projection transformation. LPP and NPE, on the other hand, consider the local nearest neighbor features of the original data points. When extracting potential information from the original data, they try to preserve the spatial distribution characteristics of the data points. Therefore, fault detection methods that consider the local structural features of data neighbors can open up new avenues for enriching the fault detection method system.

[0004] However, considering the complex characteristics of spot welding process data, the sampled data also exhibits a certain degree of temporal correlation (i.e., autocorrelation between sampled data), and the angular information between data points can also reflect some characteristics of the original data. If only the proximity features between data points are considered, the extracted potential information will be incomplete, resulting in the loss of useful information and failing to adequately describe the state of normal process data. It is conceivable that if the proximity features, temporal proximity features, and angular proximity features of the original data points are considered from multiple perspectives during data projection transformation, the extracted potential information will include more useful information reflecting the process's operational state, minimizing the risk of information loss. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a welding quality monitoring method based on multi-nearest neighbor feature analysis of resistance spot welding data, which can effectively improve the reliability of welding quality monitoring.

[0006] The technical solution adopted by this invention to solve the above-mentioned technical problems is: a welding quality monitoring method based on multi-nearest neighbor feature analysis of resistance spot welding data, characterized by including the following steps:

[0007] Step 1: Collect process data during normal resistance spot welding operation. Compile the process data collected at N sampling times into a data matrix, denoted as X. Then, standardize each column of data in X to obtain a standardized data matrix with a mean of 0 and a standard deviation of 1, denoted as X. The process data collected at each sampling time point are, in sequence, welding voltage, welding current, dynamic resistance, electrode pressure, and electrode displacement, where N≥500, X∈R. N×5 R is the set of real numbers. N×5 X represents an N×5 dimensional real number matrix, where the data in columns 1 to 5 correspond to welding voltage, welding current, dynamic resistance, electrode pressure, and electrode displacement, respectively.

[0008] Step 2: Obtain The regression coefficient matrix W is obtained through the following process:

[0009] Step 2.1: Let W denote an N×N dimensional matrix, and initialize W as an all-zero matrix; from Starting from the first row of the vector, traverse sequentially. Each row vector in the array;

[0010] Step 2.2: Set the row vector of the current traversal to... The i-th row vector in the array is denoted as . Then, considering the three aspects of spatial distance nearest neighbors, sampling time nearest neighbors, and spatial angle nearest neighbors, from... Find the match Given similar row vectors, find N that can be found together. i N row vectors; then find the N found in total i Each row vector is arranged according to its own row vector. The order of the row index numbers in the data is combined to form the data. The corresponding multiple nearest neighbor feature matrix is ​​denoted as And find the N together i Each row vector The set of row index numbers in the data is denoted as . Where i represents The row index of the row vector in the array, 1≤i≤N, where i is initially 1 and 1<N. i <N, N represents i A 5×5 dimensional real matrix;

[0011] Step 2.3: Calculation The regression coefficient vector is denoted as... Then to Normalization is performed, and the normalized regression coefficient vector is denoted as w. i , Then according to All row indices in W are used to set the elements of the same row in the i-th column vector of W to w in order. i The elements in the expression; where the superscript "T" indicates the transpose of a vector or matrix, and the symbol "||||" indicates the length of the vector. N represents i A 1×1 dimensional vector;

[0012] Step 2.4: Continue traversing The next row vector in the sequence is then processed, and the process returns to step 2.2 to continue until... After traversing all row vectors, we obtain... The regression coefficient matrix W;

[0013] Step 3: Calculation The nearest neighbor relation matrix is ​​denoted as M, where M = (I N -W)(I N -W) T Then solve the generalized eigenvalue problem. All eigenvalues ​​and their corresponding eigenvectors are given, and the eigenvectors corresponding to the smallest d eigenvalues ​​are denoted as p1, p2, ..., p. d Then set p1, p2, ..., p d A characteristic transformation matrix is ​​formed, denoted as P, where P = [p1, p2, ..., p]. d]; among them, I N Let d represent an N×N dimensional identity matrix, λ represent eigenvalues, p represent eigenvectors, and each eigenvector has a dimension of 5×1, 1 < d < 5, and p ∈ R. 5×d R 5×d Represents a 5×d-dimensional real matrix;

[0014] Step 4: Calculation The score matrix, denoted as S, Then calculate The covariance matrix is ​​denoted as Λ, Λ = S T S / (N-1); then calculate The monitoring indicator matrix, denoted as Q, Next, calculate the mean and variance of the N elements on the main diagonal of Q, denoted as U and V respectively; finally, calculate the upper control limit of the first monitoring indicator and the upper control limit of the second monitoring indicator, denoted as D respectively. lim and E lim , Where I5 ​​represents a 5×5 dimensional identity matrix, F d,N-d,α This represents the value corresponding to the F-distribution with confidence level α and degrees of freedom d and Nd, where g and h are introduced intermediate variables. This indicates that the value corresponds to a chi-square distribution with h degrees of freedom and α confidence level.

[0015] Step 5: When it is necessary to use the process data collected at the current sampling time for welding quality monitoring, represent the process data collected at the current sampling time as a 1×5 dimensional data vector, and denote it as x. new Then for x new Each data point in the dataset is standardized to obtain a 1×5 dimensional standardized data vector with a mean of 0 and a standard deviation of 1, denoted as . Recalculate The score vector and error vector are denoted as s. new and e new , Then, the first and second monitoring indicators are calculated, denoted as D and E respectively, where D = s new Λ -1 s new T , Finally, determine whether the condition D≤D is satisfied. lim And E≤E lim If the conditions are met, the welding quality of the resistance spot weld corresponding to the process data collected at the current sampling time is considered normal; if the conditions are not met, the welding quality of the resistance spot weld corresponding to the process data collected at the current sampling time is considered abnormal, and the position of the resistance spot weld is marked.

[0016] In step 2.2, N nearest neighbors were found based on three aspects: spatial distance, sampling time, and spatial angle. i The process of creating each row vector is as follows:

[0017] Step 2.2a: For spatially nearest neighbors, calculate Except Other row vectors and Spatial distance between them There are N-1 corresponding spatial distances. The j-th row vector in The spatial distance between them is denoted as Then to The corresponding N-1 spatial distances δ i,1 ,δ i,2 ,…,δ i,i-1 ,δ i,i+1 ,…,δ i,N Sort by numerical value in ascending order; then from... Select the row vectors that have spatial distances to the top k sorted rows, and use these selected row vectors as the spatial nearest neighbor row vectors, recording their row indices; where j represents The row index of the row vector in the diagram, j = 1, 2, ..., N and j ≠ i, the symbol "||||" indicates the length of the vector, which is also the Euclidean distance calculation symbol. express The j-th row vector in δ i,1 express The first row vector in The spatial distance between them, δ i,2 express The second row vector in The spatial distance between them, δ i,i-1 express The (i-1)th row vector in The spatial distance between them, δ i,i+1 express The (i+1)th row vector in The spatial distance between them, δ i,N express The Nth row vector in The spatial distance between them, where k represents the set number of nearest neighbors, and the value of k ranges from [6,12].

[0018] For sampling time proximity, based on the order of sampling times, from Find the collection from The previous at most t row vectors and the data collected from For the next t row vectors, these found row vectors are taken as the nearest neighbor row vectors at the sampling time, and their row indices are recorded; where the order of sampling times is assumed to be... If the row indices of the row vectors in the array are in ascending order, then: when t+1≤i≤Nt, the data collected from... The previous t row vectors and the data collected from For the subsequent t row vectors, when i < t+1, it is possible to find the data collected from... The following t row vectors are collected from The number of row vectors previously was 0 to t-1. When i > Nt, it is possible to find the data collected from... The previous t row vectors were collected from The number of subsequent row vectors is from 0 to t-1, where t represents the number of nearest neighbors at the set sampling time, and the value of t ranges from t∈[1,4].

[0019] For spatial angular nearest neighbors, calculate Except Other row vectors and The cosine value of the angle between them, There are N-1 corresponding angle cosine values. The j-th row vector in The cosine value of the angle between them is denoted as cosθ. i,j , Then from Find the angle cosine values ​​whose absolute value is greater than the corresponding N-1 angle cosine values. All angle cosine values; then from Select the row vectors corresponding to all the found angle cosine values, and use these selected row vectors as the spatial angle nearest neighbor row vectors, recording their row index numbers; where θ i,j express The j-th row vector in The spatial angle between them, with the superscript "T" indicating the transpose of a vector or matrix. This indicates the set spatial angle nearest neighbor threshold. The range of values ​​is

[0020] Step 2.2b: Delete duplicate row indexes from all row indexes recorded for spatial distance nearest neighbors, sampling time nearest neighbors, and spatial angle nearest neighbors, keeping only one. The set of all remaining row indexes after the deletion operation is the [database name missing]. from The middle pick travel index number belongs to All row vectors of are equal to . Similar N iRow vectors.

[0021] Compared with the prior art, the advantages of the present invention are as follows:

[0022] When extracting the normal variation characteristics of resistance spot welding process data, the method of the present invention comprehensively considers the spatial distance nearest neighbor local features, sampling time nearest neighbor local features, and spatial angle nearest neighbor local features between process data. When performing projection transformation of the original data, it can greatly reduce the risk of data feature loss. The extracted potential information can contain more useful information that can reflect the normal operation status of the process. Therefore, the method of the present invention can effectively improve the reliability of welding quality monitoring. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating the implementation of the method of the present invention. Detailed Implementation

[0024] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0025] This invention proposes a welding quality monitoring method based on multi-nearest neighbor feature analysis of resistance spot welding data, the implementation flowchart of which is shown below. Figure 1 As shown, it includes the following steps:

[0026] Step 1: Collect process data during normal resistance spot welding operation. Compile the process data collected at N sampling times into a data matrix, denoted as X. Then, standardize each column of data in X to obtain a standardized data matrix with a mean of 0 and a standard deviation of 1, denoted as X. The process data collected at each sampling time are, in sequence, welding voltage, welding current, dynamic resistance, electrode pressure, and electrode displacement, where N≥500. If N=600, X∈R N×5 R is the set of real numbers. N×5 X represents an N×5 dimensional real number matrix. The data in columns 1 to 5 correspond to five variables that can be measured in real time during resistance spot welding: welding voltage, welding current, dynamic resistance, electrode pressure, and electrode displacement.

[0027] Step 2: Obtain The regression coefficient matrix W is obtained through the following process:

[0028] Step 2.1: Let W denote an N×N dimensional matrix, and initialize W as an all-zero matrix; from Starting from the first row of the vector, traverse sequentially. Each row of vectors in the array.

[0029] Step 2.2: Set the row vector of the current traversal to... The i-th row vector in the array is denoted as . Then, considering the three aspects of spatial distance nearest neighbors, sampling time nearest neighbors, and spatial angle nearest neighbors, from... Find the match Given similar row vectors, find N that can be found together. i N row vectors; then find the N found in total i Each row vector is arranged according to its own row vector. The order of the row index numbers in the data is combined to form the data. The corresponding multiple nearest neighbor feature matrix is ​​denoted as And find the N together i Each row vector The set of row index numbers in the data is denoted as . Where i represents The row index of the row vector in the array, 1≤i≤N, where i is initially 1 and 1<N. i <N, N represents i A 5×5 dimensional real matrix.

[0030] In this embodiment, in step 2.2, N nearest neighbors are found based on three aspects: spatial distance, sampling time, and spatial angle. i The process of creating each row vector is as follows:

[0031] Step 2.2a: For spatially nearest neighbors, calculate Except Other row vectors and Spatial distance between them There are N-1 corresponding spatial distances. The j-th row vector in The spatial distance between them is denoted as δ i,j , Then to The corresponding N-1 spatial distances δ i,1 ,δ i,2 ,…,δ i,i-1 ,δ i,i+1 ,…,δ i,N Sort by numerical value in ascending order; then from... Select the row vectors that correspond to the k smallest spatial distances (i.e., the k shortest spatial distances) after sorting, and use these selected row vectors as the nearest neighbor row vectors. Record the row index of these selected row vectors; where j represents The row index of the row vector in the diagram, j = 1, 2, ..., N and j ≠ i, the symbol "||" represents the length of the vector, which is also the Euclidean distance calculation symbol. express The j-th row vector in δ i,1 express The first row vector in The spatial distance between them, δ i,2 express The second row vector in The spatial distance between them, δ i,i-1 express The (i-1)th row vector in The spatial distance between them, δ i,i+1 express The (i+1)th row vector in The spatial distance between them, δ i,N express The Nth row vector in The spatial distance between them, k represents the set number of nearest neighbors, which is a nearest neighbor parameter, and the value of k ranges from [6,12].

[0032] For sampling time proximity, based on the order of sampling times, from Find the collection from The previous at most t row vectors and the data collected from For the next t row vectors, these found row vectors are taken as the nearest neighbor row vectors at the sampling time, and their row indices are recorded; where the order of sampling times is assumed to be... If the row indices of the row vectors in the array are in ascending order, then: when t+1≤i≤Nt, the data collected from... The previous t row vectors and the data collected from For the subsequent t row vectors, when i < t+1, it is possible to find the data collected from... The following t row vectors are collected from The number of row vectors previously was 0 to t-1. When i > Nt, it is possible to find the data collected from... The previous t row vectors were collected from The number of subsequent row vectors is from 0 to t-1, where t represents the number of nearest neighbors at the set sampling time, which is a nearest neighbor parameter, and the value of t ranges from t∈[1,4].

[0033] For spatial angular nearest neighbors, calculate Except Other row vectors and The cosine value of the angle between them, There are N-1 corresponding angle cosine values. The j-th row vector in The cosine value of the angle between them is denoted as cosθ. i,j , Then from Find the angle cosine values ​​whose absolute value is greater than the corresponding N-1 angle cosine values. All angle cosine values; then from Select the row vectors corresponding to all the found angle cosine values, and use these selected row vectors as the spatial angle nearest neighbor row vectors, recording their row index numbers; where θ i,j express The j-th row vector in The spatial angle between them, with the superscript "T" indicating the transpose of a vector or matrix. This represents the set spatial angle nearest neighbor threshold, which is a nearest neighbor parameter. The range of values ​​is

[0034] Step 2.2b: Delete duplicate row indexes from all row indexes recorded for spatial distance nearest neighbors, sampling time nearest neighbors, and spatial angle nearest neighbors, keeping only one. The set of all remaining row indexes after the deletion operation is the [database name missing]. from The middle pick travel index number belongs to All row vectors of are equal to . Similar N i Row vectors.

[0035] Step 2.3: Calculation The regression coefficient vector is denoted as... Then to Normalization is performed, and the normalized regression coefficient vector is denoted as w. i , Then according to All row indices in W are used to set the elements of the same row in the i-th column vector of W to w in order. i The various elements in, such as If the row indices in W are 1, 4, 6, and 8, then the element in the first row of the i-th column vector in W is set to w. i The first element in W, the fourth element in the i-th column vector of W is set to w. i The second element in W, the element in the 6th row of the i-th column vector is set to w. i The third element in W, the element in the 8th row of the i-th column vector is set to w. i The fourth element in the matrix; where the superscript "T" indicates the transpose of a vector or matrix, and the symbol "||" indicates the length of the vector. N represents i A 1×1 dimensional vector.

[0036] Step 2.4: Continue traversing The next row vector in the sequence is then processed, and the process returns to step 2.2 to continue until... After traversing all row vectors, we obtain... The regression coefficient matrix W.

[0037] Step 3: Calculation The nearest neighbor relation matrix is ​​denoted as M, where M = (I N -W)(I N -W) T Then solve the generalized eigenvalue problem. All eigenvalues ​​and their corresponding eigenvectors are given, and the eigenvectors corresponding to the smallest d eigenvalues ​​are denoted as p1, p2, ..., p. d Then set p1, p2, ..., p d A characteristic transformation matrix is ​​formed, denoted as P, where P = [p1, p2, ..., p]. d ]; among them, I N Let d represent an N×N dimensional identity matrix, λ represent eigenvalues, p represent eigenvectors, and each eigenvector has a dimension of 5×1, where 1 < d < 5, the empirical value of d is 3 or 4, and P ∈ R. 5×d R 5×d This represents a 5×d-dimensional real matrix.

[0038] Step 4: Calculation The score matrix, denoted as S, Then calculate The covariance matrix is ​​denoted as Λ, Λ = S T S / (N-1); then calculate The monitoring indicator matrix, denoted as Q, Next, calculate the mean and variance of the N elements on the main diagonal of Q, denoted as U and V respectively; finally, calculate the upper control limit of the first monitoring indicator and the upper control limit of the second monitoring indicator, denoted as D respectively. lim and E lim , Where I5 ​​represents a 5×5 dimensional identity matrix, F d,N-d,α This represents the value corresponding to the F-distribution with confidence level α and degrees of freedom d and Nd, where α = 99%, and g and h are introduced intermediate variables. This indicates that the chi-square distribution corresponds to a degree of freedom of h and a confidence level of α.

[0039] Step 5: When it is necessary to use the process data collected at the current sampling time for welding quality monitoring, represent the process data collected at the current sampling time as a 1×5 dimensional data vector, and denote it as x. new Then for x newEach data point in the dataset is standardized to obtain a 1×5 dimensional standardized data vector with a mean of 0 and a standard deviation of 1, denoted as . Recalculate The score vector and error vector are denoted as s. new and e new , Then, the first and second monitoring indicators are calculated, denoted as D and E respectively, where D = s new Λ -1 s new T , Finally, determine whether the condition D≤D is satisfied. lim And E≤E lim If the conditions are met, the welding quality of the resistance spot weld corresponding to the process data collected at the current sampling time is considered normal; if the conditions are not met, the welding quality of the resistance spot weld corresponding to the process data collected at the current sampling time is considered abnormal, and the position of the resistance spot weld is marked. Real-time monitoring of the welding quality of resistance spot welds can be achieved by following step 5.

Claims

1. A welding quality monitoring method based on multi-nearest neighbor feature analysis of resistance spot welding data, characterized in that... Includes the following steps: Step 1: Collect process data during normal resistance spot welding operation. Compile the process data collected at N sampling times into a data matrix, denoted as X. Then, standardize each column of data in X to obtain a standardized data matrix with a mean of 0 and a standard deviation of 1, denoted as X. The process data collected at each sampling time point are, in sequence, welding voltage, welding current, dynamic resistance, electrode pressure, and electrode displacement, where N≥500, X∈R. N×5 R is the set of real numbers. N×5 X represents an N×5 dimensional real number matrix, where the data in columns 1 to 5 correspond to welding voltage, welding current, dynamic resistance, electrode pressure, and electrode displacement, respectively. Step 2: Obtain The regression coefficient matrix W is obtained through the following process: Step 2.1: Let W denote an N×N dimensional matrix, and initialize W as an all-zero matrix; from Starting from the first row of the vector, traverse sequentially. Each row vector in the array; Step 2.2: Set the row vector of the current traversal to... The i-th row vector in the array is denoted as . Then, considering the three aspects of spatial distance nearest neighbors, sampling time nearest neighbors, and spatial angle nearest neighbors, from... Find the match Given similar row vectors, find N that can be found together. i N row vectors; then find the N found in total i Each row vector is arranged according to its own row vector. The order of the row index numbers in the data is combined to form the data. The corresponding multiple nearest neighbor feature matrix is ​​denoted as And find the N together i Each row vector The set of row index numbers in the array is denoted as θ. i Where i represents The row index of the row vector in the array, 1≤i≤N, where i is initially 1 and 1<N. i <N, N represents i A 5×5 dimensional real matrix; Step 2.3: Calculation The regression coefficient vector is denoted as... Then to Normalization is performed, and the normalized regression coefficient vector is denoted as w. i , Then according to θ i All row indices in W are used to set the elements of the same row in the i-th column vector of W to w in order. i The elements in the expression; where the superscript "T" indicates the transpose of a vector or matrix, and the symbol "||||" indicates the length of a vector. N represents i A 1×1 dimensional vector; Step 2.4: Continue traversing The next row vector in the sequence is then processed, and the process returns to step 2.2 to continue until... After traversing all row vectors, we obtain... The regression coefficient matrix W; Step 3: Calculation The nearest neighbor relation matrix is ​​denoted as M, where M = (I N -W)(I N -W) T Then solve the generalized eigenvalue problem. All eigenvalues ​​and their corresponding eigenvectors are given, and the eigenvectors corresponding to the smallest d eigenvalues ​​are denoted as p1, p2, ..., p. d Then set p1, p2, ..., p d A characteristic transformation matrix is ​​formed, denoted as P, where P = [p1, p2, ..., p]. d ]; among them, I N Let d represent an N×N dimensional identity matrix, λ represent eigenvalues, p represent eigenvectors, and each eigenvector has a dimension of 5×1, 1 < d < 5, and p ∈ R. 5×d R 5×d Represents a 5×d-dimensional real matrix; Step 4: Calculation The score matrix, denoted as S, Then calculate The covariance matrix is ​​denoted as Λ, where Λ = S T S / (N-1); then calculate The monitoring indicator matrix, denoted as Q, Next, calculate the mean and variance of the N elements on the main diagonal of Q, denoted as U and V respectively; finally, calculate the upper control limit of the first monitoring indicator and the upper control limit of the second monitoring indicator, denoted as D respectively. lim and E lim , Where I5 ​​represents a 5×5 dimensional identity matrix, F d,N-d,α This represents the value corresponding to the F-distribution with confidence level α and degrees of freedom d and Nd, where g and h are introduced intermediate variables. This indicates that the value corresponds to a chi-square distribution with h degrees of freedom and α confidence level. Step 5: When it is necessary to use the process data collected at the current sampling time for welding quality monitoring, represent the process data collected at the current sampling time as a 1×5 dimensional data vector, and denote it as x. new Then for x new Each data point in the dataset is standardized to obtain a 1×5 dimensional standardized data vector with a mean of 0 and a standard deviation of 1, denoted as . Recalculate The score vector and error vector are denoted as s. new and e new , Then, the first and second monitoring indicators are calculated, denoted as D and E respectively, where D = s new Λ -1 s new T , Finally, determine whether the condition D≤D is satisfied. lim And E≤E lim If the conditions are met, the welding quality of the resistance spot weld corresponding to the process data collected at the current sampling time is considered normal; if the conditions are not met, the welding quality of the resistance spot weld corresponding to the process data collected at the current sampling time is considered abnormal, and the position of the resistance spot weld is marked.

2. The welding quality monitoring method based on multi-nearest neighbor feature analysis of resistance spot welding data according to claim 1, characterized in that... In step 2.2, N nearest neighbors were found based on three aspects: spatial distance, sampling time, and spatial angle. i The process of creating each row vector is as follows: Step 2.2a: For spatially nearest neighbors, calculate Except Other row vectors and Spatial distance between them There are N-1 corresponding spatial distances. The j-th row vector in The spatial distance between them is denoted as δ i,j , Then to The corresponding N-1 spatial distances δ i,1 ,δ i,2 ,…,δ i,i-1 ,δ i,i+1 ,…,δ i,N Sort by numerical value in ascending order; then from... Select the row vectors that have spatial distances to the top k sorted rows, and use these selected row vectors as the spatial nearest neighbor row vectors, recording their row indices; where j represents The row index of the row vector in the diagram, j = 1, 2, ..., N and j ≠ i, the symbol "||" indicates the length of the vector, which is also the Euclidean distance calculation symbol. express The j-th row vector in δ i,1 express The first row vector in The spatial distance between them, δ i,2 express The second row vector in The spatial distance between them, δ i,i-1 express The (i-1)th row vector in The spatial distance between them, δ i,i+1 Represents the (i+1)th row vector in X and The spatial distance between them, δ i,N express The Nth row vector in The spatial distance between them, where k represents the set number of nearest neighbors, and the value of k ranges from [6,12]. For sampling time proximity, based on the order of sampling times, from Find the collection from The previous at most t row vectors and the data collected from For the next t row vectors, these found row vectors are taken as the nearest neighbor row vectors at the sampling time, and their row indices are recorded; where the order of sampling times is assumed to be... If the row indices of the row vectors in the array are in ascending order, then: when t+1≤i≤Nt, the data collected from... The previous t row vectors and the data collected from For the subsequent t row vectors, when i < t+1, it is possible to find the data collected from... The following t row vectors are collected from The number of row vectors previously was 0 to t-1. When i > Nt, it is possible to find the data collected from... The previous t row vectors were collected from The number of subsequent row vectors is from 0 to t-1, where t represents the number of nearest neighbors at the set sampling time, and the value of t ranges from t∈[1,4]. For spatial angular nearest neighbors, calculate Except Other row vectors and The cosine value of the angle between them, There are N-1 corresponding angle cosine values. The j-th row vector in The cosine value of the angle between them is denoted as cosθ. i,j , Then from Find the angle cosine values ​​whose absolute value is greater than the corresponding N-1 angle cosine values. All angle cosine values; then from Select the row vectors corresponding to all the found angle cosine values, and use these selected row vectors as the spatial angle nearest neighbor row vectors, recording their row index numbers; where θ i,j express The j-th row vector in The spatial angle between them, with the superscript "T" indicating the transpose of a vector or matrix. This indicates the set spatial angle nearest neighbor threshold. The range of values ​​is Step 2.2b: Delete duplicate row indexes from all row indexes recorded for spatial distance nearest neighbors, sampling time nearest neighbors, and spatial angle nearest neighbors, leaving only one. The set of all remaining row indexes after the deletion operation is θ. i ,from The middle travel index number belongs to θ i All row vectors of are equal to . Similar N i Row vectors.

Images