A method for monitoring insert injection molding quality based on batch correlation analysis

Abstract

The present invention discloses a method for monitoring the quality of insert injection molding based on batch correlation analysis. The method selects sampling data of multiple normal production batches and performs standardization processing; obtains the relevant characteristic vector corresponding to each standardized batch data matrix by analyzing the correlation between each standardized batch data matrix and other standardized batch data matrices; obtains the upper and lower limits of characteristic changes based on the relevant characteristic vectors corresponding to all standardized batch data matrices; saves parameters for implementing online quality monitoring; utilizes the upper and lower limits of characteristic changes and the parameters for implementing online quality monitoring to implement online monitoring of the quality of insert injection molding; the method has the advantages of small calculation amount and fast response.

Description

Technical Field

[0001] The present invention relates to an intelligent monitoring technology for an industrial process, and in particular to an insert injection molding quality monitoring method based on batch correlation analysis. Background Art

[0002] In the production of automotive parts, the most commonly used method is to produce automotive inserts with specific shapes using injection molding machines. In the process of producing automotive inserts, the injection molding machine can produce the insert product in one step through four main processes: injection, pressure holding, cooling, and ejection. The entire production process is short, and any defects in the production process will directly lead to quality problems in the corresponding insert product. However, due to the extremely short production time of a single insert, even if a production anomaly is discovered, it cannot be corrected immediately. Therefore, the monitoring of insert quality is more focused on whether there are problems with the quality of the terminal insert product. With the development of industrial Internet of Things technology and distributed sensor technology, intelligent instruments such as temperature sensors, displacement sensors, and pressure sensors have been introduced into the injection molding process. Relevant production data is collected and stored in the enterprise energy management system, providing complete data for data-driven intelligent monitoring methods.

[0003] During the injection molding process of automotive inserts, there are measurement variables such as temperature, pressure, speed, flow, and position that can be collected at short intervals. These measurement variables are key factors affecting the quality of injection-molded products, and fluctuations in these measurement variables can be used to determine whether the injection molding process of automotive inserts is operating normally. Due to factors such as changes in the external environment, changes in production plans, and the inherent characteristics of the injection molding machine itself, the injection molding process of automotive inserts has multiple stable operating conditions and requires frequent changes in production conditions, resulting in multimodal characteristics of the sampled data. Therefore, traditional multivariate statistical process monitoring methods cannot be directly applied. In addition, the short injection molding process of automotive inserts further requires corresponding monitoring methods with low computational complexity and fast response, which makes a class of nonlinear modeling methods based on kernel learning or neural networks unusable.

[0004] Considering the batch nature of the injection molding process for automotive inserts, a normal batch step corresponds to the production of qualified automotive inserts, and sensors can generate sampling data for an entire production batch. The sampling data from different normal batches can reflect the normal variations in the injection molding process for automotive inserts. Therefore, processing the three-dimensional data matrix composed of sampling data from multiple normal batches of automotive inserts during the injection molding process is the first step in implementing data-driven quality monitoring. Furthermore, because a single production batch is short but contains a large amount of sampling data, it is not possible to perform quality monitoring on a single set of sampling data. Instead, the quality of the corresponding inserts can be directly determined based on the sampling data from the entire production batch. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method for monitoring the quality of insert injection molding based on batch correlation analysis. The method analyzes the correlation between the sampling data of different normal production batches in the automobile insert injection molding process, and expresses the data change characteristics of each normal production batch through a simple and effective linear transformation method. Then, by monitoring the changes in the relevant characteristic vectors of the new batch, it reflects whether there are problems with the quality of the insert products of the new batch. The method has low calculation amount and fast response.

[0006] The technical solution adopted by the present invention to solve the above technical problems is: a method for monitoring the quality of insert injection molding based on batch correlation analysis, characterized by comprising the following steps:

[0007] Step 1: Select N sampling data of normal production batches from the historical sampling database corresponding to the insert injection molding process; then arrange the sampling data of each normal production batch in the order of sampling time to form a K×M dimensional batch data matrix, and obtain N K×M dimensional batch data matrices in total. The nth batch data matrix is recorded as X n ; Wherein, N ≥ 100, K represents the number of groups of sampling data of each normal production batch, M represents the number of sensors set during the insert injection molding process, n = 1, 2, ..., N;

[0008] Step 2: Standardize each batch data matrix to obtain the corresponding standardized batch data matrix, and convert X n The corresponding standardized batch data matrix is recorded as The acquisition process is as follows: N elements at the same position in the N batch data matrix form a group of elements, with a total of K×M groups of elements, and the N elements in the k-th row and m-th column of the N batch data matrix are recorded as α 1,k,m ,α 2,k,m ,…,α N,k,m ; Then calculate the mean and standard deviation of each group of elements, and use α 1,k,m ,α 2,k,m ,…,α N,k,m The mean and standard deviation of k,m and δ k,m ; Then obtain the mean value matrix and standard deviation matrix corresponding to the N batch data matrix, which are denoted as U and Θ respectively. The element in the kth row and mth column of U is equal to μ k,m , the k-th row and m-th column element in Θ is equal to δ k,m ; Finally, Among them, k=1,2,…,K, m=1,2,…,M,α 1,k,m Represents the k-th row and m-th column element in the first batch of data matrix, α 2,k,m Represents the k-th row and m-th column element in the second batch of data matrix, α N,k,mrepresents the k-th row and m-th column element in the N-th batch data matrix, and the dimensions of U and Θ are K×M. The dimension is K×M;

[0009] Step 3: By analyzing the correlation between each standardized batch data matrix and other standardized batch data matrices, the corresponding eigenvectors of each standardized batch data matrix are obtained. The corresponding related eigenvector is recorded as w n ; Among them, w n The dimension is M×1;

[0010] Step 4: The relevant eigenvectors corresponding to the N standardized batch data matrices are recorded as w1, w2, ..., w N ; Then w1,w2,…,w N Merge into an M×N dimension related feature matrix, denoted as W, W=[w1,w2,…,w N ]; then calculate the correlation coefficient matrix of W, recorded as R, R = WW T ; Then let Φ = W T R -1 W, calculate the mean u and standard deviation σ of all elements on the main diagonal of Φ; finally calculate the upper and lower limits of the feature change, which are recorded as φ up and φ low ,φ up =u+3σ,φ low =u-3σ; where w1 represents the first standardized batch data matrix The corresponding related eigenvector, w2 represents the second standardized batch data matrix The corresponding related eigenvector, w N Represents the Nth standardized batch data matrix The corresponding related eigenvector, Φ is the introduced intermediate matrix;

[0011] Step 5: Save U, Θ, R, φ up and φ low Used to implement online quality monitoring;

[0012] Step 6: Implement online monitoring of insert injection molding quality. The specific process is: obtain the sampling data of the latest production batch of the insert injection molding process, and arrange the sampling data of the latest production batch in the order of sampling time to form a K×M-dimensional batch data matrix, recorded as X new ; then according to Get X new The corresponding standardized batch data matrix Reanalysis and The correlation between The corresponding related eigenvector is denoted as w new , w new The process of obtaining is: And order Solve the generalized eigenvalue problem L new β=λG new All eigenvalues in β and their corresponding eigenvectors, the smallest eigenvalue is denoted as λ new,min , change λ new,min The corresponding eigenvector β new,min ,according to Calculate w new ; Then calculate the quality index of the latest production batch, denoted as φ new , Finally judge φ new Whether the condition φ is met low ≤φ new ≤φ up If it is satisfied, it is considered that there is no problem with the insert molding quality in the latest production batch of the insert molding process; if it is not satisfied, it is considered that there is a problem with the insert molding quality in the latest production batch of the insert molding process; among them, Represents the standardized batch data matrix corresponding to the first batch data matrix X1, Represents the standardized batch data matrix corresponding to the second batch data matrix X2, Represents the Nth batch data matrix X N The corresponding normalized batch data matrix, G new and L new is the introduced intermediate matrix.

[0013] In the step 3, w n The process of obtaining is: And order Then solve the generalized eigenvalue problem L n β=λG n All eigenvalues in β and their corresponding eigenvectors, the smallest eigenvalue is denoted as λ min , change λ min The corresponding eigenvector is denoted as β min ; Then according to Calculate w n Among them, G n and L n is the introduced intermediate matrix, G n and L n The dimension is M×M, and the superscript "T" indicates the transpose of a vector or matrix. Represents the i-th batch data matrix X iThe corresponding standardized batch data matrix, λ represents the eigenvalue, β represents the eigenvector, i = 1, 2, ..., N and i ≠ n.

[0014] Compared with the prior art, the advantages of the present invention are:

[0015] The method of the present invention analyzes the correlation between production batches during the insert injection molding process, thereby utilizing the sampling data of the entire production batch to reflect the quality issues of the corresponding insert products. Compared to traditional methods, the method of the present invention does not process the batch data of the insert injection molding process according to the batch expansion, but instead innovatively proposes a batch standardization method to solve the batch data preprocessing problem of the insert injection molding process. More importantly, the process of analyzing the correlation between batch data matrices involved in the method of the present invention is a new data feature analysis algorithm that can specifically reflect the quality issues of insert injection molding from the perspective of inter-batch correlation. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 Flow chart for the implementation of the method of the present invention;

[0017] Figure 2 This is a schematic diagram of the five types of sensors used in the injection molding process of automotive inserts: temperature, pressure, speed, flow, and position. DETAILED DESCRIPTION

[0018] The present invention will be described in further detail below with reference to the accompanying drawings and embodiments.

[0019] The present invention proposes a method for monitoring the quality of insert injection molding based on batch correlation analysis, and its implementation flow chart is as follows: Figure 1 As shown, it includes the following steps:

[0020] Step 1: Select N sampling data of normal production batches from the historical sampling database corresponding to the insert injection molding process; then arrange the sampling data of each normal production batch in the order of sampling time to form a K×M dimensional batch data matrix, and obtain N K×M dimensional batch data matrices in total. The nth batch data matrix is recorded as X n; Wherein, N ≥ 100, such as N = 150, K represents the number of groups of sampling data for each normal production batch, and the K value is relatively large for large-sized inserts, and M represents the number of sensors set during the insert injection molding process, n = 1, 2, ..., N; Here, the historical sampling database corresponding to the insert injection molding process is composed of sampling data of multiple previous production batches, including sampling data of normal production batches and sampling data of abnormal production batches; the setting of the sensor is determined according to the measurement variables that need to be collected. Generally, it is necessary to collect temperature, pressure, speed, flow, position, etc., that is, it is necessary to set temperature sensors, pressure sensors, speed sensors, flow sensors, displacement sensors, etc. At this time, a group of sampling data includes temperature, pressure, speed, flow, position, etc.

[0021] Figure 2 Given that the injection molding process of automotive inserts involves five types of sensors, namely temperature, pressure, speed, flow, and position, sampling data of M = 13 measurement variables can be collected at a sampling interval of 20 ms, as shown in Table 1. Each production batch takes a total of 24 seconds, and K = 1200 sets of sampling data can be collected for each production batch.

[0022] Table 1 Measured variables during injection molding

[0023]

[0024] Step 2: Standardize each batch data matrix to obtain the corresponding standardized batch data matrix, and convert X n The corresponding standardized batch data matrix is recorded as The acquisition process is as follows: N elements at the same position in the N batch data matrix form a group of elements, with a total of K×M groups of elements, and the N elements in the k-th row and m-th column of the N batch data matrix are recorded as α 1,k,m ,α 2,k,m ,…,α N,k,m ; Then calculate the mean and standard deviation of each group of elements, and use α 1,k,m ,α 2,k,m ,…,α N,k,m The mean and standard deviation of k,m and δ k,m ; Then obtain the mean value matrix and standard deviation matrix corresponding to the N batch data matrix, which are denoted as U and Θ respectively. The element in the kth row and mth column of U is equal to μ k,m , the k-th row and m-th column element in Θ is equal to δ k,m ; Finally, Among them, k=1,2,…,K, m=1,2,…,M,α 1,k,m Represents the k-th row and m-th column element in the first batch of data matrix, α 2,k,mRepresents the k-th row and m-th column element in the second batch of data matrix, α N,k,m represents the k-th row and m-th column element in the N-th batch data matrix, and the dimensions of U and Θ are K×M. The dimension is K×M.

[0025] Step 3: By analyzing the correlation between each standardized batch data matrix and other standardized batch data matrices, the corresponding eigenvectors of each standardized batch data matrix are obtained. The corresponding related eigenvector is recorded as w n ; Among them, w n The dimension is M×1.

[0026] In this embodiment, in step 3, w n The process of obtaining is: And order Then solve the generalized eigenvalue problem L n β=λG n All eigenvalues in β and their corresponding eigenvectors, the smallest eigenvalue is denoted as λ min , change λ min The corresponding eigenvector is denoted as β min ; Then according to Calculate w n Among them, G n and L n is the introduced intermediate matrix, G n and L n The dimension is M×M, and the superscript "T" indicates the transpose of a vector or matrix. Represents the i-th batch data matrix X i The corresponding standardized batch data matrix, λ represents the eigenvalue, β represents the eigenvector, i = 1, 2, ..., N and i ≠ n.

[0027] It should be noted that, when analyzing a certain standardized batch data matrix (for example, the nth standardized batch data matrix ) and other standardized batch data matrices The correlation between them is to determine the corresponding correlation feature vector w n , thereby ensuring Under the premise of The correlation with other standardized batch data matrices is minimized; it can be quantified as a constrained minimization problem as shown below: After introducing the Lagrange multiplier λ, the Lagrange function can be constructed Then calculate J relative to w n And the partial derivative of λ, that is: When the partial derivatives are all equal to 0, we can get and Multiply on both sides of the equal sign After that, you can get Therefore, λ is the minimization target. Since the eigenvector in the generalized eigenvalue problem has only directional meaning but no scale meaning, the solved eigenvector can be normalized to obtain the final related eigenvector. Represents the n-1th batch data matrix X n-1 The corresponding normalized batch data matrix, Represents the n+1th batch data matrix X n+1 The corresponding standardized batch data matrix, min is the minimum function, and J represents the Lagrangian function value.

[0028] Step 4: The relevant eigenvectors corresponding to the N standardized batch data matrices are recorded as w1, w2, ..., w N ; Then w1,w2,…,w N Merge into an M×N dimension related feature matrix, denoted as W, W=[w1,w2,…,w N ]; then calculate the correlation coefficient matrix of W, recorded as R, R = WW T ; Then let Φ = W T R -1 W, calculate the mean u and standard deviation σ of all elements on the main diagonal of Φ; finally calculate the upper and lower limits of the feature change, which are recorded as φ up and φ low ,φ up =u+3σ,φ low =u-3σ; where w1 represents the first standardized batch data matrix The corresponding related eigenvector, w2 represents the second standardized batch data matrix The corresponding related eigenvector, w N Represents the Nth standardized batch data matrix The corresponding related eigenvector, Φ is the introduced intermediate matrix.

[0029] Step 5: Save U, Θ, R, φ up and φ low Used to implement online quality monitoring.

[0030] Step 6: Implement online monitoring of insert injection molding quality. The specific process is: obtain the sampling data of the latest production batch of the insert injection molding process, and arrange the sampling data of the latest production batch in the order of sampling time to form a K×M-dimensional batch data matrix, recorded as X new ; then according to Get X new The corresponding standardized batch data matrix Reanalysis and The correlation between The corresponding related eigenvector is denoted as w new , w new The process of obtaining is: And order Solve the generalized eigenvalue problem L new β=λG new All eigenvalues in β and their corresponding eigenvectors, the smallest eigenvalue is denoted as λ new,min , change λ new,min The corresponding eigenvector β new,min ,according to Calculate w new ; Then calculate the quality index of the latest production batch, denoted as φ new , Finally judge φ new Whether the condition φ is met low ≤φ new ≤φ up If it is satisfied, it is considered that there is no problem with the insert molding quality in the latest production batch of the insert molding process; if it is not satisfied, it is considered that there is a problem with the insert molding quality in the latest production batch of the insert molding process; among them, Represents the standardized batch data matrix corresponding to the first batch data matrix X1, Represents the standardized batch data matrix corresponding to the second batch data matrix X2, Represents the Nth batch data matrix X N The corresponding normalized batch data matrix, G new and L new is the introduced intermediate matrix.

Claims

1. A method for monitoring the quality of insert injection molding based on batch correlation analysis, characterized in that The following steps are involved: Step 1: Select N sampling data of normal production batches from the historical sampling database corresponding to the insert injection molding process; Then, the sampling data of each normal production batch are arranged in the order of sampling time to form a K×M dimensional batch data matrix, and a total of N K×M dimensional batch data matrices are obtained. The nth batch data matrix is recorded as X n ; Wherein, N ≥ 100, K represents the number of groups of sampling data of each normal production batch, M represents the number of sensors set during the insert injection molding process, n = 1, 2, ..., N; Step 2: Standardize each batch data matrix to obtain the corresponding standardized batch data matrix, and convert X n The corresponding standardized batch data matrix is recorded as The acquisition process is as follows: N elements at the same position in the N batch data matrix form a group of elements, with a total of K×M groups of elements, and the N elements in the k-th row and m-th column of the N batch data matrix are recorded as α 1,k,m ,α 2,k,m ,…,α N,k,m ; Then calculate the mean and standard deviation of each group of elements, and use α 1,k,m ,α 2,k,m ,…,α N,k,m The mean and standard deviation of k,m and δ k,m ; Then obtain the mean value matrix and standard deviation matrix corresponding to the N batch data matrix, which are denoted as U and Θ respectively. The element in the kth row and mth column of U is equal to μ k,m , the k-th row and m-th column element in Θ is equal to δ k,m ; Finally, Among them, k=1,2,…,K, m=1,2,…,M,α 1,k,m Represents the k-th row and m-th column element in the first batch of data matrix, α 2,k,m Represents the k-th row and m-th column element in the second batch of data matrix, α N,k,m represents the k-th row and m-th column element in the N-th batch data matrix, and the dimensions of U and Θ are K×M. The dimension is K×M; Step 3: By analyzing the correlation between each standardized batch data matrix and other standardized batch data matrices, the corresponding eigenvectors of each standardized batch data matrix are obtained. The corresponding related eigenvector is recorded as w n ; Among them, w n The dimension is M×1; Step 4: The relevant eigenvectors corresponding to the N standardized batch data matrices are recorded as w1, w2, ..., w N ; Then w1,w2,…,w N Merge into an M×N dimension related feature matrix, denoted as W, W=[w1,w2,…,w N ]; then calculate the correlation coefficient matrix of W, recorded as R, R = WW T ; Then let Φ = W T R -1 W, calculate the mean u and standard deviation σ of all elements on the main diagonal of Φ; finally calculate the upper and lower limits of the feature change, which are recorded as φ up and φ low ,φ up =u+3σ,φ low =u-3σ; where w1 represents the first standardized batch data matrix The corresponding related eigenvector, w2 represents the second standardized batch data matrix The corresponding related eigenvector, w N Represents the Nth standardized batch data matrix The corresponding related eigenvector, Φ is the introduced intermediate matrix; Step 5: Save U, Θ, R, φ up and φ low Used to implement online quality monitoring; Step 6: Implement online monitoring of insert injection molding quality. The specific process is: obtain the sampling data of the latest production batch of the insert injection molding process, and arrange the sampling data of the latest production batch in the order of sampling time to form a K×M-dimensional batch data matrix, recorded as X new ; then according to Get X new The corresponding standardized batch data matrix Reanalysis and The correlation between The corresponding related eigenvector is denoted as w new , w new The process of obtaining is: And order Solve the generalized eigenvalue problem L new β=λG new All eigenvalues in β and their corresponding eigenvectors, the smallest eigenvalue is denoted as λ new,min , change λ new,min The corresponding eigenvector β new,min ,according to Calculate w new ; Then calculate the quality index of the latest production batch, denoted as φ new , Finally judge φ new Whether the condition φ is met low ≤φ new ≤φ up If it is satisfied, it is considered that there is no problem with the insert molding quality in the latest production batch of the insert molding process; if it is not satisfied, it is considered that there is a problem with the insert molding quality in the latest production batch of the insert molding process; among them, Represents the standardized batch data matrix corresponding to the first batch data matrix X1, Represents the standardized batch data matrix corresponding to the second batch data matrix X2, Represents the Nth batch data matrix X N The corresponding normalized batch data matrix, G new and L new is the introduced intermediate matrix.

2. The method for monitoring the quality of insert injection molding based on batch correlation analysis according to claim 1, characterized in that In the step 3, w n The process of obtaining is: And order Then solve the generalized eigenvalue problem L n β=λG n All eigenvalues in β and their corresponding eigenvectors, the smallest eigenvalue is denoted as λ min , change λ min The corresponding eigenvector is denoted as β min ; Then according to Calculate w n Among them, G n and L n is the introduced intermediate matrix, G n and L n The dimension is M×M, and the superscript "T" indicates the transpose of a vector or matrix. Represents the i-th batch data matrix X i The corresponding standardized batch data matrix, λ represents the eigenvalue, β represents the eigenvector, i = 1, 2, ..., N and i ≠ n.

Images