Die-casting quality related variable importance analysis method based on multivariable standardized embedding and feature fusion

Abstract

The invention discloses a pressure casting quality related variable importance analysis method based on multivariable standardized embedding and feature fusion. The method comprises the following steps: firstly, performing data cleaning, standardization and matching on pressure casting process data and quality label data; then a die-casting variable importance identification model comprising a variable standardization embedding layer, a variable selection layer and a quality classification layer is constructed and trained, and the variable standardization embedding layer converts different types of variables into tensors in the same shape in different embedding modes according to time-varying features and discrete features of the die-casting data; the variable selection layer calculates a variable selection weight based on a feature obtained by fusing tensors of the same shape of each variable or based on a learnable weight matrix; performing weighted summation on the same-shape tensor of each variable to generate a fused implicit expression; and the quality classification layer takes the fused implicit expression as input and outputs a quality label. According to the method, multivariable standardized embedding and feature fusion are innovatively combined, and the influence degree of each process parameter on the quality can be quantified.

Description

Technical Field

[0001] The present invention relates to the fields of intelligent manufacturing and quality monitoring, and particularly to a method for analyzing the importance of die-casting quality-related variables based on multi-variable standardization embedding and feature fusion. Background Art

[0002] Die casting, as an efficient and precise metal forming process, is widely used in manufacturing fields such as automotive, aerospace, and electronics. During the die-casting process, multiple process parameters (such as pressure, temperature, stroke, etc.) and equipment state variables (such as running time, beat number, etc.) are involved, and these variables jointly determine the quality of the final product. However, the die-casting process has a high degree of non-linearity and time-variability, and there may be complex interactions between variables, resulting in quality fluctuations and defects. Traditional quality control methods mainly rely on empirical rules and the monitoring of single variables, making it difficult to comprehensively capture the complex relationships between multi-variables and unable to accurately predict and classify product quality.

[0003] In recent years, with the development of industrial big data and artificial intelligence technologies, data-driven quality analysis methods have gradually become a research hotspot. In the prior art, some methods attempt to classify die-casting quality through machine learning models, but due to the multi-source heterogeneity of die-casting data (such as the coexistence of continuous numerical data and discrete enumeration data), direct modeling often has poor effects. In addition, existing methods usually ignore the importance of variable selection, resulting in increased sensitivity of the model to redundant or irrelevant variables, reducing the classification accuracy and interpretability. Therefore, there is an urgent need for a method that can effectively process multi-variable data, extract key features, and achieve high-quality classification. Summary of the Invention

[0004] To solve the deficiencies in the prior art and improve the intelligent level and accuracy of die-casting quality monitoring, the present invention proposes a method for analyzing the importance of die-casting quality-related variables based on multi-variable standardization embedding and feature fusion. By combining data preprocessing, variable standardization embedding, feature fusion, variable selection, classification model construction, variable importance analysis, and deep learning technologies, it aims to solve the complexity and uncertainty of multi-variable data in the die-casting production process and achieve accurate classification of die-casting quality and importance analysis of related variables. The present invention is applicable to die-casting process optimization, quality monitoring, and the design and application of intelligent manufacturing systems, and has broad technical application prospects.

[0005] The object of the present invention is achieved by the following technical solutions:

[0006] A method for analyzing the importance of die-casting quality-related variables based on multi-variable standardization embedding and feature fusion, comprising the following steps:

[0007] S1: Obtain die-casting process data and quality label data within a predetermined range, and pre-process the data, including data cleaning, standardization, and matching operations;

[0008] S2: Construct and train a die-casting variable importance identification model, which includes a variable standardization embedding layer, a variable selection layer and a quality classification layer; wherein the variable standardization embedding layer is used to convert different types of variables into tensors of the same shape using different embedding methods according to the time-varying characteristics and discrete characteristics of the die-casting data; the variable selection layer calculates the variable selection weight based on the features obtained by fusion of the same-shape tensors of each variable or based on a learnable weight matrix, and then performs weighted summation on the same-shape tensors of each variable to generate a fused implicit expression; the quality classification layer takes the fused implicit expression as input and outputs a quality label close to the true value; after training, the die-casting variable importance identification model outputs the predicted quality label, and the variable selection layer outputs the importance of related variables.

[0009] Furthermore, the die-casting process data includes setting parameters, sensor acquisition parameters and statistical data;

[0010] Wherein, the setting parameters are parameters related to the die-casting process set for the die-casting machine during die-casting production;

[0011] The sensor acquisition parameters include actual measured values corresponding to the set parameters in the die-casting process, as well as other actual measured values that can be measured and monitored in the die-casting process;

[0012] The statistical data include the operating time, date, number of cycles, number of products produced and number of unqualified products produced of the die casting machine;

[0013] The quality label data is the change value of the number of unqualified products produced.

[0014] Furthermore, in the data preprocessing stage, the sample data is first divided according to the mutation difference of the number of products produced, and the sample V = {V1, V2, ..., V m}∈R m×T×1 ; Then, samples with abnormal duration, abnormal change in the number of produced samples, and abnormal change in the number of unqualified samples produced are eliminated; subsequently, the maximum sequence length is determined according to the duration of the sample, and a mask matrix is generated; for non-enumerated data, the Z-score normalization method is used for standardization; finally, data matching is performed, that is, the timestamp of the quality label data is matched with the end timestamp of the die-casting process data sample for the nearest value, and samples whose timestamp difference exceeds the preset threshold are eliminated to ensure the accuracy and consistency of the data.

[0015] Furthermore, the implementation process of the variable standardization embedding layer is as follows:

[0016] Given a sample V = {V1, V2, …, V m} ∈ R m×T×1 and an implicit expression dimension f h , where m is the number of variables in the sample, V i = {V i,1 , V i,2 , …, V i,T} ∈ R T×1 represents the sampling value of the i-th variable, and V i,t ∈ R 1 represents the sampling value of the i-th variable at time t;

[0017] The variable standardization embedding layer M vf is expressed as that is, mapping the feature dimension of each variable V i to a unified dimension size, specifically:

[0018] For continuous numerical variables, use a linear fully connected layer to expand its feature dimension to f h ;

[0019] For discrete enumeration variables, after passing through the embedding layer and the linear fully connected layer, the feature dimension is expanded to f h ;

[0020] Then, all the expanded variables are concatenated in the input order in the die-casting process data sample V = {V1, V2, …, V m} to obtain a tensor of the same shape

[0021] Furthermore, the variable selection layer M vs is expressed as that is, performing feature extraction and feature selection on the tensor V h of the same shape to generate a fused implicit variable Specifically:

[0022] (1) Use a gated residual network GRN to perform feature extraction on the sampling value h of the i-th variable in V at time t to obtain

[0023] (2) Obtain variable selection weights through the following two methods

[0024] a. Swap the dimensions of V h and flatten the last two dimensions to obtain Then generate through a gated residual network GRN, a linear fully connected layer, and a softmax function That is

[0025] b. Provide a learnable matrix w vs ∈R T×m , and generate it through the softmax function That is Variable selection weight Represents the importance of variables and is used for variable importance analysis;

[0026] (3) Use weighted summation to generate the implicit expression at time t Finally, obtain the fused implicit expression

[0027] Furthermore, the quality classification layer M c Is expressed as: That is, take the fused implicit expression As the input, convert H into a quality label probability vector Where N c Is the number of categories of quality labels, and select the label with the highest probability as the predicted quality label for output.

[0028] Furthermore, when training the die-casting variable importance identification model, the input is a sample V = {V1, V2,..., V m}∈R m×T×1 And the corresponding true quality label Among them, V is a real number matrix with T time steps, m variables, and each variable feature dimension is 1; Is a discrete enumeration value, which can be converted into a true quality label probability vector through one-hot encoding The output is the predicted quality label probability vector

[0029] By minimizing the cross-entropy loss function that measures the difference between the predicted value P and the true value Make the predicted value P as close as possible to the true value

[0030] A die-casting quality-related variable importance analysis device based on multivariate standardization embedding and feature fusion, including one or more processors for implementing the die-casting quality-related variable importance analysis method based on multivariate standardization embedding and feature fusion.

[0031] An electronic device, including:

[0032] One or more processors;

[0033] A storage device for storing one or more programs, which, when executed by the electronic device, enable the electronic device to implement a method for analyzing the importance of variables related to die-casting quality based on multivariate standardization embedding and feature fusion.

[0034] A computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements a method for analyzing the importance of variables related to die-casting quality based on multivariate standardization embedding and feature fusion.

[0035] The beneficial effects of the present invention are as follows:

[0036] The method for analyzing the importance of variables related to die-casting quality based on multivariate standardization embedding and feature fusion of the present invention can convert multi-source heterogeneous data in the die-casting process into a unified expression through the variable standardization embedding layer, solving the data compatibility problem; the variable selection layer automatically identifies key variables based on the learnable weight matrix or unified expression, reducing redundant interference and improving classification accuracy and interpretability; the classification model captures the complex relationships between multi-variables using implicit expressions to achieve accurate quality classification and reduce the misjudgment rate. At the same time, the variable selection weights provide a quantitative analysis of the impact of each variable on quality, providing a scientific basis for process optimization. This method does not rely on specific equipment or processes, is applicable to a variety of die-casting scenarios, and the modular design is convenient for expansion and integration. Through deep learning and data-driven technologies, it realizes the automation of data preprocessing, feature extraction, and classification analysis, reduces the dependence on manual experience, and significantly improves the intelligence and automation level of quality monitoring. The present invention can effectively improve the quality control ability and process optimization efficiency of die-casting production, providing important technical support for the field of intelligent manufacturing. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a schematic flow chart of the method for analyzing the importance of variables related to die-casting quality based on multivariate standardization embedding and feature fusion according to an embodiment of the present invention.

[0038] Figure 2 It is a schematic diagram of the variable standardization embedding layer and the variable selection layer according to a specific embodiment of the present invention.

[0039] Figure 3 It is a schematic diagram of variable importance obtained by applying the method of the present invention to a certain die-casting data set. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0040] The present invention will be described in detail below with reference to the drawings and preferred embodiments. The objectives and effects of the present invention will become more apparent. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0041] As Figure 1As shown, the method for analyzing the importance of variables related to die-casting quality based on multivariate standardization embedding and feature fusion provided by the present invention includes the following steps:

[0042] Step 1: Obtain die-casting process data and quality label data within a predetermined range, and preprocess the data, including data cleaning, standardization, and matching operations.

[0043] The specific implementation process is as follows: Obtain the actually collected die-casting process data and its corresponding quality label data. The die-casting process data includes set parameters, sensor acquisition parameters, and statistical data. Among them, the set parameters are the parameters related to the die-casting process set for the die-casting machine during die-casting production, such as pressure, stroke, etc.; the sensor acquisition parameters are determined by the die-casting machine brand and include the actual measured values corresponding to the set parameters during the die-casting process and other actual measured values that can be measured and monitored during the die-casting process; the statistical data covers the operating time, date, beat number, number of products produced, and number of unqualified products produced by the die-casting machine. The quality label data is the change value of the number of unqualified products produced.

[0044] In the data preprocessing stage, first divide the sample data according to the mutation difference of the number of products produced to obtain the sample V = {V1, V2, …, V m} ∈ R m×T×1 . Specifically, the production process of a product usually lasts from 45 to 90 seconds. During this period, the number of products produced remains unchanged. However, when the production of the previous product is completed and the production of the next product starts, the number of products produced will mutate. By capturing this mutation difference, the data can be reversely divided into independent samples according to the product production process. Then, samples with abnormal duration, abnormal change value of the number of produced samples, and abnormal change value of the number of unqualified produced samples are removed. Subsequently, determine the maximum sequence length according to the duration of the samples and generate a mask matrix. For non-enumerative data, use the Z-score normalization method for standardization processing. Finally, perform data matching, that is, match the timestamp of the quality label data with the end timestamp of the die-casting process data sample, and remove the samples with a time difference between the two timestamps exceeding the preset threshold to ensure the accuracy and consistency of the data.

[0045] Step 2: Construct and train a die-casting variable importance identification model, which includes a variable standardization embedding layer, a variable selection layer, and a quality classification layer; among them, the variable standardization embedding layer is used to transform different types of variables into tensors of the same shape by using different embedding methods according to the time-varying characteristics and discrete characteristics of die-casting data; the variable selection layer calculates the variable selection weights based on the features obtained by fusing the tensors of the same shape of each variable or based on a learnable weight matrix, and then performs weighted summation on the tensors of the same shape of each variable to generate a fused implicit expression; the quality classification layer takes the fused implicit expression as input and outputs a quality label close to the true value. After training, the die-casting variable importance identification model outputs the predicted quality label, and at the same time, the variable selection layer outputs the importance of the relevant variables.

[0046] The specific implementation process of the variable standardization embedding layer is as follows:

[0047] Given a sample V = {V1, V2, …, V m} ∈ R m×T×1 and the implicit expression dimension f h , where m is the number of variables in the sample, V i = {V i,1 , V i,2 , …, V i,T} ∈ R T×1 represents the sampling value of the i-th variable, and V i,t ∈ R 1 represents the sampling value of the i-th variable at time t.

[0048] Variable standardization embedding layer maps the feature dimension of each variable V i to a unified dimension size. As Figure 2 shown, for continuous numerical variables, a linear fully connected layer is used to expand its feature dimension to f h ; for discrete enumeration variables, after passing through the embedding layer and the linear fully connected layer, the feature dimension is expanded to f h . The expanded all variables are concatenated in the input order in the die-casting process data sample V = {V1, V2, …, V m} to obtain a tensor of the same shape

[0049] Variable selection layer performs feature extraction and feature selection on the tensor of the same shape V h to generate a fused implicit variable The specific process is as follows:

[0050] For the sampling value of the i-th variable in V h at time t Feature extraction is performed using a gated residual network GRN to obtain where GRN(x) = norm(9x + GELU(x)), norm is the normalization operation, GELU(x) = x·Φ(x), and Φ(x) is the cumulative distribution function of the standard normal distribution.

[0051] Variable selection weights can be obtained in the following two ways:

[0052] a. Transpose the dimensions of V h and flatten the last two dimensions to obtain Then generate i.e.,

[0053] b. Provide a learnable matrix w vs ∈R T×m and generate i.e., Variable selection weights can represent the importance of variables and be used for variable importance analysis.

[0054] For the t-th moment, use weighted summation to generate the implicit representation at the t-th moment Finally, obtain the fused implicit representation

[0055] The expression of the quality classification layer is: That is, take the fused implicit representation as the input and convert H into a quality label probability vector where N c is the number of categories of quality labels, and select the label with the highest probability as the predicted quality label for output. At the same time, the variable selection layer M vs will obtain the variable selection weights This weight will be used as the variable importance output.

[0056] During model training, the entire die-casting variable importance identification model (denoted as vf including the variable standardization embedding layer M vs and the quality classification model M c ) is jointly trained. The input is: the sample data of the die-casting process data sample V = {V1, V2,..., V } ∈ R m} and the corresponding true quality label m×T×1 and V is a real number matrix with T time steps, m variables, and each variable has a feature dimension of 1. is a discrete enumeration value, which can be transformed into a real quality label probability vector through one-hot encoding The output is the predicted quality label probability vector

[0057] By minimizing the cross-entropy loss function that measures the difference between the predicted value P and the true value make the predicted value P as close as possible to the true value

[0058] The present invention also provides a die-casting quality-related variable importance analysis device based on multivariate embedding and variable selection, including one or more processors for implementing the die-casting quality-related variable importance analysis method based on multivariate standardization embedding and feature fusion in the above embodiments. The embodiments of the die-casting quality-related variable importance analysis device based on multivariate embedding and variable selection of the present invention can be applied to any device with data processing capabilities, and the any device with data processing capabilities can be a device or apparatus such as a computer. The device embodiments can be implemented by software, or by hardware or a combination of software and hardware. Taking software implementation as an example, as a logically meaningful device, it is formed by the processor of any device with data processing capabilities reading the corresponding computer program instructions in the non-volatile memory into the memory for operation. From the hardware level, a hardware structure diagram of any device with data processing capabilities where the die-casting quality-related variable importance analysis device is located, in addition to the processor, memory, network interface, and non-volatile memory, the any device with data processing capabilities where the device is located in the embodiments usually also includes other hardware according to the actual functions of the any device with data processing capabilities, which will not be elaborated here.

[0059] An embodiment of the present invention further provides a computer-readable storage medium, on which a program is stored. When the program is executed by a processor, it implements the method for analyzing the importance of variables related to die-casting quality based on multivariate standardization embedding and feature fusion in the above embodiment. The computer-readable storage medium may be an internal storage unit of any device having data processing capabilities described in any of the foregoing embodiments, such as a hard disk or a memory. The computer-readable storage medium may also be an external storage device, such as a plug-in hard disk, a SmartMedia card (SMC), an SD card, a Flash card, etc. equipped on the device. Further, the computer-readable storage medium may also include both an internal storage unit of any device having data processing capabilities and an external storage device. The computer-readable storage medium is used to store the computer program and other programs and data required by any device having data processing capabilities, and may also be used to temporarily store the data that has been output or will be output.

[0060] The effectiveness of the present invention is verified below by combining a specific die-casting data set. This die-casting data set is collected by Internet of Things devices with a collection frequency of 1 s. There are eigenvalue representing the production quantity of die-casting products in the data set, and samples are divided by this eigenvalue. The data set is divided into a training set, a validation set and a test set in a ratio of 8:1:1. It is specified that the length of each sample is at most 60, and if it exceeds this length, 60 sampling steps are intercepted from the back to the front. After multiple rounds of training, the obtained variable selection weight distribution is as Figure 3 shown. The results show that in the die-casting process data, "BoostPrs" is the variable most related to die-casting quality.

[0061] Those of ordinary skill in the art can understand that the above are only preferred examples of the invention and are not used to limit the invention. Although the invention has been described in detail with reference to the foregoing examples, for those skilled in the art, they can still modify the technical solutions described in the foregoing examples, or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, etc. made within the spirit and principle of the invention shall be included within the protection scope of the invention.

Claims

1. A method for analyzing the importance of variables related to die-casting quality based on multi-variable standardization embedding and feature fusion, characterized in that, The steps include: S1: Obtain die-casting process data and quality label data within a predetermined range, and pre-process the data, including data cleaning, standardization, and matching operations; S2: Construct and train a die-casting variable importance identification model, which includes a variable standardization embedding layer, a variable selection layer and a quality classification layer; wherein the variable standardization embedding layer is used to convert different types of variables into tensors of the same shape using different embedding methods according to the time-varying characteristics and discrete characteristics of the die-casting data; the variable selection layer calculates the variable selection weight based on the features obtained by fusion of the same-shape tensors of each variable or based on a learnable weight matrix, and then performs weighted summation on the same-shape tensors of each variable to generate a fused implicit expression; the quality classification layer takes the fused implicit expression as input and outputs a quality label close to the true value; after training, the die-casting variable importance identification model outputs the predicted quality label, and the variable selection layer outputs the importance of related variables.

2. The importance analysis method of die-casting quality-related variables based on multivariate standardization embedding and feature fusion according to claim 1, characterized in that The die-casting process data includes setting parameters, sensor acquisition parameters and statistical data; Wherein, the setting parameters are parameters related to the die-casting process set for the die-casting machine during die-casting production; The sensor acquisition parameters include actual measured values corresponding to the set parameters in the die-casting process, as well as other actual measured values that can be measured and monitored in the die-casting process; The statistical data include the operating time, date, number of cycles, number of products produced and number of unqualified products produced of the die casting machine; The quality label data is the change value of the number of unqualified products produced.

3. The importance analysis method of die casting quality-related variables based on multi-variable standardization embedding and feature fusion according to claim 1, characterized in that In the data preprocessing stage, first, the sample data is divided according to the mutation difference of the number of products already produced, and the sample V = {V1, V2, …, V m} ∈ R m×T×1 ; Then, samples with abnormal duration, abnormal change in the number of samples produced, and abnormal change in the number of unqualified samples produced are eliminated; Subsequently, the maximum sequence length is determined according to the duration of the sample, and a mask matrix is generated. For non-enumerated data, the Z-score normalization method is used for standardization. Finally, data matching is performed, that is, the timestamp of the quality label data is matched with the end timestamp of the die-casting process data sample to the nearest value, and samples whose timestamp difference exceeds the preset threshold are eliminated to ensure the accuracy and consistency of the data.

4. The method for analyzing the importance of variables related to die-casting quality based on multivariate standardization embedding and feature fusion according to claim 1, characterized in that The implementation process of the variable standardization embedding layer is as follows: Given a sample \(V = \{V_1, V_2, \ldots, V\) m \} \in \mathbb{R} m×T×1 and an implicit expression dimension \(f\) h , where \(m\) is the number of variables in the sample, \(V\) i = \{V i,1 , V i,2 , \ldots, V i,T \} \in \mathbb{R} T×1 represents the sampling value of the \(i\)-th variable, and \(V\) i,t \in \mathbb{R} 1 represents the sampling value of the \(i\)-th variable at time \(t\); Variable standardization embedding layer M vf Denoted as That is, each variable V i The feature dimension of is mapped to a unified dimension size, specifically: For continuous numerical variables, use a linear fully connected layer to expand its feature dimension to f h ; For discrete enumerated variables, after passing through the embedding layer and the linear fully connected layer, the feature dimension is expanded to f h ; Then, splice all the expanded variables according to their input order in the die-casting process data sample V = {V1, V2, …, V m}, to obtain a tensor of the same shape 5. The die casting quality related variable importance analysis method based on multivariate standardized embedding and feature fusion according to claim 4 is characterized in that: The variable selection layer M vs is expressed as that is, for the tensor V with the same shape h feature extraction and feature selection are performed to generate a fused implicit variable Specifically (1) For V h The sampled value of the i-th variable in V at time t Feature extraction is performed using a gated residual network GRN to obtain (2) Obtain the variable selection weights in the following two ways a. Swap the dimensions of V h and flatten the last two dimensions to obtain Then generate through the gated residual network GRN, linear fully connected layer and softmax function That is b. Provide a learnable matrix w vs ∈R T×m , and generate through the softmax function That is Variable selection weights represent the importance of variables and are used for variable importance analysis; (3) Generate the implicit representation at time t using weighted summation Finally, obtain the fused implicit representation 6. The importance analysis method of die-casting quality-related variables based on multivariate standardization embedding and feature fusion according to claim 5, wherein The quality classification layer M c is expressed as: That is, taking the fused implicit expression as the input, convert H into a quality label probability vector where N c is the number of categories of quality labels, and select the label with the highest probability as the predicted quality label of the output.

7. The importance analysis method of die-casting quality-related variables based on multivariate standardization embedding and feature fusion according to claim 6, characterized in that When training the die-casting variable importance identification model, the input is a sample V of die-casting process data = {V1, V2, …, V m} ∈ R m×T×1 and the corresponding true quality label where V is a real number matrix with T time steps, m variables, and each variable feature dimension of 1; is a discrete enumeration value that can be transformed into a true quality label probability vector through one-hot encoding The output is the predicted quality label probability vector By minimizing the cross-entropy loss function that measures the difference between the predicted value P and the true value the predicted value P is made as close as possible to the true value 8. A die-casting quality-related variable importance analysis device based on multivariable standardization embedding and feature fusion, characterized in that, It comprises one or more processors for implementing the die casting quality related variable importance analysis method based on multivariate standardized embedding and feature fusion as described in any one of claims 1 to 7.

9. An electronic device, characterized in that, include: one or more processors; A storage device for storing one or more programs, which, when executed by the electronic device, enables the electronic device to implement the die-casting quality-related variable importance analysis method based on multivariate standardized embedding and feature fusion as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, A program is stored thereon, and when the program is executed by a processor, the die-casting quality-related variable importance analysis method based on multivariate standardized embedding and feature fusion as described in any one of claims 1 to 7 is implemented.

Images