Quality online prediction method for precision forming process of cartridge case

Abstract

The invention provides a cartridge case precision forming process quality online prediction method. The invention provides a prediction method combining instant learning, principal component analysis (PCA) and a minimum-maximum probability regression (MPMR). The method comprises the following steps: firstly, dynamically selecting a key sample according to the similarity between a new data point and a historical data set by utilizing a real-time learning strategy, and constructing a temporary training data set; thirdly, dimension reduction is conducted on the data through the PCA technology, the calculation complexity is reduced, and main features are reserved; afterwards, an MPMR model is adopted, and the prediction stability and reliability are improved by optimizing the probability that a prediction point falls into an allowable error range. According to the method, the cartridge weight can be accurately predicted, shooting errors caused by weight fluctuation are remarkably reduced, meanwhile, the yield and consistency of a cartridge forming process are improved, and reliable guarantee is provided for artillery shooting precision and range stability.

Description

Technical Field

[0001] The invention relates to the field of industrial process soft sensor modeling and application, and in particular to an online prediction method for the quality of a cartridge precision molding process. Background Art

[0002] The cartridge has a simple production process, low cost, and has the dual functions of a container and an energy source. It is widely used in the launch charge container of new artillery. The precision molding process of the cartridge is to coat the rolled nitrocellulose paper with a layer of energetic adhesive, and then form flexible paper tubes of various diameters, lengths, and weights after rolling, unwinding, and cutting. There are many process parameters involved in the entire process, and the molding law of the flexible paper tube is unclear. It has the characteristics of multi-process and multi-variable strong coupling, which leads to problems such as the difficulty in accurately controlling the coating amount of the energetic adhesive, inconsistent molding dimensions of the flexible paper tube, and large fluctuations in the weight of the tube, resulting in large fluctuations in the molding process quality of the cartridge and low yield rate. In the production process of the cartridge, accurate prediction of its weight plays an important role in improving product quality. This is not only the key to ensuring the consistency and stability of the weight of each batch of products, but also a necessary condition for ensuring the shooting accuracy and range of the artillery. The accuracy of the charge directly affects the ballistic performance. Weight prediction can significantly reduce the shooting error caused by weight fluctuations, thereby improving the stability and reliability of the artillery.

[0003] Weight prediction also provides important support for production process optimization and efficiency improvement. Through real-time monitoring and analysis of weight data during the production process, producers can quickly identify and solve potential problems, avoid waste and unnecessary processing steps, and thus reduce production costs. More importantly, this data-driven approach can significantly improve resource utilization efficiency and ensure the reliability and consistency of the production process. In addition, weight prediction, as an important feedback mechanism, can help producers dynamically adjust and optimize process parameters, thereby further improving the stability and consistency of product quality.

[0004] Therefore, it is particularly necessary to design a system that can predict the weight of cartridges in real time during the production process. This can not only improve product quality, but also provide a solid guarantee for efficient and stable production. Summary of the invention

[0005] The present invention proposes an online prediction method for the quality of the precision molding process of cartridges. It realizes dynamic modeling through real-time learning modeling, uses PCA to reduce the dimension of data in the production process, and adopts Minimax probabilistic regression machine to efficiently predict the weight of cartridges. This method combines the dynamic update mechanism to capture the changing characteristics of production data in real time, and improves the accuracy and robustness of weight prediction by intelligently adjusting the model parameters. While ensuring the prediction accuracy, it realizes rapid adaptation to the production process, which not only ensures the consistency and stability of product quality, but also provides strong support for optimizing production efficiency and reducing production risks.

[0006] The technical solution adopted by the present invention to achieve the above-mentioned purpose is:

[0007] A method for online prediction of the quality of a cartridge precision molding process comprises the following steps:

[0008] 1) Collecting data from the cartridge manufacturing process to form an initial data set D0;

[0009] 2) Normalize all samples in the initial data set D0 to obtain D'0;

[0010] 3) According to the Mahalanobis distance similarity criterion, select the query sample X from the initial data set D'0 new Most similar N neighbor samples, forming the training data set D train ;

[0011] 4) Using principal component analysis to analyze D train Perform feature dimensionality reduction to obtain the reduced dimensionality dataset D' train ;

[0012] 5) Use the Minimax probabilistic regression machine as the prediction model and use the dataset D' train Train them;

[0013] 6) Use the trained model to query sample X new Make predictions.

[0014] The initial data set D0 = [X, Y], X∈R N×M ,Y∈R N×L , where X is the feature of the sample, Y is the quality of the sample, N is the number of samples, M is the dimension of the sample, and L is the sample output dimension.

[0015] The step 3) is specifically as follows:

[0016] In the dataset D'0, calculate the difference between each sample and the query sample X new The Mahalanobis distance similarity is calculated, and all similarities are arranged from large to small, and the top N are selected.neighbor The samples corresponding to the similarity values ​​constitute the training data set D train .

[0017] The Mahalanobis distance similarity is specifically:

[0018]

[0019] Among them, x and y are two query samples, ∑ is the covariance matrix of the entire data set, and d M is the similarity between the two.

[0020] The step 4) comprises the following steps:

[0021] 4.1) For D train The data matrix X∈R in N×M Centralize it to get X centered , and calculate the covariance matrix ∑ of X;

[0022] 4.2) Perform eigenvalue decomposition on the covariance matrix ∑, sort the decomposed eigenvalues ​​from large to small, and select the eigenvectors corresponding to the first k largest eigenvalues ​​to form the feature matrix V K ∈ M×k ;

[0023] 4.3) V K Project to low-dimensional space to obtain the reduced-dimensional data matrix Z∈R N×k , Z=X centered V k ;

[0024] 4.4) Combine the reduced-dimensional data matrix Z with the corresponding target value to form D' train .

[0025] The eigenvalue decomposition of the covariance matrix ∑ is specifically:

[0026] ∑v i =λ i v i

[0027] Among them, λ i is the eigenvalue, v i is the corresponding eigenvector.

[0028] The step 5) is specifically as follows:

[0029] The prediction model uses the Minimax probability regression machine to make the prediction value The minimum probability within the range of the true output value y±ε, ε∈R, ε>0 is maximized, and the lower bound Ω of the minimum probability is estimated to be:

[0030]

[0031] The prediction model is:

[0032]

[0033] Among them, ε represents the allowable prediction error, is the kernel function that satisfies the Mercer condition, β i and b are model parameters.

[0034] Calculate the β i and b comprises the following steps:

[0035] (1) Divide the training data set into two categories: U and V:

[0036] U i =[y i +ε,x i1 ,...x M ] T ,i=1,2,...N

[0037] V j =[y j -ε,x j1 ,...x M ] T ,j=1,2,...N

[0038] Among them, ε represents the allowed prediction error, M represents the number of features, and the U class is defined as U = [U1, U2, ... U N ], the V class is defined as V = [V1, V2, ... V N ];

[0039] (2) Combine the U and V data to generate Z = [U1, ...U N ,V1,...V N ], calculate the Gram matrix K of Z:

[0040] K is composed of elements Here, the kernel function uses the Gaussian kernel, and then the optimization target γ is obtained by optimizing the problem:

[0041]

[0042] in, Where K u and K v Represent the first u rows and the last v rows of the Gram matrix K, respectively, N u , N v Indicates the number of U and V types of data, Represents process variables; and Specifically:

[0043]

[0044] (3) Using γ, calculate b c :

[0045]

[0046] (4) Using b c , calculate β i and b:

[0047] β i = -2ε(γ i +γ i+N )

[0048] b=-2εb c .

[0049] The present invention has the following beneficial effects and advantages:

[0050] 1. The present invention adopts a dynamic modeling mechanism of instant learning to screen local data similar to the query sample in real time during the production process, reducing the dependence on global data, enabling the prediction model to dynamically adjust the prediction results under different production batches and conditions, and significantly improving the generalization performance and robustness of the combustible cartridge weight prediction model.

[0051] 2. The present invention uses the kernel principal component analysis method to reduce the dimensionality of the dynamic training data set, which not only reduces the computational complexity but also retains the core features of the cartridge manufacturing data, thereby effectively improving the timeliness and prediction accuracy of the model.

[0052] 3. The present invention optimizes the robustness of the Minimax probabilistic regression model and achieves high-precision instant prediction for dynamic fluctuations and uncertainties in the production environment, ensuring the applicability of the model in complex production scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0054] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments.

[0055] like Figure 1 As shown, the present invention provides an online prediction method for the quality of a precision molding process of a cartridge, comprising the following steps:

[0056] (1) Collecting data from the cartridge manufacturing process to form an initial database D0 = [X, Y], X∈R N×M ,Y∈RN×L , where N is the number of samples, M is the dimension of the samples, and L is the sample output dimension;

[0057] The characteristics that can characterize the production status of the cartridge include melting temperature, coating tank temperature, coating speed, paper feeding speed, tube winding speed, etc. These parameters can be collected by sensors of the production equipment.

[0058] The data collected in step 1 include the melting temperature, coating tank temperature, coating rotation speed, paper feeding speed, tube winding speed and other characteristics recorded during the cartridge production process, which can characterize the production status of the cartridge.

[0059] (2) All samples in the initial data set D0 are normalized to obtain D'0 to eliminate the influence of feature scale differences.

[0060] (3) According to the Mahalanobis distance similarity criterion, select the query sample X from the initial data set D'0 new Most similar N neighbor samples, forming the training data set D train .

[0061] Mahalanobis Distance Similarity:

[0062]

[0063] Where x and y are two query samples, ∑ is the covariance matrix of the entire data set, and d M is the similarity between the two, d M The higher the value, the more similar the two sample data are.

[0064] In the step 3, the Mahalanobis distance is selected as the similarity criterion for instant learning modeling based on the characteristics that the data dimension is high and the features are highly correlated after dimensionality reduction during the cartridge production process.

[0065] (4) Obtain training data set D train Finally, the principal component analysis method is used to reduce the dimension of the features and extract the principal components that can reflect the key features of the cartridge process, thereby reducing the influence of noise interference and redundant features. train The data matrix X∈R in N×M Centralize it to get X centered , and calculate the covariance matrix ∑ of X at the same time.

[0066] Perform eigenvalue decomposition on the covariance matrix ∑:

[0067] ∑v i =λ i v i (2)

[0068] where λ i is the eigenvalue, v i is the corresponding eigenvector. Select the eigenvectors corresponding to the first k largest eigenvalues ​​in descending order to form the characteristic matrix V K ∈ M×k .

[0069] Project to low-dimensional space to obtain the reduced-dimensional data matrix Z∈R N×k

[0070]

[0071] The reduced data matrix Z is recombined with the corresponding target value to form D' train .

[0072] The content of step 4 includes the training data set D generated by the cartridge process through the kernel principal component analysis algorithm train Nonlinear dimensionality reduction is performed to extract the principal components that can reflect the key characteristics of the cartridge process, thereby reducing the impact of noise interference and redundant features and improving the generalization ability of the model. This process retains the nonlinear structural characteristics of the cartridge manufacturing data and lays the foundation for the construction of the subsequent prediction model.

[0073] (5) For the obtained MPMR implements model output The minimum probability of being in the pipeline of the true output value y±ε(ε∈R,ε>0) is maximized, and the lower bound Ω of the minimum probability is estimated as:

[0074]

[0075] Its regression model is:

[0076]

[0077] In the formula is the kernel function that satisfies the Mercer condition; β i and b are used as model parameters and are obtained by the MPMR algorithm.

[0078] First, the training data set is divided into two categories (U and V):

[0079]

[0080] Where ε represents the allowed prediction error and M represents the number of features.

[0081] The class U is defined as U = [U1, U2, ...U N ], and the V class is defined as V = [V1, V2, ... V NThen, combine the two to generate Z = [U1, ...U N ,V1,...V N ], calculate the Gram matrix K of Z, K consists of elements Through the optimization problem of formula (7), we can obtain γ:

[0082]

[0083] In the formula and They are obtained by the following formulas:

[0084]

[0085] After obtaining γ, b can be calculated by solving equation (10) c

[0086]

[0087] in

[0088] The analytical solution defined in equation (5) is obtained, where

[0089]

[0090] In step 5, the Minimax Probabilistic Regression Model (MPMR) is used to model the dynamic fluctuation characteristics of the cartridge manufacturing process to minimize the prediction error caused by data uncertainty. The model optimizes the regression risk in the worst case to ensure that the prediction results are highly robust and accurate under various working conditions.

[0091] (6) After training the MPMR model, new Substitute into formula (5) to obtain the predicted value Complete online prediction of cartridge precision molding process quality.

[0092] It can be seen from the above technical scheme that compared with the existing methods, the present invention discloses an online prediction method for the quality of the precision molding process of cartridges. It first realizes data dimension reduction and key feature extraction through the kernel principal component analysis algorithm, and then combines the dynamic adjustment ability of the Minimax probability regression model to accurately predict the weight of the combustible cartridge while effectively responding to dynamic fluctuations in the production process, thereby improving the robustness and adaptability of the model. Finally, the most relevant local samples are selected for dynamic training and prediction using the instant learning mechanism, which significantly improves the adaptability and prediction accuracy of the model to dynamic changes in production.

[0093] The implementation methods of each step described in this specification are closely linked, and cannot be realized independently without any link. The similarities and differences between each implementation method described can be referred to the specification part in detail. It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements not only includes those elements, but also includes other elements not clearly listed, or also includes elements inherent to such process, method, article or equipment.

[0094] The above description is only an embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, expansion, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims

1. An online prediction method for the quality of a cartridge precision molding process, characterized in that: The following steps are involved: 1) Collecting data from the cartridge manufacturing process to form an initial data set D0; 2) Normalize all samples in the initial data set D0 to obtain D'0; 3) According to the Mahalanobis distance similarity criterion, select the query sample X from the initial data set D'0 new Most similar N neighbor samples, forming the training data set D train ; 4) Using principal component analysis to analyze D train Perform feature dimensionality reduction to obtain the reduced-dimensional data set D t ' rain ; 5) Use the Minimax probabilistic regression machine as the prediction model and use the dataset D t ' rain Train them; 6) Use the trained model to query sample X new Make predictions.

2. The method for online prediction of the quality of the precision molding process of a cartridge according to claim 1 is characterized in that: The initial data set D0 = [X, Y], X∈R N×M ,Y∈R N×L , where X is the feature of the sample, Y is the quality of the sample, N is the number of samples, M is the dimension of the sample, and L is the sample output dimension.

3. The method for online prediction of the quality of the precision molding process of a cartridge according to claim 1 is characterized in that: The step 3) is specifically as follows: In the dataset D'0, calculate the difference between each sample and the query sample X new The Mahalanobis distance similarity is calculated, and all similarities are arranged from large to small, and the top N are selected. neighbor The samples corresponding to the similarity values ​​constitute the training data set D train .

4. The method for online prediction of the quality of the precision molding process of a cartridge according to claim 3 is characterized in that: The Mahalanobis distance similarity is specifically: Among them, x and y are two query samples, Σ is the covariance matrix of the entire data set, and d M is the similarity between the two.

5. The method for online prediction of the quality of the precision molding process of a cartridge according to claim 1 is characterized in that: The step 4) comprises the following steps: 4.1) For D train The data matrix X∈R in N×M Centralize it to get X centered , and calculate the covariance matrix Σ of X; 4.2) Perform eigenvalue decomposition on the covariance matrix Σ, sort the decomposed eigenvalues ​​from large to small, and select the eigenvectors corresponding to the first k largest eigenvalues ​​to form the feature matrix V K ∈ M×k ; 4.3) V K Project to low-dimensional space to obtain the reduced-dimensional data matrix Z∈R N×k , Z=X centered V k ; 4.4) Combine the reduced-dimensional data matrix Z with the corresponding target value to form D′ train .

6. The method for online prediction of the quality of the precision molding process of a cartridge according to claim 5, characterized in that: The eigenvalue decomposition of the covariance matrix Σ is specifically: ∑v i =λ i v i Among them, λ i is the eigenvalue, v i is the corresponding eigenvector.

7. The method for online prediction of the quality of the precision molding process of a cartridge according to claim 1, characterized in that: The step 5) is specifically as follows: The prediction model uses the Minimax probability regression machine to make the prediction value The minimum probability within the range of the true output value y±ε, ε∈R, ε>0 is maximized, and the lower bound Ω of the minimum probability is estimated to be: The prediction model is: Among them, ε represents the allowable prediction error, is the kernel function that satisfies the Mercer condition, β i and b are model parameters.

8. The method for online prediction of the quality of the precision molding process of a cartridge according to claim 7, characterized in that: Calculate the β i and b comprises the following steps: (1) Divide the training data set into two categories: U and V: U i =[y i +ε,x i1 ,...x M ] T ,i=1,2,...N V j =[y j −ε,x j1 ,…x M ] T ,j=1,2,...N Among them, ε represents the allowed prediction error, M represents the number of features, and the U class is defined as U = [U1, U2, ... U N ], the V class is defined as V = [V1, V2, ... V N ]; (2) Combine the U and V data to generate Z = [U1, ...U N ,V1,...V N ], calculate the Gram matrix K of Z: K is composed of elements Here, the kernel function uses the Gaussian kernel, and then the optimization target γ is obtained by optimizing the problem: in, Where K u and K v Represent the first u rows and the last v rows of the Gram matrix K, respectively, N u , N v Indicates the number of U and V types of data, Represents process variables; and Specifically: (3) Using γ, calculate b c : (4) Using b c , calculate β i and b: b i =-2ε(γ i +g i+N ) b=-2εb c 。

Images