Shielding assembly health diagnosis method based on production system balance

Abstract

The invention belongs to the technical field of health diagnosis, and particularly relates to a shielding assembly health diagnosis method based on production system balance. Comprising the steps that multi-source data in the assembly welding process are collected and arranged, and a multi-source welding characteristic parameter set of an assembly is formed; a welding data feature subset of the assembly is constructed, feature selection is conducted on the multi-source welding data of the assembly through a recursive feature elimination algorithm based on a support vector machine, feature parameters with high distinguishability are selected, useless features and redundant features are eliminated, and the assembly welding feature subset is formed; and a component health diagnosis model is trained, the model is trained through the multi-source welding feature subset of the component in an integrated learning mode, the feature subset serves as the input of the model, the welding quality serves as the output of the model, and the result whether component welding is qualified or not is obtained. The method has the beneficial effects that the diagnosis accuracy of the assembly quality is improved, and uncertain factors in the production process can be evaluated more comprehensively through integrated learning and information fusion technologies.

Description

Technical Field

[0001] This invention belongs to the field of product health diagnosis technology, specifically relating to a method for diagnosing the health of shielding components based on the balance of a production system. Background Technology

[0002] A production system refers to the information system that supports the daily business operations of an organization under normal circumstances. It includes production data, a production data processing system, and a production network. Ensuring the reliability of the shielding component production system-level processes hinges on analyzing and studying the system balance of the production system, considering health diagnostic methods for shielding components based on the production system's balance, ensuring the balance of the production system, and improving the assembly quality of the shielding components.

[0003] In modern manufacturing, the balance of the production system is crucial for ensuring product quality and production efficiency. This is especially true in the manufacturing of highly complex and precision-required product components, such as the assembly of shielding components, where the balance of the production system directly impacts the reliability and performance of the final product. However, traditional production systems often face uncertainties in the assembly process, stemming from various resources such as equipment, materials, and personnel, leading to instability in the assembly results. The configuration of the assembly line, such as the number of workstations, the setup of inspection stations, and the layout of rework points, significantly affects improving the reliability of the production line and saving production costs. Furthermore, quality control and fault diagnosis during the assembly process also face challenges, particularly in large-scale customized production environments. Accurately identifying welding quality and performing health checks becomes key to improving production efficiency and product quality.

[0004] Existing methods for diagnosing production system health largely rely on manual experience and qualitative analysis, lacking effective processing and analysis of multi-source heterogeneous data from the production process. These methods often fail to comprehensively assess the balance of the production system, struggle to achieve precise control of key process parameters during assembly, and are unable to effectively predict and diagnose potential quality problems during assembly. Therefore, there is an urgent need for a product component health diagnosis method based on the balance of the production system. This method should utilize advanced data analysis techniques to deeply mine the vast amounts of data generated during the production process, enabling real-time monitoring and intelligent diagnosis of assembly quality, thereby improving the stability of the production system and the reliability of the products. Summary of the Invention

[0005] The purpose of this invention is to provide a health diagnosis method for shielded components based on the balance of the production system, which aims to solve the problem of effectively processing and analyzing multi-source heterogeneous data in the production process, and to achieve real-time monitoring and intelligent diagnosis of the assembly quality of product components.

[0006] The technical solution of the present invention is as follows: A method for health diagnosis of shielding components based on the balance of a production system, comprising the following steps: Step 1: Collect and organize multi-source data from the component welding process to form a multi-source welding characteristic parameter set for the component; Step 2: Construct a subset of welding data features for the components. Using a recursive feature elimination algorithm based on support vector machines, feature selection is performed on the multi-source welding data of the components. Feature parameters with strong discriminative power are selected, and useless and redundant features are eliminated to form a subset of welding features for the components. Step 3: Train the component health diagnosis model. The model is trained using a subset of multi-source welding features of the component through ensemble learning. The feature subset is used as the input of the model, and the welding quality is used as the output of the model. The result of whether the component welding is qualified or not will be obtained.

[0007] Step 2 includes the following: Step 21: Find hyperplane; Step 22: Define the point closest to this hyperplane as the support vector. Assume the distance from the support vector to this hyperplane is d, then the expression for d is: Step 23: Define the distance from the two types of data points to the hyperplane as: .

[0008] Step 24: The optimization goal of SVM is to maximize the distance from the support vectors to the hyperplane while ensuring correct classification, i.e., to find the maximum margin plane. The optimization problem of SVM is transformed into: .

[0009] Step 25: If the data is linearly inseparable, SVM uses a kernel function to transform the linear classifier into a non-linear classifier. The hyperplane expression in the high-dimensional space is: .

[0010] Step 26: The kernel function used is RBF, as follows: .

[0011] Step 27: In each iteration, the component welding feature parameter with the lowest score is removed until the feature set becomes empty or a certain number of features are obtained. Then, the iteration stops. The scoring criterion is defined by the following formula, which represents the score of feature i: .

[0012] Step 3 includes the following: Step 31: Create a new training set Each time, k training samples are randomly selected from the training set using a bagging sampling method with replacement, and this process is repeated n times to eventually form n training sets.

[0013] Step 32: Train the model n new training sets correspond to n base classifiers. Each base classifier is trained sequentially using each new training set, resulting in n training models.

[0014] Step 33: Processing Results The component welding quality identification results of the n basic classifiers will be processed by voting to generate the final quality identification result.

[0015] The beneficial effects of this invention are as follows: It improves the diagnostic accuracy of assembly quality by using integrated learning and information fusion technologies to more comprehensively assess uncertainties in the production process. It enables real-time monitoring of the production process, allowing for timely detection and handling of quality issues during assembly, thus reducing rework and scrap rates. Through production system balance analysis, it optimizes assembly line configuration, improving production efficiency and resource utilization. The method of this invention is applicable to various types of production systems, possessing broad application prospects and market potential. Attached Figure Description

[0016] Figure 1 A flowchart of a method for health diagnosis of shielding components based on the balance of a production system provided by the present invention; Figure 2 This is a schematic diagram of SVM. Figure 3 This is a feature score map for each component. Detailed Implementation

[0017] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0018] A method for health diagnosis of shielding components based on the balance of a production system includes the following: (1) Data collection and preprocessing: This invention first collects multi-source data from the production process, including but not limited to equipment status, process parameters, and quality inspection results during assembly. The collected data is preprocessed, including data cleaning, normalization, and feature extraction, to ensure data quality and consistency.

[0019] (2) Feature selection and model construction: The Support Vector Machine Recursive Feature Elimination (SVM-RFE) algorithm was used to select features from the preprocessed data to identify key features that significantly affect assembly quality. Based on these key features, the Bagging algorithm was used to construct an ensemble learning model to improve the accuracy and robustness of the diagnosis.

[0020] (3) Health diagnosis and decision support: The data collected in real time during the production process is input into the constructed model, and the model outputs the health status prediction results of the components. Based on the prediction results, the system provides decision support to guide production personnel to adjust assembly parameters or take corresponding quality control measures.

[0021] (4) Production system balance analysis: This invention further analyzes the balance of the production system, optimizes the assembly line configuration by evaluating the relationship between process parameters and product quality, and ensures the stability and efficiency of the production process.

[0022] This invention employs a data-driven approach, collecting and organizing multi-source data from the component welding process to form a multi-source welding feature parameter set for the component. Then, using a recursive feature elimination algorithm based on support vector machines, it selects highly discriminative feature parameters from the multi-source welding data, eliminating useless and redundant features to form a subset of component welding features. Finally, by combining advanced machine learning techniques and production system balance analysis, it achieves accurate diagnosis of the health status of product components and optimizes the production system, thereby improving product quality and production efficiency.

[0023] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments: A method for health diagnosis of shielding components based on the balance of a production system includes the following steps: Step 1: Collect and organize multi-source data from the component welding process to form a multi-source welding characteristic parameter set for the component; The specific set of multi-source welding characteristic parameters for the components is shown in Table 1: Table SEQ Table * ARABIC 1 Component Multi-Source Welding Characteristic Parameter Set Step 2: Construct a subset of welding data features for the components. Using a recursive feature elimination algorithm based on support vector machines, select features from the multi-source welding data of the components, identify highly discriminative feature parameters, and eliminate useless features. By eliminating redundant features and reducing computational load, a subset of component welding features is ultimately formed, including the following sub-steps: Step 21: Find the maximum margin hyperplane to classify the data, i.e., find... Figure 1 Shown Such a hyperplane, in which It is a column vector composed of component welding features. Let be the normal vector of the hyperplane. The offset determines the position of the hyperplane relative to the origin; Step 22: Define the points closest to this hyperplane as support vectors. Assume the distance from the support vectors to this hyperplane is d, then the expression for d is: Step 23: Define the distance from the two types of data points to the hyperplane as: in, It is a tag used to determine the category of a component. This indicates that the sample data belongs to the positive class. This indicates that the sample data belongs to the negative class.

[0024] Step 24: The goal of SVM optimization is to maximize the distance from the support vectors to the hyperplane while ensuring correct classification, i.e., to find the maximum margin plane. The SVM optimization problem is transformed into: in, Indicates the first The feature vector of each sample This indicates the corresponding category label.

[0025] Step 25: If the data is linearly inseparable, SVM can use kernel functions to transform a linear classifier into a non-linear classifier. The hyperplane expression in high-dimensional space is: The purpose of kernel functions is to avoid direct computation. The inner product of is defined as the kernel function, which is the inner product of a high-dimensional space.

[0026] Step 26: The kernel function used in this section is RBF, as shown in the equation: in , These are two sample feature vectors in the component welding sample set. The kernel width parameter determines how well the model fits the samples.

[0027] Step 27: In each iteration, the component welding feature parameter with the lowest score is removed until the feature set becomes empty or a certain number of features are obtained, at which point the iteration stops. The scoring criterion is defined by the formula shown, and is expressed as the score of feature i: in, Representation of features Feature scores.

[0028] A subset of welding data features for the components is constructed. A recursive feature elimination algorithm based on support vector machines is used to calculate the scores of each feature parameter. The feature scores of the components are as follows: Figure 2 As shown, the first 19 are the scores of the detection features, and the last 6 are the scores of the component welding features.

[0029] Feature selection was performed on the multi-source welding data of the components. The 12 feature parameters with the highest scores were selected to form a subset of component welding feature parameters with strong distinguishability. These parameters are: hexagonal tube length M1, hexagonal tube wall thickness M2, hexagonal tube edge straightness M4, hexagonal tube inner side-to-side distance M6, assembly dimension of operating head and hexagonal tube M7, assembly dimension of lower transition joint hexagonal tube M8, component upper runout M9, component lower runout M10, pin coaxiality M13, diameter of the concave part of the pin spiral groove M14, and diameter of the convex part of the pin spiral groove M15.

[0030] Step 3: Train the component health diagnosis model. The model is trained using a multi-source welding feature subset of the component through Bagging ensemble learning. The feature subset is used as the input of the model, and the welding quality is used as the output of the model. Finally, the result of whether the component welding is qualified or not will be obtained.

[0031] Step 31: Create a new training set Each time, k training samples are randomly selected from the training set using a bagging sampling method (with replacement) to form a new training set. This process is repeated n times, resulting in n training sets. During this process, it is possible that a sample from the initial training set may appear multiple times in all n new training sets, or it is possible that a sample from the initial training set may not be selected even once in the n selections. Even if there is some overlap among the n new training sets, it is assumed that these n new training sets are independent. Step 32: Train the model n new training sets correspond to n base classifiers. One base classifier is trained sequentially using each new training set, resulting in n trained models. It is assumed that these n models are independent of each other, and a suitable base classifier is selected according to actual needs. Step 33: Processing Results The component welding quality identification results of n basic classifiers will be processed by voting to generate the final quality identification result. Bagging and SVM will be used as classifiers.

[0032] One hundred historical data sets, each consisting of a feature subset, were used as input to the model, including 50 acceptable product data points and 50 unacceptable product data points. Finally, a test set consisting of 100 additional historical data samples was selected as input to the model for testing. This test set included 10 samples of components with unacceptable welding and 90 samples of components with acceptable welding. The test results are shown in Table 2. Table SEQ * ARABIC 2 Component Welding Quality Test Results Analysis of the calculation results showed that out of 100 test samples, 95 samples were correctly identified, 5 samples were incorrectly identified, and 5 qualified samples were mistakenly classified as unqualified. The prediction accuracy of the model was 95%, and the recognition accuracy was good, which verified the feasibility of the component welding quality recognition model based on ensemble learning.

[0033] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the scope of the technology disclosed in the present invention, and such modifications or substitutions should all be covered within the scope of protection of the present invention.

Claims

1. A method for health diagnosis of shielding components based on the balance of a production system, characterized in that, Includes the following steps: Step 1: Collect and organize multi-source data from the component welding process to form a multi-source welding characteristic parameter set for the component; Step 2: Construct a subset of welding data features for the components. Using a recursive feature elimination algorithm based on support vector machines, feature selection is performed on the multi-source welding data of the components. Feature parameters with strong discriminative power are selected, and useless and redundant features are eliminated to form a subset of welding features for the components. Step 3: Train the component health diagnosis model. The model is trained using a subset of multi-source welding features of the component through ensemble learning. The feature subset is used as the input of the model, and the welding quality is used as the output of the model. The result of whether the component welding is qualified or not will be obtained.

2. The method for health diagnosis of shielding components based on the balance of a production system as described in claim 1, characterized in that, Step 2 includes the following: Step 21: Find hyperplane; Step 22: Define the point closest to this hyperplane as the support vector. Assume the distance from the support vector to this hyperplane is d, then the expression for d is: Step 23: Define the distance from the two types of data points to the hyperplane as: 。 3. The method for health diagnosis of shielding components based on the balance of a production system as described in claim 2, characterized in that, Step 2 includes the following: Step 24: The optimization goal of SVM is to maximize the distance from the support vectors to the hyperplane while ensuring correct classification, i.e., to find the maximum margin plane. The optimization problem of SVM is transformed into: 。 4. The method for health diagnosis of shielding components based on the balance of a production system as described in claim 3, characterized in that, Step 2 includes the following: Step 25: If the data is linearly inseparable, SVM uses a kernel function to transform the linear classifier into a non-linear classifier. The hyperplane expression in the high-dimensional space is: 。 5. The method for health diagnosis of shielding components based on the balance of a production system as described in claim 4, characterized in that, Step 2 includes the following: Step 26: The kernel function used is RBF, as follows: 。 6. The method for health diagnosis of shielding components based on the balance of a production system as described in claim 5, characterized in that, Step 2 includes the following: Step 27: In each iteration, the component welding feature parameter with the lowest score is removed until the feature set becomes empty or a certain number of features are obtained. Then, the iteration stops. The scoring criterion is defined by the following formula, which represents the score of feature i: 。 7. The method for health diagnosis of shielding components based on the balance of a production system as described in claim 1, characterized in that, Step 3 includes the following: Step 31: Create a new training set Each time, k training samples are randomly selected from the training set using a bagging sampling method with replacement, and this process is repeated n times to eventually form n training sets.

8. The method for health diagnosis of shielding components based on the balance of a production system as described in claim 7, characterized in that, Step 3 includes the following: Step 32: Train the model n new training sets correspond to n base classifiers. Each base classifier is trained sequentially using each new training set, resulting in n training models.

9. The method for health diagnosis of shielding components based on the balance of a production system as described in claim 8, characterized in that, Step 3 includes the following: Step 33: Processing Results The component welding quality identification results of the n basic classifiers will be processed by voting to generate the final quality identification result.

Images