Bottle preform sealing risk identification method based on fuzzy clustering

Abstract

The invention discloses a bottle preform sealing risk identification method based on fuzzy clustering, and the method comprises the steps: S1, generating a unique number for each bottle preform, collecting technological parameters and sealing performance detection data in a production process, and forming an original data table; s2, performing cleaning and Z-score standardization on the data to generate preprocessed bottle preform data; s3, performing feature extraction on the preprocessed bottle embryo data to form a feature data set of the bottle embryo sample; s4, performing clustering analysis on the feature data set by using a fuzzy C-means clustering algorithm to obtain a clustering center and a membership degree of each bottle embryo; s5, according to the membership degree and the distance between the membership degree and the risk clustering center, the sealing risk grade is judged; and S6, feeding back a risk identification result to a production management and control system to realize automatic early warning, hierarchical management and process optimization. The intelligent and automatic level of bottle preform sealing risk identification is improved, the production line quality management and control efficiency is improved, and the method is suitable for efficient and accurate quality management of the modern manufacturing industry.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent manufacturing and industrial data analysis, and in particular to a method for identifying preform sealing risks based on fuzzy clustering. The method is suitable for application scenarios such as multi-source process and quality data collection, intelligent clustering analysis, risk classification warning, and production process optimization during preform production. Background Art

[0002] In the plastics industry, preforms serve as crucial intermediates for hollow products like plastic bottles, and their sealing performance directly impacts the safety and quality of the final product. As the food, beverage, and pharmaceutical industries increasingly demand tighter packaging, identifying risks associated with preform sealing performance has become a critical component of production and quality management. Prior art testing of preform sealing performance often relies on manual spot checks or automated testing equipment based on fixed thresholds. Manual testing is inefficient, highly subjective, and prone to misjudgments and omissions due to inexperience or fatigue. Existing automated testing equipment often relies on a single or limited number of parameters (such as sealing pressure and dimensional thresholds), making it difficult to comprehensively analyze the multi-dimensional relationship between complex process parameters and sealing performance. Furthermore, its adaptability to changes in the production environment and the diversity of bottle shapes is limited.

[0003] With the increasing automation and digitization of production, production lines are able to collect a large amount of real-time data on preform process parameters and sealing performance, such as injection temperature, pressure, time, cooling time, dimensions, and sealing pressure. However, existing technologies generally lack the systematic integration and in-depth analysis of this multi-source data. While some companies have attempted to use statistical analysis methods to grade preform quality, these methods are often based solely on the mean, variance, or fixed threshold of a single variable. This approach fails to reflect the complex impact of process variations on sealing performance and is easily subject to the subjectivity and limitations of parameter settings. This makes it difficult to promptly identify potentially high-risk preforms in actual production, resulting in missed inspections of defective products or misclassification of qualified products due to overly conservative approaches, impacting production efficiency and quality control.

[0004] In recent years, the application of artificial intelligence and machine learning in industrial inspection has steadily increased. Cluster analysis, a key unsupervised learning method, can automatically group data and discover its inherent structure. However, traditional hard clustering methods such as K-means struggle to handle the ambiguity and uncertainty inherent in production data, resulting in suboptimal clustering results. The fuzzy C-means clustering algorithm assigns each sample a membership degree to a different cluster, effectively adapting to the fuzzy boundaries of preform quality differentiation in real-world processes and improving the flexibility and accuracy of risk identification. However, the current application of fuzzy clustering algorithms in preform sealing risk identification is limited, primarily focusing on theoretical research or small-scale experiments, lacking comprehensive, integrated, and systematic application in large-scale production environments.

[0005] Furthermore, existing technologies generally neglect aspects such as data cleaning, standardization, feature extraction, and visualization analysis, which are crucial for improving the quality of clustering input data and subsequent risk identification. The relationship between clustering results and sealing risk levels is not fully explored, making it difficult for existing solutions to achieve scientific risk grading and automatic early warning, failing to meet the requirements of modern intelligent manufacturing for efficient and accurate risk identification and quality traceability.

[0006] In summary, the existing technology has the following defects: lack of systematic collection and archiving of multi-source data, imperfect data preprocessing and feature analysis processes, risk identification methods rely on a single threshold or manual experience, and are difficult to adapt to the needs of intelligent risk identification under large-scale and complex process conditions; clustering methods fail to be optimized in combination with the actual production data characteristics, and the clustering results are not closely linked to the risk level, making it difficult to achieve accurate classification of sealing risks and automated early warning.

[0007] Therefore, it is urgent to propose a preform sealing risk identification method based on fuzzy clustering. Through systematic collection, standardized processing, intelligent clustering analysis and risk classification warning of multi-source data, accurate identification and graded warning of preform sealing risks can be achieved, thereby effectively improving the quality and safety management level in the preform production process. Summary of the Invention

[0008] One objective of the present invention is to propose a method for identifying preform sealing risks based on fuzzy clustering. This method fully utilizes the fuzzy C-means clustering algorithm and industrial data analysis technology, and describes in detail the entire process of systematic collection, standardized processing, intelligent clustering analysis, and risk classification and early warning of multi-source process parameters and sealing performance data. The method has the advantages of high recognition accuracy, a high degree of automation, and strong adaptability to complex production environments.

[0009] A method for identifying preform sealing risks based on fuzzy clustering according to an embodiment of the present invention includes the following steps:

[0010] S1. Generate a unique number for each preform and create an identity file. Collect and archive process parameters and sealing performance test data during the production process, and associate them to form a raw data table as the basic data set for subsequent processing;

[0011] S2. Clean each row of data in the original data table, remove missing values ​​and outliers, and then standardize the data using the Z-score standardization method to generate preprocessed preform data;

[0012] S3. Extract features from the pre-processed preform data, construct a multi-dimensional feature vector, form a feature dataset of the preform samples, and perform visual analysis on the feature dataset to facilitate early data diagnosis and feature optimization.

[0013] S4. Perform cluster analysis on the feature data set using a fuzzy C-means clustering algorithm to obtain the cluster center coordinates of each preform sample and the membership value relative to each cluster center. The iterative convergence condition is that the objective function is less than a set convergence threshold or the maximum number of iterations is reached;

[0014] S5. Determine the sealing risk level based on the membership distribution of each preform sample and its distance from the risk cluster center. Mark samples with a membership higher than a set threshold as preforms with sealing risks and assign risk level classification labels.

[0015] S6. Feedback the unique number and risk level information of the identified risky preforms to the preform production control system to achieve automatic early warning and graded management of sealing risk preforms, and provide optimization and adjustment suggestions for related process parameters.

[0016] Optionally, the S1 further includes:

[0017] S11. Assign a unique number to each preform and create an identity file. Archive all process parameters and sealing performance test data during the production process to ensure accurate and reliable data management.

[0018] S12, the process parameters include injection mold cavity temperature T m (℃), injection pressure P (MPa), injection time t inj (s), cooling time t cool (s), preform length L (mm), preform outer diameter D (mm), wall thickness H (mm);

[0019] S13. The sealing performance test data is a quantitative measurement value S (kPa) obtained by performing a sealing performance test on each preform, which is used to reflect the sealing ability of the preform under standard test conditions;

[0020] S14, the process parameters of each preform {T m ,P,t inj ,t cool , L, D, H} are associated with the corresponding quantitative measurement value S of the sealing performance and stored in a structured table format to generate an original data table. Each row represents the data information of a preform, which serves as the basic data set for subsequent data processing.

[0021] Optionally, the S2 further includes:

[0022] S21. Perform integrity check on each row of data in the original data table, remove missing data records, and perform outlier detection using the triple standard deviation method to remove abnormal data that exceeds the set threshold range;

[0023] S22, perform Z-score normalization on the data after removing outliers, specifically for each parameter x according to the formula Standardization is performed, where the parameter x is derived from the process parameters of the preform {T m ,P,t inj ,t cool , L, D, H} and the quantitative measurement value S of the sealing performance, μ is the mean value of the parameter x, σ is the standard deviation of the parameter x, and z is the standardized parameter value;

[0024] For example, for the injection mold cavity temperature T m , if a batch of original data is T m,1 =260℃,T m,2 =255℃,T m,3 =265℃,T m,4 =258℃, then the mean is The standard deviation is After Z-score standardization, it is converted into

[0025] S23, all parameters after Z-score standardization As pre-processing of preform data, it provides a data basis for subsequent feature extraction and analysis.

[0026] Optionally, the S3 further includes:

[0027] S31. Extract features from the pre-processed preform data and construct a multi-dimensional feature vector for each preform.

[0028] S32. Summarize the feature vectors F of all preform samples to form a feature dataset of preform samples. Where n is the number of samples;

[0029] S33, using t-distributed stochastic neighbor embedding (t-SNE) dimensionality reduction method to reduce the feature dataset Perform visual analysis, map multidimensional feature data into two-dimensional space, draw scatter plots and heat maps, and display the distribution, clustering trends, and outliers of each preform sample in the feature space.

[0030] Optionally, the S4 further includes:

[0031] S41. Based on the preform sample feature dataset The actual scale and distribution characteristics of the clusters are used to determine the specific value of the number of clusters c in the fuzzy C-means clustering algorithm;

[0032] S42, using K-means++ algorithm to extract the feature data set of preform samples Select c representative feature vectors as the initial cluster centers, expressed as {v1,v2,...,v c}, and for each preform sample F i Assign initial membership u ij , where i represents the i-th preform sample, j represents the j-th cluster center, and u ij Indicates the membership of the i-th sample to the j-th class, and the sum of all memberships is equal to 1, that is,

[0033] S43. Feature Dataset Cluster analysis is performed, which specifically includes the following iterative process: ① In each iteration, according to the current cluster center {v j}, calculate each preform sample F i Euclidean distance d to each cluster center ij , the formula is:

[0034] d ij =‖F i -v j ‖;

[0035] ②Update the membership u of each sample ij , the formula is:

[0036]

[0037] Where m is the fuzzification coefficient, and m>1; ③ According to the new membership, update each cluster center v j , the formula is:

[0038]

[0039] S44, repeat the cluster center and membership degree update process in step S43 until the objective function J meets the convergence condition or reaches the maximum number of iterations. The convergence condition is:

[0040]

[0041] Among them, ∈ is the convergence threshold set according to the size of the data set and actual needs;

[0042] S45. Output the cluster center coordinates v of each preform sample j and the membership value u relative to each cluster center ij , used for subsequent sealing risk level determination.

[0043] Optionally, the S5 further includes:

[0044] S51. According to the cluster analysis results, determine the cluster centers v j The cluster center corresponding to the poor sealing performance is the risk cluster center v risk ;

[0045] S52. For each preform sample F i , obtain its membership degree u relative to the risk cluster center i,risk , and preform sample F i With v risk The Euclidean distance d between i,risk , the formula is:

[0046] d i,risk =‖F i -v risk ‖;

[0047] S53, let the risk identification threshold be θ, the classification interval parameters be α1, α2, satisfying 0<α1<α2≤1; when u i,risk ≥α2, and When the preform sample F i Determined as high risk and generated a high risk label, where: For all preform samples judged as risky to the risk cluster center v risk The mean of the Euclidean distance; when α1≤u i,risk <α2, and d i,risk near When the preform sample F i It is judged as medium risk and generates a medium risk label; when θ≤u i,risk <α1, or Then the preform sample F i Determine as low risk and generate a low risk label, where δ is the tolerance parameter for distance classification; samples that do not fall within the above range are determined to be non-risk preforms;

[0048] For example, let θ = 0.7, α1 = 0.8, α2 = 0.9, δ=0.1;If a preform sample u i,risk =0.92 and d i,risk =0.45, it is high risk; if u i,risk =0.85 and d i,risk =0.52, it is medium risk; if u i,risk =0.75 and d i,risk =0.62, which is low risk.

[0049] Optionally, the S6 further includes:

[0050] S61. The unique numbers, risk level labels, risk membership values, and Euclidean distances from the risk cluster centers of the identified preforms at each risk level are collated to form a sealing risk identification result data table.

[0051] S62. Feedback the sealing risk identification result data sheet to the preform production control system in real time via a standardized data interface, industrial Ethernet, or other network communication methods, to achieve automatic early warning and hierarchical management of risky preforms.

[0052] S63. When the production control system receives the identification result of risky preforms, it automatically triggers the system warning mechanism and takes corresponding treatment measures based on the risk level: high-risk preforms are automatically rejected and arranged for manual review; medium-risk preforms are reviewed and, based on the actual situation, it is decided whether to proceed to the subsequent process or fine-tune the process parameters; low-risk preforms are automatically released to the subsequent process;

[0053] S64. The processing status and results of all risky preforms are transmitted back to the risk identification system to achieve closed-loop management. The process parameters of the relevant batches are analyzed based on the risk level statistical results, and targeted adjustment suggestions are made and fed back to production management personnel for improvement of subsequent production processes.

[0054] Optionally, the method further includes automatic archiving and tracing of historical preform production and testing data. The system can quickly retrieve the production batch, process parameters, sealing performance measurement values, and risk identification history of any preform by a unique number, thereby achieving quality traceability and data analysis of the entire life cycle of the preform.

[0055] Optionally, the sealing risk identification result data table can be managed in a hierarchical manner according to user permissions. High-risk preform information is pushed first to the production quality manager and relevant decision-makers, while ordinary operators can only view the risk level statistics of this batch, effectively improving the data security and management flexibility of the system.

[0056] Beneficial effects

[0057] By introducing a fuzzy C-means clustering algorithm and a systematic data processing process, this invention first achieves comprehensive collection and precise archiving of multi-source process parameter and sealing performance test data during preform production, overcoming the shortcomings of traditional methods in data collection and management, such as incomplete information and poor traceability. Secondly, scientific data preprocessing methods such as data cleaning, outlier removal, and Z-score normalization effectively improve the quality and consistency of the raw data, laying a solid foundation for subsequent feature extraction and cluster analysis. By constructing multidimensional feature vectors and applying the t-SNE dimensionality reduction visualization method, this invention can deeply explore the intrinsic relationship between process parameters and sealing performance, optimize the feature space structure, and thus improve the accuracy and interpretability of cluster analysis.

[0058] Furthermore, by assigning multidimensional membership to each preform sample using the fuzzy C-means clustering algorithm, the present invention not only scientifically categorizes preform sealing risks but also fully reflects the ambiguity and process fluctuations between samples in actual production, achieving flexible and intelligent risk identification. By setting appropriate risk identification thresholds and a multi-level risk grading mechanism, the present invention promptly identifies high-risk preforms, automatically outputs risk level labels and management recommendations, and provides real-time feedback to the production control system.

[0059] The present invention supports automatic early warning, sorting and process adjustment, effectively improving the accuracy, automation and management efficiency of preform sealing risk identification, greatly enhancing the intelligence and real-time level of preform sealing quality control, and meeting the actual needs of modern manufacturing for efficient, accurate, full-process quality monitoring and risk early warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0061] Figure 1 This is a schematic diagram of the process framework of a preform sealing risk identification method based on fuzzy clustering of the present invention;

[0062] Figure 2 Schematic diagram of the fuzzy C-means cluster analysis steps of the present invention;

[0063] Figure 3 1. It is a schematic diagram of the bottle preform sealing risk level determination and graded processing feedback of the present invention. DETAILED DESCRIPTION

[0064] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0065] refer to Figure 1-3 , a preform sealing risk identification method based on fuzzy clustering, including:

[0066] S1. Generate a unique number for each preform and create an identity file. Collect and archive process parameters and sealing performance test data during the production process, and associate them to form a raw data table as the basic data set for subsequent processing;

[0067] S2. Clean each row of data in the original data table, remove missing values ​​and outliers, and then standardize the data using the Z-score standardization method to generate preprocessed preform data;

[0068] S3. Extract features from the pre-processed preform data, construct a multi-dimensional feature vector, form a feature dataset of the preform samples, and perform visual analysis on the feature dataset to facilitate early data diagnosis and feature optimization.

[0069] S4. Perform cluster analysis on the feature data set using a fuzzy C-means clustering algorithm to obtain the cluster center coordinates of each preform sample and the membership value relative to each cluster center. The iterative convergence condition is that the objective function is less than a set convergence threshold or the maximum number of iterations is reached;

[0070] S5. Determine the sealing risk level based on the membership distribution of each preform sample and its distance from the risk cluster center. Mark samples with a membership higher than a set threshold as preforms with sealing risks and assign risk level classification labels.

[0071] S6. Feedback the unique number and risk level information of the identified risky preforms to the preform production control system to achieve automatic early warning and graded management of sealing risk preforms, and provide optimization and adjustment suggestions for related process parameters.

[0072] In this embodiment, the S1 further includes:

[0073] S11. Assign a unique number to each preform and create an identity file. Archive all process parameters and sealing performance test data during the production process to ensure accurate and reliable data management.

[0074] S12, the process parameters include injection mold cavity temperature T m (℃), injection pressure P (MPa), injection time t inj (s), cooling time t cool (s), preform length L (mm), preform outer diameter D (mm), wall thickness H (mm);

[0075] S13. The sealing performance test data is a quantitative measurement value S (kPa) obtained by performing a sealing performance test on each preform, which is used to reflect the sealing ability of the preform under standard test conditions;

[0076] S14, the process parameters of each preform {T m ,P,t inj ,t cool , L, D, H} are associated with the corresponding quantitative measurement value S of the sealing performance and stored in a structured table format to generate an original data table. Each row represents the data information of a preform, which serves as the basic data set for subsequent data processing.

[0077] In this embodiment, the S2 further includes:

[0078] S21. Perform integrity check on each row of data in the original data table, remove missing data records, and perform outlier detection using the triple standard deviation method to remove abnormal data that exceeds the set threshold range;

[0079] S22, perform Z-score normalization on the data after removing outliers, specifically for each parameter x according to the formula Standardization is performed, where the parameter x is derived from the process parameters of the preform {T m ,P,t inj ,t cool , L, D, H} and the quantitative measurement value S of the sealing performance, μ is the mean value of the parameter x, σ is the standard deviation of the parameter x, and z is the standardized parameter value;

[0080] For example, for the injection mold cavity temperature T m , if a batch of original data is T m,1 =260℃,T m,2 =255℃,T m,3 =265℃,T m,4 =258℃, then the mean is The standard deviation is After Z-score standardization, it is converted into

[0081] S23, all parameters after Z-score standardization As pre-processing of preform data, it provides a data basis for subsequent feature extraction and analysis.

[0082] In this embodiment, the S3 further includes:

[0083] S31. Extract features from the pre-processed preform data and construct a multi-dimensional feature vector for each preform.

[0084] S32. Summarize the feature vectors F of all preform samples to form a feature dataset of preform samples. Where n is the number of samples;

[0085] S33, using t-distributed stochastic neighbor embedding (t-SNE) dimensionality reduction method to reduce the feature dataset Perform visual analysis, map multidimensional feature data into two-dimensional space, draw scatter plots and heat maps, and display the distribution, clustering trends, and outliers of each preform sample in the feature space.

[0086] In this embodiment, the S4 further includes:

[0087] S41. Based on the preform sample feature dataset The actual scale and distribution characteristics of the clusters are used to determine the specific value of the number of clusters c in the fuzzy C-means clustering algorithm;

[0088] S42, using K-means++ algorithm to extract the feature data set of preform samples Select c representative feature vectors as the initial cluster centers, expressed as {v1,v2,...,v c}, and for each preform sample F i Assign initial membership u ij , where i represents the i-th preform sample, j represents the j-th cluster center, and u ij Indicates the membership of the i-th sample to the j-th class, and the sum of all memberships is equal to 1, that is,

[0089] S43. Feature Dataset Cluster analysis is performed, which specifically includes the following iterative process: ① In each iteration, according to the current cluster center {v j}, calculate each preform sample F i Euclidean distance d to each cluster center ij , the formula is:

[0090] d ij =‖F i -v j ‖;

[0091] ②Update the membership u of each sample ij , the formula is:

[0092]

[0093] Where m is the fuzzification coefficient, and m>1; ③ According to the new membership, update each cluster center v j , the formula is:

[0094]

[0095] S44, repeat the cluster center and membership degree update process in step S43 until the objective function J meets the convergence condition or reaches the maximum number of iterations. The convergence condition is:

[0096]

[0097] Among them, ∈ is the convergence threshold set according to the size of the data set and actual needs;

[0098] S45. Output the cluster center coordinates v of each preform sample j and the membership value u relative to each cluster centerij , used for subsequent sealing risk level determination.

[0099] In this embodiment, the S5 further includes:

[0100] S51. According to the cluster analysis results, determine the cluster centers v j The cluster center corresponding to the poor sealing performance is the risk cluster center v risk ;

[0101] S52. For each preform sample F i , obtain its membership degree u relative to the risk cluster center i,risk , and preform sample F i With v risk The Euclidean distance d between i,risk , the formula is:

[0102] d i,risk =‖F i -v risk ‖;

[0103] S53, let the risk identification threshold be θ, the classification interval parameters be α1, α2, satisfying 0<α1<α2≤1; when u i,risk ≥α2, and When the preform sample F i Determined as high risk and generated a high risk label, where: For all preform samples judged as risky to the risk cluster center v risk The mean of the Euclidean distance; when α1≤u i,risk <α2, and d i,risk near When the preform sample F i It is judged as medium risk and generates a medium risk label; when θ≤u i,risk <α1, or Then the preform sample F i Determine as low risk and generate a low risk label, where δ is the tolerance parameter for distance classification; samples that do not fall within the above range are determined to be non-risk preforms;

[0104] For example, let θ = 0.7, α1 = 0.8, α2 = 0.9, δ=0.1;If a preform sample u i,risk =0.92 and d i,risk =0.45, it is high risk; if u i,risk =0.85 and d i,risk =0.52, it is medium risk; if u i,risk =0.75 and d i,risk =0.62, which is low risk.

[0105] In this embodiment, the S6 further includes:

[0106] S61. The unique numbers, risk level labels, risk membership values, and Euclidean distances from the risk cluster centers of the identified preforms at each risk level are collated to form a sealing risk identification result data table.

[0107] S62. Feedback the sealing risk identification result data sheet to the preform production control system in real time through a standardized data interface, industrial Ethernet, or other network communication methods, to achieve automatic early warning and graded management of risky preforms;

[0108] S63. When the production control system receives the identification results of risky preforms, it automatically triggers the system's early warning mechanism and takes corresponding measures based on the risk level: high-risk preforms are automatically removed from the production line and sent to a manual re-inspection station. Quality inspectors conduct a second inspection of their sealing performance and appearance structure to confirm the cause of the defect. If high-risk preforms appear continuously, the system immediately analyzes the process fluctuations of the relevant batches and issues an alarm, suggesting adjustments to the injection molding temperature or pressure, and checking the mold status. Medium-risk preforms are assigned to sampling re-inspections by the system. Quality management personnel conduct key inspections according to the review checklist and, based on the system's process analysis results, make timely adjustments to process parameters. Low-risk preforms are directly released to the subsequent process, and the system archives their data for trend analysis and process optimization. If the sealing pressure parameters of low-risk preforms deviate from the mean for a long time, the system will issue a trend warning in advance.

[0109] S64. The processing status and results of all risky preforms are transmitted back to the risk identification system to achieve closed-loop management. The process parameters of the relevant batches are analyzed based on the risk level statistical results, and targeted adjustment suggestions are made and fed back to production management personnel for improvement of subsequent production processes.

[0110] In this embodiment, the method also includes automatic archiving and tracing of historical preform production and testing data. The system can quickly retrieve the production batch, process parameters, sealing performance measurement values, and risk identification history of any preform using a unique number, thereby achieving quality traceability and data analysis throughout the preform life cycle.

[0111] In this embodiment, the sealing risk identification result data table can be managed in a hierarchical manner according to user permissions. High-risk preform information is preferentially pushed to production quality managers and relevant decision-makers, while ordinary operators can only view the risk level statistics of this batch, effectively improving the system's data security and management flexibility.

[0112] Example 1:

[0113] To validate the practical application of this invention, it was deployed on the automated production line of a beverage packaging company with an annual output of tens of millions of PET preforms. This production line, which covers a wide range of bottle shapes, operates at a fast pace and features complex processes. In actual operation, the company has long faced issues such as insufficient precision in preform sealing performance testing, low manual inspection efficiency, limited automation, and inaccurate risk assessment in complex, multi-parameter environments. The company hopes to implement intelligent risk identification to accurately screen for sealing risks, provide graded early warnings, and optimize processes.

[0114] During the application of this invention, each preform is assigned a unique number on the production line. A data acquisition system records injection temperature, pressure, time, cooling time, preform length, outer diameter, and wall thickness in real time. The sealing pressure of the preforms is also measured after they leave the production line. All collected data is automatically archived into a structured raw data table, providing a foundation for subsequent analysis. From February to April 2024, the system collected complete data for approximately 340,000 preforms. First, the data processing module performed an integrity check and outlier removal on the raw data table, eliminating 0.6% of missing values ​​and 1.1% of outliers, ultimately obtaining high-quality data for approximately 336,000 preforms. Subsequently, all process parameter and sealing performance data were standardized using a Z-score to eliminate the influence of different parameter dimensions. The feature extraction module combined the process parameter and sealing performance data for each preform into an 8-dimensional feature vector, forming a complete preform sample feature dataset. Visual analysis of the dataset using the t-SNE dimensionality reduction method revealed a clear clustering of high-risk preforms in the feature space.

[0115] During the cluster analysis phase, a fuzzy C-means clustering algorithm is used to intelligently group samples. For example, during the week of March 2024, 112,000 preforms collected from the production line were automatically categorized into three risk groups: high-risk, medium-risk, and low-risk. High-risk preforms accounted for approximately 2.3%, medium-risk for 7.5%, and low-risk for 90.2%. The system automatically assigns a risk level label to each preform based on the degree of membership and Euclidean distance between cluster centers and assigns each of the three categories to different processing channels, achieving scientific diversion and targeted management. High-risk preforms are directly rejected by the system and sent to manual re-inspection, medium-risk preforms enter a focused sampling review process, and low-risk preforms are automatically released for downstream processing.

[0116] Compared with traditional methods, the present invention significantly improves the intelligence and automation level of risk identification. Before implementing the present invention, companies primarily relied on manual spot checks, resulting in a detection rate of only 0.04% for defective seals. Manual inspections were inefficient, and subjective misjudgments and missed detections were prominent, leading to some high-risk preforms flowing into downstream processes, posing quality risks. After implementing the present invention, the failure rate of high-risk preforms identified by the system after manual re-inspection was as high as 94.6%. The risk classification was highly consistent with the actual quality status, greatly improving the accuracy of risk identification. At the same time, the system implements automatic risk warning and sorting, shortening the entire risk identification and processing cycle to less than 2 minutes. The overall sorting and response efficiency of the production line has increased by more than 30%, significantly reducing the pressure and workload of manual inspections and ensuring efficient quality control in large-scale production environments. Table 1 shows the process parameters and sealing risk identification results of a typical batch of preforms in vertical format.

[0117] Table 1: Typical batches of preform process parameters and sealing risk identification results

[0118]

[0119] As can be seen from the above examples and data, the present invention can efficiently and accurately identify preforms with abnormal sealing performance or risks of process fluctuations, implement risk grading and automated processing, significantly improve the intelligence and real-time level of production line quality control, reduce the burden of manual inspection, and meet the management needs of modern manufacturing enterprises for high quality, high efficiency, and full-process traceability.

[0120] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A method for identifying preform sealing risks based on fuzzy clustering, characterized in that: The steps include: S1. Generate a unique number for each preform and create an identity file. Collect and archive the process parameters and sealing performance test data during the production process, and associate them to form an original data table. S2. Clean each row of data in the original data table, remove missing values ​​and outliers, and then standardize the data using the Z-score standardization method to generate preprocessed preform data; S3. Extract features from the preprocessed preform data, construct a multi-dimensional feature vector, form a feature dataset of the preform samples, and perform visual analysis on the feature dataset; S4. Perform cluster analysis on the feature data set using a fuzzy C-means clustering algorithm to obtain the cluster center coordinates of each preform sample and the membership value relative to each cluster center. The iterative convergence condition is that the objective function is less than a set convergence threshold or the maximum number of iterations is reached; S5. Determine the sealing risk level based on the membership distribution of each preform sample and its distance from the risk cluster center. Mark samples with a membership higher than a set threshold as preforms with sealing risks, and assign a graded risk level. S6. Feedback the unique number and risk level information of the identified risky preforms to the preform production control system to achieve automatic early warning and graded management of sealing risk preforms, and provide optimization and adjustment suggestions for related process parameters.

2. The method for identifying preform sealing risks based on fuzzy clustering according to claim 1, characterized in that: Said S1 further comprises: S11. Assign a unique number to each preform and create an identity file, and archive all process parameters and sealing performance test data during the production process; S12, the process parameters include injection mold cavity temperature T m (℃), injection pressure P (MPa), injection time t inj (s), cooling time t cool (s), preform length L (mm), preform outer diameter D (mm), wall thickness H (mm); S13, the sealing performance test data is a quantitative measurement value S (kPa) obtained by performing a sealing performance test on each preform; S14, the process parameters of each preform {T m ,P,t inj ,t cool , L, D, H} are associated with the corresponding quantitative measurement value S of the sealing performance and stored in a structured table format to generate an original data table, where each row represents the data information of a preform.

3. The method for identifying preform sealing risks based on fuzzy clustering according to claim 1, characterized in that: Said S2 further comprises: S21. Perform integrity check on each row of data in the original data table, remove missing data records, and perform outlier detection using the triple standard deviation method to remove abnormal data that exceeds the set threshold range; S22, after removing missing values ​​and outliers, perform Z-score standardization on the data. Specifically, for each parameter x, use the formula Standardization is performed, where the parameter x is derived from the process parameters of the preform {T m ,P,t inj ,t cool , L, D, H} and the quantitative measurement value S of the sealing performance, μ is the mean value of the parameter x, σ is the standard deviation of the parameter x, and z is the standardized parameter value; S23, all parameters after Z-score standardization As preprocessing preform data.

4. The method for identifying preform sealing risks based on fuzzy clustering according to claim 1, characterized in that: Said S3 further comprises: S31. Extract features from the pre-processed preform data and construct a multi-dimensional feature vector for each preform. S32. Summarize the feature vectors F of all preform samples to form a feature dataset of preform samples. Where n represents the number of samples; S33, using t-distributed stochastic neighbor embedding (t-SNE) dimensionality reduction method to reduce the feature dataset Perform visual analysis, map multi-dimensional feature data into two-dimensional space, and draw scatter plots and heat maps.

5. The method for identifying preform sealing risks based on fuzzy clustering according to claim 1, characterized in that: Said S4 further comprises: S41. Based on the preform sample feature dataset The actual scale and distribution characteristics of the clusters are used to determine the specific value of the number of clusters c in the fuzzy C-means clustering algorithm; S42, using K-means++ algorithm to extract the feature data set of preform samples Select c representative feature vectors as the initial cluster centers, expressed as {v1,v2,...,v c }, and for each preform sample F i Assign initial membership u ij , where i represents the i-th preform sample, j represents the j-th cluster center, and u ij Indicates the membership of the i-th sample to the j-th class, and the sum of all memberships is equal to 1, that is, S43. Feature Dataset Perform cluster analysis, which specifically includes the following iterative process: ① In each iteration, according to the current cluster center {v j }, calculate each preform sample F i Euclidean distance d to each cluster center ij , the formula is: d ij =‖F i -v j ‖; ② Update the membership degree u of each preform sample ij , the formula is: Where m is the fuzzification coefficient, and m>1; ③ According to the new membership, update each cluster center v j , the formula is: S44, repeat the updating process of cluster centers and membership degrees in step S43 until the objective function J meets the convergence condition or reaches the maximum number of iterations. The convergence condition is: Among them, ∈ is the convergence threshold set according to the size of the data set and actual needs; S45. Output the cluster center coordinates v of each preform sample j and the membership value u relative to each cluster center ij .

6. The method for identifying preform sealing risks based on fuzzy clustering according to claim 1, characterized in that: Said S5 further comprises: S51, according to the cluster analysis results, each cluster center v j The cluster center with poor sealing performance is called the risk cluster center v risk ; S52. For each preform sample F i , get its relative to the risk cluster center v risk The membership degree u i,risk , and preform sample F i With v risk The Euclidean distance d between i,risk , the formula is: d i,risk =‖F i -v risk ‖; S53, let the risk identification threshold be θ, the classification interval parameters be α1, α2, satisfying 0<α1<α2≤1; when u i,risk ≥α2, and When the preform sample F i Determined as high risk and generated a high risk label, where: For all preform samples judged as risky to the risk cluster center v risk The mean of the Euclidean distance; when α1≤u i,risk <α2, and d i,risk near When the preform sample F i It is judged as medium risk and generates a medium risk label; when θ≤u i,risk <α1, or Then the preform sample F i The sample is judged as low risk and a low risk label is generated, where δ is the tolerance parameter for distance classification. Samples that do not fall within the above range are judged as non-risk preforms.

7. The method for identifying preform sealing risks based on fuzzy clustering according to claim 1, characterized in that: Said S6 further comprises: S61. Generate a sealing risk identification result data table based on the unique number of the identified risk level preform, the risk level label, the risk membership value, and the Euclidean distance from the risk cluster center; S62. Feedback the sealing risk identification result data sheet to the preform production control system in real time through a standardized data interface, industrial Ethernet, or other network communication methods; S63. When the production control system receives the identification result of risky preforms, it automatically triggers the system warning mechanism and takes corresponding treatment measures based on the risk level: high-risk preforms are automatically rejected and arranged for manual review; medium-risk preforms are reviewed and, based on the actual situation, it is decided whether to proceed to the subsequent process or fine-tune the process parameters; low-risk preforms are automatically released to the subsequent process; S64. The processing status and results of all risky preforms are transmitted back to the risk identification system to achieve closed-loop management. The process parameters of the relevant batches are analyzed based on the risk level statistical results, and targeted adjustment suggestions are made and fed back to production management personnel for improvement of subsequent production processes.

8. The method for identifying preform sealing risks based on fuzzy clustering according to claim 1, characterized in that: The method also includes automatic archiving and tracing of historical preform production and testing data. The system can quickly retrieve the production batch, process parameters, sealing performance measurements and risk identification history of any preform by unique number.

9. The method for identifying preform sealing risks based on fuzzy clustering according to claim 1, characterized in that: The sealing risk identification result data table can be managed in a hierarchical manner according to user permissions. High-risk preform information is pushed to the production quality manager and relevant decision-makers first, and ordinary operators can only view the risk level statistics of this batch.

Images