Glass production line cold end stress detection system and method

Abstract

The invention provides a glass production line cold end stress detection system and method. Belongs to the technical field of glass production detection. The method comprises the following steps: carrying out stress-related spectral data acquisition on a cold end area of the glass production line through a multispectral stress sensing array, and preprocessing the acquired spectral data to obtain a preprocessed spectral data set; spectral feature extraction is carried out on the preprocessed spectral data set, and a spectral feature vector set is constructed; based on the spectral feature vector set, utilizing a space-time correlation analysis algorithm to extract space-time features of glass cold end stress to obtain a stress space-time feature matrix; spectral data are collected through the multispectral stress sensing array, the influence of traditional contact type detection on glass products and a production line is avoided, and the detection precision and the real-time performance are improved.

Description

Technical Field

[0001] The invention provides a cold end stress detection system and method for a glass production line, belonging to the technical field of glass production detection. Background Art

[0002] In the glass production process, the cold-end stage is crucial to glass quality. Stress is generated during the cooling process. If the stress distribution is uneven or excessive, it can cause cracks and breakage in the glass, seriously impacting product quality and production efficiency. Currently, existing methods for detecting cold-end stress in glass suffer from low accuracy, an inability to fully reflect stress distribution in real time, and difficulty accurately locating areas of abnormal stress. These issues make them difficult to meet the high-quality and efficient testing requirements of modern glass production. Therefore, developing an efficient and accurate method for detecting cold-end stress in glass production lines is of great practical significance. Summary of the Invention

[0003] The present invention provides a cold end stress detection system and method for a glass production line to solve the problems mentioned in the above background technology:

[0004] The present invention proposes a method for detecting cold end stress of a glass production line, characterized in that the method comprises:

[0005] S1: Use a multi-spectral stress sensing array to collect stress-related spectral data from the cold end area of ​​the glass production line, and preprocess the collected spectral data to obtain a preprocessed spectral dataset;

[0006] S2: Extract spectral features from the preprocessed spectral dataset and construct a spectral feature vector set. Based on the spectral feature vector set, use the spatiotemporal correlation analysis algorithm to extract the spatiotemporal characteristics of the cold end stress of the glass and obtain the stress spatiotemporal feature matrix.

[0007] S3: Collect historical data on abnormal stress cases at the cold end of glass, perform feature clustering analysis on the stress spatiotemporal feature matrix, and obtain multiple stress feature clusters. Combined with the historical data on abnormal stress cases at the cold end of glass, perform abnormal feature matching on each stress feature cluster to screen out feature clusters with potential stress anomalies, and extract the stress anomaly feature vectors of the feature clusters.

[0008] S4: Based on historical glass cold end stress anomaly case data, a stress anomaly pattern feature library is constructed. Based on the stress anomaly feature vector and the stress anomaly pattern feature library, a pattern matching algorithm is used to identify stress anomaly patterns and obtain a list of candidate stress anomaly patterns. Similarity calculation is performed on each pattern in the candidate stress anomaly pattern list to generate a stress anomaly similarity vector.

[0009] S5: Collect data on stress-related factors at the cold end of the glass production line and construct a stress-related factor dataset. Based on historical data on abnormal stress cases at the cold end of the glass and the stress-related factor dataset, use a machine learning algorithm to establish a stress anomaly probability prediction model and obtain an initial stress anomaly probability vector. Dynamically modify the initial stress anomaly probability vector based on the stress-related factor dataset to obtain stress anomaly probability distribution data.

[0010] S6: Probability sorting of the candidate stress anomaly pattern list is performed according to the stress anomaly probability distribution data to obtain a stress anomaly pattern sorting list; equipment layout data and glass transmission path data of the glass production line are obtained; based on the stress anomaly pattern sorting list, the equipment layout data and the glass transmission path data, stress anomaly area positioning analysis is performed to obtain stress anomaly area positioning result data; based on the stress anomaly area positioning result data and the stress anomaly pattern feature library, stress anomaly cause analysis is performed to generate a cold end stress detection report for the glass production line.

[0011] The present invention proposes a cold-end stress detection system for a glass production line, comprising a memory, a processor, and a computer program stored in and executable on the memory. The processor executes the program to implement any of the above-described cold-end stress detection methods for a glass production line.

[0012] The beneficial effects of the present invention are as follows: by collecting spectral data through a multi-spectral stress sensing array, the impact of traditional contact detection on glass products and production lines is avoided, and the detection accuracy and real-time performance are improved.

[0013] Combining spectral feature extraction, spatiotemporal correlation analysis, and feature clustering algorithms, it can automatically identify abnormal patterns in the stress distribution of the cold end of glass, improving the intelligent judgment capability of the detection system.

[0014] By analyzing and modeling historical abnormal cases, an abnormal pattern library containing spectral characteristics, stress distribution characteristics and defect information is formed, providing a reliable basis for subsequent pattern matching and enhancing the system's learning and adaptability.

[0015] By using pattern matching algorithms and similarity calculations, we can effectively identify whether the current stress state matches the known abnormal pattern and quantify the similarity, thereby improving the accuracy and robustness of abnormality identification.

[0016] By introducing influencing factors such as process parameters and environmental parameters and combining them with machine learning algorithms, a stress anomaly probability prediction model is established, which can be dynamically adjusted based on real-time data to improve the scientificity and practicality of the prediction.

[0017] Combining equipment layout and glass transmission path data can accurately locate the specific area where stress anomalies occur, and conduct cause analysis based on the feature library, which facilitates rapid response and troubleshooting.

[0018] The final stress detection report not only includes abnormal information, but also provides cause analysis and regional positioning, providing strong data support for production process adjustment, equipment maintenance and quality control. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 A diagram showing the steps of the method of the present invention;

[0020] Figure 2 For the present invention Figure 1 Figure 2 shows the steps in step S2. DETAILED DESCRIPTION

[0021] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0022] One embodiment of the present invention, as Figure 1 As shown, a cold end stress detection method for a glass production line comprises:

[0023] S1: Use a multi-spectral stress sensing array to collect stress-related spectral data from the cold end area of ​​the glass production line, and preprocess the collected spectral data to obtain a preprocessed spectral dataset;

[0024] S2: Extract spectral features from the preprocessed spectral dataset and construct a spectral feature vector set. Based on the spectral feature vector set, use the spatiotemporal correlation analysis algorithm to extract the spatiotemporal characteristics of the cold end stress of the glass and obtain the stress spatiotemporal feature matrix.

[0025] S3: Collect historical data on abnormal stress cases at the cold end of glass, perform feature clustering analysis on the stress spatiotemporal feature matrix, and obtain multiple stress feature clusters. Combined with the historical data on abnormal stress cases at the cold end of glass, perform abnormal feature matching on each stress feature cluster to screen out feature clusters with potential stress anomalies, and extract the stress anomaly feature vectors of the feature clusters.

[0026] S4: Based on historical glass cold end stress anomaly case data, a stress anomaly pattern feature library is constructed. Based on the stress anomaly feature vector and the stress anomaly pattern feature library, a pattern matching algorithm is used to identify stress anomaly patterns and obtain a list of candidate stress anomaly patterns. Similarity calculation is performed on each pattern in the candidate stress anomaly pattern list to generate a stress anomaly similarity vector.

[0027] S5: Collect data on stress-related factors at the cold end of the glass production line and construct a stress-related factor dataset. Based on historical data on abnormal stress cases at the cold end of the glass and the stress-related factor dataset, use a machine learning algorithm to establish a stress anomaly probability prediction model and obtain an initial stress anomaly probability vector. Dynamically modify the initial stress anomaly probability vector based on the stress-related factor dataset to obtain stress anomaly probability distribution data.

[0028] S6: Probability sorting of the candidate stress anomaly pattern list is performed according to the stress anomaly probability distribution data to obtain a stress anomaly pattern sorting list; equipment layout data and glass transmission path data of the glass production line are obtained; based on the stress anomaly pattern sorting list, the equipment layout data and the glass transmission path data, stress anomaly area positioning analysis is performed to obtain stress anomaly area positioning result data; based on the stress anomaly area positioning result data and the stress anomaly pattern feature library, stress anomaly cause analysis is performed to generate a cold end stress detection report for the glass production line.

[0029] The working principle of the above technical solution is as follows: using a multi-spectral stress sensing array to collect spectral data from the cold end area of ​​the glass production line. This data is closely related to the stress state of the glass; the collected raw spectral data is preprocessed to eliminate noise, enhance signals, etc., to obtain a higher quality preprocessed spectral data set, providing a reliable data foundation for subsequent analysis.

[0030] Spectral features are extracted from the preprocessed spectral dataset to construct a set of spectral feature vectors. These feature vectors can reflect different aspects of the stress at the cold end of the glass. Using a spatiotemporal correlation analysis algorithm, the spatiotemporal characteristics of the stress at the cold end of the glass are extracted from the spectral feature vector set to form a stress spatiotemporal feature matrix. This step helps us understand the distribution and variation of stress in time and space.

[0031] Historical data on abnormal stress at the cold end of glass was collected, including the spectral characteristics, stress distribution, and corresponding glass defect information during stress anomalies, providing a reference for anomaly feature matching. Cluster analysis was performed on the stress spatiotemporal feature matrix to obtain multiple stress feature clusters. Combined with the historical case data, anomaly feature matching was performed on each cluster to identify clusters with potential stress anomalies. The stress anomaly feature vectors for these clusters were then extracted.

[0032] Based on historical case data, a stress anomaly pattern feature library is constructed, which contains the spectral characteristics, stress distribution characteristics, and glass defect characteristics of different stress anomaly patterns. A pattern matching algorithm is used to identify a list of candidate stress anomaly patterns based on the stress anomaly feature vector and the stress anomaly pattern feature library. A similarity calculation is performed on each pattern in the candidate pattern list to generate a stress anomaly similarity vector, which is used to evaluate the degree of match between each pattern and the actual stress anomaly.

[0033] The process parameter data, environmental parameter data and other stress-related factor data of the cold end of the glass production line are collected to construct a stress-related factor dataset. Using a machine learning algorithm, a stress anomaly probability prediction model is established based on historical case data and the stress-related factor dataset to obtain an initial stress anomaly probability vector. The initial probability vector is dynamically corrected according to the stress-related factor dataset to obtain more accurate stress anomaly probability distribution data.

[0034] According to the stress anomaly probability distribution data, the candidate stress anomaly pattern list is probability-sorted to obtain a stress anomaly pattern sorting list, which facilitates the priority processing of high-probability anomaly patterns; the equipment layout data and glass transmission path data of the glass production line are obtained, and combined with the stress anomaly pattern sorting list, the stress anomaly area positioning analysis is performed to determine the specific location where the stress anomaly occurs; based on the stress anomaly area positioning result data and the stress anomaly pattern feature library, the cause of the stress anomaly is analyzed and a cold-end stress detection report of the glass production line is generated to provide decision support for production line maintenance and optimization.

[0035] The effect of the above technical solution is: through the multi-spectral stress sensing array and spatiotemporal correlation analysis algorithm, the stress data of the cold end area of ​​the glass production line can be captured and analyzed more accurately, reducing the missed detection or misjudgment caused by the shortcomings of traditional detection methods.

[0036] Through machine learning and pattern matching algorithms, stress anomaly patterns can be automatically identified, significantly reducing reliance on manual inspection and judgment and improving detection efficiency.

[0037] By combining historical data and stress-related factor data and establishing a stress anomaly probability prediction model, it is possible to predict possible stress anomalies at the cold end of the glass in advance and take preventive measures in advance, thereby reducing the risk of glass defects and downtime during the production process.

[0038] By accurately locating abnormal stress areas and analyzing the causes of the abnormalities, it helps to improve production processes and optimize equipment layout, thereby increasing the stability of the glass production line and ensuring more reliable quality of the produced glass.

[0039] By promptly identifying and locating stress anomalies, measures can be taken at an early stage to avoid the production of substandard glass, reducing defect rates and production losses.

[0040] This method comprehensively considers influencing factors such as environmental parameter data, and can dynamically adjust the stress anomaly probability prediction model to better adapt to production changes under different environmental conditions.

[0041] Cluster analysis and pattern recognition of stress characteristics can provide data support for subsequent process optimization and formulate more scientific production adjustment strategies based on historical cases and actual production conditions.

[0042] In one embodiment of the present invention, the S1 includes:

[0043] S11. Evenly arrange multi-spectral stress sensing arrays in key areas of the cold end of the glass production line; calibrate the arranged multi-spectral stress sensing arrays; measure their spectral responses under different stress conditions and establish a calibration curve between spectral response and stress;

[0044] S12. During the operation of the glass, the multispectral stress sensing array collects stress-related spectral data of the glass surface in real time; and transmits the collected spectral data to the data processing system in real time for preliminary storage and management;

[0045] S13, performing denoising processing on the collected original spectral data; and performing normalization processing on the denoised original spectral data; and then performing smoothing processing on the normalized spectral data to obtain a preprocessed spectral data set.

[0046] The working principle of this technical solution is to evenly distribute multispectral stress sensor arrays in key areas of the cold end of the glass production line, such as cooling rollers and conveyor belts. These areas are where glass stress variations are most pronounced, and the placement of the sensor arrays ensures accurate capture of stress variations on the glass surface. The deployed multispectral stress sensor arrays are then calibrated to ensure the accuracy of their measurement results. The calibration process involves measuring the sensor array's spectral response under different stress conditions and establishing a calibration curve between the spectral response and stress. This curve serves as the basis for subsequent conversion of spectral data to stress.

[0047] The glass production line is started, allowing the glass to pass through the cold end area at normal production speed. During this process, the multispectral stress sensing array collects real-time stress-related spectral data from the glass surface. This data is then transmitted to the data processing system for initial storage and management. This step ensures the timeliness and integrity of the data, providing a foundation for subsequent data processing and analysis.

[0048] Denoising is performed on the collected raw spectral data to eliminate noise and interference. This improves the signal-to-noise ratio, making subsequent analysis more accurate. Normalization is then performed on the denoised raw spectral data. Normalization brings data of varying magnitudes onto the same scale, facilitating subsequent comparison and analysis.

[0049] The normalized spectral data is smoothed to further eliminate minor fluctuations and noise. Smoothing can smooth the spectral data, improving its stability and reliability. After completing these steps, a preprocessed spectral dataset is obtained. This dataset contains denoised, normalized, and smoothed spectral data, providing a high-quality data foundation for subsequent feature extraction and stress analysis.

[0050] The effect of the above technical solution is: by evenly arranging the multi-spectral stress sensing array in the key area of ​​the cold end and calibrating it, the spectral response under different stress conditions is ensured to be accurate and stable, thereby providing high-quality stress data.

[0051] Through denoising, normalization and smoothing of spectral data, the interference of factors such as environmental noise and equipment errors is effectively reduced, making the final preprocessed spectral data set more accurate and convenient for subsequent analysis.

[0052] During the glass production process, spectral data is collected and transmitted to the data processing system in real time, ensuring continuous monitoring of the stress conditions in the cold end area of ​​the glass and timely detection of possible anomalies.

[0053] This method monitors the stress state of the cold end of the glass in real time and ensures that data collection does not interfere with the production process. It helps to perform accurate detection without affecting production efficiency and optimizes the production process.

[0054] The automated acquisition and data processing process of the multispectral stress sensing array reduces the complexity of manual operations, improves the automation level of the production line, and reduces the risks of human operations.

[0055] The collected spectral data is transmitted and stored in the data processing system in real time, and a systematic data management and storage method is established to facilitate subsequent data analysis and tracing, providing reliable historical data support.

[0056] Through standardized data preprocessing steps, the reliability of the data in subsequent spectral feature extraction and spatiotemporal feature analysis is ensured, which helps to improve the accuracy of the final stress detection results.

[0057] One embodiment of the present invention, as Figure 2 As shown, the S2 includes:

[0058] S21. Extracting stress-related spectral features from the preprocessed spectral data set; performing dimensionality reduction processing on the extracted spectral features using a feature selection algorithm; and constructing a spectral feature vector set;

[0059] S22. Processing the spectral feature vector set by a spatiotemporal correlation analysis algorithm to extract the spatiotemporal characteristics of the glass cold end stress;

[0060] S23. Organize and store the extracted spatiotemporal features according to a preset format to construct a stress spatiotemporal feature matrix, wherein the rows of the matrix represent different time points or spatial positions, the columns represent different spatiotemporal features, and the elements in the matrix represent the spatiotemporal feature values ​​corresponding to the time and spatial positions.

[0061] The working principle of the above technical solution is to extract stress-related spectral features from the preprocessed spectral data set. These features include but are not limited to spectral peaks, peak-to-valley positions, spectral bandwidth, and spectral integrated intensity. These features can reflect different aspects of glass surface stress and are an important basis for subsequent stress analysis. A feature selection algorithm is used to reduce the dimensionality of the extracted spectral features. Since the original spectral features may contain a large amount of redundant information, dimensionality reduction can remove these redundant features and retain the features most critical for stress analysis, thereby improving the efficiency and accuracy of subsequent analysis. The spectral features after dimensionality reduction are organized according to a certain format to construct a spectral feature vector set. This vector set will serve as the basic data for subsequent spatiotemporal correlation analysis.

[0062] The spectral feature vector set is processed using spatiotemporal correlation analysis algorithms, such as spatiotemporal clustering and spatiotemporal Markov models. These algorithms exploit the temporal and spatial correlations between spectral features, thereby extracting the spatiotemporal characteristics of the glass cold-end stress. The extracted spatiotemporal characteristics can reflect the temporal and spatial distribution and variation trends of stress. These characteristics are important for understanding the dynamic behavior of glass cold-end stress, predicting stress anomalies, and optimizing production processes.

[0063] The extracted spatiotemporal features are organized and stored according to a pre-set format. This step ensures the order and accessibility of the spatiotemporal features, facilitating subsequent analysis and processing. Based on the organized and stored spatiotemporal features, a stress spatiotemporal feature matrix is ​​constructed. In this matrix, rows represent different time points or spatial locations, columns represent different spatiotemporal features, and the elements in the matrix represent the spatiotemporal feature values ​​at the corresponding time and spatial locations. This matrix provides important data support for subsequent stress anomaly pattern recognition, probabilistic prediction, and regional positioning.

[0064] The effect of the above technical solution is: by extracting spectral features related to stress, it can more accurately capture the stress information of the cold end area of ​​​​the glass, providing more representative data for subsequent analysis.

[0065] The feature selection algorithm is used to reduce the dimensionality of the extracted spectral features, which reduces redundant data and makes data processing more efficient, while avoiding the computational burden and overfitting problems brought by high-dimensional data.

[0066] Through the spatiotemporal correlation analysis algorithm, the spatiotemporal characteristics of the stress at the cold end of the glass can be extracted more effectively, and the distribution law and changing trend of the stress in time and space can be comprehensively analyzed.

[0067] By extracting and analyzing spatiotemporal features, we can gain a deeper understanding of the dynamic changes and distribution characteristics of the cold-end stress of the glass, helping to identify potential production problems and anomalies.

[0068] The use of automated spatiotemporal correlation analysis algorithms reduces reliance on manual intervention, making the entire data processing process more intelligent and efficient, and reducing the impact of human factors on the results.

[0069] The extracted spatiotemporal features are organized and stored in a preset format, and a stress spatiotemporal feature matrix is ​​constructed to facilitate subsequent data query, analysis, and decision-making, thereby improving the systematicness and operability of data management.

[0070] By constructing a spatiotemporal feature matrix, accurate data support can be provided for subsequent stress anomaly prediction and process optimization, helping to formulate more scientific and targeted production adjustment strategies.

[0071] In one embodiment of the present invention, the step S23 includes:

[0072] According to the physical meaning and analysis requirements represented by spatiotemporal features, they are divided into time dimension features and space dimension features, and identifiers are defined for each type of feature;

[0073] For spatiotemporal eigenvalues ​​of different dimensions and value ranges, a unified standardization method is used to process them and map the eigenvalues ​​to specific numerical intervals;

[0074] Construct a time-space index system; in the time dimension, assign a unique index number to each time point in chronological order; in the spatial dimension, assign a corresponding index identifier to each spatial location based on the actual spatial layout of the glass cold end;

[0075] Dynamically program the dimensions of the stress spatiotemporal characteristic matrix. In the time dimension, determine the number of rows based on the time span of data acquisition and the time resolution required for analysis. In the spatial dimension, determine the number of columns based on the degree of spatial division of the cold end of the glass.

[0076] According to the time-space index system, the normalized time-space eigenvalues ​​are filled into the corresponding positions of the matrix one by one; in the case of missing data, the interpolation method is used to estimate and fill in the missing data; after all the eigenvalues ​​are filled in, a complete stress time-space characteristic matrix is ​​formed;

[0077] Perform quality verification on the constructed stress spatiotemporal characteristic matrix, and make corrections and optimizations based on the verification results;

[0078] The stress spatiotemporal characteristic matrix that has undergone quality verification and optimization is stored in a designated database or file system, and the key information of the matrix is ​​recorded to form a complete metadata record.

[0079] The working principle of this technical solution is to divide spatiotemporal features into time- and space-dimensional features based on their physical significance (e.g., the rate of change of stress over time reflects stress dynamics, and the spatial distribution gradient reflects stress heterogeneity in space) and analytical requirements (e.g., whether stress fluctuates periodically or localizes). Identifiers are defined for each feature type to facilitate accurate identification and reference during matrix construction, data processing, and analysis.

[0080] Different spatiotemporal features may have different dimensions and value ranges. Directly using raw eigenvalues ​​for matrix construction and analysis can lead to inaccurate or difficult-to-compare results. Using a unified normalization method (such as minimum-maximum normalization to map eigenvalues ​​to the [0, 1] interval) to process the eigenvalues ​​eliminates the effects of differences in dimensions and value ranges, making different features comparable within the matrix.

[0081] In the temporal dimension, each time point is assigned a unique index number in chronological order, ensuring that each time point has a unique, defined position in the matrix, facilitating the analysis and tracking of stress characteristics over time. Based on the actual spatial layout of the glass cold end, grid division and region coding are used to assign a corresponding index identifier to each spatial position, allowing the spatial position to be accurately located in the matrix and facilitating the analysis of the distribution and variation of stress in different spatial regions. This index system serves as the basis for locating matrix rows and columns, ensuring that spatiotemporal eigenvalues ​​are accurately placed in the corresponding positions in the matrix, ensuring a correspondence between the matrix structure and the actual spatiotemporal characteristic data.

[0082] The number of matrix rows is determined based on the time span of data acquisition and the required temporal resolution for analysis. The time span determines the time range covered by the matrix, while the required temporal resolution affects the accuracy of the matrix representation of each time point. The number of matrix columns is determined based on the level of spatial partitioning of the cold end of the glass. The finer the spatial partitioning, the more columns the matrix will have, enabling a more detailed representation of the spatial distribution of stress. Expansion space is reserved to accommodate future increases in data volumes or new spatiotemporal feature analysis requirements, improving the adaptability and flexibility of the matrix. Based on a temporal-spatial indexing system, the standardized spatiotemporal feature values ​​are sequentially applied to the corresponding positions in the matrix, ensuring that the matrix fully represents the stress characteristics at different temporal and spatial locations. For missing data at certain time points or spatial locations, appropriate interpolation methods (such as linear interpolation and spline interpolation) are used to estimate and fill in the gaps. This ensures the integrity and continuity of the matrix data and prevents subsequent analysis results from being affected by missing data.

[0083] Perform a quality check on the constructed stress spatiotemporal characteristic matrix to check for outliers (e.g., values ​​outside a reasonable range). Remove or replace outliers to ensure the accuracy and reliability of the matrix data. Check the data for inconsistencies (e.g., inconsistent eigenvalues ​​at the same time point or spatial location). Trace and adjust any inconsistencies to ensure data logic and consistency. Use efficient storage formats (e.g., sparse matrix storage format, especially when the matrix contains a large number of zero elements) and compression algorithms to reduce storage space usage, improve data read and write efficiency, and facilitate subsequent storage, management, and analysis of the matrix data.

[0084] The quality-verified and optimized stress spatiotemporal characteristic matrix is ​​stored in a designated database or file system to ensure data security and accessibility. Key information such as the matrix construction time, data source, and processing parameters is recorded to form a complete metadata record. This metadata record facilitates subsequent understanding, traceability, and reuse of the matrix data, making it easier for users to understand the matrix construction process and related background information.

[0085] The effect of the above technical solution is: by dividing the spatiotemporal characteristics into time dimension characteristics and space dimension characteristics, and defining identifiers for each type of characteristics, it is possible to more clearly analyze and understand the performance of stress changes in different dimensions, thereby enhancing the interpretability of the data.

[0086] A unified standardization method is used to process spatiotemporal eigenvalues ​​of different dimensions and value ranges, ensuring a unified scale for the eigenvalues ​​and effectively avoiding the deviation caused by dimensional differences between different features, thereby improving the consistency of data analysis.

[0087] By building a complete time-space index system, the precise positioning of spatiotemporal eigenvalues ​​is ensured, the dislocation or confusion of data in the matrix is ​​avoided, and the systematicness and accuracy of data management are improved.

[0088] When dynamically planning the dimensions of the spatiotemporal feature matrix, the time span and the fineness of spatial division are taken into consideration, and expansion space is reserved to ensure that the system can cope with future increases in data volume or new analysis requirements, thereby improving the system's flexibility and future adaptability.

[0089] For time points or spatial locations with missing data, reasonable interpolation methods are used to estimate and fill in the missing data, which effectively reduces the impact of missing data on the quality of the spatiotemporal feature matrix and ensures the integrity and continuity of the data.

[0090] By performing quality verification on the constructed stress spatiotemporal characteristic matrix, outliers or data inconsistencies can be discovered and corrected in a timely manner, thereby improving the accuracy and reliability of the data and ensuring the validity of the analysis results.

[0091] By optimizing the matrix storage structure and adopting sparse matrix storage format and compression algorithm, the storage space occupancy is reduced, the data reading and writing efficiency is improved, and the performance and response speed of large-scale data processing are further improved.

[0092] The optimized stress spatiotemporal characteristic matrix is ​​stored in a designated database or file system, and key information such as matrix construction time, data source, and processing parameters are recorded to form a complete metadata record to ensure data traceability and manageability.

[0093] In one embodiment of the present invention, S3 includes:

[0094] S31. Collect data on abnormal stress cases that occurred at the cold end of a glass production line over a period of time, establish a historical database of abnormal stress cases at the cold end of a glass production line, and organize and classify the collected historical case data;

[0095] S32. Using clustering algorithm, perform characteristic clustering analysis on stress spatiotemporal characteristic matrix;

[0096] S33. Combine historical glass cold end stress abnormality case data and perform abnormal feature matching on each stress feature cluster; compare the features of the cluster with the abnormal features in the historical cases and calculate the similarity; the similarity is obtained by the following formula:

[0097]

[0098] Among them, the cluster feature vector of A is [ , ,…, ]; the historical abnormal feature vector of B is [ , ,…, ]; n represents the feature dimension (such as time, position, spectral characteristics, etc.);

[0099] S34, setting a similarity threshold to screen out characteristic clusters with potential stress anomalies; for clusters whose similarity exceeds the threshold, determining whether there is stress anomaly;

[0100] S35. Extract stress anomaly feature vectors from the selected feature clusters; select the most representative features from the clusters to construct stress anomaly feature vectors.

[0101] The working principle of the above technical solution is to collect data on stress anomaly cases that occurred at the cold end of the glass production line over the past period of time. These data are an important basis for subsequent analysis. The case data covers basic information such as the time, location, glass specifications and batches of the stress anomaly, as well as key data such as the spectral characteristics, stress distribution and corresponding glass defect information when the stress anomaly occurred, which comprehensively reflects the various characteristics and impacts of stress anomalies. A historical database of glass cold-end stress anomaly cases is established to centrally store and manage the collected case data to facilitate subsequent query, call and analysis. The collected historical case data is sorted and classified, and divided according to the type and severity of the stress anomaly. This classification method helps to quickly locate stress anomaly cases of specific types or severities during subsequent analysis, thereby improving analysis efficiency and pertinence.

[0102] A clustering algorithm is used to perform feature clustering analysis on the stress spatiotemporal feature matrix. Clustering algorithms can group stress spatiotemporal data with similar characteristics into distinct clusters, thereby revealing the inherent patterns and structure of the stress spatiotemporal features. The number of clusters and similarity measurement method are determined based on the characteristics and distribution patterns of the stress spatiotemporal features. The selection of the number of clusters requires a comprehensive consideration of the complexity of the data and the analysis requirements, while the similarity measurement method determines how to measure the similarity between different data points. Clustering effectiveness metrics, such as the silhouette coefficient and the Davies-Bouldin index, are used to evaluate the quality of clustering. These metrics can reflect the quality of clustering results from different perspectives and help select the optimal clustering scheme. The characteristics and distribution of each cluster are analyzed to understand the characteristics of different stress spatiotemporal feature patterns. Through in-depth analysis of clusters, potential patterns and anomalies in the stress spatiotemporal features can be discovered, providing a basis for subsequent anomaly feature matching.

[0103] Based on historical data from abnormal glass cold-end stress cases, each stress feature cluster is matched for abnormality. The cluster's features are compared with the abnormal features in the historical cases, and similarity is calculated. This step aims to determine whether the current cluster's features are similar to those in the historical cases, thereby determining whether potential stress anomalies exist. Similarity calculation is a key step in abnormality feature matching. Using a reasonable similarity metric, the degree of similarity between cluster features and historical anomaly features can be accurately measured, providing a basis for subsequent screening and judgment. A similarity threshold is set to filter out feature clusters with potential stress anomalies. The setting of the similarity threshold needs to be adjusted based on actual conditions and analysis requirements, ensuring that true potential anomaly clusters are screened while avoiding false positives. For clusters whose similarity exceeds the threshold, a determination is made as to whether stress anomalies exist. This step, based on the similarity matching results, makes a preliminary anomaly judgment for the cluster, providing guidance for subsequent extraction and analysis of abnormal feature vectors.

[0104] Stress anomaly feature vectors are extracted from the selected characteristic clusters, and the most representative features are selected from the clusters. These representative features summarize the main characteristics and anomaly patterns of the clusters, providing key information for subsequent stress anomaly pattern recognition and analysis. The selected representative features are then constructed into stress anomaly feature vectors, which serve as important input for subsequent stress anomaly pattern recognition and analysis. This feature vector construction simplifies complex spatiotemporal stress characteristic information into a form that is easier to process and analyze, improving analysis efficiency and accuracy.

[0105] The effect of the above technical solution is: by collecting and organizing historical glass cold end stress anomaly case data, a complete database is established, which provides high-quality reference data for subsequent anomaly analysis and can more accurately identify the pattern of stress anomaly occurrence.

[0106] By organizing and classifying historical case data and dividing them according to the type and severity of stress anomalies, the data processing process is simplified and subsequent analysis work is made more systematic and efficient.

[0107] The clustering algorithm is used to perform cluster analysis on the stress spatiotemporal characteristic matrix, which can effectively identify the distribution patterns of different stress characteristics. By using the clustering validity index, the reliability and accuracy of the clustering results are guaranteed, and the depth and precision of the analysis are improved.

[0108] By matching abnormal features of each stress feature cluster and comparing them with historical cases, potential stress anomaly patterns can be quickly discovered, improving the sensitivity and reliability of anomaly detection.

[0109] Setting a similarity threshold can screen out the characteristic clusters of potential stress anomalies. By setting a reasonable threshold, the risks of misjudgment and missed judgment can be effectively reduced, and the accuracy and practicality of anomaly detection can be improved.

[0110] By extracting stress anomaly feature vectors from the screened feature clusters and selecting the most representative features, the anomaly features can be accurately extracted, providing an accurate basis for subsequent anomaly diagnosis and early warning.

[0111] This solution can cope with new types of anomalies or new analysis needs that may increase in the future. With the increase in data volume and optimization of algorithms, the system has strong scalability and flexibility and can adapt to production environments of different scales and complexities.

[0112] Through the automated clustering and abnormal feature matching process, the need for manual intervention is reduced, work efficiency is improved, and the probability of human error is reduced, thereby ensuring the stability and production efficiency of the production line.

[0113] By effectively identifying potential stress anomalies and predicting abnormal characteristics, preventive measures can be taken in advance to avoid major failures, thereby improving the reliability and production quality of the overall production line.

[0114] In one embodiment of the present invention, the S4 includes:

[0115] S41. Based on historical glass cold end stress anomaly case data, construct a stress anomaly pattern feature library; standardize the data in the feature library to unify the dimensions and ranges of different features;

[0116] S42. Select a pattern matching algorithm and optimize the algorithm parameters according to the characteristics and distribution patterns of the stress anomaly features; use the stress anomaly feature vectors and corresponding stress anomaly patterns in the historical case data to train the algorithm and adjust the algorithm parameters;

[0117] S43, using a selected pattern matching algorithm to perform stress anomaly pattern recognition based on the stress anomaly feature vector and the stress anomaly pattern feature library; matching the stress anomaly feature vector with each pattern in the feature library and calculating similarity;

[0118] S44. Obtain a candidate stress anomaly pattern list based on the similarity calculation result; perform similarity calculation on each pattern in the candidate stress anomaly pattern list to generate a stress anomaly similarity vector.

[0119] The working principle of this technical solution is to construct a stress anomaly pattern feature library based on historical glass cold-end stress anomaly case data. This feature library comprehensively covers the spectral characteristics, stress distribution characteristics, and glass defect characteristics corresponding to different stress anomaly patterns. These features describe the manifestation of stress anomalies from different perspectives, providing a rich information foundation for subsequent pattern recognition. Because different features in the feature library may have different dimensions and ranges, directly using the raw data for matching and analysis will result in inaccurate results. Therefore, the data in the feature library is standardized to unify the dimensions and ranges of different features. This makes different features comparable in subsequent pattern recognition, improving recognition accuracy and reliability.

[0120] Based on the requirements for stress anomaly pattern recognition, appropriate pattern matching algorithms, such as cosine similarity matching and Euclidean distance matching, are selected. These algorithms can measure the similarity between two vectors from different perspectives and are suitable for matching stress anomaly feature vectors. The parameters of the selected algorithm are optimized based on the characteristics and distribution patterns of the stress anomaly features. Different parameter settings affect the algorithm's performance and recognition results. By optimizing the parameters, the algorithm can better adapt to the characteristics of stress anomaly features, improving recognition accuracy and efficiency. The algorithm is trained using stress anomaly feature vectors and corresponding stress anomaly patterns from historical case data. During the training process, the algorithm parameters are continuously adjusted to enable the algorithm to more accurately identify stress anomaly patterns in historical cases, thereby improving the algorithm's generalization ability and recognition accuracy.

[0121] Based on the stress anomaly feature vector and the stress anomaly pattern signature library, a selected pattern matching algorithm is used to identify stress anomaly patterns. The stress anomaly feature vector is matched against each pattern in the signature library, and similarity is calculated to measure the proximity between the feature vector and each pattern. Similarity calculation is a core step in pattern recognition. Using a reasonable similarity measurement method, it can accurately reflect the similarity between the feature vector and the pattern, providing a basis for subsequent pattern screening and decision-making.

[0122] Based on the similarity calculation results, a list of candidate stress anomaly patterns is generated. The patterns in the list are sorted from high to low by similarity. Patterns with higher similarity are more likely to be the stress anomaly pattern corresponding to the current stress state. This sorting method helps quickly locate the most likely stress anomaly pattern, improving fault diagnosis efficiency. A similarity calculation is performed on each pattern in the candidate stress anomaly pattern list to generate a stress anomaly similarity vector. The elements in this vector represent the similarity values ​​between the current stress state and each candidate stress anomaly pattern. These similarity values ​​intuitively reflect the degree of proximity between the current stress state and each candidate pattern, providing an important quantitative basis for subsequent fault diagnosis and decision-making.

[0123] The effect of the above technical solution is: by constructing a stress anomaly pattern feature library, the system can more comprehensively store and organize the spectral characteristics, stress distribution characteristics and glass defect characteristics of different stress anomaly patterns, providing detailed and standardized data support for pattern recognition, thereby greatly improving the accuracy of pattern recognition.

[0124] By standardizing the feature library data, the dimensions and ranges of different features are unified, effectively reducing the interference between different features, making pattern matching more concise and efficient, and avoiding processing difficulties caused by data inconsistency.

[0125] By selecting appropriate pattern matching algorithms (such as cosine similarity and Euclidean distance) and optimizing them according to the characteristics and distribution patterns of stress anomaly features, it can better adapt to different types of stress anomaly patterns and improve the flexibility and robustness of pattern recognition.

[0126] By optimizing algorithm parameters and using historical case data for training, the system can complete stress anomaly pattern recognition in a shorter time, significantly improving processing efficiency and reducing response delays to real-time data.

[0127] By generating a list of candidate stress anomaly patterns based on the similarity calculation results and sorting them by similarity, the most similar patterns can be matched first, thereby reducing the risk of misidentification and ensuring that the identified anomaly patterns are more consistent with the current stress state.

[0128] By generating stress anomaly similarity vectors, an accurate basis is provided for subsequent fault diagnosis and decision-making, ensuring accurate judgment and timely response to abnormal patterns in the decision-making process, and improving the reliability and effectiveness of the decision-making process.

[0129] By accumulating rich historical data in the feature library and continuously optimizing the pattern matching algorithm, the system can continuously evolve and adapt to new production scenarios and complex changes, improving the long-term maintainability and adaptability of the system.

[0130] Through the automated stress anomaly pattern recognition process, the reliance on manual judgment is reduced, the workload of operators is alleviated, the incidence of human errors is reduced, and the stability of the production line is further improved.

[0131] Through effective pattern matching and similarity calculation, the system can predict and warn of potential stress anomalies in advance, thereby helping to take preventive measures in a timely manner, reducing production line downtime, and improving production line operation efficiency and product quality.

[0132] In one embodiment of the present invention, the step S43 includes:

[0133] Perform a pre-alignment operation on the stress anomaly feature vector and the feature vectors of each mode in the feature library. Rearrange the current stress anomaly feature vector according to the definition and order of the features in the feature library to ensure that the feature order of the two is consistent. For features that exist in the feature library but are missing in the current feature vector, a specific filling strategy is used to supplement them.

[0134] The pre-aligned stress anomaly feature vector and the pattern feature vector in the feature library are split into multiple dimensions according to feature type; for features of different dimensions, corresponding weights are assigned according to their importance in stress anomaly pattern recognition;

[0135] Using the selected pattern matching algorithm, the similarity of each dimension feature vector after splitting is calculated respectively; by calculating the similarity of each dimension, the similarity value under each dimension is obtained;

[0136] The similarity values ​​calculated from the sub-dimensions are fused according to the pre-assigned weights to obtain the comprehensive similarity values ​​between the stress anomaly feature vector and each pattern in the feature library; the fused similarity values ​​are comprehensively evaluated to preliminarily screen out several candidate patterns with the top 20% similarity;

[0137] According to the similarity distribution of different stress anomaly patterns in historical data and the characteristics of the current stress state, the similarity threshold is dynamically adjusted; the candidate patterns initially screened out are compared with the dynamic threshold, and the candidate patterns with similarity higher than the threshold are further screened out as the final set of candidate stress anomaly patterns.

[0138] The working principle of the above technical solution is as follows: pre-align the stress anomaly feature vector with the feature vectors of each pattern in the feature library, and rearrange the current stress anomaly feature vector according to the definition and order of the features in the feature library to ensure that the feature order of the two is consistent. This operation is to ensure that in the subsequent similarity calculation, the elements in the same position in different feature vectors represent the same feature meaning, avoiding calculation errors caused by inconsistent feature order. For features that exist in the feature library but are missing in the current feature vector, a specific filling strategy is used (such as filling the average or median of the feature in the historical data) to supplement them. The presence of missing values ​​will affect the accuracy of the similarity calculation. Through a reasonable filling strategy, the integrity of the feature vector can be restored as much as possible, reducing the impact of missing values ​​on the results.

[0139] The pre-aligned stress anomaly feature vectors and pattern feature vectors in the feature library are then split into multiple dimensions based on feature type (spectral features, stress distribution features, and glass defect features). Different feature types play different roles and significance in stress anomaly pattern recognition. This split allows for analysis and processing of features in different dimensions, improving the targetedness and accuracy of recognition. Weights are assigned to features of different dimensions based on their importance in stress anomaly pattern recognition. This weighting can be determined based on expert experience, historical data analysis, or machine learning algorithms (such as entropy weighting). For example, spectral features may be more critical for identifying certain stress anomaly patterns and therefore be assigned a higher weight; whereas glass defect features may have less impact in certain situations and therefore be assigned a relatively lower weight. Appropriate weighting can highlight the role of important features in similarity calculation and improve the reliability of recognition results. A selected pattern matching algorithm (such as cosine similarity matching or Euclidean distance matching) is used to calculate similarity for each dimension of the split feature vectors. For the spectral feature dimension, due to the continuity and complexity of spectral data, further data processing can be performed on the spectral features (such as Fourier transform to extract frequency domain features) before calculating the similarity to better capture the key information in the spectral features.

[0140] For stress distribution feature dimensions, considering the spatial distribution characteristics of stress, similarity metrics based on spatial distribution can be used, such as calculating the spatial correlation coefficient of stress distribution, to accurately measure the similarity between stress distribution features. For glass defect feature dimensions, appropriate similarity calculation methods are used based on the type and severity of the defect, such as encoding the defect type and calculating the similarity of the encoding, or setting different similarity thresholds based on the defect severity. By calculating similarity across different dimensions, similarity values ​​are obtained for each dimension, providing a basis for subsequent comprehensive evaluation.

[0141] The similarity values ​​calculated for each dimension are fused according to pre-assigned weights to obtain the comprehensive similarity values ​​between the stress anomaly feature vector and each pattern in the feature library. During the fusion process, different fusion strategies can be adopted, such as weighted averaging and weighted geometric averaging, with the most appropriate method selected based on the actual situation. Furthermore, considering the correlation between similarities across dimensions, methods such as principal component analysis (PCA) can be used to reduce the dimensionality of the similarity vectors, extracting key information before fusion. This reduces the impact of redundant information on the results and improves the accuracy and stability of the fusion results. The fused similarity values ​​are comprehensively evaluated to preliminarily select several candidate patterns with the top 20% similarity. The purpose of this preliminary screening is to narrow the candidate pool and improve the efficiency of subsequent analysis.

[0142] The similarity threshold is dynamically adjusted based on the similarity distribution of different stress anomaly patterns in historical data and the characteristics of the current stress state (such as stress magnitude and change trend). Dynamic threshold adjustment makes the screening process more flexible and accurate, adapting to the stress anomaly pattern identification requirements in different situations. The initially screened candidate patterns are compared with the dynamic threshold, and candidate patterns with similarities exceeding the threshold are further selected as the final set of candidate stress anomaly patterns. This final set of candidate stress anomaly patterns more accurately reflects the stress anomaly patterns that may correspond to the current stress state, providing strong support for subsequent fault diagnosis and decision-making.

[0143] The effect of the above technical solution is: by performing a pre-alignment operation on the feature vector, the consistency of the feature order between the current stress anomaly feature vector and the pattern feature vector in the feature library is ensured, the matching error caused by inconsistent order is eliminated, and the accuracy of pattern matching is greatly improved.

[0144] The use of specific filling strategies (such as using the average or median in historical data to fill missing features) avoids pattern recognition failure or misidentification due to missing features, ensuring the integrity of feature data and the stability of recognition.

[0145] By splitting the feature vector into multiple dimensions such as spectral features, stress distribution features, and glass defect features, and calculating the similarity of each dimension separately, we can capture the contribution of each feature dimension to the stress anomaly pattern in more detail, thereby improving the overall recognition accuracy.

[0146] By assigning weights to features of different dimensions and dynamically adjusting the weights based on expert experience or machine learning algorithms (such as the entropy weight method), it is possible to more flexibly adjust the influence of features on pattern recognition according to the different characteristics of stress anomaly patterns, thereby further improving recognition accuracy.

[0147] By calculating the similarity of features in different dimensions separately and then performing weighted fusion, the complexity of high-dimensional data processing is effectively reduced, the interference of redundant information is reduced, and the efficiency of the pattern recognition process is ensured.

[0148] By adopting fusion strategies such as weighted average method or weighted geometric average method to fuse the sub-dimensional similarities, we can comprehensively consider the influence of each dimension and obtain a more comprehensive and reasonable similarity value, thereby improving the reliability of pattern recognition.

[0149] By adopting dimensionality reduction methods such as principal component analysis (PCA), the similarity vector is reduced in dimensionality, which reduces the interference of redundant information, further enhances the system's adaptability to different stress states, and improves the flexibility of pattern recognition.

[0150] By comprehensively evaluating the similarity values ​​after fusion, candidate patterns with high similarity can be accurately screened out, reducing the probability of misscreening and ensuring that the selected candidate patterns are more consistent with the current stress state.

[0151] By dynamically adjusting the similarity threshold, the screening criteria can be flexibly adjusted according to the characteristics of the current stress state, making pattern screening more targeted and improving the accuracy and reliability of pattern recognition.

[0152] By preliminarily screening out the top 20% of candidate patterns based on the similarity distribution, the scope of candidate patterns can be quickly narrowed down, the efficiency of subsequent diagnosis and decision-making can be improved, and unnecessary calculations and analysis can be reduced.

[0153] In one embodiment of the present invention, the S5 includes:

[0154] S51, collecting stress-related factor data at the cold end of the glass production line; and establishing a stress-related factor data acquisition system to collect and store the factor data in real time;

[0155] S52. Based on historical glass cold end stress anomaly case data and stress correlation factor datasets, a stress anomaly probability prediction model is established using a machine learning algorithm;

[0156] S53. Preprocess and feature engineer the data; integrate stress-related factor data and stress anomaly case data to extract features that have a significant impact on stress anomaly prediction;

[0157] S54, dividing the processed data into a training set and a test set, using the training set to train the machine learning model, adjusting the model parameters, and using the test set to evaluate and verify the performance of the model; using the trained stress anomaly probability prediction model, according to the current stress correlation factor data set, to obtain an initial stress anomaly probability vector;

[0158] S55. Dynamically modify the initial stress anomaly probability vector according to the real-time changes of the stress correlation factor data set to obtain stress anomaly probability distribution data.

[0159] The working principle of the above technical solution is to collect process parameter data and environmental parameter data, as well as other stress-related factor data, from the cold end of the glass production line. Process parameters such as cooling rate, temperature gradient, and glass thickness directly affect the stress state of the glass; environmental parameters such as temperature, humidity, and air pressure also have a certain impact on the stress of the glass. Comprehensively collecting this factor data provides a rich information foundation for subsequent stress anomaly probability prediction. A stress-related factor data collection system is established to achieve real-time data collection and storage. This real-time collection enables timely access to the latest data from the cold end of the glass production line, reflecting the current state of stress-related factors. The stored data provides historical data support for subsequent data analysis and model training.

[0160] Based on historical data on abnormal stress cases at the cold end of glass and a dataset of stress-related factors, a machine learning algorithm was used to develop a stress anomaly probability prediction model. The historical case data includes information on the occurrence of stress anomalies and the corresponding factors. The machine learning algorithm can then unearth the underlying relationship between the two, thereby constructing a model capable of predicting the probability of stress anomalies. Machine learning algorithms possess powerful data processing and pattern recognition capabilities, capable of learning complex patterns and regularities from large amounts of data. Different machine learning algorithms are suitable for different types of data and problems, and selecting the appropriate algorithm can improve the model's predictive performance.

[0161] Stress-related factor data and stress anomaly case data are integrated for unified analysis and processing. This integrated data includes information on the occurrence and corresponding factors of stress anomalies, providing a complete dataset for subsequent feature extraction and model training. Data cleaning is performed to remove noise and outliers to ensure data quality. Missing values ​​are also handled, such as by using mean filling, median filling, or model-based prediction filling to mitigate the impact of missing values ​​on model training. Feature encoding is performed to convert categorical variables into numerical variables that can be processed by machine learning algorithms. Features that are important for stress anomaly prediction are extracted, and feature selection methods are used to identify the most representative features, reducing data dimensionality and redundant information, thereby improving model training efficiency and predictive performance. The processed data is divided into a training set and a test set. The training set is used for model training, enabling the model to learn patterns and regularities in the data; the test set is used to evaluate and validate the model's performance on unseen data.

[0162] Use the training set to train the machine learning model and adjust the model parameters to better fit the training data. By continuously adjusting the parameters, optimize the model's performance and improve its prediction accuracy. Use the test set to evaluate and verify the model's performance, and select the model with the best performance as the stress anomaly probability prediction model. Evaluation metrics can include accuracy, recall, and F1 value, and appropriate evaluation metrics should be selected based on actual needs. Using the trained stress anomaly probability prediction model, an initial stress anomaly probability vector is generated based on the current stress correlation factor dataset. The elements in the vector represent the initial probabilities of different regions or different stress anomaly patterns. These initial probabilities reflect the likelihood of different stress anomaly patterns occurring under the current stress correlation factor conditions.

[0163] The initial stress anomaly probability vector is dynamically revised based on real-time changes in the stress correlation factor dataset. Stress correlation factor data is constantly changing, and these changes may affect the probability of stress anomalies. By monitoring changes in stress correlation factor data in real time and adjusting the stress anomaly probability vector in a timely manner, we can more accurately reflect the current stress anomaly probability distribution. After dynamic revision, the stress anomaly probability distribution data is obtained. This data provides glass production line operators with real-time stress anomaly warning information, helping them to promptly identify potential stress anomalies and take appropriate measures to adjust and optimize, thereby ensuring the quality and stability of glass production.

[0164] The effect of the above technical solution is: by collecting process parameter and environmental parameter data at the cold end of the glass production line and establishing a stress-related factor data acquisition system, real-time data collection and storage are achieved, ensuring the comprehensiveness and timeliness of the data, and providing more accurate input data for subsequent stress anomaly probability prediction.

[0165] By using historical glass cold-end stress anomaly case data and stress-related factor datasets, combined with machine learning algorithms, a stress anomaly probability prediction model was established, which effectively reduced the errors that may be caused by traditional empirical judgment methods and improved the reliability and accuracy of the prediction.

[0166] Through data preprocessing and feature engineering, stress-related factor data and historical abnormal case data are integrated and cleaned, missing value processed, and feature encoded. This effectively extracts features that have an important impact on stress anomaly prediction, enhances the ability to process complex data, and improves the input quality of the model.

[0167] By dividing the processed data into training sets and test sets, using the training set to train the model and using the test set to verify the model performance, the model parameters can be dynamically adjusted and the model with the best performance can be selected, thereby improving the adaptability of the model and ensuring the prediction accuracy of the model under different stress states.

[0168] By using methods such as cross-validation to evaluate the performance of machine learning models, we can ensure that the model performs consistently on training data and test data, thereby effectively reducing the risk of model overfitting and making the model more generalizable and adaptable to more unknown data scenarios.

[0169] By dynamically correcting the initial stress anomaly probability vector according to the changes in the real-time stress correlation factor data set, more real-time and accurate stress anomaly probability distribution data is obtained, which improves the system's ability to quickly respond to changes and adjust prediction results in the actual production process.

[0170] Dynamically adjusting the stress anomaly probability vector and combining it with real-time data changes can make timely adjustments based on the actual situation of the production line, thereby improving the flexibility of the prediction model and ensuring that the early warning of stress anomaly patterns can accurately reflect the current production status.

[0171] Automatically generating stress anomaly probability prediction results through machine learning models reduces the frequency and dependence of human intervention, reduces subjective errors caused by manual judgment, and improves the system's degree of automation and work efficiency.

[0172] By combining machine learning and real-time data analysis, it can provide more scientific and reliable data support for production decisions, enhance the company's decision-making ability in stress anomaly prediction and production adjustment, and improve the stability and efficiency of the overall production line.

[0173] Due to the real-time collection and dynamic correction of data, the prediction model can respond to the occurrence of stress anomalies in a timely manner, provide faster early warning and response time for the production line, and reduce the impact of potential stress anomalies in the production process on glass quality and production efficiency.

[0174] In one embodiment of the present invention, the S6 includes:

[0175] S61. Probability-ranking the candidate stress anomaly pattern list according to the stress anomaly probability distribution data; rank the stress anomaly probabilities from high to low to obtain a ranked stress anomaly pattern list;

[0176] S62, obtaining equipment layout data and glass transmission path data of the glass production line; performing stress abnormality area location analysis based on the stress abnormality pattern sorting list and the equipment layout data and glass transmission path data;

[0177] S63. Analyze the cause of the stress anomaly based on the stress anomaly area positioning result data and the stress anomaly pattern feature library; and generate a cold end stress detection report for the glass production line based on the analysis results.

[0178] The working principle of the above technical solution is to probabilistically sort a list of candidate stress anomaly patterns based on stress anomaly probability distribution data. The stress anomaly probability distribution data reflects the likelihood of different stress anomaly patterns occurring under the current stress-related factors. Sorting the stress anomaly probability from high to low can intuitively show which stress anomaly patterns are most likely to occur. The resulting ranked list of stress anomaly patterns arranges the candidate patterns by probability, providing a priority for subsequent stress anomaly area location and cause analysis, allowing operators to focus on stress anomaly patterns with higher probability.

[0179] Obtain equipment layout data and glass transport path data for the glass production line. Equipment layout data includes the location and parameters of cooling equipment, conveying equipment, and testing equipment, reflecting the production line's physical structure and equipment configuration. Glass transport path data includes information such as the trajectory and speed of glass along the production line, helping to understand the flow of glass during the production process. Based on the sorted list of stress anomaly patterns, along with the equipment layout and glass transport path data, perform stress anomaly location analysis. By analyzing the relationship between stress anomaly patterns, the equipment layout, and the glass transport path, the specific areas where stress anomalies may occur are identified. For example, certain stress anomaly patterns may be associated with the location of specific cooling equipment or conveying equipment, or with specific stages in the glass transport process. Spatial analysis and path tracing can be used for location analysis. Spatial analysis can visually demonstrate the spatial relationship between stress anomaly patterns and equipment locations, while path tracing can help identify areas of glass that may be affected by stress during transport. Identifying specific areas where stress anomalies may occur provides clear targets for subsequent cause analysis and treatment.

[0180] Based on the stress anomaly location data and the stress anomaly pattern signature database, a stress anomaly cause analysis is performed. The possible causes of stress anomalies are analyzed, taking into account factors such as equipment operating status, process parameter settings, and environmental conditions. For example, equipment failure can lead to uneven cooling, resulting in stress anomalies. Improper process parameter settings, such as excessively fast or slow cooling rates, can also affect the stress state of the glass. Environmental fluctuations, such as temperature and humidity fluctuations, can also affect the stress of the glass. By comprehensively analyzing these factors, the root cause of the stress anomaly can be identified. Based on the analysis results, a cold-end stress inspection report for the glass production line is generated. The report should include a ranked list of stress anomaly patterns, the location of the stress anomaly areas, an analysis of the causes of the stress anomaly, and corresponding treatment recommendations. The ranked list of stress anomaly patterns allows operators to understand the probability of occurrence of different stress anomaly patterns; the location of the stress anomaly areas clearly identifies the specific locations where the stress anomaly may occur; the analysis of the causes of the stress anomaly reveals the root cause; and the treatment recommendations provide operators with specific solutions and measures. The test report should be clear, accurate and comprehensive, providing a decision-making basis for the maintenance and management of the glass production line, helping operators to take timely measures to eliminate stress anomalies and ensure the quality and stability of glass production.

[0181] The effect of the above technical solution is: by probabilistically sorting the candidate stress anomaly patterns according to the stress anomaly probability distribution data, high-probability stress anomaly patterns can be accurately identified, thereby effectively improving the accuracy and pertinence of stress anomaly prediction.

[0182] By sorting the stress anomaly probabilities from high to low, a stress anomaly pattern sorting list is formed, which simplifies the screening process of abnormal patterns, reduces the complexity of manual judgment, and improves work efficiency.

[0183] By combining equipment layout data, glass transmission path data, and a sorted list of stress anomaly patterns, spatial analysis, path tracing, and other methods are used to locate and analyze stress anomaly areas, ensuring accurate identification and location of abnormal areas and providing more accurate data support for subsequent processing.

[0184] By locating and analyzing stress anomaly areas, we can quickly discover potential stress anomaly areas, conduct timely troubleshooting and production line adjustments, significantly improving the response speed of the production line and reducing downtime caused by equipment failures.

[0185] By combining factors such as equipment operating status, process parameter settings, and environmental conditions for cause analysis, the possible causes of stress anomalies can be determined in a short period of time, avoiding the time wasted in traditional manual diagnosis and improving analysis efficiency.

[0186] By combining multiple factors (such as equipment status, environmental parameters, process settings, etc.), a comprehensive analysis of the causes of stress anomalies is conducted, which enhances the comprehensiveness and scientific nature of the analysis results and avoids one-sided analysis of a single factor.

[0187] By generating a cold-end stress detection report that includes stress anomaly pattern sorting, stress anomaly area location results, and stress anomaly cause analysis, it provides a clear, accurate, and comprehensive decision-making basis for production line maintenance and management, thereby improving the reliability of decision-making.

[0188] By generating detailed stress detection reports, potential stress anomalies can be identified in advance and provide a reference for formulating preventive and treatment measures, reducing the risk of production line failure during operation.

[0189] By sorting stress anomaly patterns through machine learning algorithms and combining them with automated regional positioning analysis and cause analysis, the automation level of the entire system has been significantly enhanced, reducing reliance on manual operations and improving the intelligence level of the production line.

[0190] By comprehensively analyzing stress anomaly patterns, equipment layout, and transmission paths, the accuracy and responsiveness of the fault warning system are enhanced, ensuring that the production line can receive early warning information in a timely manner before an anomaly occurs, avoiding potential production interruptions.

[0191] In one embodiment of the present invention, the step S62 includes:

[0192] Analyze the potential association between each stress anomaly pattern and different devices. Based on the analysis results, construct a device-stress anomaly pattern association matrix. The rows of the matrix represent different stress anomaly patterns, the columns represent different devices and their parameters, and the matrix elements represent the strength of the association between the device parameters and the stress anomaly pattern.

[0193] Based on the glass transmission path data, the glass movement trajectory on the production line is divided into multiple critical path segments. For each path segment, the probability of a stress anomaly pattern occurring in that path segment is evaluated. A path segment-stress anomaly probability assessment model is established. Combining the stress anomaly pattern ranking list and the equipment-stress anomaly pattern association matrix, the probability of various stress anomaly patterns occurring in each path segment is calculated.

[0194] Use equipment layout data to build a spatial topology model of the glass production line, clarifying the spatial position relationship and connection relationship between each piece of equipment, as well as the flow direction of glass between the equipment; project the stress anomaly pattern into space and analyze its relationship with the spatial distribution of equipment;

[0195] A path tracing algorithm is used to simulate the movement of glass on the production line. Combined with a stress anomaly pattern feature library, the propagation of stress within the glass is simulated. Through simulation, the specific distribution of stress anomalies on the glass and the possible affected area are determined.

[0196] The results of the equipment and stress anomaly pattern correlation analysis, the glass transmission path segmentation and stress impact assessment results, the spatial topology relationship analysis results, and the path tracing and stress propagation simulation results are integrated. A weighted comprehensive evaluation method is used to assign corresponding weights to each analysis result based on its importance and reliability.

[0197] Through comprehensive calculations, the specific areas where stress anomalies may occur are determined and these areas are prioritized.

[0198] The working principle of this technical solution is to conduct an in-depth analysis of the potential correlations between each stress anomaly pattern and different equipment. For example, for stress anomalies caused by uneven cooling, the focus is on key parameters of cooling equipment (such as cooling air knives and cooling water troughs), including position, angle, and air / water speed adjustment range. Because these parameters directly affect cooling efficiency, improper settings can easily lead to uneven cooling and, in turn, stress anomalies. For stress anomalies caused by uneven forces during conveying, the correlation is based on parameters such as the roller spacing, roller material, and transmission speed stability of the conveying equipment. Changes in these parameters affect the forces applied to the glass during conveying, thus causing stress anomalies. Based on these analysis results, a device-stress anomaly pattern correlation matrix is ​​constructed. The rows of the matrix represent different stress anomaly patterns, and the columns represent different equipment and their parameters. The matrix elements indicate the strength of the correlation between the equipment parameters and the stress anomaly pattern. This matrix format intuitively displays the correlation between equipment parameters and stress anomaly patterns, providing basic data support for subsequent analysis and location.

[0199] Based on the glass transmission path data, the glass's trajectory on the production line is divided into multiple critical path segments. For example, starting from the moment the glass enters the cooling area, it is divided into cooling stage, transition stage, and inspection stage. Different path segments have different process characteristics and stress conditions, which have different impacts on the occurrence of stress anomalies. For each path segment, the probability of a stress anomaly pattern occurring in that path segment is evaluated. Factors such as the glass's residence time, speed changes, and stress conditions in that path segment are taken into account. For example, during the cooling stage, the glass moves slowly and is cooled. If the cooling is uneven, the probability of a stress anomaly occurring in this path segment is higher. During the inspection stage, if the inspection equipment applies inappropriate force to the glass, stress anomalies may also occur. Combining the ranked list of stress anomaly patterns and the equipment-stress anomaly pattern association matrix, a path segment-stress anomaly probability assessment model is established. The probability of various stress anomaly patterns occurring in each path segment is calculated, thereby determining the potential distribution of stress anomalies along the glass transmission path.

[0200] A spatial topology model of the glass production line is constructed using equipment layout data to clarify the spatial positional relationships and connectivity between the various pieces of equipment, as well as the direction of glass flow between them. This model can intuitively display the spatial layout of the production line and the glass flow path, providing a basis for analyzing the relationship between stress anomalies and the spatial distribution of equipment. Stress anomaly patterns are spatially projected to analyze their association with the spatial distribution of equipment. For example, if a stress anomaly pattern spatially exhibits a distribution pattern centered around a particular cooling device, it can be preliminarily determined that the device may have a problem causing stress anomalies. Spatial clustering algorithms (such as the DBSCAN algorithm) are used to cluster the spatial distribution of different stress anomaly patterns, identifying high-incidence areas or equipment clusters for stress anomalies and further narrowing down the possible areas where stress anomalies may occur.

[0201] A path-tracing algorithm is used to simulate the movement of glass on the production line, recording the equipment the glass passes through at different time points and the forces, temperatures, and other effects it is subjected to. By simulating the movement of glass, we can understand the stress conditions and temperature changes experienced by the glass throughout the production process, providing detailed data for analyzing the generation of stress anomalies. Combined with a library of stress anomaly pattern features, the propagation of stress within the glass is simulated. For example, for stress anomalies caused by temperature gradients, the propagation path and intensity changes of stress from high-temperature to low-temperature areas are simulated based on the glass's thermal conductivity characteristics and temperature changes. Through simulation, the specific distribution of stress anomalies on the glass and the range of areas potentially affected are determined, providing more precise information for locating areas of stress anomalies.

[0202] The results of the equipment-stress anomaly pattern correlation analysis, the glass transmission path segmentation and stress impact assessment results, the spatial topology relationship analysis results, and the path tracing and stress propagation simulation results are integrated. A weighted comprehensive evaluation method is used to assign weights to each analysis result based on its importance and reliability. For example, the equipment-stress anomaly pattern correlation analysis results may be highly reliable and thus be given a higher weight; whereas the path tracing and stress propagation simulation results, when data is incomplete, are relatively unreliable and thus are given a lower weight. Through comprehensive calculations, specific areas where stress anomalies are likely to occur are identified and prioritized. High-priority areas indicate a greater likelihood of stress anomalies and more severe impacts, requiring priority investigation and resolution. This prioritization helps operators allocate resources more effectively and improves the efficiency of troubleshooting and resolution.

[0203] The effect of the above technical solution is: by analyzing the potential association between each stress anomaly pattern and different devices and their parameters, and constructing a device-stress anomaly pattern association matrix, the key devices and parameters that cause stress anomalies can be identified more accurately, thereby improving the accuracy of abnormal pattern diagnosis.

[0204] Dividing the production line into path segments based on glass transmission path data and evaluating the probability of stress anomalies in each path segment can reduce prediction errors caused by incomplete path analysis and improve the reliability of anomaly prediction.

[0205] By constructing a spatial topological model of the glass production line and combining it with a spatial clustering algorithm to perform cluster analysis on the spatial distribution of different stress anomaly patterns, we can effectively identify high-incidence areas or equipment clusters of stress anomalies, thereby enhancing the comprehensiveness and accuracy of the production line spatial analysis.

[0206] By simulating the movement of glass on the production line through a path tracing algorithm and combining it with a stress anomaly pattern feature library, the propagation of stress on the glass can be simulated more accurately, thereby improving the accuracy of predicting the location and range of stress anomalies.

[0207] By weighting different analysis results through the weighted comprehensive evaluation method and assigning different weights according to the importance and reliability of the analysis results, the interference of data noise can be effectively reduced and the effectiveness and reliability of the comprehensive analysis can be improved.

[0208] By comprehensively considering the correlation analysis results between equipment and stress anomaly patterns, path segment stress anomaly probability assessment, spatial topology analysis results, and path tracing simulation results, it is possible to more accurately determine areas where stress anomalies may occur, and prioritize them to ensure that high-risk areas are promptly investigated and addressed.

[0209] By prioritizing specific areas where stress anomalies may occur, production line maintenance personnel can prioritize higher-risk areas, reduce the impact of failures on the production line, and improve the accuracy and timeliness of production line maintenance.

[0210] By comprehensively analyzing multiple factors and providing clear priority sorting of stress abnormality areas, it can reduce the decision-making costs of maintenance personnel when handling faults and improve the efficiency of fault detection and handling.

[0211] By combining multiple analysis methods (correlation analysis, path segment evaluation, spatial analysis, simulation analysis, etc.), the intelligence and predictive capabilities of the entire system are enhanced, human intervention is reduced, and the system's automation level is improved.

[0212] By conducting a detailed analysis of stress abnormality areas and combining it with priority sorting, the production line can identify potential failure areas in advance and perform preventive maintenance before failure occurs, reducing the frequency of production line failures and improving management reliability.

[0213] One embodiment of the present invention provides a cold-end stress detection system for a glass production line, comprising a memory, a processor, and a computer program stored in and executable on the memory, wherein the processor executes the program to implement any of the above-described cold-end stress detection methods for a glass production line.

[0214] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A method for detecting cold end stress of a glass production line, characterized in that: The method comprises: S1: Use a multi-spectral stress sensing array to collect stress-related spectral data from the cold end area of ​​the glass production line, and preprocess the collected spectral data to obtain a preprocessed spectral dataset; S2: Extract spectral features from the preprocessed spectral dataset and construct a spectral feature vector set. Based on the spectral feature vector set, use the spatiotemporal correlation analysis algorithm to extract the spatiotemporal characteristics of the cold end stress of the glass and obtain the stress spatiotemporal feature matrix. S3: Collect historical data on abnormal stress cases at the cold end of glass, perform feature clustering analysis on the stress spatiotemporal feature matrix, and obtain multiple stress feature clusters. Combined with the historical data on abnormal stress cases at the cold end of glass, perform abnormal feature matching on each stress feature cluster to screen out feature clusters with potential stress anomalies, and extract the stress anomaly feature vectors of the feature clusters. S4: Based on historical glass cold end stress anomaly case data, a stress anomaly pattern feature library is constructed. Based on the stress anomaly feature vector and the stress anomaly pattern feature library, a pattern matching algorithm is used to identify stress anomaly patterns and obtain a list of candidate stress anomaly patterns. Similarity calculation is performed on each pattern in the candidate stress anomaly pattern list to generate a stress anomaly similarity vector. S5: Collect data on stress-related factors at the cold end of the glass production line and construct a stress-related factor dataset. Based on historical data on abnormal stress cases at the cold end of the glass and the stress-related factor dataset, use a machine learning algorithm to establish a stress anomaly probability prediction model and obtain an initial stress anomaly probability vector. Dynamically modify the initial stress anomaly probability vector based on the stress-related factor dataset to obtain stress anomaly probability distribution data. S6: Probability sorting of the candidate stress anomaly pattern list is performed according to the stress anomaly probability distribution data to obtain a stress anomaly pattern sorting list; equipment layout data and glass transmission path data of the glass production line are obtained; based on the stress anomaly pattern sorting list, the equipment layout data and the glass transmission path data, stress anomaly area positioning analysis is performed to obtain stress anomaly area positioning result data; based on the stress anomaly area positioning result data and the stress anomaly pattern feature library, stress anomaly cause analysis is performed to generate a cold end stress detection report for the glass production line.

2. A method for detecting cold end stress of a glass production line according to claim 1, characterized in that: Said S1 comprises: S11. Evenly arrange multi-spectral stress sensing arrays in key areas of the cold end of the glass production line; calibrate the arranged multi-spectral stress sensing arrays; measure their spectral responses under different stress conditions and establish a calibration curve between spectral response and stress; S12. During the operation of the glass, the multispectral stress sensing array collects stress-related spectral data of the glass surface in real time; and transmits the collected spectral data to the data processing system in real time for preliminary storage and management; S13, performing denoising processing on the collected original spectral data; and performing normalization processing on the denoised original spectral data; and then performing smoothing processing on the normalized spectral data to obtain a preprocessed spectral data set.

3. A method for detecting cold end stress of a glass production line according to claim 1, characterized in that: Said S2 comprises: S21. Extracting stress-related spectral features from the preprocessed spectral data set; performing dimensionality reduction processing on the extracted spectral features using a feature selection algorithm; and constructing a spectral feature vector set; S22. Processing the spectral feature vector set by a spatiotemporal correlation analysis algorithm to extract the spatiotemporal characteristics of the glass cold end stress; S23. Organize and store the extracted spatiotemporal features according to a preset format to construct a stress spatiotemporal feature matrix, wherein the rows of the matrix represent different time points or spatial positions, the columns represent different spatiotemporal features, and the elements in the matrix represent the spatiotemporal feature values ​​corresponding to the time and spatial positions.

4. A method for detecting cold end stress of a glass production line according to claim 3, characterized in that: Said S23 comprises: According to the physical meaning and analysis requirements represented by spatiotemporal features, they are divided into time dimension features and space dimension features, and identifiers are defined for each type of feature; For spatiotemporal eigenvalues ​​of different dimensions and value ranges, a unified standardization method is used to process them and map the eigenvalues ​​to specific numerical intervals; Construct a time-space index system; in the time dimension, assign a unique index number to each time point in chronological order; in the spatial dimension, assign a corresponding index identifier to each spatial location based on the actual spatial layout of the glass cold end; Dynamically program the dimensions of the stress spatiotemporal characteristic matrix. In the time dimension, determine the number of rows based on the time span of data acquisition and the time resolution required for analysis. In the spatial dimension, determine the number of columns based on the degree of spatial division of the cold end of the glass. According to the time-space index system, the normalized time-space eigenvalues ​​are filled into the corresponding positions of the matrix one by one; in the case of missing data, estimation and filling are performed; after all eigenvalues ​​are filled, a complete stress time-space characteristic matrix is ​​formed; Perform quality verification on the constructed stress spatiotemporal characteristic matrix, and make corrections and optimizations based on the verification results; The stress spatiotemporal characteristic matrix that has undergone quality verification and optimization is stored in a designated database or file system, and the key information of the matrix is ​​recorded to form a complete metadata record.

5. The method for detecting cold end stress of a glass production line according to claim 1, characterized in that: Said S3 comprises: S31. Collect data on abnormal stress cases that occurred at the cold end of a glass production line over a period of time, establish a historical database of abnormal stress cases at the cold end of a glass production line, and organize and classify the collected historical case data; S32. Using clustering algorithm, perform characteristic clustering analysis on stress spatiotemporal characteristic matrix; S33. Based on historical glass cold end stress abnormality case data, perform abnormal feature matching on each stress feature cluster; compare the features of the cluster with the abnormal features in the historical cases, and calculate the similarity; S34, setting a similarity threshold to screen out characteristic clusters with potential stress anomalies; for clusters whose similarity exceeds the threshold, determining whether there is stress anomaly; S35. Extract stress anomaly feature vectors from the selected feature clusters; select the most representative features from the clusters to construct stress anomaly feature vectors.

6. A method for detecting cold end stress of a glass production line according to claim 1, characterized in that: Said S4 comprises: S41. Based on historical glass cold end stress anomaly case data, construct a stress anomaly pattern feature library; standardize the data in the feature library to unify the dimensions and ranges of different features; S42. Select a pattern matching algorithm and optimize the algorithm parameters according to the characteristics and distribution patterns of the stress anomaly features; use the stress anomaly feature vectors and corresponding stress anomaly patterns in the historical case data to train the algorithm and adjust the algorithm parameters; S43, using a selected pattern matching algorithm to perform stress anomaly pattern recognition based on the stress anomaly feature vector and the stress anomaly pattern feature library; matching the stress anomaly feature vector with each pattern in the feature library and calculating similarity; S44. Obtain a candidate stress anomaly pattern list based on the similarity calculation result; perform similarity calculation on each pattern in the candidate stress anomaly pattern list to generate a stress anomaly similarity vector.

7. A method for detecting cold end stress in a glass production line according to claim 1, characterized in that: Said S5 comprises: S51, collecting stress-related factor data at the cold end of the glass production line; and establishing a stress-related factor data acquisition system to collect and store the factor data in real time; S52. Based on historical glass cold end stress anomaly case data and stress correlation factor datasets, a stress anomaly probability prediction model is established using a machine learning algorithm; S53. Preprocess and feature engineer the data; integrate stress-related factor data and stress anomaly case data to extract features that have a significant impact on stress anomaly prediction; S54, dividing the processed data into a training set and a test set, using the training set to train the machine learning model, adjusting the model parameters, and using the test set to evaluate and verify the performance of the model; using the trained stress anomaly probability prediction model, according to the current stress correlation factor data set, to obtain an initial stress anomaly probability vector; S55. Dynamically modify the initial stress anomaly probability vector according to the real-time changes of the stress correlation factor data set to obtain stress anomaly probability distribution data.

8. The method for detecting cold end stress of a glass production line according to claim 1, characterized in that: Said S6 comprises: S61. Probability-ranking the candidate stress anomaly pattern list according to the stress anomaly probability distribution data; rank the stress anomaly probabilities from high to low to obtain a ranked stress anomaly pattern list; S62, obtaining equipment layout data and glass transmission path data of the glass production line; performing stress abnormality area location analysis based on the stress abnormality pattern sorting list and the equipment layout data and glass transmission path data; S63. Analyze the cause of the stress anomaly based on the stress anomaly area positioning result data and the stress anomaly pattern feature library; and generate a cold-end stress detection report for the glass production line based on the analysis results.

9. A method for detecting cold end stress in a glass production line according to claim 8, characterized in that: The S62 includes: Analyze the potential association between each stress anomaly pattern and different devices. Based on the analysis results, construct a device-stress anomaly pattern association matrix. The rows of the matrix represent different stress anomaly patterns, the columns represent different devices and their parameters, and the matrix elements represent the strength of the association between the device parameters and the stress anomaly pattern. Based on the glass transmission path data, the glass movement trajectory on the production line is divided into multiple critical path segments. For each path segment, the probability of a stress anomaly pattern occurring in that path segment is evaluated. A path segment-stress anomaly probability assessment model is established. Combining the stress anomaly pattern ranking list and the equipment-stress anomaly pattern association matrix, the probability of various stress anomaly patterns occurring in each path segment is calculated. Use equipment layout data to build a spatial topology model of the glass production line, clarifying the spatial position relationship and connection relationship between each piece of equipment, as well as the flow direction of glass between the equipment; project the stress anomaly pattern into space and analyze its relationship with the spatial distribution of equipment; A path tracing algorithm is used to simulate the movement of glass on the production line. Combined with a stress anomaly pattern feature library, the propagation of stress within the glass is simulated. Through simulation, the specific distribution of stress anomalies on the glass and the affected area are determined. The results of the equipment and stress anomaly pattern correlation analysis, the glass transmission path segmentation and stress impact assessment results, the spatial topology relationship analysis results, and the path tracing and stress propagation simulation results are integrated. A weighted comprehensive evaluation method is used to assign corresponding weights to each analysis result based on its importance and reliability. Through comprehensive calculations, the specific areas where stress anomalies occur are determined and these areas are prioritized.

10. A cold end stress detection system for a glass production line, characterized in that: The invention comprises a memory, a processor and a computer program stored in the memory and operable on the memory, wherein the processor executes the program to implement a cold end stress detection method for a glass production line as claimed in any one of claims 1 to 9.

Images