Production equipment fault identification and early warning method and system based on video data of side end equipment

Abstract

The invention provides a production equipment fault identification and early warning method and system based on edge equipment video data, and the method comprises the steps: constructing a historical data storage structure for a compressed feature vector set, designing an index table through employing the characteristics of high time sequence correlation and high spatial correlation, and obtaining an index record which can rapidly locate the matching of an abnormal scene; extracting video clips similar to the current abnormal features from historical data storage through index records in combination with a rapid recall mechanism to obtain a historical scene fragment set highly related to the current fault; according to real-time early warning feedback, updating a feature vector set and index records in historical data storage, optimizing the accuracy of abnormal scene matching, and obtaining updated storage structure data; and for the updated storage structure data, a rapid recall mechanism and a fault feature extraction process are circularly applied, the data processing efficiency of the edge equipment in a complex environment is continuously improved, and a long-term stable fault diagnosis basis is determined.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a method for identifying and warning production equipment faults based on edge device video data. Background Art

[0002] In the industrial production sector, equipment fault identification and early warning technology based on video data is of vital importance, directly related to improving production efficiency and reducing safety risks. With the advancement of intelligent manufacturing, using edge devices to collect and analyze video data to monitor equipment status has become a key development direction in the industry. However, many current solutions have significant shortcomings in practical application, mainly reflected in the inefficient use of historical data and limited ability to accurately extract fault characteristics in complex environments. This makes it difficult to meet the timeliness and accuracy of early warning requirements.

[0003] A deeper analysis of this area reveals several core challenges. First, video data exhibits significant temporal and spatial correlations, making it difficult to quickly locate historical scenes highly relevant to the current anomaly when processing and storing large amounts of data on edge devices. A further challenge arises from building an efficient indexing and recall mechanism within resource-constrained edge environments to ensure that when an anomaly is detected, video clips of similar failures can be quickly retrieved for comparison. Failure to implement this mechanism will further lead to a lack of sufficient reference for fault diagnosis, impacting the accuracy of judgments.

[0004] Therefore, how to build a lightweight feature storage and fast recall mechanism on edge devices based on the temporal and spatial characteristics of video data to achieve efficient matching of historical fault scenarios has become a key issue that needs to be urgently addressed in this research. Summary of the Invention

[0005] The present invention provides a production equipment fault identification and early warning method based on edge device video data, which mainly includes:

[0006] In view of the strong temporal and spatial correlation characteristics of video data, a pre-established feature extraction model is used to separate key frame image sequences from the original video stream, obtain visual feature information that is highly correlated with device state changes, and obtain a preliminary feature dataset;

[0007] Based on the preliminary feature dataset, a lightweight processing method is used to compress the features. This reduces storage usage in an edge resource-constrained environment while retaining core information with strong temporal correlation, and determines the compressed feature vector set.

[0008] For the compressed feature vector set, a historical data storage structure is constructed. The index table is designed by taking advantage of the strong temporal and spatial correlation characteristics to obtain index records that can quickly locate abnormal scene matches.

[0009] By indexing records and combining them with a fast recall mechanism, video clips similar to the current abnormal features are extracted from historical data storage to obtain a collection of historical scene clips that are highly relevant to the current fault.

[0010] Analyze the fault feature extraction rules contained in the historical scene fragment collection. If the similarity between the extracted features and the current abnormal features exceeds the preset threshold, it is judged as a potential fault mode and the preliminary fault diagnosis basis is obtained;

[0011] Based on the preliminary fault diagnosis basis, the real-time features of the current video data and the feature information in the historical scene clip collection are integrated, and the fault mode is further verified through comparative analysis to determine the final fault type classification result;

[0012] Based on the final fault type classification results and the requirement for high early warning accuracy, a corresponding early warning signal is generated. Leveraging the data processing efficiency of edge devices, the signal is transmitted to the monitoring system to obtain real-time early warning feedback.

[0013] Based on real-time warning feedback, update the feature vector set and index records in the historical data storage, optimize the accuracy of abnormal scene matching, and obtain updated storage structure data;

[0014] For updated storage structure data, the rapid recall mechanism and fault feature extraction process are cyclically applied to continuously improve the data processing efficiency of edge devices in complex environments and determine long-term and stable fault diagnosis basis.

[0015] The present invention provides a production equipment fault identification and early warning system based on edge device video data, which executes the production equipment fault identification and early warning method based on edge device video data.

[0016] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0017] The present invention discloses a method for edge device fault diagnosis based on video data. In view of the temporal and spatial correlation characteristics of video data, a key frame sequence is obtained through a feature extraction model, and lightweight processing is used to compress the feature data. A historical data storage structure and index table are constructed to quickly locate abnormal scenarios. A fast recall mechanism is used to extract similar historical fragments, analyze the fault feature patterns, and compare and verify with real-time features to determine the fault type. An early warning signal is generated based on the classification results, and the storage structure is updated to continuously optimize the accuracy of abnormal matching. The present invention improves data processing efficiency in complex environments through edge computing, realizes real-time fault diagnosis and early warning of video data, and provides a long-term and stable basis for equipment maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a flow chart of a production equipment fault identification and early warning method based on edge device video data of the present invention.

[0019] Figure 2 This is a schematic diagram of a production equipment fault identification and early warning method based on edge device video data of the present invention.

[0020] Figure 3 This is another schematic diagram of a production equipment fault identification and early warning method based on edge device video data of the present invention. DETAILED DESCRIPTION

[0021] To help those skilled in the art better understand the technical solutions in this specification, the following will provide a clear and complete description of the technical solutions in the embodiments of this specification, in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of this specification, not all of them. All other embodiments derived by those skilled in the art based on the embodiments in this specification without creative effort shall fall within the scope of protection of this specification.

[0022] like Figure 1-3 In this embodiment, a production equipment fault identification and early warning method based on edge device video data may specifically include:

[0023] In step S101, based on the strong temporal correlation and high spatial correlation characteristics of video data, a pre-established feature extraction model is used to separate a key frame image sequence from the original video stream, obtain visual feature information that is highly correlated with device state changes, and obtain a preliminary feature data set.

[0024] Using a pre-built feature extraction model, keyframes are extracted from the original video to extract visual features associated with device state changes, generating a preliminary feature set. Using image sequence processing techniques, temporal correlation analysis is performed on the preliminary feature set to determine the temporal continuity of the features and obtain a temporal feature distribution. Based on the temporal feature distribution and combined with spatial correlation characteristics, the keyframes are segmented into regions to identify visual feature regions with strong spatial correlation, generating a spatial feature map. If the spatial feature map contains regions inconsistent with device state changes, deep feature extraction is performed on these regions using a convolutional neural network model to obtain a revised spatial feature set. This revised spatial feature set is then integrated with the temporal feature distribution to construct a dynamic state change trend model, determining the temporal and spatial variations of device states and identifying state change trends. Based on these state change trends, the preliminary feature set is optimized and adjusted, and cluster analysis is used to group features and identify a subset of features highly correlated with device state changes. Further validation of the feature subset and integration with the keyframes in the original video allows the final device state change feature library to be constructed, resulting in a structured dataset for subsequent analysis.

[0025] In step S102 , based on the preliminary feature data set, a lightweight processing method is used to compress the features, thereby reducing storage usage in an edge resource-constrained environment while retaining core information with strong temporal correlation, and determining a compressed feature vector set.

[0026] An initial feature dataset is obtained. Principal component analysis (PCA) is used to extract the principal components of high-dimensional features, resulting in a reduced feature subset. Features with strong temporal correlation are extracted from the reduced feature subset. If the Pearson correlation coefficient between features exceeds a preset threshold, the feature is retained to generate a temporal feature set. Variable-length coding is used to compress the feature data in the temporal feature set, resulting in an encoded feature vector. The encoded feature vector is then recompressed using Huffman coding to generate a compressed feature vector set. If the storage usage of the compressed feature vector set is below the preset storage threshold of the edge device, the set is output. If it exceeds the threshold, the variable-length coding parameters are adjusted to regenerate a compressed feature vector set. The temporal correlation of the compressed feature vector set is verified. If the difference in temporal correlation coefficient with the initial feature dataset is below a preset threshold, the final compressed feature vector set is determined. The final compressed feature vector set is stored on the edge device, and storage usage information is obtained to determine the effectiveness of storage space optimization.

[0027] In step S103, a historical data storage structure is constructed for the compressed feature vector set, and an index table is designed using the characteristics of strong temporal correlation and high spatial correlation to obtain index records that can quickly locate abnormal scene matches.

[0028] For feature vectors and historical data, a storage structure is constructed to categorize and store the compressed feature vectors according to temporal and spatial correlation characteristics. A hierarchical storage approach is used to store data in blocks for different time periods and spatial regions, resulting in a preliminary feature dataset. Based on this preliminary feature dataset, an index table is constructed. Incorporating temporal and spatial correlation characteristics, the index table is divided into multiple subtables, each corresponding to feature data for a specific time period and spatial region. The mapping relationships within the index table are then determined. A rapid location mechanism is implemented within the mapping relationships within the index table. If feature data from an anomalous scenario matches the mapping range of a subtable, matching records from the corresponding subtable are extracted to identify potential anomalous matches. Based on the potential anomalous matches identified, the temporal distribution of the anomalous matches is analyzed, incorporating temporal characteristics, to determine whether the anomalous scenario exhibits persistence or periodicity, generating temporal analysis results. Based on the temporal analysis results, incorporating spatial characteristics, the spatial distribution of the anomalous matches is analyzed. If the spatial distribution is concentrated in a specific region, the spatial clustering characteristics of the anomalous scenario are determined, resulting in spatial analysis results. Based on the analysis results from both the temporal and spatial dimensions, a support vector machine algorithm is used to comprehensively classify abnormal scenarios. The classification results are compared with preset thresholds to determine the severity of the abnormal scenario and obtain a final abnormal classification label. Based on the final abnormal classification label, a corresponding scenario analysis record is generated and associated with an index table for storage, obtaining abnormal scenario matching data for subsequent queries.

[0029] In step S104, by indexing records and combining with a fast recall mechanism, video clips similar to the current abnormal features are extracted from the historical data storage to obtain a set of historical scene clips that are highly relevant to the current fault.

[0030] Using pre-established index records, video clips related to the current anomaly are retrieved from historical data stores to determine a preliminary set of candidate clips. A rapid recall mechanism is used to compare feature similarity within this preliminary set of candidate clips, identifying a set of historical scene clips highly relevant to the current fault. Based on this set of historical scene clips, the degree of match between the anomaly features and the current fault is extracted. If the match exceeds a preset threshold, the clip is marked as a key reference clip. The corresponding historical data records are retrieved from these key reference clips, and the fault-related information contained therein is determined, forming a structured fault feature dataset. Based on this structured fault feature dataset, a support vector machine algorithm is used to classify the correlation between the current fault and the historical scenes, generating a classified correlation result. Based on this classified correlation result, the historical scene clip most similar to the current fault is extracted. If the correlation result meets preset criteria, this clip is used as the final reference. Based on the final reference, the corresponding historical solution record is retrieved to determine the appropriate handling strategy for the current fault and form an executable fault handling plan.

[0031] In step S105, the fault feature extraction rules contained in the historical scene segment set are analyzed. If the similarity between the extracted features and the current abnormal features exceeds a preset threshold, it is determined to be a potential fault mode, and a preliminary fault diagnosis basis is obtained.

[0032] Relevant data is stored in a collection of historical scene fragments, and data cleaning methods are used to remove noise and irrelevant information to obtain a processed historical scene dataset. Feature extraction techniques are applied to this processed historical scene dataset to identify the distribution patterns of fault features and determine the key fault feature set. Based on the data records of the current anomaly, the corresponding anomaly features are extracted, a feature vector of the current anomaly is constructed, and a quantitative representation of the anomaly features is obtained. If the similarity between the feature vector of the current anomaly and the fault feature set exceeds a preset threshold, it is determined to be a potential fault, and a preliminary judgment of the potential fault is obtained. Based on this preliminary judgment of the potential fault, the most similar fault mode is matched with the fault mode records in the historical scenes to determine the specific fault mode category. Based on the determined fault mode category, the diagnostic evidence data in the historical scenes is correlated to obtain preliminary diagnostic evidence related to the current anomaly, forming a diagnostic information set. Based on the diagnostic information set, a support vector machine algorithm is used to verify the correlation between the current anomaly and historical fault modes to determine the final fault diagnosis result.

[0033] Step S106 , based on the preliminary fault diagnosis basis, the real-time features of the current video data and the feature information in the historical scene segment set are integrated, and the fault mode is further verified through comparative analysis to determine the final fault type classification result.

[0034] Real-time features are acquired from video data, and image processing techniques are used to extract key information to generate a preliminary feature dataset. Based on this preliminary feature dataset, corresponding segment features are retrieved from historical scenes and compared using feature matching methods to identify a set of historical segments with high similarity. For these sets of historical segments with high similarity, real-time features are fused with segment features, and feature integration is performed using a support vector machine algorithm to determine the potential fault mode category. If the potential fault mode category is inconsistent with the preliminary diagnosis result, further differential features are extracted through comparative analysis to obtain more detailed feature distribution information. Based on this more detailed feature distribution information and combined with fault type records from historical scenes, a logistic regression algorithm is used for secondary classification to determine the final fault type. The final fault type is combined with the judgment criteria to generate a classification result, and the classification result and related feature information are stored using data storage technology. If the confidence level of the classification result falls below a preset threshold, real-time features are re-extracted from the video data, and the feature integration and classification process is repeated to obtain a more accurate classification result.

[0035] In step S107, based on the final fault type classification result and the requirement for high warning accuracy, a corresponding warning signal is generated. Through the data processing efficiency advantage of the edge device, the signal is transmitted to the monitoring system to obtain real-time warning feedback.

[0036] Using the raw data collected from edge devices, preliminary feature extraction is performed on the fault type to obtain the initial basis for the classification results. Based on the classification results, a pre-established decision tree model is used to determine the fault type and obtain a specific fault category label. If the determined fault category label meets the preset severity threshold, the warning signal generation module is triggered to determine the corresponding warning signal content. The generated warning signal is compressed and encoded by the data processing module of the edge device to obtain an optimized signal data packet. The optimized signal data packet is quickly transmitted using the signal transmission protocol and passed to the monitoring system to determine whether the transmission is complete. If the transmission is complete, the signal data packet is decoded within the monitoring system to obtain real-time warning information, which is then compared with historical data to determine the final warning feedback. Based on the final warning feedback, the monitoring parameters of the edge device are automatically adjusted to obtain an updated monitoring strategy for subsequent data collection.

[0037] Step S108: Based on the real-time warning feedback, the feature vector set and index records in the historical data storage are updated to optimize the accuracy of abnormal scene matching and obtain updated storage structure data.

[0038] Abnormal event data is obtained from real-time early warning feedback. By parsing the feedback content, key feature parameters are extracted to obtain an abnormal feature set. If there is a difference between the abnormal feature set and the feature vector set in the historical data, the feature vector set is incrementally updated to obtain an updated feature vector set. Using the updated feature vector set, the abnormal scene is classified using the k-nearest neighbor algorithm to determine the category label of the abnormal scene. Based on the category label of the abnormal scene, relevant historical records are retrieved from the index record to obtain a matching index subset. If the similarity between the index subset and the feature vector set of the current abnormal scene is lower than a preset threshold, the index record is reconstructed to obtain an optimized index record. The storage structure data is reconstructed using the optimized index record and the updated feature vector set to obtain the optimized storage structure data. Based on the optimized storage structure data, the support vector machine algorithm is used to evaluate the accuracy of the abnormal scene matching to obtain an accuracy evaluation result.

[0039] In step S109, for the updated storage structure data, the rapid recall mechanism and fault feature extraction process are cyclically applied to continuously improve the data processing efficiency of edge devices in complex environments and determine the long-term stable basis for fault diagnosis.

[0040] By monitoring data updates in the storage structure in real time, the latest data flow changes are obtained, and whether the data flow meets the preset update frequency threshold is determined, a preliminary data update status assessment is obtained. Based on this preliminary data update status assessment, a rapid recall mechanism is used to screen key data in edge devices to determine the location and distribution characteristics of abnormal data points. Based on the location and distribution characteristics of the screened abnormal data points, a fault feature extraction process is applied to isolate potential fault modes from data processing in complex environments and obtain a set of fault features. The fault feature set is classified. If the proportion of a certain feature type in the classification result exceeds a preset threshold, a support vector machine algorithm is used to conduct in-depth analysis of this feature type to determine the specific fault type. Based on the fault type determination results and combined with long-term reliable diagnostic evidence, the data processing efficiency is dynamically adjusted. Whether the adjusted efficiency reaches the preset target value is determined, and the optimized processing strategy is determined. After obtaining the optimized processing strategy, the operating status of the edge device in the complex environment is monitored in real time. If the monitored data deviates from the preset normal range, the storage structure is locally updated to obtain an adjusted data storage solution. Through the adjusted data storage solution, the rapid recall mechanism and fault feature extraction process are cyclically applied to continuously optimize data processing and fault diagnosis, and determine the final diagnostic basis and processing efficiency improvement path.

[0041] The present invention provides a production equipment fault identification and early warning system based on edge equipment video data, which executes a production equipment fault identification and early warning method based on edge equipment video data.

[0042] The present invention discloses a method for edge device fault diagnosis based on video data. In view of the temporal and spatial correlation characteristics of video data, a key frame sequence is obtained through a feature extraction model, and lightweight processing is used to compress the feature data. A historical data storage structure and index table are constructed to quickly locate abnormal scenarios. A fast recall mechanism is used to extract similar historical fragments, analyze the fault feature patterns, and compare and verify with real-time features to determine the fault type. An early warning signal is generated based on the classification results, and the storage structure is updated to continuously optimize the accuracy of abnormal matching. The present invention improves data processing efficiency in complex environments through edge computing, realizes real-time fault diagnosis and early warning of video data, and provides a long-term and stable basis for equipment maintenance.

[0043] It should be noted that the above examples are merely specific embodiments of the present invention. Obviously, the present invention is not limited to the above examples and is subject to numerous variations. All variations that can be directly derived or conceived by a person skilled in the art from the disclosure of the present invention should be considered within the scope of protection of the present invention.

Claims

1. A production equipment fault identification and early warning method based on edge equipment video data, characterized in that: The method comprises: In view of the strong temporal and spatial correlation characteristics of video data, a pre-established feature extraction model is used to separate key frame image sequences from the original video stream, obtain visual feature information that is highly correlated with device state changes, and obtain a preliminary feature dataset; Based on the preliminary feature dataset, a lightweight processing method is used to compress the features. This reduces storage usage in an edge resource-constrained environment while retaining core information with strong temporal correlation, and determines the compressed feature vector set. For the compressed feature vector set, a historical data storage structure is constructed. The index table is designed by taking advantage of the strong temporal and spatial correlation characteristics to obtain index records that can quickly locate abnormal scene matches. By indexing records and combining them with a fast recall mechanism, video clips similar to the current abnormal features are extracted from historical data storage to obtain a collection of historical scene clips that are highly relevant to the current fault. Analyze the fault feature extraction rules contained in the historical scene fragment collection. If the similarity between the extracted features and the current abnormal features exceeds the preset threshold, it is judged as a potential fault mode and the preliminary fault diagnosis basis is obtained; Based on the preliminary fault diagnosis basis, the real-time features of the current video data and the feature information in the historical scene clip collection are integrated, and the fault mode is further verified through comparative analysis to determine the final fault type classification result; Based on the final fault type classification results and the requirement for high early warning accuracy, a corresponding early warning signal is generated. Leveraging the data processing efficiency of edge devices, the signal is transmitted to the monitoring system to obtain real-time early warning feedback. Based on real-time warning feedback, update the feature vector set and index records in the historical data storage, optimize the accuracy of abnormal scene matching, and obtain updated storage structure data; For updated storage structure data, the rapid recall mechanism and fault feature extraction process are cyclically applied to continuously improve the data processing efficiency of edge devices in complex environments and determine long-term and stable fault diagnosis basis.

2. A production equipment fault identification and early warning method based on edge equipment video data according to claim 1, characterized in that: In view of the strong temporal correlation and high spatial correlation characteristics of video data, a pre-established feature extraction model is used to separate key frame image sequences from the original video stream, obtain visual feature information that is highly correlated with device state changes, and obtain a preliminary feature dataset, including: Using a pre-built feature extraction model, we extract keyframes from the original video, obtain visual features associated with device state changes, and generate a preliminary feature set. Using image sequence processing technology, we conduct time series correlation analysis on the preliminary feature set to determine the continuity of the features in the time dimension and obtain the time series feature distribution; According to the temporal feature distribution and the spatial correlation characteristics, the key frame image is divided into regions, the visual feature regions with strong spatial correlation are determined, and the spatial feature map is obtained; If there is an area in the spatial feature map that is inconsistent with the device state change, the convolutional neural network model is used to extract deep features of the area to obtain a corrected spatial feature set; Based on the corrected spatial feature set, the temporal feature distribution is integrated to build a dynamic trend model of state changes, determine the changing patterns of device states in time and space, and obtain the state change trend. Based on the state change trend, the preliminary feature set is optimized and adjusted, and the features are grouped using cluster analysis to determine the feature subset that is highly correlated with the device state change; By further verifying the feature subset and combining it with the key frame images in the original video, the final device state change feature library is constructed to obtain a structured dataset for subsequent analysis.

3. The method for identifying and warning production equipment faults based on edge equipment video data according to claim 1, characterized in that: Based on the preliminary feature dataset, a lightweight processing method is used to compress the features, reducing storage usage in an edge resource-constrained environment while retaining core information with strong temporal correlation, and determining a compressed feature vector set, including: Obtain the initial feature data set, extract the principal components of high-dimensional features through the principal component analysis algorithm, and obtain the feature subset after dimensionality reduction; Extract features with strong temporal correlation from the feature subset after dimensionality reduction. If the Pearson correlation coefficient between features exceeds the preset threshold, the feature is retained to generate a temporal feature set. For the time series feature set, the variable length coding method is used to compress the feature data to obtain the encoded feature vector; Perform secondary compression on the encoded feature vector through Huffman coding to generate a set of compressed feature vectors; If the storage occupancy of the compressed feature vector set is lower than the preset storage threshold of the edge device, the set is output; If it exceeds the threshold, the variable-length coding parameters are adjusted and the compressed feature vector set is regenerated; Based on the compressed feature vector set, its temporal correlation is verified. If the difference in temporal correlation coefficient with the initial feature data set is lower than a preset threshold, the final compressed feature vector set is determined; The final compressed feature vector set is stored on the edge device to obtain storage occupancy information and determine the storage space optimization effect.

4. The method for identifying and warning production equipment failures based on edge equipment video data according to claim 1, characterized in that: The method constructs a historical data storage structure for the compressed feature vector set, designs an index table using the characteristics of strong temporal correlation and high spatial correlation, and obtains index records that can quickly locate abnormal scene matches, including: For feature vectors and historical data, a storage structure is constructed to classify and store the compressed feature vectors according to their temporal and spatial correlation characteristics. A hierarchical storage method is used to save data in different time periods and spatial regions in blocks, thus obtaining a preliminary organized feature data set. Based on the preliminarily collated feature data set, an index table is constructed. Combining the temporal and spatial correlation characteristics, the index table is divided into multiple sub-tables. Each sub-table corresponds to the feature data of a specific time period and spatial region, and the mapping relationship of the index table is determined. A fast positioning mechanism is used for the mapping relationship of the index table. If the characteristic data of the abnormal scene is detected to match the mapping range of a sub-table, the matching records in the corresponding sub-table are extracted to obtain potential abnormal matching items; Based on the potential abnormal matching items obtained, combined with the characteristics related to time series, by analyzing the distribution pattern of abnormal matching items in the time dimension, it is determined whether the abnormal scenario has persistent or periodic characteristics, and the analysis results of the time dimension are obtained; Based on the analysis results of the time dimension and combined with the spatial-related characteristics, the distribution pattern of abnormal matching items in the spatial dimension is analyzed. If the spatial distribution is concentrated in a specific area, the spatial aggregation characteristics of the abnormal scene are determined to obtain the analysis results of the spatial dimension; Based on the analysis results of the time and space dimensions, the support vector machine algorithm is used to comprehensively classify abnormal scenarios. The classification results are compared with the preset threshold to determine the severity of the abnormal scenario and obtain the final abnormal classification label; Based on the final anomaly classification label, the corresponding scene analysis record is generated, and the analysis record is associated with the index table for storage to obtain the anomaly scene matching data for subsequent queries.

5. The method for identifying and warning production equipment faults based on edge equipment video data according to claim 1, characterized in that: The index record is combined with a fast recall mechanism to extract video clips similar to the current abnormal features from the historical data storage to obtain a set of historical scene clips that are highly relevant to the current fault, including: Through pre-established index records, video clips related to the current abnormal features are obtained from historical data storage to determine a preliminary set of candidate clips; Using a fast recall mechanism, we compare the feature similarity of the initial candidate segments to obtain a set of historical scene segments that are highly relevant to the current fault. Based on the historical scene segment set, the abnormal features are extracted to determine the matching degree with the current fault. If the matching degree is higher than the preset threshold, the segment is marked as a key reference segment. By focusing on the reference fragments, we can obtain the corresponding historical data records, determine the fault-related information contained therein, and form a structured fault feature dataset; Based on the structured fault feature data set, the support vector machine algorithm is used to classify the correlation between the current fault and the historical scenario, and the classified correlation results are obtained; Based on the classified correlation results, the historical scene segment closest to the current fault is extracted. If the correlation result meets the preset conditions, the segment is used as the final reference. Based on the final reference, obtain the corresponding historical solution records, determine the processing strategy applicable to the current fault, and form an executable fault handling plan.

6. The method for identifying and warning production equipment faults based on edge equipment video data according to claim 1, characterized in that: The above process analyzes the fault feature extraction rules contained in the historical scene segment set. If the similarity between the extracted features and the current abnormal features exceeds a preset threshold, it is determined to be a potential fault mode, and preliminary fault diagnosis basis is obtained, including: From the collection of historical scene fragments, we obtain the stored relevant data, use data cleaning methods to remove noise and irrelevant information, and obtain the processed historical scene dataset; For the processed historical scene data set, feature extraction technology is applied to identify the distribution pattern of fault characteristics and determine the key fault feature set; According to the current abnormal data record, the corresponding abnormal features are extracted, the feature vector of the current abnormality is constructed, and the quantitative representation of the abnormal features is obtained; If the similarity between the feature vector of the current anomaly and the fault feature set exceeds the preset threshold, it is determined to be a potential fault and a preliminary judgment result of the potential fault is obtained; Based on the preliminary judgment results of potential faults, combined with the fault mode records in historical scenarios, the most similar fault mode is matched to determine the specific fault mode category; By determining the fault mode category and associating the diagnostic evidence data in historical scenarios, we can obtain preliminary diagnostic evidence related to the current anomaly and form a diagnostic information set. According to the diagnostic information set, the support vector machine algorithm is used to verify the correlation between the current anomaly and the historical fault mode to determine the final fault diagnosis result.

7. The method for identifying and warning production equipment faults based on edge equipment video data according to claim 1, characterized in that: The preliminary fault diagnosis is based on the integration of real-time features of the current video data and feature information in the historical scene segment set, and the fault mode is further verified through comparative analysis to determine the final fault type classification result, including: Obtain real-time features through video data, use image processing technology to extract key information, and obtain a preliminary feature data set; Based on the preliminary feature dataset, corresponding segment features are retrieved from historical scenes, and feature matching methods are used for comparison to determine a set of historical segments with high similarity. For a set of historical segments with high similarity, real-time features and segment features are integrated, and the support vector machine algorithm is used for feature integration to determine the potential failure mode category; If the potential failure mode category is inconsistent with the preliminary diagnosis result, further differential features are extracted through comparative analysis to obtain more detailed feature distribution information; Based on more detailed feature distribution information and combined with the fault type records in historical scenarios, a logistic regression algorithm is used for secondary classification to determine the final fault type; The final fault type is combined with the judgment basis to generate the classification result, and the classification result and related feature information are saved using data storage technology; If the confidence level of the classification result is lower than the preset threshold, the real-time features are re-extracted from the video data, and the feature integration and classification process is repeated to obtain a more accurate classification result.

8. The method for identifying and warning production equipment faults based on edge equipment video data according to claim 1, characterized in that: Based on the final fault type classification results and the requirement for high early warning accuracy, a corresponding early warning signal is generated. Leveraging the data processing efficiency of edge devices, the signal is transmitted to the monitoring system to obtain real-time early warning feedback, including: Using the raw data collected from edge devices, preliminary feature extraction is performed on the fault type to obtain the initial basis for classification results. According to the classification results, the pre-established decision tree model is used to determine the fault type and obtain the specific fault category label; If the determined fault category label meets the preset severity threshold, the warning signal generation module is triggered to determine the corresponding warning signal content; The generated warning signal is compressed and encoded through the data processing module of the edge device to obtain the optimized signal data packet; The optimized signal data packets are quickly transmitted using the signal transmission protocol and delivered to the monitoring system to determine whether the transmission is complete; If the transmission is complete, the signal data packet is decoded in the monitoring system to obtain real-time warning information, which is then compared with historical data to determine the final warning feedback; Based on the final warning feedback, the monitoring parameters of the edge equipment are automatically adjusted to obtain an updated monitoring strategy, which is applied to subsequent data collection.

9. The method for identifying and warning production equipment failures based on edge equipment video data according to claim 1, characterized in that: The method of updating the feature vector set and index records in the historical data storage based on the real-time warning feedback, optimizing the accuracy of abnormal scene matching, and obtaining updated storage structure data includes: Obtain abnormal event data from real-time warning feedback, analyze the feedback content, extract key feature parameters, and obtain an abnormal feature set; If there is a difference between the abnormal feature set and the feature vector set in the historical data, the feature vector set is incrementally updated to obtain an updated feature vector set; Using the updated feature vector set, the abnormal scene is classified through the k-nearest neighbor algorithm to determine the category label of the abnormal scene; According to the category label of the abnormal scenario, relevant historical records are retrieved from the index records to obtain the matching index subset; If the similarity between the index subset and the feature vector set of the current abnormal scene is lower than a preset threshold, the index record is reconstructed to obtain an optimized index record; Reconstruct the storage structure data through the optimized index record and the updated feature vector set to obtain the optimized storage structure data; According to the optimized storage structure data, the support vector machine algorithm is used to evaluate the accuracy of abnormal scene matching and obtain the accuracy evaluation result.

10. A production equipment fault identification and early warning system based on edge equipment video data, characterized in that: Execute the production equipment fault identification and early warning method based on edge device video data as described in any one of claims 1-9.

Images