Industrial equipment real-time monitoring system based on edge computing

Abstract

The invention discloses an industrial equipment real-time monitoring system based on edge computing, and the system comprises the following steps: collecting, processing and analyzing the operation data of industrial equipment through distributed edge computing nodes, constructing a node cooperation mechanism, and completing the processing of the data; the dynamic data scheduling module is used for dynamically optimizing a data transmission path and task allocation according to node performance; the intelligent model dynamic migration module is used for adjusting lightweight model deployment in combination with node resource conditions; the multi-modal fault sensing module is used for extracting and analyzing multi-modal features by using a cross-node collaborative algorithm; the edge node self-adaptive management module is used for adjusting node operation parameters in real time and repairing an abnormal state; and the equipment relevance reasoning module is used for constructing an equipment relevance model through a graph calculation method and generating a real-time relevance graph. According to the method, efficient monitoring and anomaly detection of industrial equipment are realized through resource optimization and intelligent analysis of collaborative edge computing nodes.

Description

Technical Field

[0001] The present invention relates to the field of industrial equipment monitoring technology, and in particular to an industrial equipment real-time monitoring system based on edge computing. Background Art

[0002] With the rapid development of industrial automation and intelligent manufacturing, real-time monitoring and fault detection of industrial equipment have become an important part of ensuring production efficiency and equipment safety. However, traditional industrial monitoring systems usually rely on centralized cloud computing architecture, which has obvious performance bottlenecks when facing massive equipment data. Especially in industrial environments with limited network bandwidth and high data transmission latency, traditional systems find it difficult to achieve efficient real-time data processing and analysis.

[0003] In the existing technology, although the monitoring method based on cloud computing has powerful centralized processing capabilities, it has significant deficiencies in the following aspects: 1. Delayed response problem: Traditional monitoring systems need to upload device data to the cloud for processing. Therefore, data transmission delays directly affect the system's response speed and cannot meet the industrial field's high real-time requirements.

[0004] 2. Bandwidth limitation: The amount of data generated by industrial equipment is huge, but the network bandwidth is limited. Congestion is prone to occur during data upload, resulting in low transmission efficiency of key data.

[0005] 3. Waste of computing resources: Existing methods cannot dynamically adjust the allocation of computing tasks, node resource utilization is low, and the coordination ability between devices is insufficient, resulting in waste of computing and storage resources.

[0006] 4. Insufficient fault detection: Traditional systems lack the ability to collaboratively perceive multimodal data and are unable to efficiently extract the multi-dimensional characteristics of equipment operation, thus affecting the detection accuracy of complex fault modes.

[0007] 5. Lack of robustness: Centralized monitoring systems are sensitive to single point failures, and the system's reliability and continuous operation capabilities are weak. They are easily affected by equipment or network anomalies in industrial environments.

[0008] Therefore, how to provide a real-time monitoring system for industrial equipment based on edge computing is an issue that technical personnel in this field urgently need to solve. Summary of the invention

[0009] One purpose of the present invention is to propose a real-time monitoring system for industrial equipment based on edge computing. The present invention describes in detail a technical solution for efficiently realizing real-time monitoring and fault detection of industrial equipment through distributed edge computing node deployment, multimodal data collection and analysis, dynamic migration of intelligent models and correlation reasoning between devices. The system has the advantages of strong real-time performance, accurate fault detection, high resource utilization and strong system robustness.

[0010] The real-time monitoring system for industrial equipment based on edge computing according to an embodiment of the present invention includes the following steps: S1, distributed edge computing node deployment module, deploys multiple distributed edge computing nodes, builds a node coordination mechanism, performs distributed collection and local processing of industrial equipment operation data, and generates node performance status information; S2, dynamic data flow scheduling module, based on node performance status information, uses micro-delay optimization scheduling algorithm to dynamically adjust data priority and transmission path, and allocates data tasks to distributed edge computing nodes with matching performance; S3, intelligent model dynamic migration module, builds lightweight intelligent models according to the data tasks assigned by the dynamic data flow scheduling module, and dynamically adjusts the deployment of lightweight intelligent models based on the computing performance and load of distributed edge computing nodes; S4, multimodal fault perception module, collects multimodal data of industrial equipment through distributed edge computing nodes and collaboratively extracts fault features; S5, the distributed edge computing node adaptive management module, based on the extracted fault features, monitors the computing load and operation anomalies of the distributed edge computing nodes in real time, and uses an autonomous repair mechanism to adjust the distributed edge computing node parameters and repair the anomalies; S6, equipment correlation reasoning module, based on the extracted fault characteristics and node performance status information, uses graph computing technology to build a multi-dimensional equipment correlation model and generate a dynamic correlation graph.

[0011] Optionally, the S2 specifically includes: S21, edge node performance status collection unit, used to collect node performance status information of distributed edge computing nodes in real time, including computing power , Storage Capacity , Communication bandwidth and calculate load ,in Represents the distributed edge computing node number and generates the node performance state vector ; S22, data task priority evaluation unit, obtains data task characteristics, including task urgency , computing requirements and data size , construct the task feature vector , according to the node performance state vector and the task feature vector Calculate each data task At the node Priority value : ; in, , , is the dynamically adjusted weight factor, To calculate the load, For computing power, For communication bandwidth, the priority values ​​are organized into a data task priority matrix ; S23, the task path dynamic optimization unit is connected to the edge node performance status collection unit and the data task priority evaluation unit to collect network connection information between distributed edge computing nodes and combine the data task priority matrix , based on the micro-delay optimization scheduling algorithm, the optimal transmission path for data tasks is selected, and the path optimization goal is to minimize the data task In distributed edge computing nodes Transmission delay and execution time: ; in, For the task At the node The total time cost on For Node and The communication bandwidth between For Node and nodes The fixed transmission delay between the two paths is used, and the optimization result outputs the optimal path set. ,in, Indicates the task The optimal path of S24, task dynamic allocation unit, based on the optimal path set and data task priority matrix , assign data tasks to target distributed edge computing nodes with matching performance, and generate a task allocation table ,in Indicates the task number. Indicates the target path corresponding to the data task and monitors the execution status of the data task.

[0012] Optionally, the S3 specifically includes: S31, lightweight model building unit, connected with the task dynamic allocation unit, according to the task allocation table The data task allocation result in the data task allocation result is combined with the performance state vector of the allocated node and the task feature vector , to allocate distributed edge computing nodes Building a lightweight intelligent model , satisfying the following constraints: ; in, represents the computational complexity of the lightweight model, Indicates the remaining computing capacity of the distributed edge computing nodes, For Node The communication bandwidth For Node The computing power of For Node storage capacity, For Node The computational load; S32, the hierarchical model optimization unit, is connected with the lightweight model construction unit to form a lightweight intelligent model Decompose into base layer model and extension layer models , where the base layer model completes basic feature extraction, and the extended layer model handles deep learning calculations: ; in, is the ratio factor of the basic layer model to the computing power of distributed edge computing nodes, Distributed edge computing nodes The computing power is dynamically adjusted according to the computing load and task characteristics of the distributed edge computing nodes; S33, model dynamic migration unit, connected with task dynamic allocation unit and hierarchical model optimization unit, used for real-time monitoring of data task allocation distributed edge computing nodes When the performance status of the distributed edge computing nodes changes and causes the time cost and migration cost of the lightweight intelligent model to deviate from the threshold, the migration cost of the lightweight intelligent model is calculated. and time cost , the following migration conditions are met: ; in, For Node and The communication bandwidth between them, the migration path is based on the data task priority matrix Select and update the task assignment table after migration ; S34, the model adaptive adjustment unit is connected to the model dynamic migration unit, and according to the feedback information of the migrated data task execution, the parameters of the lightweight intelligent model are adaptively adjusted, including the model compression rate, feature selection strategy and calculation accuracy. After the adaptive adjustment, the lightweight intelligent model configuration is updated and synchronized to the task allocation table. .

[0013] Optionally, the S33 specifically includes: S331, node performance status monitoring unit, connected to the task dynamic allocation unit, combined with the node performance status vector and the task feature vector , real-time monitoring of distributed edge computing nodes The performance status changes are detected, and the monitoring results are transmitted to the migration condition judgment unit; S332, migration condition judgment unit, according to the node performance state vector and the task feature vector , determine the data task allocation distributed edge computing nodes Whether the task requirements are met. When one of the following conditions is not met, the migration operation is triggered: ; The migration condition judgment unit calculates the model migration time cost and migration cost : ; in, Assigning nodes to tasks The complexity of the above model, Distributed edge computing nodes and The communication bandwidth between is the unit data migration cost coefficient. When the following migration conditions are met, migration is started: ; S333, optimal migration path selection unit, combined with data task priority matrix , select the optimal path that meets the migration conditions, and the path selection goal is to minimize the total migration cost : ; in, For Node and nodes The fixed transmission delay between the two outputs the optimal migration path set , Representation Node To Node The set of edges in all migration paths; S334, the task allocation table updating unit is connected to the optimal migration path selection unit, based on the optimal migration path set , migrate the lightweight intelligent model to the target distributed edge computing node , update the task allocation table The distributed edge computing node binding information corresponding to the data task.

[0014] Optionally, the S4 specifically includes: S41, a multimodal data acquisition unit, which collects multimodal data of equipment operation through a multimodal sensor network deployed on the industrial equipment, wherein the sensor network includes a vibration sensor, a temperature sensor, an acoustic sensor and a current sensor, which respectively collect vibration signals , temperature signal , acoustic signal and electrical signals ,in For the sampling time point, the timestamps of the collected multimodal data are aligned to generate a time-series multimodal data set: ; in, is the total number of time points in the sampling time window; S42, a data preprocessing unit is connected to the multimodal data acquisition unit to collect the time-series multimodal data. Preprocessing is performed, and the noise in the data is suppressed by filtering algorithm to generate a denoised data set, and the denoised data set is standardized according to the amplitude to output a standardized multimodal data set : ; in, For time point The standard value of the vibration signal, For time point The standard value of the temperature signal, For time point The standard value of the acoustic signal, For time point Standard value of electrical signal; S43, a multimodal feature extraction unit, connected to the data preprocessing unit, for standardizing the multimodal data set Perform feature extraction and use a cross-node collaborative feature extraction algorithm to generate a multimodal feature vector set from the time domain, frequency domain, and spatial dimensions : ; in, Time domain features, , ; S44, the feature fusion and fault analysis unit is connected to the multimodal feature extraction unit to collect the multimodal feature vectors Input feature fusion unit, using distributed weighted fusion algorithm to generate comprehensive fault feature vector : ; in, , , are the dynamic weighting coefficients of time domain, frequency domain and spatial features respectively; Fault analysis unit combined with comprehensive fault feature vector and a standardized multimodal data set, where the standardized multimodal data set is adjusted by a collaborative optimization algorithm to generate real-time fault perception results: ; in, For real-time fault perception results, Indicates The detection probability of class faults, , is the total number of fault types.

[0015] Optionally, the S5 specifically includes: S51, node status monitoring unit, collects node operation status data in real time through distributed edge computing nodes, including computing load , memory usage , Communication bandwidth , Power consumption and node temperature ,in Indicates the node number and generates the node running state vector ; S52, the state abnormality detection unit receives the node operation state vector from the node state monitoring unit , define the node operation status threshold, and perform anomaly detection on each operation parameter. If the operation parameter does not meet the following conditions, it will be marked as abnormal: ; in, and Respectively represent the minimum and maximum values ​​of the operating parameters, and organize the detected abnormal parameters into an abnormal parameter set ; S53, node parameter adjustment unit, for the abnormal parameter set transmitted by the state abnormality detection unit , adjust the operating parameters of abnormal distributed edge computing nodes, when computing load In case of an abnormality, the computing load corresponding to the distributed edge computing nodes is dynamically adjusted, and the optimized computing load distribution is: ; in, Distributed edge computing nodes Transfer to other distributed edge computing nodes The computational load, Distributed edge computing nodes after adjustment The computational load, Distributed edge computing nodes after adjustment The computational load; When the communication bandwidth Dynamically adjust the communication bandwidth when an abnormality occurs: ; in, is the bandwidth adjustment factor, Adjusted distributed edge computing nodes Communication bandwidth; When the power consumption and node temperature When abnormal, optimize by adjusting power consumption mode and activating heat dissipation mechanism: ; in, Distributed edge computing nodes after adjustment The node temperature, Distributed edge computing nodes The upper temperature threshold, Distributed edge computing nodes after adjustment Power consumption, Distributed edge computing nodes The upper power consumption threshold; S54, autonomous repair unit, combined with node parameter adjustment unit and comprehensive fault feature vector , autonomously repair the running status of the distributed edge computing nodes, update the node running status vector, remove the abnormal parameters from the original state vector, and make the adjusted running parameters meet the following conditions: ; in, Run the state vector for the adjusted node.

[0016] Optionally, the S6 specifically includes: S61, associated data collection unit, combined with node performance state vector , comprehensive fault feature vector , and combined with the task allocation information and communication bandwidth utilization of distributed edge computing nodes, generate the interactive information matrix between devices , generate the initial set of related data : ; in, Represents a collection of device nodes. Indicates the device and Interaction information between S62, association graph construction unit, based on the initial association data set , build a device association graph ,in Represents a set of distributed edge computing nodes, represents the edge set, is the edge weight matrix, edge weight Calculated by the following formula: ; in, and are weight factors, representing the influence weights of mutual information and performance similarity, respectively. For Node and Euclidean distance of performance state vector.

[0017] S63, a graph computing processing unit, using graph computing technology to process the device association graph Perform calculations to extract multi-dimensional correlation features between devices, and optimize the correlation between devices based on distributed edge computing nodes and edge weights : ; in, Represents the node performance state vector and The correlation function between and equipment the mutual relationship between S64, a dynamic association graph generation unit dynamically updates the device association graph according to the analysis result of the graph calculation processing unit , and generate a device association graph ,in represents the dynamically adjusted edge set, represents the updated edge weight matrix.

[0018] The beneficial effects of the present invention are: (1) The present invention combines distributed edge computing node deployment and dynamic data flow scheduling technology to achieve efficient distributed processing of industrial equipment operation data and dynamic optimization of transmission paths, thereby effectively reducing data transmission delays and bandwidth occupancy and improving the response speed and system efficiency of real-time monitoring.

[0019] (2) The present invention uses an intelligent model dynamic migration module to dynamically adjust the computing task allocation of the lightweight model based on the node performance status and task priority matrix, significantly improving the resource utilization of edge computing nodes and ensuring the adaptability and stability of the system in heterogeneous industrial environments.

[0020] (3) The present invention uses a multimodal fault perception module and a collaborative feature extraction algorithm to achieve deep fusion of multimodal data and real-time fault detection, especially accurate identification of complex fault modes, thereby improving the accuracy of industrial equipment fault perception.

[0021] (4) The present invention uses a device association reasoning module to build a multi-dimensional association model between devices based on graph computing technology and dynamically generate a device association graph, thereby providing a comprehensive analysis of device interaction relationships and supporting the rapid identification and optimization of key devices and weak links.

[0022] (5) The present invention realizes dynamic adjustment of operating parameters and abnormal repair by combining fault characteristics and node operating status through the edge node adaptive management module, which significantly improves the robustness and long-term stable operation capability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings: Figure 1 This is the overall framework diagram of the real-time monitoring system for industrial equipment based on edge computing proposed by the present invention; Figure 2 This is a flow chart for implementing the intelligent model dynamic migration module proposed in the present invention; Figure 3 This is a flow chart of the operating status monitoring and autonomous repair of the edge node adaptive management module proposed by the present invention. DETAILED DESCRIPTION

[0024] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0025] refer to Figure 1-Figure 3 ,The real-time monitoring system of industrial equipment based on edge computing includes the following steps: S1, distributed edge computing node deployment module, deploys multiple distributed edge computing nodes, builds a node coordination mechanism, performs distributed collection and local processing of industrial equipment operation data, and generates node performance status information; S2, dynamic data flow scheduling module, based on node performance status information, uses micro-delay optimization scheduling algorithm to dynamically adjust data priority and transmission path, and allocates data tasks to distributed edge computing nodes with matching performance; S3, intelligent model dynamic migration module, builds lightweight intelligent models according to the data tasks assigned by the dynamic data flow scheduling module, and dynamically adjusts the deployment of lightweight intelligent models based on the computing performance and load of distributed edge computing nodes; S4, multimodal fault perception module, collects multimodal data of industrial equipment through distributed edge computing nodes and collaboratively extracts fault features; S5, the distributed edge computing node adaptive management module, based on the extracted fault features, monitors the computing load and operation anomalies of the distributed edge computing nodes in real time, and uses an autonomous repair mechanism to adjust the distributed edge computing node parameters and repair the anomalies; S6, equipment correlation reasoning module, based on the extracted fault characteristics and node performance status information, uses graph computing technology to build a multi-dimensional equipment correlation model and generate a dynamic correlation graph.

[0026] In this implementation, S2 specifically includes: S21, edge node performance status collection unit, used to collect node performance status information of distributed edge computing nodes in real time, including computing power , Storage Capacity , Communication bandwidth and calculate load ,in Represents the distributed edge computing node number and generates the node performance state vector ; S22, data task priority evaluation unit, obtains data task characteristics, including task urgency , computing requirements and data size , construct the task feature vector , according to the node performance state vector and the task feature vector Calculate each data task At the node Priority value : ; in, , , is the dynamically adjusted weight factor, To calculate the load, For computing power, For communication bandwidth, the priority values ​​are organized into a data task priority matrix ; S23, the task path dynamic optimization unit is connected to the edge node performance status collection unit and the data task priority evaluation unit to collect network connection information between distributed edge computing nodes and combine the data task priority matrix , based on the micro-delay optimization scheduling algorithm, the optimal transmission path for data tasks is selected, and the path optimization goal is to minimize the data task In distributed edge computing nodes Transmission delay and execution time: ; in, For the task At the node The total time cost on For Node and The communication bandwidth between For Node and nodes The fixed transmission delay between the two paths is used, and the optimization result outputs the optimal path set. ,in, Indicates the task The optimal path of S24, task dynamic allocation unit, based on the optimal path set and data task priority matrix , assign data tasks to target distributed edge computing nodes with matching performance, and generate a task allocation table ,in Indicates the task number. Indicates the target path corresponding to the data task and monitors the execution status of the data task.

[0027] This implementation method realizes the optimal transmission path selection and dynamic allocation of data tasks by collecting edge node performance status and evaluating data task priorities, combined with a micro-delay optimization scheduling algorithm. The task allocation table accurately matches the target node, significantly reducing the task transmission delay and execution time, improving the system's real-time performance and resource utilization efficiency, and providing a reliable guarantee for the efficient processing of distributed edge computing tasks.

[0028] In this implementation, S3 specifically includes: S31, lightweight model building unit, connected with the task dynamic allocation unit, according to the task allocation table The data task allocation result in the data task allocation result is combined with the performance state vector of the allocated node and the task feature vector , to allocate distributed edge computing nodes Building a lightweight intelligent model , satisfying the following constraints: ; in, represents the computational complexity of the lightweight model, Indicates the remaining computing capacity of the distributed edge computing nodes, For Node The communication bandwidth For Node The computing power of For Node storage capacity, For Node The computational load; S32, the hierarchical model optimization unit, is connected with the lightweight model construction unit to form a lightweight intelligent model Decompose into base layer model and extension layer models , where the base layer model completes basic feature extraction, and the extended layer model handles deep learning calculations: ; in, is the ratio factor of the basic layer model to the computing power of distributed edge computing nodes, Distributed edge computing nodes The computing power is dynamically adjusted according to the computing load and task characteristics of the distributed edge computing nodes; S33, model dynamic migration unit, connected with task dynamic allocation unit and hierarchical model optimization unit, used for real-time monitoring of data task allocation distributed edge computing nodes When the performance status of the distributed edge computing nodes changes and causes the time cost and migration cost of the lightweight intelligent model to deviate from the threshold, the migration cost of the lightweight intelligent model is calculated. and time cost , the following migration conditions are met: ; in, For Node and The communication bandwidth between them, the migration path is based on the data task priority matrix Select and update the task assignment table after migration ; S34, the model adaptive adjustment unit is connected to the model dynamic migration unit, and according to the feedback information of the migrated data task execution, the parameters of the lightweight intelligent model are adaptively adjusted, including the model compression rate, feature selection strategy and calculation accuracy. After the adaptive adjustment, the lightweight intelligent model configuration is updated and synchronized to the task allocation table. .

[0029] The present invention realizes efficient processing of tasks by distributed edge computing nodes through the construction and hierarchical optimization of lightweight models. The collaborative work of the basic layer model and the extended layer model improves the utilization rate of computing resources. At the same time, through dynamic migration and adaptive adjustment of the model, the stability of the system and the task execution efficiency in performance change scenarios are guaranteed.

[0030] In this implementation manner, the S33 specifically includes: S331, node performance status monitoring unit, connected to the task dynamic allocation unit, combined with the node performance status vector and the task feature vector , real-time monitoring of distributed edge computing nodes The performance status changes are detected, and the monitoring results are transmitted to the migration condition judgment unit; S332, migration condition judgment unit, according to the node performance state vector and the task feature vector , determine the data task allocation distributed edge computing nodes Whether the task requirements are met. When one of the following conditions is not met, the migration operation is triggered: ; The migration condition judgment unit calculates the model migration time cost and migration cost : ; in, Assigning nodes to tasks The complexity of the above model, Distributed edge computing nodes and The communication bandwidth between is the unit data migration cost coefficient. When the following migration conditions are met, migration is started: ; S333, optimal migration path selection unit, combined with data task priority matrix , select the optimal path that meets the migration conditions, and the path selection goal is to minimize the total migration cost : ; in, For Node and nodes The fixed transmission delay between the two outputs the optimal migration path set , Representation Node To Node The set of edges in all migration paths; S334, the task allocation table updating unit is connected to the optimal migration path selection unit, based on the optimal migration path set , migrate the lightweight intelligent model to the target distributed edge computing node , update the task allocation table The distributed edge computing node binding information corresponding to the data task.

[0031] This implementation method realizes efficient migration of lightweight intelligent models and dynamic update of task allocation tables by real-time monitoring of node performance status, dynamically judging migration conditions based on task characteristics, and minimizing migration costs based on the optimal path selection algorithm. This method improves the resource utilization and task execution efficiency of distributed edge computing nodes, and enhances the flexibility and adaptability of the system.

[0032] In this implementation manner, the S4 specifically includes: S41, a multimodal data acquisition unit, which collects multimodal data of equipment operation through a multimodal sensor network deployed on the industrial equipment, wherein the sensor network includes a vibration sensor, a temperature sensor, an acoustic sensor and a current sensor, which respectively collect vibration signals , temperature signal , acoustic signal and electrical signals ,in For the sampling time point, the timestamps of the collected multimodal data are aligned to generate a time-series multimodal data set: ; in, is the total number of time points in the sampling time window; S42, a data preprocessing unit is connected to the multimodal data acquisition unit to collect the time-series multimodal data Preprocessing is performed, and the noise in the data is suppressed by filtering algorithm to generate a denoised data set, and the denoised data set is standardized according to the amplitude to output a standardized multimodal data set : ; in, For time point The standard value of the vibration signal, For time point The standard value of the temperature signal, For time point The standard value of the acoustic signal, For time point Standard value of electrical signal; S43, a multimodal feature extraction unit, connected to the data preprocessing unit, for standardizing the multimodal data set Perform feature extraction and use a cross-node collaborative feature extraction algorithm to generate a multimodal feature vector set from the time domain, frequency domain, and spatial dimensions : ; in, Time domain features, , ; S44, the feature fusion and fault analysis unit is connected to the multimodal feature extraction unit to collect the multimodal feature vectors Input feature fusion unit, using distributed weighted fusion algorithm to generate comprehensive fault feature vector : ; in, , , are the dynamic weighting coefficients of time domain, frequency domain and spatial features respectively; Fault analysis unit combined with comprehensive fault feature vector and a standardized multimodal data set, where the standardized multimodal data set is adjusted by a collaborative optimization algorithm to generate real-time fault perception results: ; in, For real-time fault perception results, Indicates The detection probability of class faults, , is the total number of fault types.

[0033] This implementation method realizes efficient processing and deep coordination of multimodal signals through multimodal data acquisition, preprocessing, feature extraction and fusion analysis, generates a comprehensive fault feature vector by integrating time domain, frequency domain and spatial features, and accurately analyzes the equipment operation status with a real-time fault perception algorithm, effectively improving the accuracy and real-time performance of complex fault mode detection.

[0034] In this implementation manner, S5 specifically includes: S51, node status monitoring unit, collects node operation status data in real time through distributed edge computing nodes, including computing load , memory usage , Communication bandwidth , Power consumption and node temperature ,in Indicates the node number and generates the node running state vector ; S52, the state abnormality detection unit receives the node operation state vector from the node state monitoring unit , define the node operation status threshold, and perform anomaly detection on each operation parameter. If the operation parameter does not meet the following conditions, it will be marked as abnormal: ; in, and Respectively represent the minimum and maximum values ​​of the operating parameters, and organize the detected abnormal parameters into an abnormal parameter set ; S53, node parameter adjustment unit, for the abnormal parameter set transmitted by the state abnormality detection unit , adjust the operating parameters of abnormal distributed edge computing nodes, when computing load In case of an abnormality, the computing load corresponding to the distributed edge computing nodes is dynamically adjusted, and the optimized computing load distribution is: ; in, Distributed edge computing nodes Transfer to other distributed edge computing nodes The computational load, Distributed edge computing nodes after adjustment The computational load, Distributed edge computing nodes after adjustment The computational load; When the communication bandwidth Dynamically adjust the communication bandwidth when an abnormality occurs: ; in, is the bandwidth adjustment factor, Adjusted distributed edge computing nodes Communication bandwidth; When the power consumption and node temperature When abnormal, optimize by adjusting power consumption mode and activating heat dissipation mechanism: ; in, Distributed edge computing nodes after adjustment The node temperature, Distributed edge computing nodes The upper temperature threshold, Distributed edge computing nodes after adjustment Power consumption, Distributed edge computing nodes The upper power consumption threshold; S54, autonomous repair unit, combined with node parameter adjustment unit and comprehensive fault feature vector , autonomously repair the running status of the distributed edge computing nodes, update the node running status vector, remove the abnormal parameters from the original state vector, and make the adjusted running parameters meet the following conditions: ; in, Run the state vector for the adjusted node.

[0035] This implementation method collects and monitors the operating status of distributed edge computing nodes in real time, combines status anomaly detection and dynamic parameter adjustment, optimizes computing load, communication bandwidth, power consumption and temperature status, and completes abnormal status recovery and parameter updates through an autonomous repair mechanism, effectively improving the robustness and operating efficiency of the system and ensuring the stability and real-time performance of industrial equipment.

[0036] In this implementation manner, S6 specifically includes: S61, associated data collection unit, combined with node performance state vector , comprehensive fault feature vector , and combined with the task allocation information and communication bandwidth utilization of distributed edge computing nodes, generate the interactive information matrix between devices , generate the initial set of related data : ; in, Represents a collection of device nodes. Indicates the device and Interaction information between S62, association graph construction unit, based on the initial association data set , build a device association graph ,in Represents a set of distributed edge computing nodes, represents the edge set, is the edge weight matrix, edge weight Calculated by the following formula: ; in, and are weight factors, representing the influence weights of mutual information and performance similarity, respectively. For Node and Euclidean distance of performance state vector.

[0037] S63, a graph computing processing unit, using graph computing technology to process the device association graph Perform calculations to extract multi-dimensional correlation features between devices, and optimize the correlation between devices based on distributed edge computing nodes and edge weights : ; in, Represents the node performance state vector and The correlation function between and equipment the mutual relationship between S64, a dynamic association graph generation unit dynamically updates the device association graph according to the analysis result of the graph calculation processing unit , and generate a device association graph ,in represents the dynamically adjusted edge set, represents the updated edge weight matrix.

[0038] The present invention constructs a device association graph by combining node performance status, fault feature vectors and interaction information, and uses graph computing technology to optimize the association relationship between devices, generate a dynamic association graph, and realize multi-dimensional feature analysis and real-time update of device interactions, thereby improving the ability to accurately model and dynamically optimize the interaction relationship between devices, and effectively improving the comprehensiveness and real-time performance of system monitoring.

[0039] Embodiment 1: In order to verify the feasibility of the present invention in implementation, the present invention is applied to the industrial equipment monitoring system upgrade project of a large manufacturing enterprise. The enterprise needs to perform real-time operation monitoring and fault detection on 200 industrial equipment in its production workshop. The equipment types include machine tools, transmission equipment and temperature control equipment, which have the characteristics of complex operating environment, large amount of data and diverse failure modes. Traditional monitoring methods mainly rely on centralized cloud computing systems. Due to network bandwidth and transmission delay problems, they cannot meet real-time requirements, and the correlation analysis capabilities between devices are insufficient, resulting in low equipment fault detection accuracy and low system operation efficiency. To solve the above problems, the enterprise introduced the real-time monitoring system for industrial equipment based on edge computing of the present invention.

[0040] In actual scenarios, the present invention realizes the real-time collection and processing of industrial equipment operation data by deploying multiple distributed edge computing nodes. Each edge node covers about 10 devices, and the nodes work together to complete the dynamic scheduling of multimodal data and fault feature extraction. The specific application steps are as follows: The vibration signal, temperature signal, acoustic signal and electrical signal of the equipment are collected through the multimodal sensor network. The data collection frequency is 50 times per second, and the time synchronization module is used to align the timestamps to generate a time-series data set. After the data is processed by the edge node, the time domain, frequency domain and space domain features are extracted. The feature fusion module dynamically adjusts the fusion weight according to the equipment operation status to generate a comprehensive fault feature vector.

[0041] In terms of edge node resource management, the task dynamic scheduling module adjusts task allocation according to node performance status and task priority. For example, in a certain equipment abnormality, the node load reached 90%. The system calculated the migration cost and allocated some tasks to neighboring nodes, successfully reducing the node load to 60%. For fault detection, the fault analysis module uses a comprehensive fault feature vector combined with the equipment operation model to generate fault perception results, and generates a real-time equipment association map through the equipment association reasoning module to display the interaction relationship between devices and quickly locate key equipment. The following is the experimental data and effect verification, see Table 1-Table 3: Table 1 Industrial equipment operation monitoring data table

[0042] The results show that the single data collection time of the method of the present invention is shortened to 20% of that of the traditional method, the data transmission delay is reduced by 70%, the fault detection accuracy is improved by 14%, and the edge node resource utilization is increased by 25%.

[0043] Table 2 Fault detection data comparison

[0044] The fault detection module can respond quickly and accurately detect faults in a variety of abnormal situations, especially electrical abnormalities with a success rate of 100%.

[0045] In order to further verify the overall efficiency of the system, the experiment compared the processing efficiency of the traditional cloud computing system and the method of the present invention under different numbers of devices: Table 3 Comparison of system processing efficiency

[0046] When the number of devices doubles, the data processing time of the method of the present invention increases linearly, and the time for task allocation and association map generation is short, which reflects good scalability and real-time performance.

[0047] To sum up, the present invention realizes real-time monitoring and fault detection of industrial equipment through the deep integration of edge computing and multimodal data processing, solves the problems of high latency, low detection accuracy and waste of resources existing in traditional methods, and provides an efficient and reliable technical solution for intelligent monitoring of industrial equipment.

[0048] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. Real-time monitoring system for industrial equipment based on edge computing, characterized by: The steps include: S1, distributed edge computing node deployment module, deploys multiple distributed edge computing nodes, builds a node coordination mechanism, collects and processes industrial equipment operation data, and generates node performance status information; S2, dynamic data flow scheduling module, based on node performance status information, uses micro-delay optimization scheduling algorithm to dynamically adjust data priority and transmission path, and allocates data tasks to matching distributed edge computing nodes; S3, intelligent model dynamic migration module, builds lightweight intelligent models according to the data tasks assigned by the dynamic data flow scheduling module, and dynamically adjusts the deployment of lightweight intelligent models based on the computing performance and load of distributed edge computing nodes; S4, multimodal fault perception module, collects multimodal data of industrial equipment through distributed edge computing nodes and collaboratively extracts fault features; S5, the distributed edge computing node adaptive management module, based on the extracted fault features, monitors the computing load and operation anomalies of the distributed edge computing nodes in real time, and uses an autonomous repair mechanism to adjust the distributed edge computing node parameters and repair the anomalies; S6, equipment correlation reasoning module, based on the extracted fault characteristics and node performance status information, uses graph computing technology to build a multi-dimensional equipment correlation model and generate a dynamic correlation graph.

2. The real-time monitoring system for industrial equipment based on edge computing according to claim 1 is characterized in that: The S2 specifically includes: S21, edge node performance status collection unit, used to collect node performance status information of distributed edge computing nodes in real time, including computing power , Storage Capacity , Communication bandwidth and calculate load ,in Represents the distributed edge computing node number and generates the node performance state vector ; S22, data task priority evaluation unit, obtains data task characteristics, including task urgency , computing requirements and data size , construct the task feature vector , according to the node performance state vector and the task feature vector Calculate each data task At the node Priority value : ； in, , , is the dynamically adjusted weight factor, To calculate the load, For computing power, For communication bandwidth, the priority values ​​are organized into a data task priority matrix ; S23, the task path dynamic optimization unit is connected to the edge node performance status collection unit and the data task priority evaluation unit to collect network connection information between distributed edge computing nodes and combine the data task priority matrix , based on the micro-delay optimization scheduling algorithm, the optimal transmission path for data tasks is selected, and the path optimization goal is to minimize the data task In distributed edge computing nodes Transmission delay and execution time: ； in, For the task At the node The total time cost on For Node and The communication bandwidth between For Node and nodes The fixed transmission delay between the two paths is used, and the optimization result outputs the optimal path set. ,in, Indicates the task The optimal path of S24, task dynamic allocation unit, based on the optimal path set and data task priority matrix , assign data tasks to target distributed edge computing nodes with matching performance, and generate a task allocation table ,in Indicates the task number. Indicates the target path corresponding to the data task and monitors the execution status of the data task.

3. The real-time monitoring system for industrial equipment based on edge computing according to claim 2 is characterized in that: The S3 specifically includes: S31, lightweight model building unit, connected with the task dynamic allocation unit, according to the task allocation table The data task allocation result in the data task allocation result is combined with the performance state vector of the allocated node and the task feature vector , to allocate distributed edge computing nodes Building a lightweight intelligent model , satisfying the following constraints: ； in, represents the computational complexity of the lightweight model, Indicates the remaining computing capacity of the distributed edge computing nodes, For Node The communication bandwidth For Node The computing power of For Node storage capacity, For Node The computational load; S32, the hierarchical model optimization unit, is connected with the lightweight model construction unit to form a lightweight intelligent model Decompose into base layer model and extension layer models , where the base layer model completes basic feature extraction, and the extended layer model handles deep learning calculations: ； in, is the ratio factor of the basic layer model to the computing power of distributed edge computing nodes, Distributed edge computing nodes The computing power is dynamically adjusted according to the computing load and task characteristics of the distributed edge computing nodes; S33, model dynamic migration unit, connected with task dynamic allocation unit and hierarchical model optimization unit, used for real-time monitoring of data task allocation distributed edge computing nodes When the performance status of the distributed edge computing nodes changes and causes the time cost and migration cost of the lightweight intelligent model to deviate from the threshold, the migration cost of the lightweight intelligent model is calculated. and time cost , the following migration conditions are met: ； in, For Node and The communication bandwidth between them, the migration path is based on the data task priority matrix Select and update the task assignment table after migration ; S34, the model adaptive adjustment unit is connected to the model dynamic migration unit, and according to the feedback information of the migrated data task execution, the parameters of the lightweight intelligent model are adaptively adjusted, including the model compression rate, feature selection strategy and calculation accuracy. After the adaptive adjustment, the lightweight intelligent model configuration is updated and synchronized to the task allocation table. .

4. The real-time monitoring system for industrial equipment based on edge computing according to claim 3 is characterized in that: The S33 specifically includes: S331, node performance status monitoring unit, connected to the task dynamic allocation unit, combined with the node performance status vector and the task feature vector , real-time monitoring of distributed edge computing nodes The performance status changes are detected, and the monitoring results are transmitted to the migration condition judgment unit; S332, migration condition judgment unit, according to the node performance state vector and the task feature vector , determine the data task allocation distributed edge computing nodes Whether the task requirements are met. When one of the following conditions is not met, the migration operation is triggered: ； The migration condition judgment unit calculates the model migration time cost and migration cost : ； in, Assigning nodes to tasks The complexity of the above model, Distributed edge computing nodes and The communication bandwidth between is the unit data migration cost coefficient. When the following migration conditions are met, migration is started: ； S333, optimal migration path selection unit, combined with data task priority matrix , select the optimal path that meets the migration conditions, and the path selection goal is to minimize the total migration cost : ； in, For Node and nodes The fixed transmission delay between the two outputs the optimal migration path set , Representation Node To Node The set of edges in all migration paths; S334, the task allocation table updating unit is connected to the optimal migration path selection unit, based on the optimal migration path set , migrate the lightweight intelligent model to the target distributed edge computing node , update the task allocation table The distributed edge computing node binding information corresponding to the data task.

5. The real-time monitoring system for industrial equipment based on edge computing according to claim 2 is characterized in that: The S4 specifically includes: S41, a multimodal data acquisition unit, which collects multimodal data of equipment operation through a multimodal sensor network deployed on the industrial equipment, wherein the sensor network includes a vibration sensor, a temperature sensor, an acoustic sensor and a current sensor, which respectively collect vibration signals , temperature signal , acoustic signal and electrical signals ,in For the sampling time point, the timestamps of the collected multimodal data are aligned to generate a time-series multimodal data set: ； in, is the total number of time points in the sampling time window; S42, a data preprocessing unit is connected to the multimodal data acquisition unit to collect the time-series multimodal data Preprocessing is performed, and the noise in the data is suppressed by filtering algorithm to generate a denoised data set, and the denoised data set is standardized according to the amplitude to output a standardized multimodal data set : ； in, For time point The standard value of the vibration signal, For time point The standard value of the temperature signal, For time point The standard value of the acoustic signal, For time point Standard value of electrical signal; S43, a multimodal feature extraction unit, connected to the data preprocessing unit, for standardizing the multimodal data set Perform feature extraction and use a cross-node collaborative feature extraction algorithm to generate a multimodal feature vector set from the time domain, frequency domain, and spatial dimensions : ； in, Time domain features, , ; S44, the feature fusion and fault analysis unit is connected to the multimodal feature extraction unit to collect the multimodal feature vectors Input feature fusion unit, using distributed weighted fusion algorithm to generate comprehensive fault feature vector : ； in, , , are the dynamic weighting coefficients of time domain, frequency domain and spatial features respectively; Fault analysis unit combined with comprehensive fault feature vector and a standardized multimodal data set, where the standardized multimodal data set is adjusted by a collaborative optimization algorithm to generate real-time fault perception results: ； in, For real-time fault perception results, Indicates The detection probability of class faults, , is the total number of fault types.

6. The real-time monitoring system for industrial equipment based on edge computing according to claim 5 is characterized in that: The S5 specifically includes: S51, node status monitoring unit, collects node operation status data in real time through distributed edge computing nodes, including computing load , memory usage , Communication bandwidth , Power consumption and node temperature ,in Indicates the node number and generates the node running state vector ; S52, the state abnormality detection unit receives the node operation state vector from the node state monitoring unit , define the node operation status threshold, and perform anomaly detection on each operation parameter. If the operation parameter does not meet the following conditions, it will be marked as abnormal: ； in, and Respectively represent the minimum and maximum values ​​of the operating parameters, and organize the detected abnormal parameters into an abnormal parameter set ; S53, node parameter adjustment unit, for the abnormal parameter set transmitted by the state abnormality detection unit , adjust the operating parameters of abnormal distributed edge computing nodes, when computing load In case of an abnormality, the computing load corresponding to the distributed edge computing nodes is dynamically adjusted, and the optimized computing load distribution is: ； in, Distributed edge computing nodes Transfer to other distributed edge computing nodes The computational load, For the adjusted distributed edge computing nodes The computational load, For the adjusted distributed edge computing nodes The computational load; When the communication bandwidth Dynamically adjust the communication bandwidth when an abnormality occurs: ； in, is the bandwidth adjustment factor, Adjusted distributed edge computing nodes Communication bandwidth; When the power consumption and node temperature When abnormal, optimize by adjusting power consumption mode and activating heat dissipation mechanism: ； in, For the adjusted distributed edge computing nodes The node temperature, Distributed edge computing nodes The upper temperature threshold, For the adjusted distributed edge computing nodes Power consumption, Distributed edge computing nodes The upper power consumption threshold; S54, autonomous repair unit, combined with node parameter adjustment unit and comprehensive fault feature vector , autonomously repair the running status of the distributed edge computing nodes, update the node running status vector, remove the abnormal parameters from the original state vector, and make the adjusted running parameters meet the following conditions: ； in, Run the state vector for the adjusted node.

7. The real-time monitoring system for industrial equipment based on edge computing according to claim 5 is characterized in that: The S6 specifically includes: S61, associated data collection unit, combined with node performance state vector , comprehensive fault feature vector , and combined with the task allocation information and communication bandwidth utilization of distributed edge computing nodes, generate the interactive information matrix between devices , generate the initial set of related data : ； in, Represents a collection of device nodes. Indicates the device and Interaction information between S62, association graph construction unit, based on the initial association data set , build a device association graph ,in Represents a set of distributed edge computing nodes, represents the edge set, is the edge weight matrix, edge weight Calculated by the following formula: ； in, and are weight factors, representing the influence weights of mutual information and performance similarity, respectively. For Node and Euclidean distance of performance state vector; S63, a graph computing processing unit, using graph computing technology to process the device association graph Perform calculations to extract multi-dimensional correlation features between devices, and optimize the correlation between devices based on distributed edge computing nodes and edge weights : ； in, Represents the node performance state vector and The correlation function between and equipment the mutual relationship between S64, a dynamic association graph generation unit dynamically updates the device association graph according to the analysis result of the graph calculation processing unit , and generate a device association graph ,in represents the dynamically adjusted edge set, represents the updated edge weight matrix.

Images