A cloud-edge collaborative work digital twin monitoring modeling method and system

Abstract

The application relates to the technical field of industrial plant monitoring, in particular to a cloud-edge collaborative plant digital twin monitoring modeling method and system; the method comprises the following steps: collecting communication link data and spatial coordinate data to construct a plant topology structure diagram and construct a plant digital twin monitoring model; based on a nested graph neural network, a plant topology difference structure diagram is obtained by comparing the existing plant topology structure diagram with a historical plant topology structure diagram; a structure mapping residual function is used for structure mapping consistency verification on the plant difference topology structure diagram to generate an entity device structure change event set; a changed entity device running state data set is obtained to generate a plant performance offset index set; and the changed entity device and other entity devices directly connected with the changed entity device in the plant are monitored; the application provides a cloud-edge collaborative plant digital twin monitoring modeling method which is fused with spatial structure information and communication link features and has dynamic self-adaptive capability, and realizes real-time state monitoring of the plant.

Description

Technical Field

[0001] This invention relates to the field of industrial factory monitoring technology, specifically to a cloud-edge collaborative factory digital twin monitoring modeling method and system. Background Technology

[0002] In the process of digital transformation of manufacturing plants, digital twin monitoring models have gradually become a key technological foundation for integrating perception, modeling, analysis, and control. The aim is to achieve full-cycle perception and real-time control of equipment operating status, production processes, and spatial topology by constructing a high-precision mapping relationship between the physical space of the factory and the virtual model. Existing modeling methods mainly rely on manually configured logical connection structure diagrams or pre-set process flow diagrams for digital twin construction. This structural modeling approach is relatively static and cannot respond to dynamic evolution scenarios such as changes in the structure, position adjustments, and link changes of physical equipment. Furthermore, existing digital twin modeling methods generally neglect the integrated expression of physical connection relationships and spatial coordinate information between physical equipment in the factory, remaining only at the abstract level of logical connection topology. This makes it difficult to support dynamic adjustment modeling at the spatial level, resulting in insufficient model expression granularity, poor spatial consistency, and an inability to meet the digital twin monitoring needs of dynamically reconfigurable factory scenarios.

[0003] Specifically, in actual industrial scenarios, there are two key issues: First, existing factory topology diagrams only express the logical connections between physical devices, without integrating the operating status information and physical spatial location information of the physical devices. They cannot perceive the topology evolution caused by the addition, movement, replacement, or wiring changes of equipment, resulting in the lack of dynamic structural adaptability of the factory digital twin model, which in turn leads to inconsistencies between the virtual and real systems, affecting the accurate monitoring of the factory's operating status and fault location. Second, after changes in the equipment structure, existing modeling methods cannot actively identify other physical devices that have direct communication links or physical connections with the changed physical devices, leading to problems such as missing structural linkages, omissions in monitoring scope, or delayed responses. Summary of the Invention

[0004] The purpose of this invention is to provide a cloud-edge collaborative factory digital twin monitoring and modeling method and system to solve the two key problems mentioned in the background art.

[0005] To achieve the above objectives, the technical solution of the present invention is: a cloud-edge collaborative factory digital twin monitoring and modeling method, comprising:

[0006] S1. Collect communication link data and spatial coordinate data of factory physical equipment based on edge acquisition devices to construct a factory topology map, and upload the factory topology map to the cloud to build a digital twin monitoring model of the factory.

[0007] S2. Based on the nested graph neural network, compare the existing and historical factory topology diagrams to obtain the factory topology difference diagram. Use the structure mapping residual function to perform structure mapping consistency verification on the factory difference topology diagram, identify the changed factory physical equipment and spatial physical constraint change information, generate the physical equipment structure change event set, and upload it to the cloud to update the factory digital twin monitoring model in real time.

[0008] S3. Based on the event set of changes in the structure of physical equipment, use an edge acquisition device to obtain the dataset of the operating status of the changed physical equipment, and perform difference fitting between the dataset of the operating status of the changed physical equipment and the factory digital twin monitoring model to generate a set of factory performance deviation indicators.

[0009] S4. Based on the factory digital twin monitoring model and the factory performance deviation index set, monitor the physical equipment that has been modified and other physical equipment directly connected to it in the factory.

[0010] Preferably, in S1, the communication link data is the data on communication bandwidth, message delay, connection strength and node address mapping relationship collected by the edge acquisition device during the communication process between nodes inside the factory, and the spatial coordinate data is the three-dimensional spatial position data of each physical device in the standard coordinate system obtained by the edge acquisition device based on the factory layout calibration system.

[0011] The factory topology diagram is a spatial structure association diagram formed by integrating communication link data and spatial coordinate data. It is constructed by taking the factory physical equipment as graph nodes and the connection relationship of communication link data as graph edges. The three-dimensional spatial position parameters of the corresponding physical equipment are embedded in each graph node, and the communication link feature values ​​are recorded in the graph edges to construct a factory topology diagram with dual information of communication relationship and physical spatial structure.

[0012] Preferably, in step S1, the factory digital twin monitoring model consists of a topology layer composed of a set of structural nodes of factory physical equipment and a set of communication link edges, and a geometric layer composed of the spatial distribution status of equipment. The topology layer is used to represent the functional connection relationship between factory physical equipment, and the geometric layer is used to represent the spatial location and adjacent boundary information of the equipment in the physical space. The topology layer and the geometric layer establish a cross-layer mapping relationship through the unique identifier of the node.

[0013] The method for constructing a factory digital twin monitoring model is as follows:

[0014] The communication link data and spatial coordinate data are uploaded to the cloud. A communication topology map is constructed based on the communication link data, and a spatial entity map is generated based on the spatial coordinate data. A unique communication identifier number is established for each entity device in the communication topology map, and a spatial location identifier number for the corresponding entity device is established in the spatial entity map. The communication topology map and the spatial entity map are mapped and merged one by one based on the dual-number mapping mechanism to build a digital twin monitoring model for the factory.

[0015] Preferably, in S2, the nested graph neural network is used to extract structural features at different levels in the existing factory topology diagram and the historical factory topology diagram and construct a structural representation tensor. The nested graph neural network includes a main graph feature encoding subnetwork and a subgraph nesting aggregation module. The main graph feature encoding subnetwork is used to perform graph embedding learning on the entity device nodes and their communication relationships in the topology layer to extract communication structure features. The subgraph nesting aggregation module is used to perform convolution nesting processing on the local substructures of different spatial regions in the geometric layer to extract spatial clustering features.

[0016] The method for generating the factory topology difference structure map is as follows: using a nested graph neural network to compare the existing factory topology structure map with the historical factory topology structure map, constructing a structural representation tensor, and performing residual calculation of the structural representation tensor based on node feature alignment and spatial neighborhood similarity determination to obtain the factory topology difference structure map;

[0017] The factory topology difference diagram is used to describe changes in the direct connection relationships between physical equipment in the factory and differences in spatial structure.

[0018] Preferably, in step S2, the structure mapping residual function is used to calculate the structural variation metric value of each node in the factory topology difference structure diagram. The structure mapping residual function is defined as follows:

[0019] Perform first-order residual calculation on the structural feature tensor of each node in the current factory topology difference diagram and the historical factory topology difference diagram to obtain the structural residual vector; construct a spatial collaborative weighting factor based on the spatial coordinate relationship and edge change of each node and its first-order neighbor nodes; and calculate the structural mapping residual scalar value by weighting and fusing the structural residual vector with the corresponding spatial collaborative weighting factor.

[0020] The structure mapping consistency check specifically includes:

[0021] Calculate the structural mapping residual scalar value for all nodes in the factory topology difference structure diagram and compare it with the set structural evolution consistency judgment threshold. When the structural mapping residual scalar value exceeds the threshold, it is determined that the node has a consistency anomaly in the structural mapping. Mark the node with structural mapping anomaly as a structural change node, and identify the spatial physical constraint change information by combining its adjacent edge information, and output the set of physical equipment structural change events.

[0022] Preferably, in S2, the set of events for structural changes in physical equipment is a set of events composed of multiple event records organized according to a time series structure; each event record includes equipment identification information, structural evolution timestamp, structural evolution category, structural residual vector value, spatial collaborative weight factor value, and upstream and downstream node index information of structural changes.

[0023] Preferably, in step S3, the modified entity equipment operating status dataset is a set of multi-source operating condition monitoring data obtained by the edge acquisition device based on the equipment identification information and timestamp in the structural change event record, and calling the corresponding sensor channel.

[0024] The difference fitting is a process of calculating the feature vector residual between the current operating status data and the historical operating condition data corresponding to the physical equipment in the factory digital twin monitoring model. The residual calculation adopts a local model fine-tuning fitting mechanism, which integrates the Euclidean distance and the associated attribute offset between the current operating status vector and the historical operating condition baseline vector to calculate the residual score.

[0025] The factory performance offset index set is a set of sub-device performance offset index records constructed based on the residual score results of fitting the differences of multiple modified physical devices. The index set structure is an indexed time series format, recording the offset type, offset magnitude and offset trend corresponding to each device.

[0026] Preferably, in step S4, the other physical devices directly connected to it are physical entity nodes that have a direct connection relationship with the modified physical device in the factory topology diagram, specifically factory physical devices that have a direct communication link and a direct physical connection relationship with the modified physical device.

[0027] The monitoring method includes: collecting the operating status data of the modified physical equipment in the edge acquisition device and uploading it to the cloud to input into the factory digital twin monitoring model for real-time monitoring.

[0028] On the other hand, the present invention provides a cloud-edge collaborative factory digital twin monitoring and modeling system, including a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the computer program to implement the cloud-edge collaborative factory digital twin monitoring and modeling method described above.

[0029] Compared with the prior art, the above-mentioned technical solution of the present invention has the following beneficial technical effects:

[0030] 1. In this invention, based on explicitly mapping the data structure of edge acquisition to the spatial adjacency relationship and actual communication path between devices, the dual information of spatial configuration constraints of physical devices and communication link structure is introduced in the model construction process. This enables the constructed model to not only realistically express the physical connection relationship between devices, but also to have sensitivity to structural changes and synchronous adaptation capabilities, breaking through the limitation of traditional topology models that cannot perceive the correlation between the actual spatial layout and operating status of devices.

[0031] 2. In this invention, by modeling the direct interaction between structural change events and key equipment, the system dynamically identifies and marks adjacent equipment with strong coupling effects while updating the modeling state, and performs multi-target state collaborative monitoring based on the model's real-time response mechanism, so that the system can automatically expand the monitoring range and focus on potential impact paths even after minor structural changes. Attached Figure Description

[0032] Figure 1 This is a flowchart of one embodiment of the present invention. Detailed Implementation

[0033] Example 1, as Figure 1 As shown, the specific implementation steps of the cloud-edge collaborative factory digital twin monitoring and modeling method proposed in this invention are as follows:

[0034] S1. Collect communication link data and spatial coordinate data of factory physical equipment based on edge acquisition devices to construct a factory topology map, and upload the factory topology map to the cloud to build a digital twin monitoring model of the factory.

[0035] S2. Based on the nested graph neural network, compare the existing and historical factory topology diagrams to obtain the factory topology difference diagram. Use the structure mapping residual function to perform structure mapping consistency verification on the factory difference topology diagram, identify the changed factory physical equipment and spatial physical constraint change information, generate the physical equipment structure change event set, and upload it to the cloud to update the factory digital twin monitoring model in real time.

[0036] S3. Based on the event set of changes in the structure of physical equipment, use an edge acquisition device to obtain the dataset of the operating status of the changed physical equipment, and perform difference fitting between the dataset of the operating status of the changed physical equipment and the factory digital twin monitoring model to generate a set of factory performance deviation indicators.

[0037] S4. Based on the factory digital twin monitoring model and the factory performance deviation index set, monitor the physical equipment that has been modified and other physical equipment directly connected to it in the factory.

[0038] In this embodiment S1, the communication link data is the data on communication bandwidth, message delay, connection strength and node address mapping relationship collected by the edge acquisition device during the communication between nodes inside the factory, and the spatial coordinate data is the three-dimensional spatial position data of each physical device in the standard coordinate system obtained by the edge acquisition device based on the factory layout calibration system.

[0039] The factory topology diagram is a spatial structure association diagram formed by integrating communication link data and spatial coordinate data. It is constructed by taking the factory physical equipment as graph nodes and the connection relationship of communication link data as graph edges. The three-dimensional spatial position parameters of the corresponding physical equipment are embedded in each graph node, and the communication link feature values ​​are recorded in the graph edges to construct a factory topology diagram with dual information of communication relationship and physical spatial structure.

[0040] In this embodiment S1, the edge acquisition device is a high-frequency acquisition node deployed at the edge of the factory. It has multiple communication interfaces and positioning modules, and can realize the synchronous acquisition of factory communication network operation data and physical space layout parameters. The edge acquisition device includes at least: a network protocol parsing module, a data caching and compression module, an inertial measurement unit (IMU) or UWB positioning module, an industrial camera and a data transmission module. Through the integrated hardware and software system, the edge acquisition device can stably access various physical devices or controllers in the factory and continuously collect specified types of factory status information.

[0041] The data acquisition strategy based on edge acquisition devices uses a multi-cycle hierarchical scheduling mechanism. During the static stable period, the edge acquisition devices adopt a fixed-cycle low-frequency sampling mode, and automatically switch to a high-frequency capture mode during the state change period, dynamically adjusting the acquisition cycle and buffer capacity. Communication link data acquisition prioritizes non-intrusive methods that do not introduce additional load during daily production operations, and obtains communication messages between devices within the factory through mirror listening or protocol stack parsing. Spatial coordinate data acquisition combines benchmark calibration and inertial positioning, or integrates UWB high-precision positioning results with video image recognition to extract the spatial position information of the target device in the global factory coordinate system.

[0042] Factory physical equipment refers to physical devices or control units deployed in a factory environment that have unique identifiers and participate in production tasks. These include, but are not limited to, industrial robots, CNC equipment, conveying devices, sensors, actuators, energy management units, and edge computing gateways. All such devices can be accessed through industrial communication protocols to obtain their operating status and network connectivity. Factory communication link data is obtained by monitoring communication messages between devices through edge acquisition devices. Based on the extraction of source and destination addresses, bandwidth utilization, link latency, packet loss rate, and message integrity indicators are recorded to reflect the network connection strength and operational stability between factory physical equipment. Spatial coordinate data is obtained by spatially calibrating each physical device to a unified three-dimensional reference system in the design coordinates of drawings or the measured coordinate system, forming a consistent model with accurate positional relationships.

[0043] The construction of the factory topology diagram is completed at the edge. That is, after the edge acquisition device collects and initially processes communication link data and spatial coordinate data, it merges the communication structure and spatial structure based on local rules to construct a composite graph model containing nested communication and spatial graph structures. During the construction process, factory physical equipment is used as graph nodes, and the node attributes include the spatial three-dimensional coordinates of the equipment. Directed or undirected edges are established between two devices with communication relationships, and communication quality parameters are recorded in the edge attributes, thus obtaining a composite factory topology diagram that includes both spatial layout and network logical connection. After the graph is constructed, it is uploaded to the cloud for the initial modeling process of the factory digital twin monitoring model.

[0044] In this embodiment S1, the factory digital twin monitoring model consists of a topology layer composed of a set of factory physical equipment structural nodes and a set of communication link edges, and a geometric layer composed of the spatial distribution status of the equipment. The topology layer is used to represent the functional connection relationship between factory physical equipment, and the geometric layer is used to represent the spatial location and adjacent boundary information of the equipment in the physical space. The topology layer and the geometric layer establish a cross-layer mapping relationship through the unique identifier of the node.

[0045] The method for constructing a factory digital twin monitoring model is as follows:

[0046] The communication link data and spatial coordinate data are uploaded to the cloud. A communication topology map is constructed based on the communication link data, and a spatial entity map is generated based on the spatial coordinate data. A unique communication identifier number is established for each entity device in the communication topology map, and a spatial location identifier number for the corresponding entity device is established in the spatial entity map. The communication topology map and the spatial entity map are mapped and merged one by one based on the dual-number mapping mechanism to build a digital twin monitoring model for the factory.

[0047] In this embodiment S1, the factory digital twin monitoring model is used to construct a structure-space dual-dimensional virtual modeling structure for factory physical equipment in the cloud. This model consists of a two-layer structure composed of a topology layer and a geometry layer. The topology layer represents the communication link connections between various physical devices in the factory, including the relationship between device nodes and communication edges. This layer structure is formed by constructing a communication topology graph using communication link data. The geometry layer represents the distribution of physical devices in physical space, including device spatial coordinates, adjacency relationships, and boundary positions. This layer structure is formed by constructing a spatial entity graph using spatial coordinate data. These two layers are not independent but are mapped one-to-one through unique device identifiers to construct a bidirectional fused digital twin modeling structure. The numbering mechanism is based on edge... The acquisition device generates communication and spatial numbers based on a unified device identification rule, and maintains a synchronized mapping state through an index relationship table. During the construction process, a communication node map is first established based on the communication topology map, and then a physical space layout map is generated based on the spatial entity map. On this basis, a two-layer structure model is constructed by merging according to the numbering mapping relationship. The digital twin monitoring model is not a virtual factory structure described by traditional CAD structure or static model, but a cloud-based updatable structure model with topology dynamic perception and spatial layout matching capabilities. The above model is uploaded to the cloud platform after being calculated by edge nodes, and is stored and managed in the cloud in a two-layer structure manner for subsequent operations such as structure comparison, state fitting, and difference analysis.

[0048] In this embodiment S2, the nested graph neural network is used to extract structural features at different levels in the existing factory topology diagram and the historical factory topology diagram and construct a structural representation tensor. The nested graph neural network includes a main graph feature encoding subnetwork and a subgraph nesting aggregation module. The main graph feature encoding subnetwork is used to perform graph embedding learning on the entity device nodes and their communication relationships in the topology layer to extract communication structure features. The subgraph nesting aggregation module is used to perform convolution nesting processing on the local substructures of different spatial regions in the geometric layer to extract spatial clustering features.

[0049] The method for generating the factory topology difference structure map is as follows: using a nested graph neural network to compare the existing factory topology structure map with the historical factory topology structure map, constructing a structural representation tensor, and performing residual calculation of the structural representation tensor based on node feature alignment and spatial neighborhood similarity determination to obtain the factory topology difference structure map;

[0050] The factory topology difference diagram is used to describe changes in the direct connection relationships between physical equipment in the factory and differences in spatial structure.

[0051] In this embodiment S2, the nested graph neural network consists of an embedded structure perception network and an embedded spatial association network. The embedded structure perception network is used to learn the communication link topology relationship between each entity device node in the factory topology diagram and extract structural feature representations that reflect the node adjacency pattern and path depth. The embedded spatial association network is used to combine the three-dimensional spatial coordinate information of each node with the node physical type to model the layout similarity and functional correlation of nodes in physical space. After encoding the existing factory topology diagram and the historical factory topology diagram respectively, the nested graph neural network outputs the corresponding structural feature tensor set and spatial feature tensor set.

[0052] In the process of generating the factory topology difference structure diagram, residual calculation of the structural representation tensor is performed based on node feature alignment and spatial neighborhood similarity determination, as follows: Node feature alignment involves pairing and filtering node sets in the two topology structure diagrams based on global node identifiers, establishing a one-to-one mapping relationship, and normalizing the structural features of the nodes in the mapping pair; the calculation of spatial neighborhood similarity involves comparing the spatial relative position difference, communication link degree value change, and entity type change information between the current node and its first-order neighbor nodes within the local spatial neighborhood of each mapping node pair; based on node feature alignment and spatial neighborhood similarity, the residual calculation of the structural representation tensor is defined, that is, calculating the residual of the current node when the node is aligned with its first-order neighbor node. The previous node has a residual vector between the structural representation tensors in the existing and historical topology graphs, and a spatial neighborhood structural similarity index is introduced as a residual correction factor to form a structural mapping residual tensor. Based on the structural mapping residual tensor, a preset residual amplitude threshold function is set. When the structural mapping residual tensor of a node exceeds the threshold in any dimension, it is determined that the node has undergone a structural change event in the topology graph. Then, all nodes determined to have structural change events and their edges are aggregated to form a factory topology difference structure graph, and finally a factory topology difference structure graph reflecting the structural differences between the two factory topology graphs is obtained. All of the above processes can be executed in parallel on edge or cloud platforms.

[0053] In this embodiment S2, the structure mapping residual function is used to calculate the structural variation metric value of each node in the factory topology difference structure diagram. The structure mapping residual function is defined as follows:

[0054] Perform first-order residual calculation on the structural feature tensor of each node in the current factory topology difference diagram and the historical factory topology difference diagram to obtain the structural residual vector; construct a spatial collaborative weighting factor based on the spatial coordinate relationship and edge change of each node and its first-order neighbor nodes; and calculate the structural mapping residual scalar value by weighting and fusing the structural residual vector with the corresponding spatial collaborative weighting factor.

[0055] The structure mapping consistency check specifically includes:

[0056] Calculate the structural mapping residual scalar value for all nodes in the factory topology difference structure diagram and compare it with the set structural evolution consistency judgment threshold; when the structural mapping residual scalar value exceeds the threshold, it is determined that the node has a consistency anomaly in the structural mapping; mark the node with structural mapping anomaly as a structural change node, and identify the spatial physical constraint change information by combining its adjacent edge information, and output the set of physical equipment structural change events;

[0057] In this context, the physical device represented by the structural change node is the physical device that is being changed. In the factory topology difference structure diagram, it is a structural change node, and in the factory physical physical device, it is the physical device that is being changed.

[0058] In this embodiment S2, the first-order residual calculation refers to performing first-order difference processing on the node structural feature tensors in the current factory topology difference diagram and the historical factory topology diagram. The calculation is based on the Euclidean distance or structural tensor norm difference between the current and historical structural feature tensors of the node, forming a structural residual vector. The structural feature tensor of each node includes information such as node identifier, equipment category, sampling timestamp, and topological degree of the connection with neighboring nodes. The spatial coordinate relationship and connection change of its first-order neighbor nodes are used to identify the connection radius, direction distribution, spatial relative density, and connection existence changes of the node in the actual physical space. The connection changes include types such as new connections, broken connections, changes in connection direction, and abrupt changes in connection length.

[0059] The spatial collaborative weighting factor is a dynamic weighting coefficient used to measure the importance of the structural residual vectors of different nodes. The weighting factor consists of the following three parts: the node spatial location gradient factor (reflecting the change density of its region in the topological graph), the edge variation complexity factor (reflecting the degree of edge perturbation of its surrounding structure), and the topological nesting influence factor (reflecting its influence on the overall graph structure stability). The spatial collaborative weighting factor is calculated by the edge-side local graph neural network and dynamically weighted and updated in combination with the recent historical change frequency of the node.

[0060] The process of weighted fusion of structural residual vectors and corresponding spatial collaborative weighting factors includes: multiplying the structural residual vector elements of each node by their corresponding spatial collaborative weighting factors, and normalizing them to structural mapping residual scalar values ​​after fusion; when the structural mapping residual scalar value exceeds the structural evolution consistency judgment threshold, the node is considered to have structural evolution anomalies. This threshold is adaptively set after cloud model training and dynamically adjusted for different factory scenarios to adapt to the evolution stability characteristics of different factory topologies; further analyzing the spatial physical connection constraint relationship between adjacent nodes in conjunction with the adjacent edge information of each structural node to identify spatial physical constraint change information; the spatial physical constraint change information includes, but is not limited to: whether the connection direction in the structural pipeline is reversed, whether the connection port is replaced, the spatial angle change between supporting structures, whether the actual physical connection is detached or redundant, and whether the spatial path is compressed, bent or nonlinearly extended, etc.

[0061] In this embodiment S2, the set of events for structural changes of physical equipment is a set of events composed of multiple event records organized according to a time series structure; each event record includes equipment identification information, structural evolution timestamp, structural evolution category, structural residual vector value, spatial collaborative weight factor value, and upstream and downstream node index information of structural changes.

[0062] In this embodiment, the device identification information is used to uniquely identify the target entity device undergoing structural evolution; the structural evolution timestamp is used to record the time when the structural change event occurs; the structural evolution category is used to distinguish the type of structural evolution. The system labels different event types such as node addition, position change, connection change, and device replacement into four categories: addition, new position, connection change, and replacement, to guide the cloud modeling logic in selecting the corresponding evolution rules; the structural residual vector value is the structural feature change index calculated by the structural mapping residual function; the spatial coordination weight factor value is used to characterize the degree of influence of device structural changes on the overall stability of the topology; the upstream and downstream node index information of the structural change includes the upstream device number and downstream device number directly connected to the target device, indicating the numbering information of the upstream and downstream nodes directly connected to the target device in the topology involved in the current structural change event. This data field is directly derived from the adjacency list structure in the topology graph and can be used for automatic expansion of the monitoring range and identification of the impact chain in subsequent steps.

[0063] In this embodiment S3, the modified entity equipment operating status dataset is a set of multi-source operating condition monitoring data obtained by the edge acquisition device based on the equipment identification information and timestamp in the structural change event record, and calling the corresponding sensor channel. The multi-source operating condition monitoring data includes, but is not limited to, parameters such as real-time power, communication stability, current fluctuation, heat load and operating efficiency.

[0064] The difference fitting is a process of calculating the feature vector residual between the current operating status data and the historical operating condition data corresponding to the physical equipment in the factory digital twin monitoring model. The residual calculation adopts a local model fine-tuning fitting mechanism, which integrates the Euclidean distance and the associated attribute offset between the current operating status vector and the historical operating condition baseline vector to calculate the residual score.

[0065] The factory performance deviation index set is a set of device performance deviation index records constructed based on the residual score results of fitting the differences of multiple modified physical devices. The index set structure is an indexed time series format, recording the deviation type, deviation magnitude and deviation trend corresponding to each device. The deviation types include four categories: performance degradation, abnormal energy consumption, communication jitter and unstable operation.

[0066] In this embodiment S3, the modified entity equipment operating status dataset refers to a dataset formed by directional high-frequency sampling of the state parameters of the target equipment in the edge acquisition device, based on the target equipment set determined by the entity equipment structural change event set. The specific collected state parameters include, but are not limited to: the equipment's current operating current, voltage, temperature, vibration frequency, operating power, equipment task execution status, start / stop status, real-time load rate, and the data transmission rate and response latency of the equipment's external connection links. The operating status dataset is in time-series format, retaining high-precision timestamps and establishing sampling windows before and after the structural change events, with periodic rolling sampling performed at the edge using a sliding window method. The difference fitting refers to using the modified entity equipment operating status dataset as dynamic input and comparing it with the updated factory... In the digital twin monitoring model, the simulated state output of the device is compared dimension by dimension to calculate the state offset of each device dimension parameter. The residual calculation method adopts the orthogonal projection residual function and combines the sample stability coefficient to suppress abnormal fluctuations. In the difference fitting process, the time alignment structure of the "actual state vector sequence" and the "simulated state vector sequence" is constructed in parallel in the form of a two-way residual window. After performing sequence normalization and multi-dimensional smoothing completion on abnormal interpolation points, the offset aggregation index is obtained by using a dynamic weighted residual aggregation strategy. The factory performance offset index set consists of the difference fitting results of all modified entity devices, specifically including the mean of the operating state residual, the maximum offset amplitude, the residual change rate, the residual fluctuation variance, the frequency of state change, the fitting trend slope, and the residual reconstruction confidence of each target device.

[0067] In this embodiment S4, the other physical devices directly connected to it are physical entity nodes that have a direct connection relationship with the modified physical device in the factory topology diagram, specifically factory physical devices that have a direct communication link and a direct physical connection relationship with the modified physical device.

[0068] The monitoring method includes: collecting the operating status data of the modified physical equipment in the edge acquisition device and uploading it to the cloud to input into the factory digital twin monitoring model for real-time monitoring.

[0069] In this embodiment S4, the other physical devices directly connected to the modified physical device refer to physical entity nodes that have a direct connection relationship with the modified physical device in the factory topology diagram. The factory topology diagram is constructed by acquiring communication link data and spatial coordinate data of each physical device in the factory through an edge acquisition device. The direct connection relationship includes two types: one is a communication link connection relationship, that is, there is a direct data communication path between two physical devices, and this path can be determined by the connection status of edge nodes in the network topology layer; the other is a spatial object connection relationship.

[0070] The connection relationship refers to the physical connection, nesting, linkage, or contact between two devices. This relationship is determined by spatial coordinate data and structural layout rules. A connection is considered "directly connected" if either of the above two types of connection relationships is satisfied. In the monitoring method, the operating status data of the identified modified entity device is first collected in the edge acquisition device. This operating status data includes, but is not limited to, parameters such as power consumption, voltage, current, temperature, vibration frequency, and execution cycle. Then, the collected operating status data is uploaded to the cloud and input into the factory digital twin monitoring model. The factory digital twin monitoring model dynamically updates based on the latest uploaded factory topology diagram and equipment operating data, and constructs a local topology subgraph with the modified device as a core node. It outputs monitoring results in real time, including the status information of its directly connected devices, thus achieving dynamic identification of related equipment status anomalies that may be caused by structural changes.

[0071] Example 2: The present invention proposes a cloud-edge collaborative factory digital twin monitoring and modeling system, which is applied to the cloud-edge collaborative factory digital twin monitoring and modeling method proposed in Example 1. It includes a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the computer program to implement the cloud-edge collaborative factory digital twin monitoring and modeling method in Example 1.

[0072] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A cloud-edge collaborative factory digital twin monitoring and modeling method, characterized in that, Includes the following steps: S1. Collect communication link data and spatial coordinate data of factory physical equipment based on edge acquisition devices to construct a factory topology map, and upload the factory topology map to the cloud to build a digital twin monitoring model of the factory. The factory topology diagram is a spatial structure association diagram formed by integrating communication link data and spatial coordinate data. Its construction method is as follows: using factory physical equipment as graph nodes and communication link data connection relationships as graph edges, embedding the corresponding three-dimensional spatial position parameters of the physical equipment in each graph node, and recording communication link feature values ​​in the graph edges, thus constructing a factory topology diagram with both communication relationship and physical spatial structure information. The factory digital twin monitoring model consists of a topology layer composed of a set of factory physical equipment structure nodes and a set of communication link edges, and a geometric layer composed of the spatial distribution status of the equipment. The method for constructing the factory digital twin monitoring model is as follows: Upload communication link data and spatial coordinate data to the cloud, construct a communication topology map based on the communication link data, and generate a spatial entity map based on the spatial coordinate data; establish a unique communication identifier number for each entity device in the communication topology map, establish a corresponding spatial location identifier number for the entity device in the spatial entity map, and map and merge the communication topology map and the spatial entity map one by one based on the dual numbering mapping mechanism to construct a factory digital twin monitoring model. S2. Based on the nested graph neural network, compare the existing and historical factory topology diagrams to obtain the factory topology difference diagram. Use the structure mapping residual function to perform structure mapping consistency verification on the factory difference topology diagram, identify the changed factory physical equipment and spatial physical constraint change information, generate the physical equipment structure change event set, and upload it to the cloud to update the factory digital twin monitoring model in real time. In S2, the nested graph neural network is used to extract structural features at different levels from the existing factory topology diagram and the historical factory topology diagram and construct a structural representation tensor. The nested graph neural network includes a main graph feature encoding subnetwork and a subgraph nesting aggregation module. The method for generating the factory topology difference structure map is as follows: the nested graph neural network is used to compare the existing factory topology diagram and the historical factory topology diagram to construct a structural representation tensor. Based on node feature alignment and spatial neighborhood similarity determination, the residual calculation of the structural representation tensor is performed to obtain the factory topology difference structure map. In S2, the structure mapping residual function is used to calculate the structural variation metric value of each node in the factory topology difference structure diagram. The structure mapping residual function is defined as follows: Perform first-order residual calculation on the structural feature tensor of each node in the current factory topology difference diagram and the historical factory topology difference diagram to obtain the structural residual vector; construct a spatial collaborative weighting factor based on the spatial coordinate relationship and edge change of each node and its first-order neighbor nodes; and calculate the structural mapping residual scalar value by weighting and fusing the structural residual vector with the corresponding spatial collaborative weighting factor. S3. Based on the event set of changes in the structure of physical equipment, use an edge acquisition device to obtain the dataset of the operating status of the changed physical equipment, and perform difference fitting between the dataset of the operating status of the changed physical equipment and the factory digital twin monitoring model to generate a set of factory performance deviation indicators. S4. Based on the factory digital twin monitoring model and the factory performance deviation index set, monitor the physical equipment that has been modified and other physical equipment directly connected to it in the factory.

2. The cloud-edge collaborative factory digital twin monitoring and modeling method according to claim 1, characterized in that: In S1, the communication link data is the data on communication bandwidth, message delay, connection strength and node address mapping relationship collected by the edge acquisition device during the communication process between nodes inside the factory, and the spatial coordinate data is the three-dimensional spatial position data of each physical device in the standard coordinate system obtained by the edge acquisition device based on the factory layout calibration system.

3. The cloud-edge collaborative factory digital twin monitoring and modeling method according to claim 2, characterized in that: In S1, the topology layer is used to represent the functional connection relationship between the factory physical equipment, and the geometry layer is used to represent the spatial location and adjacent boundary information of the equipment factory physical equipment in the physical space. The topology layer and the geometry layer establish a cross-layer mapping relationship through the unique identifier of the node.

4. The cloud-edge collaborative factory digital twin monitoring and modeling method according to claim 3, characterized in that: The main graph feature encoding subnetwork is used to perform graph embedding learning on entity device nodes and their communication relationships in the topology layer to extract communication structure features. The subgraph nesting aggregation module is used to perform convolution nesting processing on the local substructures of different spatial regions in the geometric layer to extract spatial clustering features. The factory topology difference diagram is used to describe changes in the direct connection relationships between physical equipment in the factory and differences in spatial structure.

5. The cloud-edge collaborative factory digital twin monitoring and modeling method according to claim 4, characterized in that: The structure mapping consistency check specifically includes: Calculate the structural mapping residual scalar value for all nodes in the factory topology difference structure diagram and compare it with the set structural evolution consistency judgment threshold. When the structural mapping residual scalar value exceeds the threshold, it is determined that the node has a consistency anomaly in the structural mapping. Mark the node with structural mapping anomaly as a structural change node, and identify the spatial physical constraint change information by combining its adjacent edge information, and output the set of physical equipment structural change events.

6. The cloud-edge collaborative factory digital twin monitoring and modeling method according to claim 5, characterized in that: In S2, the set of events for structural changes of physical equipment is a set of events composed of multiple event records organized according to a time series structure; each event record includes equipment identification information, structural evolution timestamp, structural evolution category, structural residual vector value, spatial collaborative weight factor value, and upstream and downstream node index information of structural changes.

7. The cloud-edge collaborative factory digital twin monitoring and modeling method according to claim 6, characterized in that: In S3, the modified entity equipment operating status dataset is a multi-source operating condition monitoring data set obtained by the edge acquisition device based on the device identification information and timestamp in the structural change event record, and calling the corresponding sensor channel. The difference fitting is a process of calculating the feature vector residual between the current operating status data and the historical operating condition data corresponding to the physical equipment in the factory digital twin monitoring model. The residual calculation adopts a local model fine-tuning fitting mechanism, which integrates the Euclidean distance and the associated attribute offset between the current operating status vector and the historical operating condition baseline vector to calculate the residual score. The factory performance offset index set is a set of sub-device performance offset index records constructed based on the residual score results of fitting the differences of multiple modified physical devices. The index set structure is an indexed time series format, recording the offset type, offset magnitude and offset trend corresponding to each device.

8. The cloud-edge collaborative factory digital twin monitoring and modeling method according to claim 7, characterized in that: In S4, the other physical devices directly connected to it are physical entity nodes that have a direct connection relationship with the modified physical device in the factory topology diagram, specifically factory physical devices that have a direct communication link and a direct physical connection relationship with the modified physical device. The monitoring method includes: collecting the operating status data of the modified physical equipment in the edge acquisition device and uploading it to the cloud to input into the factory digital twin monitoring model for real-time monitoring.

9. A cloud-edge collaborative factory digital twin monitoring and modeling system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: The processor executes a computer program to implement the cloud-edge collaborative factory digital twin monitoring and modeling method as described in any one of claims 1-8.

Images