MES-based smart factory management system and method thereof

Abstract

The invention discloses a smart factory management system and method based on MES, and relates to the technical field of smart manufacturing, and the method comprises the steps: calculating a health degree score based on equipment historical maintenance records, and generating a time-space correlation multi-dimensional analysis data set containing an equipment topological structure and health degree parameters in combination with a digital twin network; performing spatial-temporal feature coupling and topological weight dynamic adjustment on the spatial-temporal correlation multi-dimensional analysis data set through a dynamic topological analysis algorithm, and generating a high-risk equipment list and an anomaly control instruction set; based on the high-risk equipment list, constructing a standardized multi-dimensional abnormal feature vector, and generating an aging weighted danger level signal through an entropy weight method; according to the method, space-time alignment of equipment operation state data and physical topology data is realized through a multi-rate Kalman filtering algorithm and an iterative nearest point algorithm, a high-fidelity digital twin network is constructed in combination with a dynamic graph convolutional network, and accurate anomaly detection and health degree evaluation are supported.

Description

Technical Field

[0001] The present invention relates to the field of intelligent manufacturing technology, and in particular to an MES-based intelligent factory management system and method thereof. Background Art

[0002] In modern intelligent manufacturing, factory management technologies based on manufacturing execution systems (MESs) have become a key means of improving production efficiency and equipment reliability. Traditional approaches typically employ condition monitoring techniques, combined with machine learning algorithms to assess equipment health and generate maintenance decisions through a rules engine. This approach enables basic anomaly detection and work order management, while also leveraging the MES to optimize production scheduling. In the application of digital twin technology, existing solutions typically construct virtual factory models through static 3D modeling and data mapping, and utilize time-series databases to store equipment operating parameters, establishing a rudimentary virtual-reality interface.

[0003] However, in terms of heterogeneous data fusion, conventional methods in existing technologies often independently process equipment operating status data and physical topology data, lacking an effective spatiotemporal alignment mechanism, resulting in limited dynamic mapping accuracy of digital twin networks; in terms of decision optimization, most methods use single-objective optimization or static weight rules, which makes it difficult to coordinate and balance multi-dimensional goals such as downtime, maintenance costs, and health improvement, and solution verification relies on physical trial and error, which is less efficient. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a smart factory management method based on MES to solve the problems of insufficient dynamic mapping accuracy of digital twin models and low efficiency of multi-objective maintenance decision-making collaboration.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In the first aspect, the present invention provides a smart factory management method based on MES, which includes real-time collection of equipment operation status data and physical topology data, spatiotemporal alignment of equipment operation status data and physical topology data through a multi-rate Kalman filter algorithm and an iterative nearest point algorithm, and generation of a digital twin network in combination with a dynamic graph convolutional network; calculating a health score based on the historical maintenance records of the equipment, and generating a spatiotemporal correlation multidimensional analysis data set containing equipment topology structure and health parameters in combination with the digital twin network; performing spatiotemporal feature coupling and dynamic adjustment of topological weights on the spatiotemporal correlation multidimensional analysis data set through a dynamic topology analysis algorithm to generate a high-risk equipment list and an abnormal control instruction set; based on the high-risk equipment list, constructing a standardized multidimensional abnormal feature vector, and generating a time-weighted hazard level signal through an entropy weight method; performing dynamic multi-objective collaborative optimization based on the standardized multidimensional abnormal feature vector, the abnormal control instruction set and the time-weighted hazard level signal through a multi-objective particle swarm optimization algorithm and a dynamic constraint solver to generate a set of alternative recovery plans, and screening the optimal recovery strategy through the digital twin network.

[0008] As a preferred solution of the MES-based smart factory management method of the present invention, wherein: the equipment operation status data includes sensor time series data, control system status data and production operation data;

[0009] The physical topology data includes device physical location information, device connection relationships, communication network topology data, and device hierarchical structure relationships.

[0010] As a preferred solution of the MES-based smart factory management method described in the present invention, the multi-rate Kalman filter algorithm and the iterative closest point algorithm are used to perform spatiotemporal alignment of the equipment operation status data and the physical topology data, and a dynamic graph convolutional network is combined to generate a digital twin network. The specific steps are as follows:

[0011] The device operating status data and physical topology data are time-synchronized using a multi-rate Kalman filter algorithm, and spatially aligned using an iterative closest point algorithm to generate a joint operating status-topology dataset.

[0012] Based on the running status-topology joint dataset, a dynamic graph convolutional network is used to perform node attribute neighborhood aggregation and dynamic optimization of edge weights to generate a digital twin network.

[0013] As a preferred solution of the MES-based smart factory management method described in the present invention, the health score is calculated based on the historical maintenance records of the equipment, and a spatiotemporal correlation multidimensional analysis data set containing equipment topology and health parameters is generated in combination with the digital twin network. The specific steps are as follows:

[0014] Based on the equipment's historical maintenance records and equipment operating status data, the equipment's health score is calculated using a hidden Markov algorithm.

[0015] Based on the health score and combined with the digital twin network, the node attributes are integrated through the graph embedding algorithm to generate a digital twin network with health.

[0016] Based on the digital twin network with health, the equipment health parameters and topology structure time series characteristics are extracted through the spatiotemporal sliding window. Combined with GraphSAGE neighborhood aggregation and Pearson correlation coefficient verification, a spatiotemporal correlation multidimensional analysis dataset is generated.

[0017] As a preferred solution of the MES-based smart factory management method described in the present invention, the following specific steps are used to dynamically adjust the spatiotemporal feature coupling and topological weight of the spatiotemporal correlation multidimensional analysis data set through a dynamic topological analysis algorithm to generate a high-risk equipment list and an abnormal control instruction set.

[0018] Based on the spatiotemporal correlation multidimensional analysis data set, the spatiotemporal topological features are fused through the dynamic topological analysis algorithm to generate a spatiotemporal coupling feature matrix with dynamic weights;

[0019] Based on the spatiotemporal coupling characteristic matrix with dynamic weights, multi-source risk propagation simulation is performed through anomaly diffusion rules to generate a list of high-risk equipment;

[0020] Based on the high-risk equipment list, the historical maintenance operation steps are matched by the maintenance type and component identifier in the maintenance record, and a dynamic graph convolutional network is combined to generate an abnormal control instruction set containing specific operation instructions.

[0021] As a preferred solution of the MES-based smart factory management method of the present invention, wherein: based on the high-risk equipment list, a standardized multi-dimensional abnormal feature vector is constructed, and a time-weighted hazard level signal is generated by the entropy weight method. The specific steps are as follows:

[0022] Based on the high-risk equipment list, multi-dimensional abnormal features are extracted using the PageRank algorithm and the dynamic time warping algorithm, and a standardized multi-dimensional abnormal feature vector is generated through Z-Score normalization.

[0023] Based on the standardized multidimensional anomaly feature vector, a time-weighted multidimensional anomaly feature vector is generated through the time-decay entropy weight method and exponentially weighted moving average;

[0024] Based on the time-weighted multidimensional anomaly feature vector, a time-weighted hazard level signal is generated through the quantile risk classification method.

[0025] As a preferred solution of the MES-based smart factory management method described in the present invention, wherein: the standardized multi-dimensional abnormal feature vector, abnormal control instruction set and timely weighted danger level signal are used to perform dynamic multi-objective collaborative optimization through a multi-objective particle swarm optimization algorithm and a dynamic constraint solver to generate a set of alternative recovery solutions, and the optimal recovery strategy is screened through a digital twin network. The specific steps are as follows:

[0026] Based on the standardized multi-dimensional anomaly feature vector, the anomaly control instruction set and the time-weighted danger level signal, a dynamic multi-objective collaborative optimization and Pareto non-dominated sorting are performed through the multi-objective particle swarm optimization algorithm to generate an initial set of alternative recovery solutions.

[0027] Using a dynamic constraint solver to perform hard screening on the initial set of candidate recovery solutions to generate a set of feasible candidate recovery solutions;

[0028] In the digital twin network, the effectiveness of the feasible alternative recovery plan set is verified through Monte Carlo simulation and multi-indicator weighted scoring, and the optimal recovery strategy execution instruction set is generated.

[0029] In the second aspect, the present invention provides a smart factory management system based on MES, including a data acquisition module, a health scoring module, an abnormal control instruction set generation module, a danger level signal generation module and an optimal recovery strategy generation module; the data acquisition module collects equipment operation status data and physical topology data in real time, and performs spatiotemporal alignment of the equipment operation status data and physical topology data through a multi-rate Kalman filter algorithm and an iterative nearest point algorithm, and generates a digital twin network in combination with a dynamic graph convolutional network; the health scoring module calculates the health score based on the equipment's historical maintenance records, and generates a spatiotemporal correlation multidimensional analysis containing the equipment topology structure and health parameters in combination with the digital twin network. The abnormal control instruction set generation module uses a dynamic topology analysis algorithm to couple the spatiotemporal characteristics and dynamically adjust the topological weights of the spatiotemporal correlation multidimensional analysis data set to generate a high-risk equipment list and an abnormal control instruction set; the hazard level signal generation module constructs a standardized multidimensional abnormal feature vector based on the high-risk equipment list, and generates a time-weighted hazard level signal through the entropy weight method; the optimal recovery strategy generation module uses a multi-objective particle swarm optimization algorithm and a dynamic constraint solver to perform dynamic multi-objective collaborative optimization based on the standardized multidimensional abnormal feature vector, the abnormal control instruction set and the time-weighted hazard level signal to generate a set of alternative recovery solutions, and screens the optimal recovery strategy through the digital twin network.

[0030] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the MES-based smart factory management method as described in the first aspect of the present invention is implemented.

[0031] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the MES-based smart factory management method as described in the first aspect of the present invention.

[0032] The beneficial effects of the present invention are: through the multi-rate Kalman filtering algorithm and the iterative nearest point algorithm, the spatiotemporal alignment of equipment operation status data and physical topology data is achieved, and a high-fidelity digital twin network is constructed in combination with a dynamic graph convolutional network to support accurate anomaly detection and health assessment; a set of alternative recovery solutions is generated through a multi-objective particle swarm optimization algorithm, and the virtual-reality collaborative robustness of the strategy is verified with the help of the digital twin network. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0034] Figure 1 This is a flowchart of the MES-based smart factory management method.

[0035] Figure 2 Schematic diagram of the MES-based smart factory management system.

[0036] Figure 3 Flowchart for generating digital twin network.

[0037] Figure 4 A flowchart for generating a recovery strategy. DETAILED DESCRIPTION

[0038] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0039] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0040] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0041] Reference Figures 1 to 4 , is an embodiment of the present invention, which provides a smart factory management method based on MES, including the following steps:

[0042] S1. Real-time collection of equipment operation status data and physical topology data. The multi-rate Kalman filter algorithm and the iterative closest point algorithm are used to align the equipment operation status data and physical topology data in time and space, and the dynamic graph convolutional network is combined to generate a digital twin network. The specific steps are as follows:

[0043] S1.1 collects equipment operation status data and physical topology data. Equipment operation status data includes sensor timing data, control system status data and production operation data. Physical topology data includes equipment physical location information, equipment connection relationship, communication network topology data and equipment hierarchical structure relationship.

[0044] It should be noted that when collecting equipment operation status data and physical topology data, the equipment operation status data obtains sensor timing data in real time through vibration sensors, temperature sensors, and current sensors deployed at key parts of the equipment, extracts control system status data through programmable logic controller instruction sequences, servo motor control signals, and valve opening and closing status logs, and collects equipment start and stop timestamps, material flow statistics, and real-time energy consumption monitoring values ​​through the manufacturing execution interface to form production operation data; physical topology data obtains equipment physical location information by analyzing the computer-aided design equipment three-dimensional coordinates and installation direction Euler angles, defines equipment connection relationships through mechanical transmission chain parameters, electrical wiring topology mapping tables, and gas-liquid pipeline network characteristic curves, describes communication network topology data through the industrial Ethernet protocol communication network node address table and wireless sensor network routing path, and constructs equipment hierarchical structure relationships through the three-level affiliation table of production line-section-equipment and the equipment unique identifier.

[0045] S1.2 uses a multi-rate Kalman filter algorithm to synchronize the device operating status data and physical topology data, and uses an iterative closest point algorithm to perform spatial registration to generate a joint operating status-topology dataset;

[0046] It should be noted that the sensor time series data in the equipment operation status data and the equipment physical position information collection timestamps in the physical topology data are time synchronized through the multi-rate Kalman filtering algorithm, and a multi-rate state transfer matrix is ​​established based on the sampling rate differences of different sensors. The time alignment of multi-source time series data is achieved through covariance matrix updating and measurement noise filtering to generate time-synchronized equipment operation status data; the equipment logical position information in the time-synchronized equipment operation status data and the equipment three-dimensional coordinates generated by computer-aided design in the physical topology data are input into the iterative nearest point algorithm, and the rigid body transformation matrix of the equipment logical position and physical space coordinates is generated through point cloud feature matching. The mechanical transmission chain parameters, electrical wiring topology mapping table and gas-liquid pipeline network characteristic curve in the equipment connection relationship are combined to complete the spatial coordinate conversion and topology association mapping to generate an operation status-topology joint data set containing time synchronization and spatial registration results.

[0047] S1.3 is based on the operating status-topology joint dataset, and uses a dynamic graph convolutional network to perform node attribute neighborhood aggregation and dynamic optimization of edge weights to generate a digital twin network.

[0048] It should be noted that the time-synchronized equipment operating status data in the operating status-topology joint dataset is mapped to dynamic graph node attributes, such as: vibration sensor timing data, temperature sensor timing data, and current sensor timing data; the equipment connection relationship in the operating status-topology joint dataset is mapped to dynamic graph edge attributes, such as: mechanical transmission chain parameters, electrical wiring topology mapping table, and gas-liquid pipeline network characteristic curve; based on the production line-section-equipment three-level affiliation table defined based on the equipment hierarchical structure relationship, the feature information of vibration sensor timing data, temperature sensor timing data, and current sensor timing data is aggregated within the node neighborhood through graph convolution operations; the edge weight parameters are dynamically adjusted according to the industrial Ethernet protocol communication network node address table and the wireless sensor network routing path in the communication network topology data; and finally, a digital twin network with equipment status mapping and anomaly detection functions is generated through multi-layer dynamic graph convolution network processing.

[0049] S2. Calculate the health score based on the historical maintenance records of the equipment, and combine it with the digital twin network to generate a spatiotemporal correlation multidimensional analysis dataset containing the equipment topology and health parameters;

[0050] S2.1 calculates the health score of the equipment based on the equipment's historical maintenance records and equipment operating status data using the Hidden Markov algorithm. The expression is as follows:

[0051] ;

[0052] in, For the device Health score at all times, is the total number of device nodes, For the The dynamic weight coefficient of each node, the value range is (0,1], represents the time decay factor, represents the discrete moment index in the time series, Representing historical moments The observation data at the current moment The exponential decay weight of the evaluation of For the moment The collected data set of equipment operation status, Is the device at the node potential operating status, Indicates that at the moment Observational data Under these conditions, the node In a specific state The posterior probability of .

[0053] It should be noted that historical maintenance records refer to the collection of equipment maintenance data from the entire lifecycle automatically collected and structured by the MES. These records primarily include core information such as equipment maintenance work orders, regular inspection and maintenance records, and breakdown repair reports. These records are automatically generated through real-time interaction between the MES and equipment management functions. They include detailed records such as maintenance operation time (accurate to the minute), maintenance type (such as preventive maintenance, predictive maintenance, breakdown repair, etc.), replacement part number (corresponding to factory material coding), standard operating procedures, operator information, and key equipment parameters before and after maintenance. Unprocessed equipment maintenance work orders, regular inspection and maintenance records, and breakdown repair report data are extracted and converted, and then processed through steps such as time standardization (conversion to a unified time format), maintenance type normalization (referenced to industry standards), part number association (matching with the equipment bill of materials), and abnormal event tagging (based on maintenance text analysis), ultimately forming a structured historical maintenance record.

[0054] It should be noted that the dynamic graph convolutional network generates the weight of each node according to the mechanical transmission chain parameters and electrical wiring topology mapping table in the equipment connection relationship The weight value range is (0,1] and reflects the importance of the node. The hidden Markov algorithm is based on the equipment operation status data after time-space alignment and the maintenance time, maintenance type and replacement component identifier in the historical maintenance record to calculate the The node status probability In order to reflect the temporal characteristics of state evolution, the historical moments Apply exponential decay weights to the state probability , where the attenuation coefficient ∈(0,1] ensures that recent observation data has a higher contribution. Finally, health assessment is achieved through double aggregation: inner summation Calculate the time-weighted state probability of a single node and sum the outer layer Complete multi-node topology fusion to obtain the device Health score at all times .

[0055] S2.2 is based on health score and combined with digital twin network, node attribute fusion is performed through graph embedding algorithm to generate digital twin network with health;

[0056] Should be explained, health score After being input into the digital twin network as node attributes, the health score is embedded into the The physical topology data of the equipment is integrated and processed. First, the physical location information of the equipment nodes in the digital twin network, the equipment connection relationship, the communication network topology data and the equipment hierarchical structure relationship are extracted as the graph structure features. Then the health score is calculated. This is mapped into node feature vectors, which together with operating status data such as vibration sensor time series data, temperature sensor time series data, and current sensor time series data, constitute multidimensional node attributes. Graph convolution operations are used to propagate features within the neighborhood defined by the device connection relationship. Mechanical transmission chain parameters, electrical wiring topology mapping tables, and gas and liquid pipeline network characteristic curves serve as edge attributes to guide information flow. The industrial Ethernet protocol communication network node address table and wireless sensor network routing paths dynamically adjust edge weight parameters to ensure that critical connections have higher propagation weights. Finally, through multi-layer graph convolutional network processing, while retaining the hierarchical structure defined by the three-level affiliation table of production line-section-equipment, a healthy digital twin network is generated that includes health scores and topological correlation characteristics.

[0057] S2.3 is based on a digital twin network with health, and extracts equipment health parameters and topology structure time series characteristics through a spatiotemporal sliding window. Combined with GraphSAGE neighborhood aggregation and Pearson correlation coefficient verification, a spatiotemporal correlation multidimensional analysis data set is generated.

[0058] It should be noted that based on the device health scores and topology data recorded in the health-related digital twin network, a spatiotemporal sliding window is applied to capture continuous time segments of data for device health parameters (health score, health degradation rate) and topology time series features (mechanical transmission chain state time series matrix, communication network load time series). The spatiotemporal sliding window slides across the entire device lifecycle timeline with a fixed step size. The data slices within the window contain synchronized sampled values ​​of health parameters and topology time series features. GraphSAGE neighborhood aggregation is used to aggregate the attributes (health score, communication network load time series) of the device nodes within the window and their neighboring nodes (neighborhood relationships are defined based on the production line-section-device three-level affiliation table in the device hierarchy and the Industrial Ethernet protocol communication network node address table in the communication network topology data) to generate a node aggregate feature vector. The aggregated feature vector is then verified using the Pearson correlation coefficient to verify the linear correlation between device health parameters and topology time series features (such as the mechanical transmission chain state time series matrix and health degradation rate). Feature pairs whose absolute values ​​of the Pearson correlation coefficient exceed a preset screening threshold are selected and associated features are retained. Finally, the screened device health parameters, topology structure time series characteristics and GraphSAGE neighborhood aggregation results are merged into a spatiotemporal correlation multidimensional analysis dataset. The fields of the spatiotemporal correlation multidimensional analysis dataset include device unique identifier, timestamp, health parameter vector, topology feature vector, neighborhood aggregation feature vector and Pearson correlation coefficient verification mark.

[0059] S3. Use dynamic topology analysis algorithms to dynamically adjust the spatiotemporal feature coupling and topology weights of the spatiotemporal correlation multidimensional analysis dataset to generate a list of high-risk equipment and an abnormal control instruction set.

[0060] S3.1 is based on the spatiotemporal correlation multidimensional analysis data set, and the spatiotemporal topological features are integrated through the dynamic topological analysis algorithm to generate a spatiotemporal coupling feature matrix with dynamic weights;

[0061] It should be noted that the spatiotemporal topological features are obtained from the health score by using the hidden Markov algorithm. Extract time dimension statistical features, including health scores The mean characteristics, variance characteristics, and trend slope characteristics of Calculated using a defined formula; using a multi-rate Kalman filter algorithm and an iterative closest point algorithm to align device operating status data with physical topology data, spatial aggregation features are extracted from mechanical transmission chain parameters, electrical wiring topology mapping tables, and gas and liquid pipeline network characteristic curves; and device connection relationships are extracted from the industrial Ethernet protocol communication network node address table and wireless sensor network routing paths. Multi-dimensional features are fused using a dynamic topology analysis algorithm, combining the trend slope features from temporal statistical features, the mechanical transmission chain parameters and electrical wiring topology mapping tables from spatial aggregation features, and the industrial Ethernet protocol communication network node address table from device connection relationships. Node weight parameters are dynamically adjusted based on the maintenance type and replacement part identifiers in maintenance records to generate a spatiotemporal coupling feature matrix with dynamic weights.

[0062] S3.2 Based on the spatiotemporal coupling characteristic matrix with dynamic weights, multi-source risk propagation simulation is performed through anomaly diffusion rules to generate a list of high-risk equipment;

[0063] It should be noted that the anomaly diffusion rule transforms the device connectivity relationships (including mechanical transmission chain parameters, electrical wiring topology mapping tables, and gas and liquid pipeline network characteristic curves) and communication network topology data (Industrial Ethernet protocol communication network node address tables and wireless sensor network routing paths) in the dynamic weighted spatiotemporal coupling feature matrix into a weighted directed graph structure. The nodes of the weighted directed graph correspond to unique device identifiers, and the edge weights are defined by the mechanical transmission chain parameters, electrical wiring topology mapping tables, or gas and liquid pipeline network characteristic curves, constraining the risk propagation path and its intensity. When the health score of a device node falls below a preset health score threshold, its abnormal state propagates risk value to neighboring nodes via the edge weights of the weighted directed graph. The propagation ratio is the product of the edge weight and the time decay factor in the time-weighted hazard level signal (e.g., the maintenance interval in the maintenance record). After each risk propagation iteration, the cumulative risk value of all device nodes is calculated, and the cross-level risk accretion effect is calculated based on the equipment hierarchy (the three-level affiliation table of production line, section, and equipment). Finally, the device nodes whose cumulative risk values ​​exceed the risk screening threshold are selected to form a list of high-risk devices containing the device unique identifier, risk level, associated maintenance type and propagation path.

[0064] S3.3 is based on the high-risk equipment list, matches historical maintenance operation steps by the maintenance type and component identifier in the maintenance record, and combines the dynamic graph convolutional network to generate an abnormal control instruction set containing specific operation instructions.

[0065] It should be noted that the historical maintenance operation steps are standardized operating procedures extracted from the standard operating instruction number field in the maintenance record, which include the maintenance task execution sequence, tool usage specifications and safety operation requirements. The specific content is determined by the association between the maintenance type code and the replacement part identifier. When the maintenance type and part identifier in the maintenance record match the historical maintenance operation steps, the maintenance type corresponds to the maintenance type code field (preventive maintenance code, predictive maintenance code, fault repair code) in the historical maintenance operation steps, and the part identifier corresponds to the replacement part identifier field (the unique part number in the factory material coding system) in the historical maintenance operation step record. The historical maintenance operation step database is jointly queried through the historical maintenance type code and replacement part identifier fields to obtain historical maintenance operation step records that fully match the current maintenance type code and replacement part identifier, including the standard operating instruction number, operation tool list and safety specification code.

[0066] Based on the unique identifiers of devices in the high-risk equipment list, a dynamic graph convolutional network extracts the hierarchical structure of equipment (a three-level affiliation table of production line, work section, and equipment) and communication network topology data (an industrial Ethernet communication network node address table and wireless sensor network routing paths) from the spatiotemporal correlation multidimensional analysis dataset. Using a neighborhood aggregation operation, the maintenance type code and replacement part identifier of the current device node are combined with the maintenance type codes and replacement part identifiers of associated neighboring nodes to generate a node feature vector. After the node feature vector is transformed by the weight matrix of the graph convolution layer, the output is an exception control instruction set containing specific operational instructions. The exception control instruction set includes codes for the sequence of operational steps, a spare parts replacement list, a tool usage list, and a safety specification execution priority. The coding rules are consistent with the standard operating instruction numbers in the maintenance records.

[0067] S4. Based on the high-risk equipment list, a standardized multi-dimensional abnormal feature vector is constructed, and a time-weighted hazard level signal is generated through the entropy weight method;

[0068] S4.1 Extract multidimensional abnormal features based on the high-risk equipment list using the PageRank algorithm and the dynamic time warping algorithm, and generate a standardized multidimensional abnormal feature vector through Z-Score normalization;

[0069] It should be noted that a device communication network topology diagram is constructed based on the industrial Ethernet protocol communication network node address table and the wireless sensor network routing path, and the influence weight value of the device node is calculated by the PageRank algorithm to generate the device node influence feature; the timing change feature is aligned by the dynamic time warping algorithm: the equipment operation status timing series is extracted from the high-risk equipment list, and the timing data length and phase deviation of different device nodes are aligned by the dynamic time warping algorithm to generate the time series alignment feature; the device node influence feature, the time series alignment feature and the trend slope feature of the health score H(t) are merged to generate a multidimensional anomaly feature; the device node influence feature, the time series alignment feature and the health score trend slope feature in the multidimensional anomaly feature are respectively subjected to Z-Score normalization processing, and the dimensional differences are eliminated and then integrated into a standardized multidimensional anomaly feature vector.

[0070] S4.2 generates a time-weighted multidimensional anomaly feature vector based on the standardized multidimensional anomaly feature vector through the time-decay entropy weight method and exponentially weighted moving average;

[0071] It should be noted that based on the health score trend slope characteristics, time series alignment characteristics, and device node influence characteristics in the standardized multidimensional abnormal feature vector, based on the standardized multidimensional abnormal feature vector, combined with the minimum disturbance term, the feature weight distribution is generated to avoid the failure of the logarithmic operation caused by the zero value; according to the feature weight distribution, the discrete degree of each feature is quantified by the information entropy formula. The larger the information entropy value, the more uniform the feature data distribution and the lower the corresponding weight; finally, the entropy weight method weight coefficient of the standardized multidimensional abnormal feature vector is generated according to the ratio of the information entropy value to the total entropy difference, combined with the health score The time decay factor defined by the trend slope feature is used to adjust the entropy weight method weight coefficient with time decay to generate the time decay entropy weight method weight coefficient; the exponential weighted moving average is applied to the time series alignment feature in the standardized multidimensional anomaly feature vector, and the weight coefficient is dynamically adjusted based on the time decay entropy weight method weight coefficient to generate the time-weighted time series alignment feature; the time decay entropy weight method weight coefficient is linearly superimposed with the time-weighted time series alignment feature, the standardized device node influence feature, and the standardized health score trend slope feature to generate the time-weighted multidimensional anomaly feature vector.

[0072] S4.3 generates a time-weighted hazard level signal based on the time-weighted multidimensional anomaly feature vector through the quantile risk classification method.

[0073] It should be noted that the time-weighted hazard level signal is generated through the quantile risk classification method: the quantile risk classification method divides the quantile threshold based on the time-weighted time series alignment feature, standardized device node influence feature, and standardized health score trend slope feature in the time-weighted multidimensional anomaly feature vector; the eigenvalue in the time-weighted multidimensional anomaly feature vector is compared with the quantile threshold, for example: the eigenvalue exceeding the 75% quantile is mapped to a high risk level signal, the eigenvalue between the 50% and 75% quantile is mapped to a medium risk level signal, and the eigenvalue below the 50% quantile is mapped to a low risk level signal; the hazard level mapping results of all device nodes are integrated to generate a time-weighted hazard level signal including the device unique identifier, hazard level label, and time-weighted eigenvalue, where the hazard level label is divided into high, medium, and low risk levels, and the time-weighted eigenvalue is composed of the weighted results of the time-weighted time series alignment feature, the standardized device node influence feature, and the standardized health score trend slope feature.

[0074] S5. Based on the standardized multi-dimensional abnormal feature vector, abnormal control instruction set and time-weighted hazard level signal, a dynamic multi-objective collaborative optimization is performed through a multi-objective particle swarm optimization algorithm and a dynamic constraint solver to generate a set of alternative recovery solutions, and the optimal recovery strategy is screened through a digital twin network.

[0075] S5.1 generates an initial set of candidate recovery solutions by performing dynamic multi-objective collaborative optimization and Pareto non-dominated sorting based on the standardized multi-dimensional anomaly feature vector, the anomaly control instruction set, and the time-sensitive weighted hazard level signal through the multi-objective particle swarm optimization algorithm;

[0076] It should be noted that the multi-objective particle swarm optimization algorithm is based on the time-weighted time series alignment features, standardized equipment node influence features, and standardized health score trend slope features in the standardized multi-dimensional anomaly feature vector, combined with the instruction type, target component identifier, operation steps, and priority parameters in the anomaly control instruction set, to define a multi-objective function of minimizing recovery time, minimizing resource consumption, and maximizing risk suppression rate, and initialize the particle swarm position and velocity; the particle position and velocity are updated through the multi-objective particle swarm optimization algorithm, and the objective function weight is dynamically adjusted in combination with the hazard level label in the time-weighted hazard level signal to generate a multi-objective optimization solution set including recovery time, resource consumption, and risk suppression rate; the multi-objective optimization solution set is Pareto non-dominated sorted to screen the non-dominated solutions on the Pareto frontier, and the operation steps and target component identifiers in the anomaly control instruction set are combined to generate an initial alternative recovery solution set including alternative solution identifiers, recovery operation steps, resource requirements, expected risk suppression rate, and priority labels.

[0077] S5.2 uses a dynamic constraint solver to perform hard screening on the initial set of candidate recovery solutions to generate a set of feasible candidate recovery solutions;

[0078] It should be noted that the dynamic constraint solver defines dynamic constraints based on the maintenance type and equipment hierarchy relationship table in the maintenance record, including the resource availability threshold generated by the quantile statistics of historical resource consumption data (such as the maximum spare parts inventory threshold), the maintenance window period dynamically set based on the production line operation cycle and maintenance type (such as the emergency repair window threshold is 4 hours), and the risk suppression rate threshold determined by the health score historical recovery effect quantile risk classification method (such as the minimum risk suppression rate lower limit is 70%); the dynamic constraint solver traverses the alternative solution identifier, recovery operation steps, resource requirements, expected risk suppression rate, and priority label in the initial alternative recovery solution set, checks whether the resource requirement is lower than the resource availability threshold, whether the recovery operation steps are within the maintenance window period, and whether the expected risk suppression rate reaches the risk suppression rate threshold; retains the alternative solutions that meet all the constraints at the same time, and generates a feasible alternative recovery solution set, whose fields are exactly the same as the initial alternative recovery solution set, including the alternative solution identifier, recovery operation steps, resource requirements, expected risk suppression rate, and priority label.

[0079] S5.3 verifies the effectiveness of the feasible alternative recovery plan set through Monte Carlo simulation and multi-index weighted scoring in the digital twin network, and generates the optimal recovery strategy execution instruction set.

[0080] It should be noted that the Monte Carlo simulation is based on the recovery operation steps, resource requirements, and expected risk suppression rate in the set of feasible alternative recovery solutions, combined with the device hierarchical structure relationship table and device operating status data in the digital twin network, to define random disturbance parameters (such as device response delay, resource scheduling fluctuation) and simulation times, and construct a recovery process simulation scenario; execute the Monte Carlo simulation to generate the recovery success rate, resource consumption deviation, and risk suppression rate deviation of each alternative solution under random disturbance, and form a Monte Carlo simulation result set; based on the Monte Carlo simulation result set, the effectiveness of the alternative solution is verified through multi-indicator weighted scoring, where the weighted scoring weight is dynamically adjusted by the priority label in the time-weighted hazard level signal, and the comprehensive score of the recovery success rate, resource consumption deviation, and risk suppression rate deviation is calculated; select the alternative solution with the highest comprehensive score, integrate the recovery operation steps, resource requirements, expected risk suppression rate, and priority label, and generate the optimal recovery strategy execution instruction set containing execution instruction identifier, target component identifier, operation step sequence, resource allocation list, and execution priority.

[0081] This embodiment also provides a smart factory management system based on MES, comprising: a data acquisition module, a health scoring module, an abnormal control instruction set generation module, a danger level signal generation module, and an optimal recovery strategy generation module;

[0082] The data acquisition module collects equipment operating status data and physical topology data in real time, aligns them in time and space through a multi-rate Kalman filter algorithm and an iterative closest point algorithm, and combines it with a dynamic graph convolutional network to generate a digital twin network.

[0083] The health scoring module calculates the health score based on the historical maintenance records of the equipment and combines it with the digital twin network to generate a multi-dimensional analysis data set containing the spatiotemporal correlation of equipment topology and health parameters;

[0084] The abnormal control instruction set generation module uses a dynamic topology analysis algorithm to couple the spatiotemporal characteristics of the spatiotemporal correlation multidimensional analysis data set and dynamically adjust the topological weight to generate a list of high-risk equipment and an abnormal control instruction set;

[0085] The hazard level signal generation module constructs a standardized multi-dimensional abnormal feature vector based on the high-risk equipment list and generates a time-weighted hazard level signal through the entropy weight method;

[0086] The optimal recovery strategy generation module uses a multi-objective particle swarm optimization algorithm and a dynamic constraint solver to perform dynamic multi-objective collaborative optimization based on standardized multi-dimensional abnormal feature vectors, abnormal control instruction sets, and timely weighted hazard level signals to generate a set of alternative recovery plans and screen the optimal recovery strategy through a digital twin network.

[0087] This embodiment also provides a computer device suitable for the MES-based smart factory management method, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the MES-based smart factory management method proposed in the above embodiment.

[0088] The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad on the computer device housing, or an external keyboard, touchpad, or mouse.

[0089] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the MES-based smart factory management method proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0090] In summary, the present invention achieves spatiotemporal alignment of equipment operating status data and physical topology data through: multi-rate Kalman filtering algorithm and iterative nearest point algorithm, combines dynamic graph convolutional network to build a high-fidelity digital twin network to support accurate anomaly detection and health assessment; generates a set of alternative recovery solutions through multi-objective particle swarm optimization algorithm, and uses digital twin network to verify the virtual-reality collaborative robustness of the strategy.

[0091] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A smart factory management method based on MES, characterized by: include, Real-time collection of equipment operating status data and physical topology data. The multi-rate Kalman filter algorithm and iterative closest point algorithm are used to align the equipment operating status data and physical topology data in time and space, and a dynamic graph convolutional network is combined to generate a digital twin network. Calculate health scores based on historical equipment maintenance records and combine them with digital twin networks to generate a spatiotemporal multidimensional analysis dataset containing equipment topology and health parameters. The dynamic topology analysis algorithm is used to couple spatiotemporal features and dynamically adjust topological weights on the spatiotemporal correlation multidimensional analysis data set to generate a list of high-risk equipment and an abnormal control instruction set. Based on the high-risk equipment list, a standardized multi-dimensional abnormal feature vector is constructed, and a time-weighted hazard level signal is generated using the entropy weight method. Based on the standardized multi-dimensional abnormal feature vector, abnormal control instruction set and timely weighted hazard level signal, dynamic multi-objective collaborative optimization is performed through the multi-objective particle swarm optimization algorithm and the dynamic constraint solver to generate a set of alternative recovery plans, and the optimal recovery strategy is screened through the digital twin network.

2. The MES-based smart factory management method according to claim 1, characterized in that: The equipment operation status data includes sensor timing data, control system status data and production operation data; The physical topology data includes device physical location information, device connection relationships, communication network topology data, and device hierarchical structure relationships.

3. The MES-based smart factory management method according to claim 2, characterized in that: The multi-rate Kalman filter algorithm and the iterative closest point algorithm are used to align the equipment operation status data and physical topology data in time and space, and a dynamic graph convolutional network is combined to generate a digital twin network. The specific steps are as follows: The device operating status data and physical topology data are time-synchronized using a multi-rate Kalman filter algorithm, and spatially aligned using an iterative closest point algorithm to generate a joint operating status-topology dataset. Based on the running status-topology joint dataset, a dynamic graph convolutional network is used to perform node attribute neighborhood aggregation and dynamic optimization of edge weights to generate a digital twin network.

4. The MES-based smart factory management method according to claim 3, characterized in that: The health score is calculated based on the historical maintenance records of the equipment, and the digital twin network is combined to generate a spatiotemporal correlation multidimensional analysis data set containing the equipment topology and health parameters. The specific steps are as follows: Based on the equipment's historical maintenance records and equipment operating status data, the equipment's health score is calculated using a hidden Markov algorithm. Based on the health score and combined with the digital twin network, the node attributes are integrated through the graph embedding algorithm to generate a digital twin network with health. Based on the digital twin network with health, the equipment health parameters and topology structure time series characteristics are extracted through the spatiotemporal sliding window. Combined with GraphSAGE neighborhood aggregation and Pearson correlation coefficient verification, a spatiotemporal correlation multidimensional analysis dataset is generated.

5. The MES-based smart factory management method according to claim 4, characterized in that: The dynamic topology analysis algorithm is used to couple the spatiotemporal characteristics and dynamically adjust the topological weight of the spatiotemporal correlation multidimensional analysis data set to generate a high-risk equipment list and an abnormal control instruction set. The specific steps are as follows: Based on the spatiotemporal correlation multidimensional analysis data set, the spatiotemporal topological features are fused through the dynamic topological analysis algorithm to generate a spatiotemporal coupling feature matrix with dynamic weights; Based on the spatiotemporal coupling characteristic matrix with dynamic weights, multi-source risk propagation simulation is performed through anomaly diffusion rules to generate a list of high-risk equipment; Based on the high-risk equipment list, the historical maintenance operation steps are matched by the maintenance type and component identifier in the maintenance record, and a dynamic graph convolutional network is combined to generate an abnormal control instruction set containing specific operation instructions.

6. The MES-based smart factory management method according to claim 5, characterized in that: Based on the high-risk equipment list, a standardized multi-dimensional abnormal feature vector is constructed, and a time-weighted hazard level signal is generated by the entropy weight method. The specific steps are as follows: Based on the high-risk equipment list, multi-dimensional abnormal features are extracted using the PageRank algorithm and the dynamic time warping algorithm, and a standardized multi-dimensional abnormal feature vector is generated through Z-Score normalization. Based on the standardized multidimensional anomaly feature vector, a time-weighted multidimensional anomaly feature vector is generated through the time-decay entropy weight method and exponentially weighted moving average; Based on the time-weighted multidimensional anomaly feature vector, a time-weighted hazard level signal is generated through the quantile risk classification method.

7. The MES-based smart factory management method according to claim 6, characterized in that: The method uses a multi-objective particle swarm optimization algorithm and a dynamic constraint solver to perform dynamic multi-objective collaborative optimization based on standardized multi-dimensional abnormal feature vectors, abnormal control instruction sets, and timely weighted danger level signals to generate a set of alternative recovery solutions, and screen the optimal recovery strategy through a digital twin network. The specific steps are as follows: Based on the standardized multi-dimensional anomaly feature vector, the anomaly control instruction set and the time-weighted danger level signal, a dynamic multi-objective collaborative optimization and Pareto non-dominated sorting are performed through the multi-objective particle swarm optimization algorithm to generate an initial set of alternative recovery solutions. Using a dynamic constraint solver to perform hard screening on the initial set of candidate recovery solutions to generate a set of feasible candidate recovery solutions; In the digital twin network, the effectiveness of the feasible alternative recovery plan set is verified through Monte Carlo simulation and multi-indicator weighted scoring, and the optimal recovery strategy execution instruction set is generated.

8. A smart factory management system based on MES, based on the smart factory management method based on MES according to any one of claims 1 to 7, characterized in that: Including data acquisition module, health scoring module, abnormal control instruction set generation module, danger level signal generation module and optimal recovery strategy generation module; The data acquisition module collects equipment operating status data and physical topology data in real time, aligns them in time and space through a multi-rate Kalman filter algorithm and an iterative closest point algorithm, and combines it with a dynamic graph convolutional network to generate a digital twin network. The health scoring module calculates the health score based on the historical maintenance records of the equipment and combines it with the digital twin network to generate a multi-dimensional analysis data set containing the spatiotemporal correlation of equipment topology and health parameters; The abnormal control instruction set generation module uses a dynamic topology analysis algorithm to couple the spatiotemporal characteristics of the spatiotemporal correlation multidimensional analysis data set and dynamically adjust the topological weight to generate a list of high-risk equipment and an abnormal control instruction set; The hazard level signal generation module constructs a standardized multi-dimensional abnormal feature vector based on the high-risk equipment list and generates a time-weighted hazard level signal through the entropy weight method; The optimal recovery strategy generation module uses a multi-objective particle swarm optimization algorithm and a dynamic constraint solver to perform dynamic multi-objective collaborative optimization based on standardized multi-dimensional abnormal feature vectors, abnormal control instruction sets, and timely weighted hazard level signals to generate a set of alternative recovery plans and screen the optimal recovery strategy through a digital twin network.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the MES-based smart factory management method described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the MES-based smart factory management method according to any one of claims 1 to 7 are implemented.

Images