Abnormal risk prediction method and device based on assembly knowledge graph, equipment and medium

Abstract

The invention discloses an abnormal risk prediction method and device based on an assembly knowledge graph, equipment and a medium, and relates to the technical field of data processing. The method comprises the steps of obtaining multi-source heterogeneous data, performing association and alignment through a unified primary key set, and performing segmentation to generate a time slice data set; and constructing an assembly domain ontology and an assembly knowledge graph mode layer, instantiating the assembly domain ontology by using the time slice data set, and constructing an assembly knowledge graph snapshot sequence. And based on the assembly knowledge graph snapshot sequence, modeling and updating the influence relationship between variables in the assembly process to obtain an assembly process influence relationship structure. And in combination with the assembly knowledge graph snapshot sequence and the assembly process influence relation structure, coding is carried out, the state representation of the next time window is predicted based on the embedding representation of the historical time window, and an abnormal risk score is calculated. When the risk exceeds a threshold value, influence path tracking is carried out, controllable decision variables are screened, candidate intervention schemes are generated, the effects of the candidate intervention schemes are evaluated, and decision suggestions are output.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and more specifically, to an anomaly risk prediction method, apparatus, equipment, and medium based on assembly knowledge graph. Background Technology

[0002] As a representative scenario of discrete manufacturing, the assembly process of industrial robot joints involves multiple steps such as press-fitting and pre-tightening of various parts including reducers and servo motors, and requires the integration of heterogeneous data from multiple sources such as MES, SCADA / PLC. Due to the influence of dynamic operating conditions such as equipment wear and order fluctuations, the assembly process is prone to quality problems such as torque exceeding limits and waveform abnormalities. Therefore, there is an urgent need for a technical solution that can uniformly model the assembly structure, process constraints, and abnormal evolution paths, and realize pre-risk warning and interpretable decision support.

[0003] Currently, various technologies exist in related fields attempting to address these needs. Threshold-based online monitoring methods achieve simple alarms by setting upper and lower limits for parameters, but struggle to capture multivariate coupling relationships. Machine learning or deep learning-based methods utilize historical data to build predictive models, but lack a structured expression of process mechanisms. Knowledge graph-based technologies focus on manufacturing knowledge management but are not deeply coupled with online anomaly detection. Meanwhile, intelligent methods oriented towards scheduling primarily focus on production efficiency, paying insufficient attention to micro-anomalies.

[0004] These existing methods all have significant shortcomings. First, they fail to integrate assembly knowledge with multi-source data within a unified framework, resulting in a disconnect between anomaly detection and decision support. Second, insufficient modeling of anomaly propagation paths and influencing relationships makes the results difficult to interpret and translate into actionable interventions. Furthermore, the lack of a closed-loop capability from risk warning to causal decision-making makes reliance on human experience inefficient. Finally, the lack of fine-grained application solutions for industrial robot joint assembly lines limits their applicability in practical deployment. Summary of the Invention

[0005] This invention provides an anomaly risk prediction method, apparatus, equipment, and medium based on assembly knowledge graph, to improve at least one of the above-mentioned technical problems.

[0006] In a first aspect, the present invention provides an anomaly risk prediction method based on assembly knowledge graph, which includes S1 to S5.

[0007] S1. Acquire multi-source heterogeneous data from the industrial site, associate and align the multi-source heterogeneous data through a unified primary key set, and divide it according to a fixed-length time window to generate a time slice dataset.

[0008] S2. Construct an assembly domain ontology and an assembly knowledge graph pattern layer. Instantiate the assembly domain ontology using the time slice dataset and construct an assembly knowledge graph snapshot sequence that scrolls with the time window.

[0009] S3. Based on the snapshot sequence of the assembly knowledge graph, model and update the influence relationship between variables in the assembly process to obtain the assembly process influence relationship structure that includes the quantified influence intensity.

[0010] S4. Combining the assembly knowledge graph snapshot sequence with the assembly process influence relationship structure, the monitoring node is encoded with graph structure features, and the state representation of the next time window is predicted based on the embedding representation of the historical time window. An abnormal risk score is calculated to identify abnormal risks.

[0011] S5. When an abnormal risk is identified that exceeds the threshold, the influence path is traced based on the influence relationship structure of the assembly process, controllable decision variables are screened, candidate intervention schemes are generated and their effects are evaluated, and decision suggestions are output.

[0012] Secondly, the present invention provides an anomaly risk prediction device based on assembly knowledge graph, which includes a segmentation module, an instantiation module, a modeling module, an identification module, and a scheme module.

[0013] The segmentation module is used to acquire multi-source heterogeneous data from industrial sites, associate and align the multi-source heterogeneous data through a unified primary key set, and segment the data according to a fixed-length time window to generate a time-slice dataset.

[0014] The instantiation module is used to construct the assembly domain ontology and the assembly knowledge graph pattern layer. It uses the time slice dataset to instantiate the assembly domain ontology and constructs a snapshot sequence of the assembly knowledge graph that scrolls with the time window.

[0015] The modeling module is used to model and update the influence relationships between variables in the assembly process based on the snapshot sequence of the assembly knowledge graph, and obtain the influence relationship structure of the assembly process including the quantified influence intensity.

[0016] The identification module is used to combine the assembly knowledge graph snapshot sequence with the assembly process influence relationship structure to encode the graph structure features of the monitoring nodes, predict the state representation of the next time window based on the embedding representation of the historical time window, and calculate the abnormal risk score to identify abnormal risks.

[0017] The solution module is used to trace the impact path based on the impact relationship structure of the assembly process when an abnormal risk is identified to exceed a threshold, screen controllable decision variables, generate candidate intervention solutions and evaluate their effects, and output decision recommendations.

[0018] Thirdly, the present invention provides an anomaly risk prediction device based on an assembly knowledge graph, comprising a processor, a memory, and a computer program stored in the memory. The computer program can be executed by the processor to implement an anomaly risk prediction method based on an assembly knowledge graph as described in any paragraph of the first aspect.

[0019] Fourthly, the present invention provides a computer-readable storage medium. The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device containing the computer-readable storage medium to perform an anomaly risk prediction method based on assembly knowledge graphs as described in any paragraph of the first aspect.

[0020] By adopting the above technical solution, the present invention can achieve the following technical effects: This invention integrates multi-source assembly data within a unified knowledge graph framework, enabling real-time monitoring and dynamic modeling of the industrial assembly process. It can not only proactively predict and warn of abnormal risks in future time windows, but also automatically trace the propagation path of anomalies, screen controllable variables, and generate intervention plans that balance quality, efficiency, and cost, thus forming an intelligent closed loop of "prediction-source tracing-decision-feedback". This significantly improves the reliability, interpretability, and decision-response efficiency of the assembly process. Attached Figure Description

[0021] Figure 1 : Model framework diagram of an anomaly risk prediction method based on assembly knowledge graph.

[0022] Figure 2 Flowchart of causal decision-making and intervention deduction. Detailed Implementation

[0023] Example 1, please refer to Figures 1 to 2 The first embodiment of the present invention provides an anomaly risk prediction method based on an assembly knowledge graph, which can be executed by an anomaly risk prediction device based on an assembly knowledge graph (hereinafter referred to as: prediction device). In particular, it is executed by one or more processors in the prediction device to implement S1 to S5.

[0024] S1. Acquire multi-source heterogeneous data from the industrial site, associate and align the multi-source heterogeneous data through a unified primary key set, and divide it according to a fixed-length time window to generate a time slice dataset.

[0025] The methods for industrial field data acquisition and integration can be summarized in the following steps: S11. Periodically or in real-time, incremental data is retrieved from the Production Execution System (MES), Supervisory Control and Data Acquisition (SCADA) system / PLC system, quality inspection system, equipment maintenance system, and document management system to form data sets.

[0026] In industrial robot joint assembly lines, this invention first uses an industrial field data acquisition and integration module to acquire assembly process-related data from a multi-source heterogeneous system, providing a unified data foundation for subsequent assembly knowledge graph construction, graph structure enhancement anomaly risk prediction, and influence relationship network modeling.

[0027] Specific data sources include, but are not limited to: MES system data includes order information (order number, product model, delivery date, etc.), process routes and operation lists, operation start / end time, operation responsible person, workstation number, work-in-process status, etc., which are used to depict the execution logic of "order - joint - operation - workstation".

[0028] Data from SCADA / PLC and field data acquisition and monitoring systems includes force-displacement curves and displacement-time curves acquired by servo presses, torque-angle curves, speed, and acceleration acquired by electric screwdrivers / servo drives, station cycle counts, robot current, torque, and position, and equipment status (running, standby, fault), etc., used to characterize the dynamic changes in process variables and equipment operating status of key assembly processes.

[0029] Quality inspection system data includes online functional test results (such as rotational resistance, clearance, noise levels, etc.), offline laboratory test results (such as life tests, vibration analysis), sampling records and their judgment labels (OK / NG, abnormality type, etc.), which are used to characterize the explicit manifestation of assembly quality results and abnormal events.

[0030] Equipment maintenance and alarm record data includes fault codes, alarm times, handling measures, spare parts replacement records, maintenance personnel, downtime, etc., which are used to describe the health status of the equipment and maintenance behavior, and provide a basis for subsequent "equipment status - abnormal event" path modeling in the influence relationship network.

[0031] Process and knowledge documentation data includes assembly work instructions, control plans, failure mode and effects analysis (FMEA), empirical adjustment rules, and typical anomaly case analysis reports. These are used to supplement explicit process knowledge, empirical rules, typical failure modes and their handling strategies, and provide a foundation for the schema and rule layers of the assembly knowledge graph.

[0032] To achieve unified modeling of the aforementioned multi-source data, this invention performs format conversion, time alignment, and primary key matching on the server side. Let the data sets from different data sources be referred to as MES data. SCADA / PLC data Quality inspection data Equipment maintenance data Documents and knowledge data .

[0033] The complete set of raw industrial site data for.

[0034] .

[0035] S12. Standardize the format and map fields of each data source, including unit unification, encoding mapping, and handling of outliers and missing values, to ensure that data from different systems can be based on a unified primary key set. Establish a connection.

[0036] S13. Define a unified set of primary keys. ,use Primary key matching and association are performed on multi-source data to generate a unified intermediate table with "order - joint - assembly process - workstation / equipment" as the main line. .

[0037] .

[0038] In the formula This indicates the order number. It indicates a joint or product identifier. Indicates process identification. Indicates workstation / equipment identification. Indicates the component serial number. Indicates the event timestamp.

[0039] To support subsequent entity-data assembly and knowledge graph instantiation, this embodiment defines a unified set of primary keys. Through primary key set It performs primary key matching and association on multi-source data to generate an intermediate unified data table.

[0040] .

[0041] In the formula For record The corresponding set of attribute fields, such as process parameters, quality results, and alarm information. The number of records; For record indexing.

[0042] S14. Timestamp Perform uniform alignment and resampling, dividing the continuous time axis into fixed lengths. Time window, in each window Internal aggregation of corresponding records generates a time-slice dataset. This is for subsequent graph snapshot construction. It provides input for time series modeling.

[0043] To facilitate subsequent anomaly risk prediction and time windowing modeling for graph structure enhancement, this invention uses a time axis of fixed length. Perform segmentation and construct discrete-time windows. In each time window Extract the corresponding intermediate data subset. And based on this, a time snapshot of the assembly knowledge graph is constructed as input.

[0044] S15. Through the data interface service, the unified intermediate table and the time slice datasets of each time window are mapped to entity nodes, attribute nodes and relation edges in the assembly knowledge graph, and written into the graph database.

[0045] Specifically, through the data interface service, the intermediate table and subsets of each time window The mapping is used to assemble entity nodes, attribute nodes, and relation edges in the knowledge graph, and written into the backend graph database to provide a unified data foundation for subsequent "assembly knowledge graph construction - graph structure enhancement anomaly risk prediction - influence relationship network modeling - causal decision-making and intervention evaluation".

[0046] Through the above steps, this invention integrates multi-source heterogeneous data that were originally scattered in MES, SCADA / PLC, quality inspection and maintenance systems under a unified key value and time axis, laying an achievable and scalable data integration foundation for future anomaly risk prediction and causal path analysis on graph structures.

[0047] S2. Construct an assembly domain ontology and an assembly knowledge graph pattern layer. Instantiate the assembly domain ontology using the time-slice dataset to construct a snapshot sequence of the assembly knowledge graph that scrolls with the time window. Preferably, S2 includes S21 to S26.

[0048] Based on the aforementioned industrial field data collection and integration results, this invention focuses on the industrial robot joint assembly business, constructing an assembly domain ontology oriented towards "graph structure-enhanced anomaly risk prediction + influence relationship network-driven causal decision-making". Building upon the traditional "equipment-process-part-quality" modeling, the ontology layer reserves information such as anomaly event representation, influence path characterization, and decision variable constraints in terms of concepts and relationships, enabling subsequent graph structure modeling, graph neural network prediction, and causal decision-making to work collaboratively within a unified semantic space.

[0049] S21. Define the entity types of the assembly domain ontology, including orders, products, parts, processes, workstations, equipment, process parameters, quality indicators, abnormal events, maintenance activities and personnel, and define the monitoring unit entity as the dynamic carrier of the key process execution process.

[0050] Specifically, this invention defines the following core entity types in the assembly domain ontology: Order, Joint, Part, Process, Station, Equipment, Parameter, Quality Index, Anomaly, Maintenance, Operator, etc.

[0051] S22. Define the relationship types between design entities, including compositional relationships and process sequence relationships representing static structures, as well as event occurrence relationships and status indication relationships representing dynamic evolution. Reserve weight attributes for influence-type relationships and controllability and cost attributes for process parameters and maintenance activity entities.

[0052] The present invention defines the following main relation types in the ontology for explicitly expressing assembly structure, process constraints, anomaly evolution and intervention paths.

[0053] BelongsTo: A component belongs to a certain joint or product.

[0054] consistsOf (composed of): A joint is composed of multiple parts.

[0055] executedAt (executed at): The process is executed at a certain workstation / equipment.

[0056] hasParameter: Process parameters associated with a process, such as the press-fit target force, holding time, and tightening torque.

[0057] leadsTo: The influence relationship between process or parameter settings and quality indicators.

[0058] Indicators: The relationship between quality indicators and abnormal events, such as "noise level exceeding the limit" indicating "abnormal fit of rotating parts".

[0059] occursOn: An abnormal event occurs on a specific order / part / workstation.

[0060] handledBy (handled by): An exception event is handled by a maintenance activity.

[0061] "Follows / Precedes" refers to the sequential relationship between processes.

[0062] compatibleWith / conflictsWith (compatible with / conflict with): Process constraints and resource compatibility relationships.

[0063] To better support "anomaly risk prediction enhanced by graph structure" and "causal decision-making driven by influence relationship network", this invention further introduces the following design points in the relationship layer.

[0064] Design Point 1 (Relationship Weights and Directional Placement): For influence-type relationships such as leadsTo and indicates, their causal direction is marked at the ontology layer, and attribute fields such as "influence strength and confidence level" are reserved. Subsequently, in the "Assembly Process Influence Relationship Modeling and Update" module, these attributes are quantified and updated in a data-driven manner, so that the directed edges in the knowledge graph have both semantic meaning and can carry numerical weights.

[0065] Design Point 2 (Decision-Related Attribute Labeling): Add attributes such as "controllable", "adjustment cost", and "safety level" to entities such as Parameter and Maintenance and their relationships. These attributes are used to filter and prioritize different intervention variables during the causal decision generation stage, achieving automatic narrowing from "all possible causes" to "executable intervention variables".

[0066] Design Point 3 (Anomaly Evolution and Path Pattern Templates): Abstract common "anomaly evolution templates" from the relationships related to Anomaly, such as "parameter drift → gradual change in quality indicators → minor anomaly → severe anomaly shutdown". Encode these templates through relationships such as indicates, leadsTo, and occurOn, so that the subsequent graph structure-enhanced anomaly risk prediction module can combine these templates to make more reasonable risk amplification and path inference when early signs are identified.

[0067] S23. Generate the schema layer of the assembly knowledge graph in the graph database, and define the node schema, edge schema, and rule set including process sequence constraints, parameter value constraints, and equipment compatibility constraints.

[0068] Generating the assembly knowledge graph schema layer: Through analysis of process documents, FMEA, expert interviews, and historical data, the entity and relationship types in the aforementioned ontology can be gradually improved, and corresponding attributes and constraint rules can be added. Based on this, this invention defines corresponding node and edge patterns in a graph database to form the schema layer of the assembly knowledge graph.

[0069] Node pattern: Defines a unified set of identifiers, main attributes, and time / batch related extended attributes for entities such as Order, Joint, Process, Station, Equipment, Parameter, Quality Index, Anomaly, Maintenance, Operator, and Monitoring Unit.

[0070] Edge pattern: Defines endpoint types, directionality, and constraints for relationships such as belongsTo, consistOf, executedAt, hasParameter, leadsTo, indicates, occursOn, handledBy, follows / precedes, compatibleWith / conflictsWith, etc.

[0071] The set of rules and constraints explicitly records process sequence constraints, parameter value constraints, equipment compatibility constraints, and safety protection constraints at the schema layer, providing a consistency verification basis for subsequent candidate intervention scheme generation and data filtering during graph neural network training.

[0072] Through the above-mentioned assembly knowledge ontology modeling and knowledge graph pattern layer construction, this invention pre-embeds structural and semantic information highly related to anomaly detection and causal decision-making at the knowledge level, enabling the subsequent modules of "entity-data assembly knowledge graph instantiation and time slicing", "assembly process influence relationship modeling and updating" and "graph-enhanced anomaly risk prediction and causal decision generation" to work collaboratively on the same graph structure, forming an overall technical solution that is both innovative and engineering-feasible.

[0073] S24, Based on a unified primary key set The time-slice dataset is instantiated into entities, and specific business object nodes and monitoring unit nodes are created in the graph database. The relationships between nodes are established based on the schema layer.

[0074] After completing the ontology and schema design in the assembly domain, this invention needs to map and instantiate actual production data into the knowledge graph to form an integrated assembly knowledge graph of "entity + data" that evolves over time, providing a unified input for subsequent assembly process influence relationship modeling and graph structure enhancement anomaly risk prediction.

[0075] First, based on a unified primary key set This involves instantiating key business objects in the production process. For each order, each joint to be assembled, each batch of parts, each process execution, each piece of equipment, and each quality inspection, a corresponding entity node is created in the graph database and bound with a unique identifier (such as order number, part number, process number, equipment number, inspection number, timestamp, etc.), which is consistent with the identifier defined in Section 2. , , , , There is a one-to-one correspondence between entity types such as Equipment, QualityIndex, Anomaly, Maintenance, and Operator.

[0076] Based on this, the present invention can also introduce a monitoring unit or similar intermediate entity according to the combination of "process + workstation + equipment + joint / order", and regard each key process execution as an independent monitoring node, which is used to carry the process variables, quality results and abnormal risk scores of the process within a specific time window, so as to facilitate subsequent time series modeling and anomaly detection based on uniform granularity of graph neural network.

[0077] S25. Perform data attribute attachment and feature extraction, attach the status information of the production execution system to the corresponding node, extract key inflection points, peak values, mean values ​​and variance statistics from the continuous waveform data collected by the data acquisition and monitoring system and encode them into process feature vectors, encode the quality inspection results and judgment labels into quality feature vectors, and attach the above feature vectors to the corresponding monitoring unit nodes.

[0078] Secondly, the numerical and status data from MES, SCADA / PLC, quality inspection, and maintenance records are linked to the aforementioned entity nodes to form an integrated "entity-data" structure. Specifically: Start / end times, process status, and responsible personnel information from the MES are associated with the corresponding order, process, workstation, and personnel nodes. Process data collected from SCADA / PLC, such as force-displacement curves, torque-angle curves, speed, acceleration, workstation cycle time, and equipment status, are extracted through feature extraction to obtain statistical quantities such as key inflection points, peak values, areas, means, and variances, and encoded as process feature vectors. Functional test results (rotational resistance, clearance, noise levels, etc.), offline test results, and judgment labels (OK / NG, abnormality type, etc.) output from the quality inspection system are encoded as quality feature vectors. Alarm codes, fault types, maintenance measures, and downtime information from equipment maintenance and alarm records are encoded as maintenance and health status features.

[0079] For continuous time series data, this invention employs a sliding time window or fixed time slice method, applying a sliding time window to each window. The signals within the window are statistically aggregated or embedded, for example, by extracting the mean, extrema, or frequency domain features within the window, or by obtaining low-dimensional embedding vectors through autoencoders / sequence encoders. Finally, these feature codes are uniformly represented as node-based... and time window Attribute vector for index And it is attached to the corresponding entity nodes (especially monitoring unit nodes) through relationship types such as "observedAtTime", "hasValue", and "hasStatus".

[0080] In matrix form, window The set of attribute vectors for all nodes within the node is as follows: This will subsequently be used as an input feature for graph structure encoding and time series prediction models.

[0081] S26. According to a fixed length Divide the timeline into segments, within each time window. Within, extract the set of active entity nodes. Relation sets and attribute vector set Construct an assembly knowledge graph snapshot corresponding to this time window. This creates a sequence of snapshots that scrolls over time.

[0082] To depict the dynamic evolution of the assembly process under different working conditions, this invention uses a time axis of fixed length. Divide the time into segments (e.g., 5 minutes or several assembly cycles) to construct discrete time windows. In each time window Internally, based on the intermediate data table Based on the entity instantiation results, extract the active entity nodes within the window and their relationships from the graph database to construct an assembly knowledge graph snapshot: In the formula For window The set of entities involved. This is the set of relationships between the aforementioned entities. It is a collection of attribute vectors.

[0083] It includes nodes related to the current production activity, such as Order, Joint, Part, Process, Station, Equipment, QualityIndex, Anomaly, Maintenance, and MonitoringUnit. It includes both static structural relationships such as belongsTo, consistOf, executedAt, hasParameter, and follows / precedes, as well as time-varying relationships related to events and states such as occurOn and handledBy.

[0084] This snapshot On the one hand, it inherits the prior information at the structural level of the assembly knowledge graph (assembly hierarchy, process sequence, resource constraints, anomaly evolution templates, etc.), and on the other hand, it reflects the real-time / near real-time observation data within the current time window. The subsequent assembly process influence relationship modeling module... Using this as input, the module estimates the strength of the influence of controllable variables on quality and anomalous events. The graph-enhanced anomalous risk prediction module then... The sequence is used to perform time-series prediction based on graph structure perception, and to calculate the future anomaly risk score of each monitoring node, thus realizing the unified modeling and prediction foundation of the "knowledge + data dual-driven" approach of this invention.

[0085] S3. Based on the snapshot sequence of the assembly knowledge graph, model and update the influence relationship between variables in the assembly process to obtain the assembly process influence relationship structure that includes the quantified influence intensity.

[0086] After completing the instantiation and time slicing of the entity-data assembly knowledge graph, this invention provides snapshots of the assembly knowledge graph based on each time window. The influence relationships between key variables in the assembly process are modeled and updated to obtain the influence relationship structure of the assembly process that evolves with changing operating conditions. This provides a structural foundation for subsequent graph-enhanced anomaly risk prediction and causal decision-making.

[0087] First, candidate variables are selected, and then samples are constructed. Within each time window... This invention, based on entity types, relation types, and process constraints in the assembly knowledge graph, in the snapshot of the assembly knowledge graph Select a set of candidate independent variables and a set of target variables .

[0088] Set of independent variables This mainly includes controllable process parameters (such as press-fit target force, holding time, tightening torque, cycle buffer time, etc.), equipment status parameters (such as equipment running / standby / fault status, current, torque, temperature, vibration index, etc.), environmental variables, and important maintenance activity markers.

[0089] target variable set This mainly includes key quality indicators (such as rotational resistance, clearance, noise level, etc.), cycle time indicators (such as workstation cycle time, work-in-process dwell time, etc.), and abnormal event indicator variables (such as abnormal pressing waveform markers, excessive noise markers, buffer blockage markers, etc.).

[0090] To ensure the interpretability and physical plausibility of the influence relationship modeling, this invention only considers influence modeling between variable pairs that already have potential physical / process relationships in the assembly knowledge graph. Specifically, for a certain target variable... Its candidate independent variable set by and Entities and their attributes derived from relationships such as leadsTo, hasParameter, indicates, and occurOn are automatically excluded from variable combinations that are clearly unrelated to the process mechanism.

[0091] After the set of variables is determined, from the snapshot Extract the corresponding variable in the window The observed values ​​within the range constitute the sample data used for modeling and subsequent estimation of influence relationships.

[0092] Then, the sparse influence relationships are estimated (simplifying the linear model). For each target variable... During the time window A simplified linear influence model is used to estimate sparse influence relationships. A regression algorithm with sparse constraints is used to automatically learn the coefficient vector from the sample data, and coefficients that are non-zero or whose absolute value exceeds a set threshold are considered evidence of significant influence.

[0093] .

[0094] In the formula, This is the set of candidate independent variables obtained by filtering from the knowledge graph relationships. The influence coefficient to be estimated is reflected in the window. Internal variables For target variable The direction and intensity of the influence. This is the residual term.

[0095] To eliminate the impact of differences in the dimensions and scales of different variables on the comparability of coefficients, this invention performs [further steps] before regression estimation. and Standardize / normalize the data (e.g., zero mean, unit variance) to make it It can be used as a dimensionless influence intensity for subsequent path contribution calculations and cross-variable comparisons.

[0096] Impact coefficient to be estimated Instead of being manually specified, these variables are automatically learned from the sample data through a regression algorithm with sparse constraints (such as a regression method using L1 regularization). This type of algorithm has the characteristic of "automatically selecting variables": for important independent variables, it yields larger non-zero coefficients. For unimportant or redundant independent variables, their coefficients are compressed to near zero or even strictly zero.

[0097] Therefore, after the solution is completed, this invention directly assigns coefficients that are non-zero or whose absolute value exceeds a set threshold. Considered "in window" Inside right Evidence that "there is a significant impact".

[0098] Finally, structural constraints and time evolution updates are performed.

[0099] Based on the coefficient vector, weighted directed edges are added or updated in the assembly knowledge graph to obtain the assembly process influence relationship structure under this time window. Furthermore, the influence relationships are modified by combining structural prior constraints and process common sense constraints.

[0100] Specifically, based on these coefficients, in the assembly knowledge graph from point to Add or update a weighted directed edge, where the edge weight can be taken as... Or its normalized value. Therefore, in each time window... A set of influence relation edge sets is obtained , with structural relation edge set They complement each other, depicting the quantitative impact relationship of "parameters / state → quality / abnormality" under the current working conditions.

[0101] To avoid pseudo-associations inconsistent with the assembly mechanism, this invention introduces structural prior constraints and process common sense constraints in the above-mentioned influence relationship modeling process. Structural prior constraints: Only variable pairs with potential physical relationships in the knowledge graph are allowed to generate influence edges. That is, only when existing ontology relations (such as leadsTo, hasParameter, indicates, occurOn, etc.) are influence edges allowed. and Only when there is a reasonable path between them will... Included It participates in coefficient estimation, while other variables are directly excluded from modeling, reducing the risk of spurious associations from the source. Process common sense constraints: For influence directions and intensities that are clearly inconsistent with process common sense (for example, given that "excessive pressing force usually increases the risk of excessive noise," but a coefficient result of "the greater the pressing force, the lower the noise") appears, this invention can suppress or correct them through threshold rules, symbolic constraints, or manual review, making the final influence relationship closer to the actual assembly mechanism.

[0102] Temporal evolution and smooth updates: As the time window rolls, it affects the relational structure. It will be continuously updated. This invention can smooth the coefficients estimated in a continuous window (e.g., by moving average or decaying weights), ensuring that the model is sensitive to new operating conditions while avoiding frequent and significant changes in the influence relationship due to short-term random fluctuations.

[0103] Through the above process, the assembly process influence relationship structure obtained by the present invention at each time window is... This approach leverages both structural and technological priors from assembly knowledge graphs and automatically identifies key influencing variables through data-driven sparse learning. Further development could include... This information is converted into weighted adjacency information in graph neural networks, enhancing the proactive anomaly detection capability of graph structure perception. On the other hand, in the "Anomaly-Oriented Impact Path Analysis and Causal Decision Generation Method," these weighted directed edges can be used to trace controllable variables backward from high-risk nodes, generating and evaluating different intervention schemes, thereby forming a causal style closed loop of "early warning—explanation—disposal recommendation."

[0104] S4. Combining the assembly knowledge graph snapshot sequence with the assembly process influence relationship structure, the monitoring nodes are encoded with graph structure features, and the state representation of the next time window is predicted based on the embedding representation of the historical time window. An anomaly risk score is calculated to identify anomaly risks. Preferably, S4 includes S41 to S46.

[0105] The graph-enhanced anomaly risk prediction module of this invention uses a sequence of assembled knowledge graph snapshots for each time window. and the assembly process influence relationship structure obtained in step S3 Based on this, multi-source assembly process data is uniformly mapped into a temporal feature representation with graph structure constraints. By predicting the deviation between the future state and the actual state, abnormal risk prediction and early warning for key processes / stations / orders are achieved, providing high-risk objects and their contextual information for subsequent causal decision-making modules.

[0106] S41, Combining the influence relationship structure of the assembly process Construct an adjacency matrix containing process structure and influence weights from the structural edge set in the assembly knowledge graph snapshot. And then normalize it.

[0107] In each time window Snapshot of assembly knowledge graph Perform graph structure feature encoding. Record the window. The set of internal nodes is The attribute vector set is It is a real number. This represents the node feature dimension / embedding dimension. Each row in the attribute vector set corresponds to the feature vector of a node (e.g., monitoring unit, equipment, process, etc.).

[0108] Combining the assembly process influence relationship structure obtained in step S3 and relation set This invention constructs an adjacency matrix that includes process structure and influence weights. And further normalized to obtain In this context, the adjacent elements corresponding to structural edges are set to 1 (or the structural weight), while the adjacent elements corresponding to influencing edges are set to... Normalized weight values. To influence the relationship edge The weights are determined by the fact that when the same pair of nodes has both structural edges and influence edges, a weighted superposition method is used to synthesize them. .in It is a structural adjacency matrix. To affect the adjacency matrix, This is the fusion coefficient.

[0109] Then, self-loops were added in the usual way, and the degree matrix was normalized to obtain the result for graph aggregation. Its elements simultaneously reflect the structural relationships and influence strength between nodes.

[0110] S42. Using a graph feature transformation structure based on neighborhood aggregation, the node features are updated layer by layer to obtain the node representation: .

[0111] .

[0112] In the formula For the first Layer node representation. For learnable transformation matrices, It is a non-linear activation function (such as ReLU). For the first Layer node representation. This represents the nodes of the first layer.

[0113] The number of graph aggregation / graph convolutional layers is: ,go through The node representation obtained after layer aggregation is denoted as .

[0114] S43. For each node corresponding to the monitored object With its recent Using the graph structure embedding sequence of each time window as input, a temporal prediction network is used to obtain the predicted embedding for the next time window. .

[0115] For each node corresponding to the monitored object (For example, "the monitoring unit for a certain order at a certain key pressing process"), can be obtained from... The corresponding row is read in the window. Encoding vector As the time window scrolls, a graph structure embedding sequence for that node can be formed. Each vector incorporates information on assembly structure relationships and influence weights.

[0116] After obtaining the graph structure embedding sequence of each monitored object, the present invention constructs a time series prediction model for each monitored object to predict the "normal state representation" of the next time window based on the recent trend of change.

[0117] Specifically, nodes Embedding in the most recent m time windows As input, using parameters Temporal prediction network (For example, sequence models based on loop structures or encoder-decoder structures) to obtain the predicted embedding for the next time window. .

[0118] .

[0119] S44. Calculate the final risk score, which includes the estimated prior risk and the residual risk. : .

[0120] In the formula, This is a priori risk calculation based on the mean and covariance of historical normal embeddings. This refers to residual risk based on actual observations. For weight fusion.

[0121] During the model training phase, this invention selects historical normal operation data as training samples and minimizes the prediction embedding. With real embedding The mean square error between them enables the time series prediction network to learn "the typical pattern of node state evolution over time under normal operating conditions".

[0122] During the online operation phase, this invention employs a two-stage risk calculation method of "pre-warning + residual calibration": in the window At the end, it can be based on predictive embedding. Calculate and estimate prior risks For example, using the mean of historical normal embeddings Covariance is used as a reference.

[0123] .

[0124] In the formula This indicates transpose. Let be the covariance matrix. It is the inverse of the covariance matrix.

[0125] when Exceeding the threshold The system should trigger an early warning and initiate the intervention assessment process. (When the window...) After the actual data arrives, the residual risk is then calculated. In the formula It is an L2 norm.

[0126] .

[0127] The final risk score is defined (or used for confirmation). .

[0128] S45, when the final risk score is... Risk warning threshold If so, it is determined that there is an abnormal risk in the next time window.

[0129] Specifically, when Exceeding the threshold When the system detects a potential anomaly in a given process / station / order, it determines that the corresponding process / station / order has an abnormal risk. Using this method, the system can provide a priori risk assessment before the next window begins and perform online calibration after actual observation, avoiding reliance solely on post-event alarms.

[0130] S46, Node Identify the location as an anomaly and retrieve the node from the assembly knowledge graph. The attribute information and its associated abnormal event entities are parsed to obtain specific abnormal type descriptions, and early warning information including prediction time, abnormal location, abnormal type and risk score is generated.

[0131] Knowing only that "a certain process has a high risk" is not enough to support subsequent causal decisions. This invention further combines the semantic and influence relationship structure in the assembly knowledge graph to refine the identification of the type and location of anomalies.

[0132] On the one hand, by utilizing the semantic path of "quality indicator - abnormal event - process / station / part" in the knowledge graph, high-risk monitoring nodes can be associated with potentially affected quality indicators and abnormal event types. For example, when the risk score of the "encoder pressing monitoring unit" node continues to rise, and it has a close connection with abnormal event nodes such as "noise value exceeding the standard" and "abnormal fit of rotating parts" in the graph, this invention can give a judgment result similar to "the pressing force waveform of a certain order in the encoder pressing process is abnormal, which may lead to excessive noise."

[0133] On the other hand, the influence relationship edge set obtained in step S3 This invention can identify key impact paths leading to a high-risk node or its downstream quality indicator nodes. This information will serve as input for the next step, "Impact Path Analysis and Causal Decision Generation Method for Anomalies," to trace back along the impact path to controllable process parameters, equipment status, or maintenance behaviors, supporting the generation of decisions on "which parameters should be adjusted and what intervention measures should be taken."

[0134] Through the above process of "graph structure feature encoding - time series prediction - risk scoring - anomaly semantic localization", this invention realizes an active anomaly detection mechanism consistent with the assembly process mechanism on the unified carrier of assembly knowledge graph. At the same time, it provides structured high-risk object and impact path information for the subsequent causal decision module, forming an integrated technical solution of "anomaly risk prediction + causal decision support".

[0135] S5. When an abnormal risk is identified that exceeds a threshold, influence path tracing is performed based on the influence relationship structure of the assembly process, controllable decision variables are screened, candidate intervention schemes are generated and their effects are evaluated, and decision recommendations are output. Preferably, S5 includes S51 to S54.

[0136] When the graph-enhanced anomaly risk prediction module is in the time window For a specific process / workstation / order corresponding node Provide anomaly warnings (i.e., anomaly risk scores) Exceeding the threshold After that, the causal decision-making module of the present invention, based on the assembly process influence relationship structure obtained in S3, By performing impact path analysis and decision-making scheme generation in the assembly knowledge graph, a causal style closed loop is achieved from "discovering an anomaly" to "providing a feasible disposal suggestion".

[0137] S51. Using the high-risk quality indicator node or abnormal event node as the endpoint node, perform a reverse search along the weighted directed edges in the influence relationship structure of the assembly process, calculate the comprehensive contribution of the path based on the influence weight of each edge on the path, and select a set of candidate cause nodes.

[0138] First, the endpoint node is the quality indicator node or the node where a high-risk event occurs. (For example, nodes with excessive noise, nodes with abnormal pressing waveforms, etc.), the influence relationship edge set in the current or most recent time window. Then, perform a reverse search from bottom to top along the weighted directed edges.

[0139] During the search process, this invention also considers prior information such as process sequence, structural hierarchy and resource constraints in the assembly knowledge graph: only under the premise of satisfying the constraints of process sequence (follows / precedes), structural composition relationship (belongsTo / consistsOf) and resource compatibility (compatibleWith / conflictsWith), reasonable paths from nodes such as process parameters / equipment status / part batch / environmental variables to quality indicators / abnormal events are retained.

[0140] For candidate cause nodes To the destination node A directed path Let the influence weight of each edge on the path be . .because The values ​​obtained from standardized / normalized variable estimation can be considered as dimensionless influence intensity. This invention defines the comprehensive contribution of the path. It is a function of the weights of each side. To influence the relationship edge The weight.

[0141] In an optional embodiment, logarithmic summation is used (i.e.: Alternatively, the path length can be normalized to represent the overall contribution of the path, thereby reducing the ranking bias caused by the path length.

[0142] The set of all paths that satisfy the constraints In this process, the path contribution is sorted to obtain a set of candidate cause nodes. and its corresponding comprehensive contribution This is used to prioritize candidate reasons. (Set) It typically includes controllable process parameters, key equipment status, specific part batches, and sensitive environmental variables.

[0143] S52. Select variables from the candidate cause node set that can be changed through parameter adjustment, process change, or maintenance operation to form a decision variable vector. .

[0144] It should be noted that not all candidate cause nodes can be directly intervened through process or maintenance measures. Therefore, this invention pre-labels variables with attributes such as "controllable," "adjustment cost," and "safety level" in the assembly knowledge graph ontology and entity attributes. During the causal decision-making stage, variables that can be changed through parameter adjustments, process modifications, or maintenance operations are selected from the candidate cause set C to form the decision variable set. Construct a decision variable vector based on the type of decision variables. The number of decision variables.

[0145] Typical components include, but are not limited to: process parameters such as press-fit target force, press-fit speed, holding time, and tightening torque; process configurations such as cycle time buffer, whether intermediate inspection procedures are used, and sampling frequency; and operation and maintenance strategies such as whether to switch to backup equipment and whether to arrange preventive maintenance.

[0146] In the subsequent decision-making process, this invention will focus on the current settings. Based on its adjustable range, several candidate intervention programs are generated. And evaluate its impact on quality and anomaly risk under the constraints of influence relationship structure and assembly knowledge graph.

[0147] S53. Generate several candidate intervention plans. Furthermore, by utilizing the influence relationship structure and local linear approximation to estimate the changing trends of key indicators, a multi-objective evaluation function is calculated. .

[0148] For objects and their neighborhoods with high current anomaly risk, this invention focuses on decision variable vectors. Construct several candidate intervention programs Strategies may include single-parameter fine-tuning, multi-parameter collaborative adjustment, partial process rearrangement, or temporary addition of testing processes.

[0149] During the scheme evaluation phase, this invention utilizes the influence relationship structure obtained in S3 and the local linear approximation to estimate the impact relationship in the scheme. Trends in key indicators This is a vector representing the change in parameters relative to the currently set parameters. (Definition) For the corresponding quality indicators and abnormal risk indicators, under the assumption of local small-scale adjustment, the following linear approximation can be used as the change vector. This is a sensitivity matrix, whose elements are derived from the influence relationship coefficients and historical data statistics, reflecting the degree of first-order influence of each decision variable on each quality / risk indicator.

[0150] In obtaining After estimation, the present invention further defines the scheme. A multi-objective evaluation function.

[0151] .

[0152] In the formula, This is a comprehensive indicator of abnormal risk. This is a production efficiency indicator. For implementation costs. This is the first weighting coefficient. This is the second weighting coefficient. This is the third weighting coefficient.

[0153] Production efficiency metrics refer to the impact of the proposed solution on production efficiency indicators such as cycle time / capacity. Implementation costs refer to the implementation costs or resource occupation (such as the costs incurred by adding testing procedures, switching equipment, and downtime maintenance). Weighting coefficients can be configured by the enterprise according to its business preferences.

[0154] In practical applications, heuristic search, local enumeration, or simple linear / integer programming methods can be used to search for solutions from the candidate set while satisfying process and safety constraints. Minimum or near-optimal solution This is a recommended strategy for handling such situations. This strategy simultaneously considers multiple objectives: "reducing the risk of anomalies," "controlling the impact on the cycle time," and "limiting implementation costs."

[0155] S54. Select the solution with the optimal evaluation function as the decision recommendation output.

[0156] The above suggestions are decision support information for production management and operation and maintenance management, and do not constitute real-time control commands for on-site equipment or robots. Finally, this invention will recommend a solution. This is mapped to actionable decision recommendations and task lists for business operations, including but not limited to: parameter formulation recommendations for MES / process parameter management systems (such as recommended values ​​and applicable ranges for press-fit target force, holding time, tightening torque, etc.); process flow configuration recommendations for production management (such as enabling / disabling intermediate inspection procedures, adjusting sampling ratios, temporarily adding inspection steps, or rearranging partial processes); and equipment switching recommendations and maintenance work order recommendations for operation and maintenance management systems. These recommendations can be pushed through interfaces with upper-level business systems such as MES, process parameter management systems, and maintenance management systems (e.g., CMMS), and approval / adoption results or actual implementation results will be recorded.

[0157] This invention simultaneously and continuously collects information such as key quality indicators, abnormal risk scores, and cycle time changes before and after implementation, in conjunction with the aforementioned... and The predicted values ​​are compared and evaluated, and the evaluation results are fed back to the assembly knowledge graph and influence relationship modeling module as feedback signals to update the sensitivity matrix M, influence relationship weights, and anomaly detection model parameters.

[0158] Through this process of "influence path tracing - decision variable screening - scheme generation and effect evaluation - suggestion push / approval - effect evaluation and feedback update", this invention realizes a causal style decision-making closed loop starting from high-risk nodes on a unified knowledge graph carrier, which tightly couples "graph structure-enhanced anomaly risk prediction" and "influence relationship network-driven causal decision-making", ensuring both the interpretability of decision suggestions and their engineering feasibility and evolvability.

[0159] Furthermore, to ensure that the "graph-structure-enhanced anomaly risk prediction + influence relationship network-driven causal decision-making" is interpretable, executable, and continuously optimized on-site, this invention designs visualization, human-computer interaction, and knowledge maintenance and feedback learning modules in the software system. The assembly knowledge graph, anomaly risk score, influence path, and decision scheme are displayed graphically to process engineers and scheduling engineers, and human feedback is used to perform closed-loop correction of knowledge and models.

[0160] The assembly knowledge graph browsing view displays entities and their relationships (such as belongTo, consistOf, executedAt, hasParameter, leadsTo, indicates, occurOn, etc.) in a graphical format, including orders, joints, parts, processes, workstations / equipment, process parameters, quality indicators, anomalies, and maintenance activities. Users can filter and expand paths by order, joint, or process link to view the structural relationships of "order—joint—critical process—workstation / equipment—quality indicator—anomaly event," and overlay the quality results, anomaly markers, and maintenance records of the current or most recent time window, providing a structural context for anomaly localization and decision interpretation.

[0161] Using an anomaly risk monitoring dashboard, based on a timeline and assembly line workstation layout, the anomaly risk score output by the graph-augmented anomaly risk prediction module is analyzed. Visualization is then implemented. This allows for the display of risk curves or heatmaps for key processes / workstations over time, reflecting how risks change over time. Spatially, nodes are categorized based on workstation layout. of Use color or size coding to highlight high-risk nodes. It also supports filtering by order, joint, or process type to easily identify areas with concentrated risk.

[0162] When the system issues an abnormal warning or provides a recommended intervention plan for a certain node At this time, the front-end interface provides an "Influence Path and Decision Explanation" view. On the assembled knowledge graph, the influence path from the candidate cause node to the quality indicator / abnormal event node is highlighted, and the influence strength of key edges is marked to explain "where the risk comes from." For candidate solutions... This demonstrates the corresponding parameter adjustments (such as pressing force, holding time, activation of testing procedures, etc.) and their expected impact on indicators such as risk, cycle time, and cost, helping engineers understand the advantages and disadvantages of different solutions. The final recommended solution is then presented. Provide a brief explanation, such as "prioritizing the adjustment of certain parameters is because they are highly sensitive to current abnormal indicators and the adjustment cost is low," to enhance the interpretability and auditability of decision-making recommendations.

[0163] The anomaly risk prediction method in this embodiment allows engineers to maintain the assembly knowledge graph and model output through an interface. Regarding knowledge maintenance, engineers can add / modify entities and relationships (such as adding fault modes, updating process constraints, maintaining equipment compatibility relationships, etc.), which are then verified and written into the graph database for subsequent impact relationship modeling and decision calculation. Regarding feedback annotation, engineers can mark the anomaly judgments and recommended solutions provided by the system with "approval / negation / partial adoption," and supplement the actual reasons and final handling measures to form structured feedback samples.

[0164] Regarding parameter updates, the server periodically updates the weights of influencing edges, anomaly risk thresholds, and sensitivity matrices based on accumulated feedback. For example, the weight of a particular influencing edge can be achieved using simple weighted smoothing.

[0165] .

[0166] In the formula The rating is calculated based on user feedback. This is the learning rate. It indicates something old.

[0167] Through the aforementioned visualization, human-computer interaction, and feedback learning mechanisms, this invention can continuously accumulate engineers' on-site experience into the assembly knowledge graph and anomaly detection / causal decision-making model, gradually improving the accuracy of anomaly detection and the rationality of decision-making, and forming a continuously evolving intelligent assembly process monitoring and decision support system.

[0168] This invention revolves around an integrated process design of "predicting the future—deducing the effects of proposed solutions—closed-loop updating." The overall architecture includes an industrial field data acquisition and integration layer, a unified representation layer of the assembly knowledge graph, a graph-enhanced anomaly risk prediction layer, a layer for constructing and updating the structure of impact relationships in the assembly process, and a causal decision-making and intervention evaluation layer. These layers are coupled using the assembly knowledge graph as a unified carrier. This unified carrier consists of a schema layer, an instantiated entity-data assembly graph, and a sequence of graph snapshots formed as the time window rolls, ensuring consistent representation of knowledge semantics, entity alignment, and temporal evolution within the same semantic space. The output of this invention is decision-making suggestions and task lists for production management and operation and maintenance management. These suggestions can be pushed through upper-level business system interfaces and implemented after approval by authorized personnel, without directly controlling the action at the field control layer.

[0169] The principle of unified representation of assembly knowledge and data is as follows: First, based on ontology modeling, orders, product structure, parts, processes, workstations / equipment, process parameters, quality indicators, abnormal events, maintenance activities, and experience rules in the industrial robot joint assembly process are abstracted into entities and relationships, constructing an assembly knowledge graph with assembly hierarchy, process sequence, resource constraints, and typical anomaly patterns. Then, multi-source time-series data, such as MES orders and work records, SCADA / PLC process parameters, online / offline quality inspection results, and equipment alarms and maintenance records, are aligned and linked with entity / relationship nodes in the knowledge graph using key identifiers such as order number, part number, process number, and timestamp, forming an integrated "entity + data" assembly diagram that evolves over time. This unified representation preserves the structural priors of the assembly process and provides a unified input for subsequent local subgraph extraction, time-slice modeling, and anomaly propagation analysis.

[0170] The principle of graph-enhanced anomaly risk prediction is as follows: Addressing the limitation of single time series sequences in characterizing inter-process coupling, this invention uses a snapshot sequence of an assembly knowledge graph as input to encode the local assembly network within a window and obtain the state embedding of the monitored nodes. Based on this, a temporal prediction network is constructed to output the prediction embedding for the next window. When running online, this invention can be applied to the window upon its closing. The distance between the data and the historical normal embedding distribution generates a "pre-emptive risk prior," triggering an early warning when the prior risk exceeds a threshold. (When the window...) After the real data arrives, the real embedding is then calculated. and The predictive residuals between the two are used to calibrate and confirm the risk score online, thereby enabling proactive detection of "anomalies that are about to occur" rather than relying solely on post-event alerts.

[0171] The principle behind the influence relationship network and anomaly propagation path modeling is as follows: To elevate anomaly risks from a "phenomenon" level to an interpretable analysis at the "mechanism" level, this invention constructs an influence relationship network for the assembly process based on the assembly knowledge graph and its temporal attributes. Specifically, using a snapshot of the assembly diagram after a time slice as input, and under the premise of satisfying prior constraints such as assembly sequence constraints and resource compatibility constraints, a relationship modeling method with sparse constraints and robustness is adopted to estimate the influence strength of controllable process parameters, equipment status, and environmental variables on quality indicators and anomaly events, forming a weighted directed relationship network. This network depicts the propagation path of "parameter – process – quality / anomaly" and can be updated through a sliding time window as operating conditions change, providing a structural basis for subsequent causal-style path tracing and intervention inference.

[0172] The principle of causal decision generation and intervention effect extrapolation is as follows: When the graph-enhanced anomaly risk prediction module indicates a high anomaly risk in a future process / workstation / order, this invention uses the high-risk node as the endpoint and searches for possible anomaly propagation paths from top to bottom under the constraints of the influence relationship network and assembly knowledge graph, forming a set of candidate cause nodes. It then selects decision variables (such as press-fit target force, press-fit speed, holding time, tightening torque, cycle time buffer time, equipment switching strategy, etc.) that can be changed through process adjustments or maintenance measures. Around these decision variables, a local intervention response model is constructed to estimate the changing trends of multi-dimensional indicators such as anomaly risk, key quality indicators, and production cycle time under different intervention schemes. Based on this, by combining linear approximation models with heuristic search / linear optimization, a set of candidate intervention schemes is generated under the premise of satisfying process and resource constraints. The intervention response model is then used to perform forward-looking effect extrapolation and multi-objective comprehensive evaluation of each scheme, thereby outputting causal-style decision recommendations and their explanatory paths that can reduce future anomaly risks while also considering capacity and cost. Through the front-end visual interface, engineers can view the assembly knowledge graph structure, key impact paths, distribution of abnormal risks, and the predictive effects of each candidate decision-making scheme. They can also interactively maintain and learn from the knowledge and models, thus forming a closed loop of abnormal risk prediction and causal decision support driven by both knowledge and data.

[0173] The principle of closed-loop monitoring and self-learning update is as follows: To improve the robustness of the system in large-scale, dynamic production environments, this invention integrates multi-level monitoring and hierarchical reasoning mechanisms at the software system level. On the one hand, the system performs hierarchical monitoring and consistency verification of data quality, model input distribution, anomaly risk scoring, and decision execution results, promptly identifying issues such as sensor noise, data loss, and operational drift. On the other hand, the system writes the key processes and results of each stage—"early warning—path interpretation—solution recommendation—execution feedback"—back to the assembly knowledge graph. Through incremental updates affecting the relational structure, revising rule constraints, and incrementally training or calibrating the prediction model, it gradually improves prediction accuracy and decision quality, forming a closed-loop optimization mechanism of "predicting the future—deriving disposal suggestions—adoption feedback—model update".

[0174] The above technical principles enable this invention to tightly couple abnormal risk prediction with causal decision support on a unified assembly knowledge graph, thereby achieving integrated intelligent assembly process monitoring and decision support from future risk prediction to intervention scheme design and effect extrapolation.

[0175] This embodiment proposes an assembly knowledge graph modeling method for industrial robot joint assembly lines. It unifies the order dimension, assembly structure dimension, process dimension, resource dimension, quality dimension, and anomaly dimension into the same graph structure, and explicitly encodes assembly sequence constraints, pairing constraints, and typical anomaly patterns, providing a unified semantic foundation for subsequent data fusion and decision analysis.

[0176] Based on multi-source assembly data, this embodiment constructs an influence relationship modeling method that integrates structural priors. By estimating the influence strength between key variables through rolling time windows, it obtains an influence relationship structure that can be updated with changes in operating conditions, laying the foundation for causal style decision analysis.

[0177] This embodiment designs a graph-enhanced anomaly risk prediction method. It utilizes the local neighborhood structure of the assembly knowledge graph and the temporal prediction network to conduct a forward-looking anomaly risk assessment of key processes / workstations / orders, achieving "early warning" rather than just alarming after anomalies occur.

[0178] This embodiment proposes a causal decision generation mechanism based on influence path analysis. It maps anomaly risk scores to causal paths in the assembly knowledge graph, automatically filters controllable variables, and constructs candidate disposal suggestions. Through a comprehensive evaluation of the solutions in terms of quality, cycle time, and cost, it outputs parameter adjustment suggestions and work order suggestions that can be adopted by production management and operation and maintenance management, and supports approval and traceability.

[0179] This embodiment integrates engineer experience into knowledge graphs and model parameter updates through human-computer interaction and feedback learning mechanisms, achieving closed-loop optimization of "risk assessment - suggestion push - adoption feedback - model update", which is conducive to forming sustainable supervision and predictive decision support capabilities in complex discrete manufacturing scenarios such as industrial robot joint assembly.

[0180] Example 2: The present invention provides an anomaly risk prediction device based on assembly knowledge graph, which includes a segmentation module, an instantiation module, a modeling module, an identification module, and a scheme module.

[0181] The segmentation module is used to acquire multi-source heterogeneous data from industrial sites, associate and align the multi-source heterogeneous data through a unified primary key set, and segment the data according to a fixed-length time window to generate a time-slice dataset.

[0182] The instantiation module is used to construct the assembly domain ontology and the assembly knowledge graph pattern layer. It uses the time slice dataset to instantiate the assembly domain ontology and constructs a snapshot sequence of the assembly knowledge graph that scrolls with the time window.

[0183] The modeling module is used to model and update the influence relationships between variables in the assembly process based on the snapshot sequence of the assembly knowledge graph, and obtain the influence relationship structure of the assembly process including the quantified influence intensity.

[0184] The identification module is used to combine the assembly knowledge graph snapshot sequence with the assembly process influence relationship structure to encode the graph structure features of the monitoring nodes, predict the state representation of the next time window based on the embedding representation of the historical time window, and calculate the abnormal risk score to identify abnormal risks.

[0185] The solution module is used to trace the impact path based on the impact relationship structure of the assembly process when an abnormal risk is identified to exceed a threshold, screen controllable decision variables, generate candidate intervention solutions and evaluate their effects, and output decision recommendations.

[0186] Example 3: This invention provides an anomaly risk prediction device based on an assembly knowledge graph, comprising a processor, a memory, and a computer program stored in the memory. The computer program can be executed by the processor to implement an anomaly risk prediction method based on an assembly knowledge graph as described in any paragraph of Example 1.

[0187] It is understood that the prediction device may be an electronic device with computing power, such as a portable laptop computer, desktop computer, server, smartphone or tablet computer.

[0188] Example 4: This invention provides a computer-readable storage medium. The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device containing the computer-readable storage medium to perform an anomaly risk prediction method based on an assembly knowledge graph as described in any paragraph of Example 1.

Claims

1. A method for predicting abnormal risk based on an assembly knowledge graph, characterized in that, Comprising: S1, acquiring multi-source heterogeneous data of an industrial site, associating and aligning the multi-source heterogeneous data through a unified primary key set, and dividing the multi-source heterogeneous data according to a fixed length time window to generate a time slice data set; S2, constructing an assembly field ontology and an assembly knowledge graph mode layer, instantiating the assembly field ontology using the time slice data set, and constructing a rolling assembly knowledge graph snapshot sequence with time windows; S3, based on the assembly knowledge graph snapshot sequence, modeling and updating the influence relationship between variables in the assembly process to obtain an assembly process influence relationship structure containing quantified influence intensity; S4, combining the assembly knowledge graph snapshot sequence and the assembly process influence relationship structure, graph structure feature coding of the monitoring node, and based on the embedding representation of the historical time window, predicting the state representation of the next time window, calculating the abnormal risk score to identify abnormal risk; S5, when the abnormal risk is identified to exceed a threshold, based on the assembly process influence relationship structure, influence path tracking is performed, controllable decision variables are screened, candidate intervention schemes are generated and their effects are evaluated, and decision suggestions are output. 2.The method of claim 1, wherein S2 Including: Defining entity types of the assembly field ontology, and defining a monitoring unit entity as a dynamic carrier of the key process execution process; Defining the relationship type between design entities, and defining the event occurrence relationship and state indication relationship representing dynamic evolution, and reserving a weight attribute for the influence class relationship, and reserving controllability and cost attributes for process parameter and maintenance activity entities; Generating a mode layer of the assembly knowledge graph in the graph database, defining node modes, edge modes, and a rule set containing process sequence constraints, parameter value constraints, and device compatibility constraints; Unified primary key set Entity instantiation is performed on the time-sliced dataset, specific business object nodes and monitoring unit nodes are created in the graph database, and the association relationship between the nodes is established according to the schema layer. Performing data attribute hanging and feature extraction, hanging the state information of the production execution system to the corresponding node, extracting key inflection points, peak values, mean values, and variance statistics from the continuous waveform data collected by the data acquisition and monitoring system to encode process feature vectors, encoding quality detection results and judgment labels into quality feature vectors, and hanging the above feature vectors to the corresponding monitoring unit node; According to the fixed length The time axis is divided, and in each time window , a set of active entity nodes , a set of relationships , and a set of attribute vectors are extracted , a snapshot of the assembled knowledge graph corresponding to the time window is constructed , forming a sequence of snapshots rolling over time. 3.The abnormal risk prediction method based on the assembled knowledge graph according to claim 1, wherein S3 Including: In each time window , according to the entity type, relationship type and process constraint in the assembly knowledge graph, a set of candidate independent variable sets and a set of target variable sets are selected in the snapshot ; For each target variable , a simplified linear influence relationship model is adopted to estimate the sparse influence relationship within the time window ; ; where, is the candidate independent variable set filtered from the knowledge graph relationship; is the influence coefficient to be estimated, reflecting the influence direction and strength of the independent variable on the target variable within the window ; is the residual term; Learning a coefficient vector from sample data through a regression algorithm with sparse constraints, and regarding non-zero or absolute values exceeding a set threshold as evidence of significant influence; According to the coefficient vector, a weighted directed edge is added or updated in the assembly knowledge graph to obtain an assembly process influence relationship structure in the time window And the influence relationship is corrected in combination with structure prior constraints and process common sense constraints. 4.The method of claim 1, wherein S4 Including: Structure of relationship between assembly process and influence and the structure edge set in the assembly knowledge graph snapshot, an adjacency matrix containing process structure and influence weight is constructed and normalized The node representation is obtained by updating the node features layer by layer using a graph feature transformation structure based on neighborhood aggregation: ; wherein is the node representation of the i-th layer; is the node representation of the i-th layer; is a learnable transformation matrix, is a nonlinear activation function. For each monitored object corresponding node , with its graph structure embedding sequence in the nearest time window as input, the next time window prediction embedding is obtained by using the time series prediction network ; Computing a final risk score that includes an estimated risk prior and a residual risk : ； wherein is the estimated risk prior based on historical normal embeddings mean and covariance calculation; is the residual risk based on real observations; is the fusion weight. 5.The method of claim 4, wherein, S4 further comprises: when the final risk score risk alert threshold an abnormal risk exists in the next time window; Node Identify the location as an anomaly and retrieve the node from the assembly knowledge graph. The attribute information and its associated abnormal event entities are parsed to obtain specific abnormal type descriptions, and early warning information including prediction time, abnormal location, abnormal type and risk score is generated. 6.The method of claim 1, wherein, Step S5 comprises: Taking a high-risk quality indicator node or an abnormal event node as a terminal node, performing reverse search along the weighted directed edges on the assembly process influence relationship structure, calculating the path comprehensive contribution degree according to the influence weight of each edge on the path, and selecting a candidate cause node set; Screening out variables that can be changed by parameter adjustment, process change or maintenance operation from the candidate cause node set to form a decision variable vector ; Generating several candidate intervention scenarios And using the impact relationship structure and the local linear approximation to estimate the change trend of the key indicators, the multi-objective evaluation function is calculated ; ； wherein is an overall abnormality risk indicator; is a production efficiency indicator; is an implementation cost; is a first weight coefficient; is a second weight coefficient; is a third weight coefficient; Selecting the scheme with the optimal evaluation function as the decision suggestion output. 7.The method of claim 1, wherein, S1 comprises: Pulling incremental data from the production execution system, the data acquisition and monitoring system, the quality detection system, the equipment maintenance system, and the document management system periodically or in real time to form data sets; Standardizing the format and mapping the fields of each data source, including unit unification, code mapping, and processing of abnormal values and missing values; Defining a uniform primary key set , by using Primary key matching and association are performed on multi-source data to generate a uniform intermediate table with "order-joint-assembly process-station / equipment" as the main line. ； wherein represents an order number; represents a joint or product identification; represents a process identification; represents a station / equipment identification; represents a part serial number; represents an event timestamp; Uniformly aligning and resampling the timestamps divides the continuous time axis into fixed length time windows, and aggregates the corresponding records within each window to generate a time-slice dataset ; The uniform intermediate table and the time-slice data sets of each time window are mapped into entity nodes, attribute nodes and relationship edges in the assembled knowledge graph through a data interface service, and are written into a graph database.

8. An abnormal risk prediction device based on an assembly knowledge graph, characterized by, Comprise: A segmentation module is configured to acquire multi-source heterogeneous data of an industrial site, associate and align the multi-source heterogeneous data through a uniform primary key set, and segment the multi-source heterogeneous data according to fixed-length time windows to generate time-slice data sets; An instantiation module is configured to construct an assembly domain ontology and an assembly knowledge graph mode layer, instantiate the assembly domain ontology using the time-slice data sets, and construct a sequence of assembly knowledge graph snapshots rolling with time windows; A modeling module is configured to model and update influence relationships between variables in an assembly process based on the sequence of assembly knowledge graph snapshots, and obtain an assembly process influence relationship structure containing quantified influence intensity; An identification module is configured to combine the sequence of assembly knowledge graph snapshots and the assembly process influence relationship structure, perform graph structure feature coding on a monitoring node, predict a state representation of a next time window based on an embedded representation of a historical time window, calculate an abnormal risk score to identify an abnormal risk, and output a decision suggestion. A scheme module is configured to perform influence path tracking based on the assembly process influence relationship structure when an abnormal risk exceeding a threshold is identified, screen controllable decision variables, generate a candidate intervention scheme and evaluate an effect of the candidate intervention scheme, and output a decision suggestion. 9.A device for predicting abnormal risk based on an assembly knowledge graph, characterized in that, The computer readable storage medium comprises a stored computer program, wherein the computer readable storage medium controls a device where the computer readable storage medium is located to execute the abnormal risk prediction method based on the assembled knowledge graph according to any one of claims 1 to 7 when the computer program runs.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium comprises a stored computer program, wherein the computer readable storage medium controls a device where the computer readable storage medium is located to execute the abnormal risk prediction method based on the assembled knowledge graph according to any one of claims 1 to 7 when the computer program runs.

Images