Production line path updating method and system combined with process-related knowledge graph

Abstract

This application discloses a production line path update method and system that combines a process association knowledge graph. The method includes: obtaining the process association knowledge graph corresponding to the target production line; performing semantic structure reverse deconstruction on the process association knowledge graph to obtain the process semantic vector space distribution state; performing negative sample topology expansion processing based on the process semantic vector space distribution state to generate a process shadow graph containing exclusive association edges; performing logical conflict scanning processing on the process shadow graph and the process association knowledge graph to determine the periphery of feasible paths and generate a production line path update constraint set; and performing process transition update processing on the process association knowledge graph according to the production line path update constraint set to obtain the production line path update result. This improves the logical rigor and semantic rationality of the production line path update operation.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a production line path update method and system that combines process association knowledge graph. Background Technology

[0002] In the field of automated manufacturing execution systems and process management, the operational logic of a production line is typically defined and maintained using a process association knowledge graph. A process association knowledge graph is a directed graph structure data representation where process nodes are entities and process flow relationship edges are associations. The process constraint attributes of process nodes define the input and output specifications of that process unit, while the flow condition descriptions of the process flow relationship edges define the triggering rules for the transfer of materials or semi-finished products between processes.

[0003] In actual production, with the iteration of equipment, adjustment of process parameters, or the need to optimize production cycle time, technicians need to add, delete, or modify the process flow relationship edges in the existing process association knowledge graph to generate updated production line paths. However, existing path update methods mostly rely on manual experience or local rules for edge operations, making it difficult to predict global process logic conflicts during the update process, and easily introducing implicit risks such as timing deadlock, thermodynamic incompatibility, or resource contention. Summary of the Invention

[0004] This application provides a production line path update method and system that combines process association knowledge graphs to overcome the problem that existing technologies are unable to predict global process logic conflicts during the update process, which can easily introduce hidden risks such as timing deadlock, thermodynamic state incompatibility, or resource contention.

[0005] This application provides a production line path update method that combines a process association knowledge graph, applied to a production line path update system. The method includes: Obtain the process association knowledge graph corresponding to the target production line; The semantic structure of the process-related knowledge graph is reversed to obtain the process semantic vector space and the process semantic embedding vectors corresponding to the process nodes in the process semantic vector space. The process semantic vector space distribution state is generated based on the process semantic vector space and the process semantic embedding vectors. Based on the spatial distribution state of the process semantic vector, negative sample topology expansion processing is performed to obtain a set of negative sample topology expansion process nodes and their corresponding virtual connection states of process disasters. A process shadow graph is generated by combining the set of negative sample topology expansion process nodes and the virtual connection states of process disasters. The process shadow graph contains the exclusive association edges between each negative sample topology expansion process node in the set of negative sample topology expansion process nodes and the process nodes in the process association knowledge graph. Logical conflict scanning is performed on the process shadow graph and the process association knowledge graph to determine the feasible path periphery of the process flow relationship edge in the process association knowledge graph and the proposed adoption path nodes contained within the feasible path periphery. A production line path update constraint set is generated based on the feasible path periphery and the proposed adoption path nodes contained within the feasible path periphery. The process association knowledge graph is updated by performing process transition update processing based on the production line path update constraint set to obtain the production line path update result. The production line path update result includes the updated process flow relationship edge and the update flow condition description information corresponding to the updated process flow relationship edge.

[0006] One embodiment of this application provides a production line path update system, including: A processor; a storage device having a computer program stored thereon; a network interface for providing network communication functions; when the computer program is executed by the processor, the processor enables the processor to implement any of the aforementioned production line path update methods that combine process association knowledge graphs.

[0007] One embodiment of this application provides a readable storage medium storing a program or instructions, which, when executed by a processor, implements the steps of the production line path update method combining process association knowledge graph.

[0008] Therefore, the embodiments of this application have the following beneficial effects: By obtaining the process association knowledge graph corresponding to the target production line and performing semantic structure reverse deconstruction processing on it to generate the process semantic vector space distribution state, discrete process symbols are mapped to a semantic manifold with continuous distance metric, thereby revealing the implicit semantic associations and distribution density between process nodes. Then, based on the process semantic vector space distribution state, negative sample topology expansion processing is performed to generate a process shadow graph containing exclusive association edges. This process shadow graph explicitly defines the logical forbidden zones in the production line reconstruction process. By performing logical conflict scanning processing on the process shadow graph and the process association knowledge graph to determine the periphery of feasible paths, a set of production line path update constraints is generated, achieving precise definition of the path change safety boundary under global semantic consistency constraints. Finally, process jump update processing is performed on the process association knowledge graph according to the production line path update constraint set. The resulting production line path update result can maintain the production line topology connectivity while avoiding process disaster risks, improving the logical rigor and semantic rationality of the production line path update operation. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a flowchart illustrating a production line path update method that combines a process association knowledge graph, as provided in an embodiment of this application.

[0011] Figure 2 This is a schematic diagram of the basic structure of a production line path update system provided in an embodiment of this application.

[0012] Figure 3 This is a functional block diagram of a production line path updating device provided in an embodiment of this application. Detailed Implementation

[0013] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0014] Please see Figure 1 , Figure 1 This is a flowchart of a production line path update method that combines a process association knowledge graph, provided in an embodiment of this application. The method can be executed by a production line path update system or jointly by a production line path update system and a server. The method may include steps 110-150.

[0015] This application provides a production line path update method that combines a process association knowledge graph. In the field of automated manufacturing execution and process management, a production line consists of a series of process nodes with strict sequential dependencies. The flow relationship between processes defines the legal movement path of materials or semi-finished products on the production line. When the production line needs to reconstruct its path due to equipment upgrades, changes in process parameters, or adjustments to quality control standards, adding or deleting paths based solely on local rules can easily introduce global process logic conflicts, such as forming deadlock loops, generating process connections that violate thermodynamic constraints, or generating unexecutable transition paths. This method performs semantic vector space deconstruction on the existing process association knowledge graph, generates a negative sample shadow graph based on spatial density distribution, and defines feasible regions driven by logic conflict scanning. Under the dual constraints of semantic continuity and topological reachability, it accurately calculates the production line path update result. This method does not depend on the morphological parameters of specific products but takes process semantics as the core driving factor. Therefore, it is applicable to various scenarios with process flow characteristics, such as discrete manufacturing, process industries, and business approval workflows. The following uses a surface mount technology production line as an example to explain in detail the implementation mechanism of each step, but this example should not be construed as a limitation on application scenarios.

[0016] Step 110: Obtain the process association knowledge graph corresponding to the target production line.

[0017] In this embodiment, the target production line refers to an electronic assembly line that includes multiple serial and parallel process units. The process association knowledge graph is a directed graph data structure stored in a graph database instance using an attribute graph model.

[0018] Each process node v in the set V of process nodes in the diagram i This corresponds to an indivisible process unit, such as solder paste printing process node v. sp , Surface Mount Technology (SMT) process node v pm Reflow soldering process node v rf Automated optical inspection process node v aoi and the process node v of the plate separation dp Each process node v i All carry process constraint attribute P i The process constraint attribute P i It is a multi-dimensional feature vector, whose dimensional components include the allowed input material status code range of the process, the output material status code range of the process, the maximum allowed waiting time window parameter of the process, and the equipment type identification code required by the process.

[0019] In the graph, each directed edge e in the set E of process flow relationships... ij This indicates that at the completion of process node v i Afterwards, the material is allowed to flow to process node v.j Each directed edge e ij Attached circulation condition description information C ij The flow condition description information C ij It is a structured conditional expression object, which includes at least the material temperature threshold range, the material position accuracy level requirement, and the test result flag requirement of the previous process. The knowledge graph associated with this process is obtained by retrieving a complete subgraph that matches the unique identifier of the target production line from the process model library of the manufacturing execution system through graph query language, and serializing it into an exchange file containing a node list, an edge list, and an attribute mapping table.

[0020] Step 120: Perform semantic structure reverse deconstruction on the process association knowledge graph to obtain the process semantic vector space and the process semantic embedding vectors corresponding to the process nodes in the process semantic vector space. Generate the process semantic vector space distribution state based on the process semantic vector space and the process semantic embedding vectors.

[0021] Since the nodes and edges in the process-related knowledge graph exist in the form of discrete symbols and structured text, they cannot directly support density analysis and logical conflict detection based on continuous spatial distance. This step aims to convert the symbolic graph structure into a low-dimensional manifold embedding that can be numerically computed, and to establish a mathematical representation of the distribution state in this manifold. The processing is further subdivided into the following sub-steps.

[0022] Step 121: Extract the process constraint attributes of each process node and the flow condition description information of each process flow relationship edge in the process association knowledge graph, and perform attribute semantic decomposition and condition semantic decomposition respectively to obtain the attribute semantic atom set containing attribute semantic atoms and their association strength, and the condition semantic atom set containing condition semantic atoms and their association strength.

[0023] First, regarding process node v i Process constraint attribute P i This attribute exists as a structured field. During attribute semantic decomposition, P... i Each field value is converted into its corresponding semantic encoding sequence in a predefined process dictionary. For example, the input material state encoding range is mapped to a concatenated set of high-dimensional sparse one-hot encoding vectors. A semantic decoupling algorithm based on nonnegative matrix factorization is applied to this concatenated result, decomposing the high-dimensional sparse vector into a weighted linear combination of several low-dimensional dense vectors. Each low-dimensional dense vector is defined as an attribute semantic atom. p The corresponding linear combination coefficients, after being normalized, are used as the attribute semantic atom association strength Str of that attribute semantic atom. p .

[0024] Therefore, each process node v i Generate a Set of property semantic atoms Ai The elements in the set are tuples (Atom) pk Str pk Secondly, regarding the process flow relationship edge e ij Flow condition description information C ij Since it contains logical expressions and numerical ranges, during conditional semantic decomposition, the logical expression is first parsed into an abstract syntax tree. Each leaf node (i.e., the basic conditional predicate, such as "temperature is less than threshold T") in the abstract syntax tree is then vectorized. A pre-trained language representation model based on a Transformer encoder structure is used to encode the predicate text, and the vector corresponding to the special classification marker position in the encoder output layer is taken as the semantic vector of the predicate. Subsequently, a non-negative matrix factorization algorithm is applied to decompose the predicate semantic vector into conditional semantic atoms. c Its corresponding conditional semantic atom association strength Str c The weighted combination. The flow relationship edge e for each process. ij Generate a Set of conditional semantic atoms Cij .

[0025] Step 122: Input the attribute semantic atom set and the condition semantic atom set into the pre-constructed graph embedding transformation structure for vectorization projection processing to obtain the initial process semantic embedding vector of the process node and the initial flow relationship semantic embedding vector of the process flow relationship edge.

[0026] The Set generated in step 121 Ai With Set Cij As input features, they are fed into a pre-constructed graph embedding transformation structure, which is a message-passing architecture containing multiple layers of graph neural networks. Specifically, its first layer is a graph isomorphic network convolutional layer, which receives each process node v i Set of semantic atom collections of attributes Ai The weighted sum vector is used as the initial node feature vector h. iinit The calculation formula is h iinit Equal to all Atom pk Multiply by its corresponding Str pk The cumulative sum. Simultaneously, this layer receives the sum of each edge e. ij Set of conditional semantic atoms Cij The weighted sum vector is used as the initial edge feature vector h. ijinit .

[0027] Graph isomorphic networks use convolutional layers to perform a non-linear transformation on the sum of node features and the features of its neighboring nodes through a learnable multilayer perceptron, thereby updating the node features. After the first convolutional layer, node v... i The feature is updated to h i_1 Edge eigenvector h ijinit Then through an independent linear transformation matrix W edge Projecting onto the same dimensional space as the node features yields the edge features h. ij_1 The second layer is a graph attention network layer, which utilizes the node features h output from the first layer. i_1 h j_1 and edge features h ij_1 Compute node v i For node v j Attention weight coefficient α ij Attention weight coefficient α ij The calculation method is as follows: first, h i_1 with h ij_1 After summing, a dot product is performed with a learnable attention weight vector a. The result is then activated by the LeakyReLU function and normalized by the SoftMax function on all neighbors j.

[0028] Subsequently, node v i The second layer output feature h i_2 It is calculated as all its neighbor nodes v j The first layer output feature h j_1 The weighted sum, where the weight is α. ij After multiple stacking layers, the final layer of the graph embedding transformation structure is a linear projection layer, which contains a weight matrix W. proj With bias vector b proj The high-dimensional intermediate node features h are transformed through linear transformation. i_L The vector is compressed into a pre-defined low-dimensional dense space R^d. The compressed d-dimensional vector is the process node v. i Initial process semantic embedding vector Emb vi Meanwhile, the edge feature h ij_L It is also input to another linear projection layer, through a linear transformation matrix W proje Compressing to the same d-dimensional space yields the initial semantic embedding vector Emb for the flow relationship. eij .

[0029] Step 123: Perform vector space position migration processing on the initial process semantic embedding vector based on the initial flow relationship semantic embedding vector, so that the process semantic embedding vector of the process node with process flow relationship edge satisfies the preset vector space distance constraint condition in the process semantic vector space.

[0030] Since the aggregation process of graph neural networks mainly relies on the similarity of node attributes, while insufficiently encoding the flow information of edges, the initial process semantic embedding vector Emb is... vi Relative positions in space may not accurately reflect the sequential dependencies between processes. This step corrects this bias through vector space position migration processing. The migration processing is based on minimizing an energy function. The first term of the energy function is defined as the attraction term, which is applied to each directed edge e in the graph. ij Calculate the embedding vectors of the two connected nodes, Emb. vi With Emb vj The distance between them is measured using Euclidean distance. The attraction term forces this distance to be approximated by an initial flow relation semantic embedding vector Emb. eij The expected distance d of the modulus expij They are close. Expected distance d expij The calculation method is to use Emb eij Input a two-layer multilayer perceptron regressor, which outputs a scalar value. This scalar value is mapped by a sigmoid function and multiplied by a preset maximum allowable distance coefficient to obtain d. expij The second term of the energy function is a repulsion term. For node pairs in the graph that are not connected by edges, the distance between their embedding vectors is calculated. The repulsion term penalizes excessively small distances to prevent nodes from different semantic clusters from collapsing in space. The Emb function is then processed using gradient descent. vi With Emb vj The vector values ​​are iteratively updated to minimize the energy function.

[0031] In each iteration, the partial derivative of the energy function with respect to each embedding vector is calculated. This partial derivative is multiplied by a preset learning rate parameter and then subtracted from the current embedding vector value to obtain the updated embedding vector. This process continues until the decrease in the energy function value is less than a preset convergence threshold. The semantic embedding vector obtained after the iteration terminates is the final vector representation that satisfies the vector space distance constraint.

[0032] Step 124: The semantic embedding vectors of the process after the vector space position migration process are spatially distributed and reorganized according to the original connection topology in the process association knowledge graph to generate the process semantic vector space.

[0033] After processing in steps 123, all process nodes obtained the corrected d-dimensional embedding vector Emb*. vi To establish a globally unified spatial reference frame, these isolated vector points need to be organized into a structured space. The specific execution method of spatial distribution reorganization is as follows: First, construct a two-dimensional matrix M with dimensions equal to the total number of process nodes multiplied by d. SpaceThe i-th row of this matrix represents the process node v. i Embedding vector Emb* vi Secondly, to preserve the topological context information of the graph, instead of simply arranging the vectors randomly by node index, the matrix row order is rearranged based on the topological sorting result of the process-related knowledge graph. A topological sorting algorithm is executed for each process node v. i Assign a topology level number L i This number represents the longest path length from the starting process node to this node. Subsequently, M... Space The row vectors in the topology are numbered L. i The nodes are arranged in ascending order, and within the same level, in descending order of their in-degree. The rearranged matrix M is then... Space This is defined as a process semantic vector space. In this vector space, the spatial distance and cosine similarity between any two points can be directly calculated and used for subsequent density analysis and semantic retrieval.

[0034] Step 125: Perform spatial density distribution analysis on the process semantic vector space to determine the spatial cluster center location information and spatial discrete boundary location information, and generate the process semantic vector space distribution state based on the spatial cluster center location information and spatial discrete boundary location information.

[0035] In the generated process semantic vector space matrix M Space In this context, the distribution of node vectors often exhibits a non-uniform manifold structure, meaning there are high-density semantic clusters and low-density sparse transition regions. Spatial density distribution analysis first involves processing the matrix M... Space Each point in Emb* vi Calculate its local density value ρ i Local density value ρ i The calculation method is as follows: using point Emb* vi Centered on a predetermined cutoff distance parameter, the distance between points Emb* and Emb in the spatial domain is statistically analyzed. vi The number of other points whose distance is less than the cutoff distance parameter is ρ. i .

[0036] Simultaneously, the high-density distance value δ for each point is calculated. i High-density distance value δ i Defined as the distance from this point to any local density value greater than ρ i The minimum Euclidean distance of the point. If the point itself is the point of maximum global density, then δ i It is defined as the maximum distance from it to any other point in space.

[0037] Then, the decision value γ for each point is calculated. i Decision value γ iDefined as ρ i With δ i The product of γ. The decision value γ i Sort the data from largest to smallest, and find the inflection point where the decision value sequence shows a significant decrease. Points before the inflection point are identified as spatial cluster center locations (Cen). k For each non-center point, it is assigned to the cluster of the nearest point with a local density value greater than its own. After all points have been clustered, the boundary region of each cluster is identified. The boundary region is defined as the set of points that belong to the cluster but contain neighboring points belonging to other clusters within its cutoff distance.

[0038] Within the boundary region, the point with the maximum local density value is identified, and this local density value is set as the density threshold for the cluster. Points within the cluster whose local density value is lower than this density threshold are marked as spatial discrete boundary location information (Bou). m The points mentioned above constitute the transitional sparse region between semantic clusters. Finally, the coordinate vectors of all cluster centers (Cen) are used to define this transitional region. k The set of all discrete boundary coordinate vectors Bou m The set of elements and the cluster label to which each point belongs together constitute the state of the semantic vector space distribution of the process. Dist This state information will be stored in the form of structured metadata, which records the member list, center vector value, and boundary point vector value of each cluster.

[0039] Step 130: Perform negative sample topology expansion processing based on the spatial distribution state of the process semantic vector to obtain the negative sample topology expansion process node set and its corresponding process disaster virtual connection state. Combine the negative sample topology expansion process node set and the process disaster virtual connection state to generate a process shadow graph. The process shadow graph contains the exclusive association edges between each negative sample topology expansion process node in the negative sample topology expansion process node set and the process nodes in the process association knowledge graph.

[0040] After clarifying the semantic distribution of normal process nodes, in order to identify the logical forbidden zones that must be avoided during subsequent path updates, this step constructs virtual process disaster nodes in low-density areas of the semantic space and establishes an exclusionary association between them and the real nodes in the original graph, thereby forming a shadow graph for logical verification. The specific processing procedure is as follows.

[0041] Step 131: Based on the vector space density distribution description information in the process semantic vector space distribution state, determine the low vector space density region in the process semantic vector space. The low vector space density region is a continuous spatial region in the process semantic vector space where the vector space density is lower than the preset vector space density threshold.

[0042] Extract the process semantic vector space distribution state generated in step 125. Dist The set of boundary points of each cluster included in Bou m and cluster density threshold. Traversal process semantic vector space matrix M Space For each embedded vector point, determine whether it belongs to an interior point of a certain cluster. If the point belongs to a cluster but is marked as a boundary point, or if the point does not belong to any cluster (i.e., a noise point), then initially mark it as a low-density candidate point.

[0043] Subsequently, a density-based spatial clustering algorithm (such as DBSCAN) was used to perform connectivity analysis on these low-density candidate points. The neighborhood radius in the algorithm parameters was set to the cutoff distance parameter used in step 125, and the minimum number of contained points was set to a preset low-density connectivity cardinality. This algorithm aggregated spatially continuous low-density candidate points into several connected regions. For each connected region, the average local density value Avg of all points within it was calculated. ρ_Region If Avg ρ_Region If the vector space density is below a preset threshold, which is dynamically determined by multiplying the global average local density value by a scaling factor less than 1, then the connected region is marked as a low vector space density region. Low The low vector space density regions mentioned above semantically correspond to "semantic vacuum zones" that are not clearly defined or are known to be high-risk in the process logic.

[0044] Step 132: Perform negative sample process node sampling processing in the low vector space density region to generate a negative sample topology extended process node set. Each negative sample topology extended process node in the negative sample topology extended process node set corresponds to a negative sample process semantic embedding vector in the process semantic vector space.

[0045] In each identified low vector space density region Low Within the region, negative sample processing nodes are sampled. The sampling process is based on a farthest-point sampling strategy, aiming to maximize the coverage of the geometry of the low-density region with the fewest possible negative sample nodes. First, in the region... Low From the set of coordinates of all points included, a point is randomly selected as the first negative sample process node v. neg_1 The coordinates of its negative sample process semantic embedding vector Emb neg_1 Secondly, for subsequent sampling points, calculate the Region. Low The coordinates of each unselected candidate point within the vector vectors of all selected negative sample process nodes and all true process nodes are embedded in the vector vector Emb*. viThe minimum Euclidean distance is calculated. The candidate point coordinates that maximize this minimum Euclidean distance are selected as the next negative sample process node v. neg_s coordinates Emb neg_s Repeat this process until the number of selected negative sample nodes reaches the preset sampling size limit Num. neg Alternatively, the minimum distance to the newly selected point may be lower than the preset minimum distance parameter.

[0046] All the virtual coordinate points generated through this process constitute the negative sample topology expansion process node set V. Neg Each coordinate point is not only a geometric location, but also implicitly represents the semantic conflict type corresponding to that region, such as a thermodynamically unreachable state or a temporal deadlock state.

[0047] Step 133: Based on the vector space distance between the semantic embedding vector of the negative sample process and the semantic embedding vector of the process node in the process association knowledge graph, determine the semantic repulsion strength parameter between each negative sample topological expansion process node and the process node in the process association knowledge graph. The semantic repulsion strength parameter is negatively correlated with the vector space distance.

[0048] For the negative sample topology expansion process node set V Neg Each node v in neg_s and its embedding vector Emb neg_s Traverse each real process node v in the process association knowledge graph i and its embedding vector Emb* vi Calculate the Euclidean distance between them, Dist. si Define the semantic rejection strength parameter Rep. si The calculation method is as follows: Rep si Dist equal to the negative of the natural constant e si The product of a scaling parameter β raised to the power of Rep si With Dist si They exhibit an exponentially decaying negative correlation. The scaling parameter β is determined by the region with low vector spatial density. Low The average local density value is dynamically determined; the lower the regional density, the smaller the β value, resulting in a slower rate of decrease in repulsive strength with distance and a larger area of ​​influence. When Dist si When Rep approaches zero si Approaching the maximum value of 1 indicates that the semantic position of the negative sample node exactly coincides with that of the real process node, at which point there is an extremely strong process conflict and semantic exclusion. When Dist si When Rep is much larger than the reciprocal of the scaling parameter β, siA value close to 0 indicates that the negative sample node is semantically far removed from the real process node, and its exclusionary effect can be ignored.

[0049] Step 134: Construct a virtual connection state of process disasters between the negative sample topology extended process node set and the process nodes in the process association knowledge graph based on the semantic exclusion strength parameter. The virtual connection state of process disasters includes the exclusionary association edges between the negative sample topology extended process nodes and the process nodes in the process association knowledge graph, as well as the process disaster type description information corresponding to the exclusionary association edges.

[0050] The semantic rejection strength parameter Rep is calculated based on step 133. si Set a preset rejection activation threshold Thresh Rep For any pair of negative sample process nodes v neg_s With actual process node v i If its semantic exclusion strength parameter Rep si Greater than Thresh Rep Then construct an exclusive associative edge e. negsi Exclusive related edge e negsi It is an undirected logical constraint edge, whose data structure includes an attribute field to store the repulsion strength value Rep. si Furthermore, the exclusive associated edge e negsi It also includes a description of the process disaster type: Disaster. si This field describes the disaster type information for the process. si The generation method is: retrieval v neg_s The region with low vector space density Low In spatial distribution state (State) Dist From the labels of several adjacent real semantic clusters, extract the key parameter names from the process constraint attributes of the corresponding process nodes, and match them with v. i The key parameter names in the process constraint attributes are combined and filled into the preset conflict description template.

[0051] For example, if the template is "Incompatibility between parameter A and parameter B leads to risk of type C", then the corresponding parameter names will be substituted to generate a specific descriptive text string. All exclusive associative edges e that meet the conditions... negsi and the Disaster it carries si The information together constitutes the virtual connection state of the process disaster. Disaster .

[0052] Step 135: Perform topological mirroring on the process association knowledge graph, embed negative sample topology to expand the process node set according to the process node mirror topology, and generate process shadow graph.

[0053] Since negative sample nodes represent "paths that should not exist," directly inserting them into the original graph would destroy the original graph's topological connectivity and semantic integrity. Therefore, a mirror layer embedding strategy is adopted, which is further divided into the following implementation steps.

[0054] Step 1351: Perform topological mirroring on the process association knowledge graph to generate a process node mirror topological structure of the process association knowledge graph. The process node mirror topological structure contains the topological position mapping relationship of each process node in the process association knowledge graph.

[0055] Create a mirror set of process nodes V corresponding one-to-one with the original process's knowledge graph process node set V. Mirror For each actual process node v i Generate a corresponding virtual mirror node v mirrori Mirror node v mirrori Inheriting only v i The node index identifies the mapping relationship and does not inherit its process constraint attributes.

[0056] Subsequently, the edge connections of the original graph are copied to the mirror layer, that is, for each directed edge e in the original graph... ij Create a path from v in the mirror layer. mirrori Pointing to v mirrorj The mirrored directed edge e mirrorij The mirrored directed edges also do not carry flow condition description information; they serve only as placeholders for the topology. The resulting mirrored node set V... Mirror With mirror edge set E Mirror Together, they constitute the mirror topology of the process nodes (Topo). Mirror .

[0057] Step 1352: The negative sample topology expansion process node set is embedded into the process node mirror topology structure according to the topological position mapping relationship in the process node mirror topology structure, so that each negative sample topology expansion process node forms an exclusive association edge correspondence with the process node with the maximum semantic repulsion strength parameter in the process association knowledge graph.

[0058] Traversing the negative sample topology to expand the process node set V Neg Each node v in neg_s In the actual process node set V, retrieve the node v. neg_s The parameter Rep with the largest semantic rejection strength max_s Real process nodes v target Since step 1351 has already generated v target mirror node v mirrortarget , will node v neg_sAdd to the mirror topology layer and build from v in the composite graph structure. neg_s Pointing to v target Exclusive related edge e neg_starget At this time, v neg_s Logically, it is "anchored" near the mirror image of the real process node with which it has the strongest conflict, thus enabling precise constraint effects on the target region during topological projection.

[0059] Step 1353: Generate a process shadow graph based on the correspondence of the exclusionary association edges and the description information of the process disaster type. The process node set of the process shadow graph includes the process node set of the negative sample topology extension and the process nodes in the process association knowledge graph. The process node relationship edge set of the process shadow graph includes the process flow relationship edge and the exclusionary association edge in the process association knowledge graph.

[0060] The final constructed process shadow graph Shadow It is a composite graph structure. Its node set V Shadow Defined as the set of true process nodes V and the set of negative sample topology-extended process nodes V Neg The union of the sets. Its edge set E. Shadow Defined as the set of edges E representing the original process flow relationships and the set of edges E representing the exclusive associations. Neg The union of the two sets. In this shadow graph, the original directed edges represent normal process flow logic, while the exclusive association edges represent logically prohibited conflict associations. By placing these two types of edges with diametrically opposed properties in the same graph structure, a unified data foundation is provided for subsequent cross-scan analysis.

[0061] Step 140: Perform logical conflict scanning on the process shadow graph and the process association knowledge graph to determine the feasible path periphery of the process flow relationship edge in the process association knowledge graph and the proposed adoption path nodes contained within the feasible path periphery. Generate a set of production line path update constraints based on the feasible path periphery and the proposed adoption path nodes contained within the feasible path periphery.

[0062] After generating the process shadow graph containing logical forbidden zone constraints, this step aims to identify the hidden high-risk paths in the original graph by detecting the topological intersections between the exclusive association edges in the shadow graph and the process flow relationship edges in the original graph, and then delineate a safe feasible region that allows path reconstruction. This process includes the following sub-steps.

[0063] Step 141: Map the exclusionary association edges in the process shadow graph to the process node connection topology of the process association knowledge graph, determine the process flow relationship edges in the process association knowledge graph that have a topological intersection relationship with the exclusionary association edges, and mark the process flow relationship edges with a topological intersection relationship as implicit exclusionary process flow relationship edges.

[0064] This step relies on a cross-validation algorithm based on graph-planar embedding. First, a two-dimensional planar layout coordinate system is generated for the node set V of the process association knowledge graph. This layout coordinate system is calculated using a force-directed layout algorithm. In this algorithm, the node repulsion force is proportional to the semantic distance between nodes, while the edge attraction force tries to bring nodes connected by edges closer together. Under this planar layout, each process flow relationship edge e in the process association knowledge graph... ij All can be achieved using a single slave node v i Coordinates pointing to node v j The coordinates are represented by directed line segments. Simultaneously, the repulsive correlation edges e in the process shadow graph are... negsi Also use a connection node v neg_s Coordinates and node v i The coordinates are represented by undirected line segments.

[0065] Subsequently, all pairs of exclusivity-related edge segments and process flow relationship edge segments are traversed, and a segment intersection determination algorithm is executed. This algorithm is based on the vector cross product notation; specifically, it calculates that if the two endpoints of an exclusivity-related edge segment are located on opposite sides of a process flow relationship edge segment, and the two endpoints of the process flow relationship edge segment are located on opposite sides of an exclusivity-related edge segment, then the two segments are considered to intersect. Once a process flow relationship edge e is detected... ij If an edge e intersects with any exclusive edge, then the flow relationship of that process is determined. ij There is a topological intersection relationship with the edge of the exclusionary association, and the edge is marked as an edge of implicit exclusionary process flow relationship. Although the marked edge seems legal in its local connection relationship, it passes through the process disaster area in the global semantic space. Therefore, it should be removed when updating the path.

[0066] Step 142: Perform path blocking analysis on the implicit exclusion process flow relationship edges to identify the process node subsets in the process association knowledge graph that cannot form a complete process flow path due to the existence of implicit exclusion process flow relationship edges, and mark the process node subsets as process logic conflict process node subsets.

[0067] Path blocking analysis is used to accurately quantify the impact of removing hidden exclusion edges on graph connectivity. The specific implementation process is as follows.

[0068] Step 1421: Extract all process flow relationship edges marked as implicitly exclusive process flow relationship edges in the process association knowledge graph, and generate a set of implicitly exclusive process flow relationship edges. Each implicitly exclusive process flow relationship edge in the set corresponds to a process node that is the starting point of the implicitly exclusive process flow relationship edge and a process node that is the ending point of the implicitly exclusive process flow relationship edge.

[0069] Traverse the edge set E of the process association knowledge graph, filter out all edge objects that were given implicit exclusion tags in step 141, and collect the references of these edge objects into a new set data structure to form the implicit exclusion process flow relationship edge set E. Hidden For E Hidden Each edge object e in hidxy Read its starting node reference v hidx Reference v to the endpoint node hid_y And it is stored as an attribute associated with the edge object.

[0070] Step 1422: Remove all implicitly exclusive process flow relationship edges from the set of implicitly exclusive process flow relationship edges in the process association knowledge graph to obtain a topological snapshot of the process association knowledge graph after removing the implicitly exclusive process flow relationship edges. The topological snapshot of the process association knowledge graph only retains process flow relationship edges that are not marked as implicitly exclusive process flow relationship edges.

[0071] Based on the adjacency list data structure of the original process association knowledge graph, a complete copy is cloned into memory to obtain the graph object. Temp In Graph Temp Execute edge removal transactions: Traverse the set E of implicit exclusion process flow relationship edges. Hidden Call the edge deletion interface of the graph data structure to remove each implicit exclusion process flow relationship edge from the Graph. Temp Logical deletion within the graph. After the deletion operation is complete, the graph... Temp The number of nodes in the graph remains the same, but the number of edges decreases, thus altering its connected component structure. The graph at this point... Temp This is a snapshot of the topology of the knowledge graph related to the process.

[0072] Step 1423: Perform connected component decomposition on the topological snapshot of the process association knowledge graph. Divide the process nodes in the topological snapshot of the process association knowledge graph into several connected components. There is at least one continuous process flow path between any two process nodes in each connected component, which is composed of the process flow relationship edge in the topological snapshot of the process association knowledge graph. There is no process flow path between different connected components.

[0073] In Graph objects Temp Above, perform connected component decomposition based on breadth-first search or a disjoint-set data structure. The process initializes an empty list of sets to store the connected components. Traverse the Graph. Temp Each unvisited process node v in startcomp , with v startcomp Perform a breadth-first search starting from the graph, only traversing along the graph during the search.Temp Traverse the existing edges and collect all visited nodes into a single node set Comp. k In the middle. After a breadth-first search is completed, Comp will be... k Add it to the list as a connected component. Repeat this process until all nodes have been visited. The final output is a set of disjoint nodes from Comp1 to Comp2. K The above sets satisfy the following: for the same set Comp k Any two nodes within the graph Temp There exists at least one path in the graph; for nodes belonging to different sets, in the graph... Temp There is no path in it.

[0074] Step 1424: Extract the starting process node from the process association knowledge graph, locate the starting connected component containing the starting process node among several connected components, mark the starting connected component as the main connected component, and mark the remaining connected components that do not contain the starting process node as isolated connected components.

[0075] Read the unique identifier v of the preset starting process node from the production line's metadata configuration. Start Iterate through the list of connected components generated in step 1423, and for each connected component Comp... k Check if its node member list contains the identifier v Start The node containing v. Start The connected components are labeled as primary connected components Comp. Main For all the rest of the list that do not contain v Start Connected components Comp Other These are uniformly labeled as isolated connected components. A primary connected component represents the set of processes that can still be started and flow normally from the production line starting point after the removal of implicit exclusion edges. An isolated connected component represents the set of processes that cannot be reached from the starting point because the critical path is cut off.

[0076] Step 1425: Extract all process nodes contained in the isolated connected components, generate a preliminary set of process nodes with process logic conflicts, and perform reverse path tracing processing on each process node in the preliminary set of process nodes with process logic conflicts to determine whether the process node is connected to the process node in the main connected component through at least one process flow path containing an edge of implicit exclusion process flow relationship before removing the implicit exclusion process flow relationship edge. If such a process flow path exists, the process node is retained in the preliminary set of process nodes with process logic conflicts; if such a process flow path does not exist, the process node is removed from the preliminary set of process nodes with process logic conflicts.

[0077] Collect the process node identifiers from all isolated connected components to form an initial set of process logic conflicting process nodes (Set). Conflict_Pre However, Set Conflict_Pre Not all nodes in the graph become isolated due to the removal of implicit exclusion edges; some nodes may have been isolated in the original graph.

[0078] To achieve precise filtering, it is necessary to perform a set... Conflict_Pre Each node v in conf Perform reverse path tracing. The process is as follows: In the original complete process association knowledge graph without removing any edges, using v conf If the endpoint is the primary connected component Comp, a depth-first search is performed in reverse. During the search, the set of edges for the current search path is maintained. If there exists at least one edge from the primary connected component Comp... Main Starting from any node in the middle and eventually reaching v conf The path, and the set of edges of this path is related to the set of edges E of the implicit exclusion process flow relationship. Hidden If the intersection of the paths is not empty (i.e., the path uses at least one implicit repulsive edge), then determine v. conf It became an isolated node because of the blocking effect of the implicit repulsive edge; therefore, it should be kept in the Set. Conflict_Pre In the middle. Conversely, if there is no path from the main connected component to v. conf If a path, or all existing paths, do not contain any implicit exclusion edges, then determine v. conf The isolated state is irrelevant to this implicit exclusion analysis, so it is removed from the Set. Conflict_Pre Removed from the middle.

[0079] Step 1426: Mark the initial set of process logic conflicting process nodes after reverse path tracing as a subset of process logic conflicting process nodes.

[0080] After filtering in step 1425, the remaining set of nodes is officially marked as the subset of process logic conflict operation nodes. Conflict The nodes in this subset represent areas in the production line that will inevitably lose connection to the main line after the removal of high-risk paths. These are logical minefields that must be bypassed or specially handled when updating paths.

[0081] Step 143: Using the subset of process nodes with conflicting process logic as boundary constraints, perform connectivity analysis on the process flow relationship edges in the process association knowledge graph to determine the connected regions of process flow relationship edges in the process association knowledge graph that do not have topological intersection relationships with the subset of process nodes with conflicting process logic, and mark the connected regions of process flow relationship edges as the periphery of feasible paths.

[0082] Obtain the subset of process logic conflicting process nodes. Conflict Subsequently, this subset is used as a "forbidden zone" constraint. This is then applied to the original process-related knowledge graph. Origin In the middle, Subset Conflict All nodes in the graph, as well as all process flow edges with nodes as their endpoints, are marked as "unavailable". This is the process for completing the marking of the graph. Origin Perform maximum connected subgraph extraction.

[0083] In detail: starting with process node v Start Starting from the source node, perform a breadth-first traversal, strictly avoiding all nodes and edges marked as "unavailable" during the traversal. At the end of the traversal, the set of all successfully visited edges constitutes the connected region of process flow relationships that does not create topological intersections with the subset of process nodes that do not conflict with the process logic. This region is topologically a region with edges v. Start A subgraph of a directed acyclic graph rooted at a node whose boundary is defined by edges pointing to "unavailable" nodes is labeled as the periphery of feasible paths. Valid The feasible path perimeter represents a topological safety space where, under the current semantic constraints, any path reconstruction operation will not violate known technological disaster risks.

[0084] Step 144: Extract the process nodes contained within the periphery of the feasible path and generate a preliminary set of nodes for the proposed adopted path. The preliminary set of nodes for the proposed adopted path contains all process nodes within the periphery of the feasible path.

[0085] Periphery of the feasible path marked in step 143 Valid In this process, the unique identifiers of all process nodes contained within the process are extracted. Specifically, this involves collecting all node objects successfully visited during the breadth-first traversal in step 143. These node identifiers are then aggregated into a set to generate a preliminary set of proposed path nodes (Set). Node_Pre Set Node_Pre It includes all nodes within the safe boundary, but this set may contain some "dangling" nodes that are located inside the safe zone but have lost their effective connection to the mainline due to pruning operations.

[0086] Step 145: Perform node reachability verification on the preliminary set of proposed path nodes. Verify whether there is at least one continuous process flow path between any two process nodes in the preliminary set of proposed path nodes, which consists of process flow relationship edges within the periphery of feasible paths. Remove process nodes that do not have continuous process flow paths from the preliminary set of proposed path nodes to obtain the proposed path nodes.

[0087] For the initial set of nodes of the proposed adoption path (Set) Node_Pre Each process node v in chk Starting with process node v Start As the source point, in the periphery outside the feasible path Valid Perform a single-source shortest path search or reachability query within the edge set. If node v chk Unable to get from v Start By Periphery Valid If a node v is reached by a path formed by its internal edges, then that node is considered valid. chk There is no continuous process flow path, so remove it from Set Node_Pre Removed from the middle.

[0088] After this round of reachability verification, the nodes retained in the set all meet two conditions: they are located within the safety boundary and have an actual, valid connection to the production line starting point. The set of retained nodes constitutes the final proposed adoption path node set. Node_Adopt .

[0089] Step 146: Based on the boundary process flow relationship edges of the feasible path and the node topology position of the node of the proposed adopted path in the process association knowledge graph, generate the production line path update topology boundary constraints and the production line path update node range constraints.

[0090] First, analyze the periphery of feasible paths. Valid The boundary is formed. Boundary process flow relationships are defined as all edges e that satisfy the following condition. bound : edge e bound The set of edges E belongs to the original process-related knowledge graph, and its starting node is located in the Periphery. Valid Inside, while the endpoint is located in the Periphery. Valid External (i.e., belonging to a subset of process nodes with conflicting process logic or other unvisited areas). Collect all such boundary edges into set E. Boundary .

[0091] Define production line path update topology boundary constraints (Cons) Boundary For: During subsequent path updates, the creation of any path crossing E is prohibited. Boundary The new process flow relationship of any edge in E is also prohibited from being modified. Boundary The original flow direction attribute of the middle edge. Secondly, the production line path update node range constraint condition (Cons). NodeRange Defined as: all operations involving adding or deleting process nodes or reconnecting process flow relationships must involve process node objects that belong to the proposed adoption path node Set. Node_Adopt .

[0092] Step 147: Combine the production line path update topology boundary constraints and the production line path update node range constraints into a production line path update constraint set.

[0093] Cons Boundary With Cons NodeRange Two constraint objects are encapsulated into a structure or dictionary data structure called Constraint. Set This set of constraints will serve as the input parameters for the subsequent transition update processing algorithm, strictly limiting the search space boundary of the solution.

[0094] Step 150: Perform process transition update processing on the process association knowledge graph according to the production line path update constraint set to obtain the production line path update result. The production line path update result includes the updated process flow relationship edge and the update flow condition description information corresponding to the updated process flow relationship edge.

[0095] After obtaining a clear set of production line path update constraints (Constraint) Set Next, this step performs specific reconstruction operations on the original graph within the constrained solution space. The reconstruction process not only removes high-risk paths but also repairs the local connectivity broken by the removal operations, ultimately generating a production line path definition that can be directly deployed. The specific implementation process includes the following steps.

[0096] Step 151: Based on the production line path update topology boundary constraints in the production line path update constraint set, determine the topology boundary range in the process association knowledge graph that allows changes to process flow relationship edges. The topology boundary range is formed by the boundary process flow relationship edges surrounding the feasible path.

[0097] Parsing Constraint Set Cons in Boundary Constraints are used to extract the set E of boundary process flow relationships. Boundary The original process is associated with a knowledge graph. Origin In the adjacency list structure, E Boundary The edge object properties are set to "locked". Define the topological boundary range (Region) that allows change operations. Change For: Starting from the initial process node v Start Starting from the point of origin, the subgraph formed by all nodes reachable through non-locked process flow edges and all non-locked edges between these nodes constitutes the periphery of feasible paths. Valid It is a subset of , but its boundaries are more clearly defined, that is, strictly defined by locked boundary edges.

[0098] Step 152: Based on the production line path update node range constraints in the production line path update constraint set, determine the range of process nodes in the process association knowledge graph that are allowed to perform process node status change operations. The range of process nodes is completely consistent with the proposed path nodes.

[0099] Parsing Constraint Set Cons in NodeRange Constraints are imposed, and a list of node ranges is extracted. This list is the set of proposed adoption path nodes generated in step 145. Node_Adopt Load this list into an in-memory node filter. Before any subsequent node change operation (such as modifying node process constraint properties, creating or deleting edges pointing to the node), this node filter must be called for validation, only performing validation if the target node involved in the operation exists in the Set. Node_Adopt Only when the condition is met will the operation be allowed to continue.

[0100] Step 153: Prune the process flow relationship edges within the topological boundary range, removing process flow relationship edges within the topological boundary range that have a topological intersection relationship with the exclusionary association edges in the process shadow graph, and obtain the pruned process flow relationship edge set.

[0101] In Region Change Internally, iterate through all process flow relationships. internal For each e internal Query the result record generated when performing cross-checking in step 141. If edge e internal If the record shows that an edge is marked as an implicitly exclusive process flow relationship edge, then a pruning operation is performed. At the data structure level, the pruning operation calls the edge deletion primitive of the graph database or graph computing framework, removing the edge e. internal Logically remove from the current working graph. The removal operation does not cascade node deletions; it only affects edge connectivity. Traverse and process all Regions. Change After removing all edges within the pruning process flow relationship set, the remaining, un-deleted process flow relationship edges constitute the pruned process flow relationship edge set E. Pruned At this point, the graph retains only those "safe edges" that do not cross any process disaster regions in the semantic vector space.

[0102] Step 154: Perform process flow path reorganization on the pruned process flow relationship edge set. Within the scope of the process node, re-establish process flow relationship edges for process nodes that have lost the continuity of process flow paths in the pruned process flow relationship edge set. The flow condition description information of the re-established process flow relationship edge is generated based on the process constraint attributes of the two process nodes of the re-established process flow relationship edge.

[0103] Pruning removes high-risk edges, but it may also sever previously legitimate local flow paths, resulting in missing incoming or outgoing edges for some nodes. Reorganization aims to repair such breaks based on the principle of semantic consistency. Its detailed execution logic is as follows.

[0104] Step 1541: Perform process flow path continuity analysis on the pruned process flow relationship edge set to determine the list of missing process nodes in the inbound edges and the list of missing process nodes in the outbound edges.

[0105] Based on the set of edges E containing only the process flow relationships after pruning Pruned A temporary directed graph structure, traversing the range of process nodes Set Node_Adopt Each process node v in chg Query node v in the temporary graph. chg in-degree value Deg in If Deg in The value of is zero, and v chg The node identifier is not equal to the starting process node v Start If the identifier is v, then v chg Record to the list of missing process nodes on the incoming edge. LackIn In the middle. Meanwhile, query node v chg out-degree value Deg out If Deg out The value of is zero, and v chg If the node identifier is not equal to the preset termination process node identifier, then v will be... chg Record to the list of missing process nodes at the outgoing edge. LackOut middle.

[0106] Step 1542: Perform process constraint attribute matching on the missing process nodes in the list of missing process nodes and the process nodes within the scope of process nodes to filter out candidate preceding process nodes that satisfy the positive process flow relationship.

[0107] For the list of missing process nodes with inbound edges (List) LackIn Each node v in lackIn Traverse the range of process nodes in Set Node_Adopt Other nodes v in cand Get v cand process constraint attribute vector P cand The "output material status code range" component is obtained simultaneously, along with v. lackIn process constraint attribute vector P lackIn The "Input Material Status Code Range" component. Perform process constraint attribute matching: calculate the overlap rate of the two status code ranges. The overlap rate is defined as the length of the intersection interval of the two ranges multiplied by v. lackInInput the ratio of the lengths of the input range intervals. If the overlap rate is greater than the preset process compatibility threshold, and v cand With v lackIn There is no v in the temporary graph. lackIn Pointing to v cand The risk of loop failure will then be v cand The candidate preceding process node is determined to satisfy the positive process flow relationship and added to the candidate preceding process node list. CandPre .

[0108] Step 1543: Based on the consistency of the process semantic embedding vector of the candidate preceding process node with the process semantic embedding vector of the process missing edge process node, determine the target preceding process node from the candidate preceding process nodes, and establish a first type of reconstructed process flow relationship edge between the target preceding process node and the process missing edge process node.

[0109] In the process semantic vector space generated in step 120, extract candidate preceding process nodes v cand Embedding vector Emb* vcand And the missing process node v at the input edge lackIn Embedding vector Emb* vlackIn Calculate the difference vector Diff between these two vectors. cand =Emb* vlackIn -Emb* vcand Simultaneously, extract all processes from the original process association knowledge graph that are not marked as implicitly exclusive and whose start and end points are both in Set. Node_Adopt The initial semantic embedding vector of the process flow relationship edge within the process flow relationship. eij Calculate the arithmetic mean vector of these vectors and use it as the Base vector for the semantics of the forward process flow. Forward Calculate the difference vector Diff. cand With reference vector Base Forward The cosine similarity is calculated by dividing the dot product of the two vectors by the product of their magnitudes.

[0110] Traversing a List CandPre Among all candidate nodes, the candidate node that maximizes the cosine similarity value is selected as the target preceding process node v. optPre At the target preceding process node v optPre With missing entry edge process node v lackIn Create a new directed edge between them and label it as a first-type reconstruction process flow relationship edge. The flow condition description information of this new edge is given by v. optPre The "output material status code range" and v lackInIt is generated by combining the intersection parameter of the "input material status code range" and a default material temperature threshold range.

[0111] Step 1544: Perform process constraint attribute matching on the missing process nodes in the missing process node list and the process nodes within the process node range to filter candidate subsequent process nodes that satisfy the positive process flow relationship.

[0112] List of missing process nodes for outgoing edges LackOut Each node v in LackOut Iterate through the Set Node_Adopt Other nodes v in cand Get v LackOut The "output material status code range" and v cand The "input material status code range" is used to calculate the overlap rate between the two. If the overlap rate is greater than the preset process compatibility threshold, and there is no loop risk, then v will be... cand Mark the node as a candidate subsequent process and add it to the candidate subsequent process node list. CandPost .

[0113] Step 1545: Based on the consistency of the process semantic embedding vector of the candidate post-process node with the process semantic embedding vector of the missing out-edge process node in the process semantic transfer direction, determine the target post-process node from the candidate post-process nodes, and establish a second type of reconstructed process flow relationship edge between the missing out-edge process node and the target post-process node.

[0114] Extract candidate successor nodes v cand Embedding vector Emb* vcand And the missing edge process node v LackOut Embedding vector Emb* vLackOut Calculate the difference vector Diff. cand =Emb* vcand -Emb* vLackOut Calculate the difference vector Diff. cand With the semantic base vector of forward process flow Forward Cosine similarity. Traverse the List. CandPost Among all candidate nodes, the candidate node that maximizes the cosine similarity value is selected as the target subsequent process node v. optPost In the missing edge process node v LackOut With the target subsequent process node v optPost Create a new directed edge between them, mark it as a second type of reconstruction process flow relationship edge, and generate its flow condition description information in the same way as in step 1543.

[0115] Step 1546: Add the first type of reconstructed process flow relationship edges and the second type of reconstructed process flow relationship edges to the pruned process flow relationship edge set to obtain the process flow relationship edge set after process flow path reorganization.

[0116] All object references of the first type of reconstructed process flow relationship edges and the second type of reconstructed process flow relationship edges created in steps 1543 and 1545 are uniformly inserted into the pruned process flow relationship edge set E. Pruned In the data container. At this time, E Pruned The set includes both the original safe edges that are retained and the reconstructed edges that are added to repair connectivity, together forming the set of process flow relationship edges after the process flow path reorganization process.

[0117] Step 155: Perform process flow closed-loop verification on the process flow knowledge graph after re-establishing process flow relationship edges. Determine whether there is an isolated process node subgraph in the process flow knowledge graph that cannot be reached from the starting process node. If there is an isolated process node subgraph, perform process flow relationship edge completion processing on the isolated process node subgraph until all process nodes in the process flow knowledge graph are on at least one complete process flow path.

[0118] After edge reorganization, a temporary directed graph is reconstructed based on the latest edge set. Starting with the initial process node v... Start Starting from the source node, perform a depth-first traversal. After the traversal, check the range of the process nodes (Set). Node_Adopt Does there exist any unvisited process node v? iso If v exists iso This indicates that the reconstructed graph still contains isolated process node subgraphs that cannot be reached from the starting point. For each such isolated node v... iso Initiate the process flow relationship completion process. The completion process first finds the distance v. iso The most recently visited node v in the topology hop count near If v near If it exists, then try in v near With v iso Establish a new edge between them, following the semantic consistency principles in steps 1542 to 1545. If the topological boundary constraints (Cons) are updated due to the production line path, [the following applies]. Boundary If a new edge cannot be established because it crosses the locked boundary, then the v... iso Record this in the anomaly report list, and include a note in the final production line path update result explaining that the node cannot be connected to the main line due to hard boundary constraints and requires manual intervention. Repeat this process until Set Node_Adopt All reachable nodes were successfully connected.

[0119] Step 156: Mark the process flow relationship edges in the process association knowledge graph after the process flow relationship edge pruning and process flow path reorganization as updated process flow relationship edges, and mark the flow condition description information corresponding to the updated process flow relationship edges as updated flow condition description information.

[0120] Traverse the final determined set of graph edges E Final All edge objects are processed. A new version identifier is generated for each edge object, and its attribute data (including start node identifier, end node identifier, and flow condition description information) is serialized into structured data rows. The collection of these data rows constitutes the formal definition of the updated process flow relationship edge. At the same time, the flow condition description information field stored in each edge object is extracted and encapsulated separately as the updated flow condition description information.

[0121] Step 157: Combine the updated process flow relationship edge and the updated flow condition description information corresponding to the updated process flow relationship edge into the production line path update result.

[0122] The two datasets generated in step 156 are assembled according to a predefined exchange format (e.g., a JSON structure containing node arrays, edge arrays, and their attribute objects) to generate the final production line path update result. Update The result fully describes a new production line process flow logic optimized under both semantic and topological constraints. This result can be parsed by downstream manufacturing execution systems to update the routing table of equipment controllers or the graphical interface of process design terminals.

[0123] Step 210: Obtain the process flow execution records of the target production line within the preset time window, and determine the actual trigger frequency distribution status of the updated process flow relationship edge based on the process flow execution records.

[0124] This step, an optional enhancement following step 150, aims to validate and correct the updated path using actual production data. From the event acquisition module of the Manufacturing Execution System (MES), the process flow execution record table is queried based on the unique identifier of the target production line and a preset time window (e.g., a natural week). Each record in this table contains a trigger timestamp and a unique identifier for the process flow relationship edge. The query results are then grouped and aggregated statistically, using the edge's unique identifier as the grouping key, to calculate the value of each updated process flow relationship edge e. new The number of triggers within the window (Freq) e . All (Freq) e e new The distribution of key-value pairs is defined as the actual trigger frequency distribution state Dist.Freq This distribution reflects the heat of each path under the actual production cycle.

[0125] Step 220: Map the actual trigger frequency distribution state to the process semantic vector space to obtain the actual flow density distribution representation information in the process semantic vector space.

[0126] For the actual trigger frequency distribution state Dist Freq For each pair of data, the process flow relationship edge e is first located in the process semantic vector space. new Initial flow relation semantic embedding vector Emb eij The spatial location. In the process semantic vector space generated in step 120, Emb eij Corresponding to the connection node embedding vector Emb* vi With Emb* vj A directed line segment. The trigger frequency Freq. e As a weighting factor for the line segment, a kernel density estimation method is used to reconstruct a smooth density field across the entire d-dimensional space. Specifically: the space is divided into grids with a preset step size. For each grid center point x, the weighted sum of the density contributions of all edge segments to that point is calculated. The density contribution of a single edge segment is determined by integrating a kernel function centered on the segment with a bandwidth parameter h. This integral value is then multiplied by the weight Freq. e After calculating for all grid points, the resulting discrete density value matrix is ​​the Map representing the actual flow density distribution. RealFlow This information is presented as a thermal field, showing the clustering areas of actual logistics in the semantic space.

[0127] Step 230: Perform spatial distribution deviation measurement on the actual flow density distribution representation information and the spatial distribution state of the process semantic vector to obtain the process flow deviation distribution representation information. The process flow deviation distribution representation information is used to describe the degree of spatial offset between the actual execution path of the process and the semantic constraints of the process-related knowledge graph.

[0128] Extract the process semantic vector space distribution state generated in step 125. Dist Based on the cluster information and the distribution of node embedding vectors within each cluster, a theoretical semantic density field Map is constructed. Semantic The semantic density field at the cluster center Cen k The highest value is found at the discrete boundary location Bou. m The decay gradually decreases. For each grid point x in the process semantic vector space, its value in the Map is read. RealFlow Density value in Real（x） and its role in Map Semantic Density value inSem（x） Calculate the deviation Dev at that point. x .

[0129] Deviation x The calculation method is as follows: first calculate Den Real（x） with Den Sem（x） The absolute value of the difference, then divide the absolute value of the difference by Den. Sem（x） The sum of this and a small constant ε is used to avoid division by zero errors. The deviation Dev for all grid points. x The data is summarized to form a deviation matrix, which represents the deviation distribution information of the process flow. Dist Dev Dist Regions with median values ​​significantly higher than the average level indicate areas through which materials frequently flow in actual production, but which are marked as low-density or boundary restricted areas in the original process design semantics. This suggests potential path optimization opportunities or unmodeled process dependencies.

[0130] Step 240: Based on the process flow deviation distribution characterization information, the repulsion strength parameter of the repulsion association edge in the process shadow map is adjusted to update the semantic repulsion strength parameter corresponding to the repulsion association edge in the virtual connection state of process disaster, and an updated process shadow map is generated according to the updated semantic repulsion strength parameter.

[0131] Traverse every exclusive correlation edge e in the process shadow graph negsi Locate the corresponding line segment in the process semantic vector space. Calculate the Dev representation information of the process flow deviation distribution of all grid points traversed by this line segment. Dist Dev in the deviation x The average value of Avg Devsi The update formula for adjusting the repulsion strength parameter is defined as: the updated semantic repulsion strength parameter Rep'. si Equal to the original semantic rejection strength parameter Rep si Multiply by 1 and Avg Devsi The sum. If the actual deviation of the region traversed by a repulsive edge is large, it indicates that the logical constraints of that region are frequently challenged in reality, and its potential process disaster risk may be higher than the estimated value. Therefore, its repulsive strength is increased. If the deviation is close to zero, the original repulsive strength remains unchanged. The calculated Rep' si Assigned to e negsi The repulsion strength attribute field is updated, and the risk level field in the associated process disaster type description information is updated synchronously. Based on the updated node set and edge set, the updated process shadow graph is regenerated. ShadowNewThis graph will be used as a new constraint source in the next path update iteration, thus forming a closed-loop optimization system that evolves continuously with feedback from actual production data.

[0132] Step 310: Obtain the production line path update results of multiple historical versions corresponding to the target production line, and extract the updated process flow relationship edges and updated flow condition description information contained in the production line path update results of each historical version.

[0133] This step, as an optional enhancement following step 150, aims to leverage historical version information to assist in current path decisions. It involves accessing the enterprise's internal process data version control repository, which uses timestamps or tags. h Manage a snapshot of the process association knowledge graph for the target production line. Retrieve a list of all historical tags associated with the target production line identifier. Read each tag sequentially. h The archived production line path update result files are then processed. Each file is deserialized and parsed to extract the set E of process flow relationship edges defined in that historical version. Histh And the set of update flow condition description information C corresponding to each edge. Histh Organize this data into a time series table by version tag.

[0134] Step 320: Perform topological difference analysis on the updated process flow relationship edges in the production line path update results of each historical version to determine the edge substitution set of process flow relationship edges that exist in the same process node interval between different versions.

[0135] For any two versions that are adjacent in time in the time series table, let's call them earlier versions V. old Compared to recent version V new Perform topology difference analysis. Compare V old The process flow relationship edge set E Vold With V new The process flow relationship edge set E Vnew Identify that it exists only in E Vold Not existing in E Vnew The set E of "removed edges" Removed and only exists in E Vnew Not existing in E Vold The set of "newly added edges" E in Added For each edge e that is removed old In E Added Find the edge e that shares the same starting or ending process node with it. new If e old Connect nodes u and v, and e newSimilarly, to connect nodes u and v, simply connect (e) old e new ) is recorded as a set of edge substitution relationships. If e old Connect u and v, and in E Added If there are two edges, u to w and w to v, then (e old (u to w, w to v) is recorded as a set of multi-hop edge substitution relationships. All identified edge substitution relationships constitute the edge substitution set Set. Sub .

[0136] Step 330: Project each set of edge substitution relationships in the edge substitution set onto the process semantic vector space to obtain the first process semantic transfer trajectory corresponding to the process flow relationship edge before substitution and the second process semantic transfer trajectory corresponding to the process flow relationship edge after substitution.

[0137] In the process semantic vector space generated in step 120, locate the nodes and edges involved in each set of edge substitution relationships. For the edge e of the process flow relationship before substitution... old Directly read its initial flow relationship semantic embedding vector Emb eold This vector is the vector representation of the semantic transfer trajectory of the first process, Vec. Old For the replaced path, if it is a direct replacement of edge e... new Then read its embedding vector Emb enew Vec, the vector representation of the semantic transfer trajectory in the second process New If it is a multi-hop alternative path (u to w, w to v), then read edge e. uw Embedded vector euw With edge e wv Embedded vector e_wv Calculate the vector sum of these two vectors as Vec New The vector sum is calculated by adding the corresponding dimension components of the two vectors.

[0138] Step 340: Based on the consistency analysis results of the vector directions of the semantic transfer trajectories of the first and second processes in the semantic vector space of the process, the edge substitution set is classified according to process semantic compatibility to obtain the inheritable edge substitution set and the process semantic conflict edge substitution set.

[0139] For the set of edge substitutions Set Sub For each set of substitution relationships, calculate its first-step semantic transfer trajectory vector Vec. Old With the semantic transfer trajectory vector Vec of the second process New Cosine similarity SubCosine similarity is calculated by dividing the dot product of the two vectors by the product of their magnitudes. A preset semantic compatibility threshold, Thresh, is set. Compat This threshold is determined by the lower quartile of the cosine similarity distribution of edge change events that have not regressed across all historical versions. If Cos... Sub Greater than Thresh Compat This indicates that although the topological structure of the edges has changed, the direction of process state transition in the semantic space is highly consistent, and the process intention has continuity. Therefore, this set of substitution relationships is categorized into the inheritable edge substitution set Set. Compat Conversely, if Cos Sub Less than or equal to Thresh Compat This indicates a significant shift in the semantic transfer direction between the old and new paths, potentially indicating a fundamental change in the process logic. Therefore, this set of substitution relationships is categorized into the process semantic conflict edge substitution set Set. ConflictSub .

[0140] Step 350: Integrate the edge substitution relationships in the inheritable edge substitution set into the process association knowledge graph corresponding to the current production line path update result, so as to generate the production line path update result after cross-version process flow path inheritance.

[0141] Extract the process association knowledge graph corresponding to the latest production line path update result. Curr Traverse the Set of inheritable edges. Compat Replace each pair of edges in the graph with a relation. For each relation, locate the graph. Curr The edge e corresponding to the replaced path in the topology curr Information on update flow conditions from historical versions (C) Histh In the middle, extract and replace the following e new Implicit parameters that were previously associated but are not explicitly documented in the current version (such as verified material wait time window caps or equipment scheduling priority masks) will be merged into the Graph as additional attributes. Curr Middle edge e curr Within the attribute set. After completing the merging of all inheritable relationships, the Graph will be... Curr The repackaged output is the production line path update result after cross-version process flow path inheritance. Final This result not only includes on-the-fly optimization based on semantic and topological constraints, but also incorporates effective empirical parameters that have been verified in actual production from historical versions, improving the robustness and practicality of path updates.

[0142] This application's embodiments obtain the process association knowledge graph corresponding to the target production line and perform semantic structure reverse deconstruction processing on it to generate the process semantic vector space distribution state. This allows discrete process symbols to be mapped into a semantic manifold with continuous distance metrics, thereby revealing the implicit semantic associations and distribution density between process nodes. Then, based on the process semantic vector space distribution state, negative sample topology expansion processing is performed to generate a process shadow graph containing exclusive association edges. This process shadow graph explicitly defines the logical forbidden zones in the production line reconstruction process. By performing logical conflict scanning processing on the process shadow graph and the process association knowledge graph to determine the periphery of feasible paths, a set of production line path update constraints is generated, achieving precise definition of the path change safety boundary under global semantic consistency constraints. Finally, process transition update processing is performed on the process association knowledge graph based on the production line path update constraint set. The resulting production line path update can maintain production line topology connectivity while avoiding process disaster risks, improving the logical rigor and semantic rationality of the production line path update operation.

[0143] It is understandable that, for the graph neural network training and energy function optimization process involved in semantic vector space generation, those skilled in the art can combine the random walk sampling strategy and negative sampling loss function in open-source graph embedding frameworks such as Word2Vec and Node2Vec, and use existing force-directed layout algorithms and t-SNE dimensionality reduction visualization methods in libraries such as PyTorch Geometric or D3.js to iteratively correct the initial embedding vector and calculate the spatial density distribution, thereby obtaining the process semantic vector space distribution state that conforms to the preset distance constraints.

[0144] For the shadow graph generation, logical conflict scanning, and path reconstruction processes involved in the embodiments of this application, those skilled in the art can refer to the path expansion algorithm and reachability analysis function in the Cypher query language of the Neo4j graph database, combined with classic graph theory algorithms such as Dijkstra's shortest path search or A* heuristic search, to perform connected component decomposition and missing path completion processing on the pruned process flow relationship edge set. Simultaneously, a process parameter matching strategy based on a rule engine can be introduced, utilizing a preset material state coding mapping table and equipment compatibility matrix to automatically generate flow condition description information for reconstructing process flow relationship edges.

[0145] For the feedback correction process based on actual flow data in the embodiments of this application, those skilled in the art can use message middleware such as Apache Kafka or RabbitMQ to collect real-time event streams of the manufacturing execution system, and use the aggregation analysis capabilities of Elasticsearch to statistically model the frequency of process flow. Furthermore, the kernel density estimation algorithm module in the Scikit-learn library can be used to map the frequency distribution to a continuous density field in the semantic space, and the repulsion strength parameter can be adaptively adjusted by calculating distribution deviation metrics such as KL divergence or Wasserstein distance.

[0146] Please see Figure 2 The figure is a schematic diagram of the basic structure of a production line path update system 200 provided in an embodiment of this application. The production line path update system 200 includes: a processor 201; a storage device 202 on which a computer program 2020 is stored; and a network interface 203 for providing network communication functions. When the computer program 2020 is executed by the processor 201, the processor 201 implements any of the production line path update methods that combine process association knowledge graphs.

[0147] Please see Figure 3 This application provides a functional block diagram of a production line path updating device, which includes: The knowledge graph acquisition module is used to acquire the process association knowledge graph corresponding to the target production line; The semantic reverse deconstruction module is used to perform semantic structure reverse deconstruction on the process association knowledge graph to obtain the process semantic vector space and the process semantic embedding vectors corresponding to the process nodes in the process semantic vector space, and generate the process semantic vector space distribution state based on the process semantic vector space and the process semantic embedding vectors. The shadow graph generation module is used to perform negative sample topology expansion processing based on the spatial distribution state of the process semantic vector, to obtain a set of negative sample topology expanded process nodes and their corresponding virtual connection states of process disasters, and to generate a process shadow graph by combining the set of negative sample topology expanded process nodes and the virtual connection states of process disasters. The process shadow graph contains the exclusive association edges between each negative sample topology expanded process node in the set of negative sample topology expanded process nodes and the process nodes in the process association knowledge graph. The logic conflict scanning module is used to perform logic conflict scanning processing on the process shadow graph and the process association knowledge graph, determine the feasible path periphery of the process flow relationship edge in the process association knowledge graph and the proposed adoption path nodes contained within the feasible path periphery, and generate a production line path update constraint set based on the feasible path periphery and the proposed adoption path nodes contained within the feasible path periphery. The production line path update module is used to perform process transition update processing on the process association knowledge graph according to the production line path update constraint set, and obtain the production line path update result. The production line path update result includes the updated process flow relationship edge and the update flow condition description information corresponding to the updated process flow relationship edge.

[0148] Based on the above, a readable storage medium is provided, on which a program or instructions are stored, and when the program or instructions are executed by a processor, the steps of the above method are implemented.

[0149] Furthermore, it should be noted that this application also provides a computer program product, which may include a computer program that can be stored in a computer-readable storage medium. The processor of the production line path update system reads the computer program from the computer-readable storage medium, and the processor can execute the computer program, causing the production line path update system to perform the aforementioned... Figure 1 The methods described in the corresponding embodiments are already known, and therefore will not be repeated here. Furthermore, the beneficial effects of using the same method will also not be repeated. For technical details not disclosed in the computer program product embodiments related to this application, please refer to the description of the method embodiments of this application.

[0150] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems or apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple, and relevant parts can be referred to the method section.

Claims

1. A production line path updating method that combines a procedure-related knowledge graph, characterized by, The method includes: Obtain the process association knowledge graph corresponding to the target production line; The semantic structure of the process-related knowledge graph is reversed to obtain the process semantic vector space and the process semantic embedding vectors corresponding to the process nodes in the process semantic vector space. The process semantic vector space distribution state is generated based on the process semantic vector space and the process semantic embedding vectors. Based on the spatial distribution state of the process semantic vector, negative sample topology expansion processing is performed to obtain a set of negative sample topology expansion process nodes and their corresponding virtual connection states of process disasters. A process shadow graph is generated by combining the set of negative sample topology expansion process nodes and the virtual connection states of process disasters. The process shadow graph contains the exclusive association edges between each negative sample topology expansion process node in the set of negative sample topology expansion process nodes and the process nodes in the process association knowledge graph. Logical conflict scanning is performed on the process shadow graph and the process association knowledge graph to determine the feasible path periphery of the process flow relationship edge in the process association knowledge graph and the proposed adoption path nodes contained within the feasible path periphery. A production line path update constraint set is generated based on the feasible path periphery and the proposed adoption path nodes contained within the feasible path periphery. The process association knowledge graph is updated by performing process transition update processing based on the production line path update constraint set to obtain the production line path update result. The production line path update result includes the updated process flow relationship edge and the update flow condition description information corresponding to the updated process flow relationship edge. The step of performing negative sample topology expansion processing based on the spatial distribution state of the process semantic vector to obtain a negative sample topology expanded process node set and its corresponding process disaster virtual connection state, and combining the negative sample topology expanded process node set and the process disaster virtual connection state to generate a process shadow map, includes: Based on the vector space density distribution description information in the process semantic vector space distribution state, a low vector space density region in the process semantic vector space is determined. The low vector space density region is a continuous spatial region in the process semantic vector space where the vector space density is lower than a preset vector space density threshold. In the low vector space density region, negative sample process node sampling is performed to generate a negative sample topology extended process node set. Each negative sample topology extended process node in the negative sample topology extended process node set corresponds to a negative sample process semantic embedding vector in the process semantic vector space. Based on the vector space distance between the semantic embedding vector of the negative sample process and the semantic embedding vector of the process node in the process association knowledge graph, the semantic repulsion strength parameter between each negative sample topology expansion process node and the process node in the process association knowledge graph is determined. The semantic repulsion strength parameter is negatively correlated with the vector space distance. Based on the semantic exclusion strength parameter, a virtual connection state of process disasters is constructed between the negative sample topology expansion process node set and the process nodes in the process association knowledge graph. The virtual connection state of process disasters includes the exclusionary association edges between the negative sample topology expansion process nodes and the process nodes in the process association knowledge graph, as well as the process disaster type description information corresponding to the exclusionary association edges. The process association knowledge graph is subjected to topological structure mirroring processing. Based on the mirrored topological structure of the process nodes, the negative sample topology is embedded to expand the process node set and generate a process shadow graph.

2. The method of claim 1, wherein, The step of performing semantic structure reverse deconstruction on the process-related knowledge graph to obtain the process semantic vector space and the process semantic embedding vectors corresponding to the process nodes in the process semantic vector space, and generating the process semantic vector space distribution state based on the process semantic vector space and the process semantic embedding vectors, includes: Extract the process constraint attributes of each process node and the flow condition description information of each process flow relationship edge in the process association knowledge graph, and perform attribute semantic decomposition and condition semantic decomposition respectively to obtain the attribute semantic atom set containing attribute semantic atoms and their attribute semantic atom association strength, and the condition semantic atom set containing condition semantic atoms and their condition semantic atom association strength. The attribute semantic atom set and the condition semantic atom set are input into a pre-constructed graph embedding transformation structure and vectorized projection processing is performed to obtain the initial process semantic embedding vector of the process node and the initial flow relationship semantic embedding vector of the process flow relationship edge. Based on the initial flow relationship semantic embedding vector, the initial process semantic embedding vector is subjected to vector space position migration processing, so that the process semantic embedding vector of the process node with process flow relationship edge satisfies the preset vector space distance constraint condition in the process semantic vector space. The process semantic embedding vectors, after vector space location migration processing, are spatially distributed and reorganized according to the original connection topology in the process association knowledge graph to generate a process semantic vector space. The spatial density distribution of the process semantic vector space is analyzed to determine the spatial cluster center location information and the spatial discrete boundary location information. Based on the spatial cluster center location information and the spatial discrete boundary location information, the process semantic vector space distribution state is generated.

3. The method of claim 1, wherein, The process of performing topological mirroring on the process-related knowledge graph, and embedding negative sample topology to expand the process node set according to the mirrored topological structure of the process nodes to generate a process shadow graph includes: The process association knowledge graph is subjected to topological structure mirroring to generate a process node mirror topological structure of the process association knowledge graph. The process node mirror topological structure contains the topological position mapping relationship of each process node in the process association knowledge graph. The negative sample topology expansion process node set is embedded into the process node mirror topology structure according to the topological position mapping relationship in the process node mirror topology structure, so that each negative sample topology expansion process node forms an exclusive association edge correspondence with the process node with the maximum semantic repulsion strength parameter in the process association knowledge graph; A process shadow graph is generated based on the correspondence of the exclusionary association edges and the description information of the process disaster type. The process node set of the process shadow graph includes the process node set of the negative sample topology expansion and the process nodes in the process association knowledge graph. The process node relationship edge set of the process shadow graph includes the process flow relationship edge in the process association knowledge graph and the exclusionary association edge.

4. The method of claim 1, wherein, The step of performing logical conflict scanning processing on the process shadow graph and the process association knowledge graph, determining the feasible path perimeter of the process flow relationship edge in the process association knowledge graph and the proposed adoption path nodes contained within the feasible path perimeter, and generating a production line path update constraint set based on the feasible path perimeter and the proposed adoption path nodes contained within the feasible path perimeter includes: Map the exclusionary association edges in the process shadow graph to the process node connection topology of the process association knowledge graph, determine the process flow relationship edges in the process association knowledge graph that have a topological intersection relationship with the exclusionary association edges, and mark the process flow relationship edges that have a topological intersection relationship as implicit exclusionary process flow relationship edges. Path blocking analysis is performed on the implicit exclusion process flow relationship edges to determine the process node subsets in the process association knowledge graph that cannot form a complete process flow path due to the existence of implicit exclusion process flow relationship edges, and the process node subsets are marked as process logic conflict process node subsets. Using the subset of process nodes with conflicting process logic as boundary constraints, the process flow relationship edges in the process association knowledge graph are analyzed for connectivity. The connected regions of process flow relationship edges in the process association knowledge graph that do not have topological intersection relationships with the subset of process nodes with conflicting process logic are determined, and the connected regions of process flow relationship edges are marked as the periphery of feasible paths. Extract the process nodes contained within the periphery of the feasible path to generate a preliminary set of proposed path nodes, which contains all process nodes within the periphery of the feasible path. The preliminary set of proposed adoption path nodes is subjected to node reachability verification to verify whether there is at least one continuous process flow path between any two process nodes in the preliminary set of proposed adoption path nodes, which is composed of process flow relationship edges within the periphery of feasible paths. Process nodes that do not have continuous process flow paths are removed from the preliminary set of proposed adoption path nodes to obtain the proposed adoption path nodes. Based on the boundary process flow relationship edges of the feasible path and the node topology position of the node of the proposed adopted path in the process association knowledge graph, generate the production line path update topology boundary constraints and the production line path update node range constraints. The production line path update topology boundary constraints and the production line path update node range constraints are combined into a production line path update constraint set.

5. The method of claim 4, wherein, The path blocking analysis of the implicit exclusion process flow relationship edges determines the subset of process nodes in the process association knowledge graph that cannot form a complete process flow path due to the existence of implicit exclusion process flow relationship edges, including: Extract all process flow relationship edges marked as implicitly exclusive process flow relationship edges from the process association knowledge graph, and generate a set of implicitly exclusive process flow relationship edges. Each implicitly exclusive process flow relationship edge in the set corresponds to a process node that is the starting point of the implicitly exclusive process flow relationship edge and a process node that is the ending point of the implicitly exclusive process flow relationship edge. Remove all implicitly exclusive process flow relationship edges from the set of implicitly exclusive process flow relationship edges in the process association knowledge graph to obtain a topological snapshot of the process association knowledge graph after removing the implicitly exclusive process flow relationship edges. The topological snapshot of the process association knowledge graph only retains process flow relationship edges that are not marked as implicitly exclusive process flow relationship edges. The process association knowledge graph topology snapshot is subjected to connected component decomposition processing, which divides the process nodes in the process association knowledge graph topology snapshot into several connected components. There is at least one continuous process flow path between any two process nodes in each connected component, which is composed of process flow relationship edges in the process association knowledge graph topology snapshot. There is no process flow path between different connected components. Extract the starting process node from the process association knowledge graph, locate the starting connected component containing the starting process node among the several connected components, mark the starting connected component as the main connected component, and mark the remaining connected components that do not contain the starting process node as isolated connected components. Extract all process nodes contained in the isolated connected component to generate a preliminary set of process logic conflict process nodes. Perform reverse path tracing processing on each process node in the preliminary set of process logic conflict process nodes to determine whether the process node is connected to the process node in the main connected component through at least one process flow path containing an edge of implicit exclusion process flow relationship before removing the implicit exclusion process flow relationship edge. If such a process flow path exists, the process node is retained in the preliminary set of process logic conflict process nodes. If such a process flow path does not exist, the process node is removed from the preliminary set of process logic conflict process nodes. The initial set of process logic conflicting process nodes after reverse path tracing is marked as a subset of process logic conflicting process nodes.

6. The method of claim 1, wherein, The step of performing process transition update processing on the process association knowledge graph based on the production line path update constraint set to obtain the production line path update result includes: Based on the production line path update topology boundary constraints in the production line path update constraint set, the topology boundary range in the process association knowledge graph that allows for changes to process flow relationship edges is determined. The topology boundary range is formed by the boundary process flow relationship edges surrounding the feasible path. Based on the production line path update node range constraints in the production line path update constraint set, determine the range of process nodes in the process association knowledge graph that are allowed to perform process node status change operations. The range of process nodes is completely consistent with the proposed path nodes. The process flow relationship edges within the topological boundary range are pruned, and process flow relationship edges within the topological boundary range that have a topological intersection relationship with the exclusionary association edge in the process shadow map are removed, resulting in a pruned set of process flow relationship edges; The process flow path reorganization process is performed on the pruned process flow relationship edge set. Within the scope of the process node, process flow relationship edges that have lost the continuity of process flow path are re-established for process nodes in the pruned process flow relationship edge set. The flow condition description information of the re-established process flow relationship edge is generated based on the process constraint attributes of the two process nodes of the re-established process flow relationship edge. After re-establishing the process flow relationship edges, perform process flow closed-loop verification on the process flow knowledge graph to determine whether there is an isolated process node subgraph that cannot be reached from the starting process node. If there is an isolated process node subgraph, perform process flow relationship edge completion on the isolated process node subgraph until all process nodes in the process flow knowledge graph are on at least one complete process flow path. The process flow relationship edges in the process association knowledge graph after the process flow relationship edge pruning and process flow path reorganization are marked as updated process flow relationship edges, and the flow condition description information corresponding to the updated process flow relationship edges is marked as updated flow condition description information. The updated process flow relationship edge and the corresponding updated flow condition description information are combined to form the production line path update result.

7. The method of claim 6, wherein, The process flow path reorganization process on the pruned process flow relationship edge set, which involves re-establishing process flow relationship edges for process nodes that have lost process flow path continuity within the pruned process flow relationship edge set within the scope of the process nodes, includes: Perform continuity analysis on the process flow relationship edge set after pruning to determine the list of missing process nodes in the inbound edge and the list of missing process nodes in the outbound edge. The process constraint attributes of the missing process nodes in the list of missing process nodes are matched with the process nodes within the scope of process nodes to filter candidate preceding process nodes that satisfy the positive process flow relationship. Based on the consistency of the process semantic embedding vector of the candidate preceding process node with the process semantic embedding vector of the process node with missing incoming edge, the target preceding process node is determined from the candidate preceding process nodes, and a first-type reconstructed process flow relationship edge is established between the target preceding process node and the process node with missing incoming edge. The process constraint attributes of the missing process nodes in the list of missing process nodes are matched with the process nodes within the scope of process nodes to filter candidate subsequent process nodes that satisfy the positive process flow relationship. Based on the consistency of the process semantic embedding vector of the candidate successor process node with the process semantic embedding vector of the missing outgoing edge process node in the process semantic transfer direction, the target successor process node is determined from the candidate successor process nodes, and a second type of reconstruction process flow relationship edge is established between the missing outgoing edge process node and the target successor process node. Add the first type of reconstructed process flow relationship edges and the second type of reconstructed process flow relationship edges to the pruned process flow relationship edge set to obtain the process flow relationship edge set after process flow path reorganization.

8. A production line path updating system characterized by comprising: include: A processor; a storage device having a computer program stored thereon; a network interface for providing network communication functions; when the computer program is executed by the processor, the processor enables the processor to implement the production line path update method combining process association knowledge graph as described in any one of claims 1-7.

9. A readable storage medium, characterized by, The readable storage medium stores a program or instructions that, when executed by a processor, implement the production line path update method combining process association knowledge graphs as described in any one of claims 1-7.

Images