A Dynamic Reasoning Method and System Based on Sparse Activation-Based Process Knowledge Graph

Abstract

To address the problems of long execution time, high computational cost, and inaccurate results in existing inference methods based on large-scale, dynamically updated process knowledge graphs, this invention proposes a dynamic inference method and system based on sparse activation of process knowledge graphs. Based on the expected goals of actual production, relationships associated with these goals are constructed in the knowledge graph. Sub-knowledge graphs associated with these goals are extracted, and the relation weight matrix of each sub-knowledge graph is calculated and assigned to it, resulting in a multi-relation weight model. Then, a sparse activation function is used to extract a sparse function activation subgraph from the multi-relation weight model, removing redundant information in the knowledge graph that is not closely related to current production, significantly reducing the number of nodes and relations in the knowledge graph. Finally, the inference algorithm is run on the sparse activation function subgraph, greatly reducing the running time and computational resources required.

Description

Technical Field

[0001] This invention relates to a dynamic reasoning method and system for process knowledge graphs. Background Technology

[0002] Process knowledge is crucial for ensuring the quality of complex products. A process knowledge graph is a structured semantic knowledge base that stores and represents entities (such as materials, equipment, and process parameters) and their relationships (such as the relationship between materials and process parameters, and the relationship between equipment and process parameters) in the form of a graph. Therefore, process knowledge graphs can help companies and researchers better understand and manage complex process knowledge, promote process innovation and optimization, and provide support for ensuring and improving the quality of complex products.

[0003] To adapt to actual industrial needs, process knowledge graphs for complex products are characterized by large scale and dynamic updates. When using existing reasoning methods based on such process knowledge graphs to obtain reasoning results that can guide process design, the following problems exist:

[0004] 1. The process reasoning algorithm takes a long time to run;

[0005] Based on large-scale process knowledge graphs, existing process reasoning algorithms search and run on the massive number of nodes and complex relationships in the global process knowledge graph. Because they need to traverse all nodes and relationships globally, even for simple reasoning tasks, the process reasoning algorithms still suffer from global traversal, low efficiency, and long processing time.

[0006] 2. The computational cost required for process reasoning algorithms is high;

[0007] As the scale of the process knowledge graph gradually expands, the number of nodes in the process knowledge graph continues to increase, resulting in an exponential growth in the number of relationships. This means that the inference algorithm needs to handle complex and massive relationships during runtime, especially multi-hop inference, which requires higher performance computing resources and results in high computational costs.

[0008] 3. The reasoning results of the process reasoning algorithm are inaccurate;

[0009] Since most process knowledge graphs are static, and existing reasoning is also mostly static, they cannot be closely aligned with actual production. Furthermore, applying static reasoning to dynamic knowledge graphs can lead to a decrease in the accuracy of the reasoning results. Summary of the Invention

[0010] To address the technical problems of existing reasoning methods, such as long running time, high computational cost, and inaccurate reasoning results when performing process reasoning based on large-scale, dynamically updated process knowledge graphs, this invention proposes a dynamic reasoning method and system based on sparse activation process knowledge graphs.

[0011] The technical solution of the present invention is as follows:

[0012] The dynamic reasoning method for process knowledge graphs based on sparse activation is unique in that it includes the following steps:

[0013] Step 1: Construct a multi-relationship weight model;

[0014] Step 1.1: Set the desired goal, and construct multiple relationships associated with the desired goal in the current processing technology knowledge graph and add relationship attributes;

[0015] Step 1.2: Calculate the weight matrix of the multiple relations between any two different nodes related to the multiple relations constructed in Step 1.1 in the current processing technology knowledge graph;

[0016] Step 1.2.1: Extract sub-knowledge graphs from the current processing technology knowledge graph that are only associated with the multiple relations, and construct a diagonal matrix between any two different nodes in each sub-knowledge graph. and an attribute feature matrix

[0017]

[0018] Among them, the diagonal matrix The size is m ij ×m ij m ij The number of efficiency optimization relationships between nodes i and j; the size of the attribute feature matrix is ​​m. ij ×1; x1, x2, ..., Let i and j be the first and second efficiency optimization relationships, respectively, ..., the m-th efficiency optimization relationship between nodes i and j. ij The attribute feature values ​​of the efficiency optimization relationship are known quantities;

[0019] Step 1.2.2: Calculate the diagonal matrix and attribute feature matrix The product of these products yields a product of size m. ij A matrix H of size ×1 ij ;

[0020] Step 1.2.3: Convert matrix H ij The elements in the matrix H are normalized, that is, the matrix H is normalized. ij Each element value in the matrix is ​​updated to its reciprocal, resulting in matrix H. i ′ j ;

[0021] Step 1.2.4: Convert matrix H i ′j Each element is subtracted from 1 to obtain the weight matrix W of the efficiency optimization relationship. ij ;

[0022] Step 1.3: Construct a multi-relationship weight model;

[0023] Assign the element values ​​of the weight matrix of the efficiency optimization relation obtained in step 1.2 to the corresponding relation in the current sub-knowledge graph, update the current sub-knowledge graph, and the sub-knowledge graph containing relation weight information after the update is the multi-relation weight model.

[0024] Step 2: Set a weight threshold, and use the weight threshold and the multi-relation weight model as input to the sparse activation function. Use the sparse activation function to extract the sparse function activation subgraph from the multi-relation weight model obtained in Step 1.

[0025] Step 3: Select an inference algorithm based on the desired goal set in Step 1.1, run the selected inference algorithm in the sparse function activation subgraph, and obtain the inference result;

[0026] Step 4: Determine whether the optimal process solution corresponding to the current reasoning result meets the production requirements. If it does, the process ends; otherwise, proceed to step 5.

[0027] Step 5: Based on the actual situation of the current production workshop, including the status of materials, cutting tools, processing features, and equipment, dynamically update the current large-scale process knowledge graph. After the update is completed, return to Step 1.

[0028] Furthermore, the desired objective mentioned in step 1.1 is efficiency optimization, quality optimization, or a combination of both; correspondingly, the multiple relationships are efficiency optimization relationships, quality optimization relationships, or a combination of both.

[0029] Furthermore, the weight threshold in step 2 is 0.7-0.9.

[0030] Furthermore, the method for extracting the sparse activation function subgraph from the multi-relation weight model obtained in step 1 in step 2 is as follows:

[0031] The sparse activation function compares the weight value of each relation in a multi-relation weight model with a preset weight threshold, and determines whether to activate the attribute feature value of each relation based on the comparison result.

[0032] If the weight value of a certain relation is greater than or equal to the weight threshold, the weight value of the relation is multiplied by 1 and used as the updated weight value to activate its attribute feature value.

[0033] If the weight value of a relation is less than the weight threshold, the weight value of the relation is multiplied by 0 and used as the updated weight value, so that its attribute feature value is not activated.

[0034] Once the weight values ​​of all relations in the multi-relation weight model have been updated, the multi-relation weight model with updated weight information is then called the sparse function activation subgraph.

[0035] Furthermore, in step 3, if the desired objective is to optimize efficiency, then the inference algorithm is an efficiency optimization algorithm; if the desired objective is to optimize quality, then the inference algorithm is a quality optimization algorithm; if the desired objective is to balance efficiency and quality, then the inference algorithm is an optimization algorithm that balances efficiency and quality.

[0036] Furthermore, the efficiency optimization algorithm is the Dijkstra's shortest path algorithm; the quality optimization algorithm is the PageRank algorithm; and the algorithm that balances efficiency and quality is either the Dijkstra's shortest path algorithm or the PageRank algorithm.

[0037] Furthermore, the method for determining whether the preferred process solution corresponding to the current reasoning result meets the production requirements in step 6 is as follows:

[0038] If the desired objective is to optimize efficiency, then based on the actual processing efficiency of the current production workshop, estimate whether the total time spent by the optimal process solution corresponding to the current reasoning result meets the customer's delivery time requirements. If yes, it indicates that the current reasoning result meets the production requirements; otherwise, it indicates that the current reasoning result does not meet the production requirements.

[0039] If the desired goal is quality optimization, then based on the actual processing capacity of the current production workshop, estimate whether the current production workshop can meet the processing accuracy required by the optimal process scheme corresponding to the current reasoning result, and determine whether the processing accuracy of the optimal process scheme can meet the processing accuracy required by the customer. If both are true, it indicates that the current reasoning result meets the production requirements; otherwise, it indicates that the current reasoning result does not meet the production requirements.

[0040] If the desired outcome is to balance efficiency and quality, then the time taken by the optimal process solution corresponding to the current reasoning result is estimated based on the actual processing efficiency of the current production workshop to see if it meets the customer's delivery time requirements. At the same time, the actual processing capacity of the current production workshop is estimated to see if the current production workshop can meet the processing accuracy required by the optimal process solution, and it is determined whether the processing accuracy of the optimal process solution can meet the customer's processing accuracy requirements. If all of the above are true, it indicates that the current reasoning result meets the production requirements; otherwise, it indicates that the current reasoning result does not meet the production requirements.

[0041] The present invention also provides a dynamic reasoning system for process knowledge graphs based on sparse activation, including a processor and a storage medium, wherein a computer program is stored on the storage medium; the special feature is that the computer program executes the above-mentioned dynamic reasoning method for process knowledge graphs based on sparse activation when it is run.

[0042] Compared with the prior art, the beneficial effects of the present invention are:

[0043] 1. This invention constructs (adds) relationships (such as efficiency optimization relationships, quality optimization relationships, or relationships that balance efficiency and quality) related to the expected goals in the actual production process within a knowledge graph. Then, it extracts sub-knowledge graphs associated with the constructed expected goals, calculates the relationship weight matrix of these sub-knowledge graphs, and assigns the elements (essentially weight values) of the relationship weight matrix to the sub-knowledge graphs, resulting in a multi-relation weight model. Next, it uses a sparse activation function to extract a sparse activation subgraph from the multi-relation weight model, removing redundant information in the process knowledge graph that is not closely related to the current production reality. This significantly reduces the number of nodes and relationships in the process knowledge graph. Finally, it runs an inference algorithm on the extracted sparse activation subgraph, eliminating the need for the inference algorithm to traverse the entire knowledge graph. Therefore, the running time of the inference algorithm is significantly reduced, and the required computational resources are greatly decreased.

[0044] 2. After obtaining the reasoning result, this invention will determine whether the reasoning result meets the production needs. If it does not, it will dynamically update the current process knowledge graph according to the actual production situation. Then, it will perform process reasoning based on the updated process knowledge graph. This process is repeated iteratively to improve the accuracy of reasoning and finally obtain the optimal process solution that meets the production needs. Attached Figure Description

[0045] Figure 1 This is a flowchart of the method of the present invention.

[0046] Figure 2 When the desired objective is efficiency optimization, the multi-relationship weight model obtained in this embodiment of the invention (the pink lines with arrows represent efficiency optimization relationships, and the numbers on the pink lines are the weights of the efficiency optimization relationships).

[0047] Figure 3 When the desired goal is quality optimization, the multi-relationship weight model obtained in this embodiment of the invention (the blue lines with arrows represent quality optimization relationships, and the numbers on the blue lines are the weights of the quality optimization relationships).

[0048] Figure 4When the desired goal is to balance efficiency and quality, the multi-relationship weight model obtained in this embodiment of the invention (the pink line with arrows represents the efficiency-preferred relationship, the blue line with arrows represents the quality-preferred relationship, and the numbers on the lines are the weights of the relationships).

[0049] Figure 5 The sparse function activation subgraph obtained in the embodiments of the present invention is when the desired objective is to optimize efficiency.

[0050] Figure 6 The sparse function activation subgraph obtained in the embodiments of the present invention is for the desired objective of quality optimization.

[0051] Figure 7 The sparse function activation subgraph obtained in the embodiments of the present invention is to achieve a balance between efficiency and quality.

[0052] Figure 8 These are visualization examples of the reasoning results obtained by the embodiments of the present invention under three different expected objectives: (a) corresponds to optimal efficiency, (b) corresponds to optimal quality, and (c) corresponds to a balance between efficiency and quality.

[0053] Figure 9 This is a schematic diagram illustrating the principle of dynamically updating the process knowledge graph in this invention. Detailed Implementation

[0054] Large-scale process knowledge graphs are characterized by numerous nodes and complex relationships. The process knowledge graph for complex aerospace products is one such example. These types of process knowledge graphs are typically constructed based on process triples containing a massive number of entities and relationships.

[0055] A machining process ternary is a canonical expression consisting of a head entity, a tail entity, and the relationships between them. A machining process ternary can be expressed in the form of (head entity) - [relationship] -> (tail entity), for example, (forging) - [efficiency optimization {weight:0.897}] -> (rough turning) and (rough turning) - [efficiency optimization {weight:0.533}] -> (heat treatment).

[0056] The machining process triple, containing a massive number of entities and relationships, is constructed based on multi-stage, multi-source, and multi-modal process data and knowledge. Multi-stage refers to the fact that the process data and knowledge involve the part design, machining, and service stages. Multi-source means that the part's process data and knowledge come from various sources, including designers, process engineers, production personnel, and quality inspectors. Multi-modal means that the process data and knowledge encompass different modalities, such as text, tables, audio, and video.

[0057] Nodes, relationships, and attributes (relationship attributes and node attributes) are important components of a process knowledge graph. Through the combination of nodes and relationships, a process knowledge graph can comprehensively display complex technological processes. In a machining process knowledge graph, nodes are typically represented by circles, and relationships between nodes are represented by directed arrows.

[0058] The following detailed description of the present invention, using the knowledge graph of the manufacturing process of an annular casing of an aerospace compressor as an example, with reference to the accompanying drawings, aims to make the technical solution and advantages of the present invention clearer and easier to understand.

[0059] Reference Figure 1 The process shown, based on the manufacturing process knowledge graph of a certain compressor annular casing in the aerospace industry, uses the following steps for process reasoning based on this invention:

[0060] Step 1: Construct a multi-relationship weight model;

[0061] Step 1.1: Construct multiple relationships in the current processing technology knowledge graph;

[0062] First, the desired objective is set as efficiency optimization, and efficiency optimization relationships are constructed in the current processing technology knowledge graph of a certain compressor annular casing in the aviation industry.

[0063] Then, an attribute is added to the constructed efficiency optimization relationship. In this embodiment, the attribute of the efficiency optimization relationship is the processing time of a certain compressor annular casing part in the aviation category, in seconds.

[0064] Step 1.2: Calculate the weight matrix of the efficiency optimization relationship between any two different nodes in the current processing technology knowledge graph that are related to the efficiency optimization relationship constructed in Step 1.1;

[0065] Step 1.2.1: Extract sub-knowledge graphs from the current processing technology knowledge graph that are only related to efficiency optimization. The extracted sub-knowledge graphs have a total of n nodes. Construct a diagonal matrix between any two different nodes among these n nodes. and an attribute feature matrix

[0066]

[0067] Among them, the diagonal matrix The size is m ij ×m ij m ij The number of efficiency optimization relationships between nodes i and j; the size of the attribute feature matrix is ​​m. ij ×1; x1, x2, ..., Let i and j be the first and second efficiency optimization relationships, respectively, ..., the m-th efficiency optimization relationship between nodes i and j. ij The attribute feature value (i.e. the specific value of the attribute; if the attribute is processing time, then the attribute feature value is the specific value of the processing time, such as 120s) of the efficiency optimization relationship is a known quantity.

[0068] To facilitate understanding, examples are provided here. Assume there are three nodes related to efficiency optimization relationships: A, B, and C. There are 4 efficiency optimization relationships between nodes A and B, 2 between nodes B and C, and 1 between nodes A and C. Then, construct a 4×4 diagonal matrix for nodes A and B, where all diagonal elements are 1 / 4 and the rest are 0; construct a 2×2 diagonal matrix for nodes B and C, where all diagonal elements are 1 / 2 and the rest are 0; construct a 1×1 diagonal matrix for nodes A and C, where only one element is 1; since there are 4 efficiency optimization relationships between nodes A and B, each efficiency optimization relationship... The attribute of each relationship is the processing time of the processing steps. Each efficiency optimization relationship has one attribute feature value. Therefore, the feature matrix between nodes A and B is a 4×1 matrix, and the element values ​​of this matrix are the attribute feature values ​​of the four efficiency optimization relationships between nodes A and B. There are two efficiency optimization relationships between nodes B and C, so the feature matrix between nodes B and C is a 2×1 matrix, and the element values ​​of this matrix are the attribute feature values ​​of the two efficiency optimization relationships between nodes B and C. There is one efficiency optimization relationship between nodes A and C, so the feature matrix between nodes A and C is a 1×1 matrix, and the element values ​​of this matrix are the attribute feature values ​​of the one efficiency optimization relationship between nodes A and C.

[0069] Step 1.2.2: Calculate the diagonal matrix and attribute feature matrix The product of these products yields a product of size m. ij A matrix H of size ×1 ij :

[0070]

[0071] in,

[0072] Step 1.2.3: Convert matrix H ij The elements in the matrix H are normalized, that is, the matrix H is normalized. ij Each element value in the matrix is ​​updated to its reciprocal, resulting in matrix H. i ′ j :

[0073]

[0074] Step 1.2.4: Convert matrix H i ′ j Each element is subtracted from 1 to obtain the weight matrix W of the efficiency optimization relationship. ij :

[0075]

[0076] Among them, W ij W is the weight matrix for the efficiency optimization relationship between nodes i and j. ij Each element in the table represents the weight of each efficiency optimization relationship between nodes i and j, i.e. For the 1st, 2nd, ..., mth nodes between nodes i and j ij The weight values ​​of the efficiency optimization relationship.

[0077] Step 1.3: Construct a multi-relationship weight model;

[0078] The elements of the weight matrix of efficiency optimization relationships obtained in step 1.2 are assigned to the corresponding efficiency optimization relationships in the current sub-knowledge graph, and the current sub-knowledge graph is updated. The sub-knowledge graph containing relationship weight information after the update is the multi-relation weight model. For example, for nodes A and B, if there are 4 efficiency optimization relationships between them, the weight matrix of the efficiency optimization relationships between nodes A and B has 4 elements. These 4 elements are the weight values ​​of the 4 efficiency optimization relationships between nodes A and B. The 4 elements of the weight matrix are then assigned to the 4 efficiency optimization relationships between nodes A and B respectively. For the remaining nodes, the same method is used to assign weight values ​​to each efficiency optimization relationship between every two nodes. The final knowledge graph with relationship weight values ​​is the constructed multi-relation weight model. Figure 2 As shown.

[0079] Step 2: Extract sparse function activation subgraphs from the multi-relation weight model obtained in Step 1;

[0080] Step 2.1: Set the weight threshold;

[0081] Considering that an excessively large weight threshold may miss key information and reduce the accuracy of the reasoning results, while an excessively small weight threshold may lead to too much redundant information and low reasoning efficiency, after extensive simulation and practical verification, a weight threshold of 0.7-0.9 is appropriate. In this embodiment, the weight threshold is set to 0.8.

[0082] Step 2.2: Apply the sparse activation function to the multi-relation weight model constructed in Step 1 to obtain the sparse function activation subgraph;

[0083] The weight values ​​of each efficiency optimization relation in the multi-relation weight model constructed in step 1 and the weight threshold set in step 2.1 are used as inputs to a sparse activation function (a well-known function that can be implemented based on one-hot encoding). The sparse activation function compares the weight values ​​of each efficiency optimization relation with the preset weight thresholds, and determines whether to activate the attribute feature values ​​of each efficiency optimization relation based on the comparison results.

[0084] If the weight value of a certain efficiency optimization relationship is greater than or equal to the weight threshold, then the weight value of the efficiency optimization relationship is multiplied by 1 and used as the updated weight value (i.e. the weight value remains unchanged) to activate its attribute feature value.

[0085] If the weight value of a certain efficiency optimization relationship is less than the weight threshold, the weight value of the efficiency optimization relationship is multiplied by 0 and used as the updated weight value (that is, the weight value is reset to 0) so that its attribute feature value is not activated.

[0086] After all the weights of the efficiency optimization relationships have been updated, the multi-relation weight model with updated weight information is now a sparse function activation subgraph, such as... Figure 5 As shown.

[0087] Step 3: Run the inference algorithm to obtain the inference result;

[0088] Step 3.1: Select the appropriate inference algorithm according to the expected goal set in Step 1.1. Since the expected goal set in Step 1.1 is efficiency optimization, the inference algorithm selected here is an efficiency optimization algorithm, such as Dijkstra's shortest path algorithm (a known algorithm).

[0089] Step 3.2: Run the inference algorithm selected in step 3.1 in the sparse function activation subgraph obtained in step 2. The inference algorithm outputs the inference result, which is the preferred process scheme that meets the expected goal.

[0090] Step 3.3: Visualize the reasoning results obtained in Step 3.2, for example, as shown below. Figure 8 As shown in Figure (a), the sparse activation probability weights of the path from the starting node (forging) to the ending node (magnetic flaw detection) are 0.886, 0.832, 0.871, and 0.876, respectively, with a total value of 3.465. It should be noted that step 3.3 is not mandatory; it is only used to more intuitively display the inference results.

[0091] Step 4: Determine whether the optimal process solution corresponding to the current reasoning result meets the production requirements. If it does, the process ends; otherwise, proceed to step 5.

[0092] The method for determining whether the current reasoning result meets the requirements of efficient production optimization is as follows:

[0093] For the efficiency optimization objective, the reasoning result corresponds to an optimal process solution. Then, by combining the actual processing efficiency of the current production workshop, we can estimate whether the total time spent by the current optimal process solution meets the customer's delivery time requirements. If it does, it means that the current reasoning result meets the production requirements; otherwise, it means that the current reasoning result does not meet the production requirements.

[0094] Step 5: Dynamically update the current large-scale process knowledge graph;

[0095] To better facilitate process optimization, this embodiment dynamically updates the current large-scale process knowledge graph based on the actual conditions of the production workshop, including the status of materials, cutting tools, machining features, and equipment. This includes adding and removing nodes in the graph, and discovering and breaking relationships. For example, for a certain part, increasing the material being processed will lead to an increase in material entity nodes in the large-scale process knowledge graph; using a new cutting tool will lead to an increase in cutting tool entity nodes, and so on. For example,... Figure 9 As shown, Figure 9 Figure (a) shows the global initial graph, which has 3 nodes: node A, node B, and node C. There are n relationships between node A and node B, and m relationships between node B and node C. Figure 9 Figure (b) is a global initial map shown in Figure (a) with node D added; Figure 9 In diagram (c), a relationship r was found between nodes A and D in diagram (b). AD The relationship r AD The initial weight is ω l A relation r was found between node B and node C. BC The relationship r BC The initial weight is ω m+1 . Figure 9 Figure (c) is a comparison Figure 9 The new map obtained after the dynamic changes and updates of the global initial map shown in Figure (a) is shown in Figure 1.

[0096] Once the current large-scale process knowledge graph has been dynamically updated, return to step 1.

[0097] The above process is an example of using the invention to perform process reasoning based on the knowledge graph of the manufacturing process of a certain compressor annular casing in the aerospace industry, with efficiency optimization as the desired goal.

[0098] When performing process reasoning based on the knowledge graph of the manufacturing process of an annular casing of an aerospace compressor in the above embodiments, the desired objective can also be set as quality optimization, or as balancing efficiency and quality.

[0099] When the desired objective is set as quality optimization, the method for process reasoning is the same as the above-described process flow, with the only difference being: the multiple relations constructed in step 1.1 are quality optimization relations, and the attributes of the quality optimization relations are related to processing quality, such as the surface roughness value or processing error of the part; the weight matrix calculated in step 1.2 is the weight matrix of the quality optimization relations, and the resulting multi-relation weight model is as follows. Figure 3 As shown; in step 2.2, the input to the sparse activation function is the weight value of each quality optimization relationship and the set weight threshold. The resulting sparse function activation subgraph is shown below. Figure 6 As shown; Step 3.1 selects and Step 3.2 runs a quality optimization algorithm, such as the PageRank algorithm. The resulting inference is the optimal process scheme that satisfies the desired goal of quality optimization. A visual demonstration is shown exemplarily as follows: Figure 8 As shown in Figure (b), the sparse activation probability weights of the path from the starting node forging to the ending node magnetic flaw detection are 0.825, 0.812, 0.881, and 0.836, respectively, and the total value of the sum of the path probability weights is 3.354. Correspondingly, the method for determining whether the current inference result meets the quality optimization production requirements is as follows: For the quality optimization target, the inference result corresponds to an optimal process scheme. Therefore, by combining the actual processing capacity of the current production workshop (e.g., the processing accuracy of the equipment), estimating whether the current production workshop can meet the processing accuracy required by the optimal process scheme, and judging whether the processing accuracy of the optimal scheme can meet the processing accuracy required by the customer, if both are true, then the current inference result meets the production requirements; otherwise, it indicates that the current inference result does not meet the production requirements.

[0100] When the desired outcome is set to balance efficiency and quality, the method for process reasoning is the same in principle and process as the method described above, with the only difference being: the multiple relations constructed in step 1.1 are relations that balance efficiency and quality (efficiency optimization relations and quality optimization relations can be constructed separately and then superimposed, or relations that balance efficiency and quality can be constructed directly), and the attributes of the relations that balance efficiency and quality are related to processing efficiency and quality, such as processing time, surface roughness values ​​of the parts, or processing errors; the weight matrix calculated in step 1.2 is the weight matrix that balances efficiency and quality, and the resulting multi-relationship weight model is as follows. Figure 4 As shown; in step 2.2, the input to the sparse activation function is the weight values ​​of each line that balance efficiency and quality, and the set weight threshold. The resulting sparse function activation subgraph is shown below. Figure 7As shown; the desired objective algorithm selected in step 3.1 and run in step 3.2 is an optimization algorithm that balances efficiency and quality, such as Dijkstra's shortest path algorithm or PageRank algorithm. The inference result obtained is the optimal process scheme that balances efficiency and quality, and is illustrated by the following visualization example. Figure 8 As shown in (c), the sparse activation probability weights of the magnetic particle inspection path from the starting node forging to the ending node are 0.825, 0.812, 0.871, and 0.836, respectively, with a total value of 3.344. Correspondingly, for a process that balances efficiency and quality objectives, the inference result corresponds to an optimal process solution. Therefore, it is necessary to estimate whether the time taken by the current optimal process solution meets the customer's delivery time requirements by considering the actual processing efficiency of the current production workshop, and simultaneously estimate whether the current production workshop can meet the processing accuracy required by the optimal process solution by considering its actual processing capacity (e.g., equipment processing accuracy). Finally, it is determined whether the processing accuracy of the optimal process solution meets the customer's requirements. If all of the above are true, then the current inference result meets the production requirements; otherwise, it indicates that the current inference result does not meet the production requirements.

[0101] The specific embodiments of the present invention have been described in detail above with reference to the accompanying drawings. Finally, it should be noted that the above description and drawings are not intended to limit the scope of protection of the present invention. Various modifications or variations made by those skilled in the art based on the technical solutions of the present invention without creative effort are still within the scope of protection of the present invention.

Claims

1. A dynamic reasoning method for process knowledge graphs based on sparse activation, characterized in that, Includes the following steps: Step 1: Construct a multi-relationship weight model; Step 1.1: Set the desired goal, and construct multiple relationships associated with the desired goal in the current processing technology knowledge graph and add relationship attributes; Step 1.2: Calculate the weight matrix of the multiple relations between any two different nodes related to the multiple relations constructed in Step 1.1 in the current processing technology knowledge graph; Step 1.2.1: Extract sub-knowledge graphs from the current processing technology knowledge graph that are only associated with the multiple relations, and construct a diagonal matrix between any two different nodes in each sub-knowledge graph. and an attribute feature matrix Among them, the diagonal matrix The size is m ij ×m ij m ij The number of efficiency optimization relationships between nodes i and j; the size of the attribute feature matrix is ​​m. ij ×1; x1, x2, ..., Let i and j be the first and second efficiency optimization relationships, respectively, ..., the m-th efficiency optimization relationship between nodes i and j. ij The attribute feature values ​​of the efficiency optimization relationship are known quantities; Step 1.2.2: Calculate the diagonal matrix and attribute feature matrix The product of these products yields a product of size m. ij A matrix H of size ×1 ij ; Step 1.2.3: Convert matrix H ij The elements in the matrix H are normalized, that is, the matrix H is normalized. ij Each element value in the matrix is ​​updated to its reciprocal, resulting in matrix H. i ′ j ; Step 1.2.4: Convert matrix H i ′ j Each element is subtracted from 1 to obtain the weight matrix W of the efficiency optimization relationship. ij ; Step 1.3: Construct a multi-relationship weight model; Assign the element values ​​of the weight matrix of the efficiency optimization relation obtained in step 1.2 to the corresponding relation in the current sub-knowledge graph, update the current sub-knowledge graph, and the sub-knowledge graph containing relation weight information after the update is the multi-relation weight model. Step 2: Set a weight threshold, and use the weight threshold and the multi-relation weight model as input to the sparse activation function. Use the sparse activation function to extract the sparse function activation subgraph from the multi-relation weight model obtained in Step 1. Step 3: Select an inference algorithm based on the desired goal set in Step 1.1, run the selected inference algorithm in the sparse function activation subgraph, and obtain the inference result; Step 4: Determine whether the optimal process solution corresponding to the current reasoning result meets the production requirements. If it does, the process ends; otherwise, proceed to step 5. Step 5: Based on the actual situation of the current production workshop, including the status of materials, cutting tools, processing features, and equipment, dynamically update the current large-scale process knowledge graph. After the update is completed, return to Step 1.

2. The dynamic reasoning method for process knowledge graphs based on sparse activation according to claim 1, characterized in that: The desired objectives mentioned in step 1.1 are efficiency optimization, quality optimization, or a combination of both; correspondingly, the multiple relationships are efficiency optimization relationships, quality optimization relationships, or a combination of both.

3. The dynamic reasoning method for process knowledge graphs based on sparse activation according to claim 2, characterized in that: The weight threshold in step 2 is 0.7-0.

9.

4. The dynamic reasoning method for process knowledge graphs based on sparse activation according to claim 3, characterized in that: The method for extracting the sparse activation function subgraph from the multi-relation weight model obtained in Step 1 in Step 2 is as follows: The sparse activation function compares the weight value of each relation in a multi-relation weight model with a preset weight threshold, and determines whether to activate the attribute feature value of each relation based on the comparison result. If the weight value of a certain relation is greater than or equal to the weight threshold, the weight value of the relation is multiplied by 1 and used as the updated weight value to activate its attribute feature value. If the weight value of a relation is less than the weight threshold, the weight value of the relation is multiplied by 0 and used as the updated weight value, so that its attribute feature value is not activated. Once the weight values ​​of all relations in the multi-relation weight model have been updated, the multi-relation weight model with updated weight information is then called the sparse function activation subgraph.

5. The dynamic reasoning method for process knowledge graphs based on sparse activation according to claim 4, characterized in that: In step 3, if the desired goal is to optimize efficiency, then the inference algorithm is an efficiency optimization algorithm; if the desired goal is to optimize quality, then the inference algorithm is a quality optimization algorithm; if the desired goal is to balance efficiency and quality, then the inference algorithm is a balance between efficiency and quality optimization.

6. The dynamic reasoning method for process knowledge graphs based on sparse activation according to claim 5, characterized in that: The efficiency-optimized algorithm is Dijkstra's shortest path algorithm; the quality-optimized algorithm is PageRank algorithm; and the algorithm that balances efficiency and quality is either Dijkstra's shortest path algorithm or PageRank algorithm.

7. The dynamic reasoning method for process knowledge graphs based on sparse activation according to claim 6, characterized in that: The method for determining whether the optimal process solution corresponding to the current reasoning result meets the production requirements in step 6 is as follows: If the desired objective is to optimize efficiency, then based on the actual processing efficiency of the current production workshop, estimate whether the total time spent by the optimal process solution corresponding to the current reasoning result meets the customer's delivery time requirements. If yes, it indicates that the current reasoning result meets the production requirements; otherwise, it indicates that the current reasoning result does not meet the production requirements. If the desired goal is quality optimization, then based on the actual processing capacity of the current production workshop, estimate whether the current production workshop can meet the processing accuracy required by the optimal process scheme corresponding to the current reasoning result, and determine whether the processing accuracy of the optimal process scheme can meet the processing accuracy required by the customer. If both are true, it indicates that the current reasoning result meets the production requirements; otherwise, it indicates that the current reasoning result does not meet the production requirements. If the desired outcome is to balance efficiency and quality, then the time taken by the optimal process solution corresponding to the current reasoning result is estimated based on the actual processing efficiency of the current production workshop to see if it meets the customer's delivery time requirements. At the same time, the actual processing capacity of the current production workshop is estimated to see if the current production workshop can meet the processing accuracy required by the optimal process solution, and it is determined whether the processing accuracy of the optimal process solution can meet the customer's processing accuracy requirements. If all of the above are true, it indicates that the current reasoning result meets the production requirements; otherwise, it indicates that the current reasoning result does not meet the production requirements.

8. A dynamic reasoning system for process knowledge graphs based on sparse activation, comprising a processor and a storage medium, wherein a computer program is stored on the storage medium; characterized in that: When the computer program is run, it executes the dynamic reasoning method for process knowledge graphs based on sparse activation as described in any one of claims 1-7.

Images