Product assembly process knowledge pushing method, system and equipment

Abstract

The invention discloses a product assembly process knowledge pushing method, system and equipment in the technical field of semantic recognition and knowledge graph construction, and the method comprises the steps: carrying out the construction of a knowledge graph through employing a trained Bert-BiLSTM-CRF model and an LLM-KE model according to the obtained product assembly process information, obtaining a first initial knowledge graph and a second initial knowledge graph; and screening the first initial knowledge graph and the second initial knowledge graph by using a structural rationality discrimination module SRDM, performing knowledge graph evaluation on the screened first initial knowledge graph and the screened second initial knowledge graph by using a trained LLM-KGE model, and constructing a product assembly process knowledge graph APKG-CP according to an evaluation result. And pushing assembly process knowledge based on Bayesian network intelligent reasoning. According to the method, knowledge pushing can be completed in different process design requirements, and the efficiency of man-machine collaborative operation is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of semantic recognition and knowledge graph construction, and in particular to a method, system and equipment for pushing product assembly process knowledge. Background Art

[0002] Complex products, such as satellites and aircraft, are characterized by complex structures, high manufacturing technology, and complex development processes. During the development of complex products, assembly accounts for 20% to 70% of the total workload, with an average of 45%. This high proportion requires process design systems to strike a balance between automation efficiency and human ergonomics. With the advent of Industry 5.0, the goal of intelligent manufacturing has shifted from technology-driven production to knowledge-driven efficiency improvements through human-machine collaboration. Within this paradigm, assembly, as the final step in complex product development, focuses not only on technological implementation but also on the deep integration of expert knowledge, human experience, and artificial intelligence to improve operational efficiency. However, in achieving efficient knowledge reuse from assembly process knowledge with complex semantic relationships, both the traditional Bidirectional Encoder Representation from Transformers (BERT) model and the more recent Large Language Model (LLM) have their own limitations. Combining LLM with BERT, leveraging their respective semantic recognition strengths for knowledge graph construction to address the problem of knowledge incompleteness during knowledge push, is a promising research direction. Summary of the Invention

[0003] The purpose of the present invention is to overcome the deficiencies in the prior art and provide a method, system and equipment for pushing product assembly process knowledge, which can achieve knowledge push in different process design requirements with relatively low hardware resources, and output assembly process knowledge that accurately matches the process requirements and can be efficiently reused, thereby improving the efficiency of human-machine collaborative operations.

[0004] In order to solve the above technical problems, the present invention is implemented by adopting the following technical solutions:

[0005] In a first aspect, the present invention provides a method for pushing product assembly process knowledge, comprising:

[0006] Based on the acquired product assembly process information, the trained Bert-BiLSTM-CRF model is used to construct a knowledge graph to obtain a first initial knowledge graph, and the trained LLM-KE model is used to construct a knowledge graph to obtain a second initial knowledge graph;

[0007] Using a structural rationality determination module SRDM to screen the first initial knowledge graph and the second initial knowledge graph to obtain a screened first initial knowledge graph and a screened second initial knowledge graph;

[0008] Using the trained LLM-KGE model, the first initial knowledge graph after screening and the second initial knowledge graph after screening are evaluated, and a product assembly process knowledge graph APKG-CP is constructed according to the evaluation results;

[0009] According to the product assembly process knowledge graph APKG-CP, product assembly process knowledge is pushed based on Bayesian network intelligent reasoning.

[0010] Optionally, the training process of the Bert-BiLSTM-CRF model, the LLM-KE model, and the LLM-KGE model includes:

[0011] Based on the acquired product assembly process information, a training knowledge graph is constructed based on human experience;

[0012] Using the text in the product assembly process information as input data for the model and the training knowledge graph as the answer when evaluating the model to construct a training set;

[0013] Set the answer results of the Bert-BiLSTM-CRF model and the distillation model LLM to the probability of each answer, input the training set into the Bert-BiLSTM-CRF model and the distillation model LLM, perform model training, and obtain the trained Bert-BiLSTM-CRF model and LLM-KE model;

[0014] The training set is expanded using the knowledge graph explicit negative samples of randomly generated false information and the error samples generated by Bert-BiLSTM-CRF and LLM-KE during the training process to obtain an expanded training set;

[0015] The answer result of the distillation model LLM is set to True or False, the expanded training set is input into the distillation model LLM, and the model is trained to obtain a trained LLM-KGE model.

[0016] Optionally, the structural rationality determination module SRDM is used to make the following preliminary error determinations on the structure of the knowledge graph:

[0017] ,

[0018] in, represents the answer of the knowledge graph, Represents the answer template, To judge the answer Does it conform to the answer template? , Express answer Conform to the answer template , Express answer Does not conform to the answer template; Represents an entity, Indicates relationship, Represents text, Represents a set of relations, Used to determine the entity in the answer Whether to extract from text , and the relationship Whether it belongs to the relationship set , Represents the entity in the answer Extracted from text And the relationship Belongs to a relation set ; To judge the answer Is it acceptable? Express answer Acceptable, Express answer Unacceptable.

[0019] Optionally, the method uses the trained LLM-KGE model to perform knowledge graph evaluation on the first initial knowledge graph after screening and the second initial knowledge graph after screening, and constructs a product assembly process knowledge graph APKG-CP according to the evaluation results, including:

[0020] Calculate the answer confidence of the Bert-BiLSTM-CRF model based on the probability distribution of each category output by the last Softmax layer of the Bert-BiLSTM-CRF model;

[0021] The answer confidence of the LLM-KE model was calculated using the multi-sampling method;

[0022] The LLM-KGE model is used to modify the answer confidence of the Bert-BiLSTM-CRF model and the LLM-KE model;

[0023] The answer confidence of the modified Bert-BiLSTM-CRF model and the LLM-KE model are weighted and fused to obtain the fused answer confidence;

[0024] According to the confidence of the fused answers and the preset threshold, answers with confidence higher than the preset threshold are selected to construct the product assembly process knowledge graph APKG-CP.

[0025] Optionally, the calculation of the answer confidence of the Bert-BiLSTM-CRF model is implemented by the following formula:

[0026] ,

[0027] in, Indicates the answer confidence of the Bert-BiLSTM-CRF model, , Indicates the The predicted probability of each category, Indicates the The predicted probability of each category, Indicates the The predicted probability of each category, Indicates the total number of categories;

[0028] The answer confidence of the LLM-KE model is calculated using the multi-sampling method by the following formula:

[0029] ,

[0030] in, represents the answer confidence of the LLM-KE model, Indicates the number of times the same result appears in the answers of the LLM-KE model, represents the total number of responses to the LLM-KE model;

[0031] The use of the LLM-KGE model to correct the answer confidence of the Bert-BiLSTM-CRF model and the LLM-KE model is achieved by the following formula:

[0032] ,

[0033] in, Represents the answer confidence of the modified Bert-BiLSTM-CRF model, Represents the confidence correction coefficient of the Bert-BiLSTM-CRF model, Indicates the number of times the same result appears in the answer of the LLM-KGE model and the answer of the Bert-BiLSTM-CRF model. Indicates the total number of answers of the Bert-BiLSTM-CRF model; represents the answer confidence of the modified LLM-KE model, represents the confidence correction coefficient of the LLM-KE model, Indicates the number of times the same result appears in the answer of the LLM-KGE model and the answer of the LLM-KE model. represents the total number of responses to the LLM-KE model;

[0034] The weighted fusion of the answer confidence of the modified Bert-BiLSTM-CRF model and the LLM-KE model is achieved by the following formula:

[0035] ,

[0036] in, represents the confidence of the answer after fusion, Represents the static weight of the Bert-BiLSTM-CRF model, Represents the static weight of the LLM-KE model.

[0037] Optionally, the product assembly process knowledge graph APKG-CP is a multi-graph structure with node labels and directed edge labels, and its expression is as follows:

[0038] ,

[0039] in, Represents a collection of entities, Represents a set of relations, Represents a collection of attributes, A set of triples representing product assembly process knowledge.

[0040] Optionally, the pushing of assembly process knowledge based on Bayesian network intelligent reasoning includes:

[0041] According to the compilation requirements of the current process content in the assembly process information, the assembly elements involved in the current process are extracted from it;

[0042] Based on the extracted assembly elements involved in the current process, the product assembly process knowledge graph APKG-CP is used for retrieval to obtain a knowledge graph retrieval network that meets the requirements, and a Bayesian network structure diagram is constructed based on this retrieval network;

[0043] Based on the Bayesian network structure diagram and the relevant data of each entity in the product assembly process knowledge graph APKG-CP, the Bayesian network parameters are learned by the EM algorithm to obtain the conditional probability table corresponding to the Bayesian network structure;

[0044] According to the Bayesian network structure and its corresponding conditional probability table, the probability of assembly elements appearing under various assembly process compilation requirements is calculated, and product assembly process knowledge is pushed.

[0045] Optionally, the Bayesian network parameter learning using the EM algorithm includes:

[0046] Initialize Bayesian network parameters ;

[0047] Repeat the following steps until the Bayesian network parameters convergence:

[0048] Use the The Bayesian network parameters of the iteration Patch the missing data in the product assembly process knowledge graph APKG-CP and obtain the expected log-likelihood function based on the patched data:

[0049] ,

[0050] in, represents the total number of samples in the product assembly process knowledge graph APKG-CP, Indicates the The sample set in the iteration, Indicates the In the iteration samples, express The set of missing variables, represents the expected log-likelihood function, Represents an entity, Represents the parameters in the Bayesian network and the patched sample Condition, missing variables The value is The conditional probability of Represents the parameters in the Bayesian network Under these conditions, the sample With missing variables The value is Probability of simultaneous occurrence;

[0051] The Bayesian network parameters are optimized by maximizing the expected log-likelihood function, and the The Bayesian network parameters of the iteration :

[0052] ,

[0053] in, Represents the maximum value function.

[0054] In a second aspect, the present invention provides a product assembly process knowledge push system, comprising:

[0055] The initial knowledge graph construction module is used to: construct a knowledge graph based on the acquired product assembly process information using the trained Bert-BiLSTM-CRF model to obtain a first initial knowledge graph, and to construct a knowledge graph based on the trained LLM-KE model to obtain a second initial knowledge graph;

[0056] A screening module is used to: use a structural rationality determination module SRDM to screen the first initial knowledge graph and the second initial knowledge graph to obtain a screened first initial knowledge graph and a screened second initial knowledge graph;

[0057] A product assembly process knowledge graph construction module is used to: use the trained LLM-KGE model to perform knowledge graph evaluation on the first initial knowledge graph after screening and the second initial knowledge graph after screening, and construct a product assembly process knowledge graph APKG-CP based on the evaluation results;

[0058] The knowledge push module is used to: push product assembly process knowledge based on the product assembly process knowledge graph APKG-CP and Bayesian network intelligent reasoning.

[0059] In a third aspect, the present invention provides a computer device, comprising:

[0060] Memory, for storing computer instructions;

[0061] A processor is configured to execute the computer instructions to implement the steps of the product assembly process knowledge pushing method described in any one of the first aspects.

[0062] Compared with the prior art, the present invention has the following beneficial effects:

[0063] 1. A dynamic weighted fusion algorithm based on LLM confidence evaluated in knowledge graphs fully leverages the strengths of the Bert model and LLM. The Bert-BiLSTM-CRF model offers advantages such as efficient task fine-tuning, accurate semantic similarity calculation, and short text classification under low resources; while LLM-KE offers advantages such as generalized understanding and creative generation. By leveraging the strengths of both models in knowledge graph construction to address their respective shortcomings, a dynamically enhanced assembly process knowledge graph is developed to address the knowledge incompleteness issue during knowledge push.

[0064] 2. The lightweight BERT model, distilled large language model, and Bayesian network are used to reduce the demand for graphics card performance. The solution can be deployed locally, reducing the risk of privacy leakage when running in the cloud.

[0065] 3. Build a Bayesian network model for assembly components, assembly features, and operations, complete knowledge push under different process design requirements, and output assembly process knowledge that accurately matches process requirements and can be efficiently reused, thereby improving the efficiency of human-machine collaborative operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 This is an overall flow chart of a method for pushing product assembly process knowledge according to an embodiment of the present invention;

[0067] Figure 2 A detailed flow chart of a method for pushing product assembly process knowledge according to an embodiment of the present invention;

[0068] Figure 3 A flowchart for constructing a product assembly process knowledge graph APKG-CP according to an embodiment of the present invention;

[0069] Figure 4 A Bayesian network structure diagram provided according to an embodiment of the present invention;

[0070] Figure 5 A substructure diagram of a knowledge graph provided according to an embodiment of the present invention;

[0071] Figure 6 A directed acyclic graph corresponding to the knowledge graph substructure provided according to an embodiment of the present invention;

[0072] Figure 7 A forward reasoning flow chart of a Bayesian network according to an embodiment of the present invention;

[0073] Figure 8 A flowchart of reverse reasoning of a Bayesian network according to an embodiment of the present invention;

[0074] Figure 9 A flow chart of bidirectional reasoning of a Bayesian network according to an embodiment of the present invention;

[0075] Figure 10 A substructure diagram of the knowledge graph for the upper cabin assembly of the payload compartment provided according to an embodiment of the present invention;

[0076] Figure 11 This is a Bayesian network structure diagram for the upper cabin assembly of the payload compartment provided according to an embodiment of the present invention. DETAILED DESCRIPTION

[0077] The technical solution of the present invention is described in detail below through the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations on the technical solution of the present invention. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0078] It should be noted that the term "and / or" in this document simply describes an association between related objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A exists alone, A and B exist simultaneously, or B exists alone. Furthermore, the character " / " in this document generally indicates that the related objects are in an "or" relationship.

[0079] Example 1:

[0080] The embodiment of the present invention discloses a method for pushing product assembly process knowledge, referring to Figure 1 As shown, the specific steps include:

[0081] S1: Based on the acquired product assembly process information, the trained Bert-BiLSTM-CRF model is used to construct a knowledge graph to obtain the first initial knowledge graph, and the trained LLM-KE model is used to construct a knowledge graph to obtain the second initial knowledge graph;

[0082] S2, using a structural rationality determination module SRDM to screen the first initial knowledge graph and the second initial knowledge graph to obtain a screened first initial knowledge graph and a screened second initial knowledge graph;

[0083] S3, using the trained LLM-KGE model to perform knowledge graph evaluation on the first initial knowledge graph after screening and the second initial knowledge graph after screening, and constructing a product assembly process knowledge graph APKG-CP according to the evaluation results;

[0084] S4, according to the product assembly process knowledge graph APKG-CP, push the product assembly process knowledge based on Bayesian network intelligent reasoning.

[0085] Specifically, in step S1, the specific text information of the workshop assembly process is obtained, and a training knowledge graph is constructed based on manual experience; the specific text of the workshop assembly process is used as the input data of the model, and the training knowledge graph is used as the answer when evaluating the model to construct the training set, validation set, and test set;

[0086] The training set is input into the Bert-BiLSTM-CRF (Bidirectional Encoder Representations from Transformers-Bidirectional Long Short-Term Memory-Conditional RandomField) for training to obtain the trained Bert-BiLSTM-CRF. The training set is input into the LLM distillation model for fine-tuning to obtain the LLM for Knowledge Extraction (LLM-KE model). The answers given by the Bert-BiLSTM-CRF model and the LLM distillation model are the probabilities of each answer. Based on the trained Bert-BiLSTM-CRF and fine-tuned LLM-KE model, the preliminary construction of the knowledge graph is achieved based on the originally unlabeled complex product assembly process information.

[0087] The training set is expanded using randomly generated knowledge graph explicit negative samples of false information and error samples generated by Bert-BiLSTM-CRF and LLM-KE during training to obtain an expanded training set. The expanded training set is used to fine-tune another distillation model LLM to obtain the knowledge graph evaluation LLM (LLM for Knowledge Graph Evaluation, LLM-KGE model). The LLM-KGE model uses a simple binary classification method to set the answer result: if correct, the answer is True; if incorrect, the answer is False.

[0088] In step S2, the structural rationality determination module SRDM is used to make the following preliminary error determinations on the structure of the knowledge graph:

[0089] ,

[0090] in, represents the answer of the knowledge graph, Represents the answer template, To judge the answer Does it conform to the answer template? , Express answer Conform to the answer template , Express answer Does not conform to the answer template; Represents an entity, Indicates relationship, Represents text, Represents a set of relations, Used to determine the entity in the answer Whether to extract from text , and the relationship Whether it belongs to the relationship set , Represents the entity in the answer Extracted from text And the relationship Belongs to a relation set ; To judge the answer Is it acceptable? Express answer Acceptable, Express answer Unacceptable.

[0091] In step S3, the trained LLM-KGE model is used to perform knowledge graph evaluation on the first initial knowledge graph after screening and the second initial knowledge graph after screening, and a product assembly process knowledge graph APKG-CP is constructed according to the evaluation results, including:

[0092] According to the probability distribution of each category output by the last Softmax layer of the Bert-BiLSTM-CRF model, the answer confidence of the Bert-BiLSTM-CRF model is calculated:

[0093] ,

[0094] in, Indicates the answer confidence of the Bert-BiLSTM-CRF model, , Indicates the The predicted probability of each category, Indicates the The predicted probability of each category, Indicates the The predicted probability of each category, Indicates the total number of categories;

[0095] The answer confidence of the LLM-KE model is calculated using the multi-sampling method:

[0096] ,

[0097] in, represents the answer confidence of the LLM-KE model, Indicates the number of times the same result appears in the answers of the LLM-KE model, represents the total number of responses to the LLM-KE model;

[0098] The LLM-KGE model is used to modify the answer confidence of the Bert-BiLSTM-CRF model and the LLM-KE model:

[0099] ,

[0100] in, Represents the answer confidence of the modified Bert-BiLSTM-CRF model, Represents the confidence correction coefficient of the Bert-BiLSTM-CRF model, Indicates the number of times the same result appears in the answer of the LLM-KGE model and the answer of the Bert-BiLSTM-CRF model. Indicates the total number of answers of the Bert-BiLSTM-CRF model; represents the answer confidence of the modified LLM-KE model, represents the confidence correction coefficient of the LLM-KE model, Indicates the number of times the same result appears in the answer of the LLM-KGE model and the answer of the LLM-KE model. represents the total number of responses to the LLM-KE model;

[0101] The answer confidence of the modified Bert-BiLSTM-CRF model and the LLM-KE model are weighted and fused to obtain the fused answer confidence:

[0102] ,

[0103] in, represents the confidence of the answer after fusion, Represents the static weight of the Bert-BiLSTM-CRF model, Represents the static weight of the LLM-KE model;

[0104] According to the confidence of the fused answers and the preset threshold, answers with confidence higher than the preset threshold are selected, and the domain knowledge graph is used to construct a commonly used top-down model. The assembly process card, which is the most core part of the assembly process file, is used as the assembly process data to construct APKG-CP. APKG-CP uses the assembly process elements in the assembly process card as nodes and the various semantic relationships between them as edges to build a structured semantic network, which facilitates the display, mining and analysis of assembly process knowledge.

[0105] The product assembly process knowledge graph APKG-CP is a multi-graph structure with node labels and directed edge labels, and its expression is as follows:

[0106] ,

[0107] in, Represents a collection of entities, Represents a set of relations, Represents a collection of attributes, A set of triples representing product assembly process knowledge.

[0108] In step S4, the assembly process knowledge is pushed based on Bayesian network intelligent reasoning, including:

[0109] S4.1, Process Preparation Requirements Extraction: When preparing different process contents, based on the preparation requirements of the current process content, extract the assembly elements such as components, assembly features, and operations involved in this process;

[0110] S4.2, Process Knowledge Graph Retrieval: Search the complex product assembly process knowledge graph based on the assembly elements required for process compilation to obtain a knowledge graph retrieval network that meets the requirements, and construct a Bayesian network structure diagram based on this retrieval network;

[0111] S4.3, Bayesian network parameter learning: After obtaining the Bayesian network structure based on the knowledge graph, the EM algorithm is used to learn the parameters based on the relevant data of each entity in the knowledge graph to obtain the conditional probability table corresponding to the Bayesian network structure;

[0112] S4.4, Bayesian network intelligent reasoning: After obtaining the Bayesian network structure diagram and conditional probability table, the reasoning characteristics of the Bayesian network are used to calculate the probability of assembly elements appearing under various process compilation requirements, thereby realizing the push of assembly process knowledge.

[0113] The Bayesian network model is constructed based on the assembly process knowledge graph, and the Bayesian network structure is designed based on the pattern layer of the assembly process knowledge graph; the Bayesian network structure uses the nodes in the knowledge graph as nodes in the Bayesian network structure to represent the event of a certain entity; the edges in the knowledge graph are used as directed edges in the Bayesian network structure to represent the connection between two events; it is necessary to ensure that the direction of the directed edge conforms to the logical relationship between the two assembly elements in the assembly process knowledge, and at the same time ensure that there is only one directed line segment between the two nodes; based on the three assembly elements of assembly components, assembly features and operations that can represent assembly operations in the knowledge graph pattern layer and the relationship between their elements, the structure of the knowledge graph (reference Figure 5 ) is converted into a directed acyclic graph (refer to Figure 6 ), so far, the Bayesian network structure based on the assembly process knowledge graph is obtained.

[0114] The directed acyclic graphical graph (DAG) and conditional probability table are the two parts that make up the Bayesian network. The directed acyclic graph can express the conditional independence and dependency relationships between each node in the network, and at the same time calculate the probability of each basic event with the help of the conditional probability table. The Bayesian network parameter learning is based on the complex product assembly process knowledge graph, and the frequency of occurrence of each entity under different conditions is counted as data for parameter learning.

[0115] The APKG-CP may have some data missing problems in practice. The Expectation Maximization (EM) algorithm is used to iterate the training data and then use the maximum expectation to solve the approximate value of the maximum likelihood estimate. An initial value of Start, initial value It can be generated randomly; after that, two steps are performed alternately: E step (Expectation Step) and M step (Maximization Step). E step is used to calculate the expectation, and M step is used to calculate the maximum likelihood estimate. Assume that t iterations have been performed and the estimate is obtained. , the t+1 iteration consists of the following E-step and M-step:

[0116] (1) Step E: Using the The Bayesian network parameters of the iteration Patch the missing data in the product assembly process knowledge graph APKG-CP and obtain the expected log-likelihood function based on the patched data:

[0117] ,

[0118] in, represents the total number of samples in the product assembly process knowledge graph APKG-CP, Indicates the The sample set in the iteration, Indicates the In the iteration samples, express The set of missing variables, represents the expected log-likelihood function, Represents an entity, Represents the parameters in the Bayesian network and the patched sample Condition, missing variables The value is The conditional probability of Represents the parameters in the Bayesian network Under these conditions, the sample With missing variables The value is Probability of simultaneous occurrence;

[0119] (2) Step M: Optimize the Bayesian network parameters by maximizing the expected log-likelihood function to obtain the first The Bayesian network parameters of the iteration :

[0120] ,

[0121] in, represents the maximum value function;

[0122] Repeat the above steps until the Bayesian network parameters Convergence is achieved, thus completing the parameter learning of the Bayesian network model.

[0123] At present, most general recommendation systems are built based on user interest models and are widely used in e-commerce platforms, short video platforms and other fields. This type of recommendation system will push out other similar projects based on the user's current search information. In the process compilation of complex products, the recommendation of specific process content is based on contextual recommendations, that is, the recommendation system can push out various types of assembly elements that are contextually related to the search content, rather than similar assembly elements. This embodiment uses the reasoning ability of the Bayesian network to realize the knowledge push of the assembly process knowledge graph, and completes the real-time push of assembly process guidance knowledge in the process compilation scenario.

[0124] The topology of a Bayesian network is a directed acyclic graph, where nodes represent variables or propositions (which can be observable variables, hidden variables, unknown parameters, etc.), and directed edges between nodes describe causal relationships or conditional dependencies. Essentially, the presence of an arrow from one node to another implies a causal relationship or conditional independence between the corresponding random variables. Figure 4 The following is a simple Bayesian network structure. to For nodes in the Bayesian network model, the directed arrows between nodes indicate the relationship between nodes. For example, the node Points to the node Indicates that there is a dependency relationship between the two, and the node With node There is no arrow connecting the two nodes to indicate that they are independent of each other. Obviously, there may be more than one directed edge between the nodes.

[0125] Two nodes are connected by a single arrow, indicating that one node is the parent node and the other is the child node. The two nodes will generate a conditional probability value. The conditional probability reflects the strength of the connection between the two nodes, such as Figure 4 In the example, conditional probabilities are used to represent nodes. and The connection strength between them. The state change of a node is deeply affected by the state of its parent node. Therefore, when the parent node presents different state combinations, the child node will show different probability distributions, thereby forming the conditional probability table of the node. The conditional probability table is shown in Table 1. Under normal circumstances, it is assumed that each node has only two states: "0" and "1." It is worth noting that regardless of how many parent nodes a node has, the sum of the probabilities of its two states is always 1, given the state of its parent nodes.

[0126] Table 1: Nodes The conditional probability table of .

[0127]

[0128] When performing probability calculations, it is necessary to follow Bayes' theorem. Before calculation, the prior probability and posterior probability must be determined first. The calculation formula for the posterior probability is called the conditional probability formula, and then the Bayes' formula can be obtained.

[0129] The reasoning described above is based on the difference between known and unknown quantities, and the reasoning mode is also different. Figures 7 to 9 As shown, it can be divided into: forward reasoning, reverse reasoning and bidirectional reasoning; among them, 、 is the cause node, 、 It is an intermediate node and also the result node of the previous level. 、 、 is the result node. Assuming that each node has only two states, "0" and "1", the reasoning process is as follows:

[0130] (1) Forward reasoning is also called causal reasoning, which is reasoning along the direction of the directed graph. In the case of the inference node The probability of occurrence, such as Figure 7 As shown. Based on the known conditions, we can find the node 、 The state probability, using the conditional probability value 、 Represented by. Then, by the total probability formula, we can further get the node The probability of occurrence is calculated as follows:

[0131] ,

[0132] in, and Indicates the node status, Representation node Under the conditions, The probability of Representation node Under the conditions, The probability of Representation node Under the conditions, The probability of Representation node Under the conditions, probability.

[0133] Taking this as an example, the conditional probability of each node can be calculated to complete forward reasoning.

[0134] (2) Reverse reasoning is also called evidential reasoning, which is reasoning in the opposite direction of the directed graph. Occurrence, through the known prior probability and conditional probability reverse reasoning to explore the reasons that lead to the occurrence of the result, such as Figure 8 As shown. By finding the node Under the conditions that occur, the cause node 、 The conditional probability of occurrence is , To compare the values ​​of the two, generate a recommendation list, and then infer the use of The reason for the occurrence. The calculation formula is as follows:

[0135] ,

[0136] in, Representation node Under the condition that the node The probability of Representation node Under the condition that the node The probability of Representation node The probability of Representation node probability;

[0137] Where, is the prior probability and is known. and Obtained by the following formula:

[0138] ,

[0139] ,

[0140] in, and Indicates the node status, express Under the conditions, The probability of Representation node Under the conditions, The probability of express Under the conditions, probability;

[0141] So far, we can find Taking this as an example, find the value of , thus completing the reverse reasoning.

[0142] (3) Bidirectional reasoning is a combination of forward reasoning and reverse reasoning, which can be used to infer the probability of the intermediate nodes. and nodes At the same time, the result of the intermediate node state is inferred through the known prior probability and conditional probability, such as Figure 9 As shown in Figure 2. Similar to forward and backward reasoning, the Bayesian formula is used to calculate the conditional probabilities of intermediate nodes, and then an inference list is generated. This will not be described in detail here.

[0143] The three Bayesian network reasoning methods described above address the need to infer the remaining required assembly elements based on the different assembly elements specified during process compilation. Therefore, the advantages of Bayesian network reasoning can be combined with the assembly process knowledge graph and applied to process compilation scenarios during assembly process design. This can provide process engineers with real-time guidance and suggestions for assembly processes that meet their requirements, improving process design efficiency.

[0144] Example 2:

[0145] The embodiment of the present invention discloses a method for pushing product assembly process knowledge, referring to Figure 2 As shown, the following steps are included:

[0146] Prepare process files, train Bert-BiLSTM-CRF, fine-tune LLM-KE and LLM-KGE, and output the final knowledge graph APKG-CP based on dynamic weighted fusion of confidence, realizing dynamic incremental completion of the existing knowledge graph; in terms of knowledge push based on Bayesian network, first execute "process step preparation requirements" and "extract assembly elements: components, assembly features, operations" according to needs, and with the support of "knowledge graph retrieval" provided by APKG-CP, "construct Bayesian network structure based on search assembly elements", adjust the "parameter learning" strategy, and select the optimal "execution parameter learning to obtain conditional probability table"; finally, implement "Bayesian network analysis" through "intelligent analysis" to obtain a "recommendation list".

[0147] like Figure 3 As shown, in this embodiment, the construction of the knowledge graph APKG-CP includes the following steps:

[0148] Through expert experience, the specific labels of relevant information in the process documents and the relationships between the labels are marked to construct a knowledge graph based on expert experience;

[0149] Based on the expert experience knowledge graph, train Bert-BiLSTM-CRF and fine-tune LLM-KE and LLM-KGE;

[0150] The initial knowledge graph information output by Bert-BiLSTM-CRF and LLM-KE is filtered by the knowledge graph structure rationality discrimination module SRDM, and the filtered information is passed to LLM-KGE to generate confidence correction coefficients and ;

[0151] Calculate the confidence of each answer of Bert-BiLSTM-CRF and LLM-KE, and multiply them by the confidence correction coefficient generated by LLM-KGE respectively and , get a certain answer and its corresponding confidence score;

[0152] Finally, by setting a threshold, high-confidence results are screened out to construct a highly accurate complex product assembly knowledge graph APKG-CP.

[0153] like Figure 5 and Figure 6As shown, in this embodiment, the process of constructing the Bayesian network structure is represented, where different colors represent different types of entities. The nodes in the knowledge graph are used as nodes in the Bayesian network structure to represent the event of the occurrence of a certain entity. The edges in the knowledge graph are used as directed edges in the Bayesian network structure to represent the connection between two events. It is worth noting that it is necessary to determine that the direction of the directed edge conforms to the logical relationship between the two assembly elements in the assembly process knowledge, and at the same time ensure that there is only one directed line segment between the two nodes. Based on the three assembly elements of assembly components, assembly features and operations that can represent assembly operations in the knowledge graph model layer and the relationship between their elements, the structure of the knowledge graph is converted into the form of a directed acyclic graph. At this point, a Bayesian network structure based on the assembly process knowledge graph is obtained.

[0154] like Figures 7-9 As shown in the figure, in this embodiment, the reasoning mode is different according to the known and unknown quantities, and can be divided into forward reasoning, reverse reasoning and bidirectional reasoning. 、 is the cause node, 、 It is an intermediate node and also the result node of the previous level. 、 、 Assume that each node has only two states, "0" and "1", and execute the reasoning process.

[0155] The present invention will be further described below with reference to specific embodiments.

[0156] This paper uses DeepSeek-R1-Distill-Qwen-1.5B as the underlying LLM model. First, to demonstrate the effectiveness of the proposed APKG-CP construction method, a preliminary experimental verification was conducted using the SemEval2010 Task 8 dataset. Then, a dataset containing assembly process cards from an anaerospace company in Shanghai (APC-AS) was constructed to construct the APKG-CP, verifying its feasibility. Finally, accurate push of the process knowledge graph was achieved based on a Bayesian network model. The dataset is described as follows:

[0157] SemEval2010 Task 8 defines 10 types of relationships between nouns: Cause-Effect, Instrument-Agency, Product-Producer, Content-Container, Entity-Origin, Entity-Destination, Component-Whole, Member-Collection, Message-Topic, and Other. When two nouns do not fall into the first nine relationships, they are classified as "Other." The dataset contains 8,000 training data points and 2,717 validation and testing data points. Each example contains a complete sentence with two labeled entities. The semantic relationship between these two entities within the sentence is one of the 10 relationship types.

[0158] The APC-AS dataset is constructed using a subset of assembly process cards as raw data. Years of digitalization have accumulated a vast amount of assembly process data from the production of numerous models and diverse structures. By organizing these process cards and extracting the names of the process steps and detailed operational details, we obtained 1,462 valid data items, totaling 98,485 words, to form the raw text set.

[0159] (1) SemEval2010 Task 8 Experiment

[0160] We evaluated entity recognition and relation classification tasks separately, using the F1 score to evaluate our models. Only completely correct predictions are counted; even a single word that differs from the annotated entity is classified as an error. Under these strict metrics, BERT-BiLSTM-CRF, LLM-KE, and KGIC-LDM (A Knowledge Graph Intelligent Completion Framework Based on Lightweight Distillation Models) achieved F1 scores of 88.6%, 65.1%, and 91.4% for entity recognition, respectively. We compared several leading algorithms for relation classification on SemEval2010 Task 8. As shown in Table 2, KGIC-LDM achieved the best classification performance compared to traditional algorithms such as CR-CNN (Classifying Relations by Ranking with Convolutional Neural Networks), FCM (Factor-based Compositional Embedding Models), and R-BERT (Relation Classification BERT). Furthermore, despite LLM-KE's relatively low score, KGIC-LDM improved upon Bert-BiLSTM-CRF and LLM-KE by 3.7% and 9.4%, respectively. This demonstrates that the dynamically weighted confidence fusion strategy effectively integrates the strengths of the two algorithms, demonstrating the feasibility and effectiveness of the proposed method for dynamic incremental knowledge graph construction.

[0161] Table 2 Comparison of relationship prediction results of different models on the SemEval2010 Task 8 dataset.

[0162]

[0163] (2) APC-AS experiment

[0164] The results were experimentally verified by SemEval2010 Task 8. The same method was used to perform the dynamic incremental construction task of APKG-CP on the APC-AS dataset.

[0165] Step S1, constructing APKG-CP;

[0166] Step S2, Process Preparation Requirements Acquisition: To verify the feasibility of the Bayesian network construction method and the effectiveness of the intelligent push technology based on the assembly process knowledge graph, a case study of the upper cabin assembly process knowledge was conducted. First, a Bayesian model was constructed based on the upper cabin assembly process knowledge graph to verify the feasibility of the push technology process. Then, a random process within the upper cabin assembly process was randomly selected to verify the accuracy of the push content.

[0167] During the process preparation, process guidance is sometimes required to write new process content. This assembly process guidance is usually summarized based on existing assembly process cases and expressed in the form of "perform XX operation on XX assembly feature of XX component". For example, "remeasure the parallelism accuracy between the upper surface of the payload bay roof and the reference platform", where the component is "payload bay roof", the assembly feature is "parallelism", and the operation is "remeasure accuracy". These three assembly elements are all entities in the knowledge graph. When the preparation requirements are proposed, one or two assembly elements are usually known, and the assembly process guidance is formed by pushing the most likely third assembly element. Based on the known assembly elements, the three reasoning modes of the Bayesian network are used to obtain the most likely results, thereby realizing the push of complete assembly process guidance.

[0168] Step S3, process knowledge graph retrieval: In the actual process preparation process, the scope of the assembly process guidance must be determined first. For example, when preparing process content related to the payload cabin upper cabin assembly, the process guidance scope should be selected as the payload cabin upper cabin assembly, and the content related to the payload cabin upper cabin assembly should be retrieved from the assembly process knowledge graph. Then, the component entities, assembly feature entities, and operation entities related to the upper cabin assembly are filtered out from the assembly process knowledge graph, and a subgraph of the assembly process knowledge graph is generated for subsequent use, such as Figure 10 shown.

[0169] Step S4, constructing a Bayesian network model: constructing a Bayesian network structure based on the structure of the assembly process knowledge graph subgraph, with the component entity as the parent node, the assembly feature entity as the first-level child node, the operation entity as the second-level child node, and the relationship between entities as the directed edge to construct a directed acyclic graph. At the same time, since there are many entities with the same name in the knowledge graph, their operations and logic are essentially the same, but they belong to different process contents. When constructing the directed acyclic graph, these entities are simplified in order to reduce the amount of calculation during parameter training and Bayesian reasoning. At the same time, if the requirements can be further refined, such as the relevant content of "assembly of the upper cabin of the payload bay" is further streamlined to the relevant content of "assembly of the upper bulkhead of the payload bay", and the retrieval requirements are refined to the assembly element part, there will be a more streamlined Bayesian network model with better reasoning effect. Taking "assembly of the upper bulkhead of the payload bay" as the process compilation requirement, the Bayesian network structure constructed based on the knowledge graph subgraph of the upper cabin assembly of the payload bay is as follows: Figure 11 shown.

[0170] After the directed acyclic graph is constructed, parameter learning is required to construct a conditional probability table. The initial dataset is formed by combining the frequency of assembly element entities in the knowledge graph subgraph with the frequency of their occurrence in the entire assembly process knowledge graph. The EM algorithm is then used to perform parameter learning on the initial dataset, and the conditional probability table is obtained after the algorithm converges. Step S5, Bayesian network reasoning: After the Bayesian network model is constructed, the status of unknown assembly elements in the current situation is calculated in combination with real-time process compilation requirements, and then sorted by probability to form a recommendation list.

[0171] After calculation, a recommendation list is generated under these conditions. Verified by process personnel, the relationship between the three assembly elements in actual assembly process compilation is "C1, F1, O3." Operation O3, with the highest probability in the recommendation list, is the optimal assembly guidance solution for actual process compilation. Similarly, after switching to other reasoning modes, the Bayesian network can still deliver results that meet the actual process compilation requirements, verifying the effectiveness of intelligent push technology. After the Bayesian network reasoning obtains the optimal solution, it can use this assembly process guidance as the search criteria, using the assembly elements in this process guidance as the search criteria, to retrieve complete process information from the assembly process knowledge graph to guide process personnel in compilation.

[0172] Therefore, through the similarity between the assembly process knowledge graph and the Bayesian network, and by utilizing the reasoning model of the Bayesian network, the assembly process knowledge graph can push out the assembly process knowledge that best meets the process compilation requirements for assembly process guidance, thereby increasing the utilization rate of assembly process knowledge and improving the quality and efficiency of process compilation.

[0173] Example 3:

[0174] Based on the same inventive concept as the first embodiment, this embodiment of the present invention discloses a product assembly process knowledge push system, including:

[0175] The initial knowledge graph construction module is used to: construct a knowledge graph based on the acquired product assembly process information using the trained Bert-BiLSTM-CRF model to obtain a first initial knowledge graph, and to construct a knowledge graph based on the trained LLM-KE model to obtain a second initial knowledge graph;

[0176] A screening module is used to: use a structural rationality determination module SRDM to screen the first initial knowledge graph and the second initial knowledge graph to obtain a screened first initial knowledge graph and a screened second initial knowledge graph;

[0177] A product assembly process knowledge graph construction module is used to: use the trained LLM-KGE model to perform knowledge graph evaluation on the first initial knowledge graph after screening and the second initial knowledge graph after screening, and construct a product assembly process knowledge graph APKG-CP based on the evaluation results;

[0178] The knowledge push module is used to push the product assembly process knowledge based on the product assembly process knowledge graph APKG-CP and Bayesian network intelligent reasoning.

[0179] The specific functional implementation of each of the above modules can be found in the relevant content of Example 1 and will not be elaborated here.

[0180] Example 4:

[0181] This embodiment provides a computer device, including:

[0182] Memory, for storing computer instructions;

[0183] The processor is configured to execute the computer instructions to implement the steps of the product assembly process knowledge pushing method described in the first embodiment.

[0184] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0185] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A system that specifies the functions of a box or boxes.

[0186] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture including an instruction system that is implemented in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0187] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0188] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, which are all protected by the present invention.

Claims

1. A method for pushing product assembly process knowledge, characterized in that: include: Based on the acquired product assembly process information, the trained Bert-BiLSTM-CRF model is used to construct a knowledge graph to obtain a first initial knowledge graph, and the trained LLM-KE model is used to construct a knowledge graph to obtain a second initial knowledge graph; Using a structural rationality determination module SRDM to screen the first initial knowledge graph and the second initial knowledge graph to obtain a screened first initial knowledge graph and a screened second initial knowledge graph; Using the trained LLM-KGE model, the first initial knowledge graph after screening and the second initial knowledge graph after screening are evaluated, and a product assembly process knowledge graph APKG-CP is constructed according to the evaluation results; According to the product assembly process knowledge graph APKG-CP, product assembly process knowledge is pushed based on Bayesian network intelligent reasoning.

2. The method for pushing product assembly process knowledge according to claim 1, characterized in that: The training process of the Bert-BiLSTM-CRF model, LLM-KE model, and LLM-KGE model includes: Based on the acquired product assembly process information, a training knowledge graph is constructed based on human experience; Using the text in the product assembly process information as input data for the model and the training knowledge graph as the answer when evaluating the model to construct a training set; Set the answer results of the Bert-BiLSTM-CRF model and the distillation model LLM to the probability of each answer, input the training set into the Bert-BiLSTM-CRF model and the distillation model LLM, perform model training, and obtain the trained Bert-BiLSTM-CRF model and LLM-KE model; The training set is expanded using the knowledge graph explicit negative samples of randomly generated false information and the error samples generated by Bert-BiLSTM-CRF and LLM-KE during the training process to obtain an expanded training set; The answer result of the distillation model LLM is set to True or False, the expanded training set is input into the distillation model LLM, and the model is trained to obtain a trained LLM-KGE model.

3. The method for pushing product assembly process knowledge according to claim 1, characterized in that: The structural rationality judgment module SRDM is used to make the following preliminary error judgments on the structure of the knowledge graph: ， in, represents the answer of the knowledge graph, Represents the answer template, To judge the answer Does it conform to the answer template? , Express answer Conform to the answer template , Express answer Does not conform to the answer template; Represents an entity, Indicates relationship, Represents text, Represents a set of relations, Used to determine the entity in the answer Whether to extract from text , and the relationship Whether it belongs to the relationship set , Represents the entity in the answer Extracted from text And the relationship Belongs to a relation set ; To judge the answer Is it acceptable? Express answer Acceptable, Express answer Unacceptable.

4. The method for pushing product assembly process knowledge according to claim 1, characterized in that: The trained LLM-KGE model is used to perform knowledge graph evaluation on the first initial knowledge graph after screening and the second initial knowledge graph after screening, and a product assembly process knowledge graph APKG-CP is constructed according to the evaluation results, including: Calculate the answer confidence of the Bert-BiLSTM-CRF model based on the probability distribution of each category output by the last Softmax layer of the Bert-BiLSTM-CRF model; The answer confidence of the LLM-KE model was calculated using the multi-sampling method; The LLM-KGE model is used to modify the answer confidence of the Bert-BiLSTM-CRF model and the LLM-KE model; The answer confidence of the modified Bert-BiLSTM-CRF model and the LLM-KE model are weighted and fused to obtain the fused answer confidence; According to the confidence of the fused answers and the preset threshold, answers with confidence higher than the preset threshold are selected to construct the product assembly process knowledge graph APKG-CP.

5. The method for pushing product assembly process knowledge according to claim 4, characterized in that: The answer confidence of the Bert-BiLSTM-CRF model is calculated using the following formula: , in, Indicates the answer confidence of the Bert-BiLSTM-CRF model, , Indicates the The predicted probability of each category, Indicates the The predicted probability of each category, Indicates the The predicted probability of each category, Indicates the total number of categories; The answer confidence of the LLM-KE model is calculated using the multi-sampling method by the following formula: , in, represents the answer confidence of the LLM-KE model, Indicates the number of times the same result appears in the answers of the LLM-KE model, represents the total number of responses to the LLM-KE model; The use of the LLM-KGE model to correct the answer confidence of the Bert-BiLSTM-CRF model and the LLM-KE model is achieved by the following formula: ， in, Represents the answer confidence of the modified Bert-BiLSTM-CRF model, Represents the confidence correction coefficient of the Bert-BiLSTM-CRF model, Indicates the number of times the same result appears in the answer of the LLM-KGE model and the answer of the Bert-BiLSTM-CRF model. Indicates the total number of answers of the Bert-BiLSTM-CRF model; represents the answer confidence of the modified LLM-KE model, represents the confidence correction coefficient of the LLM-KE model, Indicates the number of times the same result appears in the answer of the LLM-KGE model and the answer of the LLM-KE model. represents the total number of responses to the LLM-KE model; The weighted fusion of the answer confidence of the modified Bert-BiLSTM-CRF model and the LLM-KE model is achieved by the following formula: ， in, represents the confidence of the answer after fusion, Represents the static weight of the Bert-BiLSTM-CRF model, Represents the static weight of the LLM-KE model.

6. The method for pushing product assembly process knowledge according to claim 1, characterized in that: The product assembly process knowledge graph APKG-CP is a multi-graph structure with node labels and directed edge labels, and its expression is as follows: ， in, Represents a collection of entities, Represents a set of relations, Represents a collection of attributes, A set of triples representing product assembly process knowledge.

7. The method for pushing product assembly process knowledge according to claim 1, characterized in that: The pushing of assembly process knowledge based on Bayesian network intelligent reasoning includes: According to the compilation requirements of the current process content in the assembly process information, the assembly elements involved in the current process are extracted from it; Based on the extracted assembly elements involved in the current process, the product assembly process knowledge graph APKG-CP is used for retrieval to obtain a knowledge graph retrieval network that meets the requirements, and a Bayesian network structure diagram is constructed based on this retrieval network; Based on the Bayesian network structure diagram and the relevant data of each entity in the product assembly process knowledge graph APKG-CP, the Bayesian network parameters are learned by the EM algorithm to obtain the conditional probability table corresponding to the Bayesian network structure; According to the Bayesian network structure and its corresponding conditional probability table, the probability of assembly elements appearing under various assembly process compilation requirements is calculated, and product assembly process knowledge is pushed.

8. The method for pushing product assembly process knowledge according to claim 7, characterized in that: The Bayesian network parameter learning using the EM algorithm includes: Initialize Bayesian network parameters ; Repeat the following steps until the Bayesian network parameters convergence: Use the The Bayesian network parameters of the iteration Patch the missing data in the product assembly process knowledge graph APKG-CP and obtain the expected log-likelihood function based on the patched data: ， in, represents the total number of samples in the product assembly process knowledge graph APKG-CP, Indicates the The sample set in the iteration, Indicates the In the iteration samples, express The set of missing variables, represents the expected log-likelihood function, Represents an entity, Represents the parameters in the Bayesian network and the patched sample Condition, missing variables The value is The conditional probability of Represents the parameters in the Bayesian network Under these conditions, the sample With missing variables The value is Probability of simultaneous occurrence; The Bayesian network parameters are optimized by maximizing the expected log-likelihood function, and the The Bayesian network parameters of the iteration : ， in, Represents the maximum value function.

9. A product assembly process knowledge push system, characterized by: include: The initial knowledge graph construction module is used to: construct a knowledge graph based on the acquired product assembly process information using the trained Bert-BiLSTM-CRF model to obtain a first initial knowledge graph, and to construct a knowledge graph based on the trained LLM-KE model to obtain a second initial knowledge graph; A screening module is used to: use a structural rationality determination module SRDM to screen the first initial knowledge graph and the second initial knowledge graph to obtain a screened first initial knowledge graph and a screened second initial knowledge graph; A product assembly process knowledge graph construction module is used to: use the trained LLM-KGE model to perform knowledge graph evaluation on the first initial knowledge graph after screening and the second initial knowledge graph after screening, and construct a product assembly process knowledge graph APKG-CP based on the evaluation results; The knowledge push module is used to: push product assembly process knowledge based on the product assembly process knowledge graph APKG-CP and Bayesian network intelligent reasoning.

10. A computer device, characterized in that: include: Memory, for storing computer instructions; A processor is used to execute the computer instructions to implement the steps of the product assembly process knowledge pushing method according to any one of claims 1 to 8.

Images