Intelligent fault diagnosis method and device for industrial field knowledge collaboration

Abstract

The invention provides an intelligent fault diagnosis method and device for industrial field knowledge collaboration, and the method comprises the steps: carrying out the data collection and cleaning of original industrial data, and constructing the original data representation of a knowledge base; for original data representation of the knowledge base, entity and relation extraction is carried out, and a map ontology base is designed; compiling a graph construction module according to the design in the graph ontology library, and completing the construction of the industrial knowledge graph; performing text processing on the question of the user so that the system can understand the intention of the user; designing and realizing a knowledge completion inference model for performing knowledge completion on the constructed industrial knowledge graph; and designing and implementing a multi-hop knowledge reasoning model for reasoning answers required by the user in the knowledge graph. According to the intelligent fault diagnosis method oriented to industrial field knowledge collaboration, complex natural language problems can be processed, and landing and development of industrial intellectualization are promoted.

Description

Technical Field

[0001] The present invention relates to the field of industrial intelligent technology, and in particular to an intelligent fault diagnosis method and device oriented to industrial field knowledge collaboration. Background Art

[0002] In the industrial field, with the continuous advancement of science and technology and the rapid accumulation of information, more and more companies and organizations need to effectively manage and apply complex industrial knowledge to improve production efficiency, reduce costs, ensure quality, and meet the growing market demand. The intelligent fault diagnosis system for industrial knowledge collaboration is one of the solutions to this problem. It uses knowledge graphs to manage huge and complex industrial knowledge and applies industrial knowledge data to downstream tasks in the form of intelligent question-answering.

[0003] Intelligent fault diagnosis systems use deep learning, natural language processing and knowledge graph technology to intelligently organize and apply knowledge in the industrial field. However, traditional information retrieval and data analysis tools have limitations in dealing with complex problems in the industrial field because they usually cannot understand users' natural language queries and have difficulty in multi-level and multi-step reasoning. Summary of the invention

[0004] In view of the defects in the prior art, the purpose of the present invention is to provide an intelligent fault diagnosis method and device for industrial knowledge collaboration, which solves the problem that it is difficult to perform multi-level reasoning on knowledge graphs in traditional industrial scenarios and provides a reference solution for industrial fault diagnosis.

[0005] To solve the above problems, the technical solution of the present invention is:

[0006] An intelligent fault diagnosis method for industrial domain knowledge collaboration includes the following steps:

[0007] Collect and clean the original industrial data and construct the original data representation of the knowledge base;

[0008] Based on the original data representation of the knowledge base, extract entities and relationships and design the graph ontology library;

[0009] Write the graph construction module according to the design in the graph ontology library to complete the construction of the industrial knowledge graph;

[0010] Text the user's question so that the system can understand the user's intention;

[0011] Design and implement a knowledge completion reasoning model to complete the knowledge of the constructed industrial knowledge graph;

[0012] Design and implement a multi-hop knowledge reasoning model to infer the answers users need in the knowledge graph.

[0013] Preferably, the step of collecting and cleaning the original industrial data and constructing the original data representation of the knowledge base specifically includes:

[0014] Collect raw data from various industrial devices, sensors, log files, and other data sources;

[0015] Clean and preprocess the collected raw data to ensure data availability;

[0016] Structure the raw data and add tags and metadata to different data so that it can be associated with industrial equipment and parameters.

[0017] Preferably, the step of extracting entities and relationships and designing a graph ontology library for the original data representation of the knowledge base specifically includes:

[0018] Use natural language processing technology and data mining methods to extract entities and relationships from raw data;

[0019] Define an ontology, determine the entity categories and relationship types in the ontology, and design entity attributes according to the needs of the actual industrial field;

[0020] Use RDF language to create an ontology library and add entities, relationships and attributes to the ontology library;

[0021] Map the extracted entities and relations with those in the ontology library. Label the extracted entities to associate them with the corresponding entities in the ontology library.

[0022] Preferably, the step of writing a graph construction module according to the design in the graph ontology library to complete the construction of the industrial knowledge graph specifically includes:

[0023] Extract relevant information from raw industrial data, including entities, attributes, and relationships, and convert the extracted data into the format of graph data to ensure compliance with the model of the ontology library;

[0024] Establish a graph database to store the constructed knowledge graph, load the extracted and converted data into the graph database, and ensure the integrity and consistency of the data.

[0025] Preferably, the step of text processing the user's question so that the system can understand the user's intention specifically includes:

[0026] Perform text preprocessing on the questions asked by users, including removing extra spaces, punctuation marks and special characters;

[0027] Use natural language processing tools to segment text into fine-grained words and annotate them;

[0028] Perform stop word removal and word stemming.

[0029] Preferably, the step of designing and implementing a knowledge completion reasoning model for completing knowledge on the constructed industrial knowledge graph specifically includes:

[0030] Collect the constructed industrial knowledge graph data, including entities, attributes, relationships, and existing knowledge, and prepare the knowledge graph data for use in the reasoning model, including converting it into a suitable data structure;

[0031] Design a knowledge completion reasoning model.

[0032] Preferably, the knowledge completion reasoning model specifically includes:

[0033] The encoder part includes:

[0034] Relation Encoder: The role of the relation encoder is to process the relationship between entities in the knowledge graph, extract and encode these relationship information to provide context for reasoning;

[0035] Structural encoder: The structural encoder is responsible for processing the topological structure information in the knowledge graph, extracting and encoding the structural features of the subgraph, so that the model can capture the complex structural relationships between different entities and ensure that the model has a more comprehensive understanding of the overall structure of the graph;

[0036] Decoder: receives the relationship and structural features extracted by the encoder, combines the input original graph structural features, and performs reasoning through the attention mechanism;

[0037] Structure-relation attention mechanism: Combining relational and structural information, by weighting the features of different channels, it helps the model better focus on the important features in the input, thereby improving the accuracy of the reasoning process;

[0038] Reasoning process and rule extraction: The model uses the output features of the encoder to gradually infer the results of knowledge completion through the decoder, and at the same time generates a sequence of rules related to completion. The reasoning results of each step are recorded to form a complete rule chain. Finally, the model outputs the complete reasoning results and the completed knowledge graph.

[0039] Preferably, the step of designing and implementing a multi-hop knowledge reasoning model for inferring the answer required by the user in the knowledge graph specifically includes:

[0040] Design a multi-hop knowledge reasoning model;

[0041] Use the constructed knowledge graph data to train and reason about the multi-hop knowledge reasoning model;

[0042] Infer the answers the user needs in the knowledge graph and output them.

[0043] Preferably, the structure of the multi-hop knowledge reasoning model specifically includes:

[0044] Knowledge graph representation module: uses Transformer Encoder to encode the entities and relations of the knowledge graph, aiming to capture the relationship and structural information between entities in the graph;

[0045] User intention understanding module: Use the pre-trained model ALBERT-BiLSTM to extract features of user questions, and add a CRF layer to extract entities on this basis;

[0046] Multi-hop reasoning module: maps the intent of the user's question to multiple relations and entities in the knowledge graph, and gradually derives the final answer through multi-hop reasoning across multiple relations and entities. The result of each hop of reasoning is used as the input of the next hop of reasoning, and combined with the previous reasoning information, it progresses layer by layer.

[0047] Activation function and loss function: The model uses the sigmoid activation function to map the output of the reasoning process to the probability space of [0,1]. During the model training process, the predicted result is compared with the true answer, and the error is calculated through the loss function.

[0048] Furthermore, the present invention also provides an intelligent fault diagnosis device for industrial domain knowledge collaboration, comprising a processor and a memory for storing executable instructions of the processor, wherein the processor is configured to execute the intelligent fault diagnosis method for industrial domain knowledge collaboration as described above by executing the executable instructions.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] 1. The method of the present invention utilizes knowledge completion reasoning to effectively supplement the missing entities and relationships in the industrial knowledge graph, complete the missing information, and solve the problem that knowledge missing affects the effect of multi-hop knowledge reasoning.

[0051] 2. The present invention provides a multi-hop knowledge reasoning method, which is used to solve the problem that multi-level and multi-step reasoning cannot be performed in the knowledge graph under complex industrial intentions. The intermediate information is saved during the multi-hop reasoning process, which can effectively serve as an explanation of the results and enhance the interpretability of knowledge reasoning. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0053] Figure 1 It is a flowchart of the intelligent fault diagnosis method for industrial knowledge collaboration of the present invention;

[0054] Figure 2 A detailed flow chart of the intelligent fault diagnosis method for industrial knowledge collaboration of the present invention;

[0055] Figure 3 This is a diagram of the knowledge completion reasoning model architecture of the present invention;

[0056] Figure 4 A comparison chart of the results obtained by testing different product data inference completions;

[0057] Figure 5 This is an example diagram of the industrial parts failure knowledge graph entity library. DETAILED DESCRIPTION

[0058] The present invention is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those of ordinary skill in the art, several changes and improvements can also be made without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

[0059] Specifically, the present invention takes the industrial fault diagnosis knowledge graph as an example to provide an intelligent fault diagnosis method for industrial field knowledge collaboration. Figure 1 and Figure 2 As shown, the method comprises the following steps:

[0060] S1: Collect and clean the original industrial data and construct the original data representation of the knowledge base;

[0061] Specifically, the industrial fault diagnosis data mainly comes from the on-site analysis and database entry of process and quality analysts. The step S1 specifically includes the following steps:

[0062] Step 11: Collect raw data from various industrial devices, sensors, log files, and other data sources;

[0063] Step 12: Clean and preprocess the collected raw data to ensure data availability;

[0064] Step 13: Structure the raw data and add tags and metadata to different data so that they can be associated with industrial equipment and parameters.

[0065] S2: Extract entities and relationships based on the original data representation of the knowledge base and design the graph ontology library;

[0066] Specifically, after obtaining the structured original data through step S1, it is used as the standard for unstructured data annotation, the entities and relationships existing in each document are annotated, the ontology library of the graph is designed according to the data, and the ontology-data mapping is performed. The step S2 specifically includes the following steps:

[0067] Step 21: Use natural language processing technology and data mining methods to extract entities and relationships from the original data;

[0068] Step 22: Define an ontology, determine the entity categories and relationship types in the ontology according to the needs of the actual industrial field, and design entity attributes;

[0069] Step 23: Use RDF language to create an ontology library and add entities, relationships and attributes to the ontology library;

[0070] Step 24: Map the extracted entities and relations with the entities and relations in the ontology library. Label the extracted entities to associate with the corresponding entities in the ontology library.

[0071] S3: Write the graph construction module according to the design in the graph ontology library to complete the construction of the industrial knowledge graph;

[0072] Specifically, step S3 includes the following steps:

[0073] Step 31: Extract relevant information from the original industrial data, including entities, attributes and relationships, and convert the extracted data into the format of graph data to ensure that it conforms to the model of the ontology library;

[0074] Step 32: Establish a graph database to store the constructed knowledge graph, load the extracted and converted data into the graph database, and ensure the integrity and consistency of the data.

[0075] Specifically, Python is used to read the data extracted in step 31, and then the data is connected to the graph database through Python, and the extracted data is stored in the graph database using the query language, such as Figure 5The figure shows a graph data model based on a graph database, which contains multiple entities and relationships between entities, and is used to describe fault records and related information. Entity description: Responsible department: indicates the specific department or team that handles the fault. Fault category: describes the categories or types of different faults. Recorder: The individual responsible for recording fault information. Whether the line is stopped: indicates whether the fault causes the equipment to stop. Cumulative time and cumulative number of times: record the time and number of occurrences of a specific fault. Fault date: records the date or timestamp when the fault occurs. Handling method: describes the way or method to solve the fault. Fault phenomenon: describes the specific symptoms or phenomena of the fault. Equipment process: indicates the workflow or stage of the equipment when the fault occurs.

[0076] The various entity contents in a fault record are considered to have relationships. These relationships connect different entities with lines to form a complex graph that describes the association relationship between the various entities.

[0077] S4: Text-process the user’s question so that the system can understand the user’s intention;

[0078] Specifically, step S4 mainly focuses on processing the user's intention, performing word segmentation and part-of-speech tagging on the question input by the user, and then abstracting the sentence so that the system can understand the user's intention. Step S4 includes the following steps:

[0079] Step 41: Perform text preprocessing on the question raised by the user, including removing extra spaces, punctuation marks and special characters;

[0080] Step 42: Use natural language processing tools to segment the text into fine-grained words and annotate them;

[0081] Step 43: Remove stop words and extract stems.

[0082] S5: Design and implement a knowledge completion reasoning model to complete the knowledge of the constructed industrial knowledge graph;

[0083] Specifically, step S5 includes the following steps:

[0084] Step 51: Collect the constructed industrial knowledge graph data, including entities, attributes, relationships, and existing knowledge, and prepare the knowledge graph data for use in the reasoning model, including converting it into a suitable data structure;

[0085] Step 52: Design a knowledge completion reasoning model;

[0086] Select a rule extraction and knowledge completion reasoning model based on the Transformer architecture, and design the structure of the reasoning model, including the form of input data, model architecture, and reasoning strategy, such as Figure 3As shown, specifically including:

[0087] 1. Encoder part:

[0088] Relation Encoder: The role of the relation encoder is to process the relationship between entities in the knowledge graph, extract and encode this relationship information to provide context for reasoning. It helps the model fully learn the logical relationship in the graph by aggregating the relationship features in the main entity subgraph.

[0089] Structural encoder: The structural encoder is responsible for processing the topological structure information in the knowledge graph. It extracts and encodes the structural features of the subgraph, enabling the model to capture the complex structural relationships between different entities and ensuring that the model has a more comprehensive understanding of the overall structure of the graph.

[0090] 2. Decoder part: The main function of the decoder is to receive the relationship and structural features extracted by the encoder part, and combine them with the original graph structure features of the input to perform reasoning through the attention mechanism. It generates reasoning results by jointly calculating the features output by the encoder and the masked graph structure features.

[0091] 3. Structural-Relational Attention Mechanism (SR-Attention): In the model, the structural-relational attention mechanism (SR-Attention) is introduced. This mechanism combines relational and structural information, and helps the model better focus on important features in the input by weighting the features of different channels, thereby improving the accuracy of the reasoning process. This mechanism can effectively solve the problem of inaccurate reasoning in the case of relational diversity.

[0092] 4. Reasoning process and rule extraction: The model uses the output features of the encoder to gradually infer the results of knowledge completion through the decoder, and at the same time generates a sequence of rules related to completion. The reasoning results of each step are recorded to form a complete rule chain. Finally, the model outputs the complete reasoning results and the completed knowledge graph.

[0093] Figure 4 This is a comparison chart of the results obtained by testing different product data inference completions. Figure 4 The data volume comparison of different recording points is shown, showing the original data volume and the newly added data volume after inference completion. The horizontal axis represents different recording points or data sets, and the vertical axis represents the size of the data volume. Each recording point presents two columns, representing the original data volume and the newly added data volume after inference completion. Figure 4 The figure clearly shows the change in the amount of new data after reasoning completion of different record points, which helps to analyze the contribution and distribution of the completed data to the overall data volume.

[0094] S6: Design and implement a multi-hop knowledge reasoning model to infer the answers required by users in the knowledge graph;

[0095] Specifically, step S6 includes the following steps:

[0096] Step 61: Design a multi-hop knowledge reasoning model;

[0097] Select an appropriate multi-hop knowledge reasoning model, such as deep reinforcement learning, graph neural network, etc., and design the structure of the model, including input representation, multi-hop reasoning mechanism, and output answer generation.

[0098] The purpose of multi-hop knowledge reasoning is to use the knowledge of the knowledge graph to connect knowledge to user questions. After the system recognizes the user's intention, it maps it to specific entities and relationships in the graph and searches the knowledge graph to find the answer the user needs. The specific modules are as follows:

[0099] 1. Knowledge graph representation module:

[0100] Knowledge Graph Encoding: In this module, the model first needs to embed entities (nodes) and relationships (edges) in the knowledge graph. The model uses Transformer Encoder to encode entities and relationships in the graph, aiming to capture the relationship and structural information between entities in the graph. In this way, the model can generate a high-dimensional vector representation that contains entities, relationships, and their contextual information.

[0101] The basis of multi-hop reasoning: The encoding of the knowledge graph allows the model to flexibly jump based on entity relationships in subsequent reasoning, ensuring that the model can effectively retain and process the information of the previous hop in multiple reasoning steps.

[0102] 2. User intention understanding module: Use the pre-trained model ALBERT-BiLSTM to extract features of user questions, and add a CRF layer to extract entities on this basis;

[0103] ALBERT-BiLSTM model: The model extracts features from user questions through ALBERT-BiLSTM. ALBERT is a lightweight pre-trained language model that can capture the context and semantic information in user questions through BiLSTM (bidirectional long short-term memory network).

[0104] CRF layer for entity extraction: In order to further accurately extract keywords corresponding to entities in the knowledge graph from user questions, the model adds a CRF (conditional random field) layer on top of the output of ALBERT-BiLSTM. This layer is mainly used for entity recognition and topic extraction, mapping user questions to specific entities and relationships in the knowledge graph to facilitate subsequent reasoning steps.

[0105] Question semantic understanding: Through this module, the system can accurately understand the user's questions, especially when multiple entities and relationships are involved. It can gradually break down the intention in the question and clarify which relationships are the core of the question and which entities are the starting point of reasoning.

[0106] 3. Multi-hop reasoning module:

[0107] Multi-hop reasoning process: This module is responsible for mapping the intent of the user's question to multiple relations and entities in the knowledge graph, and gradually derives the final answer through multi-hop reasoning across multiple relations and entities. The result of each hop reasoning is used as the input of the next hop reasoning, and combined with the previous reasoning information, it progresses layer by layer.

[0108] Multi-layer relation encoding: In each hop of reasoning, the model learns the context of multiple relations involved in the user question and uses ALBERT-BiLSTM to capture the semantic and dependency information in these relations. In this way, the model can not only handle simple one-to-one mapping problems, but also find complex connections between multi-step relations.

[0109] Joint reasoning of entities and relations: In this process, the model will jointly input the vector representation of the user's question and the vector representation of the corresponding entities and relations in the graph to generate reasoning results. Ultimately, the answer output by the model is gradually derived based on these multi-layer information jumps.

[0110] 4. Activation function and loss function:

[0111] Sigmoid activation: The model uses the sigmoid activation function to map the output of the reasoning process to the probability space of [0,1]. This means that the result of each hop of reasoning is output in the form of probability, reflecting the model's confidence in each candidate answer.

[0112] Loss calculation: During model training, the system compares the predicted result with the true answer and calculates the error through the loss function (usually binary cross entropy). The size of the loss value reflects the accuracy of the model's reasoning about the problem. The system optimizes the model's parameters through back propagation to reduce the error.

[0113] Step 62: Use the constructed knowledge graph data to train and reason about the multi-hop knowledge reasoning model;

[0114] Training process: Use the constructed industrial knowledge graph data for training. Through the multi-hop reasoning model, the system learns how to find the appropriate path in the complex graph structure to answer the user's questions;

[0115] Reasoning process: During actual operation, the system receives questions input by users, and after being processed by the user intent understanding module, it enters a multi-hop reasoning process, gradually searches for entities and relationships in the graph, and finally finds the answer that matches the question.

[0116] Step 63: Infer the answer the user needs in the knowledge graph and output it.

[0117] Furthermore, the present invention also provides an intelligent fault diagnosis device for industrial domain knowledge collaboration, comprising a processor and a memory for storing executable instructions of the processor, wherein the processor is configured to execute the above-mentioned intelligent fault diagnosis method for industrial domain knowledge collaboration by executing the executable instructions.

[0118] The above describes the specific embodiments of the present invention. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. In the absence of conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. An intelligent fault diagnosis method for industrial knowledge collaboration, characterized in that: The method comprises the following steps: Collect and clean the original industrial data and construct the original data representation of the knowledge base; Based on the original data representation of the knowledge base, extract entities and relationships and design the graph ontology library; Write the graph construction module according to the design in the graph ontology library to complete the construction of the industrial knowledge graph; Text the user's question so that the system can understand the user's intention; Design and implement a knowledge completion reasoning model to complete the knowledge of the constructed industrial knowledge graph; Design and implement a multi-hop knowledge reasoning model to infer the answers users need in the knowledge graph.

2. The intelligent fault diagnosis method for industrial domain knowledge collaboration according to claim 1 is characterized in that: The steps of collecting and cleaning the original industrial data and constructing the original data representation of the knowledge base specifically include: Collect raw data from various industrial devices, sensors, log files, and other data sources; Clean and preprocess the collected raw data to ensure data availability; Structure the raw data and add tags and metadata to different data so that it can be associated with industrial equipment and parameters.

3. The intelligent fault diagnosis method for industrial domain knowledge collaboration according to claim 1 is characterized in that: The steps of extracting entities and relationships and designing a graph ontology library for the original data representation of the knowledge base specifically include: Use natural language processing technology and data mining methods to extract entities and relationships from raw data; Define an ontology, determine the entity categories and relationship types in the ontology, and design entity attributes according to the needs of the actual industrial field; Use RDF language to create an ontology library and add entities, relationships and attributes to the ontology library; Map the extracted entities and relations with those in the ontology library. Label the extracted entities to associate them with the corresponding entities in the ontology library.

4. The intelligent fault diagnosis method for industrial domain knowledge collaboration according to claim 1 is characterized in that: The step of writing a graph construction module according to the design in the graph ontology library to complete the construction of the industrial knowledge graph specifically includes: Extract relevant information from raw industrial data, including entities, attributes, and relationships, and convert the extracted data into the format of graph data to ensure compliance with the model of the ontology library; Establish a graph database to store the constructed knowledge graph, load the extracted and converted data into the graph database, and ensure the integrity and consistency of the data.

5. The intelligent fault diagnosis method for industrial domain knowledge collaboration according to claim 1 is characterized in that: The step of text processing the user's question so that the system can understand the user's intention specifically includes: Perform text preprocessing on the questions asked by users, including removing extra spaces, punctuation marks and special characters; Use natural language processing tools to segment text into fine-grained words and annotate them; Perform stop word removal and word stemming.

6. The intelligent fault diagnosis method for industrial domain knowledge collaboration according to claim 1 is characterized in that: The steps of designing and implementing a knowledge completion reasoning model for completing knowledge of the constructed industrial knowledge graph specifically include: Collect the constructed industrial knowledge graph data, including entities, attributes, relationships, and existing knowledge, and prepare the knowledge graph data for use in the reasoning model, including converting it into a suitable data structure; Design a knowledge completion reasoning model.

7. The intelligent fault diagnosis method for industrial domain knowledge collaboration according to claim 6 is characterized in that: The knowledge completion reasoning model specifically includes: The encoder part includes: Relation Encoder: The role of the relation encoder is to process the relationship between entities in the knowledge graph, extract and encode these relationship information to provide context for reasoning; Structural encoder: The structural encoder is responsible for processing the topological structure information in the knowledge graph, extracting and encoding the structural features of the subgraph, so that the model can capture the complex structural relationships between different entities and ensure that the model has a more comprehensive understanding of the overall structure of the graph; Decoder: receives the relationship and structural features extracted by the encoder, combines them with the input original graph structural features, and performs reasoning through the attention mechanism; Structure-relation attention mechanism: Combining relational and structural information, by weighting the features of different channels, it helps the model better focus on the important features in the input, thereby improving the accuracy of the reasoning process; Reasoning process and rule extraction: The model uses the output features of the encoder to gradually infer the results of knowledge completion through the decoder, and at the same time generates a sequence of rules related to completion. The reasoning results of each step are recorded to form a complete rule chain. Finally, the model outputs the complete reasoning results and the completed knowledge graph.

8. The intelligent fault diagnosis method for industrial domain knowledge collaboration according to claim 1 is characterized in that: The steps of designing and implementing a multi-hop knowledge reasoning model to infer the answers required by users in the knowledge graph specifically include: Design a multi-hop knowledge reasoning model; Use the constructed knowledge graph data to train and reason about the multi-hop knowledge reasoning model; Infer the answers the user needs in the knowledge graph and output them.

9. The intelligent fault diagnosis method for industrial domain knowledge collaboration according to claim 8 is characterized in that: The structure of the multi-hop knowledge reasoning model specifically includes: Knowledge graph representation module: uses Transformer Encoder to encode the entities and relations of the knowledge graph, aiming to capture the relationship and structural information between entities in the graph; User intention understanding module: Use the pre-trained model ALBERT-BiLSTM to extract features of user questions, and add a CRF layer to extract entities on this basis; Multi-hop reasoning module: maps the intent of the user's question to multiple relations and entities in the knowledge graph, and gradually derives the final answer through multi-hop reasoning across multiple relations and entities. The result of each hop of reasoning is used as the input of the next hop of reasoning, and combined with the previous reasoning information, it progresses layer by layer. Activation function and loss function: The model uses the sigmoid activation function to map the output of the reasoning process to the probability space of [0,1]. During the model training process, the predicted result is compared with the true answer, and the error is calculated through the loss function.

10. An intelligent fault diagnosis device for industrial knowledge collaboration, characterized in that: The device includes a processor and a memory for storing executable instructions of the processor, and the processor is configured to execute the intelligent fault diagnosis method for industrial field knowledge collaboration as described in any one of claims 1 to 9 by executing the executable instructions.

Images