Equipment diagnostic methods, devices and computer equipment

Abstract

This application relates to a device diagnostic method, apparatus, and computer equipment. It includes: determining the target entity in a user query request using a large model, accurately identifying the diagnostic question the user is concerned with; and performing multi-hop reasoning based on the target entity in a knowledge graph to deeply mine related knowledge and information, obtaining multiple initial reasoning paths. Based on equipment operating data, probabilistic analysis is performed on these initial reasoning paths to select the optimal target reasoning path. The large model then accurately processes the target reasoning path to obtain the diagnostic result. This method enables more comprehensive and accurate equipment fault diagnosis.

Description

Technical Field

[0001] This application relates to the field of industrial control technology, and in particular to a device diagnostic method, apparatus, and computer equipment. Background Technology

[0002] The stable operation of equipment plays a vital role in industrial production. For example, in the field of thermal power plants, various industrial equipment will inevitably experience various failures during long-term operation due to mechanical wear, improper operation, environmental factors and other factors.

[0003] Significant progress has been made in fault detection and diagnosis technologies, such as rule-based expert systems, which diagnose faults using pre-defined if-then rules and continuously maintain those rules. However, existing rule bases struggle to cover all industrial fault scenarios and require extensive manual maintenance, often presenting limitations in practical applications. Summary of the Invention

[0004] Based on this, the purpose of this application is to provide a more comprehensive and accurate method, apparatus and computer device for equipment diagnosis to solve the technical problems mentioned in the background art.

[0005] Firstly, this application provides a device diagnostic method. It includes:

[0006] Acquire user query requests and device operation data, and determine the target entity in the user query request through a large model;

[0007] Based on the target entity, multi-hop reasoning is performed in the knowledge graph to obtain multiple initial reasoning paths; the knowledge graph includes nodes representing entities and edges representing the relationships between entities.

[0008] Based on the device operation data, a probability analysis is performed on multiple initial inference paths to select the target inference path;

[0009] The diagnostic results are obtained by processing the target reasoning path using a large model.

[0010] In one embodiment, the knowledge graph construction process includes: acquiring multi-source data and a preset first prompt template; processing the multi-source data and the first prompt template using a large model to obtain extraction results; the extraction results include multiple original entities and original relationships identified from the multi-source data; and constructing a graph from the extraction results to obtain a knowledge graph.

[0011] In one embodiment, constructing a knowledge graph from the extraction results includes: standardizing the extraction results to obtain triple data constituting an initial knowledge graph; constructing a second prompt template based on the triple data; the second prompt template including known device information and the relationships between each known device information; and updating the initial knowledge graph using a large model and the second prompt template to obtain the knowledge graph.

[0012] In one embodiment, standardizing the extraction results to obtain triple data constituting the initial knowledge graph includes: preprocessing the original entities and original relations in the extraction results; the original entities include entity subjects and entity objects; the preprocessing includes at least establishing a terminology dictionary, unifying synonyms and polysemous terms; and converting the preprocessed entity subjects, original relations, and entity objects into triple data according to the triple structure.

[0013] In one embodiment, multi-hop reasoning based on the target entity in a knowledge graph is performed to obtain multiple initial reasoning paths, including: traversing the knowledge graph starting from the node where the target entity is located to determine multiple node information; the node information includes at least relationship strength and path length; when a special node is traversed, an anomaly detection algorithm is used to identify the anomaly pattern of the special node; and path scoring is performed based on the multiple node information and the anomaly pattern to obtain multiple initial reasoning paths.

[0014] In one embodiment, performing probability analysis on multiple initial inference paths based on the device operation data to select a target inference path includes: integrating each initial inference path and the device operation data through a Bayesian network to obtain multiple fault causes; constructing a decision tree associated with the device operation data; and performing posterior probability processing through the decision tree and the Bayesian network to select the initial inference path corresponding to the fault cause with a probability exceeding a preset value as the target inference path.

[0015] In one embodiment, the method further includes: obtaining optimization source information; the optimization source information includes at least user feedback, actual device processing results, diagnostic frequency of entities and relationships, information timeliness, or expert ratings; and optimizing the weights between entities and relationships in the knowledge graph based on the graph attention network and the optimization source information.

[0016] In one embodiment, the above method is applied in an intelligent agent cluster, which includes multiple intelligent agent systems that share information; the method further includes: determining the target type of the target device corresponding to the device operation data, and determining the target intelligent agent system from the multiple intelligent agent systems according to the target type; through the target intelligent agent system, realizing the process of determining the target entity in the user query request through a large model, until a diagnostic result is obtained.

[0017] Secondly, this application also provides a device diagnostic apparatus. It includes:

[0018] The data acquisition module is used to acquire user query requests and device operation data, and to determine the target entity in the user query request through a large model;

[0019] The path reasoning module is used to perform multi-hop reasoning in the knowledge graph based on the target entity to obtain multiple initial reasoning paths; the knowledge graph includes nodes representing entities and edges representing the relationships between the entities.

[0020] The result determination module is used to perform probability analysis on multiple initial inference paths based on the device operation data, and filter out the target inference path; the target inference path is processed by a large model to obtain the diagnostic result.

[0021] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the above-described device diagnostic method.

[0022] Fourthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the aforementioned device diagnostic method.

[0023] The aforementioned equipment diagnostic methods, devices, computer equipment, and readable storage media, through a large model, determine the target entity in the user's query request, accurately identifying the diagnostic questions the user is concerned with. Based on the target entity, multi-hop reasoning in a knowledge graph allows for in-depth mining of related knowledge and information, yielding multiple initial reasoning paths. Then, based on equipment operating data, probabilistic analysis is performed on these initial reasoning paths to select the optimal target reasoning path. Thus, the large model can accurately process the target reasoning path to obtain the diagnostic result. Therefore, this solution integrates multi-source data such as user query requests and equipment operating data. Through the semantic understanding capabilities of the large model and the reasoning capabilities of the knowledge graph, it can accurately select the optimal target reasoning path, achieving more comprehensive, accurate, and traceable fault diagnosis. Attached Figure Description

[0024] Figure 1 This is a diagram illustrating the application environment of a device diagnostic method in one embodiment;

[0025] Figure 2 This is a flowchart illustrating a device diagnostic method in one embodiment;

[0026] Figure 3 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0028] The device diagnostic method provided in this application embodiment can be applied to, for example, Figure 1 In the application environment shown, this is achieved through interaction between terminal 102 and server 104. Terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or placed on a cloud or other network server. Terminal 102 can obtain user query requests and device operation data. Server 104 is used to determine the target entity in the user query request using a large model, and performs multi-hop reasoning in the knowledge graph based on the target entity to obtain multiple initial reasoning paths. Server 104 is also used to perform probability analysis on the multiple initial reasoning paths based on device operation data, filter out the target reasoning path, and process the target reasoning path using a large model to obtain diagnostic results, which are then displayed by terminal 102. Terminal 102 can be, but is not limited to, various personal computers, smartphones, tablets, drones, smart vehicle devices, and portable wearable devices. Server 104 can be implemented using a standalone server or a server cluster composed of multiple servers. It can also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms.

[0029] In one embodiment, such as Figure 2 As shown, a device diagnostic method is provided, which is applied to... Figure 1 Taking a computer device including servers and terminals as an example, the process includes the following steps:

[0030] Step 202: Obtain user query requests and device operation data, and determine the target entity in the user query request through the large model.

[0031] Among them, "large model" is short for "large language model". It is a large-scale neural network language model that has gained powerful language understanding and generation capabilities on massive text data through pre-training. Large models can generate realistic text, understand complex language, and answer questions.

[0032] Specifically, the computer equipment acquires real-time operational data of the target device and uses a natural language interactive interface to perform speech recognition, converting the user's voice input into text. Then, a large model is used to perform intent recognition and entity extraction on the user input. Intent recognition is primarily used to understand the user's purpose or request type, such as determining whether the user wants to query device status, report a fault, or request maintenance guidance. Extracted target entities include, based on the request type, the device name, component name, parameter indicators, and fault type identified by the user within the interactive interface.

[0033] When the target equipment is a steam turbine, taking abnormal steam turbine vibration as an example, the maintenance personnel would input the observed abnormal phenomenon as a user query request via voice: "Steam turbine No. 1 vibration abnormality". The computer equipment would convert the voice into text and determine the intent as "report fault" through a large model, extracting the target entities as "Steam turbine No. 1" and "vibration abnormality".

[0034] Step 204: Perform multi-hop reasoning on the knowledge graph based on the target entity to obtain multiple initial reasoning paths.

[0035] Knowledge graphs consist of nodes representing entities and edges representing the relationships between entities. They are also a type of structured knowledge base, providing structured information for search, recommendation, question answering, and other functions. Multi-hop reasoning represents a process that starts from an entity and synthesizes information through multiple steps or relationship jumps to arrive at iterative, multi-level conclusions.

[0036] Specifically, multi-hop inference begins with the symptoms described by the user, such as "abnormal vibration." The computer device then locates the node containing the target entity in the knowledge graph and explores possible causes of the vibration abnormality using graph traversal algorithms such as depth-first search or breadth-first search, forming multiple initial inference paths. Each initial inference path has an initial confidence level, which decreases as the path lengthens.

[0037] Step 206: Perform probability analysis on multiple initial inference paths based on equipment operation data, select the target inference path, process the target inference path through a large model, and obtain the diagnostic results.

[0038] Among them are equipment operating data such as vibration spectrum and bearing temperature curve.

[0039] Specifically, the computer device uses a Bayesian network to integrate multiple inference paths and device operation data to perform posterior probability analysis and determine the most likely cause of the failure. The computer device then utilizes the generative capabilities of a large model to process the target inference path for the cause of the failure and outputs diagnostic results and processing suggestions in natural language.

[0040] For example, through knowledge graph reasoning and data analysis, the possible cause of the fault is determined to be "rotor imbalance". Multiple initial reasoning paths related to "rotor imbalance" are queried in the knowledge graph, and combined with the specific details of the current equipment operation data, a large model generates diagnostic results and handling suggestions in natural language, such as: "The diagnostic results show that turbine No. 1 may have a rotor imbalance problem. The following actions are recommended immediately: 1. Check for damage or fouling on the blades; 2. Conduct on-site dynamic balancing tests; 3. If conditions permit, consider reducing the unit's load. At the same time, closely monitor the vibration trend; if the vibration continues to worsen, it may be necessary to consider shutdown for maintenance."

[0041] In the aforementioned equipment diagnostic method, the target entity in the user's query request is determined through a large-scale model, accurately identifying the diagnostic issues the user is concerned about. Multi-hop reasoning based on the target entity within a knowledge graph allows for in-depth mining of related knowledge and information, yielding multiple initial reasoning paths. Then, probability analysis is performed on these initial reasoning paths based on equipment operation data to select the optimal target reasoning path. In this way, the large-scale model can accurately process the target reasoning path to obtain the diagnostic result. Therefore, this solution integrates multi-source data, including user query requests and equipment operation data. Through the semantic understanding capabilities of the large-scale model and the reasoning capabilities of the knowledge graph, it can accurately select the optimal target reasoning path, achieving a more comprehensive, accurate, and traceable fault diagnosis.

[0042] Furthermore, the natural language interface greatly enhances the usability of this solution and reduces the required level of expertise for users. Maintenance personnel can interact via voice or text to obtain diagnostic results and troubleshooting suggestions, enabling even inexperienced users to quickly and accurately diagnose and resolve faults.

[0043] In one embodiment, the knowledge graph construction process includes: acquiring multi-source data and a preset first prompt template; processing the multi-source data and the first prompt template using a large model to obtain extraction results; and constructing a graph from the extraction results to obtain a knowledge graph.

[0044] The extraction results include multiple original entities and original relationships identified from multi-source data.

[0045] Specifically, the computer equipment performs multi-source data acquisition and preprocessing, such as collecting operating data of thermal power plant equipment, equipment manuals, maintenance manuals, fault reports, and other multi-source data. Next, the structured multi-source data is cleaned and standardized, and unstructured text is converted into a processable format using technologies such as optical character recognition (OCR). Furthermore, the computer equipment is pre-designed with domain-specific prompt templates, such as "Please identify entities related to thermal power plant equipment in the following text, including equipment names, component names, parameter indicators, fault types, etc., and indicate the relationships between them. Text: [Input Text]".

[0046] The computer equipment also includes a pre-trained large model, which can be used for tasks such as named entity recognition and relation extraction. By inputting the constructed initial prompt template and preprocessed multi-source data into the large model, the model extracts information based on pre-trained knowledge and the requirements in the initial prompt template, and initially outputs the identified entities and relations. Then, regular expressions are used to parse the output of the large model, extracting entity and relation information to form a standardized output format, obtaining the original entities and original relations. Thus, a knowledge graph can be constructed based on the original entities and original relations.

[0047] For example, the multi-source data in the input text is: "An abnormality was found during a routine inspection of Unit 2's turbine generator set. The operator reported hearing abnormal noise and observing a significant increase in rotor vibration. Further inspection revealed that the turbine bearing temperature exceeded the normal operating range, reaching 95℃. The preliminary judgment is that this may be due to rotor imbalance or bearing wear. It is recommended to conduct an immediate detailed inspection and, if necessary, consider shutting down for maintenance to prevent the fault from escalating." Therefore, the extracted results are: (Unit 2 turbine generator set, found, abnormality), (abnormal noise, yes, symptom), (rotor vibration, significantly increased, none), (turbine bearing temperature, exceeds, normal operating range).

[0048] In this embodiment, by automatically integrating multi-source data and processing it using a large model and a first prompt template, the comprehensiveness and accuracy of knowledge graph construction are improved. The structured representation of knowledge graphs makes knowledge easier to discover and retrieve, supporting complex query and reasoning operations. The application of large models enhances the intelligence level of knowledge graph construction, enabling the graph to understand and process complex data relationships more deeply, thus providing comprehensive, efficient, and intelligent data support for various application scenarios. Therefore, combining multi-source data fusion, natural language processing with large model reasoning, and knowledge graphs significantly improves the accuracy, coverage, and interpretability of subsequent fault diagnosis.

[0049] In one embodiment, a computer device uses a graph database to store and manage knowledge graphs.

[0050] In one embodiment, constructing a knowledge graph from the extraction results includes: standardizing the extraction results to obtain triple data constituting the initial knowledge graph; constructing a second prompt template based on the triple data; and updating the initial knowledge graph using the large model and the second prompt template to obtain the knowledge graph.

[0051] The second prompt template includes known information about the device and the relationships between the known information about each device.

[0052] Specifically, the computer equipment converts the extracted original entities and relations according to a preset triple structure to obtain triple data that constitutes the initial knowledge graph. Next, it is necessary to determine whether the initial knowledge graph contains missing relations and perform predictive completion and updates, such as through statistical analysis, rule checking, or expert review to determine if missing relations exist.

[0053] After identifying the missing information, the computer determines the known information of the equipment based on the triplet data and infers possible relationships between the known information. For example, if the known information is: the steam turbine is the core equipment of the generator set; the rotor is a key component of the steam turbine; the bearing supports the rotor's operation; and increased vibration is a common symptom of equipment failure, the possible relationships inferred are: What relationship might exist between the steam turbine and increased vibration? What relationship might exist between the bearing and increased vibration? What relationship might exist between the rotor and increased vibration?

[0054] Then, the computer device inputs the constructed second prompt template into the large model, obtains the output results, analyzes the output of the large model, and extracts possible new relationships. For example, the large model outputs: 1. Steam turbine and increased vibration: The steam turbine may experience increased vibration, and increased vibration may be a symptom of steam turbine failure; 2. Bearing and increased vibration: Bearing failure may cause increased vibration, and increased vibration may indicate bearing problems; 3. Rotor and increased vibration: Rotor imbalance may cause increased vibration, and increased vibration may reflect abnormal rotor condition.

[0055] Finally, the computer system standardizes the new relationships output by the large model, such as: (steam turbine, may experience fault symptoms, increased vibration), (bearing failure, may cause, increased vibration), (rotor imbalance, may cause, increased vibration). The computer system then uses a rule engine or expert knowledge to verify the validity of these new relationships, adding the verified relationships to the initial knowledge graph and labeling their source and confidence level, thus obtaining the knowledge graph.

[0056] In this embodiment, the results of standardized extraction are converted into triplet data, which are then used to construct a second prompt template. The initial knowledge graph is then dynamically updated using a large model, which can significantly improve the accuracy and consistency of the knowledge graph, enabling it to respond quickly to new information and maintain the timeliness and relevance of knowledge.

[0057] In one embodiment, the computer device also incorporates an incremental update mechanism to continuously learn from and update the knowledge graph from new data. This incremental update mechanism is a dynamic and continuous process designed to maintain the timeliness and accuracy of the knowledge graph. It receives new knowledge provided by experts through a human input interface and then processes this information using modules such as change detection, knowledge extraction, consistency checks, and knowledge fusion. In practical applications, for example, if a new maintenance report is uploaded manually, the mechanism transforms it into structured knowledge and integrates it with the existing knowledge graph. Through version control and rollback mechanisms, the system can also manage the update history, ensuring the traceability and reliability of knowledge.

[0058] In one embodiment, the extraction results are standardized to obtain triple data constituting the initial knowledge graph, including: preprocessing the original entities and original relations in the extraction results; and converting the preprocessed entity subjects, original relations and entity objects into triple data according to the triple structure.

[0059] The original entities include entity subjects and entity objects; the preprocessing includes at least the establishment of a terminology dictionary, the standardization of synonyms, and the identification of polysemous terms.

[0060] Specifically, the computer equipment standardizes and preprocesses the extracted original entities and relationships, such as by unifying naming and eliminating synonyms, and then transforms them according to a triplet structure (subject, relation, object). For the thermal power plant sector, the computer equipment creates a standard terminology dictionary, including equipment names, component names, parameter specifications, and fault types. For example, equipment names might be: {steam turbine, boiler, generator...}, component names: {rotor, stator, bearing, blade...}, parameter specifications: {temperature, pressure, vibration, flow rate...}, and fault types: {imbalance, shaft misalignment, crack, corrosion...}.

[0061] Then, the large model is used to identify and unify synonyms or near-synonyms, such as "bearing overheating" -> "bearing temperature rise", "vibration" -> "vibration". Next, the naming format of the original entities and original relationships is unified. For example, the equipment name is changed to the full name: "No. 1 steam turbine" instead of "Unit 1", and the parameter indicators are given units such as "temperature (°C)" and "pressure (MPa)". Relation names use verbs or noun phrases: "cause", "includes components", "is a symptom of failure". Furthermore, terms with multiple meanings are processed. For example, "shaft" may refer to "rotor shaft" or "drive shaft", and the specific reference to "pressure" is determined according to the context, which may be "steam pressure" or "oil pressure".

[0062] Finally, the computer equipment converts the preprocessed entity subject, original relation, and entity object into triplet data, that is, into the form of (entity subject, original relation, entity object). For example: (Steam turbine generator No. 2, fault symptoms, abnormal noise), (Steam turbine generator No. 2, fault symptoms, increased vibration), (Steam turbine generator No. 2, includes components, rotor), (Steam turbine generator No. 2, includes components, bearing), (Bearing, abnormal parameters, temperature rise), (Bearing temperature, value, 95℃).

[0063] In this embodiment, by preprocessing the original entities and original relations and converting them into a standardized triplet data structure, the standardization and readability of the data can be significantly improved, making the data easier to manage and operate, and facilitating subsequent information retrieval, data analysis, and knowledge graph construction.

[0064] In one embodiment, multi-hop reasoning is performed on the knowledge graph based on the target entity to obtain multiple initial reasoning paths, including: traversing the knowledge graph starting from the node where the target entity is located to determine multiple node information; when a special node is traversed, an anomaly detection algorithm is used to identify the anomaly pattern of the special node; and path scoring is performed based on the multiple node information and the anomaly pattern to obtain multiple initial reasoning paths.

[0065] The node information includes at least the relationship strength and path length. Special nodes include those related to sensor data.

[0066] Specifically, during the traversal starting from the node containing the target entity, the computer device reads node information such as relationship strength and path length in the graph. When exploring a special node, such as "bearing temperature," it acquires sensor data related to that special node in real time. After preprocessing this sensor data, such as denoising and normalization, the computer device uses anomaly detection algorithms such as Isolation Forest or Autoencoder to identify abnormal patterns in the special nodes. The computer device considers relationship strength, path length, anomalousness of abnormal patterns, and similarity to historical cases in the graph, and determines different weight coefficients accordingly. It then uses this to perform a comprehensive path scoring based on a scoring function, and finally selects multiple initial inference paths based on the score.

[0067] For example, the scoring function is: Score=w1*R+w2*(1 / L)+w3*A+w4*S, where R is the relationship strength, L is the path length, A is the anomalousness of the abnormal pattern, S is the similarity of historical cases, and w1-w4 are their respective weight coefficients.

[0068] In this embodiment, by traversing the knowledge graph and using anomaly detection algorithms to identify abnormal patterns of special nodes, the accuracy and efficiency of anomaly detection can be improved. The path scoring mechanism can score the reasoning path based on multiple node information and abnormal patterns, which provides a more detailed and quantitative basis for decision-making and improves the accuracy of initial reasoning path determination.

[0069] In one embodiment, a probability analysis is performed on multiple initial inference paths based on device operation data to select a target inference path, including: integrating each initial inference path and device operation data through a Bayesian network to obtain multiple fault causes; constructing a decision tree associated with the device operation data; and performing posterior probability processing through the decision tree and Bayesian network to select the initial inference path corresponding to the fault cause with a probability exceeding a preset value as the target inference path.

[0070] Specifically, the computer device uses a Bayesian network to integrate multiple initial inference paths and real-time device operation data. Each inference path forms a conditional probability branch, with device operation data serving as observational evidence input into the network. Variational inference or Markov chain Monte Carlo methods are used to update the posterior probabilities, yielding the most probable cause of the failure. The computer device constructs a decision tree to refine the diagnosis, selecting the most discriminative device operation data and using historical cases for training and pruning. The probability distribution of the leaf nodes of the decision tree is then combined with the posterior probabilities of the Bayesian network to form a final ranking of multiple failure causes. Finally, the computer device extracts the initial inference paths corresponding to failure causes with probabilities exceeding preset values, inputs them into a large model, and generates a diagnostic report and treatment suggestions in natural language.

[0071] In this embodiment, by constructing a decision tree associated with device operation data, the system can make decisions based on real-time data, improving the immediacy and response speed of the decisions. By combining it with the posterior probability processing of Bayesian networks, the credibility of each fault cause can be evaluated, highlighting high-probability fault causes, thereby providing more focused and effective analysis.

[0072] In one embodiment, the above method further includes: obtaining optimized source information; and optimizing the weights between entities and relations in the knowledge graph based on the graph attention network and the optimized source information.

[0073] The optimized source information includes at least user feedback, actual equipment processing results, diagnostic frequency of entities and relationships, and information timeliness or expert ratings. User feedback refers to the acceptability of diagnostic results and their actual application effects; actual equipment processing refers to the problem-solving status after the implementation of diagnostic recommendations; diagnostic frequency of entities and relationships refers to the frequency of their appearance in successful diagnoses; information timeliness refers to the freshness of the information; and expert ratings refer to the importance scores given by domain experts to certain key relationships.

[0074] Specifically, the computer device also needs to perform feedback learning and continuous optimization, such as recording user feedback and actual processing results, and continuously optimizing the diagnostic model and knowledge graph using reinforcement learning techniques. Based on the collected optimization source information, a graph attention network (GAT) is used to dynamically adjust the weights of entities and relations in the knowledge graph. If the amplitude of oscillations significantly decreases after implementing the suggested actions, the computer device will use this positive result to reinforce paths in the knowledge graph. Simultaneously, through a contrastive learning mechanism, the computer device will analyze the differences between this successful diagnosis and previous failed diagnoses in similar situations, thereby optimizing its decision-making process.

[0075] Among them, graph attention network is a neural network model used to process graph structure data. It learns the weights between nodes through the attention mechanism, thereby obtaining the representation of the nodes.

[0076] In one embodiment, during the feedback learning and continuous optimization phase, the system not only records direct feedback from maintenance personnel but also analyzes subsequent equipment operation data. For example, if vibration amplitude significantly decreases after implementing suggested operations, the system will use this positive result to reinforce the current diagnostic path. Simultaneously, through a comparative learning mechanism, the system analyzes the differences between this successful diagnosis and previous failed diagnoses in similar situations, thereby optimizing its decision-making process.

[0077] In one embodiment, the computer device can also evaluate the effectiveness of knowledge graph optimization using the following metrics: 1. Diagnostic accuracy: The optimized knowledge graph improves the accuracy of fault diagnosis, which can be evaluated by comparing it with human diagnostic results, for example, a diagnostic accuracy rate higher than 85%. 2. Response time: Optimization should reduce the reasoning steps required for diagnosis and shorten the response time, for example, a response time of less than 5 seconds. 3. User satisfaction: Tracking user acceptance and feedback on the diagnostic results, for example, user satisfaction rate of over 90%.

[0078] In the above embodiments, through multi-dimensional feedback collection, comparative learning, and effect evaluation, continuous autonomous learning and evolution from practice are achieved. This comprehensive and closed-loop learning optimization mechanism can continuously improve its diagnostic capabilities, adapt to constantly changing equipment conditions and newly emerging fault types, and truly realize intelligent equipment fault diagnosis.

[0079] In one embodiment, the specific process of optimizing weights in a knowledge graph includes: 1. Graph structure representation: Converting the knowledge graph into a graph structure that can be processed by a graph attention network. 2. Feature vector construction: Constructing an initial feature vector for each node in the graph structure, including node type (e.g., device, symptom, cause, etc.), historical weights, usage frequency, and recent update timestamp. 3. Attention mechanism: For each node, the graph attention network calculates its attention coefficient with neighboring nodes. 4. Information aggregation: A new representation of a node is calculated by weighted sum of its neighbors. 5. Multi-head attention: Using multiple independent attention mechanisms, the results are then concatenated or averaged. 6. Feedback integration: Converting user feedback and actual device processing results into reward signals for adjusting attention weights. 7. Weight update: Updating the parameters of the graph attention network using a backpropagation algorithm to adjust the weights of entities and relationships. Therefore, weight optimization makes common faults and effective solutions more accessible in the knowledge graph, while rare or outdated information is weakened.

[0080] In one embodiment, the method further includes: determining the target type of the target device corresponding to the device operation data, and determining the target intelligent agent system from multiple intelligent agent systems according to the target type; and realizing the process of determining the target entity in the user query request through a large model through the target intelligent agent system until the diagnostic result is obtained.

[0081] This solution is applied in an intelligent agent cluster, which includes multiple intelligent agent systems that share information.

[0082] Specifically, through collaborative learning among multiple intelligent agent systems, for example, the target intelligent agent system responsible for the steam turbine may learn new vibration analysis techniques from intelligent agent systems responsible for other rotating equipment (such as fans, pumps, etc.), thereby obtaining the final diagnostic results and improving its own diagnostic capabilities. At the same time, the target intelligent agent system will also automatically generate new heuristic rules based on this diagnostic experience, such as "when the steam turbine vibration suddenly increases and is accompanied by an increase in bearing temperature, the rotor imbalance problem should be considered first".

[0083] Therefore, by introducing intelligent agent systems, each intelligent agent system is responsible for a specific type of equipment or fault, which enables the intelligent agent systems to share knowledge while maintaining their respective specialization.

[0084] In one embodiment, after receiving the diagnostic results, the user can engage in further interactive inquiries with the natural language interface based on the actual situation on-site, such as "What if it's a bearing problem? How long will it take to complete the dynamic balancing test?" The computer device will provide more detailed and personalized suggestions based on knowledge graphs and historical cases, combined with previous fault diagnosis results. This enables intelligent interaction between the powerful language understanding capabilities of the large model and maintenance personnel, and provides interpretable diagnostic results by combining structured knowledge graphs and the reasoning capabilities of the large model.

[0085] In one embodiment, if a computer device discovers a high degree of uncertainty in its understanding of vibration characteristic parameters, such as those in a specific frequency band, during the diagnostic process, it will proactively consult experts or suggest collecting more relevant data to continuously improve its own knowledge system.

[0086] In summary, this solution can be applied to an intelligent operation and maintenance management system for industrial control, mainly including the following application scenarios: (1) Real-time fault diagnosis: Operation and maintenance personnel describe the observed abnormal phenomena by voice through mobile devices or control room terminals, and quickly provide fault diagnosis and handling suggestions. (2) New employee training: Using knowledge graphs and interactive interfaces, an interactive equipment knowledge learning platform is provided for new employees. Therefore, the operation and maintenance management system can automatically construct and update the knowledge graph of the thermal power plant by integrating the operating data of thermal power plant equipment, relevant literature on the Internet and the descriptions of the operators. Operation and maintenance personnel can describe the fault phenomena by voice input, and the operation and maintenance management system combines the knowledge graph for intelligent analysis, and finally provides accurate fault diagnosis and handling suggestions, realizing intelligent diagnosis of thermal power plant equipment faults, and significantly improving the operation and maintenance efficiency and safety of thermal power plants.

[0087] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0088] Based on the same inventive concept, this application also provides a device diagnostic apparatus for implementing the device diagnostic method described above. The solution provided by this apparatus is similar to the implementation described in the above method; therefore, the specific limitations in one or more device diagnostic apparatus embodiments provided below can be found in the limitations of the device diagnostic method described above, and will not be repeated here.

[0089] In one embodiment, a device diagnostic apparatus is provided, comprising: a data acquisition module, a path reasoning module, and a result determination module, wherein:

[0090] The data acquisition module is used to acquire user query requests and device operation data, and to determine the target entity in the user query request through a large model.

[0091] The path reasoning module is used to perform multi-hop reasoning in the knowledge graph based on the target entity to obtain multiple initial reasoning paths; the knowledge graph includes nodes that represent entities and edges that represent the relationships between entities.

[0092] The result determination module is used to perform probability analysis on multiple initial inference paths based on equipment operation data, filter out the target inference path, and process the target inference path through a large model to obtain the diagnostic result.

[0093] The various modules in the above-mentioned device diagnostics can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of the computer device in hardware form or independent of it, or they can be stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0094] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 3As shown, the computer device includes a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When executed by the processor, the computer program implements a device diagnostic method. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.

[0095] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0096] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0097] In one embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0098] In one embodiment, a computer program product or computer program is provided, the computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and executes the computer instructions, causing the computer device to perform the steps in the above method embodiments.

[0099] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0100] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0101] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A device diagnostic method, characterized in that, The method includes: Acquire user query requests and device operation data, and determine the target entity in the user query request through a large model; Based on the target entity, multi-hop reasoning is performed in the knowledge graph to obtain multiple initial reasoning paths; the knowledge graph includes nodes representing entities and edges representing the relationships between entities. Based on the device operation data, a probability analysis is performed on multiple initial inference paths to select the target inference path. The target inference path is then processed by a large model to obtain the diagnostic results. The process of constructing the knowledge graph includes: Acquire multi-source data and preset first prompt templates; The multi-source data and the first prompt template are processed by a large model to obtain extraction results; the extraction results include multiple original entities and original relationships identified from the multi-source data; The extraction results are used to construct a knowledge graph, which includes: standardizing the extraction results to obtain triple data that constitutes the initial knowledge graph; constructing a second prompt template based on the triple data; the second prompt template includes known device information and the correlations between each known device information; and updating the initial knowledge graph using a large model and the second prompt template to obtain the knowledge graph. The method further includes: Obtain optimization source information; the optimization source information includes at least user feedback, actual device processing results, diagnostic frequency of entities and relationships, information timeliness, or expert scores; Based on the graph attention network and the optimization source information, the weights between entities and relations in the knowledge graph are optimized.

2. The method according to claim 1, characterized in that, The standardization of the extraction results yields triplet data constituting the initial knowledge graph, including: The original entities and original relationships in the extraction results are preprocessed; the original entities include entity subjects and entity objects; The preprocessing includes at least establishing a terminology dictionary and unifying synonyms and polysemous terms; Based on the triple structure, the preprocessed entity subject, original relation, and entity object are converted into triple data.

3. The method according to claim 1, characterized in that, The process of performing multi-hop reasoning on the knowledge graph based on the target entity to obtain multiple initial reasoning paths includes: Starting from the node where the target entity in the knowledge graph is located, traverse the graph to determine multiple node information; the node information includes at least relationship strength and path length. When a special node is encountered, an anomaly detection algorithm is used to identify the anomaly pattern of the special node. Path scoring is performed based on multiple node information and anomaly patterns to obtain multiple initial inference paths.

4. The method according to claim 1, characterized in that, The step of performing probability analysis on multiple initial inference paths based on the device operating data to select a target inference path includes: By integrating the initial inference path and device operation data through a Bayesian network, multiple causes of failure were obtained. Construct a decision tree associated with the device's operational data; By performing posterior probability processing using the decision tree and Bayesian network, the initial inference path corresponding to the fault cause with a probability exceeding a preset value is taken as the target inference path.

5. The method according to any one of claims 1 to 4, characterized in that, The method is applied in an agent cluster, which includes multiple agent systems sharing information; the method further includes: Determine the target type of the target device corresponding to the device operation data, and determine the target intelligent agent system from multiple intelligent agent systems based on the target type; The target intelligent agent system enables the process of determining the target entity in the user's query request through a large model, until a diagnostic result is obtained.

6. A device diagnostic apparatus, characterized in that, The apparatus for performing the method according to any one of claims 1 to 5 comprises: The data acquisition module is used to acquire user query requests and device operation data, and to determine the target entity in the user query request through a large model; The path reasoning module is used to perform multi-hop reasoning in the knowledge graph based on the target entity to obtain multiple initial reasoning paths; the knowledge graph includes nodes representing entities and edges representing the relationships between the entities. The result determination module is used to perform probability analysis on multiple initial inference paths based on the device operation data, and filter out the target inference path; the target inference path is processed by a large model to obtain the diagnostic result.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 5.

Images