A fault diagnosis method, system and electronic device

Abstract

This invention relates to the field of data processing technology, specifically a fault diagnosis method, system, and electronic device. The method includes: retrieving abnormal information detected in real-time equipment operation data from a knowledge graph to generate a candidate fault pattern set; assigning weights to each candidate fault pattern in the set based on evidence information and calculating a unified score corresponding to each candidate fault pattern; sorting the candidate fault patterns based on the candidate fault pattern set and the unified score to obtain a fault hypothesis list; and generating a fault diagnosis result based on the fault hypothesis list and evidence information. This invention achieves intelligent equipment diagnosis and decision support by constructing an architecture that couples a multimodal knowledge graph with a dynamic reasoning engine, overcoming the shortcomings of existing pure data-driven or pure language model solutions in terms of reliability, interpretability, and knowledge fusion depth.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a fault diagnosis method, system, and electronic device. Background Technology

[0002] In modern industrial production, unplanned downtime of critical equipment can lead to huge economic losses. Therefore, predictive maintenance of equipment has become crucial. Traditional equipment fault diagnosis systems typically rely on rule-based systems or single machine learning models. Rule-based systems require experts to manually write a large number of rules, resulting in poor flexibility and difficulty in dealing with complex and ever-changing new faults. On the other hand, single machine learning models often focus on pattern recognition at the data level, lacking the integration and utilization of deeper domain knowledge such as the equipment's physical structure, operating principles, and historical maintenance history.

[0003] In recent years, large-scale language models have demonstrated powerful capabilities in text understanding and generation. Some technical solutions have proposed applying them to the field of equipment diagnostics, allowing the model to directly output diagnostic results from structured or unstructured text input. However, such solutions have significant drawbacks: First, their decision-making process resembles a "black box," lacking interpretability and thus difficult to trust. Second, their reasoning heavily relies on statistical regularities in the training data, making them prone to "illusions" and unreliable results when faced with rare faults or equipment combinations not covered by the training data. Finally, simply processing text information makes it difficult to deeply integrate high-frequency time-series data from real-time sensors with static, structured knowledge, limiting the comprehensiveness and accuracy of the diagnosis.

[0004] Therefore, there is an urgent need in this field for an intelligent diagnostic system that can deeply integrate multi-source heterogeneous data from devices, and whose reasoning process is transparent, explainable, traceable, and highly reliable. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a fault diagnosis method, system, and electronic device. The invention retrieves abnormal information detected in real-time equipment operation data from a knowledge graph to generate a candidate fault mode set. Based on evidence information, a weight is assigned to each candidate fault mode in the set, and a unified score corresponding to each candidate fault mode is calculated. A fault hypothesis list is obtained by sorting the candidate fault mode set and the unified score. Finally, a fault diagnosis result is generated based on the fault hypothesis list and the evidence information.

[0006] The technical means employed in this invention are as follows:

[0007] A fault diagnosis method includes: retrieving information from real-time equipment operation data in a knowledge graph to generate a candidate fault pattern set; assigning weights to each candidate fault pattern in the candidate fault pattern set based on evidence information, and calculating a unified score corresponding to any candidate fault pattern; sorting the candidate fault patterns based on the candidate fault pattern set and the unified score to obtain a fault hypothesis list; and generating a fault diagnosis result based on the fault hypothesis list and evidence information.

[0008] Furthermore, the knowledge graph is constructed by performing entity extraction on device knowledge data to obtain entities; based on the entities, the correspondence between entities is obtained by using relation extraction; the attribute information of the entities is obtained by using attribute extraction; and the knowledge graph is generated by combining the correspondence between entities and the attribute information.

[0009] The equipment knowledge data comes from at least: equipment static attributes, real-time sensor time-series data, daily inspection records, full life cycle data, historical operation database, and historical maintenance records.

[0010] Furthermore, the step of retrieving candidate fault mode sets from the knowledge graph based on the abnormal information detected in the real-time operation data of the equipment specifically includes: identifying the real-time operation data of the equipment to obtain abnormal data details; extracting sensor identifiers, abnormal types, and abnormal values ​​from the abnormal data details to obtain an abnormal information structure; generating a query request that can be used in the knowledge graph based on the abnormal information structure; and using the query request to retrieve candidate fault mode sets from the knowledge graph.

[0011] Furthermore, the evidence information includes: operational deviation information, historical probability information, and graph matching information; the determination method of the evidence information includes: determining the operational deviation information based on the degree of anomaly of real-time operational data; determining the historical probability information based on historical operational records corresponding to candidate fault modes; and determining the graph matching information based on the degree of matching between the anomaly information and the knowledge graph.

[0012] Furthermore, determining the operational deviation information based on the degree of anomaly in the real-time operational data specifically includes: determining the deviation degree based on the deviation range between the abnormal data details and the preset normal range; determining the time factor based on the abnormal duration corresponding to the abnormal data details; and obtaining the operational deviation information by weighted synthesis of the deviation degree and the time factor.

[0013] Furthermore, determining the historical probability information based on the historical operation records corresponding to the candidate fault modes specifically includes: calculating the occurrence frequency of each fault mode based on the operation records matching the current operating condition in the historical operation database; for the candidate fault mode, using the ratio of the occurrence frequency of the candidate fault mode to the total number of operations as the historical probability of the candidate fault mode; and normalizing the historical probability of occurrence to obtain the historical probability information.

[0014] Furthermore, determining the graph matching information based on the degree of matching between the abnormal information and the knowledge graph specifically includes: extracting feature pattern vectors based on the abnormal information; obtaining symptom feature vectors of the fault mode based on the knowledge graph; calculating the similarity between the feature pattern vectors and the symptom feature vectors to generate a matching confidence score; and mapping the matching confidence score to graph matching information according to a preset ratio.

[0015] Furthermore, the step of generating a fault diagnosis result based on the fault hypothesis list and evidence information specifically includes: generating a diagnostic confidence score based on the unified score of each candidate fault mode in the fault hypothesis list and the corresponding evidence information; selecting a target fault mode set from the fault hypothesis list based on the diagnostic confidence score; generating a recommended combination of treatment measures based on the entities associated with the target fault mode set in the knowledge graph; and generating a fault diagnosis result based on the target fault mode set, evidence information, and recommended combination of treatment measures.

[0016] This invention also provides a fault diagnosis system based on a fault diagnosis method, comprising: a knowledge graph retrieval module, a weighted calculation module, a list sorting module, and a result generation module, wherein: the knowledge graph retrieval module is used to retrieve abnormal information detected in real-time equipment operation data from a knowledge graph to generate a set of candidate fault modes; the weighted calculation module is used to assign weights to each candidate fault mode in the set of candidate fault modes based on evidence information and calculate a unified score corresponding to any candidate fault mode; the list sorting module is used to sort the candidate fault modes based on the set of candidate fault modes and the unified score to obtain a fault hypothesis list; and the result generation module is used to generate a fault diagnosis result based on the fault hypothesis list and evidence information.

[0017] The present invention also includes an electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform a fault diagnosis method.

[0018] Compared with the prior art, the present invention has the following advantages: The fault diagnosis method provided by this invention realizes intelligent equipment diagnosis and decision support by constructing an architecture that couples a multimodal knowledge graph with a dynamic reasoning engine, thus solving the defects of pure data-driven or pure language model solutions in terms of reliability, interpretability and knowledge fusion depth in the prior art.

[0019] This invention combines real-time operational data with a knowledge graph to rapidly generate candidate fault patterns, reducing the time required for traditional manual analysis. Based on evidentiary information, candidate fault patterns are scored and ranked, improving the reliability of fault diagnosis and avoiding blind repairs. By associating equipment structure, operating status, and fault patterns through a knowledge graph, systematic fault reasoning is achieved, reducing the possibility of omissions. The fault diagnosis process relies on data-driven and algorithm-assisted methods, significantly reducing reliance on manual intervention and improving diagnostic efficiency and reliability. The final fault diagnosis results not only clearly identify the cause of the fault but also provide targeted maintenance suggestions to help maintenance personnel quickly resolve the problem. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart of the fault diagnosis method in this invention.

[0022] Figure 2 This is a schematic diagram illustrating the implementation steps of the fault diagnosis method in this invention.

[0023] Figure 3 This is a framework diagram of the fault diagnosis method in this invention.

[0024] Figure 4 This is a schematic diagram of the fault diagnosis system in this invention.

[0025] Figure 5 This is a structural diagram of the electronic device in the fault diagnosis system of the present invention. Detailed Implementation

[0026] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0027] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the present invention or its application or use. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0029] Unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps described in these embodiments do not limit the scope of the invention. It should also be understood that, for ease of description, the dimensions of the various parts shown in the drawings are not drawn to actual scale. Techniques, methods, and devices known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and devices should be considered part of the specification. In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters in the following figures denote similar items; therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0030] like Figure 1 As shown, the present invention provides a fault diagnosis method, including: Based on the anomalies detected in the real-time equipment operation data, a knowledge graph is searched to generate a set of candidate fault modes. In implementation, real-time equipment operation data refers to the dynamic data generated during equipment operation collected by sensors, monitoring systems, etc., including physical information such as temperature, pressure, current, and vibration frequency. This data reflects the equipment's operating status in real time. Anomalies refer to abnormal phenomena or indicators detected through real-time operation data, including parameters exceeding set thresholds, abnormal fluctuations, and sudden fault signals. These anomalies are usually early signs of equipment failure.

[0031] In a specific implementation, as a preferred embodiment of the present invention, the knowledge graph is constructed by performing entity extraction on the device knowledge data to obtain entities; based on the entities, the correspondence between entities is obtained by using relation extraction; the attribute information of the entities is obtained by using attribute extraction; and the knowledge graph is generated by combining the correspondence between entities and the attribute information; the device knowledge data comes from at least: device static attributes, real-time sensor time-series data, daily inspection records, full life cycle data, historical operation database, and historical maintenance records.

[0032] In implementation, a knowledge graph refers to a structured data model based on knowledge, constructing a knowledge network by associating failure modes with equipment operating status. A knowledge graph can organize equipment functions, structure, failure modes, failure causes, and corresponding solutions in a graph format, facilitating retrieval and reasoning. Candidate failure modes refer to patterns or causes that may lead to failure, retrieved from the knowledge graph based on equipment anomaly information. Candidate failure modes are related to the equipment's failure history or operating configuration.

[0033] In a specific implementation, as a preferred embodiment of the present invention, the real-time operating data of the equipment is identified to obtain abnormal data details; based on the abnormal data details, sensor identifiers, abnormal types, and abnormal values ​​are extracted to obtain an abnormal information structure; based on the abnormal information structure, a query request that can be used for a knowledge graph is generated; using the query request, a search is performed in the knowledge graph to obtain a set of candidate fault modes.

[0034] In implementation, real-time operational data of the equipment can first be processed by anomaly detection algorithms (such as threshold analysis, statistical methods, or machine learning algorithms) to identify abnormal information. Then, based on the identified anomalies, a related retrieval can be performed in a knowledge graph. The knowledge graph is constructed based on the equipment's operational patterns and fault history, enabling the mapping of anomaly information to relevant fault modes. Finally, a candidate fault mode set can be generated by combining the retrieval results.

[0035] Based on the evidence information, a weight is assigned to each candidate fault mode in the candidate fault mode set, and a unified score is calculated for each candidate fault mode. Evidence information refers to supporting data or information related to the candidate fault modes. Evidence information includes abnormal parameters of the equipment, operating environment conditions, historical records, deviations of operating parameters from standard values, etc., which are important bases for judging the credibility of fault modes.

[0036] In a specific implementation, as a preferred embodiment of the present invention, the evidence information includes: operational deviation information, historical probability information, and graph matching information; the method for determining the evidence information includes: determining the operational deviation information based on the degree of anomaly in real-time operational data; determining the historical probability information based on historical operational records corresponding to candidate fault modes; and determining the graph matching information based on the degree of matching between the anomaly information and the knowledge graph.

[0037] Weight refers to the numerical value representing the relative importance or credibility of each candidate failure mode. In implementation, the allocation of weights is usually related to the degree of matching of evidence information, the priority of relevant failure modes in the knowledge graph, and other factors.

[0038] The unified score is an evaluation value calculated by comprehensively considering the weights corresponding to each candidate failure mode. In implementation, the unified score can be used to quantify the probability of each failure mode.

[0039] First, candidate failure modes are analyzed one by one based on the evidence. The analysis process includes determining whether abnormal parameters match the characteristics of the candidate failure modes and whether the failure modes conform to the current operating conditions of the equipment. Then, weights are assigned to each candidate failure mode based on the evidence; for example, failure modes with higher matching degrees have higher weights. Finally, a uniform score is calculated for each candidate failure mode using a specific calculation method (such as weighted scoring or Bayesian inference).

[0040] In a specific implementation, as a preferred embodiment of the present invention, determining the operational deviation information based on the degree of anomaly in the real-time operational data specifically includes: determining the deviation degree based on the deviation range between the abnormal data details and the preset normal range; determining the time factor based on the abnormal duration corresponding to the abnormal data details; and obtaining the operational deviation information by weighted synthesis of the deviation degree and the time factor.

[0041] In a specific implementation, as a preferred embodiment of the present invention, determining the historical probability information based on the historical operation records corresponding to the candidate fault modes specifically includes: calculating the occurrence frequency of each fault mode based on the operation records matching the current operating condition in the historical operation database; for the candidate fault mode, using the ratio of the occurrence frequency of the candidate fault mode to the total number of operations as the historical probability of the candidate fault mode; and normalizing the historical probability of occurrence to obtain the historical probability information.

[0042] In a specific implementation, as a preferred embodiment of the present invention, determining the graph matching information based on the degree of matching between the abnormal information and the knowledge graph specifically includes: extracting feature pattern vectors based on the abnormal information; obtaining symptom feature vectors of the fault mode based on the knowledge graph; calculating the similarity between the feature pattern vectors and the symptom feature vectors to generate a matching confidence score; and mapping the matching confidence score to graph matching information according to a preset ratio.

[0043] Based on the candidate fault mode set and unified score, the candidate fault modes are sorted to obtain a fault hypothesis list; based on the fault hypothesis list and evidence information, a fault diagnosis result is generated. In a preferred embodiment of this invention, a diagnostic confidence level is generated based on the unified score and corresponding evidence information for each candidate fault mode in the fault hypothesis list; based on the diagnostic confidence level, a target fault mode set is selected from the fault hypothesis list; based on the entities associated with the target fault mode set in the knowledge graph, a recommended combination of treatment measures is generated; and based on the target fault mode set, evidence information, and recommended combination of treatment measures, a fault diagnosis result is generated.

[0044] A fault hypothesis list is a list of possible faults generated based on candidate fault modes and their uniform scores. In implementation, each fault hypothesis in the list includes a specific fault mode, its cause, and its corresponding probability assessment. First, the set of candidate fault modes is sorted according to the calculated uniform scores. Fault modes with higher uniform scores are prioritized for inclusion in the fault hypothesis list so that subsequent diagnostics can begin with the most probable mode. The generation of this list provides a clear analytical path for fault diagnosis.

[0045] Fault diagnosis results refer to the causes and solutions for faults ultimately determined based on a list of fault hypotheses and supporting evidence. In practice, fault diagnosis results typically include the fault mode, the specific faulty component, the cause of the fault, and recommended maintenance measures. First, by combining the list of fault hypotheses with real-time equipment operating data, the fault mode that best matches the supporting evidence is selected through verification and reasoning. Specifically, the cause of the fault can be further confirmed through methods such as matching with expert knowledge and simulating fault characteristics. The final fault diagnosis result may include a detailed description of the fault mode, potentially damaged equipment components, the cause of the fault, and maintenance recommendations.

[0046] like Figure 4 As shown, the present invention also provides a fault diagnosis system based on a fault diagnosis method, comprising: a knowledge graph retrieval module, a weighted calculation module, a list sorting module, and a result generation module, wherein: the knowledge graph retrieval module is used to retrieve abnormal information detected in real-time equipment operation data from a knowledge graph to generate a set of candidate fault modes; the weighted calculation module is used to assign weights to each candidate fault mode in the set of candidate fault modes based on evidence information and to calculate a unified score corresponding to any candidate fault mode; the list sorting module is used to sort the candidate fault modes based on the set of candidate fault modes and the unified score to obtain a fault hypothesis list; and the result generation module is used to generate a fault diagnosis result based on the fault hypothesis list and evidence information.

[0047] like Figure 5As shown, the present invention also includes an electronic device, at least one processor, and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform a fault diagnosis method.

[0048] The electronic device includes, but is not limited to, fixed devices and / or mobile devices. For example, fixed devices include, but are not limited to, servers, which can be cloud servers or regular servers. Mobile devices include, but are not limited to, device fault diagnosis devices, which can be mobile phones, tablets, vehicle terminals, etc. In some possible implementations, the fault diagnosis method can also be implemented by the processor calling computer-readable instructions stored in memory.

[0049] Example 1 In the process of building a knowledge graph, equipment knowledge data is the foundation, covering equipment attributes, usage status, fault conditions, maintenance records, and other content. Equipment knowledge data includes data generated or recorded during equipment design, manufacturing, operation, and maintenance.

[0050] An entity refers to a core element extracted from equipment knowledge data. It can be a specific item, person, or location, or an abstract event or state, and typically possesses independent characteristics and attributes. In this embodiment of the invention, entities may include key objects related to equipment operation and maintenance, such as equipment model, sensors, fault modes, and components.

[0051] Static attributes of equipment refer to inherent properties of the equipment that do not change during operation. In this embodiment of the invention, static attributes of equipment may include equipment model, specifications, material, design parameters, purpose, etc., which are usually determined during the equipment design or manufacturing stage.

[0052] Real-time sensor time-series data refers to dynamic data collected in real time by sensors during equipment operation. In this embodiment of the invention, real-time sensor time-series data may include data on temperature, pressure, vibration frequency, current, voltage, etc., which change over time and can reflect the operating status and health condition of the equipment.

[0053] Routine inspection records refer to inspection data recorded manually or automatically during the daily use and maintenance of equipment. In this embodiment of the invention, routine inspection records may include equipment appearance inspections, operating status records, component wear conditions, and anomaly detections.

[0054] Lifecycle data refers to relevant data of equipment throughout its entire life cycle. In this embodiment of the invention, lifecycle data may include technical parameters during the design phase, quality control data during the manufacturing phase, status data during the operation phase, maintenance records during the maintenance phase, and assessment information during the scrapping phase.

[0055] Historical maintenance records refer to the historical information on maintenance, repair, and troubleshooting operations performed on equipment during operation. In this embodiment of the invention, historical maintenance records may include maintenance content, maintenance time, maintenance personnel, cause of failure, and handling methods.

[0056] When constructing a knowledge graph, the device knowledge data is first preprocessed, including format conversion and noise reduction. Then, Natural Language Processing (NLP) techniques or machine learning methods can be used to identify and label key elements in the data, such as device name, fault mode, and sensor type, and the identified elements are defined as entities. For example, entities may include "device model A," "sensor X," and "fault mode B."

[0057] In some implementations, static attributes of the equipment can be obtained directly by reading equipment design documents, specifications, or product manuals; real-time sensor time-series data can be collected and stored in a database; daily inspection records can be extracted from manual inspection reports or automatic inspection equipment records; various data from the design, manufacturing, operation, maintenance, and scrapping stages throughout the equipment's life cycle can be integrated to obtain full life cycle data; and historical maintenance records can be retrieved from equipment maintenance logs or maintenance management systems.

[0058] The correspondence between entities refers to the logical or physical relationship between different entities in a knowledge graph. In this embodiment of the invention, the correspondence between entities may include the compositional relationship between devices and components, the association relationship between devices and fault modes, and the monitoring relationship between devices and sensors, etc.

[0059] In this embodiment of the invention, the logical or physical relationships between different entities can be analyzed first. For example, the hierarchical relationship between equipment and components can be extracted from equipment structure data; the correspondence between sensors and monitoring indicators can be extracted from sensor configuration data; and the association between fault modes and equipment components can be extracted from fault records. Subsequently, the relationships between entities can be automatically identified using algorithms such as graph structure analysis or database association queries.

[0060] Attribute information refers to the detailed description or characteristic information of each entity. In this embodiment of the invention, attribute information may include the device model, sensor acquisition range, fault mode occurrence conditions, etc., which can help define the characteristics of the entity.

[0061] In the embodiments of the present invention, the attributes of each entity can be analyzed first. Exemplarily, the design parameters and specifications can be extracted for "Device Model A"; the monitoring range and accuracy can be extracted for "Sensor X"; the occurrence conditions and symptom descriptions can be extracted for "Fault Mode B". Subsequently, entity attributes can be identified through document parsing tools, database queries or semantic analysis algorithms.

[0062] In the embodiments of the present invention, the extracted entities can be organized in the form of nodes first, and then edges can be established according to the relationships between entities to form a graph structure. Then, the attribute information can be attached to each node to enrich the content of the knowledge graph, and finally a graph database is used to store and manage the knowledge graph.

[0063] Through the integration of entities, relationships and attributes, the originally scattered device knowledge data is systematized to form a structured knowledge graph, which is convenient for quick query and analysis. The knowledge graph can intuitively display the association between the device operation state and the fault mode, provide comprehensive knowledge support for fault diagnosis, and reduce the fault analysis time. The multi-source data such as device static attributes, dynamic data, inspection records, and life cycle data are unified and integrated to fully exploit the data value and provide comprehensive support for device management. The knowledge graph can provide intelligent decision support for device faults and maintenance plans through association analysis and reasoning, and improve the reliability of device operation.

[0064] In some embodiments, the knowledge graph at least includes: an entity layer, a relationship layer and an attribute layer. Among them, in the entity layer, devices, subsystems, components, sensors, fault modes, maintenance actions, etc. can be used as entities; in the relationship layer, the relationships between entities can be defined, including "component of", "installed on", "monitored by", "may induce", "measures that can be taken", etc.; in the attribute layer, entity attributes can be recorded, such as device model, normal threshold of the sensor, typical symptom description of the fault mode, etc. Particularly, this graph integrates the structured and semi-structured knowledge in device manuals, expert experience and historical cases.

[0065] In some embodiments, according to the abnormal information detected in the device real-time operation data, a candidate fault mode set is generated by retrieving in the knowledge graph, including: identifying the device real-time operation data to obtain the details of abnormal data; extracting the sensor identifier, abnormal type and abnormal value according to the details of abnormal data to obtain the abnormal information structure; generating a query request that can be used in the knowledge graph according to the abnormal information structure; and using the query request to retrieve in the knowledge graph to obtain the candidate fault mode set.

[0066] Anomaly data details refer to specific anomaly information detected from the real-time operating data of the device. In this embodiment of the invention, anomaly data details can be a detailed record of anomaly events, which may include the time of occurrence of the anomaly, the data source, and the anomaly manifestation.

[0067] In this embodiment of the invention, the real-time operating data of the device is transmitted through a sensor acquisition system. Subsequently, anomaly detection algorithms such as threshold detection, statistical analysis, or machine learning models can be used to identify anomalies in the data. After an anomaly is detected, a detailed record of the anomaly data is generated, including the time of the event, the location of the event, the data source, and the anomaly manifestation, forming an anomaly data detail.

[0068] Sensor identifiers refer to specific sensor information indicating the source of abnormal data. In this embodiment of the invention, sensor identifiers may include the name, number, or location of the sensor, and can be used to locate the source of abnormal data.

[0069] Anomaly type refers to the nature or category of the anomaly corresponding to the abnormal data. In this embodiment of the invention, anomaly type may include "overtemperature", "overload", "signal loss" or "abnormal data fluctuation", etc.

[0070] Abnormal values ​​refer to specific numerical information detected from abnormal data. In this embodiment of the invention, abnormal values ​​may include temperature values ​​read by the sensor, the fluctuation range of current values, etc.

[0071] An anomaly information structure refers to organizing information such as sensor identifiers, anomaly types, and anomaly values ​​extracted from anomaly data details into a unified structured data format. In this embodiment of the invention, the anomaly information structure facilitates subsequent processing and knowledge graph queries.

[0072] In this embodiment of the invention, key fields can first be extracted from the abnormal data details. For example, the specific sensor causing the anomaly can be located from the source of the abnormal data, thus obtaining the sensor identifier; subsequently, by analyzing the abnormal behavior or data characteristics, it can be classified into a specific type, thus obtaining the anomaly type; furthermore, the abnormal value can be obtained by extracting the sensor data values ​​at the time of the specific anomaly. Subsequently, the extracted key fields can be structured to form a unified anomaly information structure.

[0073] A query request refers to a query statement or request generated based on anomaly information structure, used to retrieve candidate fault patterns related to anomalies from a knowledge graph. In this embodiment of the invention, query requests can be used to match anomaly types and sensor locations, thereby identifying the relevant fault causes.

[0074] In this embodiment of the invention, query statements or request formats can be designed based on the anomaly information structure to make them compatible with the query interface of the knowledge graph. For example, a query request may include the identifier of the target sensor for locating relevant parts of the device; it may also include anomaly types for matching fault modes corresponding to such anomalies in the knowledge graph; and it may include anomaly values ​​for further filtering of fault modes that meet the criteria.

[0075] In this embodiment of the invention, nodes or subgraphs related to the anomaly information are retrieved from the knowledge graph based on the generated query request. The nodes in the knowledge graph may contain information such as device components, fault modes, and anomaly characteristics, while edges represent the relationships between nodes. After retrieval, a set of candidate fault modes that meet the criteria is returned. For example, the set of candidate fault modes could be "the possible fault modes corresponding to sensor X overheating include cooling failure, excessively high ambient temperature, abnormal motor load, etc."

[0076] By structuring abnormal information, the sensors and categories of anomalies can be quickly located, shortening problem analysis time. Generating structured query requests allows for efficient use of fault mode data from the knowledge graph, achieving accurate matching and improving retrieval efficiency. Utilizing the correlation information within the knowledge graph enables comprehensive analysis of multiple possible fault modes, providing a more comprehensive reference for subsequent diagnosis. Through automatic identification of abnormal data, structuring of information, and intelligent retrieval from the knowledge graph, efficient transformation from data to knowledge is achieved, reducing manual intervention.

[0077] In some embodiments, the evidence information includes: operational deviation information, historical probability information, and graph matching information; the evidence information can be determined in the following ways: determining operational deviation information based on the degree of anomaly in real-time operational data; determining historical probability information based on historical operational records corresponding to candidate fault modes; and determining graph matching information based on the degree of matching between anomaly information and the knowledge graph.

[0078] Operational deviation information refers to the degree of deviation between the current real-time operating state of the equipment and the normal operating baseline. In this embodiment of the invention, the operational deviation information can be quantified numerically to reflect whether the equipment's operating state is abnormal and the severity of the abnormality.

[0079] Historical probability information refers to the probability of a certain failure mode occurring under similar conditions, calculated through statistical analysis of historical operating records corresponding to candidate failure modes. In this embodiment of the invention, historical probability information is a type of failure probability inference information based on historical data.

[0080] Knowledge graph matching information refers to the degree of association between abnormal information and fault patterns in the knowledge graph. For example, if an abnormal piece of information has a high similarity to a known fault pattern in the knowledge graph, then the matching information value of that pattern is large.

[0081] The degree of anomaly refers to the severity of a device's current operating data deviating from the normal range. In this embodiment of the invention, the degree of anomaly can be quantitatively analyzed by comparing real-time data with thresholds, standard values, or normal operating ranges, and can be expressed as mild anomaly, moderate anomaly, or severe anomaly.

[0082] In this embodiment of the invention, real-time operating data of the device can be collected first. By analyzing the difference between the operating data and the normal operating range, deviation information can be calculated. For example, the calculation of deviation information can be performed using methods such as absolute deviation, relative deviation, and / or multi-dimensional comprehensive scoring. Then, based on the quantified value of the deviation information, the degree of abnormality can be divided into different levels.

[0083] Historical operation records refer to historical data related to equipment operating status, fault occurrence, and maintenance. In this embodiment of the invention, historical operation records may include time-series data of various equipment operating parameters, records of past fault causes, maintenance operation logs, etc., and can serve as an important data source for analyzing historical probability information.

[0084] In this embodiment of the invention, operational data related to candidate fault modes can first be extracted from the historical operation records of the equipment. Then, for candidate fault modes under similar operating conditions, their frequency of occurrence can be statistically analyzed, and the historical probability information of each fault mode can be calculated. Finally, conditional probability can be used for further refinement.

[0085] In this embodiment of the invention, abnormal information can be matched with fault patterns in a knowledge graph. For example, the degree of matching can be determined by matching key fields of the abnormal information with corresponding fields in the knowledge graph; NLP counting analysis can be used to analyze the semantic similarity between the abnormal information and the descriptive information in the graph; and a matching score can be calculated by combining multi-dimensional data such as abnormal type, sensor location, and device status. Furthermore, the graph matching information can be determined based on the matching degree of the fault pattern. For example, the higher the matching degree of the fault pattern, the larger the value of its graph matching information.

[0086] Operational deviation information, historical probability information, and knowledge graph matching information provide multi-dimensional evidence support from real-time data, historical data, and knowledge graphs, verifying the probability of failure modes from multiple perspectives and significantly improving diagnostic accuracy. Historical operation records and the knowledge graph contain rich fault knowledge and data; through matching and statistical analysis, possible failure modes can be quickly identified, reducing manual intervention time. Operational deviation information, by quantifying the degree of anomalies, provides a basis for prioritizing the importance of failure modes, helping maintenance personnel to prioritize the resolution of serious faults. Through the combination of data-driven and knowledge-driven approaches, it can dynamically adapt to complex operating environments and failure situations, exhibiting strong versatility and scalability.

[0087] In some embodiments, determining operational deviation information based on the degree of anomaly in real-time operational data includes: determining the deviation degree based on the deviation range between the abnormal data details and the preset normal range; determining the time factor based on the abnormal duration corresponding to the abnormal data details; and obtaining operational deviation information by weighted synthesis of the deviation degree and the time factor.

[0088] The preset normal range refers to the reasonable range of equipment operating data under normal working conditions. In this embodiment of the invention, the preset normal range can be determined based on equipment design parameters, operating specifications, or historical statistical data.

[0089] Deviation refers to the absolute difference between real-time operating data and the preset normal range.

[0090] Deviation refers to a standardized quantitative result of the deviation magnitude, used to measure the severity of the current anomaly. In this embodiment of the invention, a normalized value between 0 and 1 can be generated as the deviation value by calculating the ratio of the deviation magnitude to the difference between the upper and lower limits of the normal range.

[0091] In this embodiment of the invention, a preset normal range for the device can be determined in advance. For example, the upper and lower limits of the device's normal range can be extracted from the operation manual and historical data. Then, the extent to which the abnormal values ​​in the abnormal data details exceed the normal range can be calculated. For example, the difference between the abnormal value and the normal range can be calculated. Specifically, if the value is higher than the upper limit of the normal range, the deviation range = abnormal value - upper limit of the normal range; if the value is lower than the lower limit of the normal range, the deviation range = lower limit of the normal range - abnormal value. Further, after obtaining the deviation range, the degree of deviation can be calculated using a normalization method.

[0092] The duration of an anomaly refers to the length of time an abnormal state lasts. In this embodiment of the invention, the duration of an anomaly can be defined as the time interval from the moment the anomaly is detected until the abnormal state ends or returns to the normal range.

[0093] The time factor refers to a weighting coefficient applied to the severity of an anomaly based on its duration. In this embodiment of the invention, the longer the anomaly lasts, the greater the negative impact on equipment operation and safety, and the greater the weight of the time factor.

[0094] In this embodiment of the invention, the start and end times of the anomaly can be detected, and the duration of the anomaly can be calculated. For example, the anomaly duration = anomaly end time - anomaly start time. Subsequently, a time factor can be determined based on the length of the anomaly duration. For example, the process of determining the time factor can choose a linear weighting method where the time factor increases proportionally with the duration, or a non-linear weighting method using an exponential growth model where the increase in the time factor is more significant with longer anomaly durations. Specifically, the range of the time factor can be normalized to between 0 and 1.

[0095] In this embodiment of the invention, operational deviation information can be obtained by weighted calculation combining deviation and time factors. Specifically, the weighting coefficient can be used to balance the influence of deviation and time factors, and the weight value can be set according to actual needs. Furthermore, if the deviation is large but the duration is short, the deviation information may be small; if the deviation is small but the duration is long, the deviation information may be large. In particular, the operational deviation information can be normalized into a score value for subsequent fault analysis and ranking.

[0096] Operational deviation information, by combining deviation degree and time factors, quantifies the severity of anomalies into a comprehensive score, facilitating the prioritization of multiple anomalies and allowing for the treatment of those with the greatest impact on equipment operation. Deviation degree reflects the instantaneous difference between current operating data and the normal range, while the time factor reflects the persistence of the abnormal state; combining the two provides a more comprehensive and accurate anomaly assessment. Dynamically calculating deviation information based on real-time data acquisition enables real-time monitoring of equipment operating status, providing a reliable basis for subsequent failure mode analysis and reducing ambiguity and misjudgment. By adjusting the weights of deviation degree and time factors, the method can be adapted to the operational needs of different types of equipment or different scenarios, enhancing its versatility and flexibility.

[0097] In some embodiments, determining historical probability information based on historical operation records corresponding to candidate fault modes includes: calculating the occurrence frequency of each fault mode based on operation records matching the current operating condition in the historical operation database; for candidate fault modes, using the ratio of the occurrence frequency of the candidate fault mode to the total number of operations as the historical probability of occurrence of the candidate fault mode; and normalizing the historical probability of occurrence to obtain historical probability information.

[0098] A historical operation database refers to a database that stores all relevant data from the past operation of a device. In this embodiment of the invention, the historical operation database may include operating parameters, device status, fault records, maintenance logs, etc., and can serve as an important basis for statistical analysis of fault modes.

[0099] Current operating condition refers to the operating status of the equipment at the current moment. In this embodiment of the invention, the current operating condition may include operating parameters such as temperature, pressure, and vibration frequency, as well as environmental conditions such as load, operating mode, and external temperature. The operating condition data can be used to match historical operating records.

[0100] Operation records refer to specific data entries stored in the historical operation database. In this embodiment of the invention, operation records may include the time point of equipment operation, parameter values, operating condition information, and related fault mode annotations.

[0101] Frequency of occurrence refers to the number of times a candidate fault mode occurs in historical operating records that match the current operating conditions.

[0102] In this embodiment of the invention, all operating records similar to the current operating condition can be queried from the historical operating database first. For example, the query criteria may include similar operating parameters (e.g., similar temperature, pressure, vibration, etc. within a certain error range); or similar environmental conditions (e.g., the same or similar load and operating mode). Subsequently, for each candidate fault mode, the frequency of occurrence of that mode in the matched operating records can be counted. Specifically, the counting process can be a simple count or a combination of time-weighted methods.

[0103] Historical occurrence probability refers to the ratio of the frequency of occurrence of a certain failure mode to the total number of operations. In this embodiment of the invention, historical occurrence probability can be used to represent the likelihood of the failure mode occurring under specific historical conditions.

[0104] In this embodiment of the invention, the total number of runs in the historical operation database can be obtained first. Then, the frequency of occurrence of each candidate fault mode can be compared with the total number of runs to calculate the historical probability of occurrence of the candidate fault mode. Specifically, the historical probability of occurrence reflects the likelihood of the fault mode occurring in all operation records; the higher the value, the more likely the mode is to occur under the current operating conditions.

[0105] In this embodiment of the invention, the historical occurrence probabilities of all candidate failure modes can be normalized so that their sum equals 1, facilitating subsequent analysis and comparison. Specifically, the normalized historical probability information assigns a relative probability weight to each candidate failure mode; a larger value indicates a higher probability of the mode.

[0106] By analyzing historical operating records and combining them with current operating conditions, the probability of occurrence of each candidate fault mode is calculated, providing a historical data-based reference for fault diagnosis. Historical probability information quantifies the likelihood of candidate fault modes numerically, making fault diagnosis results more intuitive and easier to understand, and supporting subsequent fault mode ranking and decision-making. Fully utilizing the large data resources in the historical operating database, frequency statistics and probability analysis quickly filter out the most likely fault modes, reducing diagnosis time and manual intervention costs. Normalized historical probability information eliminates the dimensional differences of the original probability values, facilitating horizontal comparisons among multiple candidate fault modes and ensuring more scientific and reliable diagnostic results. By combining current operating conditions with historical records, it can dynamically adapt to the operating characteristics of different equipment and environments, improving the accuracy and applicability of diagnosis.

[0107] In some embodiments, determining graph matching information based on the degree of matching between abnormal information and knowledge graph includes: extracting feature pattern vectors based on abnormal information; obtaining symptom feature vectors of fault patterns based on knowledge graph; calculating the similarity between feature pattern vectors and symptom feature vectors to generate matching confidence; and mapping the matching confidence to graph matching information according to a preset ratio.

[0108] A feature pattern vector is a vectorized representation of structured feature data extracted from anomaly information. In this embodiment of the invention, data such as anomaly type, anomaly value, sensor identifier, and time information can be appropriately encoded or embedded to form a feature pattern vector.

[0109] In this embodiment of the invention, anomaly information can be collected first, including anomaly type, anomaly value, sensor identifier, and time information. Subsequently, this information can be structured and transformed into a vectorized representation. For example, the anomaly type can be represented using one-hot encoding; the anomaly value can be processed using a normalization method; and the time information can be encoded using a time step. Finally, the above features can be combined to form a numerical vector, resulting in a feature pattern vector.

[0110] A symptom feature vector is a vectorized representation of structured feature data extracted from the symptom descriptions of failure modes in a knowledge graph. In this embodiment of the invention, information such as the abnormal features, scope of influence, and possible causes of the corresponding failure mode can be appropriately encoded or embedded to form a symptom feature vector.

[0111] In this embodiment of the invention, a symptom description node corresponding to each candidate fault mode can be found in the knowledge graph, including abnormal features, impact range, sensor location, etc. Subsequently, this symptom information can be vectorized. For example, abnormal features can be normalized, while location features are identified using embedded vectors, and then the features are combined to form a symptom feature vector.

[0112] Match confidence refers to the similarity calculation result between the feature pattern vector and the symptom feature vector, used to represent the degree of association between abnormal information and a certain fault mode in the knowledge graph. In this embodiment of the invention, the match confidence can be a numerical value, ranging from 0 to 1, with a higher value indicating a higher degree of matching.

[0113] In this embodiment of the invention, a similarity calculation method can be used to compare the degree of matching between the feature pattern vector and the symptom feature vector. For example, similarity can be calculated using methods such as cosine similarity, Euclidean distance, and Manhattan distance. Subsequently, a matching confidence score can be generated based on the calculation results. Specifically, the matching confidence score can be normalized to a range of 0 to 1.

[0114] The preset ratio refers to the conversion rule or proportional relationship used to map the matching confidence value to the graph matching information.

[0115] In this embodiment of the invention, the matching confidence score can first be proportionally mapped to transform it into graph matching information. For example, the matching confidence score can be mapped to the graph matching information. Specifically, the mapping method can be linear or nonlinear, and can be flexibly adjusted according to requirements. The resulting graph matching information can serve as an important scoring basis for candidate fault modes.

[0116] By extracting feature vectors from anomaly information and fault modes and calculating similarity, real-time anomalies can be efficiently and accurately matched with known fault modes in the knowledge graph. Matching confidence and graph matching information are expressed numerically to represent the degree of correlation, facilitating the ranking of multiple candidate fault modes and helping maintenance personnel quickly pinpoint potential fault causes. The anomaly information and knowledge graph feature vectors can be derived from multi-dimensional data, thereby improving the diagnostic capability for complex faults. Through vectorization and similarity calculation, combined with data support provided by the knowledge graph, complex fault matching tasks can be automatically completed, reducing manual intervention. By adjusting the mapping rules with preset ratios, the diagnostic needs of different devices, scenarios, and fault types can be adapted, enhancing adaptability.

[0117] In some embodiments, generating a fault diagnosis result based on a fault hypothesis list and evidence information includes: generating a diagnostic confidence score based on the uniform score of each candidate fault mode in the fault hypothesis list and the corresponding evidence information; selecting a target fault mode set from the fault hypothesis list based on the diagnostic confidence score; generating a recommended combination of treatment measures based on the entities associated with the target fault mode set in the knowledge graph; and generating a fault diagnosis result based on the target fault mode set, evidence information, and recommended combination of treatment measures.

[0118] Diagnostic confidence refers to the credibility calculated based on the uniform score and evidence information of candidate failure modes, and is used to represent the probability that each candidate failure mode is determined to be a real failure. In this embodiment of the invention, diagnostic confidence can be expressed as a percentage or a numerical probability value, which can reflect the reliability of the diagnostic results.

[0119] In this embodiment of the invention, a weighted uniform score can be generated using the feature data of each candidate fault mode in the fault hypothesis list. Subsequently, the diagnostic confidence can be calculated based on the uniform score of the fault mode and the degree of support from the evidentiary information. For example, this can be achieved through a correlation function between the uniform score and the evidentiary information. Adjust the weights of the two. Finally, the diagnostic confidence score can be normalized to between 0 and 1 to represent the confidence level of the failure mode.

[0120] The target failure mode set refers to the set of failure modes with high confidence and consistent with the evidentiary information selected from the failure hypothesis list. In this embodiment of the invention, the target failure mode set contains the failure causes most likely to cause the current anomaly.

[0121] In this embodiment of the invention, fault modes that meet the preset filtering conditions can be selected from a list of fault hypotheses to form a target fault mode set. For example, the filtering conditions can be a diagnostic confidence threshold or a TOP-K diagnostic confidence level. Subsequently, the fault modes can be sorted from high to low based on their diagnostic confidence, prioritizing the fault modes with the highest confidence. Finally, the selected fault modes can be aggregated to form the target fault mode set.

[0122] A recommended set of remedial measures refers to a set of recommended measures for resolving a fault, generated based on the entities associated with the target fault mode set in a knowledge graph. In this embodiment of the invention, the recommended set of remedial measures is a targeted solution for the target fault mode.

[0123] In this embodiment of the invention, for each fault mode in the target fault mode set, the associated handling measure entities in the knowledge graph can be queried. Subsequently, the queried handling measures can be integrated into a recommended combination to form a specific fault solution. Specifically, the recommended handling measure combinations can be prioritized based on the confidence level of the target fault mode in the set, with handling solutions for high-confidence faults being recommended first.

[0124] In this embodiment of the invention, the target fault mode set, evidence information, and recommended handling measures can be combined and integrated into a complete diagnostic report. For example, the diagnostic report may include the cause and probability of the fault, the main evidence supporting the diagnosis, and specific solutions. Subsequently, fault diagnosis results can be generated in a structured manner, facilitating quick understanding and execution by users. Finally, the diagnostic results can be presented in the form of charts, lists, or flowcharts, allowing maintenance personnel to more intuitively understand the fault situation and handling recommendations.

[0125] The diagnostic confidence score calculation integrates a unified score and evidentiary information, quantifying the credibility of failure modes from multiple perspectives to identify the most likely causes of failures and ensure more accurate and reliable diagnostic results. Ranking failure modes by diagnostic confidence score helps maintenance personnel prioritize handling higher-impact failures and optimize resource allocation. Recommended treatment combinations are directly derived from associated entities in the knowledge graph, rapidly generating targeted solutions based on historical experience and rules, reducing the time cost of manual analysis. The fault diagnosis results integrate the target set of failure modes, evidentiary information, and recommended measures to form a structured report, helping maintenance personnel quickly understand the problem and take action. From failure mode selection to recommended measure generation, the entire process is automated, significantly reducing manual intervention and improving the level of intelligence.

[0126] In some implementations, the fault diagnosis method further includes data preprocessing. Specifically, it can receive and parse input multi-source heterogeneous device data, including device static attributes, real-time sensor time-series data, daily inspection records, full life-cycle data, and historical maintenance records, and perform preprocessing processes such as data cleaning and invalid information removal.

[0127] In some implementations, this fault diagnosis method can first perform online analysis of received real-time sensor data, using predefined thresholds or adaptive statistical models to detect abnormal data points or trends. Subsequently, the preprocessed data (such as abnormal sensor IDs and symptom texts discovered during inspections) can be converted into queries on a knowledge graph to quickly match potentially associated fault patterns, resulting in a set of candidate fault patterns. Next, the fault patterns matched by the graph can be weighted and scored based on multiple evidence sources, including the degree of real-time data anomaly, historical fault occurrence rate, and the confidence level of symptom description matching. Finally, based on the weighted scores, a list of fault hypotheses sorted in descending order of probability can be generated, thereby determining the fault diagnosis result.

[0128] In some implementations, structured diagnostic reports and recommendations can be generated based on the highest probability fault assumption and its reasoning (e.g., fault mode F is matched because sensor A exceeds the threshold, and similar situations account for X% in history).

[0129] Example 2 like Figure 2 As shown, the fault diagnosis method implementation steps include: receiving and cleaning multi-source input data from the equipment; analyzing real-time operating data to identify abnormal signals; matching abnormal signals, symptom descriptions, etc., with a knowledge graph to initially identify a candidate fault set; integrating multi-dimensional evidence to perform probability fusion and weighted calculation on the candidate fault set. This multi-dimensional evidence may include real-time outliers, historical statistical probabilities, text matching degrees, etc.; generating a ranked list of fault hypotheses based on the weighted scores; and generating a diagnostic report. This diagnostic report includes the most likely fault type, cause, recommended maintenance measures, and key inference evidence.

[0130] Taking the fault diagnosis of a certain model of ball mill as an example, in the process of constructing the knowledge graph, relevant information of the ball mill equipment is abstracted into entities, relationships, and attributes. For example, the extracted entities include "ball mill body", "feed end spindle", "lubricating oil", "bearing", "temperature sensor T01", "vibration sensor V01", "bearing overheating fault", and "poor lubrication", etc.; the extracted relationships include "bearing" as a "component" of "spindle", "temperature sensor T01" used to "monitor" the operating status of the bearing, and "poor lubrication" may "induce" "bearing overheating fault", which forms a structured association between entities; at the same time, the extracted attributes include the "normal threshold upper limit" of "temperature sensor T01" being 75℃, and the typical symptoms of "bearing overheating fault" being "the bearing part is hot to the touch" and "the lubricating oil color darkens", etc., which enrich the expressive power of the knowledge graph.

[0131] Furthermore, during real-time operation, the initial input data required for diagnosis is obtained by monitoring real-time data from the equipment and inspection records. At a certain operating moment, the collected real-time data includes a reading of 85°C for "Temperature Sensor T01," a normal operating status for "Vibration Sensor V01," and the inspection record showing "Lubrication Record - Delayed and Not Replaced." Subsequently, this operational data and inspection record are structured to extract abnormal information. The reading of "Temperature Sensor T01" is far above the normal threshold (75°C) and is marked as abnormal. At the same time, Natural Language Processing (NLP) is used to parse the text information of the inspection record, extracting the semantic association between "lubrication" and "delayed and not replaced," and matching it to the entity "poor lubrication" in the knowledge graph.

[0132] Furthermore, after marking the anomaly, the knowledge graph is used to query the possible fault modes associated with the current anomaly information. Specifically, for the anomaly of temperature sensor T01, the knowledge graph returns the path "T01 → Monitoring → Bearing → Possible Cause → Bearing Overheating Fault", indicating that bearing overheating is one of the potential causes of the temperature anomaly. Simultaneously, the "poor lubrication" parsed from the inspection record is further associated with "poor lubrication → possible cause → bearing overheating fault" through the knowledge graph. Based on this, the candidate fault is initially identified as "bearing overheating fault".

[0133] Furthermore, to quantify the credibility of candidate faults, this embodiment introduces a weighted calculation method to score the fault hypothesis based on multi-dimensional evidence. First, evidence 1 is the abnormal over-threshold condition of temperature sensor T01, with a current reading of 85℃, exceeding the normal threshold by 10℃. This evidence is significantly abnormal and directly related to "bearing overheating fault," therefore its weight factor is "high." Second, evidence 2 is that the probability of "poor lubrication" historically causing "bearing overheating fault" is 35% according to the knowledge graph. This evidence is supported by historical experience, and its weight factor is set to "medium." Third, evidence 3 is that the vibration sensor V01 is operating normally, ruling out the possibility of other faults caused by severe mechanical vibration. This information further supports "bearing overheating fault" as the primary cause of the fault, and its weight factor is also set to "medium." Through the fusion calculation of the above evidence information, the diagnostic confidence of "bearing overheating fault" is calculated to be the highest, and it is ultimately determined as the target fault mode for this fault diagnosis.

[0134] Finally, after the fault mode is determined, recommended solutions are generated by further analyzing the associated handling measures through the knowledge graph. Specifically, the handling measures associated with "bearing overheating fault" in the knowledge graph include "checking the quality of the lubricating oil," "changing the lubricating oil," "checking the accuracy of the bearing temperature sensor," and "checking for wear on the bearing body." These handling measures are combined to form a set of recommended fault handling measures, which are then output in JSON format. For example, the output content could be: "Faulty": true, "Main fault": Bearing overheating fault. Possible causes: Poor lubrication or failure of the lubricating oil. Recommended troubleshooting steps: Immediately stop the machine, check the bearing condition, replace the lubricating oil, and check the lubrication system circuit. Fault confidence level: 0.87 "Key evidence": Abnormal reading of spindle temperature sensor T01 (85℃), abnormal lubrication records.

[0135] Figure 3A framework diagram of a fault diagnosis method is shown, which includes: Data acquisition involves obtaining real-time operating data, inspection records, and other historical operating data of the equipment.

[0136] Data preprocessing involves structuring and cleaning the input data.

[0137] Time-series anomaly detection identifies outliers by analyzing the time series of real-time data and marks significant deviations in the data as potential anomalies.

[0138] Graph query and matching: invoke multimodal knowledge graphs to query possible associated fault modes based on abnormal data.

[0139] Evidence weighting and fusion involves weighting the evidence information obtained from matching abnormal data and knowledge graphs to generate a confidence ranking of candidate failure modes.

[0140] Hypothesis generation and ranking: Candidate failure hypotheses are generated based on the evidence weighting results and ranked according to confidence level.

[0141] The introduction of knowledge graphs allows for the acquisition of entities, relationships, and attributes related to the current data, enabling dynamic adjustments to the diagnostic process.

[0142] The decision output, based on the output of the knowledge reasoning engine, selects the candidate failure mode with the highest confidence level and generates the final diagnostic report or solution.

[0143] Output diagnostic results in a structured format, presenting them intuitively to the operator.

[0144] like Figure 4 As shown, the system of the present invention includes: a knowledge graph retrieval module, used to retrieve abnormal information detected in the real-time operation data of the equipment in a knowledge graph and generate a set of candidate fault modes; a weighted calculation module, used to assign weights to each candidate fault mode in the set of candidate fault modes based on evidence information and calculate a unified score corresponding to any candidate fault mode; a list sorting module, used to sort the candidate fault modes and the unified score to obtain a fault hypothesis list; and a result generation module, used to generate a fault diagnosis result based on the fault hypothesis list and evidence information.

[0145] The knowledge graph retrieval module includes: a data detail submodule, used to identify real-time operating data of the equipment and obtain abnormal data details; an abnormal structure submodule, used to extract sensor identifiers, abnormal types, and abnormal values ​​based on the abnormal data details to obtain an abnormal information structure; a request generation submodule, used to generate query requests that can be used in the knowledge graph based on the abnormal information structure; and a knowledge graph retrieval submodule, used to use the query requests to search in the knowledge graph and obtain a set of candidate fault modes.

[0146] The results generation module includes: a diagnostic confidence submodule, used to generate diagnostic confidence based on the unified score and corresponding evidence information of each candidate fault mode in the fault hypothesis list; a fault screening submodule, used to filter the target fault mode set from the fault hypothesis list based on the diagnostic confidence; a recommended measures submodule, used to generate a combination of recommended treatment measures based on the entities associated with the target fault mode set in the knowledge graph; and a results integration submodule, used to generate fault diagnosis results based on the target fault mode set, evidence information, and the combination of recommended treatment measures.

[0147] The fault diagnosis system in this invention combines real-time operational data with a knowledge graph to rapidly generate candidate fault patterns, reducing the time required for traditional manual analysis. Based on evidentiary information, candidate fault patterns are scored and ranked, improving the reliability of fault diagnosis and avoiding blind repairs. By associating equipment structure, operating status, and fault patterns through the knowledge graph, systematic fault reasoning is achieved, reducing the possibility of omissions. The fault diagnosis process relies on data-driven and algorithm-assisted methods, significantly reducing reliance on manual intervention and improving diagnostic efficiency and reliability. The final fault diagnosis results not only clearly identify the cause of the fault but also provide targeted maintenance suggestions to help maintenance personnel quickly resolve problems.

[0148] Figure 5 A schematic block diagram of an example electronic device 500 that can be used to implement embodiments of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices.

[0149] The device includes a computing unit that can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) or loaded from a storage unit into random access memory (RAM). The RAM can also store various programs and data required for device operation. The computing unit, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0150] Multiple components in the device are connected to the I / O interface, including: input units such as keyboards and mice; output units such as various types of displays and speakers; storage units such as disks and optical discs; and communication units such as network interface cards (NICs), modems, and wireless transceivers. The communication unit allows the device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0151] The computing unit can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of computing units include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit performs the various methods and processes described above, such as fault diagnosis methods. For example, in some embodiments, the fault diagnosis method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and / or installed on the device via ROM and / or a communication unit. When the computer program is loaded into RAM and executed by the computing unit, one or more steps of the fault diagnosis method described above may be performed. Alternatively, in other embodiments, the computing unit may be configured to perform fault diagnosis methods by any other suitable means (e.g., by means of firmware).

[0152] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A fault diagnosis method, characterized in that, include: Based on the abnormal information detected in the real-time operation data of the equipment, a set of candidate fault modes is generated by searching the knowledge graph. Based on the evidence information, a weight is assigned to each candidate fault mode in the candidate fault mode set, and a unified score corresponding to any candidate fault mode is calculated. Based on the candidate failure mode set and the unified score, the candidate failure modes are sorted to obtain a failure hypothesis list; Based on the list of fault hypotheses and evidence information, generate fault diagnosis results.

2. The fault diagnosis method according to claim 1, characterized in that, The knowledge graph is constructed by performing entity extraction on device knowledge data to obtain entities; based on entities, the corresponding relationships between entities are obtained by using relation extraction; and the attribute information of entities is obtained by using attribute extraction. Knowledge graphs are generated by combining the correspondence and attribute information between entities. The equipment knowledge data comes from at least: equipment static attributes, real-time sensor time-series data, daily inspection records, full life cycle data, historical operation database, and historical maintenance records.

3. The fault diagnosis method according to claim 1, characterized in that, The step of retrieving candidate fault mode sets from the knowledge graph based on abnormal information detected in the real-time operating data of the equipment includes: The system identifies real-time operating data of the equipment to obtain detailed abnormal data; based on the detailed abnormal data, it extracts sensor identifiers, abnormal types, and abnormal values ​​to obtain an abnormal information structure; based on the abnormal information structure, it generates a query request that can be used in a knowledge graph; using the query request, it performs a search in the knowledge graph to obtain a set of candidate fault modes.

4. The fault diagnosis method according to claim 1, characterized in that, The evidentiary information includes: operational deviation information, historical probability information, and map matching information; the methods for determining the evidentiary information include: The operational deviation information is determined based on the degree of anomaly in real-time operational data; the historical probability information is determined based on the historical operational records corresponding to the candidate fault modes; and the graph matching information is determined based on the degree of matching between the anomaly information and the knowledge graph.

5. The fault diagnosis method according to claim 4, characterized in that, The determination of operational deviation information based on the degree of anomaly in real-time operational data specifically includes: The deviation degree is determined based on the deviation range between the abnormal data details and the preset normal range; the time factor is determined based on the abnormal duration corresponding to the abnormal data details; and the operational deviation information is obtained by weighted synthesis of the deviation degree and the time factor.

6. The fault diagnosis method according to claim 4, characterized in that, The step of determining historical probability information based on historical operation records corresponding to candidate failure modes specifically includes: Based on the operation records in the historical operation database that match the current operating conditions, the frequency of occurrence of each fault mode is counted; for candidate fault modes, the ratio of the frequency of occurrence of the candidate fault mode to the total number of operations is used as the historical probability of occurrence of the candidate fault mode; the historical probability is normalized to obtain historical probability information.

7. The fault diagnosis method according to claim 4, characterized in that, The step of determining the graph matching information based on the degree of matching between the abnormal information and the knowledge graph specifically includes: Extract feature pattern vectors from abnormal information; obtain symptom feature vectors of fault patterns from knowledge graphs; calculate the similarity between feature pattern vectors and symptom feature vectors to generate matching confidence scores; and map the matching confidence scores to knowledge graph matching information according to a preset ratio.

8. The fault diagnosis method according to claim 1, characterized in that, The process of generating fault diagnosis results based on the fault hypothesis list and evidence information specifically includes: Based on the unified score and corresponding evidence information of each candidate fault mode in the fault hypothesis list, a diagnostic confidence score is generated; based on the diagnostic confidence score, a set of target fault modes is selected from the fault hypothesis list; based on the entities associated with the set of target fault modes in the knowledge graph, a combination of recommended treatment measures is generated; based on the set of target fault modes, evidence information, and combination of recommended treatment measures, a fault diagnosis result is generated.

9. A fault diagnosis system based on the fault diagnosis method according to any one of claims 1-8, characterized in that, include: The system comprises a graph retrieval module, a weighted calculation module, a list sorting module, and a result generation module, among which: The knowledge graph retrieval module is used to retrieve abnormal information detected in the real-time operating data of the equipment in the knowledge graph and generate a set of candidate fault modes. The weighted calculation module is used to assign weights to each candidate fault mode in the candidate fault mode set based on the evidence information, and to calculate a unified score corresponding to any candidate fault mode. The list sorting module is used to sort the candidate failure modes based on the candidate failure mode set and the unified score to obtain a failure hypothesis list. The result generation module is used to generate fault diagnosis results based on the fault hypothesis list and evidence information.

10. An electronic device, comprising: At least one processor; The at least one processor is also connected in communication with a memory, wherein the memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the fault diagnosis method according to any one of claims 1-8.

Images