Motor-oriented fault mode knowledge graph construction, updating method and system thereof

Abstract

This invention discloses a method and system for constructing and updating a fault mode knowledge graph for motors. The method includes: constructing a motor fault domain ontology model: defining core concept class hierarchy, object attributes and semantic relationships, acquiring structured and unstructured data and defining data attributes, and setting logical constraints and axioms; dual-channel knowledge extraction: using a deep learning model to extract knowledge from unstructured data and a mapping method to extract knowledge from structured data, generating original triplet knowledge; knowledge fusion: integrating knowledge from different data sources through entity alignment, attribute fusion and conflict resolution to construct a high-quality, unambiguous knowledge graph and storing it in a graph database; and intelligent fault reasoning and dynamic updating based on the knowledge graph. This invention fuses multi-source data, more accurately identifies long and difficult entities and complex semantic relationships, and improves the accuracy and efficiency of fault troubleshooting through a dual-weighted algorithm.

Description

Technical Field

[0001] This invention relates to the field of knowledge graph construction technology, specifically to a method and system for constructing and updating a fault mode knowledge graph for motors. Background Technology

[0002] Electric motors (especially permanent magnet synchronous motors) are the core power source for modern industrial production lines, new energy vehicles, and aerospace equipment. Their operational reliability is directly related to the safety and efficiency of the entire system. With the advancement of intelligent manufacturing, fault diagnosis technology for motors has transformed from traditional single signal threshold monitoring to knowledge-based intelligent diagnosis.

[0003] In the full life cycle management of motors, a massive amount of fault-related data has been accumulated, which can be mainly divided into two categories: one is structured data generated during the design and manufacturing stage, such as Failure Mode and Effects Analysis (FMEA) tables, which contain standard failure modes, failure mechanisms and theoretical risk levels (such as RPN values) defined by domain experts; the other is unstructured data generated during the operation and maintenance stage, such as maintenance work orders, fault logs, expert diagnostic reports, etc., which record the actual fault phenomena, handling processes and replacement of spare parts on site.

[0004] However, existing technologies face the following serious technical challenges when using this data to build fault diagnosis systems:

[0005] First, the "island effect" of multi-source heterogeneous data leads to low knowledge utilization. Existing diagnostic systems are often too simplistic. FMEA-based systems lack support from field measurement data, resulting in a disconnect between theoretical risks and actual fault distribution (for example, theoretically high-risk faults rarely occur in practice). Log mining-based systems, on the other hand, lack expert theoretical guidance, making it difficult to discover deep-seated fault patterns and easily limited by non-standard text descriptions. Current technologies lack an effective mechanism for deep semantic fusion of "theoretical specifications at the design end" and "practical experience at the operation and maintenance end."

[0006] Second, there is a lack of deep semantic understanding and reasoning capabilities. Traditional fault knowledge bases are mostly based on keyword matching or shallow associations. For example, when searching for "excessive motor vibration," existing systems struggle to discover, through multi-hop reasoning, the deep causal chain that "bearing cage fracture" leads to "rotor dynamic imbalance," and thus causes "excessive vibration." Furthermore, the field of motors contains numerous technical terms (such as "cogging torque" and "inter-turn short circuit"), and traditional word segmentation and entity recognition methods suffer from low accuracy when faced with complex maintenance texts.

[0007] Third, the knowledge base is static and rigid, lacking a dynamic adaptive update mechanism. Existing expert system knowledge bases are typically built once. However, as the motor operating environment changes and ages, new fault modes emerge (such as new types of bearing electro-corrosion caused by improper inverter carrier frequency settings), or the frequency of certain faults drifts over time. Current technology lacks a self-growing mechanism that can automatically "learn" new knowledge from continuously added maintenance records and adjust fault association weights in real time, causing the timeliness of the diagnostic model to decrease over time.

[0008] In summary, there is an urgent need for a method to construct a motor fault mode knowledge graph that can deeply integrate data from the design and operation and maintenance ends, possess deep semantic reasoning capabilities, and be dynamically updated with business data. Summary of the Invention

[0009] To overcome the shortcomings of the above technologies, this invention provides a method and system for constructing and updating a fault mode knowledge graph for motors.

[0010] The technical solution adopted by this invention to overcome its technical problems is:

[0011] A method for constructing and updating a fault mode knowledge graph for motors, including the following steps:

[0012] Constructing an ontology model for the motor fault domain: Define a hierarchy of core concept classes to represent motor fault knowledge, define object attributes and semantic relationships, acquire structured and unstructured data and define data attributes, and then set logical constraints and axioms;

[0013] Dual-channel knowledge extraction: Based on the aforementioned motor fault domain ontology model, a deep learning model is used to extract knowledge from unstructured data, and a mapping method is used to extract knowledge from structured data to generate original triplet knowledge.

[0014] Knowledge fusion: The original triple knowledge is subjected to entity alignment, attribute fusion and conflict resolution. Knowledge from different data sources is integrated to construct a high-quality, unambiguous knowledge graph and stored in a graph database.

[0015] Intelligent fault reasoning and dynamic updating based on knowledge graph: Utilizing the constructed knowledge graph, in response to user diagnostic requests, intelligent reasoning and ranking from fault phenomena to root causes are achieved based on a preset dual-weighted algorithm; at the same time, if the work order contains new components or new fault relationships that do not exist in the knowledge graph and the extraction confidence is higher than the preset new knowledge threshold, it is incorporated into the main knowledge graph to achieve dynamic updating of the knowledge graph.

[0016] Furthermore, the core concept class hierarchy is a top-down five-dimensional concept tree structure, with the five dimensions being equipment and component class, fault phenomenon class, fault mode class, diagnosis and detection class, and maintenance decision class; the object attributes and semantic relationships include at least the compositional relationship connecting equipment and components, the causal relationship connecting fault modes and fault phenomena, the diagnostic relationship connecting fault modes and detection methods, the positional relationship connecting fault modes and equipment components, and the governance relationship connecting fault modes and maintenance measures; the data attributes include at least attributes for the fault mode class and attributes for the maintenance decision class.

[0017] Furthermore, the knowledge extraction from unstructured data using a deep learning model specifically includes:

[0018] A BERT-BiLSTM-CRF model is constructed to perform entity recognition on unstructured data. The BERT-BiLSTM-CRF model includes a BERT layer for converting unstructured data into context-related vectors, a BiLSTM layer for bidirectional extraction of sequence context features, and a CRF layer for annotating and decoding entity label sequences.

[0019] After identifying entities, an attention-based classification model is used to extract semantic relationships between entities, thereby generating triplet knowledge.

[0020] Furthermore, the knowledge extraction from structured data using a mapping method specifically includes:

[0021] Based on the preset field-entity mapping rules, the field content in the structured data is mapped to the corresponding entity in the motor fault domain ontology model;

[0022] Based on attribute extraction rules, the Risk Priority Number (RPN) in the structured data is extracted and calculated as an attribute of the failure mode node.

[0023] Based on relational construction rules, (fault mode, cause, fault phenomenon) triples and (fault mode, location, device component) triples are automatically instantiated according to the logical structure of structured data.

[0024] Furthermore, the entity alignment specifically includes:

[0025] The semantic vector representation of all entity names in the original triples is extracted using the BERT-BiLSTM-CRF model;

[0026] Calculate and determine the semantic similarity of entities. If the similarity exceeds the preset alignment threshold, merge them, retain the standard name of the entity, and record the alias.

[0027] Furthermore, the attribute fusion adopts a master-slave strategy, specifically including:

[0028] The nodes generated from structured data serve as the primary data source, preserving prior theoretical attributes, which include at least risk priority, severity, occurrence, and detectability.

[0029] The nodes generated from unstructured data are used as the data source to statistically analyze the actual frequency of failures and the average repair time.

[0030] The statistical attributes from the data source will be attached to the entity nodes determined by the main data source to achieve attribute fusion.

[0031] Furthermore, the conflict resolution specifically includes:

[0032] In case of conflicting theoretical parameters, structured data shall prevail.

[0033] In case of conflicting measured parameters, unstructured data shall prevail.

[0034] For fused data with a confidence level below the preset threshold, the data confirmed by manual review shall prevail.

[0035] Furthermore, the process of utilizing the constructed knowledge graph to respond to user diagnostic requests and achieving intelligent reasoning and ranking from fault symptoms to root causes based on a preset dual-weighted algorithm specifically includes:

[0036] Using the trained deep learning model, the query statement entered by the user is parsed, and the corresponding fault phenomenon node is located in the knowledge graph;

[0037] Starting from the fault phenomenon node, a reverse breadth-first search is performed along the causal relationship to obtain a candidate fault set;

[0038] The comprehensive recommendation score for each candidate fault is calculated using a pre-set dual weighted algorithm of "theoretical risk priority number and actual occurrence frequency".

[0039] Sort the candidate fault set from high to low according to the comprehensive recommendation score, and output the fault cause and its corresponding maintenance measures and detection methods.

[0040] Furthermore, if a work order contains new components or new fault relationships that do not exist in the knowledge graph and the extraction confidence level is higher than a preset new knowledge threshold, it is then incorporated into the main knowledge graph to achieve dynamic updates of the knowledge graph. Specifically, this includes:

[0041] Monitor the work order status of the maintenance management system in real time, and trigger the update process when the work order is completed and the cause is diagnosed;

[0042] If the cause of diagnosis in the work order exists in the existing knowledge graph, then the corresponding fault mode node is retrieved in the knowledge graph, and the actual occurrence frequency attribute of the fault mode node is cumulatively corrected.

[0043] If the entity or relation extracted from the new work order does not exist in the existing knowledge graph and the overall confidence level is higher than the preset new knowledge threshold, the corresponding new node and new edge will be automatically created in the candidate area, and the expert review process will be triggered. After the review is approved, it will be incorporated into the main knowledge graph to realize the dynamic update of the knowledge graph.

[0044] This invention also discloses a fault mode knowledge graph construction and updating system for motors, including:

[0045] The module for building the ontology model in the field of motor faults is configured to define the core concept class hierarchy used to represent knowledge of motor faults, define object attributes and semantic relationships, acquire structured and unstructured data and define data attributes, and then set logical constraints and axioms.

[0046] The dual-channel knowledge extraction module is configured to extract knowledge from unstructured data using a deep learning model and from structured data using a mapping method based on the ontology model of the motor fault domain, generating original triplet knowledge.

[0047] The knowledge fusion module is configured to perform entity alignment, attribute fusion, and conflict resolution on the original triple knowledge, and to integrate knowledge from different data sources to construct a high-quality, unambiguous knowledge graph and store it in a graph database.

[0048] The intelligent fault reasoning and dynamic update module is configured to utilize the constructed knowledge graph to respond to user diagnostic requests and achieve intelligent reasoning and ranking from fault phenomena to root causes based on a preset dual-weighted algorithm. At the same time, if the work order contains new components or new fault relationships that do not exist in the knowledge graph and the extraction confidence is higher than the preset new knowledge threshold, it is incorporated into the main knowledge graph to achieve dynamic updates of the knowledge graph.

[0049] The beneficial effects of this invention are:

[0050] 1. Multi-dimensional knowledge fusion: This invention innovatively proposes a fusion method that combines structured data from the design and manufacturing stage with unstructured data from the operation and maintenance stage. It adopts multi-source heterogeneous data, which not only retains the theoretical rigor of the expert design stage (through FMEA mapping) but also incorporates the field practical experience of the operation and maintenance stage (through text mining), thus solving the problem of one-sided information from a single data source.

[0051] 2. High-precision knowledge extraction: To address the pain points of complex technical terminology and large differences in text descriptions in the field of motors, this invention adopts the BERT-BiLSTM-CRF model. Compared with traditional dictionary matching or shallow machine learning methods, it can more accurately identify long and difficult entities and complex semantic relationships, significantly improving the utilization rate of unstructured data.

[0052] 3. Intelligent weighted reasoning: This invention abandons simple Boolean query and introduces a dual weighted sorting algorithm based on "theoretical risk priority number" and "actual occurrence frequency". This makes the diagnostic results not only indicate "what kind of fault it is", but also tell "which fault to investigate first" according to risk and probability, which greatly improves the accuracy and efficiency of on-site fault troubleshooting.

[0053] 4. Dynamic Evolution Mechanism: By introducing incremental update logic, this invention makes the knowledge graph no longer a static database, but a system that can automatically correct weights and expand new knowledge with each maintenance record, effectively solving the problems of knowledge base aging and lag. Attached Figure Description

[0054] Figure 1 This is a schematic diagram illustrating the principle of the method for constructing and updating a fault mode knowledge graph for motors as described in an embodiment of the present invention.

[0055] Figure 2 This is a schematic diagram of the principle structure of the BERT-BiLSTM-CRF model described in an embodiment of the present invention.

[0056] Figure 3 This is a performance comparison chart of different knowledge extraction models described in the embodiments of the present invention.

[0057] Figure 4 This is a comparison chart of the diagnostic accuracy of different recommendation algorithms described in the embodiments of the present invention. Detailed Implementation

[0058] To facilitate a better understanding of the present invention by those skilled in the art, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. The following are merely exemplary and do not limit the scope of protection of the present invention.

[0059] Example 1:

[0060] This invention discloses a method for constructing and updating a fault mode knowledge graph for motors, comprising the following steps:

[0061] Constructing an ontology model for the motor fault domain: Define a hierarchy of core concept classes to represent motor fault knowledge, define object attributes and semantic relationships, acquire structured and unstructured data and define data attributes, and then set logical constraints and axioms;

[0062] Dual-channel knowledge extraction: Based on the aforementioned motor fault domain ontology model, a deep learning model is used to extract knowledge from unstructured data, and a mapping method is used to extract knowledge from structured data to generate original triplet knowledge.

[0063] Knowledge fusion: The original triple knowledge is subjected to entity alignment, attribute fusion and conflict resolution. Knowledge from different data sources is integrated to construct a high-quality, unambiguous knowledge graph and stored in a graph database.

[0064] Intelligent fault reasoning and dynamic updating based on knowledge graph: Utilizing the constructed knowledge graph, in response to user diagnostic requests, intelligent reasoning and ranking from fault phenomena to root causes are achieved based on a preset dual-weighted algorithm; at the same time, if the work order contains new components or new fault relationships that do not exist in the knowledge graph and the extraction confidence is higher than the preset new knowledge threshold, it is incorporated into the main knowledge graph to achieve dynamic updating of the knowledge graph.

[0065] To better understand the above technical solution, the following will, in conjunction with the accompanying drawings and an example of a permanent magnet synchronous motor (PMSM) full lifecycle operation and maintenance management system from a motor manufacturing company, elaborate on an exemplary embodiment of the fault mode knowledge graph construction and updating method for motors described in this invention. Figure 1 As shown, this embodiment is merely an exemplary embodiment of the present invention. However, it should be understood that the present invention can be implemented in various forms and is not limited to the embodiments described herein. These embodiments are provided to enable those skilled in the art to understand the present invention more clearly and thoroughly. This method aims to address the problems in existing systems such as the fragmentation of structured data (e.g., FMEA data) generated during the design and manufacturing phase with unstructured data (e.g., maintenance logs) generated during the operation and maintenance phase, the lack of semantic reasoning capabilities in fault diagnosis, and the lag in knowledge base updates.

[0066] S1. Construct an ontology model for the motor fault domain.

[0067] In this step, the Protégé 5.5 knowledge modeling tool and the Web Ontology Language (OWL) are used, following the "seven-step method" of ontology construction (identify the domain, examine reusability, list terms, define classes and levels, define attributes, define constraints, and create instances).

[0068] The specific implementation process includes the following sub-steps:

[0069] S11. Define the core concept of class hierarchy (classes and subclasses).

[0070] To comprehensively cover knowledge about motor faults, this embodiment constructs a top-down five-dimensional concept tree structure, where all concepts are mapped to "classes" in the ontology.

[0071] (1) Equipment and Components:

[0072] Define the root node as the motor.

[0073] The first-level sub-node is defined as a permanent magnet synchronous motor, and further subdivided into second-level sub-nodes based on the physical structure:

[0074] 1. Stator system: including stator core, stator winding, and insulation layer.

[0075] 2. Rotor system: consisting of rotor core, permanent magnet, and shaft.

[0076] 3. Support system: The system includes bearings (further subdivided into inner ring, outer ring, rolling elements, and cage) and end caps.

[0077] 4. Cooling system and sensing system (including encoder, temperature sensor, etc.).

[0078] (2) Fault Phenomena:

[0079] Define observable indicators characterizing abnormal motor conditions, including:

[0080] 1. Abnormal vibration: further subdivided into high-frequency vibration and axial movement.

[0081] 2. Thermal anomalies: These are further divided into localized overheating and excessively rapid temperature rise.

[0082] 3. Electrical abnormalities: These are further divided into three-phase current imbalance and back electromotive force abnormalities.

[0083] 4. Acoustic anomalies: These are further subdivided into electromagnetic whistling and mechanical friction noise.

[0084] (3) Fault Mode Class:

[0085] Define the root cause of the phenomenon, including:

[0086] 1. Electromagnetic faults: such as inter-turn short circuits, demagnetization of magnets, and insulation breakdown.

[0087] 2. Mechanical failures: such as bearing pitting, air gap eccentricity, dynamic imbalance, and loose mounting feet.

[0088] (4) Diagnostic and testing categories:

[0089] Define the technical means of fault detection, including current characteristic analysis, vibration spectrum analysis, and infrared thermal imaging detection.

[0090] (5) Maintenance decision-making:

[0091] Define specific measures to resolve the fault, including component replacement, insulation repair (including rewinding and impregnation), and precision adjustment (including dynamic balancing and centering).

[0092] S12. Define object attributes and semantic relationships.

[0093] To endow the knowledge graph with reasoning capabilities, this sub-step defines multiple logical relationships between entities and sets the domain and range of these relationships:

[0094] 1. Composition Relationship: Define "Containing Components" to connect devices and components, such as connecting a permanent magnet synchronous motor and a stator system. This relationship is set to be transitive, that is, if a motor contains a stator system, and the stator system contains stator windings, then it can be automatically inferred that the motor contains stator windings.

[0095] 2. Causal Relationship: The domain of "causes" is the fault mode, and the value domain is the fault phenomenon. For example: (inter-turn short circuit, causes, local overheating). "Causes" is defined as the inverse relationship of "causes".

[0096] 3. Diagnostic Relationship: Define "Detected by..." to connect the fault mode with the detection method.

[0097] 4. Positional Relationship: Defines "located in", connecting the failure mode to the equipment component. For example: (pitting, located in, bearing inner ring).

[0098] 5. Governance Relationship: Defines "resolved by...", connecting failure modes with maintenance measures.

[0099] S13. Obtain structured and unstructured data and define data attributes.

[0100] This sub-step acquires structured data from the design and manufacturing phases and unstructured data generated during the operation and maintenance phases. To integrate the structured and unstructured data, this sub-step defines detailed numerical attributes for specific classes:

[0101] 1. For the fault mode class, define the following attributes:

[0102] Risk Priority Number (RPN): The data type is integer, and the data source is mapped from the FMEA table. It is used to represent the theoretical risk level.

[0103] Severity (S): Data type is integer (1-10), data source is mapped from FMEA table.

[0104] Actual frequency of occurrence: The data type is integer, the data source is statistical aggregation of operation and maintenance logs, the initial value is 0, and it is dynamically accumulated as the system runs.

[0105] Confidence level: The data type is floating point (0.0-1.0), used to characterize the reliability of the fault knowledge.

[0106] 2. For maintenance decision-making classes, define the following attributes:

[0107] Average cost: Data type is floating point.

[0108] Average time elapsed: Data type is floating point, unit is hours.

[0109] S14. Set logical constraints and axioms.

[0110] To ensure the consistency of the knowledge graph and prevent logical conflicts, this embodiment pre-defines mutual exclusion constraints in the ontology: setting that equipment components, fault phenomena, and fault modes are mutually exclusive. That is, the system restricts an entity from simultaneously belonging to the "component" class and the "fault phenomenon" class, thereby avoiding logical confusion during knowledge extraction.

[0111] S2. Dual-channel knowledge extraction.

[0112] Step S2 aims to transform the multi-source heterogeneous data (structured and unstructured data) obtained in Step S1 into computer-understandable triple knowledge that conforms to ontology specifications. To address the heterogeneity of the data sources, Step S2 employs a "dual-channel" extraction mechanism, the specific implementation process of which is as follows:

[0113] S21. Knowledge extraction from unstructured text.

[0114] For unstructured text data such as maintenance work orders, fault logs, and expert diagnostic reports, this embodiment constructs a BERT-BiLSTM-CRF model and performs entity recognition using the BERT-BiLSTM-CRF model, and combines an attention mechanism for relation extraction, such as... Figure 2 As shown. Specifically:

[0115] S211. Data Vectorization and Encoding (BERT Layer): The cleaned operation and maintenance text sequence is input into the pre-trained BERT model (the BERT model is one layer of the BERT-BiLSTM-CRF model, and can also be called the BERT layer). The BERT layer first converts the input text into a superposition of character vectors, position vectors, and paragraph vectors. Then, through the self-attention mechanism of the multi-layer Transformer encoder, the text is scanned bidirectionally. The BERT layer can dynamically adjust the semantic representation of character vectors according to the context. For example, in the texts "rotor dynamic imbalance causes vibration" and "rotor core loosening", the BERT layer can capture the subtle semantic differences of "rotor" in different contexts, solving the problem of polysemy in traditional models.

[0116] S212. Contextual Feature Extraction (BiLSTM Layer): The word vector sequence containing deep semantic information output from the BERT layer is input into a bidirectional long short-term memory (BiLSTM) network. The BiLSTM layer includes a forward LSTM and a backward LSTM. The forward LSTM processes the sequence from left to right, capturing contextual information (e.g., after "inter-turn short circuit" occurs, "overheating" is highly likely to follow). The backward LSTM processes from right to left, capturing contextual information. Finally, the hidden layer state vectors from both directions are concatenated to form a feature vector containing complete contextual dependencies.

[0117] S213. Entity Label Decoding (CRF Layer): The feature vector output from the BiLSTM layer is input into the Conditional Random Field (CRF) layer for sequence label decoding. The CRF layer applies logical constraints to the prediction results by learning the label transition probability matrix. For example, in the BIO labeling system, the label I-fault mode (mechanism middle word) must immediately follow B-fault mode (mechanism first word) or I-fault mode, and cannot appear after O (non-entity).

[0118] For example, the input text is: "The motor stator winding temperature rises abnormally due to blockage of the cooling water channel". The output sequence is: CRF decodes the optimal tag sequence and identifies the entities: [Equipment component: stator winding], [Fault mode: blockage of cooling water channel], [Fault phenomenon: abnormal temperature rise].

[0119] S214. Semantic Relation Extraction Layer: After identifying entities, an attention-based classification model is used to determine the relationship type between entities. Specifically, sentences containing entity pairs are input into the model, and the attention mechanism automatically assigns higher weights to trigger words (such as "cause", "lead to", "cause"). The attention-based classification model calculates that the entity pair (cooling water channel blockage, abnormal temperature rise) has a 98% probability of belonging to the relationship type "lead to", thus generating a triplet knowledge: (cooling water channel blockage, leads to, abnormal temperature rise).

[0120] S22. Knowledge extraction from structured data.

[0121] For structured data, which has a standardized column structure, this step uses an Extract-Transform-Load (ETL) mapper to transform table rows into knowledge graph subgraphs according to preset rules.

[0122] S221. Field-Entity Mapping Rules: Map the contents of the failure mode column (such as "bearing inner ring spalling") in the structured data to the failure mode entity in the ontology; map the contents of the failure effect column (such as "high frequency noise") to the failure phenomenon entity in the ontology; and map the contents of the component name column (such as "drive end bearing") to the equipment component entity in the ontology.

[0123] S222. Attribute extraction rules: Extract quantitative evaluation parameters from structured data and store them as numerical attributes of nodes. Extract three columns of values: severity (S), occurrence (O), and detectability (D). Calculate the risk priority number: RPN = S × O × D. Encapsulate the above values ​​into an attribute dictionary and attach it to the corresponding failure mode node.

[0124] S223. Relationship Construction Rules: Based on the logical structure of FMEA, automatically instantiate the following relations: construct (failure mode, cause, failure phenomenon) triples; construct (failure mode, location, equipment component) triples.

[0125] The following example illustrates the input data behavior: {Component: Rotor, Failure Mode: Magnet Demagnetization, Failure Effect: Reduction in Back EMF, S: 9, O: 3, D: 5}.

[0126] The output triplets are: 1. (The magnet demagnetizes, resulting in a decrease in back electromotive force) 2. (The magnet demagnetizes and is located in the rotor).

[0127] Output attribute: Demagnetize the magnet, RPN=135.

[0128] S3. Knowledge Integration.

[0129] This step aims to address the issues of entity name diversity, attribute description redundancy, and data source conflicts in the "original triples of the instance layer" generated in step S2. Through entity alignment, attribute fusion, and conflict resolution mechanisms, fragmented knowledge from different data sources is integrated into a unified, high-quality, and unambiguous knowledge graph.

[0130] The specific steps are as follows:

[0131] S31. Semantic Similarity-Based Entity Alignment: Since unstructured text data contains a large number of non-standard terms (such as describing "stator winding" as "coil" or "wound wire"), while structured data uses standard terms, entity alignment is necessary to eliminate semantic ambiguity. Specific operations include:

[0132] 1. Vectorization: Using the BERT model trained in step S21, extract the semantic vector representations of all entity names in the original triples. Let the two entity names to be aligned be... (From maintenance log) and (From structured data), and their corresponding vectors are respectively and .

[0133] 2. Similarity Calculation: The cosine similarity formula is used to calculate the semantic similarity between the two.

[0134] (1)

[0135] 3. Threshold Determination and Merging: Set the alignment threshold to... (In this embodiment, it is preferably set to 0.85).

[0136] like Then determine and Referring to the same object. The system will perform a merge operation: preserving the standard name of the symbol ontology specification (usually the one in FMEA). ),Will Stored as an "alias" attribute of the entity, and pointing to the original... All relationships redirected to .

[0137] For example, if the similarity between "stator coil" (an entity from the maintenance log) and "stator winding" (an entity from the structured data) is calculated to be 0.92, the system will automatically replace all "stator coil" nodes in the knowledge graph with "stator winding", thus eliminating redundant nodes in the knowledge graph.

[0138] S32. Attribute fusion based on "master-slave strategy".

[0139] For the same failure mode entity, structured data provides theoretical design attributes, while maintenance logs provide on-site statistical attributes. This embodiment employs a "master-slave fusion strategy" to construct a comprehensive failure profile. Specific operations include:

[0140] 1. Main data source anchoring: The nodes generated from structured data (FMEA data) are the main data source, and their prior theoretical attributes such as risk priority number (RPN), severity (S), occurrence (O) and detectability (D) are retained and assigned a high confidence label.

[0141] 2. Mounting from data source: Using nodes generated from unstructured data (maintenance logs) as slave data, count the number of times the fault mode occurs within the historical time window to generate the actual occurrence frequency attribute; count the average time of maintenance work orders corresponding to the fault to generate the average repair time (MTTR) attribute.

[0142] 3. Attribute merging: Attach the statistical attributes from the "data source" to the entity node determined by the "main data source".

[0143] For example, FMEA data: {Entity: Bearing Pitting, RPN: 160}, log data: {Entity: Bearing Peeling, Frequency: 56 times, MTTR: ​​4.5 hours}, merged node: {Entity: Bearing Pitting, Alias: Bearing Peeling, RPN: 160, Frequency: 56, MTTR: ​​4.5}.

[0144] S33. Multi-source data conflict resolution.

[0145] When different data sources contain logically conflicting descriptions of the same attribute, the conflict is handled according to a pre-defined confidence priority rule. The specific rules are as follows:

[0146] 1. Conflict of theoretical parameters: If the "estimated risk level" extracted from unstructured data (maintenance log text) is inconsistent with the "severity S" in structured data (FMEA), the structured data (FMEA) shall prevail (because FMEA represents design specifications).

[0147] 2. Measured parameter conflict: If the difference between the "estimated maintenance time" estimated in the structured data (FMEA) and the "average maintenance time" statistically calculated in the unstructured data (maintenance log text) exceeds 20%, the statistical data of the unstructured data (maintenance log text) shall prevail (because the log represents the actual situation on site), and the attribute shall be marked as "needs correction" for subsequent feedback to the design department.

[0148] 3. Manual intervention interface: For automatic fusion results with a confidence level below the threshold (such as 0.6 in this embodiment), the system marks them as "pending review" and pushes them to the expert review interface. They are then manually reviewed and confirmed before being written into the formal knowledge graph.

[0149] S34. Construct a fault knowledge graph.

[0150] The triples processed through steps S31-S33 are then imported in batches into a graph database (such as Neo4j). Specifically:

[0151] 1. Node creation: Create nodes for fault modes, fault phenomena, equipment components, etc., and write the merged attributes.

[0152] 2. Relationship creation: Create edges that cause, are located at, and contain, and add weight attributes to the edges (initial weight is set to 1.0).

[0153] 3. Index Building: Create a full-text index for the name attributes of the "Fault Phenomenon" and "Fault Mode" nodes to accelerate subsequent retrieval and inference.

[0154] S4. Intelligent fault reasoning and dynamic updating based on knowledge graphs.

[0155] Step S4 aims to leverage the constructed unified knowledge graph to respond to user diagnostic requests and provide accurate fault location; simultaneously, it uses newly added maintenance records to reverse-correct the graph weights, achieving dynamic updates to the knowledge base. The specific steps are as follows:

[0156] S41. Intelligent fault reasoning based on a dual-weighted algorithm of "theory (i.e., theoretical risk priority number, RPN value) and practice (i.e., actual occurrence frequency)": When maintenance personnel input a fault phenomenon described in natural language, the system no longer performs simple keyword matching, but rather performs probabilistic reasoning based on knowledge graph paths. The specific reasoning logic is as follows:

[0157] 1. Semantic Parsing and Anchor Point Localization: Using the BERT-BiLSTM-CRF model trained on S21, the system parses the user-input query (e.g., "There is a noticeable whistling sound at the front cover of the motor during operation, and the temperature rises rapidly"). The system identifies the fault phenomena entities [whistling] and [temperature rise], and locates the corresponding fault phenomenon nodes in the graph as the starting point for inference.

[0158] 2. Multi-hop path tracing: Starting from the located fault phenomenon node, perform a reverse breadth-first search along the edges of the "caused by..." relationship to find all reachable fault mode nodes, forming a candidate fault set. .

[0159] 3. Comprehensive Recommendation Score Calculation: To resolve the ambiguity of "one phenomenon may correspond to multiple causes," this embodiment uses the following formula to calculate the cause of each candidate fault. Recommended score :

[0160] (2)

[0161] In formula (2), This is the risk priority number for the fault in the FMEA, representing the theoretically most dangerous probability; This represents the maximum value among the risk priority numbers; The actual frequency of the fault in the knowledge graph represents the most likely probability. This represents the combined frequency of all failures. This is the adjustment coefficient, with a value range of 0-1.

[0162] 4. Sorting and Output: Sort the candidate fault set from high to low according to the comprehensive recommendation score, and output the most likely fault causes and their corresponding maintenance measures and detection methods.

[0163] For example, regarding the "whistling" phenomenon, the knowledge graph search identified two causes: "bearing cage breakage" (high RPN, low frequency) and "uneven air gap" (medium RPN, high frequency). After calculating the score, the system prioritized recommending "uneven air gap" and suggested checking the rotor eccentricity, which aligns with the actual operating conditions of an aging motor.

[0164] S42. Dynamic map update based on closed-loop feedback.

[0165] Traditional knowledge bases are fixed once built, but this embodiment constructs a closed-loop update mechanism that interfaces with the maintenance management system (CMMS), enabling the knowledge graph to have a dynamic update mechanism.

[0166] The specific update process is as follows:

[0167] 1. Event Listening: The system background monitors the work order status flow in the Maintenance Management System (CMMS) in real time. When a work order status change to "Completed" and includes the "Confirmed Reason" field, an update process is triggered.

[0168] 2. Real-time weight adjustment: This includes analyzing the diagnosed cause in the work order (e.g., "dried grease"). Retrieve the corresponding fault mode node in the knowledge graph. Perform the atomic operation: Node.Frequency = Node.Frequency + 1. Effect: As the frequency of this fault accumulates, its weight will increase in the inference formula of S41. The weighting of the components gradually increases. This means that if a certain type of fault occurs in a concentrated manner recently, the system will prioritize recommending that fault during diagnosis, thus achieving adaptation to changes in operating conditions.

[0169] 3. New Knowledge Discovery and Storage: If an entity (e.g., a new model component "ceramic bearing") or relation extracted from a new work order does not exist in the existing knowledge graph, and the overall confidence score of the model extraction is higher than the preset new knowledge threshold, the system automatically creates new nodes and edges in the candidate region of the knowledge graph. The formula for calculating the overall confidence score is as follows:

[0170] (3)

[0171] In formula (3), This represents the average probability value of the entity label sequence output by the BERT-BiLSTM-CRF model (i.e., the normalized value of the decoding score of the CRF layer). This represents the normalized exponential function probability value of the relation category output by the attention-based classification model. This is the weighting coefficient (preferably 0.5 in this embodiment).

[0172] In this embodiment, the new knowledge threshold is set to 0.95, that is, when When this happens, the system automatically creates new nodes and edges in the candidate regions of the knowledge graph, and applies the comprehensive confidence score calculated above. The "confidence level" data attribute defined in step S13 is assigned to the new node as the initial confidence level record of the knowledge point. Then, the expert review process is triggered, and domain experts are notified to review the knowledge. After the review is passed, the candidate knowledge is officially incorporated into the main knowledge graph, completing the expansion of the knowledge base and thus realizing the dynamic update of the knowledge graph.

[0173] To further verify the effectiveness, scientific validity, and advancements compared to existing technologies of the proposed method for constructing and updating a fault mode knowledge graph for motors, this embodiment uses full-lifecycle operation and maintenance data of a permanent magnet synchronous motor (PMSM) from a motor manufacturing company for full-process verification. This data includes structured data from the design and manufacturing phases (such as FMEA document data) and unstructured data generated during the operation and maintenance phases (such as historical maintenance work order records).

[0174] To visually demonstrate the complete processing flow of this invention from multi-source heterogeneous data to intelligent diagnostic results, representative sample data is selected below to focus on detailing the specific implementation process of the following steps in the method of this invention: "dual-channel knowledge extraction (step S2)," "knowledge fusion (step S3)," and "dual-weighted intelligent reasoning (step S4)."

[0175] The partial sample of structured design data (FMEA) selected in this embodiment is shown in Table 1, and the partial sample of unstructured operation and maintenance data (maintenance work orders) is shown in Table 2.

[0176] Table 1. Sample of Partial Structured Design Data

[0177]

[0178] Table 2 Sample of some unstructured operation and maintenance data

[0179]

[0180] Based on the sample data in Tables 1 and 2 above, the specific execution logic of the method in this embodiment is as follows:

[0181] 1. Dual-channel knowledge extraction:

[0182] For the stator winding inter-turn short circuit records in Table 1, the system uses the ETL mapper to automatically instantiate the fault mode node "stator inter-turn short circuit" and extracts the RPN value (135) as its prior attribute.

[0183] For the maintenance work orders in Table 2, the system uses a pre-trained BERT-BiLSTM-CRF model to perform sequence labeling on the text. For example, it identifies the entities "large current fluctuation" (fault phenomenon) and "coil turn short circuit" (fault mode) and extracts the relationship to construct a triplet (coil turn short circuit, resulting in large current fluctuation).

[0184] 2. Knowledge Integration:

[0185] Entity alignment: The system calculates semantic similarity and identifies that "stator winding" in FMEA and "coil" in work order refer to the same component, and merges "coil inter-turn short circuit" under the standard node "stator inter-turn short circuit".

[0186] Attribute fusion: The system traversed all work orders from the past three years and found that there were a total of 142 instances of "stator inter-turn short circuit", 85 instances of "bearing pitting", 56 instances of "cooling water channel blockage", and 8 instances of "permanent magnet demagnetization". The system then attached this frequency data to the corresponding nodes in the knowledge graph.

[0187] 3. Dual-weighted intelligent reasoning:

[0188] When a user inputs the query "abnormal motor current", the system calculates the comprehensive recommendation score for each candidate fault based on formula (2). In the calculation process of this embodiment, in order to ensure the comparability of values ​​in different dimensions, the parameters are normalized as follows:

[0189] (1) Normalization theory risk denominator (i.e., the maximum value in the risk priority number, : Set to the theoretical maximum value of FMEA 1000 (i.e., Severity (S) 10 × Occurrence (O) 10 × Detectability (D) 10).

[0190] (2) Normalize the denominator of the actual occurrence frequency (i.e., the sum of all fault frequencies, According to statistics, under the current query conditions, the total actual occurrence frequency of all candidate faults (including the first 4 items listed in Table 3 and other long-tail low-frequency faults) is 532 times.

[0191] (3) Adjustment coefficient ( ):set up The overall score consists of 40% theoretical risk and 60% actual frequency. This indicates a greater emphasis on actual on-site data.

[0192] Based on the above settings, the comprehensive recommendation scores for each fault mode are calculated as shown in Table 3.

[0193] Table 3. Calculation results of the comprehensive recommendation score for each failure mode

[0194]

[0195] Calculation process description: Taking "stator inter-turn short circuit as an example", the detailed calculation steps are as follows:

[0196] (1) Theoretical calculation: Divide its RPN value (135) by the maximum value (1000) to obtain the normalized value 0.135; multiply by the weight coefficient 0.4 to obtain the theoretical score component 0.054.

[0197] (2) Calculation of actual part: Divide its actual frequency (142) by the total frequency of candidate faults (532) to get the frequency ratio of about 0.267; multiply by the weight coefficient 0.6 to get the actual score component of 0.160.

[0198] (3) Summation: Add the two parts together to get 0.054 + 0.160, and finally get a comprehensive recommendation score of 0.214.

[0199] The other calculation processes are the same as above and will not be repeated here. The calculation results are shown in Table 3. Although the theoretical RPN (160) of "bearing pitting" is higher than that of "stator inter-turn short circuit" (135), the latter occurs more frequently in the field. After double weighting, the system correctly ranks "stator inter-turn short circuit" as the first recommended fault. This is in line with the field experience of "prioritizing the investigation of high-frequency faults" and also takes into account the attention to high-risk faults.

[0200] To verify the technical advantages of the BERT-BiLSTM-CRF model used in this invention when processing complex professional text in the field of motor engineering, this embodiment compares it with the traditional BiLSTM-CRF model (without pre-training layers) and the IDCNN-CRF model. Precision, recall, and F1 score were used as evaluation metrics. The experimental results are shown in Table 4. Figure 3 As shown.

[0201] Table 4. Comparison of entity recognition performance of different knowledge extraction models

[0202]

[0203] As can be seen from the data in Table 4, the BERT-BiLSTM-CRF model used in this invention improves the F1 score by about 12-15 percentage points compared with the traditional baseline model. This proves that the introduction of the BERT pre-training layer can effectively capture the deep semantic features of professional terms such as "stator inter-turn short circuit" and "cogging torque", and significantly improve the extraction quality of unstructured data.

[0204] To verify the diagnostic accuracy of the proposed dual-weighted algorithm based on "theoretical risk priority number and actual occurrence frequency," this embodiment selected 200 historical real-world fault cases for backtracking testing. The TOP-3 diagnostic accuracy (i.e., the proportion of the actual fault cause appearing in the top 3 recommendation results) of three different recommendation strategies was compared. The results are shown in Table 5. Figure 4 As shown, Strategy A is a strategy based solely on the Theoretical Risk Priority Number (RPN), Strategy B is a strategy based solely on the actual occurrence frequency, and Strategy C is the strategy of this invention that uses a dual weighted algorithm of "Theoretical Risk Priority Number and Actual Occurrence Frequency".

[0205] Table 5. Comparison of TOP-3 diagnostic accuracy for different recommendation strategies

[0206]

[0207] Experimental data show that the dual-weighted algorithm of this invention improves the TOP-3 diagnostic accuracy to 89.8%, proving that the deep integration of theoretical specifications at the design end and practical experience at the operation and maintenance end can effectively overcome the limitations of a single data source and significantly improve the intelligence level and practical value of motor fault diagnosis.

[0208] This embodiment's method successfully breaks down the barrier between structured data generated during the design and manufacturing phase and unstructured data generated during the operation and maintenance phase by constructing a standardized motor fault domain ontology model and utilizing a dual-channel mechanism of BERT-BiLSTM-CRF deep learning model and FMEA rule mapping. Employing a reasoning algorithm based on a dual weighting of "theoretical risk priority number and actual occurrence frequency," it can not only accurately locate complex faults such as "stator inter-turn short circuit" and "bearing pitting," but also dynamically adjust diagnostic strategies according to on-site operating conditions. Through closed-loop linkage with the maintenance management system, this invention achieves incremental adaptive updates of the knowledge graph, solving the problems of easy aging and high maintenance costs of traditional expert system knowledge bases, and has significant application value in the full lifecycle management of industrial motors.

[0209] II. Example 2:

[0210] This embodiment discloses a fault mode knowledge graph construction and updating system for motors, including:

[0211] The module for constructing the ontology model in the field of motor faults is configured to define the core concept class hierarchy used to represent knowledge of motor faults, define object attributes and semantic relationships, acquire structured and unstructured data and define data attributes, and then set logical constraints and axioms.

[0212] The dual-channel knowledge extraction module is configured to extract knowledge from unstructured data using a deep learning model and from structured data using a mapping method, based on the ontology model of the motor fault domain, to generate original triplet knowledge.

[0213] The knowledge fusion module is configured to perform entity alignment, attribute fusion, and conflict resolution on the original triple knowledge, and to integrate knowledge from different data sources to construct a high-quality, unambiguous knowledge graph and store it in a graph database.

[0214] The intelligent fault reasoning and dynamic update module is configured to utilize the constructed knowledge graph to respond to user diagnostic requests and achieve intelligent reasoning and ranking from fault phenomena to root causes based on a preset dual-weighted algorithm. At the same time, if the work order contains new components or new fault relationships that do not exist in the knowledge graph and the extraction confidence is higher than the preset new knowledge threshold, it is incorporated into the main knowledge graph to achieve dynamic updates of the knowledge graph.

[0215] For the system implementation, since it basically corresponds to the method implementation, the specific implementation process of the functions and roles of each module in the system can be found in the implementation process of the corresponding steps in the method described in Embodiment 1, and will not be repeated here.

[0216] The above description only outlines the basic principles and preferred embodiments of the present invention. Those skilled in the art can make many changes and modifications based on the above description, and these changes and modifications should fall within the protection scope of the present invention.

Claims

1. A motor-oriented fault mode knowledge graph construction and updating method, characterized in that, Including the following steps: Constructing an ontology model for the motor fault domain: Define a hierarchy of core concept classes to represent motor fault knowledge, define object attributes and semantic relationships, acquire structured and unstructured data and define data attributes, and then set logical constraints and axioms. The core concept class hierarchy is a top-down five-dimensional concept tree structure, with the five dimensions being equipment and component classes, fault phenomenon classes, fault mode classes, diagnosis and detection classes, and maintenance decision classes. The object attributes and semantic relationships include at least the compositional relationships connecting equipment and components, the causal relationships connecting fault modes and fault phenomena, the diagnostic relationships connecting fault modes and detection methods, the positional relationships connecting fault modes and equipment / components, and the governance relationships connecting fault modes and maintenance measures. The data attributes include at least attributes for the fault mode class and attributes for the maintenance decision class. Dual-channel knowledge extraction: Based on the aforementioned motor fault domain ontology model, a deep learning model is used to extract knowledge from unstructured data, and a mapping method is used to extract knowledge from structured data to generate original triplet knowledge. Knowledge fusion: The original triplet knowledge is subjected to entity alignment, attribute fusion, and conflict resolution. Knowledge from different data sources is integrated to construct a high-quality, unambiguous knowledge graph and stored in a graph database. The attribute fusion adopts a master-slave strategy, with nodes generated from structured data as the master data source and nodes generated from unstructured data as the slave data source. Statistical attributes from the slave data source are attached to the entity nodes determined by the master data source to achieve attribute fusion. Intelligent fault reasoning and dynamic updating based on knowledge graph: Utilizing the constructed knowledge graph, in response to user diagnostic requests, a comprehensive recommendation score is calculated for each candidate fault based on a preset dual-weighted algorithm of "theoretical risk priority number and actual occurrence frequency," enabling intelligent reasoning and ranking from fault phenomena to root causes. Simultaneously, if the work order contains new components or new fault relationships not present in the knowledge graph and the extraction confidence level is higher than the preset new knowledge threshold, it is incorporated into the main knowledge graph, achieving dynamic updating of the knowledge graph.

2. The method according to claim 1, characterized in that, The method of extracting knowledge from unstructured data using a deep learning model specifically includes: A BERT-BiLSTM-CRF model is constructed to perform entity recognition on unstructured data. The BERT-BiLSTM-CRF model includes a BERT layer for converting unstructured data into context-related vectors, a BiLSTM layer for bidirectional extraction of sequence context features, and a CRF layer for annotating and decoding entity label sequences. After identifying entities, an attention-based classification model is used to extract semantic relationships between entities, thereby generating triplet knowledge.

3. The method according to claim 1, characterized in that, The knowledge extraction from structured data using a mapping method specifically includes: Based on the preset field-entity mapping rules, the field content in the structured data is mapped to the corresponding entity in the motor fault domain ontology model; Based on attribute extraction rules, the Risk Priority Number (RPN) in the structured data is extracted and calculated as an attribute of the failure mode node. Based on relational construction rules, (fault mode, cause, fault phenomenon) triples and (fault mode, location, device component) triples are automatically instantiated according to the logical structure of structured data.

4. The method according to claim 2, characterized in that, The entity alignment specifically includes: The semantic vector representation of all entity names in the original triples is extracted using the BERT-BiLSTM-CRF model; Calculate and determine the semantic similarity of entities. If the similarity exceeds the preset alignment threshold, merge them, retain the standard name of the entity, and record the alias.

5. The method according to claim 1, characterized in that, The attribute fusion adopts a master-slave strategy, specifically including: The nodes generated from structured data serve as the primary data source, preserving prior theoretical attributes, which include at least risk priority, severity, occurrence, and detectability. The nodes generated from unstructured data are used as the data source to statistically analyze the actual frequency of failures and the average repair time. The statistical attributes from the data source will be attached to the entity nodes determined by the main data source to achieve attribute fusion.

6. The method according to claim 1, characterized in that, The conflict resolution specifically includes: In case of conflicting theoretical parameters, structured data shall prevail. In case of conflicting measured parameters, unstructured data shall prevail. For fused data with a confidence level below the preset threshold, the data confirmed by manual review shall prevail.

7. The method according to claim 1, characterized in that, The process of utilizing the constructed knowledge graph to respond to user diagnostic requests and achieving intelligent reasoning and ranking from fault symptoms to root causes based on a preset dual-weighted algorithm specifically includes: Using the trained deep learning model, the query statement entered by the user is parsed, and the corresponding fault phenomenon node is located in the knowledge graph; Starting from the fault phenomenon node, a reverse breadth-first search is performed along the causal relationship to obtain a candidate fault set; The comprehensive recommendation score for each candidate fault is calculated using a pre-set dual weighted algorithm of "theoretical risk priority number and actual occurrence frequency". Sort the candidate fault set from high to low according to the comprehensive recommendation score, and output the fault cause and its corresponding maintenance measures and detection methods.

8. The method according to claim 1, characterized in that, If a work order contains new components or new fault relationships that do not exist in the knowledge graph and the extraction confidence level is higher than a preset new knowledge threshold, it is then incorporated into the main knowledge graph to achieve dynamic updates of the knowledge graph. Specifically, this includes: Monitor the work order status of the maintenance management system in real time, and trigger the update process when the work order is completed and the cause is diagnosed; If the cause of diagnosis in the work order exists in the existing knowledge graph, then the corresponding fault mode node is retrieved in the knowledge graph, and the actual occurrence frequency attribute of the fault mode node is cumulatively corrected. If the entity or relation extracted from the new work order does not exist in the existing knowledge graph and the overall confidence level is higher than the preset new knowledge threshold, the corresponding new node and new edge will be automatically created in the candidate area, and the expert review process will be triggered. After the review is approved, it will be incorporated into the main knowledge graph to realize the dynamic update of the knowledge graph.

9. A fault mode knowledge graph construction and updating system for motors, characterized in that, include: The ontology model construction module for the motor fault domain is configured to define a core concept class hierarchy for representing motor fault knowledge, define object attributes and semantic relationships, acquire structured and unstructured data and define data attributes, and then set logical constraints and axioms. The core concept class hierarchy is a top-down five-dimensional concept tree structure, with the five dimensions being equipment and component classes, fault phenomenon classes, fault mode classes, diagnosis and detection classes, and maintenance decision classes. The object attributes and semantic relationships include at least the compositional relationships connecting equipment and components, the causal relationships connecting fault modes and fault phenomena, the diagnostic relationships connecting fault modes and detection methods, the positional relationships connecting fault modes and equipment / components, and the governance relationships connecting fault modes and maintenance measures. The data attributes include at least attributes for the fault mode class and attributes for the maintenance decision class. The dual-channel knowledge extraction module is configured to extract knowledge from unstructured data using a deep learning model and from structured data using a mapping method based on the ontology model of the motor fault domain, generating original triplet knowledge. The knowledge fusion module is configured to perform entity alignment, attribute fusion, and conflict resolution on the original triple knowledge, and to integrate knowledge from different data sources to construct a high-quality, unambiguous knowledge graph and store it in a graph database. The attribute fusion adopts a master-slave strategy, with nodes generated from structured data as the master data source and nodes generated from unstructured data as the slave data source. The statistical attributes of the slave data source are attached to the entity nodes determined by the master data source to achieve attribute fusion. The intelligent fault reasoning and dynamic update module is configured to utilize the constructed knowledge graph to respond to user diagnostic requests. Based on a preset dual-weighted algorithm of "theoretical risk priority number and actual occurrence frequency", it calculates the comprehensive recommendation score for each candidate fault, realizing intelligent reasoning and ranking from fault phenomenon to root cause. At the same time, if the work order contains new components or new fault relationships that do not exist in the knowledge graph and the extraction confidence is higher than the preset new knowledge threshold, it is incorporated into the main knowledge graph to realize dynamic updates of the knowledge graph.

Images