A wind turbine generator unit fault scanning and pushing method, device and medium

Abstract

This invention provides a method, apparatus, and medium for fault scanning and push notification of wind turbine generator sets. The method includes acquiring multi-source operating data of the target wind turbine generator set and constructing a wind power knowledge graph and a digital twin model of the wind turbine generator set. When a fault alarm state is detected in the real-time operating data, fault propagation reasoning is performed using the wind power knowledge graph to obtain the initial fault root cause. The real-time operating parameters corresponding to the initial fault root cause are input into the digital twin model of the wind turbine generator set, and a simulated vibration waveform signal is output. The similarity between the simulated vibration waveform signal and the measured waveform signal is calculated to confirm that the initial fault root cause is the target fault root cause. Semantic deduplication is performed on the fault events to which the target fault root cause belongs based on entity association rules. Equipment maintenance status attributes recorded in the wind power knowledge graph are obtained, and the deduplicated fault events are pushed according to preset push rules. This invention can improve the operation and maintenance efficiency of wind turbine generator sets.

Description

Technical Field

[0001] This invention belongs to the field of intelligent operation and maintenance technology of wind turbine generator sets, specifically involving a method, device and medium for fault scanning and push of wind turbine generator sets. Background Technology

[0002] With the continuous growth of installed wind power capacity, intelligent operation and maintenance of wind farms has become crucial for ensuring power generation efficiency and reducing operating costs. Currently, the operation monitoring of wind turbine units mainly relies on Supervisory Control and Data Acquisition (SCADA) systems for data collection and threshold alarms. However, existing technologies have many limitations in handling complex faults.

[0003] First, traditional root cause analysis relies heavily on expert experience or fixed rule engines, making it difficult to handle the complex relationships between faults, equipment, and operating conditions in wind turbines. When faults occur concurrently, the rule base becomes extremely bloated and difficult to maintain, resulting in weak reasoning capabilities and low accuracy. Second, monitoring systems generate massive amounts of real-time alarm signals, including many duplicate or fluctuating alarms caused by the same root cause, leading to information overload for operators and hindering the rapid location of serious faults. Existing technologies lack effective semantic deduplication mechanisms and cannot distinguish the severity and timeliness of alarms. Third, the logic for pushing fault information is rigid, often involving simple mass sending or fixed pushes based solely on fault codes, without considering the real-time maintenance status of the equipment (e.g., equipment under maintenance), easily causing unnecessary disruptions and missing critical information. Furthermore, fault codes, equipment status, maintenance records, and other data are scattered across multiple independent systems, forming data silos and lacking cross-data source correlation analysis capabilities, making it difficult to achieve accurate fault location and root cause tracing. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method, device and medium for fault scanning and push of wind turbine generator sets, so as to solve the problems of fault root cause reasoning relying on manual methods, redundant alarm information, rigid push logic and lack of dynamic adaptability in the prior art, and improve the operation and maintenance efficiency of wind turbine generator sets.

[0005] In a first aspect, the present invention provides a method for fault scanning and notification of wind turbine generator sets, the method comprising the following steps:

[0006] Acquire multi-source operation data of the target wind turbine generator set, and construct a knowledge graph of the wind power field and a digital twin model of the wind turbine generator set based on the multi-source operation data;

[0007] The system acquires real-time operating data of the target wind turbine generator set. When a fault alarm is detected in the real-time operating data, it uses a knowledge graph in the wind power field to perform fault propagation reasoning to obtain the initial root cause of the fault.

[0008] The real-time operating parameters corresponding to the initial fault root cause are input into the digital twin model of the wind turbine generator to trigger dynamic simulation and output the simulated vibration waveform signal.

[0009] Calculate the similarity between the simulated vibration waveform signal and the measured waveform signal in the real-time operation data, and when the similarity is greater than a preset threshold, confirm the initial fault root cause as the target fault root cause and feed the target fault root cause back to the wind power knowledge graph;

[0010] Based on entity association rules in the knowledge graph of the wind power field, semantic deduplication of the fault events to which the root cause of the target fault belongs is performed in the time dimension.

[0011] Obtain equipment maintenance status attributes recorded in the knowledge graph of the wind power field, and combine them with preset push rules to perform dynamic routing push or silent interception of deduplicated fault events.

[0012] Optionally, the process of constructing a knowledge graph for the wind power sector based on multi-source operational data includes:

[0013] Field mapping is performed on structured data from multi-source operational data to extract device entities and operational parameter entities;

[0014] For unstructured data in multi-source running data, word segmentation algorithm is used to identify entity boundaries, continuous bag-of-words model is used for coreference resolution, and semantic extraction model based on bidirectional encoder representation of pre-trained language model is used to extract implicit relationships between entities.

[0015] The extracted entities and implicit relationships are imported into a graph database to construct a knowledge graph for the wind power field.

[0016] Optionally, the process of constructing a digital twin model of a wind turbine generator includes:

[0017] Construct a dynamic model of the main bearing with five fundamental degrees of freedom and twice the number of rolling elements; the differential equations of the dynamic model are expressed as follows: ,in, This indicates the equivalent mass of the bearing object. Indicates the damping coefficient. Indicates bearing stiffness. Indicates that the bearing is in The resultant force in the direction, Indicates that the bearing is in The resultant force in the direction, express Translational acceleration in the direction, express Translational velocity in the direction, express Translational displacement in the direction, express Translational acceleration in the direction, express Translational velocity in the direction, express Translational displacement in the direction.

[0018] Optionally, the digital twin model of the wind turbine generator may also include a gearbox lumped parameter model;

[0019] In the lumped parameter model of the gearbox, the meshing frequencies of the sun gear and planet gears, as well as the meshing frequencies of the planet gears and ring gears, are set to be the same. The formula for the meshing frequency of the planet gear train is defined as follows: ;in, This indicates the meshing frequency of the planetary gear train. Indicates the rotational frequency of the planet carrier. Indicates the number of teeth on the gear ring. Indicates the rotational frequency of the sun gear. Indicates the number of teeth on the sun gear. Indicates the rotational frequency of the planetary gears. This indicates the number of teeth on the planetary gear.

[0020] Optionally, the process of using a knowledge graph in the wind power field to perform fault propagation reasoning to obtain the initial root cause of the fault includes:

[0021] By using graph neural networks to aggregate the information carried by the neighboring nodes around the target in the knowledge graph of the wind power field, the fault propagation law is learned and the root cause probability distribution of the target wind turbine generator failure is predicted.

[0022] When multiple faults occur concurrently, a Bayesian network is used in conjunction with the root cause probability distribution to calculate the posterior probability of each candidate root cause, and the candidate root cause with the highest posterior probability is taken as the initial fault root cause.

[0023] Optionally, based on entity association rules in the wind power knowledge graph, semantic deduplication of the fault events to which the target fault root cause belongs is performed in the time dimension, including:

[0024] During the waveform data parsing phase of real-time running data, a hash deduplication function is used to perform memory-level semantic deduplication on the fault code list in the waveform data field;

[0025] The list of fault codes after deduplication at the memory level is compared with the historical semantic records in the knowledge graph of the wind power field to extract the equipment identification code, fault type and occurrence time associated with the current fault event;

[0026] When the time difference between the current fault event and the historical semantic record for the same device and the same fault type is less than the duplicate judgment validity period attribute value defined in the wind power knowledge graph, the current fault event is marked as a redundant event and the push stream is blocked.

[0027] Optionally, the equipment maintenance status attributes recorded in the wind power knowledge graph are obtained, and combined with preset push rules, dynamic routing push or silent interception is performed on the deduplicated fault events, including:

[0028] When the maintenance status attribute of the corresponding equipment in the wind power knowledge graph is marked as being under maintenance, the silent interception logic is triggered to suspend the push of fault events.

[0029] When the maintenance status attribute is switched to normal status, the historical record supplementation rule is triggered to retrieve and supplement the high-priority fault events during the silent interception period.

[0030] When a fault event fails to be pushed, an exponential backoff retry rule based on a preset time interval in the knowledge graph is triggered to re-push the event. If the number of re-push attempts reaches the maximum set threshold and still fails, a circuit breaker rule is triggered to suspend the current push channel and switch to a backup push channel.

[0031] Secondly, the present invention provides a wind turbine generator fault scanning and notification device, comprising:

[0032] The data construction module is used to acquire multi-source operation data of the target wind turbine generator set, and to construct a knowledge graph in the wind power field and a digital twin model of the wind turbine generator set based on the multi-source operation data.

[0033] The reasoning and prediction module is used to acquire real-time operating data of the target wind turbine generator set, and when a fault alarm state is detected in the real-time operating data, it uses a knowledge graph in the wind power field to perform fault propagation reasoning to obtain the initial root cause of the fault.

[0034] The simulation module is used to input the real-time operating parameters corresponding to the initial fault root cause into the digital twin model of the wind turbine generator to trigger dynamic simulation and output the simulation vibration waveform signal.

[0035] The verification module is used to calculate the similarity between the simulated vibration waveform signal and the measured waveform signal in the real-time operation data. When the similarity is greater than a preset threshold, the initial fault root cause is confirmed as the target fault root cause and the target fault root cause is fed back to the wind power knowledge graph.

[0036] The deduplication module is used to perform semantic deduplication of the fault events to which the root cause of the target fault belongs in the time dimension based on entity association rules in the knowledge graph of the wind power field.

[0037] The push module is used to obtain the equipment maintenance status attributes recorded in the knowledge graph of the wind power field, and combine them with preset push rules to perform dynamic routing push or silent interception of deduplicated fault events.

[0038] Thirdly, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-mentioned wind turbine generator fault scanning and push method.

[0039] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described wind turbine generator fault scanning and push method.

[0040] The present invention has at least the following beneficial effects:

[0041] This invention constructs a knowledge graph in the wind power field, structurally accumulating core elements such as equipment, faults, and maintenance, along with their interrelationships. This solves the problem of isolated multi-source data and provides a knowledge foundation for intelligent reasoning. When a fault is detected, root cause reasoning is first performed using the knowledge graph, followed by simulation verification using an independent digital twin model. By comparing the similarity between the simulation and measured waveform signals, the reasoning results are objectively verified, effectively overcoming the illusions or inaccuracies that may occur when relying solely on rules or data models for reasoning. This significantly improves the accuracy of fault root cause localization. Subsequently, deduplication is performed in the time dimension based on semantic rules in the knowledge graph, and dynamic push notifications are generated in conjunction with the real-time maintenance status of the equipment. This reduces redundant alarms at the source, avoids interference from invalid information during the maintenance period, and achieves precise alarm classification and routing. This greatly improves operation and maintenance efficiency and fault response speed, solving the defects of alarm proliferation and rigid push logic in existing technologies. Attached Figure Description

[0042] The accompanying drawings are provided to further understand the technical solutions of the present invention and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the technical solutions of the present invention, and do not constitute a limitation on the technical solutions of the present invention.

[0043] Figure 1 This is a flowchart of a wind turbine generator fault scanning and push method according to one embodiment of this application;

[0044] Figure 2 This is a schematic diagram of a knowledge graph in the field of wind power according to one embodiment of this application;

[0045] Figure 3 This is a schematic diagram of the structure of the wind turbine generator fault scanning and push device in one embodiment of this application;

[0046] Figure 4 This is a schematic diagram of the structure of a computer device in one embodiment of this application. Detailed Implementation

[0047] The technical solution of the present invention will now be described in detail and completely with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0048] In the description of this invention, it should be noted that the terms "upper", "lower", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0049] To address the problems in existing wind turbine operation and maintenance technologies, such as reliance on manual experience for root cause analysis, redundant alarm information, rigid push logic, and lack of verification of reasoning results, this invention proposes a wind turbine fault scanning and push method.

[0050] Example 1

[0051] like Figure 1 As shown, the wind turbine generator fault scanning and push method provided by the present invention includes steps 11 to 16.

[0052] Step 11: Obtain multi-source operation data of the target wind turbine generator set, and construct a knowledge graph of the wind power field and a digital twin model of the wind turbine generator set based on the multi-source operation data.

[0053] In one feasible implementation, acquiring multi-source operational data of the target wind turbine generator set includes, but is not limited to: structured data from relational databases, such as equipment configuration tables, fault code mapping tables, scheduled maintenance records, and responsible person information; real-time streaming data from time-series databases, such as temperature, power, speed, and status words collected by SCADA systems, and vibration acceleration time-domain waveforms, spectra, and envelope spectra collected by CMS systems; and unstructured maintenance record texts, expert fault analysis reports, and traditional fault root cause reasoning rule documents. Based on this data, on the one hand, a knowledge graph in the wind power field is constructed to structure expert experience, historical cases, equipment associations, and other knowledge; on the other hand, digital twin models of key components, such as lightweight simulation models of main bearings, gearboxes, and generators, are constructed to simulate the physical response under fault conditions.

[0054] Step 12: Obtain the real-time operating data of the target wind turbine generator set, and when a fault alarm state is detected in the real-time operating data, use the knowledge graph in the wind power field to perform fault propagation reasoning to obtain the initial root cause of the fault.

[0055] In one feasible implementation, real-time operating data of the target wind turbine generator is continuously acquired, and the data stream is parsed and monitored in real time. When a fault alarm status is detected in the real-time data, for example, a code with the value "1001" is parsed from the SCADA status word stream. By querying the fault code semantic mapping dictionary in the knowledge graph, the fault phenomenon corresponding to the code, "gearbox oil temperature too high," is identified, and the fault reasoning process is triggered. Using the constructed wind power knowledge graph, fault propagation reasoning is performed through preset logical rules (such as IF-THEN rules) or graph neural networks to mine possible root cause nodes leading to the fault phenomenon, such as "gear wear," "cooling system failure," and "insufficient lubricating oil," and an initial list of fault root causes and their confidence probabilities are formed.

[0056] Step 13: Input the real-time operating parameters corresponding to the initial fault root cause into the digital twin model of the wind turbine generator to trigger dynamic simulation and output the simulated vibration waveform signal.

[0057] To verify the accuracy of the inference results, the real-time operating parameters corresponding to the initial root cause of failure with the highest confidence (e.g., "gear wear") are extracted. These parameters include, but are not limited to: spindle speed, real-time load, lubricating oil temperature, and gearbox input shaft speed.

[0058] These parameters are input into a pre-built digital twin model of the gearbox. Based on these real-time operating parameters and combined with the preset "gear wear" fault mechanism, the digital twin model dynamically simulates the vibration response that should theoretically occur under this fault state and outputs the simulated vibration waveform signal.

[0059] Step 14: Calculate the similarity between the simulated vibration waveform signal and the measured waveform signal in the real-time operation data. When the similarity is greater than the preset threshold, confirm that the initial fault root cause is the target fault root cause and feed the target fault root cause back to the wind power knowledge graph.

[0060] In one feasible implementation, the cosine similarity calculation method can be used to calculate the similarity between the simulated vibration waveform signal and the measured waveform signal in the real-time running data.

[0061] Specifically, through calculation formula

[0062]

[0063] Obtain the simulated vibration waveform signal Compared with the measured waveform signal similarity ;in, The first part represents the simulated vibration waveform signal. The amplitude of each sampling point The measured waveform signal is represented by the first... The amplitude of each sampling point Indicates the number of sampling points, molecule Used to measure the degree of matching between corresponding points of two waveforms, denominator It is used to normalize similarity, eliminate the influence of the overall amplitude of the waveform, and simply compare the similarity of shapes.

[0064] If the calculated similarity is greater than a preset threshold (e.g., 0.85), it indicates that the reasoning result of the knowledge graph highly matches the actual response of the physical entity, thus confirming that "gear wear" is indeed the target root cause of this failure. This confirmed root cause information will be fed back to the wind power knowledge graph as a new instance or to update the confidence of existing nodes, forming knowledge accumulation and iteration. If the similarity is lower than the threshold, the graph marks the reasoning result as "low confidence," returning to re-reasoning or prompting manual review.

[0065] Step 15: Based on the entity association rules in the knowledge graph of the wind power field, perform semantic deduplication of the fault events to which the root cause of the target fault belongs in the time dimension.

[0066] After confirming the root cause of the fault, semantic deduplication of the fault event will be performed based on the entity association rules defined in the wind power knowledge graph. For example, the "fault entity" in the knowledge graph defines a "duplication judgment validity period" attribute, with a default value of 24 hours. By querying the graph, it is checked whether there are historical event records for the same fault entity ("gearbox oil temperature too high") associated with the same equipment entity (such as "wind turbine No. 6") within the last 24 hours. If so, the event is marked as a "redundant event" and blocked from entering the push process, and is only recorded internally; otherwise, it is considered a "new event" and enters the push preparation stage.

[0067] Step 16: Obtain the equipment maintenance status attributes recorded in the knowledge graph of the wind power field, and combine them with the preset push rules to perform dynamic routing push or silent interception on the deduplicated fault events.

[0068] Specifically, before pushing an alarm, the system queries the "Maintenance Status" attribute of the "No. 6 Wind Turbine" entity in the wind power knowledge graph. If this attribute is marked as "Under Maintenance," a silent interception logic is triggered, suspending all alarm pushes related to this device. However, the event is still recorded as a "Silent Period Event" node and added to the graph, preserving the complete audit chain. If the device status is "Normal," the system dynamically routes the alarm based on preset push rules. For example, for an emergency alarm of level 1, such as "Gearbox Oil Temperature Too High," the system associates the "Responsible Person Entity" linked from the device entity in the knowledge graph, obtains the maintenance supervisor's mobile phone number and corporate communication account, and sends an immediate push via SMS and WeChat, requiring a response within 10 minutes. If the initial push fails, the system queries the "Retry Rule Entity" in the graph, triggering an exponential backoff retry strategy for a re-push.

[0069] It should be noted that when encountering technical problems such as reliance on manual fault root cause reasoning, redundant alarm information, and rigid push logic lacking dynamic adaptability, the conventional improvement approach for those skilled in the art is usually to simply optimize the judgment threshold or simply introduce machine learning algorithms for black-box pattern recognition. However, the wind turbine generator fault scanning and push method provided by this invention creatively breaks through the single barrier of "data flow driven" and constructs an integrated architecture of "virtual mapping (digital twin) - knowledge reasoning (graph computation) - closed-loop verification." The knowledge graph provides positive root cause reasoning with physical logical interpretability, while the digital twin model undertakes the task of reverse simulation verification. The combination of the two is not a simple patchwork of existing technologies, but rather uses the "waveform similarity" of twin simulation as conditional feedback to directly correct or confirm the graph reasoning results of the graph, completely solving the "reasoning illusion" problem inherent in pure deep learning. Furthermore, transforming the "deduplication and push mechanism of business dimensions" into "attribute node relationships of the knowledge graph" represents a disruptive reconstruction of the underlying logic of traditional monitoring software architecture, enabling adaptive and flexible decision-making based on the machine's real-time maintenance status graph.

[0070] Example 2

[0071] This embodiment, based on Embodiment 1, details the construction process of a knowledge graph in the wind power field, particularly the processing of unstructured data.

[0072] First, for structured data, the construction process is relatively straightforward. Taking relational databases as an example, it involves using the SQL statement "SELECT ... The `FROM fan_boot_maintrans_list WHERE main_code={mainCode}` function extracts device entities from the device configuration table, mapping core attributes such as main version number, device name, and site name to the "Unit Equipment" entity node in the graph. Similarly, it loads the mapping relationship between SCADA system status words and standardized fault descriptions from the `view_sc_codac` data view, constructs a global dictionary `sc_ids`, and maps status word values ​​(such as `sc_id_1001`) to structured data containing alarm level `break_level`, responsible department, and fault description, completing the semantic mapping from status words to "fault entities". Maintenance period information is extracted from the `site_unit_state_new` table and marked as the `standard_description` attribute of the "device entity".

[0073] The processing of unstructured text data such as maintenance records and expert reports is more complex.

[0074] First, the text is segmented using a word segmentation method based on the Transformer model.

[0075] For example, the sentence "Wind turbine No. 6 failed due to excessive temperature" is broken down into the word sequence "6 / wind turbine / due to / excessive / temperature / , / caused / main bearing / failure". Then, the word segmentation results are matched with a wind power terminology dictionary to identify entities and their types. The matching results yield entity nodes: [("Wind turbine No. 6", "unit equipment"), ("temperature", "operating parameters"), ("main bearing", "unit equipment")], and attributes: [("status", "excessive temperature")].

[0076] Then, the continuous bag-of-words model is used to resolve coreference issues and address the problem of ambiguous referencing in the text.

[0077] For example: "The main bearing of wind turbine No. 6 overheated, causing it to shut down. Maintenance personnel found insufficient lubrication in the main bearing." Here, "it" refers to "wind turbine No. 6," and the two instances of "main bearing" refer to the same entity. This embodiment uses the Continuous Bag-of-Words (CBOW) algorithm of the Word2vec model for training and resolution. The CBOW model predicts the center word through context (such as "unit," "transmission system," "generator," and "fault"), effectively learning word vectors on small datasets. This allows it to identify whether different expressions point to the same entity, ensuring that these entities correspond to only one node in the graph and avoiding semantic ambiguity.

[0078] Next, a semantic extraction model based on the BERT pre-trained language model (SpERT) is used to extract the implicit relationships between entities.

[0079] For example, consider a complex fault message: "Abnormal noise from the generator; endoscopic examination revealed chipping on the outer ring of the generator bearing." The handling procedure is as follows:

[0080] Word segmentation process yields a list of the smallest semantic units: [“generator”, “abnormal noise”, “through”, “endoscope”, “detection”, “discovery”, “generator”, “bearing”, “outer ring”, “chipped edge”].

[0081] These units are converted into an encoded sequence in the format of [fault description global information] + [the sequence of numeric IDs corresponding to the smallest semantic unit] + [separator].

[0082] Entity recognition and localization: Record the start and end positions of each entity in the sentence and its label.

[0083] "Generator": Identified as an equipment entity, tagged "Wind Power Equipment - Generator".

[0084] "Endoscope": Identified as a testing equipment entity, tagged "Wind Power Testing Equipment - Endoscope".

[0085] “Bearing”: Identified as an entity of the equipment component class, tagged “Wind power equipment component - bearing”.

[0086] “Break Edge”: Identified as a root cause entity of the fault, labeled “Wind Power Fault Root Cause - Break Edge”.

[0087] Relationship classification and extraction: The SpERT model predicts the relationships between entities based on their context and location information.

[0088] (“Bearing”, “Cracked Edge”): The existence of a “fault association” indicates that the bearing has experienced a chipped edge fault.

[0089] ("Endoscope", "Bearing"): The existence of a "detection" relationship indicates that the bearing was detected using an endoscope.

[0090] ("Generator", "Abnormal noise"): The existence of a "fault association" indicates that the generator is experiencing an abnormal noise fault.

[0091] Through the NLP processing flow described above, information in unstructured text is extracted into "entity-relationship-attribute" triples. Finally, all triples extracted from both structured and unstructured data are compiled into CSV files and imported in batches into the graph database Neo4j to complete the physical construction of a knowledge graph for the wind power sector, resulting in... Figure 2 The example shown is a knowledge graph of wind turbine generator failure events.

[0092] In one feasible implementation, based on the application requirements of the intelligent operation and maintenance system, 13 entity types are summarized and defined for the constructed wind power equipment fault knowledge ontology, as shown in Table 1.

[0093] Table 1.

[0094]

[0095] Eleven relationship types are defined between entities, as shown in Table 2.

[0096] Table 2.

[0097]

[0098] Based on the actual situation of the power plant, entities are sorted out and core attributes are defined for each entity. The attributes must correspond to the business data fields, as shown in Table 3.

[0099] Table 3.

[0100]

[0101] Example 3

[0102] This embodiment of the invention details the construction process and related formulas of the digital twin model of the main bearing and gearbox based on Embodiment 2.

[0103] First, a dynamic model of the main bearing is constructed. This embodiment uses a lumped parameter model with 5+2N degrees of freedom, where N is the number of rolling elements. The basic assumptions of this model are: the inner ring is driven by a constant rotational speed; the rolling elements achieve rotation and revolution under the drive of contact force; the cage is idealized to ensure that the rolling elements are evenly distributed; and the effects of clearance and lubrication are not considered for the time being. The degrees of freedom of the model are as follows:

[0104] Inner circle: translational degrees of freedom in the X and Y directions, and rotational degrees of freedom in the Z direction.

[0105] Outer ring: Translational degrees of freedom in the X and Y directions.

[0106] Each rolling element has a rotational degree of freedom and a translational degree of freedom along the bearing radial direction.

[0107] Based on Newton's second law, the system's dynamic differential equations can be expressed as: ,in, This indicates the equivalent mass of the bearing object. Indicates the damping coefficient. Indicates bearing stiffness. Indicates that the bearing is in The resultant force in the direction, Indicates that the bearing is in The resultant force in the direction (including Hertzian contact force, damping force, etc.). express Translational acceleration in the direction, express Translational velocity in the direction, express Translational displacement in the direction, express Translational acceleration in the direction, express Translational velocity in the direction, express Translational displacement in the direction.

[0108] To perform fault diagnosis, it is necessary to calculate the bearing's characteristic failure frequencies. For rolling bearings with a fixed outer ring and an inner ring that rotates with the shaft, these mainly include: inner ring failure frequency (BPFI), outer ring failure frequency (BPFO), ball failure frequency (BSF), and cage failure frequency (FTF).

[0109] In one feasible implementation, through calculation formula Obtain the spindle rotation frequency , Indicates rotational speed;

[0110] Through calculation formula Obtain the inner ring fault frequency ; Indicates the number of rolling elements. Indicates the diameter of the rolling element. Indicates the diameter of the pitch circle;

[0111] Through calculation formula Obtain the outer ring fault frequency ;

[0112] Through calculation formula Obtain the ball failure frequency ;

[0113] Through calculation formula Obtain the cage failure frequency .

[0114] In another feasible implementation, if the above-mentioned bearing parameters are not available, the bearing failure frequency can be approximated using the following formula: 、 、 、 .

[0115] Secondly, a lumped parameter model of the gearbox is constructed. Specifically, a lumped parameter model including mass, stiffness, and damping is used to approximate the vibration behavior, with the core equations as follows: ;in, This indicates the equivalent mass of the planetary gear. Indicates meshing damping, This represents the meshing stiffness, which varies periodically over time. This represents the combined force of meshing impact force and fault excitation.

[0116] For planetary gear trains, it is necessary to calculate their complex fault characteristic frequencies. The expression for the meshing frequency of a planetary gear train is as follows:

[0117]

[0118]

[0119]

[0120] in, Indicates the planetary carrier frequency. Indicates the number of teeth on the gear ring. Indicates the frequency of solar rotation. Indicates the number of teeth on the sun gear. Indicates the frequency of planetary rotation. Indicates the number of teeth on the planetary gear. Indicates the number of planetary gears. Indicates the meshing frequency.

[0121] Based on this, the characteristic frequencies of various faults can be derived, as shown in Table 4.

[0122] Table 4.

[0123]

[0124] These calculation formulas are the core basis for gearbox fault diagnosis and are all integrated into the digital twin model.

[0125] In model applications, a physical-virtual data flow fusion mechanism is established:

[0126] Physical → Virtual (Data Input): The pre-processed vibration signals (such as peak value, effective value, and amplitude at fault characteristic frequency) collected by field sensors are input into the virtual model in real time as the "real-time input parameters" of the model to calibrate the model's operating status.

[0127] Virtual to Physical (Data Output): Based on the input real-time data and the set fault type (e.g., inner ring fault), the virtual model runs the above dynamic equations and simulates the vibration acceleration signal under that fault. By comparing the similarity between the simulated signal and the measured signal, the fault type and severity can be diagnosed and verified. The trained model is integrated into a digital twin, receiving data in real time, outputting diagnostic results, and methods such as cosine similarity can be incorporated to reduce the difference between simulation and measurement, thereby improving diagnostic accuracy.

[0128] By combining the structured knowledge of knowledge graphs with the physical mechanism model of digital twins, this solution constructs a dual-source driven model of "knowledge + mechanism," laying a solid foundation for subsequent intelligent reasoning and verification. This step extracts knowledge from unstructured text and constructs a knowledge graph using NLP technology, and models the physical mechanism. Its technical effect lies in the fact that expert experience and physical laws are preserved in a computable form, solving the problems of difficult knowledge reuse and complex model construction in traditional methods.

[0129] Example 4

[0130] Based on Example 3, this embodiment details the specific implementation methods of fault root cause reasoning, semantic deduplication, and intelligent push, as well as the underlying graph semantic driving logic.

[0131] First, a Graph Neural Network (GNN) is used to learn fault propagation patterns from graph-structured data. The GNN aggregates information about neighboring nodes (such as nodes related to "gear wear," "cooling system failure," and "insufficient lubricating oil") and their edge relationships around a target node (e.g., "gearbox oil temperature too high") through a message-passing mechanism. Its core aggregation formula can be expressed as:

[0132]

[0133] in, Represents a node In the The layer's feature representation, taking the gearbox node as an example, is after... The final information carried by the gearbox node after this information exchange. It includes not only the node's... The information (such as temperature and speed) of the gearbox itself is aggregated with information from its surrounding neighboring nodes (such as "generator" and "spindle") and associated edges (such as "connection relationships"). Represents a node In the Layer feature representation. Indicates the first The update function of the layer (usually a multilayer perceptron). This represents aggregate functions, typically including SUM, MEAN, etc. Indicates the first Learnable functions of a layer. Represents a node The neighboring nodes of a node, in the knowledge graph, are related to the node. All other entities that are directly connected to the corresponding entity. Indicates the connection node with neighboring nodes The characteristics of an edge, for example in a knowledge graph, the relationship between "gearbox" and "temperature sensor" is "being monitored". The attributes of this edge (such as "monitoring accuracy" and "association strength") are... , Represents a node In the Layer feature representation.

[0134] Secondly, when multiple faults occur concurrently, Bayesian networks are introduced to handle uncertainty. The Bayesian network uses the probability distribution output by the GNN as a prior, combined with multiple observed fault phenomena (evidence), to calculate the posterior probability of each candidate root cause. For example, when both "excessive vibration" and "excessive temperature" are observed simultaneously, the Bayesian network can calculate that the posterior probability of "gear wear" is higher than that of "insufficient lubricating oil," thus outputting the candidate root cause with the highest posterior probability as the initial fault root cause. This combination of GNN and Bayesian networks balances the ability of GNN to automatically learn features from complex graph structures with the advantages of Bayesian networks in handling uncertainty and causal reasoning.

[0135] In the semantic deduplication stage, a layered deduplication mechanism was constructed. The first layer is memory-level semantic deduplication: when parsing waveform data, the `deduplicateList` hash function is used to deduplicate the fault code list extracted from the `data` field, ensuring that the same fault code will not trigger multiple alarms in a single parsing task. The second layer is persistent cross-window deduplication: the memory-deduplicated fault code list is compared with the historical semantic records of the "fault event layer" in the knowledge graph. The comparison process is based on the ternary association rule of "device-fault-time". Specifically, the graph is queried for historical event records of the same "fault entity" (e.g., "gearbox oil temperature too high") associated with the same "device entity" (e.g., "fan #6") within the last 24 hours. If the time difference between the occurrence time of the current fault event and the latest historical event is less than the "duplicate judgment validity period" attribute value defined in the knowledge graph (default 24 hours), the current event is marked as a "redundant event" (status=0), preventing it from entering the push process, and associated with historical events. Otherwise, it is marked as a "new event" (status=1) and added to the graph as a new node. This mechanism effectively solves the problem of alarm storms at the semantic level.

[0136] In the intelligent push notification process, dynamic decisions are made based on the rich semantic rules defined in the knowledge graph, thus decoupling the push logic from the code. Its core lies in establishing a triplet relationship of "alarm level - responsible person - push rule".

[0137] Alarm level semantic mapping: The knowledge graph defines an "alarm level (break_level)" attribute for each "fault entity".

[0138] Emergency Alarm (break_level=1): The "Operations Supervisor and On-site Manager" node in the association graph has a response time of 10 minutes.

[0139] Important alert (break_level=2): Associated with the "Operations Team" node, with a response time of 1 hour.

[0140] General alert (break_level=3): Associated with the "Daily Summary Push" rule node.

[0141] In one feasible implementation, when the device's "maintenance status" attribute is "normal," the corresponding responsible person node is matched according to the alarm level, and contact information such as mobile phone number and enterprise communication tool account is obtained from the associated "user entity" node. Then, the rules of the "push channel entity" are matched, such as emergency alarms being pushed simultaneously via SMS and WeChat Work, with the content being structured and populated from the "device entity" and "fault entity."

[0142] When the device's "maintenance status" attribute is "under maintenance," the "alarm pause rule" is triggered, pausing all associated alarm pushes, but the event is still entered into the database as a "quiet period event" node. When the status switches back to "normal," the "incremental scan" rule is triggered to retrieve high-priority faults associated during the quiet period and resend any missed alarms according to the rules.

[0143] When a push notification fails, the "Retry Rule Entity" in the graph is triggered, executing an exponential backoff retries (e.g., 5 seconds, 30 seconds, 5 minutes, up to 3 times). If 10 consecutive failures occur, the "Circuit Break Rule Entity" is triggered, suspending the channel for 1 hour and switching to the "Backup Push Channel Entity" (e.g., switching from SMS to WeChat Work), and generating a "Circuit Break Event Entity" for monitoring. The status of the entire process (success, failure, retrying, circuit break) is updated back to the "Fault Event Entity" attribute in the graph in real time, forming a complete audit chain.

[0144] It is worth mentioning that this invention realizes closed-loop alarm control and high availability guarantee of network channels throughout the entire operation and maintenance process; its non-obviousness is reflected in the fact that maintenance status, retry time parameters and circuit breaker mechanism are all abstracted into semantic nodes and entity relationships in a knowledge graph, so that the push strategy is no longer a one-way program instruction flow, but a flexible decision network that can be calculated and dynamically reconstructed in real time through the graph inference engine.

[0145] Example 5

[0146] This invention discloses a fault scanning and push device for wind turbine generator sets, such as... Figure 3 As shown, the device 300 includes:

[0147] The data construction module 301 is used to acquire multi-source operation data of the target wind turbine generator set, and to construct a knowledge graph in the wind power field and a digital twin model of the wind turbine generator set based on the multi-source operation data.

[0148] The reasoning and prediction module 302 is used to acquire real-time operating data of the target wind turbine generator set, and when a fault alarm state is detected in the real-time operating data, it uses a knowledge graph in the wind power field to perform fault propagation reasoning to obtain the initial fault root cause.

[0149] Simulation module 303 is used to input the real-time operating parameters corresponding to the initial fault root cause into the digital twin model of the wind turbine generator to trigger dynamic simulation and output the simulation vibration waveform signal.

[0150] The verification module 304 is used to calculate the similarity between the simulated vibration waveform signal and the measured waveform signal in the real-time operation data, and when the similarity is greater than a preset threshold, it confirms that the initial fault root cause is the target fault root cause and feeds the target fault root cause back to the wind power knowledge graph.

[0151] The deduplication module 305 is used to perform semantic deduplication of the fault events to which the target fault root cause belongs in the time dimension based on entity association rules in the knowledge graph of the wind power field.

[0152] The push module 306 is used to obtain the equipment maintenance status attributes recorded in the knowledge graph of the wind power field, and perform dynamic routing push or silent interception on the deduplicated fault events in combination with the preset push rules.

[0153] It should be noted that the information interaction and execution process between the above-mentioned devices / units are based on the same concept as the method embodiments of this application. Their specific functions and technical effects can be found in the method embodiments section, and will not be repeated here. Those skilled in the art will understand that, for ease of description and brevity, the division of the above-mentioned functional units and modules is only used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0154] like Figure 4 As shown, embodiments of the present invention provide a computer device, such as... Figure 4 As shown, the terminal device D10 of this embodiment includes: at least one processor D100 ( Figure 4 The diagram shows only one processor, a memory D101, and a computer program D102 stored in the memory D101 and executable on the at least one processor D100, wherein the processor D100 executes the computer program D102 to implement the steps in any of the above method embodiments.

[0155] Specifically, when the processor D100 executes the computer program D102, it acquires multi-source operating data of the target wind turbine generator set and constructs a wind power domain knowledge graph and a digital twin model of the wind turbine generator set based on the multi-source operating data; it acquires real-time operating data of the target wind turbine generator set, and when a fault alarm state is detected in the real-time operating data, it uses the wind power domain knowledge graph to perform fault propagation reasoning to obtain the initial fault root cause; it inputs the real-time operating parameters corresponding to the initial fault root cause into the wind turbine generator set digital twin model to trigger dynamic simulation and output simulated vibration waveform signals; it calculates the similarity between the simulated vibration waveform signal and the measured waveform signal in the real-time operating data, and when the similarity is greater than a preset threshold, it confirms that the initial fault root cause is the target fault root cause and feeds back the target fault root cause to the wind power domain knowledge graph; it performs semantic deduplication of the fault events to which the target fault root cause belongs in the time dimension based on the entity association rules in the wind power domain knowledge graph; it acquires the equipment maintenance status attributes recorded in the wind power domain knowledge graph, and performs dynamic routing push or silent interception on the deduplicated fault events in combination with preset push rules. By constructing a knowledge graph for the wind power sector, core elements such as equipment, faults, and maintenance, along with their interrelationships, are structurally stored, resolving the issue of isolated multi-source data and providing a knowledge foundation for intelligent reasoning. When a fault is detected, root cause reasoning is first performed using the knowledge graph, followed by simulation verification using an independent digital twin model. By comparing the similarity between the simulation and measured waveform signals, the reasoning results are objectively verified, effectively overcoming the illusions or inaccuracies that may occur when relying solely on rules or data models for reasoning, significantly improving the accuracy of fault root cause localization. Subsequently, deduplication is performed based on the semantic rules in the knowledge graph, and dynamic push notifications are generated in conjunction with the real-time maintenance status of the equipment. This reduces redundant alarms at the source, avoids interference from invalid information during the maintenance period, and achieves precise alarm classification and routing, greatly improving operation and maintenance efficiency and fault response speed, and solving the defects of alarm proliferation and rigid push logic in existing technologies.

[0156] The processor D100 can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0157] In some embodiments, the memory D101 may be an internal storage unit of the terminal device D10, such as a hard disk or memory of the terminal device D10. In other embodiments, the memory D101 may be an external storage device of the terminal device D10, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the terminal device D10. Furthermore, the memory D101 may include both internal and external storage units of the terminal device D10. The memory D101 is used to store the operating system, applications, bootloader, data, and other programs, such as the program code of the computer program. The memory D101 can also be used to temporarily store data that has been output or will be output.

[0158] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps described in the various method embodiments above.

[0159] This application provides a computer program product that, when run on a terminal device, enables the terminal device to implement the steps described in the various method embodiments above.

[0160] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of protection of this application is limited to these examples; within the framework of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of one or more embodiments of this application as described above, which are not provided in detail for the sake of brevity.

[0161] One or more embodiments in this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of this application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments in this application should be included within the protection scope of this application.

Claims

1. A method for fault scanning and notification of wind turbine generator sets, characterized in that, include: Acquire multi-source operation data of the target wind turbine generator set, and construct a knowledge graph of the wind power field and a digital twin model of the wind turbine generator set based on the multi-source operation data; The multi-source operational data includes structured data from relational databases, real-time streaming data from time-series databases, and unstructured data. The structured data includes equipment configuration tables, fault code mapping tables, scheduled maintenance records, and responsible personnel information. The real-time streaming data includes temperature, power, speed, and status words collected by the SCADA system, as well as vibration acceleration time-domain waveforms, spectra, and envelope spectra collected by the CMS system. The unstructured data includes maintenance record texts, expert fault analysis reports, and traditional fault root cause reasoning rule documents. The real-time operating data of the target wind turbine generator is obtained, and when a fault alarm state is detected in the real-time operating data, the fault propagation reasoning is performed using the knowledge graph in the wind power field to obtain the initial fault root cause. The real-time operating parameters corresponding to the initial fault root cause are input into the digital twin model of the wind turbine generator to trigger dynamic simulation and output the simulated vibration waveform signal. Calculate the similarity between the simulated vibration waveform signal and the measured waveform signal in the real-time operating data, and when the similarity is greater than a preset threshold, confirm that the initial fault root cause is the target fault root cause and feed the target fault root cause back to the wind power knowledge graph; Based on the entity association rules in the wind power knowledge graph, semantic deduplication is performed on the fault events to which the target fault root cause belongs in the time dimension. Obtain the equipment maintenance status attributes recorded in the wind power knowledge graph, and combine them with preset push rules to perform dynamic routing push or silent interception on the deduplicated fault events.

2. The method for fault scanning and notification of wind turbine generator sets according to claim 1, characterized in that, The process of constructing a knowledge graph for the wind power sector based on multi-source operational data includes: The structured data in the multi-source operational data is extracted by field mapping to obtain device entities and operational parameter entities; The unstructured data in the multi-source running data is identified by word segmentation algorithm, coreference resolution is performed by continuous bag-of-words model, and implicit associations between entities are extracted by semantic extraction model based on bidirectional encoder representation of pre-trained language model. The extracted entities and their implicit relationships are imported into a graph database to construct the knowledge graph for the wind power sector.

3. The method for fault scanning and notification of wind turbine generator sets according to claim 2, characterized in that, The process of building a digital twin model of a wind turbine generator includes: A dynamic model of the main bearing with five fundamental degrees of freedom and twice the number of rolling elements is constructed; the differential equations of the dynamic model are expressed as follows: ,in, This indicates the equivalent mass of the bearing object. Indicates the damping coefficient. Indicates bearing stiffness. Indicates that the bearing is in The resultant force in the direction, Indicates that the bearing is in The resultant force in the direction, express Translational acceleration in the direction, express Translational velocity in the direction, express Translational displacement in the direction, express Translational acceleration in the direction, express Translational velocity in the direction, express Translational displacement in the direction.

4. The wind turbine generator fault scanning and push method according to claim 3, characterized in that, The digital twin model of the wind turbine generator set also includes a gearbox lumped parameter model; In the lumped parameter model of the gearbox, the meshing frequencies of the sun gear and planet gears, as well as the meshing frequencies of the planet gears and ring gears, are set to be the same. The formula for the meshing frequency of the planet gear train is defined as follows: ;in, This indicates the meshing frequency of the planetary gear train. Indicates the rotational frequency of the planet carrier. Indicates the number of teeth on the gear ring. Indicates the rotational frequency of the sun gear. Indicates the number of teeth on the sun gear. Indicates the rotational frequency of the planetary gears. This indicates the number of teeth on the planetary gear.

5. The wind turbine generator fault scanning and push method according to claim 4, characterized in that, The process of using knowledge graphs in the wind power field to deduce the initial root cause of a fault through fault propagation reasoning includes: By using graph neural networks to aggregate the information carried by the neighboring nodes around the target in the knowledge graph of the wind power field, the fault propagation law is learned and the root cause probability distribution of the target wind turbine generator failure is predicted. When multiple faults occur concurrently, a Bayesian network is used in conjunction with the root cause probability distribution to calculate the posterior probability of each candidate root cause, and the candidate root cause with the highest posterior probability is taken as the initial fault root cause.

6. The wind turbine generator fault scanning and push method according to claim 5, characterized in that, Based on the entity association rules in the wind power knowledge graph, semantic deduplication is performed on the fault events to which the target fault root cause belongs in the time dimension, including: During the waveform data parsing stage of the real-time running data, the fault code list in the waveform data field is deduplicated at the memory level using a hash deduplication function; The list of fault codes after deduplication at the memory level is compared with the historical semantic records in the wind power knowledge graph to extract the equipment identification code, fault type and occurrence time associated with the current fault event. When the time difference between the current fault event and the historical semantic record for the same device and the same fault type is less than the duplicate determination validity period attribute value defined in the wind power knowledge graph, the current fault event is marked as a redundant event and the push stream is blocked.

7. The wind turbine generator fault scanning and push method according to claim 6, characterized in that, Obtain the equipment maintenance status attributes recorded in the wind power knowledge graph, and, in conjunction with preset push rules, perform dynamic routing push or silent interception on deduplicated fault events, including: When the maintenance status attribute of the corresponding equipment in the wind power knowledge graph is marked as being under maintenance, the silent interception logic is triggered to pause the push of fault events. When the maintenance status attribute is switched to the normal status, the historical record supplementation rule is triggered to retrieve and supplement the high-priority fault events during the silent interception period. When a fault event fails to be pushed, an exponential backoff retry rule based on a preset time interval in the knowledge graph is triggered to re-push the event. If the number of re-push attempts reaches the maximum set threshold and still fails, a circuit breaker rule is triggered to suspend the current push channel and switch to a backup push channel.

8. A fault scanning and push device for wind turbine generator sets, characterized in that, include: The data construction module is used to acquire multi-source operation data of the target wind turbine generator set, and construct a knowledge graph in the wind power field and a digital twin model of the wind turbine generator set based on the multi-source operation data. The multi-source operational data includes structured data from relational databases, real-time streaming data from time-series databases, and unstructured data. The structured data includes equipment configuration tables, fault code mapping tables, scheduled maintenance records, and responsible personnel information. The real-time streaming data includes temperature, power, speed, and status words collected by the SCADA system, as well as vibration acceleration time-domain waveforms, spectra, and envelope spectra collected by the CMS system. The unstructured data includes maintenance record texts, expert fault analysis reports, and traditional fault root cause reasoning rule documents. The reasoning and prediction module is used to acquire the real-time operating data of the target wind turbine generator set, and when a fault alarm state is detected in the real-time operating data, it uses the knowledge graph in the wind power field to perform fault propagation reasoning to obtain the initial fault root cause. The simulation module is used to input the real-time operating parameters corresponding to the initial fault root cause into the digital twin model of the wind turbine generator to trigger dynamic simulation and output the simulation vibration waveform signal. The verification module is used to calculate the similarity between the simulated vibration waveform signal and the measured waveform signal in the real-time operation data, and when the similarity is greater than a preset threshold, to confirm that the initial fault root cause is the target fault root cause and to feed the target fault root cause back to the wind power knowledge graph. The deduplication module is used to perform semantic deduplication of the fault events to which the target fault root cause belongs in the time dimension based on the entity association rules in the wind power knowledge graph. The push module is used to obtain the equipment maintenance status attributes recorded in the wind power knowledge graph, and, in combination with the preset push rules, to perform dynamic routing push or silent interception on the deduplicated fault events.

9. A computer device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the wind turbine generator fault scanning and push method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores a computer program, which, when executed by a processor, implements the steps of the wind turbine generator fault scanning and push method as described in any one of claims 1 to 7.

Images