A dynamic calibration method, system, device and medium for a wind turbine generator system failure monitoring system

Abstract

The application discloses a kind of wind turbine fault monitoring system dynamic calibration method, system, equipment and medium, belong to wind turbine monitoring calibration technical field, this method includes: first, based on historical data to construct fault mode library and calculate risk priority number, screen main monitoring fault mode;Second, deploy external monitoring system and construct error propagation network model;Further, the comprehensive confidence influence value of monitoring parameter is calculated in real time, identify to be calibrated parameter and generate calibration sequence;Finally, matching hardware, transmission or model level calibration operation, and according to calibration result dynamic update risk level.The application can accurately locate error source and execute targeted calibration, effectively solve the problem of calibration lag and blindness in the prior art, significantly improve the diagnostic accuracy and operation robustness of wind turbine fault monitoring system.

Description

Technical Field

[0001] This application belongs to the field of wind turbine monitoring and calibration technology, and specifically relates to a dynamic calibration method, system, equipment and medium for a wind turbine fault monitoring system. Background Technology

[0002] As the core power generation equipment of wind farms, wind turbines typically operate in complex scenarios with harsh environments and variable operating conditions. Due to the long-term effects of alternating loads, extreme weather, and grid fluctuations, key components such as gearboxes, generators, and blades are prone to failure. In order to ensure the safe and stable operation of wind turbines and reduce unplanned downtime, fault monitoring systems are widely used. Currently, fault monitoring of wind turbines mainly relies on data acquisition and monitoring systems and additional external monitoring devices. These systems deploy sensors for vibration, temperature, acoustic signatures, and displacement on key components to collect operational data in real time. Threshold alarms or simple model diagnostics are then used to determine the turbine's status. However, existing monitoring systems often face the dilemma of both high false alarm rates and missed detections in practical applications. The main reason for this is that the monitoring system's own state drifts or its performance degrades over time. Current maintenance strategies typically involve manual calibration of sensors and systems only when false alarms occur frequently or during regular maintenance. This passive calibration method has significant lag and lacks quantitative assessment of the reliability of monitoring parameters. On the one hand, the zero-point drift of the sensors themselves, interference during signal transmission, and the influence of environmental noise can lead to distorted monitoring data, triggering false alarms. On the other hand, when multiple monitoring parameters are coupled to characterize a certain fault feature, existing technologies struggle to accurately pinpoint which parameter's error caused the diagnostic bias, resulting in a lack of targeted calibration. Therefore, how to establish an intelligent dynamic calibration mechanism for wind turbine fault monitoring systems to improve the robustness and diagnostic accuracy of the monitoring systems is a technical problem that urgently needs to be solved in the field. Summary of the Invention

[0003] To address the aforementioned issues, this application provides a dynamic calibration method, system, equipment, and medium for a wind turbine fault monitoring system. This method can deeply identify patterns of family-related defects, perform accurate source tracing and diagnosis in conjunction with the actual operating environment, and has the capability for dynamic iterative updates of the knowledge base.

[0004] Firstly, this application provides a dynamic calibration method for a wind turbine fault monitoring system, the method comprising: Based on historical operating data and fault records of wind turbine units, a fault mode library is constructed, which includes system hierarchy, fault causes and impact chains. The risk priority number of each fault mode is calculated by combining severity, occurrence and undetectability. The fault modes in the fault mode library are classified according to the risk priority number, and the fault modes with the target risk level are selected as the main monitoring fault modes. For the equipment components corresponding to the main monitoring fault mode, deploy an external monitoring system that includes vibration, temperature, acoustic signature and displacement parameters, and construct an error propagation network model between fault characteristics and monitoring parameters; The monitoring data of the external monitoring system is collected in real time, the data features are extracted and input into the error propagation network model, and the comprehensive confidence impact value of each monitoring parameter relative to the main monitoring fault mode is calculated. The monitoring parameters whose comprehensive confidence impact value exceeds the dynamic threshold are defined as parameters to be calibrated. Based on the topological position of the parameters to be calibrated in the error propagation network model and the coupling strength affected by the main monitoring failure mode, a sequence of parameters to be calibrated sorted by calibration priority is generated. The corresponding calibration operation type is matched according to the sequence of parameters to be calibrated, including calling hardware-level calibration, transmission-level calibration or model-level calibration to perform calibration operations on wind turbine components; The overall confidence level impact value is recalculated based on the calibrated monitoring data. If the overall confidence level impact value is lower than the dynamic threshold, the calibration is deemed complete and monitoring data collection continues in real time. If the overall confidence level impact value still exceeds the dynamic threshold, the re-decision process for the main monitoring failure mode is triggered to update the risk level classification.

[0005] Furthermore, The calculation of the risk priority number for each failure mode, combining severity, occurrence, and undetectability, specifically includes: The weighting coefficients for severity, occurrence, and undetectability were determined using the analytic hierarchy process (AHP). The fuzzy comprehensive evaluation method was used to score each fault mode under the above three dimensions. The risk priority number for each failure mode is obtained by weighting and summing the scores with the corresponding weight coefficients.

[0006] Furthermore, The failure modes selected based on the target risk level are designated as the primary monitoring failure modes, specifically including: Set a warning threshold for the risk priority number, and define the failure mode that exceeds the warning threshold as a high-risk failure mode; Based on the historical failure frequency statistics of key components of wind turbines, high-risk failure modes with a failure frequency ranking of a preset number are selected as the main monitoring failure modes.

[0007] Furthermore, The construction of the error propagation network model for fault characteristics and monitoring parameters specifically includes: Using the fault characteristics of the main monitoring fault mode as network nodes and the measurement error of the monitoring parameters as the weight of the connection edges, a directed acyclic graph for error propagation is constructed using the Bayesian network algorithm. The directed acyclic graph of error propagation is trained based on the fault records to determine the conditional probability distribution of each node. The contribution of different monitoring parameters to the expression of fault characteristics is quantified through the conditional probability distribution, and the propagation probability of the error propagation path is determined.

[0008] Furthermore, The calculation of the overall confidence impact value of each monitoring parameter relative to the main monitoring failure mode specifically includes: Based on the error propagation network model, the degree of deviation between the real-time collected values ​​of each monitoring parameter and the preset benchmark value is obtained; The overall confidence level impact value is obtained by weighting and summing the deviation levels based on the topological position weights of each monitoring parameter in the model.

[0009] Furthermore, The generation of the sequence of parameters to be calibrated, sorted by calibration priority, specifically includes: Calculate the betweenness centrality of the parameter to be calibrated in the error propagation network model, wherein the betweenness centrality is used to characterize the criticality of the parameter in the error propagation path; The parameters to be calibrated are initially sorted according to their betweenness centrality from high to low, and the sorting results are corrected by taking into account the coupling strength affected by the main monitoring failure mode. The parameters with higher coupling strength have higher calibration priority, and finally the sequence of parameters to be calibrated is generated.

[0010] Furthermore, The step of matching the corresponding calibration operation type according to the sequence of parameters to be calibrated specifically includes: If the parameter to be calibrated is an abnormal original sensor reading, then hardware-level calibration is performed to trigger sensor zero-point drift correction or sensitivity adjustment. If the parameter to be calibrated is due to data transmission packet loss or signal interference, then a transmission-level calibration is performed, triggering a communication protocol reset or filter algorithm parameter optimization. If the parameter to be calibrated is a feature extraction bias, then a model-level calibration is performed, triggering a structural update or node parameter update of the error propagation network model.

[0011] Secondly, based on the same inventive concept, this application provides a dynamic calibration system for a wind turbine fault monitoring system. The system includes: The module constructs a fault mode library based on the historical operating data and fault records of wind turbine units. This library includes system hierarchy, fault causes, and impact chains. It also calculates the risk priority number of each fault mode by combining severity, occurrence, and undetectability. The classification and screening module classifies the fault modes in the fault mode library according to the risk priority number, and selects the fault modes with the target risk level as the main monitoring fault modes. The deployment module deploys an external monitoring system containing vibration, temperature, acoustic signature, and displacement parameters for the equipment components corresponding to the main monitoring fault mode, and constructs an error propagation network model between fault characteristics and monitoring parameters. The monitoring and calculation module collects monitoring data from the external monitoring system in real time, extracts data features and inputs them into the error propagation network model, and calculates the comprehensive confidence impact value of each monitoring parameter relative to the main monitoring fault mode. The decision production module defines the monitoring parameters whose comprehensive confidence impact value exceeds the dynamic threshold as parameters to be calibrated. Based on the topological position of the parameters to be calibrated in the error propagation network model and the coupling strength affected by the main monitoring failure mode, it generates a sequence of parameters to be calibrated sorted by calibration priority. The calibration execution module matches the corresponding calibration operation type according to the sequence of parameters to be calibrated, including calling hardware-level calibration, transmission-level calibration or model-level calibration to perform calibration operations on wind turbine components; The assessment update module recalculates the overall confidence impact value based on the calibrated monitoring data. If the overall confidence impact value is lower than the dynamic threshold, the calibration is deemed complete and monitoring data collection continues in real time. If the overall confidence impact value still exceeds the dynamic threshold, the re-decision process for the main monitoring failure mode is triggered to update the risk level classification.

[0012] Thirdly, this application also provides an electronic device, including at least one processor and at least one memory electrically connected; The memory is electrically connected to the processor, wherein the memory stores instructions that can be executed by at least one of the processors, the instructions being executed by at least one of the processors to enable at least one of the processors to perform any of the dynamic calibration methods for a wind turbine fault monitoring system as described above.

[0013] Fourthly, this application also provides a computer storage medium, wherein a computer program is stored within the computer-readable storage medium; When the computer program is executed by the processor, it implements any of the dynamic calibration methods for wind turbine fault monitoring systems described above.

[0014] Fifthly, this application also provides a computer program product, which is stored in at least one storage medium; The computer program product includes several instructions to cause at least one electronic device to execute any of the dynamic calibration methods for a wind turbine fault monitoring system as described above.

[0015] Compared with the prior art, this application has the following advantages: 1. By combining the analytic hierarchy process (AHP) and the fuzzy comprehensive evaluation method to calculate the risk priority number of the failure mode, and by comprehensively screening the main monitoring failure mode based on the historical failure frequency, high-risk key monitoring objects can be accurately identified from many failure modes. This allows limited monitoring and calibration resources to be concentrated on the most critical components and failure characteristics, thereby improving monitoring efficiency. 2. An error propagation network model of fault characteristics and monitoring parameters was constructed. Bayesian networks were used to quantify the contribution and propagation probability of parameters. The parameters to be calibrated were identified by calculating the comprehensive confidence influence value. Combined with the calibration priority sequence generated by betweenness centrality and coupling strength, the root cause parameters of monitoring deviations can be accurately located. This overcomes the problems of vague calibration targets and lack of quantitative basis in traditional methods, and realizes efficient source tracing of errors. 3. A hierarchical calibration strategy encompassing hardware, transmission, and model levels was constructed, enabling the matching of optimal calibration methods to different types of monitoring anomalies. Simultaneously, a closed-loop feedback mechanism of "monitoring-calibration-evaluation-decision" was formed through recalculation of calibrated data and comparison with dynamic thresholds. If calibration is ineffective, a re-decision process is triggered, ensuring that the monitoring system can dynamically adapt to changes in unit status and significantly reducing the false alarm rate and missed alarm rate.

[0016] Other features and advantages of this application will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating the dynamic calibration method of the wind turbine fault monitoring system according to an embodiment of this application is shown. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] Figure 1 A dynamic calibration method for a wind turbine fault monitoring system according to an embodiment of this application is shown. Figure 1 As shown in the embodiment of this application, the dynamic calibration method for the wind turbine fault monitoring system includes the following steps: S1, based on the historical operating data and fault records of wind turbine units, constructs a fault mode library that includes system hierarchy, fault causes and impact chains, and calculates the risk priority number of each fault mode by combining severity, occurrence and undetectability. S2, classify the fault modes in the fault mode library according to the risk priority number, and select the fault modes with the target risk level as the main monitoring fault modes. S3, for the equipment components corresponding to the main monitoring fault mode, deploy an external monitoring system that includes vibration, temperature, acoustic signature and displacement parameters, and construct an error propagation network model of fault characteristics and monitoring parameters; S4. Real-time acquisition of monitoring data from the external monitoring system, extraction of data features and input into the error propagation network model, calculation of the comprehensive confidence impact value of each monitoring parameter relative to the main monitoring fault mode; S5, the monitoring parameters whose comprehensive confidence impact value exceeds the dynamic threshold are defined as parameters to be calibrated. Based on the topological position of the parameters to be calibrated in the error propagation network model and the coupling strength affected by the main monitoring fault mode, a sequence of parameters to be calibrated sorted by calibration priority is generated. S6, Match the corresponding calibration operation type according to the sequence of parameters to be calibrated, including calling hardware-level calibration, transmission-level calibration or model-level calibration to perform calibration operation on wind turbine components; S7. Recalculate the overall confidence impact value based on the calibrated monitoring data. If the overall confidence impact value is lower than the dynamic threshold, the calibration is deemed complete and monitoring data continues to be collected in real time. If the overall confidence impact value still exceeds the dynamic threshold, the re-decision process for the main monitoring fault mode is triggered to update the risk level classification.

[0021] In the embodiments of this application, step S1 specifically includes: S11, the weight coefficients of severity, occurrence and undetectability are determined by the analytic hierarchy process. S12, use the fuzzy comprehensive evaluation method to score each fault mode under the above three dimensions; S13, the scoring results are weighted and summed with the corresponding weight coefficients to obtain the risk priority number of each failure mode.

[0022] In the specific implementation process, the hierarchical structure model of risk priority assessment is constructed using the analytic hierarchy process (AHP). First, a judgment matrix is ​​established with the failure mode risk level as the target layer and severity, occurrence, and undetectability as the criteria layer. The three evaluation indicators of severity, occurrence, and undetectability are compared pairwise to construct the judgment matrix. For example, for critical failures that directly affect the safe shutdown of wind turbine units, the severity weight will be given a higher priority. Then, the largest eigenvalue of the judgment matrix and its corresponding eigenvector are calculated, and the consistency is checked by the ratio of the consistency index to the random consistency index. If the consistency ratio is less than 0.1, the judgment matrix is ​​consistent. At this time, the corresponding eigenvector, after normalization, can be used as the weight coefficient vector of severity, occurrence, and undetectability. Fuzzy comprehensive evaluation method is used to address the fuzzy boundaries of evaluation indicators and the uncertainty of expert subjective judgment. First, an evaluation set is established, usually set as five levels: {very low, low, medium, high, very high}, and each level is assigned a corresponding score range (e.g., 1-10 points). For any failure mode in the failure mode library, based on historical failure records and operating conditions, the membership degree is evaluated in three dimensions: severity, occurrence, and undetectability. A fuzzy relation matrix is ​​constructed, which reflects the degree of membership of the failure mode to different evaluation levels under each evaluation indicator. Combining the weight coefficient vector determined in S11 with the fuzzy relation matrix, fuzzy synthesis operation is performed to obtain the accurate score value of each failure mode after defuzzification in the three dimensions, thus realizing the scientific quantification of qualitative indicators. The final risk quantification index is calculated based on the set of weighted coefficients obtained in S11 and the score results of each dimension obtained in S12. A risk priority number calculation model is constructed. For each failure mode, its severity score, occurrence score, and undetectability score are multiplied by their corresponding weighted coefficients and then summed. The calculation formula is as follows: Risk Priority Number = (Severity Score × Severity Weight) + (Occurrence Score × Occurrence Weight) + (Undetectability Score × Undetectability Weight); The risk priority number of each failure mode is calculated using the linear weighted model described above. This number comprehensively reflects the potential threat level of the failure mode in the operation of the wind turbine. The higher the number, the more dangerous the failure mode is, thus providing objective and quantitative data support for screening the main monitoring failure modes in the subsequent S2 step.

[0023] In the embodiments of this application, step S2 specifically includes: S21, Set a warning threshold for the risk priority number, and define a fault mode whose risk priority number exceeds the warning threshold as a high-risk fault mode; S22. Based on the historical failure frequency statistics of key components of the wind turbine, high-risk failure modes with a failure frequency ranking of a preset number are selected as the main monitoring failure modes.

[0024] In the specific implementation process, high-risk items are identified from numerous failure modes. First, based on the historical maintenance costs, safe operation standards, and industry regulations of wind turbines, statistical analysis methods (such as Pareto analysis or quantile method based on normal distribution) are used to set a warning threshold for the risk priority number. For example, the warning threshold can be set as a specific multiple of the sum of the average and standard deviation of the risk priority numbers of all failure modes, or as a critical value that includes the top 20% of failure modes in the high-risk range. Subsequently, the risk priority number calculated for each failure mode in the failure mode library is compared with the warning threshold one by one. When the risk priority number of a failure mode exceeds the warning threshold, it indicates that the failure mode has an extremely high potential hazard or probability of occurrence, and it is defined as a high-risk failure mode and included in the key monitoring candidate set. Conversely, for failure modes whose risk priority number does not exceed the warning threshold, they are temporarily marked as low-risk or medium-risk modes and are not included in the current priority monitoring scope, thereby achieving the initial screening and focusing of monitoring resources. Historical fault frequency statistics of key components of wind turbines (such as gearboxes, generators, and blades) over a set time period (e.g., the past three or five years) are retrieved to construct a fault frequency ranking. Subsequently, the high-risk fault modes selected in S21 are mapped and matched with this fault frequency ranking. To exclude occasional but serious fault interference with extremely low actual occurrence probability, a preset number of selection rules is set. For example, high-risk fault modes ranked in the top N (e.g., the top 10) or in the top M% (e.g., the top 20%) are selected. Only fault modes that simultaneously meet the dual conditions of "risk priority number exceeding the warning threshold" and "historical fault frequency ranking in a preset high position" are finally determined as the main monitoring fault modes. This selection strategy effectively eliminates low-frequency, high-risk extreme operating condition interference, enabling the subsequently constructed monitoring system and calibration model to accurately focus on the core fault characteristics of frequent and high-risk faults, thereby improving the engineering practical value of the dynamic calibration method.

[0025] In the embodiments of this application, step S3 specifically includes: S31, using the fault characteristics of the main monitoring fault mode as network nodes and the measurement error of the monitoring parameters as the weight of the connection edge, a directed acyclic graph of error propagation is constructed using the Bayesian network algorithm. S32, based on the fault records, train the directed acyclic graph of error propagation, determine the conditional probability distribution of each node, quantify the contribution of different monitoring parameters to the expression of fault characteristics through the conditional probability distribution, and determine the propagation probability of the error propagation path.

[0026] In the specific implementation process, the main monitoring fault modes selected are subjected to in-depth mechanism analysis, and physical features that can characterize the essence of the fault (such as the impact pulse characteristics caused by gearbox pitting and the harmonic distortion characteristics caused by generator eccentricity) are extracted as the parent node (root node) of the Bayesian network. Secondly, the specific monitoring parameters obtained by the external monitoring system (such as vibration acceleration, average temperature, acoustic energy band, etc.) are used as child nodes (leaf nodes). Subsequently, based on the physical mechanism of the fault occurrence and the signal propagation path, directed edges from the fault feature nodes to the monitoring parameter nodes are established to form the structural skeleton of the directed acyclic graph of error propagation. Based on this, the concept of measurement error is introduced. By utilizing the degree of deviation of the monitoring parameters from the theoretical fault characteristic values ​​in historical data, the error variance of the monitoring parameters affected by the fault characteristics is calculated and used as the initial weight of the connecting edge. This results in the construction of a network model that not only contains causal logic but also contains error propagation characteristics. Fault sample data related to the main monitoring fault mode during the historical operation of wind turbine units were collected to construct a training dataset. Maximum likelihood estimation or Bayesian estimation was used to learn the parameters of each node in the directed acyclic graph of error propagation. The conditional probability distribution of child nodes (monitoring parameters) in different states (e.g., normal, minor fault, severe fault) of a given parent node (fault feature) was calculated. The probability values ​​in the conditional probability distribution table were used to quantify the sensitivity and contribution of different monitoring parameters to capturing specific fault features (e.g., vibration parameters may contribute more to bearing fault features than temperature parameters). Simultaneously, based on the conditional dependencies between nodes in the network structure, the joint probability of reaching the monitoring parameter node from the fault feature node via intermediate nodes was calculated, thus determining the propagation probability of each error propagation path. This provides the basic parameters for probabilistic inference in subsequent calculations of the comprehensive confidence impact value.

[0027] In the embodiments of this application, step S4 specifically includes: S41, Based on the error propagation network model, obtain the degree of deviation between the real-time collected values ​​of each monitoring parameter and the preset benchmark value; S42, combining the topological position weights of each monitoring parameter in the model, the degree of deviation is weighted and summed to obtain the comprehensive confidence level influence value.

[0028] In the specific implementation process, the degree of deviation of the quantified monitoring data from the normal operating conditions is used as a basic indicator to measure whether the monitoring system itself has drifted or whether fault characteristics have appeared. First, the baseline values ​​of each monitoring parameter under the normal operating conditions of the wind turbine are preset in the error propagation network model. The baseline values ​​are usually the average values ​​obtained by statistically analyzing historical health data or the standard values ​​after standardization. Then, the current readings of vibration, temperature, acoustic signature and displacement sensors in the external monitoring system are collected in real time, and data cleaning and feature extraction are performed (such as extracting the time domain statistical features or frequency domain features of the vibration signal). The extracted real-time feature values ​​are compared with the corresponding preset baseline values, and the degree of deviation between the two is calculated. The degree of deviation can be calculated using methods such as relative deviation rate, standardized Euclidean distance or Mahalanobis distance. For example, for temperature parameters, the normalized offset relative to the baseline value can be calculated, and for vibration parameters, the difference between the signal energy spectrum and the baseline spectrum can be calculated, thereby obtaining a quantitative deviation value that can reflect the current reliability status of each monitoring parameter. By using a weighted fusion approach, the dispersed single-parameter biases are transformed into a global confidence index for the main monitoring fault mode. First, based on the topology of the error propagation network model, the connection paths between each monitoring parameter node and the main monitoring fault mode node are analyzed. Utilizing the structural characteristics of Bayesian networks, the posterior probability contribution rate or mutual information value of each monitoring parameter node relative to the main monitoring fault mode node is calculated to determine the topological position weight of each monitoring parameter in the model. This weight reflects the criticality of the parameter in characterizing a specific fault feature. Parameters located on critical propagation paths or strongly coupled with the fault mode will be assigned higher weight coefficients. Subsequently, the bias degree of each monitoring parameter calculated in S41 is weighted and summed with the corresponding topological position weight. The calculation formula can be expressed as: Comprehensive confidence impact value = Σ(deviation degree of parameter i × topological position weight of parameter i). The final comprehensive confidence impact value is a dimensionless scalar. The higher the value, the further the current monitoring data deviates from the normal mode, the lower the reliability of the monitoring system data or the greater the possibility of failure. This provides the core criterion for subsequent determination of whether to trigger the calibration process.

[0029] In the embodiments of this application, step S5 specifically includes: S51, Calculate the betweenness centrality of the parameter to be calibrated in the error propagation network model, wherein the betweenness centrality is used to characterize the criticality of the parameter in the error propagation path; S52, the parameters to be calibrated are initially sorted according to the order of betweenness centrality from high to low, and the sorting results are corrected in combination with the coupling strength affected by the main monitoring failure mode. The parameters with higher coupling strength have higher calibration priority, and finally the sequence of parameters to be calibrated is generated.

[0030] In the specific implementation process, the global influence of a single monitoring parameter in the entire error propagation network is quantified by using complex network topology indicators. First, based on the constructed directed acyclic graph of error propagation, the shortest paths between all node pairs in the network are traversed. For each node defined as a parameter to be calibrated, the proportion of the number of paths passing through that node in all shortest paths is counted to the total number of shortest paths. The calculation formula is logically as follows: betweenness centrality is equal to the sum of the number of shortest paths passing through that node divided by the sum of the number of shortest paths between all node pairs. In wind turbine monitoring scenarios, monitoring parameters with high betweenness centrality are usually located at the throat of fault feature transmission. For example, the vibration acceleration parameter of the gearbox is often the bridge node connecting early wear fault features with subsequent temperature and sound pattern changes. The higher the value of this parameter, the greater the information flow it carries in the error propagation network. Once this parameter has a measurement error, the scope and depth of its misleading influence on the overall fault diagnosis conclusion will be wider, thus making calibration more urgent. By introducing coupling factors at the fault mechanism level, the ranking results based solely on network topology are optimized and corrected to ensure that the calibration order conforms to actual physical conditions. First, based on the betweenness centrality values ​​calculated by S51, all parameters to be calibrated are arranged in descending order to generate a preliminary ranking table. Then, a coupling strength index is introduced, which reflects the degree of physical correlation between the monitoring parameters and specific main monitoring fault modes. This index can be quantified through the conditional probability distribution or mutual information entropy value in the error propagation network model (for example, the coupling strength of bearing temperature parameters to bearing wear faults is usually higher than that of ambient temperature parameters). During the correction process, the preliminary ranking table is traversed. If the coupling strength of a parameter is higher than the preset strong coupling threshold, its ranking position is moved forward. In particular, when the betweenness centrality values ​​of two parameters are similar, the order is directly determined by the magnitude of the coupling strength, with parameters with greater coupling strength having higher priority. Finally, after weighted correction by dual indices, the final sequence of parameters to be calibrated from the 1st to the Nth position is generated. This sequence considers both the global influence of the parameters in the network and their local sensitivity to specific fault modes, providing a scientific work list for subsequent accurate execution of graded calibration.

[0031] In the embodiments of this application, step S6 specifically includes: S61, if the parameter to be calibrated is an abnormal original sensor reading, then hardware-level calibration is matched to trigger sensor zero-point drift correction or sensitivity adjustment. S62, if the parameter to be calibrated is due to data transmission packet loss or signal interference, then perform a transmission-level calibration to trigger a communication protocol reset or filter algorithm parameter optimization. S63, if the parameter to be calibrated is a feature extraction bias, then match the model-level calibration and trigger the structural update or node parameter update of the error propagation network model.

[0032] In the specific implementation process, calibration is performed on the physical deviations of the sensing layer of the monitoring system. First, the specific source of abnormal parameters is identified through the error propagation network model, and its abnormal attribute is determined to be the physical performance drift of the sensor itself. For example, if the vibration sensor outputs a non-zero baseline value when the wind turbine is stopped, or the temperature sensor reading shows a non-linear deviation with the change of ambient temperature, it is determined that the original reading of the sensor is abnormal. For the zero-point drift problem, a calibration command is automatically sent to the data acquisition unit of the external monitoring system to execute the zero-point calibration program, forcing the static output value to zero or correcting it to the standard offset. For the sensitivity decrease or gain abnormality problem, the built-in standard excitation source is called or the historical calibration curve is compared to adjust the gain coefficient of the signal amplifier or the conversion coefficient of the analog-to-digital converter. This process realizes online soft calibration of the sensing hardware device or prompts the operation and maintenance personnel to remotely reset the parameters, ensuring the physical authenticity of the source data. When the monitoring system detects discontinuous data packet sequences, increased bit error rate, or high-frequency noise interference superimposed on the signal, it determines that the parameter to be calibrated belongs to the transmission layer anomaly. For data transmission packet loss or communication protocol misalignment, it automatically triggers the communication protocol reset mechanism, re-establishes the Modbus or OPC UA connection between the acquisition front-end and the host computer, resets the data transmission baud rate and parity bit, and restores the stability of the data link. For signal interference problems, it starts the transmission-level calibration module, uses digital signal processing technology to analyze the real-time data stream, and adaptively adjusts the threshold parameters of the bandpass filter, Kalman filter, or wavelet denoising algorithm according to the frequency characteristics of the interference signal (such as power frequency interference or electromagnetic pulse) to filter out noise components and restore the true signal waveform to the greatest extent, thereby eliminating the pseudo-errors introduced by the transmission link.

[0033] When the original data and transmission link are normal, but the calculated fault feature values ​​do not match the actual operating conditions (for example, the feature extraction algorithm fails to effectively track the changes in the variable speed and pitch conditions of the wind turbine), it is determined to be a feature extraction deviation, and model-level calibration is triggered: On the one hand, if the deviation is due to the mismatch of feature expression methods, the node parameters of the error propagation network model are updated, and the conditional probability distribution table of the Bayesian network nodes is retrained using the latest historical fault data to optimize the feature threshold boundary; on the other hand, if the deviation is due to the original network structure being unable to adapt to the new fault propagation mechanism, a structural update is performed, and the topology of the directed acyclic graph of error propagation is reconstructed by adding or deleting directed edges and adding or removing intermediate latent variables. This process realizes the adaptive evolution of the monitoring and diagnostic model, ensuring that the model can continuously and accurately characterize the fault features of the wind turbine under complex operating conditions.

[0034] In the embodiments of this application, step S7 specifically includes: After completing the calibration operation in step S6, the calibration process is not exited directly. Instead, the effect evaluation stage is immediately entered. Real-time data from the external monitoring system is collected again. At this time, the data stream has been processed by hardware correction, transmission filtering or model update. Using the same calculation logic as step S4, the calibrated monitoring data is input into the error propagation network model to recalculate the comprehensive confidence impact value of each monitoring parameter relative to the main monitoring fault mode, and obtain the post-calibration confidence index. Subsequently, the calibrated confidence index is compared a second time with the set dynamic threshold, and two cases are handled accordingly: In the first scenario, the calibration is deemed successful: if the recalculated comprehensive confidence impact value is reduced to below the dynamic threshold, it indicates that the calibration operation has effectively eliminated the deviation of the monitoring parameters and the status of the monitoring system has returned to normal. At this time, the calibration task is deemed complete, the system status log is updated, the current monitoring data is marked as reliable, the current calibration cycle is terminated, and the external monitoring system is controlled to continue to perform the regular real-time acquisition and monitoring tasks to maintain continuous awareness of the wind turbine's operating status. The second scenario involves calibration failure and strategy updates: If the recalculated comprehensive confidence impact value still exceeds the dynamic threshold, it reveals a deeper problem. This indicates that the current calibration operation failed to eliminate the anomaly, and it is highly likely that the actual operating state of the wind turbine has changed substantially (e.g., the fault mode has evolved). The original main monitoring fault mode definition or risk level classification is no longer applicable to the current operating conditions. In this case, it is determined that the parameter calibration has failed, triggering a re-decision process for the main monitoring fault mode. This process will go back to the aforementioned steps, re-analyze the latest historical operating data and fault records of the wind turbine, re-evaluate the risk priority of each fault mode, and update the main monitoring fault mode list and the corresponding error propagation network model parameters. This enables the monitoring strategy to dynamically iterate and upgrade with the evolution of operating conditions, ensuring that the monitoring system always remains in the optimal perception state.

[0035] Based on the same inventive concept, this application also provides a dynamic calibration system for a wind turbine fault monitoring system corresponding to the above method; The system includes: The module constructs a fault mode library based on the historical operating data and fault records of wind turbine units. This library includes system hierarchy, fault causes, and impact chains. It also calculates the risk priority number of each fault mode by combining severity, occurrence, and undetectability. The classification and screening module classifies the fault modes in the fault mode library according to the risk priority number, and selects the fault modes with the target risk level as the main monitoring fault modes. The deployment module deploys an external monitoring system containing vibration, temperature, acoustic signature, and displacement parameters for the equipment components corresponding to the main monitoring fault mode, and constructs an error propagation network model between fault characteristics and monitoring parameters. The monitoring and calculation module collects monitoring data from the external monitoring system in real time, extracts data features and inputs them into the error propagation network model, and calculates the comprehensive confidence impact value of each monitoring parameter relative to the main monitoring fault mode. The decision production module defines the monitoring parameters whose comprehensive confidence impact value exceeds the dynamic threshold as parameters to be calibrated. Based on the topological position of the parameters to be calibrated in the error propagation network model and the coupling strength affected by the main monitoring failure mode, it generates a sequence of parameters to be calibrated sorted by calibration priority. The calibration execution module matches the corresponding calibration operation type according to the sequence of parameters to be calibrated, including calling hardware-level calibration, transmission-level calibration or model-level calibration to perform calibration operations on wind turbine components; The assessment update module recalculates the overall confidence impact value based on the calibrated monitoring data. If the overall confidence impact value is lower than the dynamic threshold, the calibration is deemed complete and monitoring data collection continues in real time. If the overall confidence impact value still exceeds the dynamic threshold, the re-decision process for the main monitoring failure mode is triggered to update the risk level classification.

[0036] Based on the same inventive concept, this application also provides an electronic device. The electronic device of this application embodiment includes at least one processor and at least one memory electrically connected to the processor. The memory is electrically connected to the processor, wherein the memory stores instructions executable by the at least one processor. These instructions are executed by the at least one processor to enable the at least one processor to perform the dynamic calibration method of the wind turbine fault monitoring system as described above.

[0037] It should be noted that the electrical connections between the various units mentioned above do not necessarily represent the connections between lines. Any indirect connection method can be applied to the embodiments of this application as long as it achieves the purpose of this application.

[0038] Based on the same inventive concept, this application also provides a computer storage medium, wherein a computer program is stored in the computer-readable storage medium, and when the computer program is executed by a processor, it implements the dynamic calibration method of the wind turbine fault monitoring system as described above.

[0039] Based on the same inventive concept, this application also provides a computer program product, which is stored in at least one storage medium; the computer program product includes several instructions to cause at least one computer device to execute the dynamic calibration method of the wind turbine fault monitoring system as described above.

[0040] Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A dynamic calibration method for a wind turbine fault monitoring system, characterized in that, Build a failure mode library and calculate the risk priority number of each failure mode by combining severity, occurrence and undetectability; The fault modes in the fault mode library are classified according to the risk priority number, and the fault modes with the target risk level are selected as the main monitoring fault modes. For the main monitoring failure mode, deploy an external monitoring system and construct an error propagation network model of failure characteristics and monitoring parameters; The monitoring data of the external monitoring system is collected in real time, input into the error propagation network model, and the comprehensive confidence impact value of each monitoring parameter relative to the main monitoring fault mode is calculated. The parameters to be calibrated are defined based on the comprehensive confidence level influence value, and a sequence of parameters to be calibrated, sorted by calibration priority, is generated based on the parameters to be calibrated. The calibration operation is performed by matching the corresponding calibration operation type according to the sequence of parameters to be calibrated. Based on the calibrated monitoring data, the comprehensive confidence level impact value is recalculated and combined with the dynamic threshold. Then, the calibration is determined to be complete and real-time monitoring data collection continues, or the re-decision process of the main monitoring failure mode is triggered to update the risk level classification.

2. The method according to claim 1, characterized in that, The calculation of the risk priority number for each failure mode, combining severity, occurrence, and undetectability, specifically includes: The weighting coefficients for severity, occurrence, and undetectability were determined using the analytic hierarchy process (AHP). The fuzzy comprehensive evaluation method was used to score each fault mode across three dimensions. The risk priority number for each failure mode is obtained by weighting and summing the scores with the corresponding weight coefficients.

3. The method according to claim 2, characterized in that, The failure modes selected based on the target risk level are designated as the primary monitoring failure modes, specifically including: Set a warning threshold for the risk priority number, and define the failure mode that exceeds the warning threshold as a high-risk failure mode; Based on the historical failure frequency statistics of key components of wind turbines, high-risk failure modes with a failure frequency ranking of a preset number are selected as the main monitoring failure modes.

4. The method according to claim 3, characterized in that, The construction of the error propagation network model for fault characteristics and monitoring parameters specifically includes: Using the fault characteristics of the main monitoring fault mode as network nodes and the measurement error of the monitoring parameters as the weight of the connection edges, a directed acyclic graph for error propagation is constructed using the Bayesian network algorithm. The directed acyclic graph of error propagation is trained based on the fault records to determine the conditional probability distribution of each node. The contribution of different monitoring parameters to the expression of fault characteristics is quantified through the conditional probability distribution, and the propagation probability of the error propagation path is determined.

5. The method according to claim 4, characterized in that, The calculation of the overall confidence impact value of each monitoring parameter relative to the main monitoring failure mode specifically includes: Based on the error propagation network model, the degree of deviation between the real-time collected values ​​of each monitoring parameter and the preset benchmark value is obtained; The overall confidence level impact value is obtained by weighting and summing the deviation levels based on the topological position weights of each monitoring parameter in the model.

6. The method according to claim 5, characterized in that, The step of matching the corresponding calibration operation type according to the sequence of parameters to be calibrated specifically includes: If the parameter to be calibrated is an abnormal original sensor reading, then hardware-level calibration is performed to trigger sensor zero-point drift correction or sensitivity adjustment. If the parameter to be calibrated is due to data transmission packet loss or signal interference, then a transmission-level calibration is performed, triggering a communication protocol reset or filter algorithm parameter optimization. If the parameter to be calibrated is a feature extraction bias, then a model-level calibration is performed, triggering a structural update or node parameter update of the error propagation network model.

7. A dynamic calibration system for a wind turbine fault monitoring system, characterized in that, The system includes: The module builds a failure mode library and calculates the risk priority number of each failure mode by combining severity, occurrence, and undetectability. The classification and screening module classifies the fault modes in the fault mode library according to the risk priority number, and selects the fault modes with the target risk level as the main monitoring fault modes. Deployment and construction modules are used to deploy an external monitoring system for the main monitoring fault mode and to construct an error propagation network model of fault characteristics and monitoring parameters; The monitoring and calculation module collects monitoring data from the external monitoring system in real time, inputs it into the error propagation network model, and calculates the comprehensive confidence impact value of each monitoring parameter relative to the main monitoring fault mode. The decision-making production module defines the parameters to be calibrated based on the comprehensive confidence level influence value, and generates a sequence of parameters to be calibrated sorted by calibration priority based on the parameters to be calibrated. The calibration execution module matches the corresponding calibration operation type according to the sequence of parameters to be calibrated and executes the calibration operation. The assessment update module recalculates the overall confidence impact value based on the calibrated monitoring data and combines it with the dynamic threshold. If so, it determines that the calibration is complete and continues to collect monitoring data in real time, or triggers a re-decision process for the main monitoring failure mode to update the risk level classification.

8. An electronic device, characterized in that, Includes at least one processor and at least one memory electrically connected; The memory is electrically connected to the processor, wherein the memory stores instructions executable by at least one of the processors, the instructions being executed by at least one of the processors to enable at least one of the processors to perform the dynamic calibration method of the wind turbine fault monitoring system as described in any one of claims 1-6.

9. A computer storage medium, characterized in that, The computer-readable storage medium stores a computer program. When the computer program is executed by the processor, it implements the dynamic calibration method of the wind turbine fault monitoring system according to any one of claims 1-6.

10. A computer program product, characterized in that, The computer program product is stored in at least one storage medium; The computer program product includes several instructions for causing at least one electronic device to execute the dynamic calibration method of the wind turbine fault monitoring system according to any one of claims 1-6.

Images