Fan health assessment method, apparatus, and device

Abstract

The application discloses a kind of fan health assessment method, device and equipment, related to wind turbine fault diagnosis technical field, the method comprises: obtaining the vibration operating data of wind turbine;Determine the feature set related to the failure of each key component in wind turbine based on vibration operating data;Determine the degradation index of each original feature in feature set, and the degradation index is aggregated nonlinearly, obtain the degradation value of each key component of wind turbine;Degradation index is calculated based on health benchmark value, combined with rounding operator and zero value truncation operation;According to the degradation value of each key component, the health status of wind turbine is evaluated using dynamic time window driven multivariate parameter degradation convergence evaluation strategy, and the health assessment result is obtained.The application can improve the health status evaluation accuracy of wind turbine, and provide scientific support for fine operation management and safety production target realization of wind farm.

Description

Technical Field

[0001] This application relates to the field of wind turbine fault diagnosis technology, and in particular to a wind turbine health assessment method, apparatus and equipment. Background Technology

[0002] As a crucial component of clean energy, wind power has experienced rapid and continuous growth in installed capacity. The large-scale operation of wind turbines places higher demands on turbine health monitoring, operation and maintenance management, and fault prediction. Wind turbines are typically deployed in complex outdoor environments. During long-term operation, key components are constantly subjected to fluctuations in operating conditions and external interference. The vibration signals generated during operation are non-stationary, characterized by strong noise and drastic fluctuations in operating conditions, making them prone to performance degradation and even failure. Without effective health monitoring and fault prediction methods, not only can unplanned turbine shutdowns and equipment damage occur, but also maintenance costs surge and power generation revenue is lost, severely impacting the operational efficiency and safe production of wind farms. Therefore, research on the health status assessment and fault diagnosis of wind turbines is particularly important.

[0003] Currently, in wind farm operation and management, relevant fault monitoring and early warning technologies adopt a threshold-based early warning triggering method, that is, an alarm is triggered when the degradation value exceeds a preset threshold. However, in actual operation and monitoring scenarios, due to the complexity of the industrial site environment, the collected data is often accompanied by random fluctuations or instantaneous impacts, resulting in poor anti-interference capabilities. The traditional sliding time window method only treats the data of each day within the window with "equal weight," which cannot highlight the real-time degradation trend. This leads to "alarm lag" or "false alarm tailing" problems in operation and monitoring, affecting the timeliness of operation and maintenance decisions. Furthermore, this early warning method is relatively simple, only able to determine "normal" or "faulty" states, and cannot provide multi-dimensional status basis for wind farm operation and maintenance management. This results in poor accuracy in assessing the health status of wind turbine units, making it difficult to support the full life cycle operation and management and safe production goals of wind farms. Summary of the Invention

[0004] The purpose of this application is to provide a method, apparatus, and equipment for wind turbine health assessment, in order to solve the technical problem that the accuracy of health status assessment of wind turbine units is poor, making it difficult to support the full life cycle operation management and safe production goals of wind farms.

[0005] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a method for assessing the health of wind turbines, including: Obtain vibration operation data of wind turbine units; Based on the vibration operation data, a set of features related to the failures of each key component in the wind turbine is determined; each set of features includes at least two original features. The degradation index of each original feature in the feature set is determined, and the degradation index is nonlinearly aggregated to obtain the degradation value of each key component of the wind turbine. The degradation value is used to characterize the degree of failure degradation of the key component that deviates from the healthy operating state. The degradation index is calculated based on the health benchmark value, combined with the rounding up operator and the zero-value truncation operation. The health benchmark value is the median of the feature value sequence of each original feature in a preset time period. Based on the degradation values ​​of each key component, a dynamic time window-driven multivariate parameter degradation convergence assessment strategy is used to evaluate the health status of the wind turbine and obtain the health assessment results. The multivariate parameter degradation convergence assessment strategy includes: daily degradation classification mapping, daily status aggregation, time weighting and proportion calculation, and comprehensive proportion determination operation.

[0006] Secondly, this application provides a wind turbine health assessment device, which includes: The acquisition module is used to acquire vibration operation data of the wind turbine. The feature determination module is used to determine, based on the vibration operation data, a set of features related to the faults of various key components in the wind turbine; each feature set includes at least two original features; The degradation index calculation module is used to determine the degradation index of each original feature in the feature set, and to obtain the degradation value of each key component of the wind turbine through nonlinear aggregation of the degradation index; the degradation value is used to characterize the degree of failure degradation of the key component deviating from the healthy operating state; the degradation index is calculated based on the health benchmark value, combined with the rounding up operator and the zero-value truncation operation, and the health benchmark value is the median of the feature value sequence of each original feature in a preset time period; The health assessment module is used to assess the health status of the wind turbine based on the degradation values ​​of each key component, using a dynamic time window-driven multivariate parameter degradation convergence assessment strategy, and obtain the health assessment results. The multivariate parameter degradation convergence assessment strategy includes: daily degradation classification mapping, daily status aggregation, time weighting and proportion calculation, and comprehensive proportion determination operation.

[0007] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the wind turbine health assessment method described above.

[0008] According to the specific embodiments provided in this application, the following technical effects are disclosed: This application provides a wind turbine health assessment method, apparatus, and equipment. Compared with existing technologies, this solution extracts a multi-dimensional feature set related to the failure of each key component based on acquired vibration operation data. The feature set corresponding to each key component covers at least two original features, providing a comprehensive data foundation for subsequent wind turbine health monitoring and operation and maintenance decisions. The median of the feature value sequence with a preset time period for each original feature is used as the health benchmark value. Utilizing the median's resistance to outliers and random fluctuations, instantaneous impact interference from the industrial site is filtered out at its source. Simultaneously, the degradation index is calculated by combining the rounding up operator and zero-value truncation operation. This not only focuses on the true degradation direction and eliminates normal fluctuation interference under healthy conditions through zero-value truncation, but also achieves discretization and grading through the rounding up operation, effectively suppressing random fluctuations and instantaneous impacts, and improving... To enhance the robustness of degradation quantification, degradation values ​​for each key component are obtained through nonlinear aggregation, avoiding the abnormal dilution and artificial inflatation issues of linear summation, and accurately characterizing the degree of component degradation. Finally, a dynamic time window-driven multi-parameter degradation convergence assessment strategy is adopted, which includes daily degradation classification mapping, daily state aggregation, time weighting and proportion calculation, and comprehensive proportion judgment. Daily state aggregation smooths short-term interference, and time weighting highlights recent real-time degradation trends, solving the alarm lag and false alarm tailing problems of traditional equal-weight time windows in wind farm operation supervision. At the same time, multi-level classification judgment achieves accurate multi-state assessment, replacing the traditional two-state judgment method. This provides multi-dimensional state data for wind farm operation and maintenance management, significantly improving the accuracy and reliability of wind turbine health status assessment, and providing scientific support for the refined operation management and safety production goals of wind farms. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a schematic diagram of the application environment for the wind turbine health assessment method provided in one embodiment of this application. Figure 2 A schematic flowchart of a wind turbine health assessment method provided in an embodiment of this application; Figure 3 A schematic flowchart illustrating a method for determining basic features and component orientation features according to an embodiment of this application; Figure 4 A flowchart illustrating a wind turbine health assessment method provided in another embodiment of this application; Figure 5A schematic flowchart of a method for obtaining the degradation index of each original feature according to an embodiment of this application; Figure 6 A flowchart illustrating a method for calculating the degradation value of a part according to an embodiment of this application; Figure 7 A schematic diagram illustrating the parameter adaptive update process provided in one embodiment of this application; Figure 8 This is a schematic diagram of a fault decision-making process provided in an embodiment of this application; Figure 9 This is a schematic diagram of the health status evolution simulation of a wind turbine component provided in an embodiment of this application; Figure 10 This is a schematic diagram of real-time wind turbine data provided in an embodiment of this application; Figure 11 This application provides a wind turbine historical data and waveform spectrum diagram as an embodiment. Figure 12 This is a schematic diagram of the functional modules of a wind turbine health assessment device provided in an embodiment of this application; Figure 13 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0011] 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, and 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.

[0012] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0013] Existing wind turbine health assessment methods in related technologies still have many shortcomings in practical engineering applications, making it difficult to meet the high requirements of the wind power industry for predictive maintenance. The specific defects are as follows: First, the establishment and updating of threshold parameters lack robustness and cannot meet the long-term, refined health supervision and management needs of wind farms. Traditional feature algorithms often use arithmetic mean combined with standard deviation to set static or dynamic thresholds. However, frequent random shocks during wind turbine operation can cause random outliers in the monitoring data, and the average value is easily affected by such abnormal fluctuations, leading to a decrease in the reference value of the threshold. In addition, the adaptive update mechanism for dynamic thresholds will include the degradation data of the unit in the gradual failure stage in the baseline calculation, causing the health baseline to gradually rise as the equipment deteriorates, ultimately masking the true fault characteristics, leading to missed fault warnings, and seriously interfering with the routine health supervision and early fault prediction of wind farms.

[0014] Secondly, the feature quantification capabilities are insufficient, and the ability to characterize nonlinear anomalies is lacking, failing to provide accurate data support for wind power operation and maintenance decisions. Existing feature extraction technologies struggle to effectively extract early, weak impact features of components such as bearings and gears under strong background noise. Furthermore, the quantification process often uses simple physical quantities, failing to effectively filter out normal statistical fluctuations and distinguish the differences in the levels of equipment degradation at different stages. In terms of data aggregation, traditional methods often employ linear summation, which is prone to two extreme problems: first, severely localized scores of individual core features are "diluted" and averaged out by other normal low-scoring features, resulting in anomalies not being identified in a timely manner; second, the accumulation of small fluctuations across multiple dimensions produces "artificially high scores," making it difficult to accurately reflect the true health status of components and the entire machine, leading to distorted operation and maintenance judgments.

[0015] Third, the fault diagnosis logic has significant shortcomings, with weak anti-interference capabilities and ambiguous status classification, failing to support the management needs of wind farms for tiered predictive maintenance and optimized allocation of operation and maintenance resources. Existing diagnostic methods mostly employ fixed-threshold alarm logic, directly triggering an alarm when the monitored value exceeds a set threshold. This approach is poorly resistant to complex interference in industrial environments, and random fluctuations and instantaneous impacts in sensor data easily lead to false alarms. While some methods introduce sliding time window statistics to smooth noise, the equal weighting of all historical data within the window fails to highlight the current real-time degradation trend of the equipment, easily causing "alarm lag" in diagnostic results and creating "false alarm tails" due to the continuous influence of historical abnormal data. Furthermore, existing technologies can only make binary judgments of equipment status as "normal" or "faulty," failing to provide refined decision-making basis for predictive maintenance. They are unsuitable for the refined operation and maintenance management scenarios of the wind power industry, such as tiered early warning, priority maintenance, and spare parts scheduling, and are ill-suited to the actual needs of the wind power industry for optimized allocation of operation and maintenance resources.

[0016] In summary, current wind power health assessment technologies generally suffer from unstable baseline parameters, inaccurate degradation quantification, and imprecise diagnostic decisions, severely hindering the improvement of intelligent monitoring, fault prediction, and refined operation and maintenance management of wind farms. Therefore, there is an urgent need for a diagnostic method that can effectively offset operating condition disturbances, achieve nonlinear and sensitive evaluation, possess a time-weighted mechanism to emphasize recent degradation trends, and realize accurate mapping of multi-level states to meet the practical application needs of wind power operation and management.

[0017] To address the aforementioned shortcomings, this application provides a wind turbine health assessment method. Compared to existing technologies, this method extracts a multi-dimensional feature set related to the failures of each key component based on acquired vibration operation data. Each key component's feature set covers at least two original features, providing a comprehensive data foundation for subsequent wind turbine health assessments. Furthermore, the median of the preset time-period feature value sequence of each original feature is used as the health benchmark value. Leveraging the median's resistance to outliers and random fluctuations, instantaneous impact interference from the industrial site is filtered out at its source. Simultaneously, the degradation index is calculated by combining the rounding up operator and zero-value truncation operation. This not only focuses on the true degradation direction and eliminates normal fluctuation interference under healthy conditions through zero-value truncation, but also achieves discretization and grading through the rounding up operation, effectively suppressing random fluctuations and instantaneous impacts, and improving the robustness of degradation quantification. The degradation values ​​of key components are obtained through nonlinear aggregation, avoiding the abnormal dilution and artificial inflatation problems of linear summation, and accurately characterizing the degree of component degradation. Finally, a dynamic time window-driven multi-parameter degradation convergence assessment strategy is adopted, which includes daily degradation classification mapping, daily state aggregation, time weighting and proportion calculation, and comprehensive proportion judgment. Daily state aggregation smooths short-term interference, and time weighting highlights recent real-time degradation trends, solving the alarm lag and false alarm tailing problems of equal weight windows. At the same time, multi-level classification judgment achieves accurate multi-state assessment, replacing the traditional two-state judgment method. It can provide multi-dimensional state data for wind farm operation and maintenance management, greatly improve the accuracy and reliability of wind turbine health status assessment, and provide scientific support for the refined operation management and safe production goals of wind farms.

[0018] The wind turbine health assessment method provided in this application embodiment can be applied to, for example... Figure 1The application environment of the wind turbine health assessment method shown is as follows. This application environment includes a terminal 102, a server 104, and a data storage system. The terminal 102 communicates with the server 104 via a network. The data storage system can store the acquired vibration operation data of the wind turbine. The data storage system can be set up independently, integrated into the server 104, or placed in the cloud or on another server. After acquiring the vibration operation data of the wind turbine, the terminal 102 can send it to the server 104. After acquiring the vibration operation data of the wind turbine, the server 104 performs fault feature extraction, degradation index calculation, and health assessment operations to obtain the health assessment result. Furthermore, in some embodiments, the wind turbine health assessment method can also be implemented independently by the server 104 or the terminal 102. For example, the terminal 102 can directly perform fault feature extraction, degradation index calculation, and health assessment operations to obtain the health assessment result.

[0019] The terminal 102 can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, and smart in-vehicle devices. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices. The server 104 can be implemented using a standalone server or a server cluster composed of multiple servers, or it can be a cloud server.

[0020] In one exemplary embodiment, such as Figure 2 As shown, a wind turbine health assessment method is provided. This method is executed by a computer device, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is applied to... Figure 1 Taking server 104 as an example, the explanation includes the following steps S201 to S204. Wherein: Step S201: Obtain vibration operation data of the wind turbine.

[0021] It should be noted that the aforementioned vibration operation data refers to the vibration physical quantity data generated by the wind turbine during operation. Optionally, this vibration operation data can be acquired in real time through a data acquisition module, directly imported from external devices, or obtained from a database or blockchain. This embodiment does not impose any limitations on the method of acquiring the vibration operation data. Key components include the wind turbine's main shaft, gearbox, generator, and other core rotating parts. The data acquisition module may include a high-precision sensor system deployed at monitoring points of key components within the wind turbine, an analog-to-digital converter, and its communication interface.

[0022] The vibration operation data can be in the form of time-domain vibration waveform data, which is a continuous numerical sequence. It synchronously includes related information such as the acquisition timestamp, the corresponding monitoring point identifier, and the real-time speed of the wind turbine. Each data record covers the continuous acquisition values ​​of physical quantities such as vibration pulse, vibration amplitude, and vibration acceleration. The whole data is stored in the form of a structured dataset, which can be directly used for subsequent feature extraction calculation and analysis.

[0023] Step S202: Based on vibration operation data, determine the feature sets related to the failures of various key components in the wind turbine; each feature set includes at least two original features.

[0024] The aforementioned at least two original features include at least one basic feature and at least one component-directive feature. The basic feature reflects the overall energy level and distribution of vibration signals during wind turbine operation; the component-directive feature refers to the vibration energy characteristics of each key component at the target fault frequency. Please refer to [link to relevant documentation]. Figure 3 As shown, Figure 3 This diagram illustrates the process of extracting vibration data features according to an embodiment of this application. Through two parallel processing dimensions, feature extraction and quantitative analysis are performed on the collected vibration operation data to obtain at least two original features: basic features and component directional features. This step aims to obtain a macroscopic characterization of the vibration intensity and foundation impact characteristics of unit components by implementing industry-standard data, such as Verein Deutscher Ingenieure (VDI) standards. The aforementioned basic features are general statistical features extracted from the vibration operation data, including RMS acceleration, RMS velocity, peak acceleration, kurtosis, margin factor, spectral energy, and harmonic components. These are applicable to the overall state description of all key components, not specifically referring to a particular fault location, and mainly reflect the overall stability and foundation health level of the unit operation. Component directional features are features specifically extracted for the fault mechanisms of different key components such as the main shaft, gearbox, and generator. Combined with internal structural parameters, the theoretical fault frequency of the component is accurately derived. Subsequently, envelope demodulation processing is performed, and by calculating the energy peak value of a specific frequency band, sensitive capture of weak impact signals caused by early faults is achieved.

[0025] For example, for a spindle, the characteristic frequency of bearing failure is calculated, and after envelope demodulation processing, the damage pulse energy of the inner or outer ring of the bearing is determined. For a gearbox, the bearing failure frequency is calculated, and the bearing damage pulse energy is determined. The gear failure frequency is also calculated, and the effective value and margin factor of the fault frequency band for planetary gears and the sun gear are determined. For a generator, the electrical fault frequency is calculated, and the electrical fault frequency energy is determined. These calculations are strongly correlated with the typical failure modes of specific components, directly pointing to the deterioration and damage of a particular component or part, providing clear fault location capabilities. Each key component corresponds to a feature set. Extracting basic features ensures the comprehensiveness of overall condition monitoring, while extracting component-specific features improves the accuracy of fault location. Together, they constitute a complete feature set, providing layered and reliable feature support for subsequent deterioration index calculation, component deterioration value assessment, and fault diagnosis.

[0026] In one embodiment, based on vibration operation data, a set of features related to failures of various key components in the wind turbine is determined, including: The vibration operation data is filtered to obtain filtered vibration operation data; basic features are extracted from the filtered vibration operation data; the basic features include: effective value of acceleration and acceleration kurtosis; based on the structural parameters of the wind turbine, the filtered vibration operation data is analyzed to obtain the component orientation features corresponding to each key part of the wind turbine; based on the basic features and the component orientation features corresponding to each key part, a feature set related to the faults of each key component in the wind turbine is constructed.

[0027] Specifically, after acquiring vibration operation data, the collected vibration operation data of the wind turbine main shaft, gearbox, and generator are filtered for different speed ranges of different parts of the wind turbine. Data with speeds below a preset threshold, such as 1000 RPM, is removed. Vibration operation data with speeds below 1000 RPM are not included in subsequent diagnostics to eliminate interference from low-speed operation of the unit on feature extraction and ensure the stability of feature data, thus obtaining filtered vibration operation data. Basic features and component orientation features are then extracted from the filtered vibration operation data to provide basic monitoring of the wind turbine's operating status, suitable for early warning and long-term trend monitoring.

[0028] The aforementioned fundamental features include features related to VDI fundamental indicators and other fundamental features. When extracting fundamental features, based on filtered vibration operation data, features related to VDI fundamental indicators and other fundamental features are calculated. These features include, but are not limited to, RMS acceleration values ​​and acceleration kurtosis. Through the calculation of these fundamental features, basic monitoring of the overall operating status of wind turbine units can be achieved. These features can be directly applied to early warning of wind turbine unit operating status and can also support long-term trend monitoring of wind turbine unit operating status. The process involves obtaining filtered vibration operation data, selecting effective data segments of the vibration acceleration signal, and removing invalid noise generated during signal acquisition to ensure data validity, resulting in a processed vibration acceleration signal. Then, following general calculation standards for vibration signal analysis, the processed vibration acceleration signal is numerically calculated to obtain the effective acceleration value, which is the root mean square value of the vibration acceleration signal, reflecting the overall energy level of the vibration signal. Based on this effective acceleration value, the ratio of the fourth-order central moment of the signal to the fourth power of the root mean square value is calculated to obtain the acceleration kurtosis. This acceleration kurtosis reflects the waveform distribution characteristics of the vibration acceleration signal and is highly sensitive to the impact components in the signal.

[0029] When extracting component directional features, for specific fault types of different components of the wind turbine, such as bearings, gears, and electrical components, and in conjunction with the structural parameters of the wind turbine, the filtered vibration operation data is envelope demodulated based on a specific fault frequency range. The envelope peak value of a specific frequency band is calculated, and feature values ​​within a specified frequency range are extracted. Simultaneously, time-domain feature indicators are extracted to construct a multi-dimensional feature vector space. The basic features and component directional features are different dimensions within this multi-dimensional feature space, which is used to describe the state of key components. Specifically, for main shaft bearing faults, features such as the pulse energy of bearing inner ring damage and the pulse energy of bearing outer ring damage are extracted; for gearbox gear or bearing faults, features such as the effective value of the planetary gear fault frequency band, the effective value of the sun gear fault frequency band, the margin factor, and the pulse energy of bearing damage are extracted; for generator bearing or electrical faults, features such as bearing fault frequency energy and electrical fault frequency energy are extracted. By using these methods, background noise is reduced and fault impact features are amplified, achieving accurate fault location.

[0030] In this embodiment, the structural parameters of the wind turbine are first obtained. These structural parameters are inherent characteristics of the wind turbine, including the transmission ratio, number of gear teeth, bearing type, bearing pitch circle diameter, rolling element diameter, number of rolling elements, and pressure angle. Simultaneously, the real-time operating speed of the wind turbine is acquired. Based on these structural parameters and the real-time operating speed, the target fault characteristic frequencies of the main shaft bearing, gearbox planetary gears, gearbox sun gear, and generator bearing are calculated to determine the target fault frequency range for each component. For example, if the main shaft speed is 20 RPM, and the inner ring fault characteristic coefficient is calculated to be 12.5 based on the bearing structural parameters, the target frequency is 20 ÷ 60 × 12.5 = 4.16 Hz. Then, a bandpass filter frequency band containing the target fault characteristic frequency is set as the center. The vibration operation data after speed filtering is then subjected to bandpass filtering to filter out background noise and interference signals from non-fault frequency bands, retaining the effective signal matching the target fault characteristic frequency, thus obtaining the bandpass-filtered signal.

[0031] Next, a Hilbert transform is performed on the bandpass-filtered signal to obtain its analytical form. The magnitude of the analytical signal is calculated to obtain the vibration signal envelope. A Fast Fourier Transform is then performed on the vibration signal envelope to obtain its spectrum. The maximum amplitude corresponding to the target fault characteristic frequency is extracted from the envelope spectrum, which is the envelope peak value for a specific frequency band. Based on the target fault characteristic frequency and the envelope spectrum, feature values ​​within a specified frequency band are extracted, including fault frequency energy, damage pulse energy, effective frequency band value, and margin factor. Simultaneously, time-domain feature indicators of the vibration operation data are extracted, including time-domain statistical features and component fault-related features. Furthermore, the extracted feature values ​​within the specified frequency band and the time-domain feature indicators are integrated to construct feature vectors for the main shaft component, gearbox component, and generator component, respectively, forming a multi-dimensional feature vector space for the wind turbine.

[0032] Specifically, assuming the monitored wind turbine unit comprises three key components—the main shaft, gearbox, and generator—their corresponding feature sets are represented as X, Y, and Z, respectively. Each feature set contains multiple features extracted from each component. This feature set is represented by a feature vector, and each feature within a feature set is represented by an feature value. The feature vector for the main shaft mentioned above is: ; in, Indicates the first spindle There are eigenvalues, eigenvalues arrive This includes the basic features corresponding to the spindle and the directional features of the components.

[0033] The eigenvector corresponding to the gearbox is: ; in, Indicates the first gearbox There are eigenvalues, eigenvalues arrive This includes the basic features and component orientation features corresponding to the gearbox.

[0034] The eigenvector corresponding to the generator is: ; in, Indicates the generator's first There are eigenvalues, eigenvalues arrive This includes the basic characteristics and component orientation characteristics corresponding to the generator.

[0035] Step S203: Determine the degradation index of each original feature in the feature set, and perform nonlinear aggregation on the degradation index to obtain the degradation value of each key component of the wind turbine. The degradation value is used to characterize the degree of failure degradation of the key component that deviates from the healthy operating state. The degradation index is calculated based on the health benchmark value, combined with the rounding up operator and the zero-value truncation operation. The health benchmark value is the median of the feature value sequence of each original feature in a preset time period.

[0036] Step S204: Based on the degradation values ​​of each key component, the health status of the wind turbine is assessed using a dynamic time window-driven multivariate parameter degradation convergence assessment strategy to obtain the health assessment results. The multivariate parameter degradation convergence assessment strategy includes: daily degradation classification mapping, daily status aggregation, time weighting and proportion calculation, and comprehensive proportion determination operation.

[0037] After obtaining the feature set, the Deterioration Index (DI) is calculated for all feature sets. Based on the feature values ​​corresponding to each original feature, the continuously fluctuating original features in the feature set are transformed into discrete quantitative scores to evaluate the status of each component of the wind turbine, thereby intuitively representing the degree of deviation of each component of the wind turbine from the healthy operating state.

[0038] It should be noted that the above-mentioned degradation index is calculated using the characteristic value corresponding to the original feature, the health benchmark value, and the sensitivity adjustment factor as core parameters, combined with rounding up and zero-value truncation operations. All original features of key components of the wind turbine (main shaft component, gearbox component, generator component) are calculated using a unified formula to ensure the consistency of the evaluation logic. The health benchmark value can be understood as the typical reference value of each original feature under healthy and stable operating conditions. It is obtained by taking the median of the feature value sequence within a preset normal operating cycle and is used as a unified reference benchmark to determine whether the feature has deviated from its degradation status.

[0039] Please see Figure 4 As shown, Figure 4 This is a schematic diagram of the health assessment process for a wind turbine provided in this application embodiment. First, real-time vibration operation data of the wind turbine is acquired, and vibration operation data with speeds greater than 1000 RPM are selected. For key components such as the main shaft, gearbox, and generator components, a multi-dimensional feature extraction module extracts basic indicators and component directional features from this data as raw features. Then, a degradation index calculation module is used to calculate the degradation index, simultaneously calculating the degradation index of the basic features and the component directional features to assess the component status. During the degradation index calculation and parameter update process, an adaptive update / freeze mechanism is also added. Its core purpose is to prevent parameter drift in the continuously updated health benchmark values ​​and sensitivity adjustment factors when wind turbine components enter the degradation stage, thereby masking the component's fault characteristics and ensuring the objectivity and accuracy of the degradation index calculation, providing reliable parameter support for subsequent assessments.

[0040] After calculating the degradation index and assessing the condition of individual original features, nonlinear aggregation processing is performed on all original feature degradation indices associated with each key component to obtain the degradation value of each key component, such as the degradation value of the bearing and gear components, and thus the degradation value of the key component. This degradation value can accurately characterize the degree of degradation of the corresponding key component from its healthy operating state. Entering the fault decision and diagnosis stage, a dynamic time window-driven multi-parameter degradation convergence assessment strategy is adopted. Through four core stages—"daily degradation grading mapping," "daily condition aggregation," "time weighting and proportion calculation," and "formation of a comprehensive decision matrix"—a comprehensive degradation index that reflects the degradation trend of components is constructed. Based on this comprehensive degradation index, fault diagnosis of the wind turbine is completed, and health assessment results are output simultaneously. These results include targeted operation and maintenance suggestions corresponding to the fault type and location, ultimately forming a complete auxiliary decision-making closed loop to ensure the practicality and operability of the diagnostic results and meet the operation and maintenance needs of industrial sites.

[0041] Compared with existing technologies, this solution extracts multi-dimensional feature sets related to the failures of key components based on the acquired vibration operation data. The feature set corresponding to each key component covers at least two basic indicators, providing a comprehensive data foundation for subsequent wind turbine health assessment. Furthermore, it uses the median of the preset time period feature value sequence of each original feature as the health benchmark value, leveraging the median's resistance to outliers and random fluctuations to filter out instantaneous impact interference from the industrial site at its source. Simultaneously, it combines the rounding up operator and zero-value truncation operation to calculate the degradation index. This not only focuses on the true degradation direction and eliminates normal fluctuation interference under healthy conditions through the zero-value truncation operation, but also achieves discretization through the rounding up operation. The system effectively suppresses random fluctuations and instantaneous shocks, improving the robustness of degradation quantification. Then, nonlinear aggregation is used to obtain degradation values ​​for each key component, avoiding abnormal dilution and artificially inflated values ​​from linear summation, thus accurately characterizing the degree of component degradation. Finally, a dynamic time window-driven multi-parameter degradation convergence assessment strategy is adopted, including daily degradation classification mapping, daily state aggregation, time weighting and proportion calculation, and comprehensive proportion determination. Daily state aggregation smooths short-term interference, and time weighting highlights recent real-time degradation trends, solving the alarm lag and false alarm tailing problems of equal-weighted windows. Simultaneously, multi-level classification judgment achieves accurate multi-state assessment, replacing the traditional two-state judgment method, significantly improving the accuracy and reliability of wind turbine health status assessment.

[0042] In one embodiment, a specific implementation for determining the degradation index of each original feature in the feature set is also provided; please refer to [link to relevant documentation]. Figure 5 As shown, the method includes: Step S301: For each feature set, obtain the health baseline value and sensitivity adjustment factor of each original feature; the sensitivity adjustment factor is obtained by multiplying the median of the difference between the original feature and the health baseline value over a preset time period by a scaling coefficient.

[0043] Step S302: Subtract the feature values ​​of each original feature from the health baseline value to obtain the difference value.

[0044] Step S303: Determine the maximum value from the product of the health baseline value and the minimum value coefficient, and the sensitivity adjustment factor.

[0045] In step S304, the result of dividing the difference by the maximum value is processed by the rounding up operator and the zero-value truncation operation is performed to obtain the degradation index of each original feature.

[0046] Specifically, please see Figure 6 As shown, Figure 6This is a schematic diagram of the state scoring and anomaly determination process based on eigenvalue quantization in this application. This process transforms discrete physical features into intuitive state scores through a standardized mathematical calculation path. Specific steps include: a) Feature quantization and degradation value calculation: After the system inputs the feature value Y of the original feature, it first retrieves the health baseline value corresponding to each original feature. and sensitivity adjustment factor Then, the feature degradation index is calculated, and the eigenvalues ​​are compared with... The difference, and divide it by Then, the zero-value truncation is performed by up-rounding to obtain the degradation index of the original feature. b: Nonlinear aggregation: After obtaining the degradation index of each original feature, part-level aggregation is performed to obtain the degradation value of each part. For example, the features extracted from the internal gear ring, sun gear, and planet gear in the gearbox are all attributed to the gear part. Then, the degradation values ​​of different features belonging to the same part are aggregated and calculated to obtain the degradation value of each key part.

[0047] Based on the original features, health baseline values, and sensitivity adjustment factors in the feature vector, the degradation index of each key component on day j and the i-th original feature is obtained. It can be represented as follows: ; in, Let i be the degradation index of the i-th original feature on day j. Let i be the feature value of the i-th original feature on day j. Let i be the health baseline value of the i-th original feature. Let be the sensitivity adjustment factor for the i-th original feature. This is a rounding operator used to convert the continuous deviation of the feature value from the health benchmark value into a discrete score, max(0, ) indicates zero-value truncation, which is used to characterize that when the feature value of the original feature is less than or equal to the health baseline value, the degradation index is forcibly assigned to zero, and only the feature degradation deviation is quantified; η is the minimum value coefficient of the denominator, which can range from 1% to 5%, and is used to avoid the denominator approaching zero, which would cause numerical anomalies in the calculation results.

[0048] The aforementioned health baseline values ​​are used to characterize the standard reference values ​​of the original characteristics under healthy operating conditions, with a preset time period as the most recent. Taking the day as an example, using the most recent The median of the original feature sequence is obtained by calculating the following formula: ; in, Let i be the feature value of the i-th original feature on day j. For characteristic statistical days, The operator is used to calculate the median. Let be the health baseline value of the i-th original feature.

[0049] To address the issue of outliers in the original vibration characteristic values ​​of wind turbines due to random impacts and atypical operating conditions, this paper abandons the shortcomings of traditional arithmetic means, which are susceptible to extreme value shifts. Instead, it directly adopts the statistical median formula as a basis, and combines this with the actual data collection method of "daily characteristic value collection" in this application, limiting the data sequence to the most recent... The i-th original feature value sequence of the day is used to derive a formula applicable to the health benchmark determination of wind turbine units.

[0050] In this step, the median is used as the health benchmark value, which can effectively eliminate the influence of abnormal data such as random shocks, atypical operating conditions, and instantaneous interference, and avoid the defect that the arithmetic mean is easily affected by extreme values, so that the health benchmark value truly reflects the normal operating status of the wind turbine.

[0051] With the preset time period as the most recent Taking a day as an example, the above sensitivity adjustment factor is used to characterize the normal fluctuation range of the original characteristics in a healthy state, and adapts to the self-fluctuation characteristics of different characteristics. It can be calculated by the following formula: ; in, As a sensitivity adjustment factor, For characteristic statistical days, Let i be the feature value of the i-th original feature on day j. This is the median calculation operator, and 1.4826 is the classic statistical conversion factor. Let be the health baseline value of the i-th original feature.

[0052] This application transforms the continuous relative deviation between characteristic values ​​and health benchmark values ​​into discrete integer scores by setting an up-rounding operator in the degradation index calculation. This allows the degree of degradation to be intuitively divided into different levels, avoiding the ambiguity caused by continuous values, facilitating engineers to set clear warning thresholds, and improving the operability of fault warnings. Furthermore, through zero-value truncation, cases where characteristic values ​​are lower than the health benchmark value can be forcibly set to 0, quantifying only the positive deviation of characteristic degradation from the healthy state, completely eliminating negative interference when characteristic values ​​are better than the healthy state. This ensures that the degradation index focuses only on the degree of equipment condition deterioration, guaranteeing the specificity of the assessment results. Up-rounding small relative deviations smooths out random noise in continuous signals, and zero-value truncation filters out normal small fluctuations in the healthy state. The combined effect of these two measures prevents instantaneous interference or normal operating condition fluctuations from being misjudged as degradation, significantly improving the robustness of the degradation index and reducing the false alarm rate.

[0053] Optionally, this embodiment also provides a scheme for setting and updating parameters in the degradation index calculation.

[0054] It should be noted that the core parameters for calculating the degradation index in this application are the health benchmark value and the sensitivity adjustment factor. In order to ensure that the parameters are adapted to the operating status of the wind turbine throughout its entire life cycle, and to avoid parameter drift during the degradation stage from masking fault characteristics, an integrated control scheme with initialization setting, adaptive update and freezing mechanism is designed for the health benchmark value and the sensitivity adjustment factor.

[0055] The aforementioned preset time period includes a first preset period or a second preset period. The duration of the first preset period is longer than that of the second preset period. For example, the first preset period is the most recent three months, and the second preset period is for example, the most recent one month.

[0056] Obtain the health baseline values ​​and sensitivity adjustment factors for each raw feature, including: Determine whether a health benchmark value and sensitivity adjustment factor exist. If no health benchmark value and sensitivity adjustment factor exist, extract vibration operation data within the first preset cycle under normal operating conditions of the wind turbine, and calculate the health benchmark value and sensitivity adjustment factor based on the vibration operation data within the first preset cycle. If a health benchmark value and sensitivity adjustment factor exist, obtain the current health benchmark value and the current sensitivity adjustment factor, and determine whether state degradation has occurred. Based on the current health benchmark value, the current sensitivity adjustment factor, and the state degradation result, determine the health benchmark value and sensitivity adjustment factor.

[0057] Please see Figure 7 As shown, Figure 7 Health benchmark values ​​provided for embodiments of this application and sensitivity adjustment factor A schematic diagram of the dynamic update and iterative calibration process is provided. By establishing a data feedback mechanism that combines long-term and short-term data, the real-time nature and accuracy of diagnostic benchmarks are ensured, and a "freeze mechanism" is used to ensure the objectivity of health assessments.

[0058] Specifically, in the process of obtaining the health baseline values ​​and sensitivity adjustment factors of each original feature, taking the first preset period as the most recent three months as an example and the second preset period as the most recent month (e.g., 30 days) as an example, the parameter initialization is set as a pre-processing step for the deterioration index calculation. This is applicable to scenarios where wind turbine units are newly put into operation, undergo health assessment for the first time, or where there are no historical health baseline values ​​or sensitivity adjustment factors available.

[0059] Based on the current health baseline, current sensitivity adjustment factor, and state deterioration results, determine the health baseline and sensitivity adjustment factor, including: When the condition degradation result indicates that no condition degradation has occurred, the vibration operation data of the wind turbine is collected within the second preset cycle, and the health benchmark value and sensitivity adjustment factor are calculated based on the vibration operation data of the second preset cycle. When the condition deterioration result indicates that condition deterioration has occurred and the new health benchmark value is greater than k times the current health benchmark value, a freeze update is initiated, and the current health benchmark value and the current sensitivity adjustment factor are used as the health benchmark value and sensitivity adjustment factor, where k is a positive number; the new health benchmark value is calculated based on the vibration operation data of the wind turbine unit within the second preset cycle.

[0060] Specifically, the first step is to determine from the database whether a health benchmark exists. and sensitivity adjustment factor When health benchmarks and sensitivity factors are not available, vibration operation data for the most recent three months under normal operating conditions are selected from the historical operating data of the wind turbine. This data needs to exclude abnormal data from atypical operating conditions such as turbine start-up and shutdown, fault maintenance, and extreme weather to ensure that the data can truly reflect the characteristic fluctuation patterns of the turbine under healthy operating conditions. Based on the selected vibration operation data, the initial calculation of health benchmarks and sensitivity adjustment factors corresponding to all original characteristics is completed according to the calculation formulas for health benchmarks and sensitivity adjustment factors. The calculation results are stored in the system as initial parameters. In the subsequent deterioration index calculation process, if the health benchmarks and sensitivity adjustment factors corresponding to the original characteristics already exist in the system, they can be directly retrieved and used without repeated initialization calculations.

[0061] During the continuous execution of wind turbine health assessments, an adaptive parameter update mechanism is designed to ensure that health baseline values ​​and sensitivity adjustment factors adapt to the slow changes in the turbine's operating status (normal fluctuations under non-fault-related degradation) and achieve dynamic parameter calibration. When health baseline values ​​and sensitivity adjustment factors exist, the current health baseline value and current sensitivity adjustment factor are obtained. Subsequently, based on the current parameters, it is determined whether component degradation has occurred. If degradation is detected or baseline drift occurs, a freeze mechanism is triggered, stopping parameter updates and using the baseline parameters from the previous cycle to prevent degradation data from causing parameter drift and masking fault characteristics. If the status is determined to be stable, the vibration operation data of the wind turbine for the most recent month is extracted, the health baseline value and sensitivity adjustment factor are recalculated, and adaptive updates are performed to generate new parameter values. , The new parameters are then fed back to the dynamic parameter update process, initiating the next update cycle. This mechanism not only enables dynamic calibration of parameters under healthy conditions but also ensures the stability of parameters under deteriorated conditions, providing reliable support for the accurate calculation of the degradation index.

[0062] It should be noted that, to avoid the upward drift of parameter values ​​caused by the inclusion of degradation characteristic data in parameter update calculations after the unit components enter the degradation stage, which could mask fault characteristics and lead to inaccurate degradation index calculations, a parameter update freeze mechanism is added to the adaptive parameter update mechanism. The freeze trigger conditions and execution rules are clearly defined. During each adaptive parameter update calculation, the newly calculated health baseline value is compared with the historical health baseline value before the drift occurred. When the newly calculated health baseline value is greater than k times the current health baseline value, it is determined that the health baseline value of that characteristic item shows a significant upward drift trend, and it can be directly inferred that the corresponding unit component has entered the fault degradation stage, thus triggering the freeze condition. If these data continue to be included in the calculation, the parameter value will gradually increase over time, potentially masking fault characteristics and causing the subsequently calculated characteristic evaluation to tend towards 0. Here, k is a positive number, which can be understood as an engineering experience value and can be set in advance according to the wind turbine equipment model and operating conditions.

[0063] When the freeze condition is triggered, the adaptive update program for the health baseline value and sensitivity adjustment factor of the feature item is immediately frozen, and the update of the two parameters of the feature item using new feature data is stopped. At the same time, the health baseline value and sensitivity adjustment factor calculated in the last health cycle before the drift occurred are forcibly used as the current reference value for subsequent degradation index calculations. That is, the current health baseline value and the current sensitivity adjustment factor are used as the health baseline value and sensitivity adjustment factor until the unit completes fault repair and the corresponding components return to a healthy operating state. Then, the freeze mechanism is manually lifted and the adaptive update of parameters is restarted.

[0064] This solution establishes a healthy foundation for parameter calculations through initial settings, achieves dynamic calibration of parameters through adaptive updates, and avoids parameter drift during the degradation phase through a freezing mechanism. These three aspects work together to ensure that the health benchmark and sensitivity adjustment factor remain effective and objective throughout the entire life cycle of the unit, providing reliable parameter support for the accurate calculation of the degradation index.

[0065] Furthermore, the degradation index is nonlinearly aggregated to obtain the degradation value of each key component of the wind turbine, including: for each key part of the key component, the degradation index of each original feature associated with the key part is nonlinearly aggregated to obtain the degradation value of each key part; Euclidean distance is used to nonlinearly aggregate the degradation value of each key part to obtain the degradation value of the key component.

[0066] Specifically, after calculating the degradation index of each original feature of the wind turbine, a nonlinear aggregation method based on Euclidean distance is used to aggregate and calculate multiple degradation indices corresponding to the same critical part to obtain the degradation value of the critical part of the wind turbine; then, based on the degradation value of the critical part, the same nonlinear aggregation method is used to obtain the degradation value of the corresponding component of the wind turbine, so as to realize the graded and quantitative assessment of the degradation degree of the critical part and the whole component of the wind turbine.

[0067] Each critical component of the wind turbine, including the main shaft, gearbox, and generator, corresponds to multiple independent features, and each independent feature corresponds to a degradation index. To avoid the dilution of abnormal features and the excessive amplification of normal fluctuations caused by linear summation, this application uses Euclidean distance as the basis for nonlinear aggregation to aggregate multiple degradation indices corresponding to the same critical component. The degradation value of the critical component can be calculated using the following formula: ; Where B is the degradation value of a critical component in the wind turbine main shaft, gearbox, or generator; This is the degradation index corresponding to the s-th feature associated with this critical part. This represents the total number of features associated with this critical part.

[0068] After obtaining the degradation values ​​of critical components, the same Euclidean distance nonlinear aggregation method is used to aggregate the degradation values ​​of all critical components under the same part, thereby obtaining the degradation values ​​of the critical components of the wind turbine as a whole. These values ​​are used to characterize the degree of degradation of critical components such as the main shaft, gearbox, or generator. The degradation values ​​of critical components can be calculated using the following formula: ; Where D represents the degradation value of a critical component in the wind turbine's main shaft, gearbox, or generator. For the first The degradation value of a key component This refers to the total number of critical parts included in a critical component.

[0069] This embodiment introduces Euclidean distance for nonlinear aggregation, which can amplify the weight of abnormal degradation indices through squaring operations. This avoids the problem of insensitive early warning caused by the apportionment of severe anomalies in a single core feature being affected by other normal features in linear summation. At the same time, square root operations suppress the artificially high scores caused by small fluctuations in multiple features, reducing false alarms. This allows the degradation values ​​of key parts and components to accurately and stably reflect the true degree of degradation of the wind turbine, improving the sensitivity and reliability of fault identification. Furthermore, it enables the stable, sensitive, and accurate extraction of overall degradation information from multiple independent degradation indices, meeting the health assessment needs of wind turbines under non-stationary and high-noise operating conditions.

[0070] Understandably, in the linear summation method, if a core feature exhibits extreme degradation while other features remain normal, the contribution of this anomalous feature will be diluted by the values ​​of numerous normal features, resulting in a negligible increase in the overall score and insensitive fault warnings. However, in the Euclidean distance aggregation method, the degradation index is calculated in a squared form. The larger the anomalous feature, the more exponentially its contribution to the overall result, significantly highlighting the weight of extreme anomalous features. This allows for a rapid increase in the degradation values ​​of critical parts and components, enabling sensitive warnings of severe faults. In this embodiment, the degradation values ​​of each critical part and component obtained through step-by-step aggregation using Euclidean distance accurately reflect the true degradation state of the corresponding part or component. This prevents the degradation from being masked by a single abnormal feature, and also avoids misjudgments due to small normal fluctuations, providing a reliable quantitative basis for subsequent fault classification, fault location, and maintenance decisions.

[0071] In one embodiment, this application also provides a specific implementation method for assessing the health status of a wind turbine generator based on the degradation values ​​of various key components using a dynamic time window-driven multi-parameter degradation convergence evaluation strategy, thereby obtaining health assessment results. The method includes: Based on the degradation values ​​of all key components corresponding to each set of data within a sliding time window, a feature degradation matrix corresponding to each set of data is constructed.

[0072] The feature degradation matrix corresponding to each group of data is compared with the preset grading threshold to obtain the instantaneous degradation level of each group of data within a single day.

[0073] Summarize the instantaneous degradation levels of all groups of data within a single day, calculate the sample proportion of each instantaneous degradation level, and determine the representative status label for the single day based on the comparison between the sample proportion and the preset proportion threshold.

[0074] Within the sliding time window, the representative status labels of each day are assigned weights by the time decay factor, the weighted failure rate index is calculated, and the weighted failure rate index of each failure level is output.

[0075] The weighted percentage index of each fault level is compared with the preset percentage threshold to output the health assessment results of the wind turbine; the health assessment results include abnormal and normal states.

[0076] It should be noted that when assessing the health status of wind turbine units, a multi-parameter degradation convergence evaluation strategy driven by a dynamic time window is adopted. By setting a sliding time window as the monitoring period, the degradation values ​​of each key component within the period are integrated to construct a characteristic degradation matrix, thereby realizing continuous monitoring and fault identification of the status of wind turbine unit components.

[0077] In this embodiment, the core parameter for setting the monitoring period is the sliding time series set W. This sliding time series set can be understood as the duration of a dynamic time window, also known as a sliding time window, with the unit being days. W is a preset engineering experience value, which can be flexibly adjusted according to the operating conditions of the wind turbine and the rate of fault development, such as selecting W=7 days, 14 days, etc., to cover the typical cycle of fault development, ensuring that the continuous trend of component deterioration can be captured, and avoiding missed judgments due to the monitoring period being too short or misjudgments due to the monitoring period being too long.

[0078] After setting the sliding time series set W as the fault monitoring period, this period contains w sets of component degradation value data. The value of w is consistent with that of W, that is, one set of component degradation values ​​corresponds to each day. If W=7 days, then w=7, corresponding to 7 sets of data. Each set of data is the degradation value of the main shaft component, gearbox component, or generator component of the wind turbine obtained by Euclidean distance nonlinear aggregation as described above, ensuring that the data source is traceable and the calculation process is reproducible. In order to realize the systematic analysis of the component degradation trend within the period, the w sets of component degradation values ​​within the sliding time window W are arranged in chronological order to construct a characteristic degradation matrix. This matrix is ​​a 1×w row matrix, used to centrally characterize the continuous change law of component degradation values ​​within the monitoring period. Its expression is as follows: ; in, This is a feature degradation matrix used to integrate the degradation value data of all components within the sliding time window W; The component degradation value corresponding to the w-th data group within the sliding time window W (w is a positive integer from 1 to W) is completely consistent with the definition of the component degradation value D obtained by Euclidean distance aggregation mentioned above. It is dimensionless and is used to characterize the degree of component degradation at the corresponding time point. w is the number of component degradation value data groups included in the sliding time window W, which is equal to the value of the sliding time window duration W. For example, if W=7 days, then w=7, corresponding to the component degradation values ​​at 7 time points.

[0079] The core function of this step is to provide a standardized and continuous data foundation for subsequent degradation convergence assessment and fault identification. By integrating discrete daily component degradation values ​​into a continuous time series through the feature degradation matrix, it is easier to analyze the convergence trend and rate of change of component degradation, and thus accurately determine whether the component has failed and the stage of failure development.

[0080] After constructing the feature degradation matrix V of the sliding time window W, a multi-parameter degradation convergence evaluation strategy driven by a dynamic time window is adopted. Through a hierarchical and progressive logic of "instantaneous degradation level mapping → daily state aggregation → time-weighted calculation → threshold matching decision," accurate assessment of the wind turbine component status is achieved. First, hierarchical correlation calculations are performed on the multi-level data of the wind turbine collected daily within the preset sliding time window W (i.e., vibration operation data corresponding to key parts of the main shaft, gearbox, and generator). Combining the component directional features and basic features extracted earlier, as well as the corresponding degradation index and component degradation value, the degradation features corresponding to each set of waveform data (such as fault frequency energy, damage pulse energy, degradation index, etc.) are mapped to discrete instantaneous degradation levels according to a preset degradation level classification standard. The instantaneous degradation level is used to characterize the instantaneous operating state of the component at the time of acquisition of a single set of waveform data, ensuring that the degradation features of each set of waveform data can be accurately quantified into identifiable level labels. Second, daily state aggregation judgment is performed to generate representative state labels. Since multiple sets of waveform data are collected within a single day, the instantaneous degradation levels of different sets of data may differ. To smooth out short-term random interference and accurately characterize the overall operating pattern of components on that day, a daily-level state aggregation judgment is performed by combining the distribution ratio of instantaneous degradation levels of all samples within the day (i.e., the multiple sets of waveform data collected on a single day). Specifically, the percentage of samples for each instantaneous degradation level within a single day is statistically analyzed, and the instantaneous degradation level with the highest percentage is selected as the representative state label for that day. If multiple degradation levels have the same percentage and are the highest, the degradation level with the higher percentage is selected as the representative state label for that day. This representative state label can effectively filter out short-term random interference within the day (such as instantaneous voltage fluctuations, sudden environmental interference, etc.), truly reflect the overall operating status of components on that day, and provide reliable daily-level data support for subsequent window period trend analysis.

[0081] Subsequently, a time-weighted model based on an exponential decay function is introduced to calculate the weighted percentage of each degradation level within the window period. Considering that within the sliding time window W, the closer the daily status label is to the current moment, the better it reflects the actual degradation trend of the component, and the greater its impact on the timeliness of fault diagnosis, an exponential decay function is introduced to construct the time-weighted model. This model assigns a weight score to the representative status label of each day within the window that increases exponentially with time. The core of the time-weighted model is the exponential decay function, whose expression is derived from the classic exponential function in mathematics. The weight score increases exponentially with time approaching the current moment; that is, the weight of the daily status label at a closer moment is higher, and the weight of an earlier moment is lower, thus achieving a balance between timeliness and timeliness. Afterwards, based on the weight scores of each day, the sum of the weighted scores corresponding to each degradation level within the window is calculated. Then, the proportion of the sum of the weighted scores of each degradation level to the total sum of the weighted scores of all degradation levels is calculated to obtain the weighted percentage of each degradation level. This proportion can highlight the impact of recent degradation trends while also taking into account long-term degradation patterns within the window. Finally, by matching the weighted proportion with the preset level proportion threshold, accurate diagnostic results are output.

[0082] Specifically, please see Figure 8 As shown, after obtaining the feature degradation matrix V for the sliding time window (W days), the first stage is performed: daily degradation grading mapping. The degradation values ​​B of key parts (such as bearings and gears) are obtained by aggregating the features associated with the parts in the original features. The component degradation values ​​D are then obtained by aggregating the degradation values ​​B of each key part. Specifically, for the component degradation value corresponding to the w-th group of data within the sliding time window W... Threshold grading is performed, mapping continuous degradation values ​​to discrete degradation levels and degradation scores S. (Setting...) These correspond to the slight degradation threshold, moderate degradation threshold, and severe degradation threshold, respectively. >0, by comparing the component degradation value with the preset three-level grading threshold ( , , The system compares daily component status data and automatically maps it to four scoring tiers: normal, minor fault, general fault, and severe fault. The grading rules are as follows: like The instantaneous degradation level is marked as a severe fault; if Marked as a general fault; if Mark as a minor fault; if < Marked as normal. Let w be the feature degradation matrix of the w-th data group.

[0083] In this step, the degradation values ​​of each group of data are uniformly mapped to four levels: normal, slightly degraded, moderately degraded, and severely degraded, through the above-mentioned grading, thus achieving the initial transformation from underlying multidimensional features to standardized state assessment scores.

[0084] The second stage, daily state aggregation, is then executed. Specifically, to reduce computational burden and save space, the sample states within a 24-hour period are aggregated as the daily state, taking into account the equipment's operating cycle. The daily state is obtained by calculating the proportion of a single sample label on that day. The degradation levels of all samples within the 24 hours of a single day are statistically analyzed to determine the sample proportion of each instantaneous degradation level. Then, based on the comparison between the sample proportion of each level and a preset proportion threshold, the representative state for that day is determined, completing the daily state compression. The preset proportion threshold can be customized according to actual needs. For example, the state compression logic for day w is as follows: like Then the representative status label for day w is "Critical malfunction"; like( Then the representative status label for day w is =“General fault”; like( Then the label for day w is "Minor malfunction"; otherwise, "normal".

[0085] in, For the first The percentage of samples with a "critical fault" status within the day. This represents the percentage of samples with a "general fault" status on day w. For the first The percentage of samples with a "minor fault" status within the day.

[0086] This step is used to smooth out abnormal fluctuations in a single day caused by short-term random shocks and instantaneous noise, and outputs a deterministic label that can represent the overall operating status of the day.

[0087] The third stage involves window-level time-weighted calculation. Specifically, to reflect the fault evolution pattern and suppress the interference of outdated historical data on the current diagnosis, and to increase the weight of recent data, an exponential time decay weighting factor is introduced within the sliding time window W. This factor is used to weight and correct the representative state labels for each day within the sliding time window, preventing outdated historical data from interfering with the current judgment and ensuring that data closer to the current moment has higher decision confidence. The weight corresponding to day w is... Calculate using the following formula: ; in, The weight assigned to day w. is the decay constant, a preset integer used to control the weight decay rate; W is the total duration of the sliding time window (unit: days).

[0088] Subsequently, based on the weights, a weighted percentage index is calculated for each degradation level (severe, moderate, minor, normal) over the entire sliding time window to transform the discrete daily assessment results into a statistically significant continuous judgment criterion. The weighted percentage index can be represented as follows: ; in, for Weighted percentage index under each level The second phase consists of four labels: "Critical Fault", "General Fault", "Minor Fault", and "Normal". Let be the decision function, representing when The value is 1 when the corresponding gear is selected; otherwise, the value is 0.

[0089] Finally, the fourth stage is executed: the comprehensive decision matrix. Following the principle of prioritizing high-risk identification, a step-by-step matching logic of "from most severe to least severe, penetrating level by level" is used for the final diagnosis. The weighted proportion index is compared with the preset level proportion threshold, and the final diagnostic result is output. For example, if a three-level judgment threshold is set, the thresholds are as follows: , , , The threshold for determining a serious fault. This is the threshold for general fault diagnosis. The threshold for determining minor faults. > > In practical applications, the threshold ratios for each level can be adaptively adjusted based on the window length; a higher threshold should be set for short windows. Value, set lower when using a long window. Values, for example It is usually set to 0.2 to 0.3. It is usually set at 0.4 to 0.5. It is usually set between 0.6 and 0.8.

[0090] Then, the weighted percentage index is compared with the preset level percentage threshold. For example, first, a verification is performed. Has the threshold for determining a serious fault been reached? If the condition is not met, then compare it sequentially with the critical fault determination threshold. General fault judgment threshold Minor fault determination threshold By comparing these data, a health assessment result is obtained. This can be understood as... When the health assessment result is "serious fault", the output is "critical fault". When the health assessment result is given, it indicates a general fault; when... If the condition is met, the output health assessment result is "minor fault"; if none of the above conditions are met, it is considered normal. Through this multi-level gradient logic judgment, the system can eliminate occasional noise interference, accurately pinpoint the current fault level (severe, moderate, minor, or normal), and finally output a diagnostic conclusion, providing quantitative decision support for predictive maintenance.

[0091] Please see Figure 9 As shown, Figure 9 This is a schematic diagram of diagnostic decision-making driven by a dynamic time window, which shows the diagnostic effect of the entire evolution process of the health status of wind turbine components from normal → minor fault → general fault → severe fault. The upper figure shows the change curve of characteristic value and dynamic benchmark within a 14-day monitoring period, and the lower figure shows the evolution process of component degradation index (DI) and health status within the corresponding period.

[0092] For the verification process of adaptive benchmark and freeze mechanism, during the first 3 days of the monitoring period, the component is in a healthy and stable operating state, and the characteristic value F fluctuates smoothly. The dynamic benchmark B follows the healthy benchmark value α and performs real-time adaptive updates to ensure that the benchmark value matches the current healthy operating condition. From the 4th day onwards, the characteristic value F continues to rise, and the drift of the healthy benchmark value α exceeds 1.2 times that of the initial benchmark. The system triggers the freeze mechanism, and the dynamic benchmark B is locked at the health level before the deterioration occurred (the gray area is the benchmark freeze interval). This completely avoids the problem of the benchmark value rising synchronously with the characteristic value during the fault evolution process, which leads to the "dilution" of the deterioration index. It effectively prevents missed warnings and ensures the accuracy of deterioration quantification.

[0093] To suppress short-term random shocks and instantaneous noise in industrial settings, the system executes daily state aggregation logic: it summarizes all waveform samples collected within 24 hours each day, calculates the sample proportion P of each degradation level, determines the representative state label for the day based on the proportion threshold, and finally generates the corresponding daily state color bars in the figure below. This realizes the aggregation of discrete instantaneous data into daily stable states, effectively filtering out interference from occasional daily anomalies.

[0094] This embodiment also provides discrete quantization and state evolution characteristics of the degradation index: by using the rounding up operator and zero-value truncation operation, continuous feature deviations are transformed into discrete degradation index DI, accurately matching the evolution stage of the component state: 1) Normal phase (days 1-3): Feature values ​​fluctuate slightly around the frozen baseline, DI remains in the 0-1 range, and the daily label remains stable as "normal"; 2) Minor Fault Stage (Days 4-7): The feature value deviates slightly from the frozen baseline. The calculated DI remains in the low range of 1-3. The daily label is judged as "minor fault", realizing accurate detection of early degradation. 3) General Failure Stage (Days 8-11): The degree of characteristic deviation continues to increase, DI enters the 4-6 range, and the representative label of the day is switched to "general failure" based on the proportion, clearly reflecting the gradual process of component deterioration; 4) Severe Fault Stage (Days 12-14): The characteristic value shows an explosive growth, the DI score quickly exceeds 8 points, the system accurately captures the high-intensity fault impact, and the daily label is directly judged as "severe fault".

[0095] In the convergence process of time-weighted integrated decision-making and diagnosis, the system introduces an exponentially decaying weight model within a sliding time window to assign weights to daily status labels that increase with time (the ω value in the figure reflects the weight allocation, with recent data having significantly higher weights than historical data), and calculates the weighted percentage index for each degradation level. Based on a comprehensive decision matrix that prioritizes severity, the weighted percentage index is matched step-by-step with preset level thresholds: Between days 12 and 14, the severe fault tag receives a very high weighting due to its temporal proximity, and its weighted percentage index exceeds the severe fault determination threshold. The system ultimately outputs a diagnosis of "serious fault". This diagnostic logic filters out occasional noise on a daily basis through daily aggregation and responds sensitively to the actual degradation trend through time-weighted averaging, achieving robust and continuous evolution from early warning to fault diagnosis. This fully verifies the effectiveness and reliability of this solution in handling continuously evolving faults.

[0096] In this step, the calculated weighted proportion of each degradation level is matched step-by-step with the preset level proportion threshold. This can begin with the highest degradation level, checking if the weighted proportion index of that level reaches the corresponding threshold. If it does, the component is determined to be at that degradation level; otherwise, the next degradation level is matched sequentially until a matching level is found, completing the component status diagnosis. This step-by-step matching decision not only effectively suppresses misjudgments caused by instantaneous abnormal noise (such as severe degradation on a single day with a very low overall proportion), but also takes into account the timeliness of recent degradation trends through a time-weighted model. Furthermore, integrating long-term data within the window ensures the robustness of the diagnosis, ultimately achieving accurate diagnostic output for wind turbine component status and providing a reliable basis for subsequent operation and maintenance decisions.

[0097] The aforementioned health assessment results include abnormal and normal states, with the abnormal state including serious malfunctions, general malfunctions, and minor malfunctions.

[0098] After outputting the health assessment results of the wind turbine, the method also includes: When the health assessment result is abnormal, calculate the contribution of each critical component's degradation value; identify the critical component with the largest contribution as the primary failure component, and generate maintenance recommendations for the primary failure component.

[0099] In this embodiment, when the health assessment result is abnormal, it is necessary to analyze the contribution of the deterioration value of key parts, which is expressed by the following formula: ; in, Let be the degradation value of the z-th critical component. The system uses a identifier number to pinpoint the critical fault location within the wind turbine unit, and `ArgMax(·)` is a function that returns the index of the critical location with the highest contribution to the degradation value, enabling precise fault location. After identifying the primary fault location, based on the component type, degradation level, and historical fault mode library corresponding to that location, and matching the preset maintenance and repair strategy, the system automatically generates personalized maintenance recommendations. These recommendations include fault location, degradation level assessment, repair priority, and targeted maintenance measures (such as inspection, lubrication, and shutdown maintenance). The recommendations are then output to the maintenance terminal, completing the closed loop from fault diagnosis to maintenance guidance and providing accurate and actionable decision support for on-site maintenance personnel. The contribution of the degradation value can be understood as its contribution to the abnormal state.

[0100] For example, if a certain part contains three critical parts with degradation values ​​of B1=2.1, B2=5.3, and B3=3.7, then =2, then the second critical part is the main failure part.

[0101] This application first constructs a feature degradation matrix based on multiple sets of key component degradation values, achieving standardized integration of multi-dimensional degradation information and providing a comprehensive data foundation for graded assessment. Second, by comparing the instantaneous degradation level of each data set with preset grading thresholds, it achieves accurate mapping from continuous degradation values ​​to discrete states, providing standardized input for subsequent aggregation and judgment. Subsequently, it summarizes the instantaneous degradation levels of all data sets within a single day and calculates the sample proportion. Combined with a preset proportion threshold, it determines the representative state label for the day, effectively smoothing out daily anomalies caused by random fluctuations and instantaneous shocks in the industrial field, suppressing false alarm interference, and outputting a result reflecting the state of the day. The system generates stable labels for the actual operating status. Then, within a sliding time window, a time decay factor is used to assign weights to each daily label that increase with time, and a weighted fault ratio index is calculated to highlight the impact of recent real-time degradation trends, thus completely solving the alarm lag and false alarm tailing problems caused by traditional equal-weight time windows. Finally, the weighted ratio index of each fault level is matched with the preset level ratio threshold level by level to output a refined health assessment result that includes both abnormal and normal states. This replaces the traditional binary judgment that can only distinguish between "normal" and "faulty", greatly improving the accuracy, robustness, and engineering practicality of wind turbine health status assessment.

[0102] like Figure 10 The diagram shown is a real-time data illustration of a wind turbine according to an embodiment of this application. This illustration is used for visual monitoring of the real-time operating status of key components of a wind turbine, and is suitable for wind farm operation supervision, health assessment, and fault early warning scenarios.

[0103] In one embodiment, vibration sensors are installed at multiple key components of the wind turbine, including at least the rear bearing V of the main shaft, the front bearing H of the main shaft, the drive end of the gearbox input shaft, the intermediate shaft V of the gearbox, the non-drive end V of the gearbox high-speed shaft, the drive end V of the generator, and the non-drive end V of the generator. Each sensor collects vibration data at its corresponding measuring point in real time and uploads the vibration data to the monitoring terminal.

[0104] The monitoring terminal displays the collected data in real time. Each module in the diagram corresponds to the status information of different components, such as vibration characteristic values ​​of effective velocity and effective acceleration at different frequency bands. For each critical component, the system constructs a feature set related to the component's fault based on the collected multi-parameter vibration data. Each feature set contains at least two basic indicators to enhance the comprehensiveness of the health status assessment. For example, for monitoring the non-drive end of the high-speed shaft of the gearbox, features such as effective velocity (10-1000Hz), effective acceleration (0-10Hz), and energy values ​​within a specific frequency band can be extracted simultaneously.

[0105] In this embodiment, the system further calculates a health baseline value based on the feature value sequence of each original feature within a preset time period. The health baseline value is preferably represented by the median to reduce the impact of instantaneous shocks, outliers, and noise fluctuations on the baseline modeling. Subsequently, the real-time original features are compared with the health baseline value, and the degradation index corresponding to a single original feature is calculated by combining the rounding up operator and the zero-value truncation operation. Then, the degradation indices of multiple original features are nonlinearly aggregated to obtain the degradation value of the corresponding key component.

[0106] Furthermore, the system employs a dynamic time window-driven multi-parameter degradation convergence assessment strategy based on degradation values ​​to evaluate the health status of wind turbines. This strategy includes steps such as daily degradation level mapping, daily status aggregation, time-weighted and proportion calculation, and comprehensive proportion determination. Through this strategy, the system can map the degradation status of multi-dimensional features on a single day to a unified health level, and combine time weights to highlight recent trends, thereby solving the problems of traditional equal-weighted sliding window methods being sensitive to short-term fluctuations and insufficiently responding to continuous degradation trends.

[0107] therefore, Figure 10 The real-time data interface shown can not only intuitively display the real-time vibration status of key parts of the wind turbine, but also provide a data foundation for subsequent degradation assessment, status classification and fault early warning, thereby improving the accuracy and timeliness of wind turbine operation monitoring.

[0108] like Figure 11 The figure shown is a historical data and waveform spectrum diagram of a wind turbine provided in an embodiment of this application. It specifically shows the variation trend of the effective value of acceleration (10-2000Hz) of the non-drive end V of the high-speed shaft of the wind turbine gearbox within a preset time range, as well as the time-domain waveform diagram and frequency-domain waveform diagram at the corresponding time. This figure is used to retrospectively analyze the historical vibration state of key components of the wind turbine to help determine whether its operating state has deteriorated or is abnormal.

[0109] In one embodiment, the system continuously collects acceleration data at a vibration sensor deployed at the non-drive end V of the high-speed shaft of the gearbox, and performs statistical processing on the collected results to obtain... Figure 11 The upper part shows the trend graph of the effective acceleration value (10-2000Hz). This trend graph reflects the change of the effective acceleration value (10-2000Hz) over time at different monitoring moments. As can be seen from the graph, the effective acceleration value (10-2000Hz) at this measuring point exhibits significant fluctuations within the preset time interval, with relatively high peak values ​​appearing at some moments. This indicates that the critical component is affected by periodic loads, meshing impacts, or local abnormal excitations during operation. The system can identify whether there is a continuous accumulation of vibration energy based on this trend change, thereby determining whether the component is showing a deterioration trend.

[0110] Furthermore, the system plots the time-domain waveforms of the original vibration signals acquired at the measuring points at preset times (e.g., ...). Figure 11 As shown in the middle section, a preliminary time-domain analysis was performed. This time-domain waveform reflects the instantaneous amplitude variation characteristics of the vibration signal within the sampling time. As can be seen from the figure, the waveform exhibits significant transient fluctuations in the initial stage, subsequently becoming relatively stable overall but still accompanied by certain low-frequency oscillations and local amplitude variations, indicating that the vibration signal at this monitoring point contains a continuous mechanical excitation component. For high-speed shaft components in wind turbine gearboxes, the periodic fluctuations, impact pulses, and local irregular changes in the time-domain waveform are usually related to the gear meshing state, bearing stress state, or dynamic response of the transmission chain.

[0111] The system also performs frequency domain transformation on the time-domain vibration signal to obtain... Figure 11 The spectrum diagram shown below illustrates the energy distribution of the vibration signal at different frequencies. As can be seen, several discrete spectral peaks exist in the low-frequency and mid-to-high-frequency ranges, indicating that the vibration signal contains multiple specific frequency components. For the non-drive end V of the high-speed shaft of the gearbox, the peak distribution in the spectrum reflects the frequency response caused by rotor imbalance, gear meshing errors, local bearing damage, or other transmission system abnormalities. The system can extract original features such as characteristic frequency amplitude, harmonic amplitude, sideband characteristics, and frequency band energy based on the position, amplitude, and distribution pattern of the spectral peaks in the spectrum, and use these features as input for subsequent degradation index calculations.

[0112] In this embodiment, the system extracts a feature set related to the failure of key components based on the aforementioned historical vibration data. The median of the feature value sequence of each original feature within a preset time period is used as the health baseline value to reduce the impact of instantaneous shocks and anomalies on the baseline modeling. Subsequently, the system calculates the degradation index of each original feature using a rounding up operator and a zero-value truncation operation, and nonlinearly aggregates multiple degradation indices to obtain the degradation value of the non-drive end V of the high-speed shaft of the gearbox. Further, the system employs a dynamic time window-driven multi-parameter degradation convergence evaluation strategy to perform daily degradation grading mapping, daily state aggregation, time weighting and proportion calculation, and comprehensive proportion determination on the degradation value, thereby obtaining the health assessment result of the key component.

[0113] therefore, Figure 11 It can simultaneously reflect the degradation trend, time-domain dynamic characteristics, and frequency-domain response characteristics of key wind turbine components over historical periods, providing a basis for subsequent degradation index calculation and health status assessment based on multi-feature sets. It is suitable for wind turbine operation monitoring, condition diagnosis, and fault prediction scenarios.

[0114] Based on the same inventive concept, this application also provides a wind turbine health assessment device for implementing the aforementioned method. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations of one or more wind turbine health assessment device embodiments provided below can be found in the limitations of the wind turbine health assessment method described above, and will not be repeated here.

[0115] In one exemplary embodiment, such as Figure 12 As shown, a wind turbine health assessment device is provided, the device comprising: Module 510 is used to acquire vibration operation data of wind turbine units; The feature determination module 520 is used to determine the feature set related to the failure of each key component in the wind turbine based on vibration operation data; each feature set includes at least two original features; The degradation index calculation module 530 is used to determine the degradation index of each original feature in the feature set, and to obtain the degradation value of each key component of the wind turbine through nonlinear aggregation. The degradation value is used to characterize the degree of failure degradation of the key component that deviates from the healthy operating state. The degradation index is calculated based on the health benchmark value, combined with the rounding up operator and the zero-value truncation operation. The health benchmark value is the median of the feature value sequence of each original feature in a preset time period. The health assessment module 540 is used to assess the health status of wind turbine units based on the degradation values ​​of each key component, using a dynamic time window-driven multivariate parameter degradation convergence assessment strategy to obtain health assessment results. The multivariate parameter degradation convergence assessment strategy includes: daily degradation classification mapping, daily status aggregation, time weighting and proportion calculation, and comprehensive proportion determination operation.

[0116] As an optional implementation, the feature determination module 520 is specifically used for: The vibration operation data is filtered to obtain the filtered vibration operation data. Basic features are extracted from the filtered vibration operation data; these features include: RMS acceleration value and acceleration kurtosis. Based on the structural parameters of the wind turbine, the filtered vibration operation data were analyzed to obtain the component directivity characteristics of each key part of the wind turbine. Based on the basic characteristics and the component-oriented characteristics corresponding to each key part, a set of features related to the faults of each key component in the wind turbine is constructed.

[0117] As an optional implementation, the degradation index calculation module 530 is specifically used for: For each feature set, obtain the health baseline value and sensitivity adjustment factor of each original feature; the sensitivity adjustment factor is used to characterize the normal fluctuation range of the original feature under the health state. The sensitivity adjustment factor is obtained by multiplying the median of the difference between the original feature and the health baseline value over a preset time period with a scaling coefficient. The difference is obtained by subtracting the feature values ​​of each original feature from the health baseline value; The maximum value is determined from the product of the health baseline value and the minimum value coefficient, and the sensitivity adjustment factor; The result of dividing the difference by the maximum value is processed by the floor operator and then truncated to obtain the degradation index of each original feature.

[0118] As an optional implementation, the degradation index calculation module 530 is also used for: Determine whether health baseline values ​​and sensitivity adjustment factors exist; When there is no health benchmark and sensitivity adjustment factor, the vibration operation data of the wind turbine in the first preset period under normal operating conditions is extracted, and the health benchmark and sensitivity adjustment factor are calculated based on the vibration operation data in the first preset period. When a health baseline and sensitivity adjustment factor exist, obtain the current health baseline and current sensitivity adjustment factor, and determine whether state degradation has occurred. Based on the current health baseline, current sensitivity adjustment factor, and state degradation result, determine the health baseline and sensitivity adjustment factor.

[0119] As an optional implementation, the degradation index calculation module 530 is also used for: When the condition degradation result indicates that no condition degradation has occurred, the vibration operation data of the wind turbine is collected within the second preset cycle, and the health benchmark value and sensitivity adjustment factor are calculated based on the vibration operation data of the second preset cycle. When the condition deterioration result indicates that condition deterioration has occurred and the new health benchmark value is greater than k times the current health benchmark value, a freeze update is initiated, and the current health benchmark value and the current sensitivity adjustment factor are used as the health benchmark value and sensitivity adjustment factor, where k is a positive number; the new health benchmark value is calculated based on the vibration operation data of the wind turbine unit within the second preset cycle.

[0120] As an optional implementation, the degradation index calculation module 530 is also used for: For each critical part in the critical components, the degradation index of each original feature associated with the critical part is nonlinearly aggregated to obtain the degradation value of each critical part. The degradation values ​​of each key component are nonlinearly aggregated using Euclidean distance to obtain the degradation values ​​of the key components.

[0121] As an optional implementation, the health assessment module 540 is specifically used for: Based on the degradation values ​​of all key components corresponding to each set of data within a sliding time window, a feature degradation matrix corresponding to each set of data is constructed. The feature degradation matrix corresponding to each group of data is compared with the preset grading threshold to obtain the instantaneous degradation level of each group of data within a single day. Summarize the instantaneous degradation levels of all groups of data within a single day, calculate the sample proportion of each instantaneous degradation level, and determine the representative status label for the single day based on the comparison results of the sample proportion with the preset proportion threshold. Within the sliding time window, the representative status labels of each day are assigned weights by the time decay factor, the weighted failure ratio index is calculated, and the weighted ratio index of each failure level is output. The weighted percentage index of each fault level is compared with the preset percentage threshold to output the health assessment results of the wind turbine; the health assessment results include abnormal and normal states.

[0122] As an optional implementation, the above-described apparatus is further used for: When the health assessment result is abnormal, calculate the contribution of the deterioration value of each key part; The critical components that contribute the most to the failure are identified as the main failure points, and maintenance recommendations are generated for these main failure points.

[0123] The wind turbine health assessment device provided in this application extracts a multi-dimensional feature set related to the failure of each key component based on the acquired vibration operation data. The feature set corresponding to each key component covers at least two original features, which can provide a comprehensive data foundation for subsequent wind turbine health assessment. The device uses the median of the feature value sequence of each original feature with a preset time period as the health benchmark value. Utilizing the median's resistance to outliers and random fluctuations, it filters out instantaneous impact interference from the industrial site at its source. Simultaneously, it calculates the degradation index by combining the rounding up operator and the zero-value truncation operation. This not only focuses on the true degradation direction and eliminates normal fluctuation interference in the healthy state through the zero-value truncation operation, but also utilizes the rounding up operation... Discretized grading effectively suppresses random fluctuations and instantaneous shocks, enhancing the robustness of degradation quantification. Nonlinear aggregation yields degradation values ​​for each key component, avoiding abnormal dilution and artificially inflated values ​​from linear summation, thus accurately characterizing the degree of component degradation. Finally, a dynamic time window-driven multi-parameter degradation convergence assessment strategy is adopted, incorporating daily degradation grading mapping, daily state aggregation, time weighting and proportion calculation, and comprehensive proportion determination. Daily state aggregation smooths short-term disturbances, and time weighting highlights recent real-time degradation trends, resolving alarm lag and false alarm tailing issues associated with equal-weighted windows. Simultaneously, multi-level grading achieves accurate multi-state assessment, replacing the traditional two-state determination method and significantly improving the accuracy and reliability of wind turbine health status assessment.

[0124] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 13 As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores video tag processing data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a wind turbine health assessment method.

[0125] Those skilled in the art will understand that Figure 13The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0126] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0127] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0128] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0129] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0130] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0131] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchain. The processors involved in the embodiments provided in this application may be, but are not limited to, general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc.

[0132] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0133] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for assessing the health of a wind turbine, characterized in that, The wind turbine health assessment method includes: Obtain vibration operation data of wind turbine units; Based on the vibration operation data, a set of features related to the failures of each key component in the wind turbine is determined; each set of features includes at least two original features. The degradation index of each original feature in the feature set is determined, and the degradation index is nonlinearly aggregated to obtain the degradation value of each key component of the wind turbine. The degradation value is used to characterize the degree of failure degradation of the key component that deviates from the healthy operating state. The degradation index is calculated based on the health benchmark value, combined with the rounding up operator and the zero-value truncation operation. The health benchmark value is the median of the feature value sequence of each original feature in a preset time period. Based on the degradation values ​​of each key component, a dynamic time window-driven multivariate parameter degradation convergence assessment strategy is used to evaluate the health status of the wind turbine and obtain the health assessment results. The multivariate parameter degradation convergence assessment strategy includes: daily degradation classification mapping, daily status aggregation, time weighting and proportion calculation, and comprehensive proportion determination operation. Determining the degradation index of each original feature in the feature set includes: For each feature set, obtain the health baseline value and sensitivity adjustment factor of each original feature; the sensitivity adjustment factor is used to characterize the normal fluctuation range of the original feature under healthy conditions, and the sensitivity adjustment factor is obtained by multiplying the median of the difference between the original feature and the health baseline value in the preset time period with a scaling coefficient; The difference is obtained by subtracting the feature values ​​of each original feature from the health baseline value; The maximum value is determined from the product of the health baseline value and the minimum value coefficient, and the sensitivity adjustment factor; The result of dividing the difference by the maximum value is processed by the rounding up operator and then a zero-value truncation operation is performed to obtain the degradation index of each original feature; Based on the degradation values ​​of each key component, a dynamic time window-driven multivariate parameter degradation convergence assessment strategy is used to evaluate the health status of the wind turbine, resulting in health assessment results, including: Based on the degradation values ​​of all key components corresponding to each set of data within a sliding time window, a feature degradation matrix corresponding to each set of data is constructed. The feature degradation matrix corresponding to each group of data is compared with the preset grading threshold to obtain the instantaneous degradation level of each group of data within a single day. Summarize the instantaneous degradation levels of all groups of data within a single day, calculate the sample proportion of each instantaneous degradation level, and determine the representative status label of the single day based on the comparison result of the sample proportion with the preset proportion threshold. Within the sliding time window, the representative status labels of each day are assigned weights by the time decay factor, the weighted fault percentage index is calculated, and the weighted percentage index of each fault level is output. The weighted percentage index of each fault level is compared with the preset percentage threshold, and the health assessment result of the wind turbine is output; the health assessment result includes abnormal state and normal state.

2. The wind turbine health assessment method according to claim 1, characterized in that, The at least two original features include at least one basic feature and at least one component orientation feature; The basic characteristics are used to reflect the overall energy level and distribution of vibration signals during wind turbine operation; the component directivity characteristics refer to the vibration energy characteristics of the key components at the target fault frequency. Based on the vibration operation data, a set of features related to the failures of various key components in the wind turbine is determined, including: The vibration operation data is filtered to obtain filtered vibration operation data; Basic features are extracted from the filtered vibration operation data; the basic features include: RMS acceleration value and acceleration kurtosis. Based on the structural parameters of the wind turbine, the filtered vibration operation data is analyzed to obtain the component directivity characteristics corresponding to each key part of the wind turbine. Based on the aforementioned basic features and the component-oriented features corresponding to each key part, a feature set related to the faults of each key component in the wind turbine is constructed.

3. The wind turbine health assessment method according to claim 1, characterized in that, The preset time period includes a first preset period or a second preset period, wherein the first preset period is greater than the second preset period; Obtaining the health baseline value and sensitivity adjustment factor of each original feature, including: Determine whether health baseline values ​​and sensitivity adjustment factors exist; When there is no health benchmark and sensitivity adjustment factor, the vibration operation data of the wind turbine in the first preset period under normal operating conditions is extracted, and the health benchmark and sensitivity adjustment factor are calculated based on the vibration operation data in the first preset period. When a health baseline and a sensitivity adjustment factor exist, the current health baseline and the current sensitivity adjustment factor are obtained, and it is determined whether a state degradation has occurred. Based on the current health baseline, the current sensitivity adjustment factor, and the state degradation result, the health baseline and the sensitivity adjustment factor are determined.

4. The wind turbine health assessment method according to claim 3, characterized in that, Based on the current health baseline, the current sensitivity adjustment factor, and the state degradation results, the health baseline and sensitivity adjustment factor are determined, including: When the condition degradation result indicates that no condition degradation has occurred, the vibration operation data of the wind turbine is collected within the second preset cycle, and the health benchmark value and sensitivity adjustment factor are calculated based on the vibration operation data of the second preset cycle. When the condition deterioration result indicates that condition deterioration has occurred and the new health benchmark value is greater than k times the current health benchmark value, a freeze update is initiated, and the current health benchmark value and the current sensitivity adjustment factor are used as the health benchmark value and the sensitivity adjustment factor, where k is a positive number; the new health benchmark value is calculated based on the vibration operation data of the wind turbine unit within the second preset cycle.

5. The wind turbine health assessment method according to claim 1, characterized in that, The degradation indices are nonlinearly aggregated to obtain the degradation values ​​of each key component of the wind turbine, including: For each critical part in the critical components, the degradation index of each original feature associated with the critical part is nonlinearly aggregated to obtain the degradation value of each critical part. The degradation values ​​of each key component are obtained by nonlinear aggregation of Euclidean distance.

6. The wind turbine health assessment method according to claim 1, characterized in that, After outputting the health assessment results of the wind turbine, the method further includes: When the health assessment result is abnormal, calculate the contribution of the deterioration value of each key part; The critical components that contribute the most to the failure are identified as the main failure points, and maintenance recommendations are generated for these main failure points.

7. A wind turbine health assessment device, characterized in that, The wind turbine health assessment device includes: The acquisition module is used to acquire vibration operation data of the wind turbine. The feature determination module is used to determine, based on the vibration operation data, a set of features related to the faults of various key components in the wind turbine; each feature set includes at least two original features; The degradation index calculation module is used to determine the degradation index of each original feature in the feature set, and to obtain the degradation value of each key component of the wind turbine through nonlinear aggregation of the degradation index; the degradation value is used to characterize the degree of failure degradation of the key component deviating from the healthy operating state; the degradation index is calculated based on the health benchmark value, combined with the rounding up operator and the zero-value truncation operation, and the health benchmark value is the median of the feature value sequence of each original feature in a preset time period; The health assessment module is used to assess the health status of the wind turbine based on the degradation values ​​of each key component, using a dynamic time window-driven multivariate parameter degradation convergence assessment strategy, and obtain the health assessment results. The multivariate parameter degradation convergence assessment strategy includes: daily degradation classification mapping, daily status aggregation, time weighting and proportion calculation, and comprehensive proportion determination operation. The degradation index calculation module is specifically used for: For each feature set, obtain the health baseline value and sensitivity adjustment factor of each original feature; the sensitivity adjustment factor is used to characterize the normal fluctuation range of the original feature under healthy conditions, and the sensitivity adjustment factor is obtained by multiplying the median of the difference between the original feature and the health baseline value in the preset time period with a scaling coefficient; The difference is obtained by subtracting the feature values ​​of each original feature from the health baseline value; The maximum value is determined from the product of the health baseline value and the minimum value coefficient, and the sensitivity adjustment factor; The result of dividing the difference by the maximum value is processed by the rounding up operator and then a zero-value truncation operation is performed to obtain the degradation index of each original feature; The health assessment module is specifically used for: Based on the degradation values ​​of all key components corresponding to each set of data within a sliding time window, a feature degradation matrix corresponding to each set of data is constructed. The feature degradation matrix corresponding to each group of data is compared with the preset grading threshold to obtain the instantaneous degradation level of each group of data within a single day. Summarize the instantaneous degradation levels of all groups of data within a single day, calculate the sample proportion of each instantaneous degradation level, and determine the representative status label of the single day based on the comparison result of the sample proportion with the preset proportion threshold. Within the sliding time window, the representative status labels of each day are assigned weights by the time decay factor, the weighted fault percentage index is calculated, and the weighted percentage index of each fault level is output. The weighted percentage index of each fault level is compared with the preset percentage threshold, and the health assessment result of the wind turbine is output; the health assessment result includes abnormal state and normal state.

8. A computer device, comprising: The memory and processor contain a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the steps of the wind turbine health assessment method according to any one of claims 1-6.

Images