A wind turbine generator transmission system fault evaluation method, device, equipment and medium

Abstract

This invention relates to the field of wind power generation technology and discloses a method, device, equipment, and medium for fault assessment of wind turbine transmission systems. The method utilizes a data dimensionality reduction algorithm to reduce the dimensionality of multidimensional raw data, retaining core distinguishing features and simplifying calculations. Then, a data clustering algorithm is used to intelligently classify operating states, and a preliminary fault mode mapping is constructed by combining historical faults. Subsequently, the rationality of clustering is verified using signal source correlation coefficients, and long-term historical operating data is filtered through historical data matching rates to reduce the risk of misjudgment. Next, based on time series analysis and degradation path analysis, the coupling relationship between vibration trend slope, temperature accumulation offset, and torque decay cycle is obtained. Finally, core features are extracted through convolutional neural networks to accurately output the probability distribution and specific location of fault occurrence, thereby improving the accuracy of fault identification in the transmission system of offshore wind turbines and the ability to predict component performance degradation.

Description

Technical Field

[0001] This invention relates to the field of wind power generation technology, specifically to a method, device, equipment, and medium for fault assessment of wind turbine transmission systems. Background Technology

[0002] Offshore wind power, as a pillar industry in the clean energy sector, is crucial for ensuring energy security and achieving carbon neutrality goals through its stable operation. The transmission system, as a core component of offshore wind turbines, directly affects power generation efficiency and equipment lifespan. However, due to the complex environment, challenging data acquisition, and harsh operating conditions of offshore wind power systems, frequent failures and high maintenance costs have become key bottlenecks restricting the industry's development.

[0003] The fault monitoring methods disclosed in related technologies mostly rely on a single data source or a general model. However, the transmission system of offshore wind turbines has variable operating conditions and complex fault modes. The fault monitoring methods disclosed in related technologies are difficult to accurately identify faults, resulting in frequent false alarms and missed alarms. At the same time, they cannot provide early warning of the risk of failure caused by the performance degradation of key components. Summary of the Invention

[0004] This invention provides a method, apparatus, equipment, and medium for fault assessment of wind turbine transmission systems, in order to solve the problem that the fault monitoring methods disclosed in related technologies are difficult to accurately identify faults and provide early warning of the risk of failure caused by the performance degradation of key components.

[0005] In a first aspect, the present invention provides a method for fault assessment of a wind turbine transmission system, comprising:

[0006] Based on the acquired multidimensional raw data of the target wind turbine drive system, a data dimensionality reduction algorithm is used to reduce the dimensionality and obtain a data dimensionality reduction matrix. Then, a data clustering algorithm is used to obtain multiple operating state clustering results. Combined with the historical faults associated with each operating state cluster, a fault mode initial mapping is constructed. The multidimensional raw data includes vibration data, temperature data, and torque data. The data dimensionality reduction matrix includes the principal component of the vibration signal, the temperature distribution weight, and the torque fluctuation factor. Based on the initial mapping of the fault modes, and by integrating historical operating data, the clustering of the operating states is verified using the correlation coefficient of the signal sources, thereby obtaining a set of potential fault modes and a historical data matching rate. Based on long-term historical operating data filtered using the historical data matching rate, the vibration trend slope, temperature cumulative offset, and torque decay cycle are obtained using time series analysis. Combined with degradation path analysis, the coupling relationship between the temperature cumulative offset and the torque decay cycle is obtained, and it is determined whether the vibration trend slope exceeds a preset threshold. When the slope of the vibration trend exceeds a preset threshold, a neural network model is used to perform deep feature extraction on the principal components of the vibration signal and the temperature distribution weights to obtain the probability distribution of the fault occurrence and the location of the fault occurrence.

[0007] Through the above implementation methods, a data dimensionality reduction algorithm is used to reduce the dimensionality of multidimensional original data, retaining core distinguishing features and simplifying calculations. Then, a data clustering algorithm is used to achieve intelligent classification of operating states, and a preliminary mapping of fault modes is constructed by combining historical faults. Subsequently, the rationality of clustering is verified by the correlation coefficient of signal sources, and long-term historical operating data is screened by the historical data matching rate to reduce the risk of misjudgment. Then, based on time series analysis and degradation path analysis, the coupling relationship between vibration trend slope, temperature accumulation offset and torque decay cycle is obtained, creating conditions for early intervention. Finally, the core features are extracted through deep convolutional neural networks to accurately output the probability distribution and specific location of fault occurrence, thereby improving the accuracy of fault identification of the transmission system of offshore wind turbines and the ability to predict component performance degradation.

[0008] In one optional implementation, the step of using time series analysis to obtain the vibration trend slope, temperature cumulative offset, and torque decay period based on long-term historical operating data filtered using the historical data matching rate, combined with degradation path analysis to obtain the coupling relationship between the temperature cumulative offset and the torque decay period, and determining whether the vibration trend slope exceeds a preset threshold includes: Based on the long-term historical operating data filtered using the historical data matching rate, the vibration trend slope, temperature cumulative offset, and torque decay cycle corresponding to each time window are obtained using time series analysis. The component fatigue index is extracted, and the component performance parameter residuals are determined. Based on the residual performance parameters of the component, the fatigue index of the component is extracted using the degradation path segmentation analysis method, and the degradation path of the fatigue index is obtained. Temperature accumulation features are extracted from the degradation path to obtain the distribution pattern of temperature accumulation offset values; Based on the accumulated temperature offset value, the periodic relationship of torque decay is calculated to determine the periodic change trend; Based on the periodic change trend, the numerical characteristics of the vibration trend slope change are calculated, and it is determined whether the vibration trend slope exceeds a preset threshold.

[0009] The above implementation method uses long-term operating data filtered by historical data matching rate as a basis. Time series analysis is used to accurately extract key degradation features such as vibration trend slope, temperature cumulative offset, and torque decay cycle for each time window. The component fatigue index and performance parameter residuals are correlated to ensure the reliability and comprehensiveness of the data source for degradation analysis of each key component. Simultaneously, a complete degradation path for the fatigue index is constructed using a degradation path segmentation analysis method. The distribution pattern of temperature cumulative offset and the periodic relationship of torque decay are extracted from the path, clarifying the coupling evolution logic between multiple parameters. Furthermore, the changing characteristics of the vibration trend slope are quantified through periodic changes, enabling a scientific determination of the trend of escalating vibration anomalies. This effectively improves the foresight and accuracy of fault prediction, providing sufficient time for timely maintenance intervention by staff.

[0010] In one optional implementation, the step of using time series analysis to obtain the vibration trend slope, temperature cumulative offset, and torque decay cycle corresponding to each time window based on long-term historical operating data filtered using the historical data matching rate, and extracting the component fatigue index to determine the component performance parameter residuals includes: Based on the long-term historical operating data filtered using the historical data matching rate, time series characteristics are extracted using time series analysis, and the vibration trend slope is calculated. Based on the vibration trend slope, combined with the temperature cumulative offset analysis method, the superposition effect of vibration trend slope and temperature cumulative offset is obtained; Based on the superposition effect of the vibration trend slope and the temperature cumulative shift, the multi-dimensional change characteristics within each time window are obtained by calculating the torque decay period. Based on the multi-dimensional change characteristics within each time window, the component fatigue index is obtained using the component fatigue index calculation method. By comparing the fatigue index of the component with its performance parameters, the residual performance parameters of the component are obtained.

[0011] Through the above implementation method, based on long-term historical data filtered using historical data matching rate, time series analysis is used to accurately extract time-series features and calculate vibration trend slope. Then, by integrating the superposition effect of vibration trend slope and temperature cumulative offset, the comprehensive impact of multiple factors on the performance of wind turbine transmission system is fully considered. Furthermore, based on the superposition effect, torque decay cycle calculation is combined to form multi-dimensional change features within each time window, comprehensively capturing the dynamic evolution law of wind turbine transmission system operation status. Finally, the fatigue index of components is accurately calculated based on multi-dimensional change features, and the residual of performance parameters is determined by comparison with performance parameters, realizing the quantitative assessment of the fatigue accumulation degree and performance degradation level of key components of wind turbine transmission system.

[0012] In one optional implementation, the step of verifying the operational state clustering based on the initial fault mode mapping, integrating historical operational data, and utilizing signal source correlation coefficients to obtain a set of potential fault modes and a historical data matching rate includes: Based on the initial mapping of the fault modes, the clustering of the operating states is verified by using the correlation coefficient of the signal sources to obtain a set of potential fault modes. Based on the set of potential fault modes, each potential fault mode in the set is matched with historical fault data, and the proportion of the number of successfully matched samples to the total number of historical fault samples is calculated to obtain the historical data matching rate.

[0013] Through the above implementation methods, based on the initial mapping of fault modes, the clustering of operating states is verified by using the correlation coefficient of signal sources, quantifying the consistency of correlation between data from different monitoring, eliminating clustering results with discrete features and logical contradictions, and ensuring the reliability and relevance of the potential fault mode set. Subsequently, by performing feature matching between potential fault modes and historical fault data, the historical data matching rate is quantified by the proportion of successfully matched samples. This reflects both the historical correlation and actual probability of occurrence of potential fault modes, and provides key quantitative basis for subsequent screening of long-term operating data and optimization of fault assessment logic.

[0014] In one alternative implementation, it further includes: Based on the fault occurrence probability distribution, Bayesian inference is used to fuse the torque fluctuation factor with the historical data matching rate to obtain a quantitative risk assessment index and determine the fault risk level of the target wind turbine transmission system.

[0015] Based on the above implementation method, and taking advantage of Bayesian inference, the torque fluctuation factor and historical data matching rate are integrated to combine real-time operating status with historical experience data. At the same time, the risk assessment quantification index is obtained by integrating the fault occurrence probability distribution, torque fluctuation factor and the historical data matching rate. The abstract fault risk is transformed into a risk assessment quantification index, and the fault risk level is further determined. This can accurately reflect the severity and probability of the fault, and provide a clear and explicit risk basis for subsequent operation and maintenance decisions of the wind turbine's transmission system.

[0016] In one optional implementation, the multidimensional raw data of the acquired target wind turbine drive system is used to reduce dimensionality using a data dimensionality reduction algorithm to obtain a data dimensionality reduction matrix. Then, a data clustering algorithm is used to obtain multiple operating state clustering results. Finally, combined with historical faults associated with each operating state cluster, an initial fault mode mapping is constructed, including: Based on the multidimensional raw data of the target wind turbine transmission system, the principal components of the vibration signal, the temperature distribution weight, and the torque fluctuation factor are calculated using a data dimensionality reduction algorithm to obtain the data dimensionality reduction matrix. Based on the data dimensionality reduction matrix, the variance of the feature vector is extracted using a data clustering algorithm to obtain multiple running state clusters; For each motion state cluster, the initial fault mode mapping is obtained by combining the historical faults associated with each motion state cluster.

[0017] Through the above implementation method, for the multidimensional raw data of the target wind turbine drive system, the principal components of vibration signals, temperature distribution weights, and torque fluctuation factors are extracted using a data dimensionality reduction algorithm. While retaining each discriminative feature, the data dimensionality reduction is completed, which simplifies the subsequent calculation complexity and avoids the interference of redundant information on the analysis results. Then, based on the data dimensionality reduction matrix, the variance of the feature vector is extracted using a data clustering algorithm, and the system is divided into multiple wind turbine drive system operation state clusters. This realizes the transformation from discrete data to structured state classification and clearly defines the feature boundaries of different operation states. Furthermore, by connecting each operation state cluster with the associated historical fault data, an initial fault mode mapping is constructed, providing a clear preliminary direction for subsequent fault verification and accurate assessment.

[0018] In one alternative implementation, it further includes: Based on the acquired multidimensional raw data of the target wind turbine transmission system, wavelet transform is used to denoise the multidimensional raw data to obtain denoised multidimensional data. Based on the denoised multidimensional data, standard deviation normalization is used to obtain a normalized dataset; the normalized dataset is used to calculate the principal components of the vibration signal, the temperature distribution weight, and the torque fluctuation factor using principal component analysis.

[0019] Through the above implementation methods, wavelet transform is used to denoise the environmental interference noise contained in the multidimensional raw data of the wind turbine drive system, avoiding the interference of environmental interference noise on subsequent feature extraction and fault analysis. Then, through standard deviation normalization, the denoised multidimensional data is transformed into a standardized dataset of uniform magnitude, eliminating the calculation bias caused by the difference in data dimensions, ensuring that the data of each dimension has equal weight in the subsequent principal component analysis, and laying a standardized data foundation for the accurate calculation of the principal components of vibration signals, temperature distribution weights, and torque fluctuation factors.

[0020] Secondly, the present invention provides a fault assessment device for a wind turbine transmission system, the device comprising: The fault mapping construction module is used to reduce the dimensionality of the acquired multidimensional raw data of the target wind turbine drive system using a data dimensionality reduction algorithm to obtain a data dimensionality reduction matrix. Then, a data clustering algorithm is used to obtain multiple operating state clustering results. Combined with the historical faults associated with each operating state cluster, a fault mode initial mapping is constructed. The multidimensional raw data includes vibration data, temperature data, and torque data. The data dimensionality reduction matrix includes the principal component of the vibration signal, the temperature distribution weight, and the torque fluctuation factor. The fault mapping verification module is used to verify the clustering of the operating state based on the initial fault mode mapping, combined with historical operating data, and using the signal source correlation coefficient to obtain a set of potential fault modes and historical data matching rate. The vibration trend assessment module is used to obtain the vibration trend slope, temperature cumulative offset, and torque decay period based on long-term historical operating data filtered by the historical data matching rate using time series analysis. Combined with the degradation path analysis method, it obtains the coupling relationship between the temperature cumulative offset and the torque decay period, and determines whether the vibration trend slope exceeds a preset threshold. The fault assessment output module is used to extract deep features from the principal components of the vibration signal and the temperature distribution weights using a neural network model when the vibration trend slope exceeds a preset threshold, so as to obtain the fault occurrence probability distribution and the fault location.

[0021] Thirdly, the present invention provides an electronic device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the wind turbine drive system fault assessment method of the first aspect or any corresponding embodiment described above.

[0022] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to execute the wind turbine drive system fault assessment method of the first aspect or any corresponding embodiment described above. Attached Figure Description

[0023] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0024] Figure 1 This is a schematic diagram of the first type of fault assessment method for wind turbine transmission system according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the second process of the wind turbine transmission system fault assessment method according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the third process of the wind turbine transmission system fault assessment method according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the fourth process of the wind turbine transmission system fault assessment method according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the fifth process of the wind turbine transmission system fault assessment method according to an embodiment of the present invention; Figure 6 This is a structural block diagram of a wind turbine transmission system fault assessment device according to an embodiment of the present invention; Figure 7 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0027] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0028] The relevant technologies for fault monitoring of the transmission system of offshore wind turbines have the following main shortcomings: 1. Real-time acquisition of multi-dimensional data of the transmission system of offshore wind turbines and the technical bottlenecks limited by the marine environment; 2. A clear empirical framework has not yet been established for understanding the correlation between failure modes of the transmission system of offshore wind turbines and the performance degradation of key components. 3. The definition of the normal range of operating parameters of key components is vague, resulting in a lack of reliable basis for triggering the early warning mechanism.

[0029] To address the aforementioned shortcomings, this invention provides a fault assessment method for wind turbine drive systems. It utilizes a data dimensionality reduction algorithm to reduce the dimensionality of multidimensional raw data, preserving core distinguishing features and simplifying calculations. Then, a data clustering algorithm is used to intelligently classify operating states, and a preliminary fault mode mapping is constructed by combining historical faults. Subsequently, the rationality of clustering is verified using signal source correlation coefficients, and long-term historical operating data is filtered through historical data matching rates to reduce the risk of misjudgment. Next, based on time series analysis and degradation path analysis, the coupling relationship between vibration trend slope, temperature accumulation shift, and torque decay cycle is obtained, creating conditions for early intervention. Finally, core features are extracted through convolutional neural networks to accurately output the probability distribution and specific location of fault occurrence, improving the accuracy of fault identification in offshore wind turbine drive systems and the ability to predict component performance degradation.

[0030] According to an embodiment of the present invention, a method for fault assessment of wind turbine transmission system is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0031] This embodiment provides a method for fault assessment of wind turbine transmission systems, which can be used in the backend server of offshore wind turbines. Figure 1 This is a flowchart of a wind turbine drive system fault assessment method according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps: S101, based on the acquired multidimensional raw data of the target wind turbine drive system, a data dimensionality reduction algorithm is used to reduce the dimensionality to obtain a data dimensionality reduction matrix. Then, a data clustering algorithm is used to obtain multiple operating state clustering results. Combined with the historical faults associated with each operating state cluster, a fault mode initial mapping is constructed. The multidimensional raw data includes vibration data, temperature data, and torque data. The data dimensionality reduction matrix includes the principal component of the vibration signal, the temperature distribution weight, and the torque fluctuation factor.

[0032] Multidimensional raw data refers to various types of basic data collected from the target wind turbine's transmission system that reflect the equipment's operating status. Specifically, this includes vibration data, temperature data, and torque data. Vibration data, collected by sensors, reflects the operational stability of components such as gears and bearings, for example, whether there are abnormal vibrations caused by shaking or wear. Temperature data monitors the temperature of key parts of the transmission system, reflecting heat dissipation effectiveness or friction conditions, such as temperature changes at the gearbox and bearings. Torque data characterizes the magnitude and stability of the torque transmitted by the transmission system, reflecting whether the power transmission efficiency is normal.

[0033] Data dimensionality reduction algorithms are methods used for data dimensionality reduction and feature extraction, including principal component analysis, linear discriminant analysis (LDA), or autoencoders. In this embodiment of the invention, the data dimensionality reduction algorithm is implemented as principal component analysis. By retaining the core information of the data, redundant information is eliminated and the computational complexity is simplified. In this embodiment of the invention, the most representative key features are extracted from three types of raw data: vibration, temperature, and torque, thus avoiding computational redundancy and interference caused by multi-dimensional data.

[0034] The data dimensionality reduction matrix is ​​the result of a data dimensionality reduction algorithm. It is a structured dataset containing three core features extracted and integrated: vibration signal principal components, temperature distribution weights, and torque fluctuation factors. The vibration signal principal components are key vibration features extracted from vibration data to distinguish between normal and abnormal states of equipment, such as vibration amplitude within a specific frequency range. Temperature distribution weights are parameters that quantify the degree to which temperature in different parts of the transmission system affects the equipment's operating state; for example, the temperature weight at bearings is higher than at other non-critical parts. The torque fluctuation factor is an indicator used to characterize the stability of torque data, reflecting whether there are abnormal fluctuations in torque during power transmission, such as fluctuation amplitude and frequency.

[0035] Data clustering algorithms are machine learning algorithms that automatically divide data into multiple clusters based on the similarity of data features. Data clustering algorithms include K-means clustering, density-based spatial clustering of applications with noise (DBSCAN), and hierarchical clustering. This invention uses K-means clustering as an example. By classifying the core feature data in the data dimensionality reduction matrix, the operating status of the equipment is ultimately divided into several categories, such as normal status, slightly abnormal status, and severely abnormal status, achieving the classification from discrete data to structured status.

[0036] The operating status clustering result is the classification result of the equipment operating status output by the K-means clustering algorithm. Each operating status clustering result represents a class of operating states with similar characteristics. For example, the cluster of "vibration principal component variance < 0.1 and temperature weight distribution uniform" corresponds to "normal operating state", which is used to subsequently determine the equipment status corresponding to real-time data.

[0037] Historical faults refer to fault case data recorded during the historical operation of the transmission system of the target wind turbine, including vibration, temperature, and torque characteristics at the time of the fault, fault type, fault occurrence scenario, and other information, which are used to facilitate the subsequent construction of initial fault mode mapping.

[0038] Fault mode initial mapping is a correspondence formed by associating and matching "operating state clusters" with "historical faults". For example, "minor anomaly clustering and bearing temperature weight increase" corresponds to "early bearing wear" fault. The abstract clustering results are used to transform potential fault indications with practical engineering significance, and to preliminarily determine the fault type that may correspond to the current operating state of the equipment, providing direction for subsequent accurate verification and analysis.

[0039] For example, the initial mapping of fault modes is constructed using the K-means clustering algorithm, including: a1 sets the initial number of clusters, combines equipment mechanisms and expert experience, maps the abstract clustering results to the actual operating status, and assigns status labels to the cluster categories; The cluster categories are as follows: Cluster 1 corresponds to the normal state, characterized by "vibration principal component variance < 0.1, uniform temperature weight distribution (maximum weight < 0.3), and torque fluctuation factor < 5%"; Cluster 2 corresponds to slight anomalies, characterized by "vibration principal component variance 0.1-0.3, locally increased temperature weight (e.g., bearing position weight > 0.4), and torque fluctuation factor 5%-10%"; Cluster 3 corresponds to severe anomalies, characterized by "vibration principal component variance > 0.3, concentrated temperature weight (at a certain location > 0.6), and torque fluctuation factor > 10%".

[0040] Furthermore, by randomly selecting some clustered samples and comparing them with actual operation and maintenance records, for example, whether cluster 3 samples correspond to the state before the historical fault, the parameters of the clustering algorithm are adjusted based on the comparison results.

[0041] a2, calculate the feature mean or range for each cluster category to form a state feature template; For example, typical characteristics of "normal state" are: vibration principal component 1 mean = 0.2 ± 0.05, bearing temperature weight = 0.2 ± 0.03, torque fluctuation factor = 3% ± 1%, which are used for subsequent state matching of real-time data.

[0042] a3, by combining historical failure cases and equipment physical models, establishes a "state-failure" mapping table based on the correlation between cluster categories and historical failures, and finally forms a structured initial mapping result; among them, historical failure cases include the characteristic patterns of bearing wear in historical data, and equipment physical models include vibrations of specific frequencies caused by gear failures; Specifically, for cluster 2, which corresponds to minor anomalies, if it is accompanied by "prominent high-frequency vibration principal components and increased bearing temperature weight", the initial fault mapping is "early bearing wear"; if it is accompanied by "periodic changes in torque fluctuation factors and increased gearbox temperature weight", the initial fault mapping is "poor gear meshing". For cluster 3, which corresponds to severe anomalies, if it is accompanied by "severe fluctuations in multi-dimensional features", the initial fault mapping is "potentially fatal fault (such as shaft jamming)".

[0043] For example, a certain motion state cluster is divided into: cluster 2 corresponding to slight anomalies, and the typical characteristics of cluster 2 include: the variance of the principal component of the vibration signal is 0.25, the bearing temperature weight is 0.42, and the torque fluctuation factor is 8%; then the corresponding initial mapping of the fault modes is: early bearing wear (matching degree 65%) and poor gear meshing (matching degree 30%).

[0044] S102, based on the initial mapping of the fault modes, and by integrating historical operating data, the clustering of the operating states is verified using the correlation coefficient of the signal sources, thereby obtaining a set of potential fault modes and a historical data matching rate.

[0045] Historical operating data refers to the operating data accumulated during the historical operation of the target wind turbine drive system. It includes not only the state data such as vibration, temperature, and torque during normal operation, but also the characteristic data before and after the occurrence of a fault, as well as operation and maintenance records.

[0046] The source correlation coefficient is a numerical indicator used to quantify the degree of correlation between two or more monitoring signals. For example, the Pearson correlation coefficient. In the case of bearing wear in the transmission system of a wind turbine, vibration signals and bearing temperature signals show a strong correlation. The source correlation coefficient can be used to verify whether the characteristics of signals within a cluster under the same operating condition are consistent.

[0047] The potential fault mode set is a combination of reliable fault modes remaining after verifying the fault modes using the correlation coefficient of the signal source and eliminating those corresponding to clusters with discrete features or logical contradictions. For example, after eliminating faults that indicate "contradictions between cluster features and signal correlation," a set of reliable potential fault types such as "early bearing wear" or "poor gear meshing" is retained.

[0048] Historical data matching rate refers to the percentage of successfully matched samples out of the total number of historical failure samples, calculated by comparing the features of each failure mode in the potential failure mode set with failure cases in historical operating data. For example, if the "early bearing wear" failure matches 85 features in 100 historical failure data points, then the historical data matching rate for this mode is 85%.

[0049] For example, in the initial fault mode mapping stage, operating status data of the wind turbine's drive system equipment, such as temperature, vibration, and current signals, are first collected. Then, the correlation between each signal is calculated using the signal source correlation coefficient. Assuming 10 signals are collected, the correlation coefficient matrix is ​​calculated using the Pearson correlation coefficient. The correlation coefficient between signal A and signal B is 85, indicating a high correlation between them. Next, the K-means clustering algorithm is used to cluster the operating states, setting the number of clusters to 3. The data is divided into three categories by calculating the Euclidean distance, representing normal state, minor anomaly, and severe anomaly. When obtaining the set of potential fault modes, historical fault data is considered. Assuming the historical data contains 5 known fault modes, by matching the clustering results with the historical data, it is found that the third operating state has a 90% match with historical fault mode 1, and is preliminarily identified as potential fault mode 1. Finally, the historical data matching rate is calculated. Assuming there are 100 records in the historical data, 80 of them match potential fault mode 1, resulting in a matching rate of 80%, thus confirming the historical data matching rate corresponding to this fault mode.

[0050] S103. Based on the long-term historical operating data filtered using the historical data matching rate, the vibration trend slope, temperature cumulative offset, and torque decay cycle are obtained using time series analysis. Combined with degradation path analysis, the coupling relationship between the temperature cumulative offset and the torque decay cycle is obtained, and it is determined whether the vibration trend slope exceeds a preset threshold.

[0051] The filtered long-term historical operating data refers to the set of operating data retained after removing invalid data with mismatched features or abnormal interference from the long-term historical operating data based on the historical data matching rate as the screening standard. It is used to truly reflect the normal operation and performance degradation patterns of the transmission system equipment and to provide a reliable data source for subsequent analysis.

[0052] Time series analysis is a method of analyzing time-series data to extract the evolution characteristics of equipment status over time from long-term historical operating data and to calculate the key degradation indicators of the wind turbine drive system.

[0053] The vibration trend slope is a quantitative indicator that reflects the trend of vibration signal changes over time, calculated using time series analysis. When the vibration trend slope is positive, a larger value indicates more severe vibration in the wind turbine's transmission system; a slope close to 0 indicates a stable vibration state.

[0054] Temperature cumulative deviation refers to the cumulative deviation between the actual temperature of key parts of the wind turbine transmission system and the temperature reference value during normal operation within a certain time range. It is used to quantify the cumulative effect of long-term temperature deviation from the normal range and reflect the degree of degradation of the equipment's heat dissipation or friction condition.

[0055] The torque decay period refers to the periodic characteristic of torque stability gradually decreasing after frequency analysis of torque data using time series analysis. It is used to reflect the degradation law of torque stability during power transmission. The shorter the period, the more frequent the torque decay and the more unstable the performance.

[0056] Degradation path analysis is a segmented analysis method for the equipment performance degradation process. It divides the entire degradation process from normal operation to failure into different stages such as "normal stage - slow degradation stage - accelerated degradation stage - pre-failure stage", which facilitates the subsequent mining of the characteristic patterns and parameter correlations of each stage.

[0057] The coupling relationship refers to the quantitative relationship between the cumulative temperature offset and the torque decay cycle, which reflects the degradation logic of the equipment under the combined effect of multiple parameters. For example, for every 1°C increase in the cumulative temperature offset, the torque decay cycle is shortened by 5 hours.

[0058] The preset threshold is a judgment threshold determined by combining the upper limit of the confidence interval of the vibration trend slope under normal conditions in long-term historical operating data with the critical slope data before the occurrence of historical faults. When the actual calculated vibration trend slope exceeds this threshold, it indicates that the vibration has shown an abnormally intensified trend, and the equipment may have entered an accelerated degradation stage.

[0059] S104, when the slope of the vibration trend exceeds a preset threshold, a neural network model is used to perform deep feature extraction on the principal components of the vibration signal and the temperature distribution weights to obtain the probability distribution of the fault occurrence and the location of the fault occurrence.

[0060] Neural network models, including convolutional neural networks, recurrent neural networks (RNNs), long short-term memory networks (LSTMs), or transformer models, are used to deeply mine core feature data of device operation and accurately identify feature combinations related to faults. For example, the neural network model in this embodiment uses a convolutional neural network, whose structure includes three convolutional layers, using 32, 64, and 128 3×3 convolutional kernels respectively. Each convolutional layer is followed by a ReLU activation function and a max-pooling layer. Finally, two fully connected layers output the fault probability distribution. During network training, the Adam optimizer is used with a learning rate of 0.01, a batch size of 32, and 200 training iterations. The probability of fault occurrence at each location is calculated using the Softmax function. When the fault probability at a certain location exceeds 7, the corresponding location is marked as a potential fault point, and a diagnostic report containing the fault location coordinates and confidence level is generated, achieving precise fault localization.

[0061] The wind turbine drive system fault assessment method provided in this embodiment utilizes a data dimensionality reduction algorithm to reduce the dimensionality of multidimensional raw data, retaining core distinguishing features and simplifying calculations. Then, a data clustering algorithm is used to intelligently classify operating states, and a preliminary fault mode mapping is constructed by combining historical faults. Subsequently, the rationality of clustering is verified using signal source correlation coefficients, and long-term historical operating data is filtered through historical data matching rates to reduce the risk of misjudgment. Next, based on time series analysis and degradation path analysis, the coupling relationship between vibration trend slope, temperature accumulation shift, and torque decay cycle is obtained, creating conditions for early intervention. Finally, a neural network model is used to deeply extract core features, accurately outputting the probability distribution and specific location of fault occurrence. This improves the accuracy of fault identification in the drive system of offshore wind turbines and the ability to predict component performance degradation, overcoming the shortcomings of publicly disclosed fault monitoring methods in related technologies, which struggle to accurately identify faults and provide early warnings of fault risks caused by performance degradation of key components.

[0062] This embodiment provides a method for fault assessment of wind turbine transmission systems, which can be used in the backend server of offshore wind turbines. Figure 2 This is a flowchart of a wind turbine drive system fault assessment method according to an embodiment of the present invention, such as... Figure 2 As shown, the process includes the following steps: S201. Based on the acquired multidimensional raw data of the target wind turbine drive system, a data dimensionality reduction algorithm is used to reduce the dimensionality and obtain a data dimensionality reduction matrix. Then, a data clustering algorithm is used to obtain multiple operating state clustering results. Combined with the historical faults associated with each operating state cluster, a fault mode initial mapping is constructed. The multidimensional raw data includes vibration data, temperature data, and torque data. The data dimensionality reduction matrix includes the principal component of the vibration signal, the temperature distribution weight, and the torque fluctuation factor.

[0063] Specifically, S201 above includes: S2011, based on the multidimensional raw data of the target wind turbine transmission system, the principal components of the vibration signal, the temperature distribution weight and the torque fluctuation factor are calculated using a data dimensionality reduction algorithm to obtain the data dimensionality reduction matrix; S2012, Based on the data dimensionality reduction matrix, the variance of the feature vector is extracted using a data clustering algorithm to obtain multiple running state clusters; S2013, for each motion state cluster, combined with the historical faults associated with each running state cluster, the initial fault mode mapping is obtained.

[0064] For example, the above S201 can be implemented as follows: Based on the multidimensional raw data of the target wind turbine drive system, principal component analysis (PCA) was used to reduce the dimensionality of vibration signals, temperature distribution, and torque fluctuations. First, the vibration signal underwent Fourier transform to extract frequency domain features, forming a feature matrix. Assuming a vibration signal sampling frequency of 10kHz and a signal length of 1000 points, PCA was used to calculate the first three principal components, with contribution rates of 65%, 25%, and 10%, respectively. Next, the temperature distribution data was standardized. Assuming eight temperature sensors, each collecting 100 temperature values, PCA was used to calculate the weights of the temperature distribution, resulting in contribution rates of 70% and 20% for the first two principal components. Then, the torque fluctuation data was normalized. Assuming a torque sensor sampling frequency of 1kHz and a signal length of 500 points, PCA was used to calculate the torque fluctuation factor, resulting in contribution rates of 80% and 15% for the first two principal components. The vibration signal principal components, temperature distribution weights, and torque fluctuation factor were combined into a 3x3 data dimensionality reduction matrix. Next, the variance of the feature vectors is extracted from the data dimensionality reduction matrix. Assuming the variances are 85, 75, and 65, the operating states are clustered using the K-means clustering algorithm, with a cluster size of 3, resulting in three cluster centers representing normal operating state, minor fault state, and severe fault state, respectively. Finally, based on the clustering results, an initial fault mode mapping is obtained. Assuming that minor fault state is related to bearing wear and severe fault state is related to gear fracture, an initial fault mode mapping table is formed.

[0065] Based on the multidimensional raw data of the target wind turbine drive system, a data dimensionality reduction algorithm was used to extract the principal components of the vibration signal, the temperature distribution weight, and the torque fluctuation factor. This data dimensionality reduction was completed while retaining the discriminative features, simplifying the subsequent computational complexity and avoiding interference from redundant information in the analysis results. Then, based on the data dimensionality reduction matrix, a data clustering algorithm was used to extract the variance of the feature vectors and divide the system into multiple operating state clusters of the wind turbine drive system. This transformed the discrete data into a structured state classification and clearly defined the feature boundaries of different operating states. Finally, by connecting each operating state cluster with the associated historical fault data, an initial fault mode mapping was constructed, providing a clear preliminary direction for subsequent fault verification and accurate assessment.

[0066] Furthermore, this also includes: Based on the acquired multidimensional raw data of the target wind turbine transmission system, wavelet transform is used to denoise the multidimensional raw data to obtain denoised multidimensional data. Based on the denoised multidimensional data, standard deviation normalization is used to obtain a normalized dataset; the normalized dataset is used to calculate the principal components of the vibration signal, the temperature distribution weight, and the torque fluctuation factor through a data dimensionality reduction algorithm.

[0067] Vibration, temperature, and torque signals acquired from the drive system of the target wind turbine using a sensor network are used to construct multidimensional raw data, resulting in a raw dataset. Wavelet transform is then applied to denoise the raw dataset, yielding a denoised dataset. Standard deviation normalization is then used to process the denoised dataset, resulting in a normalized dataset.

[0068] For example, vibration, temperature, and torque signals are acquired from an offshore wind power transmission system via a sensor network. An accelerometer collects vibration signals with a frequency range of 0 to 1000 Hz, a temperature sensor collects temperature signals with an accuracy of ±1℃, and a torque sensor collects torque signals with a range of 0 to 10000 N·m. Wavelet transform is used to denoise the acquired multidimensional data, selecting the Daubechies wavelet basis function and decomposing it into 5 layers. Noise is removed using a soft thresholding method with a threshold set to 0.5, resulting in a denoised dataset. The denoised dataset is then normalized using standard deviation. The standard deviation for each signal is calculated: 12 for vibration, 5℃ for temperature, and 500 N·m for torque. Each signal value is divided by its standard deviation to obtain a normalized dataset. The normalized vibration signal range is -1 to 1, the temperature signal range is -8 to 8℃, and the torque signal range is -2 to 2 N·m.

[0069] S202, based on the initial fault mode mapping, and by integrating historical operating data, the operating state clustering is verified using the signal source correlation coefficient to obtain the potential fault mode set and the historical data matching rate. For details, please refer to [link to relevant documentation]. Figure 1 S102 of the illustrated embodiment will not be described again here.

[0070] S203: Based on long-term historical operating data filtered using the historical data matching rate, time series analysis is used to obtain the vibration trend slope, temperature cumulative offset, and torque decay period. Combined with degradation path analysis, the coupling relationship between the temperature cumulative offset and the torque decay period is obtained, and it is determined whether the vibration trend slope exceeds a preset threshold. For details, please refer to [link to relevant documentation]. Figure 1 S103 of the illustrated embodiment will not be described again here.

[0071] S204, when the vibration trend slope exceeds a preset threshold, a neural network model is used to perform deep feature extraction on the principal components of the vibration signal and the temperature distribution weights to obtain the fault occurrence probability distribution and the fault location. For details, please refer to [link to relevant documentation]. Figure 1 S104 of the illustrated embodiment will not be described again here.

[0072] The wind turbine drive system fault assessment method provided in this embodiment utilizes a data dimensionality reduction algorithm to reduce the dimensionality of multidimensional raw data, retaining core distinguishing features and simplifying calculations. Then, a data clustering algorithm is used to intelligently classify operating states, and a preliminary fault mode mapping is constructed by combining historical faults. Subsequently, the rationality of clustering is verified using signal source correlation coefficients, and long-term historical operating data is filtered through historical data matching rates to reduce the risk of misjudgment. Next, based on time series analysis and degradation path analysis, the coupling relationship between vibration trend slope, temperature accumulation shift, and torque decay cycle is obtained, creating conditions for early intervention. Finally, a neural network model is used to deeply extract core features, accurately outputting the probability distribution and specific location of fault occurrence. This improves the accuracy of fault identification in the drive system of offshore wind turbines and the ability to predict component performance degradation, overcoming the shortcomings of publicly disclosed fault monitoring methods in related technologies, which struggle to accurately identify faults and provide early warnings of fault risks caused by performance degradation of key components.

[0073] This embodiment provides a method for fault assessment of wind turbine transmission systems, which can be used in the backend server of offshore wind turbines. Figure 3 This is a flowchart of a wind turbine drive system fault assessment method according to an embodiment of the present invention, such as... Figure 3 As shown, the process includes the following steps: S301, based on the acquired multidimensional raw data of the target wind turbine drive system, a data dimensionality reduction algorithm is used to reduce the dimensionality, resulting in a data dimensionality reduction matrix. Then, a data clustering algorithm is used to obtain multiple operating state clustering results. Combined with historical faults associated with each operating state cluster, a fault mode initial mapping is constructed. The multidimensional raw data includes vibration data, temperature data, and torque data. The data dimensionality reduction matrix includes the principal components of the vibration signal, temperature distribution weights, and torque fluctuation factors. For details, please refer to [link to relevant documentation]. Figure 1 S101 of the illustrated embodiment will not be described again here.

[0074] S302, based on the initial mapping of the fault modes, and by integrating historical operating data, the clustering of the operating states is verified using the correlation coefficient of the signal sources, thereby obtaining a set of potential fault modes and a historical data matching rate.

[0075] Specifically, S302 above includes: S3021, Based on the initial mapping of the fault modes, the clustering of the operating states is verified by using the correlation coefficient of the signal sources to obtain a set of potential fault modes; S3022, Based on the set of potential fault modes, perform feature matching between each potential fault mode in the set of potential fault modes and historical fault data, calculate the proportion of the number of successfully matched samples to the total number of historical fault samples, and obtain the historical data matching rate.

[0076] Based on the initial mapping of fault modes, the clustering of operating states is verified by using the correlation coefficient of signal sources. This quantifies the consistency of the correlation between data from different monitoring systems, eliminates clustering results with discrete features and logical contradictions, and ensures the reliability and relevance of the potential fault mode set. Subsequently, by matching the potential fault modes with historical fault data, the historical data matching rate is quantified by the proportion of successfully matched samples. This reflects both the historical correlation and actual probability of occurrence of potential fault modes, and provides key quantitative basis for subsequent screening of long-term operating data and optimization of fault assessment logic.

[0077] S303: Based on long-term historical operating data filtered using the historical data matching rate, time series analysis is used to obtain the vibration trend slope, temperature cumulative offset, and torque decay period. Combined with degradation path analysis, the coupling relationship between the temperature cumulative offset and the torque decay period is obtained, and it is determined whether the vibration trend slope exceeds a preset threshold. For details, please refer to [link to relevant documentation]. Figure 1 S103 of the illustrated embodiment will not be described again here.

[0078] S304, when the vibration trend slope exceeds a preset threshold, a neural network model is used to perform deep feature extraction on the principal components of the vibration signal and the temperature distribution weights to obtain the fault occurrence probability distribution and the fault location. For details, please refer to... Figure 1 S104 of the illustrated embodiment will not be described again here.

[0079] The wind turbine drive system fault assessment method provided in this embodiment utilizes a data dimensionality reduction algorithm to reduce the dimensionality of multidimensional raw data, retaining core distinguishing features and simplifying calculations. Then, a data clustering algorithm is used to intelligently classify operating states, and a preliminary fault mode mapping is constructed by combining historical faults. Subsequently, the rationality of clustering is verified using signal source correlation coefficients, and long-term historical operating data is filtered through historical data matching rates to reduce the risk of misjudgment. Next, based on time series analysis and degradation path analysis, the coupling relationship between vibration trend slope, temperature accumulation shift, and torque decay cycle is obtained, creating conditions for early intervention. Finally, a neural network model is used to deeply extract core features, accurately outputting the probability distribution and specific location of fault occurrence. This improves the accuracy of fault identification in the drive system of offshore wind turbines and the ability to predict component performance degradation, overcoming the shortcomings of publicly disclosed fault monitoring methods in related technologies, which struggle to accurately identify faults and provide early warnings of fault risks caused by performance degradation of key components.

[0080] This embodiment provides a method for fault assessment of wind turbine transmission systems, which can be used in the backend server of offshore wind turbines. Figure 4 This is a flowchart of a wind turbine drive system fault assessment method according to an embodiment of the present invention, such as... Figure 4 As shown, the process includes the following steps: S401, based on the acquired multidimensional raw data of the target wind turbine drive system, a data dimensionality reduction algorithm is used to reduce the dimensionality, resulting in a data dimensionality reduction matrix. Then, a data clustering algorithm is used to obtain multiple operating state clustering results. Combined with historical faults associated with each operating state cluster, a fault mode initial mapping is constructed. The multidimensional raw data includes vibration data, temperature data, and torque data. The data dimensionality reduction matrix includes the principal components of the vibration signal, temperature distribution weights, and torque fluctuation factors. For details, please refer to [link to relevant documentation]. Figure 1 S101 of the illustrated embodiment will not be described again here.

[0081] S402, based on the initial fault mode mapping, and by integrating historical operating data, the operating state clustering is verified using the signal source correlation coefficient to obtain the potential fault mode set and the historical data matching rate. For details, please refer to [link to relevant documentation]. Figure 1 S102 of the illustrated embodiment will not be described again here.

[0082] S403, based on the long-term historical operating data filtered using the historical data matching rate, the vibration trend slope, temperature cumulative offset, and torque decay cycle are obtained using time series analysis. Combined with degradation path analysis, the coupling relationship between the temperature cumulative offset and the torque decay cycle is obtained, and it is determined whether the vibration trend slope exceeds a preset threshold.

[0083] Specifically, S403 includes: S4031, based on the long-term historical operating data filtered using the historical data matching rate, the vibration trend slope, temperature cumulative offset and torque decay cycle corresponding to each time window are obtained by using time series analysis, and the component fatigue index is extracted to determine the component performance parameter residuals. S4032, Based on the residual performance parameters of the component, the fatigue index of the component is extracted using the degradation path segmentation analysis method to obtain the degradation path of the fatigue index; S4033, extract temperature accumulation features from the degradation path to obtain the distribution law of temperature accumulation offset value; S4034, Based on the accumulated temperature offset value, calculate the periodic relationship of torque decay and determine the periodic change trend; S4035, based on the periodic change trend, calculate the numerical characteristics of the vibration trend slope change, and determine whether the vibration trend slope exceeds a preset threshold.

[0084] By using long-term operational data filtered through historical data matching rates as a foundation, key degradation features such as vibration trend slope, temperature cumulative offset, and torque decay cycle of each time window are accurately extracted through time series analysis. The component fatigue index and performance parameter residuals are then correlated to ensure the reliability and comprehensiveness of the data source for degradation analysis of each key component. Simultaneously, a complete degradation path of the fatigue index is constructed using a degradation path segmentation analysis method. The distribution pattern of temperature cumulative offset and the periodic relationship of torque decay are extracted from the path, clarifying the coupling evolution logic between multiple parameters. Furthermore, the changing characteristics of the vibration trend slope are quantified through periodic changes, enabling a scientific determination of the trend of escalating vibration anomalies. This effectively improves the foresight and accuracy of fault prediction, providing sufficient time for timely maintenance intervention by staff.

[0085] For example, S4031 above includes: b1. Based on the long-term historical operating data filtered using the historical data matching rate, time series features are extracted using time series analysis, and the vibration trend slope is calculated. b2, Based on the vibration trend slope, combined with the temperature cumulative offset analysis method, the superposition effect of vibration trend slope and temperature cumulative offset is obtained; b3. Based on the superposition effect of the vibration trend slope and temperature cumulative offset, the multi-dimensional change characteristics within each time window are obtained by calculating the torque decay period. b4. Based on the multi-dimensional change characteristics within each time window, the component fatigue index is obtained using the component fatigue index calculation method. b5. By comparing the component fatigue index with the performance parameters, the residual of the component performance parameters is obtained.

[0086] Based on long-term historical data filtered using historical data matching rates, time series analysis was used to accurately extract temporal features and calculate the vibration trend slope. Then, by integrating the superposition effect of the vibration trend slope and the cumulative temperature offset, the comprehensive impact of multiple factors on the performance of the wind turbine's transmission system was fully considered. Furthermore, based on the superposition effect, the torque decay cycle was calculated to form multi-dimensional change features within each time window, comprehensively capturing the dynamic evolution law of the wind turbine's transmission system operation status. Finally, the fatigue index of components was accurately calculated based on the multi-dimensional change features, and the residual of the performance parameters was determined by comparing it with the performance parameters, realizing a quantitative assessment of the fatigue accumulation degree and performance degradation level of the key components of the wind turbine's transmission system.

[0087] Furthermore, when analyzing the residual performance parameters of components, the fatigue index variation law of the component is first modeled by segmenting the degradation path. For example, a piecewise linear regression algorithm is used to fit the fatigue index, setting the segmentation point at every 1000 operating cycles. The slopes of each segment are 0.5, 0.8, and 1.2, reflecting the gradual increase in fatigue level. Then, temperature data collected by a temperature sensor is used to calculate the cumulative temperature offset. Combined with torque decay cycle data from a torque sensor, a coupling relationship between the two is established through multiple linear regression analysis. For example, for every 1°C increase in the cumulative temperature offset, the torque decay cycle shortens by 5 cycles, with a correlation coefficient of 92, indicating that temperature has a significant impact on torque decay. Next, based on vibration data collected by a vibration sensor, the slope of the vibration trend is calculated. Using a sliding window algorithm with a window size of 500 data points, a slope of 1.5 is calculated and compared with a preset threshold upper limit of 20 to determine whether the vibration trend exceeds the range. Through the above analysis, the fatigue state and performance degradation trend of the component can be accurately assessed, providing data support for component maintenance and replacement.

[0088] S404, when the vibration trend slope exceeds a preset threshold, a neural network model is used to perform deep feature extraction on the principal components of the vibration signal and the temperature distribution weights to obtain the fault occurrence probability distribution and the fault location. For details, please refer to... Figure 1 S104 of the illustrated embodiment will not be described again here.

[0089] The wind turbine drive system fault assessment method provided in this embodiment utilizes a data dimensionality reduction algorithm to reduce the dimensionality of multidimensional raw data, retaining core distinguishing features and simplifying calculations. Then, a data clustering algorithm is used to intelligently classify operating states, and a preliminary fault mode mapping is constructed by combining historical faults. Subsequently, the rationality of clustering is verified using signal source correlation coefficients, and long-term historical operating data is filtered through historical data matching rates to reduce the risk of misjudgment. Next, based on time series analysis and degradation path analysis, the coupling relationship between vibration trend slope, temperature accumulation shift, and torque decay cycle is obtained, creating conditions for early intervention. Finally, a neural network model is used to deeply extract core features, accurately outputting the probability distribution and specific location of fault occurrence. This improves the accuracy of fault identification in the drive system of offshore wind turbines and the ability to predict component performance degradation, overcoming the shortcomings of publicly disclosed fault monitoring methods in related technologies, which struggle to accurately identify faults and provide early warnings of fault risks caused by performance degradation of key components.

[0090] This embodiment provides a method for fault assessment of wind turbine transmission systems, which can be used in the backend server of offshore wind turbines. Figure 5 This is a flowchart of a wind turbine drive system fault assessment method according to an embodiment of the present invention, such as... Figure 5 As shown, the process includes the following steps: S501, based on the acquired multidimensional raw data of the target wind turbine drive system, a data dimensionality reduction algorithm is used to reduce the dimensionality, resulting in a data dimensionality reduction matrix. Then, a data clustering algorithm is used to obtain multiple operating state clustering results. Combined with historical faults associated with each operating state cluster, a fault mode initial mapping is constructed. The multidimensional raw data includes vibration data, temperature data, and torque data. The data dimensionality reduction matrix includes the principal components of the vibration signal, temperature distribution weights, and torque fluctuation factors. For details, please refer to [link to relevant documentation]. Figure 1 S101 of the illustrated embodiment will not be described again here.

[0091] S502, based on the initial fault mode mapping, and by integrating historical operating data, the operating state clustering is verified using the signal source correlation coefficient to obtain the potential fault mode set and historical data matching rate. For details, please refer to [link to relevant documentation]. Figure 1 S102 of the illustrated embodiment will not be described again here.

[0092] S503, based on long-term historical operating data filtered using the historical data matching rate, time series analysis is used to obtain the vibration trend slope, temperature cumulative offset, and torque decay period. Combined with degradation path analysis, the coupling relationship between the temperature cumulative offset and the torque decay period is obtained, and it is determined whether the vibration trend slope exceeds a preset threshold. For details, please refer to [link to relevant documentation]. Figure 1 S103 of the illustrated embodiment will not be described again here.

[0093] S504, when the vibration trend slope exceeds a preset threshold, a neural network model is used to perform deep feature extraction on the principal components of the vibration signal and the temperature distribution weights to obtain the fault occurrence probability distribution and the fault location. For details, please refer to... Figure 1 S104 of the illustrated embodiment will not be described again here.

[0094] S505, based on the fault occurrence probability distribution, Bayesian inference is used to fuse the torque fluctuation factor with the historical data matching rate to obtain a risk assessment quantitative index and determine the fault risk level of the target wind turbine transmission system.

[0095] For example, the above S505 can be implemented as follows: First, a fault occurrence probability distribution model is constructed based on historical fault data, assuming the fault probability follows a Poisson distribution with a parameter λ of 0.5, representing an average of 0.5 faults per hour. Then, through Bayesian inference, the torque fluctuation factor is fused with the historical data matching rate. The torque fluctuation factor is collected in real-time by sensors; assuming the current torque fluctuation value is 12 and the historical data matching rate is 85%, the Bayesian formula is as follows: P(A|B)=P(B|A)·P(A) / P(B), Where P(A) represents the prior probability of failure 0.5; P(B|A) is the probability of torque fluctuation factor under the condition of failure occurrence, assumed to be 8; P(B) is the marginal probability of torque fluctuation factor, assumed to be 15; the posterior probability P(A|B) is calculated to be 8·0.5 / 15≈267.

[0096] Based on historical data matching rates, a weighted average method was used with weights of 6 and 4, resulting in a final risk assessment quantification index of 267.6 + 85.4 ≈ 48. According to the preset risk level classification standard, 48 falls within the medium risk range, and the current risk level of the wind turbine drive system is determined to be medium.

[0097] The wind turbine drive system fault assessment method provided in this embodiment utilizes a data dimensionality reduction algorithm to reduce the dimensionality of multidimensional raw data, retaining core distinguishing features and simplifying calculations. Then, a data clustering algorithm is used to intelligently classify operating states, and a preliminary fault mode mapping is constructed by combining historical faults. Subsequently, the rationality of clustering is verified using signal source correlation coefficients, and long-term historical operating data is filtered through historical data matching rates to reduce the risk of misjudgment. Next, based on time series analysis and degradation path analysis, the coupling relationship between vibration trend slope, temperature accumulation shift, and torque decay cycle is obtained, creating conditions for early intervention. Finally, a neural network model is used to deeply extract core features, accurately outputting the probability distribution and specific location of fault occurrence. This improves the accuracy of fault identification in the drive system of offshore wind turbines and the ability to predict component performance degradation, overcoming the shortcomings of publicly disclosed fault monitoring methods in related technologies, which struggle to accurately identify faults and provide early warnings of fault risks caused by performance degradation of key components.

[0098] This embodiment also provides a wind turbine drive system fault assessment device, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0099] This embodiment provides a fault assessment device for a wind turbine transmission system, such as... Figure 6 As shown, it includes: The fault mapping construction module 610 is used to reduce the dimensionality of the acquired multidimensional raw data of the target wind turbine drive system using a data dimensionality reduction algorithm to obtain a data dimensionality reduction matrix. Then, a data clustering algorithm is used to obtain multiple operating state clustering results. Combined with the historical faults associated with each operating state cluster, a fault mode initial mapping is constructed. The multidimensional raw data includes vibration data, temperature data, and torque data. The data dimensionality reduction matrix includes the principal component of the vibration signal, the temperature distribution weight, and the torque fluctuation factor. The fault mapping verification module 620 is used to verify the clustering of the operating state based on the initial fault mode mapping, combined with historical operating data, and using the signal source correlation coefficient to obtain a set of potential fault modes and historical data matching rate. The vibration trend assessment module 630 is used to obtain the vibration trend slope, temperature cumulative offset and torque decay period based on long-term historical operating data filtered by the historical data matching rate using time series analysis method, and to obtain the coupling relationship between the temperature cumulative offset and the torque decay period by combining the degradation path analysis method, and to determine whether the vibration trend slope exceeds a preset threshold. The fault assessment output module 640 is used to extract deep features from the principal components of the vibration signal and the temperature distribution weights using a neural network model when the vibration trend slope exceeds a preset threshold, so as to obtain the fault occurrence probability distribution and the fault location.

[0100] In some alternative implementations, the fault mapping construction module 610 includes: The data dimensionality reduction unit is used to calculate the principal components of the vibration signal, the temperature distribution weight, and the torque fluctuation factor based on the multidimensional raw data of the target wind turbine drive system, and obtain the data dimensionality reduction matrix. The data clustering unit is used to extract the variance of the feature vector based on the data dimensionality reduction matrix using a data clustering algorithm to obtain multiple running state clusters; The fault mapping construction unit is used to obtain an initial fault mode mapping for each motion state cluster by combining the historical faults associated with each running state cluster.

[0101] In some optional implementations, the fault mapping construction module 610 further includes: The data denoising unit is used to denoise the multidimensional raw data of the acquired target wind turbine transmission system using wavelet transform to obtain denoised multidimensional data. The data normalization unit is used to process the denoised multidimensional data using standard deviation normalization to obtain a normalized dataset; the normalized dataset is used to calculate the principal components of the vibration signal, the temperature distribution weight, and the torque fluctuation factor through a data dimensionality reduction algorithm.

[0102] In some optional implementations, the fault mapping verification module 620 includes: The data verification unit is used to verify the clustering of operating states based on the initial mapping of the fault modes using the correlation coefficient of the signal sources, and to obtain a set of potential fault modes. The historical matching rate acquisition unit is used to perform feature matching between each potential fault mode in the potential fault mode set and historical fault data based on the potential fault mode set, calculate the proportion of the number of successfully matched samples to the total number of historical fault samples, and obtain the historical data matching rate.

[0103] In some alternative implementations, the vibration trend assessment module 630 includes: The time series analysis unit is used to obtain the vibration trend slope, temperature cumulative offset and torque decay cycle corresponding to each time window based on the long-term historical operating data after filtering by the historical data matching rate, and to extract the component fatigue index and determine the component performance parameter residuals. The degradation path analysis unit is used to extract the fatigue index of the component based on the residual performance parameters of the component and to obtain the degradation path of the fatigue index by using the degradation path segmentation analysis method. The temperature cumulative influence unit is used to extract temperature cumulative features from the degradation path and obtain the distribution law of temperature cumulative offset value. The torque attenuation influencing unit is used to calculate the periodic relationship of torque attenuation based on the accumulated temperature offset value and determine the periodic change trend. The vibration trend assessment unit is used to calculate the numerical characteristics of the vibration trend slope change based on the periodic change trend, and to determine whether the vibration trend slope exceeds a preset threshold.

[0104] In some optional implementations, the timing analysis unit is specifically used for: Based on the long-term historical operating data filtered using the historical data matching rate, time series characteristics are extracted using time series analysis, and the vibration trend slope is calculated. Based on the vibration trend slope, combined with the temperature cumulative offset analysis method, the superposition effect of vibration trend slope and temperature cumulative offset is obtained; Based on the superposition effect of the vibration trend slope and the temperature cumulative shift, the multi-dimensional change characteristics within each time window are obtained by calculating the torque decay period. Based on the multi-dimensional change characteristics within each time window, the component fatigue index is obtained using the component fatigue index calculation method. By comparing the fatigue index of the component with its performance parameters, the residual performance parameters of the component are obtained.

[0105] In some alternative implementations, it also includes: The fault level prediction module is used to obtain a quantitative risk assessment index by fusing the torque fluctuation factor with the historical data matching rate based on the fault occurrence probability distribution and using Bayesian inference, and to determine the fault risk level of the target wind turbine transmission system.

[0106] The wind turbine drive system fault assessment device provided in this embodiment of the invention can execute the wind turbine drive system fault assessment method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects for executing the method. Further functional descriptions of the above modules and units are the same as in the corresponding embodiments described above, and will not be repeated here.

[0107] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0108] The following is a detailed reference. Figure 7This diagram illustrates a suitable structural schematic for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 701, which can perform various appropriate actions and processes based on a program stored in read-only memory (ROM) 702 or a program loaded from memory 708 into random access memory (RAM) 703. The RAM 703 also stores various programs and data required for the operation of the electronic device. The processor 701, ROM 702, and RAM 703 are interconnected via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0109] Typically, the following devices can be connected to I / O interface 705: input devices 706 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 707 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 708 including, for example, magnetic tapes, hard disks, etc.; and communication devices 709. Communication device 709 allows electronic devices to exchange data via wireless or wired communication with other devices. Although Figure 7 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0110] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 709, or installed from a memory 708, or installed from a ROM 702. When the computer program is executed by the processor 701, it performs the functions defined in the wind turbine drive system fault assessment method of the embodiments of the present invention.

[0111] Figure 7 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0112] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the wind turbine drive system fault assessment method shown in the above embodiments is implemented.

[0113] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0114] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for fault assessment of a wind turbine transmission system, characterized in that, The method includes: Based on the acquired multidimensional raw data of the target wind turbine drive system, a data dimensionality reduction algorithm is used to reduce the dimensionality and obtain a data dimensionality reduction matrix. Then, a data clustering algorithm is used to obtain multiple operating state clustering results. Combined with the historical faults associated with each operating state cluster, a fault mode initial mapping is constructed. The multidimensional raw data includes vibration data, temperature data, and torque data. The data dimensionality reduction matrix includes the principal component of the vibration signal, the temperature distribution weight, and the torque fluctuation factor. Based on the initial mapping of the fault modes, and by integrating historical operating data, the clustering of the operating states is verified using the correlation coefficient of the signal sources, thereby obtaining a set of potential fault modes and a historical data matching rate. Based on long-term historical operating data filtered using the historical data matching rate, the vibration trend slope, temperature cumulative offset, and torque decay cycle are obtained using time series analysis. Combined with degradation path analysis, the coupling relationship between the temperature cumulative offset and the torque decay cycle is obtained, and it is determined whether the vibration trend slope exceeds a preset threshold. When the slope of the vibration trend exceeds a preset threshold, a neural network model is used to perform deep feature extraction on the principal components of the vibration signal and the temperature distribution weights to obtain the probability distribution of the fault occurrence and the location of the fault occurrence.

2. The method according to claim 1, characterized in that, Based on long-term historical operating data filtered using the historical data matching rate, time series analysis is used to obtain the vibration trend slope, temperature cumulative offset, and torque decay period. Combined with degradation path analysis, the coupling relationship between the temperature cumulative offset and the torque decay period is obtained, and it is determined whether the vibration trend slope exceeds a preset threshold, including: Based on the long-term historical operating data filtered using the historical data matching rate, the vibration trend slope, temperature cumulative offset, and torque decay cycle corresponding to each time window are obtained using time series analysis. The component fatigue index is extracted, and the component performance parameter residuals are determined. Based on the residual performance parameters of the component, the fatigue index of the component is extracted using the degradation path segmentation analysis method, and the degradation path of the fatigue index is obtained. Temperature accumulation features are extracted from the degradation path to obtain the distribution pattern of temperature accumulation offset values; Based on the accumulated temperature offset value, the periodic relationship of torque decay is calculated to determine the periodic change trend; Based on the periodic change trend, the numerical characteristics of the vibration trend slope change are calculated, and it is determined whether the vibration trend slope exceeds a preset threshold.

3. The method according to claim 2, characterized in that, Based on the long-term historical operating data filtered using the historical data matching rate, time series analysis is used to obtain the vibration trend slope, temperature cumulative offset, and torque decay cycle corresponding to each time window, and the component fatigue index is extracted to determine the component performance parameter residuals, including: Based on the long-term historical operating data filtered using the historical data matching rate, time series characteristics are extracted using time series analysis, and the vibration trend slope is calculated. Based on the vibration trend slope, combined with the temperature cumulative offset analysis method, the superposition effect of vibration trend slope and temperature cumulative offset is obtained; Based on the superposition effect of the vibration trend slope and the temperature cumulative shift, the multi-dimensional change characteristics within each time window are obtained by calculating the torque decay period. Based on the multi-dimensional change characteristics within each time window, the component fatigue index is obtained using the component fatigue index calculation method. By comparing the fatigue index of the component with its performance parameters, the residual performance parameters of the component are obtained.

4. The method according to claim 1, characterized in that, Based on the initial mapping of the fault modes, and by integrating historical operating data, the operating state clustering is verified using the signal source correlation coefficient to obtain a set of potential fault modes and a historical data matching rate, including: Based on the initial mapping of the fault modes, the clustering of the operating states is verified by using the correlation coefficient of the signal sources to obtain a set of potential fault modes. Based on the set of potential fault modes, each potential fault mode in the set is matched with historical fault data, and the proportion of the number of successfully matched samples to the total number of historical fault samples is calculated to obtain the historical data matching rate.

5. The method according to claim 1, characterized in that, Also includes: Based on the fault occurrence probability distribution, Bayesian inference is used to fuse the torque fluctuation factor with the historical data matching rate to obtain a quantitative risk assessment index and determine the fault risk level of the target wind turbine transmission system.

6. The method according to claim 1, characterized in that, The multidimensional raw data of the target wind turbine drive system is obtained, and then dimensionality reduction is performed using a data dimensionality reduction algorithm to obtain a data dimensionality reduction matrix. Next, a data clustering algorithm is used to obtain multiple operating state clustering results. Finally, combined with historical faults associated with each operating state cluster, an initial fault mode mapping is constructed, including: Based on the multidimensional raw data of the target wind turbine transmission system, the principal components of the vibration signal, the temperature distribution weight, and the torque fluctuation factor are calculated using a data dimensionality reduction algorithm to obtain the data dimensionality reduction matrix. Based on the data dimensionality reduction matrix, the variance of the feature vector is extracted using a data clustering algorithm to obtain multiple running state clusters; For each motion state cluster, the initial fault mode mapping is obtained by combining the historical faults associated with each motion state cluster.

7. The method according to claim 6, characterized in that, Also includes: Based on the acquired multidimensional raw data of the target wind turbine transmission system, wavelet transform is used to denoise the multidimensional raw data to obtain denoised multidimensional data. Based on the denoised multidimensional data, standard deviation normalization is used to obtain a normalized dataset; the normalized dataset is used to calculate the principal components of the vibration signal, the temperature distribution weight, and the torque fluctuation factor using principal component analysis.

8. A fault assessment device for a wind turbine transmission system, characterized in that, The device includes: The fault mapping construction module is used to reduce the dimensionality of the acquired multidimensional raw data of the target wind turbine drive system using a data dimensionality reduction algorithm to obtain a data dimensionality reduction matrix. Then, a data clustering algorithm is used to obtain multiple operating state clustering results. Combined with the historical faults associated with each operating state cluster, a fault mode initial mapping is constructed. The multidimensional raw data includes vibration data, temperature data, and torque data. The data dimensionality reduction matrix includes the principal component of the vibration signal, the temperature distribution weight, and the torque fluctuation factor. The fault mapping verification module is used to verify the clustering of the operating state based on the initial fault mode mapping, combined with historical operating data, and using the signal source correlation coefficient to obtain a set of potential fault modes and historical data matching rate. The vibration trend assessment module is used to obtain the vibration trend slope, temperature cumulative offset, and torque decay period based on long-term historical operating data filtered by the historical data matching rate using time series analysis. Combined with the degradation path analysis method, it obtains the coupling relationship between the temperature cumulative offset and the torque decay period, and determines whether the vibration trend slope exceeds a preset threshold. The fault assessment output module is used to extract deep features from the principal components of the vibration signal and the temperature distribution weights using a neural network model when the vibration trend slope exceeds a preset threshold, so as to obtain the fault occurrence probability distribution and the fault location.

9. An electronic device, characterized in that, include: The system includes a memory and a processor, which are interconnected. The memory stores computer instructions, and the processor executes the computer instructions to perform the wind turbine drive system fault assessment method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to execute the wind turbine drive system fault assessment method according to any one of claims 1 to 7.

Images