A wind turbine generator unit bearing fault diagnosis method and system

Abstract

The application provides a wind turbine bearing fault diagnosis method and system, relates to the technical field of bearing fault diagnosis, and collects bearing vibration signals during operation of a wind turbine; sample labeling is performed on the bearing vibration signals to obtain an operation fault label, the operation fault label is fitted with a fault impact index of the wind turbine bearing in a cross-working condition, and a fault recognition criterion is generated; weak fault characteristics between similar fault characteristics distances in the wind turbine bearing are determined according to operation health data to obtain a fault sensitive probability, a missing degradation track of the wind turbine bearing in a missing working condition is generated according to all fault sensitive probabilities, and the wind turbine bearing fault is classified and diagnosed according to the fault recognition criterion and the missing degradation track. The application can perform fault classification and diagnosis on the wind turbine bearing fault in a complex scene where the cross-working condition and the missing working condition coexist, so that the accuracy of fault diagnosis is improved.

Description

Technical Field

[0001] This application relates to the field of bearing fault diagnosis technology, and more specifically, to a method and system for diagnosing bearing faults in wind turbine generator sets. Background Technology

[0002] Bearing fault diagnosis is one of the core technologies for ensuring the safe and stable operation of wind turbine generators. It aims to collect operational health data such as vibration, temperature, and speed of the main bearing and gearbox bearings through a multi-source sensor array. Combining cross-condition operating characteristics and fault evolution patterns, it extracts multi-domain fusion features and weak fault signals to construct fault identification criteria and degradation trajectory prediction models. This fills gaps in missing operating condition data, accurately identifies fault types, severity, and evolution trends, and avoids misdiagnosis and missed diagnosis caused by operating condition fluctuations and noise interference. Bearing fault diagnosis is widely used in wind power operation and maintenance scenarios. By improving cross-condition adaptability and early fault detection capabilities, it enables early fault warning and precise location, reduces unplanned downtime losses, supports targeted maintenance decisions, and aligns with the wind power industry's trend towards intelligent and efficient operation and maintenance transformation.

[0003] However, existing wind turbine bearing fault diagnosis methods largely rely on single sensor signal acquisition, lack cross-condition adaptation mechanisms, and fail to fully exploit weak signals among similar fault characteristics. Furthermore, they neglect to complete the fault diagnosis trajectory for missing conditions, resulting in incomplete signal acquisition dimensions, one-sided fault feature extraction, and a lack of effective diagnostic basis under missing conditions. This leads to low fault identification accuracy across different operating conditions, easy omission of weak faults, and difficulty in accurately identifying bias faults, thereby increasing unplanned downtime losses and reducing unit operation and maintenance efficiency. Therefore, how to classify and diagnose wind turbine bearing faults in complex scenarios involving both cross-condition and missing conditions to improve the accuracy of fault diagnosis is a challenge facing the industry. Summary of the Invention

[0004] This application provides a method and system for diagnosing bearing faults in wind turbine generator sets, which can classify and diagnose bearing faults in wind turbine generator sets under complex scenarios where multiple operating conditions and missing operating conditions coexist, thereby improving the accuracy of fault diagnosis.

[0005] In a first aspect, this application provides a method for diagnosing bearing faults in wind turbine generator sets, the method comprising the following steps:

[0006] A multi-source sensor array is arranged at the main bearing and gearbox bearing of the wind turbine generator set to collect bearing vibration signals during the operation of the wind turbine generator set.

[0007] The bearing vibration signal is sampled and labeled to obtain the operation fault labels of various wind turbine bearings. All operation fault labels are fitted with the fault impact index of wind turbine bearings in cross-operating conditions to generate fault identification criteria of wind turbine bearings in cross-operating conditions.

[0008] Obtain operational health data of the bearings of the wind turbine generator set in operation, combine the operational health data with the weak fault characteristics between similar fault characteristics in the bearings of the wind turbine generator set to determine the fault sensitivity probability, and generate the missing degradation trajectory of the bearings of the wind turbine generator set under the condition of missing fault diagnosis based on all the fault sensitivity probabilities.

[0009] Based on the fault identification criteria and the missing degradation trajectory, the bearing of the wind turbine generator set is biased to identify the bias fault type, and the bearing fault of the wind turbine generator set is classified and diagnosed according to the bias fault type.

[0010] In conjunction with the first aspect, in certain implementations of the first aspect, the bearing vibration signal is sampled and labeled to obtain operational fault labels for various types of wind turbine generator bearings, specifically including:

[0011] Based on the bearing vibration signal, fault features are constructed from the multi-domain fusion characteristic parameters of the wind turbine generator set;

[0012] Based on the aforementioned fault characteristics, multiple fault operation modes of wind turbine generator bearings during operation are determined.

[0013] Extract the operating fault tags of the wind turbine generator bearings from each type of fault operation mode.

[0014] In conjunction with the first aspect, in some implementations of the first aspect, the cross-operating condition refers to the different operating states formed by changes in wind speed and load during the operation of the wind turbine generator set.

[0015] In conjunction with the first aspect, in certain implementations of the first aspect, determining the fault sensitivity probability by combining the operational health data with the weak fault characteristics between similar fault characteristic distances in the wind turbine generator bearings specifically includes:

[0016] Based on the aforementioned operational health data, the fault impact component of the wind turbine generator bearing vibration signal within the resonant frequency band is extracted;

[0017] Based on the aforementioned fault impact components, a weak fault feature enhancement index is determined between the distance to similar fault features.

[0018] The fault sensitivity probability is determined by using the weak fault feature enhancement index and the probability density distribution of operational health data.

[0019] In conjunction with the first aspect, in some implementations of the first aspect, the weak fault characteristics refer to the characteristic information caused by early and minor bearing faults, which have small signal amplitudes and are easily interfered with by noise.

[0020] In conjunction with the first aspect, in some implementations of the first aspect, the aforementioned fault diagnosis missing condition refers to a specific operating condition in which the bearings of the wind turbine generator set fail to collect the data required for fault diagnosis.

[0021] In conjunction with the first aspect, in certain implementations of the first aspect, the offset identification of the wind turbine generator bearing is performed based on the fault identification criteria and the missing degradation trajectory, and the offset fault types specifically include:

[0022] The vibration signal bias features of the wind turbine generator bearings are extracted from the fault identification criteria and the missing degradation trajectory, respectively.

[0023] Based on the aforementioned vibration signal bias characteristics, a bias identification model is constructed and dynamically matched with a known fault mode library.

[0024] The vibration signal is evaluated using the bias identification model, and the bias fault type is output.

[0025] In conjunction with the first aspect, in some implementations of the first aspect, the missing degradation trajectory refers to the complete degradation process curve of the bearing from normal state to fault state, simulated under fault diagnosis missing conditions.

[0026] In conjunction with the first aspect, in some implementations of the first aspect, all operational fault labels are fitted with the fault impact index of wind turbine generator bearings across operating conditions to generate fault identification criteria for wind turbine generator bearings across operating conditions. Specifically, these criteria include:

[0027] Based on all operational fault labels, vibration signal impact characteristics of wind turbine generator bearings under different operating conditions are extracted.

[0028] The failure impact index of wind turbine generator bearings under cross-operating conditions was determined based on the impact characteristics of all vibration signals.

[0029] By performing regression fitting on all operational fault labels and the fault impact index, a mapping model between fault modes and impact responses under cross-operating conditions is established.

[0030] Based on the mapping relationship model, fault identification criteria for wind turbine generator bearings under different operating conditions are generated.

[0031] Secondly, this application provides a wind turbine generator bearing fault diagnosis system for performing a wind turbine generator bearing fault diagnosis method, the fault diagnosis system comprising:

[0032] The data acquisition module is used to arrange a multi-source sensor array at the main bearing and gearbox bearing of the wind turbine generator set, and to collect bearing vibration signals during the operation of the wind turbine generator set through the multi-source sensor array.

[0033] The vibration fitting module is used to sample and label the vibration signal of the bearing to obtain the operation fault labels of multiple types of wind turbine bearings. All operation fault labels are fitted with the fault impact index of wind turbine bearings in cross-operating conditions to generate fault identification criteria of wind turbine bearings in cross-operating conditions.

[0034] The fault identification module is used to acquire the operational health data of the wind turbine bearing in operation, combine the operational health data with the weak fault characteristics between similar fault characteristics in the wind turbine bearing to determine the fault sensitivity probability, and generate the missing degradation trajectory of the wind turbine bearing under the fault diagnosis missing condition based on all the fault sensitivity probabilities.

[0035] The fault diagnosis module is used to identify the offset of the wind turbine generator bearing based on the fault identification criteria and the missing degradation trajectory, obtain the offset fault type, and classify and diagnose the wind turbine generator bearing fault according to the offset fault type.

[0036] The technical solution provided in this application has the following beneficial effects:

[0037] A multi-source sensor array is deployed at the main bearing and gearbox bearing of the wind turbine generator set to collect bearing vibration signals during wind turbine generator set operation. The bearing vibration signals are sampled and labeled to obtain multiple types of operational fault labels for the wind turbine generator set bearings. All operational fault labels are fitted with the fault impact index of the wind turbine generator set bearings across operating conditions to generate fault identification criteria for the wind turbine generator set bearings across operating conditions. Operational health data of the wind turbine generator set bearings in operation is acquired. This operational health data is combined with weak fault characteristics between similar fault feature distances in the wind turbine generator set bearings to determine fault sensitivity probabilities. Based on all fault sensitivity probabilities, a missing degradation trajectory of the wind turbine generator set bearings under fault diagnosis-deficient operating conditions is generated. Based on the fault identification criteria and the missing degradation trajectory, the wind turbine generator set bearings are biased to identify bias fault types, and the wind turbine generator set bearing faults are classified and diagnosed according to the bias fault types.

[0038] Therefore, this application demonstrates that, despite the poor cross-condition adaptability of existing wind turbine bearing fault diagnosis methods, it can achieve accurate classification and diagnosis of bearing faults. Specifically, by deploying multi-source sensor arrays of various types and locations at key locations in the main bearing and gearbox bearing, and simultaneously acquiring vibration signals from different frequency ranges while ensuring temporal correlation, it solves the problems of limited data acquisition from a single sensor and susceptibility to noise and angular deviation interference in traditional diagnosis, providing comprehensive and complete raw vibration signal data. Furthermore, by performing multi-domain feature extraction and sample labeling on the vibration signals, and combining this with cross-condition fault impact index fitting to generate fault identification criteria, it addresses the lack of cross-condition adaptability in traditional fault diagnosis. To address the issues of adaptability to operating conditions and the strong subjectivity in fault mode classification, a standardized fault judgment standard adaptable to diverse operating scenarios should be established. By acquiring multi-dimensional operational health data, combining the distance of similar fault characteristics to mine weak fault features and calculate the fault sensitivity probability, a degradation trajectory under missing operating conditions can be generated, which can solve the problems of missing operating condition data gaps and low identification of weak fault features in traditional diagnosis. By extracting bias features based on fault identification criteria and missing degradation trajectories, a dynamic matching bias identification model can be constructed and classified for diagnosis, which can solve the problems of ambiguous bias fault identification and lack of systematic diagnostic results in traditional diagnosis, and improve the comprehensiveness of wind turbine bearing fault diagnosis.

[0039] In summary, the technical solution adopted in this application can classify and diagnose bearing failures of wind turbine generator sets in complex scenarios where multiple operating conditions and missing operating conditions coexist, thereby improving the accuracy of fault diagnosis. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this embodiment of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 This is an exemplary flowchart of a method for diagnosing bearing faults in a wind turbine generator set according to this application;

[0042] Figure 2 This is a flowchart illustrating the process for determining fault identification criteria provided in this application;

[0043] Figure 3 This is a flowchart illustrating the process for determining missing degradation trajectories provided in this application;

[0044] Figure 4 This is a module structure diagram of a wind turbine generator bearing fault diagnosis system provided in this application. Detailed Implementation

[0045] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0046] This application provides a method and system for diagnosing bearing faults in wind turbine generator sets. The core of this method involves arranging a multi-source sensor array at the main bearing and gearbox bearing of the wind turbine generator set. The multi-source sensor array collects bearing vibration signals during wind turbine generator set operation. The bearing vibration signals are sampled and labeled to obtain multiple types of operational fault labels for the wind turbine generator set bearings. All operational fault labels are fitted with the fault impact index of the wind turbine generator set bearings across operating conditions to generate fault identification criteria for the wind turbine generator set bearings across operating conditions. Operational health data of the wind turbine generator set bearings in operation is acquired. This operational health data is combined with weak fault characteristics between similar fault feature distances in the wind turbine generator set bearings to determine fault sensitivity probabilities. Based on all fault sensitivity probabilities, a missing degradation trajectory of the wind turbine generator set bearings under fault diagnosis-deficient operating conditions is generated. Based on the fault identification criteria and the missing degradation trajectory, the wind turbine generator set bearings are biased to identify bias fault types, and the wind turbine generator set bearing faults are classified and diagnosed according to the bias fault types.

[0047] Example 1: To better understand the above technical solution, the following will provide a detailed description of the technical solution in conjunction with the accompanying drawings and specific implementation methods. (Refer to...) Figure 1 As shown in the figure, this is an exemplary flowchart of a wind turbine generator bearing fault diagnosis method according to this embodiment of the present application. The bearing fault diagnosis method includes the following steps:

[0048] In step S1, a multi-source sensor array is arranged at the main bearing and gearbox bearing of the wind turbine generator set, and the bearing vibration signal during the operation of the wind turbine generator set is collected through the multi-source sensor array.

[0049] In practical implementation, firstly, 2-3 sampling points are selected on the fixed and rotating ends of the main bearing of the wind turbine generator, corresponding to the faulty parts of the inner and outer rings (such as the raceway contact area). For the high-speed, medium-speed, and low-speed shaft bearings of the gearbox, 2 sampling points are respectively set on the inner and outer ring side housings. An array of piezoelectric accelerometers (to capture high-frequency fault impact signals) and magnetoelectric velocity sensors (to capture low-frequency vibration trends) is used, with the former sampling frequency set to 20-50kHz and the latter to 1-10kHz. All sensors are connected to the data acquisition unit via shielded cables, and sampling is controlled by a GPS synchronous clock to ensure that the time deviation of each sensor is less than 1ms. Continuous data is collected according to wind speed ranges (e.g., 10-15m / s, 15-20m / s, etc.), with each condition collected for 30 minutes. The collected data is stored as a timestamped time-domain vibration signal, which is used as the bearing vibration signal during wind turbine generator operation. Further details are omitted here.

[0050] It should be noted that, in this application, the multi-source sensor array refers to a combination of different types of sensors arranged in key parts of the main bearing and gearbox bearing; the bearing vibration signal refers to the information of the relative motion of the internal raceways, rolling elements, and cages of the main bearing and gearbox bearing during the operation of the wind turbine generator set due to normal friction and assembly deviations.

[0051] In step S2, the bearing vibration signal is sampled and labeled to obtain the operating fault labels of various types of wind turbine bearings. All operating fault labels are fitted with the fault impact index of wind turbine bearings in cross-operating conditions to generate fault identification criteria for wind turbine bearings in cross-operating conditions.

[0052] In this embodiment, the following steps can be used to sample and label the bearing vibration signals to obtain operational fault labels for various types of wind turbine bearings:

[0053] Based on the bearing vibration signal, fault features are constructed from the multi-domain fusion characteristic parameters of the wind turbine generator set;

[0054] Based on the aforementioned fault characteristics, multiple fault operation modes of wind turbine generator bearings during operation are determined.

[0055] Extract the operating fault tags of the wind turbine generator bearings from each type of fault operation mode.

[0056] In practice, firstly, the time-domain features (such as peak value, kurtosis, and root mean square value) of the collected bearing vibration time-domain signal are calculated to reflect the signal amplitude changes. The time-domain signal is then converted to the frequency domain using a Fast Fourier Transform (FFT) to extract characteristic frequencies (such as the fault characteristic frequencies of the bearing's inner and outer rings) and the energy proportion of the frequency band. Wavelet transform is then used to obtain time-frequency domain features (such as energy entropy and singular values ​​at different scales) to capture non-stationary impact signals. The time-domain, frequency-domain, and time-frequency-domain features are integrated to form multi-domain fusion feature parameters, and features strongly correlated with faults are selected as fault features. Next, bearing vibration signal samples with known fault types (such as inner ring wear, outer ring spalling, and rolling element cracks) are collected, and their multi-domain fault features are extracted as training data. A K-means clustering algorithm is used, with Euclidean distance between features as the similarity measure, to cluster the training data, iteratively optimizing the cluster centers until convergence. Each clustered category is matched with a known fault type to determine the fault operation mode corresponding to each category. Finally, for each identified fault operation mode, the specific fault type corresponding to the mode is determined by combining the historical maintenance records of the wind turbine generator set (such as the location of the fault and the type of parts replaced) and the analysis of fault characteristics by experts in the field. This is done by identifying the specific fault type (such as "inner ring peeling" and "outer ring wear"). A unique identifier is assigned to each fault type (such as "inner ring fault-001" and "outer ring fault-002") to form the operating fault label of the wind turbine generator set bearing.

[0057] It should be noted that, in this application, the fault features in the multi-domain fusion feature parameters refer to the parameter set that fuses time-domain, frequency-domain, and time-frequency-domain features; the fault operation mode refers to the set of stable features exhibited by the bearing under a specific fault state; and the operation fault label refers to information that identifies the specific fault type of the bearing.

[0058] Preferably, in this embodiment, all operational fault labels are fitted with the fault impact index of the wind turbine generator bearing across operating conditions to generate fault identification criteria for the wind turbine generator bearing across operating conditions, with reference to Figure 2 As shown in the figure, this is a schematic flowchart of the process for determining fault identification criteria in some embodiments of this application. In this embodiment, the determination of fault identification criteria can be achieved by the following steps:

[0059] In step S21, the vibration signal impact characteristics of the wind turbine generator bearing under different operating conditions are extracted based on all operational fault labels.

[0060] In step S22, the failure impact index of the wind turbine generator bearing under cross-operating conditions is determined based on the impact characteristics of all vibration signals.

[0061] In step S23, all operational fault labels are regressed and fitted with the fault impact index to establish a mapping relationship model between fault modes and impact responses under cross-operating conditions.

[0062] In step S24, fault identification criteria for wind turbine generator bearings under cross-operating conditions are generated based on the mapping relationship model.

[0063] In practice, firstly, operating conditions are categorized by wind speed (e.g., low, medium, high) and load (e.g., light load, full load), and bearing vibration signals with operational fault tags are screened for each type of operating condition. High-pass filtering is applied to the signals to remove low-frequency interference, and Hilbert-Huang transform is used to decompose the signals into intrinsic mode functions. Signal segments with instantaneous amplitudes greater than a set threshold in each mode are extracted as impact segments. The impact peak value, impact duration, and average interval between adjacent impacts are calculated from the impact segments as the impact characteristics of the vibration signal under that operating condition. Next, Z-score standardization is used to eliminate the dimensional influence caused by differences in operating conditions for the vibration signal impact characteristics of the same fault tag under different operating conditions. The impact peak value (reflecting impact intensity) and impact frequency (number of impacts per unit time, reflecting impact density) are selected as core features. The intensity weight and frequency weight are determined using the analytic hierarchy process (the intensity weight can be 0.6, and the frequency weight can be 0.4; this is not limited here). The weighted sum of the two standardized features is calculated to obtain the fault impact index for each sample. Then, a sample set is constructed using the fault label (e.g., inner ring fault, outer ring fault, etc., assigned different integer values ​​according to type) as the dependent variable and the corresponding cross-condition fault impact index sequence as the independent variable. A random forest regression algorithm is used to divide the sample set into a training set and a validation set in a 7:3 ratio. The training set is used to fit the model (outputting the predicted label value through voting by multiple decision trees), and the validation set is used to evaluate the model accuracy through mean squared error. The number of decision trees (50-200) is iteratively adjusted to optimize the model until the validation error stabilizes, resulting in a mapping relationship model. Finally, the trained mapping relationship model is applied to the test sample set (including samples of various faults and normal states), outputting the predicted label value and confidence level for each sample. The range of fault impact indices and the predicted confidence level threshold (e.g., confidence level ≥ 85%) corresponding to the same fault label are statistically analyzed to determine the corresponding rule of "index interval - fault label - confidence level threshold". The rules are organized into a table, for example, "when the index is in interval A and the confidence level is ≥ 85%, it is determined to be an inner ring fault," forming a cross-condition fault identification criterion.

[0064] It should be noted that, in this application, "cross-operating conditions" refers to different operating states formed by changes in wind speed and load during the operation of a wind turbine generator set; "vibration signal impact characteristics" refers to the set of impact characteristic parameters presented in the vibration signal when the bearing fails; "fault impact index" refers to the quantitative index of the vibration impact intensity and frequency caused by the bearing failure; "fault mode and impact response mapping relationship model" refers to the mathematical model describing the correlation between the operating fault label and the fault impact index; and "fault identification criteria" refers to the set of rules for determining the bearing fault type under cross-operating conditions.

[0065] In step S3, the operating health data of the wind turbine bearing in operation is obtained. The operating health data is combined with the weak fault characteristics between similar fault characteristics in the wind turbine bearing to determine the fault sensitivity probability. Based on all the fault sensitivity probabilities, the missing degradation trajectory of the wind turbine bearing under the fault diagnosis missing condition is generated.

[0066] In practice, obtaining operational health data of the bearings of a wind turbine generator set in operation can be achieved as follows: Vibration signals from the main bearing and gearbox bearings of the operating wind turbine generator set are simultaneously collected using a pre-deployed multi-source sensor array; platinum resistance temperature sensors are installed at key locations on the bearing housing to collect the bearing's operating temperature in real time; Hall effect speed sensors are installed at the shaft end to obtain the bearing's real-time speed; and the viscosity and water content data of the lubricating oil are retrieved through the unit's control system. All collected data are preprocessed: the temperature and speed data are smoothed using a moving average method; outliers in the vibration signals are removed using the 3σ criterion; and the vibration, temperature, speed, and lubricating oil parameters are correlated with the corresponding wind speed and load condition data, and integrated into a structured dataset by timestamp. This yields the operational health data, which will not be elaborated further here.

[0067] It should be noted that, in this application, operational health data refers to a multi-dimensional structured data set of the operating status of wind turbine generator bearings.

[0068] In this embodiment, determining the fault sensitivity probability by combining the operational health data with the weak fault characteristics between similar fault feature distances in the wind turbine generator bearings can be achieved through the following steps:

[0069] Based on the aforementioned operational health data, the fault impact component of the wind turbine generator bearing vibration signal within the resonant frequency band is extracted;

[0070] Based on the aforementioned fault impact components, a weak fault feature enhancement index is determined between the distance to similar fault features.

[0071] The fault sensitivity probability is determined by using the weak fault feature enhancement index and the probability density distribution of operational health data.

[0072] In practice, firstly, the vibration signals of the main bearing and gearbox bearing are separated from the operational health data. A Fast Fourier Transform (FFT) is performed on the signals, and the spectrum is analyzed to determine the resonant frequency band corresponding to the bearing structure (based on the bearing's natural frequency theory). A digital bandpass filter is designed for this frequency band to filter the vibration signals, retaining the signals within the resonant frequency band. Wavelet threshold denoising is used to eliminate residual noise in the filtered signals. By setting a dynamic threshold (based on the mean of the signal's background noise), pulse signals exceeding the threshold are detected and extracted; these are the fault impact components within the resonant frequency band. Then, a labeled database of similar fault features is retrieved, and the core parameters (peak value, kurtosis, energy) of the fault impact components within the resonant frequency band are extracted. The Euclidean distance between the current fault impact component and the similar fault feature parameters is calculated. The distance value is normalized (mapped to the 0-1 interval) and its reciprocal is taken to strengthen the weight of nearby signals. Simultaneously, the energy proportion of the current impact component within the resonant frequency band is calculated, and the weighted sum is obtained according to a set weight (reciprocal of distance 0.6, energy proportion 0.4) to obtain the weak fault feature enhancement index. Finally, historical operational health data of wind turbine generator bearings (including data on normal and various fault states) were collected. Weak fault feature enhancement indices were extracted for each state, and a probability density distribution model of the enhancement indices corresponding to different operating states was constructed using kernel density estimation. The currently calculated enhancement indices were substituted into the model to calculate the probability values ​​for each state (normal and various fault states). After removing the probability values ​​for normal states, the remaining fault state probability values ​​were normalized to obtain the fault sensitivity probability for each type of fault.

[0073] It should be noted that, in this application, "similar fault features" refers to the set of characteristic parameters corresponding to the same fault type of wind turbine generator bearing extracted from vibration signals; "weak fault features" refers to the characteristic information caused by early and minor bearing faults, which have small signal amplitudes and are easily interfered with by noise; "fault impact component" refers to the pulse signal segments presented by the bearing vibration signal within the resonant frequency band; "weak fault feature enhancement index" refers to the index that quantifies the correlation between weak fault features and similar fault features; and "fault sensitivity probability" refers to the numerical index that quantifies the probability of a certain type of fault occurring in the wind turbine generator bearing.

[0074] Preferably, in this embodiment, the missing degradation trajectory of the wind turbine generator bearing under the fault diagnosis missing condition is generated based on all fault sensitivity probabilities, with reference to... Figure 3 As shown in the figure, this is a flowchart illustrating the process of determining missing degradation trajectories in some embodiments of this application. In this embodiment, determining missing degradation trajectories can be achieved using the following steps:

[0075] In step S31, a degradation state transition model of the wind turbine generator bearing during its entire life cycle is constructed based on all fault sensitivity probabilities.

[0076] In step S32, a degradation trajectory deduction model for the missing operating condition is established based on the degradation state transition model and historical health data under the fault diagnosis missing operating condition.

[0077] In step S33, the fault mode discrimination boundary under the missing working condition is determined based on the degradation trajectory inference model and the real-time fault sensitivity probability.

[0078] In step S34, the missing degradation trajectory of the wind turbine generator bearing under the fault diagnosis missing condition is generated based on the fault mode discrimination boundary.

[0079] In practical implementation, firstly, the entire life cycle of the bearing is divided into four degradation states: normal state, minor fault, moderate fault, and severe fault. Each state is defined by the fault sensitivity probability range (e.g., normal state probability < 0.2, severe fault probability > 0.8). Historical fault sensitivity probability sequences for the entire life cycle are collected, and the number of transitions between states at adjacent times is counted. The transition probability is calculated (e.g., the number of transitions from minor fault to moderate fault / the total number of minor faults), and a Markov state transition matrix is ​​constructed to form a degradation state transition model. Next, historical health data corresponding to operating conditions with missing fault diagnoses (e.g., specific wind speed-load combinations) is selected, and the correlation between degradation features (e.g., vibration signal kurtosis, temperature change rate) and fault sensitivity probabilities is extracted. Using the degradation state transition model as a framework, a Gaussian process regression algorithm is employed. With historical feature data as input and the corresponding degradation state as output, the model is trained and the kernel function parameters are optimized through cross-validation. State transition probability constraints are integrated to construct a degradation trajectory inference model for missing operating conditions. Then, the real-time fault sensitivity probability is input into the degradation trajectory extrapolation model to obtain the probability evolution curves of different fault modes under the missing operating condition. The kernel density estimation method is used to analyze the distribution characteristics of the probability curves of each fault mode, find the intersection points and probability peak intervals of different curves, and use the probability values ​​and time points corresponding to the intersection points as preliminary boundaries. Combining critical cases of mode misjudgment in historical fault diagnosis, the boundary values ​​are adjusted (e.g., expanding the boundary range of the high misjudgment area) to determine the final fault mode discrimination boundary. Finally, based on the state sequence output by the degradation trajectory extrapolation model, the fault sensitivity probabilities at each time point are arranged along the time axis. According to the fault mode discrimination boundary, the starting point and transition point of each fault mode are marked in the sequence (the mode transition is marked when the probability crosses the boundary). Cubic spline interpolation is used to smoothly connect the state points at each time point, supplementing the transition state data within the boundary interval, forming a continuous, mode-marked, missing degradation trajectory of the wind turbine bearing under the missing fault diagnosis operating condition.

[0080] It should be noted that, in this application, the fault diagnosis missing condition refers to a specific operating condition in which the wind turbine generator bearing has not collected the data required for fault diagnosis; the degradation state transition model refers to a mathematical model describing the transition probability between various degradation states throughout the bearing's entire life cycle; the degradation trajectory inference model refers to a predictive model used to simulate the bearing degradation process under the missing condition; the real-time fault sensitivity probability refers to a numerical value that quantifies the probability of a certain type of fault occurring at present; the fault mode discrimination boundary refers to the interval that divides different fault modes on the degradation trajectory; and the missing degradation trajectory refers to the complete degradation process curve of the bearing from normal state to fault state simulated under the fault diagnosis missing condition.

[0081] In step S4, the wind turbine bearing is biased based on the fault identification criteria and the missing degradation trajectory to obtain the bias fault type, and the wind turbine bearing fault is classified and diagnosed according to the bias fault type.

[0082] In this embodiment, the offset identification of the wind turbine generator bearing based on the fault identification criteria and the missing degradation trajectory, and the determination of the offset fault type, can be achieved through the following steps:

[0083] The vibration signal bias features of the wind turbine generator bearings are extracted from the fault identification criteria and the missing degradation trajectory, respectively.

[0084] Based on the aforementioned vibration signal bias characteristics, a bias identification model is constructed and dynamically matched with a known fault mode library.

[0085] The vibration signal is evaluated using the bias identification model, and the bias fault type is output.

[0086] In practice, the standard fault impact index range and corresponding vibration characteristics (such as peak value and frequency) are extracted from the fault identification criteria, and the expected vibration characteristic sequence under the corresponding working condition is obtained from the missing degradation trajectory. The real-time acquired bearing vibration signal is compared with the standard features and expected sequence, and the amplitude difference, frequency offset, and feature occurrence time deviation are calculated. Deviation parameters that exceed the normal fluctuation range (such as difference > standard value 15%) are screened out and integrated into vibration signal bias features. Then, the known fault mode library contains bias feature templates corresponding to historical bias faults (such as faults caused by installation misalignment or uneven load). A dynamic time warping algorithm is used to calculate the similarity between the real-time vibration signal bias features and the templates in the library (by adjusting the time axis of the feature sequence to minimize the matching error), and a sliding window mechanism is introduced (the window size is updated according to the working condition) to update the weights of each mode in the model in real time (the weights of recently frequently occurring modes are increased), thus constructing a dynamically matched bias identification model. Finally, the offset features of the real-time vibration signal are input into the offset recognition model. The model calculates the similarity value between the feature and each template in the known fault mode library, and sets a similarity threshold (determined based on historical diagnostic accuracy). When the similarity value of a certain mode exceeds the threshold and is the maximum value, the fault corresponding to that mode is determined to be the current offset fault type; if none exceed the threshold, "Unidentified offset fault" is output and the feature re-extraction process is triggered, which will not be elaborated here.

[0087] It should be noted that, in this application, the vibration signal bias feature refers to the set of deviation parameters between the real-time vibration signal and the standard feature and the expected trajectory feature; the bias identification model refers to the algorithm model that can dynamically match the vibration signal bias feature with the known bias fault mode; and the bias fault type refers to the specific fault type determined due to the vibration signal feature deviating from the standard mode.

[0088] In specific implementation, the classification and diagnosis of wind turbine bearing faults based on the aforementioned bias fault types can be achieved as follows: Collect the vibration signal bias characteristics, fault sensitivity probability, and missing degradation trajectory segments corresponding to the bias fault type. Establish classification dimensions based on the fault location (main bearing / gearbox bearing) and fault mechanism (e.g., wear caused by installation misalignment, spalling caused by uneven load). Use the C4.5 decision tree algorithm, with core parameters in the bias characteristics (e.g., frequency offset, amplitude deviation rate) as classification nodes, and historical classification and diagnosis cases as training samples. Determine the node splitting order through information gain ratio to construct the classification model. Input the bias fault type to be diagnosed into the model, match it to the corresponding category along the decision path, and perform secondary verification by combining the typical evolution characteristics of this type of fault in the missing degradation trajectory (e.g., comparing the impact frequency change trend in the trajectory). Finally, output the classification and diagnosis results including the fault location, mechanism, and severity. Further details are omitted here.

[0089] It should be noted that, in this application, classification diagnosis refers to the process of systematically classifying the offset faults of wind turbine generator bearings according to preset dimensions and clarifying specific fault information.

[0090] Therefore, this application demonstrates that, despite the poor cross-condition adaptability of existing wind turbine bearing fault diagnosis methods, it can achieve accurate classification and diagnosis of bearing faults. Specifically, by deploying multi-source sensor arrays of various types and locations at key locations in the main bearing and gearbox bearing, and simultaneously acquiring vibration signals from different frequency ranges while ensuring temporal correlation, it solves the problems of limited data acquisition from a single sensor and susceptibility to noise and angular deviation interference in traditional diagnosis, providing comprehensive and complete raw vibration signal data. Furthermore, by performing multi-domain feature extraction and sample labeling on the vibration signals, and combining this with cross-condition fault impact index fitting to generate fault identification criteria, it addresses the lack of cross-condition adaptability in traditional fault diagnosis. To address the issues of adaptability to operating conditions and the strong subjectivity in fault mode classification, a standardized fault judgment standard adaptable to diverse operating scenarios should be established. By acquiring multi-dimensional operational health data, combining the distance of similar fault characteristics to mine weak fault features and calculate the fault sensitivity probability, a degradation trajectory under missing operating conditions can be generated, which can solve the problems of missing operating condition data gaps and low identification of weak fault features in traditional diagnosis. By extracting bias features based on fault identification criteria and missing degradation trajectories, a dynamic matching bias identification model can be constructed and classified for diagnosis, which can solve the problems of ambiguous bias fault identification and lack of systematic diagnostic results in traditional diagnosis, and improve the comprehensiveness of wind turbine bearing fault diagnosis.

[0091] In summary, the technical solution adopted in this application can classify and diagnose bearing failures of wind turbine generator sets in complex scenarios where multiple operating conditions and missing operating conditions coexist, thereby improving the accuracy of fault diagnosis.

[0092] Example 2: This application provides a wind turbine generator bearing fault diagnosis system, referring to... Figure 4 As shown in the figure, this is a modular structure diagram of a wind turbine generator bearing fault diagnosis system according to this embodiment of the present application. The bearing fault diagnosis system includes:

[0093] The data acquisition module 100 is used to arrange a multi-source sensor array at the main bearing and gearbox bearing of the wind turbine generator set, and to acquire bearing vibration signals during the operation of the wind turbine generator set through the multi-source sensor array.

[0094] The vibration fitting module 200 is used to sample and label the vibration signal of the bearing to obtain the operation fault labels of multiple types of wind turbine bearings, fit all the operation fault labels with the fault impact index of the wind turbine bearing in cross-operating conditions, and generate the fault identification criteria of the wind turbine bearing in cross-operating conditions.

[0095] The fault identification module 300 is used to acquire the operating health data of the wind turbine generator bearing in operation, combine the operating health data with the weak fault features between similar fault feature distances in the wind turbine generator bearing to determine the fault sensitivity probability, and generate the missing degradation trajectory of the wind turbine generator bearing under the fault diagnosis missing condition based on all the fault sensitivity probabilities.

[0096] The fault diagnosis module 400 is used to identify the offset of the wind turbine generator bearing based on the fault identification criteria and the missing degradation trajectory, obtain the offset fault type, and classify and diagnose the wind turbine generator bearing fault according to the offset fault type.

[0097] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0098] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compactdisc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0099] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

Claims

1. A method for diagnosing bearing faults in wind turbine generator sets, characterized in that, The bearing fault diagnosis method includes the following steps: A multi-source sensor array is arranged at the main bearing and gearbox bearing of the wind turbine generator set to collect bearing vibration signals during the operation of the wind turbine generator set. The bearing vibration signal is sampled and labeled to obtain the operation fault labels of various wind turbine bearings. All operation fault labels are fitted with the fault impact index of wind turbine bearings in cross-operating conditions to generate fault identification criteria of wind turbine bearings in cross-operating conditions. Obtain operational health data of the bearings of the wind turbine generator set in operation, combine the operational health data with the weak fault characteristics between similar fault characteristics in the bearings of the wind turbine generator set to determine the fault sensitivity probability, and generate the missing degradation trajectory of the bearings of the wind turbine generator set under the condition of missing fault diagnosis based on all the fault sensitivity probabilities. Specifically, determining the fault sensitivity probability by combining the operational health data with the weak fault characteristics between similar fault feature distances in the wind turbine generator bearings includes: Based on the aforementioned operational health data, the fault impact component of the wind turbine generator bearing vibration signal within the resonant frequency band is extracted; Based on the aforementioned fault impact components, a weak fault feature enhancement index is determined between the distance to similar fault features. The fault sensitivity probability is determined by using the weak fault feature enhancement index and the probability density distribution of operational health data. Based on the fault identification criteria and the missing degradation trajectory, the bearing of the wind turbine generator set is biased to identify the bias fault type, and the bearing fault of the wind turbine generator set is classified and diagnosed according to the bias fault type.

2. The method for diagnosing bearing faults in a wind turbine generator set as described in claim 1, characterized in that, The bearing vibration signals were sampled and labeled to obtain various types of operational fault labels for wind turbine generator bearings, including: Based on the bearing vibration signal, fault features are constructed from the multi-domain fusion characteristic parameters of the wind turbine generator set; Based on the aforementioned fault characteristics, multiple fault operation modes of wind turbine generator bearings during operation are determined. Extract the operating fault tags of the wind turbine generator bearings from each type of fault operation mode.

3. The method for diagnosing bearing faults in a wind turbine generator set as described in claim 1, characterized in that, The aforementioned cross-condition refers to the different operating states of a wind turbine generator set caused by changes in wind speed and load.

4. The method for diagnosing bearing faults in a wind turbine generator set as described in claim 1, characterized in that, The aforementioned weak fault characteristics refer to the characteristic information caused by early and minor bearing faults, which have small signal amplitudes and are easily interfered with by noise.

5. The method for diagnosing bearing faults in a wind turbine generator set as described in claim 1, characterized in that, The aforementioned fault diagnosis missing condition refers to a specific operating condition in which the bearings of the wind turbine generator set fail to collect the data required for fault diagnosis.

6. The method for diagnosing bearing faults in a wind turbine generator set as described in claim 1, characterized in that, Based on the fault identification criteria and the missing degradation trajectory, the bearing of the wind turbine generator is subjected to offset identification, and the offset fault types specifically include: The vibration signal bias features of the wind turbine generator bearings are extracted from the fault identification criteria and the missing degradation trajectory, respectively. Based on the aforementioned vibration signal bias characteristics, a bias identification model is constructed and dynamically matched with a known fault mode library. The vibration signal is evaluated using the bias identification model, and the bias fault type is output.

7. The method for diagnosing bearing faults in a wind turbine generator set as described in claim 1, characterized in that, The aforementioned missing degradation trajectory refers to the complete degradation process curve of the bearing from normal state to fault state, generated under the condition of missing fault diagnosis.

8. The method for diagnosing bearing faults in a wind turbine generator set as described in claim 1, characterized in that, All operational fault labels are fitted with the fault impact index of wind turbine generator bearings across operating conditions to generate fault identification criteria for wind turbine generator bearings across operating conditions. These criteria specifically include: Based on all operational fault labels, vibration signal impact characteristics of wind turbine generator bearings under different operating conditions are extracted. The failure impact index of wind turbine generator bearings under cross-operating conditions was determined based on the impact characteristics of all vibration signals. By performing regression fitting on all operational fault labels and the fault impact index, a mapping model between fault modes and impact responses under cross-operating conditions is established. Based on the mapping relationship model, fault identification criteria for wind turbine generator bearings under different operating conditions are generated.

9. A wind turbine generator bearing fault diagnosis system, used to execute a wind turbine generator bearing fault diagnosis method as described in any one of claims 1 to 8, characterized in that, The bearing fault diagnosis system includes: The data acquisition module is used to arrange a multi-source sensor array at the main bearing and gearbox bearing of the wind turbine generator set, and to collect bearing vibration signals during the operation of the wind turbine generator set through the multi-source sensor array. The vibration fitting module is used to sample and label the vibration signal of the bearing to obtain the operation fault labels of multiple types of wind turbine bearings. All operation fault labels are fitted with the fault impact index of wind turbine bearings in cross-operating conditions to generate fault identification criteria of wind turbine bearings in cross-operating conditions. The fault identification module is used to acquire the operational health data of the wind turbine bearing in operation, combine the operational health data with the weak fault characteristics between similar fault characteristics in the wind turbine bearing to determine the fault sensitivity probability, and generate the missing degradation trajectory of the wind turbine bearing under the fault diagnosis missing condition based on all the fault sensitivity probabilities. The fault diagnosis module is used to identify the offset of the wind turbine generator bearing based on the fault identification criteria and the missing degradation trajectory, obtain the offset fault type, and classify and diagnose the wind turbine generator bearing fault according to the offset fault type.

Images