Bearing fault diagnosis method for multi-source wind turbine
By integrating multiple monitoring data in the wind turbine unit, a comprehensive fault diagnosis model is established, and the problems of incomplete bearing fault diagnosis information and high misdiagnosis rate of existing technology stroke motor units are solved, and high precision and real-time fault detection and early warning are achieved. Improve the operation and maintenance efficiency and economicality of wind turbines.
Patent Information
- Application Number
- CN202411285030.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-13
- Publication Date
- 2024-11-29
- Estimated Expiration
- 2044-09-13
AI Technical Summary
In the diagnosis of bearing faults of wind turbines, there are problems such as incomplete information, high misdiagnosis rate, and difficulty in real-time monitoring, resulting in frequent missed or mis-checked situations.
Using the bearing fault diagnosis method of multi-source wind turbines, the integrated fault diagnosis model is established by integrating vibration signals, temperature data, acoustic signals and lubricant analysis results, and using machine learning and data mining technology to establish a comprehensive fault diagnosis model to realize real-time monitoring and evaluation of the health status of the bearing .
Improve the accuracy and timeliness of fault detection, reduce the risk of omissions or misjudgment of fault detection caused by a single data source, realize comprehensive and accurate status judgment of wind turbine bearings, reduce operation and maintenance costs, and improve wind turbine Operation and maintenance efficiency and economical.
Smart Images

Figure CN118794690B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of wind turbine fault diagnosis, and in particular to a method for diagnosing bearing faults of a multi-source wind turbine. Background Art
[0002] With the rapid development of renewable energy, wind power has become an important source of energy. The reliability of wind turbines directly affects the efficiency and safety of power production. Bearings are key components of wind turbines, and their failure may cause unit shutdown and economic losses. Therefore, timely and accurate diagnosis of bearing failures is crucial.
[0003] In wind turbines, bearings play a key role in transmitting loads and supporting rotating parts. The main components of wind turbines, such as the main shaft and the rotor of the generator, rely on the normal operation of bearings. Bearing failure may not only lead to a decline in equipment performance, but also cause the unit to shut down, resulting in production losses and increased maintenance costs. Common types of bearing failures include wear, fatigue cracks and insufficient lubrication. If the above failures are not detected in time, they will cause serious mechanical damage and even cause safety accidents. At present, the diagnosis of wind turbine bearing failures mainly relies on vibration analysis, temperature monitoring, acoustic signal monitoring and oil analysis. Among them, vibration analysis is to monitor the vibration signal of the bearing using a vibration sensor, and identify the fault characteristics through spectrum analysis, envelope analysis and other technologies. Although vibration analysis is an effective fault detection method, its accuracy is affected by signal noise and external interference. Temperature monitoring is to monitor the operating temperature of the bearing through a temperature sensor, and temperature abnormalities usually indicate potential faults. However, temperature changes may be caused by a variety of factors and need to be combined with other data for comprehensive analysis. Acoustic signal monitoring is to collect acoustic signals during the operation of the bearing using an acoustic sensor, and judge the fault through acoustic wave feature analysis. This method is sensitive to environmental noise and may lead to misjudgment. Oil analysis is the process of evaluating the health of bearings by analyzing the lubricating oil to check for the presence of metal particles and contaminants. This method usually requires periodic sampling and does not provide real-time monitoring.
[0004] At present, bearing fault diagnosis mainly adopts single methods such as vibration analysis, temperature monitoring, acoustic signals, etc., which often have problems such as incomplete information and high misdiagnosis rate. It is difficult to fully and accurately judge the state of the bearing, and it is easy to miss or misdiagnose. Therefore, given that a single data analysis method can no longer meet the fault diagnosis needs of modern wind turbines, a bearing fault diagnosis method based on multi-source wind turbine bearing fault diagnosis is particularly important. Summary of the invention
[0005] In view of this, the present invention proposes a multi-source wind turbine bearing fault diagnosis method, which aims to improve the detection accuracy and timeliness of wind turbine bearing faults by integrating multiple monitoring data. The method combines vibration signals, temperature data, acoustic signals and lubricating oil analysis results, and establishes a comprehensive fault diagnosis model through machine learning and data mining technology, thereby realizing real-time monitoring and evaluation of the bearing health status.
[0006] The present invention is implemented by the following technical solutions:
[0007] In a first aspect, the present invention provides a method for diagnosing bearing faults of a multi-source wind turbine, the method comprising the following steps:
[0008] Acquire multi-source heterogeneous data for bearing fault diagnosis data fusion, and after pre-processing the multi-source heterogeneous data by denoising, standardizing and processing missing values, obtain bearing sensor data that fully covers the operation of wind turbines;
[0009] The obtained bearing sensor data is subjected to feature extraction to obtain the vibration signal features, temperature data features, acoustic signal features and lubricating oil analysis features of the wind turbine bearing, and the extracted features are fused to form a comprehensive feature vector;
[0010] An unsupervised feature learning model is trained using the comprehensive feature vector to learn the characteristic patterns of the normal and faulty states of the bearing, and a bearing fault diagnosis model is established based on the trained unsupervised feature learning model;
[0011] The real-time monitoring bearing sensor data is input into the bearing fault diagnosis model, and the operating status and abnormality score of the wind turbine bearing at each monitoring moment are output;
[0012] Perform anomaly detection, identify anomaly points, and determine anomaly types based on the output operation status and anomaly scores;
[0013] Using time series analysis historical data and anomalies as input, the preset fault prediction model is applied to generate fault prediction results, and fault warning signals are generated according to the warning threshold.
[0014] As a further solution of the present invention, the acquired multi-source heterogeneous data includes vibration signals from acceleration sensors, temperature data from temperature sensors, acoustic signals from acoustic sensors, and lubricating oil analysis data from lubricating oil analysis instruments.
[0015] As a further solution of the present invention, preprocessing is performed to obtain bearing sensor data, including the following steps:
[0016] The vibration signal is denoised using wavelet transform;
[0017] Perform Z-score standardization on the denoised data to make the data mean 0 and variance 1;
[0018] The standardized data is mean filled, missing values are filled and completed to obtain the bearing sensor data.
[0019] As a further solution of the present invention, the vibration signal features, temperature data features, acoustic signal features and lubricating oil analysis features of the wind turbine bearing are extracted, and the extracted features are fused to form a comprehensive feature vector, including the following steps:
[0020] Extract the time domain features, frequency domain features and time-frequency domain features of the vibration signal;
[0021] Extract the mean and rate of change of temperature data;
[0022] Extract the sound pressure level and Mel-frequency cepstrum coefficients of the acoustic signal;
[0023] Extract lubricating oil characteristics of viscosity, acidity and water content;
[0024] The extracted feature vectors are weighted and fused to form a comprehensive feature vector.
[0025] As a further solution of the present invention, when the unsupervised feature learning model is trained using the comprehensive feature vector, the autoencoder model is trained using the comprehensive feature vector to learn the characteristic patterns of the normal and faulty states of the bearing by minimizing the reconstruction error; wherein the loss function of the autoencoder model is the reconstruction error loss function, which is defined as:
[0026] ;
[0027] In the formula, is the original feature vector, is to reconstruct the feature vector; among them, the anomaly score is calculated for the comprehensive feature vector at each monitoring moment, which is defined as:
[0028] ; and compared with preset thresholds to detect faults.
[0029] As a further solution of the present invention, the network structure of the autoencoder model includes: an input layer for receiving a comprehensive feature vector; an encoding layer for mapping the input features to a low-dimensional representation; a bottleneck layer for feature representation in the minimum dimension; a decoding layer for mapping the features of the bottleneck layer back to the original feature space; and an output layer for reconstructing the input data.
[0030] As a further solution of the present invention, before the real-time monitored bearing sensor data is input into the bearing fault diagnosis model, it also includes:
[0031] Acquire real-time monitoring data including vibration signals, temperature, acoustic signals and lubricant analysis data from bearing sensors;
[0032] The acquired real-time monitoring data is converted into the same feature vector format as the training data, a high-pass filter is used to remove low-frequency noise, the real-time monitoring data is standardized and missing values are processed;
[0033] Extract features from real-time monitoring data and generate feature vectors of real-time data;
[0034] The real-time monitored feature vector is input into the trained fault diagnosis model.
[0035] As a further solution of the present invention, when outputting the operating status and abnormality score of the wind turbine bearing at each monitoring moment, reconstruction is calculated based on the autoencoder model, the input feature vector is processed by the encoder and decoder, and the reconstructed feature vector is output; and the reconstruction error is calculated; the obtained reconstruction error is the abnormality score.
[0036] As a further solution of the present invention, when performing anomaly detection, identifying anomaly points, and determining anomaly types based on the output operating status and anomaly scores, the following steps are included:
[0037] Setting the threshold by analyzing the distribution of abnormal scores of normal samples and fault samples in the training data;
[0038] Compare the anomaly score of real-time data with the set threshold. When the anomaly score is higher than the threshold, it is detected as an anomaly and the real-time anomaly score is calculated.
[0039] When the anomaly score exceeds the threshold, the sensor data at that time point is recorded as an anomaly point;
[0040] Perform trend and amplitude analysis on the detected abnormal points to determine the abnormal points and identify whether there are periodic or sudden abnormalities;
[0041] Extract the vibration frequency and temperature variation characteristics of the abnormal point, use the labeled abnormal data to train the classification model, input the abnormal point characteristics into the trained classification model, and output the abnormal type;
[0042] Match the output of the classification model with known failure patterns to identify the specific anomaly type.
[0043] The present invention also includes a computer device, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor executes the multi-source wind turbine bearing fault diagnosis method.
[0044] The present invention also includes a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable the computer to execute the multi-source wind turbine bearing fault diagnosis method.
[0045] Compared with the prior art, the multi-source wind turbine bearing fault diagnosis method provided by the present invention has the following beneficial effects:
[0046] 1. Enhanced comprehensiveness and accuracy of data: This method comprehensively covers the operating status of wind turbine bearings by acquiring and fusing multi-source heterogeneous data, greatly improving the comprehensiveness of fault diagnosis and data accuracy, enabling the system to capture more status features and reducing the risk of fault detection omissions or misjudgments caused by a single data source.
[0047] 2. Improve fault identification capability through multi-dimensional feature extraction: This method constructs a comprehensive feature vector containing multiple information by extracting the time domain, frequency domain and time-frequency domain characteristics of the vibration signal, the mean and change rate of the temperature data, the sound pressure level and Mel-frequency cepstrum coefficients of the acoustic signal, and the viscosity, acid value and moisture content characteristics of the lubricating oil. It can better reflect the actual operating status of the bearing, enabling the diagnostic model to more accurately identify complex fault modes.
[0048] 3. The unsupervised learning model is applicable and flexible: This method uses an unsupervised feature learning model to learn the features of normal and fault states. It can effectively build a fault diagnosis model without a large number of fault samples. It learns the normal distribution of data by minimizing the reconstruction error, so that abnormal situations can be detected faster and more accurately in practical applications, and it has strong generalization ability and adaptability.
[0049] 4. Realized rapid response of real-time diagnosis and abnormality detection: This method can quickly output the operating status and abnormality score of wind turbine bearings based on real-time data monitoring, thereby achieving instant fault diagnosis and abnormality detection. Through efficient real-time monitoring and rapid response capabilities, it can effectively reduce the downtime of wind turbines caused by faults, reduce maintenance costs, and improve the operation and maintenance efficiency and economy of wind turbines.
[0050] 5. Possess sensitive anomaly detection and classification capabilities, realize fault prediction and early warning to enhance operation and maintenance safety: Based on the output anomaly score and preset threshold, this method can quickly identify anomalies and determine the type of anomalies. By analyzing the distribution of anomaly scores, the optimal detection threshold can be automatically set, further improving the sensitivity of anomaly detection. In addition, through the feature extraction of anomalies and the training of classification models, the detected anomaly types can be effectively classified and identified to provide support for accurate fault repair. This method also uses time series analysis of historical data and anomaly information to predict potential fault risks in advance and generate corresponding early warning signals, so that operation and maintenance personnel can take preventive measures before a fault occurs, which can not only reduce the risk of failure and extend the service life of wind turbines, but also significantly improve the overall safety and reliability of operation.
[0051] To sum up, the multi-source wind turbine bearing fault diagnosis method of the present invention has significant advantages in data acquisition, preprocessing, feature extraction, model training, real-time diagnosis, anomaly detection and fault prediction, and can effectively improve the operating safety and economy of wind turbines. The rapid response and high accuracy of the fault diagnosis and early warning system reduce the time and manpower investment of operation and maintenance personnel in fault troubleshooting and repair, and reduce the wind turbine downtime losses caused by faults, thereby significantly reducing the overall operation and maintenance costs and improving the economic benefits of wind farms.
[0052] The above aspects or other aspects of the present invention will be more concise and easy to understand in the following description of the embodiments. It should be understood that the above general description and the following detailed description are only exemplary and explanatory and cannot limit the present invention. BRIEF DESCRIPTION OF THE DRAWINGS
[0053] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or related technologies, the following briefly introduces the drawings required for use in the exemplary embodiments or related technical descriptions. The drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation to the present invention. In the drawings:
[0054] Figure 1 The present invention is a flowchart of a method for diagnosing bearing faults of a multi-source wind turbine generator system according to an embodiment of the present invention.
[0055] Figure 2 The present invention is a flowchart of bearing sensor data obtained by preprocessing in the multi-source wind turbine bearing fault diagnosis method according to an embodiment of the present invention.
[0056] Figure 3 The present invention is a flowchart of extracting and fusing features to form a comprehensive feature vector in a multi-source wind turbine bearing fault diagnosis method according to an embodiment of the present invention.
[0057] Figure 4 The present invention is a flowchart of performing abnormality detection, identifying abnormal points and determining abnormality types in a multi-source wind turbine bearing fault diagnosis method according to an embodiment of the present invention. DETAILED DESCRIPTION
[0058] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.
[0059] In some processes described in the specification and claims of the present invention and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that the operations may not be executed in the order in which they appear in this article or in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any execution order. In addition, the process may include more or fewer operations, and the operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., and do not represent the order of precedence, nor do they limit the "first" and "second" to be different types.
[0060] The following will be combined with the drawings in the exemplary embodiments of the present invention to clearly and completely describe the technical solutions in the exemplary embodiments of the present invention. Obviously, the exemplary embodiments described are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.
[0061] Bearing fault diagnosis mainly adopts single methods such as vibration analysis, temperature monitoring, acoustic signals, etc., which often have problems such as incomplete information and high misdiagnosis rate. It is difficult to comprehensively and accurately judge the status of the bearing, and missed detection or misdetection is prone to occur. The present invention provides a multi-source wind turbine bearing fault diagnosis method, which improves the detection accuracy and timeliness of wind turbine bearing faults by integrating multiple monitoring data, combines vibration signals, temperature data, acoustic signals and lubricating oil analysis results, and establishes a comprehensive fault diagnosis model through machine learning and data mining technology, thereby realizing real-time monitoring and evaluation of the health status of the bearing.
[0062] The technical solution of the present invention is further described below in conjunction with specific embodiments:
[0063] See also Figure 1 As shown, Figure 1A flow chart of a multi-source wind turbine bearing fault diagnosis method provided by the present invention. A multi-source wind turbine bearing fault diagnosis method provided in one embodiment of the present invention comprises the following steps:
[0064] Step S10, obtaining multi-source heterogeneous data for bearing fault diagnosis data fusion, and performing denoising, standardization and missing value processing pre-processing on the multi-source heterogeneous data to obtain bearing sensor data that fully covers the operation of the wind turbine.
[0065] In this step, the multi-source heterogeneous data obtained include but are not limited to: vibration signals from acceleration sensors, temperature data from temperature sensors, acoustic signals from acoustic sensors, and lubricating oil analysis data from lubricating oil analysis instruments.
[0066] Vibration signal: comes from the acceleration sensor, the sampling frequency is 1kHz, and the data range includes low-frequency components of 0-20Hz and high-frequency components of 20-1000Hz.
[0067] Temperature data: comes from the temperature sensor, the sampling frequency is 1Hz, and the data range is 0℃-100℃.
[0068] Acoustic signal: comes from the acoustic sensor, with a sampling frequency of 20kHz and a data range of 20Hz-10kHz.
[0069] Lubricating oil analysis data: including viscosity, acid value and water content, with a sampling frequency of 10Hz.
[0070] In this embodiment, see Figure 2 As shown, the preprocessing of the bearing sensor data includes the following steps:
[0071] Step S101: De-noise the vibration signal using wavelet transform.
[0072] In this step, Daubechies wavelet is used to transform and denoise the vibration signal, and the decomposition level is 4, where the wavelet function is:
[0073] ;
[0074] In the formula, is the wavelet transform coefficient, which indicates the vibration signal at Under the wavelet decomposition, The projection value on the wavelet basis function; is the original signal at time point The value of For the Layer, wavelet basis functions, representing the fluctuation characteristics at different scales and positions; N is the total number of sampling points of the original signal.
[0075] Step S102: Perform Z-score standardization on the denoised data to make the data mean 0 and the variance 1.
[0076] In this step, when performing Z-score standardization, the ratio of the difference between the original data value and the mean of the original data to the standard deviation of the original data is used as the standardized data value to indicate the position of the data point in the standard normal distribution.
[0077] Step S103: perform mean filling on the standardized data, fill in and complete the missing values, and obtain bearing sensor data.
[0078] Among them, the mean of the original data is used to fill the missing values to obtain the data point value after filling the missing values.
[0079] Exemplarily, when the bearing sensor data of the wind turbine operation is fully covered by the multi-source wind turbine bearing fault diagnosis method of this embodiment, the following multi-source heterogeneous data is collected in a selected wind turbine:
[0080] Vibration signal: The original data is [0.5, 0.7, 0.65, 0.8, 0.6].
[0081] Temperature data: The original data is [30℃, 32℃, 31℃, 30.75℃, 30℃].
[0082] The processing process for preprocessing multi-source heterogeneous data for denoising, standardization and missing value processing is as follows:
[0083] (1) Vibration signal: First, wavelet transform is used for denoising, and the denoised data is [0.5, 0.7, 0.65, 0.8, 0.6]; then, Z-score standardization is used. If the mean is 0.68 and the standard deviation is 0.1, the standardized data is [-1.8, 0.2, -0.3, 1.2, -0.3]; finally, the missing values are filled with the mean, and the filled data is [-1.8, 0.2, -0.3, 1.2, -0.3].
[0084] (2) Temperature data: fill in the missing values with the mean. If the mean is 30.75°C, the processed data will be [30°C, 32°C, 31°C, 30.75°C, 30°C].
[0085] The comprehensive multi-source wind turbine bearing fault diagnosis method provided in this embodiment effectively improves the data quality through data preprocessing such as wavelet transform denoising, Z-score standardization and mean padding. The processed data is integrated and fault diagnosis is performed using advanced feature learning models, which can achieve accurate fault identification and prediction and improve the operational reliability of wind turbines.
[0086] Step S20, extracting features from the obtained bearing sensor data to obtain vibration signal features, temperature data features, acoustic signal features and lubricating oil analysis features of the wind turbine bearing, and fusing the extracted features to form a comprehensive feature vector.
[0087] In this step, see Figure 3 As shown, the vibration signal features, temperature data features, acoustic signal features and lubricating oil analysis features of the wind turbine bearing are extracted, and the extracted features are fused to form a comprehensive feature vector, including the following steps:
[0088] Step S201: extracting time domain features, frequency domain features, and time-frequency domain features of the vibration signal.
[0089] In this embodiment, the time domain characteristics of the vibration signal include mean, standard deviation, and root mean square (RMS), and the calculation formula is:
[0090] ;
[0091] ;
[0092] ;
[0093] In the formula, For the The value of the data point; N is the total number of sampling points of the original signal.
[0094] When the vibration signal data set is: Then, the calculated mean is:
[0095] ;
[0096]
[0097]
[0098] ;
[0099] ;
[0100] In this embodiment, the frequency domain feature includes power spectral density (PSD), which is calculated as follows:
[0101] ;
[0102] In the formula, For time point The signal value of is the total time length of the signal; is the frequency; is an imaginary unit, , N is the total number of sampling points of the original signal, that is, the number of sampling points of the signal.
[0103] For example, when there is a signal data with a length of N=8, , calculate the power spectral density (PSD) at the frequency ;but:
[0104] ;
[0105] Then extract the time-frequency domain features, extract the wavelet coefficients through wavelet transform and get the following coefficients:
[0106] ;
[0107] The first layer wavelet coefficients:
[0108] ;
[0109] The second layer wavelet coefficients:
[0110] ;
[0111] The energy spectral density of the first-layer wavelet transform is:
[0112] .
[0113] Step S202: extract the mean and change rate of the temperature data.
[0114] In this step, assume that the temperature data is t=[30.0, 30.5, 30.8, 31.0, 31.2], unit: °C; then the mean is 30.7 °C, and the rate of change is between t2 and t3, which is (30.8 °C-30.5 °C) / 1h=0.3 °C / h.
[0115] Step S203: extracting the sound pressure level and Mel-frequency cepstrum coefficients of the acoustic signal.
[0116] In this step, assuming that the measured sound pressure P = 0.1Pa and the reference sound pressure P0 = 20uPa, the Mel frequency cepstrum coefficient is 20log 10 [0.1÷(20×10 -6 )]=73.98dB.
[0117] Step S204: extracting the viscosity, acid value and water content characteristics of the lubricating oil.
[0118] In this step, it is assumed that the characteristic values of the lubricating oil are:
[0119] Viscosity: 30 cSt;
[0120] Acid value: 1.5 mg KOH / g;
[0121] Moisture content: 0.05%.
[0122] Step S205: weighted fusion is performed on the extracted feature vectors to form a comprehensive feature vector.
[0123] In this step, the features extracted above are weighted and fused. Assume that the weights of each feature are: .
[0124] When calculating the comprehensive feature vector, the weight of each feature is multiplied by the feature vector and then weighted fused to obtain:
[0125] .
[0126] Assume that the extracted eigenvalues are as follows:
[0127] Vibration signature: [0.46,2.99]
[0128] Temperature characteristics: [30.7, 0.3]
[0129] Acoustic signature: [73.98]
[0130] Lubricant characteristics: [30, 1.5, 0.05]
[0131] The fused comprehensive feature vector is: 0.4×[0.46,2.99]+0.2×[30.7,0.3]+0.2×[73.98]+0.2×[30,1.5,0.05]=[27.12,1.856,0.01].
[0132] The comprehensive feature vector can be used as a comprehensive state representation of the wind turbine bearing and further used for analysis or prediction.
[0133] Through the above detailed steps, various data features of wind turbine bearings are systematically extracted, including the time domain, frequency domain and time-frequency domain features of vibration signals, the mean and change rate of temperature data, the sound pressure level and Mel frequency cepstrum coefficients of acoustic signals, and the viscosity, acid value and moisture content characteristics of lubricating oil. The features are weighted and fused to form a comprehensive feature vector for subsequent fault diagnosis and analysis. In this way, comprehensive status information of wind turbine bearings can be obtained, thereby improving the accuracy of fault prediction and maintenance decisions.
[0134] Step S30: Use the comprehensive feature vector to train an unsupervised feature learning model to learn the characteristic patterns of the normal and faulty states of the bearing, and establish a bearing fault diagnosis model based on the trained unsupervised feature learning model.
[0135] In this step, when the unsupervised feature learning model is trained using the comprehensive feature vector, the autoencoder model is trained using the comprehensive feature vector to learn the characteristic patterns of the normal and faulty states of the bearing by minimizing the reconstruction error; wherein the loss function of the autoencoder model is the reconstruction error loss function, which is defined as:
[0136] ;
[0137] In the formula, is the original feature vector, is to reconstruct the feature vector; among them, the anomaly score is calculated for the comprehensive feature vector at each monitoring moment, which is defined as:
[0138] ;
[0139] And compared with preset thresholds to detect faults.
[0140] During operation, the comprehensive feature vector data of the bearing is collected, and the comprehensive feature vector data includes the features in the normal state and the fault state. The comprehensive feature vector is used to train the autoencoder model. The model is optimized by minimizing the reconstruction error loss function, so that the model can effectively reconstruct the original feature vector from the low-dimensional representation, reconstruct the feature vector at each monitoring moment, and calculate the reconstruction error (i.e., the anomaly score). The anomaly score is compared with the preset threshold to determine whether there is an anomaly or a fault.
[0141] The network structure of the autoencoder model includes: an input layer for receiving a comprehensive feature vector; an encoding layer for mapping the input features to a low-dimensional representation; a bottleneck layer for representing features in the minimum dimension; a decoding layer for mapping the features of the bottleneck layer back to the original feature space; and an output layer for reconstructing the input data.
[0142] For example, assume that there is the following comprehensive feature vector data (each vector has 10 features):
[0143] ;
[0144] Use the above data to train the autoencoder, obtain the trained model, reconstruct and calculate the anomaly score.
[0145] Among them, for the test data Reconstruct and obtain the reconstructed feature vector ;like ;
[0146] Then, calculate the anomaly score:
[0147] .
[0148] Assume that the preset threshold is 0.2. If the calculated anomaly score is greater than the threshold, a fault is detected, and since 0.1 < 0.2, no fault is detected in this example.
[0149] In bearing fault diagnosis, the above steps are provided, and the autoencoder model identifies the characteristic patterns of normal state and fault state by performing unsupervised learning on the comprehensive feature vector, effectively mapping the data from high dimension to low dimension, and then mapping the low-dimensional data back to the original space. Fault detection can be achieved by calculating the anomaly score and comparing it with the preset threshold. The present invention utilizes the autoencoder to learn the characteristic pattern without labeled data, so as to effectively detect the unknown fault state.
[0150] Step S40: input the real-time monitored bearing sensor data into the bearing fault diagnosis model, and output the operating status and abnormality score of the wind turbine bearing at each monitoring moment.
[0151] In this step, before the real-time monitored bearing sensor data is input into the bearing fault diagnosis model, it also includes:
[0152] Acquire real-time monitoring data from bearing sensors, including vibration signals, temperature, acoustic signals, and lubricant analysis data;
[0153] The acquired real-time monitoring data is converted into the same feature vector format as the training data, a high-pass filter is used to remove low-frequency noise, the real-time monitoring data is standardized and missing values are processed;
[0154] Extract features from real-time monitoring data and generate feature vectors of real-time data;
[0155] The real-time monitored feature vector is input into the trained fault diagnosis model.
[0156] In this embodiment, when the operating status and abnormality score of the wind turbine bearing at each monitoring moment are output, reconstruction is calculated based on the autoencoder model, the input feature vector is processed by the encoder and decoder, and the reconstructed feature vector is output; and the reconstruction error is calculated; the obtained reconstruction error is the abnormality score.
[0157] This embodiment uses an autoencoder model to perform fault diagnosis on bearing data monitored in real time. The process includes data acquisition, preprocessing, feature extraction, model prediction, and anomaly detection. The autoencoder generates feature reconstruction through encoding and decoding processes, and calculates the reconstruction error to evaluate the degree of anomaly. Finally, it determines whether there is a fault based on the set threshold, thereby achieving effective monitoring and maintenance of wind turbine bearings.
[0158] Step S50: perform anomaly detection, identify abnormal points and determine the abnormal type based on the output operation status and abnormality score.
[0159] In this step, see Figure 4 As shown, when performing anomaly detection, identifying anomaly points, and determining anomaly types based on the output operating status and anomaly scores, the following steps are included:
[0160] Step S501, setting a threshold by analyzing the abnormal score distribution of normal samples and fault samples in the training data;
[0161] Step S502: compare the anomaly score of the real-time data with a set threshold. When the anomaly score is higher than the threshold, it is detected as an anomaly and the real-time anomaly score is calculated.
[0162] Step S503: when the abnormality score exceeds the threshold, the sensor data at this time point is recorded as an abnormal point;
[0163] Step S504: Perform trend and amplitude analysis on the detected abnormal points to determine the abnormal points, and identify whether there are periodic abnormalities or sudden abnormalities;
[0164] Step S505, extract the vibration frequency and temperature variation amplitude characteristics of the abnormal point, use the marked abnormal data to train the classification model, input the abnormal point characteristics into the trained classification model, and output the abnormal type;
[0165] Step S506: Match the output of the classification model with the known fault mode to confirm the specific abnormality type.
[0166] For example, if the anomaly score of an abnormal point in the real-time monitoring data is 0.25, the threshold is set to 0.2, and the anomaly detection judges 0.25>0.2, it is detected as abnormal, and the data at that time point is recorded; the recorded time point is t=100s, and the corresponding sensor data such as vibration signal and temperature. Then, trend and amplitude analysis is performed. The vibration signal shows that there is a trend of increasing frequency in the past 10 seconds, and the temperature data shows that the temperature has risen sharply in the past 5 seconds, with a change of 15°C; the vibration frequency f=50Hz and the temperature change amplitude of 15°C are extracted, and the feature vector [50,15] is input into the classification model, and the output abnormal type is "insufficient lubrication". Finally, the classification result matches the known fault mode "insufficient lubrication" to confirm the abnormal type.
[0167] Step S60: using the time series analysis historical data and abnormal points as input, applying a preset fault prediction model to generate a fault prediction result, and generating a fault warning signal according to a warning threshold.
[0168] In this step, historical data and detected anomalies are collected to form a time series data set. The time series data is analyzed using a preset fault prediction model. The time series data is predicted by the fault prediction model to obtain fault prediction results for future time periods. The prediction results are compared with the preset warning threshold. If the prediction results exceed the threshold, a fault warning signal is generated.
[0169] A multi-source wind turbine bearing fault diagnosis method of the present invention comprehensively covers the operating status of the wind turbine bearing by acquiring and fusing multi-source heterogeneous data, greatly improving the comprehensiveness of fault diagnosis and the accuracy of data, enabling the system to capture more state features, and reducing the risk of fault detection omission or misjudgment caused by a single data source. By extracting the time domain, frequency domain and time-frequency domain features of the vibration signal, the mean and change rate of the temperature data, the sound pressure level and Mel frequency cepstrum coefficient of the acoustic signal, and the viscosity, acid value and moisture content features of the lubricating oil, a comprehensive feature vector containing multiple information is constructed, which can better reflect the actual operating status of the bearing, so that the diagnosis model can more accurately identify complex fault modes. An unsupervised feature learning model is used to learn the features of normal and fault states, and a fault diagnosis model can be effectively constructed without a large number of fault samples. The normal distribution of data is learned by minimizing the reconstruction error, so that abnormal situations can be detected faster and more accurately in practical applications, and it has strong generalization ability and adaptability. The method can quickly output the operating status and abnormality score of the wind turbine bearing on the basis of real-time data monitoring, thereby realizing instant fault diagnosis and abnormality detection. Through efficient real-time monitoring and rapid response capabilities, the downtime of wind turbines caused by faults can be effectively reduced, maintenance costs can be reduced, and the operation and maintenance efficiency and economy of wind turbines can be improved. Based on the output anomaly score and preset threshold, the anomaly point can be quickly identified and the anomaly type can be determined. By analyzing the distribution of anomaly scores, the optimal detection threshold can be automatically set, further improving the sensitivity of anomaly detection. In addition, through the feature extraction of anomaly points and the training of classification models, the detected anomaly types can be effectively classified and identified, providing support for accurate fault repair. This method also uses time series analysis of historical data and anomaly point information to predict potential fault risks in advance and generate corresponding early warning signals, so that operation and maintenance personnel can take preventive measures before the fault occurs, which can not only reduce the risk of faults and extend the service life of wind turbines, but also significantly improve the safety and reliability of overall operation.
[0170] To sum up, the multi-source wind turbine bearing fault diagnosis method of the present invention has significant advantages in data acquisition, preprocessing, feature extraction, model training, real-time diagnosis, anomaly detection and fault prediction, and can effectively improve the operating safety and economy of wind turbines. The rapid response and high accuracy of the fault diagnosis and early warning system reduce the time and manpower investment of operation and maintenance personnel in fault troubleshooting and repair, and reduce the wind turbine downtime losses caused by faults, thereby significantly reducing the overall operation and maintenance costs and improving the economic benefits of wind farms.
[0171] It should be understood that, although described in a certain order, these steps are not necessarily performed in sequence in the above order. Unless there is clear explanation in this article, the execution of these steps does not have strict order restriction, and these steps can be performed in other orders. Moreover, a part of the steps of the present embodiment may include a plurality of steps or a plurality of stages, and these steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a part of the steps or stages in other steps or other steps.
[0172] In one embodiment, the present invention provides a multi-source wind turbine bearing fault diagnosis system, which is used to perform the multi-source wind turbine bearing fault diagnosis method, and the system includes:
[0173] Bearing sensor data acquisition module: used to acquire multi-source heterogeneous data, and perform standardization on the acquired multi-source heterogeneous data to generate bearing sensor data that fully covers the operation of wind turbines;
[0174] Feature vector extraction module: used to extract features from the bearing sensor data, extract the vibration signal features, temperature data features, acoustic signal features and lubricating oil analysis features of the wind turbine bearing, and fuse the extracted features to form a comprehensive feature vector;
[0175] Diagnostic model building module: used to train an unsupervised feature learning model using a comprehensive feature vector, learn the characteristic patterns of normal and faulty states of the bearing, and establish a bearing fault diagnosis model based on the trained unsupervised feature learning model;
[0176] Real-time diagnosis and analysis module: inputs the real-time monitoring bearing sensor data into the bearing fault diagnosis model, and outputs the operating status and abnormality score of the wind turbine bearing at each monitoring moment
[0177] Anomaly detection module: used to detect anomalies, identify abnormal points, and determine the type of anomalies based on the output operation status and anomaly scores;
[0178] Fault prediction and warning module: It is used to use time series analysis historical data and abnormal points as input, apply the preset fault prediction model to generate fault prediction results, and generate fault warning signals according to the warning threshold.
[0179] In this embodiment, the job generation system further includes a time series anomaly detection module, which is used to analyze historical data and anomalies according to the time series, perform anomaly detection, identify anomalies, and determine anomaly types according to the operating status and anomaly scores.
[0180] In this embodiment, the multi-source wind turbine bearing fault diagnosis system adopts the steps of a multi-source wind turbine bearing fault diagnosis method as described above during execution. Therefore, the operation process of the multi-source wind turbine bearing fault diagnosis system is not described in detail in this embodiment.
[0181] In one embodiment, a computer device is also provided in an embodiment of the present invention, comprising at least one processor and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor executes the steps of the multi-source wind turbine bearing fault diagnosis method.
[0182] In one embodiment, the present invention further provides a computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable the computer to execute the steps of the multi-source wind turbine bearing fault diagnosis method.
[0183] A person skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be realized by instructing the relevant hardware through a computer program represented by computer instructions, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory.
[0184] Non-volatile memory may include read-only memory, magnetic tape, floppy disk, flash memory or optical storage, etc. Volatile memory may include random access memory or external cache memory. As an illustration and not limitation, RAM may be in various forms, such as static random access memory or dynamic random access memory, etc.
[0185] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A method for diagnosing bearing faults of a multi-source wind turbine, characterized in that: The method comprises the following steps: Acquire multi-source heterogeneous data for bearing fault diagnosis data fusion, and after pre-processing the multi-source heterogeneous data by denoising, standardizing and processing missing values, obtain bearing sensor data that fully covers the operation of wind turbines; The obtained bearing sensor data is subjected to feature extraction to obtain the vibration signal features, temperature data features, acoustic signal features and lubricating oil analysis features of the wind turbine bearing, and the extracted features are fused to form a comprehensive feature vector; An unsupervised feature learning model is trained using the comprehensive feature vector to learn the characteristic patterns of the normal and faulty states of the bearing, and a bearing fault diagnosis model is established based on the trained unsupervised feature learning model; The real-time monitoring bearing sensor data is input into the bearing fault diagnosis model, and the operating status and abnormality score of the wind turbine bearing at each monitoring moment are output; Perform anomaly detection, identify anomaly points, and determine anomaly types based on the output operation status and anomaly scores; Using time series analysis historical data and abnormal points as input, applying the preset fault prediction model to generate fault prediction results, and generating fault warning signals based on the warning threshold; Before the real-time monitoring bearing sensor data is input into the bearing fault diagnosis model, it also includes: Acquire real-time monitoring data from bearing sensors, including vibration signals, temperature, acoustic signals, and lubricant analysis data; The acquired real-time monitoring data is converted into the same feature vector format as the training data, a high-pass filter is used to remove low-frequency noise, the real-time monitoring data is standardized and missing values are processed; Extract features from real-time monitoring data and generate feature vectors of real-time data; Input the real-time monitored feature vector into the trained fault diagnosis model; The extracted features are fused to form a comprehensive feature vector, including the following steps: Extract the time domain features, frequency domain features and time-frequency domain features of the vibration signal; Extract the mean and rate of change of temperature data; Extract the sound pressure level and Mel-frequency cepstrum coefficients of the acoustic signal; Extract lubricating oil characteristics of viscosity, acidity and water content; Perform weighted fusion on the extracted feature vectors to form a comprehensive feature vector; Among them, the frequency domain features include power spectrum density, and the calculation formula is: ; In the formula, For time point The signal value of is the total time length of the signal; is the frequency; is an imaginary unit, , N is the total number of sampling points of the original signal; when calculating the comprehensive feature vector, the weight of each feature is multiplied by the feature vector and then weighted fused, that is: in, .
2. The multi-source wind turbine bearing fault diagnosis method according to claim 1, characterized in that: The acquired multi-source heterogeneous data include vibration signals from acceleration sensors, temperature data from temperature sensors, acoustic signals from acoustic sensors, and lubricating oil analysis data from lubricating oil analysis instruments.
3. The multi-source wind turbine bearing fault diagnosis method according to claim 2, characterized in that: Preprocessing is performed to obtain bearing sensor data, including the following steps: The vibration signal is denoised using wavelet transform; Perform Z-score standardization on the denoised data to make the data mean 0 and variance 1; The standardized data is mean filled, missing values are filled and completed to obtain the bearing sensor data.
4. The multi-source wind turbine bearing fault diagnosis method according to claim 3, characterized in that: When the unsupervised feature learning model is trained using the comprehensive feature vector, the autoencoder model is trained using the comprehensive feature vector to learn the characteristic patterns of the normal and faulty states of the bearing by minimizing the reconstruction error; wherein the loss function of the autoencoder model is the reconstruction error loss function, which is defined as: ; In the formula, is the original eigenvector, is to reconstruct the feature vector; among them, the anomaly score is calculated for the comprehensive feature vector at each monitoring moment, which is defined as: ; And compared with preset thresholds to detect faults.
5. The multi-source wind turbine bearing fault diagnosis method according to claim 4, characterized in that: When outputting the operating status and abnormality score of the wind turbine bearing at each monitoring moment, the reconstruction is calculated based on the autoencoder model, the input feature vector is processed by the encoder and decoder, and the reconstructed feature vector is output; and the reconstruction error is calculated; the obtained reconstruction error is the abnormality score.
6. The method for diagnosing bearing faults of a multi-source wind turbine set according to claim 5, characterized in that: When performing anomaly detection, identifying anomaly points, and determining anomaly types based on the output operation status and anomaly scores, the following steps are included: Setting the threshold by analyzing the distribution of abnormal scores of normal samples and fault samples in the training data; Compare the anomaly score of real-time data with the set threshold. When the anomaly score is higher than the threshold, it is detected as an anomaly and the real-time anomaly score is calculated. When the anomaly score exceeds the threshold, the sensor data at that time point is recorded as an anomaly point; Perform trend and amplitude analysis on the detected abnormal points to determine the abnormal points and identify whether there are periodic or sudden abnormalities; Extract the characteristics of the vibration frequency and temperature change amplitude of the abnormal point, use the labeled abnormal data to train the classification model, input the abnormal point characteristics into the trained classification model, and output the abnormal type; Match the output of the classification model with known failure patterns to identify the specific anomaly type.
7. A computer device comprising: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor executes the multi-source wind turbine bearing fault diagnosis method according to any one of claims 1-6.
8. A computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable the computer to execute the multi-source wind turbine bearing fault diagnosis method according to any one of claims 1 to 6.
Citation Information
Patent Citations
Wind turbine generator vibration monitoring and fault diagnosis method
CN118327909A
