Early warning method for oil leakage fault of wind driven generator

Abstract

The invention discloses a wind driven generator oil leakage fault early warning method, which comprises the following steps: arranging vibration, temperature and oil sensors at a monitoring part, collecting multi-dimensional data for comprehensive analysis, analyzing a vibration signal by using a frequency spectrum to judge a mechanical fault, and combining a temperature trend and an oil parameter abnormal condition to realize early warning of an oil leakage fault of a wind driven generator. And carrying out classified early warning on the oil leakage fault by adopting a support vector machine algorithm. Furthermore, a Bayesian network algorithm is used to carry out accurate reasoning on the oil leakage position. According to the invention, the early warning result and the oil leakage position are uploaded to the monitoring center, and the processing suggestion is generated, so that intelligent and precise early warning of the oil leakage fault of the wind driven generator is realized. According to the multi-sensor fusion method, the accuracy and timeliness of oil leakage fault diagnosis are remarkably improved, the equipment loss and the maintenance cost are effectively reduced, and the method has important significance for guaranteeing safe and stable operation of the wind driven generator.

Description

Technical Field

[0001] The invention relates to the field of electric power systems, and in particular to an oil leakage fault early warning method for a wind turbine generator. Background Art

[0002] Wind turbines may leak oil during long-term operation, which may aggravate mechanical failures due to insufficient lubrication and poor heat dissipation, and increase maintenance costs. More seriously, if there are static sparks, it may cause greater risk of loss. The wind turbine oil leakage fault diagnosis system faces a complex technical challenge. Oil leakage can only be detected and processed after it occurs, which will cause a lag in equipment maintenance. Generally speaking, oil leakage occurs gradually during long-term use. During this process, the sealing will gradually decrease. Since water molecules are smaller than oil molecules, water vapor will first penetrate into the oil circuit, causing the water content of the oil to increase and cause pollution. When the oil does not leak over a large area, it is very necessary to detect the risk of oil leakage in a timely manner through oil parameters.

[0003] Therefore, it is necessary to conduct in-depth research on the early warning of oil leakage in wind turbines. Summary of the invention

[0004] In order to solve the problem of delayed monitoring and maintenance of oil leakage in wind turbines, the present invention provides a method for early warning of oil leakage fault in wind turbines, comprising the following steps:

[0005] A. Obtain the vibration signal, temperature data and oil parameters of the wind turbine;

[0006] B. Performing spectrum analysis on the vibration signal, and judging whether there is abnormal frequency component by comparing with a preset frequency threshold to analyze mechanical abnormality; analyzing and processing the temperature data, and judging whether the temperature shows an abnormal rising trend by comparing with a preset temperature rise slope threshold; comparing the oil parameter with a preset normal value range to judge whether there is an abnormal oil state;

[0007] C. extracting a feature vector representing the oil leakage fault according to the results of the spectrum analysis, temperature trend analysis and oil parameter comparison;

[0008] D. inputting the feature vector into a pre-trained support vector machine model, and obtaining a warning level representing the severity of the oil leakage fault through classification;

[0009] E. Based on the abnormal position and degree of abnormality of the vibration signal, temperature data, and oil parameters, a pre-built Bayesian network model is used for reasoning to obtain the posterior probability of oil leakage failure occurring at each position;

[0010] F. Upload the warning level and oil leak location to the monitoring center and display them through a human-machine interface, or use sound and light alarms to remind monitoring personnel.

[0011] Furthermore, the following steps are also included:

[0012] G. Use a big data platform to store and process the massive amount of monitoring data collected in parallel, and form an oil leakage fault knowledge base through data mining;

[0013] H. According to the characteristic patterns and evolution laws in the knowledge base, an online learning algorithm is used to optimize and update the support vector machine model and the Bayesian network model in real time.

[0014] Further, the vibration signal is subjected to spectrum analysis, including:

[0015] Performing denoising and normalization preprocessing on the vibration signal to obtain a preprocessed vibration signal;

[0016] Fast Fourier transform is used to perform spectrum analysis on the preprocessed vibration signal, and the frequency domain characteristics of the vibration signal are obtained by power spectrum density estimation;

[0017] According to a preset abnormal frequency component threshold, determining whether there is an abnormal frequency component exceeding the threshold in the frequency domain characteristics;

[0018] If there is an abnormal frequency component, a comprehensive judgment is made as to whether there is a mechanical fault at the vibration sensor part in combination with the time domain characteristics of the vibration signal; the time domain characteristics of the vibration signal include the peak-to-peak value of the vibration signal and the root mean square value of the vibration signal;

[0019] Obtain a predefined frequency range related to the oil leakage fault, and compare the abnormal frequency component with the frequency range;

[0020] If the abnormal frequency component falls within the frequency range associated with the oil leakage fault, it is determined that the mechanical fault is associated with the oil leakage fault.

[0021] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0022] The present invention discloses a method for early warning of oil leakage faults of wind turbines. The method arranges vibration, temperature and oil sensors at the monitoring position, collects multi-dimensional data for comprehensive analysis, uses spectrum analysis of vibration signals to judge mechanical faults, combines temperature trends and abnormal oil parameters, and uses support vector machine algorithms to classify and early warning of oil leakage faults. Furthermore, the Bayesian network algorithm is used to accurately infer the oil leakage location. The present invention uploads the early warning results and the oil leakage location to the monitoring center, and generates processing suggestions, thereby realizing intelligent and precise early warning of oil leakage faults of wind turbines. This multi-sensor fusion method significantly improves the accuracy and timeliness of oil leakage fault diagnosis, effectively reduces equipment losses and maintenance costs, and is of great significance to ensuring the safe and stable operation of wind turbines. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 The present invention is a flow chart of a wind turbine oil leakage fault early warning method.

[0024] Figure 2 It is a schematic diagram of a wind turbine oil leakage fault early warning method of the present invention. DETAILED DESCRIPTION

[0025] The following will describe the technical solutions in the embodiments of the present invention in detail in conjunction with the accompanying drawings in the embodiments of the present invention. The described embodiments are only part of the embodiments of the present invention.

[0026] like Figure 1-2 In this embodiment, a wind turbine oil leakage fault early warning method may specifically include:

[0027] A. Obtain the vibration signal, temperature data and oil parameters of the wind turbine;

[0028] B. Performing spectrum analysis on the vibration signal, and judging whether there is abnormal frequency component by comparing with a preset frequency threshold to analyze mechanical abnormality; analyzing and processing the temperature data, and judging whether the temperature shows an abnormal rising trend by comparing with a preset temperature rise slope threshold; comparing the oil parameter with a preset normal value range to judge whether there is an abnormal oil state;

[0029] C. extracting a feature vector representing the oil leakage fault according to the results of the spectrum analysis, temperature trend analysis and oil parameter comparison;

[0030] D. inputting the feature vector into a pre-trained support vector machine model, and obtaining a warning level representing the severity of the oil leakage fault through classification;

[0031] E. Based on the abnormal position and degree of abnormality of the vibration signal, temperature data, and oil parameters, a pre-built Bayesian network model is used for reasoning to obtain the posterior probability of oil leakage failure occurring at each position;

[0032] F. Upload the warning level and oil leak location to the monitoring center and display them through a human-machine interface, or use sound and light alarms to remind monitoring personnel.

[0033] Specifically, the structural diagram and oil circuit distribution diagram of the wind turbine are obtained, and the key monitoring parts of the oil circuit are determined according to the distribution of the key components and oil circuits of the equipment, and vibration sensors, temperature sensors and oil sensors are arranged at these parts. The vibration signal collected by the vibration sensor is obtained, and the spectrum analysis of the vibration signal is performed by fast Fourier transform. By setting the frequency threshold, it is determined whether there is an abnormal frequency component exceeding the threshold. If so, it is preliminarily determined that there may be a mechanical fault in this part. The characteristics of the abnormal frequency component are further analyzed, and combined with the abnormal conditions of temperature and oil parameters, it is determined whether the mechanical fault is related to the oil leakage fault. The temperature data collected by the temperature sensor is obtained, and the temperature data is processed by sliding average filtering. By setting the temperature rise slope threshold, it is determined whether the temperature shows an abnormal rising trend. If the temperature continues to rise abnormally, combined with the abnormal conditions of vibration and oil parameters, it is determined whether there is a temperature abnormality caused by oil leakage in this part. The parameters such as oil particle size, water content, oil grade, etc. collected by the oil sensor are obtained, and these parameters are compared with the preset normal value range. If the parameters exceed the normal range, it is determined that there is oil contamination in this part. Combined with the abnormal conditions of vibration and temperature, it is further determined whether the oil contamination is related to the oil leakage fault. According to the results of vibration spectrum analysis, temperature trend analysis and oil parameter comparison, the feature vector of the oil leakage fault is extracted, and the severity of the oil leakage fault is classified by using the pre-trained support vector machine model to obtain the oil leakage fault warning level, which is divided into three levels: general, severe and critical. According to the oil leakage fault warning level, combined with the abnormal position and degree of abnormality of vibration signals, temperature data and oil parameters, the pre-built Bayesian network model is used to infer the oil leakage location. The Bayesian network calculates the posterior probability of oil leakage at each location based on the causal relationship and conditional probability between each monitoring quantity, and selects the location with the highest probability as the most likely location of the oil leakage fault. The oil leakage fault warning level and oil leakage location are uploaded to the monitoring center through industrial Ethernet, and are intuitively displayed in the form of charts and text through the human-machine interface. At the same time, the sound and light alarm is triggered to remind the operation and maintenance personnel to go to the site for inspection and processing in time to avoid further expansion of the oil leakage fault. Hadoop and Spark and other big data platforms are used to store and parallel process the collected massive vibration signals, temperature data and oil parameters. Through data cleaning, feature extraction, correlation analysis and other steps, the characteristic patterns and evolution laws of wind turbine oil leakage faults are excavated to form an oil leakage fault knowledge base. According to the characteristic patterns and evolution laws in the oil leakage fault knowledge base, the online learning algorithm is used to optimize and update the support vector machine model and Bayesian network model in real time. When new monitoring data arrives, its features are extracted and matched with the patterns in the knowledge base. If the matching degree is low, the incremental learning process is triggered, and the new samples are used to fine-tune and update the model, continuously improving the accuracy and real-time performance of oil leakage fault diagnosis and early warning.The model performance is evaluated through methods such as cross-validation, and the evaluation results are quantified using indicators such as accuracy, recall, and F1 value.

[0034] Regarding the vibration signal parameters, the vibration signal collected by the vibration sensor of the wind turbine is obtained, and the vibration signal is denoised and normalized to obtain the preprocessed vibration signal. The preprocessed vibration signal is spectrally analyzed by fast Fourier transform, and the frequency domain characteristics of the vibration signal are obtained by power spectrum density estimation. According to the preset abnormal frequency component threshold, it is judged whether there is an abnormal frequency component exceeding the threshold in the frequency domain characteristics. If there is an abnormal frequency component, it is comprehensively judged whether there is a mechanical fault in the vibration sensor part in combination with the time domain characteristics of the vibration signal such as peak-to-peak value and root mean square value. The predefined frequency range related to the oil leakage fault is obtained, and the abnormal frequency component is compared with the frequency range. If the abnormal frequency component falls within the frequency range related to the oil leakage fault, it is judged that the mechanical fault is related to the oil leakage fault. The historical operation data of the wind turbine is collected to construct a labeled data set of the severity of the oil leakage fault. According to the correlation judgment result between the mechanical fault and the oil leakage fault, combined with the time domain characteristics of the vibration signal, the support vector machine algorithm is used to train the data set to obtain an oil leakage fault severity classification model, which is used to classify the severity of the oil leakage of the newly collected vibration signal. The temperature data of each component of the wind turbine is obtained through the temperature sensor, and the oil contamination data is obtained through the oil contamination sensor to determine whether there is temperature anomaly and oil contamination. Abnormal vibration, abnormal temperature and oil contamination are used as input factors, and the Bayesian network algorithm is used for reasoning to output the most likely location of oil leakage. The severity and location information of the oil leakage fault are uploaded to the wind turbine monitoring center through the network communication interface, triggering the corresponding early warning mechanism and notifying the operation and maintenance personnel to carry out maintenance.

[0035] Specifically, the vibration signal collected by the vibration sensor of the wind turbine is obtained, and the vibration signal is denoised by wavelet decomposition and soft threshold denoising methods. Then, the maximum and minimum normalization method is used to map the amplitude of the vibration signal to the interval [0,1]. The preprocessed vibration signal is subjected to fast Fourier transform, and the power spectrum density is estimated by the Welch method to extract the frequency domain characteristics of the vibration signal in the range of 0-5000Hz. According to the preset abnormal frequency component threshold 2g, it is judged whether there is an abnormal frequency component exceeding the threshold in the frequency domain characteristics. If there is an abnormal frequency component, combined with the time domain characteristics such as kurtosis and skewness of the vibration signal, the fuzzy comprehensive evaluation method is used to judge whether there is a mechanical fault in the vibration sensor part. The predefined frequency range of 500-2500Hz related to the oil leakage fault is obtained, and the abnormal frequency component is compared with the frequency range. If more than 80% of the abnormal frequency components fall within the frequency range related to the oil leakage fault, it is judged that the mechanical fault is related to the oil leakage fault. The historical operation data of wind turbines in the past year were obtained, and 500 groups of oil leakage fault severity were manually annotated to construct an annotated data set of oil leakage fault severity. According to the correlation judgment results between mechanical failure and oil leakage failure, combined with the time domain characteristics such as the root mean square value and waveform factor of the vibration signal, the support vector machine algorithm was used to train the data set, and the model parameters were determined through 5-fold cross validation to obtain the oil leakage fault severity classification model, with a classification accuracy of 95%.

[0036] Regarding temperature data, the temperature data of each part of the target equipment is collected through the temperature sensor to obtain the temperature data set. The temperature data set is preprocessed, and the moving average filtering algorithm is used to remove outliers and noise data to obtain the preprocessed temperature data set. According to the preprocessed temperature data set, the ARIMA time series prediction model is used to analyze the trend of the temperature of each part to obtain the temperature change trend of each part. According to the temperature change trend of each part, the threshold of abnormal temperature increase is set. If the temperature change trend exceeds the preset threshold, the temperature of the part is judged to be abnormally increased. Combined with the equipment structure diagram and component material information, the parts with abnormal temperature increase are further analyzed, and the temperature change trend, abnormal duration, part structure characteristics and other factors are comprehensively considered to determine whether the part has an oil leakage risk. For the parts judged to have an oil leakage risk, infrared thermal imaging technology is used to obtain a high-precision temperature distribution image of the part, and the image segmentation algorithm is used to extract the suspected oil leakage area. The temperature distribution data of the suspected oil leakage area is input into the pre-trained oil leakage detection convolutional neural network model to evaluate the location and severity of the oil leakage. Based on the output results of the oil leakage detection model, oil leakage alarm information is generated, the location coordinates and temperature data of the oil leakage site are marked, and the information is transmitted to the equipment maintenance management system to automatically generate equipment maintenance work orders to guide the subsequent maintenance plan formulation and maintenance task implementation.

[0037] Specifically, PT100 platinum resistance temperature sensors are installed at the gearbox, generator, transformer and other monitoring parts of the target equipment to collect real-time temperature data of each part. The data is collected continuously for 24 hours with a sampling interval of 1 minute to obtain a temperature data set containing 1440 data points. The temperature data set is preprocessed, and the spike noise in the temperature data is removed using a 5-point cubic smoothing filter algorithm. Then, by setting a threshold interval of ±3σ, the data points that exceed the threshold interval are regarded as outliers and removed to obtain the preprocessed temperature data set. According to the preprocessed temperature data set, the ARIMA (2,1,2) time series prediction model is used to predict the temperature change trend of each part in the next 6 hours. The model parameters are determined by the AIC criterion, and the mean absolute percentage error (MAPE) of the prediction results is controlled within 5%. For the temperature change trend of each part, the threshold of abnormal temperature increase is set to 2°C per hour. If the temperature change trend exceeds this threshold for 3 consecutive hours, it is judged that the temperature of the part has increased abnormally. Combined with the three-dimensional structure diagram of the equipment and the thermal conductivity information of the component materials, the finite element analysis method is used to perform thermal stress analysis on the abnormally elevated temperature parts. The temperature change trend, abnormal duration, and stress distribution of the part structure are comprehensively considered. When the stress exceeds 80% of the yield strength of the component material, the part is judged to have an oil leakage risk. For the parts judged to have an oil leakage risk, an infrared thermal imager is used to obtain a high-precision temperature distribution image of the part, and the image segmentation algorithm based on regional growth is used to extract the suspected oil leakage area. The 128*128 pixel temperature distribution data of the suspected oil leakage area is input into the pre-trained LeNet-5 oil leakage detection convolutional neural network model. The oil leakage location and severity are evaluated through feature extraction and classification recognition. The accuracy of oil leakage detection reaches more than 95%. Based on the output results of the oil leak detection model, oil leak alarm information in JSON format is generated, including the three-dimensional coordinates of the oil leak site, the oil leak area, temperature data, etc., which are transmitted to the equipment maintenance management system in real time through the WebSocket protocol, triggering the automatic reporting process of the oil leak fault, automatically generating equipment maintenance work orders, and reasonably arranging the maintenance task priority according to the severity of the fault and the equipment importance index, guiding the subsequent maintenance plan formulation and maintenance task implementation, and realizing the intelligent diagnosis and maintenance of oil leak faults.

[0038] Regarding oil parameters, obtaining oil parameter data of multiple locations, wherein the oil parameter data is collected by an oil sensor;

[0039] Compare the oil parameter data with a preset oil normal range threshold to determine whether the oil parameter of each part exceeds the normal range threshold;

[0040] If the oil parameter data exceeds the normal range threshold, combining historical data and equipment operation status, comprehensively analyzing whether the oil parameter data is abnormal data;

[0041] If the oil parameter data is abnormal data, a statistical method is used to calculate the mean and variance eigenvalues ​​of the abnormal data to construct an oil state eigenvector;

[0042] Inputting the oil state feature vector into a pre-trained support vector machine classification model to determine whether the oil state feature vector belongs to an oil contamination state;

[0043] If the oil state characteristic vector belongs to the oil contamination state, the corresponding part is marked as a suspected oil contamination part, and the specific value of the oil parameter data is recorded;

[0044] Obtain the spatial distribution of the suspected oil contamination sites and preliminarily determine the scope and extent of the oil contamination;

[0045] Inputting the oil parameter data of the suspected oil contaminated part into a pre-trained K-means clustering model to obtain a plurality of cluster centers;

[0046] Matching the cluster center with a typical oil leakage pattern in a pre-established oil leakage fault pattern library, calculating the similarity, and judging the nature of the oil leakage fault according to the matching result;

[0047] Combined with the equipment structure layout diagram and the oil flow path diagram, analyze the positional relationship and connectivity of the suspected oil contaminated parts to infer the specific location where the oil leakage occurs;

[0048] Combining the results of cluster analysis and structural analysis, the most suspicious location of the oil leak is determined;

[0049] If it is determined to be an oil leakage fault, an alarm signal will be generated and sent to the monitoring system, and the relevant maintenance personnel will be automatically notified via SMS and email, prompting them to go to the site for inspection and repair in time.

[0050] Specifically, the oil parameter data of multiple parts are collected through the oil sensor to obtain the numerical information reflecting the oil state. The collected oil parameter data are compared with the normal range threshold of the oil preset by the system to determine whether the oil parameter of each part exceeds the normal range. For the parts that exceed the threshold, the historical data and the equipment operation status are combined to comprehensively analyze whether it is abnormal data. The suspected abnormal oil parameter data are further analyzed, and the mean, variance and other characteristic values ​​of each parameter are calculated by statistical methods to construct the oil state characteristic vector. The support vector machine (SVM) algorithm is used to classify the characteristic vector to determine whether it belongs to the oil contamination state. According to the SVM classification results, the parts judged to be in the oil contamination state are marked as suspected oil contamination parts, and the specific values ​​of the oil parameters are recorded. The spatial distribution of the suspected oil contamination parts is analyzed to preliminarily determine the scope and degree of oil contamination. The K-means clustering algorithm is used to perform cluster analysis on the oil parameter data of the suspected oil contamination parts. The cluster center is matched with the typical oil leakage mode in the pre-established oil leakage fault mode library, the similarity is calculated, and the possibility of oil leakage fault is judged according to the matching results. Combined with the equipment structure layout diagram and the oil flow path diagram, the position relationship and connectivity of the suspected oil contaminated parts are analyzed to infer the specific parts where the oil leakage is most likely to occur. Considering the results of cluster analysis and structural analysis, the most suspicious location of the oil leakage fault is determined. The equipment part information of the suspected oil leakage fault is sent to the monitoring system in the form of an alarm signal to trigger the equipment maintenance warning. At the same time, the relevant maintenance personnel are automatically notified through SMS and email, prompting them to go to the site for inspection and maintenance in time to avoid the expansion of the oil leakage fault. In the system initialization stage, according to the technical parameters and operating experience of the equipment, the normal range thresholds of various oil parameters are pre-set as the basis for judgment. At the same time, by analyzing the equipment structure layout diagram and the oil flow path diagram, the spatial topological relationship between the oil contaminated part and the equipment structure is established to provide support for subsequent fault location.

[0051] In step C, the vibration signal of the device is obtained, and the spectrum analysis of the vibration signal is performed using the fast Fourier transform (FFT) to obtain the spectrum characteristics of the device. If the spectrum characteristics have abnormal frequencies, it is judged that the device may have a fault. The temperature data of the device is obtained, the temperature data is smoothed using the moving average method, and then the first-order difference of the temperature data is calculated to obtain the temperature change rate of the device. If the temperature change rate exceeds the preset threshold, it is judged that the device may have a fault. The oil parameters of the device are obtained, including the viscosity, acid value, moisture content, etc. of the oil, and the oil parameters are compared with the standard values ​​when the device is working normally. If the oil parameters are beyond the normal range, it is judged that the device may have a fault.

[0052] In step D, the abnormal frequency, temperature change rate and oil parameters are used as input features to construct a support vector machine classification model. The optimal model parameters are selected through grid search and cross-validation, and the model is trained using the labeled historical fault data to establish an oil leakage fault classifier. The trained support vector machine classifier is used to classify the current state of the equipment to obtain the oil leakage fault warning level of the equipment. If the warning level exceeds the preset threshold, the oil leakage fault alarm is triggered. For equipment with a lower warning level, regular inspection and maintenance are arranged; for equipment with a higher warning level, timely repair or replacement of parts is arranged. The monitoring data such as the vibration signal, temperature data and oil parameters of the equipment, as well as the oil leakage fault warning level, maintenance plan and other information are stored in a relational database, and a reasonable data table structure and index are established to facilitate subsequent data analysis and query. By regularly extracting the historical monitoring data and maintenance records of the equipment from the database, using data mining algorithms to analyze the failure mode and fault symptoms of the equipment, optimizing the feature selection and parameter setting of the support vector machine classification model, and improving the accuracy and reliability of the oil leakage fault warning. At the same time, the optimized model is applied to online equipment monitoring to achieve real-time warning and diagnosis of equipment failures.

[0053] In step E, according to the warning level, the PyMC3 tool is used to establish a Bayesian network model, with abnormal data as input nodes and oil leak locations as output nodes. By training the model, the conditional probability relationship between the input node and the output node is obtained. The current abnormal data is input into the trained model, and the most likely oil leak location is calculated using conditional probability reasoning. According to the reasoning results and the specific values ​​of the abnormal data, the severity of the oil leak fault is analyzed. If the severity of the fault exceeds the preset threshold, the alarm mechanism is triggered and the maintenance personnel are notified to handle it. Otherwise, continue to monitor the equipment status and perform fault analysis regularly. All relevant data and fault analysis results are recorded in the fault log for subsequent maintenance and diagnosis reference. Through this process, the oil leak fault warning model is continuously optimized to improve the accuracy of fault prediction.

[0054] Specifically, by installing vibration sensors, temperature sensors and oil parameter sensors on the equipment, the vibration signals, temperature data and oil viscosity, acid value, moisture content and other parameters of the equipment are collected in real time. The collected data is processed using statistical analysis methods to calculate the standard deviation and mean of the data. For example, the standard deviation of the vibration signal is normally 5m / s 2 , the current measured standard deviation is 1m / s 2, more than 2 times the normal value, it is judged that the vibration data is abnormal; the normal value of the mean temperature data is 65℃, and the current measured mean is 89℃, which is more than 3 times the normal value, so it is judged that the temperature data is abnormal; the normal value of the standard deviation of oil viscosity is 8mPa·s, and the current measured standard deviation is 3mPa·s, which is more than 8 times the normal value, so it is judged that the oil viscosity data is abnormal. According to the specific value of the abnormal data and the preset fault level threshold, the oil leakage fault warning level is divided into three levels: mild, moderate and severe. The PyMC3 tool is used to establish a Bayesian network model, with vibration anomalies, temperature anomalies, and oil viscosity anomalies as input nodes, and the oil leakage location as the output node. The model is trained through a large amount of historical fault data to obtain the conditional probability table between the input node and the output node. The vibration abnormality of the current equipment is 1m / s 2 , the temperature abnormal value is 89℃, and the oil viscosity abnormal value is 3mPa·s. These abnormal values ​​are input into the trained Bayesian network model. Through conditional probability reasoning calculation, it is found that the most likely oil leakage location is the oil tank connection, and the fault severity is moderate. Since the fault severity exceeds the preset mild threshold, the system automatically triggers the alarm mechanism, sends SMS and email notifications to the on-duty personnel, and pops up a warning prompt box on the monitoring interface. After receiving the notification, the on-duty personnel handle the fault according to the plan process and record the fault situation. The system automatically writes the abnormal data, fault location, fault level, processing results and other information of this fault into the fault log database for subsequent fault trend analysis and preventive maintenance decision-making reference. Through the continuous optimization and iteration of the oil leakage fault early warning process, the accuracy and reliability of the early warning model are continuously improved, providing a strong guarantee for the safe operation of the equipment.

[0055] Finally, after receiving the uploaded data, the monitoring center transmits it to the human-machine interface for real-time display and triggers the warning level judgment program;

[0056] The above embodiments are only used to illustrate the technical solution of the present invention but not to limit it. The present invention is described in detail with reference to the preferred embodiments. It should be understood by those skilled in the art that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention, and should be included in the scope of the claims of the present invention.

Claims

1. A wind turbine oil leakage fault early warning method, characterized in that: The following steps are involved: A. Obtain the vibration signal, temperature data and oil parameters of the wind turbine; B. Performing spectrum analysis on the vibration signal, and determining whether there is an abnormal frequency component by comparing it with a preset frequency threshold to analyze mechanical abnormality; Analyze and process the temperature data, and determine whether the temperature shows an abnormal rising trend by comparing it with a preset temperature rise slope threshold; compare the oil parameters with a preset normal value range to determine whether there is an abnormal oil state; C. extracting a feature vector representing the oil leakage fault according to the results of the spectrum analysis, temperature trend analysis and oil parameter comparison; D. inputting the feature vector into a pre-trained support vector machine model, and obtaining a warning level representing the severity of the oil leakage fault through classification; E. Based on the abnormal position and degree of abnormality of the vibration signal, temperature data, and oil parameters, a pre-built Bayesian network model is used for reasoning to obtain the posterior probability of oil leakage failure occurring at each position; F. Upload the warning level and oil leak location to the monitoring center and display them through a human-machine interface, or use sound and light alarms to remind monitoring personnel.

2. A wind turbine oil leakage fault early warning method according to claim 1, characterized in that: It also includes the following steps: G. Use a big data platform to store and process the massive amount of monitoring data collected in parallel, and form an oil leakage fault knowledge base through data mining; H. According to the characteristic patterns and evolution laws in the knowledge base, an online learning algorithm is used to optimize and update the support vector machine model and the Bayesian network model in real time.

3. The wind turbine oil leakage fault early warning method according to claim 1 is characterized in that: Performing spectrum analysis on the vibration signal, comprising: Performing denoising and normalization preprocessing on the vibration signal to obtain a preprocessed vibration signal; Fast Fourier transform is used to perform spectrum analysis on the preprocessed vibration signal, and the frequency domain characteristics of the vibration signal are obtained by power spectrum density estimation; According to a preset abnormal frequency component threshold, determining whether there is an abnormal frequency component exceeding the threshold in the frequency domain characteristics; If there is an abnormal frequency component, a comprehensive judgment is made as to whether there is a mechanical fault at the vibration sensor part in combination with the time domain characteristics of the vibration signal; the time domain characteristics of the vibration signal include the peak-to-peak value of the vibration signal and the root mean square value of the vibration signal; Obtain a predefined frequency range related to the oil leakage fault, and compare the abnormal frequency component with the frequency range; If the abnormal frequency component falls within the frequency range associated with the oil leakage fault, it is determined that the mechanical fault is associated with the oil leakage fault.

4. The wind turbine oil leakage fault early warning method according to claim 1 is characterized in that: Analyze and process temperature data, including: Preprocess the temperature data set, use the moving average filtering algorithm to remove outliers and noise data, and obtain the preprocessed temperature data set; According to the preprocessed temperature data set, the ARIMA time series prediction model is used to analyze the temperature trend of each part to obtain the temperature change trend of each part; According to the temperature change trend of each part, a threshold value of abnormal temperature increase is set. If the temperature change trend exceeds the preset threshold, it is judged that the temperature of the part has increased abnormally; Combined with the preset location information of the temperature collection points, further analysis is conducted on the parts where the temperature rises abnormally. Factors such as the temperature change trend, the duration of the abnormality, and the structural characteristics of the part are comprehensively considered to determine whether there is an oil leakage risk in this part.

5. The wind turbine oil leakage fault early warning method according to claim 1 is characterized in that: The analysis of abnormal oil condition includes the following methods: Compare the oil parameter data with a preset oil normal range threshold to determine whether the oil parameter of each part exceeds the normal range threshold; If the oil parameter data exceeds the normal range threshold, combining historical data and equipment operation status, comprehensively analyzing whether the oil parameter data is abnormal data; If the oil parameter data is abnormal data, then the mean and variance eigenvalues ​​of the abnormal data are calculated to construct an oil state eigenvector; The oil state feature vector is input into a pre-trained support vector machine classification model to determine whether the oil state feature vector is an abnormal state vector.

6. The method according to claim 1, characterized in that The classification in step D includes: The optimal model parameters are selected through grid search and cross-validation, and the model is trained using labeled historical fault data to establish an oil leakage fault classifier; The trained support vector machine classifier is used to classify the current status of the equipment and obtain the oil leakage fault warning level of the equipment.

7. The method according to claim 1, characterized in that In step E, it also includes: According to the warning level, the Bayesian network model was established using the PyMC3 tool, with the abnormal data as the input node and the oil spill location as the output node; By training the model, the conditional probability relationship between the input node and the output node is obtained; The current anomaly data is input into the trained model and conditional probability inference is used to calculate the most likely oil leak location.

Images