Analysis method for main bearing of wind generating set, medium and computer equipment

Abstract

The invention provides an analysis method for a main bearing of a wind generating set, a medium and computer equipment. The analysis method comprises the steps that real-time monitoring data of at least one of the wind speed, the generated power and the generator rotating speed of the wind generating set and at least one of the front bearing temperature and the rear bearing temperature of the wind generating set are obtained; obtaining original feature data from the real-time monitoring data and calculating reconstructed feature data based on the original feature data; inputting the original feature data and the reconstructed feature data into a pre-trained gradient lifting model to obtain a first diagnosis result; inputting the original feature data and the reconstructed feature data into a pre-trained regression model to obtain a first prediction result representing the service life of the main bearing of the wind generating set; and determining fault information and service life information of the main bearing of the wind generating set based on the first diagnosis result and the first prediction result.

Description

Technical Field

[0001] The present application relates to the field of wind power, and more specifically, to an analysis method, medium and computer equipment for a main bearing of a wind turbine generator set. Background Art

[0002] In wind turbines, the health of the main bearing, as an important component, directly affects the operating efficiency and safety of the entire system. The failure of the main bearing is divided into the following different stages: abnormal acoustic emission; abnormal acoustic emission, increased noise, signs of increased total vibration, abnormal grease detection; abnormal acoustic emission, obvious noise, a significant increase in total vibration, obvious fault frequency, abnormal grease detection, and increased temperature; abnormal acoustic emission, obvious noise, a significant increase in total vibration, obvious fault frequency, abnormal grease detection, and obvious abnormal temperature; frequent triggering of high temperature, speed comparison fault, wear of proximity switch, obvious abnormal noise, obvious vibration, etc.

[0003] Existing fault identification methods for fan main bearings mainly rely on the above-mentioned abnormal noise, abnormal vibration, excessive bearing temperature, grease leakage at the seal, initial signs of wear and grease state changes.

[0004] However, monitoring methods based on these characterizations are insufficient in terms of the accuracy of fault identification and life prediction. Summary of the invention

[0005] One of the objectives of the present disclosure is to provide an analysis method capable of improving the accuracy of fault identification and life prediction of a main bearing of a wind turbine generator set.

[0006] According to one aspect of the present disclosure, there is provided an analysis method for a main bearing of a wind turbine generator set, the analysis method comprising: acquiring real-time monitoring data of the wind turbine generator set including at least one of wind speed, power generation and generator speed and at least one of a front bearing temperature and a rear bearing temperature of the wind turbine generator set; acquiring original feature data from the real-time monitoring data and calculating reconstructed feature data based on the original feature data; inputting the original feature data and the reconstructed feature data into a pre-trained gradient boosting model to obtain a first diagnosis result characterizing whether there is a fault in the main bearing of the wind turbine generator set and the service life of the main bearing of the wind turbine generator set; inputting the original feature data and the reconstructed feature data into a pre-trained regression model to obtain a first prediction result characterizing the service life of the main bearing of the wind turbine generator set; and determining the fault information and service life information of the main bearing of the wind turbine generator set based on the first diagnosis result and the first prediction result.

[0007] Optionally, the step of determining the fault information and service life information of the main bearing of the wind turbine generator set based on the first diagnostic result and the first prediction result may include: stacking and fusing the prediction result in the first diagnostic result and the first prediction result to obtain a prediction fusion result, and determining the service life of the main bearing of the wind turbine generator set based on the prediction fusion result.

[0008] Optionally, the analysis method may also include: obtaining detection data of lubricating grease of the main bearing of the wind turbine generator set and on-site feedback data of the main bearing of the wind turbine generator set, and the step of determining the fault information and service life information of the main bearing of the wind turbine generator set based on the first diagnostic result and the first prediction result includes: determining the fault state and service life of the main bearing of the wind turbine generator set according to the detection data, on-site feedback data and prediction fusion results.

[0009] Optionally, the detection data may include metal content in lubricating grease, and the field feedback data may include at least one of axial clearance data and axial displacement data of a main bearing of the wind turbine generator set.

[0010] Optionally, the gradient boosting model may include an XGBoost model, and the regression model may include a Lasso regression model.

[0011] Optionally, the regression model can be trained and tested in the following manner: obtaining historical monitoring data of the wind turbine generator set including at least one of wind speed, generated power and generator speed, and at least one of the front bearing temperature and the rear bearing temperature of the wind turbine generator set; obtaining original feature variables from the historical monitoring data and calculating reconstructed feature variables based on the original feature variables; dividing the combination of the original feature variables and the reconstructed feature variables into a first training set and a first test set; training the regression model using the first training set, making predictions on the first test set, and evaluating the performance of the regression model based on the accuracy.

[0012] Optionally, the gradient boosting model can be trained and tested in the following manner: the gradient boosting model is trained using a first training set, tested on a first test set, and the performance of the gradient boosting model is evaluated based on accuracy.

[0013] Optionally, during the training process of the regression model and the gradient boosting model, the gradient boosting model determines the weight of each of the original feature variables and the reconstructed feature variables; during the training process of the regression model and the gradient boosting model, the original feature variables and the reconstructed feature variables are screened according to the weights, and the regression model and the gradient boosting model are retrained using the screened feature variables.

[0014] Optionally, the analysis method may also include: using the Shapley value method to evaluate the contribution of each of the original feature data and the reconstructed feature data to the first diagnostic result of the gradient boosting model and the first prediction result of the regression model, and screening out at least one of the original feature data and the reconstructed feature data based on the contribution, and during the training process of the gradient boosting model and the regression model, re-training the gradient boosting model and the regression model using the feature variables corresponding to the screened feature data.

[0015] According to a second aspect of the present disclosure, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a program or an instruction, and when the program or the instruction is executed by a processor, the processor is prompted to execute the above-mentioned analysis method.

[0016] According to a third aspect of the present disclosure, a computer device is provided, the computer device comprising a memory and a processor, the memory storing a program or instructions, and when the program or instructions are executed by the processor, the processor is prompted to execute the above analysis method.

[0017] The analysis method according to the embodiment of the present disclosure can narrow the scope of boarding inspection, improve the efficiency of inspection, and reduce the cost of inspection. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is a flowchart showing an analysis method according to the first embodiment of the present disclosure.

[0019] Figure 2 It is a graph showing the change of the average bearing temperature with the ambient temperature, generator speed, power, and grid connection duration.

[0020] Figure 3 A graph showing ambient temperature, generator speed, power, front bearing temperature 1, front bearing temperature 2, and front bearing temperature difference.

[0021] Figure 4 is a flow chart showing the training method and testing method of the gradient boosting model of the present disclosure.

[0022] Figure 5 is a ranking graph showing the importance of some features determined by the XGBoost model.

[0023] Figure 6 is a flow chart showing the training method and the testing method of the regression model of the present disclosure.

[0024] Figure 7 It is a graph showing the contribution of some features analyzed by the Shapley value method. DETAILED DESCRIPTION

[0025] The following detailed description is provided to help gain a comprehensive understanding of the methods, devices and / or systems described herein. However, the order of operations described herein is only an example and is not limited to those orders set forth herein, but may be equivalently replaced or changed except for operations that must occur or be performed in a specific order. In addition, for greater clarity and simplicity, the description of content known in the art will be omitted or simplified.

[0026] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as those generally understood by a person of ordinary skill in the art to which the present disclosure belongs after understanding the present disclosure. Unless explicitly defined as such herein, terms (such as those defined in a general dictionary) should be interpreted as having a meaning consistent with their meaning in the context of the relevant art and the present disclosure, and should not be interpreted in an idealized or overly formal manner.

[0027] Unless otherwise specified, the same reference numerals generally refer to the same elements (e.g., components, steps, and methods). Reference numerals described in the previous embodiments may be omitted if they appear again in the subsequent embodiments. In addition, the technical features described in different or the same embodiments may be combined in any manner, as long as the combined embodiments or technical solutions are complete and can solve the technical problems of the present application or achieve the technical effects described or not described in the present disclosure but can be determined based on the above complete technical solutions.

[0028] The present invention realizes fault identification and life prediction of the main bearing of a wind turbine generator set by fusing multi-source data through a machine learning model based on supervisory control and data acquisition (SCADA) data, and can improve the accuracy of fault identification and life prediction of the main bearing of a wind turbine generator set.

[0029] Figure 1 is a flowchart showing an analysis method according to a first embodiment of the present disclosure, Figure 2 It is a graph showing the change of average bearing temperature with ambient temperature, generator speed, power, and grid connection duration. Figure 3 A graph showing ambient temperature, generator speed, power, front bearing temperature 1, front bearing temperature 2, and front bearing temperature difference.

[0030] The analysis method of the first embodiment of the present disclosure may be used as an analysis method for a main bearing of a wind turbine generator set, and may include steps S110 , S120 , S130 , S140 , and S150 .

[0031] Reference Figure 1In step S110, real-time monitoring data of the wind turbine generator set including at least one of wind speed, power generation and generator speed and at least one of front bearing temperature and rear bearing temperature of the wind turbine generator set is obtained.

[0032] The real-time monitoring data here can be the data of the day of the analysis time, but this is only an example, and the real-time monitoring data can also be the data within a few hours before the analysis time. In contrast, the historical monitoring data can be the data of multiple months (e.g., six months) before the month of the analysis time.

[0033] As an example, various data about a wind turbine generator system may be obtained through a SCADA system.

[0034] The various data collected and recorded by the SCADA system may include temperature data of the main bearing, generator, gearbox, etc. Temperature is one of the important indicators reflecting the health of the equipment, and abnormal temperature changes often indicate potential problems. For example, an abnormal increase in the temperature of the main bearing may indicate insufficient lubrication or increased wear.

[0035] The various data collected and recorded by the SCADA system may also include wind speed, power generation of wind turbines, rotation speed (e.g., generator speed), etc. The data collected by the SCADA system can provide important reference for the operation and maintenance of wind turbines.

[0036] As an example, real-time monitoring data of the wind turbine generator set including at least one of wind speed, power generation and generator speed and at least one of front bearing temperature and rear bearing temperature of the wind turbine generator set may be directly acquired from the SCADA system.

[0037] Specifically, the types of data acquired from SCADA may be as shown in Table 1 below.

[0038] Table 1 SCADA data measurement points

[0039] Measuring point name Chinese rectime time wtid Fan No. TurSt Unit operating status WTUR_Temp Ambient temperature WNAC_Temp Cabin temperature Wlq Wind speed PwrAt power WGENSpd Generator speed overspblade1 Generator speed 1 BackAxis1 Rear bearing temperature 1 BackAxis2 Rear bearing temperature 2 FrontAxis1 Front bearing temperature 1 FrontAxis2 Front bearing temperature 2

[0040] In Table 1, rectime: time, the timestamp of recording data, used to identify the specific time of data collection; wtid: wind turbine number, which identifies the unique number of a wind turbine generator set, which is convenient for distinguishing different wind turbine generator sets; TurSt: unit operating status, indicating the current working status of the wind turbine generator set, such as power generation, standby, maintenance, etc.; WTUR_Temp: ambient temperature, refers to the temperature of the environment around the wind turbine generator set, which affects the operating efficiency and safety of the unit; WNAC_Temp: cabin temperature, refers to the temperature inside the cabin of the wind turbine generator set; WSpd: wind speed, the wind speed at the location of the wind turbine generator set, which is an important factor in determining whether the wind turbine generator set can generate electricity and the amount of electricity generated; PwrAt: power, the power generated by the wind turbine generator set, which reflects the working efficiency of the wind turbine generator set; WGENSpd: generator speed, refers to the speed of the generator; overspblade1: generator speed 1, refers to the speed measurement value related to the blade, or the speed measurement of the generator under specific conditions; BackAxis1: rear bearing temperature 1, refers to the temperature of the bearing located at the rear of the wind turbine generator set, which is used to monitor the health of the bearing. BackAxis2: Back bearing temperature 2, refers to the temperature of another position or another group of back bearings. FrontAxis1: Front bearing temperature 1, refers to the temperature of the bearing located at the front of the wind turbine, and is also used to monitor the health of the bearing; FrontAxis2: Front bearing temperature 2, may refer to the temperature of another position or another group of front bearings. Both the front bearing and the rear bearing can be part of the main bearing of the wind turbine, and the main bearing can be simply referred to as the bearing.

[0041] The acquired real-time monitoring data and the historical monitoring data mentioned below may include at least one of the data in Table 1 above.

[0042] In step S120, original feature data is acquired from the real-time monitoring data and reconstructed feature data is calculated based on the original feature data. As an example, the original feature data here may include wind speed data, power generation data, rotation speed data, and the like.

[0043] Reference Figure 3 , Figure 3 It shows that the ambient temperature is around 40°C, the rotation speed is in the range of 2RPM to 10RPM, the power generation power is in the range of 0 to 5000kw, the front bearing temperature 1 is in the range of 30°C to 60°C, the front bearing temperature 2 is in the range of 30°C to 60°C, and the front bearing temperature difference is in the range of -10°C to 10°C.

[0044] The reconstructed characteristic data can be data obtained based on the original characteristic data, such as data obtained by preprocessing. Preprocessing may include: screening data that conforms to business logic (for example, wind speed, wind turbine status, power, etc.), non-unit maintenance process data, data that can more clearly identify abnormal characteristic conditions (pitch, yaw, full power, continuous grid connection, etc.), data transformation, etc.

[0045] In addition, reconstructed feature data can be obtained through feature engineering. Related algorithms for feature engineering may include: feature selection (selecting the most influential features from the original data set and removing irrelevant or redundant features), for example, based on statistical tests (such as chi-square test, t-test), model-based feature importance (such as decision tree, random forest), recursive feature elimination (RFE), etc.; feature construction (creating new features based on existing features), for example, bearing temperature difference, temperature rise, continuous sequence recognition, integral calculation, etc. can be obtained through feature construction. In addition, feature engineering may also include feature scaling, missing value processing, etc.

[0046] As an example, the feature data can be reconstructed by using the trapezoidal rule, numerical integration to find the area, etc. The trapezoidal rule uses a trapezoid to estimate the area under the curve, which is equivalent to approximating the integrand as a straight line function. The integrand is approximated as a trapezoid. To obtain a more accurate value, the interval for which the product is required can be divided into multiple small intervals.

[0047] Numerical integration to obtain area means: in the calculation coordinate system, the area from the curve to the coordinate axis. This method can be used to calculate the coverage area of ​​relevant data curves such as bearing temperature, temperature rise, temperature difference, etc., so as to measure the accumulated heat of the bearing under different working conditions, and perform time series longitudinal comparison and horizontal comparison across units. However, the present disclosure is not limited to this, and various reconstructed feature data can also be obtained through other methods.

[0048] For example, the difference between the front and rear bearing temperatures can be reconstructed using the two original feature data BackAxis1 and FrontAxis1 (reconstructed feature data); the change in the rear bearing temperature can be reconstructed using BackAxis1(t)-BackAxis1(t-1); BackAxis1 can be reconstructed by numerically integrating the area to obtain the cumulative change in the rear bearing temperature over a period of time, and so on.

[0049] The reconstructed characteristic data may include at least one of basic index data, bearing temperature and temperature rise index data, bearing temperature continuous increase subsequence index data and each parameter integral data.

[0050] Basic indicator data may include the following time-related characteristics: grid-connection start time, which refers to the time when the wind turbine is connected to the grid; grid-connection end time, which refers to the time when the wind turbine stops supplying power to the grid; grid-connection duration, which refers to the length of time the wind turbine continues to work during the grid-connection period; average ambient temperature, which reflects the climatic conditions of the area where the wind turbine is located; maximum generator speed, which is the maximum rotation speed reached by the generator; average generator speed, which is the average rotation speed of the generator during the entire grid-connection period; maximum unit power, which is the maximum power output generated by the wind turbine during the grid-connection period; average unit power, which is the average power output of the wind turbine during the grid-connection period.

[0051] The bearing temperature and temperature rise index data may include the following temperature-related characteristics: maximum bearing temperature, the highest temperature reached by the bearing during the grid-connected period; minimum bearing temperature, the lowest temperature reached by the bearing during the grid-connected period; average bearing temperature, the average temperature of the bearing during the grid-connected period; maximum bearing temperature rise, the maximum increase in bearing temperature relative to the initial temperature; minimum bearing temperature rise, the minimum increase in bearing temperature relative to the initial temperature; average bearing temperature rise, the average increase in bearing temperature relative to the initial temperature.

[0052] The bearing temperature continuously increasing subsequence indicator data may include the following continuously increasing related features: the duration of the bearing temperature continuously increasing subsequence, the initial value of the bearing temperature continuously increasing, the maximum value of the bearing temperature continuously increasing and the absolute increment of the bearing temperature continuously increasing.

[0053] The integrals of each parameter may include: wind speed integral, wind speed corresponding temperature integral, ambient temperature integral, power integral, front bearing temperature 1 integral, front bearing temperature 2 integral, rear bearing temperature 1 integral, rear bearing temperature 2 integral, front bearing temperature 1 temperature rise integral, front bearing temperature 2 temperature rise integral, etc. The integrals of each parameter can be calculated by obtaining the area through the above numerical integration.

[0054] Through the above-mentioned reconstructed characteristic data, when evaluating the relevant temperature of the main bearing, the combined effects of factors such as ambient temperature, power generation power, and operating time can be comprehensively considered, thereby improving the accuracy of fault identification and life prediction.

[0055] Bearing related influencing factors can refer to Figure 2 The average bearing temperature changes with the ambient temperature, generator speed, power generation, and duration of grid connection. As the ambient temperature, generator speed, power generation, and duration of grid connection increase, the average bearing temperature gradually increases. Under stable and normal operation, the average bearing temperature will remain stable.

[0056] Reference Figure 1In step S130, the original feature data and the reconstructed feature data are input into a pre-trained gradient boosting model to obtain a first diagnostic result characterizing whether there is a fault in the main bearing of the wind turbine generator set and the service life of the main bearing of the wind turbine generator set.

[0057] As described above, the original feature data and the reconstructed feature data can be feature data that fuses various types of data, and the original feature data and the reconstructed feature data can constitute multidimensional feature data. The dimensions of the original feature data and the reconstructed feature data can be at least thirty-five, so that the accuracy of the fault diagnosis and life prediction of the model can be increased. However, the higher the dimension, the larger the amount of data, which will increase the length of the model calculation. As a compromise, the dimension of the original feature data can be greater than or equal to 5, the dimension of the reconstructed feature data can be greater than or equal to 10, and the dimension of the feature data input to the gradient boosting model and the regression model can be between 10 and 20.

[0058] The gradient boosting model may be any of various models widely used in machine learning and data mining. The gradient boosting model may include LightGBM, CatBoost, XGBoost, etc. However, these are only examples, and other classification and regression algorithms may also be used.

[0059] As an example, XGBoost is implemented through an efficient gradient boosting tree. The core idea is to build a new weak learner (usually a decision tree) in each round of iteration by minimizing the residual sum of squares, and add the new learner to the current model to improve the predictive ability of the overall model.

[0060] When it is necessary to analyze the main bearing of a wind turbine generator set, the original feature data and the reconstructed feature data as described above can be input into the pre-trained XGBoost model to obtain a first diagnostic result characterizing whether there is a fault in the main bearing of the wind turbine generator set and the service life of the main bearing of the wind turbine generator set.

[0061] As an example, at least one of the original feature data and at least one of the reconstructed feature data may be input into a pre-trained XGBoost model to obtain a first diagnostic result. At least one of the original features, at least one of the time-related features, at least one of the temperature-related features, at least one of the continuously increasing related features, and at least one of the integrals of each parameter may be input into a pre-trained XGBoost model to obtain a first diagnostic result characterizing whether there is a fault in the main bearing of the wind turbine generator set and the service life of the main bearing of the wind turbine generator set.

[0062] As an example, the result characterizing the presence of a fault in the main bearing of a wind turbine generator set can be represented as 1, the result characterizing the absence of a fault in the main bearing of a wind turbine generator set can be represented as 0, and the first diagnostic result characterizing the service life of the main bearing of a wind turbine generator set can be represented as a specific number of months, years, etc. For example, the XGBoost model can output two results, 1 and 8, respectively, where 1 indicates that the main bearing of the wind turbine generator set has a fault and its service life is 8 months. However, this is only an example, and the result characterizing whether the main bearing of a wind turbine generator set has a fault can be represented as a probability between 0 and 1. For example, when the output result is 0.7, it indicates that the probability of a fault in the main bearing of a wind turbine generator set is 0.7.

[0063] Reference Figure 1 In step S140, the original feature data and the reconstructed feature data are input into a pre-trained regression model to obtain a first prediction result characterizing the service life of a main bearing of a wind turbine generator set.

[0064] The regression model of the present disclosure may include a Lasso regression model, a ridge regression model, a support vector machine, a random forest, and the like.

[0065] The Lasso regression model is a linear regression model that improves the model by adding an L1 regularization term to the ordinary least squares method. The Lasso regression model prevents overfitting by reducing the complexity of the model, and it can compress some coefficients to an absolute value of zero, thereby achieving variable selection.

[0066] At least one of the original feature data and at least one of the reconstructed feature data may be input into a pre-trained Lasso regression model to obtain a first prediction result. At least one of the original features, at least one of the time-related features, at least one of the temperature-related features, at least one of the continuously increasing related features, and at least one of the integrals of each parameter (or integral features) may be input into a pre-trained Lasso regression model to obtain a first prediction result characterizing the service life of the main bearing of the wind turbine generator set.

[0067] As an example, the first prediction result characterizing the service life of the main bearing of the wind turbine generator set may be expressed as a specific number of months, years, etc. For example, the Lasso regression model may output a result, such as 9, indicating that the service life of the main bearing of the wind turbine generator set is 9 months.

[0068] Reference Figure 1 In step S150, the fault information and service life information of the main bearing of the wind turbine generator set are determined based on the first diagnosis result and the first prediction result. For example, according to the first diagnosis result and the first prediction result, it can be determined that the main bearing of the wind turbine generator set has a fault and its service life is between 8 months and 9 months.

[0069] As an example, the step of determining the fault state and service life of the main bearing of the wind turbine generator set based on the first diagnosis result and the first prediction result may include: stacking and fusing the prediction result in the first diagnosis result and the first prediction result to obtain a prediction fusion result, and determining the service life of the main bearing of the wind turbine generator set according to the prediction fusion result. The stacking fusion here refers to the result obtained by fusing the prediction result based on the gradient boosting model and the first prediction result of the regression model.

[0070] For example, the prediction fusion result can be obtained by stacking and fusion of averaging, weighted averaging, etc. As an example, the prediction results output by the XGBoost model and the Lasso regression model can be averaged, weighted averaged, etc. to obtain the prediction fusion result.

[0071] For example, the prediction result output by XGBoost is 8, and the prediction result output by Lasso regression model is 9. The average prediction result is 8.5. Furthermore, the weight of XGBoost model is 0.6, and the weight of Lasso regression model is 0.4. The average prediction result obtained by weighted average is 8.4. However, this is just an example, and stacking and fusion can also be performed in other ways.

[0072] It should be noted that the prediction fusion result can also be used to determine whether there is a fault in the main bearing of the wind turbine generator set. For example, when the difference between the prediction fusion result and the life value at the current moment in the preset life curve exceeds the preset threshold, it is determined that the main bearing of the wind turbine generator set has a fault. For example, the design life value of the main bearing of the wind turbine generator set at the current moment is 30 months, while the prediction fusion result is 8.5 months, and the difference between the two is 21.5 months (far exceeding the preset threshold of 6).

[0073] The analysis method disclosed in the present invention can also perform comprehensive analysis and judgment based on detection data and on-site feedback data, thereby further improving the accuracy of fault identification and life prediction.

[0074] The analysis method also includes: obtaining detection data of lubricating grease of the main bearing of the wind turbine generator set and on-site feedback data of the main bearing of the wind turbine generator set, and the step of determining the fault information and service life information of the main bearing of the wind turbine generator set based on the first diagnostic result and the first prediction result includes: determining the fault state and service life of the main bearing of the wind turbine generator set according to the detection data, on-site feedback data and prediction fusion results.

[0075] That is to say, the fault status and service life of the main bearing of the wind turbine generator set are determined on the basis of the prediction fusion results and further combined with the detection data of the lubricating grease of the main bearing of the wind turbine generator set and the on-site feedback data of the main bearing of the wind turbine generator set.

[0076] As an example, the detection data includes the metal content in lubricating grease, and the field feedback data includes at least one of the axial clearance data and the axial displacement data of the main bearing of the wind turbine generator set.

[0077] For example, when testing lubricating grease, the iron content (mg / kg) in the grease can be tested first. When the iron content in the lubricating grease is between 1500 and 2000 (mg / kg), it needs to be paid attention to. When the iron content exceeds 2000 mg / kg, it means that the situation is serious, and when it exceeds 5000 mg / kg, it needs to be dealt with as soon as possible. As an example, the iron content in the middle and late stages of bearing abnormality can generally reach tens of thousands of mg / kg. As an example, the development of other metal content and water content and other elements can also be paid attention to, so as to comprehensively judge whether the main bearing is faulty.

[0078] In addition to the metal content in lubricating grease, the detection data may also include various vibration monitoring data, noise monitoring data, etc. In other words, a comprehensive judgment can be made based on the model prediction and in combination with various vibration monitoring data, noise monitoring data, etc.

[0079] The axial clearance as the on-site feedback data refers to the distance between the inner ring of the main bearing and the journal. It is an indicator to measure the assembly accuracy and operating status of the main bearing. Reasonable axial clearance can ensure the normal operation of the main bearing. Excessive axial clearance may lead to increased vibration, increased noise and poor lubrication, which will affect the service life and reliability of the equipment. Too small axial clearance may lead to increased friction and even cause jamming or burning.

[0080] The axial displacement as field feedback data can refer to the movement of the main bearing in the axial direction, including radial displacement and axial displacement. The axial displacement reflects the change in the position of the main bearing in the axial direction, which may be caused by factors such as installation error, thermal expansion, load change, etc. Large axial displacement may cause the main bearing to bear additional stress, accelerate wear, and even cause bearing failure. Therefore, monitoring the axial displacement of the main bearing can help determine whether there are potential problems, such as loose bearings, bent shafts, insufficient axial preload, etc.

[0081] In addition to the axial clearance data and axial displacement data of the main bearing, the on-site feedback data can also include auditory data, visual data, etc. For example, the abnormal noise heard by the maintenance personnel, the picture of whether there is oil leakage, etc.

[0082] On the basis of the prediction fusion result, the fault state and service life of the main bearing of the wind turbine generator set can be determined by further combining the detection data of the lubricating grease of the main bearing of the wind turbine generator set and the on-site feedback data of the main bearing of the wind turbine generator set. For example, when the detection data and the on-site feedback data both indicate that the main bearing of the wind turbine generator set is faulty, it is determined that the main bearing of the wind turbine generator set is faulty.

[0083] Figure 4 is a flowchart showing the training method and testing method of the gradient boosting model of the present disclosure, Figure 5 is a ranking diagram showing the importance of some features determined by the XGBoost model, Figure 6 is a flowchart showing a training method and a testing method of a regression model disclosed herein, Figure 7 It is a graph showing the contribution of some features analyzed by the Shapley value method.

[0084] Reference Figure 4 , the training and testing of the gradient boosting model may include step S210, step S220, step S230 and step S240.

[0085] Similarly, refer to Figure 4 , the training and testing of the gradient boosting model may include step S210, step S220, step S230 and step S240.

[0086] In step S210, historical monitoring data of the wind generator set including at least one of wind speed, power generation and generator rotation speed and at least one of front bearing temperature and rear bearing temperature of the wind generator set is acquired.

[0087] The data type, acquisition method, preprocessing method, etc. of the historical monitoring data here can be the same as the above-mentioned real-time monitoring data. As mentioned above, the real-time monitoring data can be the data of the day of the analysis time, or the data within a few hours before the analysis time, etc., while the historical monitoring data can be the data of multiple months (for example, six months) before the analysis month. In other words, there are significant differences between the historical monitoring data and the real-time monitoring data in terms of time span and time interval from the current analysis time.

[0088] The historical monitoring data may also include at least one of the various types of data in Table 1 above, and may be preprocessed by the preprocessing manner as shown above.

[0089] In step S220 , original feature variables are acquired from historical monitoring data and reconstructed feature variables are calculated based on the original feature variables.

[0090] The original feature variables and the reconstructed feature variables here may correspond to the original feature data and the reconstructed feature data respectively, and the reconstructed feature variables may also be obtained on the basis of the original feature variables by feature engineering. The original feature variables may be the variables listed in Table 1 above, and the reconstructed feature variables may include at least one of the time-related features, at least one of the temperature-related features, at least one of the continuously increasing related features, and at least one of the integrals of the parameters as described above.

[0091] In step S230, the combination of the original feature variables and the reconstructed feature variables is divided into a first training set and a first test set, which can be divided in a ratio of 80% for the training set and 20% for the test set.

[0092] In step S240, the gradient boosting model is trained using the first training set, tested on the first test set, and the performance of the gradient boosting model is evaluated based on the accuracy.

[0093] Taking XGBoost as an example, when training the XGBoost model, you can set the learner parameters (hyperparameters), among which the more important parameters (hyperparameters) include: nrounds (number of iterations), eta (learning rate), max_depth (maximum tree depth), subsample (sample ratio for each iteration), colsample_bytree (feature ratio for each split), gamma (minimum loss reduction required for node splitting), lambda (L2 regularization weight), alpha (L1 regularization weight), booster (boosting method), objective (learning goal). Then, through grid search, hyperparameter optimization and test verification are performed, and finally appropriate parameters are set to achieve the optimal level of model accuracy and recall rate.

[0094] The gradient boosting model may determine the weight of each of the original feature variables and the reconstructed feature variables, that is, the importance of each feature variable.

[0095] Reference Figure 5 , the horizontal axis represents the weight (i.e., importance) of the characteristic variable, and the vertical axis represents the name of the characteristic variable. The weight of the characteristic variable gradually decreases in the following order: front bearing average temperature 2, front bearing average temperature 1, front bearing temperature 1, rear bearing average temperature 2, rear bearing temperature 1, cabin temperature, rear bearing temperature 2, front bearing temperature rise 2, front bearing temperature rise average 2, generator speed, front bearing temperature rise average 1, front bearing temperature rise 1, rear bearing temperature average 1, rear bearing temperature rise average 1, rear bearing temperature rise average 2, front bearing temperature 2, rear bearing temperature rise 1, rear bearing temperature rise average 2, front bearing temperature 2, rear bearing temperature rise 1, rear bearing temperature rise 2.

[0096] During the training process of the gradient boosting model, the original feature variables and the reconstructed feature variables can be screened according to the above weights, and the gradient boosting model can be retrained using the screened feature variables.

[0097] As an example, the features whose weights are less than a predetermined threshold (e.g., 0.02) in the above feature variables can be removed, and then the feature variables that are greater than or equal to 0.02 can be screened out, that is, front bearing average temperature 2, front bearing average temperature 1, front bearing temperature 1, rear bearing average temperature 2, rear bearing temperature 1, cabin temperature, rear bearing temperature 2, front bearing temperature rise 2, and then retraining is performed using the screened feature variables, thereby reducing the amount of calculation and increasing the training speed. The dimension of the data input to the model is reduced during retraining.

[0098] Similarly, refer to Figure 6 , the training and testing of the regression model may include step S310, step S320, step S330 and step S340.

[0099] In step S310, historical monitoring data of the wind generator set including at least one of wind speed, power generation and generator speed and at least one of front bearing temperature and rear bearing temperature of the wind generator set is acquired.

[0100] In step S320 , original feature variables are acquired from historical monitoring data and reconstructed feature variables are calculated based on the original feature variables.

[0101] In step S330, the combination of the original feature variables and the reconstructed feature variables is divided into a first training set and a first test set, which can be divided into 80% for the training set and 20% for the test set, or 90% for the training set and 10% for the test set.

[0102] In step S340, the regression model is trained using the first training set, tested on the first test set, and the performance of the regression model is evaluated based on the accuracy.

[0103] Taking the Lasso regression model as an example, when training the Lasso regression model (lifetime prediction), cross-validation can be used to select the best lambda value (λ value), refit the model based on the obtained optimal lambda value, and make predictions on the first test set, and calculate the mean square error (MSE) to evaluate the model performance. Through verification, the optimal lambda value improves the accuracy of the model by at least 20%, and the model accuracy of the final test data is above 90%.

[0104] During the training process of the regression model, the original feature variables and the reconstructed feature variables may also be screened according to the above weights, and the regression model may be retrained using the screened feature variables.

[0105] The analysis method disclosed herein may further include: using the Shapley value method to evaluate the contribution of each of the original feature data and the reconstructed feature data to the first diagnostic result of the gradient boosting model and the first prediction result of the regression model, and screening out at least one of the original feature data and the reconstructed feature data based on the contribution, and during the training process of the gradient boosting model and the regression model, retraining the gradient boosting model and the regression model using the feature variables corresponding to the screened feature data.

[0106] The Shapley value method is mainly used to solve the problem of interest distribution among all parties in cooperative games. The most important thing is that it reflects the contribution of all parties to the overall goal of cooperation and avoids egalitarianism in distribution. It is more reasonable and fair than the distribution method based on resource input value or resource allocation efficiency or a combination of the two.

[0107] The Shapley value can be calculated by the following formula (1):

[0108]

[0109] ·φ i (v): Shapley value of the ith participant

[0110] N: Number of participants

[0111] · A set of participants with a specific order R

[0112] · Contributions of a set of actors with a specific order R

[0113] · The i-th participant's joint contribution after participating in a set of participants of a specific order R

[0114] When calculating the contribution of each of the original feature data and the reconstructed feature data to the first diagnostic result of the gradient boosting model and the first prediction result of the regression model, existing Shapley value analysis tools (eg, Matlab and Pathon libraries) may be used for calculation.

[0115] The Shapley value or SHAP value (i.e., contribution) of some feature data can be referred to Figure 7 From the attached Figure 7It can be seen from the changes in the SHAP values ​​of front bearing temperature 2, front bearing temperature rise 1, front bearing temperature 1, rear bearing temperature 1, rear bearing temperature average 2, cabin temperature, rear bearing temperature 2, and front bearing temperature rise 1. Specifically, the contribution of each feature data to the accuracy can be determined according to the maximum value of the SHAP value of each feature data. The contribution of front bearing temperature 2, front bearing temperature rise 1, front bearing temperature 1, rear bearing temperature 1, rear bearing temperature average 2, cabin temperature, rear bearing temperature 2, and front bearing temperature rise 1 can be: 15, 11, 4, 2, 7, 7, 4, 1, respectively.

[0116] Can be based on Figure 7 The maximum value of SHAP of each feature data in the model is used to determine the contribution of each feature data, so as to obtain the contribution ranking of each feature data. The top feature data can be selected according to the ranking result. During the training process, the top feature data or feature variables can be selected for model training and testing. The Shapley value analysis method can be used to explain the final results and can also be used to optimize the training of the model. By combining the Shapley value analysis method with the XGBoost model, important feature data or feature variables can be screened to improve the accuracy of fault identification and life prediction. Shapley value analysis can not only be used in the actual prediction process, but also can be used to analyze the test results and guide the training of the model.

[0117] The results predicted by the model can be combined with the grease detection results and on-site feedback results for comprehensive judgment, thereby obtaining the comprehensive fault judgment results and life prediction results of the unit.

[0118] Based on the above model disclosed in the present invention, more than 500 normal units and more than 10 abnormal units were finally tested, and the final accuracy of the model exceeded 80%, which can provide effective support for fault diagnosis and life prediction.

[0119] At least one of the multiple modules performing the above steps may be implemented by an AI model. Functions associated with AI may be performed by a non-volatile memory, a volatile memory, and a processor.

[0120] The processor may include one or more processors. In this case, the one or more processors may be general-purpose processors, such as a central processing unit (CPU), an application processor (AP), etc., a processor used only for graphics (such as a graphics processing unit (GPU), a visual processing unit (VPU) and / or an AI-specific processor (such as a neural processing unit (NPU)).

[0121] One or more processors control the processing of input data according to predefined operating rules or artificial intelligence (AI) models stored in non-volatile memory and volatile memory. The predefined operating rules or artificial intelligence models can be provided by training or learning. Here, providing by learning means that by applying a learning algorithm to multiple learning data, a predefined operating rule or AI model with desired characteristics is formed. Learning can be performed in the device itself that executes AI according to an embodiment, and / or can be implemented by a separate server / device / system.

[0122] It should be noted that the control method according to the exemplary embodiment of the present disclosure can completely rely on the operation of computer programs or instructions to realize the corresponding functions, that is, each device corresponds to each step in the functional architecture of the computer program, so that the entire system is called through a special software package (for example, lib library) to realize the corresponding function.

[0123] The computer-readable storage medium according to an embodiment of the present disclosure may store a program or an instruction, and when the program or the instruction is executed by a processor, the processor is prompted to perform the above-mentioned analysis method.

[0124] Furthermore, according to an exemplary embodiment of the present disclosure, a computer device including a processor and a readable medium storing computer program instructions may be provided, wherein the instructions, when executed by the processor, prompt the processor to perform at least one of the above steps.

[0125] The analysis method according to the embodiment of the present disclosure can improve the accuracy of fault identification and life prediction of the main bearing of a wind turbine generator set.

[0126] The analysis method according to the embodiment of the present disclosure deeply integrates the experience of business experts and the optimal parameters, hyperparameters, and models driven by data based on different technologies, so that it can be truly applied to the fault identification and life prediction of the main bearing of the wind turbine.

[0127] The analysis method according to the embodiment of the present disclosure can narrow the scope of boarding inspection, improve the efficiency of inspection, and reduce the cost of inspection.

[0128] The specific implementation methods of the present disclosure are described in detail above. Although some embodiments have been shown and described, those skilled in the art should understand that these embodiments can be combined, modified and improved without departing from the principles and spirit of the present disclosure whose scope is defined by the claims and their equivalents. These combinations, modifications and improvements should also be within the scope of protection of the present disclosure.

Claims

1. An analysis method for a main bearing of a wind turbine generator set, characterized in that: The analysis method comprises: Acquire real-time monitoring data of the wind generator set including at least one of wind speed, power generation and generator speed and at least one of front bearing temperature and rear bearing temperature of the wind generator set; Acquire original feature data from the real-time monitoring data and calculate reconstructed feature data based on the original feature data; Inputting the original feature data and the reconstructed feature data into a pre-trained gradient boosting model to obtain a first diagnostic result characterizing whether a main bearing of the wind turbine generator set has a fault and a service life of the main bearing of the wind turbine generator set; Inputting the original feature data and the reconstructed feature data into a pre-trained regression model to obtain a first prediction result characterizing the service life of the main bearing of the wind turbine generator set; Fault information and service life information of the main bearing of the wind turbine generator set are determined based on the first diagnosis result and the first prediction result.

2. The analysis method for the main bearing of a wind turbine generator set according to claim 1, characterized in that: The step of determining the fault information and service life information of the main bearing of the wind turbine generator set based on the first diagnostic result and the first prediction result includes: stacking and fusing the prediction result in the first diagnostic result and the first prediction result to obtain a prediction fusion result, and determining the service life of the main bearing of the wind turbine generator set according to the prediction fusion result.

3. The analysis method for the main bearing of a wind turbine generator set according to claim 2, characterized in that: The analysis method further comprises: obtaining detection data of lubricating grease of the main bearing of the wind turbine generator set and on-site feedback data of the main bearing of the wind turbine generator set, The step of determining the fault information and service life information of the main bearing of the wind turbine generator set based on the first diagnostic result and the first prediction result includes: determining the fault state and service life of the main bearing of the wind turbine generator set according to the detection data, the field feedback data and the prediction fusion result.

4. The analysis method for the main bearing of a wind turbine generator set according to claim 3, characterized in that: The detection data includes the metal content in the lubricating grease, and the field feedback data includes at least one of the axial clearance data and the axial displacement data of the main bearing of the wind turbine generator set.

5. The analysis method for the main bearing of a wind turbine generator set according to claim 1, characterized in that: The gradient boosting model includes an XGBoost model, and the regression model includes a Lasso regression model.

6. The analysis method for the main bearing of a wind turbine generator set according to claim 1, characterized in that: The regression model is trained and tested as follows: Acquire historical monitoring data of the wind turbine generator set including at least one of wind speed, power generation and generator speed and at least one of front bearing temperature and rear bearing temperature of the wind turbine generator set; Acquire original feature variables from the historical monitoring data and calculate reconstructed feature variables based on the original feature variables; Dividing the combination of the original feature variables and the reconstructed feature variables into a first training set and a first test set; The regression model is trained using the first training set, prediction is performed on the first test set, and performance of the regression model is evaluated based on accuracy.

7. The analysis method for the main bearing of a wind turbine generator set according to claim 6, characterized in that: The gradient boosting model is trained and tested as follows: The gradient boosting model is trained using the first training set, tested on the first test set, and the performance of the gradient boosting model is evaluated based on accuracy.

8. The analysis method for the main bearing of a wind turbine generator set according to claim 7, characterized in that: During the training of the regression model and the gradient boosting model, the gradient boosting model determines a weight for each of the original feature variables and the reconstructed feature variables; During the training process of the regression model and the gradient boosting model, the original feature variables and the reconstructed feature variables are screened according to the weights, and the regression model and the gradient boosting model are retrained using the screened feature variables.

9. The analysis method for the main bearing of a wind turbine generator set according to claim 1, characterized in that: The analysis method further comprises: The Shapley value method is used to evaluate the contribution of each of the original feature data and the reconstructed feature data to the first diagnostic result of the gradient boosting model and the first prediction result of the regression model, and at least one of the original feature data and the reconstructed feature data is screened out based on the contribution. During the training process of the gradient boosting model and the regression model, the gradient boosting model and the regression model are retrained using the feature variables corresponding to the screened feature data.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a program or an instruction, and when the program or the instruction is executed by a processor, the processor is prompted to perform the analysis method according to any one of claims 1 to 9.

11. A computer device, characterized in that: The computer device comprises a memory and a processor, wherein the memory stores a program or an instruction, and when the program or the instruction is executed by the processor, the processor is prompted to perform the analysis method according to any one of claims 1 to 9.

Images