Wind power bearing base vibration intelligent diagnosis method

Abstract

The invention belongs to the technical field of fault diagnosis of mechanical parts, and particularly provides an intelligent diagnosis method for vibration of a wind power bearing base, mainly through a multi-source monitoring data acquisition and dual-mode switching strategy, in combination with stiffness, modal and other feature calculation and residual analysis, high-resolution impedance, strain and acoustic emission data are fused, and the fault diagnosis accuracy of the wind power bearing base is improved. And generating a comprehensive diagnosis result through the decision fusion model. The problem that damage types and accurate positions such as bolt looseness and microcracks of the wind power bearing base are difficult to distinguish in health monitoring of the wind power bearing base structure in the prior art is effectively solved, suspected abnormal paths are accurately positioned, the damage types are determined, unnecessary shutdown inspection and maintenance losses are reduced, damage development is restrained in time, and the service life of the wind power bearing base is prolonged. Serious structural faults are avoided, and diagnosis intelligence and reliability of the wind power bearing base are improved.

Description

Technical Field

[0001] This application belongs to the field of mechanical component fault diagnosis technology, and specifically relates to an intelligent method for diagnosing vibration of wind turbine bearing base. Background Technology

[0002] In the health monitoring of wind turbine bearing base structures, sensors are typically installed at preset locations on the base to periodically collect vibration and excitation signals. The extracted features are then compared with preset thresholds or benchmark models to directly determine whether the structure is abnormal.

[0003] However, when the above diagnostic method is in operation, if there is a loose bolt or micro-crack in a certain part of the wind turbine bearing base structure, and the initial monitoring mode is still used to observe the specific location, the subsequent data obtained will be difficult to effectively distinguish the specific damage type and precise location. This can easily lead to unnecessary downtime for inspection and maintenance, waste of maintenance resources, or failure to stop the damage from developing due to improper handling, which may eventually lead to more serious structural failures. Summary of the Invention

[0004] This application provides an intelligent vibration diagnosis method for wind turbine bearing bases, which effectively solves the problem in the prior art that it is difficult to distinguish the damage types and precise locations of wind turbine bearing base structure health monitoring such as loose bolts and microcracks. It achieves accurate location of suspected abnormal paths, clarifies the damage type, reduces unnecessary downtime for inspection and maintenance, promptly curbs damage development, avoids serious structural failures, and improves the intelligence and reliability of diagnosis.

[0005] To achieve the above objectives, this application adopts the following technical solution:

[0006] Firstly, this application provides a method for intelligent diagnosis of vibration of wind turbine bearing base, including:

[0007] Acquire multi-source monitoring data, including active excitation response signals from piezoelectric sensors along the base load transmission path collected in the first monitoring mode, a first vibration signal from the main bearing, and a second vibration signal from the flange connecting the base and the tower. Based on the multi-source monitoring data, calculate the local stiffness change index, main bearing vibration energy value, and overall modal frequency offset for each monitoring path. Use a preset health benchmark relationship model to generate model prediction values ​​for the local stiffness change index of each monitoring path. Calculate the residual between the model prediction values ​​and the measured values ​​of the local stiffness change index, and analyze the trend of the residual's variation. The deviation between the overall modal frequency offset trends of the period is determined. When the residual of any monitoring path continuously exceeds the preset first threshold and the deviation exceeds the preset second threshold, the monitoring path is determined to be a suspected abnormal path. The piezoelectric sensor on the suspected abnormal path is instructed to switch from the first monitoring mode to the second monitoring mode. High-resolution impedance data collected by the piezoelectric sensor using the second monitoring mode, as well as strain data and acoustic emission data from the suspected abnormal path are acquired. Based on the residual change trend, high-resolution impedance data, strain data and acoustic emission data, a comprehensive diagnostic result of the suspected abnormal path is generated using a preset decision fusion model.

[0008] Furthermore, based on multi-source monitoring data, the local stiffness change index, main bearing vibration energy value, and overall modal frequency offset of each monitoring path are calculated, including: extracting the path transfer function of each monitoring path from the active excitation response signal, constructing a mechanical impedance network spectrum based on the transfer function, calculating the dynamic time warping distance of the mechanical impedance network spectrum between adjacent monitoring cycles as the local stiffness change index; calculating the energy value of the first vibration signal within a preset frequency band as the main bearing vibration energy value; extracting at least one modal frequency related to the base support structure from the second vibration signal, and calculating the offset of the modal frequency relative to the reference frequency as the overall modal frequency offset.

[0009] Furthermore, the health baseline relationship model is used to characterize the statistical law of the local stiffness change index changing with the vibration energy value of the main bearing and the overall modal frequency offset under healthy conditions.

[0010] Furthermore, the piezoelectric sensor on the suspected abnormal path of the instruction is switched from the first monitoring mode to the second monitoring mode, including: the monitoring mode of the piezoelectric sensor on the suspected abnormal path of the instruction is switched from the periodic wide-area excitation inspection mode to the directional high-frequency focusing monitoring mode.

[0011] Furthermore, a comprehensive diagnostic result for suspected abnormal paths is generated using a pre-defined decision fusion model, including: extracting the guided wave propagation attenuation coefficient and signal energy attenuation spectrum from high-resolution impedance data as high-resolution impedance features; calculating the cyclic strain energy density distribution based on strain data as the first auxiliary feature; and analyzing the impact count rate and amplitude distribution based on acoustic emission data as the second auxiliary feature; inputting the residual change trend, high-resolution impedance features, first auxiliary feature, and second auxiliary feature into the pre-defined decision fusion model to generate a comprehensive diagnostic result for suspected abnormal paths.

[0012] Furthermore, the decision fusion model is used to assign weights to the high-resolution impedance features, the first auxiliary features, and the second auxiliary features based on the residual change trend, and then weighted and fused to obtain a fused feature vector. Based on the fused feature vector, a comprehensive diagnostic result containing the damage type and damage level of the suspected abnormal path is generated.

[0013] Furthermore, weights are assigned to the high-resolution impedance feature, the first auxiliary feature, and the second auxiliary feature based on the residual change trend, including: determining the feature pattern of the residual change trend, where the feature pattern includes slow linear growth and abrupt changes; when the feature pattern of the residual change trend is slow linear growth, the weight assigned to the first auxiliary feature is greater than the weight assigned to the high-resolution impedance feature and the second auxiliary feature; when the feature pattern of the residual change trend is abrupt changes, the weight assigned to the second auxiliary feature is greater than the weight assigned to the high-resolution impedance feature and the first auxiliary feature.

[0014] Furthermore, a comprehensive diagnostic result containing the damage type and damage level of the suspected abnormal path is generated based on the fused feature vector, including: matching and comparing the fused feature vector with a preset diagnostic mapping table to obtain a comparison result; the diagnostic mapping table stores multiple standard weighted feature vectors, as well as the damage type and damage level corresponding to each standard weighted feature vector; and determining the damage type and damage level of the suspected abnormal path based on the comparison result.

[0015] Furthermore, the fused feature vector is matched and compared with a preset diagnostic mapping table to obtain the comparison results, including: calculating the Euclidean distance between the fused feature vector and each standard weighted feature vector, and selecting the standard weighted feature vector with the smallest Euclidean distance as the matching result.

[0016] Secondly, this application provides a wind turbine bearing base vibration intelligent diagnostic system, comprising:

[0017] Data acquisition module: used to acquire multi-source monitoring data, which includes the active excitation response signals of each monitoring path collected by the piezoelectric sensor on the load transmission path of the base in the first monitoring mode, the first vibration signal from the main bearing, and the second vibration signal from the flange connecting the base and the tower.

[0018] Feature Calculation and Prediction Module: This module is used to calculate the local stiffness change index, main bearing vibration energy value, and overall modal frequency offset of each monitoring path based on multi-source monitoring data. It also generates model prediction values ​​of the local stiffness change index of each monitoring path using a preset health benchmark relationship model.

[0019] Anomaly detection and mode switching module: It is used to calculate the residual between the model prediction value and the measured value of the local stiffness change index, and to calculate the deviation between the trend of the residual change and the trend of the overall modal frequency offset during the same period. When the residual of any monitoring path continuously exceeds the preset first threshold and the deviation exceeds the preset second threshold, the monitoring path is determined to be a suspected abnormal path, and the piezoelectric sensor on the suspected abnormal path is instructed to switch from the first monitoring mode to the second monitoring mode.

[0020] Directional data acquisition module: used to acquire high-resolution impedance data collected by the piezoelectric sensor using the second monitoring mode, as well as strain data and acoustic emission data from suspected abnormal paths.

[0021] Fusion Diagnosis Module: Based on residual change trends, high-resolution impedance data, strain data, and acoustic emission data, this module uses a pre-defined decision fusion model to generate comprehensive diagnostic results for suspected abnormal paths.

[0022] Thirdly, this application provides a computer program product, including a computer program or instructions, which, when executed by a processor, implement the steps of an intelligent diagnostic method for vibration of a wind turbine bearing base.

[0023] The beneficial effects of this application are:

[0024] This application utilizes a multi-source monitoring data acquisition and dual-mode switching strategy, combined with stiffness, modal and other characteristic calculations and residual analysis, and integrates high-resolution impedance, strain and acoustic emission data. Through a decision fusion model, a comprehensive diagnostic result is generated, effectively solving the problem in existing technologies where health monitoring of wind turbine bearing base structures is difficult to distinguish the damage types and precise locations of wind turbine bearing base bolt loosening, microcracks and other damage types. It achieves accurate location of suspected abnormal paths, clarifies damage types, reduces unnecessary downtime for inspection and maintenance, promptly curbs damage development, avoids serious structural failures, and improves the intelligence and reliability of diagnosis.

[0025] Other features and advantages of this application will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures pointed out in the description and the accompanying drawings. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0027] Figure 1 A flowchart illustrating the intelligent vibration diagnosis method for wind turbine bearing bases of this application is shown.

[0028] Figure 2 This paper illustrates a flowchart of the process for generating comprehensive diagnostic results of suspected abnormal paths using a decision fusion model, as described in this application.

[0029] Figure 3 A schematic diagram of the process for assigning weights to high-resolution impedance features, first auxiliary features, and second auxiliary features in this application is shown.

[0030] Figure 4 The diagram illustrates the process of generating damage type and damage level based on fused feature vectors according to this application. Detailed Implementation

[0031] To address the problems raised in the background technology, this application adopts a multi-source monitoring data acquisition and dual-mode switching strategy, combined with stiffness, modal and other feature calculations and residual analysis, and integrates high-resolution impedance, strain and acoustic emission data. The comprehensive diagnostic results are generated through a decision fusion model, which effectively improves the diagnostic intelligence and reliability of wind turbine bearing base structures.

[0032] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0033] In some embodiments, such as Figure 1 As shown, this application provides a smart diagnostic method for vibration of wind turbine bearing base, including:

[0034] S1. Acquire multi-source monitoring data, which includes the active excitation response signals of each monitoring path collected by the piezoelectric sensor on the load transmission path of the base in the first monitoring mode, the first vibration signal from the main bearing, and the second vibration signal from the flange connecting the base and the tower.

[0035] The first monitoring mode refers to a conventional, periodic inspection mode. In this mode, a specific piezoelectric element in the array is designated as the exciter, emitting a sweep frequency signal or pulse signal within a specific frequency range as the active excitation signal. The other piezoelectric elements in the array act as sensors, synchronously receiving the active excitation response signal obtained after the active excitation signal propagates through the structure. The active excitation response signal reflects the frequency response characteristics changes experienced by the signal along a specific propagation path. This change is closely related to the local mechanical impedance of the structure along the path. The mechanical impedance itself is directly related to the stiffness, damping, and mass distribution of the structure. By processing the active excitation response signal, it is possible to highly sensitively capture minute stiffness changes or early damage in local areas of the structure, such as welds or bolted connections.

[0036] The first vibration signal is collected by a vibration acceleration sensor installed on the main bearing housing. It is used to record the actual working vibration generated by the rotation of the wind turbine and gear transmission during normal operation of the wind turbine, and transmitted to the base through the main bearing. It directly reflects the intensity of the current operating conditions, such as the load size, whether there is imbalance or early failure characteristics of gears and bearings.

[0037] The second vibration signal is collected by a triaxial vibration acceleration sensor installed at the connecting flange, which is used to record the overall dynamic response of the entire tower base support structure under the combined action of environmental excitations, such as wind load and background micro-vibration, and operational excitations.

[0038] S2. Calculate the local stiffness change index, main bearing vibration energy value, and overall modal frequency offset for each monitoring path based on multi-source monitoring data, and generate model prediction values ​​of the local stiffness change index for each monitoring path using a preset health benchmark relationship model.

[0039] The local stiffness variation index is used to characterize the relative degree of change in the local stiffness of the corresponding monitoring path structure. The main bearing vibration energy value is used to characterize the dynamic load level acting on the base under the current working condition. The overall modal frequency offset reflects the change in the overall stiffness of the entire tower base support structure, such as stiffness changes caused by temperature changes or overall damage.

[0040] S3. Calculate the residual between the predicted value of the model and the measured value of the local stiffness change index, and the deviation between the trend of the calculated residual and the trend of the overall modal frequency offset during the same period. When the residual of any monitoring path continuously exceeds the preset first threshold and the deviation exceeds the preset second threshold, the monitoring path is determined to be a suspected abnormal path, and the piezoelectric sensor on the suspected abnormal path is instructed to switch from the first monitoring mode to the second monitoring mode.

[0041] For each monitoring path, the difference between the measured value and the model prediction of its local stiffness change index is the residual. A linear fit is performed on the residual sequence over a continuous time window to obtain the slope of the change. As a trend of residual change, a linear fit is performed on the overall modal frequency shift sequence within the same time window, and the resulting slope is... As a trend of change in the overall modal frequency offset.

[0042] The deviation D between the residual variation trend and the overall modal frequency offset variation trend is calculated to quantify the inconsistency in the rate and direction of their changes. Refer to the formula: The greater the deviation, the more it represents... The local change trend represented by and from The more significant the difference in the overall trend of change in the representation.

[0043] The first threshold reflects the maximum allowable instantaneous deviation of local stiffness changes from the healthy baseline, while the second threshold reflects the maximum acceptable trend deviation between the local change trend and the overall structural change trend. The first threshold can be set based on the standard deviation of the healthy data residual sequence, and the second threshold can be set based on the distribution characteristics of the healthy data deviation sequence. For example, the first threshold can be set to three times the standard deviation of the mean of the healthy residual sequence, and the second threshold can be set to the 95th percentile of the healthy deviation sequence.

[0044] When the residual of a certain monitoring path continuously exceeds the first threshold and the deviation exceeds the second threshold, it means that the local characteristics of the path have not only changed abnormally, but this change cannot be explained by the normal changes in the overall structure, strongly suggesting the existence of local damage.

[0045] S4. Acquire high-resolution impedance data collected by the piezoelectric sensor using the second monitoring mode, as well as strain data and acoustic emission data from the suspected abnormal path.

[0046] Specifically, once a suspected abnormal path is identified, the piezoelectric sensor on the suspected abnormal path is instructed to switch from the first monitoring mode to the second monitoring mode. The second monitoring mode is an enhanced monitoring mode compared to the first monitoring mode. Specifically, it involves higher frequency, higher energy excitation, and higher sampling rate signal acquisition for the specific suspected abnormal path, thereby obtaining richer response signals with a higher signal-to-noise ratio, i.e., high-resolution impedance data. At the same time, the system simultaneously activates high-frequency strain gauges and acoustic emission sensors deployed near the physical location of the suspected abnormal path to collect strain data and acoustic emission data, respectively.

[0047] S5. Based on the residual change trend, high-resolution impedance data, strain data and acoustic emission data, a comprehensive diagnostic result of suspected abnormal paths is generated using a preset decision fusion model.

[0048] For example, on day T, the system controls the piezoelectric sensor array arranged on the base support rib to operate, sequentially exciting and receiving signals from all 50 preset monitoring paths, and collecting vibration signals from the main bearing and the tower flange. After processing the data, it is found that the measured value of the local stiffness change index of the third monitoring path of support rib A is 0.735, the current vibration energy value of the main bearing is 0.8, the overall modal frequency offset is -0.05, the model prediction value of the local stiffness change index generated by the health benchmark relationship model is 0.485, the calculated residual is 0.25, and its absolute value is 0.25. Analyzing the data of the last 8 days, the absolute value sequence of the residual for this path is [0.18, 0.19, 0.19]. [20, 0.21, 0.22, 0.23, 0.24, 0.25] has exceeded the preset first threshold of 0.20 for three consecutive days. At the same time, the deviation between its residual change trend and the overall modal frequency offset change trend is calculated to be 0.45, exceeding the preset second threshold of 0.1. Therefore, the monitoring path is determined to be a suspected abnormal path. The piezoelectric sensor related to the path is then instructed to switch from the first monitoring mode to the second monitoring mode, and a start-up command is sent to control the activation of the strain gauge and acoustic emission sensor at that location. Finally, the system integrates all data and outputs a comprehensive diagnostic result through the decision fusion model: "Location: Support rib A-path 3; Damage type: Crack; Damage level: Moderate".

[0049] In some embodiments, calculating the local stiffness change index, main bearing vibration energy value, and overall modal frequency offset for each monitoring path based on multi-source monitoring data includes:

[0050] The path transfer function of each monitoring path in the active excitation response signal is extracted, and a mechanical impedance network spectrum is constructed based on the transfer function. The dynamic time warping distance of the mechanical impedance network spectrum between adjacent monitoring cycles is calculated as the local stiffness change index. The energy value of the first vibration signal in the preset frequency band is calculated as the vibration energy value of the main bearing. At least one modal frequency related to the base support structure is extracted from the second vibration signal, and the offset of the modal frequency relative to the reference frequency is calculated as the overall modal frequency offset.

[0051] Specifically, for each monitoring path, the path transfer function is extracted. The path transfer function describes the proportional relationship between the sensor-received signal and the exciter-transmitted signal in the frequency domain, reflecting the frequency response characteristics of the signal propagating in a specific path. The frequency response characteristics are directly related to the physical characteristics of the structure along the path, such as stiffness, mass, and damping. After calculating the transfer functions of all paths, a mechanical impedance network spectrum reflecting the dynamic characteristics of the entire monitoring area structure can be constructed. The dynamic time warping distance between the mechanical impedance network spectrum obtained in the current monitoring cycle and the spectrum obtained in the previous monitoring cycle is calculated. This dynamic time warping distance is used as the local stiffness change index of the monitoring path in this cycle.

[0052] The first vibration signal is preprocessed, and the energy value of the first vibration signal is calculated as the vibration energy value of the main bearing within a preset frequency band that can reflect the main load characteristics, such as the frequency band related to the fan rotation frequency and gear meshing frequency.

[0053] Modal parameter identification algorithms, such as peak picking, are applied to the second vibration signal to directly analyze and generate the main modes. Each main mode contains its inherent frequency value and mode shape vector. Based on prior knowledge of the dynamics of the base and tower structure, a preset target frequency range is invoked, such as the lower frequency band of 1-5Hz, which corresponds to the typical range of the first-order overall bending frequency of the base structure. The frequency values ​​of each mode are compared with the target frequency range one by one. The mode shape vectors of each mode whose frequency values ​​fall within the target frequency range are matched with the pre-stored expected shape vectors of the main overall deformation of the base structure. For example, the cosine similarity is calculated, and the mode with the highest mode shape vector matching degree is selected. This mode is the target mode closely related to the vibration of the base support structure. The difference between the current frequency of this mode and a reference frequency measured in a healthy state is calculated as the overall modal frequency offset.

[0054] For example, for the third monitoring path of support rib A, the system extracts the path transfer function obtained from this monitoring, constructs a mechanical impedance network spectrum based on the transfer function, calculates the dynamic time warping distance of the mechanical impedance network spectrum between adjacent monitoring cycles, and finally obtains a local stiffness change index of 0.735; calculates the energy value of the first vibration signal collected on day T within the preset frequency band of 0.5Hz to 50Hz, and obtains a main bearing vibration energy value of 0.8; extracts the first-order modal frequency related to the base support structure from the second vibration signal collected on day T, which is 2.45Hz, and calculates the offset of this modal frequency relative to the reference frequency of 2.50Hz as the overall modal frequency offset of -0.05Hz.

[0055] In some embodiments, a preset health baseline relationship model is used to characterize the statistical law of the local stiffness change index changing with the vibration energy value of the main bearing and the overall modal frequency shift under a healthy state. Its expression can be defined as: in, The value represents the predicted local stiffness variation index, BE represents the main bearing vibration energy value, and MFD represents the global modal frequency shift. , These represent the regression coefficients of BE and MFD, respectively, reflecting... Sensitivity to BE and MFD respectively Represents a constant term.

[0056] During the initial healthy operation phase of the wind turbine, a large number of multi-source monitoring data samples were collected. Each sample contained the measured value of the local stiffness change index, the main bearing vibration energy value, and the overall modal frequency shift acquired at the same monitoring time. The model was trained using the multi-source monitoring data samples: the main bearing vibration energy value and the overall modal frequency shift from the multi-source monitoring data samples were used as inputs, and the corresponding measured value of the local stiffness change index was used as the target output. Least squares method was used for multiple linear regression to solve for the... , and The specific numerical value. After training, for any new input BE and MFD, the model can output a value expected under a healthy state. .

[0057] For example, if the vibration energy value of the main bearing is 0.8 and the overall modal frequency shift is -0.05Hz on day T, then the model-predicted local stiffness change index is [value missing]. .

[0058] In some embodiments, the piezoelectric sensor on the suspected abnormal path is switched from a first monitoring mode to a second monitoring mode, including: the monitoring mode of the piezoelectric sensor on the suspected abnormal path is switched from a periodic wide-area excitation inspection mode to a directional high-frequency focusing monitoring mode.

[0059] In the periodic wide-area excitation inspection mode, the system automatically controls the exciter and sensor to complete a complete process of active excitation signal transmission and excitation response signal reception for all preset monitoring paths at fixed time intervals, such as every 24 hours. In this mode, the monitoring efficiency is high, but the time resolution and signal strength of each measurement may be limited.

[0060] In the directional high-frequency focusing monitoring mode, the system concentrates resources to conduct high-density and high-intensity monitoring of suspected abnormal paths. For example, the monitoring interval for suspected abnormal paths is significantly shortened from once every 24 hours to once per minute, and the energy of the excitation signal is increased, such as increasing the excitation signal energy from 50V to 100V, in order to stimulate a stronger response signal and improve the signal-to-noise ratio.

[0061] In some embodiments, such as Figure 2 As shown, a comprehensive diagnostic result for suspected abnormal paths is generated using a pre-defined decision fusion model, including:

[0062] S5.1. Extract the guided wave propagation attenuation coefficient and signal energy attenuation spectrum from the high-resolution impedance data as high-resolution impedance features; calculate the cyclic strain energy density distribution based on the strain data as the first auxiliary feature; and analyze the impact count rate and amplitude distribution based on the acoustic emission data as the second auxiliary feature.

[0063] S5.2. Input the residual change trend, high-resolution impedance characteristics, first auxiliary characteristics and second auxiliary characteristics into the preset decision fusion model to generate a comprehensive diagnostic result of suspected abnormal paths.

[0064] For example, if the guided wave propagation attenuation coefficient is 0.68 and the signal energy attenuation spectrum is -22dB extracted from high-resolution impedance data, then the high-resolution impedance characteristics are [0.68, -22]. If the cyclic strain energy density distribution is calculated to be 0.085J / m³ based on strain data, then the first auxiliary characteristic is [0.085]. If the impact count rate is 15 times / second and the amplitude distribution is 58dB based on acoustic emission data analysis, then the second auxiliary characteristic is [15, 58].

[0065] In some embodiments, the decision fusion model is used to assign weights to the high-resolution impedance features, the first auxiliary features, and the second auxiliary features according to the residual change trend, and to perform weighted fusion to obtain a fusion feature vector. Based on the fusion feature vector, a comprehensive diagnostic result containing the damage type and damage level of the suspected abnormal path is generated.

[0066] Reference formula: ;in, Represents the fused feature vector. This represents a high-resolution impedance eigenvector. Represents the first auxiliary feature vector. Represents the second auxiliary feature vector. , , Represent , , The weight.

[0067] The decision fusion model generates a comprehensive diagnostic result based on the fusion feature vector, which includes the damage type and damage level of the suspected abnormal path. The damage types include cracks, loosening, corrosion, etc., and the damage levels include minor, moderate, and severe.

[0068] In some embodiments, such as Figure 3 As shown, weights are assigned to the high-resolution impedance feature, the first auxiliary feature, and the second auxiliary feature based on the residual change trend. This includes: determining the feature pattern of the residual change trend, where the feature pattern includes slow linear growth and abrupt changes; when the feature pattern of the residual change trend is slow linear growth, the weight assigned to the first auxiliary feature is greater than the weight assigned to the high-resolution impedance feature and the second auxiliary feature; when the feature pattern of the residual change trend is abrupt changes, the weight assigned to the second auxiliary feature is greater than the weight assigned to the high-resolution impedance feature and the first auxiliary feature.

[0069] Specifically, trend analysis can be performed on the residual values ​​over the most recent 5-10 periods. When the slope of the residual value changes... When the slope is greater than zero and less than the preset slope threshold, the characteristic pattern of the residual change trend is determined to be slow linear growth. The slope threshold can be set according to the statistical distribution characteristics of the slope of the residual sequence in the historical data of healthy operation. For example, it can be set to three standard deviations of the mean of the absolute value of the slope sequence. When the residual change in adjacent periods exceeds the preset change threshold, the characteristic pattern of the residual change trend is determined to be abrupt change. The change threshold can be set according to the statistical distribution characteristics of the residual change in adjacent periods in the historical data of healthy operation. For example, it can be set to three standard deviations of the mean of the absolute value of the change sequence.

[0070] If the residual change trend shows a slow linear growth pattern, it may indicate a gradual damage mode, such as the accumulation of fatigue damage. For this damage mode, the cyclic strain energy density, which reflects the internal energy dissipation and plastic deformation of the material, may be more indicative and can be assigned a relatively high weight to the first auxiliary feature. If the residual change trend shows an abrupt change, it may indicate a sudden damage event, such as the instantaneous initiation or propagation of microcracks. For this damage mode, the acoustic emission feature, which is extremely sensitive to fracture events, may be more indicative and can be assigned a relatively high weight to the second auxiliary feature. High-resolution impedance features are usually sensitive to various types of structural changes and can be set to a base value or dynamically fine-tuned according to the situation of other features.

[0071] For example, when the characteristic pattern of the residual change trend is slow linear growth, , , The values ​​can be 0.2, 0.5, and 0.3 respectively; when the characteristic pattern of the residual change trend shows an abrupt change, , , The values ​​can be 0.2, 0.3, and 0.5 respectively; for other modes, the default weight allocation can be used, such as... , , The values ​​can be 0.4, 0.3, and 0.3 respectively.

[0072] For example, if , , If the values ​​are 0.2, 0.5, and 0.3 respectively, then the calculated fusion feature vectors are... [0.136, -4.4, 0.0425, 4.5, 17.4].

[0073] In some embodiments, such as Figure 4 As shown, a comprehensive diagnostic result containing the damage type and damage level of the suspected abnormal path is generated based on the fused feature vector. This includes: matching and comparing the fused feature vector with a preset diagnostic mapping table to obtain a comparison result; the diagnostic mapping table stores multiple standard weighted feature vectors, as well as the damage type and damage level corresponding to each standard weighted feature vector; and determining the damage type and damage level of the suspected abnormal path based on the comparison result.

[0074] In some embodiments, the fused feature vector is matched and compared with a preset diagnostic mapping table to obtain a comparison result, including: calculating the Euclidean distance between the fused feature vector and each standard weighted feature vector, and selecting the standard weighted feature vector with the smallest Euclidean distance as the matching result.

[0075] For example, if the standard vector corresponding to "moderate crack" is [0.15, -4.0, 0.05, 4.8, 18.0], and the standard vector corresponding to "slight loosening" is [0.05, -1.5, 0.01, 2.0, 10.0], calculate... The distance to the "moderate crack" vector is 0.781, and the distance to the "slight loosening" vector is approximately 8.332. Therefore, the damage type of this suspected abnormal path is "crack", and the damage level is "moderate".

[0076] In some embodiments, this application provides a wind turbine bearing base vibration intelligent diagnostic system, including:

[0077] Data acquisition module: used to acquire multi-source monitoring data, which includes the active excitation response signals of each monitoring path collected by the piezoelectric sensor on the load transmission path of the base in the first monitoring mode, the first vibration signal from the main bearing, and the second vibration signal from the flange connecting the base and the tower.

[0078] Feature Calculation and Prediction Module: This module is used to calculate the local stiffness change index, main bearing vibration energy value, and overall modal frequency offset of each monitoring path based on multi-source monitoring data. It also generates model prediction values ​​of the local stiffness change index of each monitoring path using a preset health benchmark relationship model.

[0079] Anomaly detection and mode switching module: It is used to calculate the residual between the model prediction value and the measured value of the local stiffness change index, and to calculate the deviation between the trend of the residual change and the trend of the overall modal frequency offset during the same period. When the residual of any monitoring path continuously exceeds the preset first threshold and the deviation exceeds the preset second threshold, the monitoring path is determined to be a suspected abnormal path, and the piezoelectric sensor on the suspected abnormal path is instructed to switch from the first monitoring mode to the second monitoring mode.

[0080] Directional data acquisition module: used to acquire high-resolution impedance data collected by the piezoelectric sensor using the second monitoring mode, as well as strain data and acoustic emission data from suspected abnormal paths.

[0081] Fusion Diagnosis Module: Based on residual change trends, high-resolution impedance data, strain data, and acoustic emission data, this module uses a pre-defined decision fusion model to generate comprehensive diagnostic results for suspected abnormal paths.

[0082] In some embodiments, this application provides a computer program product, including a computer program or instructions, which, when executed by a processor, implement the steps of a method for intelligent diagnosis of vibration of a wind turbine bearing base.

[0083] It should be noted that, in this application, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0084] Any references to memory, storage, database, or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory.

[0085] Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for intelligent diagnosis of vibration of wind turbine bearing base, characterized in that, include: Acquire multi-source monitoring data, which includes the active excitation response signals of each monitoring path collected by the piezoelectric sensor on the load transmission path of the base in the preset first monitoring mode, the first vibration signal from the main bearing, and the second vibration signal from the flange connecting the base and the tower. Based on the multi-source monitoring data, the local stiffness change index, main bearing vibration energy value, and overall modal frequency offset of each monitoring path are calculated. The model prediction value of the local stiffness change index of each monitoring path is generated using a preset health benchmark relationship model. The residual between the model prediction value and the measured value of the local stiffness change index is calculated, and the deviation between the trend of the residual change and the trend of the overall modal frequency offset during the same period is calculated. When the residual of any monitoring path continuously exceeds the preset first threshold and the deviation exceeds the preset second threshold, the monitoring path is determined to be a suspected abnormal path, and the piezoelectric sensor on the suspected abnormal path is instructed to switch from the first monitoring mode to the second monitoring mode. Acquire high-resolution impedance data collected by the piezoelectric sensor using the second monitoring mode, as well as strain data and acoustic emission data from suspected abnormal paths; Based on the residual change trend, high-resolution impedance data, strain data, and acoustic emission data, a comprehensive diagnostic result for suspected abnormal paths is generated using a preset decision fusion model.

2. The method according to claim 1, characterized in that, Based on the multi-source monitoring data, the local stiffness change index, main bearing vibration energy value, and overall modal frequency offset of each monitoring path are calculated, including: The path transfer function of each monitoring path in the active excitation response signal is extracted, and a mechanical impedance network spectrum is constructed based on the transfer function. The dynamic time warping distance of the mechanical impedance network spectrum between adjacent monitoring cycles is calculated as the local stiffness change index. The energy value of the first vibration signal in the preset frequency band is calculated as the vibration energy value of the main bearing. At least one modal frequency related to the base support structure is extracted from the second vibration signal, and the offset of the modal frequency relative to the reference frequency is calculated as the overall modal frequency offset.

3. The method according to claim 1, characterized in that, The health benchmark relationship model is used to characterize the statistical law of the local stiffness change index changing with the vibration energy value of the main bearing and the overall modal frequency offset under a healthy state.

4. The method according to claim 1, characterized in that, The instruction switches the piezoelectric sensor on the suspected abnormal path from the first monitoring mode to the second monitoring mode, including: The monitoring mode of the piezoelectric sensor on the suspected abnormal path was switched from the periodic wide-area excitation inspection mode to the directional high-frequency focused monitoring mode.

5. The method according to claim 1, characterized in that, A comprehensive diagnostic result for suspected abnormal paths is generated using a pre-defined decision fusion model, including: The guided wave propagation attenuation coefficient and signal energy attenuation spectrum are extracted from the high-resolution impedance data as high-resolution impedance features; the cyclic strain energy density distribution is calculated based on the strain data as a first auxiliary feature; and the impact count rate and amplitude distribution are analyzed based on the acoustic emission data as a second auxiliary feature. The residual change trend, high-resolution impedance characteristics, first auxiliary characteristics, and second auxiliary characteristics are input into a preset decision fusion model to generate a comprehensive diagnostic result for the suspected abnormal path.

6. The method according to claim 5, characterized in that, The decision fusion model is used to assign weights to the high-resolution impedance features, the first auxiliary features, and the second auxiliary features according to the residual change trend, and then perform weighted fusion to obtain a fusion feature vector. Based on the fusion feature vector, a comprehensive diagnostic result containing the damage type and damage level of the suspected abnormal path is generated.

7. The method according to claim 6, characterized in that, Weights are assigned to the high-resolution impedance features, the first auxiliary feature, and the second auxiliary feature based on the residual variation trend, including: Identify characteristic patterns of residual change trends, including slow linear growth and abrupt changes; When the characteristic pattern of the residual change trend is slow linear growth, the weight assigned to the first auxiliary feature is greater than the weight assigned to the high-resolution impedance feature and the second auxiliary feature; when the characteristic pattern of the residual change trend is abrupt change, the weight assigned to the second auxiliary feature is greater than the weight assigned to the high-resolution impedance feature and the first auxiliary feature.

8. The method according to claim 6, characterized in that, A comprehensive diagnostic result is generated based on the fused feature vector, including the damage type and damage level of the suspected abnormal path, including: The fused feature vector is matched and compared with a preset diagnostic mapping table to obtain the comparison results; the diagnostic mapping table stores multiple standard weighted feature vectors, as well as the damage type and damage level corresponding to each standard weighted feature vector; Based on the comparison results, the damage type and damage level of the suspected abnormal path are determined.

9. The method according to claim 8, characterized in that, The fused feature vector is matched and compared with a pre-defined diagnostic mapping table to obtain the comparison results, including: Calculate the Euclidean distance between the fused feature vector and each standard weighted feature vector, and select the standard weighted feature vector with the smallest Euclidean distance as the matching result.

Images