Fan operation abnormal vibration monitoring method and system based on multi-sensor fusion

Abstract

The application discloses a fan operation abnormal vibration monitoring method and system based on multi-sensor fusion, relates to the technical field of fan operation monitoring, and comprises the following steps: deploying multiple types of sensors on a fan, collecting sensor data in real time, adopting IEEE1588 protocol for sensor data synchronization processing, combining wavelet denoising for pretreatment, and then extracting characteristic data by using a principal component PCA combined independent component ICA analysis method; the application considers the actual conditions of offshore wind farms by deploying multiple types of sensors on the fan, finely processes and extracts features after collecting multi-source signals in real time, accurately monitors and reliably identifies abnormal vibration conditions of the fan in a complex environment in a coastal or offshore wind farm, so that early abnormal features of fan operation can be captured in time, normal changes caused by environmental factors and real fault abnormalities can be effectively distinguished, and early fault discovery is further realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fan operation monitoring, in particular to a fan operation abnormal vibration monitoring method and system based on multi-sensor fusion. BACKGROUND

[0002] The environment where the coastal or offshore fan is located is extremely harsh, and it is long-term in the complex environment of high salt fog, high humidity and strong wind and wave during operation, and such environmental factors will affect the vibration characteristics of the fan.

[0003] For example, on the one hand, the periodic changes of strong wind and tide on the sea will have a combined effect on the fan, the strong wind makes the fan blade vibrate, and the tide rise and fall changes the hydrodynamic environment around the fan foundation, the interaction of the two may cause the modulation phenomenon of low-frequency vibration of the fan, and such low-frequency modulation will mask the early fault characteristics of the fan, and the traditional monitoring method is difficult to detect the early signs of failure, resulting in failure to discover in time, increasing the risk of fan damage; on the other hand, the fan blade is usually made of glass fiber / carbon fiber composite material, and such composite material has hygroscopicity, when running in the coastal high humidity environment for a long time, the blade will absorb water, causing the material elastic modulus to change, the density to increase, the damping characteristics to enhance the toughening effect, and then change the natural frequency and vibration response characteristics of the blade, and also change the propagation speed and attenuation performance of the sound wave in the acoustic monitoring of the blade, and many changes may affect the system to diagnose the blade modal change caused by humidity saturation as internal structure delamination, web cracking and other damage, thereby causing false abnormal alarm, making it difficult for operation and maintenance personnel to accurately judge whether the fan has really failed, increasing the operation and maintenance cost and workload.

[0004] The prior art has the following disadvantages: the existing fan vibration monitoring technology has the following limitations: first, the existing technology mainly uses a single type of sensor, lacks deep fusion of multi-physical field data, and cannot fully obtain the information of the fan operation state; second, the environmental impact factors are not considered, resulting in a high false alarm rate, which brings great trouble to operation and maintenance work; finally, it lacks self-adaptive learning ability and cannot adapt to the dynamic changes of the offshore environment, so that the monitoring effect of the monitoring system under different environmental conditions is unstable; therefore, it is urgent to propose a fan operation abnormal vibration monitoring method and system based on multi-sensor fusion, which can accurately monitor and reliably identify the abnormal vibration of the fan in the complex environment of the coastal or offshore wind farm, effectively distinguish between normal changes caused by environmental factors and real fault abnormalities, thereby reducing the false alarm rate, realizing early detection and accurate diagnosis of faults, and improving the safety and reliability of fan operation.

[0005] The above information disclosed in the BACKGROUND section is only used to enhance the understanding of the background of the present disclosure, and therefore it can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0006] The purpose of the present application is to provide a fan operation abnormal vibration monitoring method and system based on multi-sensor fusion. The present application extracts feature data by deploying multiple types of sensors on the fan, establishes a scene feature template library for complex offshore environments, uses physical modeling combined with data-driven modeling to establish an accurate scene model, and comprehensively uses vibration analysis technology, acoustic detection technology and machine learning diagnosis algorithm to diagnose faults for quantified abnormal features, to solve the problems in the above background technology.

[0007] To achieve the above purpose, the present application provides the following technical scheme: a fan operation abnormal vibration monitoring method based on multi-sensor fusion, comprising the following steps:

[0008] S1, deploying multiple types of sensors on the fan, collecting real-time sensor data, using IEEE1588 protocol for sensor data synchronization processing, combining wavelet denoising for preprocessing, and using principal component PCA combined with independent component ICA analysis method to extract feature data;

[0009] S2, establishing a scene feature template library for complex offshore environments, comparing the extracted feature data with the scene feature template library, using cosine similarity algorithm to identify the scene where the current fan is located, defining the feature threshold of the abnormal early warning mechanism, and triggering the abnormal early warning when the extracted feature data exceeds the feature threshold;

[0010] S3, according to the scene identified by the extracted feature data, using physical modeling combined with data-driven modeling to establish an accurate scene model, inputting the extracted feature data into the corresponding scene model, simulating the normal operation state, comparing the actual data to quantify the abnormal feature index, refining the scene classification, setting the abnormal feature threshold, and determining the abnormal type and severity according to the comparison of the quantified abnormal feature index;

[0011] S4, comprehensively using vibration analysis technology, acoustic detection technology and machine learning diagnosis algorithm to diagnose faults for quantified abnormal features, generating accurate fault positioning diagnosis results, developing a scene adaptive mechanism, adaptively adjusting the diagnosis parameters according to the scene characteristics, and automatically triggering the corresponding processing decision;

[0012] S5, based on the system feedback of the diagnosis results and processing decisions, establishing a double-loop feedback mechanism to continuously optimize the system performance, while building a knowledge base, regularly analyzing and updating the knowledge base, and optimizing the monitoring and diagnosis capability of the system.

[0013] Optionally, the collection of each sensor data is as follows:

[0014] Counting the types of sensors and the number of corresponding sensor types deployed on the wind turbine, wherein the sensor types include acceleration sensors, displacement sensors, acoustic sensors, anemometers, tide sensors, and temperature and humidity sensors;

[0015] Numbering the types and quantities of the counted multi-type sensors to mark each sensor deployed on the wind turbine;

[0016] Starting the multi-type sensors to collect data in a continuous time period to form time-series sensor data;

[0017] Integrating the sensor data into a multi-sensor data set.

[0018] Optionally, the data synchronization processing and wavelet denoising preprocessing steps are as follows:

[0019] For each sensor, there is a time offset and network transmission delay between the local timestamp and the master clock timestamp;

[0020] Using the interleaved signaling mechanism of the IEEE1588 PTP protocol for the sensor data, the time offset and network transmission delay are calculated;

[0021] Based on the time offset and network transmission delay, the sensor data of the time series is corrected, the timestamps of all sensor data are unified to a unified timestamp on a master clock, and a time-aligned multi-sensor data set is generated;

[0022] Wavelet denoising preprocessing is performed on the time-aligned multi-sensor data set, a wavelet basis function is selected, and wavelet transform is performed to decompose the wavelet coefficients into different scales;

[0023] A threshold function is applied to the wavelet coefficients to remove small-amplitude wavelet coefficients to obtain processed detail coefficients;

[0024] Wavelet reconstruction is performed using the processed detail coefficients to obtain a denoised multi-sensor data set.

[0025] Optionally, the feature data extraction steps are as follows:

[0026] The multi-sensor data set after synchronization and denoising processing is standardized and organized into an observation matrix;

[0027] Based on the observation matrix, a covariance matrix is calculated using principal component analysis (PCA) method;

[0028] Eigenvalue decomposition is performed on the covariance matrix to obtain eigenvalues;

[0029] The eigenvectors corresponding to the first N largest eigenvalues are selected as principal components, and the observation matrix after standardization is projected onto the principal components to obtain a principal component score matrix;

[0030] The principal component score matrix is input into an independent component analysis (ICA) method, and a FastICA algorithm is used to iteratively solve a separation matrix. The FastICA algorithm needs to initialize the separation matrix and iteratively update the initialized separation matrix until convergence, so as to obtain an independent component matrix;

[0031] The independent component matrix is used as an eigenvector to form extracted feature data.

[0032] Optionally, the establishment of the scene feature template library is as follows:

[0033] For long-term historical data of the fan in a healthy state, and different scenes covering strong wind-tide combined action, blade moisture absorption and composite scenes, multi-sensor data in the actual fan running process is obtained.

[0034] Still using the principal component analysis (PCA) combined with the independent component analysis (ICA) method, the eigenvectors of the data collected under each scene are extracted.

[0035] The extracted eigenvectors are analyzed by the unsupervised clustering analysis K-Means method to match the feature mode.

[0036] The eigenvectors are divided into three clusters, and the centroids of each cluster constitute a scene feature template, including a strong wind-tide combined action feature mode, a blade moisture absorption feature mode and a composite scene feature mode.

[0037] For each scene feature template, the Euclidean distance distribution of all eigenvectors in the cluster to the centroid is calculated, and a dynamic threshold is set to determine whether the data points of the new eigenvector are abnormal or not.

[0038] The statistical distribution of the data points of all eigenvectors falling into each cluster and the corresponding scene feature template is analyzed, and the data points of the abnormal scene are marked.

[0039] Optionally, the steps of identifying the scene by the cosine similarity algorithm are as follows:

[0040] The cosine similarity of the data points of the new eigenvector and each template in the scene feature template library is calculated.

[0041] The scene corresponding to the template with the highest similarity is selected as the scene where the current fan is located.

[0042] After identifying the scene, the Euclidean distance between the data points of the new eigenvector and the centroid of the scene feature template is calculated.

[0043] The Euclidean distance is compared with a dynamic threshold, and when the Euclidean distance exceeds the dynamic threshold, the scene anomaly is marked and an abnormal early warning is triggered.

[0044] Optionally, the precise scene model establishment step is as follows:

[0045] A physical modeling method and a data-driven modeling method are used to establish a wind turbine-ocean environment coupling model and a moisture absorption-mechanical property change model.

[0046] Among them, for the establishment of the wind turbine-ocean environment coupling model:

[0047] The kinetic behavior of the wind turbine under the combined action of strong wind and tide is collected, including the environmental information vector and the state vector of the wind turbine operation;

[0048] Based on the environmental information vector and the state vector, a force vector function generated by strong wind is constructed to calculate the aerodynamic load vector of the air dynamic force acting on the wind turbine, and a force vector function generated by tide is constructed to calculate the water dynamic load vector of the wave and current acting on the wind turbine foundation structure;

[0049] Based on the aerodynamic load vector and the water dynamic load vector, combined with the data-driven inertia force, damping force and elastic force of the wind turbine operation, the coupled kinetic behavior model is obtained;

[0050] For the moisture absorption-mechanical property change model:

[0051] The moisture concentration absorbed in the wind turbine blade composite material, the environmental temperature and humidity data are collected;

[0052] According to the moisture diffusion coefficient of the moisture diffusion dynamics theory, the Fick's second law is used to construct the diffusion equation of the moisture in the blade from the high concentration area to the low concentration area in the composite material, the moisture concentration field is calculated, and the boundary condition is defined according to the environmental humidity;

[0053] Based on the moisture concentration field, combined with the equivalent elastic modulus and damping coefficient of the blade material, the moisture absorption-mechanical property change model is established to obtain the mechanical property degradation relationship between the mechanical property and the moisture absorption amount;

[0054] The established wind turbine-ocean environment coupling model and moisture absorption-mechanical property change model are used to simulate the running state of the wind turbine in the scene, and the characteristic vector predicted by the scene model is output.

[0055] Optionally, the generation step of the diagnosis result is as follows:

[0056] The quantified abnormal feature index is received, combined with the extracted feature data, and according to the possible fault types and normal states in the wind turbine operation state, a state recognition framework is constructed.

[0057] The vibration analysis technology, acoustic detection technology and machine learning diagnosis algorithm are used as the evidence source BPA of fault diagnosis, and the fault diagnosis probabilities of different state types in the state recognition framework are allocated, including generating the vibration analysis evidence source by analyzing the frequency spectrum, envelope spectrum and modal parameters of the vibration signal;

[0058] The acoustic detection evidence source is generated by analyzing the amplitude, count, energy and positioning information of the acoustic emission signal;

[0059] The machine learning evidence source is generated by inputting the feature data vector and abnormal feature index extracted in real time during the operation of the fan into the probability distribution output by a pre-trained multi-class XGBoost classifier.

[0060] The vibration analysis evidence source, acoustic detection evidence source and machine learning evidence source are integrated using the combination rule of D-S evidence theory decision-level fusion to obtain a comprehensive quality function, and a processing decision is made according to the comprehensive quality function.

[0061] The diagnostic parameters of the recognition scene are dynamically adjusted using a scene adaptive mechanism, the scene characteristics are mapped with the diagnostic parameters, and a diagnostic result is generated.

[0062] Optionally, the quantization step of the abnormal feature index is as follows:

[0063] The actual measured synchronous denoised feature data is compared and analyzed with the predicted feature vector output by the accurate scene model simulation of the normal operation state of the fan, and the abnormal feature index of the low-frequency modulation index LMI, the moisture absorption influence factor MIF and the composite abnormal index CAI is quantitatively calculated.

[0064] For the low-frequency modulation index LMI, the residual signal of the difference between the real-time monitored feature data and the predicted feature vector output by the accurate scene model is subjected to Hilbert transform, the envelope line is calculated, the envelope line is subjected to frequency spectrum analysis, the frequency spectrum of the actual monitored signal and the frequency spectrum of the simulation signal of the accurate scene model are obtained, and the energy affected by the tide / wave characteristic frequency is found, and the low-frequency modulation index LMI is calculated.

[0065] For the moisture absorption influence factor MIF, the weight matrix is set according to the prior knowledge of the humidity sensitivity of different scene characteristics, and the moisture absorption influence factor MIF is calculated based on the residual signal using Mahalanobis distance.

[0066] For the composite abnormal index CAI, the composite abnormal index CAI is calculated using the weighted average method based on the abnormal indexes of the low-frequency modulation index LMI and the moisture absorption influence factor MIF.

[0067] The fan operation abnormal vibration monitoring system based on multi-sensor fusion comprises a multi-source data acquisition module of deployed sensors: a plurality of types of sensors of acceleration sensors, displacement sensors, acoustic sensors, anemometers, tidal sensors and temperature and humidity sensors are respectively deployed on the cabin, the tower drum and the blades of the fan, IEEE1588 protocol is used to synchronize the microsecond level time of all sensors, and sensor data about vibration, acoustics and environment of the fan in the coastal / sea operation state is collected in real time;

[0068] A data preprocessing and feature extraction module: receives sensor data, uses wavelet denoising combined with principal component analysis joint PCA-ICA algorithm to extract feature data;

[0069] A scene modeling and abnormal feature quantification module: constructs a scene feature template library, identifies scenes by comparing feature data, establishes a scene model by using physical modeling combined with data-driven modeling method, quantifies abnormal feature indexes, further refines scene classification, and sets an abnormal feature threshold to evaluate the type and severity of the abnormality;

[0070] A multi-technology fusion diagnosis fault module: comprehensively uses vibration analysis technology, acoustic detection technology and machine learning diagnosis algorithm to diagnose faults of abnormal feature indexes, develops a scene adaptive mechanism, adjusts diagnosis parameters according to scene characteristics, and triggers corresponding processing decisions;

[0071] A feedback optimization knowledge base management module: establishes a double-loop feedback mechanism, feeds back diagnosis results and processing decisions to each module of the system, constructs and updates a knowledge base, and then adjusts sensor data acquisition strategies, feature data extraction methods, scene model construction and abnormal threshold setting according to feedback information.

[0072] In the above technical solution, the technical effects and advantages provided by the present application are as follows:

[0073] The present application considers the actual conditions of offshore wind farms by deploying multiple types of sensors on the wind turbine, and after real-time collection of multi-source signals, fine processing and feature extraction are performed to accurately monitor and reliably identify abnormal vibration conditions of the wind turbine in the complex environment of the coastal or offshore wind farm, so that early abnormal features of the wind turbine operation can be captured in time, normal changes caused by environmental factors and real fault abnormalities can be effectively distinguished, and early fault detection is further realized to provide sufficient time for maintenance and repair of the wind turbine; by establishing a scene feature template library and quantifying abnormal feature indicators, the influence of environmental factors and real structural damage can be effectively distinguished, the false alarm rate is greatly reduced, and the adaptability of the wind turbine system in the complex environment of the coast / sea is improved; a multi-technology fusion diagnosis strategy is adopted to organically combine vibration analysis, acoustic detection and machine learning algorithms, and according to multi-evidence fusion, accurate positioning and identification of faults are realized, the accuracy of fault diagnosis is improved, and strong guarantee is provided for safe and reliable operation of the offshore wind turbine; and through a double-loop feedback optimization mechanism and a periodic model updating mechanism, the monitoring performance of the wind turbine system is improved, and the fault case database is enriched to adapt to the dynamic changes of the offshore environment. BRIEF DESCRIPTION OF DRAWINGS

[0074] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced. Obviously, the drawings in the following description only represent some embodiments described in the present application, and other drawings can also be obtained by those skilled in the art based on these drawings.

[0075] Figure 1 The flowchart of the wind turbine operation abnormal vibration monitoring method of the present application.

[0076] Figure 2 The module block diagram of the wind turbine operation abnormal vibration monitoring system of the present application. DETAILED DESCRIPTION

[0077] Example implementations will now be described more fully with reference to the accompanying drawings. Example implementations may, however, be implemented in many different forms and should not be construed as limited to the examples set forth herein; rather, these example implementations are provided so that this disclosure will be thorough and complete, and will fully convey the inventive aspects to those skilled in the art.

[0078] Example 1

[0079] The present application provides a wind turbine operation abnormal vibration monitoring method based on multi-sensor fusion as shown in Figure 1 The present application provides a wind turbine operation abnormal vibration monitoring method based on multi-sensor fusion as shown in

[0080] S1, a plurality of types of sensors are disposed on the wind turbine, including acceleration sensors installed on the nacelle, tower and blades, displacement sensors installed on the foundation, acoustic sensors installed on the blades, and anemometers, tide sensors and temperature and humidity sensors for monitoring the environment, real-time collection of sensor data, sensor data synchronization processing using IEEE1588 protocol, preprocessing using wavelet denoising, and then using principal component PCA combined with independent component ICA analysis method to extract feature data;

[0081] Specifically, the collection steps of each sensor data are as follows:

[0082] For the plurality of types of sensors disposed on the wind turbine, the number of sensor types and corresponding sensor types is counted, wherein the sensor types are acceleration sensors, displacement sensors, acoustic sensors, anemometers, tide sensors and temperature and humidity sensors, respectively labeled as ;

[0083] The total number of corresponding sensor type statistics is ;

[0084] The types and quantities of the counted plurality of types of sensors are numbered to mark each sensor disposed on the wind turbine, for example, the first acceleration sensor installed on the wind turbine is marked as , and the th acceleration sensor installed on the wind turbine is marked as ;

[0085] The first displacement sensor installed on the wind turbine is marked as , and the th displacement sensor installed on the wind turbine is marked as ;

[0086] The first acoustic sensor installed on the wind turbine is marked as , and the th acoustic sensor installed on the wind turbine is marked as ;

[0087] The first anemometer installed on the wind turbine is marked as , and the th anemometer installed on the wind turbine is marked as ;

[0088] The first tide sensor installed on the wind turbine is marked as , and the th tide sensor installed on the wind turbine is marked as ;

[0089] The first temperature sensor installed on the wind turbine is marked as a first temperature sensor installed on the fan, marked as ; ;

[0090] a first humidity sensor installed on the fan, marked as ; ; ;

[0091] starting the multi-type sensor to collect data in a continuous time to form time series of each sensor data, wherein the expression of each sensor data is , , , , , , , and , wherein represents the acceleration sensor data collected by the statistical acceleration sensors at the th sampling moment, represents the displacement sensor data collected by the statistical displacement sensors at the th sampling moment, represents the acoustic sensor data collected by the statistical acoustic sensors at the th sampling moment, represents the anemograph data collected by the statistical anemographs at the th sampling moment, represents the tidal sensor data collected by the statistical tidal sensors at the th sampling moment, represents the temperature sensor data collected by the statistical temperature sensors at the th sampling moment, represents the humidity sensor data collected by the statistical humidity sensors at the th sampling moment, represents the total number of sampling moments, represents the time stamp of the sampling point;

[0092] ​​​​​​​The sensor data is integrated into a multi-sensor data set, wherein the expression of the multi-sensor data set is , wherein, is expressed as the multi-sensor data set.

[0093] Specifically, the data synchronization processing and wavelet denoising preprocessing steps are as follows:

[0094] For each sensor, there is a time offset and network transmission delay between the local timestamp and the master clock timestamp;

[0095] The time offset and network transmission delay are calculated by using the interleaved signaling mechanism of the IEEE1588 PTP protocol for the sensor data, wherein the calculation formula of the time offset is , wherein, is expressed as the time offset, is expressed as the time when the slave clock receives the synchronization message, is expressed as the timestamp when the master clock sends the synchronization message request, is expressed as the time when the master clock receives the delay request message, is expressed as the time when the slave clock sends the delay request message; and the calculation formula of the network transmission delay is , wherein, is expressed as the network transmission delay;

[0096] Based on the time offset and network transmission delay, the sensor data of the time sequence is corrected, and the timestamps of all sensor data are unified to a unified timestamp on a master clock, which is used to eliminate the timing error caused by communication delay and sampling start time difference, and generate a time-aligned multi-sensor data set, wherein the expression of the unified timestamp is , wherein, is expressed as the corrected unified timestamp, is expressed as the local timestamp of the slave clock; and the expression of the time-aligned multi-sensor data set is , and , , , , , , , wherein, is expressed as the time-aligned multi-sensor data set, is respectively expressed as the time-aligned sensor data;

[0097] The time-aligned multi-sensor data set is subjected to wavelet denoising preprocessing, a wavelet basis function is selected, and the wavelet coefficients of different scales are decomposed through wavelet transform, wherein the expression of the wavelet coefficient is , wherein, denoted as wavelet coefficients, denoted as scale parameters, denoted as shift parameters, denoted as mother wavelet functions, denoted as complex conjugate of mother wavelet functions;

[0098] applying a threshold function to the wavelet coefficients to remove small amplitude wavelet coefficients to obtain processed detail coefficients, wherein the threshold function is expressed as , wherein, denoted as new wavelet coefficients after threshold processing, denoted as soft threshold function, denoted as threshold distinguishing signal and noise;

[0099] performing wavelet reconstruction using the processed detail coefficients to obtain a denoised multi-sensor data set, wherein the denoised multi-sensor data set is expressed as , wherein, denoted as output denoised multi-sensor data set, denoted as wavelet admissibility constant.

[0100] Specifically, the feature data extraction step is as follows:

[0101] performing standardization processing on the multi-sensor data set after synchronization and denoising processing, and organizing it into an observation matrix, wherein the observation matrix is expressed as , wherein, denoted as observation matrix, denoted as number of sampling points, denoted as the th sampling point, denoted as matrix transpose;

[0102] based on the observation matrix, calculating a covariance matrix using a principal component analysis (PCA) method, wherein the covariance matrix is calculated according to the formula , and , wherein, denoted as covariance matrix, denoted as mean vector, denoted as the th sampling point, ;

[0103] performing eigenvalue decomposition on the covariance matrix to obtain eigenvalues, wherein the covariance matrix decomposition is expressed as , wherein, denoted as eigenvalue matrix, is expressed as a feature vector matrix;

[0104] The feature vectors corresponding to the top N largest eigenvalues are selected as principal components, and the observation matrix after normalization is projected onto the principal components to obtain a principal component score matrix, wherein the expression of the principal component score matrix is , wherein, is expressed as a principal component score matrix, is expressed as a set of directions of the top N principal components, and N represents the number of retained principal components;

[0105] The principal component score matrix is input into an independent component ICA analysis method, and the goal of ICA is to find a separation matrix that satisfies the independent component matrix equal to the product of the principal component score matrix and the separation matrix;

[0106] The FastICA algorithm is used to iteratively solve the separation matrix, and the FastICA algorithm needs to initialize the separation matrix first, and iteratively update the initialized separation matrix until convergence, so as to obtain the independent component matrix, wherein the expression of the independent component matrix is , wherein, is expressed as an independent component matrix, is expressed as a separation matrix;

[0107] The independent component matrix is taken as a feature vector to form extracted feature data.

[0108] S2, a scene feature template library is established for complex marine environments, including strong wind-tide combined feature mode, blade moisture absorption feature mode, and composite scene feature mode, the extracted feature data is compared with the scene feature template library, a cosine similarity algorithm is used to identify the scene where the current fan is located, a feature threshold of an abnormal early warning mechanism is defined, and the abnormal early warning is triggered when the extracted feature data exceeds the feature threshold;

[0109] Specifically, the establishment steps of the scene feature template library are as follows:

[0110] For long-term historical data of the fan in a healthy state, and covering different scenes of strong wind-tide combination, blade moisture absorption and composite scene, multi-sensor data in the actual fan running process is obtained;

[0111] Still using the principal component PCA combined with the independent component ICA analysis method, the feature vectors of the data collected under each scene are extracted;

[0112] The extracted feature vectors are analyzed by clustering analysis combined with the unsupervised clustering analysis K-Means method to match the feature mode;

[0113] The feature vectors are divided into 3 clusters, and the centroid of each cluster constitutes a scene feature template, including a strong wind-tide combined action feature mode, a blade moisture absorption feature mode, and a composite scene feature mode.

[0114] For each scene feature template, the Euclidean distance distribution of all feature vectors in the cluster to the centroid is calculated, and a dynamic threshold is set to determine whether the data points of the new feature vector and the scene feature template are abnormal;

[0115] The statistical distribution of all feature vectors in each cluster and the corresponding scene feature template is analyzed, and the data points of the abnormal scene are marked.

[0116] Specifically, the steps of the cosine similarity algorithm for identifying the scene are as follows:

[0117] The cosine similarity of the data points of the new feature vector and each template in the scene feature template library is calculated;

[0118] The scene corresponding to the template with the highest similarity is selected as the scene where the current fan is located;

[0119] After identifying the scene, the Euclidean distance between the data points of the new feature vector and the centroid of the scene feature template is calculated;

[0120] The Euclidean distance is compared with the dynamic threshold, and when the Euclidean distance exceeds the dynamic threshold, the scene is marked as abnormal and an abnormal warning is triggered.

[0121] S3, according to the scene identified by the extracted feature data, a precise scene model is established by using physical modeling combined with data-driven modeling method, including establishing a fan-ocean environment coupling model for the strong wind-tide combined action scene, establishing a moisture absorption-mechanical property change model for the blade moisture absorption scene, and using XGBoost algorithm to establish a data-driven compensation model to correct the model error. The extracted feature data is input into the corresponding scene model to simulate the normal operating state. The actual data is compared to quantify the abnormal feature indexes, including the low frequency modulation index LMI, the moisture absorption influence factor MIF and the composite abnormal index CAI, to refine the scene classification, set the abnormal feature threshold, and determine the abnormal type and severity according to the comparison of the quantified abnormal feature indexes;

[0122] Specifically, the steps of the precise scene model establishment are as follows:

[0123] The fan-ocean environment coupling model and the moisture absorption-mechanical property change model are established by using physical modeling combined with data-driven modeling method.

[0124] Among them, for establishing the fan-ocean environment coupling model:

[0125] The dynamic behavior of the wind turbine under the combined action of strong wind and tide is collected, including the environmental information vector of wind speed, wind direction, tidal height, and effective wave height, and the state vector of the wind turbine operation of the displacement / velocity / acceleration of the tower top, cabin acceleration, blade pitch angle, and blade tip displacement, wherein the environmental information vector and the state vector of the wind turbine operation are respectively denoted as and , and , wherein is denoted as the environmental information vector, is denoted as the wind speed, is denoted as the wind direction, is denoted as the tidal height, is denoted as the effective wave height, is denoted as the transpose, , wherein is denoted as the state vector of the wind turbine operation, is denoted as the tower top displacement, is denoted as the tower top velocity, is denoted as the tower top acceleration, is denoted as the cabin acceleration, is denoted as the blade pitch angle, is denoted as the blade tip displacement;

[0126] Based on the environmental information vector and the state vector, a force vector function generated by strong wind is constructed to calculate the aerodynamic load vector of the air dynamic force acting on the wind turbine, and a force vector function generated by tide is constructed to calculate the hydrodynamic load vector of the wave and current force acting on the wind turbine foundation structure, wherein the calculation formula of the aerodynamic load vector is , wherein is denoted as the aerodynamic load vector, is denoted as the force vector function generated by strong wind;

[0127] The calculation formula of the hydrodynamic load vector is , wherein is denoted as the hydrodynamic load vector, is denoted as the force vector function generated by tide;

[0128] Based on the aerodynamic load vector and the hydrodynamic load vector, the data driving of the inertia force, the damping force, and the elastic force of the wind turbine operation is combined to obtain a coupled dynamic behavior model to describe the motion trajectory of the system under the action of internal and external forces, wherein the expression of the dynamic behavior model is , wherein is denoted as the mass matrix of the inertia force of the wind turbine operation, is denoted as the damping matrix of the damping force of the wind turbine operation, a stiffness matrix representing the elastic force of the fan operating, a second order derivative of a state vector representing the fan operating, a first order derivative of a state vector representing the fan operating;

[0129] for the hygro-mechanical property change model:

[0130] collecting the moisture concentration absorbed at the time instant, at the time instant, the moisture concentration absorbed at the time instant, the environmental temperature and humidity data, wherein the moisture concentration absorbed at the time instant, at the time instant, is calibrated as ;

[0131] a moisture diffusion coefficient according to the theory of moisture diffusion kinetics, a diffusion equation of the blade moisture inside the composite material from a high concentration area to a low concentration area is constructed using Fick's second law, a moisture concentration field is calculated, and a boundary condition is defined according to the environmental humidity, wherein the diffusion equation expression constructed by Fick's second law is , wherein, represents the diffusion law of the moisture concentration over time, represents the divergence operator, represents the theoretical moisture diffusion coefficient, represents the gradient operator of the moisture concentration;

[0132] based on the moisture concentration field, combined with the equivalent elastic modulus and damping coefficient of the blade material, a hygro-mechanical property change model is established, a mechanical property degradation relationship between the mechanical property and the moisture absorption amount is obtained, and the vibration response of the fan after absorbing moisture in the coastal humid air is described, wherein the expression of the hygro-mechanical property change model is , wherein, represents the equivalent elastic modulus, represents the elastic modulus of the material in the dry state, represents the degradation coefficient of the elastic modulus, represents the equivalent elastic modulus, represents the damping coefficient of the material in the dry state, represents the degradation coefficient of the damping;

[0133] the established fan-ocean environment coupling model and the hygro-mechanical property change model are used to simulate the operating state of the fan in the scene, and a scene model predicted feature vector is output, wherein the expression of the scene model predicted feature vector is , wherein, represents the scene model predicted feature vector.

[0134] It is further supplemented that the establishment steps of the data-driven compensation model are as follows:

[0135] The feature vector predicted by the scene model and the feature vector actually extracted from the historical data of the fan operation health state are acquired as a training sample set;

[0136] Based on the training sample set, the data-driven compensation model is trained using the XGBoost algorithm, and the prediction error value of the data-driven compensation model is output, wherein the expression of the data-driven compensation model is , and , wherein, represents the prediction error value of the model, represents the feature vector actually extracted from the historical data of the fan operation health state, represents the feature vector predicted by the scene model, represents the data-driven compensation result of the prediction error value, represents the XGBoost regression algorithm, represents the tree structure and weight parameters in the XGBoost algorithm;

[0137] The tree structure and weight parameters of the data-driven compensation model are learned and optimized by defining the number of decision trees and minimizing the objective function, and the trained data-driven compensation model is obtained, wherein the expression of the minimized objective function is , wherein, represents the minimized loss function, represents the number of decision trees, represents the data-driven compensation result output by the th decision tree, represents the regularization term of the model to prevent overfitting;

[0138] The feature vector predicted by the scene model is input into the trained data-driven compensation model, and the prediction of the error value of the compensated scene model is output, which further refines the scene model, wherein the expression of the refined scene model is , wherein, represents the refined scene model after compensating the scene model.

[0139] Specifically, the quantification steps of the abnormal feature indicators are as follows:

[0140] The actual measured synchronous denoised feature data is compared with the predicted feature vector output by the refined scene model simulating the normal operation state of the fan The abnormal characteristic indexes of the low-frequency modulation index LMI, the moisture absorption influence factor MIF and the composite abnormal index CAI are quantitatively calculated through comparative analysis;

[0141] For the low-frequency modulation index LMI, whether the low-frequency modulation strength caused by the tide / wave in the actual signal exceeds the normal range predicted by the model is described. The Hilbert transform is performed on the residual signal of the difference between the real-time monitored characteristic data and the predicted characteristic vector of the accurate scene model output, the envelope line is obtained, the frequency spectrum of the actual monitoring signal and the frequency spectrum of the accurate scene model simulation signal are obtained through frequency spectrum analysis on the envelope line, and the energy affected by the characteristic frequency of the tide / wave is found. The low-frequency modulation index LMI is calculated, wherein the calculation formula of the low-frequency modulation index LMI is , and , wherein is the low-frequency modulation index, is the characteristic frequency of the tide / wave, is the frequency bandwidth, is the power spectral density of the envelope line, is the weight function of the modulation frequency, is the envelope line of the residual signal, is the residual signal;

[0142] For the moisture absorption influence factor MIF, the degree of deviation of the blade vibration characteristic from the expected baseline under the current humidity condition is quantified. The weight matrix is set according to the prior knowledge of the humidity sensitivity of different scene characteristics, and the moisture absorption influence factor MIF is calculated based on the residual signal using the Mahalanobis distance, wherein the calculation formula of the moisture absorption influence factor MIF is , wherein is the moisture absorption influence factor, is the weight matrix set according to the prior knowledge of the humidity sensitivity of different scene characteristics, is the Mahalanobis distance calculation symbol;

[0143] For the composite abnormal index CAI, multiple abnormal indexes are integrated to calculate the overall abnormal composite score. The composite abnormal index CAI is calculated using the weighted average method based on the abnormal indexes of the low-frequency modulation index LMI and the moisture absorption influence factor MIF, wherein the calculation formula of the composite abnormal index CAI is , and , wherein is the composite abnormal index, is the weight coefficient corresponding to the low-frequency modulation index LMI, is the weight coefficient corresponding to the moisture absorption influence factor MIF.

[0144] It is further supplemented that the determination steps of the abnormal type and the severity are as follows:

[0145] The threshold values of the low-frequency modulation index, the moisture absorption influence factor and the composite abnormality index are defined as , , ;

[0146] The abnormality characteristic indexes of the quantified low-frequency modulation index LMI, the moisture absorption influence factor MIF and the composite abnormality index CAI are compared with the threshold values of the low-frequency modulation index, the moisture absorption influence factor and the composite abnormality index to determine the abnormality type, including when and , it is judged that the abnormality type is structural loosening or aerodynamic asymmetry caused by strong wind-tide modulation;

[0147] when and , it is judged that the abnormality type is material performance degradation or damage caused by blade moisture absorption;

[0148] when composite type abnormality;

[0149] The severity of the abnormality is determined according to the ratio of the abnormality characteristic index to the threshold value, and the severity level is divided according to the severity of the abnormality, including mild abnormality, moderate abnormality and severe abnormality, wherein the calculation formula of the severity of the abnormality is , wherein represents the severity of the abnormality, represents the maximum value symbol, when , it is divided into mild abnormality;

[0150] when , it is divided into moderate abnormality;

[0151] when , it is divided into severe abnormality.

[0152] S4, comprehensively use vibration analysis technology, acoustic detection technology and machine learning diagnosis algorithm to diagnose the quantified abnormality characteristics, generate precise fault positioning diagnosis results, develop scene adaptive mechanism, and automatically trigger corresponding processing decisions according to scene characteristics and adaptively adjust diagnosis parameters;

[0153] Specifically, the generation steps of the diagnosis result are as follows:

[0154] Receive the quantified abnormality characteristic index, combine the extracted feature data, list and count the possible fault types and normal states in the fan running state, and construct a state recognition framework;

[0155] The vibration analysis technology, the acoustic detection technology and the machine learning diagnosis algorithm are taken as the evidence source BPA of fault diagnosis, and the fault diagnosis probabilities of different state types in the state recognition framework are allocated, including generating the vibration analysis evidence source by analyzing the frequency spectrum, envelope spectrum and modal parameters of the vibration signal;

[0156] The acoustic detection evidence source is generated by analyzing the amplitude, count, energy and positioning information of the acoustic emission signal;

[0157] The machine learning evidence source is generated by inputting the feature data vector and the abnormal feature index extracted in real time during the operation of the fan into the probability distribution output by a pre-trained multi-class XGBooster.

[0158] The vibration analysis evidence source, the acoustic detection evidence source and the machine learning evidence source are comprehensively fused by using the combination rule of the decision-level fusion of the D-S evidence theory to obtain a comprehensive quality function, and a processing decision is made according to the comprehensive quality function.

[0159] The diagnostic parameters of the recognition scene are dynamically adjusted by using a scene adaptive mechanism, the scene features are mapped with the diagnostic parameters, and a final diagnosis result containing the determined fault type, diagnosis confidence and fault location is generated.

[0160] S5, based on the system feedback of the diagnosis result and the processing decision, a double-loop feedback mechanism is established to continuously optimize the system performance, the inner loop feedback is to monitor the fan operation result in real time to adjust the sensor data acquisition and feature data extraction, the outer loop feedback is to periodically update the model parameters and the knowledge base, and a knowledge base including a case base, a model base and a rule base is constructed, the knowledge base is analyzed and updated regularly, and the monitoring and diagnosis capabilities of the system are optimized.

[0161] Specifically, the establishment steps of the double-loop feedback mechanism are as follows:

[0162] Inner loop feedback: the data acquisition and feature extraction strategy of the front end of the system is optimized in real time according to the diagnosis confidence in the diagnosis result, including the sensor sampling rate, the data sampling time window length, the feature selection weight vector and the filtering parameter;

[0163] Outer loop feedback: the system model is periodically retrained using the historical fault data accumulated during the operation of the fan, in each update period, the model parameters are updated using the gradient descent optimization algorithm according to the feature data of the real-time fan operation, and the knowledge base is updated according to the new fault diagnosis cases and operation data.

[0164] Specifically, the construction steps of the knowledge base are as follows: the knowledge base comprises a case base, a model base and a rule base, wherein, for the case base: the fault cases occurring in the running process of the fan are stored in the case base in a structured form, and each fault case is a tuple, which includes the environment, characteristics, index data, scene, diagnosis result containing confidence, processing decision and running result after decision processing when the fault occurs, and the fault case is automatically stored in the database through the verified data in the outer loop feedback process;

[0165] For the model base: different versions of historical models under different scenes are stored, including but not limited to accurate scene models, data-driven compensation models, clustering analysis K-Means models, principal component PCA joint independent component ICA analysis models, XGBoost models and machine learning models, etc., after the model is retrained in the outer loop feedback each time, the new model and performance are stored in the model base as a new version, otherwise, the system can roll back to a historical version with better performance;

[0166] For the rule base: the logical rules of diagnosis and decision are stored, the decision tree or association rule algorithm is used to learn the frequently occurring "condition-result" patterns from the fault cases and characteristic data in the case base, generate candidate rules, and after the candidate rules are audited, corrected and confirmed by the personnel in the field, the candidate rules are formally added to the rule base, and the rule base is regularly combed, redundant rules are merged, and conflict rules are solved.

[0167] Embodiment 2

[0168] The application provides a fan running abnormal vibration monitoring system based on multi-sensor fusion as shown in Figure 2 The fan running abnormal vibration monitoring system based on multi-sensor fusion comprises a deployment sensor to collect multi-source data module: a plurality of types of sensors of acceleration sensors, displacement sensors, acoustic sensors, anemometers, tide sensors and temperature and humidity sensors are respectively arranged on the cabin, the tower drum and the blades of the fan, IEEE1588 protocol is used to synchronize the microsecond-level time of all the sensors, and sensor data about vibration, acoustics and environment of the fan in the coastal / sea running state is collected in real time;

[0169] The data preprocessing and feature extraction module: receives the sensor data, adopts wavelet denoising combined with principal component analysis joint PCA-ICA algorithm to extract characteristic data;

[0170] The scene modeling and abnormal feature quantification module: a scene characteristic template library is constructed, the scene is identified by comparing the characteristic data, a scene model is established by using physical modeling combined with data-driven modeling method, abnormal feature indexes are quantified, scene classification is further refined, and an abnormal type and severity are evaluated by setting an abnormal feature threshold value;

[0171] Multi-technology fusion diagnostic fault module: comprehensively use vibration analysis technology, acoustic detection technology and machine learning diagnostic algorithm to diagnose abnormal characteristic indicators, develop scene adaptive mechanism, adjust diagnostic parameters according to scene characteristics, and trigger corresponding processing decisions;

[0172] Feedback optimization knowledge base management module: a double-loop feedback mechanism is established to feed back diagnostic results and processing decisions to each module of the system, and after building and updating the knowledge base, the sensor data acquisition strategy, feature data extraction method, scene model construction and abnormal threshold setting are adjusted according to the feedback information.

[0173] It is further supplemented that multiple types of sensors are installed on the fan and networked to ensure that the sensors can normally collect data and transmit the data to the monitoring system.

[0174] Deploy edge computing devices to realize data acquisition and preliminary processing, and the edge computing devices can pre-process the collected data locally, reduce data transmission volume, and improve the response speed of the system.

[0175] Establish a cloud computing platform for deep data analysis and intelligent decision-making, and the cloud computing platform has strong computing and storage capabilities, and can analyze and process a large amount of data to support the decision-making of the system.

[0176] Develop a human-computer interaction interface to display monitoring results and alarm information, and the human-computer interaction interface can facilitate the operation and maintenance personnel to real-time understand the running state of the fan and timely handle abnormal conditions.

[0177] Periodically calibrate the system and update the model to ensure the accuracy and reliability of the system, and the system calibration can ensure the measurement accuracy of the sensor, and the model updating can make the system adapt to the changes of the offshore environment.

[0178] The wind turbine operation abnormal vibration monitoring system based on multi-sensor fusion provided by the embodiment of the application is realized by the above-mentioned wind turbine operation abnormal vibration monitoring method based on multi-sensor fusion, and the specific method and process of the wind turbine operation abnormal vibration monitoring system based on multi-sensor fusion are described in the above-mentioned embodiment of the wind turbine operation abnormal vibration monitoring method based on multi-sensor fusion, which will not be repeated here.

[0179] The above formulas are dimensionless values, and the formulas are obtained by software simulation of a large amount of data to obtain a formula of the nearest real situation, and the preset parameters in the formula are set by a person skilled in the art according to the actual situation.

[0180] The above-described embodiments can be implemented in part or in whole through software, hardware, firmware or any combination thereof. When implemented in software, the above-described embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When loaded and executed by a computer, the computer instructions or computer programs cause the computer to perform the processes or functions described in the embodiments of the present application. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable apparatus. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, from a website site, a computer, a server or a data center to another website site, a computer, a server or a data center through a wired or wireless (for example, infrared, wireless, microwave, etc.) manner. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, a data center, etc. containing one or more available medium collections. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state disk.

[0181] It should be understood that the size of the sequence number of each process described above in various embodiments of the present application does not mean the order of execution, and the execution order of each process should be determined by its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0182] Those of ordinary skill in the art can realize that the units and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be realized in electronic hardware, or a combination of computer software and electronic hardware. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0183] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for monitoring abnormal vibration during wind turbine operation based on multi-sensor fusion, characterized in that, Includes the following steps: S1. Deploy multiple types of sensors on the wind turbine to collect data from each sensor in real time. Use the IEEE1588 protocol to synchronize and process the sensor data. Combine wavelet denoising for preprocessing and then use principal component analysis (PCA) and independent component analysis (ICA) to extract feature data. S2. Establish a scene feature template library for complex marine environments, compare the extracted feature data with the scene feature template library, use the cosine similarity algorithm to identify the scene where the wind turbine is located, define the feature threshold of the abnormal early warning mechanism, and trigger the abnormal early warning when the extracted feature data exceeds the feature threshold. S3. Based on the scene identified by the extracted feature data, a precise scene model is established using a combination of physical modeling and data-driven modeling. The extracted feature data is input into the corresponding scene model to simulate normal operation. Abnormal feature indicators are quantified by comparing with actual data, scene classification is refined, and abnormal feature thresholds are set. The abnormality type and severity are determined based on the comparison and quantification of abnormal feature indicators. The steps for establishing the precise scene model are as follows: A coupled model of wind turbine and marine environment and a model of moisture absorption and mechanical performance change were established by using physical modeling combined with data-driven modeling. Among them, for establishing a coupled wind turbine-ocean environment model: Collect the dynamic behavior of the wind turbine under the combined action of strong wind and tide, including environmental information vectors and the state vector of wind turbine operation; Based on environmental information vectors and state vectors, a force vector function generated by strong winds is constructed to calculate the aerodynamic load vector of the aerodynamic forces acting on the wind turbine. At the same time, a force vector function generated by tidal forces is constructed to calculate the hydrodynamic load vector of the forces of waves and ocean currents acting on the wind turbine foundation structure. Based on aerodynamic load vectors and hydrodynamic load vectors, and combined with data-driven inertial forces, damping forces, and elastic forces during wind turbine operation, a coupled dynamic behavior model is obtained. For the hygroscopic-mechanical property change model: Collect data on the moisture concentration, ambient temperature, and humidity absorbed by the composite material of the wind turbine blades; Based on the theoretical water diffusion coefficient of water diffusion dynamics, Fick's second law is used to construct the diffusion equation of water in the leaf from the high concentration region to the low concentration region inside the composite material, calculate the water concentration field, and define the boundary conditions according to the ambient humidity. Based on the moisture concentration field, combined with the equivalent elastic modulus and damping coefficient of the blade material, a moisture absorption-mechanical property change model is established to obtain the mechanical property degradation relationship between mechanical properties and moisture absorption. The established wind turbine-ocean environment coupling model and moisture absorption-mechanical performance change model will simulate the operating state of the wind turbine in the scenario and output the feature vector predicted by the scenario model. S4. By comprehensively utilizing vibration analysis technology, acoustic detection technology, and machine learning diagnostic algorithms, fault diagnosis is performed on quantified abnormal features to generate diagnostic results that accurately locate the fault. A scene adaptive mechanism is developed to adaptively adjust diagnostic parameters according to scene characteristics and automatically trigger corresponding processing decisions. S5. Based on the system feedback of diagnostic results and processing decisions, establish a dual-loop feedback mechanism to continuously optimize system performance. At the same time, build a knowledge base, analyze and update the knowledge base regularly, and optimize the system's monitoring and diagnostic capabilities.

2. The method for monitoring abnormal vibration of wind turbine operation based on multi-sensor fusion according to claim 1, characterized in that, The steps for collecting data from each sensor are as follows: For the various types of sensors deployed on the wind turbine, the sensor types and the corresponding number of sensor types are counted. Among them, the sensor types are divided into acceleration sensors, displacement sensors, acoustic sensors, anemometers, tidal sensors and temperature and humidity sensors. The various types of sensors were statistically analyzed and assigned type and quantity numbers to identify each sensor deployed on the wind turbine. Multiple types of sensors are activated to collect data over a continuous period of time, forming a time series of sensor data. Data from various sensors is integrated into a multi-sensor dataset.

3. The method for monitoring abnormal vibration of wind turbine operation based on multi-sensor fusion according to claim 2, characterized in that, The steps of data synchronization processing and wavelet denoising preprocessing are as follows: For each sensor, there is a time offset and network transmission delay between the local timestamp and the master clock timestamp; The time offset and network transmission delay are calculated by using the interleaved signaling mechanism of the IEEE 1588PTP protocol on the data from each sensor. Based on time offset and network transmission delay correction of time series data from each sensor, the timestamps of all sensor data are unified to a unified timestamp on a master clock to generate a time-aligned multi-sensor dataset. Wavelet denoising preprocessing is performed on time-aligned multi-sensor datasets. Wavelet basis functions are selected and decomposed into wavelet coefficients of different scales through wavelet transform. Apply a threshold function to the wavelet coefficients to remove wavelet coefficients with small amplitudes, and obtain the processed detail coefficients. Wavelet reconstruction is performed using the processed detail coefficients to obtain the denoised multi-sensor dataset.

4. The method for monitoring abnormal vibration of wind turbine operation based on multi-sensor fusion according to claim 3, characterized in that, The steps for extracting the feature data are as follows: The multi-sensor dataset, after synchronization and denoising, is standardized and organized into an observation matrix; Based on the observation matrix, the covariance matrix is ​​calculated using principal component analysis (PCA). Eigenvalue decomposition is performed on the covariance matrix to obtain eigenvalues; Select the eigenvectors corresponding to the top N largest eigenvalues ​​as principal components, and project the standardized observation matrix onto the principal components to obtain the principal component score matrix. The principal component score matrix is ​​input into the independent component ICA analysis method, and the FastICA algorithm is used to iteratively solve the separation matrix. The FastICA algorithm requires initializing the separation matrix first, and then iteratively updating the initial separation matrix until convergence, thus obtaining the independent component matrix. The independent component matrix is ​​used as the feature vector to form the extracted feature data.

5. The method for monitoring abnormal vibration of wind turbine operation based on multi-sensor fusion according to claim 1, characterized in that, The steps for establishing the scene feature template library are as follows: Based on long-term historical data of wind turbines in a healthy state, and covering different scenarios such as combined strong wind and tidal effects, blade moisture absorption, and complex scenarios, multi-sensor data were obtained during the actual operation of wind turbines. We still use the principal component analysis (PCA) combined with independent component analysis (ICA) to extract feature vectors from the data collected for each scenario. The extracted feature vectors are clustered using the unsupervised clustering method K-Means to match feature patterns; The feature vectors are divided into three clusters, and the centroid of each cluster constitutes a scene feature template, including the strong wind-tidal combined effect feature pattern, the leaf moisture absorption feature pattern, and the composite scene feature pattern. For each scene feature template, calculate the Euclidean distance distribution from all feature vectors within the cluster to the centroid, and set a dynamic threshold to determine whether the data points of the new feature vector are abnormal with respect to the scene feature template. Analyze the statistical distribution of all feature vector data points falling into each cluster with their corresponding scene feature templates, and mark the data points of abnormal scenes.

6. The method for monitoring abnormal vibration of wind turbine operation based on multi-sensor fusion according to claim 1, characterized in that, The steps of the cosine similarity algorithm for scene recognition are as follows: Calculate the cosine similarity between the data points of the new feature vector and each template in the scene feature template library; Select the scene corresponding to the template with the highest similarity as the scene where the wind turbine is currently located; After identifying the scene, calculate the Euclidean distance between the data points of the new feature vector and the centroid of the scene feature template; The Euclidean distance is compared with a dynamic threshold. When the Euclidean distance exceeds the dynamic threshold, the scene is marked as abnormal and an abnormality warning is triggered.

7. The method for monitoring abnormal vibration of wind turbine operation based on multi-sensor fusion according to claim 1, characterized in that, The steps for generating the diagnostic results are as follows: Receive quantified abnormal feature indicators, combine them with extracted feature data, and list and statistically analyze the types of faults and normal states that may occur in the operation of the wind turbine to build a state recognition framework. Vibration analysis technology, acoustic detection technology, and machine learning diagnostic algorithms are used as evidence sources (BPA) for fault diagnosis. Fault diagnosis probabilities are assigned to different state types in the state recognition framework, including generating vibration analysis evidence sources by analyzing the spectrum, envelope spectrum, and modal parameters of vibration signals. By analyzing the amplitude, count, energy, and location information of acoustic emission signals, acoustic detection evidence sources are generated. By inputting the feature data vector and abnormal feature indicators extracted in real time during wind turbine operation into the probability distribution output by a pre-trained multi-class XGBooster, a source of machine learning evidence is generated. The decision-level fusion combination rule using DS evidence theory integrates vibration analysis evidence sources, acoustic detection evidence sources, and machine learning evidence sources to obtain a comprehensive quality function, and then formulates processing decisions based on the comprehensive quality function. The diagnostic parameters for scene recognition are dynamically adjusted using a scene adaptation mechanism, the relationship between scene features and diagnostic parameters is mapped, and diagnostic results are generated.

8. The method for monitoring abnormal vibration of wind turbine operation based on multi-sensor fusion according to claim 1, characterized in that, The quantification steps for the abnormal feature indicators are as follows: Real-time online monitoring of coastal / offshore wind turbine operation is conducted. The actual measured synchronously denoised feature data is compared and analyzed with the predicted feature vector output after simulating the normal operation of the wind turbine by a precise scenario model. Abnormal feature indicators of low frequency modulation index (LMI), moisture absorption influence factor (MIF), and composite anomaly index (CAI) are quantitatively calculated. For the low-frequency modulation index (LMI), the residual signal of the difference between the real-time monitored feature data and the predicted feature vector output by the accurate scene model is subjected to Hilbert transform to obtain the envelope. The envelope is then subjected to spectral analysis to obtain the spectrum of the actual monitored signal and the spectrum of the accurate scene model simulated signal. The energy affected by the tidal / wave characteristic frequency is found, and the low-frequency modulation index (LMI) is calculated. For the moisture absorption impact factor (MIF), a weight matrix is ​​set based on prior knowledge of the humidity sensitivity of different scene characteristics, and the moisture absorption impact factor (MIF) is calculated based on the residual signal using Mahalanobis distance. For the composite anomaly index CAI, the weighted average method is used to calculate the composite anomaly index CAI based on the low-frequency modulation index LMI and the moisture absorption effect factor MIF.

9. A wind turbine abnormal operation vibration monitoring system based on multi-sensor fusion, implemented by the wind turbine abnormal operation vibration monitoring method based on multi-sensor fusion as described in any one of claims 1-8, characterized in that, This includes a multi-source data acquisition module that deploys sensors: various types of sensors, such as acceleration sensors, displacement sensors, acoustic sensors, anemometers, tidal sensors, and temperature and humidity sensors, are deployed on the nacelle, tower, and blades of the wind turbine. The microsecond-level time of all sensors is synchronized using the IEEE1588 protocol, and sensor data on vibration, acoustics, and environment are collected in real time when the wind turbine is operating in coastal / offshore conditions. Data preprocessing and feature extraction module: Receives sensor data and extracts feature data using wavelet denoising combined with principal component analysis and PCA-ICA algorithm; Scene modeling and anomaly feature quantification module: Construct a scene feature template library, identify scenes by comparing feature data, establish scene models using physical modeling combined with data-driven modeling, quantify anomaly feature indicators, further refine scene classification, and set anomaly feature thresholds to assess anomaly type and severity; Multi-technology integrated fault diagnosis module: It comprehensively utilizes vibration analysis technology, acoustic detection technology and machine learning diagnostic algorithms to diagnose faults based on abnormal characteristic indicators, develops a scene adaptive mechanism, adjusts diagnostic parameters according to scene characteristics, and triggers corresponding processing decisions. Feedback optimization knowledge base management module: Establish a dual-loop feedback mechanism to feed back diagnostic results and processing decisions to various modules of the system. After building and updating the knowledge base, adjust sensor data acquisition strategies, feature data extraction methods, scene model construction, and anomaly threshold settings based on the feedback information.

Images